1 /* 2 * linux/fs/ext2/xattr.c 3 * 4 * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de> 5 * 6 * Fix by Harrison Xing <harrison@mountainviewdata.com>. 7 * Extended attributes for symlinks and special files added per 8 * suggestion of Luka Renko <luka.renko@hermes.si>. 9 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>, 10 * Red Hat Inc. 11 * 12 */ 13 14 /* 15 * Extended attributes are stored on disk blocks allocated outside of 16 * any inode. The i_file_acl field is then made to point to this allocated 17 * block. If all extended attributes of an inode are identical, these 18 * inodes may share the same extended attribute block. Such situations 19 * are automatically detected by keeping a cache of recent attribute block 20 * numbers and hashes over the block's contents in memory. 21 * 22 * 23 * Extended attribute block layout: 24 * 25 * +------------------+ 26 * | header | 27 * | entry 1 | | 28 * | entry 2 | | growing downwards 29 * | entry 3 | v 30 * | four null bytes | 31 * | . . . | 32 * | value 1 | ^ 33 * | value 3 | | growing upwards 34 * | value 2 | | 35 * +------------------+ 36 * 37 * The block header is followed by multiple entry descriptors. These entry 38 * descriptors are variable in size, and aligned to EXT2_XATTR_PAD 39 * byte boundaries. The entry descriptors are sorted by attribute name, 40 * so that two extended attribute blocks can be compared efficiently. 41 * 42 * Attribute values are aligned to the end of the block, stored in 43 * no specific order. They are also padded to EXT2_XATTR_PAD byte 44 * boundaries. No additional gaps are left between them. 45 * 46 * Locking strategy 47 * ---------------- 48 * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem. 49 * EA blocks are only changed if they are exclusive to an inode, so 50 * holding xattr_sem also means that nothing but the EA block's reference 51 * count will change. Multiple writers to an EA block are synchronized 52 * by the bh lock. No more than a single bh lock is held at any time 53 * to avoid deadlocks. 54 */ 55 56 #include <linux/buffer_head.h> 57 #include <linux/init.h> 58 #include <linux/slab.h> 59 #include <linux/mbcache.h> 60 #include <linux/quotaops.h> 61 #include <linux/rwsem.h> 62 #include <linux/security.h> 63 #include "ext2.h" 64 #include "xattr.h" 65 #include "acl.h" 66 67 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data)) 68 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr)) 69 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1) 70 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0) 71 72 #ifdef EXT2_XATTR_DEBUG 73 # define ea_idebug(inode, f...) do { \ 74 printk(KERN_DEBUG "inode %s:%ld: ", \ 75 inode->i_sb->s_id, inode->i_ino); \ 76 printk(f); \ 77 printk("\n"); \ 78 } while (0) 79 # define ea_bdebug(bh, f...) do { \ 80 printk(KERN_DEBUG "block %pg:%lu: ", \ 81 bh->b_bdev, (unsigned long) bh->b_blocknr); \ 82 printk(f); \ 83 printk("\n"); \ 84 } while (0) 85 #else 86 # define ea_idebug(f...) 87 # define ea_bdebug(f...) 88 #endif 89 90 static int ext2_xattr_set2(struct inode *, struct buffer_head *, 91 struct ext2_xattr_header *); 92 93 static int ext2_xattr_cache_insert(struct buffer_head *); 94 static struct buffer_head *ext2_xattr_cache_find(struct inode *, 95 struct ext2_xattr_header *); 96 static void ext2_xattr_rehash(struct ext2_xattr_header *, 97 struct ext2_xattr_entry *); 98 99 static struct mb_cache *ext2_xattr_cache; 100 101 static const struct xattr_handler *ext2_xattr_handler_map[] = { 102 [EXT2_XATTR_INDEX_USER] = &ext2_xattr_user_handler, 103 #ifdef CONFIG_EXT2_FS_POSIX_ACL 104 [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler, 105 [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler, 106 #endif 107 [EXT2_XATTR_INDEX_TRUSTED] = &ext2_xattr_trusted_handler, 108 #ifdef CONFIG_EXT2_FS_SECURITY 109 [EXT2_XATTR_INDEX_SECURITY] = &ext2_xattr_security_handler, 110 #endif 111 }; 112 113 const struct xattr_handler *ext2_xattr_handlers[] = { 114 &ext2_xattr_user_handler, 115 &ext2_xattr_trusted_handler, 116 #ifdef CONFIG_EXT2_FS_POSIX_ACL 117 &posix_acl_access_xattr_handler, 118 &posix_acl_default_xattr_handler, 119 #endif 120 #ifdef CONFIG_EXT2_FS_SECURITY 121 &ext2_xattr_security_handler, 122 #endif 123 NULL 124 }; 125 126 static inline const struct xattr_handler * 127 ext2_xattr_handler(int name_index) 128 { 129 const struct xattr_handler *handler = NULL; 130 131 if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map)) 132 handler = ext2_xattr_handler_map[name_index]; 133 return handler; 134 } 135 136 /* 137 * ext2_xattr_get() 138 * 139 * Copy an extended attribute into the buffer 140 * provided, or compute the buffer size required. 141 * Buffer is NULL to compute the size of the buffer required. 142 * 143 * Returns a negative error number on failure, or the number of bytes 144 * used / required on success. 145 */ 146 int 147 ext2_xattr_get(struct inode *inode, int name_index, const char *name, 148 void *buffer, size_t buffer_size) 149 { 150 struct buffer_head *bh = NULL; 151 struct ext2_xattr_entry *entry; 152 size_t name_len, size; 153 char *end; 154 int error; 155 156 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld", 157 name_index, name, buffer, (long)buffer_size); 158 159 if (name == NULL) 160 return -EINVAL; 161 name_len = strlen(name); 162 if (name_len > 255) 163 return -ERANGE; 164 165 down_read(&EXT2_I(inode)->xattr_sem); 166 error = -ENODATA; 167 if (!EXT2_I(inode)->i_file_acl) 168 goto cleanup; 169 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 170 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 171 error = -EIO; 172 if (!bh) 173 goto cleanup; 174 ea_bdebug(bh, "b_count=%d, refcount=%d", 175 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 176 end = bh->b_data + bh->b_size; 177 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 178 HDR(bh)->h_blocks != cpu_to_le32(1)) { 179 bad_block: ext2_error(inode->i_sb, "ext2_xattr_get", 180 "inode %ld: bad block %d", inode->i_ino, 181 EXT2_I(inode)->i_file_acl); 182 error = -EIO; 183 goto cleanup; 184 } 185 186 /* find named attribute */ 187 entry = FIRST_ENTRY(bh); 188 while (!IS_LAST_ENTRY(entry)) { 189 struct ext2_xattr_entry *next = 190 EXT2_XATTR_NEXT(entry); 191 if ((char *)next >= end) 192 goto bad_block; 193 if (name_index == entry->e_name_index && 194 name_len == entry->e_name_len && 195 memcmp(name, entry->e_name, name_len) == 0) 196 goto found; 197 entry = next; 198 } 199 if (ext2_xattr_cache_insert(bh)) 200 ea_idebug(inode, "cache insert failed"); 201 error = -ENODATA; 202 goto cleanup; 203 found: 204 /* check the buffer size */ 205 if (entry->e_value_block != 0) 206 goto bad_block; 207 size = le32_to_cpu(entry->e_value_size); 208 if (size > inode->i_sb->s_blocksize || 209 le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize) 210 goto bad_block; 211 212 if (ext2_xattr_cache_insert(bh)) 213 ea_idebug(inode, "cache insert failed"); 214 if (buffer) { 215 error = -ERANGE; 216 if (size > buffer_size) 217 goto cleanup; 218 /* return value of attribute */ 219 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs), 220 size); 221 } 222 error = size; 223 224 cleanup: 225 brelse(bh); 226 up_read(&EXT2_I(inode)->xattr_sem); 227 228 return error; 229 } 230 231 /* 232 * ext2_xattr_list() 233 * 234 * Copy a list of attribute names into the buffer 235 * provided, or compute the buffer size required. 236 * Buffer is NULL to compute the size of the buffer required. 237 * 238 * Returns a negative error number on failure, or the number of bytes 239 * used / required on success. 240 */ 241 static int 242 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size) 243 { 244 struct inode *inode = d_inode(dentry); 245 struct buffer_head *bh = NULL; 246 struct ext2_xattr_entry *entry; 247 char *end; 248 size_t rest = buffer_size; 249 int error; 250 251 ea_idebug(inode, "buffer=%p, buffer_size=%ld", 252 buffer, (long)buffer_size); 253 254 down_read(&EXT2_I(inode)->xattr_sem); 255 error = 0; 256 if (!EXT2_I(inode)->i_file_acl) 257 goto cleanup; 258 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 259 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 260 error = -EIO; 261 if (!bh) 262 goto cleanup; 263 ea_bdebug(bh, "b_count=%d, refcount=%d", 264 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 265 end = bh->b_data + bh->b_size; 266 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 267 HDR(bh)->h_blocks != cpu_to_le32(1)) { 268 bad_block: ext2_error(inode->i_sb, "ext2_xattr_list", 269 "inode %ld: bad block %d", inode->i_ino, 270 EXT2_I(inode)->i_file_acl); 271 error = -EIO; 272 goto cleanup; 273 } 274 275 /* check the on-disk data structure */ 276 entry = FIRST_ENTRY(bh); 277 while (!IS_LAST_ENTRY(entry)) { 278 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry); 279 280 if ((char *)next >= end) 281 goto bad_block; 282 entry = next; 283 } 284 if (ext2_xattr_cache_insert(bh)) 285 ea_idebug(inode, "cache insert failed"); 286 287 /* list the attribute names */ 288 for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry); 289 entry = EXT2_XATTR_NEXT(entry)) { 290 const struct xattr_handler *handler = 291 ext2_xattr_handler(entry->e_name_index); 292 293 if (handler) { 294 size_t size = handler->list(handler, dentry, buffer, 295 rest, entry->e_name, 296 entry->e_name_len); 297 if (buffer) { 298 if (size > rest) { 299 error = -ERANGE; 300 goto cleanup; 301 } 302 buffer += size; 303 } 304 rest -= size; 305 } 306 } 307 error = buffer_size - rest; /* total size */ 308 309 cleanup: 310 brelse(bh); 311 up_read(&EXT2_I(inode)->xattr_sem); 312 313 return error; 314 } 315 316 /* 317 * Inode operation listxattr() 318 * 319 * d_inode(dentry)->i_mutex: don't care 320 */ 321 ssize_t 322 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size) 323 { 324 return ext2_xattr_list(dentry, buffer, size); 325 } 326 327 /* 328 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is 329 * not set, set it. 330 */ 331 static void ext2_xattr_update_super_block(struct super_block *sb) 332 { 333 if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR)) 334 return; 335 336 spin_lock(&EXT2_SB(sb)->s_lock); 337 EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR); 338 spin_unlock(&EXT2_SB(sb)->s_lock); 339 mark_buffer_dirty(EXT2_SB(sb)->s_sbh); 340 } 341 342 /* 343 * ext2_xattr_set() 344 * 345 * Create, replace or remove an extended attribute for this inode. Value 346 * is NULL to remove an existing extended attribute, and non-NULL to 347 * either replace an existing extended attribute, or create a new extended 348 * attribute. The flags XATTR_REPLACE and XATTR_CREATE 349 * specify that an extended attribute must exist and must not exist 350 * previous to the call, respectively. 351 * 352 * Returns 0, or a negative error number on failure. 353 */ 354 int 355 ext2_xattr_set(struct inode *inode, int name_index, const char *name, 356 const void *value, size_t value_len, int flags) 357 { 358 struct super_block *sb = inode->i_sb; 359 struct buffer_head *bh = NULL; 360 struct ext2_xattr_header *header = NULL; 361 struct ext2_xattr_entry *here, *last; 362 size_t name_len, free, min_offs = sb->s_blocksize; 363 int not_found = 1, error; 364 char *end; 365 366 /* 367 * header -- Points either into bh, or to a temporarily 368 * allocated buffer. 369 * here -- The named entry found, or the place for inserting, within 370 * the block pointed to by header. 371 * last -- Points right after the last named entry within the block 372 * pointed to by header. 373 * min_offs -- The offset of the first value (values are aligned 374 * towards the end of the block). 375 * end -- Points right after the block pointed to by header. 376 */ 377 378 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld", 379 name_index, name, value, (long)value_len); 380 381 if (value == NULL) 382 value_len = 0; 383 if (name == NULL) 384 return -EINVAL; 385 name_len = strlen(name); 386 if (name_len > 255 || value_len > sb->s_blocksize) 387 return -ERANGE; 388 down_write(&EXT2_I(inode)->xattr_sem); 389 if (EXT2_I(inode)->i_file_acl) { 390 /* The inode already has an extended attribute block. */ 391 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl); 392 error = -EIO; 393 if (!bh) 394 goto cleanup; 395 ea_bdebug(bh, "b_count=%d, refcount=%d", 396 atomic_read(&(bh->b_count)), 397 le32_to_cpu(HDR(bh)->h_refcount)); 398 header = HDR(bh); 399 end = bh->b_data + bh->b_size; 400 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 401 header->h_blocks != cpu_to_le32(1)) { 402 bad_block: ext2_error(sb, "ext2_xattr_set", 403 "inode %ld: bad block %d", inode->i_ino, 404 EXT2_I(inode)->i_file_acl); 405 error = -EIO; 406 goto cleanup; 407 } 408 /* Find the named attribute. */ 409 here = FIRST_ENTRY(bh); 410 while (!IS_LAST_ENTRY(here)) { 411 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here); 412 if ((char *)next >= end) 413 goto bad_block; 414 if (!here->e_value_block && here->e_value_size) { 415 size_t offs = le16_to_cpu(here->e_value_offs); 416 if (offs < min_offs) 417 min_offs = offs; 418 } 419 not_found = name_index - here->e_name_index; 420 if (!not_found) 421 not_found = name_len - here->e_name_len; 422 if (!not_found) 423 not_found = memcmp(name, here->e_name,name_len); 424 if (not_found <= 0) 425 break; 426 here = next; 427 } 428 last = here; 429 /* We still need to compute min_offs and last. */ 430 while (!IS_LAST_ENTRY(last)) { 431 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last); 432 if ((char *)next >= end) 433 goto bad_block; 434 if (!last->e_value_block && last->e_value_size) { 435 size_t offs = le16_to_cpu(last->e_value_offs); 436 if (offs < min_offs) 437 min_offs = offs; 438 } 439 last = next; 440 } 441 442 /* Check whether we have enough space left. */ 443 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32); 444 } else { 445 /* We will use a new extended attribute block. */ 446 free = sb->s_blocksize - 447 sizeof(struct ext2_xattr_header) - sizeof(__u32); 448 here = last = NULL; /* avoid gcc uninitialized warning. */ 449 } 450 451 if (not_found) { 452 /* Request to remove a nonexistent attribute? */ 453 error = -ENODATA; 454 if (flags & XATTR_REPLACE) 455 goto cleanup; 456 error = 0; 457 if (value == NULL) 458 goto cleanup; 459 } else { 460 /* Request to create an existing attribute? */ 461 error = -EEXIST; 462 if (flags & XATTR_CREATE) 463 goto cleanup; 464 if (!here->e_value_block && here->e_value_size) { 465 size_t size = le32_to_cpu(here->e_value_size); 466 467 if (le16_to_cpu(here->e_value_offs) + size > 468 sb->s_blocksize || size > sb->s_blocksize) 469 goto bad_block; 470 free += EXT2_XATTR_SIZE(size); 471 } 472 free += EXT2_XATTR_LEN(name_len); 473 } 474 error = -ENOSPC; 475 if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len)) 476 goto cleanup; 477 478 /* Here we know that we can set the new attribute. */ 479 480 if (header) { 481 struct mb_cache_entry *ce; 482 483 /* assert(header == HDR(bh)); */ 484 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, 485 bh->b_blocknr); 486 lock_buffer(bh); 487 if (header->h_refcount == cpu_to_le32(1)) { 488 ea_bdebug(bh, "modifying in-place"); 489 if (ce) 490 mb_cache_entry_free(ce); 491 /* keep the buffer locked while modifying it. */ 492 } else { 493 int offset; 494 495 if (ce) 496 mb_cache_entry_release(ce); 497 unlock_buffer(bh); 498 ea_bdebug(bh, "cloning"); 499 header = kmalloc(bh->b_size, GFP_KERNEL); 500 error = -ENOMEM; 501 if (header == NULL) 502 goto cleanup; 503 memcpy(header, HDR(bh), bh->b_size); 504 header->h_refcount = cpu_to_le32(1); 505 506 offset = (char *)here - bh->b_data; 507 here = ENTRY((char *)header + offset); 508 offset = (char *)last - bh->b_data; 509 last = ENTRY((char *)header + offset); 510 } 511 } else { 512 /* Allocate a buffer where we construct the new block. */ 513 header = kzalloc(sb->s_blocksize, GFP_KERNEL); 514 error = -ENOMEM; 515 if (header == NULL) 516 goto cleanup; 517 end = (char *)header + sb->s_blocksize; 518 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC); 519 header->h_blocks = header->h_refcount = cpu_to_le32(1); 520 last = here = ENTRY(header+1); 521 } 522 523 /* Iff we are modifying the block in-place, bh is locked here. */ 524 525 if (not_found) { 526 /* Insert the new name. */ 527 size_t size = EXT2_XATTR_LEN(name_len); 528 size_t rest = (char *)last - (char *)here; 529 memmove((char *)here + size, here, rest); 530 memset(here, 0, size); 531 here->e_name_index = name_index; 532 here->e_name_len = name_len; 533 memcpy(here->e_name, name, name_len); 534 } else { 535 if (!here->e_value_block && here->e_value_size) { 536 char *first_val = (char *)header + min_offs; 537 size_t offs = le16_to_cpu(here->e_value_offs); 538 char *val = (char *)header + offs; 539 size_t size = EXT2_XATTR_SIZE( 540 le32_to_cpu(here->e_value_size)); 541 542 if (size == EXT2_XATTR_SIZE(value_len)) { 543 /* The old and the new value have the same 544 size. Just replace. */ 545 here->e_value_size = cpu_to_le32(value_len); 546 memset(val + size - EXT2_XATTR_PAD, 0, 547 EXT2_XATTR_PAD); /* Clear pad bytes. */ 548 memcpy(val, value, value_len); 549 goto skip_replace; 550 } 551 552 /* Remove the old value. */ 553 memmove(first_val + size, first_val, val - first_val); 554 memset(first_val, 0, size); 555 here->e_value_offs = 0; 556 min_offs += size; 557 558 /* Adjust all value offsets. */ 559 last = ENTRY(header+1); 560 while (!IS_LAST_ENTRY(last)) { 561 size_t o = le16_to_cpu(last->e_value_offs); 562 if (!last->e_value_block && o < offs) 563 last->e_value_offs = 564 cpu_to_le16(o + size); 565 last = EXT2_XATTR_NEXT(last); 566 } 567 } 568 if (value == NULL) { 569 /* Remove the old name. */ 570 size_t size = EXT2_XATTR_LEN(name_len); 571 last = ENTRY((char *)last - size); 572 memmove(here, (char*)here + size, 573 (char*)last - (char*)here); 574 memset(last, 0, size); 575 } 576 } 577 578 if (value != NULL) { 579 /* Insert the new value. */ 580 here->e_value_size = cpu_to_le32(value_len); 581 if (value_len) { 582 size_t size = EXT2_XATTR_SIZE(value_len); 583 char *val = (char *)header + min_offs - size; 584 here->e_value_offs = 585 cpu_to_le16((char *)val - (char *)header); 586 memset(val + size - EXT2_XATTR_PAD, 0, 587 EXT2_XATTR_PAD); /* Clear the pad bytes. */ 588 memcpy(val, value, value_len); 589 } 590 } 591 592 skip_replace: 593 if (IS_LAST_ENTRY(ENTRY(header+1))) { 594 /* This block is now empty. */ 595 if (bh && header == HDR(bh)) 596 unlock_buffer(bh); /* we were modifying in-place. */ 597 error = ext2_xattr_set2(inode, bh, NULL); 598 } else { 599 ext2_xattr_rehash(header, here); 600 if (bh && header == HDR(bh)) 601 unlock_buffer(bh); /* we were modifying in-place. */ 602 error = ext2_xattr_set2(inode, bh, header); 603 } 604 605 cleanup: 606 brelse(bh); 607 if (!(bh && header == HDR(bh))) 608 kfree(header); 609 up_write(&EXT2_I(inode)->xattr_sem); 610 611 return error; 612 } 613 614 /* 615 * Second half of ext2_xattr_set(): Update the file system. 616 */ 617 static int 618 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh, 619 struct ext2_xattr_header *header) 620 { 621 struct super_block *sb = inode->i_sb; 622 struct buffer_head *new_bh = NULL; 623 int error; 624 625 if (header) { 626 new_bh = ext2_xattr_cache_find(inode, header); 627 if (new_bh) { 628 /* We found an identical block in the cache. */ 629 if (new_bh == old_bh) { 630 ea_bdebug(new_bh, "keeping this block"); 631 } else { 632 /* The old block is released after updating 633 the inode. */ 634 ea_bdebug(new_bh, "reusing block"); 635 636 error = dquot_alloc_block(inode, 1); 637 if (error) { 638 unlock_buffer(new_bh); 639 goto cleanup; 640 } 641 le32_add_cpu(&HDR(new_bh)->h_refcount, 1); 642 ea_bdebug(new_bh, "refcount now=%d", 643 le32_to_cpu(HDR(new_bh)->h_refcount)); 644 } 645 unlock_buffer(new_bh); 646 } else if (old_bh && header == HDR(old_bh)) { 647 /* Keep this block. No need to lock the block as we 648 don't need to change the reference count. */ 649 new_bh = old_bh; 650 get_bh(new_bh); 651 ext2_xattr_cache_insert(new_bh); 652 } else { 653 /* We need to allocate a new block */ 654 ext2_fsblk_t goal = ext2_group_first_block_no(sb, 655 EXT2_I(inode)->i_block_group); 656 int block = ext2_new_block(inode, goal, &error); 657 if (error) 658 goto cleanup; 659 ea_idebug(inode, "creating block %d", block); 660 661 new_bh = sb_getblk(sb, block); 662 if (unlikely(!new_bh)) { 663 ext2_free_blocks(inode, block, 1); 664 mark_inode_dirty(inode); 665 error = -ENOMEM; 666 goto cleanup; 667 } 668 lock_buffer(new_bh); 669 memcpy(new_bh->b_data, header, new_bh->b_size); 670 set_buffer_uptodate(new_bh); 671 unlock_buffer(new_bh); 672 ext2_xattr_cache_insert(new_bh); 673 674 ext2_xattr_update_super_block(sb); 675 } 676 mark_buffer_dirty(new_bh); 677 if (IS_SYNC(inode)) { 678 sync_dirty_buffer(new_bh); 679 error = -EIO; 680 if (buffer_req(new_bh) && !buffer_uptodate(new_bh)) 681 goto cleanup; 682 } 683 } 684 685 /* Update the inode. */ 686 EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0; 687 inode->i_ctime = CURRENT_TIME_SEC; 688 if (IS_SYNC(inode)) { 689 error = sync_inode_metadata(inode, 1); 690 /* In case sync failed due to ENOSPC the inode was actually 691 * written (only some dirty data were not) so we just proceed 692 * as if nothing happened and cleanup the unused block */ 693 if (error && error != -ENOSPC) { 694 if (new_bh && new_bh != old_bh) { 695 dquot_free_block_nodirty(inode, 1); 696 mark_inode_dirty(inode); 697 } 698 goto cleanup; 699 } 700 } else 701 mark_inode_dirty(inode); 702 703 error = 0; 704 if (old_bh && old_bh != new_bh) { 705 struct mb_cache_entry *ce; 706 707 /* 708 * If there was an old block and we are no longer using it, 709 * release the old block. 710 */ 711 ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev, 712 old_bh->b_blocknr); 713 lock_buffer(old_bh); 714 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) { 715 /* Free the old block. */ 716 if (ce) 717 mb_cache_entry_free(ce); 718 ea_bdebug(old_bh, "freeing"); 719 ext2_free_blocks(inode, old_bh->b_blocknr, 1); 720 mark_inode_dirty(inode); 721 /* We let our caller release old_bh, so we 722 * need to duplicate the buffer before. */ 723 get_bh(old_bh); 724 bforget(old_bh); 725 } else { 726 /* Decrement the refcount only. */ 727 le32_add_cpu(&HDR(old_bh)->h_refcount, -1); 728 if (ce) 729 mb_cache_entry_release(ce); 730 dquot_free_block_nodirty(inode, 1); 731 mark_inode_dirty(inode); 732 mark_buffer_dirty(old_bh); 733 ea_bdebug(old_bh, "refcount now=%d", 734 le32_to_cpu(HDR(old_bh)->h_refcount)); 735 } 736 unlock_buffer(old_bh); 737 } 738 739 cleanup: 740 brelse(new_bh); 741 742 return error; 743 } 744 745 /* 746 * ext2_xattr_delete_inode() 747 * 748 * Free extended attribute resources associated with this inode. This 749 * is called immediately before an inode is freed. 750 */ 751 void 752 ext2_xattr_delete_inode(struct inode *inode) 753 { 754 struct buffer_head *bh = NULL; 755 struct mb_cache_entry *ce; 756 757 down_write(&EXT2_I(inode)->xattr_sem); 758 if (!EXT2_I(inode)->i_file_acl) 759 goto cleanup; 760 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 761 if (!bh) { 762 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 763 "inode %ld: block %d read error", inode->i_ino, 764 EXT2_I(inode)->i_file_acl); 765 goto cleanup; 766 } 767 ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count))); 768 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 769 HDR(bh)->h_blocks != cpu_to_le32(1)) { 770 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 771 "inode %ld: bad block %d", inode->i_ino, 772 EXT2_I(inode)->i_file_acl); 773 goto cleanup; 774 } 775 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr); 776 lock_buffer(bh); 777 if (HDR(bh)->h_refcount == cpu_to_le32(1)) { 778 if (ce) 779 mb_cache_entry_free(ce); 780 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1); 781 get_bh(bh); 782 bforget(bh); 783 unlock_buffer(bh); 784 } else { 785 le32_add_cpu(&HDR(bh)->h_refcount, -1); 786 if (ce) 787 mb_cache_entry_release(ce); 788 ea_bdebug(bh, "refcount now=%d", 789 le32_to_cpu(HDR(bh)->h_refcount)); 790 unlock_buffer(bh); 791 mark_buffer_dirty(bh); 792 if (IS_SYNC(inode)) 793 sync_dirty_buffer(bh); 794 dquot_free_block_nodirty(inode, 1); 795 } 796 EXT2_I(inode)->i_file_acl = 0; 797 798 cleanup: 799 brelse(bh); 800 up_write(&EXT2_I(inode)->xattr_sem); 801 } 802 803 /* 804 * ext2_xattr_put_super() 805 * 806 * This is called when a file system is unmounted. 807 */ 808 void 809 ext2_xattr_put_super(struct super_block *sb) 810 { 811 mb_cache_shrink(sb->s_bdev); 812 } 813 814 815 /* 816 * ext2_xattr_cache_insert() 817 * 818 * Create a new entry in the extended attribute cache, and insert 819 * it unless such an entry is already in the cache. 820 * 821 * Returns 0, or a negative error number on failure. 822 */ 823 static int 824 ext2_xattr_cache_insert(struct buffer_head *bh) 825 { 826 __u32 hash = le32_to_cpu(HDR(bh)->h_hash); 827 struct mb_cache_entry *ce; 828 int error; 829 830 ce = mb_cache_entry_alloc(ext2_xattr_cache, GFP_NOFS); 831 if (!ce) 832 return -ENOMEM; 833 error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash); 834 if (error) { 835 mb_cache_entry_free(ce); 836 if (error == -EBUSY) { 837 ea_bdebug(bh, "already in cache (%d cache entries)", 838 atomic_read(&ext2_xattr_cache->c_entry_count)); 839 error = 0; 840 } 841 } else { 842 ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash, 843 atomic_read(&ext2_xattr_cache->c_entry_count)); 844 mb_cache_entry_release(ce); 845 } 846 return error; 847 } 848 849 /* 850 * ext2_xattr_cmp() 851 * 852 * Compare two extended attribute blocks for equality. 853 * 854 * Returns 0 if the blocks are equal, 1 if they differ, and 855 * a negative error number on errors. 856 */ 857 static int 858 ext2_xattr_cmp(struct ext2_xattr_header *header1, 859 struct ext2_xattr_header *header2) 860 { 861 struct ext2_xattr_entry *entry1, *entry2; 862 863 entry1 = ENTRY(header1+1); 864 entry2 = ENTRY(header2+1); 865 while (!IS_LAST_ENTRY(entry1)) { 866 if (IS_LAST_ENTRY(entry2)) 867 return 1; 868 if (entry1->e_hash != entry2->e_hash || 869 entry1->e_name_index != entry2->e_name_index || 870 entry1->e_name_len != entry2->e_name_len || 871 entry1->e_value_size != entry2->e_value_size || 872 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len)) 873 return 1; 874 if (entry1->e_value_block != 0 || entry2->e_value_block != 0) 875 return -EIO; 876 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs), 877 (char *)header2 + le16_to_cpu(entry2->e_value_offs), 878 le32_to_cpu(entry1->e_value_size))) 879 return 1; 880 881 entry1 = EXT2_XATTR_NEXT(entry1); 882 entry2 = EXT2_XATTR_NEXT(entry2); 883 } 884 if (!IS_LAST_ENTRY(entry2)) 885 return 1; 886 return 0; 887 } 888 889 /* 890 * ext2_xattr_cache_find() 891 * 892 * Find an identical extended attribute block. 893 * 894 * Returns a locked buffer head to the block found, or NULL if such 895 * a block was not found or an error occurred. 896 */ 897 static struct buffer_head * 898 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header) 899 { 900 __u32 hash = le32_to_cpu(header->h_hash); 901 struct mb_cache_entry *ce; 902 903 if (!header->h_hash) 904 return NULL; /* never share */ 905 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash); 906 again: 907 ce = mb_cache_entry_find_first(ext2_xattr_cache, inode->i_sb->s_bdev, 908 hash); 909 while (ce) { 910 struct buffer_head *bh; 911 912 if (IS_ERR(ce)) { 913 if (PTR_ERR(ce) == -EAGAIN) 914 goto again; 915 break; 916 } 917 918 bh = sb_bread(inode->i_sb, ce->e_block); 919 if (!bh) { 920 ext2_error(inode->i_sb, "ext2_xattr_cache_find", 921 "inode %ld: block %ld read error", 922 inode->i_ino, (unsigned long) ce->e_block); 923 } else { 924 lock_buffer(bh); 925 if (le32_to_cpu(HDR(bh)->h_refcount) > 926 EXT2_XATTR_REFCOUNT_MAX) { 927 ea_idebug(inode, "block %ld refcount %d>%d", 928 (unsigned long) ce->e_block, 929 le32_to_cpu(HDR(bh)->h_refcount), 930 EXT2_XATTR_REFCOUNT_MAX); 931 } else if (!ext2_xattr_cmp(header, HDR(bh))) { 932 ea_bdebug(bh, "b_count=%d", 933 atomic_read(&(bh->b_count))); 934 mb_cache_entry_release(ce); 935 return bh; 936 } 937 unlock_buffer(bh); 938 brelse(bh); 939 } 940 ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash); 941 } 942 return NULL; 943 } 944 945 #define NAME_HASH_SHIFT 5 946 #define VALUE_HASH_SHIFT 16 947 948 /* 949 * ext2_xattr_hash_entry() 950 * 951 * Compute the hash of an extended attribute. 952 */ 953 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header, 954 struct ext2_xattr_entry *entry) 955 { 956 __u32 hash = 0; 957 char *name = entry->e_name; 958 int n; 959 960 for (n=0; n < entry->e_name_len; n++) { 961 hash = (hash << NAME_HASH_SHIFT) ^ 962 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^ 963 *name++; 964 } 965 966 if (entry->e_value_block == 0 && entry->e_value_size != 0) { 967 __le32 *value = (__le32 *)((char *)header + 968 le16_to_cpu(entry->e_value_offs)); 969 for (n = (le32_to_cpu(entry->e_value_size) + 970 EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) { 971 hash = (hash << VALUE_HASH_SHIFT) ^ 972 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^ 973 le32_to_cpu(*value++); 974 } 975 } 976 entry->e_hash = cpu_to_le32(hash); 977 } 978 979 #undef NAME_HASH_SHIFT 980 #undef VALUE_HASH_SHIFT 981 982 #define BLOCK_HASH_SHIFT 16 983 984 /* 985 * ext2_xattr_rehash() 986 * 987 * Re-compute the extended attribute hash value after an entry has changed. 988 */ 989 static void ext2_xattr_rehash(struct ext2_xattr_header *header, 990 struct ext2_xattr_entry *entry) 991 { 992 struct ext2_xattr_entry *here; 993 __u32 hash = 0; 994 995 ext2_xattr_hash_entry(header, entry); 996 here = ENTRY(header+1); 997 while (!IS_LAST_ENTRY(here)) { 998 if (!here->e_hash) { 999 /* Block is not shared if an entry's hash value == 0 */ 1000 hash = 0; 1001 break; 1002 } 1003 hash = (hash << BLOCK_HASH_SHIFT) ^ 1004 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^ 1005 le32_to_cpu(here->e_hash); 1006 here = EXT2_XATTR_NEXT(here); 1007 } 1008 header->h_hash = cpu_to_le32(hash); 1009 } 1010 1011 #undef BLOCK_HASH_SHIFT 1012 1013 int __init 1014 init_ext2_xattr(void) 1015 { 1016 ext2_xattr_cache = mb_cache_create("ext2_xattr", 6); 1017 if (!ext2_xattr_cache) 1018 return -ENOMEM; 1019 return 0; 1020 } 1021 1022 void 1023 exit_ext2_xattr(void) 1024 { 1025 mb_cache_destroy(ext2_xattr_cache); 1026 } 1027