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