1 /* 2 * linux/fs/reiserfs/xattr.c 3 * 4 * Copyright (c) 2002 by Jeff Mahoney, <jeffm@suse.com> 5 * 6 */ 7 8 /* 9 * In order to implement EA/ACLs in a clean, backwards compatible manner, 10 * they are implemented as files in a "private" directory. 11 * Each EA is in it's own file, with the directory layout like so (/ is assumed 12 * to be relative to fs root). Inside the /.reiserfs_priv/xattrs directory, 13 * directories named using the capital-hex form of the objectid and 14 * generation number are used. Inside each directory are individual files 15 * named with the name of the extended attribute. 16 * 17 * So, for objectid 12648430, we could have: 18 * /.reiserfs_priv/xattrs/C0FFEE.0/system.posix_acl_access 19 * /.reiserfs_priv/xattrs/C0FFEE.0/system.posix_acl_default 20 * /.reiserfs_priv/xattrs/C0FFEE.0/user.Content-Type 21 * .. or similar. 22 * 23 * The file contents are the text of the EA. The size is known based on the 24 * stat data describing the file. 25 * 26 * In the case of system.posix_acl_access and system.posix_acl_default, since 27 * these are special cases for filesystem ACLs, they are interpreted by the 28 * kernel, in addition, they are negatively and positively cached and attached 29 * to the inode so that unnecessary lookups are avoided. 30 * 31 * Locking works like so: 32 * Directory components (xattr root, xattr dir) are protectd by their i_mutex. 33 * The xattrs themselves are protected by the xattr_sem. 34 */ 35 36 #include <linux/reiserfs_fs.h> 37 #include <linux/capability.h> 38 #include <linux/dcache.h> 39 #include <linux/namei.h> 40 #include <linux/errno.h> 41 #include <linux/fs.h> 42 #include <linux/file.h> 43 #include <linux/pagemap.h> 44 #include <linux/xattr.h> 45 #include <linux/reiserfs_xattr.h> 46 #include <linux/reiserfs_acl.h> 47 #include <asm/uaccess.h> 48 #include <net/checksum.h> 49 #include <linux/smp_lock.h> 50 #include <linux/stat.h> 51 #include <linux/quotaops.h> 52 53 #define PRIVROOT_NAME ".reiserfs_priv" 54 #define XAROOT_NAME "xattrs" 55 56 static struct reiserfs_xattr_handler *find_xattr_handler_prefix(const char *); 57 58 /* Helpers for inode ops. We do this so that we don't have all the VFS 59 * overhead and also for proper i_mutex annotation. 60 * dir->i_mutex must be held for all of them. */ 61 static int xattr_create(struct inode *dir, struct dentry *dentry, int mode) 62 { 63 BUG_ON(!mutex_is_locked(&dir->i_mutex)); 64 DQUOT_INIT(dir); 65 return dir->i_op->create(dir, dentry, mode, NULL); 66 } 67 68 static int xattr_mkdir(struct inode *dir, struct dentry *dentry, int mode) 69 { 70 BUG_ON(!mutex_is_locked(&dir->i_mutex)); 71 DQUOT_INIT(dir); 72 return dir->i_op->mkdir(dir, dentry, mode); 73 } 74 75 /* We use I_MUTEX_CHILD here to silence lockdep. It's safe because xattr 76 * mutation ops aren't called during rename or splace, which are the 77 * only other users of I_MUTEX_CHILD. It violates the ordering, but that's 78 * better than allocating another subclass just for this code. */ 79 static int xattr_unlink(struct inode *dir, struct dentry *dentry) 80 { 81 int error; 82 BUG_ON(!mutex_is_locked(&dir->i_mutex)); 83 DQUOT_INIT(dir); 84 85 mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_CHILD); 86 error = dir->i_op->unlink(dir, dentry); 87 mutex_unlock(&dentry->d_inode->i_mutex); 88 89 if (!error) 90 d_delete(dentry); 91 return error; 92 } 93 94 static int xattr_rmdir(struct inode *dir, struct dentry *dentry) 95 { 96 int error; 97 BUG_ON(!mutex_is_locked(&dir->i_mutex)); 98 DQUOT_INIT(dir); 99 100 mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_CHILD); 101 dentry_unhash(dentry); 102 error = dir->i_op->rmdir(dir, dentry); 103 if (!error) 104 dentry->d_inode->i_flags |= S_DEAD; 105 mutex_unlock(&dentry->d_inode->i_mutex); 106 if (!error) 107 d_delete(dentry); 108 dput(dentry); 109 110 return error; 111 } 112 113 114 #define xattr_may_create(flags) (!flags || flags & XATTR_CREATE) 115 116 /* Returns and possibly creates the xattr dir. */ 117 static struct dentry *lookup_or_create_dir(struct dentry *parent, 118 const char *name, int flags) 119 { 120 struct dentry *dentry; 121 BUG_ON(!parent); 122 123 dentry = lookup_one_len(name, parent, strlen(name)); 124 if (IS_ERR(dentry)) 125 return dentry; 126 else if (!dentry->d_inode) { 127 int err = -ENODATA; 128 129 if (xattr_may_create(flags)) { 130 mutex_lock_nested(&parent->d_inode->i_mutex, 131 I_MUTEX_XATTR); 132 err = xattr_mkdir(parent->d_inode, dentry, 0700); 133 mutex_unlock(&parent->d_inode->i_mutex); 134 } 135 136 if (err) { 137 dput(dentry); 138 dentry = ERR_PTR(err); 139 } 140 } 141 142 return dentry; 143 } 144 145 static struct dentry *open_xa_root(struct super_block *sb, int flags) 146 { 147 struct dentry *privroot = REISERFS_SB(sb)->priv_root; 148 if (!privroot) 149 return ERR_PTR(-ENODATA); 150 return lookup_or_create_dir(privroot, XAROOT_NAME, flags); 151 } 152 153 static struct dentry *open_xa_dir(const struct inode *inode, int flags) 154 { 155 struct dentry *xaroot, *xadir; 156 char namebuf[17]; 157 158 xaroot = open_xa_root(inode->i_sb, flags); 159 if (IS_ERR(xaroot)) 160 return xaroot; 161 162 snprintf(namebuf, sizeof(namebuf), "%X.%X", 163 le32_to_cpu(INODE_PKEY(inode)->k_objectid), 164 inode->i_generation); 165 166 xadir = lookup_or_create_dir(xaroot, namebuf, flags); 167 dput(xaroot); 168 return xadir; 169 170 } 171 172 /* 173 * this is very similar to fs/reiserfs/dir.c:reiserfs_readdir, but 174 * we need to drop the path before calling the filldir struct. That 175 * would be a big performance hit to the non-xattr case, so I've copied 176 * the whole thing for now. --clm 177 * 178 * the big difference is that I go backwards through the directory, 179 * and don't mess with f->f_pos, but the idea is the same. Do some 180 * action on each and every entry in the directory. 181 * 182 * we're called with i_mutex held, so there are no worries about the directory 183 * changing underneath us. 184 */ 185 static int __xattr_readdir(struct inode *inode, void *dirent, filldir_t filldir) 186 { 187 struct cpu_key pos_key; /* key of current position in the directory (key of directory entry) */ 188 INITIALIZE_PATH(path_to_entry); 189 struct buffer_head *bh; 190 int entry_num; 191 struct item_head *ih, tmp_ih; 192 int search_res; 193 char *local_buf; 194 loff_t next_pos; 195 char small_buf[32]; /* avoid kmalloc if we can */ 196 struct reiserfs_de_head *deh; 197 int d_reclen; 198 char *d_name; 199 off_t d_off; 200 ino_t d_ino; 201 struct reiserfs_dir_entry de; 202 203 /* form key for search the next directory entry using f_pos field of 204 file structure */ 205 next_pos = max_reiserfs_offset(inode); 206 207 while (1) { 208 research: 209 if (next_pos <= DOT_DOT_OFFSET) 210 break; 211 make_cpu_key(&pos_key, inode, next_pos, TYPE_DIRENTRY, 3); 212 213 search_res = 214 search_by_entry_key(inode->i_sb, &pos_key, &path_to_entry, 215 &de); 216 if (search_res == IO_ERROR) { 217 // FIXME: we could just skip part of directory which could 218 // not be read 219 pathrelse(&path_to_entry); 220 return -EIO; 221 } 222 223 if (search_res == NAME_NOT_FOUND) 224 de.de_entry_num--; 225 226 set_de_name_and_namelen(&de); 227 entry_num = de.de_entry_num; 228 deh = &(de.de_deh[entry_num]); 229 230 bh = de.de_bh; 231 ih = de.de_ih; 232 233 if (!is_direntry_le_ih(ih)) { 234 reiserfs_error(inode->i_sb, "jdm-20000", 235 "not direntry %h", ih); 236 break; 237 } 238 copy_item_head(&tmp_ih, ih); 239 240 /* we must have found item, that is item of this directory, */ 241 RFALSE(COMP_SHORT_KEYS(&(ih->ih_key), &pos_key), 242 "vs-9000: found item %h does not match to dir we readdir %K", 243 ih, &pos_key); 244 245 if (deh_offset(deh) <= DOT_DOT_OFFSET) { 246 break; 247 } 248 249 /* look for the previous entry in the directory */ 250 next_pos = deh_offset(deh) - 1; 251 252 if (!de_visible(deh)) 253 /* it is hidden entry */ 254 continue; 255 256 d_reclen = entry_length(bh, ih, entry_num); 257 d_name = B_I_DEH_ENTRY_FILE_NAME(bh, ih, deh); 258 d_off = deh_offset(deh); 259 d_ino = deh_objectid(deh); 260 261 if (!d_name[d_reclen - 1]) 262 d_reclen = strlen(d_name); 263 264 if (d_reclen > REISERFS_MAX_NAME(inode->i_sb->s_blocksize)) { 265 /* too big to send back to VFS */ 266 continue; 267 } 268 269 /* Ignore the .reiserfs_priv entry */ 270 if (reiserfs_xattrs(inode->i_sb) && 271 !old_format_only(inode->i_sb) && 272 deh_objectid(deh) == 273 le32_to_cpu(INODE_PKEY 274 (REISERFS_SB(inode->i_sb)->priv_root->d_inode)-> 275 k_objectid)) 276 continue; 277 278 if (d_reclen <= 32) { 279 local_buf = small_buf; 280 } else { 281 local_buf = kmalloc(d_reclen, GFP_NOFS); 282 if (!local_buf) { 283 pathrelse(&path_to_entry); 284 return -ENOMEM; 285 } 286 if (item_moved(&tmp_ih, &path_to_entry)) { 287 kfree(local_buf); 288 289 /* sigh, must retry. Do this same offset again */ 290 next_pos = d_off; 291 goto research; 292 } 293 } 294 295 // Note, that we copy name to user space via temporary 296 // buffer (local_buf) because filldir will block if 297 // user space buffer is swapped out. At that time 298 // entry can move to somewhere else 299 memcpy(local_buf, d_name, d_reclen); 300 301 /* the filldir function might need to start transactions, 302 * or do who knows what. Release the path now that we've 303 * copied all the important stuff out of the deh 304 */ 305 pathrelse(&path_to_entry); 306 307 if (filldir(dirent, local_buf, d_reclen, d_off, d_ino, 308 DT_UNKNOWN) < 0) { 309 if (local_buf != small_buf) { 310 kfree(local_buf); 311 } 312 goto end; 313 } 314 if (local_buf != small_buf) { 315 kfree(local_buf); 316 } 317 } /* while */ 318 319 end: 320 pathrelse(&path_to_entry); 321 return 0; 322 } 323 324 /* 325 * this could be done with dedicated readdir ops for the xattr files, 326 * but I want to get something working asap 327 * this is stolen from vfs_readdir 328 * 329 */ 330 static 331 int xattr_readdir(struct inode *inode, filldir_t filler, void *buf) 332 { 333 int res = -ENOENT; 334 if (!IS_DEADDIR(inode)) { 335 lock_kernel(); 336 res = __xattr_readdir(inode, buf, filler); 337 unlock_kernel(); 338 } 339 return res; 340 } 341 342 /* expects xadir->d_inode->i_mutex to be locked */ 343 static int 344 __reiserfs_xattr_del(struct dentry *xadir, const char *name, int namelen) 345 { 346 struct dentry *dentry; 347 struct inode *dir = xadir->d_inode; 348 int err = 0; 349 struct reiserfs_xattr_handler *xah; 350 351 dentry = lookup_one_len(name, xadir, namelen); 352 if (IS_ERR(dentry)) { 353 err = PTR_ERR(dentry); 354 goto out; 355 } else if (!dentry->d_inode) { 356 err = -ENODATA; 357 goto out_file; 358 } 359 360 /* Skip directories.. */ 361 if (S_ISDIR(dentry->d_inode->i_mode)) 362 goto out_file; 363 364 if (!IS_PRIVATE(dentry->d_inode)) { 365 reiserfs_error(dir->i_sb, "jdm-20003", 366 "OID %08x [%.*s/%.*s] doesn't have " 367 "priv flag set [parent is %sset].", 368 le32_to_cpu(INODE_PKEY(dentry->d_inode)-> 369 k_objectid), xadir->d_name.len, 370 xadir->d_name.name, namelen, name, 371 IS_PRIVATE(xadir->d_inode) ? "" : 372 "not "); 373 dput(dentry); 374 return -EIO; 375 } 376 377 /* Deletion pre-operation */ 378 xah = find_xattr_handler_prefix(name); 379 if (xah && xah->del) { 380 err = xah->del(dentry->d_inode, name); 381 if (err) 382 goto out; 383 } 384 385 err = xattr_unlink(dir, dentry); 386 387 out_file: 388 dput(dentry); 389 390 out: 391 return err; 392 } 393 394 /* The following are side effects of other operations that aren't explicitly 395 * modifying extended attributes. This includes operations such as permissions 396 * or ownership changes, object deletions, etc. */ 397 398 static int 399 reiserfs_delete_xattrs_filler(void *buf, const char *name, int namelen, 400 loff_t offset, u64 ino, unsigned int d_type) 401 { 402 struct dentry *xadir = (struct dentry *)buf; 403 404 return __reiserfs_xattr_del(xadir, name, namelen); 405 406 } 407 408 /* This is called w/ inode->i_mutex downed */ 409 int reiserfs_delete_xattrs(struct inode *inode) 410 { 411 int err = -ENODATA; 412 struct dentry *dir, *root; 413 struct reiserfs_transaction_handle th; 414 int blocks = JOURNAL_PER_BALANCE_CNT * 2 + 2 + 415 4 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb); 416 417 /* Skip out, an xattr has no xattrs associated with it */ 418 if (IS_PRIVATE(inode) || get_inode_sd_version(inode) == STAT_DATA_V1) 419 return 0; 420 421 dir = open_xa_dir(inode, XATTR_REPLACE); 422 if (IS_ERR(dir)) { 423 err = PTR_ERR(dir); 424 goto out; 425 } else if (!dir->d_inode) { 426 dput(dir); 427 goto out; 428 } 429 430 mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_XATTR); 431 err = xattr_readdir(dir->d_inode, reiserfs_delete_xattrs_filler, dir); 432 mutex_unlock(&dir->d_inode->i_mutex); 433 if (err) { 434 dput(dir); 435 goto out; 436 } 437 438 root = dget(dir->d_parent); 439 dput(dir); 440 441 /* We start a transaction here to avoid a ABBA situation 442 * between the xattr root's i_mutex and the journal lock. 443 * Inode creation will inherit an ACL, which requires a 444 * lookup. The lookup locks the xattr root i_mutex with a 445 * transaction open. Inode deletion takes teh xattr root 446 * i_mutex to delete the directory and then starts a 447 * transaction inside it. Boom. This doesn't incur much 448 * additional overhead since the reiserfs_rmdir transaction 449 * will just nest inside the outer transaction. */ 450 err = journal_begin(&th, inode->i_sb, blocks); 451 if (!err) { 452 int jerror; 453 mutex_lock_nested(&root->d_inode->i_mutex, I_MUTEX_XATTR); 454 err = xattr_rmdir(root->d_inode, dir); 455 jerror = journal_end(&th, inode->i_sb, blocks); 456 mutex_unlock(&root->d_inode->i_mutex); 457 err = jerror ?: err; 458 } 459 460 dput(root); 461 out: 462 if (!err) 463 REISERFS_I(inode)->i_flags = 464 REISERFS_I(inode)->i_flags & ~i_has_xattr_dir; 465 else 466 reiserfs_warning(inode->i_sb, "jdm-20004", 467 "Couldn't remove all xattrs (%d)\n", err); 468 return err; 469 } 470 471 struct reiserfs_chown_buf { 472 struct inode *inode; 473 struct dentry *xadir; 474 struct iattr *attrs; 475 }; 476 477 /* XXX: If there is a better way to do this, I'd love to hear about it */ 478 static int 479 reiserfs_chown_xattrs_filler(void *buf, const char *name, int namelen, 480 loff_t offset, u64 ino, unsigned int d_type) 481 { 482 struct reiserfs_chown_buf *chown_buf = (struct reiserfs_chown_buf *)buf; 483 struct dentry *xafile, *xadir = chown_buf->xadir; 484 struct iattr *attrs = chown_buf->attrs; 485 int err = 0; 486 487 xafile = lookup_one_len(name, xadir, namelen); 488 if (IS_ERR(xafile)) 489 return PTR_ERR(xafile); 490 else if (!xafile->d_inode) { 491 dput(xafile); 492 return -ENODATA; 493 } 494 495 if (!S_ISDIR(xafile->d_inode->i_mode)) { 496 mutex_lock_nested(&xafile->d_inode->i_mutex, I_MUTEX_CHILD); 497 err = reiserfs_setattr(xafile, attrs); 498 mutex_unlock(&xafile->d_inode->i_mutex); 499 } 500 dput(xafile); 501 502 return err; 503 } 504 505 int reiserfs_chown_xattrs(struct inode *inode, struct iattr *attrs) 506 { 507 struct dentry *dir; 508 int err = 0; 509 struct reiserfs_chown_buf buf; 510 unsigned int ia_valid = attrs->ia_valid; 511 512 /* Skip out, an xattr has no xattrs associated with it */ 513 if (IS_PRIVATE(inode) || get_inode_sd_version(inode) == STAT_DATA_V1) 514 return 0; 515 516 dir = open_xa_dir(inode, XATTR_REPLACE); 517 if (IS_ERR(dir)) { 518 if (PTR_ERR(dir) != -ENODATA) 519 err = PTR_ERR(dir); 520 goto out; 521 } else if (!dir->d_inode) 522 goto out_dir; 523 524 attrs->ia_valid &= (ATTR_UID | ATTR_GID | ATTR_CTIME); 525 buf.xadir = dir; 526 buf.attrs = attrs; 527 buf.inode = inode; 528 529 mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_XATTR); 530 err = xattr_readdir(dir->d_inode, reiserfs_chown_xattrs_filler, &buf); 531 532 if (!err) 533 err = reiserfs_setattr(dir, attrs); 534 mutex_unlock(&dir->d_inode->i_mutex); 535 536 attrs->ia_valid = ia_valid; 537 out_dir: 538 dput(dir); 539 out: 540 if (err) 541 reiserfs_warning(inode->i_sb, "jdm-20007", 542 "Couldn't chown all xattrs (%d)\n", err); 543 return err; 544 } 545 546 #ifdef CONFIG_REISERFS_FS_XATTR 547 static struct reiserfs_xattr_handler *find_xattr_handler_prefix(const char 548 *prefix); 549 550 /* Returns a dentry corresponding to a specific extended attribute file 551 * for the inode. If flags allow, the file is created. Otherwise, a 552 * valid or negative dentry, or an error is returned. */ 553 static struct dentry *get_xa_file_dentry(const struct inode *inode, 554 const char *name, int flags) 555 { 556 struct dentry *xadir, *xafile; 557 int err = 0; 558 559 xadir = open_xa_dir(inode, flags); 560 if (IS_ERR(xadir)) 561 return ERR_CAST(xadir); 562 563 xafile = lookup_one_len(name, xadir, strlen(name)); 564 if (IS_ERR(xafile)) { 565 err = PTR_ERR(xafile); 566 goto out; 567 } 568 569 if (xafile->d_inode && (flags & XATTR_CREATE)) 570 err = -EEXIST; 571 572 if (!xafile->d_inode) { 573 err = -ENODATA; 574 if (xattr_may_create(flags)) { 575 mutex_lock_nested(&xadir->d_inode->i_mutex, 576 I_MUTEX_XATTR); 577 err = xattr_create(xadir->d_inode, xafile, 578 0700|S_IFREG); 579 mutex_unlock(&xadir->d_inode->i_mutex); 580 } 581 } 582 583 if (err) 584 dput(xafile); 585 out: 586 dput(xadir); 587 if (err) 588 return ERR_PTR(err); 589 return xafile; 590 } 591 592 /* Internal operations on file data */ 593 static inline void reiserfs_put_page(struct page *page) 594 { 595 kunmap(page); 596 page_cache_release(page); 597 } 598 599 static struct page *reiserfs_get_page(struct inode *dir, size_t n) 600 { 601 struct address_space *mapping = dir->i_mapping; 602 struct page *page; 603 /* We can deadlock if we try to free dentries, 604 and an unlink/rmdir has just occured - GFP_NOFS avoids this */ 605 mapping_set_gfp_mask(mapping, GFP_NOFS); 606 page = read_mapping_page(mapping, n >> PAGE_CACHE_SHIFT, NULL); 607 if (!IS_ERR(page)) { 608 kmap(page); 609 if (PageError(page)) 610 goto fail; 611 } 612 return page; 613 614 fail: 615 reiserfs_put_page(page); 616 return ERR_PTR(-EIO); 617 } 618 619 static inline __u32 xattr_hash(const char *msg, int len) 620 { 621 return csum_partial(msg, len, 0); 622 } 623 624 int reiserfs_commit_write(struct file *f, struct page *page, 625 unsigned from, unsigned to); 626 int reiserfs_prepare_write(struct file *f, struct page *page, 627 unsigned from, unsigned to); 628 629 630 /* Generic extended attribute operations that can be used by xa plugins */ 631 632 /* 633 * inode->i_mutex: down 634 */ 635 int 636 reiserfs_xattr_set(struct inode *inode, const char *name, const void *buffer, 637 size_t buffer_size, int flags) 638 { 639 int err = 0; 640 struct dentry *dentry; 641 struct page *page; 642 char *data; 643 size_t file_pos = 0; 644 size_t buffer_pos = 0; 645 struct iattr newattrs; 646 __u32 xahash = 0; 647 648 if (get_inode_sd_version(inode) == STAT_DATA_V1) 649 return -EOPNOTSUPP; 650 651 if (!buffer) 652 return reiserfs_xattr_del(inode, name); 653 654 dentry = get_xa_file_dentry(inode, name, flags); 655 if (IS_ERR(dentry)) { 656 err = PTR_ERR(dentry); 657 goto out; 658 } 659 660 down_write(&REISERFS_I(inode)->i_xattr_sem); 661 662 xahash = xattr_hash(buffer, buffer_size); 663 REISERFS_I(inode)->i_flags |= i_has_xattr_dir; 664 665 /* Resize it so we're ok to write there */ 666 newattrs.ia_size = buffer_size; 667 newattrs.ia_ctime = current_fs_time(inode->i_sb); 668 newattrs.ia_valid = ATTR_SIZE | ATTR_CTIME; 669 mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_XATTR); 670 down_write(&dentry->d_inode->i_alloc_sem); 671 err = reiserfs_setattr(dentry, &newattrs); 672 up_write(&dentry->d_inode->i_alloc_sem); 673 mutex_unlock(&dentry->d_inode->i_mutex); 674 if (err) 675 goto out_filp; 676 677 while (buffer_pos < buffer_size || buffer_pos == 0) { 678 size_t chunk; 679 size_t skip = 0; 680 size_t page_offset = (file_pos & (PAGE_CACHE_SIZE - 1)); 681 if (buffer_size - buffer_pos > PAGE_CACHE_SIZE) 682 chunk = PAGE_CACHE_SIZE; 683 else 684 chunk = buffer_size - buffer_pos; 685 686 page = reiserfs_get_page(dentry->d_inode, file_pos); 687 if (IS_ERR(page)) { 688 err = PTR_ERR(page); 689 goto out_filp; 690 } 691 692 lock_page(page); 693 data = page_address(page); 694 695 if (file_pos == 0) { 696 struct reiserfs_xattr_header *rxh; 697 skip = file_pos = sizeof(struct reiserfs_xattr_header); 698 if (chunk + skip > PAGE_CACHE_SIZE) 699 chunk = PAGE_CACHE_SIZE - skip; 700 rxh = (struct reiserfs_xattr_header *)data; 701 rxh->h_magic = cpu_to_le32(REISERFS_XATTR_MAGIC); 702 rxh->h_hash = cpu_to_le32(xahash); 703 } 704 705 err = reiserfs_prepare_write(NULL, page, page_offset, 706 page_offset + chunk + skip); 707 if (!err) { 708 if (buffer) 709 memcpy(data + skip, buffer + buffer_pos, chunk); 710 err = reiserfs_commit_write(NULL, page, page_offset, 711 page_offset + chunk + 712 skip); 713 } 714 unlock_page(page); 715 reiserfs_put_page(page); 716 buffer_pos += chunk; 717 file_pos += chunk; 718 skip = 0; 719 if (err || buffer_size == 0 || !buffer) 720 break; 721 } 722 723 /* We can't mark the inode dirty if it's not hashed. This is the case 724 * when we're inheriting the default ACL. If we dirty it, the inode 725 * gets marked dirty, but won't (ever) make it onto the dirty list until 726 * it's synced explicitly to clear I_DIRTY. This is bad. */ 727 if (!hlist_unhashed(&inode->i_hash)) { 728 inode->i_ctime = CURRENT_TIME_SEC; 729 mark_inode_dirty(inode); 730 } 731 732 out_filp: 733 up_write(&REISERFS_I(inode)->i_xattr_sem); 734 dput(dentry); 735 736 out: 737 return err; 738 } 739 740 /* 741 * inode->i_mutex: down 742 */ 743 int 744 reiserfs_xattr_get(const struct inode *inode, const char *name, void *buffer, 745 size_t buffer_size) 746 { 747 ssize_t err = 0; 748 struct dentry *dentry; 749 size_t isize; 750 size_t file_pos = 0; 751 size_t buffer_pos = 0; 752 struct page *page; 753 __u32 hash = 0; 754 755 if (name == NULL) 756 return -EINVAL; 757 758 /* We can't have xattrs attached to v1 items since they don't have 759 * generation numbers */ 760 if (get_inode_sd_version(inode) == STAT_DATA_V1) 761 return -EOPNOTSUPP; 762 763 dentry = get_xa_file_dentry(inode, name, XATTR_REPLACE); 764 if (IS_ERR(dentry)) { 765 err = PTR_ERR(dentry); 766 goto out; 767 } 768 769 down_read(&REISERFS_I(inode)->i_xattr_sem); 770 771 isize = i_size_read(dentry->d_inode); 772 REISERFS_I(inode)->i_flags |= i_has_xattr_dir; 773 774 /* Just return the size needed */ 775 if (buffer == NULL) { 776 err = isize - sizeof(struct reiserfs_xattr_header); 777 goto out_unlock; 778 } 779 780 if (buffer_size < isize - sizeof(struct reiserfs_xattr_header)) { 781 err = -ERANGE; 782 goto out_unlock; 783 } 784 785 while (file_pos < isize) { 786 size_t chunk; 787 char *data; 788 size_t skip = 0; 789 if (isize - file_pos > PAGE_CACHE_SIZE) 790 chunk = PAGE_CACHE_SIZE; 791 else 792 chunk = isize - file_pos; 793 794 page = reiserfs_get_page(dentry->d_inode, file_pos); 795 if (IS_ERR(page)) { 796 err = PTR_ERR(page); 797 goto out_unlock; 798 } 799 800 lock_page(page); 801 data = page_address(page); 802 if (file_pos == 0) { 803 struct reiserfs_xattr_header *rxh = 804 (struct reiserfs_xattr_header *)data; 805 skip = file_pos = sizeof(struct reiserfs_xattr_header); 806 chunk -= skip; 807 /* Magic doesn't match up.. */ 808 if (rxh->h_magic != cpu_to_le32(REISERFS_XATTR_MAGIC)) { 809 unlock_page(page); 810 reiserfs_put_page(page); 811 reiserfs_warning(inode->i_sb, "jdm-20001", 812 "Invalid magic for xattr (%s) " 813 "associated with %k", name, 814 INODE_PKEY(inode)); 815 err = -EIO; 816 goto out_unlock; 817 } 818 hash = le32_to_cpu(rxh->h_hash); 819 } 820 memcpy(buffer + buffer_pos, data + skip, chunk); 821 unlock_page(page); 822 reiserfs_put_page(page); 823 file_pos += chunk; 824 buffer_pos += chunk; 825 skip = 0; 826 } 827 err = isize - sizeof(struct reiserfs_xattr_header); 828 829 if (xattr_hash(buffer, isize - sizeof(struct reiserfs_xattr_header)) != 830 hash) { 831 reiserfs_warning(inode->i_sb, "jdm-20002", 832 "Invalid hash for xattr (%s) associated " 833 "with %k", name, INODE_PKEY(inode)); 834 err = -EIO; 835 } 836 837 out_unlock: 838 up_read(&REISERFS_I(inode)->i_xattr_sem); 839 dput(dentry); 840 841 out: 842 return err; 843 } 844 845 int reiserfs_xattr_del(struct inode *inode, const char *name) 846 { 847 struct dentry *dir; 848 int err; 849 850 dir = open_xa_dir(inode, XATTR_REPLACE); 851 if (IS_ERR(dir)) { 852 err = PTR_ERR(dir); 853 goto out; 854 } 855 856 mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_XATTR); 857 err = __reiserfs_xattr_del(dir, name, strlen(name)); 858 mutex_unlock(&dir->d_inode->i_mutex); 859 dput(dir); 860 861 if (!err) { 862 inode->i_ctime = CURRENT_TIME_SEC; 863 mark_inode_dirty(inode); 864 } 865 866 out: 867 return err; 868 } 869 870 /* Actual operations that are exported to VFS-land */ 871 /* 872 * Inode operation getxattr() 873 */ 874 ssize_t 875 reiserfs_getxattr(struct dentry * dentry, const char *name, void *buffer, 876 size_t size) 877 { 878 struct reiserfs_xattr_handler *xah = find_xattr_handler_prefix(name); 879 int err; 880 881 if (!xah || !reiserfs_xattrs(dentry->d_sb) || 882 get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1) 883 return -EOPNOTSUPP; 884 885 err = xah->get(dentry->d_inode, name, buffer, size); 886 return err; 887 } 888 889 /* 890 * Inode operation setxattr() 891 * 892 * dentry->d_inode->i_mutex down 893 */ 894 int 895 reiserfs_setxattr(struct dentry *dentry, const char *name, const void *value, 896 size_t size, int flags) 897 { 898 struct reiserfs_xattr_handler *xah = find_xattr_handler_prefix(name); 899 int err; 900 901 if (!xah || !reiserfs_xattrs(dentry->d_sb) || 902 get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1) 903 return -EOPNOTSUPP; 904 905 err = xah->set(dentry->d_inode, name, value, size, flags); 906 return err; 907 } 908 909 /* 910 * Inode operation removexattr() 911 * 912 * dentry->d_inode->i_mutex down 913 */ 914 int reiserfs_removexattr(struct dentry *dentry, const char *name) 915 { 916 int err; 917 struct reiserfs_xattr_handler *xah = find_xattr_handler_prefix(name); 918 919 if (!xah || !reiserfs_xattrs(dentry->d_sb) || 920 get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1) 921 return -EOPNOTSUPP; 922 923 err = reiserfs_xattr_del(dentry->d_inode, name); 924 925 dentry->d_inode->i_ctime = CURRENT_TIME_SEC; 926 mark_inode_dirty(dentry->d_inode); 927 928 return err; 929 } 930 931 /* This is what filldir will use: 932 * r_pos will always contain the amount of space required for the entire 933 * list. If r_pos becomes larger than r_size, we need more space and we 934 * return an error indicating this. If r_pos is less than r_size, then we've 935 * filled the buffer successfully and we return success */ 936 struct reiserfs_listxattr_buf { 937 int r_pos; 938 int r_size; 939 char *r_buf; 940 struct inode *r_inode; 941 }; 942 943 static int 944 reiserfs_listxattr_filler(void *buf, const char *name, int namelen, 945 loff_t offset, u64 ino, unsigned int d_type) 946 { 947 struct reiserfs_listxattr_buf *b = (struct reiserfs_listxattr_buf *)buf; 948 int len = 0; 949 if (name[0] != '.' 950 || (namelen != 1 && (name[1] != '.' || namelen != 2))) { 951 struct reiserfs_xattr_handler *xah = 952 find_xattr_handler_prefix(name); 953 if (!xah) 954 return 0; /* Unsupported xattr name, skip it */ 955 956 /* We call ->list() twice because the operation isn't required to just 957 * return the name back - we want to make sure we have enough space */ 958 len += xah->list(b->r_inode, name, namelen, NULL); 959 960 if (len) { 961 if (b->r_pos + len + 1 <= b->r_size) { 962 char *p = b->r_buf + b->r_pos; 963 p += xah->list(b->r_inode, name, namelen, p); 964 *p++ = '\0'; 965 } 966 b->r_pos += len + 1; 967 } 968 } 969 970 return 0; 971 } 972 973 /* 974 * Inode operation listxattr() 975 */ 976 ssize_t reiserfs_listxattr(struct dentry * dentry, char *buffer, size_t size) 977 { 978 struct dentry *dir; 979 int err = 0; 980 struct reiserfs_listxattr_buf buf; 981 982 if (!dentry->d_inode) 983 return -EINVAL; 984 985 if (!reiserfs_xattrs(dentry->d_sb) || 986 get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1) 987 return -EOPNOTSUPP; 988 989 dir = open_xa_dir(dentry->d_inode, XATTR_REPLACE); 990 if (IS_ERR(dir)) { 991 err = PTR_ERR(dir); 992 if (err == -ENODATA) 993 err = 0; /* Not an error if there aren't any xattrs */ 994 goto out; 995 } 996 997 buf.r_buf = buffer; 998 buf.r_size = buffer ? size : 0; 999 buf.r_pos = 0; 1000 buf.r_inode = dentry->d_inode; 1001 1002 REISERFS_I(dentry->d_inode)->i_flags |= i_has_xattr_dir; 1003 1004 mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_XATTR); 1005 err = xattr_readdir(dir->d_inode, reiserfs_listxattr_filler, &buf); 1006 mutex_unlock(&dir->d_inode->i_mutex); 1007 1008 if (!err) { 1009 if (buf.r_pos > buf.r_size && buffer != NULL) 1010 err = -ERANGE; 1011 else 1012 err = buf.r_pos; 1013 } 1014 1015 dput(dir); 1016 out: 1017 return err; 1018 } 1019 1020 /* This is the implementation for the xattr plugin infrastructure */ 1021 static LIST_HEAD(xattr_handlers); 1022 static DEFINE_RWLOCK(handler_lock); 1023 1024 static struct reiserfs_xattr_handler *find_xattr_handler_prefix(const char 1025 *prefix) 1026 { 1027 struct reiserfs_xattr_handler *xah = NULL; 1028 struct list_head *p; 1029 1030 read_lock(&handler_lock); 1031 list_for_each(p, &xattr_handlers) { 1032 xah = list_entry(p, struct reiserfs_xattr_handler, handlers); 1033 if (strncmp(xah->prefix, prefix, strlen(xah->prefix)) == 0) 1034 break; 1035 xah = NULL; 1036 } 1037 1038 read_unlock(&handler_lock); 1039 return xah; 1040 } 1041 1042 static void __unregister_handlers(void) 1043 { 1044 struct reiserfs_xattr_handler *xah; 1045 struct list_head *p, *tmp; 1046 1047 list_for_each_safe(p, tmp, &xattr_handlers) { 1048 xah = list_entry(p, struct reiserfs_xattr_handler, handlers); 1049 if (xah->exit) 1050 xah->exit(); 1051 1052 list_del_init(p); 1053 } 1054 INIT_LIST_HEAD(&xattr_handlers); 1055 } 1056 1057 int __init reiserfs_xattr_register_handlers(void) 1058 { 1059 int err = 0; 1060 struct reiserfs_xattr_handler *xah; 1061 struct list_head *p; 1062 1063 write_lock(&handler_lock); 1064 1065 /* If we're already initialized, nothing to do */ 1066 if (!list_empty(&xattr_handlers)) { 1067 write_unlock(&handler_lock); 1068 return 0; 1069 } 1070 1071 /* Add the handlers */ 1072 list_add_tail(&user_handler.handlers, &xattr_handlers); 1073 list_add_tail(&trusted_handler.handlers, &xattr_handlers); 1074 #ifdef CONFIG_REISERFS_FS_SECURITY 1075 list_add_tail(&security_handler.handlers, &xattr_handlers); 1076 #endif 1077 #ifdef CONFIG_REISERFS_FS_POSIX_ACL 1078 list_add_tail(&posix_acl_access_handler.handlers, &xattr_handlers); 1079 list_add_tail(&posix_acl_default_handler.handlers, &xattr_handlers); 1080 #endif 1081 1082 /* Run initializers, if available */ 1083 list_for_each(p, &xattr_handlers) { 1084 xah = list_entry(p, struct reiserfs_xattr_handler, handlers); 1085 if (xah->init) { 1086 err = xah->init(); 1087 if (err) { 1088 list_del_init(p); 1089 break; 1090 } 1091 } 1092 } 1093 1094 /* Clean up other handlers, if any failed */ 1095 if (err) 1096 __unregister_handlers(); 1097 1098 write_unlock(&handler_lock); 1099 return err; 1100 } 1101 1102 void reiserfs_xattr_unregister_handlers(void) 1103 { 1104 write_lock(&handler_lock); 1105 __unregister_handlers(); 1106 write_unlock(&handler_lock); 1107 } 1108 1109 static int reiserfs_check_acl(struct inode *inode, int mask) 1110 { 1111 struct posix_acl *acl; 1112 int error = -EAGAIN; /* do regular unix permission checks by default */ 1113 1114 acl = reiserfs_get_acl(inode, ACL_TYPE_ACCESS); 1115 1116 if (acl) { 1117 if (!IS_ERR(acl)) { 1118 error = posix_acl_permission(inode, acl, mask); 1119 posix_acl_release(acl); 1120 } else if (PTR_ERR(acl) != -ENODATA) 1121 error = PTR_ERR(acl); 1122 } 1123 1124 return error; 1125 } 1126 1127 int reiserfs_permission(struct inode *inode, int mask) 1128 { 1129 /* 1130 * We don't do permission checks on the internal objects. 1131 * Permissions are determined by the "owning" object. 1132 */ 1133 if (IS_PRIVATE(inode)) 1134 return 0; 1135 /* 1136 * Stat data v1 doesn't support ACLs. 1137 */ 1138 if (get_inode_sd_version(inode) == STAT_DATA_V1) 1139 return generic_permission(inode, mask, NULL); 1140 else 1141 return generic_permission(inode, mask, reiserfs_check_acl); 1142 } 1143 1144 static int create_privroot(struct dentry *dentry) 1145 { 1146 int err; 1147 struct inode *inode = dentry->d_parent->d_inode; 1148 mutex_lock_nested(&inode->i_mutex, I_MUTEX_XATTR); 1149 err = xattr_mkdir(inode, dentry, 0700); 1150 mutex_unlock(&inode->i_mutex); 1151 if (err) { 1152 dput(dentry); 1153 dentry = NULL; 1154 } 1155 1156 if (dentry && dentry->d_inode) 1157 reiserfs_info(dentry->d_sb, "Created %s - reserved for xattr " 1158 "storage.\n", PRIVROOT_NAME); 1159 1160 return err; 1161 } 1162 1163 static int xattr_mount_check(struct super_block *s) 1164 { 1165 /* We need generation numbers to ensure that the oid mapping is correct 1166 * v3.5 filesystems don't have them. */ 1167 if (!old_format_only(s)) { 1168 set_bit(REISERFS_XATTRS, &(REISERFS_SB(s)->s_mount_opt)); 1169 } else if (reiserfs_xattrs_optional(s)) { 1170 /* Old format filesystem, but optional xattrs have been enabled 1171 * at mount time. Error out. */ 1172 reiserfs_warning(s, "jdm-20005", 1173 "xattrs/ACLs not supported on pre v3.6 " 1174 "format filesystem. Failing mount."); 1175 return -EOPNOTSUPP; 1176 } else { 1177 /* Old format filesystem, but no optional xattrs have 1178 * been enabled. This means we silently disable xattrs 1179 * on the filesystem. */ 1180 clear_bit(REISERFS_XATTRS, &(REISERFS_SB(s)->s_mount_opt)); 1181 } 1182 1183 return 0; 1184 } 1185 1186 #else 1187 int __init reiserfs_xattr_register_handlers(void) { return 0; } 1188 void reiserfs_xattr_unregister_handlers(void) {} 1189 #endif 1190 1191 /* This will catch lookups from the fs root to .reiserfs_priv */ 1192 static int 1193 xattr_lookup_poison(struct dentry *dentry, struct qstr *q1, struct qstr *name) 1194 { 1195 struct dentry *priv_root = REISERFS_SB(dentry->d_sb)->priv_root; 1196 if (name->len == priv_root->d_name.len && 1197 name->hash == priv_root->d_name.hash && 1198 !memcmp(name->name, priv_root->d_name.name, name->len)) { 1199 return -ENOENT; 1200 } else if (q1->len == name->len && 1201 !memcmp(q1->name, name->name, name->len)) 1202 return 0; 1203 return 1; 1204 } 1205 1206 static struct dentry_operations xattr_lookup_poison_ops = { 1207 .d_compare = xattr_lookup_poison, 1208 }; 1209 1210 /* We need to take a copy of the mount flags since things like 1211 * MS_RDONLY don't get set until *after* we're called. 1212 * mount_flags != mount_options */ 1213 int reiserfs_xattr_init(struct super_block *s, int mount_flags) 1214 { 1215 int err = 0; 1216 1217 #ifdef CONFIG_REISERFS_FS_XATTR 1218 err = xattr_mount_check(s); 1219 if (err) 1220 goto error; 1221 #endif 1222 1223 /* If we don't have the privroot located yet - go find it */ 1224 if (!REISERFS_SB(s)->priv_root) { 1225 struct dentry *dentry; 1226 dentry = lookup_one_len(PRIVROOT_NAME, s->s_root, 1227 strlen(PRIVROOT_NAME)); 1228 if (!IS_ERR(dentry)) { 1229 #ifdef CONFIG_REISERFS_FS_XATTR 1230 if (!(mount_flags & MS_RDONLY) && !dentry->d_inode) 1231 err = create_privroot(dentry); 1232 #endif 1233 if (!dentry->d_inode) { 1234 dput(dentry); 1235 dentry = NULL; 1236 } 1237 } else 1238 err = PTR_ERR(dentry); 1239 1240 if (!err && dentry) { 1241 s->s_root->d_op = &xattr_lookup_poison_ops; 1242 dentry->d_inode->i_flags |= S_PRIVATE; 1243 REISERFS_SB(s)->priv_root = dentry; 1244 #ifdef CONFIG_REISERFS_FS_XATTR 1245 /* xattrs are unavailable */ 1246 } else if (!(mount_flags & MS_RDONLY)) { 1247 /* If we're read-only it just means that the dir 1248 * hasn't been created. Not an error -- just no 1249 * xattrs on the fs. We'll check again if we 1250 * go read-write */ 1251 reiserfs_warning(s, "jdm-20006", 1252 "xattrs/ACLs enabled and couldn't " 1253 "find/create .reiserfs_priv. " 1254 "Failing mount."); 1255 err = -EOPNOTSUPP; 1256 #endif 1257 } 1258 } 1259 1260 #ifdef CONFIG_REISERFS_FS_XATTR 1261 error: 1262 if (err) { 1263 clear_bit(REISERFS_XATTRS, &(REISERFS_SB(s)->s_mount_opt)); 1264 clear_bit(REISERFS_XATTRS_USER, &(REISERFS_SB(s)->s_mount_opt)); 1265 clear_bit(REISERFS_POSIXACL, &(REISERFS_SB(s)->s_mount_opt)); 1266 } 1267 #endif 1268 1269 /* The super_block MS_POSIXACL must mirror the (no)acl mount option. */ 1270 s->s_flags = s->s_flags & ~MS_POSIXACL; 1271 #ifdef CONFIG_REISERFS_FS_POSIX_ACL 1272 if (reiserfs_posixacl(s)) 1273 s->s_flags |= MS_POSIXACL; 1274 #endif 1275 1276 return err; 1277 } 1278