1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/xattr.c 4 * 5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de> 6 * 7 * Fix by Harrison Xing <harrison@mountainviewdata.com>. 8 * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>. 9 * Extended attributes for symlinks and special files added per 10 * suggestion of Luka Renko <luka.renko@hermes.si>. 11 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>, 12 * Red Hat Inc. 13 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz 14 * and Andreas Gruenbacher <agruen@suse.de>. 15 */ 16 17 /* 18 * Extended attributes are stored directly in inodes (on file systems with 19 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl 20 * field contains the block number if an inode uses an additional block. All 21 * attributes must fit in the inode and one additional block. Blocks that 22 * contain the identical set of attributes may be shared among several inodes. 23 * Identical blocks are detected by keeping a cache of blocks that have 24 * recently been accessed. 25 * 26 * The attributes in inodes and on blocks have a different header; the entries 27 * are stored in the same format: 28 * 29 * +------------------+ 30 * | header | 31 * | entry 1 | | 32 * | entry 2 | | growing downwards 33 * | entry 3 | v 34 * | four null bytes | 35 * | . . . | 36 * | value 1 | ^ 37 * | value 3 | | growing upwards 38 * | value 2 | | 39 * +------------------+ 40 * 41 * The header is followed by multiple entry descriptors. In disk blocks, the 42 * entry descriptors are kept sorted. In inodes, they are unsorted. The 43 * attribute values are aligned to the end of the block in no specific order. 44 * 45 * Locking strategy 46 * ---------------- 47 * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem. 48 * EA blocks are only changed if they are exclusive to an inode, so 49 * holding xattr_sem also means that nothing but the EA block's reference 50 * count can change. Multiple writers to the same block are synchronized 51 * by the buffer lock. 52 */ 53 54 #include <linux/init.h> 55 #include <linux/fs.h> 56 #include <linux/slab.h> 57 #include <linux/mbcache.h> 58 #include <linux/quotaops.h> 59 #include <linux/iversion.h> 60 #include "ext4_jbd2.h" 61 #include "ext4.h" 62 #include "xattr.h" 63 #include "acl.h" 64 65 #ifdef EXT4_XATTR_DEBUG 66 # define ea_idebug(inode, fmt, ...) \ 67 printk(KERN_DEBUG "inode %s:%lu: " fmt "\n", \ 68 inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__) 69 # define ea_bdebug(bh, fmt, ...) \ 70 printk(KERN_DEBUG "block %pg:%lu: " fmt "\n", \ 71 bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__) 72 #else 73 # define ea_idebug(inode, fmt, ...) no_printk(fmt, ##__VA_ARGS__) 74 # define ea_bdebug(bh, fmt, ...) no_printk(fmt, ##__VA_ARGS__) 75 #endif 76 77 static void ext4_xattr_block_cache_insert(struct mb_cache *, 78 struct buffer_head *); 79 static struct buffer_head * 80 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *, 81 struct mb_cache_entry **); 82 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value, 83 size_t value_count); 84 static void ext4_xattr_rehash(struct ext4_xattr_header *); 85 86 static const struct xattr_handler * const ext4_xattr_handler_map[] = { 87 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler, 88 #ifdef CONFIG_EXT4_FS_POSIX_ACL 89 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler, 90 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler, 91 #endif 92 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler, 93 #ifdef CONFIG_EXT4_FS_SECURITY 94 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler, 95 #endif 96 [EXT4_XATTR_INDEX_HURD] = &ext4_xattr_hurd_handler, 97 }; 98 99 const struct xattr_handler *ext4_xattr_handlers[] = { 100 &ext4_xattr_user_handler, 101 &ext4_xattr_trusted_handler, 102 #ifdef CONFIG_EXT4_FS_POSIX_ACL 103 &posix_acl_access_xattr_handler, 104 &posix_acl_default_xattr_handler, 105 #endif 106 #ifdef CONFIG_EXT4_FS_SECURITY 107 &ext4_xattr_security_handler, 108 #endif 109 &ext4_xattr_hurd_handler, 110 NULL 111 }; 112 113 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \ 114 inode->i_sb->s_fs_info)->s_ea_block_cache) 115 116 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \ 117 inode->i_sb->s_fs_info)->s_ea_inode_cache) 118 119 static int 120 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array, 121 struct inode *inode); 122 123 #ifdef CONFIG_LOCKDEP 124 void ext4_xattr_inode_set_class(struct inode *ea_inode) 125 { 126 lockdep_set_subclass(&ea_inode->i_rwsem, 1); 127 } 128 #endif 129 130 static __le32 ext4_xattr_block_csum(struct inode *inode, 131 sector_t block_nr, 132 struct ext4_xattr_header *hdr) 133 { 134 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 135 __u32 csum; 136 __le64 dsk_block_nr = cpu_to_le64(block_nr); 137 __u32 dummy_csum = 0; 138 int offset = offsetof(struct ext4_xattr_header, h_checksum); 139 140 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr, 141 sizeof(dsk_block_nr)); 142 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset); 143 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum)); 144 offset += sizeof(dummy_csum); 145 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset, 146 EXT4_BLOCK_SIZE(inode->i_sb) - offset); 147 148 return cpu_to_le32(csum); 149 } 150 151 static int ext4_xattr_block_csum_verify(struct inode *inode, 152 struct buffer_head *bh) 153 { 154 struct ext4_xattr_header *hdr = BHDR(bh); 155 int ret = 1; 156 157 if (ext4_has_metadata_csum(inode->i_sb)) { 158 lock_buffer(bh); 159 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode, 160 bh->b_blocknr, hdr)); 161 unlock_buffer(bh); 162 } 163 return ret; 164 } 165 166 static void ext4_xattr_block_csum_set(struct inode *inode, 167 struct buffer_head *bh) 168 { 169 if (ext4_has_metadata_csum(inode->i_sb)) 170 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode, 171 bh->b_blocknr, BHDR(bh)); 172 } 173 174 static inline const struct xattr_handler * 175 ext4_xattr_handler(int name_index) 176 { 177 const struct xattr_handler *handler = NULL; 178 179 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map)) 180 handler = ext4_xattr_handler_map[name_index]; 181 return handler; 182 } 183 184 static int 185 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end, 186 void *value_start) 187 { 188 struct ext4_xattr_entry *e = entry; 189 190 /* Find the end of the names list */ 191 while (!IS_LAST_ENTRY(e)) { 192 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e); 193 if ((void *)next >= end) 194 return -EFSCORRUPTED; 195 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len) 196 return -EFSCORRUPTED; 197 e = next; 198 } 199 200 /* Check the values */ 201 while (!IS_LAST_ENTRY(entry)) { 202 u32 size = le32_to_cpu(entry->e_value_size); 203 204 if (size > EXT4_XATTR_SIZE_MAX) 205 return -EFSCORRUPTED; 206 207 if (size != 0 && entry->e_value_inum == 0) { 208 u16 offs = le16_to_cpu(entry->e_value_offs); 209 void *value; 210 211 /* 212 * The value cannot overlap the names, and the value 213 * with padding cannot extend beyond 'end'. Check both 214 * the padded and unpadded sizes, since the size may 215 * overflow to 0 when adding padding. 216 */ 217 if (offs > end - value_start) 218 return -EFSCORRUPTED; 219 value = value_start + offs; 220 if (value < (void *)e + sizeof(u32) || 221 size > end - value || 222 EXT4_XATTR_SIZE(size) > end - value) 223 return -EFSCORRUPTED; 224 } 225 entry = EXT4_XATTR_NEXT(entry); 226 } 227 228 return 0; 229 } 230 231 static inline int 232 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh, 233 const char *function, unsigned int line) 234 { 235 int error = -EFSCORRUPTED; 236 237 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) || 238 BHDR(bh)->h_blocks != cpu_to_le32(1)) 239 goto errout; 240 if (buffer_verified(bh)) 241 return 0; 242 243 error = -EFSBADCRC; 244 if (!ext4_xattr_block_csum_verify(inode, bh)) 245 goto errout; 246 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size, 247 bh->b_data); 248 errout: 249 if (error) 250 __ext4_error_inode(inode, function, line, 0, -error, 251 "corrupted xattr block %llu", 252 (unsigned long long) bh->b_blocknr); 253 else 254 set_buffer_verified(bh); 255 return error; 256 } 257 258 #define ext4_xattr_check_block(inode, bh) \ 259 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__) 260 261 262 static int 263 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header, 264 void *end, const char *function, unsigned int line) 265 { 266 int error = -EFSCORRUPTED; 267 268 if (end - (void *)header < sizeof(*header) + sizeof(u32) || 269 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC))) 270 goto errout; 271 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header)); 272 errout: 273 if (error) 274 __ext4_error_inode(inode, function, line, 0, -error, 275 "corrupted in-inode xattr"); 276 return error; 277 } 278 279 #define xattr_check_inode(inode, header, end) \ 280 __xattr_check_inode((inode), (header), (end), __func__, __LINE__) 281 282 static int 283 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry, 284 void *end, int name_index, const char *name, int sorted) 285 { 286 struct ext4_xattr_entry *entry, *next; 287 size_t name_len; 288 int cmp = 1; 289 290 if (name == NULL) 291 return -EINVAL; 292 name_len = strlen(name); 293 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) { 294 next = EXT4_XATTR_NEXT(entry); 295 if ((void *) next >= end) { 296 EXT4_ERROR_INODE(inode, "corrupted xattr entries"); 297 return -EFSCORRUPTED; 298 } 299 cmp = name_index - entry->e_name_index; 300 if (!cmp) 301 cmp = name_len - entry->e_name_len; 302 if (!cmp) 303 cmp = memcmp(name, entry->e_name, name_len); 304 if (cmp <= 0 && (sorted || cmp == 0)) 305 break; 306 } 307 *pentry = entry; 308 return cmp ? -ENODATA : 0; 309 } 310 311 static u32 312 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size) 313 { 314 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size); 315 } 316 317 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode) 318 { 319 return ((u64)ea_inode->i_ctime.tv_sec << 32) | 320 (u32) inode_peek_iversion_raw(ea_inode); 321 } 322 323 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count) 324 { 325 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32); 326 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff); 327 } 328 329 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode) 330 { 331 return (u32)ea_inode->i_atime.tv_sec; 332 } 333 334 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash) 335 { 336 ea_inode->i_atime.tv_sec = hash; 337 } 338 339 /* 340 * Read the EA value from an inode. 341 */ 342 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size) 343 { 344 int blocksize = 1 << ea_inode->i_blkbits; 345 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits; 346 int tail_size = (size % blocksize) ?: blocksize; 347 struct buffer_head *bhs_inline[8]; 348 struct buffer_head **bhs = bhs_inline; 349 int i, ret; 350 351 if (bh_count > ARRAY_SIZE(bhs_inline)) { 352 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS); 353 if (!bhs) 354 return -ENOMEM; 355 } 356 357 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count, 358 true /* wait */, bhs); 359 if (ret) 360 goto free_bhs; 361 362 for (i = 0; i < bh_count; i++) { 363 /* There shouldn't be any holes in ea_inode. */ 364 if (!bhs[i]) { 365 ret = -EFSCORRUPTED; 366 goto put_bhs; 367 } 368 memcpy((char *)buf + blocksize * i, bhs[i]->b_data, 369 i < bh_count - 1 ? blocksize : tail_size); 370 } 371 ret = 0; 372 put_bhs: 373 for (i = 0; i < bh_count; i++) 374 brelse(bhs[i]); 375 free_bhs: 376 if (bhs != bhs_inline) 377 kfree(bhs); 378 return ret; 379 } 380 381 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec) 382 383 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino, 384 u32 ea_inode_hash, struct inode **ea_inode) 385 { 386 struct inode *inode; 387 int err; 388 389 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL); 390 if (IS_ERR(inode)) { 391 err = PTR_ERR(inode); 392 ext4_error(parent->i_sb, 393 "error while reading EA inode %lu err=%d", ea_ino, 394 err); 395 return err; 396 } 397 398 if (is_bad_inode(inode)) { 399 ext4_error(parent->i_sb, 400 "error while reading EA inode %lu is_bad_inode", 401 ea_ino); 402 err = -EIO; 403 goto error; 404 } 405 406 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { 407 ext4_error(parent->i_sb, 408 "EA inode %lu does not have EXT4_EA_INODE_FL flag", 409 ea_ino); 410 err = -EINVAL; 411 goto error; 412 } 413 414 ext4_xattr_inode_set_class(inode); 415 416 /* 417 * Check whether this is an old Lustre-style xattr inode. Lustre 418 * implementation does not have hash validation, rather it has a 419 * backpointer from ea_inode to the parent inode. 420 */ 421 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) && 422 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino && 423 inode->i_generation == parent->i_generation) { 424 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE); 425 ext4_xattr_inode_set_ref(inode, 1); 426 } else { 427 inode_lock(inode); 428 inode->i_flags |= S_NOQUOTA; 429 inode_unlock(inode); 430 } 431 432 *ea_inode = inode; 433 return 0; 434 error: 435 iput(inode); 436 return err; 437 } 438 439 static int 440 ext4_xattr_inode_verify_hashes(struct inode *ea_inode, 441 struct ext4_xattr_entry *entry, void *buffer, 442 size_t size) 443 { 444 u32 hash; 445 446 /* Verify stored hash matches calculated hash. */ 447 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size); 448 if (hash != ext4_xattr_inode_get_hash(ea_inode)) 449 return -EFSCORRUPTED; 450 451 if (entry) { 452 __le32 e_hash, tmp_data; 453 454 /* Verify entry hash. */ 455 tmp_data = cpu_to_le32(hash); 456 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len, 457 &tmp_data, 1); 458 if (e_hash != entry->e_hash) 459 return -EFSCORRUPTED; 460 } 461 return 0; 462 } 463 464 /* 465 * Read xattr value from the EA inode. 466 */ 467 static int 468 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry, 469 void *buffer, size_t size) 470 { 471 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode); 472 struct inode *ea_inode; 473 int err; 474 475 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum), 476 le32_to_cpu(entry->e_hash), &ea_inode); 477 if (err) { 478 ea_inode = NULL; 479 goto out; 480 } 481 482 if (i_size_read(ea_inode) != size) { 483 ext4_warning_inode(ea_inode, 484 "ea_inode file size=%llu entry size=%zu", 485 i_size_read(ea_inode), size); 486 err = -EFSCORRUPTED; 487 goto out; 488 } 489 490 err = ext4_xattr_inode_read(ea_inode, buffer, size); 491 if (err) 492 goto out; 493 494 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) { 495 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer, 496 size); 497 if (err) { 498 ext4_warning_inode(ea_inode, 499 "EA inode hash validation failed"); 500 goto out; 501 } 502 503 if (ea_inode_cache) 504 mb_cache_entry_create(ea_inode_cache, GFP_NOFS, 505 ext4_xattr_inode_get_hash(ea_inode), 506 ea_inode->i_ino, true /* reusable */); 507 } 508 out: 509 iput(ea_inode); 510 return err; 511 } 512 513 static int 514 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name, 515 void *buffer, size_t buffer_size) 516 { 517 struct buffer_head *bh = NULL; 518 struct ext4_xattr_entry *entry; 519 size_t size; 520 void *end; 521 int error; 522 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 523 524 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld", 525 name_index, name, buffer, (long)buffer_size); 526 527 if (!EXT4_I(inode)->i_file_acl) 528 return -ENODATA; 529 ea_idebug(inode, "reading block %llu", 530 (unsigned long long)EXT4_I(inode)->i_file_acl); 531 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 532 if (IS_ERR(bh)) 533 return PTR_ERR(bh); 534 ea_bdebug(bh, "b_count=%d, refcount=%d", 535 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount)); 536 error = ext4_xattr_check_block(inode, bh); 537 if (error) 538 goto cleanup; 539 ext4_xattr_block_cache_insert(ea_block_cache, bh); 540 entry = BFIRST(bh); 541 end = bh->b_data + bh->b_size; 542 error = xattr_find_entry(inode, &entry, end, name_index, name, 1); 543 if (error) 544 goto cleanup; 545 size = le32_to_cpu(entry->e_value_size); 546 error = -ERANGE; 547 if (unlikely(size > EXT4_XATTR_SIZE_MAX)) 548 goto cleanup; 549 if (buffer) { 550 if (size > buffer_size) 551 goto cleanup; 552 if (entry->e_value_inum) { 553 error = ext4_xattr_inode_get(inode, entry, buffer, 554 size); 555 if (error) 556 goto cleanup; 557 } else { 558 u16 offset = le16_to_cpu(entry->e_value_offs); 559 void *p = bh->b_data + offset; 560 561 if (unlikely(p + size > end)) 562 goto cleanup; 563 memcpy(buffer, p, size); 564 } 565 } 566 error = size; 567 568 cleanup: 569 brelse(bh); 570 return error; 571 } 572 573 int 574 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name, 575 void *buffer, size_t buffer_size) 576 { 577 struct ext4_xattr_ibody_header *header; 578 struct ext4_xattr_entry *entry; 579 struct ext4_inode *raw_inode; 580 struct ext4_iloc iloc; 581 size_t size; 582 void *end; 583 int error; 584 585 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR)) 586 return -ENODATA; 587 error = ext4_get_inode_loc(inode, &iloc); 588 if (error) 589 return error; 590 raw_inode = ext4_raw_inode(&iloc); 591 header = IHDR(inode, raw_inode); 592 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 593 error = xattr_check_inode(inode, header, end); 594 if (error) 595 goto cleanup; 596 entry = IFIRST(header); 597 error = xattr_find_entry(inode, &entry, end, name_index, name, 0); 598 if (error) 599 goto cleanup; 600 size = le32_to_cpu(entry->e_value_size); 601 error = -ERANGE; 602 if (unlikely(size > EXT4_XATTR_SIZE_MAX)) 603 goto cleanup; 604 if (buffer) { 605 if (size > buffer_size) 606 goto cleanup; 607 if (entry->e_value_inum) { 608 error = ext4_xattr_inode_get(inode, entry, buffer, 609 size); 610 if (error) 611 goto cleanup; 612 } else { 613 u16 offset = le16_to_cpu(entry->e_value_offs); 614 void *p = (void *)IFIRST(header) + offset; 615 616 if (unlikely(p + size > end)) 617 goto cleanup; 618 memcpy(buffer, p, size); 619 } 620 } 621 error = size; 622 623 cleanup: 624 brelse(iloc.bh); 625 return error; 626 } 627 628 /* 629 * ext4_xattr_get() 630 * 631 * Copy an extended attribute into the buffer 632 * provided, or compute the buffer size required. 633 * Buffer is NULL to compute the size of the buffer required. 634 * 635 * Returns a negative error number on failure, or the number of bytes 636 * used / required on success. 637 */ 638 int 639 ext4_xattr_get(struct inode *inode, int name_index, const char *name, 640 void *buffer, size_t buffer_size) 641 { 642 int error; 643 644 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 645 return -EIO; 646 647 if (strlen(name) > 255) 648 return -ERANGE; 649 650 down_read(&EXT4_I(inode)->xattr_sem); 651 error = ext4_xattr_ibody_get(inode, name_index, name, buffer, 652 buffer_size); 653 if (error == -ENODATA) 654 error = ext4_xattr_block_get(inode, name_index, name, buffer, 655 buffer_size); 656 up_read(&EXT4_I(inode)->xattr_sem); 657 return error; 658 } 659 660 static int 661 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry, 662 char *buffer, size_t buffer_size) 663 { 664 size_t rest = buffer_size; 665 666 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) { 667 const struct xattr_handler *handler = 668 ext4_xattr_handler(entry->e_name_index); 669 670 if (handler && (!handler->list || handler->list(dentry))) { 671 const char *prefix = handler->prefix ?: handler->name; 672 size_t prefix_len = strlen(prefix); 673 size_t size = prefix_len + entry->e_name_len + 1; 674 675 if (buffer) { 676 if (size > rest) 677 return -ERANGE; 678 memcpy(buffer, prefix, prefix_len); 679 buffer += prefix_len; 680 memcpy(buffer, entry->e_name, entry->e_name_len); 681 buffer += entry->e_name_len; 682 *buffer++ = 0; 683 } 684 rest -= size; 685 } 686 } 687 return buffer_size - rest; /* total size */ 688 } 689 690 static int 691 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size) 692 { 693 struct inode *inode = d_inode(dentry); 694 struct buffer_head *bh = NULL; 695 int error; 696 697 ea_idebug(inode, "buffer=%p, buffer_size=%ld", 698 buffer, (long)buffer_size); 699 700 if (!EXT4_I(inode)->i_file_acl) 701 return 0; 702 ea_idebug(inode, "reading block %llu", 703 (unsigned long long)EXT4_I(inode)->i_file_acl); 704 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 705 if (IS_ERR(bh)) 706 return PTR_ERR(bh); 707 ea_bdebug(bh, "b_count=%d, refcount=%d", 708 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount)); 709 error = ext4_xattr_check_block(inode, bh); 710 if (error) 711 goto cleanup; 712 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh); 713 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, 714 buffer_size); 715 cleanup: 716 brelse(bh); 717 return error; 718 } 719 720 static int 721 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size) 722 { 723 struct inode *inode = d_inode(dentry); 724 struct ext4_xattr_ibody_header *header; 725 struct ext4_inode *raw_inode; 726 struct ext4_iloc iloc; 727 void *end; 728 int error; 729 730 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR)) 731 return 0; 732 error = ext4_get_inode_loc(inode, &iloc); 733 if (error) 734 return error; 735 raw_inode = ext4_raw_inode(&iloc); 736 header = IHDR(inode, raw_inode); 737 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 738 error = xattr_check_inode(inode, header, end); 739 if (error) 740 goto cleanup; 741 error = ext4_xattr_list_entries(dentry, IFIRST(header), 742 buffer, buffer_size); 743 744 cleanup: 745 brelse(iloc.bh); 746 return error; 747 } 748 749 /* 750 * Inode operation listxattr() 751 * 752 * d_inode(dentry)->i_rwsem: don't care 753 * 754 * Copy a list of attribute names into the buffer 755 * provided, or compute the buffer size required. 756 * Buffer is NULL to compute the size of the buffer required. 757 * 758 * Returns a negative error number on failure, or the number of bytes 759 * used / required on success. 760 */ 761 ssize_t 762 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size) 763 { 764 int ret, ret2; 765 766 down_read(&EXT4_I(d_inode(dentry))->xattr_sem); 767 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size); 768 if (ret < 0) 769 goto errout; 770 if (buffer) { 771 buffer += ret; 772 buffer_size -= ret; 773 } 774 ret = ext4_xattr_block_list(dentry, buffer, buffer_size); 775 if (ret < 0) 776 goto errout; 777 ret += ret2; 778 errout: 779 up_read(&EXT4_I(d_inode(dentry))->xattr_sem); 780 return ret; 781 } 782 783 /* 784 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is 785 * not set, set it. 786 */ 787 static void ext4_xattr_update_super_block(handle_t *handle, 788 struct super_block *sb) 789 { 790 if (ext4_has_feature_xattr(sb)) 791 return; 792 793 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); 794 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) { 795 lock_buffer(EXT4_SB(sb)->s_sbh); 796 ext4_set_feature_xattr(sb); 797 ext4_superblock_csum_set(sb); 798 unlock_buffer(EXT4_SB(sb)->s_sbh); 799 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); 800 } 801 } 802 803 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage) 804 { 805 struct ext4_iloc iloc = { .bh = NULL }; 806 struct buffer_head *bh = NULL; 807 struct ext4_inode *raw_inode; 808 struct ext4_xattr_ibody_header *header; 809 struct ext4_xattr_entry *entry; 810 qsize_t ea_inode_refs = 0; 811 void *end; 812 int ret; 813 814 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem); 815 816 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { 817 ret = ext4_get_inode_loc(inode, &iloc); 818 if (ret) 819 goto out; 820 raw_inode = ext4_raw_inode(&iloc); 821 header = IHDR(inode, raw_inode); 822 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 823 ret = xattr_check_inode(inode, header, end); 824 if (ret) 825 goto out; 826 827 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry); 828 entry = EXT4_XATTR_NEXT(entry)) 829 if (entry->e_value_inum) 830 ea_inode_refs++; 831 } 832 833 if (EXT4_I(inode)->i_file_acl) { 834 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 835 if (IS_ERR(bh)) { 836 ret = PTR_ERR(bh); 837 bh = NULL; 838 goto out; 839 } 840 841 ret = ext4_xattr_check_block(inode, bh); 842 if (ret) 843 goto out; 844 845 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry); 846 entry = EXT4_XATTR_NEXT(entry)) 847 if (entry->e_value_inum) 848 ea_inode_refs++; 849 } 850 *usage = ea_inode_refs + 1; 851 ret = 0; 852 out: 853 brelse(iloc.bh); 854 brelse(bh); 855 return ret; 856 } 857 858 static inline size_t round_up_cluster(struct inode *inode, size_t length) 859 { 860 struct super_block *sb = inode->i_sb; 861 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits + 862 inode->i_blkbits); 863 size_t mask = ~(cluster_size - 1); 864 865 return (length + cluster_size - 1) & mask; 866 } 867 868 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len) 869 { 870 int err; 871 872 err = dquot_alloc_inode(inode); 873 if (err) 874 return err; 875 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len)); 876 if (err) 877 dquot_free_inode(inode); 878 return err; 879 } 880 881 static void ext4_xattr_inode_free_quota(struct inode *parent, 882 struct inode *ea_inode, 883 size_t len) 884 { 885 if (ea_inode && 886 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) 887 return; 888 dquot_free_space_nodirty(parent, round_up_cluster(parent, len)); 889 dquot_free_inode(parent); 890 } 891 892 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode, 893 struct buffer_head *block_bh, size_t value_len, 894 bool is_create) 895 { 896 int credits; 897 int blocks; 898 899 /* 900 * 1) Owner inode update 901 * 2) Ref count update on old xattr block 902 * 3) new xattr block 903 * 4) block bitmap update for new xattr block 904 * 5) group descriptor for new xattr block 905 * 6) block bitmap update for old xattr block 906 * 7) group descriptor for old block 907 * 908 * 6 & 7 can happen if we have two racing threads T_a and T_b 909 * which are each trying to set an xattr on inodes I_a and I_b 910 * which were both initially sharing an xattr block. 911 */ 912 credits = 7; 913 914 /* Quota updates. */ 915 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb); 916 917 /* 918 * In case of inline data, we may push out the data to a block, 919 * so we need to reserve credits for this eventuality 920 */ 921 if (inode && ext4_has_inline_data(inode)) 922 credits += ext4_writepage_trans_blocks(inode) + 1; 923 924 /* We are done if ea_inode feature is not enabled. */ 925 if (!ext4_has_feature_ea_inode(sb)) 926 return credits; 927 928 /* New ea_inode, inode map, block bitmap, group descriptor. */ 929 credits += 4; 930 931 /* Data blocks. */ 932 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits; 933 934 /* Indirection block or one level of extent tree. */ 935 blocks += 1; 936 937 /* Block bitmap and group descriptor updates for each block. */ 938 credits += blocks * 2; 939 940 /* Blocks themselves. */ 941 credits += blocks; 942 943 if (!is_create) { 944 /* Dereference ea_inode holding old xattr value. 945 * Old ea_inode, inode map, block bitmap, group descriptor. 946 */ 947 credits += 4; 948 949 /* Data blocks for old ea_inode. */ 950 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits; 951 952 /* Indirection block or one level of extent tree for old 953 * ea_inode. 954 */ 955 blocks += 1; 956 957 /* Block bitmap and group descriptor updates for each block. */ 958 credits += blocks * 2; 959 } 960 961 /* We may need to clone the existing xattr block in which case we need 962 * to increment ref counts for existing ea_inodes referenced by it. 963 */ 964 if (block_bh) { 965 struct ext4_xattr_entry *entry = BFIRST(block_bh); 966 967 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) 968 if (entry->e_value_inum) 969 /* Ref count update on ea_inode. */ 970 credits += 1; 971 } 972 return credits; 973 } 974 975 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode, 976 int ref_change) 977 { 978 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode); 979 struct ext4_iloc iloc; 980 s64 ref_count; 981 u32 hash; 982 int ret; 983 984 inode_lock(ea_inode); 985 986 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc); 987 if (ret) 988 goto out; 989 990 ref_count = ext4_xattr_inode_get_ref(ea_inode); 991 ref_count += ref_change; 992 ext4_xattr_inode_set_ref(ea_inode, ref_count); 993 994 if (ref_change > 0) { 995 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld", 996 ea_inode->i_ino, ref_count); 997 998 if (ref_count == 1) { 999 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u", 1000 ea_inode->i_ino, ea_inode->i_nlink); 1001 1002 set_nlink(ea_inode, 1); 1003 ext4_orphan_del(handle, ea_inode); 1004 1005 if (ea_inode_cache) { 1006 hash = ext4_xattr_inode_get_hash(ea_inode); 1007 mb_cache_entry_create(ea_inode_cache, 1008 GFP_NOFS, hash, 1009 ea_inode->i_ino, 1010 true /* reusable */); 1011 } 1012 } 1013 } else { 1014 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld", 1015 ea_inode->i_ino, ref_count); 1016 1017 if (ref_count == 0) { 1018 WARN_ONCE(ea_inode->i_nlink != 1, 1019 "EA inode %lu i_nlink=%u", 1020 ea_inode->i_ino, ea_inode->i_nlink); 1021 1022 clear_nlink(ea_inode); 1023 ext4_orphan_add(handle, ea_inode); 1024 1025 if (ea_inode_cache) { 1026 hash = ext4_xattr_inode_get_hash(ea_inode); 1027 mb_cache_entry_delete(ea_inode_cache, hash, 1028 ea_inode->i_ino); 1029 } 1030 } 1031 } 1032 1033 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc); 1034 if (ret) 1035 ext4_warning_inode(ea_inode, 1036 "ext4_mark_iloc_dirty() failed ret=%d", ret); 1037 out: 1038 inode_unlock(ea_inode); 1039 return ret; 1040 } 1041 1042 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode) 1043 { 1044 return ext4_xattr_inode_update_ref(handle, ea_inode, 1); 1045 } 1046 1047 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode) 1048 { 1049 return ext4_xattr_inode_update_ref(handle, ea_inode, -1); 1050 } 1051 1052 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent, 1053 struct ext4_xattr_entry *first) 1054 { 1055 struct inode *ea_inode; 1056 struct ext4_xattr_entry *entry; 1057 struct ext4_xattr_entry *failed_entry; 1058 unsigned int ea_ino; 1059 int err, saved_err; 1060 1061 for (entry = first; !IS_LAST_ENTRY(entry); 1062 entry = EXT4_XATTR_NEXT(entry)) { 1063 if (!entry->e_value_inum) 1064 continue; 1065 ea_ino = le32_to_cpu(entry->e_value_inum); 1066 err = ext4_xattr_inode_iget(parent, ea_ino, 1067 le32_to_cpu(entry->e_hash), 1068 &ea_inode); 1069 if (err) 1070 goto cleanup; 1071 err = ext4_xattr_inode_inc_ref(handle, ea_inode); 1072 if (err) { 1073 ext4_warning_inode(ea_inode, "inc ref error %d", err); 1074 iput(ea_inode); 1075 goto cleanup; 1076 } 1077 iput(ea_inode); 1078 } 1079 return 0; 1080 1081 cleanup: 1082 saved_err = err; 1083 failed_entry = entry; 1084 1085 for (entry = first; entry != failed_entry; 1086 entry = EXT4_XATTR_NEXT(entry)) { 1087 if (!entry->e_value_inum) 1088 continue; 1089 ea_ino = le32_to_cpu(entry->e_value_inum); 1090 err = ext4_xattr_inode_iget(parent, ea_ino, 1091 le32_to_cpu(entry->e_hash), 1092 &ea_inode); 1093 if (err) { 1094 ext4_warning(parent->i_sb, 1095 "cleanup ea_ino %u iget error %d", ea_ino, 1096 err); 1097 continue; 1098 } 1099 err = ext4_xattr_inode_dec_ref(handle, ea_inode); 1100 if (err) 1101 ext4_warning_inode(ea_inode, "cleanup dec ref error %d", 1102 err); 1103 iput(ea_inode); 1104 } 1105 return saved_err; 1106 } 1107 1108 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode, 1109 struct buffer_head *bh, bool block_csum, bool dirty) 1110 { 1111 int error; 1112 1113 if (bh && dirty) { 1114 if (block_csum) 1115 ext4_xattr_block_csum_set(inode, bh); 1116 error = ext4_handle_dirty_metadata(handle, NULL, bh); 1117 if (error) { 1118 ext4_warning(inode->i_sb, "Handle metadata (error %d)", 1119 error); 1120 return error; 1121 } 1122 } 1123 return 0; 1124 } 1125 1126 static void 1127 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent, 1128 struct buffer_head *bh, 1129 struct ext4_xattr_entry *first, bool block_csum, 1130 struct ext4_xattr_inode_array **ea_inode_array, 1131 int extra_credits, bool skip_quota) 1132 { 1133 struct inode *ea_inode; 1134 struct ext4_xattr_entry *entry; 1135 bool dirty = false; 1136 unsigned int ea_ino; 1137 int err; 1138 int credits; 1139 1140 /* One credit for dec ref on ea_inode, one for orphan list addition, */ 1141 credits = 2 + extra_credits; 1142 1143 for (entry = first; !IS_LAST_ENTRY(entry); 1144 entry = EXT4_XATTR_NEXT(entry)) { 1145 if (!entry->e_value_inum) 1146 continue; 1147 ea_ino = le32_to_cpu(entry->e_value_inum); 1148 err = ext4_xattr_inode_iget(parent, ea_ino, 1149 le32_to_cpu(entry->e_hash), 1150 &ea_inode); 1151 if (err) 1152 continue; 1153 1154 err = ext4_expand_inode_array(ea_inode_array, ea_inode); 1155 if (err) { 1156 ext4_warning_inode(ea_inode, 1157 "Expand inode array err=%d", err); 1158 iput(ea_inode); 1159 continue; 1160 } 1161 1162 err = ext4_journal_ensure_credits_fn(handle, credits, credits, 1163 ext4_free_metadata_revoke_credits(parent->i_sb, 1), 1164 ext4_xattr_restart_fn(handle, parent, bh, block_csum, 1165 dirty)); 1166 if (err < 0) { 1167 ext4_warning_inode(ea_inode, "Ensure credits err=%d", 1168 err); 1169 continue; 1170 } 1171 if (err > 0) { 1172 err = ext4_journal_get_write_access(handle, bh); 1173 if (err) { 1174 ext4_warning_inode(ea_inode, 1175 "Re-get write access err=%d", 1176 err); 1177 continue; 1178 } 1179 } 1180 1181 err = ext4_xattr_inode_dec_ref(handle, ea_inode); 1182 if (err) { 1183 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d", 1184 err); 1185 continue; 1186 } 1187 1188 if (!skip_quota) 1189 ext4_xattr_inode_free_quota(parent, ea_inode, 1190 le32_to_cpu(entry->e_value_size)); 1191 1192 /* 1193 * Forget about ea_inode within the same transaction that 1194 * decrements the ref count. This avoids duplicate decrements in 1195 * case the rest of the work spills over to subsequent 1196 * transactions. 1197 */ 1198 entry->e_value_inum = 0; 1199 entry->e_value_size = 0; 1200 1201 dirty = true; 1202 } 1203 1204 if (dirty) { 1205 /* 1206 * Note that we are deliberately skipping csum calculation for 1207 * the final update because we do not expect any journal 1208 * restarts until xattr block is freed. 1209 */ 1210 1211 err = ext4_handle_dirty_metadata(handle, NULL, bh); 1212 if (err) 1213 ext4_warning_inode(parent, 1214 "handle dirty metadata err=%d", err); 1215 } 1216 } 1217 1218 /* 1219 * Release the xattr block BH: If the reference count is > 1, decrement it; 1220 * otherwise free the block. 1221 */ 1222 static void 1223 ext4_xattr_release_block(handle_t *handle, struct inode *inode, 1224 struct buffer_head *bh, 1225 struct ext4_xattr_inode_array **ea_inode_array, 1226 int extra_credits) 1227 { 1228 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 1229 u32 hash, ref; 1230 int error = 0; 1231 1232 BUFFER_TRACE(bh, "get_write_access"); 1233 error = ext4_journal_get_write_access(handle, bh); 1234 if (error) 1235 goto out; 1236 1237 lock_buffer(bh); 1238 hash = le32_to_cpu(BHDR(bh)->h_hash); 1239 ref = le32_to_cpu(BHDR(bh)->h_refcount); 1240 if (ref == 1) { 1241 ea_bdebug(bh, "refcount now=0; freeing"); 1242 /* 1243 * This must happen under buffer lock for 1244 * ext4_xattr_block_set() to reliably detect freed block 1245 */ 1246 if (ea_block_cache) 1247 mb_cache_entry_delete(ea_block_cache, hash, 1248 bh->b_blocknr); 1249 get_bh(bh); 1250 unlock_buffer(bh); 1251 1252 if (ext4_has_feature_ea_inode(inode->i_sb)) 1253 ext4_xattr_inode_dec_ref_all(handle, inode, bh, 1254 BFIRST(bh), 1255 true /* block_csum */, 1256 ea_inode_array, 1257 extra_credits, 1258 true /* skip_quota */); 1259 ext4_free_blocks(handle, inode, bh, 0, 1, 1260 EXT4_FREE_BLOCKS_METADATA | 1261 EXT4_FREE_BLOCKS_FORGET); 1262 } else { 1263 ref--; 1264 BHDR(bh)->h_refcount = cpu_to_le32(ref); 1265 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) { 1266 struct mb_cache_entry *ce; 1267 1268 if (ea_block_cache) { 1269 ce = mb_cache_entry_get(ea_block_cache, hash, 1270 bh->b_blocknr); 1271 if (ce) { 1272 ce->e_reusable = 1; 1273 mb_cache_entry_put(ea_block_cache, ce); 1274 } 1275 } 1276 } 1277 1278 ext4_xattr_block_csum_set(inode, bh); 1279 /* 1280 * Beware of this ugliness: Releasing of xattr block references 1281 * from different inodes can race and so we have to protect 1282 * from a race where someone else frees the block (and releases 1283 * its journal_head) before we are done dirtying the buffer. In 1284 * nojournal mode this race is harmless and we actually cannot 1285 * call ext4_handle_dirty_metadata() with locked buffer as 1286 * that function can call sync_dirty_buffer() so for that case 1287 * we handle the dirtying after unlocking the buffer. 1288 */ 1289 if (ext4_handle_valid(handle)) 1290 error = ext4_handle_dirty_metadata(handle, inode, bh); 1291 unlock_buffer(bh); 1292 if (!ext4_handle_valid(handle)) 1293 error = ext4_handle_dirty_metadata(handle, inode, bh); 1294 if (IS_SYNC(inode)) 1295 ext4_handle_sync(handle); 1296 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1)); 1297 ea_bdebug(bh, "refcount now=%d; releasing", 1298 le32_to_cpu(BHDR(bh)->h_refcount)); 1299 } 1300 out: 1301 ext4_std_error(inode->i_sb, error); 1302 return; 1303 } 1304 1305 /* 1306 * Find the available free space for EAs. This also returns the total number of 1307 * bytes used by EA entries. 1308 */ 1309 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last, 1310 size_t *min_offs, void *base, int *total) 1311 { 1312 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { 1313 if (!last->e_value_inum && last->e_value_size) { 1314 size_t offs = le16_to_cpu(last->e_value_offs); 1315 if (offs < *min_offs) 1316 *min_offs = offs; 1317 } 1318 if (total) 1319 *total += EXT4_XATTR_LEN(last->e_name_len); 1320 } 1321 return (*min_offs - ((void *)last - base) - sizeof(__u32)); 1322 } 1323 1324 /* 1325 * Write the value of the EA in an inode. 1326 */ 1327 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode, 1328 const void *buf, int bufsize) 1329 { 1330 struct buffer_head *bh = NULL; 1331 unsigned long block = 0; 1332 int blocksize = ea_inode->i_sb->s_blocksize; 1333 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits; 1334 int csize, wsize = 0; 1335 int ret = 0, ret2 = 0; 1336 int retries = 0; 1337 1338 retry: 1339 while (ret >= 0 && ret < max_blocks) { 1340 struct ext4_map_blocks map; 1341 map.m_lblk = block += ret; 1342 map.m_len = max_blocks -= ret; 1343 1344 ret = ext4_map_blocks(handle, ea_inode, &map, 1345 EXT4_GET_BLOCKS_CREATE); 1346 if (ret <= 0) { 1347 ext4_mark_inode_dirty(handle, ea_inode); 1348 if (ret == -ENOSPC && 1349 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) { 1350 ret = 0; 1351 goto retry; 1352 } 1353 break; 1354 } 1355 } 1356 1357 if (ret < 0) 1358 return ret; 1359 1360 block = 0; 1361 while (wsize < bufsize) { 1362 brelse(bh); 1363 csize = (bufsize - wsize) > blocksize ? blocksize : 1364 bufsize - wsize; 1365 bh = ext4_getblk(handle, ea_inode, block, 0); 1366 if (IS_ERR(bh)) 1367 return PTR_ERR(bh); 1368 if (!bh) { 1369 WARN_ON_ONCE(1); 1370 EXT4_ERROR_INODE(ea_inode, 1371 "ext4_getblk() return bh = NULL"); 1372 return -EFSCORRUPTED; 1373 } 1374 ret = ext4_journal_get_write_access(handle, bh); 1375 if (ret) 1376 goto out; 1377 1378 memcpy(bh->b_data, buf, csize); 1379 set_buffer_uptodate(bh); 1380 ext4_handle_dirty_metadata(handle, ea_inode, bh); 1381 1382 buf += csize; 1383 wsize += csize; 1384 block += 1; 1385 } 1386 1387 inode_lock(ea_inode); 1388 i_size_write(ea_inode, wsize); 1389 ext4_update_i_disksize(ea_inode, wsize); 1390 inode_unlock(ea_inode); 1391 1392 ret2 = ext4_mark_inode_dirty(handle, ea_inode); 1393 if (unlikely(ret2 && !ret)) 1394 ret = ret2; 1395 1396 out: 1397 brelse(bh); 1398 1399 return ret; 1400 } 1401 1402 /* 1403 * Create an inode to store the value of a large EA. 1404 */ 1405 static struct inode *ext4_xattr_inode_create(handle_t *handle, 1406 struct inode *inode, u32 hash) 1407 { 1408 struct inode *ea_inode = NULL; 1409 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) }; 1410 int err; 1411 1412 /* 1413 * Let the next inode be the goal, so we try and allocate the EA inode 1414 * in the same group, or nearby one. 1415 */ 1416 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode, 1417 S_IFREG | 0600, NULL, inode->i_ino + 1, owner, 1418 EXT4_EA_INODE_FL); 1419 if (!IS_ERR(ea_inode)) { 1420 ea_inode->i_op = &ext4_file_inode_operations; 1421 ea_inode->i_fop = &ext4_file_operations; 1422 ext4_set_aops(ea_inode); 1423 ext4_xattr_inode_set_class(ea_inode); 1424 unlock_new_inode(ea_inode); 1425 ext4_xattr_inode_set_ref(ea_inode, 1); 1426 ext4_xattr_inode_set_hash(ea_inode, hash); 1427 err = ext4_mark_inode_dirty(handle, ea_inode); 1428 if (!err) 1429 err = ext4_inode_attach_jinode(ea_inode); 1430 if (err) { 1431 iput(ea_inode); 1432 return ERR_PTR(err); 1433 } 1434 1435 /* 1436 * Xattr inodes are shared therefore quota charging is performed 1437 * at a higher level. 1438 */ 1439 dquot_free_inode(ea_inode); 1440 dquot_drop(ea_inode); 1441 inode_lock(ea_inode); 1442 ea_inode->i_flags |= S_NOQUOTA; 1443 inode_unlock(ea_inode); 1444 } 1445 1446 return ea_inode; 1447 } 1448 1449 static struct inode * 1450 ext4_xattr_inode_cache_find(struct inode *inode, const void *value, 1451 size_t value_len, u32 hash) 1452 { 1453 struct inode *ea_inode; 1454 struct mb_cache_entry *ce; 1455 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode); 1456 void *ea_data; 1457 1458 if (!ea_inode_cache) 1459 return NULL; 1460 1461 ce = mb_cache_entry_find_first(ea_inode_cache, hash); 1462 if (!ce) 1463 return NULL; 1464 1465 ea_data = kvmalloc(value_len, GFP_KERNEL); 1466 if (!ea_data) { 1467 mb_cache_entry_put(ea_inode_cache, ce); 1468 return NULL; 1469 } 1470 1471 while (ce) { 1472 ea_inode = ext4_iget(inode->i_sb, ce->e_value, 1473 EXT4_IGET_NORMAL); 1474 if (!IS_ERR(ea_inode) && 1475 !is_bad_inode(ea_inode) && 1476 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) && 1477 i_size_read(ea_inode) == value_len && 1478 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) && 1479 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data, 1480 value_len) && 1481 !memcmp(value, ea_data, value_len)) { 1482 mb_cache_entry_touch(ea_inode_cache, ce); 1483 mb_cache_entry_put(ea_inode_cache, ce); 1484 kvfree(ea_data); 1485 return ea_inode; 1486 } 1487 1488 if (!IS_ERR(ea_inode)) 1489 iput(ea_inode); 1490 ce = mb_cache_entry_find_next(ea_inode_cache, ce); 1491 } 1492 kvfree(ea_data); 1493 return NULL; 1494 } 1495 1496 /* 1497 * Add value of the EA in an inode. 1498 */ 1499 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode, 1500 const void *value, size_t value_len, 1501 struct inode **ret_inode) 1502 { 1503 struct inode *ea_inode; 1504 u32 hash; 1505 int err; 1506 1507 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len); 1508 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash); 1509 if (ea_inode) { 1510 err = ext4_xattr_inode_inc_ref(handle, ea_inode); 1511 if (err) { 1512 iput(ea_inode); 1513 return err; 1514 } 1515 1516 *ret_inode = ea_inode; 1517 return 0; 1518 } 1519 1520 /* Create an inode for the EA value */ 1521 ea_inode = ext4_xattr_inode_create(handle, inode, hash); 1522 if (IS_ERR(ea_inode)) 1523 return PTR_ERR(ea_inode); 1524 1525 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len); 1526 if (err) { 1527 ext4_xattr_inode_dec_ref(handle, ea_inode); 1528 iput(ea_inode); 1529 return err; 1530 } 1531 1532 if (EA_INODE_CACHE(inode)) 1533 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash, 1534 ea_inode->i_ino, true /* reusable */); 1535 1536 *ret_inode = ea_inode; 1537 return 0; 1538 } 1539 1540 /* 1541 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode 1542 * feature is enabled. 1543 */ 1544 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U) 1545 1546 static int ext4_xattr_set_entry(struct ext4_xattr_info *i, 1547 struct ext4_xattr_search *s, 1548 handle_t *handle, struct inode *inode, 1549 bool is_block) 1550 { 1551 struct ext4_xattr_entry *last, *next; 1552 struct ext4_xattr_entry *here = s->here; 1553 size_t min_offs = s->end - s->base, name_len = strlen(i->name); 1554 int in_inode = i->in_inode; 1555 struct inode *old_ea_inode = NULL; 1556 struct inode *new_ea_inode = NULL; 1557 size_t old_size, new_size; 1558 int ret; 1559 1560 /* Space used by old and new values. */ 1561 old_size = (!s->not_found && !here->e_value_inum) ? 1562 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0; 1563 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0; 1564 1565 /* 1566 * Optimization for the simple case when old and new values have the 1567 * same padded sizes. Not applicable if external inodes are involved. 1568 */ 1569 if (new_size && new_size == old_size) { 1570 size_t offs = le16_to_cpu(here->e_value_offs); 1571 void *val = s->base + offs; 1572 1573 here->e_value_size = cpu_to_le32(i->value_len); 1574 if (i->value == EXT4_ZERO_XATTR_VALUE) { 1575 memset(val, 0, new_size); 1576 } else { 1577 memcpy(val, i->value, i->value_len); 1578 /* Clear padding bytes. */ 1579 memset(val + i->value_len, 0, new_size - i->value_len); 1580 } 1581 goto update_hash; 1582 } 1583 1584 /* Compute min_offs and last. */ 1585 last = s->first; 1586 for (; !IS_LAST_ENTRY(last); last = next) { 1587 next = EXT4_XATTR_NEXT(last); 1588 if ((void *)next >= s->end) { 1589 EXT4_ERROR_INODE(inode, "corrupted xattr entries"); 1590 ret = -EFSCORRUPTED; 1591 goto out; 1592 } 1593 if (!last->e_value_inum && last->e_value_size) { 1594 size_t offs = le16_to_cpu(last->e_value_offs); 1595 if (offs < min_offs) 1596 min_offs = offs; 1597 } 1598 } 1599 1600 /* Check whether we have enough space. */ 1601 if (i->value) { 1602 size_t free; 1603 1604 free = min_offs - ((void *)last - s->base) - sizeof(__u32); 1605 if (!s->not_found) 1606 free += EXT4_XATTR_LEN(name_len) + old_size; 1607 1608 if (free < EXT4_XATTR_LEN(name_len) + new_size) { 1609 ret = -ENOSPC; 1610 goto out; 1611 } 1612 1613 /* 1614 * If storing the value in an external inode is an option, 1615 * reserve space for xattr entries/names in the external 1616 * attribute block so that a long value does not occupy the 1617 * whole space and prevent futher entries being added. 1618 */ 1619 if (ext4_has_feature_ea_inode(inode->i_sb) && 1620 new_size && is_block && 1621 (min_offs + old_size - new_size) < 1622 EXT4_XATTR_BLOCK_RESERVE(inode)) { 1623 ret = -ENOSPC; 1624 goto out; 1625 } 1626 } 1627 1628 /* 1629 * Getting access to old and new ea inodes is subject to failures. 1630 * Finish that work before doing any modifications to the xattr data. 1631 */ 1632 if (!s->not_found && here->e_value_inum) { 1633 ret = ext4_xattr_inode_iget(inode, 1634 le32_to_cpu(here->e_value_inum), 1635 le32_to_cpu(here->e_hash), 1636 &old_ea_inode); 1637 if (ret) { 1638 old_ea_inode = NULL; 1639 goto out; 1640 } 1641 } 1642 if (i->value && in_inode) { 1643 WARN_ON_ONCE(!i->value_len); 1644 1645 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len); 1646 if (ret) 1647 goto out; 1648 1649 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value, 1650 i->value_len, 1651 &new_ea_inode); 1652 if (ret) { 1653 new_ea_inode = NULL; 1654 ext4_xattr_inode_free_quota(inode, NULL, i->value_len); 1655 goto out; 1656 } 1657 } 1658 1659 if (old_ea_inode) { 1660 /* We are ready to release ref count on the old_ea_inode. */ 1661 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode); 1662 if (ret) { 1663 /* Release newly required ref count on new_ea_inode. */ 1664 if (new_ea_inode) { 1665 int err; 1666 1667 err = ext4_xattr_inode_dec_ref(handle, 1668 new_ea_inode); 1669 if (err) 1670 ext4_warning_inode(new_ea_inode, 1671 "dec ref new_ea_inode err=%d", 1672 err); 1673 ext4_xattr_inode_free_quota(inode, new_ea_inode, 1674 i->value_len); 1675 } 1676 goto out; 1677 } 1678 1679 ext4_xattr_inode_free_quota(inode, old_ea_inode, 1680 le32_to_cpu(here->e_value_size)); 1681 } 1682 1683 /* No failures allowed past this point. */ 1684 1685 if (!s->not_found && here->e_value_size && !here->e_value_inum) { 1686 /* Remove the old value. */ 1687 void *first_val = s->base + min_offs; 1688 size_t offs = le16_to_cpu(here->e_value_offs); 1689 void *val = s->base + offs; 1690 1691 memmove(first_val + old_size, first_val, val - first_val); 1692 memset(first_val, 0, old_size); 1693 min_offs += old_size; 1694 1695 /* Adjust all value offsets. */ 1696 last = s->first; 1697 while (!IS_LAST_ENTRY(last)) { 1698 size_t o = le16_to_cpu(last->e_value_offs); 1699 1700 if (!last->e_value_inum && 1701 last->e_value_size && o < offs) 1702 last->e_value_offs = cpu_to_le16(o + old_size); 1703 last = EXT4_XATTR_NEXT(last); 1704 } 1705 } 1706 1707 if (!i->value) { 1708 /* Remove old name. */ 1709 size_t size = EXT4_XATTR_LEN(name_len); 1710 1711 last = ENTRY((void *)last - size); 1712 memmove(here, (void *)here + size, 1713 (void *)last - (void *)here + sizeof(__u32)); 1714 memset(last, 0, size); 1715 } else if (s->not_found) { 1716 /* Insert new name. */ 1717 size_t size = EXT4_XATTR_LEN(name_len); 1718 size_t rest = (void *)last - (void *)here + sizeof(__u32); 1719 1720 memmove((void *)here + size, here, rest); 1721 memset(here, 0, size); 1722 here->e_name_index = i->name_index; 1723 here->e_name_len = name_len; 1724 memcpy(here->e_name, i->name, name_len); 1725 } else { 1726 /* This is an update, reset value info. */ 1727 here->e_value_inum = 0; 1728 here->e_value_offs = 0; 1729 here->e_value_size = 0; 1730 } 1731 1732 if (i->value) { 1733 /* Insert new value. */ 1734 if (in_inode) { 1735 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino); 1736 } else if (i->value_len) { 1737 void *val = s->base + min_offs - new_size; 1738 1739 here->e_value_offs = cpu_to_le16(min_offs - new_size); 1740 if (i->value == EXT4_ZERO_XATTR_VALUE) { 1741 memset(val, 0, new_size); 1742 } else { 1743 memcpy(val, i->value, i->value_len); 1744 /* Clear padding bytes. */ 1745 memset(val + i->value_len, 0, 1746 new_size - i->value_len); 1747 } 1748 } 1749 here->e_value_size = cpu_to_le32(i->value_len); 1750 } 1751 1752 update_hash: 1753 if (i->value) { 1754 __le32 hash = 0; 1755 1756 /* Entry hash calculation. */ 1757 if (in_inode) { 1758 __le32 crc32c_hash; 1759 1760 /* 1761 * Feed crc32c hash instead of the raw value for entry 1762 * hash calculation. This is to avoid walking 1763 * potentially long value buffer again. 1764 */ 1765 crc32c_hash = cpu_to_le32( 1766 ext4_xattr_inode_get_hash(new_ea_inode)); 1767 hash = ext4_xattr_hash_entry(here->e_name, 1768 here->e_name_len, 1769 &crc32c_hash, 1); 1770 } else if (is_block) { 1771 __le32 *value = s->base + le16_to_cpu( 1772 here->e_value_offs); 1773 1774 hash = ext4_xattr_hash_entry(here->e_name, 1775 here->e_name_len, value, 1776 new_size >> 2); 1777 } 1778 here->e_hash = hash; 1779 } 1780 1781 if (is_block) 1782 ext4_xattr_rehash((struct ext4_xattr_header *)s->base); 1783 1784 ret = 0; 1785 out: 1786 iput(old_ea_inode); 1787 iput(new_ea_inode); 1788 return ret; 1789 } 1790 1791 struct ext4_xattr_block_find { 1792 struct ext4_xattr_search s; 1793 struct buffer_head *bh; 1794 }; 1795 1796 static int 1797 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i, 1798 struct ext4_xattr_block_find *bs) 1799 { 1800 struct super_block *sb = inode->i_sb; 1801 int error; 1802 1803 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld", 1804 i->name_index, i->name, i->value, (long)i->value_len); 1805 1806 if (EXT4_I(inode)->i_file_acl) { 1807 /* The inode already has an extended attribute block. */ 1808 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 1809 if (IS_ERR(bs->bh)) { 1810 error = PTR_ERR(bs->bh); 1811 bs->bh = NULL; 1812 return error; 1813 } 1814 ea_bdebug(bs->bh, "b_count=%d, refcount=%d", 1815 atomic_read(&(bs->bh->b_count)), 1816 le32_to_cpu(BHDR(bs->bh)->h_refcount)); 1817 error = ext4_xattr_check_block(inode, bs->bh); 1818 if (error) 1819 return error; 1820 /* Find the named attribute. */ 1821 bs->s.base = BHDR(bs->bh); 1822 bs->s.first = BFIRST(bs->bh); 1823 bs->s.end = bs->bh->b_data + bs->bh->b_size; 1824 bs->s.here = bs->s.first; 1825 error = xattr_find_entry(inode, &bs->s.here, bs->s.end, 1826 i->name_index, i->name, 1); 1827 if (error && error != -ENODATA) 1828 return error; 1829 bs->s.not_found = error; 1830 } 1831 return 0; 1832 } 1833 1834 static int 1835 ext4_xattr_block_set(handle_t *handle, struct inode *inode, 1836 struct ext4_xattr_info *i, 1837 struct ext4_xattr_block_find *bs) 1838 { 1839 struct super_block *sb = inode->i_sb; 1840 struct buffer_head *new_bh = NULL; 1841 struct ext4_xattr_search s_copy = bs->s; 1842 struct ext4_xattr_search *s = &s_copy; 1843 struct mb_cache_entry *ce = NULL; 1844 int error = 0; 1845 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 1846 struct inode *ea_inode = NULL, *tmp_inode; 1847 size_t old_ea_inode_quota = 0; 1848 unsigned int ea_ino; 1849 1850 1851 #define header(x) ((struct ext4_xattr_header *)(x)) 1852 1853 if (s->base) { 1854 BUFFER_TRACE(bs->bh, "get_write_access"); 1855 error = ext4_journal_get_write_access(handle, bs->bh); 1856 if (error) 1857 goto cleanup; 1858 lock_buffer(bs->bh); 1859 1860 if (header(s->base)->h_refcount == cpu_to_le32(1)) { 1861 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash); 1862 1863 /* 1864 * This must happen under buffer lock for 1865 * ext4_xattr_block_set() to reliably detect modified 1866 * block 1867 */ 1868 if (ea_block_cache) 1869 mb_cache_entry_delete(ea_block_cache, hash, 1870 bs->bh->b_blocknr); 1871 ea_bdebug(bs->bh, "modifying in-place"); 1872 error = ext4_xattr_set_entry(i, s, handle, inode, 1873 true /* is_block */); 1874 ext4_xattr_block_csum_set(inode, bs->bh); 1875 unlock_buffer(bs->bh); 1876 if (error == -EFSCORRUPTED) 1877 goto bad_block; 1878 if (!error) 1879 error = ext4_handle_dirty_metadata(handle, 1880 inode, 1881 bs->bh); 1882 if (error) 1883 goto cleanup; 1884 goto inserted; 1885 } else { 1886 int offset = (char *)s->here - bs->bh->b_data; 1887 1888 unlock_buffer(bs->bh); 1889 ea_bdebug(bs->bh, "cloning"); 1890 s->base = kmalloc(bs->bh->b_size, GFP_NOFS); 1891 error = -ENOMEM; 1892 if (s->base == NULL) 1893 goto cleanup; 1894 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size); 1895 s->first = ENTRY(header(s->base)+1); 1896 header(s->base)->h_refcount = cpu_to_le32(1); 1897 s->here = ENTRY(s->base + offset); 1898 s->end = s->base + bs->bh->b_size; 1899 1900 /* 1901 * If existing entry points to an xattr inode, we need 1902 * to prevent ext4_xattr_set_entry() from decrementing 1903 * ref count on it because the reference belongs to the 1904 * original block. In this case, make the entry look 1905 * like it has an empty value. 1906 */ 1907 if (!s->not_found && s->here->e_value_inum) { 1908 ea_ino = le32_to_cpu(s->here->e_value_inum); 1909 error = ext4_xattr_inode_iget(inode, ea_ino, 1910 le32_to_cpu(s->here->e_hash), 1911 &tmp_inode); 1912 if (error) 1913 goto cleanup; 1914 1915 if (!ext4_test_inode_state(tmp_inode, 1916 EXT4_STATE_LUSTRE_EA_INODE)) { 1917 /* 1918 * Defer quota free call for previous 1919 * inode until success is guaranteed. 1920 */ 1921 old_ea_inode_quota = le32_to_cpu( 1922 s->here->e_value_size); 1923 } 1924 iput(tmp_inode); 1925 1926 s->here->e_value_inum = 0; 1927 s->here->e_value_size = 0; 1928 } 1929 } 1930 } else { 1931 /* Allocate a buffer where we construct the new block. */ 1932 s->base = kzalloc(sb->s_blocksize, GFP_NOFS); 1933 error = -ENOMEM; 1934 if (s->base == NULL) 1935 goto cleanup; 1936 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC); 1937 header(s->base)->h_blocks = cpu_to_le32(1); 1938 header(s->base)->h_refcount = cpu_to_le32(1); 1939 s->first = ENTRY(header(s->base)+1); 1940 s->here = ENTRY(header(s->base)+1); 1941 s->end = s->base + sb->s_blocksize; 1942 } 1943 1944 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */); 1945 if (error == -EFSCORRUPTED) 1946 goto bad_block; 1947 if (error) 1948 goto cleanup; 1949 1950 if (i->value && s->here->e_value_inum) { 1951 /* 1952 * A ref count on ea_inode has been taken as part of the call to 1953 * ext4_xattr_set_entry() above. We would like to drop this 1954 * extra ref but we have to wait until the xattr block is 1955 * initialized and has its own ref count on the ea_inode. 1956 */ 1957 ea_ino = le32_to_cpu(s->here->e_value_inum); 1958 error = ext4_xattr_inode_iget(inode, ea_ino, 1959 le32_to_cpu(s->here->e_hash), 1960 &ea_inode); 1961 if (error) { 1962 ea_inode = NULL; 1963 goto cleanup; 1964 } 1965 } 1966 1967 inserted: 1968 if (!IS_LAST_ENTRY(s->first)) { 1969 new_bh = ext4_xattr_block_cache_find(inode, header(s->base), 1970 &ce); 1971 if (new_bh) { 1972 /* We found an identical block in the cache. */ 1973 if (new_bh == bs->bh) 1974 ea_bdebug(new_bh, "keeping"); 1975 else { 1976 u32 ref; 1977 1978 WARN_ON_ONCE(dquot_initialize_needed(inode)); 1979 1980 /* The old block is released after updating 1981 the inode. */ 1982 error = dquot_alloc_block(inode, 1983 EXT4_C2B(EXT4_SB(sb), 1)); 1984 if (error) 1985 goto cleanup; 1986 BUFFER_TRACE(new_bh, "get_write_access"); 1987 error = ext4_journal_get_write_access(handle, 1988 new_bh); 1989 if (error) 1990 goto cleanup_dquot; 1991 lock_buffer(new_bh); 1992 /* 1993 * We have to be careful about races with 1994 * freeing, rehashing or adding references to 1995 * xattr block. Once we hold buffer lock xattr 1996 * block's state is stable so we can check 1997 * whether the block got freed / rehashed or 1998 * not. Since we unhash mbcache entry under 1999 * buffer lock when freeing / rehashing xattr 2000 * block, checking whether entry is still 2001 * hashed is reliable. Same rules hold for 2002 * e_reusable handling. 2003 */ 2004 if (hlist_bl_unhashed(&ce->e_hash_list) || 2005 !ce->e_reusable) { 2006 /* 2007 * Undo everything and check mbcache 2008 * again. 2009 */ 2010 unlock_buffer(new_bh); 2011 dquot_free_block(inode, 2012 EXT4_C2B(EXT4_SB(sb), 2013 1)); 2014 brelse(new_bh); 2015 mb_cache_entry_put(ea_block_cache, ce); 2016 ce = NULL; 2017 new_bh = NULL; 2018 goto inserted; 2019 } 2020 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1; 2021 BHDR(new_bh)->h_refcount = cpu_to_le32(ref); 2022 if (ref >= EXT4_XATTR_REFCOUNT_MAX) 2023 ce->e_reusable = 0; 2024 ea_bdebug(new_bh, "reusing; refcount now=%d", 2025 ref); 2026 ext4_xattr_block_csum_set(inode, new_bh); 2027 unlock_buffer(new_bh); 2028 error = ext4_handle_dirty_metadata(handle, 2029 inode, 2030 new_bh); 2031 if (error) 2032 goto cleanup_dquot; 2033 } 2034 mb_cache_entry_touch(ea_block_cache, ce); 2035 mb_cache_entry_put(ea_block_cache, ce); 2036 ce = NULL; 2037 } else if (bs->bh && s->base == bs->bh->b_data) { 2038 /* We were modifying this block in-place. */ 2039 ea_bdebug(bs->bh, "keeping this block"); 2040 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh); 2041 new_bh = bs->bh; 2042 get_bh(new_bh); 2043 } else { 2044 /* We need to allocate a new block */ 2045 ext4_fsblk_t goal, block; 2046 2047 WARN_ON_ONCE(dquot_initialize_needed(inode)); 2048 2049 goal = ext4_group_first_block_no(sb, 2050 EXT4_I(inode)->i_block_group); 2051 2052 /* non-extent files can't have physical blocks past 2^32 */ 2053 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 2054 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; 2055 2056 block = ext4_new_meta_blocks(handle, inode, goal, 0, 2057 NULL, &error); 2058 if (error) 2059 goto cleanup; 2060 2061 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 2062 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS); 2063 2064 ea_idebug(inode, "creating block %llu", 2065 (unsigned long long)block); 2066 2067 new_bh = sb_getblk(sb, block); 2068 if (unlikely(!new_bh)) { 2069 error = -ENOMEM; 2070 getblk_failed: 2071 ext4_free_blocks(handle, inode, NULL, block, 1, 2072 EXT4_FREE_BLOCKS_METADATA); 2073 goto cleanup; 2074 } 2075 error = ext4_xattr_inode_inc_ref_all(handle, inode, 2076 ENTRY(header(s->base)+1)); 2077 if (error) 2078 goto getblk_failed; 2079 if (ea_inode) { 2080 /* Drop the extra ref on ea_inode. */ 2081 error = ext4_xattr_inode_dec_ref(handle, 2082 ea_inode); 2083 if (error) 2084 ext4_warning_inode(ea_inode, 2085 "dec ref error=%d", 2086 error); 2087 iput(ea_inode); 2088 ea_inode = NULL; 2089 } 2090 2091 lock_buffer(new_bh); 2092 error = ext4_journal_get_create_access(handle, new_bh); 2093 if (error) { 2094 unlock_buffer(new_bh); 2095 error = -EIO; 2096 goto getblk_failed; 2097 } 2098 memcpy(new_bh->b_data, s->base, new_bh->b_size); 2099 ext4_xattr_block_csum_set(inode, new_bh); 2100 set_buffer_uptodate(new_bh); 2101 unlock_buffer(new_bh); 2102 ext4_xattr_block_cache_insert(ea_block_cache, new_bh); 2103 error = ext4_handle_dirty_metadata(handle, inode, 2104 new_bh); 2105 if (error) 2106 goto cleanup; 2107 } 2108 } 2109 2110 if (old_ea_inode_quota) 2111 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota); 2112 2113 /* Update the inode. */ 2114 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0; 2115 2116 /* Drop the previous xattr block. */ 2117 if (bs->bh && bs->bh != new_bh) { 2118 struct ext4_xattr_inode_array *ea_inode_array = NULL; 2119 2120 ext4_xattr_release_block(handle, inode, bs->bh, 2121 &ea_inode_array, 2122 0 /* extra_credits */); 2123 ext4_xattr_inode_array_free(ea_inode_array); 2124 } 2125 error = 0; 2126 2127 cleanup: 2128 if (ea_inode) { 2129 int error2; 2130 2131 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode); 2132 if (error2) 2133 ext4_warning_inode(ea_inode, "dec ref error=%d", 2134 error2); 2135 2136 /* If there was an error, revert the quota charge. */ 2137 if (error) 2138 ext4_xattr_inode_free_quota(inode, ea_inode, 2139 i_size_read(ea_inode)); 2140 iput(ea_inode); 2141 } 2142 if (ce) 2143 mb_cache_entry_put(ea_block_cache, ce); 2144 brelse(new_bh); 2145 if (!(bs->bh && s->base == bs->bh->b_data)) 2146 kfree(s->base); 2147 2148 return error; 2149 2150 cleanup_dquot: 2151 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1)); 2152 goto cleanup; 2153 2154 bad_block: 2155 EXT4_ERROR_INODE(inode, "bad block %llu", 2156 EXT4_I(inode)->i_file_acl); 2157 goto cleanup; 2158 2159 #undef header 2160 } 2161 2162 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i, 2163 struct ext4_xattr_ibody_find *is) 2164 { 2165 struct ext4_xattr_ibody_header *header; 2166 struct ext4_inode *raw_inode; 2167 int error; 2168 2169 if (EXT4_I(inode)->i_extra_isize == 0) 2170 return 0; 2171 raw_inode = ext4_raw_inode(&is->iloc); 2172 header = IHDR(inode, raw_inode); 2173 is->s.base = is->s.first = IFIRST(header); 2174 is->s.here = is->s.first; 2175 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 2176 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { 2177 error = xattr_check_inode(inode, header, is->s.end); 2178 if (error) 2179 return error; 2180 /* Find the named attribute. */ 2181 error = xattr_find_entry(inode, &is->s.here, is->s.end, 2182 i->name_index, i->name, 0); 2183 if (error && error != -ENODATA) 2184 return error; 2185 is->s.not_found = error; 2186 } 2187 return 0; 2188 } 2189 2190 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode, 2191 struct ext4_xattr_info *i, 2192 struct ext4_xattr_ibody_find *is) 2193 { 2194 struct ext4_xattr_ibody_header *header; 2195 struct ext4_xattr_search *s = &is->s; 2196 int error; 2197 2198 if (EXT4_I(inode)->i_extra_isize == 0) 2199 return -ENOSPC; 2200 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */); 2201 if (error) 2202 return error; 2203 header = IHDR(inode, ext4_raw_inode(&is->iloc)); 2204 if (!IS_LAST_ENTRY(s->first)) { 2205 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC); 2206 ext4_set_inode_state(inode, EXT4_STATE_XATTR); 2207 } else { 2208 header->h_magic = cpu_to_le32(0); 2209 ext4_clear_inode_state(inode, EXT4_STATE_XATTR); 2210 } 2211 return 0; 2212 } 2213 2214 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode, 2215 struct ext4_xattr_info *i, 2216 struct ext4_xattr_ibody_find *is) 2217 { 2218 struct ext4_xattr_ibody_header *header; 2219 struct ext4_xattr_search *s = &is->s; 2220 int error; 2221 2222 if (EXT4_I(inode)->i_extra_isize == 0) 2223 return -ENOSPC; 2224 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */); 2225 if (error) 2226 return error; 2227 header = IHDR(inode, ext4_raw_inode(&is->iloc)); 2228 if (!IS_LAST_ENTRY(s->first)) { 2229 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC); 2230 ext4_set_inode_state(inode, EXT4_STATE_XATTR); 2231 } else { 2232 header->h_magic = cpu_to_le32(0); 2233 ext4_clear_inode_state(inode, EXT4_STATE_XATTR); 2234 } 2235 return 0; 2236 } 2237 2238 static int ext4_xattr_value_same(struct ext4_xattr_search *s, 2239 struct ext4_xattr_info *i) 2240 { 2241 void *value; 2242 2243 /* When e_value_inum is set the value is stored externally. */ 2244 if (s->here->e_value_inum) 2245 return 0; 2246 if (le32_to_cpu(s->here->e_value_size) != i->value_len) 2247 return 0; 2248 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs); 2249 return !memcmp(value, i->value, i->value_len); 2250 } 2251 2252 static struct buffer_head *ext4_xattr_get_block(struct inode *inode) 2253 { 2254 struct buffer_head *bh; 2255 int error; 2256 2257 if (!EXT4_I(inode)->i_file_acl) 2258 return NULL; 2259 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 2260 if (IS_ERR(bh)) 2261 return bh; 2262 error = ext4_xattr_check_block(inode, bh); 2263 if (error) { 2264 brelse(bh); 2265 return ERR_PTR(error); 2266 } 2267 return bh; 2268 } 2269 2270 /* 2271 * ext4_xattr_set_handle() 2272 * 2273 * Create, replace or remove an extended attribute for this inode. Value 2274 * is NULL to remove an existing extended attribute, and non-NULL to 2275 * either replace an existing extended attribute, or create a new extended 2276 * attribute. The flags XATTR_REPLACE and XATTR_CREATE 2277 * specify that an extended attribute must exist and must not exist 2278 * previous to the call, respectively. 2279 * 2280 * Returns 0, or a negative error number on failure. 2281 */ 2282 int 2283 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index, 2284 const char *name, const void *value, size_t value_len, 2285 int flags) 2286 { 2287 struct ext4_xattr_info i = { 2288 .name_index = name_index, 2289 .name = name, 2290 .value = value, 2291 .value_len = value_len, 2292 .in_inode = 0, 2293 }; 2294 struct ext4_xattr_ibody_find is = { 2295 .s = { .not_found = -ENODATA, }, 2296 }; 2297 struct ext4_xattr_block_find bs = { 2298 .s = { .not_found = -ENODATA, }, 2299 }; 2300 int no_expand; 2301 int error; 2302 2303 if (!name) 2304 return -EINVAL; 2305 if (strlen(name) > 255) 2306 return -ERANGE; 2307 2308 ext4_write_lock_xattr(inode, &no_expand); 2309 2310 /* Check journal credits under write lock. */ 2311 if (ext4_handle_valid(handle)) { 2312 struct buffer_head *bh; 2313 int credits; 2314 2315 bh = ext4_xattr_get_block(inode); 2316 if (IS_ERR(bh)) { 2317 error = PTR_ERR(bh); 2318 goto cleanup; 2319 } 2320 2321 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh, 2322 value_len, 2323 flags & XATTR_CREATE); 2324 brelse(bh); 2325 2326 if (jbd2_handle_buffer_credits(handle) < credits) { 2327 error = -ENOSPC; 2328 goto cleanup; 2329 } 2330 } 2331 2332 error = ext4_reserve_inode_write(handle, inode, &is.iloc); 2333 if (error) 2334 goto cleanup; 2335 2336 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) { 2337 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc); 2338 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); 2339 ext4_clear_inode_state(inode, EXT4_STATE_NEW); 2340 } 2341 2342 error = ext4_xattr_ibody_find(inode, &i, &is); 2343 if (error) 2344 goto cleanup; 2345 if (is.s.not_found) 2346 error = ext4_xattr_block_find(inode, &i, &bs); 2347 if (error) 2348 goto cleanup; 2349 if (is.s.not_found && bs.s.not_found) { 2350 error = -ENODATA; 2351 if (flags & XATTR_REPLACE) 2352 goto cleanup; 2353 error = 0; 2354 if (!value) 2355 goto cleanup; 2356 } else { 2357 error = -EEXIST; 2358 if (flags & XATTR_CREATE) 2359 goto cleanup; 2360 } 2361 2362 if (!value) { 2363 if (!is.s.not_found) 2364 error = ext4_xattr_ibody_set(handle, inode, &i, &is); 2365 else if (!bs.s.not_found) 2366 error = ext4_xattr_block_set(handle, inode, &i, &bs); 2367 } else { 2368 error = 0; 2369 /* Xattr value did not change? Save us some work and bail out */ 2370 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i)) 2371 goto cleanup; 2372 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i)) 2373 goto cleanup; 2374 2375 if (ext4_has_feature_ea_inode(inode->i_sb) && 2376 (EXT4_XATTR_SIZE(i.value_len) > 2377 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize))) 2378 i.in_inode = 1; 2379 retry_inode: 2380 error = ext4_xattr_ibody_set(handle, inode, &i, &is); 2381 if (!error && !bs.s.not_found) { 2382 i.value = NULL; 2383 error = ext4_xattr_block_set(handle, inode, &i, &bs); 2384 } else if (error == -ENOSPC) { 2385 if (EXT4_I(inode)->i_file_acl && !bs.s.base) { 2386 brelse(bs.bh); 2387 bs.bh = NULL; 2388 error = ext4_xattr_block_find(inode, &i, &bs); 2389 if (error) 2390 goto cleanup; 2391 } 2392 error = ext4_xattr_block_set(handle, inode, &i, &bs); 2393 if (!error && !is.s.not_found) { 2394 i.value = NULL; 2395 error = ext4_xattr_ibody_set(handle, inode, &i, 2396 &is); 2397 } else if (error == -ENOSPC) { 2398 /* 2399 * Xattr does not fit in the block, store at 2400 * external inode if possible. 2401 */ 2402 if (ext4_has_feature_ea_inode(inode->i_sb) && 2403 !i.in_inode) { 2404 i.in_inode = 1; 2405 goto retry_inode; 2406 } 2407 } 2408 } 2409 } 2410 if (!error) { 2411 ext4_xattr_update_super_block(handle, inode->i_sb); 2412 inode->i_ctime = current_time(inode); 2413 if (!value) 2414 no_expand = 0; 2415 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); 2416 /* 2417 * The bh is consumed by ext4_mark_iloc_dirty, even with 2418 * error != 0. 2419 */ 2420 is.iloc.bh = NULL; 2421 if (IS_SYNC(inode)) 2422 ext4_handle_sync(handle); 2423 } 2424 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR); 2425 2426 cleanup: 2427 brelse(is.iloc.bh); 2428 brelse(bs.bh); 2429 ext4_write_unlock_xattr(inode, &no_expand); 2430 return error; 2431 } 2432 2433 int ext4_xattr_set_credits(struct inode *inode, size_t value_len, 2434 bool is_create, int *credits) 2435 { 2436 struct buffer_head *bh; 2437 int err; 2438 2439 *credits = 0; 2440 2441 if (!EXT4_SB(inode->i_sb)->s_journal) 2442 return 0; 2443 2444 down_read(&EXT4_I(inode)->xattr_sem); 2445 2446 bh = ext4_xattr_get_block(inode); 2447 if (IS_ERR(bh)) { 2448 err = PTR_ERR(bh); 2449 } else { 2450 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh, 2451 value_len, is_create); 2452 brelse(bh); 2453 err = 0; 2454 } 2455 2456 up_read(&EXT4_I(inode)->xattr_sem); 2457 return err; 2458 } 2459 2460 /* 2461 * ext4_xattr_set() 2462 * 2463 * Like ext4_xattr_set_handle, but start from an inode. This extended 2464 * attribute modification is a filesystem transaction by itself. 2465 * 2466 * Returns 0, or a negative error number on failure. 2467 */ 2468 int 2469 ext4_xattr_set(struct inode *inode, int name_index, const char *name, 2470 const void *value, size_t value_len, int flags) 2471 { 2472 handle_t *handle; 2473 struct super_block *sb = inode->i_sb; 2474 int error, retries = 0; 2475 int credits; 2476 2477 error = dquot_initialize(inode); 2478 if (error) 2479 return error; 2480 2481 retry: 2482 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE, 2483 &credits); 2484 if (error) 2485 return error; 2486 2487 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits); 2488 if (IS_ERR(handle)) { 2489 error = PTR_ERR(handle); 2490 } else { 2491 int error2; 2492 2493 error = ext4_xattr_set_handle(handle, inode, name_index, name, 2494 value, value_len, flags); 2495 error2 = ext4_journal_stop(handle); 2496 if (error == -ENOSPC && 2497 ext4_should_retry_alloc(sb, &retries)) 2498 goto retry; 2499 if (error == 0) 2500 error = error2; 2501 } 2502 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR); 2503 2504 return error; 2505 } 2506 2507 /* 2508 * Shift the EA entries in the inode to create space for the increased 2509 * i_extra_isize. 2510 */ 2511 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry, 2512 int value_offs_shift, void *to, 2513 void *from, size_t n) 2514 { 2515 struct ext4_xattr_entry *last = entry; 2516 int new_offs; 2517 2518 /* We always shift xattr headers further thus offsets get lower */ 2519 BUG_ON(value_offs_shift > 0); 2520 2521 /* Adjust the value offsets of the entries */ 2522 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { 2523 if (!last->e_value_inum && last->e_value_size) { 2524 new_offs = le16_to_cpu(last->e_value_offs) + 2525 value_offs_shift; 2526 last->e_value_offs = cpu_to_le16(new_offs); 2527 } 2528 } 2529 /* Shift the entries by n bytes */ 2530 memmove(to, from, n); 2531 } 2532 2533 /* 2534 * Move xattr pointed to by 'entry' from inode into external xattr block 2535 */ 2536 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode, 2537 struct ext4_inode *raw_inode, 2538 struct ext4_xattr_entry *entry) 2539 { 2540 struct ext4_xattr_ibody_find *is = NULL; 2541 struct ext4_xattr_block_find *bs = NULL; 2542 char *buffer = NULL, *b_entry_name = NULL; 2543 size_t value_size = le32_to_cpu(entry->e_value_size); 2544 struct ext4_xattr_info i = { 2545 .value = NULL, 2546 .value_len = 0, 2547 .name_index = entry->e_name_index, 2548 .in_inode = !!entry->e_value_inum, 2549 }; 2550 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode); 2551 int error; 2552 2553 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS); 2554 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS); 2555 buffer = kmalloc(value_size, GFP_NOFS); 2556 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS); 2557 if (!is || !bs || !buffer || !b_entry_name) { 2558 error = -ENOMEM; 2559 goto out; 2560 } 2561 2562 is->s.not_found = -ENODATA; 2563 bs->s.not_found = -ENODATA; 2564 is->iloc.bh = NULL; 2565 bs->bh = NULL; 2566 2567 /* Save the entry name and the entry value */ 2568 if (entry->e_value_inum) { 2569 error = ext4_xattr_inode_get(inode, entry, buffer, value_size); 2570 if (error) 2571 goto out; 2572 } else { 2573 size_t value_offs = le16_to_cpu(entry->e_value_offs); 2574 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size); 2575 } 2576 2577 memcpy(b_entry_name, entry->e_name, entry->e_name_len); 2578 b_entry_name[entry->e_name_len] = '\0'; 2579 i.name = b_entry_name; 2580 2581 error = ext4_get_inode_loc(inode, &is->iloc); 2582 if (error) 2583 goto out; 2584 2585 error = ext4_xattr_ibody_find(inode, &i, is); 2586 if (error) 2587 goto out; 2588 2589 /* Remove the chosen entry from the inode */ 2590 error = ext4_xattr_ibody_set(handle, inode, &i, is); 2591 if (error) 2592 goto out; 2593 2594 i.value = buffer; 2595 i.value_len = value_size; 2596 error = ext4_xattr_block_find(inode, &i, bs); 2597 if (error) 2598 goto out; 2599 2600 /* Add entry which was removed from the inode into the block */ 2601 error = ext4_xattr_block_set(handle, inode, &i, bs); 2602 if (error) 2603 goto out; 2604 error = 0; 2605 out: 2606 kfree(b_entry_name); 2607 kfree(buffer); 2608 if (is) 2609 brelse(is->iloc.bh); 2610 if (bs) 2611 brelse(bs->bh); 2612 kfree(is); 2613 kfree(bs); 2614 2615 return error; 2616 } 2617 2618 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode, 2619 struct ext4_inode *raw_inode, 2620 int isize_diff, size_t ifree, 2621 size_t bfree, int *total_ino) 2622 { 2623 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode); 2624 struct ext4_xattr_entry *small_entry; 2625 struct ext4_xattr_entry *entry; 2626 struct ext4_xattr_entry *last; 2627 unsigned int entry_size; /* EA entry size */ 2628 unsigned int total_size; /* EA entry size + value size */ 2629 unsigned int min_total_size; 2630 int error; 2631 2632 while (isize_diff > ifree) { 2633 entry = NULL; 2634 small_entry = NULL; 2635 min_total_size = ~0U; 2636 last = IFIRST(header); 2637 /* Find the entry best suited to be pushed into EA block */ 2638 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { 2639 /* never move system.data out of the inode */ 2640 if ((last->e_name_len == 4) && 2641 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) && 2642 !memcmp(last->e_name, "data", 4)) 2643 continue; 2644 total_size = EXT4_XATTR_LEN(last->e_name_len); 2645 if (!last->e_value_inum) 2646 total_size += EXT4_XATTR_SIZE( 2647 le32_to_cpu(last->e_value_size)); 2648 if (total_size <= bfree && 2649 total_size < min_total_size) { 2650 if (total_size + ifree < isize_diff) { 2651 small_entry = last; 2652 } else { 2653 entry = last; 2654 min_total_size = total_size; 2655 } 2656 } 2657 } 2658 2659 if (entry == NULL) { 2660 if (small_entry == NULL) 2661 return -ENOSPC; 2662 entry = small_entry; 2663 } 2664 2665 entry_size = EXT4_XATTR_LEN(entry->e_name_len); 2666 total_size = entry_size; 2667 if (!entry->e_value_inum) 2668 total_size += EXT4_XATTR_SIZE( 2669 le32_to_cpu(entry->e_value_size)); 2670 error = ext4_xattr_move_to_block(handle, inode, raw_inode, 2671 entry); 2672 if (error) 2673 return error; 2674 2675 *total_ino -= entry_size; 2676 ifree += total_size; 2677 bfree -= total_size; 2678 } 2679 2680 return 0; 2681 } 2682 2683 /* 2684 * Expand an inode by new_extra_isize bytes when EAs are present. 2685 * Returns 0 on success or negative error number on failure. 2686 */ 2687 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize, 2688 struct ext4_inode *raw_inode, handle_t *handle) 2689 { 2690 struct ext4_xattr_ibody_header *header; 2691 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 2692 static unsigned int mnt_count; 2693 size_t min_offs; 2694 size_t ifree, bfree; 2695 int total_ino; 2696 void *base, *end; 2697 int error = 0, tried_min_extra_isize = 0; 2698 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize); 2699 int isize_diff; /* How much do we need to grow i_extra_isize */ 2700 2701 retry: 2702 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize; 2703 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) 2704 return 0; 2705 2706 header = IHDR(inode, raw_inode); 2707 2708 /* 2709 * Check if enough free space is available in the inode to shift the 2710 * entries ahead by new_extra_isize. 2711 */ 2712 2713 base = IFIRST(header); 2714 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 2715 min_offs = end - base; 2716 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32); 2717 2718 error = xattr_check_inode(inode, header, end); 2719 if (error) 2720 goto cleanup; 2721 2722 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino); 2723 if (ifree >= isize_diff) 2724 goto shift; 2725 2726 /* 2727 * Enough free space isn't available in the inode, check if 2728 * EA block can hold new_extra_isize bytes. 2729 */ 2730 if (EXT4_I(inode)->i_file_acl) { 2731 struct buffer_head *bh; 2732 2733 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 2734 if (IS_ERR(bh)) { 2735 error = PTR_ERR(bh); 2736 goto cleanup; 2737 } 2738 error = ext4_xattr_check_block(inode, bh); 2739 if (error) { 2740 brelse(bh); 2741 goto cleanup; 2742 } 2743 base = BHDR(bh); 2744 end = bh->b_data + bh->b_size; 2745 min_offs = end - base; 2746 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base, 2747 NULL); 2748 brelse(bh); 2749 if (bfree + ifree < isize_diff) { 2750 if (!tried_min_extra_isize && s_min_extra_isize) { 2751 tried_min_extra_isize++; 2752 new_extra_isize = s_min_extra_isize; 2753 goto retry; 2754 } 2755 error = -ENOSPC; 2756 goto cleanup; 2757 } 2758 } else { 2759 bfree = inode->i_sb->s_blocksize; 2760 } 2761 2762 error = ext4_xattr_make_inode_space(handle, inode, raw_inode, 2763 isize_diff, ifree, bfree, 2764 &total_ino); 2765 if (error) { 2766 if (error == -ENOSPC && !tried_min_extra_isize && 2767 s_min_extra_isize) { 2768 tried_min_extra_isize++; 2769 new_extra_isize = s_min_extra_isize; 2770 goto retry; 2771 } 2772 goto cleanup; 2773 } 2774 shift: 2775 /* Adjust the offsets and shift the remaining entries ahead */ 2776 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize 2777 - new_extra_isize, (void *)raw_inode + 2778 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize, 2779 (void *)header, total_ino); 2780 EXT4_I(inode)->i_extra_isize = new_extra_isize; 2781 2782 cleanup: 2783 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) { 2784 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.", 2785 inode->i_ino); 2786 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count); 2787 } 2788 return error; 2789 } 2790 2791 #define EIA_INCR 16 /* must be 2^n */ 2792 #define EIA_MASK (EIA_INCR - 1) 2793 2794 /* Add the large xattr @inode into @ea_inode_array for deferred iput(). 2795 * If @ea_inode_array is new or full it will be grown and the old 2796 * contents copied over. 2797 */ 2798 static int 2799 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array, 2800 struct inode *inode) 2801 { 2802 if (*ea_inode_array == NULL) { 2803 /* 2804 * Start with 15 inodes, so it fits into a power-of-two size. 2805 * If *ea_inode_array is NULL, this is essentially offsetof() 2806 */ 2807 (*ea_inode_array) = 2808 kmalloc(offsetof(struct ext4_xattr_inode_array, 2809 inodes[EIA_MASK]), 2810 GFP_NOFS); 2811 if (*ea_inode_array == NULL) 2812 return -ENOMEM; 2813 (*ea_inode_array)->count = 0; 2814 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) { 2815 /* expand the array once all 15 + n * 16 slots are full */ 2816 struct ext4_xattr_inode_array *new_array = NULL; 2817 int count = (*ea_inode_array)->count; 2818 2819 /* if new_array is NULL, this is essentially offsetof() */ 2820 new_array = kmalloc( 2821 offsetof(struct ext4_xattr_inode_array, 2822 inodes[count + EIA_INCR]), 2823 GFP_NOFS); 2824 if (new_array == NULL) 2825 return -ENOMEM; 2826 memcpy(new_array, *ea_inode_array, 2827 offsetof(struct ext4_xattr_inode_array, inodes[count])); 2828 kfree(*ea_inode_array); 2829 *ea_inode_array = new_array; 2830 } 2831 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode; 2832 return 0; 2833 } 2834 2835 /* 2836 * ext4_xattr_delete_inode() 2837 * 2838 * Free extended attribute resources associated with this inode. Traverse 2839 * all entries and decrement reference on any xattr inodes associated with this 2840 * inode. This is called immediately before an inode is freed. We have exclusive 2841 * access to the inode. If an orphan inode is deleted it will also release its 2842 * references on xattr block and xattr inodes. 2843 */ 2844 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode, 2845 struct ext4_xattr_inode_array **ea_inode_array, 2846 int extra_credits) 2847 { 2848 struct buffer_head *bh = NULL; 2849 struct ext4_xattr_ibody_header *header; 2850 struct ext4_iloc iloc = { .bh = NULL }; 2851 struct ext4_xattr_entry *entry; 2852 struct inode *ea_inode; 2853 int error; 2854 2855 error = ext4_journal_ensure_credits(handle, extra_credits, 2856 ext4_free_metadata_revoke_credits(inode->i_sb, 1)); 2857 if (error < 0) { 2858 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error); 2859 goto cleanup; 2860 } 2861 2862 if (ext4_has_feature_ea_inode(inode->i_sb) && 2863 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { 2864 2865 error = ext4_get_inode_loc(inode, &iloc); 2866 if (error) { 2867 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error); 2868 goto cleanup; 2869 } 2870 2871 error = ext4_journal_get_write_access(handle, iloc.bh); 2872 if (error) { 2873 EXT4_ERROR_INODE(inode, "write access (error %d)", 2874 error); 2875 goto cleanup; 2876 } 2877 2878 header = IHDR(inode, ext4_raw_inode(&iloc)); 2879 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC)) 2880 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh, 2881 IFIRST(header), 2882 false /* block_csum */, 2883 ea_inode_array, 2884 extra_credits, 2885 false /* skip_quota */); 2886 } 2887 2888 if (EXT4_I(inode)->i_file_acl) { 2889 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 2890 if (IS_ERR(bh)) { 2891 error = PTR_ERR(bh); 2892 if (error == -EIO) { 2893 EXT4_ERROR_INODE_ERR(inode, EIO, 2894 "block %llu read error", 2895 EXT4_I(inode)->i_file_acl); 2896 } 2897 bh = NULL; 2898 goto cleanup; 2899 } 2900 error = ext4_xattr_check_block(inode, bh); 2901 if (error) 2902 goto cleanup; 2903 2904 if (ext4_has_feature_ea_inode(inode->i_sb)) { 2905 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry); 2906 entry = EXT4_XATTR_NEXT(entry)) { 2907 if (!entry->e_value_inum) 2908 continue; 2909 error = ext4_xattr_inode_iget(inode, 2910 le32_to_cpu(entry->e_value_inum), 2911 le32_to_cpu(entry->e_hash), 2912 &ea_inode); 2913 if (error) 2914 continue; 2915 ext4_xattr_inode_free_quota(inode, ea_inode, 2916 le32_to_cpu(entry->e_value_size)); 2917 iput(ea_inode); 2918 } 2919 2920 } 2921 2922 ext4_xattr_release_block(handle, inode, bh, ea_inode_array, 2923 extra_credits); 2924 /* 2925 * Update i_file_acl value in the same transaction that releases 2926 * block. 2927 */ 2928 EXT4_I(inode)->i_file_acl = 0; 2929 error = ext4_mark_inode_dirty(handle, inode); 2930 if (error) { 2931 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)", 2932 error); 2933 goto cleanup; 2934 } 2935 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR); 2936 } 2937 error = 0; 2938 cleanup: 2939 brelse(iloc.bh); 2940 brelse(bh); 2941 return error; 2942 } 2943 2944 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array) 2945 { 2946 int idx; 2947 2948 if (ea_inode_array == NULL) 2949 return; 2950 2951 for (idx = 0; idx < ea_inode_array->count; ++idx) 2952 iput(ea_inode_array->inodes[idx]); 2953 kfree(ea_inode_array); 2954 } 2955 2956 /* 2957 * ext4_xattr_block_cache_insert() 2958 * 2959 * Create a new entry in the extended attribute block cache, and insert 2960 * it unless such an entry is already in the cache. 2961 * 2962 * Returns 0, or a negative error number on failure. 2963 */ 2964 static void 2965 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache, 2966 struct buffer_head *bh) 2967 { 2968 struct ext4_xattr_header *header = BHDR(bh); 2969 __u32 hash = le32_to_cpu(header->h_hash); 2970 int reusable = le32_to_cpu(header->h_refcount) < 2971 EXT4_XATTR_REFCOUNT_MAX; 2972 int error; 2973 2974 if (!ea_block_cache) 2975 return; 2976 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash, 2977 bh->b_blocknr, reusable); 2978 if (error) { 2979 if (error == -EBUSY) 2980 ea_bdebug(bh, "already in cache"); 2981 } else 2982 ea_bdebug(bh, "inserting [%x]", (int)hash); 2983 } 2984 2985 /* 2986 * ext4_xattr_cmp() 2987 * 2988 * Compare two extended attribute blocks for equality. 2989 * 2990 * Returns 0 if the blocks are equal, 1 if they differ, and 2991 * a negative error number on errors. 2992 */ 2993 static int 2994 ext4_xattr_cmp(struct ext4_xattr_header *header1, 2995 struct ext4_xattr_header *header2) 2996 { 2997 struct ext4_xattr_entry *entry1, *entry2; 2998 2999 entry1 = ENTRY(header1+1); 3000 entry2 = ENTRY(header2+1); 3001 while (!IS_LAST_ENTRY(entry1)) { 3002 if (IS_LAST_ENTRY(entry2)) 3003 return 1; 3004 if (entry1->e_hash != entry2->e_hash || 3005 entry1->e_name_index != entry2->e_name_index || 3006 entry1->e_name_len != entry2->e_name_len || 3007 entry1->e_value_size != entry2->e_value_size || 3008 entry1->e_value_inum != entry2->e_value_inum || 3009 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len)) 3010 return 1; 3011 if (!entry1->e_value_inum && 3012 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs), 3013 (char *)header2 + le16_to_cpu(entry2->e_value_offs), 3014 le32_to_cpu(entry1->e_value_size))) 3015 return 1; 3016 3017 entry1 = EXT4_XATTR_NEXT(entry1); 3018 entry2 = EXT4_XATTR_NEXT(entry2); 3019 } 3020 if (!IS_LAST_ENTRY(entry2)) 3021 return 1; 3022 return 0; 3023 } 3024 3025 /* 3026 * ext4_xattr_block_cache_find() 3027 * 3028 * Find an identical extended attribute block. 3029 * 3030 * Returns a pointer to the block found, or NULL if such a block was 3031 * not found or an error occurred. 3032 */ 3033 static struct buffer_head * 3034 ext4_xattr_block_cache_find(struct inode *inode, 3035 struct ext4_xattr_header *header, 3036 struct mb_cache_entry **pce) 3037 { 3038 __u32 hash = le32_to_cpu(header->h_hash); 3039 struct mb_cache_entry *ce; 3040 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 3041 3042 if (!ea_block_cache) 3043 return NULL; 3044 if (!header->h_hash) 3045 return NULL; /* never share */ 3046 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash); 3047 ce = mb_cache_entry_find_first(ea_block_cache, hash); 3048 while (ce) { 3049 struct buffer_head *bh; 3050 3051 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO); 3052 if (IS_ERR(bh)) { 3053 if (PTR_ERR(bh) == -ENOMEM) 3054 return NULL; 3055 bh = NULL; 3056 EXT4_ERROR_INODE(inode, "block %lu read error", 3057 (unsigned long)ce->e_value); 3058 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) { 3059 *pce = ce; 3060 return bh; 3061 } 3062 brelse(bh); 3063 ce = mb_cache_entry_find_next(ea_block_cache, ce); 3064 } 3065 return NULL; 3066 } 3067 3068 #define NAME_HASH_SHIFT 5 3069 #define VALUE_HASH_SHIFT 16 3070 3071 /* 3072 * ext4_xattr_hash_entry() 3073 * 3074 * Compute the hash of an extended attribute. 3075 */ 3076 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value, 3077 size_t value_count) 3078 { 3079 __u32 hash = 0; 3080 3081 while (name_len--) { 3082 hash = (hash << NAME_HASH_SHIFT) ^ 3083 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^ 3084 *name++; 3085 } 3086 while (value_count--) { 3087 hash = (hash << VALUE_HASH_SHIFT) ^ 3088 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^ 3089 le32_to_cpu(*value++); 3090 } 3091 return cpu_to_le32(hash); 3092 } 3093 3094 #undef NAME_HASH_SHIFT 3095 #undef VALUE_HASH_SHIFT 3096 3097 #define BLOCK_HASH_SHIFT 16 3098 3099 /* 3100 * ext4_xattr_rehash() 3101 * 3102 * Re-compute the extended attribute hash value after an entry has changed. 3103 */ 3104 static void ext4_xattr_rehash(struct ext4_xattr_header *header) 3105 { 3106 struct ext4_xattr_entry *here; 3107 __u32 hash = 0; 3108 3109 here = ENTRY(header+1); 3110 while (!IS_LAST_ENTRY(here)) { 3111 if (!here->e_hash) { 3112 /* Block is not shared if an entry's hash value == 0 */ 3113 hash = 0; 3114 break; 3115 } 3116 hash = (hash << BLOCK_HASH_SHIFT) ^ 3117 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^ 3118 le32_to_cpu(here->e_hash); 3119 here = EXT4_XATTR_NEXT(here); 3120 } 3121 header->h_hash = cpu_to_le32(hash); 3122 } 3123 3124 #undef BLOCK_HASH_SHIFT 3125 3126 #define HASH_BUCKET_BITS 10 3127 3128 struct mb_cache * 3129 ext4_xattr_create_cache(void) 3130 { 3131 return mb_cache_create(HASH_BUCKET_BITS); 3132 } 3133 3134 void ext4_xattr_destroy_cache(struct mb_cache *cache) 3135 { 3136 if (cache) 3137 mb_cache_destroy(cache); 3138 } 3139 3140