Loading...
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
77static void ext4_xattr_block_cache_insert(struct mb_cache *,
78 struct buffer_head *);
79static struct buffer_head *
80ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
81 struct mb_cache_entry **);
82static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
83 size_t value_count);
84static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value,
85 size_t value_count);
86static void ext4_xattr_rehash(struct ext4_xattr_header *);
87
88static const struct xattr_handler * const ext4_xattr_handler_map[] = {
89 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
90#ifdef CONFIG_EXT4_FS_POSIX_ACL
91 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
92 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
93#endif
94 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
95#ifdef CONFIG_EXT4_FS_SECURITY
96 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
97#endif
98 [EXT4_XATTR_INDEX_HURD] = &ext4_xattr_hurd_handler,
99};
100
101const struct xattr_handler *ext4_xattr_handlers[] = {
102 &ext4_xattr_user_handler,
103 &ext4_xattr_trusted_handler,
104#ifdef CONFIG_EXT4_FS_POSIX_ACL
105 &posix_acl_access_xattr_handler,
106 &posix_acl_default_xattr_handler,
107#endif
108#ifdef CONFIG_EXT4_FS_SECURITY
109 &ext4_xattr_security_handler,
110#endif
111 &ext4_xattr_hurd_handler,
112 NULL
113};
114
115#define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
116 inode->i_sb->s_fs_info)->s_ea_block_cache)
117
118#define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
119 inode->i_sb->s_fs_info)->s_ea_inode_cache)
120
121static int
122ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
123 struct inode *inode);
124
125#ifdef CONFIG_LOCKDEP
126void ext4_xattr_inode_set_class(struct inode *ea_inode)
127{
128 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
129}
130#endif
131
132static __le32 ext4_xattr_block_csum(struct inode *inode,
133 sector_t block_nr,
134 struct ext4_xattr_header *hdr)
135{
136 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
137 __u32 csum;
138 __le64 dsk_block_nr = cpu_to_le64(block_nr);
139 __u32 dummy_csum = 0;
140 int offset = offsetof(struct ext4_xattr_header, h_checksum);
141
142 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
143 sizeof(dsk_block_nr));
144 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
145 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
146 offset += sizeof(dummy_csum);
147 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
148 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
149
150 return cpu_to_le32(csum);
151}
152
153static int ext4_xattr_block_csum_verify(struct inode *inode,
154 struct buffer_head *bh)
155{
156 struct ext4_xattr_header *hdr = BHDR(bh);
157 int ret = 1;
158
159 if (ext4_has_metadata_csum(inode->i_sb)) {
160 lock_buffer(bh);
161 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
162 bh->b_blocknr, hdr));
163 unlock_buffer(bh);
164 }
165 return ret;
166}
167
168static void ext4_xattr_block_csum_set(struct inode *inode,
169 struct buffer_head *bh)
170{
171 if (ext4_has_metadata_csum(inode->i_sb))
172 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
173 bh->b_blocknr, BHDR(bh));
174}
175
176static inline const struct xattr_handler *
177ext4_xattr_handler(int name_index)
178{
179 const struct xattr_handler *handler = NULL;
180
181 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
182 handler = ext4_xattr_handler_map[name_index];
183 return handler;
184}
185
186static int
187ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
188 void *value_start)
189{
190 struct ext4_xattr_entry *e = entry;
191
192 /* Find the end of the names list */
193 while (!IS_LAST_ENTRY(e)) {
194 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
195 if ((void *)next >= end)
196 return -EFSCORRUPTED;
197 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
198 return -EFSCORRUPTED;
199 e = next;
200 }
201
202 /* Check the values */
203 while (!IS_LAST_ENTRY(entry)) {
204 u32 size = le32_to_cpu(entry->e_value_size);
205
206 if (size > EXT4_XATTR_SIZE_MAX)
207 return -EFSCORRUPTED;
208
209 if (size != 0 && entry->e_value_inum == 0) {
210 u16 offs = le16_to_cpu(entry->e_value_offs);
211 void *value;
212
213 /*
214 * The value cannot overlap the names, and the value
215 * with padding cannot extend beyond 'end'. Check both
216 * the padded and unpadded sizes, since the size may
217 * overflow to 0 when adding padding.
218 */
219 if (offs > end - value_start)
220 return -EFSCORRUPTED;
221 value = value_start + offs;
222 if (value < (void *)e + sizeof(u32) ||
223 size > end - value ||
224 EXT4_XATTR_SIZE(size) > end - value)
225 return -EFSCORRUPTED;
226 }
227 entry = EXT4_XATTR_NEXT(entry);
228 }
229
230 return 0;
231}
232
233static inline int
234__ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
235 const char *function, unsigned int line)
236{
237 int error = -EFSCORRUPTED;
238
239 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
240 BHDR(bh)->h_blocks != cpu_to_le32(1))
241 goto errout;
242 if (buffer_verified(bh))
243 return 0;
244
245 error = -EFSBADCRC;
246 if (!ext4_xattr_block_csum_verify(inode, bh))
247 goto errout;
248 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
249 bh->b_data);
250errout:
251 if (error)
252 __ext4_error_inode(inode, function, line, 0, -error,
253 "corrupted xattr block %llu",
254 (unsigned long long) bh->b_blocknr);
255 else
256 set_buffer_verified(bh);
257 return error;
258}
259
260#define ext4_xattr_check_block(inode, bh) \
261 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
262
263
264static int
265__xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
266 void *end, const char *function, unsigned int line)
267{
268 int error = -EFSCORRUPTED;
269
270 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
271 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
272 goto errout;
273 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
274errout:
275 if (error)
276 __ext4_error_inode(inode, function, line, 0, -error,
277 "corrupted in-inode xattr");
278 return error;
279}
280
281#define xattr_check_inode(inode, header, end) \
282 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
283
284static int
285xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
286 void *end, int name_index, const char *name, int sorted)
287{
288 struct ext4_xattr_entry *entry, *next;
289 size_t name_len;
290 int cmp = 1;
291
292 if (name == NULL)
293 return -EINVAL;
294 name_len = strlen(name);
295 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
296 next = EXT4_XATTR_NEXT(entry);
297 if ((void *) next >= end) {
298 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
299 return -EFSCORRUPTED;
300 }
301 cmp = name_index - entry->e_name_index;
302 if (!cmp)
303 cmp = name_len - entry->e_name_len;
304 if (!cmp)
305 cmp = memcmp(name, entry->e_name, name_len);
306 if (cmp <= 0 && (sorted || cmp == 0))
307 break;
308 }
309 *pentry = entry;
310 return cmp ? -ENODATA : 0;
311}
312
313static u32
314ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
315{
316 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
317}
318
319static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
320{
321 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
322 (u32) inode_peek_iversion_raw(ea_inode);
323}
324
325static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
326{
327 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
328 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
329}
330
331static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
332{
333 return (u32)ea_inode->i_atime.tv_sec;
334}
335
336static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
337{
338 ea_inode->i_atime.tv_sec = hash;
339}
340
341/*
342 * Read the EA value from an inode.
343 */
344static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
345{
346 int blocksize = 1 << ea_inode->i_blkbits;
347 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
348 int tail_size = (size % blocksize) ?: blocksize;
349 struct buffer_head *bhs_inline[8];
350 struct buffer_head **bhs = bhs_inline;
351 int i, ret;
352
353 if (bh_count > ARRAY_SIZE(bhs_inline)) {
354 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
355 if (!bhs)
356 return -ENOMEM;
357 }
358
359 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
360 true /* wait */, bhs);
361 if (ret)
362 goto free_bhs;
363
364 for (i = 0; i < bh_count; i++) {
365 /* There shouldn't be any holes in ea_inode. */
366 if (!bhs[i]) {
367 ret = -EFSCORRUPTED;
368 goto put_bhs;
369 }
370 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
371 i < bh_count - 1 ? blocksize : tail_size);
372 }
373 ret = 0;
374put_bhs:
375 for (i = 0; i < bh_count; i++)
376 brelse(bhs[i]);
377free_bhs:
378 if (bhs != bhs_inline)
379 kfree(bhs);
380 return ret;
381}
382
383#define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
384
385static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
386 u32 ea_inode_hash, struct inode **ea_inode)
387{
388 struct inode *inode;
389 int err;
390
391 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
392 if (IS_ERR(inode)) {
393 err = PTR_ERR(inode);
394 ext4_error(parent->i_sb,
395 "error while reading EA inode %lu err=%d", ea_ino,
396 err);
397 return err;
398 }
399
400 if (is_bad_inode(inode)) {
401 ext4_error(parent->i_sb,
402 "error while reading EA inode %lu is_bad_inode",
403 ea_ino);
404 err = -EIO;
405 goto error;
406 }
407
408 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
409 ext4_error(parent->i_sb,
410 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
411 ea_ino);
412 err = -EINVAL;
413 goto error;
414 }
415
416 ext4_xattr_inode_set_class(inode);
417
418 /*
419 * Check whether this is an old Lustre-style xattr inode. Lustre
420 * implementation does not have hash validation, rather it has a
421 * backpointer from ea_inode to the parent inode.
422 */
423 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
424 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
425 inode->i_generation == parent->i_generation) {
426 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
427 ext4_xattr_inode_set_ref(inode, 1);
428 } else {
429 inode_lock(inode);
430 inode->i_flags |= S_NOQUOTA;
431 inode_unlock(inode);
432 }
433
434 *ea_inode = inode;
435 return 0;
436error:
437 iput(inode);
438 return err;
439}
440
441/* Remove entry from mbcache when EA inode is getting evicted */
442void ext4_evict_ea_inode(struct inode *inode)
443{
444 struct mb_cache_entry *oe;
445
446 if (!EA_INODE_CACHE(inode))
447 return;
448 /* Wait for entry to get unused so that we can remove it */
449 while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
450 ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
451 mb_cache_entry_wait_unused(oe);
452 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
453 }
454}
455
456static int
457ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
458 struct ext4_xattr_entry *entry, void *buffer,
459 size_t size)
460{
461 u32 hash;
462
463 /* Verify stored hash matches calculated hash. */
464 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
465 if (hash != ext4_xattr_inode_get_hash(ea_inode))
466 return -EFSCORRUPTED;
467
468 if (entry) {
469 __le32 e_hash, tmp_data;
470
471 /* Verify entry hash. */
472 tmp_data = cpu_to_le32(hash);
473 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
474 &tmp_data, 1);
475 /* All good? */
476 if (e_hash == entry->e_hash)
477 return 0;
478
479 /*
480 * Not good. Maybe the entry hash was calculated
481 * using the buggy signed char version?
482 */
483 e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
484 &tmp_data, 1);
485 /* Still no match - bad */
486 if (e_hash != entry->e_hash)
487 return -EFSCORRUPTED;
488
489 /* Let people know about old hash */
490 pr_warn_once("ext4: filesystem with signed xattr name hash");
491 }
492 return 0;
493}
494
495/*
496 * Read xattr value from the EA inode.
497 */
498static int
499ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
500 void *buffer, size_t size)
501{
502 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
503 struct inode *ea_inode;
504 int err;
505
506 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
507 le32_to_cpu(entry->e_hash), &ea_inode);
508 if (err) {
509 ea_inode = NULL;
510 goto out;
511 }
512
513 if (i_size_read(ea_inode) != size) {
514 ext4_warning_inode(ea_inode,
515 "ea_inode file size=%llu entry size=%zu",
516 i_size_read(ea_inode), size);
517 err = -EFSCORRUPTED;
518 goto out;
519 }
520
521 err = ext4_xattr_inode_read(ea_inode, buffer, size);
522 if (err)
523 goto out;
524
525 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
526 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
527 size);
528 if (err) {
529 ext4_warning_inode(ea_inode,
530 "EA inode hash validation failed");
531 goto out;
532 }
533
534 if (ea_inode_cache)
535 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
536 ext4_xattr_inode_get_hash(ea_inode),
537 ea_inode->i_ino, true /* reusable */);
538 }
539out:
540 iput(ea_inode);
541 return err;
542}
543
544static int
545ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
546 void *buffer, size_t buffer_size)
547{
548 struct buffer_head *bh = NULL;
549 struct ext4_xattr_entry *entry;
550 size_t size;
551 void *end;
552 int error;
553 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
554
555 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
556 name_index, name, buffer, (long)buffer_size);
557
558 if (!EXT4_I(inode)->i_file_acl)
559 return -ENODATA;
560 ea_idebug(inode, "reading block %llu",
561 (unsigned long long)EXT4_I(inode)->i_file_acl);
562 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
563 if (IS_ERR(bh))
564 return PTR_ERR(bh);
565 ea_bdebug(bh, "b_count=%d, refcount=%d",
566 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
567 error = ext4_xattr_check_block(inode, bh);
568 if (error)
569 goto cleanup;
570 ext4_xattr_block_cache_insert(ea_block_cache, bh);
571 entry = BFIRST(bh);
572 end = bh->b_data + bh->b_size;
573 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
574 if (error)
575 goto cleanup;
576 size = le32_to_cpu(entry->e_value_size);
577 error = -ERANGE;
578 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
579 goto cleanup;
580 if (buffer) {
581 if (size > buffer_size)
582 goto cleanup;
583 if (entry->e_value_inum) {
584 error = ext4_xattr_inode_get(inode, entry, buffer,
585 size);
586 if (error)
587 goto cleanup;
588 } else {
589 u16 offset = le16_to_cpu(entry->e_value_offs);
590 void *p = bh->b_data + offset;
591
592 if (unlikely(p + size > end))
593 goto cleanup;
594 memcpy(buffer, p, size);
595 }
596 }
597 error = size;
598
599cleanup:
600 brelse(bh);
601 return error;
602}
603
604int
605ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
606 void *buffer, size_t buffer_size)
607{
608 struct ext4_xattr_ibody_header *header;
609 struct ext4_xattr_entry *entry;
610 struct ext4_inode *raw_inode;
611 struct ext4_iloc iloc;
612 size_t size;
613 void *end;
614 int error;
615
616 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
617 return -ENODATA;
618 error = ext4_get_inode_loc(inode, &iloc);
619 if (error)
620 return error;
621 raw_inode = ext4_raw_inode(&iloc);
622 header = IHDR(inode, raw_inode);
623 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
624 error = xattr_check_inode(inode, header, end);
625 if (error)
626 goto cleanup;
627 entry = IFIRST(header);
628 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
629 if (error)
630 goto cleanup;
631 size = le32_to_cpu(entry->e_value_size);
632 error = -ERANGE;
633 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
634 goto cleanup;
635 if (buffer) {
636 if (size > buffer_size)
637 goto cleanup;
638 if (entry->e_value_inum) {
639 error = ext4_xattr_inode_get(inode, entry, buffer,
640 size);
641 if (error)
642 goto cleanup;
643 } else {
644 u16 offset = le16_to_cpu(entry->e_value_offs);
645 void *p = (void *)IFIRST(header) + offset;
646
647 if (unlikely(p + size > end))
648 goto cleanup;
649 memcpy(buffer, p, size);
650 }
651 }
652 error = size;
653
654cleanup:
655 brelse(iloc.bh);
656 return error;
657}
658
659/*
660 * ext4_xattr_get()
661 *
662 * Copy an extended attribute into the buffer
663 * provided, or compute the buffer size required.
664 * Buffer is NULL to compute the size of the buffer required.
665 *
666 * Returns a negative error number on failure, or the number of bytes
667 * used / required on success.
668 */
669int
670ext4_xattr_get(struct inode *inode, int name_index, const char *name,
671 void *buffer, size_t buffer_size)
672{
673 int error;
674
675 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
676 return -EIO;
677
678 if (strlen(name) > 255)
679 return -ERANGE;
680
681 down_read(&EXT4_I(inode)->xattr_sem);
682 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
683 buffer_size);
684 if (error == -ENODATA)
685 error = ext4_xattr_block_get(inode, name_index, name, buffer,
686 buffer_size);
687 up_read(&EXT4_I(inode)->xattr_sem);
688 return error;
689}
690
691static int
692ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
693 char *buffer, size_t buffer_size)
694{
695 size_t rest = buffer_size;
696
697 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
698 const struct xattr_handler *handler =
699 ext4_xattr_handler(entry->e_name_index);
700
701 if (handler && (!handler->list || handler->list(dentry))) {
702 const char *prefix = handler->prefix ?: handler->name;
703 size_t prefix_len = strlen(prefix);
704 size_t size = prefix_len + entry->e_name_len + 1;
705
706 if (buffer) {
707 if (size > rest)
708 return -ERANGE;
709 memcpy(buffer, prefix, prefix_len);
710 buffer += prefix_len;
711 memcpy(buffer, entry->e_name, entry->e_name_len);
712 buffer += entry->e_name_len;
713 *buffer++ = 0;
714 }
715 rest -= size;
716 }
717 }
718 return buffer_size - rest; /* total size */
719}
720
721static int
722ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
723{
724 struct inode *inode = d_inode(dentry);
725 struct buffer_head *bh = NULL;
726 int error;
727
728 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
729 buffer, (long)buffer_size);
730
731 if (!EXT4_I(inode)->i_file_acl)
732 return 0;
733 ea_idebug(inode, "reading block %llu",
734 (unsigned long long)EXT4_I(inode)->i_file_acl);
735 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
736 if (IS_ERR(bh))
737 return PTR_ERR(bh);
738 ea_bdebug(bh, "b_count=%d, refcount=%d",
739 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
740 error = ext4_xattr_check_block(inode, bh);
741 if (error)
742 goto cleanup;
743 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
744 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
745 buffer_size);
746cleanup:
747 brelse(bh);
748 return error;
749}
750
751static int
752ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
753{
754 struct inode *inode = d_inode(dentry);
755 struct ext4_xattr_ibody_header *header;
756 struct ext4_inode *raw_inode;
757 struct ext4_iloc iloc;
758 void *end;
759 int error;
760
761 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
762 return 0;
763 error = ext4_get_inode_loc(inode, &iloc);
764 if (error)
765 return error;
766 raw_inode = ext4_raw_inode(&iloc);
767 header = IHDR(inode, raw_inode);
768 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
769 error = xattr_check_inode(inode, header, end);
770 if (error)
771 goto cleanup;
772 error = ext4_xattr_list_entries(dentry, IFIRST(header),
773 buffer, buffer_size);
774
775cleanup:
776 brelse(iloc.bh);
777 return error;
778}
779
780/*
781 * Inode operation listxattr()
782 *
783 * d_inode(dentry)->i_rwsem: don't care
784 *
785 * Copy a list of attribute names into the buffer
786 * provided, or compute the buffer size required.
787 * Buffer is NULL to compute the size of the buffer required.
788 *
789 * Returns a negative error number on failure, or the number of bytes
790 * used / required on success.
791 */
792ssize_t
793ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
794{
795 int ret, ret2;
796
797 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
798 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
799 if (ret < 0)
800 goto errout;
801 if (buffer) {
802 buffer += ret;
803 buffer_size -= ret;
804 }
805 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
806 if (ret < 0)
807 goto errout;
808 ret += ret2;
809errout:
810 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
811 return ret;
812}
813
814/*
815 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
816 * not set, set it.
817 */
818static void ext4_xattr_update_super_block(handle_t *handle,
819 struct super_block *sb)
820{
821 if (ext4_has_feature_xattr(sb))
822 return;
823
824 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
825 if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
826 EXT4_JTR_NONE) == 0) {
827 lock_buffer(EXT4_SB(sb)->s_sbh);
828 ext4_set_feature_xattr(sb);
829 ext4_superblock_csum_set(sb);
830 unlock_buffer(EXT4_SB(sb)->s_sbh);
831 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
832 }
833}
834
835int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
836{
837 struct ext4_iloc iloc = { .bh = NULL };
838 struct buffer_head *bh = NULL;
839 struct ext4_inode *raw_inode;
840 struct ext4_xattr_ibody_header *header;
841 struct ext4_xattr_entry *entry;
842 qsize_t ea_inode_refs = 0;
843 void *end;
844 int ret;
845
846 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
847
848 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
849 ret = ext4_get_inode_loc(inode, &iloc);
850 if (ret)
851 goto out;
852 raw_inode = ext4_raw_inode(&iloc);
853 header = IHDR(inode, raw_inode);
854 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
855 ret = xattr_check_inode(inode, header, end);
856 if (ret)
857 goto out;
858
859 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
860 entry = EXT4_XATTR_NEXT(entry))
861 if (entry->e_value_inum)
862 ea_inode_refs++;
863 }
864
865 if (EXT4_I(inode)->i_file_acl) {
866 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
867 if (IS_ERR(bh)) {
868 ret = PTR_ERR(bh);
869 bh = NULL;
870 goto out;
871 }
872
873 ret = ext4_xattr_check_block(inode, bh);
874 if (ret)
875 goto out;
876
877 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
878 entry = EXT4_XATTR_NEXT(entry))
879 if (entry->e_value_inum)
880 ea_inode_refs++;
881 }
882 *usage = ea_inode_refs + 1;
883 ret = 0;
884out:
885 brelse(iloc.bh);
886 brelse(bh);
887 return ret;
888}
889
890static inline size_t round_up_cluster(struct inode *inode, size_t length)
891{
892 struct super_block *sb = inode->i_sb;
893 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
894 inode->i_blkbits);
895 size_t mask = ~(cluster_size - 1);
896
897 return (length + cluster_size - 1) & mask;
898}
899
900static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
901{
902 int err;
903
904 err = dquot_alloc_inode(inode);
905 if (err)
906 return err;
907 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
908 if (err)
909 dquot_free_inode(inode);
910 return err;
911}
912
913static void ext4_xattr_inode_free_quota(struct inode *parent,
914 struct inode *ea_inode,
915 size_t len)
916{
917 if (ea_inode &&
918 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
919 return;
920 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
921 dquot_free_inode(parent);
922}
923
924int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
925 struct buffer_head *block_bh, size_t value_len,
926 bool is_create)
927{
928 int credits;
929 int blocks;
930
931 /*
932 * 1) Owner inode update
933 * 2) Ref count update on old xattr block
934 * 3) new xattr block
935 * 4) block bitmap update for new xattr block
936 * 5) group descriptor for new xattr block
937 * 6) block bitmap update for old xattr block
938 * 7) group descriptor for old block
939 *
940 * 6 & 7 can happen if we have two racing threads T_a and T_b
941 * which are each trying to set an xattr on inodes I_a and I_b
942 * which were both initially sharing an xattr block.
943 */
944 credits = 7;
945
946 /* Quota updates. */
947 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
948
949 /*
950 * In case of inline data, we may push out the data to a block,
951 * so we need to reserve credits for this eventuality
952 */
953 if (inode && ext4_has_inline_data(inode))
954 credits += ext4_writepage_trans_blocks(inode) + 1;
955
956 /* We are done if ea_inode feature is not enabled. */
957 if (!ext4_has_feature_ea_inode(sb))
958 return credits;
959
960 /* New ea_inode, inode map, block bitmap, group descriptor. */
961 credits += 4;
962
963 /* Data blocks. */
964 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
965
966 /* Indirection block or one level of extent tree. */
967 blocks += 1;
968
969 /* Block bitmap and group descriptor updates for each block. */
970 credits += blocks * 2;
971
972 /* Blocks themselves. */
973 credits += blocks;
974
975 if (!is_create) {
976 /* Dereference ea_inode holding old xattr value.
977 * Old ea_inode, inode map, block bitmap, group descriptor.
978 */
979 credits += 4;
980
981 /* Data blocks for old ea_inode. */
982 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
983
984 /* Indirection block or one level of extent tree for old
985 * ea_inode.
986 */
987 blocks += 1;
988
989 /* Block bitmap and group descriptor updates for each block. */
990 credits += blocks * 2;
991 }
992
993 /* We may need to clone the existing xattr block in which case we need
994 * to increment ref counts for existing ea_inodes referenced by it.
995 */
996 if (block_bh) {
997 struct ext4_xattr_entry *entry = BFIRST(block_bh);
998
999 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1000 if (entry->e_value_inum)
1001 /* Ref count update on ea_inode. */
1002 credits += 1;
1003 }
1004 return credits;
1005}
1006
1007static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1008 int ref_change)
1009{
1010 struct ext4_iloc iloc;
1011 s64 ref_count;
1012 int ret;
1013
1014 inode_lock(ea_inode);
1015
1016 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1017 if (ret)
1018 goto out;
1019
1020 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1021 ref_count += ref_change;
1022 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1023
1024 if (ref_change > 0) {
1025 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1026 ea_inode->i_ino, ref_count);
1027
1028 if (ref_count == 1) {
1029 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1030 ea_inode->i_ino, ea_inode->i_nlink);
1031
1032 set_nlink(ea_inode, 1);
1033 ext4_orphan_del(handle, ea_inode);
1034 }
1035 } else {
1036 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1037 ea_inode->i_ino, ref_count);
1038
1039 if (ref_count == 0) {
1040 WARN_ONCE(ea_inode->i_nlink != 1,
1041 "EA inode %lu i_nlink=%u",
1042 ea_inode->i_ino, ea_inode->i_nlink);
1043
1044 clear_nlink(ea_inode);
1045 ext4_orphan_add(handle, ea_inode);
1046 }
1047 }
1048
1049 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1050 if (ret)
1051 ext4_warning_inode(ea_inode,
1052 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1053out:
1054 inode_unlock(ea_inode);
1055 return ret;
1056}
1057
1058static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1059{
1060 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1061}
1062
1063static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1064{
1065 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1066}
1067
1068static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1069 struct ext4_xattr_entry *first)
1070{
1071 struct inode *ea_inode;
1072 struct ext4_xattr_entry *entry;
1073 struct ext4_xattr_entry *failed_entry;
1074 unsigned int ea_ino;
1075 int err, saved_err;
1076
1077 for (entry = first; !IS_LAST_ENTRY(entry);
1078 entry = EXT4_XATTR_NEXT(entry)) {
1079 if (!entry->e_value_inum)
1080 continue;
1081 ea_ino = le32_to_cpu(entry->e_value_inum);
1082 err = ext4_xattr_inode_iget(parent, ea_ino,
1083 le32_to_cpu(entry->e_hash),
1084 &ea_inode);
1085 if (err)
1086 goto cleanup;
1087 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1088 if (err) {
1089 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1090 iput(ea_inode);
1091 goto cleanup;
1092 }
1093 iput(ea_inode);
1094 }
1095 return 0;
1096
1097cleanup:
1098 saved_err = err;
1099 failed_entry = entry;
1100
1101 for (entry = first; entry != failed_entry;
1102 entry = EXT4_XATTR_NEXT(entry)) {
1103 if (!entry->e_value_inum)
1104 continue;
1105 ea_ino = le32_to_cpu(entry->e_value_inum);
1106 err = ext4_xattr_inode_iget(parent, ea_ino,
1107 le32_to_cpu(entry->e_hash),
1108 &ea_inode);
1109 if (err) {
1110 ext4_warning(parent->i_sb,
1111 "cleanup ea_ino %u iget error %d", ea_ino,
1112 err);
1113 continue;
1114 }
1115 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1116 if (err)
1117 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1118 err);
1119 iput(ea_inode);
1120 }
1121 return saved_err;
1122}
1123
1124static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1125 struct buffer_head *bh, bool block_csum, bool dirty)
1126{
1127 int error;
1128
1129 if (bh && dirty) {
1130 if (block_csum)
1131 ext4_xattr_block_csum_set(inode, bh);
1132 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1133 if (error) {
1134 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1135 error);
1136 return error;
1137 }
1138 }
1139 return 0;
1140}
1141
1142static void
1143ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1144 struct buffer_head *bh,
1145 struct ext4_xattr_entry *first, bool block_csum,
1146 struct ext4_xattr_inode_array **ea_inode_array,
1147 int extra_credits, bool skip_quota)
1148{
1149 struct inode *ea_inode;
1150 struct ext4_xattr_entry *entry;
1151 bool dirty = false;
1152 unsigned int ea_ino;
1153 int err;
1154 int credits;
1155
1156 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1157 credits = 2 + extra_credits;
1158
1159 for (entry = first; !IS_LAST_ENTRY(entry);
1160 entry = EXT4_XATTR_NEXT(entry)) {
1161 if (!entry->e_value_inum)
1162 continue;
1163 ea_ino = le32_to_cpu(entry->e_value_inum);
1164 err = ext4_xattr_inode_iget(parent, ea_ino,
1165 le32_to_cpu(entry->e_hash),
1166 &ea_inode);
1167 if (err)
1168 continue;
1169
1170 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1171 if (err) {
1172 ext4_warning_inode(ea_inode,
1173 "Expand inode array err=%d", err);
1174 iput(ea_inode);
1175 continue;
1176 }
1177
1178 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1179 ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1180 ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1181 dirty));
1182 if (err < 0) {
1183 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1184 err);
1185 continue;
1186 }
1187 if (err > 0) {
1188 err = ext4_journal_get_write_access(handle,
1189 parent->i_sb, bh, EXT4_JTR_NONE);
1190 if (err) {
1191 ext4_warning_inode(ea_inode,
1192 "Re-get write access err=%d",
1193 err);
1194 continue;
1195 }
1196 }
1197
1198 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1199 if (err) {
1200 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1201 err);
1202 continue;
1203 }
1204
1205 if (!skip_quota)
1206 ext4_xattr_inode_free_quota(parent, ea_inode,
1207 le32_to_cpu(entry->e_value_size));
1208
1209 /*
1210 * Forget about ea_inode within the same transaction that
1211 * decrements the ref count. This avoids duplicate decrements in
1212 * case the rest of the work spills over to subsequent
1213 * transactions.
1214 */
1215 entry->e_value_inum = 0;
1216 entry->e_value_size = 0;
1217
1218 dirty = true;
1219 }
1220
1221 if (dirty) {
1222 /*
1223 * Note that we are deliberately skipping csum calculation for
1224 * the final update because we do not expect any journal
1225 * restarts until xattr block is freed.
1226 */
1227
1228 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1229 if (err)
1230 ext4_warning_inode(parent,
1231 "handle dirty metadata err=%d", err);
1232 }
1233}
1234
1235/*
1236 * Release the xattr block BH: If the reference count is > 1, decrement it;
1237 * otherwise free the block.
1238 */
1239static void
1240ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1241 struct buffer_head *bh,
1242 struct ext4_xattr_inode_array **ea_inode_array,
1243 int extra_credits)
1244{
1245 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1246 u32 hash, ref;
1247 int error = 0;
1248
1249 BUFFER_TRACE(bh, "get_write_access");
1250 error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1251 EXT4_JTR_NONE);
1252 if (error)
1253 goto out;
1254
1255retry_ref:
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 struct mb_cache_entry *oe;
1267
1268 oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1269 bh->b_blocknr);
1270 if (oe) {
1271 unlock_buffer(bh);
1272 mb_cache_entry_wait_unused(oe);
1273 mb_cache_entry_put(ea_block_cache, oe);
1274 goto retry_ref;
1275 }
1276 }
1277 get_bh(bh);
1278 unlock_buffer(bh);
1279
1280 if (ext4_has_feature_ea_inode(inode->i_sb))
1281 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1282 BFIRST(bh),
1283 true /* block_csum */,
1284 ea_inode_array,
1285 extra_credits,
1286 true /* skip_quota */);
1287 ext4_free_blocks(handle, inode, bh, 0, 1,
1288 EXT4_FREE_BLOCKS_METADATA |
1289 EXT4_FREE_BLOCKS_FORGET);
1290 } else {
1291 ref--;
1292 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1293 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1294 struct mb_cache_entry *ce;
1295
1296 if (ea_block_cache) {
1297 ce = mb_cache_entry_get(ea_block_cache, hash,
1298 bh->b_blocknr);
1299 if (ce) {
1300 set_bit(MBE_REUSABLE_B, &ce->e_flags);
1301 mb_cache_entry_put(ea_block_cache, ce);
1302 }
1303 }
1304 }
1305
1306 ext4_xattr_block_csum_set(inode, bh);
1307 /*
1308 * Beware of this ugliness: Releasing of xattr block references
1309 * from different inodes can race and so we have to protect
1310 * from a race where someone else frees the block (and releases
1311 * its journal_head) before we are done dirtying the buffer. In
1312 * nojournal mode this race is harmless and we actually cannot
1313 * call ext4_handle_dirty_metadata() with locked buffer as
1314 * that function can call sync_dirty_buffer() so for that case
1315 * we handle the dirtying after unlocking the buffer.
1316 */
1317 if (ext4_handle_valid(handle))
1318 error = ext4_handle_dirty_metadata(handle, inode, bh);
1319 unlock_buffer(bh);
1320 if (!ext4_handle_valid(handle))
1321 error = ext4_handle_dirty_metadata(handle, inode, bh);
1322 if (IS_SYNC(inode))
1323 ext4_handle_sync(handle);
1324 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1325 ea_bdebug(bh, "refcount now=%d; releasing",
1326 le32_to_cpu(BHDR(bh)->h_refcount));
1327 }
1328out:
1329 ext4_std_error(inode->i_sb, error);
1330 return;
1331}
1332
1333/*
1334 * Find the available free space for EAs. This also returns the total number of
1335 * bytes used by EA entries.
1336 */
1337static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1338 size_t *min_offs, void *base, int *total)
1339{
1340 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1341 if (!last->e_value_inum && last->e_value_size) {
1342 size_t offs = le16_to_cpu(last->e_value_offs);
1343 if (offs < *min_offs)
1344 *min_offs = offs;
1345 }
1346 if (total)
1347 *total += EXT4_XATTR_LEN(last->e_name_len);
1348 }
1349 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1350}
1351
1352/*
1353 * Write the value of the EA in an inode.
1354 */
1355static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1356 const void *buf, int bufsize)
1357{
1358 struct buffer_head *bh = NULL;
1359 unsigned long block = 0;
1360 int blocksize = ea_inode->i_sb->s_blocksize;
1361 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1362 int csize, wsize = 0;
1363 int ret = 0, ret2 = 0;
1364 int retries = 0;
1365
1366retry:
1367 while (ret >= 0 && ret < max_blocks) {
1368 struct ext4_map_blocks map;
1369 map.m_lblk = block += ret;
1370 map.m_len = max_blocks -= ret;
1371
1372 ret = ext4_map_blocks(handle, ea_inode, &map,
1373 EXT4_GET_BLOCKS_CREATE);
1374 if (ret <= 0) {
1375 ext4_mark_inode_dirty(handle, ea_inode);
1376 if (ret == -ENOSPC &&
1377 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1378 ret = 0;
1379 goto retry;
1380 }
1381 break;
1382 }
1383 }
1384
1385 if (ret < 0)
1386 return ret;
1387
1388 block = 0;
1389 while (wsize < bufsize) {
1390 brelse(bh);
1391 csize = (bufsize - wsize) > blocksize ? blocksize :
1392 bufsize - wsize;
1393 bh = ext4_getblk(handle, ea_inode, block, 0);
1394 if (IS_ERR(bh))
1395 return PTR_ERR(bh);
1396 if (!bh) {
1397 WARN_ON_ONCE(1);
1398 EXT4_ERROR_INODE(ea_inode,
1399 "ext4_getblk() return bh = NULL");
1400 return -EFSCORRUPTED;
1401 }
1402 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1403 EXT4_JTR_NONE);
1404 if (ret)
1405 goto out;
1406
1407 memcpy(bh->b_data, buf, csize);
1408 set_buffer_uptodate(bh);
1409 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1410
1411 buf += csize;
1412 wsize += csize;
1413 block += 1;
1414 }
1415
1416 inode_lock(ea_inode);
1417 i_size_write(ea_inode, wsize);
1418 ext4_update_i_disksize(ea_inode, wsize);
1419 inode_unlock(ea_inode);
1420
1421 ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1422 if (unlikely(ret2 && !ret))
1423 ret = ret2;
1424
1425out:
1426 brelse(bh);
1427
1428 return ret;
1429}
1430
1431/*
1432 * Create an inode to store the value of a large EA.
1433 */
1434static struct inode *ext4_xattr_inode_create(handle_t *handle,
1435 struct inode *inode, u32 hash)
1436{
1437 struct inode *ea_inode = NULL;
1438 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1439 int err;
1440
1441 /*
1442 * Let the next inode be the goal, so we try and allocate the EA inode
1443 * in the same group, or nearby one.
1444 */
1445 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1446 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1447 EXT4_EA_INODE_FL);
1448 if (!IS_ERR(ea_inode)) {
1449 ea_inode->i_op = &ext4_file_inode_operations;
1450 ea_inode->i_fop = &ext4_file_operations;
1451 ext4_set_aops(ea_inode);
1452 ext4_xattr_inode_set_class(ea_inode);
1453 unlock_new_inode(ea_inode);
1454 ext4_xattr_inode_set_ref(ea_inode, 1);
1455 ext4_xattr_inode_set_hash(ea_inode, hash);
1456 err = ext4_mark_inode_dirty(handle, ea_inode);
1457 if (!err)
1458 err = ext4_inode_attach_jinode(ea_inode);
1459 if (err) {
1460 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1461 ext4_warning_inode(ea_inode,
1462 "cleanup dec ref error %d", err);
1463 iput(ea_inode);
1464 return ERR_PTR(err);
1465 }
1466
1467 /*
1468 * Xattr inodes are shared therefore quota charging is performed
1469 * at a higher level.
1470 */
1471 dquot_free_inode(ea_inode);
1472 dquot_drop(ea_inode);
1473 inode_lock(ea_inode);
1474 ea_inode->i_flags |= S_NOQUOTA;
1475 inode_unlock(ea_inode);
1476 }
1477
1478 return ea_inode;
1479}
1480
1481static struct inode *
1482ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1483 size_t value_len, u32 hash)
1484{
1485 struct inode *ea_inode;
1486 struct mb_cache_entry *ce;
1487 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1488 void *ea_data;
1489
1490 if (!ea_inode_cache)
1491 return NULL;
1492
1493 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1494 if (!ce)
1495 return NULL;
1496
1497 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1498 !(current->flags & PF_MEMALLOC_NOFS));
1499
1500 ea_data = kvmalloc(value_len, GFP_KERNEL);
1501 if (!ea_data) {
1502 mb_cache_entry_put(ea_inode_cache, ce);
1503 return NULL;
1504 }
1505
1506 while (ce) {
1507 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1508 EXT4_IGET_NORMAL);
1509 if (!IS_ERR(ea_inode) &&
1510 !is_bad_inode(ea_inode) &&
1511 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1512 i_size_read(ea_inode) == value_len &&
1513 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1514 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1515 value_len) &&
1516 !memcmp(value, ea_data, value_len)) {
1517 mb_cache_entry_touch(ea_inode_cache, ce);
1518 mb_cache_entry_put(ea_inode_cache, ce);
1519 kvfree(ea_data);
1520 return ea_inode;
1521 }
1522
1523 if (!IS_ERR(ea_inode))
1524 iput(ea_inode);
1525 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1526 }
1527 kvfree(ea_data);
1528 return NULL;
1529}
1530
1531/*
1532 * Add value of the EA in an inode.
1533 */
1534static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1535 const void *value, size_t value_len,
1536 struct inode **ret_inode)
1537{
1538 struct inode *ea_inode;
1539 u32 hash;
1540 int err;
1541
1542 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1543 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1544 if (ea_inode) {
1545 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1546 if (err) {
1547 iput(ea_inode);
1548 return err;
1549 }
1550
1551 *ret_inode = ea_inode;
1552 return 0;
1553 }
1554
1555 /* Create an inode for the EA value */
1556 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1557 if (IS_ERR(ea_inode))
1558 return PTR_ERR(ea_inode);
1559
1560 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1561 if (err) {
1562 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1563 ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1564 iput(ea_inode);
1565 return err;
1566 }
1567
1568 if (EA_INODE_CACHE(inode))
1569 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1570 ea_inode->i_ino, true /* reusable */);
1571
1572 *ret_inode = ea_inode;
1573 return 0;
1574}
1575
1576/*
1577 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1578 * feature is enabled.
1579 */
1580#define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1581
1582static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1583 struct ext4_xattr_search *s,
1584 handle_t *handle, struct inode *inode,
1585 bool is_block)
1586{
1587 struct ext4_xattr_entry *last, *next;
1588 struct ext4_xattr_entry *here = s->here;
1589 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1590 int in_inode = i->in_inode;
1591 struct inode *old_ea_inode = NULL;
1592 struct inode *new_ea_inode = NULL;
1593 size_t old_size, new_size;
1594 int ret;
1595
1596 /* Space used by old and new values. */
1597 old_size = (!s->not_found && !here->e_value_inum) ?
1598 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1599 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1600
1601 /*
1602 * Optimization for the simple case when old and new values have the
1603 * same padded sizes. Not applicable if external inodes are involved.
1604 */
1605 if (new_size && new_size == old_size) {
1606 size_t offs = le16_to_cpu(here->e_value_offs);
1607 void *val = s->base + offs;
1608
1609 here->e_value_size = cpu_to_le32(i->value_len);
1610 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1611 memset(val, 0, new_size);
1612 } else {
1613 memcpy(val, i->value, i->value_len);
1614 /* Clear padding bytes. */
1615 memset(val + i->value_len, 0, new_size - i->value_len);
1616 }
1617 goto update_hash;
1618 }
1619
1620 /* Compute min_offs and last. */
1621 last = s->first;
1622 for (; !IS_LAST_ENTRY(last); last = next) {
1623 next = EXT4_XATTR_NEXT(last);
1624 if ((void *)next >= s->end) {
1625 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1626 ret = -EFSCORRUPTED;
1627 goto out;
1628 }
1629 if (!last->e_value_inum && last->e_value_size) {
1630 size_t offs = le16_to_cpu(last->e_value_offs);
1631 if (offs < min_offs)
1632 min_offs = offs;
1633 }
1634 }
1635
1636 /* Check whether we have enough space. */
1637 if (i->value) {
1638 size_t free;
1639
1640 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1641 if (!s->not_found)
1642 free += EXT4_XATTR_LEN(name_len) + old_size;
1643
1644 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1645 ret = -ENOSPC;
1646 goto out;
1647 }
1648
1649 /*
1650 * If storing the value in an external inode is an option,
1651 * reserve space for xattr entries/names in the external
1652 * attribute block so that a long value does not occupy the
1653 * whole space and prevent further entries being added.
1654 */
1655 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1656 new_size && is_block &&
1657 (min_offs + old_size - new_size) <
1658 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1659 ret = -ENOSPC;
1660 goto out;
1661 }
1662 }
1663
1664 /*
1665 * Getting access to old and new ea inodes is subject to failures.
1666 * Finish that work before doing any modifications to the xattr data.
1667 */
1668 if (!s->not_found && here->e_value_inum) {
1669 ret = ext4_xattr_inode_iget(inode,
1670 le32_to_cpu(here->e_value_inum),
1671 le32_to_cpu(here->e_hash),
1672 &old_ea_inode);
1673 if (ret) {
1674 old_ea_inode = NULL;
1675 goto out;
1676 }
1677 }
1678 if (i->value && in_inode) {
1679 WARN_ON_ONCE(!i->value_len);
1680
1681 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1682 if (ret)
1683 goto out;
1684
1685 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1686 i->value_len,
1687 &new_ea_inode);
1688 if (ret) {
1689 new_ea_inode = NULL;
1690 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1691 goto out;
1692 }
1693 }
1694
1695 if (old_ea_inode) {
1696 /* We are ready to release ref count on the old_ea_inode. */
1697 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1698 if (ret) {
1699 /* Release newly required ref count on new_ea_inode. */
1700 if (new_ea_inode) {
1701 int err;
1702
1703 err = ext4_xattr_inode_dec_ref(handle,
1704 new_ea_inode);
1705 if (err)
1706 ext4_warning_inode(new_ea_inode,
1707 "dec ref new_ea_inode err=%d",
1708 err);
1709 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1710 i->value_len);
1711 }
1712 goto out;
1713 }
1714
1715 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1716 le32_to_cpu(here->e_value_size));
1717 }
1718
1719 /* No failures allowed past this point. */
1720
1721 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1722 /* Remove the old value. */
1723 void *first_val = s->base + min_offs;
1724 size_t offs = le16_to_cpu(here->e_value_offs);
1725 void *val = s->base + offs;
1726
1727 memmove(first_val + old_size, first_val, val - first_val);
1728 memset(first_val, 0, old_size);
1729 min_offs += old_size;
1730
1731 /* Adjust all value offsets. */
1732 last = s->first;
1733 while (!IS_LAST_ENTRY(last)) {
1734 size_t o = le16_to_cpu(last->e_value_offs);
1735
1736 if (!last->e_value_inum &&
1737 last->e_value_size && o < offs)
1738 last->e_value_offs = cpu_to_le16(o + old_size);
1739 last = EXT4_XATTR_NEXT(last);
1740 }
1741 }
1742
1743 if (!i->value) {
1744 /* Remove old name. */
1745 size_t size = EXT4_XATTR_LEN(name_len);
1746
1747 last = ENTRY((void *)last - size);
1748 memmove(here, (void *)here + size,
1749 (void *)last - (void *)here + sizeof(__u32));
1750 memset(last, 0, size);
1751 } else if (s->not_found) {
1752 /* Insert new name. */
1753 size_t size = EXT4_XATTR_LEN(name_len);
1754 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1755
1756 memmove((void *)here + size, here, rest);
1757 memset(here, 0, size);
1758 here->e_name_index = i->name_index;
1759 here->e_name_len = name_len;
1760 memcpy(here->e_name, i->name, name_len);
1761 } else {
1762 /* This is an update, reset value info. */
1763 here->e_value_inum = 0;
1764 here->e_value_offs = 0;
1765 here->e_value_size = 0;
1766 }
1767
1768 if (i->value) {
1769 /* Insert new value. */
1770 if (in_inode) {
1771 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1772 } else if (i->value_len) {
1773 void *val = s->base + min_offs - new_size;
1774
1775 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1776 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1777 memset(val, 0, new_size);
1778 } else {
1779 memcpy(val, i->value, i->value_len);
1780 /* Clear padding bytes. */
1781 memset(val + i->value_len, 0,
1782 new_size - i->value_len);
1783 }
1784 }
1785 here->e_value_size = cpu_to_le32(i->value_len);
1786 }
1787
1788update_hash:
1789 if (i->value) {
1790 __le32 hash = 0;
1791
1792 /* Entry hash calculation. */
1793 if (in_inode) {
1794 __le32 crc32c_hash;
1795
1796 /*
1797 * Feed crc32c hash instead of the raw value for entry
1798 * hash calculation. This is to avoid walking
1799 * potentially long value buffer again.
1800 */
1801 crc32c_hash = cpu_to_le32(
1802 ext4_xattr_inode_get_hash(new_ea_inode));
1803 hash = ext4_xattr_hash_entry(here->e_name,
1804 here->e_name_len,
1805 &crc32c_hash, 1);
1806 } else if (is_block) {
1807 __le32 *value = s->base + le16_to_cpu(
1808 here->e_value_offs);
1809
1810 hash = ext4_xattr_hash_entry(here->e_name,
1811 here->e_name_len, value,
1812 new_size >> 2);
1813 }
1814 here->e_hash = hash;
1815 }
1816
1817 if (is_block)
1818 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1819
1820 ret = 0;
1821out:
1822 iput(old_ea_inode);
1823 iput(new_ea_inode);
1824 return ret;
1825}
1826
1827struct ext4_xattr_block_find {
1828 struct ext4_xattr_search s;
1829 struct buffer_head *bh;
1830};
1831
1832static int
1833ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1834 struct ext4_xattr_block_find *bs)
1835{
1836 struct super_block *sb = inode->i_sb;
1837 int error;
1838
1839 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1840 i->name_index, i->name, i->value, (long)i->value_len);
1841
1842 if (EXT4_I(inode)->i_file_acl) {
1843 /* The inode already has an extended attribute block. */
1844 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1845 if (IS_ERR(bs->bh)) {
1846 error = PTR_ERR(bs->bh);
1847 bs->bh = NULL;
1848 return error;
1849 }
1850 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1851 atomic_read(&(bs->bh->b_count)),
1852 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1853 error = ext4_xattr_check_block(inode, bs->bh);
1854 if (error)
1855 return error;
1856 /* Find the named attribute. */
1857 bs->s.base = BHDR(bs->bh);
1858 bs->s.first = BFIRST(bs->bh);
1859 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1860 bs->s.here = bs->s.first;
1861 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1862 i->name_index, i->name, 1);
1863 if (error && error != -ENODATA)
1864 return error;
1865 bs->s.not_found = error;
1866 }
1867 return 0;
1868}
1869
1870static int
1871ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1872 struct ext4_xattr_info *i,
1873 struct ext4_xattr_block_find *bs)
1874{
1875 struct super_block *sb = inode->i_sb;
1876 struct buffer_head *new_bh = NULL;
1877 struct ext4_xattr_search s_copy = bs->s;
1878 struct ext4_xattr_search *s = &s_copy;
1879 struct mb_cache_entry *ce = NULL;
1880 int error = 0;
1881 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1882 struct inode *ea_inode = NULL, *tmp_inode;
1883 size_t old_ea_inode_quota = 0;
1884 unsigned int ea_ino;
1885
1886
1887#define header(x) ((struct ext4_xattr_header *)(x))
1888
1889 if (s->base) {
1890 int offset = (char *)s->here - bs->bh->b_data;
1891
1892 BUFFER_TRACE(bs->bh, "get_write_access");
1893 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1894 EXT4_JTR_NONE);
1895 if (error)
1896 goto cleanup;
1897 lock_buffer(bs->bh);
1898
1899 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1900 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1901
1902 /*
1903 * This must happen under buffer lock for
1904 * ext4_xattr_block_set() to reliably detect modified
1905 * block
1906 */
1907 if (ea_block_cache) {
1908 struct mb_cache_entry *oe;
1909
1910 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1911 hash, bs->bh->b_blocknr);
1912 if (oe) {
1913 /*
1914 * Xattr block is getting reused. Leave
1915 * it alone.
1916 */
1917 mb_cache_entry_put(ea_block_cache, oe);
1918 goto clone_block;
1919 }
1920 }
1921 ea_bdebug(bs->bh, "modifying in-place");
1922 error = ext4_xattr_set_entry(i, s, handle, inode,
1923 true /* is_block */);
1924 ext4_xattr_block_csum_set(inode, bs->bh);
1925 unlock_buffer(bs->bh);
1926 if (error == -EFSCORRUPTED)
1927 goto bad_block;
1928 if (!error)
1929 error = ext4_handle_dirty_metadata(handle,
1930 inode,
1931 bs->bh);
1932 if (error)
1933 goto cleanup;
1934 goto inserted;
1935 }
1936clone_block:
1937 unlock_buffer(bs->bh);
1938 ea_bdebug(bs->bh, "cloning");
1939 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1940 error = -ENOMEM;
1941 if (s->base == NULL)
1942 goto cleanup;
1943 s->first = ENTRY(header(s->base)+1);
1944 header(s->base)->h_refcount = cpu_to_le32(1);
1945 s->here = ENTRY(s->base + offset);
1946 s->end = s->base + bs->bh->b_size;
1947
1948 /*
1949 * If existing entry points to an xattr inode, we need
1950 * to prevent ext4_xattr_set_entry() from decrementing
1951 * ref count on it because the reference belongs to the
1952 * original block. In this case, make the entry look
1953 * like it has an empty value.
1954 */
1955 if (!s->not_found && s->here->e_value_inum) {
1956 ea_ino = le32_to_cpu(s->here->e_value_inum);
1957 error = ext4_xattr_inode_iget(inode, ea_ino,
1958 le32_to_cpu(s->here->e_hash),
1959 &tmp_inode);
1960 if (error)
1961 goto cleanup;
1962
1963 if (!ext4_test_inode_state(tmp_inode,
1964 EXT4_STATE_LUSTRE_EA_INODE)) {
1965 /*
1966 * Defer quota free call for previous
1967 * inode until success is guaranteed.
1968 */
1969 old_ea_inode_quota = le32_to_cpu(
1970 s->here->e_value_size);
1971 }
1972 iput(tmp_inode);
1973
1974 s->here->e_value_inum = 0;
1975 s->here->e_value_size = 0;
1976 }
1977 } else {
1978 /* Allocate a buffer where we construct the new block. */
1979 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1980 error = -ENOMEM;
1981 if (s->base == NULL)
1982 goto cleanup;
1983 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1984 header(s->base)->h_blocks = cpu_to_le32(1);
1985 header(s->base)->h_refcount = cpu_to_le32(1);
1986 s->first = ENTRY(header(s->base)+1);
1987 s->here = ENTRY(header(s->base)+1);
1988 s->end = s->base + sb->s_blocksize;
1989 }
1990
1991 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1992 if (error == -EFSCORRUPTED)
1993 goto bad_block;
1994 if (error)
1995 goto cleanup;
1996
1997 if (i->value && s->here->e_value_inum) {
1998 /*
1999 * A ref count on ea_inode has been taken as part of the call to
2000 * ext4_xattr_set_entry() above. We would like to drop this
2001 * extra ref but we have to wait until the xattr block is
2002 * initialized and has its own ref count on the ea_inode.
2003 */
2004 ea_ino = le32_to_cpu(s->here->e_value_inum);
2005 error = ext4_xattr_inode_iget(inode, ea_ino,
2006 le32_to_cpu(s->here->e_hash),
2007 &ea_inode);
2008 if (error) {
2009 ea_inode = NULL;
2010 goto cleanup;
2011 }
2012 }
2013
2014inserted:
2015 if (!IS_LAST_ENTRY(s->first)) {
2016 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2017 &ce);
2018 if (new_bh) {
2019 /* We found an identical block in the cache. */
2020 if (new_bh == bs->bh)
2021 ea_bdebug(new_bh, "keeping");
2022 else {
2023 u32 ref;
2024
2025 WARN_ON_ONCE(dquot_initialize_needed(inode));
2026
2027 /* The old block is released after updating
2028 the inode. */
2029 error = dquot_alloc_block(inode,
2030 EXT4_C2B(EXT4_SB(sb), 1));
2031 if (error)
2032 goto cleanup;
2033 BUFFER_TRACE(new_bh, "get_write_access");
2034 error = ext4_journal_get_write_access(
2035 handle, sb, new_bh,
2036 EXT4_JTR_NONE);
2037 if (error)
2038 goto cleanup_dquot;
2039 lock_buffer(new_bh);
2040 /*
2041 * We have to be careful about races with
2042 * adding references to xattr block. Once we
2043 * hold buffer lock xattr block's state is
2044 * stable so we can check the additional
2045 * reference fits.
2046 */
2047 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2048 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2049 /*
2050 * Undo everything and check mbcache
2051 * again.
2052 */
2053 unlock_buffer(new_bh);
2054 dquot_free_block(inode,
2055 EXT4_C2B(EXT4_SB(sb),
2056 1));
2057 brelse(new_bh);
2058 mb_cache_entry_put(ea_block_cache, ce);
2059 ce = NULL;
2060 new_bh = NULL;
2061 goto inserted;
2062 }
2063 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2064 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2065 clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2066 ea_bdebug(new_bh, "reusing; refcount now=%d",
2067 ref);
2068 ext4_xattr_block_csum_set(inode, new_bh);
2069 unlock_buffer(new_bh);
2070 error = ext4_handle_dirty_metadata(handle,
2071 inode,
2072 new_bh);
2073 if (error)
2074 goto cleanup_dquot;
2075 }
2076 mb_cache_entry_touch(ea_block_cache, ce);
2077 mb_cache_entry_put(ea_block_cache, ce);
2078 ce = NULL;
2079 } else if (bs->bh && s->base == bs->bh->b_data) {
2080 /* We were modifying this block in-place. */
2081 ea_bdebug(bs->bh, "keeping this block");
2082 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2083 new_bh = bs->bh;
2084 get_bh(new_bh);
2085 } else {
2086 /* We need to allocate a new block */
2087 ext4_fsblk_t goal, block;
2088
2089 WARN_ON_ONCE(dquot_initialize_needed(inode));
2090
2091 goal = ext4_group_first_block_no(sb,
2092 EXT4_I(inode)->i_block_group);
2093 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2094 NULL, &error);
2095 if (error)
2096 goto cleanup;
2097
2098 ea_idebug(inode, "creating block %llu",
2099 (unsigned long long)block);
2100
2101 new_bh = sb_getblk(sb, block);
2102 if (unlikely(!new_bh)) {
2103 error = -ENOMEM;
2104getblk_failed:
2105 ext4_free_blocks(handle, inode, NULL, block, 1,
2106 EXT4_FREE_BLOCKS_METADATA);
2107 goto cleanup;
2108 }
2109 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2110 ENTRY(header(s->base)+1));
2111 if (error)
2112 goto getblk_failed;
2113 if (ea_inode) {
2114 /* Drop the extra ref on ea_inode. */
2115 error = ext4_xattr_inode_dec_ref(handle,
2116 ea_inode);
2117 if (error)
2118 ext4_warning_inode(ea_inode,
2119 "dec ref error=%d",
2120 error);
2121 iput(ea_inode);
2122 ea_inode = NULL;
2123 }
2124
2125 lock_buffer(new_bh);
2126 error = ext4_journal_get_create_access(handle, sb,
2127 new_bh, EXT4_JTR_NONE);
2128 if (error) {
2129 unlock_buffer(new_bh);
2130 error = -EIO;
2131 goto getblk_failed;
2132 }
2133 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2134 ext4_xattr_block_csum_set(inode, new_bh);
2135 set_buffer_uptodate(new_bh);
2136 unlock_buffer(new_bh);
2137 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2138 error = ext4_handle_dirty_metadata(handle, inode,
2139 new_bh);
2140 if (error)
2141 goto cleanup;
2142 }
2143 }
2144
2145 if (old_ea_inode_quota)
2146 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2147
2148 /* Update the inode. */
2149 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2150
2151 /* Drop the previous xattr block. */
2152 if (bs->bh && bs->bh != new_bh) {
2153 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2154
2155 ext4_xattr_release_block(handle, inode, bs->bh,
2156 &ea_inode_array,
2157 0 /* extra_credits */);
2158 ext4_xattr_inode_array_free(ea_inode_array);
2159 }
2160 error = 0;
2161
2162cleanup:
2163 if (ea_inode) {
2164 int error2;
2165
2166 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2167 if (error2)
2168 ext4_warning_inode(ea_inode, "dec ref error=%d",
2169 error2);
2170
2171 /* If there was an error, revert the quota charge. */
2172 if (error)
2173 ext4_xattr_inode_free_quota(inode, ea_inode,
2174 i_size_read(ea_inode));
2175 iput(ea_inode);
2176 }
2177 if (ce)
2178 mb_cache_entry_put(ea_block_cache, ce);
2179 brelse(new_bh);
2180 if (!(bs->bh && s->base == bs->bh->b_data))
2181 kfree(s->base);
2182
2183 return error;
2184
2185cleanup_dquot:
2186 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2187 goto cleanup;
2188
2189bad_block:
2190 EXT4_ERROR_INODE(inode, "bad block %llu",
2191 EXT4_I(inode)->i_file_acl);
2192 goto cleanup;
2193
2194#undef header
2195}
2196
2197int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2198 struct ext4_xattr_ibody_find *is)
2199{
2200 struct ext4_xattr_ibody_header *header;
2201 struct ext4_inode *raw_inode;
2202 int error;
2203
2204 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2205 return 0;
2206
2207 raw_inode = ext4_raw_inode(&is->iloc);
2208 header = IHDR(inode, raw_inode);
2209 is->s.base = is->s.first = IFIRST(header);
2210 is->s.here = is->s.first;
2211 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2212 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2213 error = xattr_check_inode(inode, header, is->s.end);
2214 if (error)
2215 return error;
2216 /* Find the named attribute. */
2217 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2218 i->name_index, i->name, 0);
2219 if (error && error != -ENODATA)
2220 return error;
2221 is->s.not_found = error;
2222 }
2223 return 0;
2224}
2225
2226int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2227 struct ext4_xattr_info *i,
2228 struct ext4_xattr_ibody_find *is)
2229{
2230 struct ext4_xattr_ibody_header *header;
2231 struct ext4_xattr_search *s = &is->s;
2232 int error;
2233
2234 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2235 return -ENOSPC;
2236
2237 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2238 if (error)
2239 return error;
2240 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2241 if (!IS_LAST_ENTRY(s->first)) {
2242 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2243 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2244 } else {
2245 header->h_magic = cpu_to_le32(0);
2246 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2247 }
2248 return 0;
2249}
2250
2251static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2252 struct ext4_xattr_info *i)
2253{
2254 void *value;
2255
2256 /* When e_value_inum is set the value is stored externally. */
2257 if (s->here->e_value_inum)
2258 return 0;
2259 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2260 return 0;
2261 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2262 return !memcmp(value, i->value, i->value_len);
2263}
2264
2265static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2266{
2267 struct buffer_head *bh;
2268 int error;
2269
2270 if (!EXT4_I(inode)->i_file_acl)
2271 return NULL;
2272 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2273 if (IS_ERR(bh))
2274 return bh;
2275 error = ext4_xattr_check_block(inode, bh);
2276 if (error) {
2277 brelse(bh);
2278 return ERR_PTR(error);
2279 }
2280 return bh;
2281}
2282
2283/*
2284 * ext4_xattr_set_handle()
2285 *
2286 * Create, replace or remove an extended attribute for this inode. Value
2287 * is NULL to remove an existing extended attribute, and non-NULL to
2288 * either replace an existing extended attribute, or create a new extended
2289 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2290 * specify that an extended attribute must exist and must not exist
2291 * previous to the call, respectively.
2292 *
2293 * Returns 0, or a negative error number on failure.
2294 */
2295int
2296ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2297 const char *name, const void *value, size_t value_len,
2298 int flags)
2299{
2300 struct ext4_xattr_info i = {
2301 .name_index = name_index,
2302 .name = name,
2303 .value = value,
2304 .value_len = value_len,
2305 .in_inode = 0,
2306 };
2307 struct ext4_xattr_ibody_find is = {
2308 .s = { .not_found = -ENODATA, },
2309 };
2310 struct ext4_xattr_block_find bs = {
2311 .s = { .not_found = -ENODATA, },
2312 };
2313 int no_expand;
2314 int error;
2315
2316 if (!name)
2317 return -EINVAL;
2318 if (strlen(name) > 255)
2319 return -ERANGE;
2320
2321 ext4_write_lock_xattr(inode, &no_expand);
2322
2323 /* Check journal credits under write lock. */
2324 if (ext4_handle_valid(handle)) {
2325 struct buffer_head *bh;
2326 int credits;
2327
2328 bh = ext4_xattr_get_block(inode);
2329 if (IS_ERR(bh)) {
2330 error = PTR_ERR(bh);
2331 goto cleanup;
2332 }
2333
2334 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2335 value_len,
2336 flags & XATTR_CREATE);
2337 brelse(bh);
2338
2339 if (jbd2_handle_buffer_credits(handle) < credits) {
2340 error = -ENOSPC;
2341 goto cleanup;
2342 }
2343 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2344 }
2345
2346 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2347 if (error)
2348 goto cleanup;
2349
2350 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2351 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2352 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2353 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2354 }
2355
2356 error = ext4_xattr_ibody_find(inode, &i, &is);
2357 if (error)
2358 goto cleanup;
2359 if (is.s.not_found)
2360 error = ext4_xattr_block_find(inode, &i, &bs);
2361 if (error)
2362 goto cleanup;
2363 if (is.s.not_found && bs.s.not_found) {
2364 error = -ENODATA;
2365 if (flags & XATTR_REPLACE)
2366 goto cleanup;
2367 error = 0;
2368 if (!value)
2369 goto cleanup;
2370 } else {
2371 error = -EEXIST;
2372 if (flags & XATTR_CREATE)
2373 goto cleanup;
2374 }
2375
2376 if (!value) {
2377 if (!is.s.not_found)
2378 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2379 else if (!bs.s.not_found)
2380 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2381 } else {
2382 error = 0;
2383 /* Xattr value did not change? Save us some work and bail out */
2384 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2385 goto cleanup;
2386 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2387 goto cleanup;
2388
2389 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2390 (EXT4_XATTR_SIZE(i.value_len) >
2391 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2392 i.in_inode = 1;
2393retry_inode:
2394 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2395 if (!error && !bs.s.not_found) {
2396 i.value = NULL;
2397 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2398 } else if (error == -ENOSPC) {
2399 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2400 brelse(bs.bh);
2401 bs.bh = NULL;
2402 error = ext4_xattr_block_find(inode, &i, &bs);
2403 if (error)
2404 goto cleanup;
2405 }
2406 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2407 if (!error && !is.s.not_found) {
2408 i.value = NULL;
2409 error = ext4_xattr_ibody_set(handle, inode, &i,
2410 &is);
2411 } else if (error == -ENOSPC) {
2412 /*
2413 * Xattr does not fit in the block, store at
2414 * external inode if possible.
2415 */
2416 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2417 i.value_len && !i.in_inode) {
2418 i.in_inode = 1;
2419 goto retry_inode;
2420 }
2421 }
2422 }
2423 }
2424 if (!error) {
2425 ext4_xattr_update_super_block(handle, inode->i_sb);
2426 inode->i_ctime = current_time(inode);
2427 inode_inc_iversion(inode);
2428 if (!value)
2429 no_expand = 0;
2430 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2431 /*
2432 * The bh is consumed by ext4_mark_iloc_dirty, even with
2433 * error != 0.
2434 */
2435 is.iloc.bh = NULL;
2436 if (IS_SYNC(inode))
2437 ext4_handle_sync(handle);
2438 }
2439 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2440
2441cleanup:
2442 brelse(is.iloc.bh);
2443 brelse(bs.bh);
2444 ext4_write_unlock_xattr(inode, &no_expand);
2445 return error;
2446}
2447
2448int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2449 bool is_create, int *credits)
2450{
2451 struct buffer_head *bh;
2452 int err;
2453
2454 *credits = 0;
2455
2456 if (!EXT4_SB(inode->i_sb)->s_journal)
2457 return 0;
2458
2459 down_read(&EXT4_I(inode)->xattr_sem);
2460
2461 bh = ext4_xattr_get_block(inode);
2462 if (IS_ERR(bh)) {
2463 err = PTR_ERR(bh);
2464 } else {
2465 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2466 value_len, is_create);
2467 brelse(bh);
2468 err = 0;
2469 }
2470
2471 up_read(&EXT4_I(inode)->xattr_sem);
2472 return err;
2473}
2474
2475/*
2476 * ext4_xattr_set()
2477 *
2478 * Like ext4_xattr_set_handle, but start from an inode. This extended
2479 * attribute modification is a filesystem transaction by itself.
2480 *
2481 * Returns 0, or a negative error number on failure.
2482 */
2483int
2484ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2485 const void *value, size_t value_len, int flags)
2486{
2487 handle_t *handle;
2488 struct super_block *sb = inode->i_sb;
2489 int error, retries = 0;
2490 int credits;
2491
2492 error = dquot_initialize(inode);
2493 if (error)
2494 return error;
2495
2496retry:
2497 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2498 &credits);
2499 if (error)
2500 return error;
2501
2502 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2503 if (IS_ERR(handle)) {
2504 error = PTR_ERR(handle);
2505 } else {
2506 int error2;
2507
2508 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2509 value, value_len, flags);
2510 error2 = ext4_journal_stop(handle);
2511 if (error == -ENOSPC &&
2512 ext4_should_retry_alloc(sb, &retries))
2513 goto retry;
2514 if (error == 0)
2515 error = error2;
2516 }
2517 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2518
2519 return error;
2520}
2521
2522/*
2523 * Shift the EA entries in the inode to create space for the increased
2524 * i_extra_isize.
2525 */
2526static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2527 int value_offs_shift, void *to,
2528 void *from, size_t n)
2529{
2530 struct ext4_xattr_entry *last = entry;
2531 int new_offs;
2532
2533 /* We always shift xattr headers further thus offsets get lower */
2534 BUG_ON(value_offs_shift > 0);
2535
2536 /* Adjust the value offsets of the entries */
2537 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2538 if (!last->e_value_inum && last->e_value_size) {
2539 new_offs = le16_to_cpu(last->e_value_offs) +
2540 value_offs_shift;
2541 last->e_value_offs = cpu_to_le16(new_offs);
2542 }
2543 }
2544 /* Shift the entries by n bytes */
2545 memmove(to, from, n);
2546}
2547
2548/*
2549 * Move xattr pointed to by 'entry' from inode into external xattr block
2550 */
2551static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2552 struct ext4_inode *raw_inode,
2553 struct ext4_xattr_entry *entry)
2554{
2555 struct ext4_xattr_ibody_find *is = NULL;
2556 struct ext4_xattr_block_find *bs = NULL;
2557 char *buffer = NULL, *b_entry_name = NULL;
2558 size_t value_size = le32_to_cpu(entry->e_value_size);
2559 struct ext4_xattr_info i = {
2560 .value = NULL,
2561 .value_len = 0,
2562 .name_index = entry->e_name_index,
2563 .in_inode = !!entry->e_value_inum,
2564 };
2565 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2566 int error;
2567
2568 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2569 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2570 buffer = kvmalloc(value_size, GFP_NOFS);
2571 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2572 if (!is || !bs || !buffer || !b_entry_name) {
2573 error = -ENOMEM;
2574 goto out;
2575 }
2576
2577 is->s.not_found = -ENODATA;
2578 bs->s.not_found = -ENODATA;
2579 is->iloc.bh = NULL;
2580 bs->bh = NULL;
2581
2582 /* Save the entry name and the entry value */
2583 if (entry->e_value_inum) {
2584 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2585 if (error)
2586 goto out;
2587 } else {
2588 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2589 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2590 }
2591
2592 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2593 b_entry_name[entry->e_name_len] = '\0';
2594 i.name = b_entry_name;
2595
2596 error = ext4_get_inode_loc(inode, &is->iloc);
2597 if (error)
2598 goto out;
2599
2600 error = ext4_xattr_ibody_find(inode, &i, is);
2601 if (error)
2602 goto out;
2603
2604 /* Remove the chosen entry from the inode */
2605 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2606 if (error)
2607 goto out;
2608
2609 i.value = buffer;
2610 i.value_len = value_size;
2611 error = ext4_xattr_block_find(inode, &i, bs);
2612 if (error)
2613 goto out;
2614
2615 /* Add entry which was removed from the inode into the block */
2616 error = ext4_xattr_block_set(handle, inode, &i, bs);
2617 if (error)
2618 goto out;
2619 error = 0;
2620out:
2621 kfree(b_entry_name);
2622 kvfree(buffer);
2623 if (is)
2624 brelse(is->iloc.bh);
2625 if (bs)
2626 brelse(bs->bh);
2627 kfree(is);
2628 kfree(bs);
2629
2630 return error;
2631}
2632
2633static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2634 struct ext4_inode *raw_inode,
2635 int isize_diff, size_t ifree,
2636 size_t bfree, int *total_ino)
2637{
2638 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2639 struct ext4_xattr_entry *small_entry;
2640 struct ext4_xattr_entry *entry;
2641 struct ext4_xattr_entry *last;
2642 unsigned int entry_size; /* EA entry size */
2643 unsigned int total_size; /* EA entry size + value size */
2644 unsigned int min_total_size;
2645 int error;
2646
2647 while (isize_diff > ifree) {
2648 entry = NULL;
2649 small_entry = NULL;
2650 min_total_size = ~0U;
2651 last = IFIRST(header);
2652 /* Find the entry best suited to be pushed into EA block */
2653 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2654 /* never move system.data out of the inode */
2655 if ((last->e_name_len == 4) &&
2656 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2657 !memcmp(last->e_name, "data", 4))
2658 continue;
2659 total_size = EXT4_XATTR_LEN(last->e_name_len);
2660 if (!last->e_value_inum)
2661 total_size += EXT4_XATTR_SIZE(
2662 le32_to_cpu(last->e_value_size));
2663 if (total_size <= bfree &&
2664 total_size < min_total_size) {
2665 if (total_size + ifree < isize_diff) {
2666 small_entry = last;
2667 } else {
2668 entry = last;
2669 min_total_size = total_size;
2670 }
2671 }
2672 }
2673
2674 if (entry == NULL) {
2675 if (small_entry == NULL)
2676 return -ENOSPC;
2677 entry = small_entry;
2678 }
2679
2680 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2681 total_size = entry_size;
2682 if (!entry->e_value_inum)
2683 total_size += EXT4_XATTR_SIZE(
2684 le32_to_cpu(entry->e_value_size));
2685 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2686 entry);
2687 if (error)
2688 return error;
2689
2690 *total_ino -= entry_size;
2691 ifree += total_size;
2692 bfree -= total_size;
2693 }
2694
2695 return 0;
2696}
2697
2698/*
2699 * Expand an inode by new_extra_isize bytes when EAs are present.
2700 * Returns 0 on success or negative error number on failure.
2701 */
2702int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2703 struct ext4_inode *raw_inode, handle_t *handle)
2704{
2705 struct ext4_xattr_ibody_header *header;
2706 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2707 static unsigned int mnt_count;
2708 size_t min_offs;
2709 size_t ifree, bfree;
2710 int total_ino;
2711 void *base, *end;
2712 int error = 0, tried_min_extra_isize = 0;
2713 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2714 int isize_diff; /* How much do we need to grow i_extra_isize */
2715
2716retry:
2717 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2718 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2719 return 0;
2720
2721 header = IHDR(inode, raw_inode);
2722
2723 /*
2724 * Check if enough free space is available in the inode to shift the
2725 * entries ahead by new_extra_isize.
2726 */
2727
2728 base = IFIRST(header);
2729 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2730 min_offs = end - base;
2731 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2732
2733 error = xattr_check_inode(inode, header, end);
2734 if (error)
2735 goto cleanup;
2736
2737 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2738 if (ifree >= isize_diff)
2739 goto shift;
2740
2741 /*
2742 * Enough free space isn't available in the inode, check if
2743 * EA block can hold new_extra_isize bytes.
2744 */
2745 if (EXT4_I(inode)->i_file_acl) {
2746 struct buffer_head *bh;
2747
2748 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2749 if (IS_ERR(bh)) {
2750 error = PTR_ERR(bh);
2751 goto cleanup;
2752 }
2753 error = ext4_xattr_check_block(inode, bh);
2754 if (error) {
2755 brelse(bh);
2756 goto cleanup;
2757 }
2758 base = BHDR(bh);
2759 end = bh->b_data + bh->b_size;
2760 min_offs = end - base;
2761 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2762 NULL);
2763 brelse(bh);
2764 if (bfree + ifree < isize_diff) {
2765 if (!tried_min_extra_isize && s_min_extra_isize) {
2766 tried_min_extra_isize++;
2767 new_extra_isize = s_min_extra_isize;
2768 goto retry;
2769 }
2770 error = -ENOSPC;
2771 goto cleanup;
2772 }
2773 } else {
2774 bfree = inode->i_sb->s_blocksize;
2775 }
2776
2777 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2778 isize_diff, ifree, bfree,
2779 &total_ino);
2780 if (error) {
2781 if (error == -ENOSPC && !tried_min_extra_isize &&
2782 s_min_extra_isize) {
2783 tried_min_extra_isize++;
2784 new_extra_isize = s_min_extra_isize;
2785 goto retry;
2786 }
2787 goto cleanup;
2788 }
2789shift:
2790 /* Adjust the offsets and shift the remaining entries ahead */
2791 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2792 - new_extra_isize, (void *)raw_inode +
2793 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2794 (void *)header, total_ino);
2795 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2796
2797cleanup:
2798 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2799 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2800 inode->i_ino);
2801 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2802 }
2803 return error;
2804}
2805
2806#define EIA_INCR 16 /* must be 2^n */
2807#define EIA_MASK (EIA_INCR - 1)
2808
2809/* Add the large xattr @inode into @ea_inode_array for deferred iput().
2810 * If @ea_inode_array is new or full it will be grown and the old
2811 * contents copied over.
2812 */
2813static int
2814ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2815 struct inode *inode)
2816{
2817 if (*ea_inode_array == NULL) {
2818 /*
2819 * Start with 15 inodes, so it fits into a power-of-two size.
2820 * If *ea_inode_array is NULL, this is essentially offsetof()
2821 */
2822 (*ea_inode_array) =
2823 kmalloc(offsetof(struct ext4_xattr_inode_array,
2824 inodes[EIA_MASK]),
2825 GFP_NOFS);
2826 if (*ea_inode_array == NULL)
2827 return -ENOMEM;
2828 (*ea_inode_array)->count = 0;
2829 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2830 /* expand the array once all 15 + n * 16 slots are full */
2831 struct ext4_xattr_inode_array *new_array = NULL;
2832 int count = (*ea_inode_array)->count;
2833
2834 /* if new_array is NULL, this is essentially offsetof() */
2835 new_array = kmalloc(
2836 offsetof(struct ext4_xattr_inode_array,
2837 inodes[count + EIA_INCR]),
2838 GFP_NOFS);
2839 if (new_array == NULL)
2840 return -ENOMEM;
2841 memcpy(new_array, *ea_inode_array,
2842 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2843 kfree(*ea_inode_array);
2844 *ea_inode_array = new_array;
2845 }
2846 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2847 return 0;
2848}
2849
2850/*
2851 * ext4_xattr_delete_inode()
2852 *
2853 * Free extended attribute resources associated with this inode. Traverse
2854 * all entries and decrement reference on any xattr inodes associated with this
2855 * inode. This is called immediately before an inode is freed. We have exclusive
2856 * access to the inode. If an orphan inode is deleted it will also release its
2857 * references on xattr block and xattr inodes.
2858 */
2859int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2860 struct ext4_xattr_inode_array **ea_inode_array,
2861 int extra_credits)
2862{
2863 struct buffer_head *bh = NULL;
2864 struct ext4_xattr_ibody_header *header;
2865 struct ext4_iloc iloc = { .bh = NULL };
2866 struct ext4_xattr_entry *entry;
2867 struct inode *ea_inode;
2868 int error;
2869
2870 error = ext4_journal_ensure_credits(handle, extra_credits,
2871 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2872 if (error < 0) {
2873 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2874 goto cleanup;
2875 }
2876
2877 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2878 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2879
2880 error = ext4_get_inode_loc(inode, &iloc);
2881 if (error) {
2882 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2883 goto cleanup;
2884 }
2885
2886 error = ext4_journal_get_write_access(handle, inode->i_sb,
2887 iloc.bh, EXT4_JTR_NONE);
2888 if (error) {
2889 EXT4_ERROR_INODE(inode, "write access (error %d)",
2890 error);
2891 goto cleanup;
2892 }
2893
2894 header = IHDR(inode, ext4_raw_inode(&iloc));
2895 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2896 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2897 IFIRST(header),
2898 false /* block_csum */,
2899 ea_inode_array,
2900 extra_credits,
2901 false /* skip_quota */);
2902 }
2903
2904 if (EXT4_I(inode)->i_file_acl) {
2905 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2906 if (IS_ERR(bh)) {
2907 error = PTR_ERR(bh);
2908 if (error == -EIO) {
2909 EXT4_ERROR_INODE_ERR(inode, EIO,
2910 "block %llu read error",
2911 EXT4_I(inode)->i_file_acl);
2912 }
2913 bh = NULL;
2914 goto cleanup;
2915 }
2916 error = ext4_xattr_check_block(inode, bh);
2917 if (error)
2918 goto cleanup;
2919
2920 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2921 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2922 entry = EXT4_XATTR_NEXT(entry)) {
2923 if (!entry->e_value_inum)
2924 continue;
2925 error = ext4_xattr_inode_iget(inode,
2926 le32_to_cpu(entry->e_value_inum),
2927 le32_to_cpu(entry->e_hash),
2928 &ea_inode);
2929 if (error)
2930 continue;
2931 ext4_xattr_inode_free_quota(inode, ea_inode,
2932 le32_to_cpu(entry->e_value_size));
2933 iput(ea_inode);
2934 }
2935
2936 }
2937
2938 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2939 extra_credits);
2940 /*
2941 * Update i_file_acl value in the same transaction that releases
2942 * block.
2943 */
2944 EXT4_I(inode)->i_file_acl = 0;
2945 error = ext4_mark_inode_dirty(handle, inode);
2946 if (error) {
2947 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2948 error);
2949 goto cleanup;
2950 }
2951 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2952 }
2953 error = 0;
2954cleanup:
2955 brelse(iloc.bh);
2956 brelse(bh);
2957 return error;
2958}
2959
2960void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2961{
2962 int idx;
2963
2964 if (ea_inode_array == NULL)
2965 return;
2966
2967 for (idx = 0; idx < ea_inode_array->count; ++idx)
2968 iput(ea_inode_array->inodes[idx]);
2969 kfree(ea_inode_array);
2970}
2971
2972/*
2973 * ext4_xattr_block_cache_insert()
2974 *
2975 * Create a new entry in the extended attribute block cache, and insert
2976 * it unless such an entry is already in the cache.
2977 *
2978 * Returns 0, or a negative error number on failure.
2979 */
2980static void
2981ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2982 struct buffer_head *bh)
2983{
2984 struct ext4_xattr_header *header = BHDR(bh);
2985 __u32 hash = le32_to_cpu(header->h_hash);
2986 int reusable = le32_to_cpu(header->h_refcount) <
2987 EXT4_XATTR_REFCOUNT_MAX;
2988 int error;
2989
2990 if (!ea_block_cache)
2991 return;
2992 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2993 bh->b_blocknr, reusable);
2994 if (error) {
2995 if (error == -EBUSY)
2996 ea_bdebug(bh, "already in cache");
2997 } else
2998 ea_bdebug(bh, "inserting [%x]", (int)hash);
2999}
3000
3001/*
3002 * ext4_xattr_cmp()
3003 *
3004 * Compare two extended attribute blocks for equality.
3005 *
3006 * Returns 0 if the blocks are equal, 1 if they differ, and
3007 * a negative error number on errors.
3008 */
3009static int
3010ext4_xattr_cmp(struct ext4_xattr_header *header1,
3011 struct ext4_xattr_header *header2)
3012{
3013 struct ext4_xattr_entry *entry1, *entry2;
3014
3015 entry1 = ENTRY(header1+1);
3016 entry2 = ENTRY(header2+1);
3017 while (!IS_LAST_ENTRY(entry1)) {
3018 if (IS_LAST_ENTRY(entry2))
3019 return 1;
3020 if (entry1->e_hash != entry2->e_hash ||
3021 entry1->e_name_index != entry2->e_name_index ||
3022 entry1->e_name_len != entry2->e_name_len ||
3023 entry1->e_value_size != entry2->e_value_size ||
3024 entry1->e_value_inum != entry2->e_value_inum ||
3025 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3026 return 1;
3027 if (!entry1->e_value_inum &&
3028 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3029 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3030 le32_to_cpu(entry1->e_value_size)))
3031 return 1;
3032
3033 entry1 = EXT4_XATTR_NEXT(entry1);
3034 entry2 = EXT4_XATTR_NEXT(entry2);
3035 }
3036 if (!IS_LAST_ENTRY(entry2))
3037 return 1;
3038 return 0;
3039}
3040
3041/*
3042 * ext4_xattr_block_cache_find()
3043 *
3044 * Find an identical extended attribute block.
3045 *
3046 * Returns a pointer to the block found, or NULL if such a block was
3047 * not found or an error occurred.
3048 */
3049static struct buffer_head *
3050ext4_xattr_block_cache_find(struct inode *inode,
3051 struct ext4_xattr_header *header,
3052 struct mb_cache_entry **pce)
3053{
3054 __u32 hash = le32_to_cpu(header->h_hash);
3055 struct mb_cache_entry *ce;
3056 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3057
3058 if (!ea_block_cache)
3059 return NULL;
3060 if (!header->h_hash)
3061 return NULL; /* never share */
3062 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3063 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3064 while (ce) {
3065 struct buffer_head *bh;
3066
3067 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3068 if (IS_ERR(bh)) {
3069 if (PTR_ERR(bh) == -ENOMEM)
3070 return NULL;
3071 bh = NULL;
3072 EXT4_ERROR_INODE(inode, "block %lu read error",
3073 (unsigned long)ce->e_value);
3074 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3075 *pce = ce;
3076 return bh;
3077 }
3078 brelse(bh);
3079 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3080 }
3081 return NULL;
3082}
3083
3084#define NAME_HASH_SHIFT 5
3085#define VALUE_HASH_SHIFT 16
3086
3087/*
3088 * ext4_xattr_hash_entry()
3089 *
3090 * Compute the hash of an extended attribute.
3091 */
3092static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3093 size_t value_count)
3094{
3095 __u32 hash = 0;
3096
3097 while (name_len--) {
3098 hash = (hash << NAME_HASH_SHIFT) ^
3099 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3100 (unsigned char)*name++;
3101 }
3102 while (value_count--) {
3103 hash = (hash << VALUE_HASH_SHIFT) ^
3104 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3105 le32_to_cpu(*value++);
3106 }
3107 return cpu_to_le32(hash);
3108}
3109
3110/*
3111 * ext4_xattr_hash_entry_signed()
3112 *
3113 * Compute the hash of an extended attribute incorrectly.
3114 */
3115static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3116{
3117 __u32 hash = 0;
3118
3119 while (name_len--) {
3120 hash = (hash << NAME_HASH_SHIFT) ^
3121 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3122 (signed char)*name++;
3123 }
3124 while (value_count--) {
3125 hash = (hash << VALUE_HASH_SHIFT) ^
3126 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3127 le32_to_cpu(*value++);
3128 }
3129 return cpu_to_le32(hash);
3130}
3131
3132#undef NAME_HASH_SHIFT
3133#undef VALUE_HASH_SHIFT
3134
3135#define BLOCK_HASH_SHIFT 16
3136
3137/*
3138 * ext4_xattr_rehash()
3139 *
3140 * Re-compute the extended attribute hash value after an entry has changed.
3141 */
3142static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3143{
3144 struct ext4_xattr_entry *here;
3145 __u32 hash = 0;
3146
3147 here = ENTRY(header+1);
3148 while (!IS_LAST_ENTRY(here)) {
3149 if (!here->e_hash) {
3150 /* Block is not shared if an entry's hash value == 0 */
3151 hash = 0;
3152 break;
3153 }
3154 hash = (hash << BLOCK_HASH_SHIFT) ^
3155 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3156 le32_to_cpu(here->e_hash);
3157 here = EXT4_XATTR_NEXT(here);
3158 }
3159 header->h_hash = cpu_to_le32(hash);
3160}
3161
3162#undef BLOCK_HASH_SHIFT
3163
3164#define HASH_BUCKET_BITS 10
3165
3166struct mb_cache *
3167ext4_xattr_create_cache(void)
3168{
3169 return mb_cache_create(HASH_BUCKET_BITS);
3170}
3171
3172void ext4_xattr_destroy_cache(struct mb_cache *cache)
3173{
3174 if (cache)
3175 mb_cache_destroy(cache);
3176}
3177
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
77static void ext4_xattr_block_cache_insert(struct mb_cache *,
78 struct buffer_head *);
79static struct buffer_head *
80ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
81 struct mb_cache_entry **);
82static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
83 size_t value_count);
84static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value,
85 size_t value_count);
86static void ext4_xattr_rehash(struct ext4_xattr_header *);
87
88static const struct xattr_handler * const ext4_xattr_handler_map[] = {
89 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
90#ifdef CONFIG_EXT4_FS_POSIX_ACL
91 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &nop_posix_acl_access,
92 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &nop_posix_acl_default,
93#endif
94 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
95#ifdef CONFIG_EXT4_FS_SECURITY
96 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
97#endif
98 [EXT4_XATTR_INDEX_HURD] = &ext4_xattr_hurd_handler,
99};
100
101const struct xattr_handler * const ext4_xattr_handlers[] = {
102 &ext4_xattr_user_handler,
103 &ext4_xattr_trusted_handler,
104#ifdef CONFIG_EXT4_FS_SECURITY
105 &ext4_xattr_security_handler,
106#endif
107 &ext4_xattr_hurd_handler,
108 NULL
109};
110
111#define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
112 inode->i_sb->s_fs_info)->s_ea_block_cache)
113
114#define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
115 inode->i_sb->s_fs_info)->s_ea_inode_cache)
116
117static int
118ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
119 struct inode *inode);
120
121#ifdef CONFIG_LOCKDEP
122void ext4_xattr_inode_set_class(struct inode *ea_inode)
123{
124 struct ext4_inode_info *ei = EXT4_I(ea_inode);
125
126 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
127 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */
128 lockdep_set_subclass(&ei->i_data_sem, I_DATA_SEM_EA);
129}
130#endif
131
132static __le32 ext4_xattr_block_csum(struct inode *inode,
133 sector_t block_nr,
134 struct ext4_xattr_header *hdr)
135{
136 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
137 __u32 csum;
138 __le64 dsk_block_nr = cpu_to_le64(block_nr);
139 __u32 dummy_csum = 0;
140 int offset = offsetof(struct ext4_xattr_header, h_checksum);
141
142 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
143 sizeof(dsk_block_nr));
144 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
145 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
146 offset += sizeof(dummy_csum);
147 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
148 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
149
150 return cpu_to_le32(csum);
151}
152
153static int ext4_xattr_block_csum_verify(struct inode *inode,
154 struct buffer_head *bh)
155{
156 struct ext4_xattr_header *hdr = BHDR(bh);
157 int ret = 1;
158
159 if (ext4_has_metadata_csum(inode->i_sb)) {
160 lock_buffer(bh);
161 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
162 bh->b_blocknr, hdr));
163 unlock_buffer(bh);
164 }
165 return ret;
166}
167
168static void ext4_xattr_block_csum_set(struct inode *inode,
169 struct buffer_head *bh)
170{
171 if (ext4_has_metadata_csum(inode->i_sb))
172 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
173 bh->b_blocknr, BHDR(bh));
174}
175
176static inline const char *ext4_xattr_prefix(int name_index,
177 struct dentry *dentry)
178{
179 const struct xattr_handler *handler = NULL;
180
181 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
182 handler = ext4_xattr_handler_map[name_index];
183
184 if (!xattr_handler_can_list(handler, dentry))
185 return NULL;
186
187 return xattr_prefix(handler);
188}
189
190static int
191check_xattrs(struct inode *inode, struct buffer_head *bh,
192 struct ext4_xattr_entry *entry, void *end, void *value_start,
193 const char *function, unsigned int line)
194{
195 struct ext4_xattr_entry *e = entry;
196 int err = -EFSCORRUPTED;
197 char *err_str;
198
199 if (bh) {
200 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
201 BHDR(bh)->h_blocks != cpu_to_le32(1)) {
202 err_str = "invalid header";
203 goto errout;
204 }
205 if (buffer_verified(bh))
206 return 0;
207 if (!ext4_xattr_block_csum_verify(inode, bh)) {
208 err = -EFSBADCRC;
209 err_str = "invalid checksum";
210 goto errout;
211 }
212 } else {
213 struct ext4_xattr_ibody_header *header = value_start;
214
215 header -= 1;
216 if (end - (void *)header < sizeof(*header) + sizeof(u32)) {
217 err_str = "in-inode xattr block too small";
218 goto errout;
219 }
220 if (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
221 err_str = "bad magic number in in-inode xattr";
222 goto errout;
223 }
224 }
225
226 /* Find the end of the names list */
227 while (!IS_LAST_ENTRY(e)) {
228 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
229 if ((void *)next >= end) {
230 err_str = "e_name out of bounds";
231 goto errout;
232 }
233 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len) {
234 err_str = "bad e_name length";
235 goto errout;
236 }
237 e = next;
238 }
239
240 /* Check the values */
241 while (!IS_LAST_ENTRY(entry)) {
242 u32 size = le32_to_cpu(entry->e_value_size);
243 unsigned long ea_ino = le32_to_cpu(entry->e_value_inum);
244
245 if (!ext4_has_feature_ea_inode(inode->i_sb) && ea_ino) {
246 err_str = "ea_inode specified without ea_inode feature enabled";
247 goto errout;
248 }
249 if (ea_ino && ((ea_ino == EXT4_ROOT_INO) ||
250 !ext4_valid_inum(inode->i_sb, ea_ino))) {
251 err_str = "invalid ea_ino";
252 goto errout;
253 }
254 if (size > EXT4_XATTR_SIZE_MAX) {
255 err_str = "e_value size too large";
256 goto errout;
257 }
258
259 if (size != 0 && entry->e_value_inum == 0) {
260 u16 offs = le16_to_cpu(entry->e_value_offs);
261 void *value;
262
263 /*
264 * The value cannot overlap the names, and the value
265 * with padding cannot extend beyond 'end'. Check both
266 * the padded and unpadded sizes, since the size may
267 * overflow to 0 when adding padding.
268 */
269 if (offs > end - value_start) {
270 err_str = "e_value out of bounds";
271 goto errout;
272 }
273 value = value_start + offs;
274 if (value < (void *)e + sizeof(u32) ||
275 size > end - value ||
276 EXT4_XATTR_SIZE(size) > end - value) {
277 err_str = "overlapping e_value ";
278 goto errout;
279 }
280 }
281 entry = EXT4_XATTR_NEXT(entry);
282 }
283 if (bh)
284 set_buffer_verified(bh);
285 return 0;
286
287errout:
288 if (bh)
289 __ext4_error_inode(inode, function, line, 0, -err,
290 "corrupted xattr block %llu: %s",
291 (unsigned long long) bh->b_blocknr,
292 err_str);
293 else
294 __ext4_error_inode(inode, function, line, 0, -err,
295 "corrupted in-inode xattr: %s", err_str);
296 return err;
297}
298
299static inline int
300__ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
301 const char *function, unsigned int line)
302{
303 return check_xattrs(inode, bh, BFIRST(bh), bh->b_data + bh->b_size,
304 bh->b_data, function, line);
305}
306
307#define ext4_xattr_check_block(inode, bh) \
308 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
309
310
311static inline int
312__xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
313 void *end, const char *function, unsigned int line)
314{
315 return check_xattrs(inode, NULL, IFIRST(header), end, IFIRST(header),
316 function, line);
317}
318
319#define xattr_check_inode(inode, header, end) \
320 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
321
322static int
323xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
324 void *end, int name_index, const char *name, int sorted)
325{
326 struct ext4_xattr_entry *entry, *next;
327 size_t name_len;
328 int cmp = 1;
329
330 if (name == NULL)
331 return -EINVAL;
332 name_len = strlen(name);
333 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
334 next = EXT4_XATTR_NEXT(entry);
335 if ((void *) next >= end) {
336 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
337 return -EFSCORRUPTED;
338 }
339 cmp = name_index - entry->e_name_index;
340 if (!cmp)
341 cmp = name_len - entry->e_name_len;
342 if (!cmp)
343 cmp = memcmp(name, entry->e_name, name_len);
344 if (cmp <= 0 && (sorted || cmp == 0))
345 break;
346 }
347 *pentry = entry;
348 return cmp ? -ENODATA : 0;
349}
350
351static u32
352ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
353{
354 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
355}
356
357static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
358{
359 return ((u64) inode_get_ctime_sec(ea_inode) << 32) |
360 (u32) inode_peek_iversion_raw(ea_inode);
361}
362
363static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
364{
365 inode_set_ctime(ea_inode, (u32)(ref_count >> 32), 0);
366 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
367}
368
369static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
370{
371 return (u32) inode_get_atime_sec(ea_inode);
372}
373
374static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
375{
376 inode_set_atime(ea_inode, hash, 0);
377}
378
379/*
380 * Read the EA value from an inode.
381 */
382static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
383{
384 int blocksize = 1 << ea_inode->i_blkbits;
385 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
386 int tail_size = (size % blocksize) ?: blocksize;
387 struct buffer_head *bhs_inline[8];
388 struct buffer_head **bhs = bhs_inline;
389 int i, ret;
390
391 if (bh_count > ARRAY_SIZE(bhs_inline)) {
392 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
393 if (!bhs)
394 return -ENOMEM;
395 }
396
397 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
398 true /* wait */, bhs);
399 if (ret)
400 goto free_bhs;
401
402 for (i = 0; i < bh_count; i++) {
403 /* There shouldn't be any holes in ea_inode. */
404 if (!bhs[i]) {
405 ret = -EFSCORRUPTED;
406 goto put_bhs;
407 }
408 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
409 i < bh_count - 1 ? blocksize : tail_size);
410 }
411 ret = 0;
412put_bhs:
413 for (i = 0; i < bh_count; i++)
414 brelse(bhs[i]);
415free_bhs:
416 if (bhs != bhs_inline)
417 kfree(bhs);
418 return ret;
419}
420
421#define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode_get_mtime_sec(inode)))
422
423static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
424 u32 ea_inode_hash, struct inode **ea_inode)
425{
426 struct inode *inode;
427 int err;
428
429 /*
430 * We have to check for this corruption early as otherwise
431 * iget_locked() could wait indefinitely for the state of our
432 * parent inode.
433 */
434 if (parent->i_ino == ea_ino) {
435 ext4_error(parent->i_sb,
436 "Parent and EA inode have the same ino %lu", ea_ino);
437 return -EFSCORRUPTED;
438 }
439
440 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_EA_INODE);
441 if (IS_ERR(inode)) {
442 err = PTR_ERR(inode);
443 ext4_error(parent->i_sb,
444 "error while reading EA inode %lu err=%d", ea_ino,
445 err);
446 return err;
447 }
448 ext4_xattr_inode_set_class(inode);
449
450 /*
451 * Check whether this is an old Lustre-style xattr inode. Lustre
452 * implementation does not have hash validation, rather it has a
453 * backpointer from ea_inode to the parent inode.
454 */
455 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
456 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
457 inode->i_generation == parent->i_generation) {
458 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
459 ext4_xattr_inode_set_ref(inode, 1);
460 } else {
461 inode_lock_nested(inode, I_MUTEX_XATTR);
462 inode->i_flags |= S_NOQUOTA;
463 inode_unlock(inode);
464 }
465
466 *ea_inode = inode;
467 return 0;
468}
469
470/* Remove entry from mbcache when EA inode is getting evicted */
471void ext4_evict_ea_inode(struct inode *inode)
472{
473 struct mb_cache_entry *oe;
474
475 if (!EA_INODE_CACHE(inode))
476 return;
477 /* Wait for entry to get unused so that we can remove it */
478 while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
479 ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
480 mb_cache_entry_wait_unused(oe);
481 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
482 }
483}
484
485static int
486ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
487 struct ext4_xattr_entry *entry, void *buffer,
488 size_t size)
489{
490 u32 hash;
491
492 /* Verify stored hash matches calculated hash. */
493 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
494 if (hash != ext4_xattr_inode_get_hash(ea_inode))
495 return -EFSCORRUPTED;
496
497 if (entry) {
498 __le32 e_hash, tmp_data;
499
500 /* Verify entry hash. */
501 tmp_data = cpu_to_le32(hash);
502 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
503 &tmp_data, 1);
504 /* All good? */
505 if (e_hash == entry->e_hash)
506 return 0;
507
508 /*
509 * Not good. Maybe the entry hash was calculated
510 * using the buggy signed char version?
511 */
512 e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
513 &tmp_data, 1);
514 /* Still no match - bad */
515 if (e_hash != entry->e_hash)
516 return -EFSCORRUPTED;
517
518 /* Let people know about old hash */
519 pr_warn_once("ext4: filesystem with signed xattr name hash");
520 }
521 return 0;
522}
523
524/*
525 * Read xattr value from the EA inode.
526 */
527static int
528ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
529 void *buffer, size_t size)
530{
531 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
532 struct inode *ea_inode;
533 int err;
534
535 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
536 le32_to_cpu(entry->e_hash), &ea_inode);
537 if (err) {
538 ea_inode = NULL;
539 goto out;
540 }
541
542 if (i_size_read(ea_inode) != size) {
543 ext4_warning_inode(ea_inode,
544 "ea_inode file size=%llu entry size=%zu",
545 i_size_read(ea_inode), size);
546 err = -EFSCORRUPTED;
547 goto out;
548 }
549
550 err = ext4_xattr_inode_read(ea_inode, buffer, size);
551 if (err)
552 goto out;
553
554 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
555 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
556 size);
557 if (err) {
558 ext4_warning_inode(ea_inode,
559 "EA inode hash validation failed");
560 goto out;
561 }
562
563 if (ea_inode_cache)
564 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
565 ext4_xattr_inode_get_hash(ea_inode),
566 ea_inode->i_ino, true /* reusable */);
567 }
568out:
569 iput(ea_inode);
570 return err;
571}
572
573static int
574ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
575 void *buffer, size_t buffer_size)
576{
577 struct buffer_head *bh = NULL;
578 struct ext4_xattr_entry *entry;
579 size_t size;
580 void *end;
581 int error;
582 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
583
584 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
585 name_index, name, buffer, (long)buffer_size);
586
587 if (!EXT4_I(inode)->i_file_acl)
588 return -ENODATA;
589 ea_idebug(inode, "reading block %llu",
590 (unsigned long long)EXT4_I(inode)->i_file_acl);
591 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
592 if (IS_ERR(bh))
593 return PTR_ERR(bh);
594 ea_bdebug(bh, "b_count=%d, refcount=%d",
595 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
596 error = ext4_xattr_check_block(inode, bh);
597 if (error)
598 goto cleanup;
599 ext4_xattr_block_cache_insert(ea_block_cache, bh);
600 entry = BFIRST(bh);
601 end = bh->b_data + bh->b_size;
602 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
603 if (error)
604 goto cleanup;
605 size = le32_to_cpu(entry->e_value_size);
606 error = -ERANGE;
607 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
608 goto cleanup;
609 if (buffer) {
610 if (size > buffer_size)
611 goto cleanup;
612 if (entry->e_value_inum) {
613 error = ext4_xattr_inode_get(inode, entry, buffer,
614 size);
615 if (error)
616 goto cleanup;
617 } else {
618 u16 offset = le16_to_cpu(entry->e_value_offs);
619 void *p = bh->b_data + offset;
620
621 if (unlikely(p + size > end))
622 goto cleanup;
623 memcpy(buffer, p, size);
624 }
625 }
626 error = size;
627
628cleanup:
629 brelse(bh);
630 return error;
631}
632
633int
634ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
635 void *buffer, size_t buffer_size)
636{
637 struct ext4_xattr_ibody_header *header;
638 struct ext4_xattr_entry *entry;
639 struct ext4_inode *raw_inode;
640 struct ext4_iloc iloc;
641 size_t size;
642 void *end;
643 int error;
644
645 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
646 return -ENODATA;
647 error = ext4_get_inode_loc(inode, &iloc);
648 if (error)
649 return error;
650 raw_inode = ext4_raw_inode(&iloc);
651 header = IHDR(inode, raw_inode);
652 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
653 error = xattr_check_inode(inode, header, end);
654 if (error)
655 goto cleanup;
656 entry = IFIRST(header);
657 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
658 if (error)
659 goto cleanup;
660 size = le32_to_cpu(entry->e_value_size);
661 error = -ERANGE;
662 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
663 goto cleanup;
664 if (buffer) {
665 if (size > buffer_size)
666 goto cleanup;
667 if (entry->e_value_inum) {
668 error = ext4_xattr_inode_get(inode, entry, buffer,
669 size);
670 if (error)
671 goto cleanup;
672 } else {
673 u16 offset = le16_to_cpu(entry->e_value_offs);
674 void *p = (void *)IFIRST(header) + offset;
675
676 if (unlikely(p + size > end))
677 goto cleanup;
678 memcpy(buffer, p, size);
679 }
680 }
681 error = size;
682
683cleanup:
684 brelse(iloc.bh);
685 return error;
686}
687
688/*
689 * ext4_xattr_get()
690 *
691 * Copy an extended attribute into the buffer
692 * provided, or compute the buffer size required.
693 * Buffer is NULL to compute the size of the buffer required.
694 *
695 * Returns a negative error number on failure, or the number of bytes
696 * used / required on success.
697 */
698int
699ext4_xattr_get(struct inode *inode, int name_index, const char *name,
700 void *buffer, size_t buffer_size)
701{
702 int error;
703
704 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
705 return -EIO;
706
707 if (strlen(name) > 255)
708 return -ERANGE;
709
710 down_read(&EXT4_I(inode)->xattr_sem);
711 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
712 buffer_size);
713 if (error == -ENODATA)
714 error = ext4_xattr_block_get(inode, name_index, name, buffer,
715 buffer_size);
716 up_read(&EXT4_I(inode)->xattr_sem);
717 return error;
718}
719
720static int
721ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
722 char *buffer, size_t buffer_size)
723{
724 size_t rest = buffer_size;
725
726 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
727 const char *prefix;
728
729 prefix = ext4_xattr_prefix(entry->e_name_index, dentry);
730 if (prefix) {
731 size_t prefix_len = strlen(prefix);
732 size_t size = prefix_len + entry->e_name_len + 1;
733
734 if (buffer) {
735 if (size > rest)
736 return -ERANGE;
737 memcpy(buffer, prefix, prefix_len);
738 buffer += prefix_len;
739 memcpy(buffer, entry->e_name, entry->e_name_len);
740 buffer += entry->e_name_len;
741 *buffer++ = 0;
742 }
743 rest -= size;
744 }
745 }
746 return buffer_size - rest; /* total size */
747}
748
749static int
750ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
751{
752 struct inode *inode = d_inode(dentry);
753 struct buffer_head *bh = NULL;
754 int error;
755
756 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
757 buffer, (long)buffer_size);
758
759 if (!EXT4_I(inode)->i_file_acl)
760 return 0;
761 ea_idebug(inode, "reading block %llu",
762 (unsigned long long)EXT4_I(inode)->i_file_acl);
763 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
764 if (IS_ERR(bh))
765 return PTR_ERR(bh);
766 ea_bdebug(bh, "b_count=%d, refcount=%d",
767 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
768 error = ext4_xattr_check_block(inode, bh);
769 if (error)
770 goto cleanup;
771 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
772 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
773 buffer_size);
774cleanup:
775 brelse(bh);
776 return error;
777}
778
779static int
780ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
781{
782 struct inode *inode = d_inode(dentry);
783 struct ext4_xattr_ibody_header *header;
784 struct ext4_inode *raw_inode;
785 struct ext4_iloc iloc;
786 void *end;
787 int error;
788
789 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
790 return 0;
791 error = ext4_get_inode_loc(inode, &iloc);
792 if (error)
793 return error;
794 raw_inode = ext4_raw_inode(&iloc);
795 header = IHDR(inode, raw_inode);
796 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
797 error = xattr_check_inode(inode, header, end);
798 if (error)
799 goto cleanup;
800 error = ext4_xattr_list_entries(dentry, IFIRST(header),
801 buffer, buffer_size);
802
803cleanup:
804 brelse(iloc.bh);
805 return error;
806}
807
808/*
809 * Inode operation listxattr()
810 *
811 * d_inode(dentry)->i_rwsem: don't care
812 *
813 * Copy a list of attribute names into the buffer
814 * provided, or compute the buffer size required.
815 * Buffer is NULL to compute the size of the buffer required.
816 *
817 * Returns a negative error number on failure, or the number of bytes
818 * used / required on success.
819 */
820ssize_t
821ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
822{
823 int ret, ret2;
824
825 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
826 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
827 if (ret < 0)
828 goto errout;
829 if (buffer) {
830 buffer += ret;
831 buffer_size -= ret;
832 }
833 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
834 if (ret < 0)
835 goto errout;
836 ret += ret2;
837errout:
838 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
839 return ret;
840}
841
842/*
843 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
844 * not set, set it.
845 */
846static void ext4_xattr_update_super_block(handle_t *handle,
847 struct super_block *sb)
848{
849 if (ext4_has_feature_xattr(sb))
850 return;
851
852 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
853 if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
854 EXT4_JTR_NONE) == 0) {
855 lock_buffer(EXT4_SB(sb)->s_sbh);
856 ext4_set_feature_xattr(sb);
857 ext4_superblock_csum_set(sb);
858 unlock_buffer(EXT4_SB(sb)->s_sbh);
859 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
860 }
861}
862
863int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
864{
865 struct ext4_iloc iloc = { .bh = NULL };
866 struct buffer_head *bh = NULL;
867 struct ext4_inode *raw_inode;
868 struct ext4_xattr_ibody_header *header;
869 struct ext4_xattr_entry *entry;
870 qsize_t ea_inode_refs = 0;
871 void *end;
872 int ret;
873
874 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
875
876 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
877 ret = ext4_get_inode_loc(inode, &iloc);
878 if (ret)
879 goto out;
880 raw_inode = ext4_raw_inode(&iloc);
881 header = IHDR(inode, raw_inode);
882 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
883 ret = xattr_check_inode(inode, header, end);
884 if (ret)
885 goto out;
886
887 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
888 entry = EXT4_XATTR_NEXT(entry))
889 if (entry->e_value_inum)
890 ea_inode_refs++;
891 }
892
893 if (EXT4_I(inode)->i_file_acl) {
894 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
895 if (IS_ERR(bh)) {
896 ret = PTR_ERR(bh);
897 bh = NULL;
898 goto out;
899 }
900
901 ret = ext4_xattr_check_block(inode, bh);
902 if (ret)
903 goto out;
904
905 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
906 entry = EXT4_XATTR_NEXT(entry))
907 if (entry->e_value_inum)
908 ea_inode_refs++;
909 }
910 *usage = ea_inode_refs + 1;
911 ret = 0;
912out:
913 brelse(iloc.bh);
914 brelse(bh);
915 return ret;
916}
917
918static inline size_t round_up_cluster(struct inode *inode, size_t length)
919{
920 struct super_block *sb = inode->i_sb;
921 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
922 inode->i_blkbits);
923 size_t mask = ~(cluster_size - 1);
924
925 return (length + cluster_size - 1) & mask;
926}
927
928static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
929{
930 int err;
931
932 err = dquot_alloc_inode(inode);
933 if (err)
934 return err;
935 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
936 if (err)
937 dquot_free_inode(inode);
938 return err;
939}
940
941static void ext4_xattr_inode_free_quota(struct inode *parent,
942 struct inode *ea_inode,
943 size_t len)
944{
945 if (ea_inode &&
946 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
947 return;
948 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
949 dquot_free_inode(parent);
950}
951
952int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
953 struct buffer_head *block_bh, size_t value_len,
954 bool is_create)
955{
956 int credits;
957 int blocks;
958
959 /*
960 * 1) Owner inode update
961 * 2) Ref count update on old xattr block
962 * 3) new xattr block
963 * 4) block bitmap update for new xattr block
964 * 5) group descriptor for new xattr block
965 * 6) block bitmap update for old xattr block
966 * 7) group descriptor for old block
967 *
968 * 6 & 7 can happen if we have two racing threads T_a and T_b
969 * which are each trying to set an xattr on inodes I_a and I_b
970 * which were both initially sharing an xattr block.
971 */
972 credits = 7;
973
974 /* Quota updates. */
975 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
976
977 /*
978 * In case of inline data, we may push out the data to a block,
979 * so we need to reserve credits for this eventuality
980 */
981 if (inode && ext4_has_inline_data(inode))
982 credits += ext4_writepage_trans_blocks(inode) + 1;
983
984 /* We are done if ea_inode feature is not enabled. */
985 if (!ext4_has_feature_ea_inode(sb))
986 return credits;
987
988 /* New ea_inode, inode map, block bitmap, group descriptor. */
989 credits += 4;
990
991 /* Data blocks. */
992 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
993
994 /* Indirection block or one level of extent tree. */
995 blocks += 1;
996
997 /* Block bitmap and group descriptor updates for each block. */
998 credits += blocks * 2;
999
1000 /* Blocks themselves. */
1001 credits += blocks;
1002
1003 if (!is_create) {
1004 /* Dereference ea_inode holding old xattr value.
1005 * Old ea_inode, inode map, block bitmap, group descriptor.
1006 */
1007 credits += 4;
1008
1009 /* Data blocks for old ea_inode. */
1010 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
1011
1012 /* Indirection block or one level of extent tree for old
1013 * ea_inode.
1014 */
1015 blocks += 1;
1016
1017 /* Block bitmap and group descriptor updates for each block. */
1018 credits += blocks * 2;
1019 }
1020
1021 /* We may need to clone the existing xattr block in which case we need
1022 * to increment ref counts for existing ea_inodes referenced by it.
1023 */
1024 if (block_bh) {
1025 struct ext4_xattr_entry *entry = BFIRST(block_bh);
1026
1027 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1028 if (entry->e_value_inum)
1029 /* Ref count update on ea_inode. */
1030 credits += 1;
1031 }
1032 return credits;
1033}
1034
1035static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1036 int ref_change)
1037{
1038 struct ext4_iloc iloc;
1039 s64 ref_count;
1040 int ret;
1041
1042 inode_lock_nested(ea_inode, I_MUTEX_XATTR);
1043
1044 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1045 if (ret)
1046 goto out;
1047
1048 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1049 ref_count += ref_change;
1050 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1051
1052 if (ref_change > 0) {
1053 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1054 ea_inode->i_ino, ref_count);
1055
1056 if (ref_count == 1) {
1057 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1058 ea_inode->i_ino, ea_inode->i_nlink);
1059
1060 set_nlink(ea_inode, 1);
1061 ext4_orphan_del(handle, ea_inode);
1062 }
1063 } else {
1064 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1065 ea_inode->i_ino, ref_count);
1066
1067 if (ref_count == 0) {
1068 WARN_ONCE(ea_inode->i_nlink != 1,
1069 "EA inode %lu i_nlink=%u",
1070 ea_inode->i_ino, ea_inode->i_nlink);
1071
1072 clear_nlink(ea_inode);
1073 ext4_orphan_add(handle, ea_inode);
1074 }
1075 }
1076
1077 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1078 if (ret)
1079 ext4_warning_inode(ea_inode,
1080 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1081out:
1082 inode_unlock(ea_inode);
1083 return ret;
1084}
1085
1086static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1087{
1088 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1089}
1090
1091static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1092{
1093 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1094}
1095
1096static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1097 struct ext4_xattr_entry *first)
1098{
1099 struct inode *ea_inode;
1100 struct ext4_xattr_entry *entry;
1101 struct ext4_xattr_entry *failed_entry;
1102 unsigned int ea_ino;
1103 int err, saved_err;
1104
1105 for (entry = first; !IS_LAST_ENTRY(entry);
1106 entry = EXT4_XATTR_NEXT(entry)) {
1107 if (!entry->e_value_inum)
1108 continue;
1109 ea_ino = le32_to_cpu(entry->e_value_inum);
1110 err = ext4_xattr_inode_iget(parent, ea_ino,
1111 le32_to_cpu(entry->e_hash),
1112 &ea_inode);
1113 if (err)
1114 goto cleanup;
1115 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1116 if (err) {
1117 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1118 iput(ea_inode);
1119 goto cleanup;
1120 }
1121 iput(ea_inode);
1122 }
1123 return 0;
1124
1125cleanup:
1126 saved_err = err;
1127 failed_entry = entry;
1128
1129 for (entry = first; entry != failed_entry;
1130 entry = EXT4_XATTR_NEXT(entry)) {
1131 if (!entry->e_value_inum)
1132 continue;
1133 ea_ino = le32_to_cpu(entry->e_value_inum);
1134 err = ext4_xattr_inode_iget(parent, ea_ino,
1135 le32_to_cpu(entry->e_hash),
1136 &ea_inode);
1137 if (err) {
1138 ext4_warning(parent->i_sb,
1139 "cleanup ea_ino %u iget error %d", ea_ino,
1140 err);
1141 continue;
1142 }
1143 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1144 if (err)
1145 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1146 err);
1147 iput(ea_inode);
1148 }
1149 return saved_err;
1150}
1151
1152static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1153 struct buffer_head *bh, bool block_csum, bool dirty)
1154{
1155 int error;
1156
1157 if (bh && dirty) {
1158 if (block_csum)
1159 ext4_xattr_block_csum_set(inode, bh);
1160 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1161 if (error) {
1162 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1163 error);
1164 return error;
1165 }
1166 }
1167 return 0;
1168}
1169
1170static void
1171ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1172 struct buffer_head *bh,
1173 struct ext4_xattr_entry *first, bool block_csum,
1174 struct ext4_xattr_inode_array **ea_inode_array,
1175 int extra_credits, bool skip_quota)
1176{
1177 struct inode *ea_inode;
1178 struct ext4_xattr_entry *entry;
1179 bool dirty = false;
1180 unsigned int ea_ino;
1181 int err;
1182 int credits;
1183
1184 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1185 credits = 2 + extra_credits;
1186
1187 for (entry = first; !IS_LAST_ENTRY(entry);
1188 entry = EXT4_XATTR_NEXT(entry)) {
1189 if (!entry->e_value_inum)
1190 continue;
1191 ea_ino = le32_to_cpu(entry->e_value_inum);
1192 err = ext4_xattr_inode_iget(parent, ea_ino,
1193 le32_to_cpu(entry->e_hash),
1194 &ea_inode);
1195 if (err)
1196 continue;
1197
1198 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1199 if (err) {
1200 ext4_warning_inode(ea_inode,
1201 "Expand inode array err=%d", err);
1202 iput(ea_inode);
1203 continue;
1204 }
1205
1206 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1207 ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1208 ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1209 dirty));
1210 if (err < 0) {
1211 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1212 err);
1213 continue;
1214 }
1215 if (err > 0) {
1216 err = ext4_journal_get_write_access(handle,
1217 parent->i_sb, bh, EXT4_JTR_NONE);
1218 if (err) {
1219 ext4_warning_inode(ea_inode,
1220 "Re-get write access err=%d",
1221 err);
1222 continue;
1223 }
1224 }
1225
1226 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1227 if (err) {
1228 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1229 err);
1230 continue;
1231 }
1232
1233 if (!skip_quota)
1234 ext4_xattr_inode_free_quota(parent, ea_inode,
1235 le32_to_cpu(entry->e_value_size));
1236
1237 /*
1238 * Forget about ea_inode within the same transaction that
1239 * decrements the ref count. This avoids duplicate decrements in
1240 * case the rest of the work spills over to subsequent
1241 * transactions.
1242 */
1243 entry->e_value_inum = 0;
1244 entry->e_value_size = 0;
1245
1246 dirty = true;
1247 }
1248
1249 if (dirty) {
1250 /*
1251 * Note that we are deliberately skipping csum calculation for
1252 * the final update because we do not expect any journal
1253 * restarts until xattr block is freed.
1254 */
1255
1256 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1257 if (err)
1258 ext4_warning_inode(parent,
1259 "handle dirty metadata err=%d", err);
1260 }
1261}
1262
1263/*
1264 * Release the xattr block BH: If the reference count is > 1, decrement it;
1265 * otherwise free the block.
1266 */
1267static void
1268ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1269 struct buffer_head *bh,
1270 struct ext4_xattr_inode_array **ea_inode_array,
1271 int extra_credits)
1272{
1273 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1274 u32 hash, ref;
1275 int error = 0;
1276
1277 BUFFER_TRACE(bh, "get_write_access");
1278 error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1279 EXT4_JTR_NONE);
1280 if (error)
1281 goto out;
1282
1283retry_ref:
1284 lock_buffer(bh);
1285 hash = le32_to_cpu(BHDR(bh)->h_hash);
1286 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1287 if (ref == 1) {
1288 ea_bdebug(bh, "refcount now=0; freeing");
1289 /*
1290 * This must happen under buffer lock for
1291 * ext4_xattr_block_set() to reliably detect freed block
1292 */
1293 if (ea_block_cache) {
1294 struct mb_cache_entry *oe;
1295
1296 oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1297 bh->b_blocknr);
1298 if (oe) {
1299 unlock_buffer(bh);
1300 mb_cache_entry_wait_unused(oe);
1301 mb_cache_entry_put(ea_block_cache, oe);
1302 goto retry_ref;
1303 }
1304 }
1305 get_bh(bh);
1306 unlock_buffer(bh);
1307
1308 if (ext4_has_feature_ea_inode(inode->i_sb))
1309 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1310 BFIRST(bh),
1311 true /* block_csum */,
1312 ea_inode_array,
1313 extra_credits,
1314 true /* skip_quota */);
1315 ext4_free_blocks(handle, inode, bh, 0, 1,
1316 EXT4_FREE_BLOCKS_METADATA |
1317 EXT4_FREE_BLOCKS_FORGET);
1318 } else {
1319 ref--;
1320 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1321 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1322 struct mb_cache_entry *ce;
1323
1324 if (ea_block_cache) {
1325 ce = mb_cache_entry_get(ea_block_cache, hash,
1326 bh->b_blocknr);
1327 if (ce) {
1328 set_bit(MBE_REUSABLE_B, &ce->e_flags);
1329 mb_cache_entry_put(ea_block_cache, ce);
1330 }
1331 }
1332 }
1333
1334 ext4_xattr_block_csum_set(inode, bh);
1335 /*
1336 * Beware of this ugliness: Releasing of xattr block references
1337 * from different inodes can race and so we have to protect
1338 * from a race where someone else frees the block (and releases
1339 * its journal_head) before we are done dirtying the buffer. In
1340 * nojournal mode this race is harmless and we actually cannot
1341 * call ext4_handle_dirty_metadata() with locked buffer as
1342 * that function can call sync_dirty_buffer() so for that case
1343 * we handle the dirtying after unlocking the buffer.
1344 */
1345 if (ext4_handle_valid(handle))
1346 error = ext4_handle_dirty_metadata(handle, inode, bh);
1347 unlock_buffer(bh);
1348 if (!ext4_handle_valid(handle))
1349 error = ext4_handle_dirty_metadata(handle, inode, bh);
1350 if (IS_SYNC(inode))
1351 ext4_handle_sync(handle);
1352 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1353 ea_bdebug(bh, "refcount now=%d; releasing",
1354 le32_to_cpu(BHDR(bh)->h_refcount));
1355 }
1356out:
1357 ext4_std_error(inode->i_sb, error);
1358 return;
1359}
1360
1361/*
1362 * Find the available free space for EAs. This also returns the total number of
1363 * bytes used by EA entries.
1364 */
1365static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1366 size_t *min_offs, void *base, int *total)
1367{
1368 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1369 if (!last->e_value_inum && last->e_value_size) {
1370 size_t offs = le16_to_cpu(last->e_value_offs);
1371 if (offs < *min_offs)
1372 *min_offs = offs;
1373 }
1374 if (total)
1375 *total += EXT4_XATTR_LEN(last->e_name_len);
1376 }
1377 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1378}
1379
1380/*
1381 * Write the value of the EA in an inode.
1382 */
1383static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1384 const void *buf, int bufsize)
1385{
1386 struct buffer_head *bh = NULL;
1387 unsigned long block = 0;
1388 int blocksize = ea_inode->i_sb->s_blocksize;
1389 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1390 int csize, wsize = 0;
1391 int ret = 0, ret2 = 0;
1392 int retries = 0;
1393
1394retry:
1395 while (ret >= 0 && ret < max_blocks) {
1396 struct ext4_map_blocks map;
1397 map.m_lblk = block += ret;
1398 map.m_len = max_blocks -= ret;
1399
1400 ret = ext4_map_blocks(handle, ea_inode, &map,
1401 EXT4_GET_BLOCKS_CREATE);
1402 if (ret <= 0) {
1403 ext4_mark_inode_dirty(handle, ea_inode);
1404 if (ret == -ENOSPC &&
1405 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1406 ret = 0;
1407 goto retry;
1408 }
1409 break;
1410 }
1411 }
1412
1413 if (ret < 0)
1414 return ret;
1415
1416 block = 0;
1417 while (wsize < bufsize) {
1418 brelse(bh);
1419 csize = (bufsize - wsize) > blocksize ? blocksize :
1420 bufsize - wsize;
1421 bh = ext4_getblk(handle, ea_inode, block, 0);
1422 if (IS_ERR(bh))
1423 return PTR_ERR(bh);
1424 if (!bh) {
1425 WARN_ON_ONCE(1);
1426 EXT4_ERROR_INODE(ea_inode,
1427 "ext4_getblk() return bh = NULL");
1428 return -EFSCORRUPTED;
1429 }
1430 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1431 EXT4_JTR_NONE);
1432 if (ret)
1433 goto out;
1434
1435 memcpy(bh->b_data, buf, csize);
1436 /*
1437 * Zero out block tail to avoid writing uninitialized memory
1438 * to disk.
1439 */
1440 if (csize < blocksize)
1441 memset(bh->b_data + csize, 0, blocksize - csize);
1442 set_buffer_uptodate(bh);
1443 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1444
1445 buf += csize;
1446 wsize += csize;
1447 block += 1;
1448 }
1449
1450 inode_lock(ea_inode);
1451 i_size_write(ea_inode, wsize);
1452 ext4_update_i_disksize(ea_inode, wsize);
1453 inode_unlock(ea_inode);
1454
1455 ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1456 if (unlikely(ret2 && !ret))
1457 ret = ret2;
1458
1459out:
1460 brelse(bh);
1461
1462 return ret;
1463}
1464
1465/*
1466 * Create an inode to store the value of a large EA.
1467 */
1468static struct inode *ext4_xattr_inode_create(handle_t *handle,
1469 struct inode *inode, u32 hash)
1470{
1471 struct inode *ea_inode = NULL;
1472 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1473 int err;
1474
1475 if (inode->i_sb->s_root == NULL) {
1476 ext4_warning(inode->i_sb,
1477 "refuse to create EA inode when umounting");
1478 WARN_ON(1);
1479 return ERR_PTR(-EINVAL);
1480 }
1481
1482 /*
1483 * Let the next inode be the goal, so we try and allocate the EA inode
1484 * in the same group, or nearby one.
1485 */
1486 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1487 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1488 EXT4_EA_INODE_FL);
1489 if (!IS_ERR(ea_inode)) {
1490 ea_inode->i_op = &ext4_file_inode_operations;
1491 ea_inode->i_fop = &ext4_file_operations;
1492 ext4_set_aops(ea_inode);
1493 ext4_xattr_inode_set_class(ea_inode);
1494 unlock_new_inode(ea_inode);
1495 ext4_xattr_inode_set_ref(ea_inode, 1);
1496 ext4_xattr_inode_set_hash(ea_inode, hash);
1497 err = ext4_mark_inode_dirty(handle, ea_inode);
1498 if (!err)
1499 err = ext4_inode_attach_jinode(ea_inode);
1500 if (err) {
1501 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1502 ext4_warning_inode(ea_inode,
1503 "cleanup dec ref error %d", err);
1504 iput(ea_inode);
1505 return ERR_PTR(err);
1506 }
1507
1508 /*
1509 * Xattr inodes are shared therefore quota charging is performed
1510 * at a higher level.
1511 */
1512 dquot_free_inode(ea_inode);
1513 dquot_drop(ea_inode);
1514 inode_lock(ea_inode);
1515 ea_inode->i_flags |= S_NOQUOTA;
1516 inode_unlock(ea_inode);
1517 }
1518
1519 return ea_inode;
1520}
1521
1522static struct inode *
1523ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1524 size_t value_len, u32 hash)
1525{
1526 struct inode *ea_inode;
1527 struct mb_cache_entry *ce;
1528 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1529 void *ea_data;
1530
1531 if (!ea_inode_cache)
1532 return NULL;
1533
1534 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1535 if (!ce)
1536 return NULL;
1537
1538 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1539 !(current->flags & PF_MEMALLOC_NOFS));
1540
1541 ea_data = kvmalloc(value_len, GFP_KERNEL);
1542 if (!ea_data) {
1543 mb_cache_entry_put(ea_inode_cache, ce);
1544 return NULL;
1545 }
1546
1547 while (ce) {
1548 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1549 EXT4_IGET_EA_INODE);
1550 if (IS_ERR(ea_inode))
1551 goto next_entry;
1552 ext4_xattr_inode_set_class(ea_inode);
1553 if (i_size_read(ea_inode) == value_len &&
1554 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1555 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1556 value_len) &&
1557 !memcmp(value, ea_data, value_len)) {
1558 mb_cache_entry_touch(ea_inode_cache, ce);
1559 mb_cache_entry_put(ea_inode_cache, ce);
1560 kvfree(ea_data);
1561 return ea_inode;
1562 }
1563 iput(ea_inode);
1564 next_entry:
1565 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1566 }
1567 kvfree(ea_data);
1568 return NULL;
1569}
1570
1571/*
1572 * Add value of the EA in an inode.
1573 */
1574static struct inode *ext4_xattr_inode_lookup_create(handle_t *handle,
1575 struct inode *inode, const void *value, size_t value_len)
1576{
1577 struct inode *ea_inode;
1578 u32 hash;
1579 int err;
1580
1581 /* Account inode & space to quota even if sharing... */
1582 err = ext4_xattr_inode_alloc_quota(inode, value_len);
1583 if (err)
1584 return ERR_PTR(err);
1585
1586 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1587 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1588 if (ea_inode) {
1589 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1590 if (err)
1591 goto out_err;
1592 return ea_inode;
1593 }
1594
1595 /* Create an inode for the EA value */
1596 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1597 if (IS_ERR(ea_inode)) {
1598 ext4_xattr_inode_free_quota(inode, NULL, value_len);
1599 return ea_inode;
1600 }
1601
1602 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1603 if (err) {
1604 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1605 ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1606 goto out_err;
1607 }
1608
1609 if (EA_INODE_CACHE(inode))
1610 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1611 ea_inode->i_ino, true /* reusable */);
1612 return ea_inode;
1613out_err:
1614 iput(ea_inode);
1615 ext4_xattr_inode_free_quota(inode, NULL, value_len);
1616 return ERR_PTR(err);
1617}
1618
1619/*
1620 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1621 * feature is enabled.
1622 */
1623#define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1624
1625static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1626 struct ext4_xattr_search *s,
1627 handle_t *handle, struct inode *inode,
1628 struct inode *new_ea_inode,
1629 bool is_block)
1630{
1631 struct ext4_xattr_entry *last, *next;
1632 struct ext4_xattr_entry *here = s->here;
1633 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1634 int in_inode = i->in_inode;
1635 struct inode *old_ea_inode = NULL;
1636 size_t old_size, new_size;
1637 int ret;
1638
1639 /* Space used by old and new values. */
1640 old_size = (!s->not_found && !here->e_value_inum) ?
1641 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1642 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1643
1644 /*
1645 * Optimization for the simple case when old and new values have the
1646 * same padded sizes. Not applicable if external inodes are involved.
1647 */
1648 if (new_size && new_size == old_size) {
1649 size_t offs = le16_to_cpu(here->e_value_offs);
1650 void *val = s->base + offs;
1651
1652 here->e_value_size = cpu_to_le32(i->value_len);
1653 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1654 memset(val, 0, new_size);
1655 } else {
1656 memcpy(val, i->value, i->value_len);
1657 /* Clear padding bytes. */
1658 memset(val + i->value_len, 0, new_size - i->value_len);
1659 }
1660 goto update_hash;
1661 }
1662
1663 /* Compute min_offs and last. */
1664 last = s->first;
1665 for (; !IS_LAST_ENTRY(last); last = next) {
1666 next = EXT4_XATTR_NEXT(last);
1667 if ((void *)next >= s->end) {
1668 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1669 ret = -EFSCORRUPTED;
1670 goto out;
1671 }
1672 if (!last->e_value_inum && last->e_value_size) {
1673 size_t offs = le16_to_cpu(last->e_value_offs);
1674 if (offs < min_offs)
1675 min_offs = offs;
1676 }
1677 }
1678
1679 /* Check whether we have enough space. */
1680 if (i->value) {
1681 size_t free;
1682
1683 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1684 if (!s->not_found)
1685 free += EXT4_XATTR_LEN(name_len) + old_size;
1686
1687 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1688 ret = -ENOSPC;
1689 goto out;
1690 }
1691
1692 /*
1693 * If storing the value in an external inode is an option,
1694 * reserve space for xattr entries/names in the external
1695 * attribute block so that a long value does not occupy the
1696 * whole space and prevent further entries being added.
1697 */
1698 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1699 new_size && is_block &&
1700 (min_offs + old_size - new_size) <
1701 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1702 ret = -ENOSPC;
1703 goto out;
1704 }
1705 }
1706
1707 /*
1708 * Getting access to old and new ea inodes is subject to failures.
1709 * Finish that work before doing any modifications to the xattr data.
1710 */
1711 if (!s->not_found && here->e_value_inum) {
1712 ret = ext4_xattr_inode_iget(inode,
1713 le32_to_cpu(here->e_value_inum),
1714 le32_to_cpu(here->e_hash),
1715 &old_ea_inode);
1716 if (ret) {
1717 old_ea_inode = NULL;
1718 goto out;
1719 }
1720
1721 /* We are ready to release ref count on the old_ea_inode. */
1722 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1723 if (ret)
1724 goto out;
1725
1726 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1727 le32_to_cpu(here->e_value_size));
1728 }
1729
1730 /* No failures allowed past this point. */
1731
1732 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1733 /* Remove the old value. */
1734 void *first_val = s->base + min_offs;
1735 size_t offs = le16_to_cpu(here->e_value_offs);
1736 void *val = s->base + offs;
1737
1738 memmove(first_val + old_size, first_val, val - first_val);
1739 memset(first_val, 0, old_size);
1740 min_offs += old_size;
1741
1742 /* Adjust all value offsets. */
1743 last = s->first;
1744 while (!IS_LAST_ENTRY(last)) {
1745 size_t o = le16_to_cpu(last->e_value_offs);
1746
1747 if (!last->e_value_inum &&
1748 last->e_value_size && o < offs)
1749 last->e_value_offs = cpu_to_le16(o + old_size);
1750 last = EXT4_XATTR_NEXT(last);
1751 }
1752 }
1753
1754 if (!i->value) {
1755 /* Remove old name. */
1756 size_t size = EXT4_XATTR_LEN(name_len);
1757
1758 last = ENTRY((void *)last - size);
1759 memmove(here, (void *)here + size,
1760 (void *)last - (void *)here + sizeof(__u32));
1761 memset(last, 0, size);
1762
1763 /*
1764 * Update i_inline_off - moved ibody region might contain
1765 * system.data attribute. Handling a failure here won't
1766 * cause other complications for setting an xattr.
1767 */
1768 if (!is_block && ext4_has_inline_data(inode)) {
1769 ret = ext4_find_inline_data_nolock(inode);
1770 if (ret) {
1771 ext4_warning_inode(inode,
1772 "unable to update i_inline_off");
1773 goto out;
1774 }
1775 }
1776 } else if (s->not_found) {
1777 /* Insert new name. */
1778 size_t size = EXT4_XATTR_LEN(name_len);
1779 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1780
1781 memmove((void *)here + size, here, rest);
1782 memset(here, 0, size);
1783 here->e_name_index = i->name_index;
1784 here->e_name_len = name_len;
1785 memcpy(here->e_name, i->name, name_len);
1786 } else {
1787 /* This is an update, reset value info. */
1788 here->e_value_inum = 0;
1789 here->e_value_offs = 0;
1790 here->e_value_size = 0;
1791 }
1792
1793 if (i->value) {
1794 /* Insert new value. */
1795 if (in_inode) {
1796 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1797 } else if (i->value_len) {
1798 void *val = s->base + min_offs - new_size;
1799
1800 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1801 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1802 memset(val, 0, new_size);
1803 } else {
1804 memcpy(val, i->value, i->value_len);
1805 /* Clear padding bytes. */
1806 memset(val + i->value_len, 0,
1807 new_size - i->value_len);
1808 }
1809 }
1810 here->e_value_size = cpu_to_le32(i->value_len);
1811 }
1812
1813update_hash:
1814 if (i->value) {
1815 __le32 hash = 0;
1816
1817 /* Entry hash calculation. */
1818 if (in_inode) {
1819 __le32 crc32c_hash;
1820
1821 /*
1822 * Feed crc32c hash instead of the raw value for entry
1823 * hash calculation. This is to avoid walking
1824 * potentially long value buffer again.
1825 */
1826 crc32c_hash = cpu_to_le32(
1827 ext4_xattr_inode_get_hash(new_ea_inode));
1828 hash = ext4_xattr_hash_entry(here->e_name,
1829 here->e_name_len,
1830 &crc32c_hash, 1);
1831 } else if (is_block) {
1832 __le32 *value = s->base + le16_to_cpu(
1833 here->e_value_offs);
1834
1835 hash = ext4_xattr_hash_entry(here->e_name,
1836 here->e_name_len, value,
1837 new_size >> 2);
1838 }
1839 here->e_hash = hash;
1840 }
1841
1842 if (is_block)
1843 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1844
1845 ret = 0;
1846out:
1847 iput(old_ea_inode);
1848 return ret;
1849}
1850
1851struct ext4_xattr_block_find {
1852 struct ext4_xattr_search s;
1853 struct buffer_head *bh;
1854};
1855
1856static int
1857ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1858 struct ext4_xattr_block_find *bs)
1859{
1860 struct super_block *sb = inode->i_sb;
1861 int error;
1862
1863 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1864 i->name_index, i->name, i->value, (long)i->value_len);
1865
1866 if (EXT4_I(inode)->i_file_acl) {
1867 /* The inode already has an extended attribute block. */
1868 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1869 if (IS_ERR(bs->bh)) {
1870 error = PTR_ERR(bs->bh);
1871 bs->bh = NULL;
1872 return error;
1873 }
1874 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1875 atomic_read(&(bs->bh->b_count)),
1876 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1877 error = ext4_xattr_check_block(inode, bs->bh);
1878 if (error)
1879 return error;
1880 /* Find the named attribute. */
1881 bs->s.base = BHDR(bs->bh);
1882 bs->s.first = BFIRST(bs->bh);
1883 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1884 bs->s.here = bs->s.first;
1885 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1886 i->name_index, i->name, 1);
1887 if (error && error != -ENODATA)
1888 return error;
1889 bs->s.not_found = error;
1890 }
1891 return 0;
1892}
1893
1894static int
1895ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1896 struct ext4_xattr_info *i,
1897 struct ext4_xattr_block_find *bs)
1898{
1899 struct super_block *sb = inode->i_sb;
1900 struct buffer_head *new_bh = NULL;
1901 struct ext4_xattr_search s_copy = bs->s;
1902 struct ext4_xattr_search *s = &s_copy;
1903 struct mb_cache_entry *ce = NULL;
1904 int error = 0;
1905 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1906 struct inode *ea_inode = NULL, *tmp_inode;
1907 size_t old_ea_inode_quota = 0;
1908 unsigned int ea_ino;
1909
1910#define header(x) ((struct ext4_xattr_header *)(x))
1911
1912 /* If we need EA inode, prepare it before locking the buffer */
1913 if (i->value && i->in_inode) {
1914 WARN_ON_ONCE(!i->value_len);
1915
1916 ea_inode = ext4_xattr_inode_lookup_create(handle, inode,
1917 i->value, i->value_len);
1918 if (IS_ERR(ea_inode)) {
1919 error = PTR_ERR(ea_inode);
1920 ea_inode = NULL;
1921 goto cleanup;
1922 }
1923 }
1924
1925 if (s->base) {
1926 int offset = (char *)s->here - bs->bh->b_data;
1927
1928 BUFFER_TRACE(bs->bh, "get_write_access");
1929 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1930 EXT4_JTR_NONE);
1931 if (error)
1932 goto cleanup;
1933
1934 lock_buffer(bs->bh);
1935
1936 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1937 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1938
1939 /*
1940 * This must happen under buffer lock for
1941 * ext4_xattr_block_set() to reliably detect modified
1942 * block
1943 */
1944 if (ea_block_cache) {
1945 struct mb_cache_entry *oe;
1946
1947 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1948 hash, bs->bh->b_blocknr);
1949 if (oe) {
1950 /*
1951 * Xattr block is getting reused. Leave
1952 * it alone.
1953 */
1954 mb_cache_entry_put(ea_block_cache, oe);
1955 goto clone_block;
1956 }
1957 }
1958 ea_bdebug(bs->bh, "modifying in-place");
1959 error = ext4_xattr_set_entry(i, s, handle, inode,
1960 ea_inode, true /* is_block */);
1961 ext4_xattr_block_csum_set(inode, bs->bh);
1962 unlock_buffer(bs->bh);
1963 if (error == -EFSCORRUPTED)
1964 goto bad_block;
1965 if (!error)
1966 error = ext4_handle_dirty_metadata(handle,
1967 inode,
1968 bs->bh);
1969 if (error)
1970 goto cleanup;
1971 goto inserted;
1972 }
1973clone_block:
1974 unlock_buffer(bs->bh);
1975 ea_bdebug(bs->bh, "cloning");
1976 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1977 error = -ENOMEM;
1978 if (s->base == NULL)
1979 goto cleanup;
1980 s->first = ENTRY(header(s->base)+1);
1981 header(s->base)->h_refcount = cpu_to_le32(1);
1982 s->here = ENTRY(s->base + offset);
1983 s->end = s->base + bs->bh->b_size;
1984
1985 /*
1986 * If existing entry points to an xattr inode, we need
1987 * to prevent ext4_xattr_set_entry() from decrementing
1988 * ref count on it because the reference belongs to the
1989 * original block. In this case, make the entry look
1990 * like it has an empty value.
1991 */
1992 if (!s->not_found && s->here->e_value_inum) {
1993 ea_ino = le32_to_cpu(s->here->e_value_inum);
1994 error = ext4_xattr_inode_iget(inode, ea_ino,
1995 le32_to_cpu(s->here->e_hash),
1996 &tmp_inode);
1997 if (error)
1998 goto cleanup;
1999
2000 if (!ext4_test_inode_state(tmp_inode,
2001 EXT4_STATE_LUSTRE_EA_INODE)) {
2002 /*
2003 * Defer quota free call for previous
2004 * inode until success is guaranteed.
2005 */
2006 old_ea_inode_quota = le32_to_cpu(
2007 s->here->e_value_size);
2008 }
2009 iput(tmp_inode);
2010
2011 s->here->e_value_inum = 0;
2012 s->here->e_value_size = 0;
2013 }
2014 } else {
2015 /* Allocate a buffer where we construct the new block. */
2016 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
2017 error = -ENOMEM;
2018 if (s->base == NULL)
2019 goto cleanup;
2020 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2021 header(s->base)->h_blocks = cpu_to_le32(1);
2022 header(s->base)->h_refcount = cpu_to_le32(1);
2023 s->first = ENTRY(header(s->base)+1);
2024 s->here = ENTRY(header(s->base)+1);
2025 s->end = s->base + sb->s_blocksize;
2026 }
2027
2028 error = ext4_xattr_set_entry(i, s, handle, inode, ea_inode,
2029 true /* is_block */);
2030 if (error == -EFSCORRUPTED)
2031 goto bad_block;
2032 if (error)
2033 goto cleanup;
2034
2035inserted:
2036 if (!IS_LAST_ENTRY(s->first)) {
2037 new_bh = ext4_xattr_block_cache_find(inode, header(s->base), &ce);
2038 if (IS_ERR(new_bh)) {
2039 error = PTR_ERR(new_bh);
2040 new_bh = NULL;
2041 goto cleanup;
2042 }
2043
2044 if (new_bh) {
2045 /* We found an identical block in the cache. */
2046 if (new_bh == bs->bh)
2047 ea_bdebug(new_bh, "keeping");
2048 else {
2049 u32 ref;
2050
2051#ifdef EXT4_XATTR_DEBUG
2052 WARN_ON_ONCE(dquot_initialize_needed(inode));
2053#endif
2054 /* The old block is released after updating
2055 the inode. */
2056 error = dquot_alloc_block(inode,
2057 EXT4_C2B(EXT4_SB(sb), 1));
2058 if (error)
2059 goto cleanup;
2060 BUFFER_TRACE(new_bh, "get_write_access");
2061 error = ext4_journal_get_write_access(
2062 handle, sb, new_bh,
2063 EXT4_JTR_NONE);
2064 if (error)
2065 goto cleanup_dquot;
2066 lock_buffer(new_bh);
2067 /*
2068 * We have to be careful about races with
2069 * adding references to xattr block. Once we
2070 * hold buffer lock xattr block's state is
2071 * stable so we can check the additional
2072 * reference fits.
2073 */
2074 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2075 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2076 /*
2077 * Undo everything and check mbcache
2078 * again.
2079 */
2080 unlock_buffer(new_bh);
2081 dquot_free_block(inode,
2082 EXT4_C2B(EXT4_SB(sb),
2083 1));
2084 brelse(new_bh);
2085 mb_cache_entry_put(ea_block_cache, ce);
2086 ce = NULL;
2087 new_bh = NULL;
2088 goto inserted;
2089 }
2090 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2091 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2092 clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2093 ea_bdebug(new_bh, "reusing; refcount now=%d",
2094 ref);
2095 ext4_xattr_block_csum_set(inode, new_bh);
2096 unlock_buffer(new_bh);
2097 error = ext4_handle_dirty_metadata(handle,
2098 inode,
2099 new_bh);
2100 if (error)
2101 goto cleanup_dquot;
2102 }
2103 mb_cache_entry_touch(ea_block_cache, ce);
2104 mb_cache_entry_put(ea_block_cache, ce);
2105 ce = NULL;
2106 } else if (bs->bh && s->base == bs->bh->b_data) {
2107 /* We were modifying this block in-place. */
2108 ea_bdebug(bs->bh, "keeping this block");
2109 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2110 new_bh = bs->bh;
2111 get_bh(new_bh);
2112 } else {
2113 /* We need to allocate a new block */
2114 ext4_fsblk_t goal, block;
2115
2116#ifdef EXT4_XATTR_DEBUG
2117 WARN_ON_ONCE(dquot_initialize_needed(inode));
2118#endif
2119 goal = ext4_group_first_block_no(sb,
2120 EXT4_I(inode)->i_block_group);
2121 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2122 NULL, &error);
2123 if (error)
2124 goto cleanup;
2125
2126 ea_idebug(inode, "creating block %llu",
2127 (unsigned long long)block);
2128
2129 new_bh = sb_getblk(sb, block);
2130 if (unlikely(!new_bh)) {
2131 error = -ENOMEM;
2132getblk_failed:
2133 ext4_free_blocks(handle, inode, NULL, block, 1,
2134 EXT4_FREE_BLOCKS_METADATA);
2135 goto cleanup;
2136 }
2137 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2138 ENTRY(header(s->base)+1));
2139 if (error)
2140 goto getblk_failed;
2141 if (ea_inode) {
2142 /* Drop the extra ref on ea_inode. */
2143 error = ext4_xattr_inode_dec_ref(handle,
2144 ea_inode);
2145 if (error)
2146 ext4_warning_inode(ea_inode,
2147 "dec ref error=%d",
2148 error);
2149 iput(ea_inode);
2150 ea_inode = NULL;
2151 }
2152
2153 lock_buffer(new_bh);
2154 error = ext4_journal_get_create_access(handle, sb,
2155 new_bh, EXT4_JTR_NONE);
2156 if (error) {
2157 unlock_buffer(new_bh);
2158 error = -EIO;
2159 goto getblk_failed;
2160 }
2161 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2162 ext4_xattr_block_csum_set(inode, new_bh);
2163 set_buffer_uptodate(new_bh);
2164 unlock_buffer(new_bh);
2165 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2166 error = ext4_handle_dirty_metadata(handle, inode,
2167 new_bh);
2168 if (error)
2169 goto cleanup;
2170 }
2171 }
2172
2173 if (old_ea_inode_quota)
2174 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2175
2176 /* Update the inode. */
2177 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2178
2179 /* Drop the previous xattr block. */
2180 if (bs->bh && bs->bh != new_bh) {
2181 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2182
2183 ext4_xattr_release_block(handle, inode, bs->bh,
2184 &ea_inode_array,
2185 0 /* extra_credits */);
2186 ext4_xattr_inode_array_free(ea_inode_array);
2187 }
2188 error = 0;
2189
2190cleanup:
2191 if (ea_inode) {
2192 if (error) {
2193 int error2;
2194
2195 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2196 if (error2)
2197 ext4_warning_inode(ea_inode, "dec ref error=%d",
2198 error2);
2199 ext4_xattr_inode_free_quota(inode, ea_inode,
2200 i_size_read(ea_inode));
2201 }
2202 iput(ea_inode);
2203 }
2204 if (ce)
2205 mb_cache_entry_put(ea_block_cache, ce);
2206 brelse(new_bh);
2207 if (!(bs->bh && s->base == bs->bh->b_data))
2208 kfree(s->base);
2209
2210 return error;
2211
2212cleanup_dquot:
2213 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2214 goto cleanup;
2215
2216bad_block:
2217 EXT4_ERROR_INODE(inode, "bad block %llu",
2218 EXT4_I(inode)->i_file_acl);
2219 goto cleanup;
2220
2221#undef header
2222}
2223
2224int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2225 struct ext4_xattr_ibody_find *is)
2226{
2227 struct ext4_xattr_ibody_header *header;
2228 struct ext4_inode *raw_inode;
2229 int error;
2230
2231 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2232 return 0;
2233
2234 raw_inode = ext4_raw_inode(&is->iloc);
2235 header = IHDR(inode, raw_inode);
2236 is->s.base = is->s.first = IFIRST(header);
2237 is->s.here = is->s.first;
2238 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2239 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2240 error = xattr_check_inode(inode, header, is->s.end);
2241 if (error)
2242 return error;
2243 /* Find the named attribute. */
2244 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2245 i->name_index, i->name, 0);
2246 if (error && error != -ENODATA)
2247 return error;
2248 is->s.not_found = error;
2249 }
2250 return 0;
2251}
2252
2253int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2254 struct ext4_xattr_info *i,
2255 struct ext4_xattr_ibody_find *is)
2256{
2257 struct ext4_xattr_ibody_header *header;
2258 struct ext4_xattr_search *s = &is->s;
2259 struct inode *ea_inode = NULL;
2260 int error;
2261
2262 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2263 return -ENOSPC;
2264
2265 /* If we need EA inode, prepare it before locking the buffer */
2266 if (i->value && i->in_inode) {
2267 WARN_ON_ONCE(!i->value_len);
2268
2269 ea_inode = ext4_xattr_inode_lookup_create(handle, inode,
2270 i->value, i->value_len);
2271 if (IS_ERR(ea_inode))
2272 return PTR_ERR(ea_inode);
2273 }
2274 error = ext4_xattr_set_entry(i, s, handle, inode, ea_inode,
2275 false /* is_block */);
2276 if (error) {
2277 if (ea_inode) {
2278 int error2;
2279
2280 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2281 if (error2)
2282 ext4_warning_inode(ea_inode, "dec ref error=%d",
2283 error2);
2284
2285 ext4_xattr_inode_free_quota(inode, ea_inode,
2286 i_size_read(ea_inode));
2287 iput(ea_inode);
2288 }
2289 return error;
2290 }
2291 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2292 if (!IS_LAST_ENTRY(s->first)) {
2293 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2294 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2295 } else {
2296 header->h_magic = cpu_to_le32(0);
2297 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2298 }
2299 iput(ea_inode);
2300 return 0;
2301}
2302
2303static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2304 struct ext4_xattr_info *i)
2305{
2306 void *value;
2307
2308 /* When e_value_inum is set the value is stored externally. */
2309 if (s->here->e_value_inum)
2310 return 0;
2311 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2312 return 0;
2313 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2314 return !memcmp(value, i->value, i->value_len);
2315}
2316
2317static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2318{
2319 struct buffer_head *bh;
2320 int error;
2321
2322 if (!EXT4_I(inode)->i_file_acl)
2323 return NULL;
2324 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2325 if (IS_ERR(bh))
2326 return bh;
2327 error = ext4_xattr_check_block(inode, bh);
2328 if (error) {
2329 brelse(bh);
2330 return ERR_PTR(error);
2331 }
2332 return bh;
2333}
2334
2335/*
2336 * ext4_xattr_set_handle()
2337 *
2338 * Create, replace or remove an extended attribute for this inode. Value
2339 * is NULL to remove an existing extended attribute, and non-NULL to
2340 * either replace an existing extended attribute, or create a new extended
2341 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2342 * specify that an extended attribute must exist and must not exist
2343 * previous to the call, respectively.
2344 *
2345 * Returns 0, or a negative error number on failure.
2346 */
2347int
2348ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2349 const char *name, const void *value, size_t value_len,
2350 int flags)
2351{
2352 struct ext4_xattr_info i = {
2353 .name_index = name_index,
2354 .name = name,
2355 .value = value,
2356 .value_len = value_len,
2357 .in_inode = 0,
2358 };
2359 struct ext4_xattr_ibody_find is = {
2360 .s = { .not_found = -ENODATA, },
2361 };
2362 struct ext4_xattr_block_find bs = {
2363 .s = { .not_found = -ENODATA, },
2364 };
2365 int no_expand;
2366 int error;
2367
2368 if (!name)
2369 return -EINVAL;
2370 if (strlen(name) > 255)
2371 return -ERANGE;
2372
2373 ext4_write_lock_xattr(inode, &no_expand);
2374
2375 /* Check journal credits under write lock. */
2376 if (ext4_handle_valid(handle)) {
2377 struct buffer_head *bh;
2378 int credits;
2379
2380 bh = ext4_xattr_get_block(inode);
2381 if (IS_ERR(bh)) {
2382 error = PTR_ERR(bh);
2383 goto cleanup;
2384 }
2385
2386 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2387 value_len,
2388 flags & XATTR_CREATE);
2389 brelse(bh);
2390
2391 if (jbd2_handle_buffer_credits(handle) < credits) {
2392 error = -ENOSPC;
2393 goto cleanup;
2394 }
2395 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2396 }
2397
2398 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2399 if (error)
2400 goto cleanup;
2401
2402 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2403 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2404 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2405 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2406 }
2407
2408 error = ext4_xattr_ibody_find(inode, &i, &is);
2409 if (error)
2410 goto cleanup;
2411 if (is.s.not_found)
2412 error = ext4_xattr_block_find(inode, &i, &bs);
2413 if (error)
2414 goto cleanup;
2415 if (is.s.not_found && bs.s.not_found) {
2416 error = -ENODATA;
2417 if (flags & XATTR_REPLACE)
2418 goto cleanup;
2419 error = 0;
2420 if (!value)
2421 goto cleanup;
2422 } else {
2423 error = -EEXIST;
2424 if (flags & XATTR_CREATE)
2425 goto cleanup;
2426 }
2427
2428 if (!value) {
2429 if (!is.s.not_found)
2430 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2431 else if (!bs.s.not_found)
2432 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2433 } else {
2434 error = 0;
2435 /* Xattr value did not change? Save us some work and bail out */
2436 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2437 goto cleanup;
2438 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2439 goto cleanup;
2440
2441 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2442 (EXT4_XATTR_SIZE(i.value_len) >
2443 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2444 i.in_inode = 1;
2445retry_inode:
2446 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2447 if (!error && !bs.s.not_found) {
2448 i.value = NULL;
2449 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2450 } else if (error == -ENOSPC) {
2451 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2452 brelse(bs.bh);
2453 bs.bh = NULL;
2454 error = ext4_xattr_block_find(inode, &i, &bs);
2455 if (error)
2456 goto cleanup;
2457 }
2458 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2459 if (!error && !is.s.not_found) {
2460 i.value = NULL;
2461 error = ext4_xattr_ibody_set(handle, inode, &i,
2462 &is);
2463 } else if (error == -ENOSPC) {
2464 /*
2465 * Xattr does not fit in the block, store at
2466 * external inode if possible.
2467 */
2468 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2469 i.value_len && !i.in_inode) {
2470 i.in_inode = 1;
2471 goto retry_inode;
2472 }
2473 }
2474 }
2475 }
2476 if (!error) {
2477 ext4_xattr_update_super_block(handle, inode->i_sb);
2478 inode_set_ctime_current(inode);
2479 inode_inc_iversion(inode);
2480 if (!value)
2481 no_expand = 0;
2482 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2483 /*
2484 * The bh is consumed by ext4_mark_iloc_dirty, even with
2485 * error != 0.
2486 */
2487 is.iloc.bh = NULL;
2488 if (IS_SYNC(inode))
2489 ext4_handle_sync(handle);
2490 }
2491 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2492
2493cleanup:
2494 brelse(is.iloc.bh);
2495 brelse(bs.bh);
2496 ext4_write_unlock_xattr(inode, &no_expand);
2497 return error;
2498}
2499
2500int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2501 bool is_create, int *credits)
2502{
2503 struct buffer_head *bh;
2504 int err;
2505
2506 *credits = 0;
2507
2508 if (!EXT4_SB(inode->i_sb)->s_journal)
2509 return 0;
2510
2511 down_read(&EXT4_I(inode)->xattr_sem);
2512
2513 bh = ext4_xattr_get_block(inode);
2514 if (IS_ERR(bh)) {
2515 err = PTR_ERR(bh);
2516 } else {
2517 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2518 value_len, is_create);
2519 brelse(bh);
2520 err = 0;
2521 }
2522
2523 up_read(&EXT4_I(inode)->xattr_sem);
2524 return err;
2525}
2526
2527/*
2528 * ext4_xattr_set()
2529 *
2530 * Like ext4_xattr_set_handle, but start from an inode. This extended
2531 * attribute modification is a filesystem transaction by itself.
2532 *
2533 * Returns 0, or a negative error number on failure.
2534 */
2535int
2536ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2537 const void *value, size_t value_len, int flags)
2538{
2539 handle_t *handle;
2540 struct super_block *sb = inode->i_sb;
2541 int error, retries = 0;
2542 int credits;
2543
2544 error = dquot_initialize(inode);
2545 if (error)
2546 return error;
2547
2548retry:
2549 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2550 &credits);
2551 if (error)
2552 return error;
2553
2554 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2555 if (IS_ERR(handle)) {
2556 error = PTR_ERR(handle);
2557 } else {
2558 int error2;
2559
2560 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2561 value, value_len, flags);
2562 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR,
2563 handle);
2564 error2 = ext4_journal_stop(handle);
2565 if (error == -ENOSPC &&
2566 ext4_should_retry_alloc(sb, &retries))
2567 goto retry;
2568 if (error == 0)
2569 error = error2;
2570 }
2571
2572 return error;
2573}
2574
2575/*
2576 * Shift the EA entries in the inode to create space for the increased
2577 * i_extra_isize.
2578 */
2579static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2580 int value_offs_shift, void *to,
2581 void *from, size_t n)
2582{
2583 struct ext4_xattr_entry *last = entry;
2584 int new_offs;
2585
2586 /* We always shift xattr headers further thus offsets get lower */
2587 BUG_ON(value_offs_shift > 0);
2588
2589 /* Adjust the value offsets of the entries */
2590 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2591 if (!last->e_value_inum && last->e_value_size) {
2592 new_offs = le16_to_cpu(last->e_value_offs) +
2593 value_offs_shift;
2594 last->e_value_offs = cpu_to_le16(new_offs);
2595 }
2596 }
2597 /* Shift the entries by n bytes */
2598 memmove(to, from, n);
2599}
2600
2601/*
2602 * Move xattr pointed to by 'entry' from inode into external xattr block
2603 */
2604static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2605 struct ext4_inode *raw_inode,
2606 struct ext4_xattr_entry *entry)
2607{
2608 struct ext4_xattr_ibody_find *is = NULL;
2609 struct ext4_xattr_block_find *bs = NULL;
2610 char *buffer = NULL, *b_entry_name = NULL;
2611 size_t value_size = le32_to_cpu(entry->e_value_size);
2612 struct ext4_xattr_info i = {
2613 .value = NULL,
2614 .value_len = 0,
2615 .name_index = entry->e_name_index,
2616 .in_inode = !!entry->e_value_inum,
2617 };
2618 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2619 int needs_kvfree = 0;
2620 int error;
2621
2622 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2623 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2624 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2625 if (!is || !bs || !b_entry_name) {
2626 error = -ENOMEM;
2627 goto out;
2628 }
2629
2630 is->s.not_found = -ENODATA;
2631 bs->s.not_found = -ENODATA;
2632 is->iloc.bh = NULL;
2633 bs->bh = NULL;
2634
2635 /* Save the entry name and the entry value */
2636 if (entry->e_value_inum) {
2637 buffer = kvmalloc(value_size, GFP_NOFS);
2638 if (!buffer) {
2639 error = -ENOMEM;
2640 goto out;
2641 }
2642 needs_kvfree = 1;
2643 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2644 if (error)
2645 goto out;
2646 } else {
2647 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2648 buffer = (void *)IFIRST(header) + value_offs;
2649 }
2650
2651 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2652 b_entry_name[entry->e_name_len] = '\0';
2653 i.name = b_entry_name;
2654
2655 error = ext4_get_inode_loc(inode, &is->iloc);
2656 if (error)
2657 goto out;
2658
2659 error = ext4_xattr_ibody_find(inode, &i, is);
2660 if (error)
2661 goto out;
2662
2663 i.value = buffer;
2664 i.value_len = value_size;
2665 error = ext4_xattr_block_find(inode, &i, bs);
2666 if (error)
2667 goto out;
2668
2669 /* Move ea entry from the inode into the block */
2670 error = ext4_xattr_block_set(handle, inode, &i, bs);
2671 if (error)
2672 goto out;
2673
2674 /* Remove the chosen entry from the inode */
2675 i.value = NULL;
2676 i.value_len = 0;
2677 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2678
2679out:
2680 kfree(b_entry_name);
2681 if (needs_kvfree && buffer)
2682 kvfree(buffer);
2683 if (is)
2684 brelse(is->iloc.bh);
2685 if (bs)
2686 brelse(bs->bh);
2687 kfree(is);
2688 kfree(bs);
2689
2690 return error;
2691}
2692
2693static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2694 struct ext4_inode *raw_inode,
2695 int isize_diff, size_t ifree,
2696 size_t bfree, int *total_ino)
2697{
2698 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2699 struct ext4_xattr_entry *small_entry;
2700 struct ext4_xattr_entry *entry;
2701 struct ext4_xattr_entry *last;
2702 unsigned int entry_size; /* EA entry size */
2703 unsigned int total_size; /* EA entry size + value size */
2704 unsigned int min_total_size;
2705 int error;
2706
2707 while (isize_diff > ifree) {
2708 entry = NULL;
2709 small_entry = NULL;
2710 min_total_size = ~0U;
2711 last = IFIRST(header);
2712 /* Find the entry best suited to be pushed into EA block */
2713 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2714 /* never move system.data out of the inode */
2715 if ((last->e_name_len == 4) &&
2716 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2717 !memcmp(last->e_name, "data", 4))
2718 continue;
2719 total_size = EXT4_XATTR_LEN(last->e_name_len);
2720 if (!last->e_value_inum)
2721 total_size += EXT4_XATTR_SIZE(
2722 le32_to_cpu(last->e_value_size));
2723 if (total_size <= bfree &&
2724 total_size < min_total_size) {
2725 if (total_size + ifree < isize_diff) {
2726 small_entry = last;
2727 } else {
2728 entry = last;
2729 min_total_size = total_size;
2730 }
2731 }
2732 }
2733
2734 if (entry == NULL) {
2735 if (small_entry == NULL)
2736 return -ENOSPC;
2737 entry = small_entry;
2738 }
2739
2740 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2741 total_size = entry_size;
2742 if (!entry->e_value_inum)
2743 total_size += EXT4_XATTR_SIZE(
2744 le32_to_cpu(entry->e_value_size));
2745 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2746 entry);
2747 if (error)
2748 return error;
2749
2750 *total_ino -= entry_size;
2751 ifree += total_size;
2752 bfree -= total_size;
2753 }
2754
2755 return 0;
2756}
2757
2758/*
2759 * Expand an inode by new_extra_isize bytes when EAs are present.
2760 * Returns 0 on success or negative error number on failure.
2761 */
2762int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2763 struct ext4_inode *raw_inode, handle_t *handle)
2764{
2765 struct ext4_xattr_ibody_header *header;
2766 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2767 static unsigned int mnt_count;
2768 size_t min_offs;
2769 size_t ifree, bfree;
2770 int total_ino;
2771 void *base, *end;
2772 int error = 0, tried_min_extra_isize = 0;
2773 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2774 int isize_diff; /* How much do we need to grow i_extra_isize */
2775
2776retry:
2777 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2778 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2779 return 0;
2780
2781 header = IHDR(inode, raw_inode);
2782
2783 /*
2784 * Check if enough free space is available in the inode to shift the
2785 * entries ahead by new_extra_isize.
2786 */
2787
2788 base = IFIRST(header);
2789 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2790 min_offs = end - base;
2791 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2792
2793 error = xattr_check_inode(inode, header, end);
2794 if (error)
2795 goto cleanup;
2796
2797 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2798 if (ifree >= isize_diff)
2799 goto shift;
2800
2801 /*
2802 * Enough free space isn't available in the inode, check if
2803 * EA block can hold new_extra_isize bytes.
2804 */
2805 if (EXT4_I(inode)->i_file_acl) {
2806 struct buffer_head *bh;
2807
2808 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2809 if (IS_ERR(bh)) {
2810 error = PTR_ERR(bh);
2811 goto cleanup;
2812 }
2813 error = ext4_xattr_check_block(inode, bh);
2814 if (error) {
2815 brelse(bh);
2816 goto cleanup;
2817 }
2818 base = BHDR(bh);
2819 end = bh->b_data + bh->b_size;
2820 min_offs = end - base;
2821 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2822 NULL);
2823 brelse(bh);
2824 if (bfree + ifree < isize_diff) {
2825 if (!tried_min_extra_isize && s_min_extra_isize) {
2826 tried_min_extra_isize++;
2827 new_extra_isize = s_min_extra_isize;
2828 goto retry;
2829 }
2830 error = -ENOSPC;
2831 goto cleanup;
2832 }
2833 } else {
2834 bfree = inode->i_sb->s_blocksize;
2835 }
2836
2837 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2838 isize_diff, ifree, bfree,
2839 &total_ino);
2840 if (error) {
2841 if (error == -ENOSPC && !tried_min_extra_isize &&
2842 s_min_extra_isize) {
2843 tried_min_extra_isize++;
2844 new_extra_isize = s_min_extra_isize;
2845 goto retry;
2846 }
2847 goto cleanup;
2848 }
2849shift:
2850 /* Adjust the offsets and shift the remaining entries ahead */
2851 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2852 - new_extra_isize, (void *)raw_inode +
2853 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2854 (void *)header, total_ino);
2855 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2856
2857 if (ext4_has_inline_data(inode))
2858 error = ext4_find_inline_data_nolock(inode);
2859
2860cleanup:
2861 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2862 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2863 inode->i_ino);
2864 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2865 }
2866 return error;
2867}
2868
2869#define EIA_INCR 16 /* must be 2^n */
2870#define EIA_MASK (EIA_INCR - 1)
2871
2872/* Add the large xattr @inode into @ea_inode_array for deferred iput().
2873 * If @ea_inode_array is new or full it will be grown and the old
2874 * contents copied over.
2875 */
2876static int
2877ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2878 struct inode *inode)
2879{
2880 if (*ea_inode_array == NULL) {
2881 /*
2882 * Start with 15 inodes, so it fits into a power-of-two size.
2883 */
2884 (*ea_inode_array) = kmalloc(
2885 struct_size(*ea_inode_array, inodes, EIA_MASK),
2886 GFP_NOFS);
2887 if (*ea_inode_array == NULL)
2888 return -ENOMEM;
2889 (*ea_inode_array)->count = 0;
2890 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2891 /* expand the array once all 15 + n * 16 slots are full */
2892 struct ext4_xattr_inode_array *new_array = NULL;
2893
2894 new_array = kmalloc(
2895 struct_size(*ea_inode_array, inodes,
2896 (*ea_inode_array)->count + EIA_INCR),
2897 GFP_NOFS);
2898 if (new_array == NULL)
2899 return -ENOMEM;
2900 memcpy(new_array, *ea_inode_array,
2901 struct_size(*ea_inode_array, inodes,
2902 (*ea_inode_array)->count));
2903 kfree(*ea_inode_array);
2904 *ea_inode_array = new_array;
2905 }
2906 (*ea_inode_array)->count++;
2907 (*ea_inode_array)->inodes[(*ea_inode_array)->count - 1] = inode;
2908 return 0;
2909}
2910
2911/*
2912 * ext4_xattr_delete_inode()
2913 *
2914 * Free extended attribute resources associated with this inode. Traverse
2915 * all entries and decrement reference on any xattr inodes associated with this
2916 * inode. This is called immediately before an inode is freed. We have exclusive
2917 * access to the inode. If an orphan inode is deleted it will also release its
2918 * references on xattr block and xattr inodes.
2919 */
2920int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2921 struct ext4_xattr_inode_array **ea_inode_array,
2922 int extra_credits)
2923{
2924 struct buffer_head *bh = NULL;
2925 struct ext4_xattr_ibody_header *header;
2926 struct ext4_iloc iloc = { .bh = NULL };
2927 struct ext4_xattr_entry *entry;
2928 struct inode *ea_inode;
2929 int error;
2930
2931 error = ext4_journal_ensure_credits(handle, extra_credits,
2932 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2933 if (error < 0) {
2934 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2935 goto cleanup;
2936 }
2937
2938 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2939 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2940
2941 error = ext4_get_inode_loc(inode, &iloc);
2942 if (error) {
2943 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2944 goto cleanup;
2945 }
2946
2947 error = ext4_journal_get_write_access(handle, inode->i_sb,
2948 iloc.bh, EXT4_JTR_NONE);
2949 if (error) {
2950 EXT4_ERROR_INODE(inode, "write access (error %d)",
2951 error);
2952 goto cleanup;
2953 }
2954
2955 header = IHDR(inode, ext4_raw_inode(&iloc));
2956 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2957 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2958 IFIRST(header),
2959 false /* block_csum */,
2960 ea_inode_array,
2961 extra_credits,
2962 false /* skip_quota */);
2963 }
2964
2965 if (EXT4_I(inode)->i_file_acl) {
2966 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2967 if (IS_ERR(bh)) {
2968 error = PTR_ERR(bh);
2969 if (error == -EIO) {
2970 EXT4_ERROR_INODE_ERR(inode, EIO,
2971 "block %llu read error",
2972 EXT4_I(inode)->i_file_acl);
2973 }
2974 bh = NULL;
2975 goto cleanup;
2976 }
2977 error = ext4_xattr_check_block(inode, bh);
2978 if (error)
2979 goto cleanup;
2980
2981 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2982 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2983 entry = EXT4_XATTR_NEXT(entry)) {
2984 if (!entry->e_value_inum)
2985 continue;
2986 error = ext4_xattr_inode_iget(inode,
2987 le32_to_cpu(entry->e_value_inum),
2988 le32_to_cpu(entry->e_hash),
2989 &ea_inode);
2990 if (error)
2991 continue;
2992 ext4_xattr_inode_free_quota(inode, ea_inode,
2993 le32_to_cpu(entry->e_value_size));
2994 iput(ea_inode);
2995 }
2996
2997 }
2998
2999 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
3000 extra_credits);
3001 /*
3002 * Update i_file_acl value in the same transaction that releases
3003 * block.
3004 */
3005 EXT4_I(inode)->i_file_acl = 0;
3006 error = ext4_mark_inode_dirty(handle, inode);
3007 if (error) {
3008 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
3009 error);
3010 goto cleanup;
3011 }
3012 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
3013 }
3014 error = 0;
3015cleanup:
3016 brelse(iloc.bh);
3017 brelse(bh);
3018 return error;
3019}
3020
3021void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
3022{
3023 int idx;
3024
3025 if (ea_inode_array == NULL)
3026 return;
3027
3028 for (idx = 0; idx < ea_inode_array->count; ++idx)
3029 iput(ea_inode_array->inodes[idx]);
3030 kfree(ea_inode_array);
3031}
3032
3033/*
3034 * ext4_xattr_block_cache_insert()
3035 *
3036 * Create a new entry in the extended attribute block cache, and insert
3037 * it unless such an entry is already in the cache.
3038 */
3039static void
3040ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
3041 struct buffer_head *bh)
3042{
3043 struct ext4_xattr_header *header = BHDR(bh);
3044 __u32 hash = le32_to_cpu(header->h_hash);
3045 int reusable = le32_to_cpu(header->h_refcount) <
3046 EXT4_XATTR_REFCOUNT_MAX;
3047 int error;
3048
3049 if (!ea_block_cache)
3050 return;
3051 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3052 bh->b_blocknr, reusable);
3053 if (error) {
3054 if (error == -EBUSY)
3055 ea_bdebug(bh, "already in cache");
3056 } else
3057 ea_bdebug(bh, "inserting [%x]", (int)hash);
3058}
3059
3060/*
3061 * ext4_xattr_cmp()
3062 *
3063 * Compare two extended attribute blocks for equality.
3064 *
3065 * Returns 0 if the blocks are equal, 1 if they differ.
3066 */
3067static int
3068ext4_xattr_cmp(struct ext4_xattr_header *header1,
3069 struct ext4_xattr_header *header2)
3070{
3071 struct ext4_xattr_entry *entry1, *entry2;
3072
3073 entry1 = ENTRY(header1+1);
3074 entry2 = ENTRY(header2+1);
3075 while (!IS_LAST_ENTRY(entry1)) {
3076 if (IS_LAST_ENTRY(entry2))
3077 return 1;
3078 if (entry1->e_hash != entry2->e_hash ||
3079 entry1->e_name_index != entry2->e_name_index ||
3080 entry1->e_name_len != entry2->e_name_len ||
3081 entry1->e_value_size != entry2->e_value_size ||
3082 entry1->e_value_inum != entry2->e_value_inum ||
3083 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3084 return 1;
3085 if (!entry1->e_value_inum &&
3086 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3087 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3088 le32_to_cpu(entry1->e_value_size)))
3089 return 1;
3090
3091 entry1 = EXT4_XATTR_NEXT(entry1);
3092 entry2 = EXT4_XATTR_NEXT(entry2);
3093 }
3094 if (!IS_LAST_ENTRY(entry2))
3095 return 1;
3096 return 0;
3097}
3098
3099/*
3100 * ext4_xattr_block_cache_find()
3101 *
3102 * Find an identical extended attribute block.
3103 *
3104 * Returns a pointer to the block found, or NULL if such a block was not
3105 * found, or an error pointer if an error occurred while reading ea block.
3106 */
3107static struct buffer_head *
3108ext4_xattr_block_cache_find(struct inode *inode,
3109 struct ext4_xattr_header *header,
3110 struct mb_cache_entry **pce)
3111{
3112 __u32 hash = le32_to_cpu(header->h_hash);
3113 struct mb_cache_entry *ce;
3114 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3115
3116 if (!ea_block_cache)
3117 return NULL;
3118 if (!header->h_hash)
3119 return NULL; /* never share */
3120 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3121 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3122 while (ce) {
3123 struct buffer_head *bh;
3124
3125 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3126 if (IS_ERR(bh)) {
3127 if (PTR_ERR(bh) != -ENOMEM)
3128 EXT4_ERROR_INODE(inode, "block %lu read error",
3129 (unsigned long)ce->e_value);
3130 mb_cache_entry_put(ea_block_cache, ce);
3131 return bh;
3132 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3133 *pce = ce;
3134 return bh;
3135 }
3136 brelse(bh);
3137 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3138 }
3139 return NULL;
3140}
3141
3142#define NAME_HASH_SHIFT 5
3143#define VALUE_HASH_SHIFT 16
3144
3145/*
3146 * ext4_xattr_hash_entry()
3147 *
3148 * Compute the hash of an extended attribute.
3149 */
3150static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3151 size_t value_count)
3152{
3153 __u32 hash = 0;
3154
3155 while (name_len--) {
3156 hash = (hash << NAME_HASH_SHIFT) ^
3157 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3158 (unsigned char)*name++;
3159 }
3160 while (value_count--) {
3161 hash = (hash << VALUE_HASH_SHIFT) ^
3162 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3163 le32_to_cpu(*value++);
3164 }
3165 return cpu_to_le32(hash);
3166}
3167
3168/*
3169 * ext4_xattr_hash_entry_signed()
3170 *
3171 * Compute the hash of an extended attribute incorrectly.
3172 */
3173static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3174{
3175 __u32 hash = 0;
3176
3177 while (name_len--) {
3178 hash = (hash << NAME_HASH_SHIFT) ^
3179 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3180 (signed char)*name++;
3181 }
3182 while (value_count--) {
3183 hash = (hash << VALUE_HASH_SHIFT) ^
3184 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3185 le32_to_cpu(*value++);
3186 }
3187 return cpu_to_le32(hash);
3188}
3189
3190#undef NAME_HASH_SHIFT
3191#undef VALUE_HASH_SHIFT
3192
3193#define BLOCK_HASH_SHIFT 16
3194
3195/*
3196 * ext4_xattr_rehash()
3197 *
3198 * Re-compute the extended attribute hash value after an entry has changed.
3199 */
3200static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3201{
3202 struct ext4_xattr_entry *here;
3203 __u32 hash = 0;
3204
3205 here = ENTRY(header+1);
3206 while (!IS_LAST_ENTRY(here)) {
3207 if (!here->e_hash) {
3208 /* Block is not shared if an entry's hash value == 0 */
3209 hash = 0;
3210 break;
3211 }
3212 hash = (hash << BLOCK_HASH_SHIFT) ^
3213 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3214 le32_to_cpu(here->e_hash);
3215 here = EXT4_XATTR_NEXT(here);
3216 }
3217 header->h_hash = cpu_to_le32(hash);
3218}
3219
3220#undef BLOCK_HASH_SHIFT
3221
3222#define HASH_BUCKET_BITS 10
3223
3224struct mb_cache *
3225ext4_xattr_create_cache(void)
3226{
3227 return mb_cache_create(HASH_BUCKET_BITS);
3228}
3229
3230void ext4_xattr_destroy_cache(struct mb_cache *cache)
3231{
3232 if (cache)
3233 mb_cache_destroy(cache);
3234}
3235