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 alligned 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/module.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 "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 char b[BDEVNAME_SIZE]; \ 81 printk(KERN_DEBUG "block %s:%lu: ", \ 82 bdevname(bh->b_bdev, b), \ 83 (unsigned long) bh->b_blocknr); \ 84 printk(f); \ 85 printk("\n"); \ 86 } while (0) 87 #else 88 # define ea_idebug(f...) 89 # define ea_bdebug(f...) 90 #endif 91 92 static int ext2_xattr_set2(struct inode *, struct buffer_head *, 93 struct ext2_xattr_header *); 94 95 static int ext2_xattr_cache_insert(struct buffer_head *); 96 static struct buffer_head *ext2_xattr_cache_find(struct inode *, 97 struct ext2_xattr_header *); 98 static void ext2_xattr_rehash(struct ext2_xattr_header *, 99 struct ext2_xattr_entry *); 100 101 static struct mb_cache *ext2_xattr_cache; 102 103 static struct xattr_handler *ext2_xattr_handler_map[] = { 104 [EXT2_XATTR_INDEX_USER] = &ext2_xattr_user_handler, 105 #ifdef CONFIG_EXT2_FS_POSIX_ACL 106 [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS] = &ext2_xattr_acl_access_handler, 107 [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext2_xattr_acl_default_handler, 108 #endif 109 [EXT2_XATTR_INDEX_TRUSTED] = &ext2_xattr_trusted_handler, 110 #ifdef CONFIG_EXT2_FS_SECURITY 111 [EXT2_XATTR_INDEX_SECURITY] = &ext2_xattr_security_handler, 112 #endif 113 }; 114 115 struct xattr_handler *ext2_xattr_handlers[] = { 116 &ext2_xattr_user_handler, 117 &ext2_xattr_trusted_handler, 118 #ifdef CONFIG_EXT2_FS_POSIX_ACL 119 &ext2_xattr_acl_access_handler, 120 &ext2_xattr_acl_default_handler, 121 #endif 122 #ifdef CONFIG_EXT2_FS_SECURITY 123 &ext2_xattr_security_handler, 124 #endif 125 NULL 126 }; 127 128 static inline struct xattr_handler * 129 ext2_xattr_handler(int name_index) 130 { 131 struct xattr_handler *handler = NULL; 132 133 if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map)) 134 handler = ext2_xattr_handler_map[name_index]; 135 return handler; 136 } 137 138 /* 139 * ext2_xattr_get() 140 * 141 * Copy an extended attribute into the buffer 142 * provided, or compute the buffer size required. 143 * Buffer is NULL to compute the size of the buffer required. 144 * 145 * Returns a negative error number on failure, or the number of bytes 146 * used / required on success. 147 */ 148 int 149 ext2_xattr_get(struct inode *inode, int name_index, const char *name, 150 void *buffer, size_t buffer_size) 151 { 152 struct buffer_head *bh = NULL; 153 struct ext2_xattr_entry *entry; 154 size_t name_len, size; 155 char *end; 156 int error; 157 158 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld", 159 name_index, name, buffer, (long)buffer_size); 160 161 if (name == NULL) 162 return -EINVAL; 163 down_read(&EXT2_I(inode)->xattr_sem); 164 error = -ENODATA; 165 if (!EXT2_I(inode)->i_file_acl) 166 goto cleanup; 167 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 168 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 169 error = -EIO; 170 if (!bh) 171 goto cleanup; 172 ea_bdebug(bh, "b_count=%d, refcount=%d", 173 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 174 end = bh->b_data + bh->b_size; 175 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 176 HDR(bh)->h_blocks != cpu_to_le32(1)) { 177 bad_block: ext2_error(inode->i_sb, "ext2_xattr_get", 178 "inode %ld: bad block %d", inode->i_ino, 179 EXT2_I(inode)->i_file_acl); 180 error = -EIO; 181 goto cleanup; 182 } 183 /* find named attribute */ 184 name_len = strlen(name); 185 186 error = -ERANGE; 187 if (name_len > 255) 188 goto cleanup; 189 entry = FIRST_ENTRY(bh); 190 while (!IS_LAST_ENTRY(entry)) { 191 struct ext2_xattr_entry *next = 192 EXT2_XATTR_NEXT(entry); 193 if ((char *)next >= end) 194 goto bad_block; 195 if (name_index == entry->e_name_index && 196 name_len == entry->e_name_len && 197 memcmp(name, entry->e_name, name_len) == 0) 198 goto found; 199 entry = next; 200 } 201 /* Check the remaining name entries */ 202 while (!IS_LAST_ENTRY(entry)) { 203 struct ext2_xattr_entry *next = 204 EXT2_XATTR_NEXT(entry); 205 if ((char *)next >= end) 206 goto bad_block; 207 entry = next; 208 } 209 if (ext2_xattr_cache_insert(bh)) 210 ea_idebug(inode, "cache insert failed"); 211 error = -ENODATA; 212 goto cleanup; 213 found: 214 /* check the buffer size */ 215 if (entry->e_value_block != 0) 216 goto bad_block; 217 size = le32_to_cpu(entry->e_value_size); 218 if (size > inode->i_sb->s_blocksize || 219 le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize) 220 goto bad_block; 221 222 if (ext2_xattr_cache_insert(bh)) 223 ea_idebug(inode, "cache insert failed"); 224 if (buffer) { 225 error = -ERANGE; 226 if (size > buffer_size) 227 goto cleanup; 228 /* return value of attribute */ 229 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs), 230 size); 231 } 232 error = size; 233 234 cleanup: 235 brelse(bh); 236 up_read(&EXT2_I(inode)->xattr_sem); 237 238 return error; 239 } 240 241 /* 242 * ext2_xattr_list() 243 * 244 * Copy a list of attribute names into the buffer 245 * provided, or compute the buffer size required. 246 * Buffer is NULL to compute the size of the buffer required. 247 * 248 * Returns a negative error number on failure, or the number of bytes 249 * used / required on success. 250 */ 251 static int 252 ext2_xattr_list(struct inode *inode, char *buffer, size_t buffer_size) 253 { 254 struct buffer_head *bh = NULL; 255 struct ext2_xattr_entry *entry; 256 char *end; 257 size_t rest = buffer_size; 258 int error; 259 260 ea_idebug(inode, "buffer=%p, buffer_size=%ld", 261 buffer, (long)buffer_size); 262 263 down_read(&EXT2_I(inode)->xattr_sem); 264 error = 0; 265 if (!EXT2_I(inode)->i_file_acl) 266 goto cleanup; 267 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 268 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 269 error = -EIO; 270 if (!bh) 271 goto cleanup; 272 ea_bdebug(bh, "b_count=%d, refcount=%d", 273 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 274 end = bh->b_data + bh->b_size; 275 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 276 HDR(bh)->h_blocks != cpu_to_le32(1)) { 277 bad_block: ext2_error(inode->i_sb, "ext2_xattr_list", 278 "inode %ld: bad block %d", inode->i_ino, 279 EXT2_I(inode)->i_file_acl); 280 error = -EIO; 281 goto cleanup; 282 } 283 284 /* check the on-disk data structure */ 285 entry = FIRST_ENTRY(bh); 286 while (!IS_LAST_ENTRY(entry)) { 287 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry); 288 289 if ((char *)next >= end) 290 goto bad_block; 291 entry = next; 292 } 293 if (ext2_xattr_cache_insert(bh)) 294 ea_idebug(inode, "cache insert failed"); 295 296 /* list the attribute names */ 297 for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry); 298 entry = EXT2_XATTR_NEXT(entry)) { 299 struct xattr_handler *handler = 300 ext2_xattr_handler(entry->e_name_index); 301 302 if (handler) { 303 size_t size = handler->list(inode, buffer, rest, 304 entry->e_name, 305 entry->e_name_len); 306 if (buffer) { 307 if (size > rest) { 308 error = -ERANGE; 309 goto cleanup; 310 } 311 buffer += size; 312 } 313 rest -= size; 314 } 315 } 316 error = buffer_size - rest; /* total size */ 317 318 cleanup: 319 brelse(bh); 320 up_read(&EXT2_I(inode)->xattr_sem); 321 322 return error; 323 } 324 325 /* 326 * Inode operation listxattr() 327 * 328 * dentry->d_inode->i_mutex: don't care 329 */ 330 ssize_t 331 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size) 332 { 333 return ext2_xattr_list(dentry->d_inode, buffer, size); 334 } 335 336 /* 337 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is 338 * not set, set it. 339 */ 340 static void ext2_xattr_update_super_block(struct super_block *sb) 341 { 342 if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR)) 343 return; 344 345 lock_super(sb); 346 EXT2_SB(sb)->s_es->s_feature_compat |= 347 cpu_to_le32(EXT2_FEATURE_COMPAT_EXT_ATTR); 348 sb->s_dirt = 1; 349 mark_buffer_dirty(EXT2_SB(sb)->s_sbh); 350 unlock_super(sb); 351 } 352 353 /* 354 * ext2_xattr_set() 355 * 356 * Create, replace or remove an extended attribute for this inode. Buffer 357 * is NULL to remove an existing extended attribute, and non-NULL to 358 * either replace an existing extended attribute, or create a new extended 359 * attribute. The flags XATTR_REPLACE and XATTR_CREATE 360 * specify that an extended attribute must exist and must not exist 361 * previous to the call, respectively. 362 * 363 * Returns 0, or a negative error number on failure. 364 */ 365 int 366 ext2_xattr_set(struct inode *inode, int name_index, const char *name, 367 const void *value, size_t value_len, int flags) 368 { 369 struct super_block *sb = inode->i_sb; 370 struct buffer_head *bh = NULL; 371 struct ext2_xattr_header *header = NULL; 372 struct ext2_xattr_entry *here, *last; 373 size_t name_len, free, min_offs = sb->s_blocksize; 374 int not_found = 1, error; 375 char *end; 376 377 /* 378 * header -- Points either into bh, or to a temporarily 379 * allocated buffer. 380 * here -- The named entry found, or the place for inserting, within 381 * the block pointed to by header. 382 * last -- Points right after the last named entry within the block 383 * pointed to by header. 384 * min_offs -- The offset of the first value (values are aligned 385 * towards the end of the block). 386 * end -- Points right after the block pointed to by header. 387 */ 388 389 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld", 390 name_index, name, value, (long)value_len); 391 392 if (value == NULL) 393 value_len = 0; 394 if (name == NULL) 395 return -EINVAL; 396 name_len = strlen(name); 397 if (name_len > 255 || value_len > sb->s_blocksize) 398 return -ERANGE; 399 down_write(&EXT2_I(inode)->xattr_sem); 400 if (EXT2_I(inode)->i_file_acl) { 401 /* The inode already has an extended attribute block. */ 402 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl); 403 error = -EIO; 404 if (!bh) 405 goto cleanup; 406 ea_bdebug(bh, "b_count=%d, refcount=%d", 407 atomic_read(&(bh->b_count)), 408 le32_to_cpu(HDR(bh)->h_refcount)); 409 header = HDR(bh); 410 end = bh->b_data + bh->b_size; 411 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 412 header->h_blocks != cpu_to_le32(1)) { 413 bad_block: ext2_error(sb, "ext2_xattr_set", 414 "inode %ld: bad block %d", inode->i_ino, 415 EXT2_I(inode)->i_file_acl); 416 error = -EIO; 417 goto cleanup; 418 } 419 /* Find the named attribute. */ 420 here = FIRST_ENTRY(bh); 421 while (!IS_LAST_ENTRY(here)) { 422 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here); 423 if ((char *)next >= end) 424 goto bad_block; 425 if (!here->e_value_block && here->e_value_size) { 426 size_t offs = le16_to_cpu(here->e_value_offs); 427 if (offs < min_offs) 428 min_offs = offs; 429 } 430 not_found = name_index - here->e_name_index; 431 if (!not_found) 432 not_found = name_len - here->e_name_len; 433 if (!not_found) 434 not_found = memcmp(name, here->e_name,name_len); 435 if (not_found <= 0) 436 break; 437 here = next; 438 } 439 last = here; 440 /* We still need to compute min_offs and last. */ 441 while (!IS_LAST_ENTRY(last)) { 442 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last); 443 if ((char *)next >= end) 444 goto bad_block; 445 if (!last->e_value_block && last->e_value_size) { 446 size_t offs = le16_to_cpu(last->e_value_offs); 447 if (offs < min_offs) 448 min_offs = offs; 449 } 450 last = next; 451 } 452 453 /* Check whether we have enough space left. */ 454 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32); 455 } else { 456 /* We will use a new extended attribute block. */ 457 free = sb->s_blocksize - 458 sizeof(struct ext2_xattr_header) - sizeof(__u32); 459 here = last = NULL; /* avoid gcc uninitialized warning. */ 460 } 461 462 if (not_found) { 463 /* Request to remove a nonexistent attribute? */ 464 error = -ENODATA; 465 if (flags & XATTR_REPLACE) 466 goto cleanup; 467 error = 0; 468 if (value == NULL) 469 goto cleanup; 470 } else { 471 /* Request to create an existing attribute? */ 472 error = -EEXIST; 473 if (flags & XATTR_CREATE) 474 goto cleanup; 475 if (!here->e_value_block && here->e_value_size) { 476 size_t size = le32_to_cpu(here->e_value_size); 477 478 if (le16_to_cpu(here->e_value_offs) + size > 479 sb->s_blocksize || size > sb->s_blocksize) 480 goto bad_block; 481 free += EXT2_XATTR_SIZE(size); 482 } 483 free += EXT2_XATTR_LEN(name_len); 484 } 485 error = -ENOSPC; 486 if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len)) 487 goto cleanup; 488 489 /* Here we know that we can set the new attribute. */ 490 491 if (header) { 492 struct mb_cache_entry *ce; 493 494 /* assert(header == HDR(bh)); */ 495 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, 496 bh->b_blocknr); 497 lock_buffer(bh); 498 if (header->h_refcount == cpu_to_le32(1)) { 499 ea_bdebug(bh, "modifying in-place"); 500 if (ce) 501 mb_cache_entry_free(ce); 502 /* keep the buffer locked while modifying it. */ 503 } else { 504 int offset; 505 506 if (ce) 507 mb_cache_entry_release(ce); 508 unlock_buffer(bh); 509 ea_bdebug(bh, "cloning"); 510 header = kmalloc(bh->b_size, GFP_KERNEL); 511 error = -ENOMEM; 512 if (header == NULL) 513 goto cleanup; 514 memcpy(header, HDR(bh), bh->b_size); 515 header->h_refcount = cpu_to_le32(1); 516 517 offset = (char *)here - bh->b_data; 518 here = ENTRY((char *)header + offset); 519 offset = (char *)last - bh->b_data; 520 last = ENTRY((char *)header + offset); 521 } 522 } else { 523 /* Allocate a buffer where we construct the new block. */ 524 header = kmalloc(sb->s_blocksize, GFP_KERNEL); 525 error = -ENOMEM; 526 if (header == NULL) 527 goto cleanup; 528 memset(header, 0, sb->s_blocksize); 529 end = (char *)header + sb->s_blocksize; 530 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC); 531 header->h_blocks = header->h_refcount = cpu_to_le32(1); 532 last = here = ENTRY(header+1); 533 } 534 535 /* Iff we are modifying the block in-place, bh is locked here. */ 536 537 if (not_found) { 538 /* Insert the new name. */ 539 size_t size = EXT2_XATTR_LEN(name_len); 540 size_t rest = (char *)last - (char *)here; 541 memmove((char *)here + size, here, rest); 542 memset(here, 0, size); 543 here->e_name_index = name_index; 544 here->e_name_len = name_len; 545 memcpy(here->e_name, name, name_len); 546 } else { 547 if (!here->e_value_block && here->e_value_size) { 548 char *first_val = (char *)header + min_offs; 549 size_t offs = le16_to_cpu(here->e_value_offs); 550 char *val = (char *)header + offs; 551 size_t size = EXT2_XATTR_SIZE( 552 le32_to_cpu(here->e_value_size)); 553 554 if (size == EXT2_XATTR_SIZE(value_len)) { 555 /* The old and the new value have the same 556 size. Just replace. */ 557 here->e_value_size = cpu_to_le32(value_len); 558 memset(val + size - EXT2_XATTR_PAD, 0, 559 EXT2_XATTR_PAD); /* Clear pad bytes. */ 560 memcpy(val, value, value_len); 561 goto skip_replace; 562 } 563 564 /* Remove the old value. */ 565 memmove(first_val + size, first_val, val - first_val); 566 memset(first_val, 0, size); 567 here->e_value_offs = 0; 568 min_offs += size; 569 570 /* Adjust all value offsets. */ 571 last = ENTRY(header+1); 572 while (!IS_LAST_ENTRY(last)) { 573 size_t o = le16_to_cpu(last->e_value_offs); 574 if (!last->e_value_block && o < offs) 575 last->e_value_offs = 576 cpu_to_le16(o + size); 577 last = EXT2_XATTR_NEXT(last); 578 } 579 } 580 if (value == NULL) { 581 /* Remove the old name. */ 582 size_t size = EXT2_XATTR_LEN(name_len); 583 last = ENTRY((char *)last - size); 584 memmove(here, (char*)here + size, 585 (char*)last - (char*)here); 586 memset(last, 0, size); 587 } 588 } 589 590 if (value != NULL) { 591 /* Insert the new value. */ 592 here->e_value_size = cpu_to_le32(value_len); 593 if (value_len) { 594 size_t size = EXT2_XATTR_SIZE(value_len); 595 char *val = (char *)header + min_offs - size; 596 here->e_value_offs = 597 cpu_to_le16((char *)val - (char *)header); 598 memset(val + size - EXT2_XATTR_PAD, 0, 599 EXT2_XATTR_PAD); /* Clear the pad bytes. */ 600 memcpy(val, value, value_len); 601 } 602 } 603 604 skip_replace: 605 if (IS_LAST_ENTRY(ENTRY(header+1))) { 606 /* This block is now empty. */ 607 if (bh && header == HDR(bh)) 608 unlock_buffer(bh); /* we were modifying in-place. */ 609 error = ext2_xattr_set2(inode, bh, NULL); 610 } else { 611 ext2_xattr_rehash(header, here); 612 if (bh && header == HDR(bh)) 613 unlock_buffer(bh); /* we were modifying in-place. */ 614 error = ext2_xattr_set2(inode, bh, header); 615 } 616 617 cleanup: 618 brelse(bh); 619 if (!(bh && header == HDR(bh))) 620 kfree(header); 621 up_write(&EXT2_I(inode)->xattr_sem); 622 623 return error; 624 } 625 626 /* 627 * Second half of ext2_xattr_set(): Update the file system. 628 */ 629 static int 630 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh, 631 struct ext2_xattr_header *header) 632 { 633 struct super_block *sb = inode->i_sb; 634 struct buffer_head *new_bh = NULL; 635 int error; 636 637 if (header) { 638 new_bh = ext2_xattr_cache_find(inode, header); 639 if (new_bh) { 640 /* We found an identical block in the cache. */ 641 if (new_bh == old_bh) { 642 ea_bdebug(new_bh, "keeping this block"); 643 } else { 644 /* The old block is released after updating 645 the inode. */ 646 ea_bdebug(new_bh, "reusing block"); 647 648 error = -EDQUOT; 649 if (DQUOT_ALLOC_BLOCK(inode, 1)) { 650 unlock_buffer(new_bh); 651 goto cleanup; 652 } 653 HDR(new_bh)->h_refcount = cpu_to_le32(1 + 654 le32_to_cpu(HDR(new_bh)->h_refcount)); 655 ea_bdebug(new_bh, "refcount now=%d", 656 le32_to_cpu(HDR(new_bh)->h_refcount)); 657 } 658 unlock_buffer(new_bh); 659 } else if (old_bh && header == HDR(old_bh)) { 660 /* Keep this block. No need to lock the block as we 661 don't need to change the reference count. */ 662 new_bh = old_bh; 663 get_bh(new_bh); 664 ext2_xattr_cache_insert(new_bh); 665 } else { 666 /* We need to allocate a new block */ 667 int goal = le32_to_cpu(EXT2_SB(sb)->s_es-> 668 s_first_data_block) + 669 EXT2_I(inode)->i_block_group * 670 EXT2_BLOCKS_PER_GROUP(sb); 671 int block = ext2_new_block(inode, goal, 672 NULL, NULL, &error); 673 if (error) 674 goto cleanup; 675 ea_idebug(inode, "creating block %d", block); 676 677 new_bh = sb_getblk(sb, block); 678 if (!new_bh) { 679 ext2_free_blocks(inode, block, 1); 680 error = -EIO; 681 goto cleanup; 682 } 683 lock_buffer(new_bh); 684 memcpy(new_bh->b_data, header, new_bh->b_size); 685 set_buffer_uptodate(new_bh); 686 unlock_buffer(new_bh); 687 ext2_xattr_cache_insert(new_bh); 688 689 ext2_xattr_update_super_block(sb); 690 } 691 mark_buffer_dirty(new_bh); 692 if (IS_SYNC(inode)) { 693 sync_dirty_buffer(new_bh); 694 error = -EIO; 695 if (buffer_req(new_bh) && !buffer_uptodate(new_bh)) 696 goto cleanup; 697 } 698 } 699 700 /* Update the inode. */ 701 EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0; 702 inode->i_ctime = CURRENT_TIME_SEC; 703 if (IS_SYNC(inode)) { 704 error = ext2_sync_inode (inode); 705 /* In case sync failed due to ENOSPC the inode was actually 706 * written (only some dirty data were not) so we just proceed 707 * as if nothing happened and cleanup the unused block */ 708 if (error && error != -ENOSPC) { 709 if (new_bh && new_bh != old_bh) 710 DQUOT_FREE_BLOCK(inode, 1); 711 goto cleanup; 712 } 713 } else 714 mark_inode_dirty(inode); 715 716 error = 0; 717 if (old_bh && old_bh != new_bh) { 718 struct mb_cache_entry *ce; 719 720 /* 721 * If there was an old block and we are no longer using it, 722 * release the old block. 723 */ 724 ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev, 725 old_bh->b_blocknr); 726 lock_buffer(old_bh); 727 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) { 728 /* Free the old block. */ 729 if (ce) 730 mb_cache_entry_free(ce); 731 ea_bdebug(old_bh, "freeing"); 732 ext2_free_blocks(inode, old_bh->b_blocknr, 1); 733 /* We let our caller release old_bh, so we 734 * need to duplicate the buffer before. */ 735 get_bh(old_bh); 736 bforget(old_bh); 737 } else { 738 /* Decrement the refcount only. */ 739 HDR(old_bh)->h_refcount = cpu_to_le32( 740 le32_to_cpu(HDR(old_bh)->h_refcount) - 1); 741 if (ce) 742 mb_cache_entry_release(ce); 743 DQUOT_FREE_BLOCK(inode, 1); 744 mark_buffer_dirty(old_bh); 745 ea_bdebug(old_bh, "refcount now=%d", 746 le32_to_cpu(HDR(old_bh)->h_refcount)); 747 } 748 unlock_buffer(old_bh); 749 } 750 751 cleanup: 752 brelse(new_bh); 753 754 return error; 755 } 756 757 /* 758 * ext2_xattr_delete_inode() 759 * 760 * Free extended attribute resources associated with this inode. This 761 * is called immediately before an inode is freed. 762 */ 763 void 764 ext2_xattr_delete_inode(struct inode *inode) 765 { 766 struct buffer_head *bh = NULL; 767 struct mb_cache_entry *ce; 768 769 down_write(&EXT2_I(inode)->xattr_sem); 770 if (!EXT2_I(inode)->i_file_acl) 771 goto cleanup; 772 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 773 if (!bh) { 774 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 775 "inode %ld: block %d read error", inode->i_ino, 776 EXT2_I(inode)->i_file_acl); 777 goto cleanup; 778 } 779 ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count))); 780 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 781 HDR(bh)->h_blocks != cpu_to_le32(1)) { 782 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 783 "inode %ld: bad block %d", inode->i_ino, 784 EXT2_I(inode)->i_file_acl); 785 goto cleanup; 786 } 787 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr); 788 lock_buffer(bh); 789 if (HDR(bh)->h_refcount == cpu_to_le32(1)) { 790 if (ce) 791 mb_cache_entry_free(ce); 792 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1); 793 get_bh(bh); 794 bforget(bh); 795 unlock_buffer(bh); 796 } else { 797 HDR(bh)->h_refcount = cpu_to_le32( 798 le32_to_cpu(HDR(bh)->h_refcount) - 1); 799 if (ce) 800 mb_cache_entry_release(ce); 801 ea_bdebug(bh, "refcount now=%d", 802 le32_to_cpu(HDR(bh)->h_refcount)); 803 unlock_buffer(bh); 804 mark_buffer_dirty(bh); 805 if (IS_SYNC(inode)) 806 sync_dirty_buffer(bh); 807 DQUOT_FREE_BLOCK(inode, 1); 808 } 809 EXT2_I(inode)->i_file_acl = 0; 810 811 cleanup: 812 brelse(bh); 813 up_write(&EXT2_I(inode)->xattr_sem); 814 } 815 816 /* 817 * ext2_xattr_put_super() 818 * 819 * This is called when a file system is unmounted. 820 */ 821 void 822 ext2_xattr_put_super(struct super_block *sb) 823 { 824 mb_cache_shrink(sb->s_bdev); 825 } 826 827 828 /* 829 * ext2_xattr_cache_insert() 830 * 831 * Create a new entry in the extended attribute cache, and insert 832 * it unless such an entry is already in the cache. 833 * 834 * Returns 0, or a negative error number on failure. 835 */ 836 static int 837 ext2_xattr_cache_insert(struct buffer_head *bh) 838 { 839 __u32 hash = le32_to_cpu(HDR(bh)->h_hash); 840 struct mb_cache_entry *ce; 841 int error; 842 843 ce = mb_cache_entry_alloc(ext2_xattr_cache); 844 if (!ce) 845 return -ENOMEM; 846 error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, &hash); 847 if (error) { 848 mb_cache_entry_free(ce); 849 if (error == -EBUSY) { 850 ea_bdebug(bh, "already in cache (%d cache entries)", 851 atomic_read(&ext2_xattr_cache->c_entry_count)); 852 error = 0; 853 } 854 } else { 855 ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash, 856 atomic_read(&ext2_xattr_cache->c_entry_count)); 857 mb_cache_entry_release(ce); 858 } 859 return error; 860 } 861 862 /* 863 * ext2_xattr_cmp() 864 * 865 * Compare two extended attribute blocks for equality. 866 * 867 * Returns 0 if the blocks are equal, 1 if they differ, and 868 * a negative error number on errors. 869 */ 870 static int 871 ext2_xattr_cmp(struct ext2_xattr_header *header1, 872 struct ext2_xattr_header *header2) 873 { 874 struct ext2_xattr_entry *entry1, *entry2; 875 876 entry1 = ENTRY(header1+1); 877 entry2 = ENTRY(header2+1); 878 while (!IS_LAST_ENTRY(entry1)) { 879 if (IS_LAST_ENTRY(entry2)) 880 return 1; 881 if (entry1->e_hash != entry2->e_hash || 882 entry1->e_name_index != entry2->e_name_index || 883 entry1->e_name_len != entry2->e_name_len || 884 entry1->e_value_size != entry2->e_value_size || 885 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len)) 886 return 1; 887 if (entry1->e_value_block != 0 || entry2->e_value_block != 0) 888 return -EIO; 889 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs), 890 (char *)header2 + le16_to_cpu(entry2->e_value_offs), 891 le32_to_cpu(entry1->e_value_size))) 892 return 1; 893 894 entry1 = EXT2_XATTR_NEXT(entry1); 895 entry2 = EXT2_XATTR_NEXT(entry2); 896 } 897 if (!IS_LAST_ENTRY(entry2)) 898 return 1; 899 return 0; 900 } 901 902 /* 903 * ext2_xattr_cache_find() 904 * 905 * Find an identical extended attribute block. 906 * 907 * Returns a locked buffer head to the block found, or NULL if such 908 * a block was not found or an error occurred. 909 */ 910 static struct buffer_head * 911 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header) 912 { 913 __u32 hash = le32_to_cpu(header->h_hash); 914 struct mb_cache_entry *ce; 915 916 if (!header->h_hash) 917 return NULL; /* never share */ 918 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash); 919 again: 920 ce = mb_cache_entry_find_first(ext2_xattr_cache, 0, 921 inode->i_sb->s_bdev, hash); 922 while (ce) { 923 struct buffer_head *bh; 924 925 if (IS_ERR(ce)) { 926 if (PTR_ERR(ce) == -EAGAIN) 927 goto again; 928 break; 929 } 930 931 bh = sb_bread(inode->i_sb, ce->e_block); 932 if (!bh) { 933 ext2_error(inode->i_sb, "ext2_xattr_cache_find", 934 "inode %ld: block %ld read error", 935 inode->i_ino, (unsigned long) ce->e_block); 936 } else { 937 lock_buffer(bh); 938 if (le32_to_cpu(HDR(bh)->h_refcount) > 939 EXT2_XATTR_REFCOUNT_MAX) { 940 ea_idebug(inode, "block %ld refcount %d>%d", 941 (unsigned long) ce->e_block, 942 le32_to_cpu(HDR(bh)->h_refcount), 943 EXT2_XATTR_REFCOUNT_MAX); 944 } else if (!ext2_xattr_cmp(header, HDR(bh))) { 945 ea_bdebug(bh, "b_count=%d", 946 atomic_read(&(bh->b_count))); 947 mb_cache_entry_release(ce); 948 return bh; 949 } 950 unlock_buffer(bh); 951 brelse(bh); 952 } 953 ce = mb_cache_entry_find_next(ce, 0, inode->i_sb->s_bdev, hash); 954 } 955 return NULL; 956 } 957 958 #define NAME_HASH_SHIFT 5 959 #define VALUE_HASH_SHIFT 16 960 961 /* 962 * ext2_xattr_hash_entry() 963 * 964 * Compute the hash of an extended attribute. 965 */ 966 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header, 967 struct ext2_xattr_entry *entry) 968 { 969 __u32 hash = 0; 970 char *name = entry->e_name; 971 int n; 972 973 for (n=0; n < entry->e_name_len; n++) { 974 hash = (hash << NAME_HASH_SHIFT) ^ 975 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^ 976 *name++; 977 } 978 979 if (entry->e_value_block == 0 && entry->e_value_size != 0) { 980 __le32 *value = (__le32 *)((char *)header + 981 le16_to_cpu(entry->e_value_offs)); 982 for (n = (le32_to_cpu(entry->e_value_size) + 983 EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) { 984 hash = (hash << VALUE_HASH_SHIFT) ^ 985 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^ 986 le32_to_cpu(*value++); 987 } 988 } 989 entry->e_hash = cpu_to_le32(hash); 990 } 991 992 #undef NAME_HASH_SHIFT 993 #undef VALUE_HASH_SHIFT 994 995 #define BLOCK_HASH_SHIFT 16 996 997 /* 998 * ext2_xattr_rehash() 999 * 1000 * Re-compute the extended attribute hash value after an entry has changed. 1001 */ 1002 static void ext2_xattr_rehash(struct ext2_xattr_header *header, 1003 struct ext2_xattr_entry *entry) 1004 { 1005 struct ext2_xattr_entry *here; 1006 __u32 hash = 0; 1007 1008 ext2_xattr_hash_entry(header, entry); 1009 here = ENTRY(header+1); 1010 while (!IS_LAST_ENTRY(here)) { 1011 if (!here->e_hash) { 1012 /* Block is not shared if an entry's hash value == 0 */ 1013 hash = 0; 1014 break; 1015 } 1016 hash = (hash << BLOCK_HASH_SHIFT) ^ 1017 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^ 1018 le32_to_cpu(here->e_hash); 1019 here = EXT2_XATTR_NEXT(here); 1020 } 1021 header->h_hash = cpu_to_le32(hash); 1022 } 1023 1024 #undef BLOCK_HASH_SHIFT 1025 1026 int __init 1027 init_ext2_xattr(void) 1028 { 1029 ext2_xattr_cache = mb_cache_create("ext2_xattr", NULL, 1030 sizeof(struct mb_cache_entry) + 1031 sizeof(((struct mb_cache_entry *) 0)->e_indexes[0]), 1, 6); 1032 if (!ext2_xattr_cache) 1033 return -ENOMEM; 1034 return 0; 1035 } 1036 1037 void 1038 exit_ext2_xattr(void) 1039 { 1040 mb_cache_destroy(ext2_xattr_cache); 1041 } 1042