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 }; 97 98 const struct xattr_handler *ext4_xattr_handlers[] = { 99 &ext4_xattr_user_handler, 100 &ext4_xattr_trusted_handler, 101 #ifdef CONFIG_EXT4_FS_POSIX_ACL 102 &posix_acl_access_xattr_handler, 103 &posix_acl_default_xattr_handler, 104 #endif 105 #ifdef CONFIG_EXT4_FS_SECURITY 106 &ext4_xattr_security_handler, 107 #endif 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) 116 117 static int 118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array, 119 struct inode *inode); 120 121 #ifdef CONFIG_LOCKDEP 122 void ext4_xattr_inode_set_class(struct inode *ea_inode) 123 { 124 lockdep_set_subclass(&ea_inode->i_rwsem, 1); 125 } 126 #endif 127 128 static __le32 ext4_xattr_block_csum(struct inode *inode, 129 sector_t block_nr, 130 struct ext4_xattr_header *hdr) 131 { 132 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 133 __u32 csum; 134 __le64 dsk_block_nr = cpu_to_le64(block_nr); 135 __u32 dummy_csum = 0; 136 int offset = offsetof(struct ext4_xattr_header, h_checksum); 137 138 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr, 139 sizeof(dsk_block_nr)); 140 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset); 141 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum)); 142 offset += sizeof(dummy_csum); 143 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset, 144 EXT4_BLOCK_SIZE(inode->i_sb) - offset); 145 146 return cpu_to_le32(csum); 147 } 148 149 static int ext4_xattr_block_csum_verify(struct inode *inode, 150 struct buffer_head *bh) 151 { 152 struct ext4_xattr_header *hdr = BHDR(bh); 153 int ret = 1; 154 155 if (ext4_has_metadata_csum(inode->i_sb)) { 156 lock_buffer(bh); 157 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode, 158 bh->b_blocknr, hdr)); 159 unlock_buffer(bh); 160 } 161 return ret; 162 } 163 164 static void ext4_xattr_block_csum_set(struct inode *inode, 165 struct buffer_head *bh) 166 { 167 if (ext4_has_metadata_csum(inode->i_sb)) 168 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode, 169 bh->b_blocknr, BHDR(bh)); 170 } 171 172 static inline const struct xattr_handler * 173 ext4_xattr_handler(int name_index) 174 { 175 const struct xattr_handler *handler = NULL; 176 177 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map)) 178 handler = ext4_xattr_handler_map[name_index]; 179 return handler; 180 } 181 182 static int 183 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end, 184 void *value_start) 185 { 186 struct ext4_xattr_entry *e = entry; 187 188 /* Find the end of the names list */ 189 while (!IS_LAST_ENTRY(e)) { 190 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e); 191 if ((void *)next >= end) 192 return -EFSCORRUPTED; 193 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len) 194 return -EFSCORRUPTED; 195 e = next; 196 } 197 198 /* Check the values */ 199 while (!IS_LAST_ENTRY(entry)) { 200 u32 size = le32_to_cpu(entry->e_value_size); 201 202 if (size > EXT4_XATTR_SIZE_MAX) 203 return -EFSCORRUPTED; 204 205 if (size != 0 && entry->e_value_inum == 0) { 206 u16 offs = le16_to_cpu(entry->e_value_offs); 207 void *value; 208 209 /* 210 * The value cannot overlap the names, and the value 211 * with padding cannot extend beyond 'end'. Check both 212 * the padded and unpadded sizes, since the size may 213 * overflow to 0 when adding padding. 214 */ 215 if (offs > end - value_start) 216 return -EFSCORRUPTED; 217 value = value_start + offs; 218 if (value < (void *)e + sizeof(u32) || 219 size > end - value || 220 EXT4_XATTR_SIZE(size) > end - value) 221 return -EFSCORRUPTED; 222 } 223 entry = EXT4_XATTR_NEXT(entry); 224 } 225 226 return 0; 227 } 228 229 static inline int 230 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh, 231 const char *function, unsigned int line) 232 { 233 int error = -EFSCORRUPTED; 234 235 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) || 236 BHDR(bh)->h_blocks != cpu_to_le32(1)) 237 goto errout; 238 if (buffer_verified(bh)) 239 return 0; 240 241 error = -EFSBADCRC; 242 if (!ext4_xattr_block_csum_verify(inode, bh)) 243 goto errout; 244 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size, 245 bh->b_data); 246 errout: 247 if (error) 248 __ext4_error_inode(inode, function, line, 0, 249 "corrupted xattr block %llu", 250 (unsigned long long) bh->b_blocknr); 251 else 252 set_buffer_verified(bh); 253 return error; 254 } 255 256 #define ext4_xattr_check_block(inode, bh) \ 257 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__) 258 259 260 static int 261 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header, 262 void *end, const char *function, unsigned int line) 263 { 264 int error = -EFSCORRUPTED; 265 266 if (end - (void *)header < sizeof(*header) + sizeof(u32) || 267 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC))) 268 goto errout; 269 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header)); 270 errout: 271 if (error) 272 __ext4_error_inode(inode, function, line, 0, 273 "corrupted in-inode xattr"); 274 return error; 275 } 276 277 #define xattr_check_inode(inode, header, end) \ 278 __xattr_check_inode((inode), (header), (end), __func__, __LINE__) 279 280 static int 281 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry, 282 void *end, int name_index, const char *name, int sorted) 283 { 284 struct ext4_xattr_entry *entry, *next; 285 size_t name_len; 286 int cmp = 1; 287 288 if (name == NULL) 289 return -EINVAL; 290 name_len = strlen(name); 291 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) { 292 next = EXT4_XATTR_NEXT(entry); 293 if ((void *) next >= end) { 294 EXT4_ERROR_INODE(inode, "corrupted xattr entries"); 295 return -EFSCORRUPTED; 296 } 297 cmp = name_index - entry->e_name_index; 298 if (!cmp) 299 cmp = name_len - entry->e_name_len; 300 if (!cmp) 301 cmp = memcmp(name, entry->e_name, name_len); 302 if (cmp <= 0 && (sorted || cmp == 0)) 303 break; 304 } 305 *pentry = entry; 306 return cmp ? -ENODATA : 0; 307 } 308 309 static u32 310 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size) 311 { 312 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size); 313 } 314 315 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode) 316 { 317 return ((u64)ea_inode->i_ctime.tv_sec << 32) | 318 (u32) inode_peek_iversion_raw(ea_inode); 319 } 320 321 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count) 322 { 323 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32); 324 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff); 325 } 326 327 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode) 328 { 329 return (u32)ea_inode->i_atime.tv_sec; 330 } 331 332 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash) 333 { 334 ea_inode->i_atime.tv_sec = hash; 335 } 336 337 /* 338 * Read the EA value from an inode. 339 */ 340 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size) 341 { 342 int blocksize = 1 << ea_inode->i_blkbits; 343 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits; 344 int tail_size = (size % blocksize) ?: blocksize; 345 struct buffer_head *bhs_inline[8]; 346 struct buffer_head **bhs = bhs_inline; 347 int i, ret; 348 349 if (bh_count > ARRAY_SIZE(bhs_inline)) { 350 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS); 351 if (!bhs) 352 return -ENOMEM; 353 } 354 355 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count, 356 true /* wait */, bhs); 357 if (ret) 358 goto free_bhs; 359 360 for (i = 0; i < bh_count; i++) { 361 /* There shouldn't be any holes in ea_inode. */ 362 if (!bhs[i]) { 363 ret = -EFSCORRUPTED; 364 goto put_bhs; 365 } 366 memcpy((char *)buf + blocksize * i, bhs[i]->b_data, 367 i < bh_count - 1 ? blocksize : tail_size); 368 } 369 ret = 0; 370 put_bhs: 371 for (i = 0; i < bh_count; i++) 372 brelse(bhs[i]); 373 free_bhs: 374 if (bhs != bhs_inline) 375 kfree(bhs); 376 return ret; 377 } 378 379 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec) 380 381 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino, 382 u32 ea_inode_hash, struct inode **ea_inode) 383 { 384 struct inode *inode; 385 int err; 386 387 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL); 388 if (IS_ERR(inode)) { 389 err = PTR_ERR(inode); 390 ext4_error(parent->i_sb, 391 "error while reading EA inode %lu err=%d", ea_ino, 392 err); 393 return err; 394 } 395 396 if (is_bad_inode(inode)) { 397 ext4_error(parent->i_sb, 398 "error while reading EA inode %lu is_bad_inode", 399 ea_ino); 400 err = -EIO; 401 goto error; 402 } 403 404 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { 405 ext4_error(parent->i_sb, 406 "EA inode %lu does not have EXT4_EA_INODE_FL flag", 407 ea_ino); 408 err = -EINVAL; 409 goto error; 410 } 411 412 ext4_xattr_inode_set_class(inode); 413 414 /* 415 * Check whether this is an old Lustre-style xattr inode. Lustre 416 * implementation does not have hash validation, rather it has a 417 * backpointer from ea_inode to the parent inode. 418 */ 419 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) && 420 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino && 421 inode->i_generation == parent->i_generation) { 422 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE); 423 ext4_xattr_inode_set_ref(inode, 1); 424 } else { 425 inode_lock(inode); 426 inode->i_flags |= S_NOQUOTA; 427 inode_unlock(inode); 428 } 429 430 *ea_inode = inode; 431 return 0; 432 error: 433 iput(inode); 434 return err; 435 } 436 437 static int 438 ext4_xattr_inode_verify_hashes(struct inode *ea_inode, 439 struct ext4_xattr_entry *entry, void *buffer, 440 size_t size) 441 { 442 u32 hash; 443 444 /* Verify stored hash matches calculated hash. */ 445 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size); 446 if (hash != ext4_xattr_inode_get_hash(ea_inode)) 447 return -EFSCORRUPTED; 448 449 if (entry) { 450 __le32 e_hash, tmp_data; 451 452 /* Verify entry hash. */ 453 tmp_data = cpu_to_le32(hash); 454 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len, 455 &tmp_data, 1); 456 if (e_hash != entry->e_hash) 457 return -EFSCORRUPTED; 458 } 459 return 0; 460 } 461 462 /* 463 * Read xattr value from the EA inode. 464 */ 465 static int 466 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry, 467 void *buffer, size_t size) 468 { 469 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode); 470 struct inode *ea_inode; 471 int err; 472 473 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum), 474 le32_to_cpu(entry->e_hash), &ea_inode); 475 if (err) { 476 ea_inode = NULL; 477 goto out; 478 } 479 480 if (i_size_read(ea_inode) != size) { 481 ext4_warning_inode(ea_inode, 482 "ea_inode file size=%llu entry size=%zu", 483 i_size_read(ea_inode), size); 484 err = -EFSCORRUPTED; 485 goto out; 486 } 487 488 err = ext4_xattr_inode_read(ea_inode, buffer, size); 489 if (err) 490 goto out; 491 492 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) { 493 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer, 494 size); 495 if (err) { 496 ext4_warning_inode(ea_inode, 497 "EA inode hash validation failed"); 498 goto out; 499 } 500 501 if (ea_inode_cache) 502 mb_cache_entry_create(ea_inode_cache, GFP_NOFS, 503 ext4_xattr_inode_get_hash(ea_inode), 504 ea_inode->i_ino, true /* reusable */); 505 } 506 out: 507 iput(ea_inode); 508 return err; 509 } 510 511 static int 512 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name, 513 void *buffer, size_t buffer_size) 514 { 515 struct buffer_head *bh = NULL; 516 struct ext4_xattr_entry *entry; 517 size_t size; 518 void *end; 519 int error; 520 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 521 522 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld", 523 name_index, name, buffer, (long)buffer_size); 524 525 if (!EXT4_I(inode)->i_file_acl) 526 return -ENODATA; 527 ea_idebug(inode, "reading block %llu", 528 (unsigned long long)EXT4_I(inode)->i_file_acl); 529 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 530 if (IS_ERR(bh)) 531 return PTR_ERR(bh); 532 ea_bdebug(bh, "b_count=%d, refcount=%d", 533 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount)); 534 error = ext4_xattr_check_block(inode, bh); 535 if (error) 536 goto cleanup; 537 ext4_xattr_block_cache_insert(ea_block_cache, bh); 538 entry = BFIRST(bh); 539 end = bh->b_data + bh->b_size; 540 error = xattr_find_entry(inode, &entry, end, name_index, name, 1); 541 if (error) 542 goto cleanup; 543 size = le32_to_cpu(entry->e_value_size); 544 error = -ERANGE; 545 if (unlikely(size > EXT4_XATTR_SIZE_MAX)) 546 goto cleanup; 547 if (buffer) { 548 if (size > buffer_size) 549 goto cleanup; 550 if (entry->e_value_inum) { 551 error = ext4_xattr_inode_get(inode, entry, buffer, 552 size); 553 if (error) 554 goto cleanup; 555 } else { 556 u16 offset = le16_to_cpu(entry->e_value_offs); 557 void *p = bh->b_data + offset; 558 559 if (unlikely(p + size > end)) 560 goto cleanup; 561 memcpy(buffer, p, size); 562 } 563 } 564 error = size; 565 566 cleanup: 567 brelse(bh); 568 return error; 569 } 570 571 int 572 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name, 573 void *buffer, size_t buffer_size) 574 { 575 struct ext4_xattr_ibody_header *header; 576 struct ext4_xattr_entry *entry; 577 struct ext4_inode *raw_inode; 578 struct ext4_iloc iloc; 579 size_t size; 580 void *end; 581 int error; 582 583 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR)) 584 return -ENODATA; 585 error = ext4_get_inode_loc(inode, &iloc); 586 if (error) 587 return error; 588 raw_inode = ext4_raw_inode(&iloc); 589 header = IHDR(inode, raw_inode); 590 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 591 error = xattr_check_inode(inode, header, end); 592 if (error) 593 goto cleanup; 594 entry = IFIRST(header); 595 error = xattr_find_entry(inode, &entry, end, name_index, name, 0); 596 if (error) 597 goto cleanup; 598 size = le32_to_cpu(entry->e_value_size); 599 error = -ERANGE; 600 if (unlikely(size > EXT4_XATTR_SIZE_MAX)) 601 goto cleanup; 602 if (buffer) { 603 if (size > buffer_size) 604 goto cleanup; 605 if (entry->e_value_inum) { 606 error = ext4_xattr_inode_get(inode, entry, buffer, 607 size); 608 if (error) 609 goto cleanup; 610 } else { 611 u16 offset = le16_to_cpu(entry->e_value_offs); 612 void *p = (void *)IFIRST(header) + offset; 613 614 if (unlikely(p + size > end)) 615 goto cleanup; 616 memcpy(buffer, p, size); 617 } 618 } 619 error = size; 620 621 cleanup: 622 brelse(iloc.bh); 623 return error; 624 } 625 626 /* 627 * ext4_xattr_get() 628 * 629 * Copy an extended attribute into the buffer 630 * provided, or compute the buffer size required. 631 * Buffer is NULL to compute the size of the buffer required. 632 * 633 * Returns a negative error number on failure, or the number of bytes 634 * used / required on success. 635 */ 636 int 637 ext4_xattr_get(struct inode *inode, int name_index, const char *name, 638 void *buffer, size_t buffer_size) 639 { 640 int error; 641 642 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) 643 return -EIO; 644 645 if (strlen(name) > 255) 646 return -ERANGE; 647 648 down_read(&EXT4_I(inode)->xattr_sem); 649 error = ext4_xattr_ibody_get(inode, name_index, name, buffer, 650 buffer_size); 651 if (error == -ENODATA) 652 error = ext4_xattr_block_get(inode, name_index, name, buffer, 653 buffer_size); 654 up_read(&EXT4_I(inode)->xattr_sem); 655 return error; 656 } 657 658 static int 659 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry, 660 char *buffer, size_t buffer_size) 661 { 662 size_t rest = buffer_size; 663 664 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) { 665 const struct xattr_handler *handler = 666 ext4_xattr_handler(entry->e_name_index); 667 668 if (handler && (!handler->list || handler->list(dentry))) { 669 const char *prefix = handler->prefix ?: handler->name; 670 size_t prefix_len = strlen(prefix); 671 size_t size = prefix_len + entry->e_name_len + 1; 672 673 if (buffer) { 674 if (size > rest) 675 return -ERANGE; 676 memcpy(buffer, prefix, prefix_len); 677 buffer += prefix_len; 678 memcpy(buffer, entry->e_name, entry->e_name_len); 679 buffer += entry->e_name_len; 680 *buffer++ = 0; 681 } 682 rest -= size; 683 } 684 } 685 return buffer_size - rest; /* total size */ 686 } 687 688 static int 689 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size) 690 { 691 struct inode *inode = d_inode(dentry); 692 struct buffer_head *bh = NULL; 693 int error; 694 695 ea_idebug(inode, "buffer=%p, buffer_size=%ld", 696 buffer, (long)buffer_size); 697 698 if (!EXT4_I(inode)->i_file_acl) 699 return 0; 700 ea_idebug(inode, "reading block %llu", 701 (unsigned long long)EXT4_I(inode)->i_file_acl); 702 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 703 if (IS_ERR(bh)) 704 return PTR_ERR(bh); 705 ea_bdebug(bh, "b_count=%d, refcount=%d", 706 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount)); 707 error = ext4_xattr_check_block(inode, bh); 708 if (error) 709 goto cleanup; 710 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh); 711 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, 712 buffer_size); 713 cleanup: 714 brelse(bh); 715 return error; 716 } 717 718 static int 719 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size) 720 { 721 struct inode *inode = d_inode(dentry); 722 struct ext4_xattr_ibody_header *header; 723 struct ext4_inode *raw_inode; 724 struct ext4_iloc iloc; 725 void *end; 726 int error; 727 728 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR)) 729 return 0; 730 error = ext4_get_inode_loc(inode, &iloc); 731 if (error) 732 return error; 733 raw_inode = ext4_raw_inode(&iloc); 734 header = IHDR(inode, raw_inode); 735 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 736 error = xattr_check_inode(inode, header, end); 737 if (error) 738 goto cleanup; 739 error = ext4_xattr_list_entries(dentry, IFIRST(header), 740 buffer, buffer_size); 741 742 cleanup: 743 brelse(iloc.bh); 744 return error; 745 } 746 747 /* 748 * Inode operation listxattr() 749 * 750 * d_inode(dentry)->i_rwsem: don't care 751 * 752 * Copy a list of attribute names into the buffer 753 * provided, or compute the buffer size required. 754 * Buffer is NULL to compute the size of the buffer required. 755 * 756 * Returns a negative error number on failure, or the number of bytes 757 * used / required on success. 758 */ 759 ssize_t 760 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size) 761 { 762 int ret, ret2; 763 764 down_read(&EXT4_I(d_inode(dentry))->xattr_sem); 765 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size); 766 if (ret < 0) 767 goto errout; 768 if (buffer) { 769 buffer += ret; 770 buffer_size -= ret; 771 } 772 ret = ext4_xattr_block_list(dentry, buffer, buffer_size); 773 if (ret < 0) 774 goto errout; 775 ret += ret2; 776 errout: 777 up_read(&EXT4_I(d_inode(dentry))->xattr_sem); 778 return ret; 779 } 780 781 /* 782 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is 783 * not set, set it. 784 */ 785 static void ext4_xattr_update_super_block(handle_t *handle, 786 struct super_block *sb) 787 { 788 if (ext4_has_feature_xattr(sb)) 789 return; 790 791 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); 792 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) { 793 ext4_set_feature_xattr(sb); 794 ext4_handle_dirty_super(handle, sb); 795 } 796 } 797 798 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage) 799 { 800 struct ext4_iloc iloc = { .bh = NULL }; 801 struct buffer_head *bh = NULL; 802 struct ext4_inode *raw_inode; 803 struct ext4_xattr_ibody_header *header; 804 struct ext4_xattr_entry *entry; 805 qsize_t ea_inode_refs = 0; 806 void *end; 807 int ret; 808 809 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem); 810 811 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { 812 ret = ext4_get_inode_loc(inode, &iloc); 813 if (ret) 814 goto out; 815 raw_inode = ext4_raw_inode(&iloc); 816 header = IHDR(inode, raw_inode); 817 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 818 ret = xattr_check_inode(inode, header, end); 819 if (ret) 820 goto out; 821 822 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry); 823 entry = EXT4_XATTR_NEXT(entry)) 824 if (entry->e_value_inum) 825 ea_inode_refs++; 826 } 827 828 if (EXT4_I(inode)->i_file_acl) { 829 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 830 if (IS_ERR(bh)) { 831 ret = PTR_ERR(bh); 832 bh = NULL; 833 goto out; 834 } 835 836 ret = ext4_xattr_check_block(inode, bh); 837 if (ret) 838 goto out; 839 840 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry); 841 entry = EXT4_XATTR_NEXT(entry)) 842 if (entry->e_value_inum) 843 ea_inode_refs++; 844 } 845 *usage = ea_inode_refs + 1; 846 ret = 0; 847 out: 848 brelse(iloc.bh); 849 brelse(bh); 850 return ret; 851 } 852 853 static inline size_t round_up_cluster(struct inode *inode, size_t length) 854 { 855 struct super_block *sb = inode->i_sb; 856 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits + 857 inode->i_blkbits); 858 size_t mask = ~(cluster_size - 1); 859 860 return (length + cluster_size - 1) & mask; 861 } 862 863 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len) 864 { 865 int err; 866 867 err = dquot_alloc_inode(inode); 868 if (err) 869 return err; 870 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len)); 871 if (err) 872 dquot_free_inode(inode); 873 return err; 874 } 875 876 static void ext4_xattr_inode_free_quota(struct inode *parent, 877 struct inode *ea_inode, 878 size_t len) 879 { 880 if (ea_inode && 881 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) 882 return; 883 dquot_free_space_nodirty(parent, round_up_cluster(parent, len)); 884 dquot_free_inode(parent); 885 } 886 887 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode, 888 struct buffer_head *block_bh, size_t value_len, 889 bool is_create) 890 { 891 int credits; 892 int blocks; 893 894 /* 895 * 1) Owner inode update 896 * 2) Ref count update on old xattr block 897 * 3) new xattr block 898 * 4) block bitmap update for new xattr block 899 * 5) group descriptor for new xattr block 900 * 6) block bitmap update for old xattr block 901 * 7) group descriptor for old block 902 * 903 * 6 & 7 can happen if we have two racing threads T_a and T_b 904 * which are each trying to set an xattr on inodes I_a and I_b 905 * which were both initially sharing an xattr block. 906 */ 907 credits = 7; 908 909 /* Quota updates. */ 910 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb); 911 912 /* 913 * In case of inline data, we may push out the data to a block, 914 * so we need to reserve credits for this eventuality 915 */ 916 if (inode && ext4_has_inline_data(inode)) 917 credits += ext4_writepage_trans_blocks(inode) + 1; 918 919 /* We are done if ea_inode feature is not enabled. */ 920 if (!ext4_has_feature_ea_inode(sb)) 921 return credits; 922 923 /* New ea_inode, inode map, block bitmap, group descriptor. */ 924 credits += 4; 925 926 /* Data blocks. */ 927 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits; 928 929 /* Indirection block or one level of extent tree. */ 930 blocks += 1; 931 932 /* Block bitmap and group descriptor updates for each block. */ 933 credits += blocks * 2; 934 935 /* Blocks themselves. */ 936 credits += blocks; 937 938 if (!is_create) { 939 /* Dereference ea_inode holding old xattr value. 940 * Old ea_inode, inode map, block bitmap, group descriptor. 941 */ 942 credits += 4; 943 944 /* Data blocks for old ea_inode. */ 945 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits; 946 947 /* Indirection block or one level of extent tree for old 948 * ea_inode. 949 */ 950 blocks += 1; 951 952 /* Block bitmap and group descriptor updates for each block. */ 953 credits += blocks * 2; 954 } 955 956 /* We may need to clone the existing xattr block in which case we need 957 * to increment ref counts for existing ea_inodes referenced by it. 958 */ 959 if (block_bh) { 960 struct ext4_xattr_entry *entry = BFIRST(block_bh); 961 962 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) 963 if (entry->e_value_inum) 964 /* Ref count update on ea_inode. */ 965 credits += 1; 966 } 967 return credits; 968 } 969 970 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode, 971 int credits, struct buffer_head *bh, 972 bool dirty, bool block_csum) 973 { 974 int error; 975 976 if (!ext4_handle_valid(handle)) 977 return 0; 978 979 if (handle->h_buffer_credits >= credits) 980 return 0; 981 982 error = ext4_journal_extend(handle, credits - handle->h_buffer_credits); 983 if (!error) 984 return 0; 985 if (error < 0) { 986 ext4_warning(inode->i_sb, "Extend journal (error %d)", error); 987 return error; 988 } 989 990 if (bh && dirty) { 991 if (block_csum) 992 ext4_xattr_block_csum_set(inode, bh); 993 error = ext4_handle_dirty_metadata(handle, NULL, bh); 994 if (error) { 995 ext4_warning(inode->i_sb, "Handle metadata (error %d)", 996 error); 997 return error; 998 } 999 } 1000 1001 error = ext4_journal_restart(handle, credits); 1002 if (error) { 1003 ext4_warning(inode->i_sb, "Restart journal (error %d)", error); 1004 return error; 1005 } 1006 1007 if (bh) { 1008 error = ext4_journal_get_write_access(handle, bh); 1009 if (error) { 1010 ext4_warning(inode->i_sb, 1011 "Get write access failed (error %d)", 1012 error); 1013 return error; 1014 } 1015 } 1016 return 0; 1017 } 1018 1019 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode, 1020 int ref_change) 1021 { 1022 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode); 1023 struct ext4_iloc iloc; 1024 s64 ref_count; 1025 u32 hash; 1026 int ret; 1027 1028 inode_lock(ea_inode); 1029 1030 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc); 1031 if (ret) 1032 goto out; 1033 1034 ref_count = ext4_xattr_inode_get_ref(ea_inode); 1035 ref_count += ref_change; 1036 ext4_xattr_inode_set_ref(ea_inode, ref_count); 1037 1038 if (ref_change > 0) { 1039 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld", 1040 ea_inode->i_ino, ref_count); 1041 1042 if (ref_count == 1) { 1043 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u", 1044 ea_inode->i_ino, ea_inode->i_nlink); 1045 1046 set_nlink(ea_inode, 1); 1047 ext4_orphan_del(handle, ea_inode); 1048 1049 if (ea_inode_cache) { 1050 hash = ext4_xattr_inode_get_hash(ea_inode); 1051 mb_cache_entry_create(ea_inode_cache, 1052 GFP_NOFS, hash, 1053 ea_inode->i_ino, 1054 true /* reusable */); 1055 } 1056 } 1057 } else { 1058 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld", 1059 ea_inode->i_ino, ref_count); 1060 1061 if (ref_count == 0) { 1062 WARN_ONCE(ea_inode->i_nlink != 1, 1063 "EA inode %lu i_nlink=%u", 1064 ea_inode->i_ino, ea_inode->i_nlink); 1065 1066 clear_nlink(ea_inode); 1067 ext4_orphan_add(handle, ea_inode); 1068 1069 if (ea_inode_cache) { 1070 hash = ext4_xattr_inode_get_hash(ea_inode); 1071 mb_cache_entry_delete(ea_inode_cache, hash, 1072 ea_inode->i_ino); 1073 } 1074 } 1075 } 1076 1077 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc); 1078 if (ret) 1079 ext4_warning_inode(ea_inode, 1080 "ext4_mark_iloc_dirty() failed ret=%d", ret); 1081 out: 1082 inode_unlock(ea_inode); 1083 return ret; 1084 } 1085 1086 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode) 1087 { 1088 return ext4_xattr_inode_update_ref(handle, ea_inode, 1); 1089 } 1090 1091 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode) 1092 { 1093 return ext4_xattr_inode_update_ref(handle, ea_inode, -1); 1094 } 1095 1096 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent, 1097 struct ext4_xattr_entry *first) 1098 { 1099 struct inode *ea_inode; 1100 struct ext4_xattr_entry *entry; 1101 struct ext4_xattr_entry *failed_entry; 1102 unsigned int ea_ino; 1103 int err, saved_err; 1104 1105 for (entry = first; !IS_LAST_ENTRY(entry); 1106 entry = EXT4_XATTR_NEXT(entry)) { 1107 if (!entry->e_value_inum) 1108 continue; 1109 ea_ino = le32_to_cpu(entry->e_value_inum); 1110 err = ext4_xattr_inode_iget(parent, ea_ino, 1111 le32_to_cpu(entry->e_hash), 1112 &ea_inode); 1113 if (err) 1114 goto cleanup; 1115 err = ext4_xattr_inode_inc_ref(handle, ea_inode); 1116 if (err) { 1117 ext4_warning_inode(ea_inode, "inc ref error %d", err); 1118 iput(ea_inode); 1119 goto cleanup; 1120 } 1121 iput(ea_inode); 1122 } 1123 return 0; 1124 1125 cleanup: 1126 saved_err = err; 1127 failed_entry = entry; 1128 1129 for (entry = first; entry != failed_entry; 1130 entry = EXT4_XATTR_NEXT(entry)) { 1131 if (!entry->e_value_inum) 1132 continue; 1133 ea_ino = le32_to_cpu(entry->e_value_inum); 1134 err = ext4_xattr_inode_iget(parent, ea_ino, 1135 le32_to_cpu(entry->e_hash), 1136 &ea_inode); 1137 if (err) { 1138 ext4_warning(parent->i_sb, 1139 "cleanup ea_ino %u iget error %d", ea_ino, 1140 err); 1141 continue; 1142 } 1143 err = ext4_xattr_inode_dec_ref(handle, ea_inode); 1144 if (err) 1145 ext4_warning_inode(ea_inode, "cleanup dec ref error %d", 1146 err); 1147 iput(ea_inode); 1148 } 1149 return saved_err; 1150 } 1151 1152 static void 1153 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent, 1154 struct buffer_head *bh, 1155 struct ext4_xattr_entry *first, bool block_csum, 1156 struct ext4_xattr_inode_array **ea_inode_array, 1157 int extra_credits, bool skip_quota) 1158 { 1159 struct inode *ea_inode; 1160 struct ext4_xattr_entry *entry; 1161 bool dirty = false; 1162 unsigned int ea_ino; 1163 int err; 1164 int credits; 1165 1166 /* One credit for dec ref on ea_inode, one for orphan list addition, */ 1167 credits = 2 + extra_credits; 1168 1169 for (entry = first; !IS_LAST_ENTRY(entry); 1170 entry = EXT4_XATTR_NEXT(entry)) { 1171 if (!entry->e_value_inum) 1172 continue; 1173 ea_ino = le32_to_cpu(entry->e_value_inum); 1174 err = ext4_xattr_inode_iget(parent, ea_ino, 1175 le32_to_cpu(entry->e_hash), 1176 &ea_inode); 1177 if (err) 1178 continue; 1179 1180 err = ext4_expand_inode_array(ea_inode_array, ea_inode); 1181 if (err) { 1182 ext4_warning_inode(ea_inode, 1183 "Expand inode array err=%d", err); 1184 iput(ea_inode); 1185 continue; 1186 } 1187 1188 err = ext4_xattr_ensure_credits(handle, parent, credits, bh, 1189 dirty, block_csum); 1190 if (err) { 1191 ext4_warning_inode(ea_inode, "Ensure credits err=%d", 1192 err); 1193 continue; 1194 } 1195 1196 err = ext4_xattr_inode_dec_ref(handle, ea_inode); 1197 if (err) { 1198 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d", 1199 err); 1200 continue; 1201 } 1202 1203 if (!skip_quota) 1204 ext4_xattr_inode_free_quota(parent, ea_inode, 1205 le32_to_cpu(entry->e_value_size)); 1206 1207 /* 1208 * Forget about ea_inode within the same transaction that 1209 * decrements the ref count. This avoids duplicate decrements in 1210 * case the rest of the work spills over to subsequent 1211 * transactions. 1212 */ 1213 entry->e_value_inum = 0; 1214 entry->e_value_size = 0; 1215 1216 dirty = true; 1217 } 1218 1219 if (dirty) { 1220 /* 1221 * Note that we are deliberately skipping csum calculation for 1222 * the final update because we do not expect any journal 1223 * restarts until xattr block is freed. 1224 */ 1225 1226 err = ext4_handle_dirty_metadata(handle, NULL, bh); 1227 if (err) 1228 ext4_warning_inode(parent, 1229 "handle dirty metadata err=%d", err); 1230 } 1231 } 1232 1233 /* 1234 * Release the xattr block BH: If the reference count is > 1, decrement it; 1235 * otherwise free the block. 1236 */ 1237 static void 1238 ext4_xattr_release_block(handle_t *handle, struct inode *inode, 1239 struct buffer_head *bh, 1240 struct ext4_xattr_inode_array **ea_inode_array, 1241 int extra_credits) 1242 { 1243 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 1244 u32 hash, ref; 1245 int error = 0; 1246 1247 BUFFER_TRACE(bh, "get_write_access"); 1248 error = ext4_journal_get_write_access(handle, bh); 1249 if (error) 1250 goto out; 1251 1252 lock_buffer(bh); 1253 hash = le32_to_cpu(BHDR(bh)->h_hash); 1254 ref = le32_to_cpu(BHDR(bh)->h_refcount); 1255 if (ref == 1) { 1256 ea_bdebug(bh, "refcount now=0; freeing"); 1257 /* 1258 * This must happen under buffer lock for 1259 * ext4_xattr_block_set() to reliably detect freed block 1260 */ 1261 if (ea_block_cache) 1262 mb_cache_entry_delete(ea_block_cache, hash, 1263 bh->b_blocknr); 1264 get_bh(bh); 1265 unlock_buffer(bh); 1266 1267 if (ext4_has_feature_ea_inode(inode->i_sb)) 1268 ext4_xattr_inode_dec_ref_all(handle, inode, bh, 1269 BFIRST(bh), 1270 true /* block_csum */, 1271 ea_inode_array, 1272 extra_credits, 1273 true /* skip_quota */); 1274 ext4_free_blocks(handle, inode, bh, 0, 1, 1275 EXT4_FREE_BLOCKS_METADATA | 1276 EXT4_FREE_BLOCKS_FORGET); 1277 } else { 1278 ref--; 1279 BHDR(bh)->h_refcount = cpu_to_le32(ref); 1280 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) { 1281 struct mb_cache_entry *ce; 1282 1283 if (ea_block_cache) { 1284 ce = mb_cache_entry_get(ea_block_cache, hash, 1285 bh->b_blocknr); 1286 if (ce) { 1287 ce->e_reusable = 1; 1288 mb_cache_entry_put(ea_block_cache, ce); 1289 } 1290 } 1291 } 1292 1293 ext4_xattr_block_csum_set(inode, bh); 1294 /* 1295 * Beware of this ugliness: Releasing of xattr block references 1296 * from different inodes can race and so we have to protect 1297 * from a race where someone else frees the block (and releases 1298 * its journal_head) before we are done dirtying the buffer. In 1299 * nojournal mode this race is harmless and we actually cannot 1300 * call ext4_handle_dirty_metadata() with locked buffer as 1301 * that function can call sync_dirty_buffer() so for that case 1302 * we handle the dirtying after unlocking the buffer. 1303 */ 1304 if (ext4_handle_valid(handle)) 1305 error = ext4_handle_dirty_metadata(handle, inode, bh); 1306 unlock_buffer(bh); 1307 if (!ext4_handle_valid(handle)) 1308 error = ext4_handle_dirty_metadata(handle, inode, bh); 1309 if (IS_SYNC(inode)) 1310 ext4_handle_sync(handle); 1311 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1)); 1312 ea_bdebug(bh, "refcount now=%d; releasing", 1313 le32_to_cpu(BHDR(bh)->h_refcount)); 1314 } 1315 out: 1316 ext4_std_error(inode->i_sb, error); 1317 return; 1318 } 1319 1320 /* 1321 * Find the available free space for EAs. This also returns the total number of 1322 * bytes used by EA entries. 1323 */ 1324 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last, 1325 size_t *min_offs, void *base, int *total) 1326 { 1327 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { 1328 if (!last->e_value_inum && last->e_value_size) { 1329 size_t offs = le16_to_cpu(last->e_value_offs); 1330 if (offs < *min_offs) 1331 *min_offs = offs; 1332 } 1333 if (total) 1334 *total += EXT4_XATTR_LEN(last->e_name_len); 1335 } 1336 return (*min_offs - ((void *)last - base) - sizeof(__u32)); 1337 } 1338 1339 /* 1340 * Write the value of the EA in an inode. 1341 */ 1342 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode, 1343 const void *buf, int bufsize) 1344 { 1345 struct buffer_head *bh = NULL; 1346 unsigned long block = 0; 1347 int blocksize = ea_inode->i_sb->s_blocksize; 1348 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits; 1349 int csize, wsize = 0; 1350 int ret = 0; 1351 int retries = 0; 1352 1353 retry: 1354 while (ret >= 0 && ret < max_blocks) { 1355 struct ext4_map_blocks map; 1356 map.m_lblk = block += ret; 1357 map.m_len = max_blocks -= ret; 1358 1359 ret = ext4_map_blocks(handle, ea_inode, &map, 1360 EXT4_GET_BLOCKS_CREATE); 1361 if (ret <= 0) { 1362 ext4_mark_inode_dirty(handle, ea_inode); 1363 if (ret == -ENOSPC && 1364 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) { 1365 ret = 0; 1366 goto retry; 1367 } 1368 break; 1369 } 1370 } 1371 1372 if (ret < 0) 1373 return ret; 1374 1375 block = 0; 1376 while (wsize < bufsize) { 1377 if (bh != NULL) 1378 brelse(bh); 1379 csize = (bufsize - wsize) > blocksize ? blocksize : 1380 bufsize - wsize; 1381 bh = ext4_getblk(handle, ea_inode, block, 0); 1382 if (IS_ERR(bh)) 1383 return PTR_ERR(bh); 1384 if (!bh) { 1385 WARN_ON_ONCE(1); 1386 EXT4_ERROR_INODE(ea_inode, 1387 "ext4_getblk() return bh = NULL"); 1388 return -EFSCORRUPTED; 1389 } 1390 ret = ext4_journal_get_write_access(handle, bh); 1391 if (ret) 1392 goto out; 1393 1394 memcpy(bh->b_data, buf, csize); 1395 set_buffer_uptodate(bh); 1396 ext4_handle_dirty_metadata(handle, ea_inode, bh); 1397 1398 buf += csize; 1399 wsize += csize; 1400 block += 1; 1401 } 1402 1403 inode_lock(ea_inode); 1404 i_size_write(ea_inode, wsize); 1405 ext4_update_i_disksize(ea_inode, wsize); 1406 inode_unlock(ea_inode); 1407 1408 ext4_mark_inode_dirty(handle, ea_inode); 1409 1410 out: 1411 brelse(bh); 1412 1413 return ret; 1414 } 1415 1416 /* 1417 * Create an inode to store the value of a large EA. 1418 */ 1419 static struct inode *ext4_xattr_inode_create(handle_t *handle, 1420 struct inode *inode, u32 hash) 1421 { 1422 struct inode *ea_inode = NULL; 1423 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) }; 1424 int err; 1425 1426 /* 1427 * Let the next inode be the goal, so we try and allocate the EA inode 1428 * in the same group, or nearby one. 1429 */ 1430 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode, 1431 S_IFREG | 0600, NULL, inode->i_ino + 1, owner, 1432 EXT4_EA_INODE_FL); 1433 if (!IS_ERR(ea_inode)) { 1434 ea_inode->i_op = &ext4_file_inode_operations; 1435 ea_inode->i_fop = &ext4_file_operations; 1436 ext4_set_aops(ea_inode); 1437 ext4_xattr_inode_set_class(ea_inode); 1438 unlock_new_inode(ea_inode); 1439 ext4_xattr_inode_set_ref(ea_inode, 1); 1440 ext4_xattr_inode_set_hash(ea_inode, hash); 1441 err = ext4_mark_inode_dirty(handle, ea_inode); 1442 if (!err) 1443 err = ext4_inode_attach_jinode(ea_inode); 1444 if (err) { 1445 iput(ea_inode); 1446 return ERR_PTR(err); 1447 } 1448 1449 /* 1450 * Xattr inodes are shared therefore quota charging is performed 1451 * at a higher level. 1452 */ 1453 dquot_free_inode(ea_inode); 1454 dquot_drop(ea_inode); 1455 inode_lock(ea_inode); 1456 ea_inode->i_flags |= S_NOQUOTA; 1457 inode_unlock(ea_inode); 1458 } 1459 1460 return ea_inode; 1461 } 1462 1463 static struct inode * 1464 ext4_xattr_inode_cache_find(struct inode *inode, const void *value, 1465 size_t value_len, u32 hash) 1466 { 1467 struct inode *ea_inode; 1468 struct mb_cache_entry *ce; 1469 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode); 1470 void *ea_data; 1471 1472 if (!ea_inode_cache) 1473 return NULL; 1474 1475 ce = mb_cache_entry_find_first(ea_inode_cache, hash); 1476 if (!ce) 1477 return NULL; 1478 1479 ea_data = ext4_kvmalloc(value_len, GFP_NOFS); 1480 if (!ea_data) { 1481 mb_cache_entry_put(ea_inode_cache, ce); 1482 return NULL; 1483 } 1484 1485 while (ce) { 1486 ea_inode = ext4_iget(inode->i_sb, ce->e_value, 1487 EXT4_IGET_NORMAL); 1488 if (!IS_ERR(ea_inode) && 1489 !is_bad_inode(ea_inode) && 1490 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) && 1491 i_size_read(ea_inode) == value_len && 1492 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) && 1493 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data, 1494 value_len) && 1495 !memcmp(value, ea_data, value_len)) { 1496 mb_cache_entry_touch(ea_inode_cache, ce); 1497 mb_cache_entry_put(ea_inode_cache, ce); 1498 kvfree(ea_data); 1499 return ea_inode; 1500 } 1501 1502 if (!IS_ERR(ea_inode)) 1503 iput(ea_inode); 1504 ce = mb_cache_entry_find_next(ea_inode_cache, ce); 1505 } 1506 kvfree(ea_data); 1507 return NULL; 1508 } 1509 1510 /* 1511 * Add value of the EA in an inode. 1512 */ 1513 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode, 1514 const void *value, size_t value_len, 1515 struct inode **ret_inode) 1516 { 1517 struct inode *ea_inode; 1518 u32 hash; 1519 int err; 1520 1521 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len); 1522 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash); 1523 if (ea_inode) { 1524 err = ext4_xattr_inode_inc_ref(handle, ea_inode); 1525 if (err) { 1526 iput(ea_inode); 1527 return err; 1528 } 1529 1530 *ret_inode = ea_inode; 1531 return 0; 1532 } 1533 1534 /* Create an inode for the EA value */ 1535 ea_inode = ext4_xattr_inode_create(handle, inode, hash); 1536 if (IS_ERR(ea_inode)) 1537 return PTR_ERR(ea_inode); 1538 1539 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len); 1540 if (err) { 1541 ext4_xattr_inode_dec_ref(handle, ea_inode); 1542 iput(ea_inode); 1543 return err; 1544 } 1545 1546 if (EA_INODE_CACHE(inode)) 1547 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash, 1548 ea_inode->i_ino, true /* reusable */); 1549 1550 *ret_inode = ea_inode; 1551 return 0; 1552 } 1553 1554 /* 1555 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode 1556 * feature is enabled. 1557 */ 1558 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U) 1559 1560 static int ext4_xattr_set_entry(struct ext4_xattr_info *i, 1561 struct ext4_xattr_search *s, 1562 handle_t *handle, struct inode *inode, 1563 bool is_block) 1564 { 1565 struct ext4_xattr_entry *last, *next; 1566 struct ext4_xattr_entry *here = s->here; 1567 size_t min_offs = s->end - s->base, name_len = strlen(i->name); 1568 int in_inode = i->in_inode; 1569 struct inode *old_ea_inode = NULL; 1570 struct inode *new_ea_inode = NULL; 1571 size_t old_size, new_size; 1572 int ret; 1573 1574 /* Space used by old and new values. */ 1575 old_size = (!s->not_found && !here->e_value_inum) ? 1576 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0; 1577 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0; 1578 1579 /* 1580 * Optimization for the simple case when old and new values have the 1581 * same padded sizes. Not applicable if external inodes are involved. 1582 */ 1583 if (new_size && new_size == old_size) { 1584 size_t offs = le16_to_cpu(here->e_value_offs); 1585 void *val = s->base + offs; 1586 1587 here->e_value_size = cpu_to_le32(i->value_len); 1588 if (i->value == EXT4_ZERO_XATTR_VALUE) { 1589 memset(val, 0, new_size); 1590 } else { 1591 memcpy(val, i->value, i->value_len); 1592 /* Clear padding bytes. */ 1593 memset(val + i->value_len, 0, new_size - i->value_len); 1594 } 1595 goto update_hash; 1596 } 1597 1598 /* Compute min_offs and last. */ 1599 last = s->first; 1600 for (; !IS_LAST_ENTRY(last); last = next) { 1601 next = EXT4_XATTR_NEXT(last); 1602 if ((void *)next >= s->end) { 1603 EXT4_ERROR_INODE(inode, "corrupted xattr entries"); 1604 ret = -EFSCORRUPTED; 1605 goto out; 1606 } 1607 if (!last->e_value_inum && last->e_value_size) { 1608 size_t offs = le16_to_cpu(last->e_value_offs); 1609 if (offs < min_offs) 1610 min_offs = offs; 1611 } 1612 } 1613 1614 /* Check whether we have enough space. */ 1615 if (i->value) { 1616 size_t free; 1617 1618 free = min_offs - ((void *)last - s->base) - sizeof(__u32); 1619 if (!s->not_found) 1620 free += EXT4_XATTR_LEN(name_len) + old_size; 1621 1622 if (free < EXT4_XATTR_LEN(name_len) + new_size) { 1623 ret = -ENOSPC; 1624 goto out; 1625 } 1626 1627 /* 1628 * If storing the value in an external inode is an option, 1629 * reserve space for xattr entries/names in the external 1630 * attribute block so that a long value does not occupy the 1631 * whole space and prevent futher entries being added. 1632 */ 1633 if (ext4_has_feature_ea_inode(inode->i_sb) && 1634 new_size && is_block && 1635 (min_offs + old_size - new_size) < 1636 EXT4_XATTR_BLOCK_RESERVE(inode)) { 1637 ret = -ENOSPC; 1638 goto out; 1639 } 1640 } 1641 1642 /* 1643 * Getting access to old and new ea inodes is subject to failures. 1644 * Finish that work before doing any modifications to the xattr data. 1645 */ 1646 if (!s->not_found && here->e_value_inum) { 1647 ret = ext4_xattr_inode_iget(inode, 1648 le32_to_cpu(here->e_value_inum), 1649 le32_to_cpu(here->e_hash), 1650 &old_ea_inode); 1651 if (ret) { 1652 old_ea_inode = NULL; 1653 goto out; 1654 } 1655 } 1656 if (i->value && in_inode) { 1657 WARN_ON_ONCE(!i->value_len); 1658 1659 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len); 1660 if (ret) 1661 goto out; 1662 1663 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value, 1664 i->value_len, 1665 &new_ea_inode); 1666 if (ret) { 1667 new_ea_inode = NULL; 1668 ext4_xattr_inode_free_quota(inode, NULL, i->value_len); 1669 goto out; 1670 } 1671 } 1672 1673 if (old_ea_inode) { 1674 /* We are ready to release ref count on the old_ea_inode. */ 1675 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode); 1676 if (ret) { 1677 /* Release newly required ref count on new_ea_inode. */ 1678 if (new_ea_inode) { 1679 int err; 1680 1681 err = ext4_xattr_inode_dec_ref(handle, 1682 new_ea_inode); 1683 if (err) 1684 ext4_warning_inode(new_ea_inode, 1685 "dec ref new_ea_inode err=%d", 1686 err); 1687 ext4_xattr_inode_free_quota(inode, new_ea_inode, 1688 i->value_len); 1689 } 1690 goto out; 1691 } 1692 1693 ext4_xattr_inode_free_quota(inode, old_ea_inode, 1694 le32_to_cpu(here->e_value_size)); 1695 } 1696 1697 /* No failures allowed past this point. */ 1698 1699 if (!s->not_found && here->e_value_size && here->e_value_offs) { 1700 /* Remove the old value. */ 1701 void *first_val = s->base + min_offs; 1702 size_t offs = le16_to_cpu(here->e_value_offs); 1703 void *val = s->base + offs; 1704 1705 memmove(first_val + old_size, first_val, val - first_val); 1706 memset(first_val, 0, old_size); 1707 min_offs += old_size; 1708 1709 /* Adjust all value offsets. */ 1710 last = s->first; 1711 while (!IS_LAST_ENTRY(last)) { 1712 size_t o = le16_to_cpu(last->e_value_offs); 1713 1714 if (!last->e_value_inum && 1715 last->e_value_size && o < offs) 1716 last->e_value_offs = cpu_to_le16(o + old_size); 1717 last = EXT4_XATTR_NEXT(last); 1718 } 1719 } 1720 1721 if (!i->value) { 1722 /* Remove old name. */ 1723 size_t size = EXT4_XATTR_LEN(name_len); 1724 1725 last = ENTRY((void *)last - size); 1726 memmove(here, (void *)here + size, 1727 (void *)last - (void *)here + sizeof(__u32)); 1728 memset(last, 0, size); 1729 } else if (s->not_found) { 1730 /* Insert new name. */ 1731 size_t size = EXT4_XATTR_LEN(name_len); 1732 size_t rest = (void *)last - (void *)here + sizeof(__u32); 1733 1734 memmove((void *)here + size, here, rest); 1735 memset(here, 0, size); 1736 here->e_name_index = i->name_index; 1737 here->e_name_len = name_len; 1738 memcpy(here->e_name, i->name, name_len); 1739 } else { 1740 /* This is an update, reset value info. */ 1741 here->e_value_inum = 0; 1742 here->e_value_offs = 0; 1743 here->e_value_size = 0; 1744 } 1745 1746 if (i->value) { 1747 /* Insert new value. */ 1748 if (in_inode) { 1749 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino); 1750 } else if (i->value_len) { 1751 void *val = s->base + min_offs - new_size; 1752 1753 here->e_value_offs = cpu_to_le16(min_offs - new_size); 1754 if (i->value == EXT4_ZERO_XATTR_VALUE) { 1755 memset(val, 0, new_size); 1756 } else { 1757 memcpy(val, i->value, i->value_len); 1758 /* Clear padding bytes. */ 1759 memset(val + i->value_len, 0, 1760 new_size - i->value_len); 1761 } 1762 } 1763 here->e_value_size = cpu_to_le32(i->value_len); 1764 } 1765 1766 update_hash: 1767 if (i->value) { 1768 __le32 hash = 0; 1769 1770 /* Entry hash calculation. */ 1771 if (in_inode) { 1772 __le32 crc32c_hash; 1773 1774 /* 1775 * Feed crc32c hash instead of the raw value for entry 1776 * hash calculation. This is to avoid walking 1777 * potentially long value buffer again. 1778 */ 1779 crc32c_hash = cpu_to_le32( 1780 ext4_xattr_inode_get_hash(new_ea_inode)); 1781 hash = ext4_xattr_hash_entry(here->e_name, 1782 here->e_name_len, 1783 &crc32c_hash, 1); 1784 } else if (is_block) { 1785 __le32 *value = s->base + le16_to_cpu( 1786 here->e_value_offs); 1787 1788 hash = ext4_xattr_hash_entry(here->e_name, 1789 here->e_name_len, value, 1790 new_size >> 2); 1791 } 1792 here->e_hash = hash; 1793 } 1794 1795 if (is_block) 1796 ext4_xattr_rehash((struct ext4_xattr_header *)s->base); 1797 1798 ret = 0; 1799 out: 1800 iput(old_ea_inode); 1801 iput(new_ea_inode); 1802 return ret; 1803 } 1804 1805 struct ext4_xattr_block_find { 1806 struct ext4_xattr_search s; 1807 struct buffer_head *bh; 1808 }; 1809 1810 static int 1811 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i, 1812 struct ext4_xattr_block_find *bs) 1813 { 1814 struct super_block *sb = inode->i_sb; 1815 int error; 1816 1817 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld", 1818 i->name_index, i->name, i->value, (long)i->value_len); 1819 1820 if (EXT4_I(inode)->i_file_acl) { 1821 /* The inode already has an extended attribute block. */ 1822 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 1823 if (IS_ERR(bs->bh)) 1824 return PTR_ERR(bs->bh); 1825 ea_bdebug(bs->bh, "b_count=%d, refcount=%d", 1826 atomic_read(&(bs->bh->b_count)), 1827 le32_to_cpu(BHDR(bs->bh)->h_refcount)); 1828 error = ext4_xattr_check_block(inode, bs->bh); 1829 if (error) 1830 return error; 1831 /* Find the named attribute. */ 1832 bs->s.base = BHDR(bs->bh); 1833 bs->s.first = BFIRST(bs->bh); 1834 bs->s.end = bs->bh->b_data + bs->bh->b_size; 1835 bs->s.here = bs->s.first; 1836 error = xattr_find_entry(inode, &bs->s.here, bs->s.end, 1837 i->name_index, i->name, 1); 1838 if (error && error != -ENODATA) 1839 return error; 1840 bs->s.not_found = error; 1841 } 1842 return 0; 1843 } 1844 1845 static int 1846 ext4_xattr_block_set(handle_t *handle, struct inode *inode, 1847 struct ext4_xattr_info *i, 1848 struct ext4_xattr_block_find *bs) 1849 { 1850 struct super_block *sb = inode->i_sb; 1851 struct buffer_head *new_bh = NULL; 1852 struct ext4_xattr_search s_copy = bs->s; 1853 struct ext4_xattr_search *s = &s_copy; 1854 struct mb_cache_entry *ce = NULL; 1855 int error = 0; 1856 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 1857 struct inode *ea_inode = NULL, *tmp_inode; 1858 size_t old_ea_inode_quota = 0; 1859 unsigned int ea_ino; 1860 1861 1862 #define header(x) ((struct ext4_xattr_header *)(x)) 1863 1864 if (s->base) { 1865 BUFFER_TRACE(bs->bh, "get_write_access"); 1866 error = ext4_journal_get_write_access(handle, bs->bh); 1867 if (error) 1868 goto cleanup; 1869 lock_buffer(bs->bh); 1870 1871 if (header(s->base)->h_refcount == cpu_to_le32(1)) { 1872 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash); 1873 1874 /* 1875 * This must happen under buffer lock for 1876 * ext4_xattr_block_set() to reliably detect modified 1877 * block 1878 */ 1879 if (ea_block_cache) 1880 mb_cache_entry_delete(ea_block_cache, hash, 1881 bs->bh->b_blocknr); 1882 ea_bdebug(bs->bh, "modifying in-place"); 1883 error = ext4_xattr_set_entry(i, s, handle, inode, 1884 true /* is_block */); 1885 ext4_xattr_block_csum_set(inode, bs->bh); 1886 unlock_buffer(bs->bh); 1887 if (error == -EFSCORRUPTED) 1888 goto bad_block; 1889 if (!error) 1890 error = ext4_handle_dirty_metadata(handle, 1891 inode, 1892 bs->bh); 1893 if (error) 1894 goto cleanup; 1895 goto inserted; 1896 } else { 1897 int offset = (char *)s->here - bs->bh->b_data; 1898 1899 unlock_buffer(bs->bh); 1900 ea_bdebug(bs->bh, "cloning"); 1901 s->base = kmalloc(bs->bh->b_size, GFP_NOFS); 1902 error = -ENOMEM; 1903 if (s->base == NULL) 1904 goto cleanup; 1905 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size); 1906 s->first = ENTRY(header(s->base)+1); 1907 header(s->base)->h_refcount = cpu_to_le32(1); 1908 s->here = ENTRY(s->base + offset); 1909 s->end = s->base + bs->bh->b_size; 1910 1911 /* 1912 * If existing entry points to an xattr inode, we need 1913 * to prevent ext4_xattr_set_entry() from decrementing 1914 * ref count on it because the reference belongs to the 1915 * original block. In this case, make the entry look 1916 * like it has an empty value. 1917 */ 1918 if (!s->not_found && s->here->e_value_inum) { 1919 ea_ino = le32_to_cpu(s->here->e_value_inum); 1920 error = ext4_xattr_inode_iget(inode, ea_ino, 1921 le32_to_cpu(s->here->e_hash), 1922 &tmp_inode); 1923 if (error) 1924 goto cleanup; 1925 1926 if (!ext4_test_inode_state(tmp_inode, 1927 EXT4_STATE_LUSTRE_EA_INODE)) { 1928 /* 1929 * Defer quota free call for previous 1930 * inode until success is guaranteed. 1931 */ 1932 old_ea_inode_quota = le32_to_cpu( 1933 s->here->e_value_size); 1934 } 1935 iput(tmp_inode); 1936 1937 s->here->e_value_inum = 0; 1938 s->here->e_value_size = 0; 1939 } 1940 } 1941 } else { 1942 /* Allocate a buffer where we construct the new block. */ 1943 s->base = kzalloc(sb->s_blocksize, GFP_NOFS); 1944 /* assert(header == s->base) */ 1945 error = -ENOMEM; 1946 if (s->base == NULL) 1947 goto cleanup; 1948 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC); 1949 header(s->base)->h_blocks = cpu_to_le32(1); 1950 header(s->base)->h_refcount = cpu_to_le32(1); 1951 s->first = ENTRY(header(s->base)+1); 1952 s->here = ENTRY(header(s->base)+1); 1953 s->end = s->base + sb->s_blocksize; 1954 } 1955 1956 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */); 1957 if (error == -EFSCORRUPTED) 1958 goto bad_block; 1959 if (error) 1960 goto cleanup; 1961 1962 if (i->value && s->here->e_value_inum) { 1963 /* 1964 * A ref count on ea_inode has been taken as part of the call to 1965 * ext4_xattr_set_entry() above. We would like to drop this 1966 * extra ref but we have to wait until the xattr block is 1967 * initialized and has its own ref count on the ea_inode. 1968 */ 1969 ea_ino = le32_to_cpu(s->here->e_value_inum); 1970 error = ext4_xattr_inode_iget(inode, ea_ino, 1971 le32_to_cpu(s->here->e_hash), 1972 &ea_inode); 1973 if (error) { 1974 ea_inode = NULL; 1975 goto cleanup; 1976 } 1977 } 1978 1979 inserted: 1980 if (!IS_LAST_ENTRY(s->first)) { 1981 new_bh = ext4_xattr_block_cache_find(inode, header(s->base), 1982 &ce); 1983 if (new_bh) { 1984 /* We found an identical block in the cache. */ 1985 if (new_bh == bs->bh) 1986 ea_bdebug(new_bh, "keeping"); 1987 else { 1988 u32 ref; 1989 1990 WARN_ON_ONCE(dquot_initialize_needed(inode)); 1991 1992 /* The old block is released after updating 1993 the inode. */ 1994 error = dquot_alloc_block(inode, 1995 EXT4_C2B(EXT4_SB(sb), 1)); 1996 if (error) 1997 goto cleanup; 1998 BUFFER_TRACE(new_bh, "get_write_access"); 1999 error = ext4_journal_get_write_access(handle, 2000 new_bh); 2001 if (error) 2002 goto cleanup_dquot; 2003 lock_buffer(new_bh); 2004 /* 2005 * We have to be careful about races with 2006 * freeing, rehashing or adding references to 2007 * xattr block. Once we hold buffer lock xattr 2008 * block's state is stable so we can check 2009 * whether the block got freed / rehashed or 2010 * not. Since we unhash mbcache entry under 2011 * buffer lock when freeing / rehashing xattr 2012 * block, checking whether entry is still 2013 * hashed is reliable. Same rules hold for 2014 * e_reusable handling. 2015 */ 2016 if (hlist_bl_unhashed(&ce->e_hash_list) || 2017 !ce->e_reusable) { 2018 /* 2019 * Undo everything and check mbcache 2020 * again. 2021 */ 2022 unlock_buffer(new_bh); 2023 dquot_free_block(inode, 2024 EXT4_C2B(EXT4_SB(sb), 2025 1)); 2026 brelse(new_bh); 2027 mb_cache_entry_put(ea_block_cache, ce); 2028 ce = NULL; 2029 new_bh = NULL; 2030 goto inserted; 2031 } 2032 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1; 2033 BHDR(new_bh)->h_refcount = cpu_to_le32(ref); 2034 if (ref >= EXT4_XATTR_REFCOUNT_MAX) 2035 ce->e_reusable = 0; 2036 ea_bdebug(new_bh, "reusing; refcount now=%d", 2037 ref); 2038 ext4_xattr_block_csum_set(inode, new_bh); 2039 unlock_buffer(new_bh); 2040 error = ext4_handle_dirty_metadata(handle, 2041 inode, 2042 new_bh); 2043 if (error) 2044 goto cleanup_dquot; 2045 } 2046 mb_cache_entry_touch(ea_block_cache, ce); 2047 mb_cache_entry_put(ea_block_cache, ce); 2048 ce = NULL; 2049 } else if (bs->bh && s->base == bs->bh->b_data) { 2050 /* We were modifying this block in-place. */ 2051 ea_bdebug(bs->bh, "keeping this block"); 2052 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh); 2053 new_bh = bs->bh; 2054 get_bh(new_bh); 2055 } else { 2056 /* We need to allocate a new block */ 2057 ext4_fsblk_t goal, block; 2058 2059 WARN_ON_ONCE(dquot_initialize_needed(inode)); 2060 2061 goal = ext4_group_first_block_no(sb, 2062 EXT4_I(inode)->i_block_group); 2063 2064 /* non-extent files can't have physical blocks past 2^32 */ 2065 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 2066 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; 2067 2068 block = ext4_new_meta_blocks(handle, inode, goal, 0, 2069 NULL, &error); 2070 if (error) 2071 goto cleanup; 2072 2073 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 2074 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS); 2075 2076 ea_idebug(inode, "creating block %llu", 2077 (unsigned long long)block); 2078 2079 new_bh = sb_getblk(sb, block); 2080 if (unlikely(!new_bh)) { 2081 error = -ENOMEM; 2082 getblk_failed: 2083 ext4_free_blocks(handle, inode, NULL, block, 1, 2084 EXT4_FREE_BLOCKS_METADATA); 2085 goto cleanup; 2086 } 2087 error = ext4_xattr_inode_inc_ref_all(handle, inode, 2088 ENTRY(header(s->base)+1)); 2089 if (error) 2090 goto getblk_failed; 2091 if (ea_inode) { 2092 /* Drop the extra ref on ea_inode. */ 2093 error = ext4_xattr_inode_dec_ref(handle, 2094 ea_inode); 2095 if (error) 2096 ext4_warning_inode(ea_inode, 2097 "dec ref error=%d", 2098 error); 2099 iput(ea_inode); 2100 ea_inode = NULL; 2101 } 2102 2103 lock_buffer(new_bh); 2104 error = ext4_journal_get_create_access(handle, new_bh); 2105 if (error) { 2106 unlock_buffer(new_bh); 2107 error = -EIO; 2108 goto getblk_failed; 2109 } 2110 memcpy(new_bh->b_data, s->base, new_bh->b_size); 2111 ext4_xattr_block_csum_set(inode, new_bh); 2112 set_buffer_uptodate(new_bh); 2113 unlock_buffer(new_bh); 2114 ext4_xattr_block_cache_insert(ea_block_cache, new_bh); 2115 error = ext4_handle_dirty_metadata(handle, inode, 2116 new_bh); 2117 if (error) 2118 goto cleanup; 2119 } 2120 } 2121 2122 if (old_ea_inode_quota) 2123 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota); 2124 2125 /* Update the inode. */ 2126 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0; 2127 2128 /* Drop the previous xattr block. */ 2129 if (bs->bh && bs->bh != new_bh) { 2130 struct ext4_xattr_inode_array *ea_inode_array = NULL; 2131 2132 ext4_xattr_release_block(handle, inode, bs->bh, 2133 &ea_inode_array, 2134 0 /* extra_credits */); 2135 ext4_xattr_inode_array_free(ea_inode_array); 2136 } 2137 error = 0; 2138 2139 cleanup: 2140 if (ea_inode) { 2141 int error2; 2142 2143 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode); 2144 if (error2) 2145 ext4_warning_inode(ea_inode, "dec ref error=%d", 2146 error2); 2147 2148 /* If there was an error, revert the quota charge. */ 2149 if (error) 2150 ext4_xattr_inode_free_quota(inode, ea_inode, 2151 i_size_read(ea_inode)); 2152 iput(ea_inode); 2153 } 2154 if (ce) 2155 mb_cache_entry_put(ea_block_cache, ce); 2156 brelse(new_bh); 2157 if (!(bs->bh && s->base == bs->bh->b_data)) 2158 kfree(s->base); 2159 2160 return error; 2161 2162 cleanup_dquot: 2163 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1)); 2164 goto cleanup; 2165 2166 bad_block: 2167 EXT4_ERROR_INODE(inode, "bad block %llu", 2168 EXT4_I(inode)->i_file_acl); 2169 goto cleanup; 2170 2171 #undef header 2172 } 2173 2174 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i, 2175 struct ext4_xattr_ibody_find *is) 2176 { 2177 struct ext4_xattr_ibody_header *header; 2178 struct ext4_inode *raw_inode; 2179 int error; 2180 2181 if (EXT4_I(inode)->i_extra_isize == 0) 2182 return 0; 2183 raw_inode = ext4_raw_inode(&is->iloc); 2184 header = IHDR(inode, raw_inode); 2185 is->s.base = is->s.first = IFIRST(header); 2186 is->s.here = is->s.first; 2187 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 2188 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { 2189 error = xattr_check_inode(inode, header, is->s.end); 2190 if (error) 2191 return error; 2192 /* Find the named attribute. */ 2193 error = xattr_find_entry(inode, &is->s.here, is->s.end, 2194 i->name_index, i->name, 0); 2195 if (error && error != -ENODATA) 2196 return error; 2197 is->s.not_found = error; 2198 } 2199 return 0; 2200 } 2201 2202 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode, 2203 struct ext4_xattr_info *i, 2204 struct ext4_xattr_ibody_find *is) 2205 { 2206 struct ext4_xattr_ibody_header *header; 2207 struct ext4_xattr_search *s = &is->s; 2208 int error; 2209 2210 if (EXT4_I(inode)->i_extra_isize == 0) 2211 return -ENOSPC; 2212 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */); 2213 if (error) 2214 return error; 2215 header = IHDR(inode, ext4_raw_inode(&is->iloc)); 2216 if (!IS_LAST_ENTRY(s->first)) { 2217 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC); 2218 ext4_set_inode_state(inode, EXT4_STATE_XATTR); 2219 } else { 2220 header->h_magic = cpu_to_le32(0); 2221 ext4_clear_inode_state(inode, EXT4_STATE_XATTR); 2222 } 2223 return 0; 2224 } 2225 2226 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode, 2227 struct ext4_xattr_info *i, 2228 struct ext4_xattr_ibody_find *is) 2229 { 2230 struct ext4_xattr_ibody_header *header; 2231 struct ext4_xattr_search *s = &is->s; 2232 int error; 2233 2234 if (EXT4_I(inode)->i_extra_isize == 0) 2235 return -ENOSPC; 2236 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */); 2237 if (error) 2238 return error; 2239 header = IHDR(inode, ext4_raw_inode(&is->iloc)); 2240 if (!IS_LAST_ENTRY(s->first)) { 2241 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC); 2242 ext4_set_inode_state(inode, EXT4_STATE_XATTR); 2243 } else { 2244 header->h_magic = cpu_to_le32(0); 2245 ext4_clear_inode_state(inode, EXT4_STATE_XATTR); 2246 } 2247 return 0; 2248 } 2249 2250 static int ext4_xattr_value_same(struct ext4_xattr_search *s, 2251 struct ext4_xattr_info *i) 2252 { 2253 void *value; 2254 2255 /* When e_value_inum is set the value is stored externally. */ 2256 if (s->here->e_value_inum) 2257 return 0; 2258 if (le32_to_cpu(s->here->e_value_size) != i->value_len) 2259 return 0; 2260 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs); 2261 return !memcmp(value, i->value, i->value_len); 2262 } 2263 2264 static struct buffer_head *ext4_xattr_get_block(struct inode *inode) 2265 { 2266 struct buffer_head *bh; 2267 int error; 2268 2269 if (!EXT4_I(inode)->i_file_acl) 2270 return NULL; 2271 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 2272 if (IS_ERR(bh)) 2273 return bh; 2274 error = ext4_xattr_check_block(inode, bh); 2275 if (error) { 2276 brelse(bh); 2277 return ERR_PTR(error); 2278 } 2279 return bh; 2280 } 2281 2282 /* 2283 * ext4_xattr_set_handle() 2284 * 2285 * Create, replace or remove an extended attribute for this inode. Value 2286 * is NULL to remove an existing extended attribute, and non-NULL to 2287 * either replace an existing extended attribute, or create a new extended 2288 * attribute. The flags XATTR_REPLACE and XATTR_CREATE 2289 * specify that an extended attribute must exist and must not exist 2290 * previous to the call, respectively. 2291 * 2292 * Returns 0, or a negative error number on failure. 2293 */ 2294 int 2295 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index, 2296 const char *name, const void *value, size_t value_len, 2297 int flags) 2298 { 2299 struct ext4_xattr_info i = { 2300 .name_index = name_index, 2301 .name = name, 2302 .value = value, 2303 .value_len = value_len, 2304 .in_inode = 0, 2305 }; 2306 struct ext4_xattr_ibody_find is = { 2307 .s = { .not_found = -ENODATA, }, 2308 }; 2309 struct ext4_xattr_block_find bs = { 2310 .s = { .not_found = -ENODATA, }, 2311 }; 2312 int no_expand; 2313 int error; 2314 2315 if (!name) 2316 return -EINVAL; 2317 if (strlen(name) > 255) 2318 return -ERANGE; 2319 2320 ext4_write_lock_xattr(inode, &no_expand); 2321 2322 /* Check journal credits under write lock. */ 2323 if (ext4_handle_valid(handle)) { 2324 struct buffer_head *bh; 2325 int credits; 2326 2327 bh = ext4_xattr_get_block(inode); 2328 if (IS_ERR(bh)) { 2329 error = PTR_ERR(bh); 2330 goto cleanup; 2331 } 2332 2333 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh, 2334 value_len, 2335 flags & XATTR_CREATE); 2336 brelse(bh); 2337 2338 if (!ext4_handle_has_enough_credits(handle, credits)) { 2339 error = -ENOSPC; 2340 goto cleanup; 2341 } 2342 } 2343 2344 error = ext4_reserve_inode_write(handle, inode, &is.iloc); 2345 if (error) 2346 goto cleanup; 2347 2348 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) { 2349 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc); 2350 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); 2351 ext4_clear_inode_state(inode, EXT4_STATE_NEW); 2352 } 2353 2354 error = ext4_xattr_ibody_find(inode, &i, &is); 2355 if (error) 2356 goto cleanup; 2357 if (is.s.not_found) 2358 error = ext4_xattr_block_find(inode, &i, &bs); 2359 if (error) 2360 goto cleanup; 2361 if (is.s.not_found && bs.s.not_found) { 2362 error = -ENODATA; 2363 if (flags & XATTR_REPLACE) 2364 goto cleanup; 2365 error = 0; 2366 if (!value) 2367 goto cleanup; 2368 } else { 2369 error = -EEXIST; 2370 if (flags & XATTR_CREATE) 2371 goto cleanup; 2372 } 2373 2374 if (!value) { 2375 if (!is.s.not_found) 2376 error = ext4_xattr_ibody_set(handle, inode, &i, &is); 2377 else if (!bs.s.not_found) 2378 error = ext4_xattr_block_set(handle, inode, &i, &bs); 2379 } else { 2380 error = 0; 2381 /* Xattr value did not change? Save us some work and bail out */ 2382 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i)) 2383 goto cleanup; 2384 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i)) 2385 goto cleanup; 2386 2387 if (ext4_has_feature_ea_inode(inode->i_sb) && 2388 (EXT4_XATTR_SIZE(i.value_len) > 2389 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize))) 2390 i.in_inode = 1; 2391 retry_inode: 2392 error = ext4_xattr_ibody_set(handle, inode, &i, &is); 2393 if (!error && !bs.s.not_found) { 2394 i.value = NULL; 2395 error = ext4_xattr_block_set(handle, inode, &i, &bs); 2396 } else if (error == -ENOSPC) { 2397 if (EXT4_I(inode)->i_file_acl && !bs.s.base) { 2398 brelse(bs.bh); 2399 bs.bh = NULL; 2400 error = ext4_xattr_block_find(inode, &i, &bs); 2401 if (error) 2402 goto cleanup; 2403 } 2404 error = ext4_xattr_block_set(handle, inode, &i, &bs); 2405 if (!error && !is.s.not_found) { 2406 i.value = NULL; 2407 error = ext4_xattr_ibody_set(handle, inode, &i, 2408 &is); 2409 } else if (error == -ENOSPC) { 2410 /* 2411 * Xattr does not fit in the block, store at 2412 * external inode if possible. 2413 */ 2414 if (ext4_has_feature_ea_inode(inode->i_sb) && 2415 !i.in_inode) { 2416 i.in_inode = 1; 2417 goto retry_inode; 2418 } 2419 } 2420 } 2421 } 2422 if (!error) { 2423 ext4_xattr_update_super_block(handle, inode->i_sb); 2424 inode->i_ctime = current_time(inode); 2425 if (!value) 2426 no_expand = 0; 2427 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); 2428 /* 2429 * The bh is consumed by ext4_mark_iloc_dirty, even with 2430 * error != 0. 2431 */ 2432 is.iloc.bh = NULL; 2433 if (IS_SYNC(inode)) 2434 ext4_handle_sync(handle); 2435 } 2436 2437 cleanup: 2438 brelse(is.iloc.bh); 2439 brelse(bs.bh); 2440 ext4_write_unlock_xattr(inode, &no_expand); 2441 return error; 2442 } 2443 2444 int ext4_xattr_set_credits(struct inode *inode, size_t value_len, 2445 bool is_create, int *credits) 2446 { 2447 struct buffer_head *bh; 2448 int err; 2449 2450 *credits = 0; 2451 2452 if (!EXT4_SB(inode->i_sb)->s_journal) 2453 return 0; 2454 2455 down_read(&EXT4_I(inode)->xattr_sem); 2456 2457 bh = ext4_xattr_get_block(inode); 2458 if (IS_ERR(bh)) { 2459 err = PTR_ERR(bh); 2460 } else { 2461 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh, 2462 value_len, is_create); 2463 brelse(bh); 2464 err = 0; 2465 } 2466 2467 up_read(&EXT4_I(inode)->xattr_sem); 2468 return err; 2469 } 2470 2471 /* 2472 * ext4_xattr_set() 2473 * 2474 * Like ext4_xattr_set_handle, but start from an inode. This extended 2475 * attribute modification is a filesystem transaction by itself. 2476 * 2477 * Returns 0, or a negative error number on failure. 2478 */ 2479 int 2480 ext4_xattr_set(struct inode *inode, int name_index, const char *name, 2481 const void *value, size_t value_len, int flags) 2482 { 2483 handle_t *handle; 2484 struct super_block *sb = inode->i_sb; 2485 int error, retries = 0; 2486 int credits; 2487 2488 error = dquot_initialize(inode); 2489 if (error) 2490 return error; 2491 2492 retry: 2493 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE, 2494 &credits); 2495 if (error) 2496 return error; 2497 2498 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits); 2499 if (IS_ERR(handle)) { 2500 error = PTR_ERR(handle); 2501 } else { 2502 int error2; 2503 2504 error = ext4_xattr_set_handle(handle, inode, name_index, name, 2505 value, value_len, flags); 2506 error2 = ext4_journal_stop(handle); 2507 if (error == -ENOSPC && 2508 ext4_should_retry_alloc(sb, &retries)) 2509 goto retry; 2510 if (error == 0) 2511 error = error2; 2512 } 2513 2514 return error; 2515 } 2516 2517 /* 2518 * Shift the EA entries in the inode to create space for the increased 2519 * i_extra_isize. 2520 */ 2521 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry, 2522 int value_offs_shift, void *to, 2523 void *from, size_t n) 2524 { 2525 struct ext4_xattr_entry *last = entry; 2526 int new_offs; 2527 2528 /* We always shift xattr headers further thus offsets get lower */ 2529 BUG_ON(value_offs_shift > 0); 2530 2531 /* Adjust the value offsets of the entries */ 2532 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { 2533 if (!last->e_value_inum && last->e_value_size) { 2534 new_offs = le16_to_cpu(last->e_value_offs) + 2535 value_offs_shift; 2536 last->e_value_offs = cpu_to_le16(new_offs); 2537 } 2538 } 2539 /* Shift the entries by n bytes */ 2540 memmove(to, from, n); 2541 } 2542 2543 /* 2544 * Move xattr pointed to by 'entry' from inode into external xattr block 2545 */ 2546 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode, 2547 struct ext4_inode *raw_inode, 2548 struct ext4_xattr_entry *entry) 2549 { 2550 struct ext4_xattr_ibody_find *is = NULL; 2551 struct ext4_xattr_block_find *bs = NULL; 2552 char *buffer = NULL, *b_entry_name = NULL; 2553 size_t value_size = le32_to_cpu(entry->e_value_size); 2554 struct ext4_xattr_info i = { 2555 .value = NULL, 2556 .value_len = 0, 2557 .name_index = entry->e_name_index, 2558 .in_inode = !!entry->e_value_inum, 2559 }; 2560 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode); 2561 int error; 2562 2563 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS); 2564 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS); 2565 buffer = kmalloc(value_size, GFP_NOFS); 2566 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS); 2567 if (!is || !bs || !buffer || !b_entry_name) { 2568 error = -ENOMEM; 2569 goto out; 2570 } 2571 2572 is->s.not_found = -ENODATA; 2573 bs->s.not_found = -ENODATA; 2574 is->iloc.bh = NULL; 2575 bs->bh = NULL; 2576 2577 /* Save the entry name and the entry value */ 2578 if (entry->e_value_inum) { 2579 error = ext4_xattr_inode_get(inode, entry, buffer, value_size); 2580 if (error) 2581 goto out; 2582 } else { 2583 size_t value_offs = le16_to_cpu(entry->e_value_offs); 2584 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size); 2585 } 2586 2587 memcpy(b_entry_name, entry->e_name, entry->e_name_len); 2588 b_entry_name[entry->e_name_len] = '\0'; 2589 i.name = b_entry_name; 2590 2591 error = ext4_get_inode_loc(inode, &is->iloc); 2592 if (error) 2593 goto out; 2594 2595 error = ext4_xattr_ibody_find(inode, &i, is); 2596 if (error) 2597 goto out; 2598 2599 /* Remove the chosen entry from the inode */ 2600 error = ext4_xattr_ibody_set(handle, inode, &i, is); 2601 if (error) 2602 goto out; 2603 2604 i.value = buffer; 2605 i.value_len = value_size; 2606 error = ext4_xattr_block_find(inode, &i, bs); 2607 if (error) 2608 goto out; 2609 2610 /* Add entry which was removed from the inode into the block */ 2611 error = ext4_xattr_block_set(handle, inode, &i, bs); 2612 if (error) 2613 goto out; 2614 error = 0; 2615 out: 2616 kfree(b_entry_name); 2617 kfree(buffer); 2618 if (is) 2619 brelse(is->iloc.bh); 2620 if (bs) 2621 brelse(bs->bh); 2622 kfree(is); 2623 kfree(bs); 2624 2625 return error; 2626 } 2627 2628 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode, 2629 struct ext4_inode *raw_inode, 2630 int isize_diff, size_t ifree, 2631 size_t bfree, int *total_ino) 2632 { 2633 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode); 2634 struct ext4_xattr_entry *small_entry; 2635 struct ext4_xattr_entry *entry; 2636 struct ext4_xattr_entry *last; 2637 unsigned int entry_size; /* EA entry size */ 2638 unsigned int total_size; /* EA entry size + value size */ 2639 unsigned int min_total_size; 2640 int error; 2641 2642 while (isize_diff > ifree) { 2643 entry = NULL; 2644 small_entry = NULL; 2645 min_total_size = ~0U; 2646 last = IFIRST(header); 2647 /* Find the entry best suited to be pushed into EA block */ 2648 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { 2649 /* never move system.data out of the inode */ 2650 if ((last->e_name_len == 4) && 2651 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) && 2652 !memcmp(last->e_name, "data", 4)) 2653 continue; 2654 total_size = EXT4_XATTR_LEN(last->e_name_len); 2655 if (!last->e_value_inum) 2656 total_size += EXT4_XATTR_SIZE( 2657 le32_to_cpu(last->e_value_size)); 2658 if (total_size <= bfree && 2659 total_size < min_total_size) { 2660 if (total_size + ifree < isize_diff) { 2661 small_entry = last; 2662 } else { 2663 entry = last; 2664 min_total_size = total_size; 2665 } 2666 } 2667 } 2668 2669 if (entry == NULL) { 2670 if (small_entry == NULL) 2671 return -ENOSPC; 2672 entry = small_entry; 2673 } 2674 2675 entry_size = EXT4_XATTR_LEN(entry->e_name_len); 2676 total_size = entry_size; 2677 if (!entry->e_value_inum) 2678 total_size += EXT4_XATTR_SIZE( 2679 le32_to_cpu(entry->e_value_size)); 2680 error = ext4_xattr_move_to_block(handle, inode, raw_inode, 2681 entry); 2682 if (error) 2683 return error; 2684 2685 *total_ino -= entry_size; 2686 ifree += total_size; 2687 bfree -= total_size; 2688 } 2689 2690 return 0; 2691 } 2692 2693 /* 2694 * Expand an inode by new_extra_isize bytes when EAs are present. 2695 * Returns 0 on success or negative error number on failure. 2696 */ 2697 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize, 2698 struct ext4_inode *raw_inode, handle_t *handle) 2699 { 2700 struct ext4_xattr_ibody_header *header; 2701 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 2702 static unsigned int mnt_count; 2703 size_t min_offs; 2704 size_t ifree, bfree; 2705 int total_ino; 2706 void *base, *end; 2707 int error = 0, tried_min_extra_isize = 0; 2708 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize); 2709 int isize_diff; /* How much do we need to grow i_extra_isize */ 2710 2711 retry: 2712 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize; 2713 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) 2714 return 0; 2715 2716 header = IHDR(inode, raw_inode); 2717 2718 /* 2719 * Check if enough free space is available in the inode to shift the 2720 * entries ahead by new_extra_isize. 2721 */ 2722 2723 base = IFIRST(header); 2724 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 2725 min_offs = end - base; 2726 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32); 2727 2728 error = xattr_check_inode(inode, header, end); 2729 if (error) 2730 goto cleanup; 2731 2732 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino); 2733 if (ifree >= isize_diff) 2734 goto shift; 2735 2736 /* 2737 * Enough free space isn't available in the inode, check if 2738 * EA block can hold new_extra_isize bytes. 2739 */ 2740 if (EXT4_I(inode)->i_file_acl) { 2741 struct buffer_head *bh; 2742 2743 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 2744 if (IS_ERR(bh)) { 2745 error = PTR_ERR(bh); 2746 goto cleanup; 2747 } 2748 error = ext4_xattr_check_block(inode, bh); 2749 if (error) { 2750 brelse(bh); 2751 goto cleanup; 2752 } 2753 base = BHDR(bh); 2754 end = bh->b_data + bh->b_size; 2755 min_offs = end - base; 2756 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base, 2757 NULL); 2758 brelse(bh); 2759 if (bfree + ifree < isize_diff) { 2760 if (!tried_min_extra_isize && s_min_extra_isize) { 2761 tried_min_extra_isize++; 2762 new_extra_isize = s_min_extra_isize; 2763 goto retry; 2764 } 2765 error = -ENOSPC; 2766 goto cleanup; 2767 } 2768 } else { 2769 bfree = inode->i_sb->s_blocksize; 2770 } 2771 2772 error = ext4_xattr_make_inode_space(handle, inode, raw_inode, 2773 isize_diff, ifree, bfree, 2774 &total_ino); 2775 if (error) { 2776 if (error == -ENOSPC && !tried_min_extra_isize && 2777 s_min_extra_isize) { 2778 tried_min_extra_isize++; 2779 new_extra_isize = s_min_extra_isize; 2780 goto retry; 2781 } 2782 goto cleanup; 2783 } 2784 shift: 2785 /* Adjust the offsets and shift the remaining entries ahead */ 2786 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize 2787 - new_extra_isize, (void *)raw_inode + 2788 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize, 2789 (void *)header, total_ino); 2790 EXT4_I(inode)->i_extra_isize = new_extra_isize; 2791 2792 cleanup: 2793 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) { 2794 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.", 2795 inode->i_ino); 2796 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count); 2797 } 2798 return error; 2799 } 2800 2801 #define EIA_INCR 16 /* must be 2^n */ 2802 #define EIA_MASK (EIA_INCR - 1) 2803 2804 /* Add the large xattr @inode into @ea_inode_array for deferred iput(). 2805 * If @ea_inode_array is new or full it will be grown and the old 2806 * contents copied over. 2807 */ 2808 static int 2809 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array, 2810 struct inode *inode) 2811 { 2812 if (*ea_inode_array == NULL) { 2813 /* 2814 * Start with 15 inodes, so it fits into a power-of-two size. 2815 * If *ea_inode_array is NULL, this is essentially offsetof() 2816 */ 2817 (*ea_inode_array) = 2818 kmalloc(offsetof(struct ext4_xattr_inode_array, 2819 inodes[EIA_MASK]), 2820 GFP_NOFS); 2821 if (*ea_inode_array == NULL) 2822 return -ENOMEM; 2823 (*ea_inode_array)->count = 0; 2824 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) { 2825 /* expand the array once all 15 + n * 16 slots are full */ 2826 struct ext4_xattr_inode_array *new_array = NULL; 2827 int count = (*ea_inode_array)->count; 2828 2829 /* if new_array is NULL, this is essentially offsetof() */ 2830 new_array = kmalloc( 2831 offsetof(struct ext4_xattr_inode_array, 2832 inodes[count + EIA_INCR]), 2833 GFP_NOFS); 2834 if (new_array == NULL) 2835 return -ENOMEM; 2836 memcpy(new_array, *ea_inode_array, 2837 offsetof(struct ext4_xattr_inode_array, inodes[count])); 2838 kfree(*ea_inode_array); 2839 *ea_inode_array = new_array; 2840 } 2841 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode; 2842 return 0; 2843 } 2844 2845 /* 2846 * ext4_xattr_delete_inode() 2847 * 2848 * Free extended attribute resources associated with this inode. Traverse 2849 * all entries and decrement reference on any xattr inodes associated with this 2850 * inode. This is called immediately before an inode is freed. We have exclusive 2851 * access to the inode. If an orphan inode is deleted it will also release its 2852 * references on xattr block and xattr inodes. 2853 */ 2854 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode, 2855 struct ext4_xattr_inode_array **ea_inode_array, 2856 int extra_credits) 2857 { 2858 struct buffer_head *bh = NULL; 2859 struct ext4_xattr_ibody_header *header; 2860 struct ext4_iloc iloc = { .bh = NULL }; 2861 struct ext4_xattr_entry *entry; 2862 struct inode *ea_inode; 2863 int error; 2864 2865 error = ext4_xattr_ensure_credits(handle, inode, extra_credits, 2866 NULL /* bh */, 2867 false /* dirty */, 2868 false /* block_csum */); 2869 if (error) { 2870 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error); 2871 goto cleanup; 2872 } 2873 2874 if (ext4_has_feature_ea_inode(inode->i_sb) && 2875 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { 2876 2877 error = ext4_get_inode_loc(inode, &iloc); 2878 if (error) { 2879 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error); 2880 goto cleanup; 2881 } 2882 2883 error = ext4_journal_get_write_access(handle, iloc.bh); 2884 if (error) { 2885 EXT4_ERROR_INODE(inode, "write access (error %d)", 2886 error); 2887 goto cleanup; 2888 } 2889 2890 header = IHDR(inode, ext4_raw_inode(&iloc)); 2891 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC)) 2892 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh, 2893 IFIRST(header), 2894 false /* block_csum */, 2895 ea_inode_array, 2896 extra_credits, 2897 false /* skip_quota */); 2898 } 2899 2900 if (EXT4_I(inode)->i_file_acl) { 2901 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 2902 if (IS_ERR(bh)) { 2903 error = PTR_ERR(bh); 2904 if (error == -EIO) 2905 EXT4_ERROR_INODE(inode, "block %llu read error", 2906 EXT4_I(inode)->i_file_acl); 2907 bh = NULL; 2908 goto cleanup; 2909 } 2910 error = ext4_xattr_check_block(inode, bh); 2911 if (error) 2912 goto cleanup; 2913 2914 if (ext4_has_feature_ea_inode(inode->i_sb)) { 2915 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry); 2916 entry = EXT4_XATTR_NEXT(entry)) { 2917 if (!entry->e_value_inum) 2918 continue; 2919 error = ext4_xattr_inode_iget(inode, 2920 le32_to_cpu(entry->e_value_inum), 2921 le32_to_cpu(entry->e_hash), 2922 &ea_inode); 2923 if (error) 2924 continue; 2925 ext4_xattr_inode_free_quota(inode, ea_inode, 2926 le32_to_cpu(entry->e_value_size)); 2927 iput(ea_inode); 2928 } 2929 2930 } 2931 2932 ext4_xattr_release_block(handle, inode, bh, ea_inode_array, 2933 extra_credits); 2934 /* 2935 * Update i_file_acl value in the same transaction that releases 2936 * block. 2937 */ 2938 EXT4_I(inode)->i_file_acl = 0; 2939 error = ext4_mark_inode_dirty(handle, inode); 2940 if (error) { 2941 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)", 2942 error); 2943 goto cleanup; 2944 } 2945 } 2946 error = 0; 2947 cleanup: 2948 brelse(iloc.bh); 2949 brelse(bh); 2950 return error; 2951 } 2952 2953 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array) 2954 { 2955 int idx; 2956 2957 if (ea_inode_array == NULL) 2958 return; 2959 2960 for (idx = 0; idx < ea_inode_array->count; ++idx) 2961 iput(ea_inode_array->inodes[idx]); 2962 kfree(ea_inode_array); 2963 } 2964 2965 /* 2966 * ext4_xattr_block_cache_insert() 2967 * 2968 * Create a new entry in the extended attribute block cache, and insert 2969 * it unless such an entry is already in the cache. 2970 * 2971 * Returns 0, or a negative error number on failure. 2972 */ 2973 static void 2974 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache, 2975 struct buffer_head *bh) 2976 { 2977 struct ext4_xattr_header *header = BHDR(bh); 2978 __u32 hash = le32_to_cpu(header->h_hash); 2979 int reusable = le32_to_cpu(header->h_refcount) < 2980 EXT4_XATTR_REFCOUNT_MAX; 2981 int error; 2982 2983 if (!ea_block_cache) 2984 return; 2985 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash, 2986 bh->b_blocknr, reusable); 2987 if (error) { 2988 if (error == -EBUSY) 2989 ea_bdebug(bh, "already in cache"); 2990 } else 2991 ea_bdebug(bh, "inserting [%x]", (int)hash); 2992 } 2993 2994 /* 2995 * ext4_xattr_cmp() 2996 * 2997 * Compare two extended attribute blocks for equality. 2998 * 2999 * Returns 0 if the blocks are equal, 1 if they differ, and 3000 * a negative error number on errors. 3001 */ 3002 static int 3003 ext4_xattr_cmp(struct ext4_xattr_header *header1, 3004 struct ext4_xattr_header *header2) 3005 { 3006 struct ext4_xattr_entry *entry1, *entry2; 3007 3008 entry1 = ENTRY(header1+1); 3009 entry2 = ENTRY(header2+1); 3010 while (!IS_LAST_ENTRY(entry1)) { 3011 if (IS_LAST_ENTRY(entry2)) 3012 return 1; 3013 if (entry1->e_hash != entry2->e_hash || 3014 entry1->e_name_index != entry2->e_name_index || 3015 entry1->e_name_len != entry2->e_name_len || 3016 entry1->e_value_size != entry2->e_value_size || 3017 entry1->e_value_inum != entry2->e_value_inum || 3018 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len)) 3019 return 1; 3020 if (!entry1->e_value_inum && 3021 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs), 3022 (char *)header2 + le16_to_cpu(entry2->e_value_offs), 3023 le32_to_cpu(entry1->e_value_size))) 3024 return 1; 3025 3026 entry1 = EXT4_XATTR_NEXT(entry1); 3027 entry2 = EXT4_XATTR_NEXT(entry2); 3028 } 3029 if (!IS_LAST_ENTRY(entry2)) 3030 return 1; 3031 return 0; 3032 } 3033 3034 /* 3035 * ext4_xattr_block_cache_find() 3036 * 3037 * Find an identical extended attribute block. 3038 * 3039 * Returns a pointer to the block found, or NULL if such a block was 3040 * not found or an error occurred. 3041 */ 3042 static struct buffer_head * 3043 ext4_xattr_block_cache_find(struct inode *inode, 3044 struct ext4_xattr_header *header, 3045 struct mb_cache_entry **pce) 3046 { 3047 __u32 hash = le32_to_cpu(header->h_hash); 3048 struct mb_cache_entry *ce; 3049 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 3050 3051 if (!ea_block_cache) 3052 return NULL; 3053 if (!header->h_hash) 3054 return NULL; /* never share */ 3055 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash); 3056 ce = mb_cache_entry_find_first(ea_block_cache, hash); 3057 while (ce) { 3058 struct buffer_head *bh; 3059 3060 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO); 3061 if (IS_ERR(bh)) { 3062 if (PTR_ERR(bh) == -ENOMEM) 3063 return NULL; 3064 bh = NULL; 3065 EXT4_ERROR_INODE(inode, "block %lu read error", 3066 (unsigned long)ce->e_value); 3067 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) { 3068 *pce = ce; 3069 return bh; 3070 } 3071 brelse(bh); 3072 ce = mb_cache_entry_find_next(ea_block_cache, ce); 3073 } 3074 return NULL; 3075 } 3076 3077 #define NAME_HASH_SHIFT 5 3078 #define VALUE_HASH_SHIFT 16 3079 3080 /* 3081 * ext4_xattr_hash_entry() 3082 * 3083 * Compute the hash of an extended attribute. 3084 */ 3085 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value, 3086 size_t value_count) 3087 { 3088 __u32 hash = 0; 3089 3090 while (name_len--) { 3091 hash = (hash << NAME_HASH_SHIFT) ^ 3092 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^ 3093 *name++; 3094 } 3095 while (value_count--) { 3096 hash = (hash << VALUE_HASH_SHIFT) ^ 3097 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^ 3098 le32_to_cpu(*value++); 3099 } 3100 return cpu_to_le32(hash); 3101 } 3102 3103 #undef NAME_HASH_SHIFT 3104 #undef VALUE_HASH_SHIFT 3105 3106 #define BLOCK_HASH_SHIFT 16 3107 3108 /* 3109 * ext4_xattr_rehash() 3110 * 3111 * Re-compute the extended attribute hash value after an entry has changed. 3112 */ 3113 static void ext4_xattr_rehash(struct ext4_xattr_header *header) 3114 { 3115 struct ext4_xattr_entry *here; 3116 __u32 hash = 0; 3117 3118 here = ENTRY(header+1); 3119 while (!IS_LAST_ENTRY(here)) { 3120 if (!here->e_hash) { 3121 /* Block is not shared if an entry's hash value == 0 */ 3122 hash = 0; 3123 break; 3124 } 3125 hash = (hash << BLOCK_HASH_SHIFT) ^ 3126 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^ 3127 le32_to_cpu(here->e_hash); 3128 here = EXT4_XATTR_NEXT(here); 3129 } 3130 header->h_hash = cpu_to_le32(hash); 3131 } 3132 3133 #undef BLOCK_HASH_SHIFT 3134 3135 #define HASH_BUCKET_BITS 10 3136 3137 struct mb_cache * 3138 ext4_xattr_create_cache(void) 3139 { 3140 return mb_cache_create(HASH_BUCKET_BITS); 3141 } 3142 3143 void ext4_xattr_destroy_cache(struct mb_cache *cache) 3144 { 3145 if (cache) 3146 mb_cache_destroy(cache); 3147 } 3148 3149