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 <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 char b[BDEVNAME_SIZE]; \ 82 printk(KERN_DEBUG "block %s:%lu: ", \ 83 bdevname(bh->b_bdev, b), \ 84 (unsigned long) bh->b_blocknr); \ 85 printk(f); \ 86 printk("\n"); \ 87 } while (0) 88#else 89# define ea_idebug(f...) 90# define ea_bdebug(f...) 91#endif 92 93static int ext2_xattr_set2(struct inode *, struct buffer_head *, 94 struct ext2_xattr_header *); 95 96static int ext2_xattr_cache_insert(struct buffer_head *); 97static struct buffer_head *ext2_xattr_cache_find(struct inode *, 98 struct ext2_xattr_header *); 99static void ext2_xattr_rehash(struct ext2_xattr_header *, 100 struct ext2_xattr_entry *); 101 102static struct mb_cache *ext2_xattr_cache; 103 104static const struct xattr_handler *ext2_xattr_handler_map[] = { 105 [EXT2_XATTR_INDEX_USER] = &ext2_xattr_user_handler, 106#ifdef CONFIG_EXT2_FS_POSIX_ACL 107 [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS] = &ext2_xattr_acl_access_handler, 108 [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext2_xattr_acl_default_handler, 109#endif 110 [EXT2_XATTR_INDEX_TRUSTED] = &ext2_xattr_trusted_handler, 111#ifdef CONFIG_EXT2_FS_SECURITY 112 [EXT2_XATTR_INDEX_SECURITY] = &ext2_xattr_security_handler, 113#endif 114}; 115 116const struct xattr_handler *ext2_xattr_handlers[] = { 117 &ext2_xattr_user_handler, 118 &ext2_xattr_trusted_handler, 119#ifdef CONFIG_EXT2_FS_POSIX_ACL 120 &ext2_xattr_acl_access_handler, 121 &ext2_xattr_acl_default_handler, 122#endif 123#ifdef CONFIG_EXT2_FS_SECURITY 124 &ext2_xattr_security_handler, 125#endif 126 NULL 127}; 128 129static inline const struct xattr_handler * 130ext2_xattr_handler(int name_index) 131{ 132 const struct xattr_handler *handler = NULL; 133 134 if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map)) 135 handler = ext2_xattr_handler_map[name_index]; 136 return handler; 137} 138 139/* 140 * ext2_xattr_get() 141 * 142 * Copy an extended attribute into the buffer 143 * provided, or compute the buffer size required. 144 * Buffer is NULL to compute the size of the buffer required. 145 * 146 * Returns a negative error number on failure, or the number of bytes 147 * used / required on success. 148 */ 149int 150ext2_xattr_get(struct inode *inode, int name_index, const char *name, 151 void *buffer, size_t buffer_size) 152{ 153 struct buffer_head *bh = NULL; 154 struct ext2_xattr_entry *entry; 155 size_t name_len, size; 156 char *end; 157 int error; 158 159 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld", 160 name_index, name, buffer, (long)buffer_size); 161 162 if (name == NULL) 163 return -EINVAL; 164 down_read(&EXT2_I(inode)->xattr_sem); 165 error = -ENODATA; 166 if (!EXT2_I(inode)->i_file_acl) 167 goto cleanup; 168 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 169 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 170 error = -EIO; 171 if (!bh) 172 goto cleanup; 173 ea_bdebug(bh, "b_count=%d, refcount=%d", 174 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 175 end = bh->b_data + bh->b_size; 176 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 177 HDR(bh)->h_blocks != cpu_to_le32(1)) { 178bad_block: ext2_error(inode->i_sb, "ext2_xattr_get", 179 "inode %ld: bad block %d", inode->i_ino, 180 EXT2_I(inode)->i_file_acl); 181 error = -EIO; 182 goto cleanup; 183 } 184 /* find named attribute */ 185 name_len = strlen(name); 186 187 error = -ERANGE; 188 if (name_len > 255) 189 goto cleanup; 190 entry = FIRST_ENTRY(bh); 191 while (!IS_LAST_ENTRY(entry)) { 192 struct ext2_xattr_entry *next = 193 EXT2_XATTR_NEXT(entry); 194 if ((char *)next >= end) 195 goto bad_block; 196 if (name_index == entry->e_name_index && 197 name_len == entry->e_name_len && 198 memcmp(name, entry->e_name, name_len) == 0) 199 goto found; 200 entry = next; 201 } 202 /* Check the remaining name entries */ 203 while (!IS_LAST_ENTRY(entry)) { 204 struct ext2_xattr_entry *next = 205 EXT2_XATTR_NEXT(entry); 206 if ((char *)next >= end) 207 goto bad_block; 208 entry = next; 209 } 210 if (ext2_xattr_cache_insert(bh)) 211 ea_idebug(inode, "cache insert failed"); 212 error = -ENODATA; 213 goto cleanup; 214found: 215 /* check the buffer size */ 216 if (entry->e_value_block != 0) 217 goto bad_block; 218 size = le32_to_cpu(entry->e_value_size); 219 if (size > inode->i_sb->s_blocksize || 220 le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize) 221 goto bad_block; 222 223 if (ext2_xattr_cache_insert(bh)) 224 ea_idebug(inode, "cache insert failed"); 225 if (buffer) { 226 error = -ERANGE; 227 if (size > buffer_size) 228 goto cleanup; 229 /* return value of attribute */ 230 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs), 231 size); 232 } 233 error = size; 234 235cleanup: 236 brelse(bh); 237 up_read(&EXT2_I(inode)->xattr_sem); 238 239 return error; 240} 241 242/* 243 * ext2_xattr_list() 244 * 245 * Copy a list of attribute names into the buffer 246 * provided, or compute the buffer size required. 247 * Buffer is NULL to compute the size of the buffer required. 248 * 249 * Returns a negative error number on failure, or the number of bytes 250 * used / required on success. 251 */ 252static int 253ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size) 254{ 255 struct inode *inode = dentry->d_inode; 256 struct buffer_head *bh = NULL; 257 struct ext2_xattr_entry *entry; 258 char *end; 259 size_t rest = buffer_size; 260 int error; 261 262 ea_idebug(inode, "buffer=%p, buffer_size=%ld", 263 buffer, (long)buffer_size); 264 265 down_read(&EXT2_I(inode)->xattr_sem); 266 error = 0; 267 if (!EXT2_I(inode)->i_file_acl) 268 goto cleanup; 269 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl); 270 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 271 error = -EIO; 272 if (!bh) 273 goto cleanup; 274 ea_bdebug(bh, "b_count=%d, refcount=%d", 275 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount)); 276 end = bh->b_data + bh->b_size; 277 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 278 HDR(bh)->h_blocks != cpu_to_le32(1)) { 279bad_block: ext2_error(inode->i_sb, "ext2_xattr_list", 280 "inode %ld: bad block %d", inode->i_ino, 281 EXT2_I(inode)->i_file_acl); 282 error = -EIO; 283 goto cleanup; 284 } 285 286 /* check the on-disk data structure */ 287 entry = FIRST_ENTRY(bh); 288 while (!IS_LAST_ENTRY(entry)) { 289 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry); 290 291 if ((char *)next >= end) 292 goto bad_block; 293 entry = next; 294 } 295 if (ext2_xattr_cache_insert(bh)) 296 ea_idebug(inode, "cache insert failed"); 297 298 /* list the attribute names */ 299 for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry); 300 entry = EXT2_XATTR_NEXT(entry)) { 301 const struct xattr_handler *handler = 302 ext2_xattr_handler(entry->e_name_index); 303 304 if (handler) { 305 size_t size = handler->list(dentry, buffer, rest, 306 entry->e_name, 307 entry->e_name_len, 308 handler->flags); 309 if (buffer) { 310 if (size > rest) { 311 error = -ERANGE; 312 goto cleanup; 313 } 314 buffer += size; 315 } 316 rest -= size; 317 } 318 } 319 error = buffer_size - rest; /* total size */ 320 321cleanup: 322 brelse(bh); 323 up_read(&EXT2_I(inode)->xattr_sem); 324 325 return error; 326} 327 328/* 329 * Inode operation listxattr() 330 * 331 * dentry->d_inode->i_mutex: don't care 332 */ 333ssize_t 334ext2_listxattr(struct dentry *dentry, char *buffer, size_t size) 335{ 336 return ext2_xattr_list(dentry, buffer, size); 337} 338 339/* 340 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is 341 * not set, set it. 342 */ 343static void ext2_xattr_update_super_block(struct super_block *sb) 344{ 345 if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR)) 346 return; 347 348 spin_lock(&EXT2_SB(sb)->s_lock); 349 EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR); 350 spin_unlock(&EXT2_SB(sb)->s_lock); 351 sb->s_dirt = 1; 352 mark_buffer_dirty(EXT2_SB(sb)->s_sbh); 353} 354 355/* 356 * ext2_xattr_set() 357 * 358 * Create, replace or remove an extended attribute for this inode. Buffer 359 * is NULL to remove an existing extended attribute, and non-NULL to 360 * either replace an existing extended attribute, or create a new extended 361 * attribute. The flags XATTR_REPLACE and XATTR_CREATE 362 * specify that an extended attribute must exist and must not exist 363 * previous to the call, respectively. 364 * 365 * Returns 0, or a negative error number on failure. 366 */ 367int 368ext2_xattr_set(struct inode *inode, int name_index, const char *name, 369 const void *value, size_t value_len, int flags) 370{ 371 struct super_block *sb = inode->i_sb; 372 struct buffer_head *bh = NULL; 373 struct ext2_xattr_header *header = NULL; 374 struct ext2_xattr_entry *here, *last; 375 size_t name_len, free, min_offs = sb->s_blocksize; 376 int not_found = 1, error; 377 char *end; 378 379 /* 380 * header -- Points either into bh, or to a temporarily 381 * allocated buffer. 382 * here -- The named entry found, or the place for inserting, within 383 * the block pointed to by header. 384 * last -- Points right after the last named entry within the block 385 * pointed to by header. 386 * min_offs -- The offset of the first value (values are aligned 387 * towards the end of the block). 388 * end -- Points right after the block pointed to by header. 389 */ 390 391 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld", 392 name_index, name, value, (long)value_len); 393 394 if (value == NULL) 395 value_len = 0; 396 if (name == NULL) 397 return -EINVAL; 398 name_len = strlen(name); 399 if (name_len > 255 || value_len > sb->s_blocksize) 400 return -ERANGE; 401 down_write(&EXT2_I(inode)->xattr_sem); 402 if (EXT2_I(inode)->i_file_acl) { 403 /* The inode already has an extended attribute block. */ 404 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl); 405 error = -EIO; 406 if (!bh) 407 goto cleanup; 408 ea_bdebug(bh, "b_count=%d, refcount=%d", 409 atomic_read(&(bh->b_count)), 410 le32_to_cpu(HDR(bh)->h_refcount)); 411 header = HDR(bh); 412 end = bh->b_data + bh->b_size; 413 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 414 header->h_blocks != cpu_to_le32(1)) { 415bad_block: ext2_error(sb, "ext2_xattr_set", 416 "inode %ld: bad block %d", inode->i_ino, 417 EXT2_I(inode)->i_file_acl); 418 error = -EIO; 419 goto cleanup; 420 } 421 /* Find the named attribute. */ 422 here = FIRST_ENTRY(bh); 423 while (!IS_LAST_ENTRY(here)) { 424 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here); 425 if ((char *)next >= end) 426 goto bad_block; 427 if (!here->e_value_block && here->e_value_size) { 428 size_t offs = le16_to_cpu(here->e_value_offs); 429 if (offs < min_offs) 430 min_offs = offs; 431 } 432 not_found = name_index - here->e_name_index; 433 if (!not_found) 434 not_found = name_len - here->e_name_len; 435 if (!not_found) 436 not_found = memcmp(name, here->e_name,name_len); 437 if (not_found <= 0) 438 break; 439 here = next; 440 } 441 last = here; 442 /* We still need to compute min_offs and last. */ 443 while (!IS_LAST_ENTRY(last)) { 444 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last); 445 if ((char *)next >= end) 446 goto bad_block; 447 if (!last->e_value_block && last->e_value_size) { 448 size_t offs = le16_to_cpu(last->e_value_offs); 449 if (offs < min_offs) 450 min_offs = offs; 451 } 452 last = next; 453 } 454 455 /* Check whether we have enough space left. */ 456 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32); 457 } else { 458 /* We will use a new extended attribute block. */ 459 free = sb->s_blocksize - 460 sizeof(struct ext2_xattr_header) - sizeof(__u32); 461 here = last = NULL; /* avoid gcc uninitialized warning. */ 462 } 463 464 if (not_found) { 465 /* Request to remove a nonexistent attribute? */ 466 error = -ENODATA; 467 if (flags & XATTR_REPLACE) 468 goto cleanup; 469 error = 0; 470 if (value == NULL) 471 goto cleanup; 472 } else { 473 /* Request to create an existing attribute? */ 474 error = -EEXIST; 475 if (flags & XATTR_CREATE) 476 goto cleanup; 477 if (!here->e_value_block && here->e_value_size) { 478 size_t size = le32_to_cpu(here->e_value_size); 479 480 if (le16_to_cpu(here->e_value_offs) + size > 481 sb->s_blocksize || size > sb->s_blocksize) 482 goto bad_block; 483 free += EXT2_XATTR_SIZE(size); 484 } 485 free += EXT2_XATTR_LEN(name_len); 486 } 487 error = -ENOSPC; 488 if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len)) 489 goto cleanup; 490 491 /* Here we know that we can set the new attribute. */ 492 493 if (header) { 494 struct mb_cache_entry *ce; 495 496 /* assert(header == HDR(bh)); */ 497 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, 498 bh->b_blocknr); 499 lock_buffer(bh); 500 if (header->h_refcount == cpu_to_le32(1)) { 501 ea_bdebug(bh, "modifying in-place"); 502 if (ce) 503 mb_cache_entry_free(ce); 504 /* keep the buffer locked while modifying it. */ 505 } else { 506 int offset; 507 508 if (ce) 509 mb_cache_entry_release(ce); 510 unlock_buffer(bh); 511 ea_bdebug(bh, "cloning"); 512 header = kmalloc(bh->b_size, GFP_KERNEL); 513 error = -ENOMEM; 514 if (header == NULL) 515 goto cleanup; 516 memcpy(header, HDR(bh), bh->b_size); 517 header->h_refcount = cpu_to_le32(1); 518 519 offset = (char *)here - bh->b_data; 520 here = ENTRY((char *)header + offset); 521 offset = (char *)last - bh->b_data; 522 last = ENTRY((char *)header + offset); 523 } 524 } else { 525 /* Allocate a buffer where we construct the new block. */ 526 header = kzalloc(sb->s_blocksize, GFP_KERNEL); 527 error = -ENOMEM; 528 if (header == NULL) 529 goto cleanup; 530 end = (char *)header + sb->s_blocksize; 531 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC); 532 header->h_blocks = header->h_refcount = cpu_to_le32(1); 533 last = here = ENTRY(header+1); 534 } 535 536 /* Iff we are modifying the block in-place, bh is locked here. */ 537 538 if (not_found) { 539 /* Insert the new name. */ 540 size_t size = EXT2_XATTR_LEN(name_len); 541 size_t rest = (char *)last - (char *)here; 542 memmove((char *)here + size, here, rest); 543 memset(here, 0, size); 544 here->e_name_index = name_index; 545 here->e_name_len = name_len; 546 memcpy(here->e_name, name, name_len); 547 } else { 548 if (!here->e_value_block && here->e_value_size) { 549 char *first_val = (char *)header + min_offs; 550 size_t offs = le16_to_cpu(here->e_value_offs); 551 char *val = (char *)header + offs; 552 size_t size = EXT2_XATTR_SIZE( 553 le32_to_cpu(here->e_value_size)); 554 555 if (size == EXT2_XATTR_SIZE(value_len)) { 556 /* The old and the new value have the same 557 size. Just replace. */ 558 here->e_value_size = cpu_to_le32(value_len); 559 memset(val + size - EXT2_XATTR_PAD, 0, 560 EXT2_XATTR_PAD); /* Clear pad bytes. */ 561 memcpy(val, value, value_len); 562 goto skip_replace; 563 } 564 565 /* Remove the old value. */ 566 memmove(first_val + size, first_val, val - first_val); 567 memset(first_val, 0, size); 568 here->e_value_offs = 0; 569 min_offs += size; 570 571 /* Adjust all value offsets. */ 572 last = ENTRY(header+1); 573 while (!IS_LAST_ENTRY(last)) { 574 size_t o = le16_to_cpu(last->e_value_offs); 575 if (!last->e_value_block && o < offs) 576 last->e_value_offs = 577 cpu_to_le16(o + size); 578 last = EXT2_XATTR_NEXT(last); 579 } 580 } 581 if (value == NULL) { 582 /* Remove the old name. */ 583 size_t size = EXT2_XATTR_LEN(name_len); 584 last = ENTRY((char *)last - size); 585 memmove(here, (char*)here + size, 586 (char*)last - (char*)here); 587 memset(last, 0, size); 588 } 589 } 590 591 if (value != NULL) { 592 /* Insert the new value. */ 593 here->e_value_size = cpu_to_le32(value_len); 594 if (value_len) { 595 size_t size = EXT2_XATTR_SIZE(value_len); 596 char *val = (char *)header + min_offs - size; 597 here->e_value_offs = 598 cpu_to_le16((char *)val - (char *)header); 599 memset(val + size - EXT2_XATTR_PAD, 0, 600 EXT2_XATTR_PAD); /* Clear the pad bytes. */ 601 memcpy(val, value, value_len); 602 } 603 } 604 605skip_replace: 606 if (IS_LAST_ENTRY(ENTRY(header+1))) { 607 /* This block is now empty. */ 608 if (bh && header == HDR(bh)) 609 unlock_buffer(bh); /* we were modifying in-place. */ 610 error = ext2_xattr_set2(inode, bh, NULL); 611 } else { 612 ext2_xattr_rehash(header, here); 613 if (bh && header == HDR(bh)) 614 unlock_buffer(bh); /* we were modifying in-place. */ 615 error = ext2_xattr_set2(inode, bh, header); 616 } 617 618cleanup: 619 brelse(bh); 620 if (!(bh && header == HDR(bh))) 621 kfree(header); 622 up_write(&EXT2_I(inode)->xattr_sem); 623 624 return error; 625} 626 627/* 628 * Second half of ext2_xattr_set(): Update the file system. 629 */ 630static int 631ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh, 632 struct ext2_xattr_header *header) 633{ 634 struct super_block *sb = inode->i_sb; 635 struct buffer_head *new_bh = NULL; 636 int error; 637 638 if (header) { 639 new_bh = ext2_xattr_cache_find(inode, header); 640 if (new_bh) { 641 /* We found an identical block in the cache. */ 642 if (new_bh == old_bh) { 643 ea_bdebug(new_bh, "keeping this block"); 644 } else { 645 /* The old block is released after updating 646 the inode. */ 647 ea_bdebug(new_bh, "reusing block"); 648 649 error = dquot_alloc_block(inode, 1); 650 if (error) { 651 unlock_buffer(new_bh); 652 goto cleanup; 653 } 654 le32_add_cpu(&HDR(new_bh)->h_refcount, 1); 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 ext2_fsblk_t goal = ext2_group_first_block_no(sb, 668 EXT2_I(inode)->i_block_group); 669 int block = ext2_new_block(inode, goal, &error); 670 if (error) 671 goto cleanup; 672 ea_idebug(inode, "creating block %d", block); 673 674 new_bh = sb_getblk(sb, block); 675 if (!new_bh) { 676 ext2_free_blocks(inode, block, 1); 677 mark_inode_dirty(inode); 678 error = -EIO; 679 goto cleanup; 680 } 681 lock_buffer(new_bh); 682 memcpy(new_bh->b_data, header, new_bh->b_size); 683 set_buffer_uptodate(new_bh); 684 unlock_buffer(new_bh); 685 ext2_xattr_cache_insert(new_bh); 686 687 ext2_xattr_update_super_block(sb); 688 } 689 mark_buffer_dirty(new_bh); 690 if (IS_SYNC(inode)) { 691 sync_dirty_buffer(new_bh); 692 error = -EIO; 693 if (buffer_req(new_bh) && !buffer_uptodate(new_bh)) 694 goto cleanup; 695 } 696 } 697 698 /* Update the inode. */ 699 EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0; 700 inode->i_ctime = CURRENT_TIME_SEC; 701 if (IS_SYNC(inode)) { 702 error = ext2_sync_inode (inode); 703 /* In case sync failed due to ENOSPC the inode was actually 704 * written (only some dirty data were not) so we just proceed 705 * as if nothing happened and cleanup the unused block */ 706 if (error && error != -ENOSPC) { 707 if (new_bh && new_bh != old_bh) { 708 dquot_free_block_nodirty(inode, 1); 709 mark_inode_dirty(inode); 710 } 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 mark_inode_dirty(inode); 734 /* We let our caller release old_bh, so we 735 * need to duplicate the buffer before. */ 736 get_bh(old_bh); 737 bforget(old_bh); 738 } else { 739 /* Decrement the refcount only. */ 740 le32_add_cpu(&HDR(old_bh)->h_refcount, -1); 741 if (ce) 742 mb_cache_entry_release(ce); 743 dquot_free_block_nodirty(inode, 1); 744 mark_inode_dirty(inode); 745 mark_buffer_dirty(old_bh); 746 ea_bdebug(old_bh, "refcount now=%d", 747 le32_to_cpu(HDR(old_bh)->h_refcount)); 748 } 749 unlock_buffer(old_bh); 750 } 751 752cleanup: 753 brelse(new_bh); 754 755 return error; 756} 757 758/* 759 * ext2_xattr_delete_inode() 760 * 761 * Free extended attribute resources associated with this inode. This 762 * is called immediately before an inode is freed. 763 */ 764void 765ext2_xattr_delete_inode(struct inode *inode) 766{ 767 struct buffer_head *bh = NULL; 768 struct mb_cache_entry *ce; 769 770 down_write(&EXT2_I(inode)->xattr_sem); 771 if (!EXT2_I(inode)->i_file_acl) 772 goto cleanup; 773 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl); 774 if (!bh) { 775 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 776 "inode %ld: block %d read error", inode->i_ino, 777 EXT2_I(inode)->i_file_acl); 778 goto cleanup; 779 } 780 ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count))); 781 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) || 782 HDR(bh)->h_blocks != cpu_to_le32(1)) { 783 ext2_error(inode->i_sb, "ext2_xattr_delete_inode", 784 "inode %ld: bad block %d", inode->i_ino, 785 EXT2_I(inode)->i_file_acl); 786 goto cleanup; 787 } 788 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr); 789 lock_buffer(bh); 790 if (HDR(bh)->h_refcount == cpu_to_le32(1)) { 791 if (ce) 792 mb_cache_entry_free(ce); 793 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1); 794 get_bh(bh); 795 bforget(bh); 796 unlock_buffer(bh); 797 } else { 798 le32_add_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_nodirty(inode, 1); 808 } 809 EXT2_I(inode)->i_file_acl = 0; 810 811cleanup: 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 */ 821void 822ext2_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 */ 836static int 837ext2_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, GFP_NOFS); 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 */ 870static int 871ext2_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 */ 910static struct buffer_head * 911ext2_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); 919again: 920 ce = mb_cache_entry_find_first(ext2_xattr_cache, inode->i_sb->s_bdev, 921 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, 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 */ 966static 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 */ 1002static 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 1026int __init 1027init_ext2_xattr(void) 1028{ 1029 ext2_xattr_cache = mb_cache_create("ext2_xattr", 6); 1030 if (!ext2_xattr_cache) 1031 return -ENOMEM; 1032 return 0; 1033} 1034 1035void 1036exit_ext2_xattr(void) 1037{ 1038 mb_cache_destroy(ext2_xattr_cache); 1039} 1040