projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'mm-hotfixes-stable-2025-07-11-16-16' of git://git.kernel.org/pub/scm/linux...
[linux-2.6-block.git] / fs / ext4 / xattr.c
... / ...
CommitLineData
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->s_csum_seed, (__u8 *)&dsk_block_nr,
143 sizeof(dsk_block_nr));
144 csum = ext4_chksum(csum, (__u8 *)hdr, offset);
145 csum = ext4_chksum(csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
146 offset += sizeof(dummy_csum);
147 csum = ext4_chksum(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_feature_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_feature_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
311int
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
319static int
320xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
321 void *end, int name_index, const char *name, int sorted)
322{
323 struct ext4_xattr_entry *entry, *next;
324 size_t name_len;
325 int cmp = 1;
326
327 if (name == NULL)
328 return -EINVAL;
329 name_len = strlen(name);
330 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
331 next = EXT4_XATTR_NEXT(entry);
332 if ((void *) next >= end) {
333 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
334 return -EFSCORRUPTED;
335 }
336 cmp = name_index - entry->e_name_index;
337 if (!cmp)
338 cmp = name_len - entry->e_name_len;
339 if (!cmp)
340 cmp = memcmp(name, entry->e_name, name_len);
341 if (cmp <= 0 && (sorted || cmp == 0))
342 break;
343 }
344 *pentry = entry;
345 return cmp ? -ENODATA : 0;
346}
347
348static u32
349ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
350{
351 return ext4_chksum(sbi->s_csum_seed, buffer, size);
352}
353
354static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
355{
356 return ((u64) inode_get_ctime_sec(ea_inode) << 32) |
357 (u32) inode_peek_iversion_raw(ea_inode);
358}
359
360static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
361{
362 inode_set_ctime(ea_inode, (u32)(ref_count >> 32), 0);
363 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
364}
365
366static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
367{
368 return (u32) inode_get_atime_sec(ea_inode);
369}
370
371static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
372{
373 inode_set_atime(ea_inode, hash, 0);
374}
375
376/*
377 * Read the EA value from an inode.
378 */
379static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
380{
381 int blocksize = 1 << ea_inode->i_blkbits;
382 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
383 int tail_size = (size % blocksize) ?: blocksize;
384 struct buffer_head *bhs_inline[8];
385 struct buffer_head **bhs = bhs_inline;
386 int i, ret;
387
388 if (bh_count > ARRAY_SIZE(bhs_inline)) {
389 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
390 if (!bhs)
391 return -ENOMEM;
392 }
393
394 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
395 true /* wait */, bhs);
396 if (ret)
397 goto free_bhs;
398
399 for (i = 0; i < bh_count; i++) {
400 /* There shouldn't be any holes in ea_inode. */
401 if (!bhs[i]) {
402 ret = -EFSCORRUPTED;
403 goto put_bhs;
404 }
405 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
406 i < bh_count - 1 ? blocksize : tail_size);
407 }
408 ret = 0;
409put_bhs:
410 for (i = 0; i < bh_count; i++)
411 brelse(bhs[i]);
412free_bhs:
413 if (bhs != bhs_inline)
414 kfree(bhs);
415 return ret;
416}
417
418#define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode_get_mtime_sec(inode)))
419
420static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
421 u32 ea_inode_hash, struct inode **ea_inode)
422{
423 struct inode *inode;
424 int err;
425
426 /*
427 * We have to check for this corruption early as otherwise
428 * iget_locked() could wait indefinitely for the state of our
429 * parent inode.
430 */
431 if (parent->i_ino == ea_ino) {
432 ext4_error(parent->i_sb,
433 "Parent and EA inode have the same ino %lu", ea_ino);
434 return -EFSCORRUPTED;
435 }
436
437 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_EA_INODE);
438 if (IS_ERR(inode)) {
439 err = PTR_ERR(inode);
440 ext4_error(parent->i_sb,
441 "error while reading EA inode %lu err=%d", ea_ino,
442 err);
443 return err;
444 }
445 ext4_xattr_inode_set_class(inode);
446
447 /*
448 * Check whether this is an old Lustre-style xattr inode. Lustre
449 * implementation does not have hash validation, rather it has a
450 * backpointer from ea_inode to the parent inode.
451 */
452 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
453 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
454 inode->i_generation == parent->i_generation) {
455 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
456 ext4_xattr_inode_set_ref(inode, 1);
457 } else {
458 inode_lock_nested(inode, I_MUTEX_XATTR);
459 inode->i_flags |= S_NOQUOTA;
460 inode_unlock(inode);
461 }
462
463 *ea_inode = inode;
464 return 0;
465}
466
467/* Remove entry from mbcache when EA inode is getting evicted */
468void ext4_evict_ea_inode(struct inode *inode)
469{
470 struct mb_cache_entry *oe;
471
472 if (!EA_INODE_CACHE(inode))
473 return;
474 /* Wait for entry to get unused so that we can remove it */
475 while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
476 ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
477 mb_cache_entry_wait_unused(oe);
478 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
479 }
480}
481
482static int
483ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
484 struct ext4_xattr_entry *entry, void *buffer,
485 size_t size)
486{
487 u32 hash;
488
489 /* Verify stored hash matches calculated hash. */
490 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
491 if (hash != ext4_xattr_inode_get_hash(ea_inode))
492 return -EFSCORRUPTED;
493
494 if (entry) {
495 __le32 e_hash, tmp_data;
496
497 /* Verify entry hash. */
498 tmp_data = cpu_to_le32(hash);
499 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
500 &tmp_data, 1);
501 /* All good? */
502 if (e_hash == entry->e_hash)
503 return 0;
504
505 /*
506 * Not good. Maybe the entry hash was calculated
507 * using the buggy signed char version?
508 */
509 e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
510 &tmp_data, 1);
511 /* Still no match - bad */
512 if (e_hash != entry->e_hash)
513 return -EFSCORRUPTED;
514
515 /* Let people know about old hash */
516 pr_warn_once("ext4: filesystem with signed xattr name hash");
517 }
518 return 0;
519}
520
521/*
522 * Read xattr value from the EA inode.
523 */
524static int
525ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
526 void *buffer, size_t size)
527{
528 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
529 struct inode *ea_inode;
530 int err;
531
532 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
533 le32_to_cpu(entry->e_hash), &ea_inode);
534 if (err) {
535 ea_inode = NULL;
536 goto out;
537 }
538
539 if (i_size_read(ea_inode) != size) {
540 ext4_warning_inode(ea_inode,
541 "ea_inode file size=%llu entry size=%zu",
542 i_size_read(ea_inode), size);
543 err = -EFSCORRUPTED;
544 goto out;
545 }
546
547 err = ext4_xattr_inode_read(ea_inode, buffer, size);
548 if (err)
549 goto out;
550
551 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
552 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
553 size);
554 if (err) {
555 ext4_warning_inode(ea_inode,
556 "EA inode hash validation failed");
557 goto out;
558 }
559
560 if (ea_inode_cache)
561 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
562 ext4_xattr_inode_get_hash(ea_inode),
563 ea_inode->i_ino, true /* reusable */);
564 }
565out:
566 iput(ea_inode);
567 return err;
568}
569
570static int
571ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
572 void *buffer, size_t buffer_size)
573{
574 struct buffer_head *bh = NULL;
575 struct ext4_xattr_entry *entry;
576 size_t size;
577 void *end;
578 int error;
579 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
580
581 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
582 name_index, name, buffer, (long)buffer_size);
583
584 if (!EXT4_I(inode)->i_file_acl)
585 return -ENODATA;
586 ea_idebug(inode, "reading block %llu",
587 (unsigned long long)EXT4_I(inode)->i_file_acl);
588 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
589 if (IS_ERR(bh))
590 return PTR_ERR(bh);
591 ea_bdebug(bh, "b_count=%d, refcount=%d",
592 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
593 error = ext4_xattr_check_block(inode, bh);
594 if (error)
595 goto cleanup;
596 ext4_xattr_block_cache_insert(ea_block_cache, bh);
597 entry = BFIRST(bh);
598 end = bh->b_data + bh->b_size;
599 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
600 if (error)
601 goto cleanup;
602 size = le32_to_cpu(entry->e_value_size);
603 error = -ERANGE;
604 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
605 goto cleanup;
606 if (buffer) {
607 if (size > buffer_size)
608 goto cleanup;
609 if (entry->e_value_inum) {
610 error = ext4_xattr_inode_get(inode, entry, buffer,
611 size);
612 if (error)
613 goto cleanup;
614 } else {
615 u16 offset = le16_to_cpu(entry->e_value_offs);
616 void *p = bh->b_data + offset;
617
618 if (unlikely(p + size > end))
619 goto cleanup;
620 memcpy(buffer, p, size);
621 }
622 }
623 error = size;
624
625cleanup:
626 brelse(bh);
627 return error;
628}
629
630int
631ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
632 void *buffer, size_t buffer_size)
633{
634 struct ext4_xattr_ibody_header *header;
635 struct ext4_xattr_entry *entry;
636 struct ext4_inode *raw_inode;
637 struct ext4_iloc iloc;
638 size_t size;
639 void *end;
640 int error;
641
642 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
643 return -ENODATA;
644 error = ext4_get_inode_loc(inode, &iloc);
645 if (error)
646 return error;
647 raw_inode = ext4_raw_inode(&iloc);
648 header = IHDR(inode, raw_inode);
649 end = ITAIL(inode, raw_inode);
650 entry = IFIRST(header);
651 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
652 if (error)
653 goto cleanup;
654 size = le32_to_cpu(entry->e_value_size);
655 error = -ERANGE;
656 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
657 goto cleanup;
658 if (buffer) {
659 if (size > buffer_size)
660 goto cleanup;
661 if (entry->e_value_inum) {
662 error = ext4_xattr_inode_get(inode, entry, buffer,
663 size);
664 if (error)
665 goto cleanup;
666 } else {
667 u16 offset = le16_to_cpu(entry->e_value_offs);
668 void *p = (void *)IFIRST(header) + offset;
669
670 if (unlikely(p + size > end))
671 goto cleanup;
672 memcpy(buffer, p, size);
673 }
674 }
675 error = size;
676
677cleanup:
678 brelse(iloc.bh);
679 return error;
680}
681
682/*
683 * ext4_xattr_get()
684 *
685 * Copy an extended attribute into the buffer
686 * provided, or compute the buffer size required.
687 * Buffer is NULL to compute the size of the buffer required.
688 *
689 * Returns a negative error number on failure, or the number of bytes
690 * used / required on success.
691 */
692int
693ext4_xattr_get(struct inode *inode, int name_index, const char *name,
694 void *buffer, size_t buffer_size)
695{
696 int error;
697
698 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
699 return -EIO;
700
701 if (strlen(name) > 255)
702 return -ERANGE;
703
704 down_read(&EXT4_I(inode)->xattr_sem);
705 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
706 buffer_size);
707 if (error == -ENODATA)
708 error = ext4_xattr_block_get(inode, name_index, name, buffer,
709 buffer_size);
710 up_read(&EXT4_I(inode)->xattr_sem);
711 return error;
712}
713
714static int
715ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
716 char *buffer, size_t buffer_size)
717{
718 size_t rest = buffer_size;
719
720 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
721 const char *prefix;
722
723 prefix = ext4_xattr_prefix(entry->e_name_index, dentry);
724 if (prefix) {
725 size_t prefix_len = strlen(prefix);
726 size_t size = prefix_len + entry->e_name_len + 1;
727
728 if (buffer) {
729 if (size > rest)
730 return -ERANGE;
731 memcpy(buffer, prefix, prefix_len);
732 buffer += prefix_len;
733 memcpy(buffer, entry->e_name, entry->e_name_len);
734 buffer += entry->e_name_len;
735 *buffer++ = 0;
736 }
737 rest -= size;
738 }
739 }
740 return buffer_size - rest; /* total size */
741}
742
743static int
744ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
745{
746 struct inode *inode = d_inode(dentry);
747 struct buffer_head *bh = NULL;
748 int error;
749
750 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
751 buffer, (long)buffer_size);
752
753 if (!EXT4_I(inode)->i_file_acl)
754 return 0;
755 ea_idebug(inode, "reading block %llu",
756 (unsigned long long)EXT4_I(inode)->i_file_acl);
757 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
758 if (IS_ERR(bh))
759 return PTR_ERR(bh);
760 ea_bdebug(bh, "b_count=%d, refcount=%d",
761 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
762 error = ext4_xattr_check_block(inode, bh);
763 if (error)
764 goto cleanup;
765 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
766 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
767 buffer_size);
768cleanup:
769 brelse(bh);
770 return error;
771}
772
773static int
774ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
775{
776 struct inode *inode = d_inode(dentry);
777 struct ext4_xattr_ibody_header *header;
778 struct ext4_inode *raw_inode;
779 struct ext4_iloc iloc;
780 int error;
781
782 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
783 return 0;
784 error = ext4_get_inode_loc(inode, &iloc);
785 if (error)
786 return error;
787 raw_inode = ext4_raw_inode(&iloc);
788 header = IHDR(inode, raw_inode);
789 error = ext4_xattr_list_entries(dentry, IFIRST(header),
790 buffer, buffer_size);
791
792 brelse(iloc.bh);
793 return error;
794}
795
796/*
797 * Inode operation listxattr()
798 *
799 * d_inode(dentry)->i_rwsem: don't care
800 *
801 * Copy a list of attribute names into the buffer
802 * provided, or compute the buffer size required.
803 * Buffer is NULL to compute the size of the buffer required.
804 *
805 * Returns a negative error number on failure, or the number of bytes
806 * used / required on success.
807 */
808ssize_t
809ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
810{
811 int ret, ret2;
812
813 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
814 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
815 if (ret < 0)
816 goto errout;
817 if (buffer) {
818 buffer += ret;
819 buffer_size -= ret;
820 }
821 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
822 if (ret < 0)
823 goto errout;
824 ret += ret2;
825errout:
826 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
827 return ret;
828}
829
830/*
831 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
832 * not set, set it.
833 */
834static void ext4_xattr_update_super_block(handle_t *handle,
835 struct super_block *sb)
836{
837 if (ext4_has_feature_xattr(sb))
838 return;
839
840 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
841 if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
842 EXT4_JTR_NONE) == 0) {
843 lock_buffer(EXT4_SB(sb)->s_sbh);
844 ext4_set_feature_xattr(sb);
845 ext4_superblock_csum_set(sb);
846 unlock_buffer(EXT4_SB(sb)->s_sbh);
847 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
848 }
849}
850
851int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
852{
853 struct ext4_iloc iloc = { .bh = NULL };
854 struct buffer_head *bh = NULL;
855 struct ext4_inode *raw_inode;
856 struct ext4_xattr_ibody_header *header;
857 struct ext4_xattr_entry *entry;
858 qsize_t ea_inode_refs = 0;
859 int ret;
860
861 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
862
863 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
864 ret = ext4_get_inode_loc(inode, &iloc);
865 if (ret)
866 goto out;
867 raw_inode = ext4_raw_inode(&iloc);
868 header = IHDR(inode, raw_inode);
869
870 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
871 entry = EXT4_XATTR_NEXT(entry))
872 if (entry->e_value_inum)
873 ea_inode_refs++;
874 }
875
876 if (EXT4_I(inode)->i_file_acl) {
877 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
878 if (IS_ERR(bh)) {
879 ret = PTR_ERR(bh);
880 bh = NULL;
881 goto out;
882 }
883
884 ret = ext4_xattr_check_block(inode, bh);
885 if (ret)
886 goto out;
887
888 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
889 entry = EXT4_XATTR_NEXT(entry))
890 if (entry->e_value_inum)
891 ea_inode_refs++;
892 }
893 *usage = ea_inode_refs + 1;
894 ret = 0;
895out:
896 brelse(iloc.bh);
897 brelse(bh);
898 return ret;
899}
900
901static inline size_t round_up_cluster(struct inode *inode, size_t length)
902{
903 struct super_block *sb = inode->i_sb;
904 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
905 inode->i_blkbits);
906 size_t mask = ~(cluster_size - 1);
907
908 return (length + cluster_size - 1) & mask;
909}
910
911static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
912{
913 int err;
914
915 err = dquot_alloc_inode(inode);
916 if (err)
917 return err;
918 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
919 if (err)
920 dquot_free_inode(inode);
921 return err;
922}
923
924static void ext4_xattr_inode_free_quota(struct inode *parent,
925 struct inode *ea_inode,
926 size_t len)
927{
928 if (ea_inode &&
929 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
930 return;
931 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
932 dquot_free_inode(parent);
933}
934
935int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
936 struct buffer_head *block_bh, size_t value_len,
937 bool is_create)
938{
939 int credits;
940 int blocks;
941
942 /*
943 * 1) Owner inode update
944 * 2) Ref count update on old xattr block
945 * 3) new xattr block
946 * 4) block bitmap update for new xattr block
947 * 5) group descriptor for new xattr block
948 * 6) block bitmap update for old xattr block
949 * 7) group descriptor for old block
950 *
951 * 6 & 7 can happen if we have two racing threads T_a and T_b
952 * which are each trying to set an xattr on inodes I_a and I_b
953 * which were both initially sharing an xattr block.
954 */
955 credits = 7;
956
957 /* Quota updates. */
958 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
959
960 /*
961 * In case of inline data, we may push out the data to a block,
962 * so we need to reserve credits for this eventuality
963 */
964 if (inode && ext4_has_inline_data(inode))
965 credits += ext4_writepage_trans_blocks(inode) + 1;
966
967 /* We are done if ea_inode feature is not enabled. */
968 if (!ext4_has_feature_ea_inode(sb))
969 return credits;
970
971 /* New ea_inode, inode map, block bitmap, group descriptor. */
972 credits += 4;
973
974 /* Data blocks. */
975 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
976
977 /* Indirection block or one level of extent tree. */
978 blocks += 1;
979
980 /* Block bitmap and group descriptor updates for each block. */
981 credits += blocks * 2;
982
983 /* Blocks themselves. */
984 credits += blocks;
985
986 if (!is_create) {
987 /* Dereference ea_inode holding old xattr value.
988 * Old ea_inode, inode map, block bitmap, group descriptor.
989 */
990 credits += 4;
991
992 /* Data blocks for old ea_inode. */
993 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
994
995 /* Indirection block or one level of extent tree for old
996 * ea_inode.
997 */
998 blocks += 1;
999
1000 /* Block bitmap and group descriptor updates for each block. */
1001 credits += blocks * 2;
1002 }
1003
1004 /* We may need to clone the existing xattr block in which case we need
1005 * to increment ref counts for existing ea_inodes referenced by it.
1006 */
1007 if (block_bh) {
1008 struct ext4_xattr_entry *entry = BFIRST(block_bh);
1009
1010 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1011 if (entry->e_value_inum)
1012 /* Ref count update on ea_inode. */
1013 credits += 1;
1014 }
1015 return credits;
1016}
1017
1018static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1019 int ref_change)
1020{
1021 struct ext4_iloc iloc;
1022 s64 ref_count;
1023 int ret;
1024
1025 inode_lock_nested(ea_inode, I_MUTEX_XATTR);
1026
1027 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1028 if (ret)
1029 goto out;
1030
1031 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1032 ref_count += ref_change;
1033 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1034
1035 if (ref_change > 0) {
1036 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1037 ea_inode->i_ino, ref_count);
1038
1039 if (ref_count == 1) {
1040 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1041 ea_inode->i_ino, ea_inode->i_nlink);
1042
1043 set_nlink(ea_inode, 1);
1044 ext4_orphan_del(handle, ea_inode);
1045 }
1046 } else {
1047 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1048 ea_inode->i_ino, ref_count);
1049
1050 if (ref_count == 0) {
1051 WARN_ONCE(ea_inode->i_nlink != 1,
1052 "EA inode %lu i_nlink=%u",
1053 ea_inode->i_ino, ea_inode->i_nlink);
1054
1055 clear_nlink(ea_inode);
1056 ext4_orphan_add(handle, ea_inode);
1057 }
1058 }
1059
1060 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1061 if (ret)
1062 ext4_warning_inode(ea_inode,
1063 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1064out:
1065 inode_unlock(ea_inode);
1066 return ret;
1067}
1068
1069static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1070{
1071 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1072}
1073
1074static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1075{
1076 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1077}
1078
1079static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1080 struct ext4_xattr_entry *first)
1081{
1082 struct inode *ea_inode;
1083 struct ext4_xattr_entry *entry;
1084 struct ext4_xattr_entry *failed_entry;
1085 unsigned int ea_ino;
1086 int err, saved_err;
1087
1088 for (entry = first; !IS_LAST_ENTRY(entry);
1089 entry = EXT4_XATTR_NEXT(entry)) {
1090 if (!entry->e_value_inum)
1091 continue;
1092 ea_ino = le32_to_cpu(entry->e_value_inum);
1093 err = ext4_xattr_inode_iget(parent, ea_ino,
1094 le32_to_cpu(entry->e_hash),
1095 &ea_inode);
1096 if (err)
1097 goto cleanup;
1098 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1099 if (err) {
1100 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1101 iput(ea_inode);
1102 goto cleanup;
1103 }
1104 iput(ea_inode);
1105 }
1106 return 0;
1107
1108cleanup:
1109 saved_err = err;
1110 failed_entry = entry;
1111
1112 for (entry = first; entry != failed_entry;
1113 entry = EXT4_XATTR_NEXT(entry)) {
1114 if (!entry->e_value_inum)
1115 continue;
1116 ea_ino = le32_to_cpu(entry->e_value_inum);
1117 err = ext4_xattr_inode_iget(parent, ea_ino,
1118 le32_to_cpu(entry->e_hash),
1119 &ea_inode);
1120 if (err) {
1121 ext4_warning(parent->i_sb,
1122 "cleanup ea_ino %u iget error %d", ea_ino,
1123 err);
1124 continue;
1125 }
1126 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1127 if (err)
1128 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1129 err);
1130 iput(ea_inode);
1131 }
1132 return saved_err;
1133}
1134
1135static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1136 struct buffer_head *bh, bool block_csum, bool dirty)
1137{
1138 int error;
1139
1140 if (bh && dirty) {
1141 if (block_csum)
1142 ext4_xattr_block_csum_set(inode, bh);
1143 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1144 if (error) {
1145 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1146 error);
1147 return error;
1148 }
1149 }
1150 return 0;
1151}
1152
1153static void
1154ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1155 struct buffer_head *bh,
1156 struct ext4_xattr_entry *first, bool block_csum,
1157 struct ext4_xattr_inode_array **ea_inode_array,
1158 int extra_credits, bool skip_quota)
1159{
1160 struct inode *ea_inode;
1161 struct ext4_xattr_entry *entry;
1162 struct ext4_iloc iloc;
1163 bool dirty = false;
1164 unsigned int ea_ino;
1165 int err;
1166 int credits;
1167 void *end;
1168
1169 if (block_csum)
1170 end = (void *)bh->b_data + bh->b_size;
1171 else {
1172 ext4_get_inode_loc(parent, &iloc);
1173 end = (void *)ext4_raw_inode(&iloc) + EXT4_SB(parent->i_sb)->s_inode_size;
1174 }
1175
1176 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1177 credits = 2 + extra_credits;
1178
1179 for (entry = first; (void *)entry < end && !IS_LAST_ENTRY(entry);
1180 entry = EXT4_XATTR_NEXT(entry)) {
1181 if (!entry->e_value_inum)
1182 continue;
1183 ea_ino = le32_to_cpu(entry->e_value_inum);
1184 err = ext4_xattr_inode_iget(parent, ea_ino,
1185 le32_to_cpu(entry->e_hash),
1186 &ea_inode);
1187 if (err)
1188 continue;
1189
1190 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1191 if (err) {
1192 ext4_warning_inode(ea_inode,
1193 "Expand inode array err=%d", err);
1194 iput(ea_inode);
1195 continue;
1196 }
1197
1198 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1199 ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1200 ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1201 dirty));
1202 if (err < 0) {
1203 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1204 err);
1205 continue;
1206 }
1207 if (err > 0) {
1208 err = ext4_journal_get_write_access(handle,
1209 parent->i_sb, bh, EXT4_JTR_NONE);
1210 if (err) {
1211 ext4_warning_inode(ea_inode,
1212 "Re-get write access err=%d",
1213 err);
1214 continue;
1215 }
1216 }
1217
1218 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1219 if (err) {
1220 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1221 err);
1222 continue;
1223 }
1224
1225 if (!skip_quota)
1226 ext4_xattr_inode_free_quota(parent, ea_inode,
1227 le32_to_cpu(entry->e_value_size));
1228
1229 /*
1230 * Forget about ea_inode within the same transaction that
1231 * decrements the ref count. This avoids duplicate decrements in
1232 * case the rest of the work spills over to subsequent
1233 * transactions.
1234 */
1235 entry->e_value_inum = 0;
1236 entry->e_value_size = 0;
1237
1238 dirty = true;
1239 }
1240
1241 if (dirty) {
1242 /*
1243 * Note that we are deliberately skipping csum calculation for
1244 * the final update because we do not expect any journal
1245 * restarts until xattr block is freed.
1246 */
1247
1248 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1249 if (err)
1250 ext4_warning_inode(parent,
1251 "handle dirty metadata err=%d", err);
1252 }
1253}
1254
1255/*
1256 * Release the xattr block BH: If the reference count is > 1, decrement it;
1257 * otherwise free the block.
1258 */
1259static void
1260ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1261 struct buffer_head *bh,
1262 struct ext4_xattr_inode_array **ea_inode_array,
1263 int extra_credits)
1264{
1265 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1266 u32 hash, ref;
1267 int error = 0;
1268
1269 BUFFER_TRACE(bh, "get_write_access");
1270 error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1271 EXT4_JTR_NONE);
1272 if (error)
1273 goto out;
1274
1275retry_ref:
1276 lock_buffer(bh);
1277 hash = le32_to_cpu(BHDR(bh)->h_hash);
1278 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1279 if (ref == 1) {
1280 ea_bdebug(bh, "refcount now=0; freeing");
1281 /*
1282 * This must happen under buffer lock for
1283 * ext4_xattr_block_set() to reliably detect freed block
1284 */
1285 if (ea_block_cache) {
1286 struct mb_cache_entry *oe;
1287
1288 oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1289 bh->b_blocknr);
1290 if (oe) {
1291 unlock_buffer(bh);
1292 mb_cache_entry_wait_unused(oe);
1293 mb_cache_entry_put(ea_block_cache, oe);
1294 goto retry_ref;
1295 }
1296 }
1297 get_bh(bh);
1298 unlock_buffer(bh);
1299
1300 if (ext4_has_feature_ea_inode(inode->i_sb))
1301 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1302 BFIRST(bh),
1303 true /* block_csum */,
1304 ea_inode_array,
1305 extra_credits,
1306 true /* skip_quota */);
1307 ext4_free_blocks(handle, inode, bh, 0, 1,
1308 EXT4_FREE_BLOCKS_METADATA |
1309 EXT4_FREE_BLOCKS_FORGET);
1310 } else {
1311 ref--;
1312 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1313 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1314 struct mb_cache_entry *ce;
1315
1316 if (ea_block_cache) {
1317 ce = mb_cache_entry_get(ea_block_cache, hash,
1318 bh->b_blocknr);
1319 if (ce) {
1320 set_bit(MBE_REUSABLE_B, &ce->e_flags);
1321 mb_cache_entry_put(ea_block_cache, ce);
1322 }
1323 }
1324 }
1325
1326 ext4_xattr_block_csum_set(inode, bh);
1327 /*
1328 * Beware of this ugliness: Releasing of xattr block references
1329 * from different inodes can race and so we have to protect
1330 * from a race where someone else frees the block (and releases
1331 * its journal_head) before we are done dirtying the buffer. In
1332 * nojournal mode this race is harmless and we actually cannot
1333 * call ext4_handle_dirty_metadata() with locked buffer as
1334 * that function can call sync_dirty_buffer() so for that case
1335 * we handle the dirtying after unlocking the buffer.
1336 */
1337 if (ext4_handle_valid(handle))
1338 error = ext4_handle_dirty_metadata(handle, inode, bh);
1339 unlock_buffer(bh);
1340 if (!ext4_handle_valid(handle))
1341 error = ext4_handle_dirty_metadata(handle, inode, bh);
1342 if (IS_SYNC(inode))
1343 ext4_handle_sync(handle);
1344 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1345 ea_bdebug(bh, "refcount now=%d; releasing",
1346 le32_to_cpu(BHDR(bh)->h_refcount));
1347 }
1348out:
1349 ext4_std_error(inode->i_sb, error);
1350 return;
1351}
1352
1353/*
1354 * Find the available free space for EAs. This also returns the total number of
1355 * bytes used by EA entries.
1356 */
1357static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1358 size_t *min_offs, void *base, int *total)
1359{
1360 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1361 if (!last->e_value_inum && last->e_value_size) {
1362 size_t offs = le16_to_cpu(last->e_value_offs);
1363 if (offs < *min_offs)
1364 *min_offs = offs;
1365 }
1366 if (total)
1367 *total += EXT4_XATTR_LEN(last->e_name_len);
1368 }
1369 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1370}
1371
1372/*
1373 * Write the value of the EA in an inode.
1374 */
1375static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1376 const void *buf, int bufsize)
1377{
1378 struct buffer_head *bh = NULL;
1379 unsigned long block = 0;
1380 int blocksize = ea_inode->i_sb->s_blocksize;
1381 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1382 int csize, wsize = 0;
1383 int ret = 0, ret2 = 0;
1384 int retries = 0;
1385
1386retry:
1387 while (ret >= 0 && ret < max_blocks) {
1388 struct ext4_map_blocks map;
1389 map.m_lblk = block += ret;
1390 map.m_len = max_blocks -= ret;
1391
1392 ret = ext4_map_blocks(handle, ea_inode, &map,
1393 EXT4_GET_BLOCKS_CREATE);
1394 if (ret <= 0) {
1395 ext4_mark_inode_dirty(handle, ea_inode);
1396 if (ret == -ENOSPC &&
1397 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1398 ret = 0;
1399 goto retry;
1400 }
1401 break;
1402 }
1403 }
1404
1405 if (ret < 0)
1406 return ret;
1407
1408 block = 0;
1409 while (wsize < bufsize) {
1410 brelse(bh);
1411 csize = (bufsize - wsize) > blocksize ? blocksize :
1412 bufsize - wsize;
1413 bh = ext4_getblk(handle, ea_inode, block, 0);
1414 if (IS_ERR(bh))
1415 return PTR_ERR(bh);
1416 if (!bh) {
1417 WARN_ON_ONCE(1);
1418 EXT4_ERROR_INODE(ea_inode,
1419 "ext4_getblk() return bh = NULL");
1420 return -EFSCORRUPTED;
1421 }
1422 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1423 EXT4_JTR_NONE);
1424 if (ret)
1425 goto out;
1426
1427 memcpy(bh->b_data, buf, csize);
1428 /*
1429 * Zero out block tail to avoid writing uninitialized memory
1430 * to disk.
1431 */
1432 if (csize < blocksize)
1433 memset(bh->b_data + csize, 0, blocksize - csize);
1434 set_buffer_uptodate(bh);
1435 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1436
1437 buf += csize;
1438 wsize += csize;
1439 block += 1;
1440 }
1441
1442 inode_lock(ea_inode);
1443 i_size_write(ea_inode, wsize);
1444 ext4_update_i_disksize(ea_inode, wsize);
1445 inode_unlock(ea_inode);
1446
1447 ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1448 if (unlikely(ret2 && !ret))
1449 ret = ret2;
1450
1451out:
1452 brelse(bh);
1453
1454 return ret;
1455}
1456
1457/*
1458 * Create an inode to store the value of a large EA.
1459 */
1460static struct inode *ext4_xattr_inode_create(handle_t *handle,
1461 struct inode *inode, u32 hash)
1462{
1463 struct inode *ea_inode = NULL;
1464 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1465 int err;
1466
1467 if (inode->i_sb->s_root == NULL) {
1468 ext4_warning(inode->i_sb,
1469 "refuse to create EA inode when umounting");
1470 WARN_ON(1);
1471 return ERR_PTR(-EINVAL);
1472 }
1473
1474 /*
1475 * Let the next inode be the goal, so we try and allocate the EA inode
1476 * in the same group, or nearby one.
1477 */
1478 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1479 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1480 EXT4_EA_INODE_FL);
1481 if (!IS_ERR(ea_inode)) {
1482 ea_inode->i_op = &ext4_file_inode_operations;
1483 ea_inode->i_fop = &ext4_file_operations;
1484 ext4_set_aops(ea_inode);
1485 ext4_xattr_inode_set_class(ea_inode);
1486 unlock_new_inode(ea_inode);
1487 ext4_xattr_inode_set_ref(ea_inode, 1);
1488 ext4_xattr_inode_set_hash(ea_inode, hash);
1489 err = ext4_mark_inode_dirty(handle, ea_inode);
1490 if (!err)
1491 err = ext4_inode_attach_jinode(ea_inode);
1492 if (err) {
1493 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1494 ext4_warning_inode(ea_inode,
1495 "cleanup dec ref error %d", err);
1496 iput(ea_inode);
1497 return ERR_PTR(err);
1498 }
1499
1500 /*
1501 * Xattr inodes are shared therefore quota charging is performed
1502 * at a higher level.
1503 */
1504 dquot_free_inode(ea_inode);
1505 dquot_drop(ea_inode);
1506 inode_lock(ea_inode);
1507 ea_inode->i_flags |= S_NOQUOTA;
1508 inode_unlock(ea_inode);
1509 }
1510
1511 return ea_inode;
1512}
1513
1514static struct inode *
1515ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1516 size_t value_len, u32 hash)
1517{
1518 struct inode *ea_inode;
1519 struct mb_cache_entry *ce;
1520 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1521 void *ea_data;
1522
1523 if (!ea_inode_cache)
1524 return NULL;
1525
1526 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1527 if (!ce)
1528 return NULL;
1529
1530 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1531 !(current->flags & PF_MEMALLOC_NOFS));
1532
1533 ea_data = kvmalloc(value_len, GFP_KERNEL);
1534 if (!ea_data) {
1535 mb_cache_entry_put(ea_inode_cache, ce);
1536 return NULL;
1537 }
1538
1539 while (ce) {
1540 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1541 EXT4_IGET_EA_INODE);
1542 if (IS_ERR(ea_inode))
1543 goto next_entry;
1544 ext4_xattr_inode_set_class(ea_inode);
1545 if (i_size_read(ea_inode) == value_len &&
1546 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1547 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1548 value_len) &&
1549 !memcmp(value, ea_data, value_len)) {
1550 mb_cache_entry_touch(ea_inode_cache, ce);
1551 mb_cache_entry_put(ea_inode_cache, ce);
1552 kvfree(ea_data);
1553 return ea_inode;
1554 }
1555 iput(ea_inode);
1556 next_entry:
1557 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1558 }
1559 kvfree(ea_data);
1560 return NULL;
1561}
1562
1563/*
1564 * Add value of the EA in an inode.
1565 */
1566static struct inode *ext4_xattr_inode_lookup_create(handle_t *handle,
1567 struct inode *inode, const void *value, size_t value_len)
1568{
1569 struct inode *ea_inode;
1570 u32 hash;
1571 int err;
1572
1573 /* Account inode & space to quota even if sharing... */
1574 err = ext4_xattr_inode_alloc_quota(inode, value_len);
1575 if (err)
1576 return ERR_PTR(err);
1577
1578 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1579 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1580 if (ea_inode) {
1581 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1582 if (err)
1583 goto out_err;
1584 return ea_inode;
1585 }
1586
1587 /* Create an inode for the EA value */
1588 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1589 if (IS_ERR(ea_inode)) {
1590 ext4_xattr_inode_free_quota(inode, NULL, value_len);
1591 return ea_inode;
1592 }
1593
1594 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1595 if (err) {
1596 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1597 ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1598 goto out_err;
1599 }
1600
1601 if (EA_INODE_CACHE(inode))
1602 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1603 ea_inode->i_ino, true /* reusable */);
1604 return ea_inode;
1605out_err:
1606 iput(ea_inode);
1607 ext4_xattr_inode_free_quota(inode, NULL, value_len);
1608 return ERR_PTR(err);
1609}
1610
1611/*
1612 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1613 * feature is enabled.
1614 */
1615#define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1616
1617static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1618 struct ext4_xattr_search *s,
1619 handle_t *handle, struct inode *inode,
1620 struct inode *new_ea_inode,
1621 bool is_block)
1622{
1623 struct ext4_xattr_entry *last, *next;
1624 struct ext4_xattr_entry *here = s->here;
1625 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1626 int in_inode = i->in_inode;
1627 struct inode *old_ea_inode = NULL;
1628 size_t old_size, new_size;
1629 int ret;
1630
1631 /* Space used by old and new values. */
1632 old_size = (!s->not_found && !here->e_value_inum) ?
1633 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1634 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1635
1636 /*
1637 * Optimization for the simple case when old and new values have the
1638 * same padded sizes. Not applicable if external inodes are involved.
1639 */
1640 if (new_size && new_size == old_size) {
1641 size_t offs = le16_to_cpu(here->e_value_offs);
1642 void *val = s->base + offs;
1643
1644 here->e_value_size = cpu_to_le32(i->value_len);
1645 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1646 memset(val, 0, new_size);
1647 } else {
1648 memcpy(val, i->value, i->value_len);
1649 /* Clear padding bytes. */
1650 memset(val + i->value_len, 0, new_size - i->value_len);
1651 }
1652 goto update_hash;
1653 }
1654
1655 /* Compute min_offs and last. */
1656 last = s->first;
1657 for (; !IS_LAST_ENTRY(last); last = next) {
1658 next = EXT4_XATTR_NEXT(last);
1659 if ((void *)next >= s->end) {
1660 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1661 ret = -EFSCORRUPTED;
1662 goto out;
1663 }
1664 if (!last->e_value_inum && last->e_value_size) {
1665 size_t offs = le16_to_cpu(last->e_value_offs);
1666 if (offs < min_offs)
1667 min_offs = offs;
1668 }
1669 }
1670
1671 /* Check whether we have enough space. */
1672 if (i->value) {
1673 size_t free;
1674
1675 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1676 if (!s->not_found)
1677 free += EXT4_XATTR_LEN(name_len) + old_size;
1678
1679 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1680 ret = -ENOSPC;
1681 goto out;
1682 }
1683
1684 /*
1685 * If storing the value in an external inode is an option,
1686 * reserve space for xattr entries/names in the external
1687 * attribute block so that a long value does not occupy the
1688 * whole space and prevent further entries being added.
1689 */
1690 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1691 new_size && is_block &&
1692 (min_offs + old_size - new_size) <
1693 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1694 ret = -ENOSPC;
1695 goto out;
1696 }
1697 }
1698
1699 /*
1700 * Getting access to old and new ea inodes is subject to failures.
1701 * Finish that work before doing any modifications to the xattr data.
1702 */
1703 if (!s->not_found && here->e_value_inum) {
1704 ret = ext4_xattr_inode_iget(inode,
1705 le32_to_cpu(here->e_value_inum),
1706 le32_to_cpu(here->e_hash),
1707 &old_ea_inode);
1708 if (ret) {
1709 old_ea_inode = NULL;
1710 goto out;
1711 }
1712
1713 /* We are ready to release ref count on the old_ea_inode. */
1714 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1715 if (ret)
1716 goto out;
1717
1718 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1719 le32_to_cpu(here->e_value_size));
1720 }
1721
1722 /* No failures allowed past this point. */
1723
1724 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1725 /* Remove the old value. */
1726 void *first_val = s->base + min_offs;
1727 size_t offs = le16_to_cpu(here->e_value_offs);
1728 void *val = s->base + offs;
1729
1730 memmove(first_val + old_size, first_val, val - first_val);
1731 memset(first_val, 0, old_size);
1732 min_offs += old_size;
1733
1734 /* Adjust all value offsets. */
1735 last = s->first;
1736 while (!IS_LAST_ENTRY(last)) {
1737 size_t o = le16_to_cpu(last->e_value_offs);
1738
1739 if (!last->e_value_inum &&
1740 last->e_value_size && o < offs)
1741 last->e_value_offs = cpu_to_le16(o + old_size);
1742 last = EXT4_XATTR_NEXT(last);
1743 }
1744 }
1745
1746 if (!i->value) {
1747 /* Remove old name. */
1748 size_t size = EXT4_XATTR_LEN(name_len);
1749
1750 last = ENTRY((void *)last - size);
1751 memmove(here, (void *)here + size,
1752 (void *)last - (void *)here + sizeof(__u32));
1753 memset(last, 0, size);
1754
1755 /*
1756 * Update i_inline_off - moved ibody region might contain
1757 * system.data attribute. Handling a failure here won't
1758 * cause other complications for setting an xattr.
1759 */
1760 if (!is_block && ext4_has_inline_data(inode)) {
1761 ret = ext4_find_inline_data_nolock(inode);
1762 if (ret) {
1763 ext4_warning_inode(inode,
1764 "unable to update i_inline_off");
1765 goto out;
1766 }
1767 }
1768 } else if (s->not_found) {
1769 /* Insert new name. */
1770 size_t size = EXT4_XATTR_LEN(name_len);
1771 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1772
1773 memmove((void *)here + size, here, rest);
1774 memset(here, 0, size);
1775 here->e_name_index = i->name_index;
1776 here->e_name_len = name_len;
1777 memcpy(here->e_name, i->name, name_len);
1778 } else {
1779 /* This is an update, reset value info. */
1780 here->e_value_inum = 0;
1781 here->e_value_offs = 0;
1782 here->e_value_size = 0;
1783 }
1784
1785 if (i->value) {
1786 /* Insert new value. */
1787 if (in_inode) {
1788 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1789 } else if (i->value_len) {
1790 void *val = s->base + min_offs - new_size;
1791
1792 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1793 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1794 memset(val, 0, new_size);
1795 } else {
1796 memcpy(val, i->value, i->value_len);
1797 /* Clear padding bytes. */
1798 memset(val + i->value_len, 0,
1799 new_size - i->value_len);
1800 }
1801 }
1802 here->e_value_size = cpu_to_le32(i->value_len);
1803 }
1804
1805update_hash:
1806 if (i->value) {
1807 __le32 hash = 0;
1808
1809 /* Entry hash calculation. */
1810 if (in_inode) {
1811 __le32 crc32c_hash;
1812
1813 /*
1814 * Feed crc32c hash instead of the raw value for entry
1815 * hash calculation. This is to avoid walking
1816 * potentially long value buffer again.
1817 */
1818 crc32c_hash = cpu_to_le32(
1819 ext4_xattr_inode_get_hash(new_ea_inode));
1820 hash = ext4_xattr_hash_entry(here->e_name,
1821 here->e_name_len,
1822 &crc32c_hash, 1);
1823 } else if (is_block) {
1824 __le32 *value = s->base + le16_to_cpu(
1825 here->e_value_offs);
1826
1827 hash = ext4_xattr_hash_entry(here->e_name,
1828 here->e_name_len, value,
1829 new_size >> 2);
1830 }
1831 here->e_hash = hash;
1832 }
1833
1834 if (is_block)
1835 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1836
1837 ret = 0;
1838out:
1839 iput(old_ea_inode);
1840 return ret;
1841}
1842
1843struct ext4_xattr_block_find {
1844 struct ext4_xattr_search s;
1845 struct buffer_head *bh;
1846};
1847
1848static int
1849ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1850 struct ext4_xattr_block_find *bs)
1851{
1852 struct super_block *sb = inode->i_sb;
1853 int error;
1854
1855 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1856 i->name_index, i->name, i->value, (long)i->value_len);
1857
1858 if (EXT4_I(inode)->i_file_acl) {
1859 /* The inode already has an extended attribute block. */
1860 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1861 if (IS_ERR(bs->bh)) {
1862 error = PTR_ERR(bs->bh);
1863 bs->bh = NULL;
1864 return error;
1865 }
1866 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1867 atomic_read(&(bs->bh->b_count)),
1868 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1869 error = ext4_xattr_check_block(inode, bs->bh);
1870 if (error)
1871 return error;
1872 /* Find the named attribute. */
1873 bs->s.base = BHDR(bs->bh);
1874 bs->s.first = BFIRST(bs->bh);
1875 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1876 bs->s.here = bs->s.first;
1877 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1878 i->name_index, i->name, 1);
1879 if (error && error != -ENODATA)
1880 return error;
1881 bs->s.not_found = error;
1882 }
1883 return 0;
1884}
1885
1886static int
1887ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1888 struct ext4_xattr_info *i,
1889 struct ext4_xattr_block_find *bs)
1890{
1891 struct super_block *sb = inode->i_sb;
1892 struct buffer_head *new_bh = NULL;
1893 struct ext4_xattr_search s_copy = bs->s;
1894 struct ext4_xattr_search *s = &s_copy;
1895 struct mb_cache_entry *ce = NULL;
1896 int error = 0;
1897 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1898 struct inode *ea_inode = NULL, *tmp_inode;
1899 size_t old_ea_inode_quota = 0;
1900 unsigned int ea_ino;
1901
1902#define header(x) ((struct ext4_xattr_header *)(x))
1903
1904 /* If we need EA inode, prepare it before locking the buffer */
1905 if (i->value && i->in_inode) {
1906 WARN_ON_ONCE(!i->value_len);
1907
1908 ea_inode = ext4_xattr_inode_lookup_create(handle, inode,
1909 i->value, i->value_len);
1910 if (IS_ERR(ea_inode)) {
1911 error = PTR_ERR(ea_inode);
1912 ea_inode = NULL;
1913 goto cleanup;
1914 }
1915 }
1916
1917 if (s->base) {
1918 int offset = (char *)s->here - bs->bh->b_data;
1919
1920 BUFFER_TRACE(bs->bh, "get_write_access");
1921 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1922 EXT4_JTR_NONE);
1923 if (error)
1924 goto cleanup;
1925
1926 lock_buffer(bs->bh);
1927
1928 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1929 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1930
1931 /*
1932 * This must happen under buffer lock for
1933 * ext4_xattr_block_set() to reliably detect modified
1934 * block
1935 */
1936 if (ea_block_cache) {
1937 struct mb_cache_entry *oe;
1938
1939 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1940 hash, bs->bh->b_blocknr);
1941 if (oe) {
1942 /*
1943 * Xattr block is getting reused. Leave
1944 * it alone.
1945 */
1946 mb_cache_entry_put(ea_block_cache, oe);
1947 goto clone_block;
1948 }
1949 }
1950 ea_bdebug(bs->bh, "modifying in-place");
1951 error = ext4_xattr_set_entry(i, s, handle, inode,
1952 ea_inode, true /* is_block */);
1953 ext4_xattr_block_csum_set(inode, bs->bh);
1954 unlock_buffer(bs->bh);
1955 if (error == -EFSCORRUPTED)
1956 goto bad_block;
1957 if (!error)
1958 error = ext4_handle_dirty_metadata(handle,
1959 inode,
1960 bs->bh);
1961 if (error)
1962 goto cleanup;
1963 goto inserted;
1964 }
1965clone_block:
1966 unlock_buffer(bs->bh);
1967 ea_bdebug(bs->bh, "cloning");
1968 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1969 error = -ENOMEM;
1970 if (s->base == NULL)
1971 goto cleanup;
1972 s->first = ENTRY(header(s->base)+1);
1973 header(s->base)->h_refcount = cpu_to_le32(1);
1974 s->here = ENTRY(s->base + offset);
1975 s->end = s->base + bs->bh->b_size;
1976
1977 /*
1978 * If existing entry points to an xattr inode, we need
1979 * to prevent ext4_xattr_set_entry() from decrementing
1980 * ref count on it because the reference belongs to the
1981 * original block. In this case, make the entry look
1982 * like it has an empty value.
1983 */
1984 if (!s->not_found && s->here->e_value_inum) {
1985 ea_ino = le32_to_cpu(s->here->e_value_inum);
1986 error = ext4_xattr_inode_iget(inode, ea_ino,
1987 le32_to_cpu(s->here->e_hash),
1988 &tmp_inode);
1989 if (error)
1990 goto cleanup;
1991
1992 if (!ext4_test_inode_state(tmp_inode,
1993 EXT4_STATE_LUSTRE_EA_INODE)) {
1994 /*
1995 * Defer quota free call for previous
1996 * inode until success is guaranteed.
1997 */
1998 old_ea_inode_quota = le32_to_cpu(
1999 s->here->e_value_size);
2000 }
2001 iput(tmp_inode);
2002
2003 s->here->e_value_inum = 0;
2004 s->here->e_value_size = 0;
2005 }
2006 } else {
2007 /* Allocate a buffer where we construct the new block. */
2008 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
2009 error = -ENOMEM;
2010 if (s->base == NULL)
2011 goto cleanup;
2012 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2013 header(s->base)->h_blocks = cpu_to_le32(1);
2014 header(s->base)->h_refcount = cpu_to_le32(1);
2015 s->first = ENTRY(header(s->base)+1);
2016 s->here = ENTRY(header(s->base)+1);
2017 s->end = s->base + sb->s_blocksize;
2018 }
2019
2020 error = ext4_xattr_set_entry(i, s, handle, inode, ea_inode,
2021 true /* is_block */);
2022 if (error == -EFSCORRUPTED)
2023 goto bad_block;
2024 if (error)
2025 goto cleanup;
2026
2027inserted:
2028 if (!IS_LAST_ENTRY(s->first)) {
2029 new_bh = ext4_xattr_block_cache_find(inode, header(s->base), &ce);
2030 if (IS_ERR(new_bh)) {
2031 error = PTR_ERR(new_bh);
2032 new_bh = NULL;
2033 goto cleanup;
2034 }
2035
2036 if (new_bh) {
2037 /* We found an identical block in the cache. */
2038 if (new_bh == bs->bh)
2039 ea_bdebug(new_bh, "keeping");
2040 else {
2041 u32 ref;
2042
2043#ifdef EXT4_XATTR_DEBUG
2044 WARN_ON_ONCE(dquot_initialize_needed(inode));
2045#endif
2046 /* The old block is released after updating
2047 the inode. */
2048 error = dquot_alloc_block(inode,
2049 EXT4_C2B(EXT4_SB(sb), 1));
2050 if (error)
2051 goto cleanup;
2052 BUFFER_TRACE(new_bh, "get_write_access");
2053 error = ext4_journal_get_write_access(
2054 handle, sb, new_bh,
2055 EXT4_JTR_NONE);
2056 if (error)
2057 goto cleanup_dquot;
2058 lock_buffer(new_bh);
2059 /*
2060 * We have to be careful about races with
2061 * adding references to xattr block. Once we
2062 * hold buffer lock xattr block's state is
2063 * stable so we can check the additional
2064 * reference fits.
2065 */
2066 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2067 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2068 /*
2069 * Undo everything and check mbcache
2070 * again.
2071 */
2072 unlock_buffer(new_bh);
2073 dquot_free_block(inode,
2074 EXT4_C2B(EXT4_SB(sb),
2075 1));
2076 brelse(new_bh);
2077 mb_cache_entry_put(ea_block_cache, ce);
2078 ce = NULL;
2079 new_bh = NULL;
2080 goto inserted;
2081 }
2082 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2083 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2084 clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2085 ea_bdebug(new_bh, "reusing; refcount now=%d",
2086 ref);
2087 ext4_xattr_block_csum_set(inode, new_bh);
2088 unlock_buffer(new_bh);
2089 error = ext4_handle_dirty_metadata(handle,
2090 inode,
2091 new_bh);
2092 if (error)
2093 goto cleanup_dquot;
2094 }
2095 mb_cache_entry_touch(ea_block_cache, ce);
2096 mb_cache_entry_put(ea_block_cache, ce);
2097 ce = NULL;
2098 } else if (bs->bh && s->base == bs->bh->b_data) {
2099 /* We were modifying this block in-place. */
2100 ea_bdebug(bs->bh, "keeping this block");
2101 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2102 new_bh = bs->bh;
2103 get_bh(new_bh);
2104 } else {
2105 /* We need to allocate a new block */
2106 ext4_fsblk_t goal, block;
2107
2108#ifdef EXT4_XATTR_DEBUG
2109 WARN_ON_ONCE(dquot_initialize_needed(inode));
2110#endif
2111 goal = ext4_group_first_block_no(sb,
2112 EXT4_I(inode)->i_block_group);
2113 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2114 NULL, &error);
2115 if (error)
2116 goto cleanup;
2117
2118 ea_idebug(inode, "creating block %llu",
2119 (unsigned long long)block);
2120
2121 new_bh = sb_getblk(sb, block);
2122 if (unlikely(!new_bh)) {
2123 error = -ENOMEM;
2124getblk_failed:
2125 ext4_free_blocks(handle, inode, NULL, block, 1,
2126 EXT4_FREE_BLOCKS_METADATA);
2127 goto cleanup;
2128 }
2129 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2130 ENTRY(header(s->base)+1));
2131 if (error)
2132 goto getblk_failed;
2133 if (ea_inode) {
2134 /* Drop the extra ref on ea_inode. */
2135 error = ext4_xattr_inode_dec_ref(handle,
2136 ea_inode);
2137 if (error)
2138 ext4_warning_inode(ea_inode,
2139 "dec ref error=%d",
2140 error);
2141 iput(ea_inode);
2142 ea_inode = NULL;
2143 }
2144
2145 lock_buffer(new_bh);
2146 error = ext4_journal_get_create_access(handle, sb,
2147 new_bh, EXT4_JTR_NONE);
2148 if (error) {
2149 unlock_buffer(new_bh);
2150 error = -EIO;
2151 goto getblk_failed;
2152 }
2153 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2154 ext4_xattr_block_csum_set(inode, new_bh);
2155 set_buffer_uptodate(new_bh);
2156 unlock_buffer(new_bh);
2157 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2158 error = ext4_handle_dirty_metadata(handle, inode,
2159 new_bh);
2160 if (error)
2161 goto cleanup;
2162 }
2163 }
2164
2165 if (old_ea_inode_quota)
2166 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2167
2168 /* Update the inode. */
2169 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2170
2171 /* Drop the previous xattr block. */
2172 if (bs->bh && bs->bh != new_bh) {
2173 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2174
2175 ext4_xattr_release_block(handle, inode, bs->bh,
2176 &ea_inode_array,
2177 0 /* extra_credits */);
2178 ext4_xattr_inode_array_free(ea_inode_array);
2179 }
2180 error = 0;
2181
2182cleanup:
2183 if (ea_inode) {
2184 if (error) {
2185 int error2;
2186
2187 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2188 if (error2)
2189 ext4_warning_inode(ea_inode, "dec ref error=%d",
2190 error2);
2191 ext4_xattr_inode_free_quota(inode, ea_inode,
2192 i_size_read(ea_inode));
2193 }
2194 iput(ea_inode);
2195 }
2196 if (ce)
2197 mb_cache_entry_put(ea_block_cache, ce);
2198 brelse(new_bh);
2199 if (!(bs->bh && s->base == bs->bh->b_data))
2200 kfree(s->base);
2201
2202 return error;
2203
2204cleanup_dquot:
2205 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2206 goto cleanup;
2207
2208bad_block:
2209 EXT4_ERROR_INODE(inode, "bad block %llu",
2210 EXT4_I(inode)->i_file_acl);
2211 goto cleanup;
2212
2213#undef header
2214}
2215
2216int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2217 struct ext4_xattr_ibody_find *is)
2218{
2219 struct ext4_xattr_ibody_header *header;
2220 struct ext4_inode *raw_inode;
2221 int error;
2222
2223 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2224 return 0;
2225
2226 raw_inode = ext4_raw_inode(&is->iloc);
2227 header = IHDR(inode, raw_inode);
2228 is->s.base = is->s.first = IFIRST(header);
2229 is->s.here = is->s.first;
2230 is->s.end = ITAIL(inode, raw_inode);
2231 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2232 /* Find the named attribute. */
2233 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2234 i->name_index, i->name, 0);
2235 if (error && error != -ENODATA)
2236 return error;
2237 is->s.not_found = error;
2238 }
2239 return 0;
2240}
2241
2242int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2243 struct ext4_xattr_info *i,
2244 struct ext4_xattr_ibody_find *is)
2245{
2246 struct ext4_xattr_ibody_header *header;
2247 struct ext4_xattr_search *s = &is->s;
2248 struct inode *ea_inode = NULL;
2249 int error;
2250
2251 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2252 return -ENOSPC;
2253
2254 /* If we need EA inode, prepare it before locking the buffer */
2255 if (i->value && i->in_inode) {
2256 WARN_ON_ONCE(!i->value_len);
2257
2258 ea_inode = ext4_xattr_inode_lookup_create(handle, inode,
2259 i->value, i->value_len);
2260 if (IS_ERR(ea_inode))
2261 return PTR_ERR(ea_inode);
2262 }
2263 error = ext4_xattr_set_entry(i, s, handle, inode, ea_inode,
2264 false /* is_block */);
2265 if (error) {
2266 if (ea_inode) {
2267 int error2;
2268
2269 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2270 if (error2)
2271 ext4_warning_inode(ea_inode, "dec ref error=%d",
2272 error2);
2273
2274 ext4_xattr_inode_free_quota(inode, ea_inode,
2275 i_size_read(ea_inode));
2276 iput(ea_inode);
2277 }
2278 return error;
2279 }
2280 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2281 if (!IS_LAST_ENTRY(s->first)) {
2282 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2283 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2284 } else {
2285 header->h_magic = cpu_to_le32(0);
2286 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2287 }
2288 iput(ea_inode);
2289 return 0;
2290}
2291
2292static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2293 struct ext4_xattr_info *i)
2294{
2295 void *value;
2296
2297 /* When e_value_inum is set the value is stored externally. */
2298 if (s->here->e_value_inum)
2299 return 0;
2300 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2301 return 0;
2302 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2303 return !memcmp(value, i->value, i->value_len);
2304}
2305
2306static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2307{
2308 struct buffer_head *bh;
2309 int error;
2310
2311 if (!EXT4_I(inode)->i_file_acl)
2312 return NULL;
2313 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2314 if (IS_ERR(bh))
2315 return bh;
2316 error = ext4_xattr_check_block(inode, bh);
2317 if (error) {
2318 brelse(bh);
2319 return ERR_PTR(error);
2320 }
2321 return bh;
2322}
2323
2324/*
2325 * ext4_xattr_set_handle()
2326 *
2327 * Create, replace or remove an extended attribute for this inode. Value
2328 * is NULL to remove an existing extended attribute, and non-NULL to
2329 * either replace an existing extended attribute, or create a new extended
2330 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2331 * specify that an extended attribute must exist and must not exist
2332 * previous to the call, respectively.
2333 *
2334 * Returns 0, or a negative error number on failure.
2335 */
2336int
2337ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2338 const char *name, const void *value, size_t value_len,
2339 int flags)
2340{
2341 struct ext4_xattr_info i = {
2342 .name_index = name_index,
2343 .name = name,
2344 .value = value,
2345 .value_len = value_len,
2346 .in_inode = 0,
2347 };
2348 struct ext4_xattr_ibody_find is = {
2349 .s = { .not_found = -ENODATA, },
2350 };
2351 struct ext4_xattr_block_find bs = {
2352 .s = { .not_found = -ENODATA, },
2353 };
2354 int no_expand;
2355 int error;
2356
2357 if (!name)
2358 return -EINVAL;
2359 if (strlen(name) > 255)
2360 return -ERANGE;
2361
2362 ext4_write_lock_xattr(inode, &no_expand);
2363
2364 /* Check journal credits under write lock. */
2365 if (ext4_handle_valid(handle)) {
2366 struct buffer_head *bh;
2367 int credits;
2368
2369 bh = ext4_xattr_get_block(inode);
2370 if (IS_ERR(bh)) {
2371 error = PTR_ERR(bh);
2372 goto cleanup;
2373 }
2374
2375 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2376 value_len,
2377 flags & XATTR_CREATE);
2378 brelse(bh);
2379
2380 if (jbd2_handle_buffer_credits(handle) < credits) {
2381 error = -ENOSPC;
2382 goto cleanup;
2383 }
2384 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2385 }
2386
2387 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2388 if (error)
2389 goto cleanup;
2390
2391 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2392 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2393 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2394 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2395 }
2396
2397 error = ext4_xattr_ibody_find(inode, &i, &is);
2398 if (error)
2399 goto cleanup;
2400 if (is.s.not_found)
2401 error = ext4_xattr_block_find(inode, &i, &bs);
2402 if (error)
2403 goto cleanup;
2404 if (is.s.not_found && bs.s.not_found) {
2405 error = -ENODATA;
2406 if (flags & XATTR_REPLACE)
2407 goto cleanup;
2408 error = 0;
2409 if (!value)
2410 goto cleanup;
2411 } else {
2412 error = -EEXIST;
2413 if (flags & XATTR_CREATE)
2414 goto cleanup;
2415 }
2416
2417 if (!value) {
2418 if (!is.s.not_found)
2419 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2420 else if (!bs.s.not_found)
2421 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2422 } else {
2423 error = 0;
2424 /* Xattr value did not change? Save us some work and bail out */
2425 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2426 goto cleanup;
2427 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2428 goto cleanup;
2429
2430 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2431 (EXT4_XATTR_SIZE(i.value_len) >
2432 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2433 i.in_inode = 1;
2434retry_inode:
2435 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2436 if (!error && !bs.s.not_found) {
2437 i.value = NULL;
2438 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2439 } else if (error == -ENOSPC) {
2440 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2441 brelse(bs.bh);
2442 bs.bh = NULL;
2443 error = ext4_xattr_block_find(inode, &i, &bs);
2444 if (error)
2445 goto cleanup;
2446 }
2447 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2448 if (!error && !is.s.not_found) {
2449 i.value = NULL;
2450 error = ext4_xattr_ibody_set(handle, inode, &i,
2451 &is);
2452 } else if (error == -ENOSPC) {
2453 /*
2454 * Xattr does not fit in the block, store at
2455 * external inode if possible.
2456 */
2457 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2458 i.value_len && !i.in_inode) {
2459 i.in_inode = 1;
2460 goto retry_inode;
2461 }
2462 }
2463 }
2464 }
2465 if (!error) {
2466 ext4_xattr_update_super_block(handle, inode->i_sb);
2467 inode_set_ctime_current(inode);
2468 inode_inc_iversion(inode);
2469 if (!value)
2470 no_expand = 0;
2471 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2472 /*
2473 * The bh is consumed by ext4_mark_iloc_dirty, even with
2474 * error != 0.
2475 */
2476 is.iloc.bh = NULL;
2477 if (IS_SYNC(inode))
2478 ext4_handle_sync(handle);
2479 }
2480 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2481
2482cleanup:
2483 brelse(is.iloc.bh);
2484 brelse(bs.bh);
2485 ext4_write_unlock_xattr(inode, &no_expand);
2486 return error;
2487}
2488
2489int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2490 bool is_create, int *credits)
2491{
2492 struct buffer_head *bh;
2493 int err;
2494
2495 *credits = 0;
2496
2497 if (!EXT4_SB(inode->i_sb)->s_journal)
2498 return 0;
2499
2500 down_read(&EXT4_I(inode)->xattr_sem);
2501
2502 bh = ext4_xattr_get_block(inode);
2503 if (IS_ERR(bh)) {
2504 err = PTR_ERR(bh);
2505 } else {
2506 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2507 value_len, is_create);
2508 brelse(bh);
2509 err = 0;
2510 }
2511
2512 up_read(&EXT4_I(inode)->xattr_sem);
2513 return err;
2514}
2515
2516/*
2517 * ext4_xattr_set()
2518 *
2519 * Like ext4_xattr_set_handle, but start from an inode. This extended
2520 * attribute modification is a filesystem transaction by itself.
2521 *
2522 * Returns 0, or a negative error number on failure.
2523 */
2524int
2525ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2526 const void *value, size_t value_len, int flags)
2527{
2528 handle_t *handle;
2529 struct super_block *sb = inode->i_sb;
2530 int error, retries = 0;
2531 int credits;
2532
2533 error = dquot_initialize(inode);
2534 if (error)
2535 return error;
2536
2537retry:
2538 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2539 &credits);
2540 if (error)
2541 return error;
2542
2543 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2544 if (IS_ERR(handle)) {
2545 error = PTR_ERR(handle);
2546 } else {
2547 int error2;
2548
2549 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2550 value, value_len, flags);
2551 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR,
2552 handle);
2553 error2 = ext4_journal_stop(handle);
2554 if (error == -ENOSPC &&
2555 ext4_should_retry_alloc(sb, &retries))
2556 goto retry;
2557 if (error == 0)
2558 error = error2;
2559 }
2560
2561 return error;
2562}
2563
2564/*
2565 * Shift the EA entries in the inode to create space for the increased
2566 * i_extra_isize.
2567 */
2568static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2569 int value_offs_shift, void *to,
2570 void *from, size_t n)
2571{
2572 struct ext4_xattr_entry *last = entry;
2573 int new_offs;
2574
2575 /* We always shift xattr headers further thus offsets get lower */
2576 BUG_ON(value_offs_shift > 0);
2577
2578 /* Adjust the value offsets of the entries */
2579 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2580 if (!last->e_value_inum && last->e_value_size) {
2581 new_offs = le16_to_cpu(last->e_value_offs) +
2582 value_offs_shift;
2583 last->e_value_offs = cpu_to_le16(new_offs);
2584 }
2585 }
2586 /* Shift the entries by n bytes */
2587 memmove(to, from, n);
2588}
2589
2590/*
2591 * Move xattr pointed to by 'entry' from inode into external xattr block
2592 */
2593static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2594 struct ext4_inode *raw_inode,
2595 struct ext4_xattr_entry *entry)
2596{
2597 struct ext4_xattr_ibody_find *is = NULL;
2598 struct ext4_xattr_block_find *bs = NULL;
2599 char *buffer = NULL, *b_entry_name = NULL;
2600 size_t value_size = le32_to_cpu(entry->e_value_size);
2601 struct ext4_xattr_info i = {
2602 .value = NULL,
2603 .value_len = 0,
2604 .name_index = entry->e_name_index,
2605 .in_inode = !!entry->e_value_inum,
2606 };
2607 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2608 int needs_kvfree = 0;
2609 int error;
2610
2611 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2612 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2613 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2614 if (!is || !bs || !b_entry_name) {
2615 error = -ENOMEM;
2616 goto out;
2617 }
2618
2619 is->s.not_found = -ENODATA;
2620 bs->s.not_found = -ENODATA;
2621 is->iloc.bh = NULL;
2622 bs->bh = NULL;
2623
2624 /* Save the entry name and the entry value */
2625 if (entry->e_value_inum) {
2626 buffer = kvmalloc(value_size, GFP_NOFS);
2627 if (!buffer) {
2628 error = -ENOMEM;
2629 goto out;
2630 }
2631 needs_kvfree = 1;
2632 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2633 if (error)
2634 goto out;
2635 } else {
2636 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2637 buffer = (void *)IFIRST(header) + value_offs;
2638 }
2639
2640 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2641 b_entry_name[entry->e_name_len] = '\0';
2642 i.name = b_entry_name;
2643
2644 error = ext4_get_inode_loc(inode, &is->iloc);
2645 if (error)
2646 goto out;
2647
2648 error = ext4_xattr_ibody_find(inode, &i, is);
2649 if (error)
2650 goto out;
2651
2652 i.value = buffer;
2653 i.value_len = value_size;
2654 error = ext4_xattr_block_find(inode, &i, bs);
2655 if (error)
2656 goto out;
2657
2658 /* Move ea entry from the inode into the block */
2659 error = ext4_xattr_block_set(handle, inode, &i, bs);
2660 if (error)
2661 goto out;
2662
2663 /* Remove the chosen entry from the inode */
2664 i.value = NULL;
2665 i.value_len = 0;
2666 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2667
2668out:
2669 kfree(b_entry_name);
2670 if (needs_kvfree && buffer)
2671 kvfree(buffer);
2672 if (is)
2673 brelse(is->iloc.bh);
2674 if (bs)
2675 brelse(bs->bh);
2676 kfree(is);
2677 kfree(bs);
2678
2679 return error;
2680}
2681
2682static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2683 struct ext4_inode *raw_inode,
2684 int isize_diff, size_t ifree,
2685 size_t bfree, int *total_ino)
2686{
2687 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2688 struct ext4_xattr_entry *small_entry;
2689 struct ext4_xattr_entry *entry;
2690 struct ext4_xattr_entry *last;
2691 unsigned int entry_size; /* EA entry size */
2692 unsigned int total_size; /* EA entry size + value size */
2693 unsigned int min_total_size;
2694 int error;
2695
2696 while (isize_diff > ifree) {
2697 entry = NULL;
2698 small_entry = NULL;
2699 min_total_size = ~0U;
2700 last = IFIRST(header);
2701 /* Find the entry best suited to be pushed into EA block */
2702 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2703 /* never move system.data out of the inode */
2704 if ((last->e_name_len == 4) &&
2705 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2706 !memcmp(last->e_name, "data", 4))
2707 continue;
2708 total_size = EXT4_XATTR_LEN(last->e_name_len);
2709 if (!last->e_value_inum)
2710 total_size += EXT4_XATTR_SIZE(
2711 le32_to_cpu(last->e_value_size));
2712 if (total_size <= bfree &&
2713 total_size < min_total_size) {
2714 if (total_size + ifree < isize_diff) {
2715 small_entry = last;
2716 } else {
2717 entry = last;
2718 min_total_size = total_size;
2719 }
2720 }
2721 }
2722
2723 if (entry == NULL) {
2724 if (small_entry == NULL)
2725 return -ENOSPC;
2726 entry = small_entry;
2727 }
2728
2729 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2730 total_size = entry_size;
2731 if (!entry->e_value_inum)
2732 total_size += EXT4_XATTR_SIZE(
2733 le32_to_cpu(entry->e_value_size));
2734 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2735 entry);
2736 if (error)
2737 return error;
2738
2739 *total_ino -= entry_size;
2740 ifree += total_size;
2741 bfree -= total_size;
2742 }
2743
2744 return 0;
2745}
2746
2747/*
2748 * Expand an inode by new_extra_isize bytes when EAs are present.
2749 * Returns 0 on success or negative error number on failure.
2750 */
2751int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2752 struct ext4_inode *raw_inode, handle_t *handle)
2753{
2754 struct ext4_xattr_ibody_header *header;
2755 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2756 static unsigned int mnt_count;
2757 size_t min_offs;
2758 size_t ifree, bfree;
2759 int total_ino;
2760 void *base, *end;
2761 int error = 0, tried_min_extra_isize = 0;
2762 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2763 int isize_diff; /* How much do we need to grow i_extra_isize */
2764
2765retry:
2766 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2767 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2768 return 0;
2769
2770 header = IHDR(inode, raw_inode);
2771
2772 /*
2773 * Check if enough free space is available in the inode to shift the
2774 * entries ahead by new_extra_isize.
2775 */
2776
2777 base = IFIRST(header);
2778 end = ITAIL(inode, raw_inode);
2779 min_offs = end - base;
2780 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2781
2782 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2783 if (ifree >= isize_diff)
2784 goto shift;
2785
2786 /*
2787 * Enough free space isn't available in the inode, check if
2788 * EA block can hold new_extra_isize bytes.
2789 */
2790 if (EXT4_I(inode)->i_file_acl) {
2791 struct buffer_head *bh;
2792
2793 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2794 if (IS_ERR(bh)) {
2795 error = PTR_ERR(bh);
2796 goto cleanup;
2797 }
2798 error = ext4_xattr_check_block(inode, bh);
2799 if (error) {
2800 brelse(bh);
2801 goto cleanup;
2802 }
2803 base = BHDR(bh);
2804 end = bh->b_data + bh->b_size;
2805 min_offs = end - base;
2806 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2807 NULL);
2808 brelse(bh);
2809 if (bfree + ifree < isize_diff) {
2810 if (!tried_min_extra_isize && s_min_extra_isize) {
2811 tried_min_extra_isize++;
2812 new_extra_isize = s_min_extra_isize;
2813 goto retry;
2814 }
2815 error = -ENOSPC;
2816 goto cleanup;
2817 }
2818 } else {
2819 bfree = inode->i_sb->s_blocksize;
2820 }
2821
2822 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2823 isize_diff, ifree, bfree,
2824 &total_ino);
2825 if (error) {
2826 if (error == -ENOSPC && !tried_min_extra_isize &&
2827 s_min_extra_isize) {
2828 tried_min_extra_isize++;
2829 new_extra_isize = s_min_extra_isize;
2830 goto retry;
2831 }
2832 goto cleanup;
2833 }
2834shift:
2835 /* Adjust the offsets and shift the remaining entries ahead */
2836 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2837 - new_extra_isize, (void *)raw_inode +
2838 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2839 (void *)header, total_ino);
2840 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2841
2842 if (ext4_has_inline_data(inode))
2843 error = ext4_find_inline_data_nolock(inode);
2844
2845cleanup:
2846 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2847 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2848 inode->i_ino);
2849 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2850 }
2851 return error;
2852}
2853
2854#define EIA_INCR 16 /* must be 2^n */
2855#define EIA_MASK (EIA_INCR - 1)
2856
2857/* Add the large xattr @inode into @ea_inode_array for deferred iput().
2858 * If @ea_inode_array is new or full it will be grown and the old
2859 * contents copied over.
2860 */
2861static int
2862ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2863 struct inode *inode)
2864{
2865 if (*ea_inode_array == NULL) {
2866 /*
2867 * Start with 15 inodes, so it fits into a power-of-two size.
2868 */
2869 (*ea_inode_array) = kmalloc(
2870 struct_size(*ea_inode_array, inodes, EIA_MASK),
2871 GFP_NOFS);
2872 if (*ea_inode_array == NULL)
2873 return -ENOMEM;
2874 (*ea_inode_array)->count = 0;
2875 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2876 /* expand the array once all 15 + n * 16 slots are full */
2877 struct ext4_xattr_inode_array *new_array = NULL;
2878
2879 new_array = kmalloc(
2880 struct_size(*ea_inode_array, inodes,
2881 (*ea_inode_array)->count + EIA_INCR),
2882 GFP_NOFS);
2883 if (new_array == NULL)
2884 return -ENOMEM;
2885 memcpy(new_array, *ea_inode_array,
2886 struct_size(*ea_inode_array, inodes,
2887 (*ea_inode_array)->count));
2888 kfree(*ea_inode_array);
2889 *ea_inode_array = new_array;
2890 }
2891 (*ea_inode_array)->count++;
2892 (*ea_inode_array)->inodes[(*ea_inode_array)->count - 1] = inode;
2893 return 0;
2894}
2895
2896/*
2897 * ext4_xattr_delete_inode()
2898 *
2899 * Free extended attribute resources associated with this inode. Traverse
2900 * all entries and decrement reference on any xattr inodes associated with this
2901 * inode. This is called immediately before an inode is freed. We have exclusive
2902 * access to the inode. If an orphan inode is deleted it will also release its
2903 * references on xattr block and xattr inodes.
2904 */
2905int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2906 struct ext4_xattr_inode_array **ea_inode_array,
2907 int extra_credits)
2908{
2909 struct buffer_head *bh = NULL;
2910 struct ext4_xattr_ibody_header *header;
2911 struct ext4_iloc iloc = { .bh = NULL };
2912 struct ext4_xattr_entry *entry;
2913 struct inode *ea_inode;
2914 int error;
2915
2916 error = ext4_journal_ensure_credits(handle, extra_credits,
2917 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2918 if (error < 0) {
2919 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2920 goto cleanup;
2921 }
2922
2923 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2924 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2925
2926 error = ext4_get_inode_loc(inode, &iloc);
2927 if (error) {
2928 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2929 goto cleanup;
2930 }
2931
2932 error = ext4_journal_get_write_access(handle, inode->i_sb,
2933 iloc.bh, EXT4_JTR_NONE);
2934 if (error) {
2935 EXT4_ERROR_INODE(inode, "write access (error %d)",
2936 error);
2937 goto cleanup;
2938 }
2939
2940 header = IHDR(inode, ext4_raw_inode(&iloc));
2941 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2942 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2943 IFIRST(header),
2944 false /* block_csum */,
2945 ea_inode_array,
2946 extra_credits,
2947 false /* skip_quota */);
2948 }
2949
2950 if (EXT4_I(inode)->i_file_acl) {
2951 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2952 if (IS_ERR(bh)) {
2953 error = PTR_ERR(bh);
2954 if (error == -EIO) {
2955 EXT4_ERROR_INODE_ERR(inode, EIO,
2956 "block %llu read error",
2957 EXT4_I(inode)->i_file_acl);
2958 }
2959 bh = NULL;
2960 goto cleanup;
2961 }
2962 error = ext4_xattr_check_block(inode, bh);
2963 if (error)
2964 goto cleanup;
2965
2966 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2967 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2968 entry = EXT4_XATTR_NEXT(entry)) {
2969 if (!entry->e_value_inum)
2970 continue;
2971 error = ext4_xattr_inode_iget(inode,
2972 le32_to_cpu(entry->e_value_inum),
2973 le32_to_cpu(entry->e_hash),
2974 &ea_inode);
2975 if (error)
2976 continue;
2977 ext4_xattr_inode_free_quota(inode, ea_inode,
2978 le32_to_cpu(entry->e_value_size));
2979 iput(ea_inode);
2980 }
2981
2982 }
2983
2984 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2985 extra_credits);
2986 /*
2987 * Update i_file_acl value in the same transaction that releases
2988 * block.
2989 */
2990 EXT4_I(inode)->i_file_acl = 0;
2991 error = ext4_mark_inode_dirty(handle, inode);
2992 if (error) {
2993 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2994 error);
2995 goto cleanup;
2996 }
2997 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2998 }
2999 error = 0;
3000cleanup:
3001 brelse(iloc.bh);
3002 brelse(bh);
3003 return error;
3004}
3005
3006void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
3007{
3008 int idx;
3009
3010 if (ea_inode_array == NULL)
3011 return;
3012
3013 for (idx = 0; idx < ea_inode_array->count; ++idx)
3014 iput(ea_inode_array->inodes[idx]);
3015 kfree(ea_inode_array);
3016}
3017
3018/*
3019 * ext4_xattr_block_cache_insert()
3020 *
3021 * Create a new entry in the extended attribute block cache, and insert
3022 * it unless such an entry is already in the cache.
3023 */
3024static void
3025ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
3026 struct buffer_head *bh)
3027{
3028 struct ext4_xattr_header *header = BHDR(bh);
3029 __u32 hash = le32_to_cpu(header->h_hash);
3030 int reusable = le32_to_cpu(header->h_refcount) <
3031 EXT4_XATTR_REFCOUNT_MAX;
3032 int error;
3033
3034 if (!ea_block_cache)
3035 return;
3036 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3037 bh->b_blocknr, reusable);
3038 if (error) {
3039 if (error == -EBUSY)
3040 ea_bdebug(bh, "already in cache");
3041 } else
3042 ea_bdebug(bh, "inserting [%x]", (int)hash);
3043}
3044
3045/*
3046 * ext4_xattr_cmp()
3047 *
3048 * Compare two extended attribute blocks for equality.
3049 *
3050 * Returns 0 if the blocks are equal, 1 if they differ.
3051 */
3052static int
3053ext4_xattr_cmp(struct ext4_xattr_header *header1,
3054 struct ext4_xattr_header *header2)
3055{
3056 struct ext4_xattr_entry *entry1, *entry2;
3057
3058 entry1 = ENTRY(header1+1);
3059 entry2 = ENTRY(header2+1);
3060 while (!IS_LAST_ENTRY(entry1)) {
3061 if (IS_LAST_ENTRY(entry2))
3062 return 1;
3063 if (entry1->e_hash != entry2->e_hash ||
3064 entry1->e_name_index != entry2->e_name_index ||
3065 entry1->e_name_len != entry2->e_name_len ||
3066 entry1->e_value_size != entry2->e_value_size ||
3067 entry1->e_value_inum != entry2->e_value_inum ||
3068 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3069 return 1;
3070 if (!entry1->e_value_inum &&
3071 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3072 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3073 le32_to_cpu(entry1->e_value_size)))
3074 return 1;
3075
3076 entry1 = EXT4_XATTR_NEXT(entry1);
3077 entry2 = EXT4_XATTR_NEXT(entry2);
3078 }
3079 if (!IS_LAST_ENTRY(entry2))
3080 return 1;
3081 return 0;
3082}
3083
3084/*
3085 * ext4_xattr_block_cache_find()
3086 *
3087 * Find an identical extended attribute block.
3088 *
3089 * Returns a pointer to the block found, or NULL if such a block was not
3090 * found, or an error pointer if an error occurred while reading ea block.
3091 */
3092static struct buffer_head *
3093ext4_xattr_block_cache_find(struct inode *inode,
3094 struct ext4_xattr_header *header,
3095 struct mb_cache_entry **pce)
3096{
3097 __u32 hash = le32_to_cpu(header->h_hash);
3098 struct mb_cache_entry *ce;
3099 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3100
3101 if (!ea_block_cache)
3102 return NULL;
3103 if (!header->h_hash)
3104 return NULL; /* never share */
3105 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3106 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3107 while (ce) {
3108 struct buffer_head *bh;
3109
3110 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3111 if (IS_ERR(bh)) {
3112 if (PTR_ERR(bh) != -ENOMEM)
3113 EXT4_ERROR_INODE(inode, "block %lu read error",
3114 (unsigned long)ce->e_value);
3115 mb_cache_entry_put(ea_block_cache, ce);
3116 return bh;
3117 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3118 *pce = ce;
3119 return bh;
3120 }
3121 brelse(bh);
3122 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3123 }
3124 return NULL;
3125}
3126
3127#define NAME_HASH_SHIFT 5
3128#define VALUE_HASH_SHIFT 16
3129
3130/*
3131 * ext4_xattr_hash_entry()
3132 *
3133 * Compute the hash of an extended attribute.
3134 */
3135static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3136 size_t value_count)
3137{
3138 __u32 hash = 0;
3139
3140 while (name_len--) {
3141 hash = (hash << NAME_HASH_SHIFT) ^
3142 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3143 (unsigned char)*name++;
3144 }
3145 while (value_count--) {
3146 hash = (hash << VALUE_HASH_SHIFT) ^
3147 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3148 le32_to_cpu(*value++);
3149 }
3150 return cpu_to_le32(hash);
3151}
3152
3153/*
3154 * ext4_xattr_hash_entry_signed()
3155 *
3156 * Compute the hash of an extended attribute incorrectly.
3157 */
3158static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3159{
3160 __u32 hash = 0;
3161
3162 while (name_len--) {
3163 hash = (hash << NAME_HASH_SHIFT) ^
3164 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3165 (signed char)*name++;
3166 }
3167 while (value_count--) {
3168 hash = (hash << VALUE_HASH_SHIFT) ^
3169 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3170 le32_to_cpu(*value++);
3171 }
3172 return cpu_to_le32(hash);
3173}
3174
3175#undef NAME_HASH_SHIFT
3176#undef VALUE_HASH_SHIFT
3177
3178#define BLOCK_HASH_SHIFT 16
3179
3180/*
3181 * ext4_xattr_rehash()
3182 *
3183 * Re-compute the extended attribute hash value after an entry has changed.
3184 */
3185static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3186{
3187 struct ext4_xattr_entry *here;
3188 __u32 hash = 0;
3189
3190 here = ENTRY(header+1);
3191 while (!IS_LAST_ENTRY(here)) {
3192 if (!here->e_hash) {
3193 /* Block is not shared if an entry's hash value == 0 */
3194 hash = 0;
3195 break;
3196 }
3197 hash = (hash << BLOCK_HASH_SHIFT) ^
3198 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3199 le32_to_cpu(here->e_hash);
3200 here = EXT4_XATTR_NEXT(here);
3201 }
3202 header->h_hash = cpu_to_le32(hash);
3203}
3204
3205#undef BLOCK_HASH_SHIFT
3206
3207#define HASH_BUCKET_BITS 10
3208
3209struct mb_cache *
3210ext4_xattr_create_cache(void)
3211{
3212 return mb_cache_create(HASH_BUCKET_BITS);
3213}
3214
3215void ext4_xattr_destroy_cache(struct mb_cache *cache)
3216{
3217 if (cache)
3218 mb_cache_destroy(cache);
3219}
3220
Linux 6.x block layer and io_uring tree(s)