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, sb, EXT4_SB(sb)->s_sbh, 795 EXT4_JTR_NONE) == 0) { 796 lock_buffer(EXT4_SB(sb)->s_sbh); 797 ext4_set_feature_xattr(sb); 798 ext4_superblock_csum_set(sb); 799 unlock_buffer(EXT4_SB(sb)->s_sbh); 800 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); 801 } 802 } 803 804 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage) 805 { 806 struct ext4_iloc iloc = { .bh = NULL }; 807 struct buffer_head *bh = NULL; 808 struct ext4_inode *raw_inode; 809 struct ext4_xattr_ibody_header *header; 810 struct ext4_xattr_entry *entry; 811 qsize_t ea_inode_refs = 0; 812 void *end; 813 int ret; 814 815 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem); 816 817 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { 818 ret = ext4_get_inode_loc(inode, &iloc); 819 if (ret) 820 goto out; 821 raw_inode = ext4_raw_inode(&iloc); 822 header = IHDR(inode, raw_inode); 823 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 824 ret = xattr_check_inode(inode, header, end); 825 if (ret) 826 goto out; 827 828 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry); 829 entry = EXT4_XATTR_NEXT(entry)) 830 if (entry->e_value_inum) 831 ea_inode_refs++; 832 } 833 834 if (EXT4_I(inode)->i_file_acl) { 835 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 836 if (IS_ERR(bh)) { 837 ret = PTR_ERR(bh); 838 bh = NULL; 839 goto out; 840 } 841 842 ret = ext4_xattr_check_block(inode, bh); 843 if (ret) 844 goto out; 845 846 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry); 847 entry = EXT4_XATTR_NEXT(entry)) 848 if (entry->e_value_inum) 849 ea_inode_refs++; 850 } 851 *usage = ea_inode_refs + 1; 852 ret = 0; 853 out: 854 brelse(iloc.bh); 855 brelse(bh); 856 return ret; 857 } 858 859 static inline size_t round_up_cluster(struct inode *inode, size_t length) 860 { 861 struct super_block *sb = inode->i_sb; 862 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits + 863 inode->i_blkbits); 864 size_t mask = ~(cluster_size - 1); 865 866 return (length + cluster_size - 1) & mask; 867 } 868 869 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len) 870 { 871 int err; 872 873 err = dquot_alloc_inode(inode); 874 if (err) 875 return err; 876 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len)); 877 if (err) 878 dquot_free_inode(inode); 879 return err; 880 } 881 882 static void ext4_xattr_inode_free_quota(struct inode *parent, 883 struct inode *ea_inode, 884 size_t len) 885 { 886 if (ea_inode && 887 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) 888 return; 889 dquot_free_space_nodirty(parent, round_up_cluster(parent, len)); 890 dquot_free_inode(parent); 891 } 892 893 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode, 894 struct buffer_head *block_bh, size_t value_len, 895 bool is_create) 896 { 897 int credits; 898 int blocks; 899 900 /* 901 * 1) Owner inode update 902 * 2) Ref count update on old xattr block 903 * 3) new xattr block 904 * 4) block bitmap update for new xattr block 905 * 5) group descriptor for new xattr block 906 * 6) block bitmap update for old xattr block 907 * 7) group descriptor for old block 908 * 909 * 6 & 7 can happen if we have two racing threads T_a and T_b 910 * which are each trying to set an xattr on inodes I_a and I_b 911 * which were both initially sharing an xattr block. 912 */ 913 credits = 7; 914 915 /* Quota updates. */ 916 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb); 917 918 /* 919 * In case of inline data, we may push out the data to a block, 920 * so we need to reserve credits for this eventuality 921 */ 922 if (inode && ext4_has_inline_data(inode)) 923 credits += ext4_writepage_trans_blocks(inode) + 1; 924 925 /* We are done if ea_inode feature is not enabled. */ 926 if (!ext4_has_feature_ea_inode(sb)) 927 return credits; 928 929 /* New ea_inode, inode map, block bitmap, group descriptor. */ 930 credits += 4; 931 932 /* Data blocks. */ 933 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits; 934 935 /* Indirection block or one level of extent tree. */ 936 blocks += 1; 937 938 /* Block bitmap and group descriptor updates for each block. */ 939 credits += blocks * 2; 940 941 /* Blocks themselves. */ 942 credits += blocks; 943 944 if (!is_create) { 945 /* Dereference ea_inode holding old xattr value. 946 * Old ea_inode, inode map, block bitmap, group descriptor. 947 */ 948 credits += 4; 949 950 /* Data blocks for old ea_inode. */ 951 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits; 952 953 /* Indirection block or one level of extent tree for old 954 * ea_inode. 955 */ 956 blocks += 1; 957 958 /* Block bitmap and group descriptor updates for each block. */ 959 credits += blocks * 2; 960 } 961 962 /* We may need to clone the existing xattr block in which case we need 963 * to increment ref counts for existing ea_inodes referenced by it. 964 */ 965 if (block_bh) { 966 struct ext4_xattr_entry *entry = BFIRST(block_bh); 967 968 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) 969 if (entry->e_value_inum) 970 /* Ref count update on ea_inode. */ 971 credits += 1; 972 } 973 return credits; 974 } 975 976 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode, 977 int ref_change) 978 { 979 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode); 980 struct ext4_iloc iloc; 981 s64 ref_count; 982 u32 hash; 983 int ret; 984 985 inode_lock(ea_inode); 986 987 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc); 988 if (ret) 989 goto out; 990 991 ref_count = ext4_xattr_inode_get_ref(ea_inode); 992 ref_count += ref_change; 993 ext4_xattr_inode_set_ref(ea_inode, ref_count); 994 995 if (ref_change > 0) { 996 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld", 997 ea_inode->i_ino, ref_count); 998 999 if (ref_count == 1) { 1000 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u", 1001 ea_inode->i_ino, ea_inode->i_nlink); 1002 1003 set_nlink(ea_inode, 1); 1004 ext4_orphan_del(handle, ea_inode); 1005 1006 if (ea_inode_cache) { 1007 hash = ext4_xattr_inode_get_hash(ea_inode); 1008 mb_cache_entry_create(ea_inode_cache, 1009 GFP_NOFS, hash, 1010 ea_inode->i_ino, 1011 true /* reusable */); 1012 } 1013 } 1014 } else { 1015 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld", 1016 ea_inode->i_ino, ref_count); 1017 1018 if (ref_count == 0) { 1019 WARN_ONCE(ea_inode->i_nlink != 1, 1020 "EA inode %lu i_nlink=%u", 1021 ea_inode->i_ino, ea_inode->i_nlink); 1022 1023 clear_nlink(ea_inode); 1024 ext4_orphan_add(handle, ea_inode); 1025 1026 if (ea_inode_cache) { 1027 hash = ext4_xattr_inode_get_hash(ea_inode); 1028 mb_cache_entry_delete(ea_inode_cache, hash, 1029 ea_inode->i_ino); 1030 } 1031 } 1032 } 1033 1034 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc); 1035 if (ret) 1036 ext4_warning_inode(ea_inode, 1037 "ext4_mark_iloc_dirty() failed ret=%d", ret); 1038 out: 1039 inode_unlock(ea_inode); 1040 return ret; 1041 } 1042 1043 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode) 1044 { 1045 return ext4_xattr_inode_update_ref(handle, ea_inode, 1); 1046 } 1047 1048 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode) 1049 { 1050 return ext4_xattr_inode_update_ref(handle, ea_inode, -1); 1051 } 1052 1053 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent, 1054 struct ext4_xattr_entry *first) 1055 { 1056 struct inode *ea_inode; 1057 struct ext4_xattr_entry *entry; 1058 struct ext4_xattr_entry *failed_entry; 1059 unsigned int ea_ino; 1060 int err, saved_err; 1061 1062 for (entry = first; !IS_LAST_ENTRY(entry); 1063 entry = EXT4_XATTR_NEXT(entry)) { 1064 if (!entry->e_value_inum) 1065 continue; 1066 ea_ino = le32_to_cpu(entry->e_value_inum); 1067 err = ext4_xattr_inode_iget(parent, ea_ino, 1068 le32_to_cpu(entry->e_hash), 1069 &ea_inode); 1070 if (err) 1071 goto cleanup; 1072 err = ext4_xattr_inode_inc_ref(handle, ea_inode); 1073 if (err) { 1074 ext4_warning_inode(ea_inode, "inc ref error %d", err); 1075 iput(ea_inode); 1076 goto cleanup; 1077 } 1078 iput(ea_inode); 1079 } 1080 return 0; 1081 1082 cleanup: 1083 saved_err = err; 1084 failed_entry = entry; 1085 1086 for (entry = first; entry != failed_entry; 1087 entry = EXT4_XATTR_NEXT(entry)) { 1088 if (!entry->e_value_inum) 1089 continue; 1090 ea_ino = le32_to_cpu(entry->e_value_inum); 1091 err = ext4_xattr_inode_iget(parent, ea_ino, 1092 le32_to_cpu(entry->e_hash), 1093 &ea_inode); 1094 if (err) { 1095 ext4_warning(parent->i_sb, 1096 "cleanup ea_ino %u iget error %d", ea_ino, 1097 err); 1098 continue; 1099 } 1100 err = ext4_xattr_inode_dec_ref(handle, ea_inode); 1101 if (err) 1102 ext4_warning_inode(ea_inode, "cleanup dec ref error %d", 1103 err); 1104 iput(ea_inode); 1105 } 1106 return saved_err; 1107 } 1108 1109 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode, 1110 struct buffer_head *bh, bool block_csum, bool dirty) 1111 { 1112 int error; 1113 1114 if (bh && dirty) { 1115 if (block_csum) 1116 ext4_xattr_block_csum_set(inode, bh); 1117 error = ext4_handle_dirty_metadata(handle, NULL, bh); 1118 if (error) { 1119 ext4_warning(inode->i_sb, "Handle metadata (error %d)", 1120 error); 1121 return error; 1122 } 1123 } 1124 return 0; 1125 } 1126 1127 static void 1128 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent, 1129 struct buffer_head *bh, 1130 struct ext4_xattr_entry *first, bool block_csum, 1131 struct ext4_xattr_inode_array **ea_inode_array, 1132 int extra_credits, bool skip_quota) 1133 { 1134 struct inode *ea_inode; 1135 struct ext4_xattr_entry *entry; 1136 bool dirty = false; 1137 unsigned int ea_ino; 1138 int err; 1139 int credits; 1140 1141 /* One credit for dec ref on ea_inode, one for orphan list addition, */ 1142 credits = 2 + extra_credits; 1143 1144 for (entry = first; !IS_LAST_ENTRY(entry); 1145 entry = EXT4_XATTR_NEXT(entry)) { 1146 if (!entry->e_value_inum) 1147 continue; 1148 ea_ino = le32_to_cpu(entry->e_value_inum); 1149 err = ext4_xattr_inode_iget(parent, ea_ino, 1150 le32_to_cpu(entry->e_hash), 1151 &ea_inode); 1152 if (err) 1153 continue; 1154 1155 err = ext4_expand_inode_array(ea_inode_array, ea_inode); 1156 if (err) { 1157 ext4_warning_inode(ea_inode, 1158 "Expand inode array err=%d", err); 1159 iput(ea_inode); 1160 continue; 1161 } 1162 1163 err = ext4_journal_ensure_credits_fn(handle, credits, credits, 1164 ext4_free_metadata_revoke_credits(parent->i_sb, 1), 1165 ext4_xattr_restart_fn(handle, parent, bh, block_csum, 1166 dirty)); 1167 if (err < 0) { 1168 ext4_warning_inode(ea_inode, "Ensure credits err=%d", 1169 err); 1170 continue; 1171 } 1172 if (err > 0) { 1173 err = ext4_journal_get_write_access(handle, 1174 parent->i_sb, bh, EXT4_JTR_NONE); 1175 if (err) { 1176 ext4_warning_inode(ea_inode, 1177 "Re-get write access err=%d", 1178 err); 1179 continue; 1180 } 1181 } 1182 1183 err = ext4_xattr_inode_dec_ref(handle, ea_inode); 1184 if (err) { 1185 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d", 1186 err); 1187 continue; 1188 } 1189 1190 if (!skip_quota) 1191 ext4_xattr_inode_free_quota(parent, ea_inode, 1192 le32_to_cpu(entry->e_value_size)); 1193 1194 /* 1195 * Forget about ea_inode within the same transaction that 1196 * decrements the ref count. This avoids duplicate decrements in 1197 * case the rest of the work spills over to subsequent 1198 * transactions. 1199 */ 1200 entry->e_value_inum = 0; 1201 entry->e_value_size = 0; 1202 1203 dirty = true; 1204 } 1205 1206 if (dirty) { 1207 /* 1208 * Note that we are deliberately skipping csum calculation for 1209 * the final update because we do not expect any journal 1210 * restarts until xattr block is freed. 1211 */ 1212 1213 err = ext4_handle_dirty_metadata(handle, NULL, bh); 1214 if (err) 1215 ext4_warning_inode(parent, 1216 "handle dirty metadata err=%d", err); 1217 } 1218 } 1219 1220 /* 1221 * Release the xattr block BH: If the reference count is > 1, decrement it; 1222 * otherwise free the block. 1223 */ 1224 static void 1225 ext4_xattr_release_block(handle_t *handle, struct inode *inode, 1226 struct buffer_head *bh, 1227 struct ext4_xattr_inode_array **ea_inode_array, 1228 int extra_credits) 1229 { 1230 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 1231 u32 hash, ref; 1232 int error = 0; 1233 1234 BUFFER_TRACE(bh, "get_write_access"); 1235 error = ext4_journal_get_write_access(handle, inode->i_sb, bh, 1236 EXT4_JTR_NONE); 1237 if (error) 1238 goto out; 1239 1240 lock_buffer(bh); 1241 hash = le32_to_cpu(BHDR(bh)->h_hash); 1242 ref = le32_to_cpu(BHDR(bh)->h_refcount); 1243 if (ref == 1) { 1244 ea_bdebug(bh, "refcount now=0; freeing"); 1245 /* 1246 * This must happen under buffer lock for 1247 * ext4_xattr_block_set() to reliably detect freed block 1248 */ 1249 if (ea_block_cache) 1250 mb_cache_entry_delete(ea_block_cache, hash, 1251 bh->b_blocknr); 1252 get_bh(bh); 1253 unlock_buffer(bh); 1254 1255 if (ext4_has_feature_ea_inode(inode->i_sb)) 1256 ext4_xattr_inode_dec_ref_all(handle, inode, bh, 1257 BFIRST(bh), 1258 true /* block_csum */, 1259 ea_inode_array, 1260 extra_credits, 1261 true /* skip_quota */); 1262 ext4_free_blocks(handle, inode, bh, 0, 1, 1263 EXT4_FREE_BLOCKS_METADATA | 1264 EXT4_FREE_BLOCKS_FORGET); 1265 } else { 1266 ref--; 1267 BHDR(bh)->h_refcount = cpu_to_le32(ref); 1268 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) { 1269 struct mb_cache_entry *ce; 1270 1271 if (ea_block_cache) { 1272 ce = mb_cache_entry_get(ea_block_cache, hash, 1273 bh->b_blocknr); 1274 if (ce) { 1275 ce->e_reusable = 1; 1276 mb_cache_entry_put(ea_block_cache, ce); 1277 } 1278 } 1279 } 1280 1281 ext4_xattr_block_csum_set(inode, bh); 1282 /* 1283 * Beware of this ugliness: Releasing of xattr block references 1284 * from different inodes can race and so we have to protect 1285 * from a race where someone else frees the block (and releases 1286 * its journal_head) before we are done dirtying the buffer. In 1287 * nojournal mode this race is harmless and we actually cannot 1288 * call ext4_handle_dirty_metadata() with locked buffer as 1289 * that function can call sync_dirty_buffer() so for that case 1290 * we handle the dirtying after unlocking the buffer. 1291 */ 1292 if (ext4_handle_valid(handle)) 1293 error = ext4_handle_dirty_metadata(handle, inode, bh); 1294 unlock_buffer(bh); 1295 if (!ext4_handle_valid(handle)) 1296 error = ext4_handle_dirty_metadata(handle, inode, bh); 1297 if (IS_SYNC(inode)) 1298 ext4_handle_sync(handle); 1299 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1)); 1300 ea_bdebug(bh, "refcount now=%d; releasing", 1301 le32_to_cpu(BHDR(bh)->h_refcount)); 1302 } 1303 out: 1304 ext4_std_error(inode->i_sb, error); 1305 return; 1306 } 1307 1308 /* 1309 * Find the available free space for EAs. This also returns the total number of 1310 * bytes used by EA entries. 1311 */ 1312 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last, 1313 size_t *min_offs, void *base, int *total) 1314 { 1315 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { 1316 if (!last->e_value_inum && last->e_value_size) { 1317 size_t offs = le16_to_cpu(last->e_value_offs); 1318 if (offs < *min_offs) 1319 *min_offs = offs; 1320 } 1321 if (total) 1322 *total += EXT4_XATTR_LEN(last->e_name_len); 1323 } 1324 return (*min_offs - ((void *)last - base) - sizeof(__u32)); 1325 } 1326 1327 /* 1328 * Write the value of the EA in an inode. 1329 */ 1330 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode, 1331 const void *buf, int bufsize) 1332 { 1333 struct buffer_head *bh = NULL; 1334 unsigned long block = 0; 1335 int blocksize = ea_inode->i_sb->s_blocksize; 1336 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits; 1337 int csize, wsize = 0; 1338 int ret = 0, ret2 = 0; 1339 int retries = 0; 1340 1341 retry: 1342 while (ret >= 0 && ret < max_blocks) { 1343 struct ext4_map_blocks map; 1344 map.m_lblk = block += ret; 1345 map.m_len = max_blocks -= ret; 1346 1347 ret = ext4_map_blocks(handle, ea_inode, &map, 1348 EXT4_GET_BLOCKS_CREATE); 1349 if (ret <= 0) { 1350 ext4_mark_inode_dirty(handle, ea_inode); 1351 if (ret == -ENOSPC && 1352 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) { 1353 ret = 0; 1354 goto retry; 1355 } 1356 break; 1357 } 1358 } 1359 1360 if (ret < 0) 1361 return ret; 1362 1363 block = 0; 1364 while (wsize < bufsize) { 1365 brelse(bh); 1366 csize = (bufsize - wsize) > blocksize ? blocksize : 1367 bufsize - wsize; 1368 bh = ext4_getblk(handle, ea_inode, block, 0); 1369 if (IS_ERR(bh)) 1370 return PTR_ERR(bh); 1371 if (!bh) { 1372 WARN_ON_ONCE(1); 1373 EXT4_ERROR_INODE(ea_inode, 1374 "ext4_getblk() return bh = NULL"); 1375 return -EFSCORRUPTED; 1376 } 1377 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh, 1378 EXT4_JTR_NONE); 1379 if (ret) 1380 goto out; 1381 1382 memcpy(bh->b_data, buf, csize); 1383 set_buffer_uptodate(bh); 1384 ext4_handle_dirty_metadata(handle, ea_inode, bh); 1385 1386 buf += csize; 1387 wsize += csize; 1388 block += 1; 1389 } 1390 1391 inode_lock(ea_inode); 1392 i_size_write(ea_inode, wsize); 1393 ext4_update_i_disksize(ea_inode, wsize); 1394 inode_unlock(ea_inode); 1395 1396 ret2 = ext4_mark_inode_dirty(handle, ea_inode); 1397 if (unlikely(ret2 && !ret)) 1398 ret = ret2; 1399 1400 out: 1401 brelse(bh); 1402 1403 return ret; 1404 } 1405 1406 /* 1407 * Create an inode to store the value of a large EA. 1408 */ 1409 static struct inode *ext4_xattr_inode_create(handle_t *handle, 1410 struct inode *inode, u32 hash) 1411 { 1412 struct inode *ea_inode = NULL; 1413 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) }; 1414 int err; 1415 1416 /* 1417 * Let the next inode be the goal, so we try and allocate the EA inode 1418 * in the same group, or nearby one. 1419 */ 1420 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode, 1421 S_IFREG | 0600, NULL, inode->i_ino + 1, owner, 1422 EXT4_EA_INODE_FL); 1423 if (!IS_ERR(ea_inode)) { 1424 ea_inode->i_op = &ext4_file_inode_operations; 1425 ea_inode->i_fop = &ext4_file_operations; 1426 ext4_set_aops(ea_inode); 1427 ext4_xattr_inode_set_class(ea_inode); 1428 unlock_new_inode(ea_inode); 1429 ext4_xattr_inode_set_ref(ea_inode, 1); 1430 ext4_xattr_inode_set_hash(ea_inode, hash); 1431 err = ext4_mark_inode_dirty(handle, ea_inode); 1432 if (!err) 1433 err = ext4_inode_attach_jinode(ea_inode); 1434 if (err) { 1435 iput(ea_inode); 1436 return ERR_PTR(err); 1437 } 1438 1439 /* 1440 * Xattr inodes are shared therefore quota charging is performed 1441 * at a higher level. 1442 */ 1443 dquot_free_inode(ea_inode); 1444 dquot_drop(ea_inode); 1445 inode_lock(ea_inode); 1446 ea_inode->i_flags |= S_NOQUOTA; 1447 inode_unlock(ea_inode); 1448 } 1449 1450 return ea_inode; 1451 } 1452 1453 static struct inode * 1454 ext4_xattr_inode_cache_find(struct inode *inode, const void *value, 1455 size_t value_len, u32 hash) 1456 { 1457 struct inode *ea_inode; 1458 struct mb_cache_entry *ce; 1459 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode); 1460 void *ea_data; 1461 1462 if (!ea_inode_cache) 1463 return NULL; 1464 1465 ce = mb_cache_entry_find_first(ea_inode_cache, hash); 1466 if (!ce) 1467 return NULL; 1468 1469 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) && 1470 !(current->flags & PF_MEMALLOC_NOFS)); 1471 1472 ea_data = kvmalloc(value_len, GFP_KERNEL); 1473 if (!ea_data) { 1474 mb_cache_entry_put(ea_inode_cache, ce); 1475 return NULL; 1476 } 1477 1478 while (ce) { 1479 ea_inode = ext4_iget(inode->i_sb, ce->e_value, 1480 EXT4_IGET_NORMAL); 1481 if (!IS_ERR(ea_inode) && 1482 !is_bad_inode(ea_inode) && 1483 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) && 1484 i_size_read(ea_inode) == value_len && 1485 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) && 1486 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data, 1487 value_len) && 1488 !memcmp(value, ea_data, value_len)) { 1489 mb_cache_entry_touch(ea_inode_cache, ce); 1490 mb_cache_entry_put(ea_inode_cache, ce); 1491 kvfree(ea_data); 1492 return ea_inode; 1493 } 1494 1495 if (!IS_ERR(ea_inode)) 1496 iput(ea_inode); 1497 ce = mb_cache_entry_find_next(ea_inode_cache, ce); 1498 } 1499 kvfree(ea_data); 1500 return NULL; 1501 } 1502 1503 /* 1504 * Add value of the EA in an inode. 1505 */ 1506 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode, 1507 const void *value, size_t value_len, 1508 struct inode **ret_inode) 1509 { 1510 struct inode *ea_inode; 1511 u32 hash; 1512 int err; 1513 1514 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len); 1515 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash); 1516 if (ea_inode) { 1517 err = ext4_xattr_inode_inc_ref(handle, ea_inode); 1518 if (err) { 1519 iput(ea_inode); 1520 return err; 1521 } 1522 1523 *ret_inode = ea_inode; 1524 return 0; 1525 } 1526 1527 /* Create an inode for the EA value */ 1528 ea_inode = ext4_xattr_inode_create(handle, inode, hash); 1529 if (IS_ERR(ea_inode)) 1530 return PTR_ERR(ea_inode); 1531 1532 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len); 1533 if (err) { 1534 ext4_xattr_inode_dec_ref(handle, ea_inode); 1535 iput(ea_inode); 1536 return err; 1537 } 1538 1539 if (EA_INODE_CACHE(inode)) 1540 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash, 1541 ea_inode->i_ino, true /* reusable */); 1542 1543 *ret_inode = ea_inode; 1544 return 0; 1545 } 1546 1547 /* 1548 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode 1549 * feature is enabled. 1550 */ 1551 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U) 1552 1553 static int ext4_xattr_set_entry(struct ext4_xattr_info *i, 1554 struct ext4_xattr_search *s, 1555 handle_t *handle, struct inode *inode, 1556 bool is_block) 1557 { 1558 struct ext4_xattr_entry *last, *next; 1559 struct ext4_xattr_entry *here = s->here; 1560 size_t min_offs = s->end - s->base, name_len = strlen(i->name); 1561 int in_inode = i->in_inode; 1562 struct inode *old_ea_inode = NULL; 1563 struct inode *new_ea_inode = NULL; 1564 size_t old_size, new_size; 1565 int ret; 1566 1567 /* Space used by old and new values. */ 1568 old_size = (!s->not_found && !here->e_value_inum) ? 1569 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0; 1570 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0; 1571 1572 /* 1573 * Optimization for the simple case when old and new values have the 1574 * same padded sizes. Not applicable if external inodes are involved. 1575 */ 1576 if (new_size && new_size == old_size) { 1577 size_t offs = le16_to_cpu(here->e_value_offs); 1578 void *val = s->base + offs; 1579 1580 here->e_value_size = cpu_to_le32(i->value_len); 1581 if (i->value == EXT4_ZERO_XATTR_VALUE) { 1582 memset(val, 0, new_size); 1583 } else { 1584 memcpy(val, i->value, i->value_len); 1585 /* Clear padding bytes. */ 1586 memset(val + i->value_len, 0, new_size - i->value_len); 1587 } 1588 goto update_hash; 1589 } 1590 1591 /* Compute min_offs and last. */ 1592 last = s->first; 1593 for (; !IS_LAST_ENTRY(last); last = next) { 1594 next = EXT4_XATTR_NEXT(last); 1595 if ((void *)next >= s->end) { 1596 EXT4_ERROR_INODE(inode, "corrupted xattr entries"); 1597 ret = -EFSCORRUPTED; 1598 goto out; 1599 } 1600 if (!last->e_value_inum && last->e_value_size) { 1601 size_t offs = le16_to_cpu(last->e_value_offs); 1602 if (offs < min_offs) 1603 min_offs = offs; 1604 } 1605 } 1606 1607 /* Check whether we have enough space. */ 1608 if (i->value) { 1609 size_t free; 1610 1611 free = min_offs - ((void *)last - s->base) - sizeof(__u32); 1612 if (!s->not_found) 1613 free += EXT4_XATTR_LEN(name_len) + old_size; 1614 1615 if (free < EXT4_XATTR_LEN(name_len) + new_size) { 1616 ret = -ENOSPC; 1617 goto out; 1618 } 1619 1620 /* 1621 * If storing the value in an external inode is an option, 1622 * reserve space for xattr entries/names in the external 1623 * attribute block so that a long value does not occupy the 1624 * whole space and prevent further entries being added. 1625 */ 1626 if (ext4_has_feature_ea_inode(inode->i_sb) && 1627 new_size && is_block && 1628 (min_offs + old_size - new_size) < 1629 EXT4_XATTR_BLOCK_RESERVE(inode)) { 1630 ret = -ENOSPC; 1631 goto out; 1632 } 1633 } 1634 1635 /* 1636 * Getting access to old and new ea inodes is subject to failures. 1637 * Finish that work before doing any modifications to the xattr data. 1638 */ 1639 if (!s->not_found && here->e_value_inum) { 1640 ret = ext4_xattr_inode_iget(inode, 1641 le32_to_cpu(here->e_value_inum), 1642 le32_to_cpu(here->e_hash), 1643 &old_ea_inode); 1644 if (ret) { 1645 old_ea_inode = NULL; 1646 goto out; 1647 } 1648 } 1649 if (i->value && in_inode) { 1650 WARN_ON_ONCE(!i->value_len); 1651 1652 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len); 1653 if (ret) 1654 goto out; 1655 1656 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value, 1657 i->value_len, 1658 &new_ea_inode); 1659 if (ret) { 1660 new_ea_inode = NULL; 1661 ext4_xattr_inode_free_quota(inode, NULL, i->value_len); 1662 goto out; 1663 } 1664 } 1665 1666 if (old_ea_inode) { 1667 /* We are ready to release ref count on the old_ea_inode. */ 1668 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode); 1669 if (ret) { 1670 /* Release newly required ref count on new_ea_inode. */ 1671 if (new_ea_inode) { 1672 int err; 1673 1674 err = ext4_xattr_inode_dec_ref(handle, 1675 new_ea_inode); 1676 if (err) 1677 ext4_warning_inode(new_ea_inode, 1678 "dec ref new_ea_inode err=%d", 1679 err); 1680 ext4_xattr_inode_free_quota(inode, new_ea_inode, 1681 i->value_len); 1682 } 1683 goto out; 1684 } 1685 1686 ext4_xattr_inode_free_quota(inode, old_ea_inode, 1687 le32_to_cpu(here->e_value_size)); 1688 } 1689 1690 /* No failures allowed past this point. */ 1691 1692 if (!s->not_found && here->e_value_size && !here->e_value_inum) { 1693 /* Remove the old value. */ 1694 void *first_val = s->base + min_offs; 1695 size_t offs = le16_to_cpu(here->e_value_offs); 1696 void *val = s->base + offs; 1697 1698 memmove(first_val + old_size, first_val, val - first_val); 1699 memset(first_val, 0, old_size); 1700 min_offs += old_size; 1701 1702 /* Adjust all value offsets. */ 1703 last = s->first; 1704 while (!IS_LAST_ENTRY(last)) { 1705 size_t o = le16_to_cpu(last->e_value_offs); 1706 1707 if (!last->e_value_inum && 1708 last->e_value_size && o < offs) 1709 last->e_value_offs = cpu_to_le16(o + old_size); 1710 last = EXT4_XATTR_NEXT(last); 1711 } 1712 } 1713 1714 if (!i->value) { 1715 /* Remove old name. */ 1716 size_t size = EXT4_XATTR_LEN(name_len); 1717 1718 last = ENTRY((void *)last - size); 1719 memmove(here, (void *)here + size, 1720 (void *)last - (void *)here + sizeof(__u32)); 1721 memset(last, 0, size); 1722 } else if (s->not_found) { 1723 /* Insert new name. */ 1724 size_t size = EXT4_XATTR_LEN(name_len); 1725 size_t rest = (void *)last - (void *)here + sizeof(__u32); 1726 1727 memmove((void *)here + size, here, rest); 1728 memset(here, 0, size); 1729 here->e_name_index = i->name_index; 1730 here->e_name_len = name_len; 1731 memcpy(here->e_name, i->name, name_len); 1732 } else { 1733 /* This is an update, reset value info. */ 1734 here->e_value_inum = 0; 1735 here->e_value_offs = 0; 1736 here->e_value_size = 0; 1737 } 1738 1739 if (i->value) { 1740 /* Insert new value. */ 1741 if (in_inode) { 1742 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino); 1743 } else if (i->value_len) { 1744 void *val = s->base + min_offs - new_size; 1745 1746 here->e_value_offs = cpu_to_le16(min_offs - new_size); 1747 if (i->value == EXT4_ZERO_XATTR_VALUE) { 1748 memset(val, 0, new_size); 1749 } else { 1750 memcpy(val, i->value, i->value_len); 1751 /* Clear padding bytes. */ 1752 memset(val + i->value_len, 0, 1753 new_size - i->value_len); 1754 } 1755 } 1756 here->e_value_size = cpu_to_le32(i->value_len); 1757 } 1758 1759 update_hash: 1760 if (i->value) { 1761 __le32 hash = 0; 1762 1763 /* Entry hash calculation. */ 1764 if (in_inode) { 1765 __le32 crc32c_hash; 1766 1767 /* 1768 * Feed crc32c hash instead of the raw value for entry 1769 * hash calculation. This is to avoid walking 1770 * potentially long value buffer again. 1771 */ 1772 crc32c_hash = cpu_to_le32( 1773 ext4_xattr_inode_get_hash(new_ea_inode)); 1774 hash = ext4_xattr_hash_entry(here->e_name, 1775 here->e_name_len, 1776 &crc32c_hash, 1); 1777 } else if (is_block) { 1778 __le32 *value = s->base + le16_to_cpu( 1779 here->e_value_offs); 1780 1781 hash = ext4_xattr_hash_entry(here->e_name, 1782 here->e_name_len, value, 1783 new_size >> 2); 1784 } 1785 here->e_hash = hash; 1786 } 1787 1788 if (is_block) 1789 ext4_xattr_rehash((struct ext4_xattr_header *)s->base); 1790 1791 ret = 0; 1792 out: 1793 iput(old_ea_inode); 1794 iput(new_ea_inode); 1795 return ret; 1796 } 1797 1798 struct ext4_xattr_block_find { 1799 struct ext4_xattr_search s; 1800 struct buffer_head *bh; 1801 }; 1802 1803 static int 1804 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i, 1805 struct ext4_xattr_block_find *bs) 1806 { 1807 struct super_block *sb = inode->i_sb; 1808 int error; 1809 1810 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld", 1811 i->name_index, i->name, i->value, (long)i->value_len); 1812 1813 if (EXT4_I(inode)->i_file_acl) { 1814 /* The inode already has an extended attribute block. */ 1815 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 1816 if (IS_ERR(bs->bh)) { 1817 error = PTR_ERR(bs->bh); 1818 bs->bh = NULL; 1819 return error; 1820 } 1821 ea_bdebug(bs->bh, "b_count=%d, refcount=%d", 1822 atomic_read(&(bs->bh->b_count)), 1823 le32_to_cpu(BHDR(bs->bh)->h_refcount)); 1824 error = ext4_xattr_check_block(inode, bs->bh); 1825 if (error) 1826 return error; 1827 /* Find the named attribute. */ 1828 bs->s.base = BHDR(bs->bh); 1829 bs->s.first = BFIRST(bs->bh); 1830 bs->s.end = bs->bh->b_data + bs->bh->b_size; 1831 bs->s.here = bs->s.first; 1832 error = xattr_find_entry(inode, &bs->s.here, bs->s.end, 1833 i->name_index, i->name, 1); 1834 if (error && error != -ENODATA) 1835 return error; 1836 bs->s.not_found = error; 1837 } 1838 return 0; 1839 } 1840 1841 static int 1842 ext4_xattr_block_set(handle_t *handle, struct inode *inode, 1843 struct ext4_xattr_info *i, 1844 struct ext4_xattr_block_find *bs) 1845 { 1846 struct super_block *sb = inode->i_sb; 1847 struct buffer_head *new_bh = NULL; 1848 struct ext4_xattr_search s_copy = bs->s; 1849 struct ext4_xattr_search *s = &s_copy; 1850 struct mb_cache_entry *ce = NULL; 1851 int error = 0; 1852 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 1853 struct inode *ea_inode = NULL, *tmp_inode; 1854 size_t old_ea_inode_quota = 0; 1855 unsigned int ea_ino; 1856 1857 1858 #define header(x) ((struct ext4_xattr_header *)(x)) 1859 1860 if (s->base) { 1861 BUFFER_TRACE(bs->bh, "get_write_access"); 1862 error = ext4_journal_get_write_access(handle, sb, bs->bh, 1863 EXT4_JTR_NONE); 1864 if (error) 1865 goto cleanup; 1866 lock_buffer(bs->bh); 1867 1868 if (header(s->base)->h_refcount == cpu_to_le32(1)) { 1869 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash); 1870 1871 /* 1872 * This must happen under buffer lock for 1873 * ext4_xattr_block_set() to reliably detect modified 1874 * block 1875 */ 1876 if (ea_block_cache) 1877 mb_cache_entry_delete(ea_block_cache, hash, 1878 bs->bh->b_blocknr); 1879 ea_bdebug(bs->bh, "modifying in-place"); 1880 error = ext4_xattr_set_entry(i, s, handle, inode, 1881 true /* is_block */); 1882 ext4_xattr_block_csum_set(inode, bs->bh); 1883 unlock_buffer(bs->bh); 1884 if (error == -EFSCORRUPTED) 1885 goto bad_block; 1886 if (!error) 1887 error = ext4_handle_dirty_metadata(handle, 1888 inode, 1889 bs->bh); 1890 if (error) 1891 goto cleanup; 1892 goto inserted; 1893 } else { 1894 int offset = (char *)s->here - bs->bh->b_data; 1895 1896 unlock_buffer(bs->bh); 1897 ea_bdebug(bs->bh, "cloning"); 1898 s->base = kmalloc(bs->bh->b_size, GFP_NOFS); 1899 error = -ENOMEM; 1900 if (s->base == NULL) 1901 goto cleanup; 1902 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size); 1903 s->first = ENTRY(header(s->base)+1); 1904 header(s->base)->h_refcount = cpu_to_le32(1); 1905 s->here = ENTRY(s->base + offset); 1906 s->end = s->base + bs->bh->b_size; 1907 1908 /* 1909 * If existing entry points to an xattr inode, we need 1910 * to prevent ext4_xattr_set_entry() from decrementing 1911 * ref count on it because the reference belongs to the 1912 * original block. In this case, make the entry look 1913 * like it has an empty value. 1914 */ 1915 if (!s->not_found && s->here->e_value_inum) { 1916 ea_ino = le32_to_cpu(s->here->e_value_inum); 1917 error = ext4_xattr_inode_iget(inode, ea_ino, 1918 le32_to_cpu(s->here->e_hash), 1919 &tmp_inode); 1920 if (error) 1921 goto cleanup; 1922 1923 if (!ext4_test_inode_state(tmp_inode, 1924 EXT4_STATE_LUSTRE_EA_INODE)) { 1925 /* 1926 * Defer quota free call for previous 1927 * inode until success is guaranteed. 1928 */ 1929 old_ea_inode_quota = le32_to_cpu( 1930 s->here->e_value_size); 1931 } 1932 iput(tmp_inode); 1933 1934 s->here->e_value_inum = 0; 1935 s->here->e_value_size = 0; 1936 } 1937 } 1938 } else { 1939 /* Allocate a buffer where we construct the new block. */ 1940 s->base = kzalloc(sb->s_blocksize, GFP_NOFS); 1941 error = -ENOMEM; 1942 if (s->base == NULL) 1943 goto cleanup; 1944 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC); 1945 header(s->base)->h_blocks = cpu_to_le32(1); 1946 header(s->base)->h_refcount = cpu_to_le32(1); 1947 s->first = ENTRY(header(s->base)+1); 1948 s->here = ENTRY(header(s->base)+1); 1949 s->end = s->base + sb->s_blocksize; 1950 } 1951 1952 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */); 1953 if (error == -EFSCORRUPTED) 1954 goto bad_block; 1955 if (error) 1956 goto cleanup; 1957 1958 if (i->value && s->here->e_value_inum) { 1959 /* 1960 * A ref count on ea_inode has been taken as part of the call to 1961 * ext4_xattr_set_entry() above. We would like to drop this 1962 * extra ref but we have to wait until the xattr block is 1963 * initialized and has its own ref count on the ea_inode. 1964 */ 1965 ea_ino = le32_to_cpu(s->here->e_value_inum); 1966 error = ext4_xattr_inode_iget(inode, ea_ino, 1967 le32_to_cpu(s->here->e_hash), 1968 &ea_inode); 1969 if (error) { 1970 ea_inode = NULL; 1971 goto cleanup; 1972 } 1973 } 1974 1975 inserted: 1976 if (!IS_LAST_ENTRY(s->first)) { 1977 new_bh = ext4_xattr_block_cache_find(inode, header(s->base), 1978 &ce); 1979 if (new_bh) { 1980 /* We found an identical block in the cache. */ 1981 if (new_bh == bs->bh) 1982 ea_bdebug(new_bh, "keeping"); 1983 else { 1984 u32 ref; 1985 1986 WARN_ON_ONCE(dquot_initialize_needed(inode)); 1987 1988 /* The old block is released after updating 1989 the inode. */ 1990 error = dquot_alloc_block(inode, 1991 EXT4_C2B(EXT4_SB(sb), 1)); 1992 if (error) 1993 goto cleanup; 1994 BUFFER_TRACE(new_bh, "get_write_access"); 1995 error = ext4_journal_get_write_access( 1996 handle, sb, new_bh, 1997 EXT4_JTR_NONE); 1998 if (error) 1999 goto cleanup_dquot; 2000 lock_buffer(new_bh); 2001 /* 2002 * We have to be careful about races with 2003 * freeing, rehashing or adding references to 2004 * xattr block. Once we hold buffer lock xattr 2005 * block's state is stable so we can check 2006 * whether the block got freed / rehashed or 2007 * not. Since we unhash mbcache entry under 2008 * buffer lock when freeing / rehashing xattr 2009 * block, checking whether entry is still 2010 * hashed is reliable. Same rules hold for 2011 * e_reusable handling. 2012 */ 2013 if (hlist_bl_unhashed(&ce->e_hash_list) || 2014 !ce->e_reusable) { 2015 /* 2016 * Undo everything and check mbcache 2017 * again. 2018 */ 2019 unlock_buffer(new_bh); 2020 dquot_free_block(inode, 2021 EXT4_C2B(EXT4_SB(sb), 2022 1)); 2023 brelse(new_bh); 2024 mb_cache_entry_put(ea_block_cache, ce); 2025 ce = NULL; 2026 new_bh = NULL; 2027 goto inserted; 2028 } 2029 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1; 2030 BHDR(new_bh)->h_refcount = cpu_to_le32(ref); 2031 if (ref >= EXT4_XATTR_REFCOUNT_MAX) 2032 ce->e_reusable = 0; 2033 ea_bdebug(new_bh, "reusing; refcount now=%d", 2034 ref); 2035 ext4_xattr_block_csum_set(inode, new_bh); 2036 unlock_buffer(new_bh); 2037 error = ext4_handle_dirty_metadata(handle, 2038 inode, 2039 new_bh); 2040 if (error) 2041 goto cleanup_dquot; 2042 } 2043 mb_cache_entry_touch(ea_block_cache, ce); 2044 mb_cache_entry_put(ea_block_cache, ce); 2045 ce = NULL; 2046 } else if (bs->bh && s->base == bs->bh->b_data) { 2047 /* We were modifying this block in-place. */ 2048 ea_bdebug(bs->bh, "keeping this block"); 2049 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh); 2050 new_bh = bs->bh; 2051 get_bh(new_bh); 2052 } else { 2053 /* We need to allocate a new block */ 2054 ext4_fsblk_t goal, block; 2055 2056 WARN_ON_ONCE(dquot_initialize_needed(inode)); 2057 2058 goal = ext4_group_first_block_no(sb, 2059 EXT4_I(inode)->i_block_group); 2060 2061 /* non-extent files can't have physical blocks past 2^32 */ 2062 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 2063 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; 2064 2065 block = ext4_new_meta_blocks(handle, inode, goal, 0, 2066 NULL, &error); 2067 if (error) 2068 goto cleanup; 2069 2070 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 2071 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS); 2072 2073 ea_idebug(inode, "creating block %llu", 2074 (unsigned long long)block); 2075 2076 new_bh = sb_getblk(sb, block); 2077 if (unlikely(!new_bh)) { 2078 error = -ENOMEM; 2079 getblk_failed: 2080 ext4_free_blocks(handle, inode, NULL, block, 1, 2081 EXT4_FREE_BLOCKS_METADATA); 2082 goto cleanup; 2083 } 2084 error = ext4_xattr_inode_inc_ref_all(handle, inode, 2085 ENTRY(header(s->base)+1)); 2086 if (error) 2087 goto getblk_failed; 2088 if (ea_inode) { 2089 /* Drop the extra ref on ea_inode. */ 2090 error = ext4_xattr_inode_dec_ref(handle, 2091 ea_inode); 2092 if (error) 2093 ext4_warning_inode(ea_inode, 2094 "dec ref error=%d", 2095 error); 2096 iput(ea_inode); 2097 ea_inode = NULL; 2098 } 2099 2100 lock_buffer(new_bh); 2101 error = ext4_journal_get_create_access(handle, sb, 2102 new_bh, EXT4_JTR_NONE); 2103 if (error) { 2104 unlock_buffer(new_bh); 2105 error = -EIO; 2106 goto getblk_failed; 2107 } 2108 memcpy(new_bh->b_data, s->base, new_bh->b_size); 2109 ext4_xattr_block_csum_set(inode, new_bh); 2110 set_buffer_uptodate(new_bh); 2111 unlock_buffer(new_bh); 2112 ext4_xattr_block_cache_insert(ea_block_cache, new_bh); 2113 error = ext4_handle_dirty_metadata(handle, inode, 2114 new_bh); 2115 if (error) 2116 goto cleanup; 2117 } 2118 } 2119 2120 if (old_ea_inode_quota) 2121 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota); 2122 2123 /* Update the inode. */ 2124 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0; 2125 2126 /* Drop the previous xattr block. */ 2127 if (bs->bh && bs->bh != new_bh) { 2128 struct ext4_xattr_inode_array *ea_inode_array = NULL; 2129 2130 ext4_xattr_release_block(handle, inode, bs->bh, 2131 &ea_inode_array, 2132 0 /* extra_credits */); 2133 ext4_xattr_inode_array_free(ea_inode_array); 2134 } 2135 error = 0; 2136 2137 cleanup: 2138 if (ea_inode) { 2139 int error2; 2140 2141 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode); 2142 if (error2) 2143 ext4_warning_inode(ea_inode, "dec ref error=%d", 2144 error2); 2145 2146 /* If there was an error, revert the quota charge. */ 2147 if (error) 2148 ext4_xattr_inode_free_quota(inode, ea_inode, 2149 i_size_read(ea_inode)); 2150 iput(ea_inode); 2151 } 2152 if (ce) 2153 mb_cache_entry_put(ea_block_cache, ce); 2154 brelse(new_bh); 2155 if (!(bs->bh && s->base == bs->bh->b_data)) 2156 kfree(s->base); 2157 2158 return error; 2159 2160 cleanup_dquot: 2161 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1)); 2162 goto cleanup; 2163 2164 bad_block: 2165 EXT4_ERROR_INODE(inode, "bad block %llu", 2166 EXT4_I(inode)->i_file_acl); 2167 goto cleanup; 2168 2169 #undef header 2170 } 2171 2172 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i, 2173 struct ext4_xattr_ibody_find *is) 2174 { 2175 struct ext4_xattr_ibody_header *header; 2176 struct ext4_inode *raw_inode; 2177 int error; 2178 2179 if (EXT4_I(inode)->i_extra_isize == 0) 2180 return 0; 2181 raw_inode = ext4_raw_inode(&is->iloc); 2182 header = IHDR(inode, raw_inode); 2183 is->s.base = is->s.first = IFIRST(header); 2184 is->s.here = is->s.first; 2185 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 2186 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { 2187 error = xattr_check_inode(inode, header, is->s.end); 2188 if (error) 2189 return error; 2190 /* Find the named attribute. */ 2191 error = xattr_find_entry(inode, &is->s.here, is->s.end, 2192 i->name_index, i->name, 0); 2193 if (error && error != -ENODATA) 2194 return error; 2195 is->s.not_found = error; 2196 } 2197 return 0; 2198 } 2199 2200 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode, 2201 struct ext4_xattr_info *i, 2202 struct ext4_xattr_ibody_find *is) 2203 { 2204 struct ext4_xattr_ibody_header *header; 2205 struct ext4_xattr_search *s = &is->s; 2206 int error; 2207 2208 if (EXT4_I(inode)->i_extra_isize == 0) 2209 return -ENOSPC; 2210 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */); 2211 if (error) 2212 return error; 2213 header = IHDR(inode, ext4_raw_inode(&is->iloc)); 2214 if (!IS_LAST_ENTRY(s->first)) { 2215 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC); 2216 ext4_set_inode_state(inode, EXT4_STATE_XATTR); 2217 } else { 2218 header->h_magic = cpu_to_le32(0); 2219 ext4_clear_inode_state(inode, EXT4_STATE_XATTR); 2220 } 2221 return 0; 2222 } 2223 2224 static int ext4_xattr_value_same(struct ext4_xattr_search *s, 2225 struct ext4_xattr_info *i) 2226 { 2227 void *value; 2228 2229 /* When e_value_inum is set the value is stored externally. */ 2230 if (s->here->e_value_inum) 2231 return 0; 2232 if (le32_to_cpu(s->here->e_value_size) != i->value_len) 2233 return 0; 2234 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs); 2235 return !memcmp(value, i->value, i->value_len); 2236 } 2237 2238 static struct buffer_head *ext4_xattr_get_block(struct inode *inode) 2239 { 2240 struct buffer_head *bh; 2241 int error; 2242 2243 if (!EXT4_I(inode)->i_file_acl) 2244 return NULL; 2245 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 2246 if (IS_ERR(bh)) 2247 return bh; 2248 error = ext4_xattr_check_block(inode, bh); 2249 if (error) { 2250 brelse(bh); 2251 return ERR_PTR(error); 2252 } 2253 return bh; 2254 } 2255 2256 /* 2257 * ext4_xattr_set_handle() 2258 * 2259 * Create, replace or remove an extended attribute for this inode. Value 2260 * is NULL to remove an existing extended attribute, and non-NULL to 2261 * either replace an existing extended attribute, or create a new extended 2262 * attribute. The flags XATTR_REPLACE and XATTR_CREATE 2263 * specify that an extended attribute must exist and must not exist 2264 * previous to the call, respectively. 2265 * 2266 * Returns 0, or a negative error number on failure. 2267 */ 2268 int 2269 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index, 2270 const char *name, const void *value, size_t value_len, 2271 int flags) 2272 { 2273 struct ext4_xattr_info i = { 2274 .name_index = name_index, 2275 .name = name, 2276 .value = value, 2277 .value_len = value_len, 2278 .in_inode = 0, 2279 }; 2280 struct ext4_xattr_ibody_find is = { 2281 .s = { .not_found = -ENODATA, }, 2282 }; 2283 struct ext4_xattr_block_find bs = { 2284 .s = { .not_found = -ENODATA, }, 2285 }; 2286 int no_expand; 2287 int error; 2288 2289 if (!name) 2290 return -EINVAL; 2291 if (strlen(name) > 255) 2292 return -ERANGE; 2293 2294 ext4_write_lock_xattr(inode, &no_expand); 2295 2296 /* Check journal credits under write lock. */ 2297 if (ext4_handle_valid(handle)) { 2298 struct buffer_head *bh; 2299 int credits; 2300 2301 bh = ext4_xattr_get_block(inode); 2302 if (IS_ERR(bh)) { 2303 error = PTR_ERR(bh); 2304 goto cleanup; 2305 } 2306 2307 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh, 2308 value_len, 2309 flags & XATTR_CREATE); 2310 brelse(bh); 2311 2312 if (jbd2_handle_buffer_credits(handle) < credits) { 2313 error = -ENOSPC; 2314 goto cleanup; 2315 } 2316 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS)); 2317 } 2318 2319 error = ext4_reserve_inode_write(handle, inode, &is.iloc); 2320 if (error) 2321 goto cleanup; 2322 2323 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) { 2324 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc); 2325 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); 2326 ext4_clear_inode_state(inode, EXT4_STATE_NEW); 2327 } 2328 2329 error = ext4_xattr_ibody_find(inode, &i, &is); 2330 if (error) 2331 goto cleanup; 2332 if (is.s.not_found) 2333 error = ext4_xattr_block_find(inode, &i, &bs); 2334 if (error) 2335 goto cleanup; 2336 if (is.s.not_found && bs.s.not_found) { 2337 error = -ENODATA; 2338 if (flags & XATTR_REPLACE) 2339 goto cleanup; 2340 error = 0; 2341 if (!value) 2342 goto cleanup; 2343 } else { 2344 error = -EEXIST; 2345 if (flags & XATTR_CREATE) 2346 goto cleanup; 2347 } 2348 2349 if (!value) { 2350 if (!is.s.not_found) 2351 error = ext4_xattr_ibody_set(handle, inode, &i, &is); 2352 else if (!bs.s.not_found) 2353 error = ext4_xattr_block_set(handle, inode, &i, &bs); 2354 } else { 2355 error = 0; 2356 /* Xattr value did not change? Save us some work and bail out */ 2357 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i)) 2358 goto cleanup; 2359 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i)) 2360 goto cleanup; 2361 2362 if (ext4_has_feature_ea_inode(inode->i_sb) && 2363 (EXT4_XATTR_SIZE(i.value_len) > 2364 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize))) 2365 i.in_inode = 1; 2366 retry_inode: 2367 error = ext4_xattr_ibody_set(handle, inode, &i, &is); 2368 if (!error && !bs.s.not_found) { 2369 i.value = NULL; 2370 error = ext4_xattr_block_set(handle, inode, &i, &bs); 2371 } else if (error == -ENOSPC) { 2372 if (EXT4_I(inode)->i_file_acl && !bs.s.base) { 2373 brelse(bs.bh); 2374 bs.bh = NULL; 2375 error = ext4_xattr_block_find(inode, &i, &bs); 2376 if (error) 2377 goto cleanup; 2378 } 2379 error = ext4_xattr_block_set(handle, inode, &i, &bs); 2380 if (!error && !is.s.not_found) { 2381 i.value = NULL; 2382 error = ext4_xattr_ibody_set(handle, inode, &i, 2383 &is); 2384 } else if (error == -ENOSPC) { 2385 /* 2386 * Xattr does not fit in the block, store at 2387 * external inode if possible. 2388 */ 2389 if (ext4_has_feature_ea_inode(inode->i_sb) && 2390 i.value_len && !i.in_inode) { 2391 i.in_inode = 1; 2392 goto retry_inode; 2393 } 2394 } 2395 } 2396 } 2397 if (!error) { 2398 ext4_xattr_update_super_block(handle, inode->i_sb); 2399 inode->i_ctime = current_time(inode); 2400 if (!value) 2401 no_expand = 0; 2402 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); 2403 /* 2404 * The bh is consumed by ext4_mark_iloc_dirty, even with 2405 * error != 0. 2406 */ 2407 is.iloc.bh = NULL; 2408 if (IS_SYNC(inode)) 2409 ext4_handle_sync(handle); 2410 } 2411 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR); 2412 2413 cleanup: 2414 brelse(is.iloc.bh); 2415 brelse(bs.bh); 2416 ext4_write_unlock_xattr(inode, &no_expand); 2417 return error; 2418 } 2419 2420 int ext4_xattr_set_credits(struct inode *inode, size_t value_len, 2421 bool is_create, int *credits) 2422 { 2423 struct buffer_head *bh; 2424 int err; 2425 2426 *credits = 0; 2427 2428 if (!EXT4_SB(inode->i_sb)->s_journal) 2429 return 0; 2430 2431 down_read(&EXT4_I(inode)->xattr_sem); 2432 2433 bh = ext4_xattr_get_block(inode); 2434 if (IS_ERR(bh)) { 2435 err = PTR_ERR(bh); 2436 } else { 2437 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh, 2438 value_len, is_create); 2439 brelse(bh); 2440 err = 0; 2441 } 2442 2443 up_read(&EXT4_I(inode)->xattr_sem); 2444 return err; 2445 } 2446 2447 /* 2448 * ext4_xattr_set() 2449 * 2450 * Like ext4_xattr_set_handle, but start from an inode. This extended 2451 * attribute modification is a filesystem transaction by itself. 2452 * 2453 * Returns 0, or a negative error number on failure. 2454 */ 2455 int 2456 ext4_xattr_set(struct inode *inode, int name_index, const char *name, 2457 const void *value, size_t value_len, int flags) 2458 { 2459 handle_t *handle; 2460 struct super_block *sb = inode->i_sb; 2461 int error, retries = 0; 2462 int credits; 2463 2464 error = dquot_initialize(inode); 2465 if (error) 2466 return error; 2467 2468 retry: 2469 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE, 2470 &credits); 2471 if (error) 2472 return error; 2473 2474 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits); 2475 if (IS_ERR(handle)) { 2476 error = PTR_ERR(handle); 2477 } else { 2478 int error2; 2479 2480 error = ext4_xattr_set_handle(handle, inode, name_index, name, 2481 value, value_len, flags); 2482 error2 = ext4_journal_stop(handle); 2483 if (error == -ENOSPC && 2484 ext4_should_retry_alloc(sb, &retries)) 2485 goto retry; 2486 if (error == 0) 2487 error = error2; 2488 } 2489 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR); 2490 2491 return error; 2492 } 2493 2494 /* 2495 * Shift the EA entries in the inode to create space for the increased 2496 * i_extra_isize. 2497 */ 2498 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry, 2499 int value_offs_shift, void *to, 2500 void *from, size_t n) 2501 { 2502 struct ext4_xattr_entry *last = entry; 2503 int new_offs; 2504 2505 /* We always shift xattr headers further thus offsets get lower */ 2506 BUG_ON(value_offs_shift > 0); 2507 2508 /* Adjust the value offsets of the entries */ 2509 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { 2510 if (!last->e_value_inum && last->e_value_size) { 2511 new_offs = le16_to_cpu(last->e_value_offs) + 2512 value_offs_shift; 2513 last->e_value_offs = cpu_to_le16(new_offs); 2514 } 2515 } 2516 /* Shift the entries by n bytes */ 2517 memmove(to, from, n); 2518 } 2519 2520 /* 2521 * Move xattr pointed to by 'entry' from inode into external xattr block 2522 */ 2523 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode, 2524 struct ext4_inode *raw_inode, 2525 struct ext4_xattr_entry *entry) 2526 { 2527 struct ext4_xattr_ibody_find *is = NULL; 2528 struct ext4_xattr_block_find *bs = NULL; 2529 char *buffer = NULL, *b_entry_name = NULL; 2530 size_t value_size = le32_to_cpu(entry->e_value_size); 2531 struct ext4_xattr_info i = { 2532 .value = NULL, 2533 .value_len = 0, 2534 .name_index = entry->e_name_index, 2535 .in_inode = !!entry->e_value_inum, 2536 }; 2537 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode); 2538 int error; 2539 2540 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS); 2541 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS); 2542 buffer = kmalloc(value_size, GFP_NOFS); 2543 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS); 2544 if (!is || !bs || !buffer || !b_entry_name) { 2545 error = -ENOMEM; 2546 goto out; 2547 } 2548 2549 is->s.not_found = -ENODATA; 2550 bs->s.not_found = -ENODATA; 2551 is->iloc.bh = NULL; 2552 bs->bh = NULL; 2553 2554 /* Save the entry name and the entry value */ 2555 if (entry->e_value_inum) { 2556 error = ext4_xattr_inode_get(inode, entry, buffer, value_size); 2557 if (error) 2558 goto out; 2559 } else { 2560 size_t value_offs = le16_to_cpu(entry->e_value_offs); 2561 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size); 2562 } 2563 2564 memcpy(b_entry_name, entry->e_name, entry->e_name_len); 2565 b_entry_name[entry->e_name_len] = '\0'; 2566 i.name = b_entry_name; 2567 2568 error = ext4_get_inode_loc(inode, &is->iloc); 2569 if (error) 2570 goto out; 2571 2572 error = ext4_xattr_ibody_find(inode, &i, is); 2573 if (error) 2574 goto out; 2575 2576 /* Remove the chosen entry from the inode */ 2577 error = ext4_xattr_ibody_set(handle, inode, &i, is); 2578 if (error) 2579 goto out; 2580 2581 i.value = buffer; 2582 i.value_len = value_size; 2583 error = ext4_xattr_block_find(inode, &i, bs); 2584 if (error) 2585 goto out; 2586 2587 /* Add entry which was removed from the inode into the block */ 2588 error = ext4_xattr_block_set(handle, inode, &i, bs); 2589 if (error) 2590 goto out; 2591 error = 0; 2592 out: 2593 kfree(b_entry_name); 2594 kfree(buffer); 2595 if (is) 2596 brelse(is->iloc.bh); 2597 if (bs) 2598 brelse(bs->bh); 2599 kfree(is); 2600 kfree(bs); 2601 2602 return error; 2603 } 2604 2605 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode, 2606 struct ext4_inode *raw_inode, 2607 int isize_diff, size_t ifree, 2608 size_t bfree, int *total_ino) 2609 { 2610 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode); 2611 struct ext4_xattr_entry *small_entry; 2612 struct ext4_xattr_entry *entry; 2613 struct ext4_xattr_entry *last; 2614 unsigned int entry_size; /* EA entry size */ 2615 unsigned int total_size; /* EA entry size + value size */ 2616 unsigned int min_total_size; 2617 int error; 2618 2619 while (isize_diff > ifree) { 2620 entry = NULL; 2621 small_entry = NULL; 2622 min_total_size = ~0U; 2623 last = IFIRST(header); 2624 /* Find the entry best suited to be pushed into EA block */ 2625 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) { 2626 /* never move system.data out of the inode */ 2627 if ((last->e_name_len == 4) && 2628 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) && 2629 !memcmp(last->e_name, "data", 4)) 2630 continue; 2631 total_size = EXT4_XATTR_LEN(last->e_name_len); 2632 if (!last->e_value_inum) 2633 total_size += EXT4_XATTR_SIZE( 2634 le32_to_cpu(last->e_value_size)); 2635 if (total_size <= bfree && 2636 total_size < min_total_size) { 2637 if (total_size + ifree < isize_diff) { 2638 small_entry = last; 2639 } else { 2640 entry = last; 2641 min_total_size = total_size; 2642 } 2643 } 2644 } 2645 2646 if (entry == NULL) { 2647 if (small_entry == NULL) 2648 return -ENOSPC; 2649 entry = small_entry; 2650 } 2651 2652 entry_size = EXT4_XATTR_LEN(entry->e_name_len); 2653 total_size = entry_size; 2654 if (!entry->e_value_inum) 2655 total_size += EXT4_XATTR_SIZE( 2656 le32_to_cpu(entry->e_value_size)); 2657 error = ext4_xattr_move_to_block(handle, inode, raw_inode, 2658 entry); 2659 if (error) 2660 return error; 2661 2662 *total_ino -= entry_size; 2663 ifree += total_size; 2664 bfree -= total_size; 2665 } 2666 2667 return 0; 2668 } 2669 2670 /* 2671 * Expand an inode by new_extra_isize bytes when EAs are present. 2672 * Returns 0 on success or negative error number on failure. 2673 */ 2674 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize, 2675 struct ext4_inode *raw_inode, handle_t *handle) 2676 { 2677 struct ext4_xattr_ibody_header *header; 2678 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 2679 static unsigned int mnt_count; 2680 size_t min_offs; 2681 size_t ifree, bfree; 2682 int total_ino; 2683 void *base, *end; 2684 int error = 0, tried_min_extra_isize = 0; 2685 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize); 2686 int isize_diff; /* How much do we need to grow i_extra_isize */ 2687 2688 retry: 2689 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize; 2690 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) 2691 return 0; 2692 2693 header = IHDR(inode, raw_inode); 2694 2695 /* 2696 * Check if enough free space is available in the inode to shift the 2697 * entries ahead by new_extra_isize. 2698 */ 2699 2700 base = IFIRST(header); 2701 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size; 2702 min_offs = end - base; 2703 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32); 2704 2705 error = xattr_check_inode(inode, header, end); 2706 if (error) 2707 goto cleanup; 2708 2709 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino); 2710 if (ifree >= isize_diff) 2711 goto shift; 2712 2713 /* 2714 * Enough free space isn't available in the inode, check if 2715 * EA block can hold new_extra_isize bytes. 2716 */ 2717 if (EXT4_I(inode)->i_file_acl) { 2718 struct buffer_head *bh; 2719 2720 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 2721 if (IS_ERR(bh)) { 2722 error = PTR_ERR(bh); 2723 goto cleanup; 2724 } 2725 error = ext4_xattr_check_block(inode, bh); 2726 if (error) { 2727 brelse(bh); 2728 goto cleanup; 2729 } 2730 base = BHDR(bh); 2731 end = bh->b_data + bh->b_size; 2732 min_offs = end - base; 2733 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base, 2734 NULL); 2735 brelse(bh); 2736 if (bfree + ifree < isize_diff) { 2737 if (!tried_min_extra_isize && s_min_extra_isize) { 2738 tried_min_extra_isize++; 2739 new_extra_isize = s_min_extra_isize; 2740 goto retry; 2741 } 2742 error = -ENOSPC; 2743 goto cleanup; 2744 } 2745 } else { 2746 bfree = inode->i_sb->s_blocksize; 2747 } 2748 2749 error = ext4_xattr_make_inode_space(handle, inode, raw_inode, 2750 isize_diff, ifree, bfree, 2751 &total_ino); 2752 if (error) { 2753 if (error == -ENOSPC && !tried_min_extra_isize && 2754 s_min_extra_isize) { 2755 tried_min_extra_isize++; 2756 new_extra_isize = s_min_extra_isize; 2757 goto retry; 2758 } 2759 goto cleanup; 2760 } 2761 shift: 2762 /* Adjust the offsets and shift the remaining entries ahead */ 2763 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize 2764 - new_extra_isize, (void *)raw_inode + 2765 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize, 2766 (void *)header, total_ino); 2767 EXT4_I(inode)->i_extra_isize = new_extra_isize; 2768 2769 cleanup: 2770 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) { 2771 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.", 2772 inode->i_ino); 2773 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count); 2774 } 2775 return error; 2776 } 2777 2778 #define EIA_INCR 16 /* must be 2^n */ 2779 #define EIA_MASK (EIA_INCR - 1) 2780 2781 /* Add the large xattr @inode into @ea_inode_array for deferred iput(). 2782 * If @ea_inode_array is new or full it will be grown and the old 2783 * contents copied over. 2784 */ 2785 static int 2786 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array, 2787 struct inode *inode) 2788 { 2789 if (*ea_inode_array == NULL) { 2790 /* 2791 * Start with 15 inodes, so it fits into a power-of-two size. 2792 * If *ea_inode_array is NULL, this is essentially offsetof() 2793 */ 2794 (*ea_inode_array) = 2795 kmalloc(offsetof(struct ext4_xattr_inode_array, 2796 inodes[EIA_MASK]), 2797 GFP_NOFS); 2798 if (*ea_inode_array == NULL) 2799 return -ENOMEM; 2800 (*ea_inode_array)->count = 0; 2801 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) { 2802 /* expand the array once all 15 + n * 16 slots are full */ 2803 struct ext4_xattr_inode_array *new_array = NULL; 2804 int count = (*ea_inode_array)->count; 2805 2806 /* if new_array is NULL, this is essentially offsetof() */ 2807 new_array = kmalloc( 2808 offsetof(struct ext4_xattr_inode_array, 2809 inodes[count + EIA_INCR]), 2810 GFP_NOFS); 2811 if (new_array == NULL) 2812 return -ENOMEM; 2813 memcpy(new_array, *ea_inode_array, 2814 offsetof(struct ext4_xattr_inode_array, inodes[count])); 2815 kfree(*ea_inode_array); 2816 *ea_inode_array = new_array; 2817 } 2818 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode; 2819 return 0; 2820 } 2821 2822 /* 2823 * ext4_xattr_delete_inode() 2824 * 2825 * Free extended attribute resources associated with this inode. Traverse 2826 * all entries and decrement reference on any xattr inodes associated with this 2827 * inode. This is called immediately before an inode is freed. We have exclusive 2828 * access to the inode. If an orphan inode is deleted it will also release its 2829 * references on xattr block and xattr inodes. 2830 */ 2831 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode, 2832 struct ext4_xattr_inode_array **ea_inode_array, 2833 int extra_credits) 2834 { 2835 struct buffer_head *bh = NULL; 2836 struct ext4_xattr_ibody_header *header; 2837 struct ext4_iloc iloc = { .bh = NULL }; 2838 struct ext4_xattr_entry *entry; 2839 struct inode *ea_inode; 2840 int error; 2841 2842 error = ext4_journal_ensure_credits(handle, extra_credits, 2843 ext4_free_metadata_revoke_credits(inode->i_sb, 1)); 2844 if (error < 0) { 2845 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error); 2846 goto cleanup; 2847 } 2848 2849 if (ext4_has_feature_ea_inode(inode->i_sb) && 2850 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { 2851 2852 error = ext4_get_inode_loc(inode, &iloc); 2853 if (error) { 2854 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error); 2855 goto cleanup; 2856 } 2857 2858 error = ext4_journal_get_write_access(handle, inode->i_sb, 2859 iloc.bh, EXT4_JTR_NONE); 2860 if (error) { 2861 EXT4_ERROR_INODE(inode, "write access (error %d)", 2862 error); 2863 goto cleanup; 2864 } 2865 2866 header = IHDR(inode, ext4_raw_inode(&iloc)); 2867 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC)) 2868 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh, 2869 IFIRST(header), 2870 false /* block_csum */, 2871 ea_inode_array, 2872 extra_credits, 2873 false /* skip_quota */); 2874 } 2875 2876 if (EXT4_I(inode)->i_file_acl) { 2877 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO); 2878 if (IS_ERR(bh)) { 2879 error = PTR_ERR(bh); 2880 if (error == -EIO) { 2881 EXT4_ERROR_INODE_ERR(inode, EIO, 2882 "block %llu read error", 2883 EXT4_I(inode)->i_file_acl); 2884 } 2885 bh = NULL; 2886 goto cleanup; 2887 } 2888 error = ext4_xattr_check_block(inode, bh); 2889 if (error) 2890 goto cleanup; 2891 2892 if (ext4_has_feature_ea_inode(inode->i_sb)) { 2893 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry); 2894 entry = EXT4_XATTR_NEXT(entry)) { 2895 if (!entry->e_value_inum) 2896 continue; 2897 error = ext4_xattr_inode_iget(inode, 2898 le32_to_cpu(entry->e_value_inum), 2899 le32_to_cpu(entry->e_hash), 2900 &ea_inode); 2901 if (error) 2902 continue; 2903 ext4_xattr_inode_free_quota(inode, ea_inode, 2904 le32_to_cpu(entry->e_value_size)); 2905 iput(ea_inode); 2906 } 2907 2908 } 2909 2910 ext4_xattr_release_block(handle, inode, bh, ea_inode_array, 2911 extra_credits); 2912 /* 2913 * Update i_file_acl value in the same transaction that releases 2914 * block. 2915 */ 2916 EXT4_I(inode)->i_file_acl = 0; 2917 error = ext4_mark_inode_dirty(handle, inode); 2918 if (error) { 2919 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)", 2920 error); 2921 goto cleanup; 2922 } 2923 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR); 2924 } 2925 error = 0; 2926 cleanup: 2927 brelse(iloc.bh); 2928 brelse(bh); 2929 return error; 2930 } 2931 2932 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array) 2933 { 2934 int idx; 2935 2936 if (ea_inode_array == NULL) 2937 return; 2938 2939 for (idx = 0; idx < ea_inode_array->count; ++idx) 2940 iput(ea_inode_array->inodes[idx]); 2941 kfree(ea_inode_array); 2942 } 2943 2944 /* 2945 * ext4_xattr_block_cache_insert() 2946 * 2947 * Create a new entry in the extended attribute block cache, and insert 2948 * it unless such an entry is already in the cache. 2949 * 2950 * Returns 0, or a negative error number on failure. 2951 */ 2952 static void 2953 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache, 2954 struct buffer_head *bh) 2955 { 2956 struct ext4_xattr_header *header = BHDR(bh); 2957 __u32 hash = le32_to_cpu(header->h_hash); 2958 int reusable = le32_to_cpu(header->h_refcount) < 2959 EXT4_XATTR_REFCOUNT_MAX; 2960 int error; 2961 2962 if (!ea_block_cache) 2963 return; 2964 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash, 2965 bh->b_blocknr, reusable); 2966 if (error) { 2967 if (error == -EBUSY) 2968 ea_bdebug(bh, "already in cache"); 2969 } else 2970 ea_bdebug(bh, "inserting [%x]", (int)hash); 2971 } 2972 2973 /* 2974 * ext4_xattr_cmp() 2975 * 2976 * Compare two extended attribute blocks for equality. 2977 * 2978 * Returns 0 if the blocks are equal, 1 if they differ, and 2979 * a negative error number on errors. 2980 */ 2981 static int 2982 ext4_xattr_cmp(struct ext4_xattr_header *header1, 2983 struct ext4_xattr_header *header2) 2984 { 2985 struct ext4_xattr_entry *entry1, *entry2; 2986 2987 entry1 = ENTRY(header1+1); 2988 entry2 = ENTRY(header2+1); 2989 while (!IS_LAST_ENTRY(entry1)) { 2990 if (IS_LAST_ENTRY(entry2)) 2991 return 1; 2992 if (entry1->e_hash != entry2->e_hash || 2993 entry1->e_name_index != entry2->e_name_index || 2994 entry1->e_name_len != entry2->e_name_len || 2995 entry1->e_value_size != entry2->e_value_size || 2996 entry1->e_value_inum != entry2->e_value_inum || 2997 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len)) 2998 return 1; 2999 if (!entry1->e_value_inum && 3000 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs), 3001 (char *)header2 + le16_to_cpu(entry2->e_value_offs), 3002 le32_to_cpu(entry1->e_value_size))) 3003 return 1; 3004 3005 entry1 = EXT4_XATTR_NEXT(entry1); 3006 entry2 = EXT4_XATTR_NEXT(entry2); 3007 } 3008 if (!IS_LAST_ENTRY(entry2)) 3009 return 1; 3010 return 0; 3011 } 3012 3013 /* 3014 * ext4_xattr_block_cache_find() 3015 * 3016 * Find an identical extended attribute block. 3017 * 3018 * Returns a pointer to the block found, or NULL if such a block was 3019 * not found or an error occurred. 3020 */ 3021 static struct buffer_head * 3022 ext4_xattr_block_cache_find(struct inode *inode, 3023 struct ext4_xattr_header *header, 3024 struct mb_cache_entry **pce) 3025 { 3026 __u32 hash = le32_to_cpu(header->h_hash); 3027 struct mb_cache_entry *ce; 3028 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 3029 3030 if (!ea_block_cache) 3031 return NULL; 3032 if (!header->h_hash) 3033 return NULL; /* never share */ 3034 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash); 3035 ce = mb_cache_entry_find_first(ea_block_cache, hash); 3036 while (ce) { 3037 struct buffer_head *bh; 3038 3039 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO); 3040 if (IS_ERR(bh)) { 3041 if (PTR_ERR(bh) == -ENOMEM) 3042 return NULL; 3043 bh = NULL; 3044 EXT4_ERROR_INODE(inode, "block %lu read error", 3045 (unsigned long)ce->e_value); 3046 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) { 3047 *pce = ce; 3048 return bh; 3049 } 3050 brelse(bh); 3051 ce = mb_cache_entry_find_next(ea_block_cache, ce); 3052 } 3053 return NULL; 3054 } 3055 3056 #define NAME_HASH_SHIFT 5 3057 #define VALUE_HASH_SHIFT 16 3058 3059 /* 3060 * ext4_xattr_hash_entry() 3061 * 3062 * Compute the hash of an extended attribute. 3063 */ 3064 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value, 3065 size_t value_count) 3066 { 3067 __u32 hash = 0; 3068 3069 while (name_len--) { 3070 hash = (hash << NAME_HASH_SHIFT) ^ 3071 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^ 3072 *name++; 3073 } 3074 while (value_count--) { 3075 hash = (hash << VALUE_HASH_SHIFT) ^ 3076 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^ 3077 le32_to_cpu(*value++); 3078 } 3079 return cpu_to_le32(hash); 3080 } 3081 3082 #undef NAME_HASH_SHIFT 3083 #undef VALUE_HASH_SHIFT 3084 3085 #define BLOCK_HASH_SHIFT 16 3086 3087 /* 3088 * ext4_xattr_rehash() 3089 * 3090 * Re-compute the extended attribute hash value after an entry has changed. 3091 */ 3092 static void ext4_xattr_rehash(struct ext4_xattr_header *header) 3093 { 3094 struct ext4_xattr_entry *here; 3095 __u32 hash = 0; 3096 3097 here = ENTRY(header+1); 3098 while (!IS_LAST_ENTRY(here)) { 3099 if (!here->e_hash) { 3100 /* Block is not shared if an entry's hash value == 0 */ 3101 hash = 0; 3102 break; 3103 } 3104 hash = (hash << BLOCK_HASH_SHIFT) ^ 3105 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^ 3106 le32_to_cpu(here->e_hash); 3107 here = EXT4_XATTR_NEXT(here); 3108 } 3109 header->h_hash = cpu_to_le32(hash); 3110 } 3111 3112 #undef BLOCK_HASH_SHIFT 3113 3114 #define HASH_BUCKET_BITS 10 3115 3116 struct mb_cache * 3117 ext4_xattr_create_cache(void) 3118 { 3119 return mb_cache_create(HASH_BUCKET_BITS); 3120 } 3121 3122 void ext4_xattr_destroy_cache(struct mb_cache *cache) 3123 { 3124 if (cache) 3125 mb_cache_destroy(cache); 3126 } 3127 3128