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 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 const 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 const 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 const struct xattr_handler * 129 ext2_xattr_handler(int name_index) 130 { 131 const 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 name_len = strlen(name); 164 if (name_len > 255) 165 return -ERANGE; 166 167 down_read(&EXT2_I(inode)->xattr_sem); 168 error = -ENODATA; 169 if (!EXT2_I(inode)->i_file_acl) 170 goto cleanup; 171 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 172 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 173 error = -EIO; 174 if (!bh) 175 goto cleanup; 176 ea_bdebug(bh, "b_count=%d, refcount=%d", 177 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 178 end = bh->b_data + bh->b_size; 179 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 180 HDR(bh)->h_blocks != cpu_to_le32(1)) { 181 bad_block: ext2_error(inode->i_sb, "ext2_xattr_get", 182 "inode %ld: bad block %d", inode->i_ino, 183 EXT2_I(inode)->i_file_acl); 184 error = -EIO; 185 goto cleanup; 186 } 187 188 /* find named attribute */ 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 if (ext2_xattr_cache_insert(bh)) 202 ea_idebug(inode, "cache insert failed"); 203 error = -ENODATA; 204 goto cleanup; 205 found: 206 /* check the buffer size */ 207 if (entry->e_value_block != 0) 208 goto bad_block; 209 size = le32_to_cpu(entry->e_value_size); 210 if (size > inode->i_sb->s_blocksize || 211 le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize) 212 goto bad_block; 213 214 if (ext2_xattr_cache_insert(bh)) 215 ea_idebug(inode, "cache insert failed"); 216 if (buffer) { 217 error = -ERANGE; 218 if (size > buffer_size) 219 goto cleanup; 220 /* return value of attribute */ 221 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs), 222 size); 223 } 224 error = size; 225 226 cleanup: 227 brelse(bh); 228 up_read(&EXT2_I(inode)->xattr_sem); 229 230 return error; 231 } 232 233 /* 234 * ext2_xattr_list() 235 * 236 * Copy a list of attribute names into the buffer 237 * provided, or compute the buffer size required. 238 * Buffer is NULL to compute the size of the buffer required. 239 * 240 * Returns a negative error number on failure, or the number of bytes 241 * used / required on success. 242 */ 243 static int 244 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size) 245 { 246 struct inode *inode = dentry->d_inode; 247 struct buffer_head *bh = NULL; 248 struct ext2_xattr_entry *entry; 249 char *end; 250 size_t rest = buffer_size; 251 int error; 252 253 ea_idebug(inode, "buffer=%p, buffer_size=%ld", 254 buffer, (long)buffer_size); 255 256 down_read(&EXT2_I(inode)->xattr_sem); 257 error = 0; 258 if (!EXT2_I(inode)->i_file_acl) 259 goto cleanup; 260 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 261 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 262 error = -EIO; 263 if (!bh) 264 goto cleanup; 265 ea_bdebug(bh, "b_count=%d, refcount=%d", 266 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 267 end = bh->b_data + bh->b_size; 268 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 269 HDR(bh)->h_blocks != cpu_to_le32(1)) { 270 bad_block: ext2_error(inode->i_sb, "ext2_xattr_list", 271 "inode %ld: bad block %d", inode->i_ino, 272 EXT2_I(inode)->i_file_acl); 273 error = -EIO; 274 goto cleanup; 275 } 276 277 /* check the on-disk data structure */ 278 entry = FIRST_ENTRY(bh); 279 while (!IS_LAST_ENTRY(entry)) { 280 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry); 281 282 if ((char *)next >= end) 283 goto bad_block; 284 entry = next; 285 } 286 if (ext2_xattr_cache_insert(bh)) 287 ea_idebug(inode, "cache insert failed"); 288 289 /* list the attribute names */ 290 for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry); 291 entry = EXT2_XATTR_NEXT(entry)) { 292 const struct xattr_handler *handler = 293 ext2_xattr_handler(entry->e_name_index); 294 295 if (handler) { 296 size_t size = handler->list(dentry, buffer, rest, 297 entry->e_name, 298 entry->e_name_len, 299 handler->flags); 300 if (buffer) { 301 if (size > rest) { 302 error = -ERANGE; 303 goto cleanup; 304 } 305 buffer += size; 306 } 307 rest -= size; 308 } 309 } 310 error = buffer_size - rest; /* total size */ 311 312 cleanup: 313 brelse(bh); 314 up_read(&EXT2_I(inode)->xattr_sem); 315 316 return error; 317 } 318 319 /* 320 * Inode operation listxattr() 321 * 322 * dentry->d_inode->i_mutex: don't care 323 */ 324 ssize_t 325 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size) 326 { 327 return ext2_xattr_list(dentry, buffer, size); 328 } 329 330 /* 331 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is 332 * not set, set it. 333 */ 334 static void ext2_xattr_update_super_block(struct super_block *sb) 335 { 336 if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR)) 337 return; 338 339 spin_lock(&EXT2_SB(sb)->s_lock); 340 EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR); 341 spin_unlock(&EXT2_SB(sb)->s_lock); 342 sb->s_dirt = 1; 343 mark_buffer_dirty(EXT2_SB(sb)->s_sbh); 344 } 345 346 /* 347 * ext2_xattr_set() 348 * 349 * Create, replace or remove an extended attribute for this inode. Value 350 * is NULL to remove an existing extended attribute, and non-NULL to 351 * either replace an existing extended attribute, or create a new extended 352 * attribute. The flags XATTR_REPLACE and XATTR_CREATE 353 * specify that an extended attribute must exist and must not exist 354 * previous to the call, respectively. 355 * 356 * Returns 0, or a negative error number on failure. 357 */ 358 int 359 ext2_xattr_set(struct inode *inode, int name_index, const char *name, 360 const void *value, size_t value_len, int flags) 361 { 362 struct super_block *sb = inode->i_sb; 363 struct buffer_head *bh = NULL; 364 struct ext2_xattr_header *header = NULL; 365 struct ext2_xattr_entry *here, *last; 366 size_t name_len, free, min_offs = sb->s_blocksize; 367 int not_found = 1, error; 368 char *end; 369 370 /* 371 * header -- Points either into bh, or to a temporarily 372 * allocated buffer. 373 * here -- The named entry found, or the place for inserting, within 374 * the block pointed to by header. 375 * last -- Points right after the last named entry within the block 376 * pointed to by header. 377 * min_offs -- The offset of the first value (values are aligned 378 * towards the end of the block). 379 * end -- Points right after the block pointed to by header. 380 */ 381 382 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld", 383 name_index, name, value, (long)value_len); 384 385 if (value == NULL) 386 value_len = 0; 387 if (name == NULL) 388 return -EINVAL; 389 name_len = strlen(name); 390 if (name_len > 255 || value_len > sb->s_blocksize) 391 return -ERANGE; 392 down_write(&EXT2_I(inode)->xattr_sem); 393 if (EXT2_I(inode)->i_file_acl) { 394 /* The inode already has an extended attribute block. */ 395 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl); 396 error = -EIO; 397 if (!bh) 398 goto cleanup; 399 ea_bdebug(bh, "b_count=%d, refcount=%d", 400 atomic_read(&(bh->b_count)), 401 le32_to_cpu(HDR(bh)->h_refcount)); 402 header = HDR(bh); 403 end = bh->b_data + bh->b_size; 404 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 405 header->h_blocks != cpu_to_le32(1)) { 406 bad_block: ext2_error(sb, "ext2_xattr_set", 407 "inode %ld: bad block %d", inode->i_ino, 408 EXT2_I(inode)->i_file_acl); 409 error = -EIO; 410 goto cleanup; 411 } 412 /* Find the named attribute. */ 413 here = FIRST_ENTRY(bh); 414 while (!IS_LAST_ENTRY(here)) { 415 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here); 416 if ((char *)next >= end) 417 goto bad_block; 418 if (!here->e_value_block && here->e_value_size) { 419 size_t offs = le16_to_cpu(here->e_value_offs); 420 if (offs < min_offs) 421 min_offs = offs; 422 } 423 not_found = name_index - here->e_name_index; 424 if (!not_found) 425 not_found = name_len - here->e_name_len; 426 if (!not_found) 427 not_found = memcmp(name, here->e_name,name_len); 428 if (not_found <= 0) 429 break; 430 here = next; 431 } 432 last = here; 433 /* We still need to compute min_offs and last. */ 434 while (!IS_LAST_ENTRY(last)) { 435 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last); 436 if ((char *)next >= end) 437 goto bad_block; 438 if (!last->e_value_block && last->e_value_size) { 439 size_t offs = le16_to_cpu(last->e_value_offs); 440 if (offs < min_offs) 441 min_offs = offs; 442 } 443 last = next; 444 } 445 446 /* Check whether we have enough space left. */ 447 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32); 448 } else { 449 /* We will use a new extended attribute block. */ 450 free = sb->s_blocksize - 451 sizeof(struct ext2_xattr_header) - sizeof(__u32); 452 here = last = NULL; /* avoid gcc uninitialized warning. */ 453 } 454 455 if (not_found) { 456 /* Request to remove a nonexistent attribute? */ 457 error = -ENODATA; 458 if (flags & XATTR_REPLACE) 459 goto cleanup; 460 error = 0; 461 if (value == NULL) 462 goto cleanup; 463 } else { 464 /* Request to create an existing attribute? */ 465 error = -EEXIST; 466 if (flags & XATTR_CREATE) 467 goto cleanup; 468 if (!here->e_value_block && here->e_value_size) { 469 size_t size = le32_to_cpu(here->e_value_size); 470 471 if (le16_to_cpu(here->e_value_offs) + size > 472 sb->s_blocksize || size > sb->s_blocksize) 473 goto bad_block; 474 free += EXT2_XATTR_SIZE(size); 475 } 476 free += EXT2_XATTR_LEN(name_len); 477 } 478 error = -ENOSPC; 479 if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len)) 480 goto cleanup; 481 482 /* Here we know that we can set the new attribute. */ 483 484 if (header) { 485 struct mb_cache_entry *ce; 486 487 /* assert(header == HDR(bh)); */ 488 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, 489 bh->b_blocknr); 490 lock_buffer(bh); 491 if (header->h_refcount == cpu_to_le32(1)) { 492 ea_bdebug(bh, "modifying in-place"); 493 if (ce) 494 mb_cache_entry_free(ce); 495 /* keep the buffer locked while modifying it. */ 496 } else { 497 int offset; 498 499 if (ce) 500 mb_cache_entry_release(ce); 501 unlock_buffer(bh); 502 ea_bdebug(bh, "cloning"); 503 header = kmalloc(bh->b_size, GFP_KERNEL); 504 error = -ENOMEM; 505 if (header == NULL) 506 goto cleanup; 507 memcpy(header, HDR(bh), bh->b_size); 508 header->h_refcount = cpu_to_le32(1); 509 510 offset = (char *)here - bh->b_data; 511 here = ENTRY((char *)header + offset); 512 offset = (char *)last - bh->b_data; 513 last = ENTRY((char *)header + offset); 514 } 515 } else { 516 /* Allocate a buffer where we construct the new block. */ 517 header = kzalloc(sb->s_blocksize, GFP_KERNEL); 518 error = -ENOMEM; 519 if (header == NULL) 520 goto cleanup; 521 end = (char *)header + sb->s_blocksize; 522 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC); 523 header->h_blocks = header->h_refcount = cpu_to_le32(1); 524 last = here = ENTRY(header+1); 525 } 526 527 /* Iff we are modifying the block in-place, bh is locked here. */ 528 529 if (not_found) { 530 /* Insert the new name. */ 531 size_t size = EXT2_XATTR_LEN(name_len); 532 size_t rest = (char *)last - (char *)here; 533 memmove((char *)here + size, here, rest); 534 memset(here, 0, size); 535 here->e_name_index = name_index; 536 here->e_name_len = name_len; 537 memcpy(here->e_name, name, name_len); 538 } else { 539 if (!here->e_value_block && here->e_value_size) { 540 char *first_val = (char *)header + min_offs; 541 size_t offs = le16_to_cpu(here->e_value_offs); 542 char *val = (char *)header + offs; 543 size_t size = EXT2_XATTR_SIZE( 544 le32_to_cpu(here->e_value_size)); 545 546 if (size == EXT2_XATTR_SIZE(value_len)) { 547 /* The old and the new value have the same 548 size. Just replace. */ 549 here->e_value_size = cpu_to_le32(value_len); 550 memset(val + size - EXT2_XATTR_PAD, 0, 551 EXT2_XATTR_PAD); /* Clear pad bytes. */ 552 memcpy(val, value, value_len); 553 goto skip_replace; 554 } 555 556 /* Remove the old value. */ 557 memmove(first_val + size, first_val, val - first_val); 558 memset(first_val, 0, size); 559 here->e_value_offs = 0; 560 min_offs += size; 561 562 /* Adjust all value offsets. */ 563 last = ENTRY(header+1); 564 while (!IS_LAST_ENTRY(last)) { 565 size_t o = le16_to_cpu(last->e_value_offs); 566 if (!last->e_value_block && o < offs) 567 last->e_value_offs = 568 cpu_to_le16(o + size); 569 last = EXT2_XATTR_NEXT(last); 570 } 571 } 572 if (value == NULL) { 573 /* Remove the old name. */ 574 size_t size = EXT2_XATTR_LEN(name_len); 575 last = ENTRY((char *)last - size); 576 memmove(here, (char*)here + size, 577 (char*)last - (char*)here); 578 memset(last, 0, size); 579 } 580 } 581 582 if (value != NULL) { 583 /* Insert the new value. */ 584 here->e_value_size = cpu_to_le32(value_len); 585 if (value_len) { 586 size_t size = EXT2_XATTR_SIZE(value_len); 587 char *val = (char *)header + min_offs - size; 588 here->e_value_offs = 589 cpu_to_le16((char *)val - (char *)header); 590 memset(val + size - EXT2_XATTR_PAD, 0, 591 EXT2_XATTR_PAD); /* Clear the pad bytes. */ 592 memcpy(val, value, value_len); 593 } 594 } 595 596 skip_replace: 597 if (IS_LAST_ENTRY(ENTRY(header+1))) { 598 /* This block is now empty. */ 599 if (bh && header == HDR(bh)) 600 unlock_buffer(bh); /* we were modifying in-place. */ 601 error = ext2_xattr_set2(inode, bh, NULL); 602 } else { 603 ext2_xattr_rehash(header, here); 604 if (bh && header == HDR(bh)) 605 unlock_buffer(bh); /* we were modifying in-place. */ 606 error = ext2_xattr_set2(inode, bh, header); 607 } 608 609 cleanup: 610 brelse(bh); 611 if (!(bh && header == HDR(bh))) 612 kfree(header); 613 up_write(&EXT2_I(inode)->xattr_sem); 614 615 return error; 616 } 617 618 /* 619 * Second half of ext2_xattr_set(): Update the file system. 620 */ 621 static int 622 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh, 623 struct ext2_xattr_header *header) 624 { 625 struct super_block *sb = inode->i_sb; 626 struct buffer_head *new_bh = NULL; 627 int error; 628 629 if (header) { 630 new_bh = ext2_xattr_cache_find(inode, header); 631 if (new_bh) { 632 /* We found an identical block in the cache. */ 633 if (new_bh == old_bh) { 634 ea_bdebug(new_bh, "keeping this block"); 635 } else { 636 /* The old block is released after updating 637 the inode. */ 638 ea_bdebug(new_bh, "reusing block"); 639 640 error = dquot_alloc_block(inode, 1); 641 if (error) { 642 unlock_buffer(new_bh); 643 goto cleanup; 644 } 645 le32_add_cpu(&HDR(new_bh)->h_refcount, 1); 646 ea_bdebug(new_bh, "refcount now=%d", 647 le32_to_cpu(HDR(new_bh)->h_refcount)); 648 } 649 unlock_buffer(new_bh); 650 } else if (old_bh && header == HDR(old_bh)) { 651 /* Keep this block. No need to lock the block as we 652 don't need to change the reference count. */ 653 new_bh = old_bh; 654 get_bh(new_bh); 655 ext2_xattr_cache_insert(new_bh); 656 } else { 657 /* We need to allocate a new block */ 658 ext2_fsblk_t goal = ext2_group_first_block_no(sb, 659 EXT2_I(inode)->i_block_group); 660 int block = ext2_new_block(inode, goal, &error); 661 if (error) 662 goto cleanup; 663 ea_idebug(inode, "creating block %d", block); 664 665 new_bh = sb_getblk(sb, block); 666 if (!new_bh) { 667 ext2_free_blocks(inode, block, 1); 668 mark_inode_dirty(inode); 669 error = -EIO; 670 goto cleanup; 671 } 672 lock_buffer(new_bh); 673 memcpy(new_bh->b_data, header, new_bh->b_size); 674 set_buffer_uptodate(new_bh); 675 unlock_buffer(new_bh); 676 ext2_xattr_cache_insert(new_bh); 677 678 ext2_xattr_update_super_block(sb); 679 } 680 mark_buffer_dirty(new_bh); 681 if (IS_SYNC(inode)) { 682 sync_dirty_buffer(new_bh); 683 error = -EIO; 684 if (buffer_req(new_bh) && !buffer_uptodate(new_bh)) 685 goto cleanup; 686 } 687 } 688 689 /* Update the inode. */ 690 EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0; 691 inode->i_ctime = CURRENT_TIME_SEC; 692 if (IS_SYNC(inode)) { 693 error = sync_inode_metadata(inode, 1); 694 /* In case sync failed due to ENOSPC the inode was actually 695 * written (only some dirty data were not) so we just proceed 696 * as if nothing happened and cleanup the unused block */ 697 if (error && error != -ENOSPC) { 698 if (new_bh && new_bh != old_bh) { 699 dquot_free_block_nodirty(inode, 1); 700 mark_inode_dirty(inode); 701 } 702 goto cleanup; 703 } 704 } else 705 mark_inode_dirty(inode); 706 707 error = 0; 708 if (old_bh && old_bh != new_bh) { 709 struct mb_cache_entry *ce; 710 711 /* 712 * If there was an old block and we are no longer using it, 713 * release the old block. 714 */ 715 ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev, 716 old_bh->b_blocknr); 717 lock_buffer(old_bh); 718 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) { 719 /* Free the old block. */ 720 if (ce) 721 mb_cache_entry_free(ce); 722 ea_bdebug(old_bh, "freeing"); 723 ext2_free_blocks(inode, old_bh->b_blocknr, 1); 724 mark_inode_dirty(inode); 725 /* We let our caller release old_bh, so we 726 * need to duplicate the buffer before. */ 727 get_bh(old_bh); 728 bforget(old_bh); 729 } else { 730 /* Decrement the refcount only. */ 731 le32_add_cpu(&HDR(old_bh)->h_refcount, -1); 732 if (ce) 733 mb_cache_entry_release(ce); 734 dquot_free_block_nodirty(inode, 1); 735 mark_inode_dirty(inode); 736 mark_buffer_dirty(old_bh); 737 ea_bdebug(old_bh, "refcount now=%d", 738 le32_to_cpu(HDR(old_bh)->h_refcount)); 739 } 740 unlock_buffer(old_bh); 741 } 742 743 cleanup: 744 brelse(new_bh); 745 746 return error; 747 } 748 749 /* 750 * ext2_xattr_delete_inode() 751 * 752 * Free extended attribute resources associated with this inode. This 753 * is called immediately before an inode is freed. 754 */ 755 void 756 ext2_xattr_delete_inode(struct inode *inode) 757 { 758 struct buffer_head *bh = NULL; 759 struct mb_cache_entry *ce; 760 761 down_write(&EXT2_I(inode)->xattr_sem); 762 if (!EXT2_I(inode)->i_file_acl) 763 goto cleanup; 764 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 765 if (!bh) { 766 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 767 "inode %ld: block %d read error", inode->i_ino, 768 EXT2_I(inode)->i_file_acl); 769 goto cleanup; 770 } 771 ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count))); 772 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 773 HDR(bh)->h_blocks != cpu_to_le32(1)) { 774 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 775 "inode %ld: bad block %d", inode->i_ino, 776 EXT2_I(inode)->i_file_acl); 777 goto cleanup; 778 } 779 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr); 780 lock_buffer(bh); 781 if (HDR(bh)->h_refcount == cpu_to_le32(1)) { 782 if (ce) 783 mb_cache_entry_free(ce); 784 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1); 785 get_bh(bh); 786 bforget(bh); 787 unlock_buffer(bh); 788 } else { 789 le32_add_cpu(&HDR(bh)->h_refcount, -1); 790 if (ce) 791 mb_cache_entry_release(ce); 792 ea_bdebug(bh, "refcount now=%d", 793 le32_to_cpu(HDR(bh)->h_refcount)); 794 unlock_buffer(bh); 795 mark_buffer_dirty(bh); 796 if (IS_SYNC(inode)) 797 sync_dirty_buffer(bh); 798 dquot_free_block_nodirty(inode, 1); 799 } 800 EXT2_I(inode)->i_file_acl = 0; 801 802 cleanup: 803 brelse(bh); 804 up_write(&EXT2_I(inode)->xattr_sem); 805 } 806 807 /* 808 * ext2_xattr_put_super() 809 * 810 * This is called when a file system is unmounted. 811 */ 812 void 813 ext2_xattr_put_super(struct super_block *sb) 814 { 815 mb_cache_shrink(sb->s_bdev); 816 } 817 818 819 /* 820 * ext2_xattr_cache_insert() 821 * 822 * Create a new entry in the extended attribute cache, and insert 823 * it unless such an entry is already in the cache. 824 * 825 * Returns 0, or a negative error number on failure. 826 */ 827 static int 828 ext2_xattr_cache_insert(struct buffer_head *bh) 829 { 830 __u32 hash = le32_to_cpu(HDR(bh)->h_hash); 831 struct mb_cache_entry *ce; 832 int error; 833 834 ce = mb_cache_entry_alloc(ext2_xattr_cache, GFP_NOFS); 835 if (!ce) 836 return -ENOMEM; 837 error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash); 838 if (error) { 839 mb_cache_entry_free(ce); 840 if (error == -EBUSY) { 841 ea_bdebug(bh, "already in cache (%d cache entries)", 842 atomic_read(&ext2_xattr_cache->c_entry_count)); 843 error = 0; 844 } 845 } else { 846 ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash, 847 atomic_read(&ext2_xattr_cache->c_entry_count)); 848 mb_cache_entry_release(ce); 849 } 850 return error; 851 } 852 853 /* 854 * ext2_xattr_cmp() 855 * 856 * Compare two extended attribute blocks for equality. 857 * 858 * Returns 0 if the blocks are equal, 1 if they differ, and 859 * a negative error number on errors. 860 */ 861 static int 862 ext2_xattr_cmp(struct ext2_xattr_header *header1, 863 struct ext2_xattr_header *header2) 864 { 865 struct ext2_xattr_entry *entry1, *entry2; 866 867 entry1 = ENTRY(header1+1); 868 entry2 = ENTRY(header2+1); 869 while (!IS_LAST_ENTRY(entry1)) { 870 if (IS_LAST_ENTRY(entry2)) 871 return 1; 872 if (entry1->e_hash != entry2->e_hash || 873 entry1->e_name_index != entry2->e_name_index || 874 entry1->e_name_len != entry2->e_name_len || 875 entry1->e_value_size != entry2->e_value_size || 876 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len)) 877 return 1; 878 if (entry1->e_value_block != 0 || entry2->e_value_block != 0) 879 return -EIO; 880 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs), 881 (char *)header2 + le16_to_cpu(entry2->e_value_offs), 882 le32_to_cpu(entry1->e_value_size))) 883 return 1; 884 885 entry1 = EXT2_XATTR_NEXT(entry1); 886 entry2 = EXT2_XATTR_NEXT(entry2); 887 } 888 if (!IS_LAST_ENTRY(entry2)) 889 return 1; 890 return 0; 891 } 892 893 /* 894 * ext2_xattr_cache_find() 895 * 896 * Find an identical extended attribute block. 897 * 898 * Returns a locked buffer head to the block found, or NULL if such 899 * a block was not found or an error occurred. 900 */ 901 static struct buffer_head * 902 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header) 903 { 904 __u32 hash = le32_to_cpu(header->h_hash); 905 struct mb_cache_entry *ce; 906 907 if (!header->h_hash) 908 return NULL; /* never share */ 909 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash); 910 again: 911 ce = mb_cache_entry_find_first(ext2_xattr_cache, inode->i_sb->s_bdev, 912 hash); 913 while (ce) { 914 struct buffer_head *bh; 915 916 if (IS_ERR(ce)) { 917 if (PTR_ERR(ce) == -EAGAIN) 918 goto again; 919 break; 920 } 921 922 bh = sb_bread(inode->i_sb, ce->e_block); 923 if (!bh) { 924 ext2_error(inode->i_sb, "ext2_xattr_cache_find", 925 "inode %ld: block %ld read error", 926 inode->i_ino, (unsigned long) ce->e_block); 927 } else { 928 lock_buffer(bh); 929 if (le32_to_cpu(HDR(bh)->h_refcount) > 930 EXT2_XATTR_REFCOUNT_MAX) { 931 ea_idebug(inode, "block %ld refcount %d>%d", 932 (unsigned long) ce->e_block, 933 le32_to_cpu(HDR(bh)->h_refcount), 934 EXT2_XATTR_REFCOUNT_MAX); 935 } else if (!ext2_xattr_cmp(header, HDR(bh))) { 936 ea_bdebug(bh, "b_count=%d", 937 atomic_read(&(bh->b_count))); 938 mb_cache_entry_release(ce); 939 return bh; 940 } 941 unlock_buffer(bh); 942 brelse(bh); 943 } 944 ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash); 945 } 946 return NULL; 947 } 948 949 #define NAME_HASH_SHIFT 5 950 #define VALUE_HASH_SHIFT 16 951 952 /* 953 * ext2_xattr_hash_entry() 954 * 955 * Compute the hash of an extended attribute. 956 */ 957 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header, 958 struct ext2_xattr_entry *entry) 959 { 960 __u32 hash = 0; 961 char *name = entry->e_name; 962 int n; 963 964 for (n=0; n < entry->e_name_len; n++) { 965 hash = (hash << NAME_HASH_SHIFT) ^ 966 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^ 967 *name++; 968 } 969 970 if (entry->e_value_block == 0 && entry->e_value_size != 0) { 971 __le32 *value = (__le32 *)((char *)header + 972 le16_to_cpu(entry->e_value_offs)); 973 for (n = (le32_to_cpu(entry->e_value_size) + 974 EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) { 975 hash = (hash << VALUE_HASH_SHIFT) ^ 976 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^ 977 le32_to_cpu(*value++); 978 } 979 } 980 entry->e_hash = cpu_to_le32(hash); 981 } 982 983 #undef NAME_HASH_SHIFT 984 #undef VALUE_HASH_SHIFT 985 986 #define BLOCK_HASH_SHIFT 16 987 988 /* 989 * ext2_xattr_rehash() 990 * 991 * Re-compute the extended attribute hash value after an entry has changed. 992 */ 993 static void ext2_xattr_rehash(struct ext2_xattr_header *header, 994 struct ext2_xattr_entry *entry) 995 { 996 struct ext2_xattr_entry *here; 997 __u32 hash = 0; 998 999 ext2_xattr_hash_entry(header, entry); 1000 here = ENTRY(header+1); 1001 while (!IS_LAST_ENTRY(here)) { 1002 if (!here->e_hash) { 1003 /* Block is not shared if an entry's hash value == 0 */ 1004 hash = 0; 1005 break; 1006 } 1007 hash = (hash << BLOCK_HASH_SHIFT) ^ 1008 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^ 1009 le32_to_cpu(here->e_hash); 1010 here = EXT2_XATTR_NEXT(here); 1011 } 1012 header->h_hash = cpu_to_le32(hash); 1013 } 1014 1015 #undef BLOCK_HASH_SHIFT 1016 1017 int __init 1018 init_ext2_xattr(void) 1019 { 1020 ext2_xattr_cache = mb_cache_create("ext2_xattr", 6); 1021 if (!ext2_xattr_cache) 1022 return -ENOMEM; 1023 return 0; 1024 } 1025 1026 void 1027 exit_ext2_xattr(void) 1028 { 1029 mb_cache_destroy(ext2_xattr_cache); 1030 } 1031