]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/ext4/xattr.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: hns3: Fix for information of phydev lost problem when down/up
[mirror_ubuntu-bionic-kernel.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 "ext4_jbd2.h"
60 #include "ext4.h"
61 #include "xattr.h"
62 #include "acl.h"
63
64 #ifdef EXT4_XATTR_DEBUG
65 # define ea_idebug(inode, fmt, ...) \
66 printk(KERN_DEBUG "inode %s:%lu: " fmt "\n", \
67 inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
68 # define ea_bdebug(bh, fmt, ...) \
69 printk(KERN_DEBUG "block %pg:%lu: " fmt "\n", \
70 bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
71 #else
72 # define ea_idebug(inode, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
73 # define ea_bdebug(bh, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
74 #endif
75
76 static void ext4_xattr_block_cache_insert(struct mb_cache *,
77 struct buffer_head *);
78 static struct buffer_head *
79 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
80 struct mb_cache_entry **);
81 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
82 size_t value_count);
83 static void ext4_xattr_rehash(struct ext4_xattr_header *);
84
85 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
86 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
87 #ifdef CONFIG_EXT4_FS_POSIX_ACL
88 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
89 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
90 #endif
91 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
92 #ifdef CONFIG_EXT4_FS_SECURITY
93 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
94 #endif
95 };
96
97 const struct xattr_handler *ext4_xattr_handlers[] = {
98 &ext4_xattr_user_handler,
99 &ext4_xattr_trusted_handler,
100 #ifdef CONFIG_EXT4_FS_POSIX_ACL
101 &posix_acl_access_xattr_handler,
102 &posix_acl_default_xattr_handler,
103 #endif
104 #ifdef CONFIG_EXT4_FS_SECURITY
105 &ext4_xattr_security_handler,
106 #endif
107 NULL
108 };
109
110 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
111 inode->i_sb->s_fs_info)->s_ea_block_cache)
112
113 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
114 inode->i_sb->s_fs_info)->s_ea_inode_cache)
115
116 static int
117 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
118 struct inode *inode);
119
120 #ifdef CONFIG_LOCKDEP
121 void ext4_xattr_inode_set_class(struct inode *ea_inode)
122 {
123 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
124 }
125 #endif
126
127 static __le32 ext4_xattr_block_csum(struct inode *inode,
128 sector_t block_nr,
129 struct ext4_xattr_header *hdr)
130 {
131 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
132 __u32 csum;
133 __le64 dsk_block_nr = cpu_to_le64(block_nr);
134 __u32 dummy_csum = 0;
135 int offset = offsetof(struct ext4_xattr_header, h_checksum);
136
137 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
138 sizeof(dsk_block_nr));
139 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
140 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
141 offset += sizeof(dummy_csum);
142 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
143 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
144
145 return cpu_to_le32(csum);
146 }
147
148 static int ext4_xattr_block_csum_verify(struct inode *inode,
149 struct buffer_head *bh)
150 {
151 struct ext4_xattr_header *hdr = BHDR(bh);
152 int ret = 1;
153
154 if (ext4_has_metadata_csum(inode->i_sb)) {
155 lock_buffer(bh);
156 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
157 bh->b_blocknr, hdr));
158 unlock_buffer(bh);
159 }
160 return ret;
161 }
162
163 static void ext4_xattr_block_csum_set(struct inode *inode,
164 struct buffer_head *bh)
165 {
166 if (ext4_has_metadata_csum(inode->i_sb))
167 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
168 bh->b_blocknr, BHDR(bh));
169 }
170
171 static inline const struct xattr_handler *
172 ext4_xattr_handler(int name_index)
173 {
174 const struct xattr_handler *handler = NULL;
175
176 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
177 handler = ext4_xattr_handler_map[name_index];
178 return handler;
179 }
180
181 static int
182 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
183 void *value_start)
184 {
185 struct ext4_xattr_entry *e = entry;
186
187 /* Find the end of the names list */
188 while (!IS_LAST_ENTRY(e)) {
189 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
190 if ((void *)next >= end)
191 return -EFSCORRUPTED;
192 e = next;
193 }
194
195 /* Check the values */
196 while (!IS_LAST_ENTRY(entry)) {
197 u32 size = le32_to_cpu(entry->e_value_size);
198
199 if (size > EXT4_XATTR_SIZE_MAX)
200 return -EFSCORRUPTED;
201
202 if (size != 0 && entry->e_value_inum == 0) {
203 u16 offs = le16_to_cpu(entry->e_value_offs);
204 void *value;
205
206 /*
207 * The value cannot overlap the names, and the value
208 * with padding cannot extend beyond 'end'. Check both
209 * the padded and unpadded sizes, since the size may
210 * overflow to 0 when adding padding.
211 */
212 if (offs > end - value_start)
213 return -EFSCORRUPTED;
214 value = value_start + offs;
215 if (value < (void *)e + sizeof(u32) ||
216 size > end - value ||
217 EXT4_XATTR_SIZE(size) > end - value)
218 return -EFSCORRUPTED;
219 }
220 entry = EXT4_XATTR_NEXT(entry);
221 }
222
223 return 0;
224 }
225
226 static inline int
227 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
228 const char *function, unsigned int line)
229 {
230 int error = -EFSCORRUPTED;
231
232 if (buffer_verified(bh))
233 return 0;
234
235 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
236 BHDR(bh)->h_blocks != cpu_to_le32(1))
237 goto errout;
238 error = -EFSBADCRC;
239 if (!ext4_xattr_block_csum_verify(inode, bh))
240 goto errout;
241 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
242 bh->b_data);
243 errout:
244 if (error)
245 __ext4_error_inode(inode, function, line, 0,
246 "corrupted xattr block %llu",
247 (unsigned long long) bh->b_blocknr);
248 else
249 set_buffer_verified(bh);
250 return error;
251 }
252
253 #define ext4_xattr_check_block(inode, bh) \
254 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
255
256
257 static int
258 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
259 void *end, const char *function, unsigned int line)
260 {
261 int error = -EFSCORRUPTED;
262
263 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
264 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
265 goto errout;
266 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
267 errout:
268 if (error)
269 __ext4_error_inode(inode, function, line, 0,
270 "corrupted in-inode xattr");
271 return error;
272 }
273
274 #define xattr_check_inode(inode, header, end) \
275 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
276
277 static int
278 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
279 void *end, int name_index, const char *name, int sorted)
280 {
281 struct ext4_xattr_entry *entry, *next;
282 size_t name_len;
283 int cmp = 1;
284
285 if (name == NULL)
286 return -EINVAL;
287 name_len = strlen(name);
288 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
289 next = EXT4_XATTR_NEXT(entry);
290 if ((void *) next >= end) {
291 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
292 return -EFSCORRUPTED;
293 }
294 cmp = name_index - entry->e_name_index;
295 if (!cmp)
296 cmp = name_len - entry->e_name_len;
297 if (!cmp)
298 cmp = memcmp(name, entry->e_name, name_len);
299 if (cmp <= 0 && (sorted || cmp == 0))
300 break;
301 }
302 *pentry = entry;
303 return cmp ? -ENODATA : 0;
304 }
305
306 static u32
307 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
308 {
309 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
310 }
311
312 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
313 {
314 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
315 ((u32)ea_inode->i_version);
316 }
317
318 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
319 {
320 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
321 ea_inode->i_version = (u32)ref_count;
322 }
323
324 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
325 {
326 return (u32)ea_inode->i_atime.tv_sec;
327 }
328
329 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
330 {
331 ea_inode->i_atime.tv_sec = hash;
332 }
333
334 /*
335 * Read the EA value from an inode.
336 */
337 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
338 {
339 int blocksize = 1 << ea_inode->i_blkbits;
340 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
341 int tail_size = (size % blocksize) ?: blocksize;
342 struct buffer_head *bhs_inline[8];
343 struct buffer_head **bhs = bhs_inline;
344 int i, ret;
345
346 if (bh_count > ARRAY_SIZE(bhs_inline)) {
347 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
348 if (!bhs)
349 return -ENOMEM;
350 }
351
352 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
353 true /* wait */, bhs);
354 if (ret)
355 goto free_bhs;
356
357 for (i = 0; i < bh_count; i++) {
358 /* There shouldn't be any holes in ea_inode. */
359 if (!bhs[i]) {
360 ret = -EFSCORRUPTED;
361 goto put_bhs;
362 }
363 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
364 i < bh_count - 1 ? blocksize : tail_size);
365 }
366 ret = 0;
367 put_bhs:
368 for (i = 0; i < bh_count; i++)
369 brelse(bhs[i]);
370 free_bhs:
371 if (bhs != bhs_inline)
372 kfree(bhs);
373 return ret;
374 }
375
376 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
377
378 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
379 u32 ea_inode_hash, struct inode **ea_inode)
380 {
381 struct inode *inode;
382 int err;
383
384 inode = ext4_iget(parent->i_sb, ea_ino);
385 if (IS_ERR(inode)) {
386 err = PTR_ERR(inode);
387 ext4_error(parent->i_sb,
388 "error while reading EA inode %lu err=%d", ea_ino,
389 err);
390 return err;
391 }
392
393 if (is_bad_inode(inode)) {
394 ext4_error(parent->i_sb,
395 "error while reading EA inode %lu is_bad_inode",
396 ea_ino);
397 err = -EIO;
398 goto error;
399 }
400
401 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
402 ext4_error(parent->i_sb,
403 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
404 ea_ino);
405 err = -EINVAL;
406 goto error;
407 }
408
409 ext4_xattr_inode_set_class(inode);
410
411 /*
412 * Check whether this is an old Lustre-style xattr inode. Lustre
413 * implementation does not have hash validation, rather it has a
414 * backpointer from ea_inode to the parent inode.
415 */
416 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
417 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
418 inode->i_generation == parent->i_generation) {
419 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
420 ext4_xattr_inode_set_ref(inode, 1);
421 } else {
422 inode_lock(inode);
423 inode->i_flags |= S_NOQUOTA;
424 inode_unlock(inode);
425 }
426
427 *ea_inode = inode;
428 return 0;
429 error:
430 iput(inode);
431 return err;
432 }
433
434 static int
435 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
436 struct ext4_xattr_entry *entry, void *buffer,
437 size_t size)
438 {
439 u32 hash;
440
441 /* Verify stored hash matches calculated hash. */
442 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
443 if (hash != ext4_xattr_inode_get_hash(ea_inode))
444 return -EFSCORRUPTED;
445
446 if (entry) {
447 __le32 e_hash, tmp_data;
448
449 /* Verify entry hash. */
450 tmp_data = cpu_to_le32(hash);
451 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
452 &tmp_data, 1);
453 if (e_hash != entry->e_hash)
454 return -EFSCORRUPTED;
455 }
456 return 0;
457 }
458
459 /*
460 * Read xattr value from the EA inode.
461 */
462 static int
463 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
464 void *buffer, size_t size)
465 {
466 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
467 struct inode *ea_inode;
468 int err;
469
470 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
471 le32_to_cpu(entry->e_hash), &ea_inode);
472 if (err) {
473 ea_inode = NULL;
474 goto out;
475 }
476
477 if (i_size_read(ea_inode) != size) {
478 ext4_warning_inode(ea_inode,
479 "ea_inode file size=%llu entry size=%zu",
480 i_size_read(ea_inode), size);
481 err = -EFSCORRUPTED;
482 goto out;
483 }
484
485 err = ext4_xattr_inode_read(ea_inode, buffer, size);
486 if (err)
487 goto out;
488
489 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
490 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
491 size);
492 if (err) {
493 ext4_warning_inode(ea_inode,
494 "EA inode hash validation failed");
495 goto out;
496 }
497
498 if (ea_inode_cache)
499 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
500 ext4_xattr_inode_get_hash(ea_inode),
501 ea_inode->i_ino, true /* reusable */);
502 }
503 out:
504 iput(ea_inode);
505 return err;
506 }
507
508 static int
509 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
510 void *buffer, size_t buffer_size)
511 {
512 struct buffer_head *bh = NULL;
513 struct ext4_xattr_entry *entry;
514 size_t size;
515 void *end;
516 int error;
517 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
518
519 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
520 name_index, name, buffer, (long)buffer_size);
521
522 error = -ENODATA;
523 if (!EXT4_I(inode)->i_file_acl)
524 goto cleanup;
525 ea_idebug(inode, "reading block %llu",
526 (unsigned long long)EXT4_I(inode)->i_file_acl);
527 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
528 if (!bh)
529 goto cleanup;
530 ea_bdebug(bh, "b_count=%d, refcount=%d",
531 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
532 error = ext4_xattr_check_block(inode, bh);
533 if (error)
534 goto cleanup;
535 ext4_xattr_block_cache_insert(ea_block_cache, bh);
536 entry = BFIRST(bh);
537 end = bh->b_data + bh->b_size;
538 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
539 if (error)
540 goto cleanup;
541 size = le32_to_cpu(entry->e_value_size);
542 error = -ERANGE;
543 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
544 goto cleanup;
545 if (buffer) {
546 if (size > buffer_size)
547 goto cleanup;
548 if (entry->e_value_inum) {
549 error = ext4_xattr_inode_get(inode, entry, buffer,
550 size);
551 if (error)
552 goto cleanup;
553 } else {
554 u16 offset = le16_to_cpu(entry->e_value_offs);
555 void *p = bh->b_data + offset;
556
557 if (unlikely(p + size > end))
558 goto cleanup;
559 memcpy(buffer, p, size);
560 }
561 }
562 error = size;
563
564 cleanup:
565 brelse(bh);
566 return error;
567 }
568
569 int
570 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
571 void *buffer, size_t buffer_size)
572 {
573 struct ext4_xattr_ibody_header *header;
574 struct ext4_xattr_entry *entry;
575 struct ext4_inode *raw_inode;
576 struct ext4_iloc iloc;
577 size_t size;
578 void *end;
579 int error;
580
581 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
582 return -ENODATA;
583 error = ext4_get_inode_loc(inode, &iloc);
584 if (error)
585 return error;
586 raw_inode = ext4_raw_inode(&iloc);
587 header = IHDR(inode, raw_inode);
588 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
589 error = xattr_check_inode(inode, header, end);
590 if (error)
591 goto cleanup;
592 entry = IFIRST(header);
593 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
594 if (error)
595 goto cleanup;
596 size = le32_to_cpu(entry->e_value_size);
597 error = -ERANGE;
598 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
599 goto cleanup;
600 if (buffer) {
601 if (size > buffer_size)
602 goto cleanup;
603 if (entry->e_value_inum) {
604 error = ext4_xattr_inode_get(inode, entry, buffer,
605 size);
606 if (error)
607 goto cleanup;
608 } else {
609 u16 offset = le16_to_cpu(entry->e_value_offs);
610 void *p = (void *)IFIRST(header) + offset;
611
612 if (unlikely(p + size > end))
613 goto cleanup;
614 memcpy(buffer, p, size);
615 }
616 }
617 error = size;
618
619 cleanup:
620 brelse(iloc.bh);
621 return error;
622 }
623
624 /*
625 * ext4_xattr_get()
626 *
627 * Copy an extended attribute into the buffer
628 * provided, or compute the buffer size required.
629 * Buffer is NULL to compute the size of the buffer required.
630 *
631 * Returns a negative error number on failure, or the number of bytes
632 * used / required on success.
633 */
634 int
635 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
636 void *buffer, size_t buffer_size)
637 {
638 int error;
639
640 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
641 return -EIO;
642
643 if (strlen(name) > 255)
644 return -ERANGE;
645
646 down_read(&EXT4_I(inode)->xattr_sem);
647 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
648 buffer_size);
649 if (error == -ENODATA)
650 error = ext4_xattr_block_get(inode, name_index, name, buffer,
651 buffer_size);
652 up_read(&EXT4_I(inode)->xattr_sem);
653 return error;
654 }
655
656 static int
657 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
658 char *buffer, size_t buffer_size)
659 {
660 size_t rest = buffer_size;
661
662 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
663 const struct xattr_handler *handler =
664 ext4_xattr_handler(entry->e_name_index);
665
666 if (handler && (!handler->list || handler->list(dentry))) {
667 const char *prefix = handler->prefix ?: handler->name;
668 size_t prefix_len = strlen(prefix);
669 size_t size = prefix_len + entry->e_name_len + 1;
670
671 if (buffer) {
672 if (size > rest)
673 return -ERANGE;
674 memcpy(buffer, prefix, prefix_len);
675 buffer += prefix_len;
676 memcpy(buffer, entry->e_name, entry->e_name_len);
677 buffer += entry->e_name_len;
678 *buffer++ = 0;
679 }
680 rest -= size;
681 }
682 }
683 return buffer_size - rest; /* total size */
684 }
685
686 static int
687 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
688 {
689 struct inode *inode = d_inode(dentry);
690 struct buffer_head *bh = NULL;
691 int error;
692
693 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
694 buffer, (long)buffer_size);
695
696 error = 0;
697 if (!EXT4_I(inode)->i_file_acl)
698 goto cleanup;
699 ea_idebug(inode, "reading block %llu",
700 (unsigned long long)EXT4_I(inode)->i_file_acl);
701 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
702 error = -EIO;
703 if (!bh)
704 goto cleanup;
705 ea_bdebug(bh, "b_count=%d, refcount=%d",
706 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
707 error = ext4_xattr_check_block(inode, bh);
708 if (error)
709 goto cleanup;
710 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
711 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
712
713 cleanup:
714 brelse(bh);
715
716 return error;
717 }
718
719 static int
720 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
721 {
722 struct inode *inode = d_inode(dentry);
723 struct ext4_xattr_ibody_header *header;
724 struct ext4_inode *raw_inode;
725 struct ext4_iloc iloc;
726 void *end;
727 int error;
728
729 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
730 return 0;
731 error = ext4_get_inode_loc(inode, &iloc);
732 if (error)
733 return error;
734 raw_inode = ext4_raw_inode(&iloc);
735 header = IHDR(inode, raw_inode);
736 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
737 error = xattr_check_inode(inode, header, end);
738 if (error)
739 goto cleanup;
740 error = ext4_xattr_list_entries(dentry, IFIRST(header),
741 buffer, buffer_size);
742
743 cleanup:
744 brelse(iloc.bh);
745 return error;
746 }
747
748 /*
749 * Inode operation listxattr()
750 *
751 * d_inode(dentry)->i_rwsem: don't care
752 *
753 * Copy a list of attribute names into the buffer
754 * provided, or compute the buffer size required.
755 * Buffer is NULL to compute the size of the buffer required.
756 *
757 * Returns a negative error number on failure, or the number of bytes
758 * used / required on success.
759 */
760 ssize_t
761 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
762 {
763 int ret, ret2;
764
765 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
766 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
767 if (ret < 0)
768 goto errout;
769 if (buffer) {
770 buffer += ret;
771 buffer_size -= ret;
772 }
773 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
774 if (ret < 0)
775 goto errout;
776 ret += ret2;
777 errout:
778 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
779 return ret;
780 }
781
782 /*
783 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
784 * not set, set it.
785 */
786 static void ext4_xattr_update_super_block(handle_t *handle,
787 struct super_block *sb)
788 {
789 if (ext4_has_feature_xattr(sb))
790 return;
791
792 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
793 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
794 ext4_set_feature_xattr(sb);
795 ext4_handle_dirty_super(handle, sb);
796 }
797 }
798
799 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
800 {
801 struct ext4_iloc iloc = { .bh = NULL };
802 struct buffer_head *bh = NULL;
803 struct ext4_inode *raw_inode;
804 struct ext4_xattr_ibody_header *header;
805 struct ext4_xattr_entry *entry;
806 qsize_t ea_inode_refs = 0;
807 void *end;
808 int ret;
809
810 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
811
812 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
813 ret = ext4_get_inode_loc(inode, &iloc);
814 if (ret)
815 goto out;
816 raw_inode = ext4_raw_inode(&iloc);
817 header = IHDR(inode, raw_inode);
818 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
819 ret = xattr_check_inode(inode, header, end);
820 if (ret)
821 goto out;
822
823 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
824 entry = EXT4_XATTR_NEXT(entry))
825 if (entry->e_value_inum)
826 ea_inode_refs++;
827 }
828
829 if (EXT4_I(inode)->i_file_acl) {
830 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
831 if (!bh) {
832 ret = -EIO;
833 goto out;
834 }
835
836 ret = ext4_xattr_check_block(inode, bh);
837 if (ret)
838 goto out;
839
840 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
841 entry = EXT4_XATTR_NEXT(entry))
842 if (entry->e_value_inum)
843 ea_inode_refs++;
844 }
845 *usage = ea_inode_refs + 1;
846 ret = 0;
847 out:
848 brelse(iloc.bh);
849 brelse(bh);
850 return ret;
851 }
852
853 static inline size_t round_up_cluster(struct inode *inode, size_t length)
854 {
855 struct super_block *sb = inode->i_sb;
856 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
857 inode->i_blkbits);
858 size_t mask = ~(cluster_size - 1);
859
860 return (length + cluster_size - 1) & mask;
861 }
862
863 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
864 {
865 int err;
866
867 err = dquot_alloc_inode(inode);
868 if (err)
869 return err;
870 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
871 if (err)
872 dquot_free_inode(inode);
873 return err;
874 }
875
876 static void ext4_xattr_inode_free_quota(struct inode *parent,
877 struct inode *ea_inode,
878 size_t len)
879 {
880 if (ea_inode &&
881 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
882 return;
883 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
884 dquot_free_inode(parent);
885 }
886
887 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
888 struct buffer_head *block_bh, size_t value_len,
889 bool is_create)
890 {
891 int credits;
892 int blocks;
893
894 /*
895 * 1) Owner inode update
896 * 2) Ref count update on old xattr block
897 * 3) new xattr block
898 * 4) block bitmap update for new xattr block
899 * 5) group descriptor for new xattr block
900 * 6) block bitmap update for old xattr block
901 * 7) group descriptor for old block
902 *
903 * 6 & 7 can happen if we have two racing threads T_a and T_b
904 * which are each trying to set an xattr on inodes I_a and I_b
905 * which were both initially sharing an xattr block.
906 */
907 credits = 7;
908
909 /* Quota updates. */
910 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
911
912 /*
913 * In case of inline data, we may push out the data to a block,
914 * so we need to reserve credits for this eventuality
915 */
916 if (inode && ext4_has_inline_data(inode))
917 credits += ext4_writepage_trans_blocks(inode) + 1;
918
919 /* We are done if ea_inode feature is not enabled. */
920 if (!ext4_has_feature_ea_inode(sb))
921 return credits;
922
923 /* New ea_inode, inode map, block bitmap, group descriptor. */
924 credits += 4;
925
926 /* Data blocks. */
927 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
928
929 /* Indirection block or one level of extent tree. */
930 blocks += 1;
931
932 /* Block bitmap and group descriptor updates for each block. */
933 credits += blocks * 2;
934
935 /* Blocks themselves. */
936 credits += blocks;
937
938 if (!is_create) {
939 /* Dereference ea_inode holding old xattr value.
940 * Old ea_inode, inode map, block bitmap, group descriptor.
941 */
942 credits += 4;
943
944 /* Data blocks for old ea_inode. */
945 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
946
947 /* Indirection block or one level of extent tree for old
948 * ea_inode.
949 */
950 blocks += 1;
951
952 /* Block bitmap and group descriptor updates for each block. */
953 credits += blocks * 2;
954 }
955
956 /* We may need to clone the existing xattr block in which case we need
957 * to increment ref counts for existing ea_inodes referenced by it.
958 */
959 if (block_bh) {
960 struct ext4_xattr_entry *entry = BFIRST(block_bh);
961
962 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
963 if (entry->e_value_inum)
964 /* Ref count update on ea_inode. */
965 credits += 1;
966 }
967 return credits;
968 }
969
970 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
971 int credits, struct buffer_head *bh,
972 bool dirty, bool block_csum)
973 {
974 int error;
975
976 if (!ext4_handle_valid(handle))
977 return 0;
978
979 if (handle->h_buffer_credits >= credits)
980 return 0;
981
982 error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
983 if (!error)
984 return 0;
985 if (error < 0) {
986 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
987 return error;
988 }
989
990 if (bh && dirty) {
991 if (block_csum)
992 ext4_xattr_block_csum_set(inode, bh);
993 error = ext4_handle_dirty_metadata(handle, NULL, bh);
994 if (error) {
995 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
996 error);
997 return error;
998 }
999 }
1000
1001 error = ext4_journal_restart(handle, credits);
1002 if (error) {
1003 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1004 return error;
1005 }
1006
1007 if (bh) {
1008 error = ext4_journal_get_write_access(handle, bh);
1009 if (error) {
1010 ext4_warning(inode->i_sb,
1011 "Get write access failed (error %d)",
1012 error);
1013 return error;
1014 }
1015 }
1016 return 0;
1017 }
1018
1019 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1020 int ref_change)
1021 {
1022 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1023 struct ext4_iloc iloc;
1024 s64 ref_count;
1025 u32 hash;
1026 int ret;
1027
1028 inode_lock(ea_inode);
1029
1030 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1031 if (ret) {
1032 iloc.bh = NULL;
1033 goto out;
1034 }
1035
1036 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1037 ref_count += ref_change;
1038 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1039
1040 if (ref_change > 0) {
1041 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1042 ea_inode->i_ino, ref_count);
1043
1044 if (ref_count == 1) {
1045 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1046 ea_inode->i_ino, ea_inode->i_nlink);
1047
1048 set_nlink(ea_inode, 1);
1049 ext4_orphan_del(handle, ea_inode);
1050
1051 if (ea_inode_cache) {
1052 hash = ext4_xattr_inode_get_hash(ea_inode);
1053 mb_cache_entry_create(ea_inode_cache,
1054 GFP_NOFS, hash,
1055 ea_inode->i_ino,
1056 true /* reusable */);
1057 }
1058 }
1059 } else {
1060 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1061 ea_inode->i_ino, ref_count);
1062
1063 if (ref_count == 0) {
1064 WARN_ONCE(ea_inode->i_nlink != 1,
1065 "EA inode %lu i_nlink=%u",
1066 ea_inode->i_ino, ea_inode->i_nlink);
1067
1068 clear_nlink(ea_inode);
1069 ext4_orphan_add(handle, ea_inode);
1070
1071 if (ea_inode_cache) {
1072 hash = ext4_xattr_inode_get_hash(ea_inode);
1073 mb_cache_entry_delete(ea_inode_cache, hash,
1074 ea_inode->i_ino);
1075 }
1076 }
1077 }
1078
1079 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1080 iloc.bh = NULL;
1081 if (ret)
1082 ext4_warning_inode(ea_inode,
1083 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1084 out:
1085 brelse(iloc.bh);
1086 inode_unlock(ea_inode);
1087 return ret;
1088 }
1089
1090 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1091 {
1092 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1093 }
1094
1095 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1096 {
1097 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1098 }
1099
1100 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1101 struct ext4_xattr_entry *first)
1102 {
1103 struct inode *ea_inode;
1104 struct ext4_xattr_entry *entry;
1105 struct ext4_xattr_entry *failed_entry;
1106 unsigned int ea_ino;
1107 int err, saved_err;
1108
1109 for (entry = first; !IS_LAST_ENTRY(entry);
1110 entry = EXT4_XATTR_NEXT(entry)) {
1111 if (!entry->e_value_inum)
1112 continue;
1113 ea_ino = le32_to_cpu(entry->e_value_inum);
1114 err = ext4_xattr_inode_iget(parent, ea_ino,
1115 le32_to_cpu(entry->e_hash),
1116 &ea_inode);
1117 if (err)
1118 goto cleanup;
1119 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1120 if (err) {
1121 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1122 iput(ea_inode);
1123 goto cleanup;
1124 }
1125 iput(ea_inode);
1126 }
1127 return 0;
1128
1129 cleanup:
1130 saved_err = err;
1131 failed_entry = entry;
1132
1133 for (entry = first; entry != failed_entry;
1134 entry = EXT4_XATTR_NEXT(entry)) {
1135 if (!entry->e_value_inum)
1136 continue;
1137 ea_ino = le32_to_cpu(entry->e_value_inum);
1138 err = ext4_xattr_inode_iget(parent, ea_ino,
1139 le32_to_cpu(entry->e_hash),
1140 &ea_inode);
1141 if (err) {
1142 ext4_warning(parent->i_sb,
1143 "cleanup ea_ino %u iget error %d", ea_ino,
1144 err);
1145 continue;
1146 }
1147 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1148 if (err)
1149 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1150 err);
1151 iput(ea_inode);
1152 }
1153 return saved_err;
1154 }
1155
1156 static void
1157 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1158 struct buffer_head *bh,
1159 struct ext4_xattr_entry *first, bool block_csum,
1160 struct ext4_xattr_inode_array **ea_inode_array,
1161 int extra_credits, bool skip_quota)
1162 {
1163 struct inode *ea_inode;
1164 struct ext4_xattr_entry *entry;
1165 bool dirty = false;
1166 unsigned int ea_ino;
1167 int err;
1168 int credits;
1169
1170 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1171 credits = 2 + extra_credits;
1172
1173 for (entry = first; !IS_LAST_ENTRY(entry);
1174 entry = EXT4_XATTR_NEXT(entry)) {
1175 if (!entry->e_value_inum)
1176 continue;
1177 ea_ino = le32_to_cpu(entry->e_value_inum);
1178 err = ext4_xattr_inode_iget(parent, ea_ino,
1179 le32_to_cpu(entry->e_hash),
1180 &ea_inode);
1181 if (err)
1182 continue;
1183
1184 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1185 if (err) {
1186 ext4_warning_inode(ea_inode,
1187 "Expand inode array err=%d", err);
1188 iput(ea_inode);
1189 continue;
1190 }
1191
1192 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1193 dirty, block_csum);
1194 if (err) {
1195 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1196 err);
1197 continue;
1198 }
1199
1200 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1201 if (err) {
1202 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1203 err);
1204 continue;
1205 }
1206
1207 if (!skip_quota)
1208 ext4_xattr_inode_free_quota(parent, ea_inode,
1209 le32_to_cpu(entry->e_value_size));
1210
1211 /*
1212 * Forget about ea_inode within the same transaction that
1213 * decrements the ref count. This avoids duplicate decrements in
1214 * case the rest of the work spills over to subsequent
1215 * transactions.
1216 */
1217 entry->e_value_inum = 0;
1218 entry->e_value_size = 0;
1219
1220 dirty = true;
1221 }
1222
1223 if (dirty) {
1224 /*
1225 * Note that we are deliberately skipping csum calculation for
1226 * the final update because we do not expect any journal
1227 * restarts until xattr block is freed.
1228 */
1229
1230 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1231 if (err)
1232 ext4_warning_inode(parent,
1233 "handle dirty metadata err=%d", err);
1234 }
1235 }
1236
1237 /*
1238 * Release the xattr block BH: If the reference count is > 1, decrement it;
1239 * otherwise free the block.
1240 */
1241 static void
1242 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1243 struct buffer_head *bh,
1244 struct ext4_xattr_inode_array **ea_inode_array,
1245 int extra_credits)
1246 {
1247 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1248 u32 hash, ref;
1249 int error = 0;
1250
1251 BUFFER_TRACE(bh, "get_write_access");
1252 error = ext4_journal_get_write_access(handle, bh);
1253 if (error)
1254 goto out;
1255
1256 lock_buffer(bh);
1257 hash = le32_to_cpu(BHDR(bh)->h_hash);
1258 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1259 if (ref == 1) {
1260 ea_bdebug(bh, "refcount now=0; freeing");
1261 /*
1262 * This must happen under buffer lock for
1263 * ext4_xattr_block_set() to reliably detect freed block
1264 */
1265 if (ea_block_cache)
1266 mb_cache_entry_delete(ea_block_cache, hash,
1267 bh->b_blocknr);
1268 get_bh(bh);
1269 unlock_buffer(bh);
1270
1271 if (ext4_has_feature_ea_inode(inode->i_sb))
1272 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1273 BFIRST(bh),
1274 true /* block_csum */,
1275 ea_inode_array,
1276 extra_credits,
1277 true /* skip_quota */);
1278 ext4_free_blocks(handle, inode, bh, 0, 1,
1279 EXT4_FREE_BLOCKS_METADATA |
1280 EXT4_FREE_BLOCKS_FORGET);
1281 } else {
1282 ref--;
1283 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1284 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1285 struct mb_cache_entry *ce;
1286
1287 if (ea_block_cache) {
1288 ce = mb_cache_entry_get(ea_block_cache, hash,
1289 bh->b_blocknr);
1290 if (ce) {
1291 ce->e_reusable = 1;
1292 mb_cache_entry_put(ea_block_cache, ce);
1293 }
1294 }
1295 }
1296
1297 ext4_xattr_block_csum_set(inode, bh);
1298 /*
1299 * Beware of this ugliness: Releasing of xattr block references
1300 * from different inodes can race and so we have to protect
1301 * from a race where someone else frees the block (and releases
1302 * its journal_head) before we are done dirtying the buffer. In
1303 * nojournal mode this race is harmless and we actually cannot
1304 * call ext4_handle_dirty_metadata() with locked buffer as
1305 * that function can call sync_dirty_buffer() so for that case
1306 * we handle the dirtying after unlocking the buffer.
1307 */
1308 if (ext4_handle_valid(handle))
1309 error = ext4_handle_dirty_metadata(handle, inode, bh);
1310 unlock_buffer(bh);
1311 if (!ext4_handle_valid(handle))
1312 error = ext4_handle_dirty_metadata(handle, inode, bh);
1313 if (IS_SYNC(inode))
1314 ext4_handle_sync(handle);
1315 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1316 ea_bdebug(bh, "refcount now=%d; releasing",
1317 le32_to_cpu(BHDR(bh)->h_refcount));
1318 }
1319 out:
1320 ext4_std_error(inode->i_sb, error);
1321 return;
1322 }
1323
1324 /*
1325 * Find the available free space for EAs. This also returns the total number of
1326 * bytes used by EA entries.
1327 */
1328 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1329 size_t *min_offs, void *base, int *total)
1330 {
1331 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1332 if (!last->e_value_inum && last->e_value_size) {
1333 size_t offs = le16_to_cpu(last->e_value_offs);
1334 if (offs < *min_offs)
1335 *min_offs = offs;
1336 }
1337 if (total)
1338 *total += EXT4_XATTR_LEN(last->e_name_len);
1339 }
1340 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1341 }
1342
1343 /*
1344 * Write the value of the EA in an inode.
1345 */
1346 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1347 const void *buf, int bufsize)
1348 {
1349 struct buffer_head *bh = NULL;
1350 unsigned long block = 0;
1351 int blocksize = ea_inode->i_sb->s_blocksize;
1352 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1353 int csize, wsize = 0;
1354 int ret = 0;
1355 int retries = 0;
1356
1357 retry:
1358 while (ret >= 0 && ret < max_blocks) {
1359 struct ext4_map_blocks map;
1360 map.m_lblk = block += ret;
1361 map.m_len = max_blocks -= ret;
1362
1363 ret = ext4_map_blocks(handle, ea_inode, &map,
1364 EXT4_GET_BLOCKS_CREATE);
1365 if (ret <= 0) {
1366 ext4_mark_inode_dirty(handle, ea_inode);
1367 if (ret == -ENOSPC &&
1368 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1369 ret = 0;
1370 goto retry;
1371 }
1372 break;
1373 }
1374 }
1375
1376 if (ret < 0)
1377 return ret;
1378
1379 block = 0;
1380 while (wsize < bufsize) {
1381 if (bh != NULL)
1382 brelse(bh);
1383 csize = (bufsize - wsize) > blocksize ? blocksize :
1384 bufsize - wsize;
1385 bh = ext4_getblk(handle, ea_inode, block, 0);
1386 if (IS_ERR(bh))
1387 return PTR_ERR(bh);
1388 ret = ext4_journal_get_write_access(handle, bh);
1389 if (ret)
1390 goto out;
1391
1392 memcpy(bh->b_data, buf, csize);
1393 set_buffer_uptodate(bh);
1394 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1395
1396 buf += csize;
1397 wsize += csize;
1398 block += 1;
1399 }
1400
1401 inode_lock(ea_inode);
1402 i_size_write(ea_inode, wsize);
1403 ext4_update_i_disksize(ea_inode, wsize);
1404 inode_unlock(ea_inode);
1405
1406 ext4_mark_inode_dirty(handle, ea_inode);
1407
1408 out:
1409 brelse(bh);
1410
1411 return ret;
1412 }
1413
1414 /*
1415 * Create an inode to store the value of a large EA.
1416 */
1417 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1418 struct inode *inode, u32 hash)
1419 {
1420 struct inode *ea_inode = NULL;
1421 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1422 int err;
1423
1424 /*
1425 * Let the next inode be the goal, so we try and allocate the EA inode
1426 * in the same group, or nearby one.
1427 */
1428 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1429 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1430 EXT4_EA_INODE_FL);
1431 if (!IS_ERR(ea_inode)) {
1432 ea_inode->i_op = &ext4_file_inode_operations;
1433 ea_inode->i_fop = &ext4_file_operations;
1434 ext4_set_aops(ea_inode);
1435 ext4_xattr_inode_set_class(ea_inode);
1436 unlock_new_inode(ea_inode);
1437 ext4_xattr_inode_set_ref(ea_inode, 1);
1438 ext4_xattr_inode_set_hash(ea_inode, hash);
1439 err = ext4_mark_inode_dirty(handle, ea_inode);
1440 if (!err)
1441 err = ext4_inode_attach_jinode(ea_inode);
1442 if (err) {
1443 iput(ea_inode);
1444 return ERR_PTR(err);
1445 }
1446
1447 /*
1448 * Xattr inodes are shared therefore quota charging is performed
1449 * at a higher level.
1450 */
1451 dquot_free_inode(ea_inode);
1452 dquot_drop(ea_inode);
1453 inode_lock(ea_inode);
1454 ea_inode->i_flags |= S_NOQUOTA;
1455 inode_unlock(ea_inode);
1456 }
1457
1458 return ea_inode;
1459 }
1460
1461 static struct inode *
1462 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1463 size_t value_len, u32 hash)
1464 {
1465 struct inode *ea_inode;
1466 struct mb_cache_entry *ce;
1467 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1468 void *ea_data;
1469
1470 if (!ea_inode_cache)
1471 return NULL;
1472
1473 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1474 if (!ce)
1475 return NULL;
1476
1477 ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1478 if (!ea_data) {
1479 mb_cache_entry_put(ea_inode_cache, ce);
1480 return NULL;
1481 }
1482
1483 while (ce) {
1484 ea_inode = ext4_iget(inode->i_sb, ce->e_value);
1485 if (!IS_ERR(ea_inode) &&
1486 !is_bad_inode(ea_inode) &&
1487 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1488 i_size_read(ea_inode) == value_len &&
1489 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1490 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1491 value_len) &&
1492 !memcmp(value, ea_data, value_len)) {
1493 mb_cache_entry_touch(ea_inode_cache, ce);
1494 mb_cache_entry_put(ea_inode_cache, ce);
1495 kvfree(ea_data);
1496 return ea_inode;
1497 }
1498
1499 if (!IS_ERR(ea_inode))
1500 iput(ea_inode);
1501 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1502 }
1503 kvfree(ea_data);
1504 return NULL;
1505 }
1506
1507 /*
1508 * Add value of the EA in an inode.
1509 */
1510 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1511 const void *value, size_t value_len,
1512 struct inode **ret_inode)
1513 {
1514 struct inode *ea_inode;
1515 u32 hash;
1516 int err;
1517
1518 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1519 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1520 if (ea_inode) {
1521 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1522 if (err) {
1523 iput(ea_inode);
1524 return err;
1525 }
1526
1527 *ret_inode = ea_inode;
1528 return 0;
1529 }
1530
1531 /* Create an inode for the EA value */
1532 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1533 if (IS_ERR(ea_inode))
1534 return PTR_ERR(ea_inode);
1535
1536 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1537 if (err) {
1538 ext4_xattr_inode_dec_ref(handle, ea_inode);
1539 iput(ea_inode);
1540 return err;
1541 }
1542
1543 if (EA_INODE_CACHE(inode))
1544 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1545 ea_inode->i_ino, true /* reusable */);
1546
1547 *ret_inode = ea_inode;
1548 return 0;
1549 }
1550
1551 /*
1552 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1553 * feature is enabled.
1554 */
1555 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1556
1557 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1558 struct ext4_xattr_search *s,
1559 handle_t *handle, struct inode *inode,
1560 bool is_block)
1561 {
1562 struct ext4_xattr_entry *last;
1563 struct ext4_xattr_entry *here = s->here;
1564 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1565 int in_inode = i->in_inode;
1566 struct inode *old_ea_inode = NULL;
1567 struct inode *new_ea_inode = NULL;
1568 size_t old_size, new_size;
1569 int ret;
1570
1571 /* Space used by old and new values. */
1572 old_size = (!s->not_found && !here->e_value_inum) ?
1573 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1574 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1575
1576 /*
1577 * Optimization for the simple case when old and new values have the
1578 * same padded sizes. Not applicable if external inodes are involved.
1579 */
1580 if (new_size && new_size == old_size) {
1581 size_t offs = le16_to_cpu(here->e_value_offs);
1582 void *val = s->base + offs;
1583
1584 here->e_value_size = cpu_to_le32(i->value_len);
1585 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1586 memset(val, 0, new_size);
1587 } else {
1588 memcpy(val, i->value, i->value_len);
1589 /* Clear padding bytes. */
1590 memset(val + i->value_len, 0, new_size - i->value_len);
1591 }
1592 goto update_hash;
1593 }
1594
1595 /* Compute min_offs and last. */
1596 last = s->first;
1597 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1598 if (!last->e_value_inum && last->e_value_size) {
1599 size_t offs = le16_to_cpu(last->e_value_offs);
1600 if (offs < min_offs)
1601 min_offs = offs;
1602 }
1603 }
1604
1605 /* Check whether we have enough space. */
1606 if (i->value) {
1607 size_t free;
1608
1609 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1610 if (!s->not_found)
1611 free += EXT4_XATTR_LEN(name_len) + old_size;
1612
1613 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1614 ret = -ENOSPC;
1615 goto out;
1616 }
1617
1618 /*
1619 * If storing the value in an external inode is an option,
1620 * reserve space for xattr entries/names in the external
1621 * attribute block so that a long value does not occupy the
1622 * whole space and prevent futher entries being added.
1623 */
1624 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1625 new_size && is_block &&
1626 (min_offs + old_size - new_size) <
1627 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1628 ret = -ENOSPC;
1629 goto out;
1630 }
1631 }
1632
1633 /*
1634 * Getting access to old and new ea inodes is subject to failures.
1635 * Finish that work before doing any modifications to the xattr data.
1636 */
1637 if (!s->not_found && here->e_value_inum) {
1638 ret = ext4_xattr_inode_iget(inode,
1639 le32_to_cpu(here->e_value_inum),
1640 le32_to_cpu(here->e_hash),
1641 &old_ea_inode);
1642 if (ret) {
1643 old_ea_inode = NULL;
1644 goto out;
1645 }
1646 }
1647 if (i->value && in_inode) {
1648 WARN_ON_ONCE(!i->value_len);
1649
1650 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1651 if (ret)
1652 goto out;
1653
1654 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1655 i->value_len,
1656 &new_ea_inode);
1657 if (ret) {
1658 new_ea_inode = NULL;
1659 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1660 goto out;
1661 }
1662 }
1663
1664 if (old_ea_inode) {
1665 /* We are ready to release ref count on the old_ea_inode. */
1666 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1667 if (ret) {
1668 /* Release newly required ref count on new_ea_inode. */
1669 if (new_ea_inode) {
1670 int err;
1671
1672 err = ext4_xattr_inode_dec_ref(handle,
1673 new_ea_inode);
1674 if (err)
1675 ext4_warning_inode(new_ea_inode,
1676 "dec ref new_ea_inode err=%d",
1677 err);
1678 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1679 i->value_len);
1680 }
1681 goto out;
1682 }
1683
1684 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1685 le32_to_cpu(here->e_value_size));
1686 }
1687
1688 /* No failures allowed past this point. */
1689
1690 if (!s->not_found && here->e_value_size && here->e_value_offs) {
1691 /* Remove the old value. */
1692 void *first_val = s->base + min_offs;
1693 size_t offs = le16_to_cpu(here->e_value_offs);
1694 void *val = s->base + offs;
1695
1696 memmove(first_val + old_size, first_val, val - first_val);
1697 memset(first_val, 0, old_size);
1698 min_offs += old_size;
1699
1700 /* Adjust all value offsets. */
1701 last = s->first;
1702 while (!IS_LAST_ENTRY(last)) {
1703 size_t o = le16_to_cpu(last->e_value_offs);
1704
1705 if (!last->e_value_inum &&
1706 last->e_value_size && o < offs)
1707 last->e_value_offs = cpu_to_le16(o + old_size);
1708 last = EXT4_XATTR_NEXT(last);
1709 }
1710 }
1711
1712 if (!i->value) {
1713 /* Remove old name. */
1714 size_t size = EXT4_XATTR_LEN(name_len);
1715
1716 last = ENTRY((void *)last - size);
1717 memmove(here, (void *)here + size,
1718 (void *)last - (void *)here + sizeof(__u32));
1719 memset(last, 0, size);
1720 } else if (s->not_found) {
1721 /* Insert new name. */
1722 size_t size = EXT4_XATTR_LEN(name_len);
1723 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1724
1725 memmove((void *)here + size, here, rest);
1726 memset(here, 0, size);
1727 here->e_name_index = i->name_index;
1728 here->e_name_len = name_len;
1729 memcpy(here->e_name, i->name, name_len);
1730 } else {
1731 /* This is an update, reset value info. */
1732 here->e_value_inum = 0;
1733 here->e_value_offs = 0;
1734 here->e_value_size = 0;
1735 }
1736
1737 if (i->value) {
1738 /* Insert new value. */
1739 if (in_inode) {
1740 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1741 } else if (i->value_len) {
1742 void *val = s->base + min_offs - new_size;
1743
1744 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1745 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1746 memset(val, 0, new_size);
1747 } else {
1748 memcpy(val, i->value, i->value_len);
1749 /* Clear padding bytes. */
1750 memset(val + i->value_len, 0,
1751 new_size - i->value_len);
1752 }
1753 }
1754 here->e_value_size = cpu_to_le32(i->value_len);
1755 }
1756
1757 update_hash:
1758 if (i->value) {
1759 __le32 hash = 0;
1760
1761 /* Entry hash calculation. */
1762 if (in_inode) {
1763 __le32 crc32c_hash;
1764
1765 /*
1766 * Feed crc32c hash instead of the raw value for entry
1767 * hash calculation. This is to avoid walking
1768 * potentially long value buffer again.
1769 */
1770 crc32c_hash = cpu_to_le32(
1771 ext4_xattr_inode_get_hash(new_ea_inode));
1772 hash = ext4_xattr_hash_entry(here->e_name,
1773 here->e_name_len,
1774 &crc32c_hash, 1);
1775 } else if (is_block) {
1776 __le32 *value = s->base + le16_to_cpu(
1777 here->e_value_offs);
1778
1779 hash = ext4_xattr_hash_entry(here->e_name,
1780 here->e_name_len, value,
1781 new_size >> 2);
1782 }
1783 here->e_hash = hash;
1784 }
1785
1786 if (is_block)
1787 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1788
1789 ret = 0;
1790 out:
1791 iput(old_ea_inode);
1792 iput(new_ea_inode);
1793 return ret;
1794 }
1795
1796 struct ext4_xattr_block_find {
1797 struct ext4_xattr_search s;
1798 struct buffer_head *bh;
1799 };
1800
1801 static int
1802 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1803 struct ext4_xattr_block_find *bs)
1804 {
1805 struct super_block *sb = inode->i_sb;
1806 int error;
1807
1808 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1809 i->name_index, i->name, i->value, (long)i->value_len);
1810
1811 if (EXT4_I(inode)->i_file_acl) {
1812 /* The inode already has an extended attribute block. */
1813 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
1814 error = -EIO;
1815 if (!bs->bh)
1816 goto cleanup;
1817 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1818 atomic_read(&(bs->bh->b_count)),
1819 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1820 error = ext4_xattr_check_block(inode, bs->bh);
1821 if (error)
1822 goto cleanup;
1823 /* Find the named attribute. */
1824 bs->s.base = BHDR(bs->bh);
1825 bs->s.first = BFIRST(bs->bh);
1826 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1827 bs->s.here = bs->s.first;
1828 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1829 i->name_index, i->name, 1);
1830 if (error && error != -ENODATA)
1831 goto cleanup;
1832 bs->s.not_found = error;
1833 }
1834 error = 0;
1835
1836 cleanup:
1837 return error;
1838 }
1839
1840 static int
1841 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1842 struct ext4_xattr_info *i,
1843 struct ext4_xattr_block_find *bs)
1844 {
1845 struct super_block *sb = inode->i_sb;
1846 struct buffer_head *new_bh = NULL;
1847 struct ext4_xattr_search s_copy = bs->s;
1848 struct ext4_xattr_search *s = &s_copy;
1849 struct mb_cache_entry *ce = NULL;
1850 int error = 0;
1851 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1852 struct inode *ea_inode = NULL, *tmp_inode;
1853 size_t old_ea_inode_quota = 0;
1854 unsigned int ea_ino;
1855
1856
1857 #define header(x) ((struct ext4_xattr_header *)(x))
1858
1859 if (s->base) {
1860 BUFFER_TRACE(bs->bh, "get_write_access");
1861 error = ext4_journal_get_write_access(handle, bs->bh);
1862 if (error)
1863 goto cleanup;
1864 lock_buffer(bs->bh);
1865
1866 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1867 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1868
1869 /*
1870 * This must happen under buffer lock for
1871 * ext4_xattr_block_set() to reliably detect modified
1872 * block
1873 */
1874 if (ea_block_cache)
1875 mb_cache_entry_delete(ea_block_cache, hash,
1876 bs->bh->b_blocknr);
1877 ea_bdebug(bs->bh, "modifying in-place");
1878 error = ext4_xattr_set_entry(i, s, handle, inode,
1879 true /* is_block */);
1880 ext4_xattr_block_csum_set(inode, bs->bh);
1881 unlock_buffer(bs->bh);
1882 if (error == -EFSCORRUPTED)
1883 goto bad_block;
1884 if (!error)
1885 error = ext4_handle_dirty_metadata(handle,
1886 inode,
1887 bs->bh);
1888 if (error)
1889 goto cleanup;
1890 goto inserted;
1891 } else {
1892 int offset = (char *)s->here - bs->bh->b_data;
1893
1894 unlock_buffer(bs->bh);
1895 ea_bdebug(bs->bh, "cloning");
1896 s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1897 error = -ENOMEM;
1898 if (s->base == NULL)
1899 goto cleanup;
1900 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1901 s->first = ENTRY(header(s->base)+1);
1902 header(s->base)->h_refcount = cpu_to_le32(1);
1903 s->here = ENTRY(s->base + offset);
1904 s->end = s->base + bs->bh->b_size;
1905
1906 /*
1907 * If existing entry points to an xattr inode, we need
1908 * to prevent ext4_xattr_set_entry() from decrementing
1909 * ref count on it because the reference belongs to the
1910 * original block. In this case, make the entry look
1911 * like it has an empty value.
1912 */
1913 if (!s->not_found && s->here->e_value_inum) {
1914 ea_ino = le32_to_cpu(s->here->e_value_inum);
1915 error = ext4_xattr_inode_iget(inode, ea_ino,
1916 le32_to_cpu(s->here->e_hash),
1917 &tmp_inode);
1918 if (error)
1919 goto cleanup;
1920
1921 if (!ext4_test_inode_state(tmp_inode,
1922 EXT4_STATE_LUSTRE_EA_INODE)) {
1923 /*
1924 * Defer quota free call for previous
1925 * inode until success is guaranteed.
1926 */
1927 old_ea_inode_quota = le32_to_cpu(
1928 s->here->e_value_size);
1929 }
1930 iput(tmp_inode);
1931
1932 s->here->e_value_inum = 0;
1933 s->here->e_value_size = 0;
1934 }
1935 }
1936 } else {
1937 /* Allocate a buffer where we construct the new block. */
1938 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1939 /* assert(header == s->base) */
1940 error = -ENOMEM;
1941 if (s->base == NULL)
1942 goto cleanup;
1943 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1944 header(s->base)->h_blocks = cpu_to_le32(1);
1945 header(s->base)->h_refcount = cpu_to_le32(1);
1946 s->first = ENTRY(header(s->base)+1);
1947 s->here = ENTRY(header(s->base)+1);
1948 s->end = s->base + sb->s_blocksize;
1949 }
1950
1951 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1952 if (error == -EFSCORRUPTED)
1953 goto bad_block;
1954 if (error)
1955 goto cleanup;
1956
1957 if (i->value && s->here->e_value_inum) {
1958 /*
1959 * A ref count on ea_inode has been taken as part of the call to
1960 * ext4_xattr_set_entry() above. We would like to drop this
1961 * extra ref but we have to wait until the xattr block is
1962 * initialized and has its own ref count on the ea_inode.
1963 */
1964 ea_ino = le32_to_cpu(s->here->e_value_inum);
1965 error = ext4_xattr_inode_iget(inode, ea_ino,
1966 le32_to_cpu(s->here->e_hash),
1967 &ea_inode);
1968 if (error) {
1969 ea_inode = NULL;
1970 goto cleanup;
1971 }
1972 }
1973
1974 inserted:
1975 if (!IS_LAST_ENTRY(s->first)) {
1976 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1977 &ce);
1978 if (new_bh) {
1979 /* We found an identical block in the cache. */
1980 if (new_bh == bs->bh)
1981 ea_bdebug(new_bh, "keeping");
1982 else {
1983 u32 ref;
1984
1985 WARN_ON_ONCE(dquot_initialize_needed(inode));
1986
1987 /* The old block is released after updating
1988 the inode. */
1989 error = dquot_alloc_block(inode,
1990 EXT4_C2B(EXT4_SB(sb), 1));
1991 if (error)
1992 goto cleanup;
1993 BUFFER_TRACE(new_bh, "get_write_access");
1994 error = ext4_journal_get_write_access(handle,
1995 new_bh);
1996 if (error)
1997 goto cleanup_dquot;
1998 lock_buffer(new_bh);
1999 /*
2000 * We have to be careful about races with
2001 * freeing, rehashing or adding references to
2002 * xattr block. Once we hold buffer lock xattr
2003 * block's state is stable so we can check
2004 * whether the block got freed / rehashed or
2005 * not. Since we unhash mbcache entry under
2006 * buffer lock when freeing / rehashing xattr
2007 * block, checking whether entry is still
2008 * hashed is reliable. Same rules hold for
2009 * e_reusable handling.
2010 */
2011 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2012 !ce->e_reusable) {
2013 /*
2014 * Undo everything and check mbcache
2015 * again.
2016 */
2017 unlock_buffer(new_bh);
2018 dquot_free_block(inode,
2019 EXT4_C2B(EXT4_SB(sb),
2020 1));
2021 brelse(new_bh);
2022 mb_cache_entry_put(ea_block_cache, ce);
2023 ce = NULL;
2024 new_bh = NULL;
2025 goto inserted;
2026 }
2027 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2028 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2029 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2030 ce->e_reusable = 0;
2031 ea_bdebug(new_bh, "reusing; refcount now=%d",
2032 ref);
2033 ext4_xattr_block_csum_set(inode, new_bh);
2034 unlock_buffer(new_bh);
2035 error = ext4_handle_dirty_metadata(handle,
2036 inode,
2037 new_bh);
2038 if (error)
2039 goto cleanup_dquot;
2040 }
2041 mb_cache_entry_touch(ea_block_cache, ce);
2042 mb_cache_entry_put(ea_block_cache, ce);
2043 ce = NULL;
2044 } else if (bs->bh && s->base == bs->bh->b_data) {
2045 /* We were modifying this block in-place. */
2046 ea_bdebug(bs->bh, "keeping this block");
2047 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2048 new_bh = bs->bh;
2049 get_bh(new_bh);
2050 } else {
2051 /* We need to allocate a new block */
2052 ext4_fsblk_t goal, block;
2053
2054 WARN_ON_ONCE(dquot_initialize_needed(inode));
2055
2056 goal = ext4_group_first_block_no(sb,
2057 EXT4_I(inode)->i_block_group);
2058
2059 /* non-extent files can't have physical blocks past 2^32 */
2060 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2061 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2062
2063 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2064 NULL, &error);
2065 if (error)
2066 goto cleanup;
2067
2068 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2069 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2070
2071 ea_idebug(inode, "creating block %llu",
2072 (unsigned long long)block);
2073
2074 new_bh = sb_getblk(sb, block);
2075 if (unlikely(!new_bh)) {
2076 error = -ENOMEM;
2077 getblk_failed:
2078 ext4_free_blocks(handle, inode, NULL, block, 1,
2079 EXT4_FREE_BLOCKS_METADATA);
2080 goto cleanup;
2081 }
2082 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2083 ENTRY(header(s->base)+1));
2084 if (error)
2085 goto getblk_failed;
2086 if (ea_inode) {
2087 /* Drop the extra ref on ea_inode. */
2088 error = ext4_xattr_inode_dec_ref(handle,
2089 ea_inode);
2090 if (error)
2091 ext4_warning_inode(ea_inode,
2092 "dec ref error=%d",
2093 error);
2094 iput(ea_inode);
2095 ea_inode = NULL;
2096 }
2097
2098 lock_buffer(new_bh);
2099 error = ext4_journal_get_create_access(handle, new_bh);
2100 if (error) {
2101 unlock_buffer(new_bh);
2102 error = -EIO;
2103 goto getblk_failed;
2104 }
2105 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2106 ext4_xattr_block_csum_set(inode, new_bh);
2107 set_buffer_uptodate(new_bh);
2108 unlock_buffer(new_bh);
2109 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2110 error = ext4_handle_dirty_metadata(handle, inode,
2111 new_bh);
2112 if (error)
2113 goto cleanup;
2114 }
2115 }
2116
2117 if (old_ea_inode_quota)
2118 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2119
2120 /* Update the inode. */
2121 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2122
2123 /* Drop the previous xattr block. */
2124 if (bs->bh && bs->bh != new_bh) {
2125 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2126
2127 ext4_xattr_release_block(handle, inode, bs->bh,
2128 &ea_inode_array,
2129 0 /* extra_credits */);
2130 ext4_xattr_inode_array_free(ea_inode_array);
2131 }
2132 error = 0;
2133
2134 cleanup:
2135 if (ea_inode) {
2136 int error2;
2137
2138 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2139 if (error2)
2140 ext4_warning_inode(ea_inode, "dec ref error=%d",
2141 error2);
2142
2143 /* If there was an error, revert the quota charge. */
2144 if (error)
2145 ext4_xattr_inode_free_quota(inode, ea_inode,
2146 i_size_read(ea_inode));
2147 iput(ea_inode);
2148 }
2149 if (ce)
2150 mb_cache_entry_put(ea_block_cache, ce);
2151 brelse(new_bh);
2152 if (!(bs->bh && s->base == bs->bh->b_data))
2153 kfree(s->base);
2154
2155 return error;
2156
2157 cleanup_dquot:
2158 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2159 goto cleanup;
2160
2161 bad_block:
2162 EXT4_ERROR_INODE(inode, "bad block %llu",
2163 EXT4_I(inode)->i_file_acl);
2164 goto cleanup;
2165
2166 #undef header
2167 }
2168
2169 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2170 struct ext4_xattr_ibody_find *is)
2171 {
2172 struct ext4_xattr_ibody_header *header;
2173 struct ext4_inode *raw_inode;
2174 int error;
2175
2176 if (EXT4_I(inode)->i_extra_isize == 0)
2177 return 0;
2178 raw_inode = ext4_raw_inode(&is->iloc);
2179 header = IHDR(inode, raw_inode);
2180 is->s.base = is->s.first = IFIRST(header);
2181 is->s.here = is->s.first;
2182 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2183 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2184 error = xattr_check_inode(inode, header, is->s.end);
2185 if (error)
2186 return error;
2187 /* Find the named attribute. */
2188 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2189 i->name_index, i->name, 0);
2190 if (error && error != -ENODATA)
2191 return error;
2192 is->s.not_found = error;
2193 }
2194 return 0;
2195 }
2196
2197 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2198 struct ext4_xattr_info *i,
2199 struct ext4_xattr_ibody_find *is)
2200 {
2201 struct ext4_xattr_ibody_header *header;
2202 struct ext4_xattr_search *s = &is->s;
2203 int error;
2204
2205 if (EXT4_I(inode)->i_extra_isize == 0)
2206 return -ENOSPC;
2207 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2208 if (error) {
2209 if (error == -ENOSPC &&
2210 ext4_has_inline_data(inode)) {
2211 error = ext4_try_to_evict_inline_data(handle, inode,
2212 EXT4_XATTR_LEN(strlen(i->name) +
2213 EXT4_XATTR_SIZE(i->value_len)));
2214 if (error)
2215 return error;
2216 error = ext4_xattr_ibody_find(inode, i, is);
2217 if (error)
2218 return error;
2219 error = ext4_xattr_set_entry(i, s, handle, inode,
2220 false /* is_block */);
2221 }
2222 if (error)
2223 return error;
2224 }
2225 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2226 if (!IS_LAST_ENTRY(s->first)) {
2227 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2228 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2229 } else {
2230 header->h_magic = cpu_to_le32(0);
2231 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2232 }
2233 return 0;
2234 }
2235
2236 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2237 struct ext4_xattr_info *i,
2238 struct ext4_xattr_ibody_find *is)
2239 {
2240 struct ext4_xattr_ibody_header *header;
2241 struct ext4_xattr_search *s = &is->s;
2242 int error;
2243
2244 if (EXT4_I(inode)->i_extra_isize == 0)
2245 return -ENOSPC;
2246 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2247 if (error)
2248 return error;
2249 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2250 if (!IS_LAST_ENTRY(s->first)) {
2251 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2252 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2253 } else {
2254 header->h_magic = cpu_to_le32(0);
2255 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2256 }
2257 return 0;
2258 }
2259
2260 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2261 struct ext4_xattr_info *i)
2262 {
2263 void *value;
2264
2265 /* When e_value_inum is set the value is stored externally. */
2266 if (s->here->e_value_inum)
2267 return 0;
2268 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2269 return 0;
2270 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2271 return !memcmp(value, i->value, i->value_len);
2272 }
2273
2274 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2275 {
2276 struct buffer_head *bh;
2277 int error;
2278
2279 if (!EXT4_I(inode)->i_file_acl)
2280 return NULL;
2281 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2282 if (!bh)
2283 return ERR_PTR(-EIO);
2284 error = ext4_xattr_check_block(inode, bh);
2285 if (error)
2286 return ERR_PTR(error);
2287 return bh;
2288 }
2289
2290 /*
2291 * ext4_xattr_set_handle()
2292 *
2293 * Create, replace or remove an extended attribute for this inode. Value
2294 * is NULL to remove an existing extended attribute, and non-NULL to
2295 * either replace an existing extended attribute, or create a new extended
2296 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2297 * specify that an extended attribute must exist and must not exist
2298 * previous to the call, respectively.
2299 *
2300 * Returns 0, or a negative error number on failure.
2301 */
2302 int
2303 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2304 const char *name, const void *value, size_t value_len,
2305 int flags)
2306 {
2307 struct ext4_xattr_info i = {
2308 .name_index = name_index,
2309 .name = name,
2310 .value = value,
2311 .value_len = value_len,
2312 .in_inode = 0,
2313 };
2314 struct ext4_xattr_ibody_find is = {
2315 .s = { .not_found = -ENODATA, },
2316 };
2317 struct ext4_xattr_block_find bs = {
2318 .s = { .not_found = -ENODATA, },
2319 };
2320 int no_expand;
2321 int error;
2322
2323 if (!name)
2324 return -EINVAL;
2325 if (strlen(name) > 255)
2326 return -ERANGE;
2327
2328 ext4_write_lock_xattr(inode, &no_expand);
2329
2330 /* Check journal credits under write lock. */
2331 if (ext4_handle_valid(handle)) {
2332 struct buffer_head *bh;
2333 int credits;
2334
2335 bh = ext4_xattr_get_block(inode);
2336 if (IS_ERR(bh)) {
2337 error = PTR_ERR(bh);
2338 goto cleanup;
2339 }
2340
2341 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2342 value_len,
2343 flags & XATTR_CREATE);
2344 brelse(bh);
2345
2346 if (!ext4_handle_has_enough_credits(handle, credits)) {
2347 error = -ENOSPC;
2348 goto cleanup;
2349 }
2350 }
2351
2352 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2353 if (error)
2354 goto cleanup;
2355
2356 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2357 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2358 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2359 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2360 }
2361
2362 error = ext4_xattr_ibody_find(inode, &i, &is);
2363 if (error)
2364 goto cleanup;
2365 if (is.s.not_found)
2366 error = ext4_xattr_block_find(inode, &i, &bs);
2367 if (error)
2368 goto cleanup;
2369 if (is.s.not_found && bs.s.not_found) {
2370 error = -ENODATA;
2371 if (flags & XATTR_REPLACE)
2372 goto cleanup;
2373 error = 0;
2374 if (!value)
2375 goto cleanup;
2376 } else {
2377 error = -EEXIST;
2378 if (flags & XATTR_CREATE)
2379 goto cleanup;
2380 }
2381
2382 if (!value) {
2383 if (!is.s.not_found)
2384 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2385 else if (!bs.s.not_found)
2386 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2387 } else {
2388 error = 0;
2389 /* Xattr value did not change? Save us some work and bail out */
2390 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2391 goto cleanup;
2392 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2393 goto cleanup;
2394
2395 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2396 (EXT4_XATTR_SIZE(i.value_len) >
2397 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2398 i.in_inode = 1;
2399 retry_inode:
2400 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2401 if (!error && !bs.s.not_found) {
2402 i.value = NULL;
2403 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2404 } else if (error == -ENOSPC) {
2405 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2406 error = ext4_xattr_block_find(inode, &i, &bs);
2407 if (error)
2408 goto cleanup;
2409 }
2410 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2411 if (!error && !is.s.not_found) {
2412 i.value = NULL;
2413 error = ext4_xattr_ibody_set(handle, inode, &i,
2414 &is);
2415 } else if (error == -ENOSPC) {
2416 /*
2417 * Xattr does not fit in the block, store at
2418 * external inode if possible.
2419 */
2420 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2421 !i.in_inode) {
2422 i.in_inode = 1;
2423 goto retry_inode;
2424 }
2425 }
2426 }
2427 }
2428 if (!error) {
2429 ext4_xattr_update_super_block(handle, inode->i_sb);
2430 inode->i_ctime = current_time(inode);
2431 if (!value)
2432 no_expand = 0;
2433 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2434 /*
2435 * The bh is consumed by ext4_mark_iloc_dirty, even with
2436 * error != 0.
2437 */
2438 is.iloc.bh = NULL;
2439 if (IS_SYNC(inode))
2440 ext4_handle_sync(handle);
2441 }
2442
2443 cleanup:
2444 brelse(is.iloc.bh);
2445 brelse(bs.bh);
2446 ext4_write_unlock_xattr(inode, &no_expand);
2447 return error;
2448 }
2449
2450 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2451 bool is_create, int *credits)
2452 {
2453 struct buffer_head *bh;
2454 int err;
2455
2456 *credits = 0;
2457
2458 if (!EXT4_SB(inode->i_sb)->s_journal)
2459 return 0;
2460
2461 down_read(&EXT4_I(inode)->xattr_sem);
2462
2463 bh = ext4_xattr_get_block(inode);
2464 if (IS_ERR(bh)) {
2465 err = PTR_ERR(bh);
2466 } else {
2467 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2468 value_len, is_create);
2469 brelse(bh);
2470 err = 0;
2471 }
2472
2473 up_read(&EXT4_I(inode)->xattr_sem);
2474 return err;
2475 }
2476
2477 /*
2478 * ext4_xattr_set()
2479 *
2480 * Like ext4_xattr_set_handle, but start from an inode. This extended
2481 * attribute modification is a filesystem transaction by itself.
2482 *
2483 * Returns 0, or a negative error number on failure.
2484 */
2485 int
2486 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2487 const void *value, size_t value_len, int flags)
2488 {
2489 handle_t *handle;
2490 struct super_block *sb = inode->i_sb;
2491 int error, retries = 0;
2492 int credits;
2493
2494 error = dquot_initialize(inode);
2495 if (error)
2496 return error;
2497
2498 retry:
2499 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2500 &credits);
2501 if (error)
2502 return error;
2503
2504 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2505 if (IS_ERR(handle)) {
2506 error = PTR_ERR(handle);
2507 } else {
2508 int error2;
2509
2510 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2511 value, value_len, flags);
2512 error2 = ext4_journal_stop(handle);
2513 if (error == -ENOSPC &&
2514 ext4_should_retry_alloc(sb, &retries))
2515 goto retry;
2516 if (error == 0)
2517 error = error2;
2518 }
2519
2520 return error;
2521 }
2522
2523 /*
2524 * Shift the EA entries in the inode to create space for the increased
2525 * i_extra_isize.
2526 */
2527 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2528 int value_offs_shift, void *to,
2529 void *from, size_t n)
2530 {
2531 struct ext4_xattr_entry *last = entry;
2532 int new_offs;
2533
2534 /* We always shift xattr headers further thus offsets get lower */
2535 BUG_ON(value_offs_shift > 0);
2536
2537 /* Adjust the value offsets of the entries */
2538 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2539 if (!last->e_value_inum && last->e_value_size) {
2540 new_offs = le16_to_cpu(last->e_value_offs) +
2541 value_offs_shift;
2542 last->e_value_offs = cpu_to_le16(new_offs);
2543 }
2544 }
2545 /* Shift the entries by n bytes */
2546 memmove(to, from, n);
2547 }
2548
2549 /*
2550 * Move xattr pointed to by 'entry' from inode into external xattr block
2551 */
2552 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2553 struct ext4_inode *raw_inode,
2554 struct ext4_xattr_entry *entry)
2555 {
2556 struct ext4_xattr_ibody_find *is = NULL;
2557 struct ext4_xattr_block_find *bs = NULL;
2558 char *buffer = NULL, *b_entry_name = NULL;
2559 size_t value_size = le32_to_cpu(entry->e_value_size);
2560 struct ext4_xattr_info i = {
2561 .value = NULL,
2562 .value_len = 0,
2563 .name_index = entry->e_name_index,
2564 .in_inode = !!entry->e_value_inum,
2565 };
2566 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2567 int error;
2568
2569 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2570 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2571 buffer = kmalloc(value_size, GFP_NOFS);
2572 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2573 if (!is || !bs || !buffer || !b_entry_name) {
2574 error = -ENOMEM;
2575 goto out;
2576 }
2577
2578 is->s.not_found = -ENODATA;
2579 bs->s.not_found = -ENODATA;
2580 is->iloc.bh = NULL;
2581 bs->bh = NULL;
2582
2583 /* Save the entry name and the entry value */
2584 if (entry->e_value_inum) {
2585 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2586 if (error)
2587 goto out;
2588 } else {
2589 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2590 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2591 }
2592
2593 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2594 b_entry_name[entry->e_name_len] = '\0';
2595 i.name = b_entry_name;
2596
2597 error = ext4_get_inode_loc(inode, &is->iloc);
2598 if (error)
2599 goto out;
2600
2601 error = ext4_xattr_ibody_find(inode, &i, is);
2602 if (error)
2603 goto out;
2604
2605 /* Remove the chosen entry from the inode */
2606 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2607 if (error)
2608 goto out;
2609
2610 i.value = buffer;
2611 i.value_len = value_size;
2612 error = ext4_xattr_block_find(inode, &i, bs);
2613 if (error)
2614 goto out;
2615
2616 /* Add entry which was removed from the inode into the block */
2617 error = ext4_xattr_block_set(handle, inode, &i, bs);
2618 if (error)
2619 goto out;
2620 error = 0;
2621 out:
2622 kfree(b_entry_name);
2623 kfree(buffer);
2624 if (is)
2625 brelse(is->iloc.bh);
2626 kfree(is);
2627 kfree(bs);
2628
2629 return error;
2630 }
2631
2632 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2633 struct ext4_inode *raw_inode,
2634 int isize_diff, size_t ifree,
2635 size_t bfree, int *total_ino)
2636 {
2637 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2638 struct ext4_xattr_entry *small_entry;
2639 struct ext4_xattr_entry *entry;
2640 struct ext4_xattr_entry *last;
2641 unsigned int entry_size; /* EA entry size */
2642 unsigned int total_size; /* EA entry size + value size */
2643 unsigned int min_total_size;
2644 int error;
2645
2646 while (isize_diff > ifree) {
2647 entry = NULL;
2648 small_entry = NULL;
2649 min_total_size = ~0U;
2650 last = IFIRST(header);
2651 /* Find the entry best suited to be pushed into EA block */
2652 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2653 total_size = EXT4_XATTR_LEN(last->e_name_len);
2654 if (!last->e_value_inum)
2655 total_size += EXT4_XATTR_SIZE(
2656 le32_to_cpu(last->e_value_size));
2657 if (total_size <= bfree &&
2658 total_size < min_total_size) {
2659 if (total_size + ifree < isize_diff) {
2660 small_entry = last;
2661 } else {
2662 entry = last;
2663 min_total_size = total_size;
2664 }
2665 }
2666 }
2667
2668 if (entry == NULL) {
2669 if (small_entry == NULL)
2670 return -ENOSPC;
2671 entry = small_entry;
2672 }
2673
2674 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2675 total_size = entry_size;
2676 if (!entry->e_value_inum)
2677 total_size += EXT4_XATTR_SIZE(
2678 le32_to_cpu(entry->e_value_size));
2679 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2680 entry);
2681 if (error)
2682 return error;
2683
2684 *total_ino -= entry_size;
2685 ifree += total_size;
2686 bfree -= total_size;
2687 }
2688
2689 return 0;
2690 }
2691
2692 /*
2693 * Expand an inode by new_extra_isize bytes when EAs are present.
2694 * Returns 0 on success or negative error number on failure.
2695 */
2696 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2697 struct ext4_inode *raw_inode, handle_t *handle)
2698 {
2699 struct ext4_xattr_ibody_header *header;
2700 struct buffer_head *bh;
2701 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2702 static unsigned int mnt_count;
2703 size_t min_offs;
2704 size_t ifree, bfree;
2705 int total_ino;
2706 void *base, *end;
2707 int error = 0, tried_min_extra_isize = 0;
2708 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2709 int isize_diff; /* How much do we need to grow i_extra_isize */
2710
2711 retry:
2712 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2713 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2714 return 0;
2715
2716 header = IHDR(inode, raw_inode);
2717
2718 /*
2719 * Check if enough free space is available in the inode to shift the
2720 * entries ahead by new_extra_isize.
2721 */
2722
2723 base = IFIRST(header);
2724 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2725 min_offs = end - base;
2726 total_ino = sizeof(struct ext4_xattr_ibody_header);
2727
2728 error = xattr_check_inode(inode, header, end);
2729 if (error)
2730 goto cleanup;
2731
2732 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2733 if (ifree >= isize_diff)
2734 goto shift;
2735
2736 /*
2737 * Enough free space isn't available in the inode, check if
2738 * EA block can hold new_extra_isize bytes.
2739 */
2740 if (EXT4_I(inode)->i_file_acl) {
2741 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2742 error = -EIO;
2743 if (!bh)
2744 goto cleanup;
2745 error = ext4_xattr_check_block(inode, bh);
2746 if (error)
2747 goto cleanup;
2748 base = BHDR(bh);
2749 end = bh->b_data + bh->b_size;
2750 min_offs = end - base;
2751 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2752 NULL);
2753 brelse(bh);
2754 if (bfree + ifree < isize_diff) {
2755 if (!tried_min_extra_isize && s_min_extra_isize) {
2756 tried_min_extra_isize++;
2757 new_extra_isize = s_min_extra_isize;
2758 goto retry;
2759 }
2760 error = -ENOSPC;
2761 goto cleanup;
2762 }
2763 } else {
2764 bfree = inode->i_sb->s_blocksize;
2765 }
2766
2767 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2768 isize_diff, ifree, bfree,
2769 &total_ino);
2770 if (error) {
2771 if (error == -ENOSPC && !tried_min_extra_isize &&
2772 s_min_extra_isize) {
2773 tried_min_extra_isize++;
2774 new_extra_isize = s_min_extra_isize;
2775 goto retry;
2776 }
2777 goto cleanup;
2778 }
2779 shift:
2780 /* Adjust the offsets and shift the remaining entries ahead */
2781 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2782 - new_extra_isize, (void *)raw_inode +
2783 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2784 (void *)header, total_ino);
2785 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2786
2787 cleanup:
2788 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2789 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2790 inode->i_ino);
2791 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2792 }
2793 return error;
2794 }
2795
2796 #define EIA_INCR 16 /* must be 2^n */
2797 #define EIA_MASK (EIA_INCR - 1)
2798
2799 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2800 * If @ea_inode_array is new or full it will be grown and the old
2801 * contents copied over.
2802 */
2803 static int
2804 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2805 struct inode *inode)
2806 {
2807 if (*ea_inode_array == NULL) {
2808 /*
2809 * Start with 15 inodes, so it fits into a power-of-two size.
2810 * If *ea_inode_array is NULL, this is essentially offsetof()
2811 */
2812 (*ea_inode_array) =
2813 kmalloc(offsetof(struct ext4_xattr_inode_array,
2814 inodes[EIA_MASK]),
2815 GFP_NOFS);
2816 if (*ea_inode_array == NULL)
2817 return -ENOMEM;
2818 (*ea_inode_array)->count = 0;
2819 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2820 /* expand the array once all 15 + n * 16 slots are full */
2821 struct ext4_xattr_inode_array *new_array = NULL;
2822 int count = (*ea_inode_array)->count;
2823
2824 /* if new_array is NULL, this is essentially offsetof() */
2825 new_array = kmalloc(
2826 offsetof(struct ext4_xattr_inode_array,
2827 inodes[count + EIA_INCR]),
2828 GFP_NOFS);
2829 if (new_array == NULL)
2830 return -ENOMEM;
2831 memcpy(new_array, *ea_inode_array,
2832 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2833 kfree(*ea_inode_array);
2834 *ea_inode_array = new_array;
2835 }
2836 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2837 return 0;
2838 }
2839
2840 /*
2841 * ext4_xattr_delete_inode()
2842 *
2843 * Free extended attribute resources associated with this inode. Traverse
2844 * all entries and decrement reference on any xattr inodes associated with this
2845 * inode. This is called immediately before an inode is freed. We have exclusive
2846 * access to the inode. If an orphan inode is deleted it will also release its
2847 * references on xattr block and xattr inodes.
2848 */
2849 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2850 struct ext4_xattr_inode_array **ea_inode_array,
2851 int extra_credits)
2852 {
2853 struct buffer_head *bh = NULL;
2854 struct ext4_xattr_ibody_header *header;
2855 struct ext4_iloc iloc = { .bh = NULL };
2856 struct ext4_xattr_entry *entry;
2857 struct inode *ea_inode;
2858 int error;
2859
2860 error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2861 NULL /* bh */,
2862 false /* dirty */,
2863 false /* block_csum */);
2864 if (error) {
2865 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2866 goto cleanup;
2867 }
2868
2869 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2870 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2871
2872 error = ext4_get_inode_loc(inode, &iloc);
2873 if (error) {
2874 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2875 goto cleanup;
2876 }
2877
2878 error = ext4_journal_get_write_access(handle, iloc.bh);
2879 if (error) {
2880 EXT4_ERROR_INODE(inode, "write access (error %d)",
2881 error);
2882 goto cleanup;
2883 }
2884
2885 header = IHDR(inode, ext4_raw_inode(&iloc));
2886 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2887 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2888 IFIRST(header),
2889 false /* block_csum */,
2890 ea_inode_array,
2891 extra_credits,
2892 false /* skip_quota */);
2893 }
2894
2895 if (EXT4_I(inode)->i_file_acl) {
2896 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2897 if (!bh) {
2898 EXT4_ERROR_INODE(inode, "block %llu read error",
2899 EXT4_I(inode)->i_file_acl);
2900 error = -EIO;
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 }
2939 error = 0;
2940 cleanup:
2941 brelse(iloc.bh);
2942 brelse(bh);
2943 return error;
2944 }
2945
2946 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2947 {
2948 int idx;
2949
2950 if (ea_inode_array == NULL)
2951 return;
2952
2953 for (idx = 0; idx < ea_inode_array->count; ++idx)
2954 iput(ea_inode_array->inodes[idx]);
2955 kfree(ea_inode_array);
2956 }
2957
2958 /*
2959 * ext4_xattr_block_cache_insert()
2960 *
2961 * Create a new entry in the extended attribute block cache, and insert
2962 * it unless such an entry is already in the cache.
2963 *
2964 * Returns 0, or a negative error number on failure.
2965 */
2966 static void
2967 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2968 struct buffer_head *bh)
2969 {
2970 struct ext4_xattr_header *header = BHDR(bh);
2971 __u32 hash = le32_to_cpu(header->h_hash);
2972 int reusable = le32_to_cpu(header->h_refcount) <
2973 EXT4_XATTR_REFCOUNT_MAX;
2974 int error;
2975
2976 if (!ea_block_cache)
2977 return;
2978 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2979 bh->b_blocknr, reusable);
2980 if (error) {
2981 if (error == -EBUSY)
2982 ea_bdebug(bh, "already in cache");
2983 } else
2984 ea_bdebug(bh, "inserting [%x]", (int)hash);
2985 }
2986
2987 /*
2988 * ext4_xattr_cmp()
2989 *
2990 * Compare two extended attribute blocks for equality.
2991 *
2992 * Returns 0 if the blocks are equal, 1 if they differ, and
2993 * a negative error number on errors.
2994 */
2995 static int
2996 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2997 struct ext4_xattr_header *header2)
2998 {
2999 struct ext4_xattr_entry *entry1, *entry2;
3000
3001 entry1 = ENTRY(header1+1);
3002 entry2 = ENTRY(header2+1);
3003 while (!IS_LAST_ENTRY(entry1)) {
3004 if (IS_LAST_ENTRY(entry2))
3005 return 1;
3006 if (entry1->e_hash != entry2->e_hash ||
3007 entry1->e_name_index != entry2->e_name_index ||
3008 entry1->e_name_len != entry2->e_name_len ||
3009 entry1->e_value_size != entry2->e_value_size ||
3010 entry1->e_value_inum != entry2->e_value_inum ||
3011 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3012 return 1;
3013 if (!entry1->e_value_inum &&
3014 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3015 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3016 le32_to_cpu(entry1->e_value_size)))
3017 return 1;
3018
3019 entry1 = EXT4_XATTR_NEXT(entry1);
3020 entry2 = EXT4_XATTR_NEXT(entry2);
3021 }
3022 if (!IS_LAST_ENTRY(entry2))
3023 return 1;
3024 return 0;
3025 }
3026
3027 /*
3028 * ext4_xattr_block_cache_find()
3029 *
3030 * Find an identical extended attribute block.
3031 *
3032 * Returns a pointer to the block found, or NULL if such a block was
3033 * not found or an error occurred.
3034 */
3035 static struct buffer_head *
3036 ext4_xattr_block_cache_find(struct inode *inode,
3037 struct ext4_xattr_header *header,
3038 struct mb_cache_entry **pce)
3039 {
3040 __u32 hash = le32_to_cpu(header->h_hash);
3041 struct mb_cache_entry *ce;
3042 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3043
3044 if (!ea_block_cache)
3045 return NULL;
3046 if (!header->h_hash)
3047 return NULL; /* never share */
3048 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3049 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3050 while (ce) {
3051 struct buffer_head *bh;
3052
3053 bh = sb_bread(inode->i_sb, ce->e_value);
3054 if (!bh) {
3055 EXT4_ERROR_INODE(inode, "block %lu read error",
3056 (unsigned long)ce->e_value);
3057 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3058 *pce = ce;
3059 return bh;
3060 }
3061 brelse(bh);
3062 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3063 }
3064 return NULL;
3065 }
3066
3067 #define NAME_HASH_SHIFT 5
3068 #define VALUE_HASH_SHIFT 16
3069
3070 /*
3071 * ext4_xattr_hash_entry()
3072 *
3073 * Compute the hash of an extended attribute.
3074 */
3075 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3076 size_t value_count)
3077 {
3078 __u32 hash = 0;
3079
3080 while (name_len--) {
3081 hash = (hash << NAME_HASH_SHIFT) ^
3082 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3083 *name++;
3084 }
3085 while (value_count--) {
3086 hash = (hash << VALUE_HASH_SHIFT) ^
3087 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3088 le32_to_cpu(*value++);
3089 }
3090 return cpu_to_le32(hash);
3091 }
3092
3093 #undef NAME_HASH_SHIFT
3094 #undef VALUE_HASH_SHIFT
3095
3096 #define BLOCK_HASH_SHIFT 16
3097
3098 /*
3099 * ext4_xattr_rehash()
3100 *
3101 * Re-compute the extended attribute hash value after an entry has changed.
3102 */
3103 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3104 {
3105 struct ext4_xattr_entry *here;
3106 __u32 hash = 0;
3107
3108 here = ENTRY(header+1);
3109 while (!IS_LAST_ENTRY(here)) {
3110 if (!here->e_hash) {
3111 /* Block is not shared if an entry's hash value == 0 */
3112 hash = 0;
3113 break;
3114 }
3115 hash = (hash << BLOCK_HASH_SHIFT) ^
3116 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3117 le32_to_cpu(here->e_hash);
3118 here = EXT4_XATTR_NEXT(here);
3119 }
3120 header->h_hash = cpu_to_le32(hash);
3121 }
3122
3123 #undef BLOCK_HASH_SHIFT
3124
3125 #define HASH_BUCKET_BITS 10
3126
3127 struct mb_cache *
3128 ext4_xattr_create_cache(void)
3129 {
3130 return mb_cache_create(HASH_BUCKET_BITS);
3131 }
3132
3133 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3134 {
3135 if (cache)
3136 mb_cache_destroy(cache);
3137 }
3138
ubuntu-bionic-kernel mirror