fs/reiserfs/xattr.c

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