1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext2/xattr.c 4 * 5 * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de> 6 * 7 * Fix by Harrison Xing <harrison@mountainviewdata.com>. 8 * Extended attributes for symlinks and special files added per 9 * suggestion of Luka Renko <luka.renko@hermes.si>. 10 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>, 11 * Red Hat Inc. 12 * 13 */ 14 15 /* 16 * Extended attributes are stored on disk blocks allocated outside of 17 * any inode. The i_file_acl field is then made to point to this allocated 18 * block. If all extended attributes of an inode are identical, these 19 * inodes may share the same extended attribute block. Such situations 20 * are automatically detected by keeping a cache of recent attribute block 21 * numbers and hashes over the block's contents in memory. 22 * 23 * 24 * Extended attribute block layout: 25 * 26 * +------------------+ 27 * | header | 28 * | entry 1 | | 29 * | entry 2 | | growing downwards 30 * | entry 3 | v 31 * | four null bytes | 32 * | . . . | 33 * | value 1 | ^ 34 * | value 3 | | growing upwards 35 * | value 2 | | 36 * +------------------+ 37 * 38 * The block header is followed by multiple entry descriptors. These entry 39 * descriptors are variable in size, and aligned to EXT2_XATTR_PAD 40 * byte boundaries. The entry descriptors are sorted by attribute name, 41 * so that two extended attribute blocks can be compared efficiently. 42 * 43 * Attribute values are aligned to the end of the block, stored in 44 * no specific order. They are also padded to EXT2_XATTR_PAD byte 45 * boundaries. No additional gaps are left between them. 46 * 47 * Locking strategy 48 * ---------------- 49 * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem. 50 * EA blocks are only changed if they are exclusive to an inode, so 51 * holding xattr_sem also means that nothing but the EA block's reference 52 * count will change. Multiple writers to an EA block are synchronized 53 * by the bh lock. No more than a single bh lock is held at any time 54 * to avoid deadlocks. 55 */ 56 57 #include <linux/buffer_head.h> 58 #include <linux/init.h> 59 #include <linux/slab.h> 60 #include <linux/mbcache.h> 61 #include <linux/quotaops.h> 62 #include <linux/rwsem.h> 63 #include <linux/security.h> 64 #include "ext2.h" 65 #include "xattr.h" 66 #include "acl.h" 67 68 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data)) 69 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr)) 70 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1) 71 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0) 72 73 #ifdef EXT2_XATTR_DEBUG 74 # define ea_idebug(inode, f...) do { \ 75 printk(KERN_DEBUG "inode %s:%ld: ", \ 76 inode->i_sb->s_id, inode->i_ino); \ 77 printk(f); \ 78 printk("\n"); \ 79 } while (0) 80 # define ea_bdebug(bh, f...) do { \ 81 printk(KERN_DEBUG "block %pg:%lu: ", \ 82 bh->b_bdev, (unsigned long) bh->b_blocknr); \ 83 printk(f); \ 84 printk("\n"); \ 85 } while (0) 86 #else 87 # define ea_idebug(f...) 88 # define ea_bdebug(f...) 89 #endif 90 91 static int ext2_xattr_set2(struct inode *, struct buffer_head *, 92 struct ext2_xattr_header *); 93 94 static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *); 95 static struct buffer_head *ext2_xattr_cache_find(struct inode *, 96 struct ext2_xattr_header *); 97 static void ext2_xattr_rehash(struct ext2_xattr_header *, 98 struct ext2_xattr_entry *); 99 100 static const struct xattr_handler *ext2_xattr_handler_map[] = { 101 [EXT2_XATTR_INDEX_USER] = &ext2_xattr_user_handler, 102 #ifdef CONFIG_EXT2_FS_POSIX_ACL 103 [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler, 104 [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler, 105 #endif 106 [EXT2_XATTR_INDEX_TRUSTED] = &ext2_xattr_trusted_handler, 107 #ifdef CONFIG_EXT2_FS_SECURITY 108 [EXT2_XATTR_INDEX_SECURITY] = &ext2_xattr_security_handler, 109 #endif 110 }; 111 112 const struct xattr_handler *ext2_xattr_handlers[] = { 113 &ext2_xattr_user_handler, 114 &ext2_xattr_trusted_handler, 115 #ifdef CONFIG_EXT2_FS_POSIX_ACL 116 &posix_acl_access_xattr_handler, 117 &posix_acl_default_xattr_handler, 118 #endif 119 #ifdef CONFIG_EXT2_FS_SECURITY 120 &ext2_xattr_security_handler, 121 #endif 122 NULL 123 }; 124 125 #define EA_BLOCK_CACHE(inode) (EXT2_SB(inode->i_sb)->s_ea_block_cache) 126 127 static inline const struct xattr_handler * 128 ext2_xattr_handler(int name_index) 129 { 130 const struct xattr_handler *handler = NULL; 131 132 if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map)) 133 handler = ext2_xattr_handler_map[name_index]; 134 return handler; 135 } 136 137 static bool 138 ext2_xattr_header_valid(struct ext2_xattr_header *header) 139 { 140 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 141 header->h_blocks != cpu_to_le32(1)) 142 return false; 143 144 return true; 145 } 146 147 static bool 148 ext2_xattr_entry_valid(struct ext2_xattr_entry *entry, size_t end_offs) 149 { 150 size_t size; 151 152 if (entry->e_value_block != 0) 153 return false; 154 155 size = le32_to_cpu(entry->e_value_size); 156 if (size > end_offs || 157 le16_to_cpu(entry->e_value_offs) + size > end_offs) 158 return false; 159 160 return true; 161 } 162 163 /* 164 * ext2_xattr_get() 165 * 166 * Copy an extended attribute into the buffer 167 * provided, or compute the buffer size required. 168 * Buffer is NULL to compute the size of the buffer required. 169 * 170 * Returns a negative error number on failure, or the number of bytes 171 * used / required on success. 172 */ 173 int 174 ext2_xattr_get(struct inode *inode, int name_index, const char *name, 175 void *buffer, size_t buffer_size) 176 { 177 struct buffer_head *bh = NULL; 178 struct ext2_xattr_entry *entry; 179 size_t name_len, size; 180 char *end; 181 int error; 182 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 183 184 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld", 185 name_index, name, buffer, (long)buffer_size); 186 187 if (name == NULL) 188 return -EINVAL; 189 name_len = strlen(name); 190 if (name_len > 255) 191 return -ERANGE; 192 193 down_read(&EXT2_I(inode)->xattr_sem); 194 error = -ENODATA; 195 if (!EXT2_I(inode)->i_file_acl) 196 goto cleanup; 197 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 198 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 199 error = -EIO; 200 if (!bh) 201 goto cleanup; 202 ea_bdebug(bh, "b_count=%d, refcount=%d", 203 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 204 end = bh->b_data + bh->b_size; 205 if (!ext2_xattr_header_valid(HDR(bh))) { 206 bad_block: 207 ext2_error(inode->i_sb, "ext2_xattr_get", 208 "inode %ld: bad block %d", inode->i_ino, 209 EXT2_I(inode)->i_file_acl); 210 error = -EIO; 211 goto cleanup; 212 } 213 214 /* find named attribute */ 215 entry = FIRST_ENTRY(bh); 216 while (!IS_LAST_ENTRY(entry)) { 217 struct ext2_xattr_entry *next = 218 EXT2_XATTR_NEXT(entry); 219 if ((char *)next >= end) 220 goto bad_block; 221 if (!ext2_xattr_entry_valid(entry, inode->i_sb->s_blocksize)) 222 goto bad_block; 223 if (name_index == entry->e_name_index && 224 name_len == entry->e_name_len && 225 memcmp(name, entry->e_name, name_len) == 0) 226 goto found; 227 entry = next; 228 } 229 if (ext2_xattr_cache_insert(ea_block_cache, bh)) 230 ea_idebug(inode, "cache insert failed"); 231 error = -ENODATA; 232 goto cleanup; 233 found: 234 size = le32_to_cpu(entry->e_value_size); 235 if (ext2_xattr_cache_insert(ea_block_cache, bh)) 236 ea_idebug(inode, "cache insert failed"); 237 if (buffer) { 238 error = -ERANGE; 239 if (size > buffer_size) 240 goto cleanup; 241 /* return value of attribute */ 242 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs), 243 size); 244 } 245 error = size; 246 247 cleanup: 248 brelse(bh); 249 up_read(&EXT2_I(inode)->xattr_sem); 250 251 return error; 252 } 253 254 /* 255 * ext2_xattr_list() 256 * 257 * Copy a list of attribute names into the buffer 258 * provided, or compute the buffer size required. 259 * Buffer is NULL to compute the size of the buffer required. 260 * 261 * Returns a negative error number on failure, or the number of bytes 262 * used / required on success. 263 */ 264 static int 265 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size) 266 { 267 struct inode *inode = d_inode(dentry); 268 struct buffer_head *bh = NULL; 269 struct ext2_xattr_entry *entry; 270 char *end; 271 size_t rest = buffer_size; 272 int error; 273 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 274 275 ea_idebug(inode, "buffer=%p, buffer_size=%ld", 276 buffer, (long)buffer_size); 277 278 down_read(&EXT2_I(inode)->xattr_sem); 279 error = 0; 280 if (!EXT2_I(inode)->i_file_acl) 281 goto cleanup; 282 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 283 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 284 error = -EIO; 285 if (!bh) 286 goto cleanup; 287 ea_bdebug(bh, "b_count=%d, refcount=%d", 288 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 289 end = bh->b_data + bh->b_size; 290 if (!ext2_xattr_header_valid(HDR(bh))) { 291 bad_block: 292 ext2_error(inode->i_sb, "ext2_xattr_list", 293 "inode %ld: bad block %d", inode->i_ino, 294 EXT2_I(inode)->i_file_acl); 295 error = -EIO; 296 goto cleanup; 297 } 298 299 /* check the on-disk data structure */ 300 entry = FIRST_ENTRY(bh); 301 while (!IS_LAST_ENTRY(entry)) { 302 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry); 303 304 if ((char *)next >= end) 305 goto bad_block; 306 if (!ext2_xattr_entry_valid(entry, inode->i_sb->s_blocksize)) 307 goto bad_block; 308 entry = next; 309 } 310 if (ext2_xattr_cache_insert(ea_block_cache, bh)) 311 ea_idebug(inode, "cache insert failed"); 312 313 /* list the attribute names */ 314 for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry); 315 entry = EXT2_XATTR_NEXT(entry)) { 316 const struct xattr_handler *handler = 317 ext2_xattr_handler(entry->e_name_index); 318 319 if (handler && (!handler->list || handler->list(dentry))) { 320 const char *prefix = handler->prefix ?: handler->name; 321 size_t prefix_len = strlen(prefix); 322 size_t size = prefix_len + entry->e_name_len + 1; 323 324 if (buffer) { 325 if (size > rest) { 326 error = -ERANGE; 327 goto cleanup; 328 } 329 memcpy(buffer, prefix, prefix_len); 330 buffer += prefix_len; 331 memcpy(buffer, entry->e_name, entry->e_name_len); 332 buffer += entry->e_name_len; 333 *buffer++ = 0; 334 } 335 rest -= size; 336 } 337 } 338 error = buffer_size - rest; /* total size */ 339 340 cleanup: 341 brelse(bh); 342 up_read(&EXT2_I(inode)->xattr_sem); 343 344 return error; 345 } 346 347 /* 348 * Inode operation listxattr() 349 * 350 * d_inode(dentry)->i_mutex: don't care 351 */ 352 ssize_t 353 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size) 354 { 355 return ext2_xattr_list(dentry, buffer, size); 356 } 357 358 /* 359 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is 360 * not set, set it. 361 */ 362 static void ext2_xattr_update_super_block(struct super_block *sb) 363 { 364 if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR)) 365 return; 366 367 spin_lock(&EXT2_SB(sb)->s_lock); 368 ext2_update_dynamic_rev(sb); 369 EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR); 370 spin_unlock(&EXT2_SB(sb)->s_lock); 371 mark_buffer_dirty(EXT2_SB(sb)->s_sbh); 372 } 373 374 /* 375 * ext2_xattr_set() 376 * 377 * Create, replace or remove an extended attribute for this inode. Value 378 * is NULL to remove an existing extended attribute, and non-NULL to 379 * either replace an existing extended attribute, or create a new extended 380 * attribute. The flags XATTR_REPLACE and XATTR_CREATE 381 * specify that an extended attribute must exist and must not exist 382 * previous to the call, respectively. 383 * 384 * Returns 0, or a negative error number on failure. 385 */ 386 int 387 ext2_xattr_set(struct inode *inode, int name_index, const char *name, 388 const void *value, size_t value_len, int flags) 389 { 390 struct super_block *sb = inode->i_sb; 391 struct buffer_head *bh = NULL; 392 struct ext2_xattr_header *header = NULL; 393 struct ext2_xattr_entry *here, *last; 394 size_t name_len, free, min_offs = sb->s_blocksize; 395 int not_found = 1, error; 396 char *end; 397 398 /* 399 * header -- Points either into bh, or to a temporarily 400 * allocated buffer. 401 * here -- The named entry found, or the place for inserting, within 402 * the block pointed to by header. 403 * last -- Points right after the last named entry within the block 404 * pointed to by header. 405 * min_offs -- The offset of the first value (values are aligned 406 * towards the end of the block). 407 * end -- Points right after the block pointed to by header. 408 */ 409 410 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld", 411 name_index, name, value, (long)value_len); 412 413 if (value == NULL) 414 value_len = 0; 415 if (name == NULL) 416 return -EINVAL; 417 name_len = strlen(name); 418 if (name_len > 255 || value_len > sb->s_blocksize) 419 return -ERANGE; 420 down_write(&EXT2_I(inode)->xattr_sem); 421 if (EXT2_I(inode)->i_file_acl) { 422 /* The inode already has an extended attribute block. */ 423 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl); 424 error = -EIO; 425 if (!bh) 426 goto cleanup; 427 ea_bdebug(bh, "b_count=%d, refcount=%d", 428 atomic_read(&(bh->b_count)), 429 le32_to_cpu(HDR(bh)->h_refcount)); 430 header = HDR(bh); 431 end = bh->b_data + bh->b_size; 432 if (!ext2_xattr_header_valid(header)) { 433 bad_block: 434 ext2_error(sb, "ext2_xattr_set", 435 "inode %ld: bad block %d", inode->i_ino, 436 EXT2_I(inode)->i_file_acl); 437 error = -EIO; 438 goto cleanup; 439 } 440 /* 441 * Find the named attribute. If not found, 'here' will point 442 * to entry where the new attribute should be inserted to 443 * maintain sorting. 444 */ 445 last = FIRST_ENTRY(bh); 446 while (!IS_LAST_ENTRY(last)) { 447 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last); 448 if ((char *)next >= end) 449 goto bad_block; 450 if (!ext2_xattr_entry_valid(last, sb->s_blocksize)) 451 goto bad_block; 452 if (last->e_value_size) { 453 size_t offs = le16_to_cpu(last->e_value_offs); 454 if (offs < min_offs) 455 min_offs = offs; 456 } 457 if (not_found > 0) { 458 not_found = name_index - last->e_name_index; 459 if (!not_found) 460 not_found = name_len - last->e_name_len; 461 if (!not_found) { 462 not_found = memcmp(name, last->e_name, 463 name_len); 464 } 465 if (not_found <= 0) 466 here = last; 467 } 468 last = next; 469 } 470 if (not_found > 0) 471 here = last; 472 473 /* Check whether we have enough space left. */ 474 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32); 475 } else { 476 /* We will use a new extended attribute block. */ 477 free = sb->s_blocksize - 478 sizeof(struct ext2_xattr_header) - sizeof(__u32); 479 here = last = NULL; /* avoid gcc uninitialized warning. */ 480 } 481 482 if (not_found) { 483 /* Request to remove a nonexistent attribute? */ 484 error = -ENODATA; 485 if (flags & XATTR_REPLACE) 486 goto cleanup; 487 error = 0; 488 if (value == NULL) 489 goto cleanup; 490 } else { 491 /* Request to create an existing attribute? */ 492 error = -EEXIST; 493 if (flags & XATTR_CREATE) 494 goto cleanup; 495 free += EXT2_XATTR_SIZE(le32_to_cpu(here->e_value_size)); 496 free += EXT2_XATTR_LEN(name_len); 497 } 498 error = -ENOSPC; 499 if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len)) 500 goto cleanup; 501 502 /* Here we know that we can set the new attribute. */ 503 504 if (header) { 505 /* assert(header == HDR(bh)); */ 506 lock_buffer(bh); 507 if (header->h_refcount == cpu_to_le32(1)) { 508 __u32 hash = le32_to_cpu(header->h_hash); 509 510 ea_bdebug(bh, "modifying in-place"); 511 /* 512 * This must happen under buffer lock for 513 * ext2_xattr_set2() to reliably detect modified block 514 */ 515 mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash, 516 bh->b_blocknr); 517 518 /* keep the buffer locked while modifying it. */ 519 } else { 520 int offset; 521 522 unlock_buffer(bh); 523 ea_bdebug(bh, "cloning"); 524 header = kmalloc(bh->b_size, GFP_KERNEL); 525 error = -ENOMEM; 526 if (header == NULL) 527 goto cleanup; 528 memcpy(header, HDR(bh), bh->b_size); 529 header->h_refcount = cpu_to_le32(1); 530 531 offset = (char *)here - bh->b_data; 532 here = ENTRY((char *)header + offset); 533 offset = (char *)last - bh->b_data; 534 last = ENTRY((char *)header + offset); 535 } 536 } else { 537 /* Allocate a buffer where we construct the new block. */ 538 header = kzalloc(sb->s_blocksize, GFP_KERNEL); 539 error = -ENOMEM; 540 if (header == NULL) 541 goto cleanup; 542 end = (char *)header + sb->s_blocksize; 543 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC); 544 header->h_blocks = header->h_refcount = cpu_to_le32(1); 545 last = here = ENTRY(header+1); 546 } 547 548 /* Iff we are modifying the block in-place, bh is locked here. */ 549 550 if (not_found) { 551 /* Insert the new name. */ 552 size_t size = EXT2_XATTR_LEN(name_len); 553 size_t rest = (char *)last - (char *)here; 554 memmove((char *)here + size, here, rest); 555 memset(here, 0, size); 556 here->e_name_index = name_index; 557 here->e_name_len = name_len; 558 memcpy(here->e_name, name, name_len); 559 } else { 560 if (here->e_value_size) { 561 char *first_val = (char *)header + min_offs; 562 size_t offs = le16_to_cpu(here->e_value_offs); 563 char *val = (char *)header + offs; 564 size_t size = EXT2_XATTR_SIZE( 565 le32_to_cpu(here->e_value_size)); 566 567 if (size == EXT2_XATTR_SIZE(value_len)) { 568 /* The old and the new value have the same 569 size. Just replace. */ 570 here->e_value_size = cpu_to_le32(value_len); 571 memset(val + size - EXT2_XATTR_PAD, 0, 572 EXT2_XATTR_PAD); /* Clear pad bytes. */ 573 memcpy(val, value, value_len); 574 goto skip_replace; 575 } 576 577 /* Remove the old value. */ 578 memmove(first_val + size, first_val, val - first_val); 579 memset(first_val, 0, size); 580 here->e_value_offs = 0; 581 min_offs += size; 582 583 /* Adjust all value offsets. */ 584 last = ENTRY(header+1); 585 while (!IS_LAST_ENTRY(last)) { 586 size_t o = le16_to_cpu(last->e_value_offs); 587 if (o < offs) 588 last->e_value_offs = 589 cpu_to_le16(o + size); 590 last = EXT2_XATTR_NEXT(last); 591 } 592 } 593 if (value == NULL) { 594 /* Remove the old name. */ 595 size_t size = EXT2_XATTR_LEN(name_len); 596 last = ENTRY((char *)last - size); 597 memmove(here, (char*)here + size, 598 (char*)last - (char*)here); 599 memset(last, 0, size); 600 } 601 } 602 603 if (value != NULL) { 604 /* Insert the new value. */ 605 here->e_value_size = cpu_to_le32(value_len); 606 if (value_len) { 607 size_t size = EXT2_XATTR_SIZE(value_len); 608 char *val = (char *)header + min_offs - size; 609 here->e_value_offs = 610 cpu_to_le16((char *)val - (char *)header); 611 memset(val + size - EXT2_XATTR_PAD, 0, 612 EXT2_XATTR_PAD); /* Clear the pad bytes. */ 613 memcpy(val, value, value_len); 614 } 615 } 616 617 skip_replace: 618 if (IS_LAST_ENTRY(ENTRY(header+1))) { 619 /* This block is now empty. */ 620 if (bh && header == HDR(bh)) 621 unlock_buffer(bh); /* we were modifying in-place. */ 622 error = ext2_xattr_set2(inode, bh, NULL); 623 } else { 624 ext2_xattr_rehash(header, here); 625 if (bh && header == HDR(bh)) 626 unlock_buffer(bh); /* we were modifying in-place. */ 627 error = ext2_xattr_set2(inode, bh, header); 628 } 629 630 cleanup: 631 if (!(bh && header == HDR(bh))) 632 kfree(header); 633 brelse(bh); 634 up_write(&EXT2_I(inode)->xattr_sem); 635 636 return error; 637 } 638 639 /* 640 * Second half of ext2_xattr_set(): Update the file system. 641 */ 642 static int 643 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh, 644 struct ext2_xattr_header *header) 645 { 646 struct super_block *sb = inode->i_sb; 647 struct buffer_head *new_bh = NULL; 648 int error; 649 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 650 651 if (header) { 652 new_bh = ext2_xattr_cache_find(inode, header); 653 if (new_bh) { 654 /* We found an identical block in the cache. */ 655 if (new_bh == old_bh) { 656 ea_bdebug(new_bh, "keeping this block"); 657 } else { 658 /* The old block is released after updating 659 the inode. */ 660 ea_bdebug(new_bh, "reusing block"); 661 662 error = dquot_alloc_block(inode, 1); 663 if (error) { 664 unlock_buffer(new_bh); 665 goto cleanup; 666 } 667 le32_add_cpu(&HDR(new_bh)->h_refcount, 1); 668 ea_bdebug(new_bh, "refcount now=%d", 669 le32_to_cpu(HDR(new_bh)->h_refcount)); 670 } 671 unlock_buffer(new_bh); 672 } else if (old_bh && header == HDR(old_bh)) { 673 /* Keep this block. No need to lock the block as we 674 don't need to change the reference count. */ 675 new_bh = old_bh; 676 get_bh(new_bh); 677 ext2_xattr_cache_insert(ea_block_cache, new_bh); 678 } else { 679 /* We need to allocate a new block */ 680 ext2_fsblk_t goal = ext2_group_first_block_no(sb, 681 EXT2_I(inode)->i_block_group); 682 int block = ext2_new_block(inode, goal, &error); 683 if (error) 684 goto cleanup; 685 ea_idebug(inode, "creating block %d", block); 686 687 new_bh = sb_getblk(sb, block); 688 if (unlikely(!new_bh)) { 689 ext2_free_blocks(inode, block, 1); 690 mark_inode_dirty(inode); 691 error = -ENOMEM; 692 goto cleanup; 693 } 694 lock_buffer(new_bh); 695 memcpy(new_bh->b_data, header, new_bh->b_size); 696 set_buffer_uptodate(new_bh); 697 unlock_buffer(new_bh); 698 ext2_xattr_cache_insert(ea_block_cache, new_bh); 699 700 ext2_xattr_update_super_block(sb); 701 } 702 mark_buffer_dirty(new_bh); 703 if (IS_SYNC(inode)) { 704 sync_dirty_buffer(new_bh); 705 error = -EIO; 706 if (buffer_req(new_bh) && !buffer_uptodate(new_bh)) 707 goto cleanup; 708 } 709 } 710 711 /* Update the inode. */ 712 EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0; 713 inode->i_ctime = current_time(inode); 714 if (IS_SYNC(inode)) { 715 error = sync_inode_metadata(inode, 1); 716 /* In case sync failed due to ENOSPC the inode was actually 717 * written (only some dirty data were not) so we just proceed 718 * as if nothing happened and cleanup the unused block */ 719 if (error && error != -ENOSPC) { 720 if (new_bh && new_bh != old_bh) { 721 dquot_free_block_nodirty(inode, 1); 722 mark_inode_dirty(inode); 723 } 724 goto cleanup; 725 } 726 } else 727 mark_inode_dirty(inode); 728 729 error = 0; 730 if (old_bh && old_bh != new_bh) { 731 /* 732 * If there was an old block and we are no longer using it, 733 * release the old block. 734 */ 735 lock_buffer(old_bh); 736 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) { 737 __u32 hash = le32_to_cpu(HDR(old_bh)->h_hash); 738 739 /* 740 * This must happen under buffer lock for 741 * ext2_xattr_set2() to reliably detect freed block 742 */ 743 mb_cache_entry_delete(ea_block_cache, hash, 744 old_bh->b_blocknr); 745 /* Free the old block. */ 746 ea_bdebug(old_bh, "freeing"); 747 ext2_free_blocks(inode, old_bh->b_blocknr, 1); 748 mark_inode_dirty(inode); 749 /* We let our caller release old_bh, so we 750 * need to duplicate the buffer before. */ 751 get_bh(old_bh); 752 bforget(old_bh); 753 } else { 754 /* Decrement the refcount only. */ 755 le32_add_cpu(&HDR(old_bh)->h_refcount, -1); 756 dquot_free_block_nodirty(inode, 1); 757 mark_inode_dirty(inode); 758 mark_buffer_dirty(old_bh); 759 ea_bdebug(old_bh, "refcount now=%d", 760 le32_to_cpu(HDR(old_bh)->h_refcount)); 761 } 762 unlock_buffer(old_bh); 763 } 764 765 cleanup: 766 brelse(new_bh); 767 768 return error; 769 } 770 771 /* 772 * ext2_xattr_delete_inode() 773 * 774 * Free extended attribute resources associated with this inode. This 775 * is called immediately before an inode is freed. 776 */ 777 void 778 ext2_xattr_delete_inode(struct inode *inode) 779 { 780 struct buffer_head *bh = NULL; 781 struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb); 782 783 down_write(&EXT2_I(inode)->xattr_sem); 784 if (!EXT2_I(inode)->i_file_acl) 785 goto cleanup; 786 787 if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 0)) { 788 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 789 "inode %ld: xattr block %d is out of data blocks range", 790 inode->i_ino, EXT2_I(inode)->i_file_acl); 791 goto cleanup; 792 } 793 794 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 795 if (!bh) { 796 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 797 "inode %ld: block %d read error", inode->i_ino, 798 EXT2_I(inode)->i_file_acl); 799 goto cleanup; 800 } 801 ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count))); 802 if (!ext2_xattr_header_valid(HDR(bh))) { 803 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 804 "inode %ld: bad block %d", inode->i_ino, 805 EXT2_I(inode)->i_file_acl); 806 goto cleanup; 807 } 808 lock_buffer(bh); 809 if (HDR(bh)->h_refcount == cpu_to_le32(1)) { 810 __u32 hash = le32_to_cpu(HDR(bh)->h_hash); 811 812 /* 813 * This must happen under buffer lock for ext2_xattr_set2() to 814 * reliably detect freed block 815 */ 816 mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash, 817 bh->b_blocknr); 818 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1); 819 get_bh(bh); 820 bforget(bh); 821 unlock_buffer(bh); 822 } else { 823 le32_add_cpu(&HDR(bh)->h_refcount, -1); 824 ea_bdebug(bh, "refcount now=%d", 825 le32_to_cpu(HDR(bh)->h_refcount)); 826 unlock_buffer(bh); 827 mark_buffer_dirty(bh); 828 if (IS_SYNC(inode)) 829 sync_dirty_buffer(bh); 830 dquot_free_block_nodirty(inode, 1); 831 } 832 EXT2_I(inode)->i_file_acl = 0; 833 834 cleanup: 835 brelse(bh); 836 up_write(&EXT2_I(inode)->xattr_sem); 837 } 838 839 /* 840 * ext2_xattr_cache_insert() 841 * 842 * Create a new entry in the extended attribute cache, and insert 843 * it unless such an entry is already in the cache. 844 * 845 * Returns 0, or a negative error number on failure. 846 */ 847 static int 848 ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh) 849 { 850 __u32 hash = le32_to_cpu(HDR(bh)->h_hash); 851 int error; 852 853 error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr, 854 true); 855 if (error) { 856 if (error == -EBUSY) { 857 ea_bdebug(bh, "already in cache (%d cache entries)", 858 atomic_read(&ext2_xattr_cache->c_entry_count)); 859 error = 0; 860 } 861 } else 862 ea_bdebug(bh, "inserting [%x]", (int)hash); 863 return error; 864 } 865 866 /* 867 * ext2_xattr_cmp() 868 * 869 * Compare two extended attribute blocks for equality. 870 * 871 * Returns 0 if the blocks are equal, 1 if they differ, and 872 * a negative error number on errors. 873 */ 874 static int 875 ext2_xattr_cmp(struct ext2_xattr_header *header1, 876 struct ext2_xattr_header *header2) 877 { 878 struct ext2_xattr_entry *entry1, *entry2; 879 880 entry1 = ENTRY(header1+1); 881 entry2 = ENTRY(header2+1); 882 while (!IS_LAST_ENTRY(entry1)) { 883 if (IS_LAST_ENTRY(entry2)) 884 return 1; 885 if (entry1->e_hash != entry2->e_hash || 886 entry1->e_name_index != entry2->e_name_index || 887 entry1->e_name_len != entry2->e_name_len || 888 entry1->e_value_size != entry2->e_value_size || 889 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len)) 890 return 1; 891 if (entry1->e_value_block != 0 || entry2->e_value_block != 0) 892 return -EIO; 893 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs), 894 (char *)header2 + le16_to_cpu(entry2->e_value_offs), 895 le32_to_cpu(entry1->e_value_size))) 896 return 1; 897 898 entry1 = EXT2_XATTR_NEXT(entry1); 899 entry2 = EXT2_XATTR_NEXT(entry2); 900 } 901 if (!IS_LAST_ENTRY(entry2)) 902 return 1; 903 return 0; 904 } 905 906 /* 907 * ext2_xattr_cache_find() 908 * 909 * Find an identical extended attribute block. 910 * 911 * Returns a locked buffer head to the block found, or NULL if such 912 * a block was not found or an error occurred. 913 */ 914 static struct buffer_head * 915 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header) 916 { 917 __u32 hash = le32_to_cpu(header->h_hash); 918 struct mb_cache_entry *ce; 919 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode); 920 921 if (!header->h_hash) 922 return NULL; /* never share */ 923 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash); 924 again: 925 ce = mb_cache_entry_find_first(ea_block_cache, hash); 926 while (ce) { 927 struct buffer_head *bh; 928 929 bh = sb_bread(inode->i_sb, ce->e_value); 930 if (!bh) { 931 ext2_error(inode->i_sb, "ext2_xattr_cache_find", 932 "inode %ld: block %ld read error", 933 inode->i_ino, (unsigned long) ce->e_value); 934 } else { 935 lock_buffer(bh); 936 /* 937 * We have to be careful about races with freeing or 938 * rehashing of xattr block. Once we hold buffer lock 939 * xattr block's state is stable so we can check 940 * whether the block got freed / rehashed or not. 941 * Since we unhash mbcache entry under buffer lock when 942 * freeing / rehashing xattr block, checking whether 943 * entry is still hashed is reliable. 944 */ 945 if (hlist_bl_unhashed(&ce->e_hash_list)) { 946 mb_cache_entry_put(ea_block_cache, ce); 947 unlock_buffer(bh); 948 brelse(bh); 949 goto again; 950 } else if (le32_to_cpu(HDR(bh)->h_refcount) > 951 EXT2_XATTR_REFCOUNT_MAX) { 952 ea_idebug(inode, "block %ld refcount %d>%d", 953 (unsigned long) ce->e_value, 954 le32_to_cpu(HDR(bh)->h_refcount), 955 EXT2_XATTR_REFCOUNT_MAX); 956 } else if (!ext2_xattr_cmp(header, HDR(bh))) { 957 ea_bdebug(bh, "b_count=%d", 958 atomic_read(&(bh->b_count))); 959 mb_cache_entry_touch(ea_block_cache, ce); 960 mb_cache_entry_put(ea_block_cache, ce); 961 return bh; 962 } 963 unlock_buffer(bh); 964 brelse(bh); 965 } 966 ce = mb_cache_entry_find_next(ea_block_cache, ce); 967 } 968 return NULL; 969 } 970 971 #define NAME_HASH_SHIFT 5 972 #define VALUE_HASH_SHIFT 16 973 974 /* 975 * ext2_xattr_hash_entry() 976 * 977 * Compute the hash of an extended attribute. 978 */ 979 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header, 980 struct ext2_xattr_entry *entry) 981 { 982 __u32 hash = 0; 983 char *name = entry->e_name; 984 int n; 985 986 for (n=0; n < entry->e_name_len; n++) { 987 hash = (hash << NAME_HASH_SHIFT) ^ 988 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^ 989 *name++; 990 } 991 992 if (entry->e_value_block == 0 && entry->e_value_size != 0) { 993 __le32 *value = (__le32 *)((char *)header + 994 le16_to_cpu(entry->e_value_offs)); 995 for (n = (le32_to_cpu(entry->e_value_size) + 996 EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) { 997 hash = (hash << VALUE_HASH_SHIFT) ^ 998 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^ 999 le32_to_cpu(*value++); 1000 } 1001 } 1002 entry->e_hash = cpu_to_le32(hash); 1003 } 1004 1005 #undef NAME_HASH_SHIFT 1006 #undef VALUE_HASH_SHIFT 1007 1008 #define BLOCK_HASH_SHIFT 16 1009 1010 /* 1011 * ext2_xattr_rehash() 1012 * 1013 * Re-compute the extended attribute hash value after an entry has changed. 1014 */ 1015 static void ext2_xattr_rehash(struct ext2_xattr_header *header, 1016 struct ext2_xattr_entry *entry) 1017 { 1018 struct ext2_xattr_entry *here; 1019 __u32 hash = 0; 1020 1021 ext2_xattr_hash_entry(header, entry); 1022 here = ENTRY(header+1); 1023 while (!IS_LAST_ENTRY(here)) { 1024 if (!here->e_hash) { 1025 /* Block is not shared if an entry's hash value == 0 */ 1026 hash = 0; 1027 break; 1028 } 1029 hash = (hash << BLOCK_HASH_SHIFT) ^ 1030 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^ 1031 le32_to_cpu(here->e_hash); 1032 here = EXT2_XATTR_NEXT(here); 1033 } 1034 header->h_hash = cpu_to_le32(hash); 1035 } 1036 1037 #undef BLOCK_HASH_SHIFT 1038 1039 #define HASH_BUCKET_BITS 10 1040 1041 struct mb_cache *ext2_xattr_create_cache(void) 1042 { 1043 return mb_cache_create(HASH_BUCKET_BITS); 1044 } 1045 1046 void ext2_xattr_destroy_cache(struct mb_cache *cache) 1047 { 1048 if (cache) 1049 mb_cache_destroy(cache); 1050 } 1051