fs/open.c

   1 /*
   2  *  linux/fs/open.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/string.h>
   8 #include <linux/mm.h>
   9 #include <linux/file.h>
  10 #include <linux/fdtable.h>
  11 #include <linux/quotaops.h>
  12 #include <linux/fsnotify.h>
  13 #include <linux/module.h>
  14 #include <linux/slab.h>
  15 #include <linux/tty.h>
  16 #include <linux/namei.h>
  17 #include <linux/backing-dev.h>
  18 #include <linux/capability.h>
  19 #include <linux/securebits.h>
  20 #include <linux/security.h>
  21 #include <linux/mount.h>
  22 #include <linux/vfs.h>
  23 #include <linux/fcntl.h>
  24 #include <asm/uaccess.h>
  25 #include <linux/fs.h>
  26 #include <linux/personality.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/rcupdate.h>
  30 #include <linux/audit.h>
  31 #include <linux/falloc.h>
  32
  33 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  34 {
  35         int retval = -ENODEV;
  36
  37         if (dentry) {
  38                 retval = -ENOSYS;
  39                 if (dentry->d_sb->s_op->statfs) {
  40                         memset(buf, 0, sizeof(*buf));
  41                         retval = security_sb_statfs(dentry);
  42                         if (retval)
  43                                 return retval;
  44                         retval = dentry->d_sb->s_op->statfs(dentry, buf);
  45                         if (retval == 0 && buf->f_frsize == 0)
  46                                 buf->f_frsize = buf->f_bsize;
  47                 }
  48         }
  49         return retval;
  50 }
  51
  52 EXPORT_SYMBOL(vfs_statfs);
  53
  54 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
  55 {
  56         struct kstatfs st;
  57         int retval;
  58
  59         retval = vfs_statfs(dentry, &st);
  60         if (retval)
  61                 return retval;
  62
  63         if (sizeof(*buf) == sizeof(st))
  64                 memcpy(buf, &st, sizeof(st));
  65         else {
  66                 if (sizeof buf->f_blocks == 4) {
  67                         if ((st.f_blocks | st.f_bfree | st.f_bavail) &
  68                             0xffffffff00000000ULL)
  69                                 return -EOVERFLOW;
  70                         /*
  71                          * f_files and f_ffree may be -1; it's okay to stuff
  72                          * that into 32 bits
  73                          */
  74                         if (st.f_files != -1 &&
  75                             (st.f_files & 0xffffffff00000000ULL))
  76                                 return -EOVERFLOW;
  77                         if (st.f_ffree != -1 &&
  78                             (st.f_ffree & 0xffffffff00000000ULL))
  79                                 return -EOVERFLOW;
  80                 }
  81
  82                 buf->f_type = st.f_type;
  83                 buf->f_bsize = st.f_bsize;
  84                 buf->f_blocks = st.f_blocks;
  85                 buf->f_bfree = st.f_bfree;
  86                 buf->f_bavail = st.f_bavail;
  87                 buf->f_files = st.f_files;
  88                 buf->f_ffree = st.f_ffree;
  89                 buf->f_fsid = st.f_fsid;
  90                 buf->f_namelen = st.f_namelen;
  91                 buf->f_frsize = st.f_frsize;
  92                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  93         }
  94         return 0;
  95 }
  96
  97 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
  98 {
  99         struct kstatfs st;
 100         int retval;
 101
 102         retval = vfs_statfs(dentry, &st);
 103         if (retval)
 104                 return retval;
 105
 106         if (sizeof(*buf) == sizeof(st))
 107                 memcpy(buf, &st, sizeof(st));
 108         else {
 109                 buf->f_type = st.f_type;
 110                 buf->f_bsize = st.f_bsize;
 111                 buf->f_blocks = st.f_blocks;
 112                 buf->f_bfree = st.f_bfree;
 113                 buf->f_bavail = st.f_bavail;
 114                 buf->f_files = st.f_files;
 115                 buf->f_ffree = st.f_ffree;
 116                 buf->f_fsid = st.f_fsid;
 117                 buf->f_namelen = st.f_namelen;
 118                 buf->f_frsize = st.f_frsize;
 119                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 120         }
 121         return 0;
 122 }
 123
 124 asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
 125 {
 126         struct nameidata nd;
 127         int error;
 128
 129         error = user_path_walk(path, &nd);
 130         if (!error) {
 131                 struct statfs tmp;
 132                 error = vfs_statfs_native(nd.path.dentry, &tmp);
 133                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 134                         error = -EFAULT;
 135                 path_put(&nd.path);
 136         }
 137         return error;
 138 }
 139
 140
 141 asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
 142 {
 143         struct nameidata nd;
 144         long error;
 145
 146         if (sz != sizeof(*buf))
 147                 return -EINVAL;
 148         error = user_path_walk(path, &nd);
 149         if (!error) {
 150                 struct statfs64 tmp;
 151                 error = vfs_statfs64(nd.path.dentry, &tmp);
 152                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 153                         error = -EFAULT;
 154                 path_put(&nd.path);
 155         }
 156         return error;
 157 }
 158
 159
 160 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
 161 {
 162         struct file * file;
 163         struct statfs tmp;
 164         int error;
 165
 166         error = -EBADF;
 167         file = fget(fd);
 168         if (!file)
 169                 goto out;
 170         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 171         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 172                 error = -EFAULT;
 173         fput(file);
 174 out:
 175         return error;
 176 }
 177
 178 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
 179 {
 180         struct file * file;
 181         struct statfs64 tmp;
 182         int error;
 183
 184         if (sz != sizeof(*buf))
 185                 return -EINVAL;
 186
 187         error = -EBADF;
 188         file = fget(fd);
 189         if (!file)
 190                 goto out;
 191         error = vfs_statfs64(file->f_path.dentry, &tmp);
 192         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 193                 error = -EFAULT;
 194         fput(file);
 195 out:
 196         return error;
 197 }
 198
 199 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 200         struct file *filp)
 201 {
 202         int err;
 203         struct iattr newattrs;
 204
 205         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 206         if (length < 0)
 207                 return -EINVAL;
 208
 209         newattrs.ia_size = length;
 210         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 211         if (filp) {
 212                 newattrs.ia_file = filp;
 213                 newattrs.ia_valid |= ATTR_FILE;
 214         }
 215
 216         /* Remove suid/sgid on truncate too */
 217         newattrs.ia_valid |= should_remove_suid(dentry);
 218
 219         mutex_lock(&dentry->d_inode->i_mutex);
 220         err = notify_change(dentry, &newattrs);
 221         mutex_unlock(&dentry->d_inode->i_mutex);
 222         return err;
 223 }
 224
 225 static long do_sys_truncate(const char __user * path, loff_t length)
 226 {
 227         struct nameidata nd;
 228         struct inode * inode;
 229         int error;
 230
 231         error = -EINVAL;
 232         if (length < 0) /* sorry, but loff_t says... */
 233                 goto out;
 234
 235         error = user_path_walk(path, &nd);
 236         if (error)
 237                 goto out;
 238         inode = nd.path.dentry->d_inode;
 239
 240         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 241         error = -EISDIR;
 242         if (S_ISDIR(inode->i_mode))
 243                 goto dput_and_out;
 244
 245         error = -EINVAL;
 246         if (!S_ISREG(inode->i_mode))
 247                 goto dput_and_out;
 248
 249         error = mnt_want_write(nd.path.mnt);
 250         if (error)
 251                 goto dput_and_out;
 252
 253         error = vfs_permission(&nd, MAY_WRITE);
 254         if (error)
 255                 goto mnt_drop_write_and_out;
 256
 257         error = -EPERM;
 258         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 259                 goto mnt_drop_write_and_out;
 260
 261         error = get_write_access(inode);
 262         if (error)
 263                 goto mnt_drop_write_and_out;
 264
 265         /*
 266          * Make sure that there are no leases.  get_write_access() protects
 267          * against the truncate racing with a lease-granting setlease().
 268          */
 269         error = break_lease(inode, FMODE_WRITE);
 270         if (error)
 271                 goto put_write_and_out;
 272
 273         error = locks_verify_truncate(inode, NULL, length);
 274         if (!error) {
 275                 DQUOT_INIT(inode);
 276                 error = do_truncate(nd.path.dentry, length, 0, NULL);
 277         }
 278
 279 put_write_and_out:
 280         put_write_access(inode);
 281 mnt_drop_write_and_out:
 282         mnt_drop_write(nd.path.mnt);
 283 dput_and_out:
 284         path_put(&nd.path);
 285 out:
 286         return error;
 287 }
 288
 289 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 290 {
 291         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 292         return do_sys_truncate(path, (long)length);
 293 }
 294
 295 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 296 {
 297         struct inode * inode;
 298         struct dentry *dentry;
 299         struct file * file;
 300         int error;
 301
 302         error = -EINVAL;
 303         if (length < 0)
 304                 goto out;
 305         error = -EBADF;
 306         file = fget(fd);
 307         if (!file)
 308                 goto out;
 309
 310         /* explicitly opened as large or we are on 64-bit box */
 311         if (file->f_flags & O_LARGEFILE)
 312                 small = 0;
 313
 314         dentry = file->f_path.dentry;
 315         inode = dentry->d_inode;
 316         error = -EINVAL;
 317         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 318                 goto out_putf;
 319
 320         error = -EINVAL;
 321         /* Cannot ftruncate over 2^31 bytes without large file support */
 322         if (small && length > MAX_NON_LFS)
 323                 goto out_putf;
 324
 325         error = -EPERM;
 326         if (IS_APPEND(inode))
 327                 goto out_putf;
 328
 329         error = locks_verify_truncate(inode, file, length);
 330         if (!error)
 331                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 332 out_putf:
 333         fput(file);
 334 out:
 335         return error;
 336 }
 337
 338 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 339 {
 340         long ret = do_sys_ftruncate(fd, length, 1);
 341         /* avoid REGPARM breakage on x86: */
 342         asmlinkage_protect(2, ret, fd, length);
 343         return ret;
 344 }
 345
 346 /* LFS versions of truncate are only needed on 32 bit machines */
 347 #if BITS_PER_LONG == 32
 348 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 349 {
 350         return do_sys_truncate(path, length);
 351 }
 352
 353 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 354 {
 355         long ret = do_sys_ftruncate(fd, length, 0);
 356         /* avoid REGPARM breakage on x86: */
 357         asmlinkage_protect(2, ret, fd, length);
 358         return ret;
 359 }
 360 #endif
 361
 362 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 363 {
 364         struct file *file;
 365         struct inode *inode;
 366         long ret = -EINVAL;
 367
 368         if (offset < 0 || len <= 0)
 369                 goto out;
 370
 371         /* Return error if mode is not supported */
 372         ret = -EOPNOTSUPP;
 373         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 374                 goto out;
 375
 376         ret = -EBADF;
 377         file = fget(fd);
 378         if (!file)
 379                 goto out;
 380         if (!(file->f_mode & FMODE_WRITE))
 381                 goto out_fput;
 382         /*
 383          * Revalidate the write permissions, in case security policy has
 384          * changed since the files were opened.
 385          */
 386         ret = security_file_permission(file, MAY_WRITE);
 387         if (ret)
 388                 goto out_fput;
 389
 390         inode = file->f_path.dentry->d_inode;
 391
 392         ret = -ESPIPE;
 393         if (S_ISFIFO(inode->i_mode))
 394                 goto out_fput;
 395
 396         ret = -ENODEV;
 397         /*
 398          * Let individual file system decide if it supports preallocation
 399          * for directories or not.
 400          */
 401         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 402                 goto out_fput;
 403
 404         ret = -EFBIG;
 405         /* Check for wrap through zero too */
 406         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 407                 goto out_fput;
 408
 409         if (inode->i_op && inode->i_op->fallocate)
 410                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 411         else
 412                 ret = -EOPNOTSUPP;
 413
 414 out_fput:
 415         fput(file);
 416 out:
 417         return ret;
 418 }
 419
 420 /*
 421  * access() needs to use the real uid/gid, not the effective uid/gid.
 422  * We do this by temporarily clearing all FS-related capabilities and
 423  * switching the fsuid/fsgid around to the real ones.
 424  */
 425 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 426 {
 427         struct nameidata nd;
 428         int old_fsuid, old_fsgid;
 429         kernel_cap_t uninitialized_var(old_cap);  /* !SECURE_NO_SETUID_FIXUP */
 430         int res;
 431
 432         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 433                 return -EINVAL;
 434
 435         old_fsuid = current->fsuid;
 436         old_fsgid = current->fsgid;
 437
 438         current->fsuid = current->uid;
 439         current->fsgid = current->gid;
 440
 441         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 442                 /*
 443                  * Clear the capabilities if we switch to a non-root user
 444                  */
 445 #ifndef CONFIG_SECURITY_FILE_CAPABILITIES
 446                 /*
 447                  * FIXME: There is a race here against sys_capset.  The
 448                  * capabilities can change yet we will restore the old
 449                  * value below.  We should hold task_capabilities_lock,
 450                  * but we cannot because user_path_walk can sleep.
 451                  */
 452 #endif /* ndef CONFIG_SECURITY_FILE_CAPABILITIES */
 453                 if (current->uid)
 454                         old_cap = cap_set_effective(__cap_empty_set);
 455                 else
 456                         old_cap = cap_set_effective(current->cap_permitted);
 457         }
 458
 459         res = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 460         if (res)
 461                 goto out;
 462
 463         res = vfs_permission(&nd, mode);
 464         /* SuS v2 requires we report a read only fs too */
 465         if(res || !(mode & S_IWOTH) ||
 466            special_file(nd.path.dentry->d_inode->i_mode))
 467                 goto out_path_release;
 468         /*
 469          * This is a rare case where using __mnt_is_readonly()
 470          * is OK without a mnt_want/drop_write() pair.  Since
 471          * no actual write to the fs is performed here, we do
 472          * not need to telegraph to that to anyone.
 473          *
 474          * By doing this, we accept that this access is
 475          * inherently racy and know that the fs may change
 476          * state before we even see this result.
 477          */
 478         if (__mnt_is_readonly(nd.path.mnt))
 479                 res = -EROFS;
 480
 481 out_path_release:
 482         path_put(&nd.path);
 483 out:
 484         current->fsuid = old_fsuid;
 485         current->fsgid = old_fsgid;
 486
 487         if (!issecure(SECURE_NO_SETUID_FIXUP))
 488                 cap_set_effective(old_cap);
 489
 490         return res;
 491 }
 492
 493 asmlinkage long sys_access(const char __user *filename, int mode)
 494 {
 495         return sys_faccessat(AT_FDCWD, filename, mode);
 496 }
 497
 498 asmlinkage long sys_chdir(const char __user * filename)
 499 {
 500         struct nameidata nd;
 501         int error;
 502
 503         error = __user_walk(filename,
 504                             LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 505         if (error)
 506                 goto out;
 507
 508         error = vfs_permission(&nd, MAY_EXEC);
 509         if (error)
 510                 goto dput_and_out;
 511
 512         set_fs_pwd(current->fs, &nd.path);
 513
 514 dput_and_out:
 515         path_put(&nd.path);
 516 out:
 517         return error;
 518 }
 519
 520 asmlinkage long sys_fchdir(unsigned int fd)
 521 {
 522         struct file *file;
 523         struct inode *inode;
 524         int error;
 525
 526         error = -EBADF;
 527         file = fget(fd);
 528         if (!file)
 529                 goto out;
 530
 531         inode = file->f_path.dentry->d_inode;
 532
 533         error = -ENOTDIR;
 534         if (!S_ISDIR(inode->i_mode))
 535                 goto out_putf;
 536
 537         error = file_permission(file, MAY_EXEC);
 538         if (!error)
 539                 set_fs_pwd(current->fs, &file->f_path);
 540 out_putf:
 541         fput(file);
 542 out:
 543         return error;
 544 }
 545
 546 asmlinkage long sys_chroot(const char __user * filename)
 547 {
 548         struct nameidata nd;
 549         int error;
 550
 551         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 552         if (error)
 553                 goto out;
 554
 555         error = vfs_permission(&nd, MAY_EXEC);
 556         if (error)
 557                 goto dput_and_out;
 558
 559         error = -EPERM;
 560         if (!capable(CAP_SYS_CHROOT))
 561                 goto dput_and_out;
 562
 563         set_fs_root(current->fs, &nd.path);
 564         set_fs_altroot();
 565         error = 0;
 566 dput_and_out:
 567         path_put(&nd.path);
 568 out:
 569         return error;
 570 }
 571
 572 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 573 {
 574         struct inode * inode;
 575         struct dentry * dentry;
 576         struct file * file;
 577         int err = -EBADF;
 578         struct iattr newattrs;
 579
 580         file = fget(fd);
 581         if (!file)
 582                 goto out;
 583
 584         dentry = file->f_path.dentry;
 585         inode = dentry->d_inode;
 586
 587         audit_inode(NULL, dentry);
 588
 589         err = mnt_want_write(file->f_path.mnt);
 590         if (err)
 591                 goto out_putf;
 592         err = -EPERM;
 593         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 594                 goto out_drop_write;
 595         mutex_lock(&inode->i_mutex);
 596         if (mode == (mode_t) -1)
 597                 mode = inode->i_mode;
 598         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 599         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 600         err = notify_change(dentry, &newattrs);
 601         mutex_unlock(&inode->i_mutex);
 602
 603 out_drop_write:
 604         mnt_drop_write(file->f_path.mnt);
 605 out_putf:
 606         fput(file);
 607 out:
 608         return err;
 609 }
 610
 611 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 612                              mode_t mode)
 613 {
 614         struct nameidata nd;
 615         struct inode * inode;
 616         int error;
 617         struct iattr newattrs;
 618
 619         error = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
 620         if (error)
 621                 goto out;
 622         inode = nd.path.dentry->d_inode;
 623
 624         error = mnt_want_write(nd.path.mnt);
 625         if (error)
 626                 goto dput_and_out;
 627
 628         error = -EPERM;
 629         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 630                 goto out_drop_write;
 631
 632         mutex_lock(&inode->i_mutex);
 633         if (mode == (mode_t) -1)
 634                 mode = inode->i_mode;
 635         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 636         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 637         error = notify_change(nd.path.dentry, &newattrs);
 638         mutex_unlock(&inode->i_mutex);
 639
 640 out_drop_write:
 641         mnt_drop_write(nd.path.mnt);
 642 dput_and_out:
 643         path_put(&nd.path);
 644 out:
 645         return error;
 646 }
 647
 648 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 649 {
 650         return sys_fchmodat(AT_FDCWD, filename, mode);
 651 }
 652
 653 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 654 {
 655         struct inode * inode;
 656         int error;
 657         struct iattr newattrs;
 658
 659         error = -ENOENT;
 660         if (!(inode = dentry->d_inode)) {
 661                 printk(KERN_ERR "chown_common: NULL inode\n");
 662                 goto out;
 663         }
 664         error = -EPERM;
 665         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 666                 goto out;
 667         newattrs.ia_valid =  ATTR_CTIME;
 668         if (user != (uid_t) -1) {
 669                 newattrs.ia_valid |= ATTR_UID;
 670                 newattrs.ia_uid = user;
 671         }
 672         if (group != (gid_t) -1) {
 673                 newattrs.ia_valid |= ATTR_GID;
 674                 newattrs.ia_gid = group;
 675         }
 676         if (!S_ISDIR(inode->i_mode))
 677                 newattrs.ia_valid |=
 678                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 679         mutex_lock(&inode->i_mutex);
 680         error = notify_change(dentry, &newattrs);
 681         mutex_unlock(&inode->i_mutex);
 682 out:
 683         return error;
 684 }
 685
 686 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 687 {
 688         struct nameidata nd;
 689         int error;
 690
 691         error = user_path_walk(filename, &nd);
 692         if (error)
 693                 goto out;
 694         error = mnt_want_write(nd.path.mnt);
 695         if (error)
 696                 goto out_release;
 697         error = chown_common(nd.path.dentry, user, group);
 698         mnt_drop_write(nd.path.mnt);
 699 out_release:
 700         path_put(&nd.path);
 701 out:
 702         return error;
 703 }
 704
 705 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 706                              gid_t group, int flag)
 707 {
 708         struct nameidata nd;
 709         int error = -EINVAL;
 710         int follow;
 711
 712         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 713                 goto out;
 714
 715         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 716         error = __user_walk_fd(dfd, filename, follow, &nd);
 717         if (error)
 718                 goto out;
 719         error = mnt_want_write(nd.path.mnt);
 720         if (error)
 721                 goto out_release;
 722         error = chown_common(nd.path.dentry, user, group);
 723         mnt_drop_write(nd.path.mnt);
 724 out_release:
 725         path_put(&nd.path);
 726 out:
 727         return error;
 728 }
 729
 730 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 731 {
 732         struct nameidata nd;
 733         int error;
 734
 735         error = user_path_walk_link(filename, &nd);
 736         if (error)
 737                 goto out;
 738         error = mnt_want_write(nd.path.mnt);
 739         if (error)
 740                 goto out_release;
 741         error = chown_common(nd.path.dentry, user, group);
 742         mnt_drop_write(nd.path.mnt);
 743 out_release:
 744         path_put(&nd.path);
 745 out:
 746         return error;
 747 }
 748
 749
 750 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 751 {
 752         struct file * file;
 753         int error = -EBADF;
 754         struct dentry * dentry;
 755
 756         file = fget(fd);
 757         if (!file)
 758                 goto out;
 759
 760         error = mnt_want_write(file->f_path.mnt);
 761         if (error)
 762                 goto out_fput;
 763         dentry = file->f_path.dentry;
 764         audit_inode(NULL, dentry);
 765         error = chown_common(dentry, user, group);
 766         mnt_drop_write(file->f_path.mnt);
 767 out_fput:
 768         fput(file);
 769 out:
 770         return error;
 771 }
 772
 773 /*
 774  * You have to be very careful that these write
 775  * counts get cleaned up in error cases and
 776  * upon __fput().  This should probably never
 777  * be called outside of __dentry_open().
 778  */
 779 static inline int __get_file_write_access(struct inode *inode,
 780                                           struct vfsmount *mnt)
 781 {
 782         int error;
 783         error = get_write_access(inode);
 784         if (error)
 785                 return error;
 786         /*
 787          * Do not take mount writer counts on
 788          * special files since no writes to
 789          * the mount itself will occur.
 790          */
 791         if (!special_file(inode->i_mode)) {
 792                 /*
 793                  * Balanced in __fput()
 794                  */
 795                 error = mnt_want_write(mnt);
 796                 if (error)
 797                         put_write_access(inode);
 798         }
 799         return error;
 800 }
 801
 802 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 803                                         int flags, struct file *f,
 804                                         int (*open)(struct inode *, struct file *))
 805 {
 806         struct inode *inode;
 807         int error;
 808
 809         f->f_flags = flags;
 810         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 811                                 FMODE_PREAD | FMODE_PWRITE;
 812         inode = dentry->d_inode;
 813         if (f->f_mode & FMODE_WRITE) {
 814                 error = __get_file_write_access(inode, mnt);
 815                 if (error)
 816                         goto cleanup_file;
 817                 if (!special_file(inode->i_mode))
 818                         file_take_write(f);
 819         }
 820
 821         f->f_mapping = inode->i_mapping;
 822         f->f_path.dentry = dentry;
 823         f->f_path.mnt = mnt;
 824         f->f_pos = 0;
 825         f->f_op = fops_get(inode->i_fop);
 826         file_move(f, &inode->i_sb->s_files);
 827
 828         error = security_dentry_open(f);
 829         if (error)
 830                 goto cleanup_all;
 831
 832         if (!open && f->f_op)
 833                 open = f->f_op->open;
 834         if (open) {
 835                 error = open(inode, f);
 836                 if (error)
 837                         goto cleanup_all;
 838         }
 839
 840         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 841
 842         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 843
 844         /* NB: we're sure to have correct a_ops only after f_op->open */
 845         if (f->f_flags & O_DIRECT) {
 846                 if (!f->f_mapping->a_ops ||
 847                     ((!f->f_mapping->a_ops->direct_IO) &&
 848                     (!f->f_mapping->a_ops->get_xip_mem))) {
 849                         fput(f);
 850                         f = ERR_PTR(-EINVAL);
 851                 }
 852         }
 853
 854         return f;
 855
 856 cleanup_all:
 857         fops_put(f->f_op);
 858         if (f->f_mode & FMODE_WRITE) {
 859                 put_write_access(inode);
 860                 if (!special_file(inode->i_mode)) {
 861                         /*
 862                          * We don't consider this a real
 863                          * mnt_want/drop_write() pair
 864                          * because it all happenend right
 865                          * here, so just reset the state.
 866                          */
 867                         file_reset_write(f);
 868                         mnt_drop_write(mnt);
 869                 }
 870         }
 871         file_kill(f);
 872         f->f_path.dentry = NULL;
 873         f->f_path.mnt = NULL;
 874 cleanup_file:
 875         put_filp(f);
 876         dput(dentry);
 877         mntput(mnt);
 878         return ERR_PTR(error);
 879 }
 880
 881 /**
 882  * lookup_instantiate_filp - instantiates the open intent filp
 883  * @nd: pointer to nameidata
 884  * @dentry: pointer to dentry
 885  * @open: open callback
 886  *
 887  * Helper for filesystems that want to use lookup open intents and pass back
 888  * a fully instantiated struct file to the caller.
 889  * This function is meant to be called from within a filesystem's
 890  * lookup method.
 891  * Beware of calling it for non-regular files! Those ->open methods might block
 892  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 893  * leading to a deadlock, as nobody can open that fifo anymore, because
 894  * another process to open fifo will block on locked parent when doing lookup).
 895  * Note that in case of error, nd->intent.open.file is destroyed, but the
 896  * path information remains valid.
 897  * If the open callback is set to NULL, then the standard f_op->open()
 898  * filesystem callback is substituted.
 899  */
 900 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 901                 int (*open)(struct inode *, struct file *))
 902 {
 903         if (IS_ERR(nd->intent.open.file))
 904                 goto out;
 905         if (IS_ERR(dentry))
 906                 goto out_err;
 907         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 908                                              nd->intent.open.flags - 1,
 909                                              nd->intent.open.file,
 910                                              open);
 911 out:
 912         return nd->intent.open.file;
 913 out_err:
 914         release_open_intent(nd);
 915         nd->intent.open.file = (struct file *)dentry;
 916         goto out;
 917 }
 918 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 919
 920 /**
 921  * nameidata_to_filp - convert a nameidata to an open filp.
 922  * @nd: pointer to nameidata
 923  * @flags: open flags
 924  *
 925  * Note that this function destroys the original nameidata
 926  */
 927 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 928 {
 929         struct file *filp;
 930
 931         /* Pick up the filp from the open intent */
 932         filp = nd->intent.open.file;
 933         /* Has the filesystem initialised the file for us? */
 934         if (filp->f_path.dentry == NULL)
 935                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 936                                      NULL);
 937         else
 938                 path_put(&nd->path);
 939         return filp;
 940 }
 941
 942 /*
 943  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 944  * error.
 945  */
 946 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 947 {
 948         int error;
 949         struct file *f;
 950
 951         /*
 952          * We must always pass in a valid mount pointer.   Historically
 953          * callers got away with not passing it, but we must enforce this at
 954          * the earliest possible point now to avoid strange problems deep in the
 955          * filesystem stack.
 956          */
 957         if (!mnt) {
 958                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 959                 dump_stack();
 960                 return ERR_PTR(-EINVAL);
 961         }
 962
 963         error = -ENFILE;
 964         f = get_empty_filp();
 965         if (f == NULL) {
 966                 dput(dentry);
 967                 mntput(mnt);
 968                 return ERR_PTR(error);
 969         }
 970
 971         return __dentry_open(dentry, mnt, flags, f, NULL);
 972 }
 973 EXPORT_SYMBOL(dentry_open);
 974
 975 /*
 976  * Find an empty file descriptor entry, and mark it busy.
 977  */
 978 int get_unused_fd_flags(int flags)
 979 {
 980         struct files_struct * files = current->files;
 981         int fd, error;
 982         struct fdtable *fdt;
 983
 984         error = -EMFILE;
 985         spin_lock(&files->file_lock);
 986
 987 repeat:
 988         fdt = files_fdtable(files);
 989         fd = find_next_zero_bit(fdt->open_fds->fds_bits, fdt->max_fds,
 990                                 files->next_fd);
 991
 992         /*
 993          * N.B. For clone tasks sharing a files structure, this test
 994          * will limit the total number of files that can be opened.
 995          */
 996         if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 997                 goto out;
 998
 999         /* Do we need to expand the fd array or fd set?  */
1000         error = expand_files(files, fd);
1001         if (error < 0)
1002                 goto out;
1003
1004         if (error) {
1005                 /*
1006                  * If we needed to expand the fs array we
1007                  * might have blocked - try again.
1008                  */
1009                 error = -EMFILE;
1010                 goto repeat;
1011         }
1012
1013         FD_SET(fd, fdt->open_fds);
1014         if (flags & O_CLOEXEC)
1015                 FD_SET(fd, fdt->close_on_exec);
1016         else
1017                 FD_CLR(fd, fdt->close_on_exec);
1018         files->next_fd = fd + 1;
1019 #if 1
1020         /* Sanity check */
1021         if (fdt->fd[fd] != NULL) {
1022                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
1023                 fdt->fd[fd] = NULL;
1024         }
1025 #endif
1026         error = fd;
1027
1028 out:
1029         spin_unlock(&files->file_lock);
1030         return error;
1031 }
1032
1033 int get_unused_fd(void)
1034 {
1035         return get_unused_fd_flags(0);
1036 }
1037
1038 EXPORT_SYMBOL(get_unused_fd);
1039
1040 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
1041 {
1042         struct fdtable *fdt = files_fdtable(files);
1043         __FD_CLR(fd, fdt->open_fds);
1044         if (fd < files->next_fd)
1045                 files->next_fd = fd;
1046 }
1047
1048 void put_unused_fd(unsigned int fd)
1049 {
1050         struct files_struct *files = current->files;
1051         spin_lock(&files->file_lock);
1052         __put_unused_fd(files, fd);
1053         spin_unlock(&files->file_lock);
1054 }
1055
1056 EXPORT_SYMBOL(put_unused_fd);
1057
1058 /*
1059  * Install a file pointer in the fd array.
1060  *
1061  * The VFS is full of places where we drop the files lock between
1062  * setting the open_fds bitmap and installing the file in the file
1063  * array.  At any such point, we are vulnerable to a dup2() race
1064  * installing a file in the array before us.  We need to detect this and
1065  * fput() the struct file we are about to overwrite in this case.
1066  *
1067  * It should never happen - if we allow dup2() do it, _really_ bad things
1068  * will follow.
1069  */
1070
1071 void fd_install(unsigned int fd, struct file *file)
1072 {
1073         struct files_struct *files = current->files;
1074         struct fdtable *fdt;
1075         spin_lock(&files->file_lock);
1076         fdt = files_fdtable(files);
1077         BUG_ON(fdt->fd[fd] != NULL);
1078         rcu_assign_pointer(fdt->fd[fd], file);
1079         spin_unlock(&files->file_lock);
1080 }
1081
1082 EXPORT_SYMBOL(fd_install);
1083
1084 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1085 {
1086         char *tmp = getname(filename);
1087         int fd = PTR_ERR(tmp);
1088
1089         if (!IS_ERR(tmp)) {
1090                 fd = get_unused_fd_flags(flags);
1091                 if (fd >= 0) {
1092                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1093                         if (IS_ERR(f)) {
1094                                 put_unused_fd(fd);
1095                                 fd = PTR_ERR(f);
1096                         } else {
1097                                 fsnotify_open(f->f_path.dentry);
1098                                 fd_install(fd, f);
1099                         }
1100                 }
1101                 putname(tmp);
1102         }
1103         return fd;
1104 }
1105
1106 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1107 {
1108         long ret;
1109
1110         if (force_o_largefile())
1111                 flags |= O_LARGEFILE;
1112
1113         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1114         /* avoid REGPARM breakage on x86: */
1115         asmlinkage_protect(3, ret, filename, flags, mode);
1116         return ret;
1117 }
1118
1119 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1120                            int mode)
1121 {
1122         long ret;
1123
1124         if (force_o_largefile())
1125                 flags |= O_LARGEFILE;
1126
1127         ret = do_sys_open(dfd, filename, flags, mode);
1128         /* avoid REGPARM breakage on x86: */
1129         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1130         return ret;
1131 }
1132
1133 #ifndef __alpha__
1134
1135 /*
1136  * For backward compatibility?  Maybe this should be moved
1137  * into arch/i386 instead?
1138  */
1139 asmlinkage long sys_creat(const char __user * pathname, int mode)
1140 {
1141         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1142 }
1143
1144 #endif
1145
1146 /*
1147  * "id" is the POSIX thread ID. We use the
1148  * files pointer for this..
1149  */
1150 int filp_close(struct file *filp, fl_owner_t id)
1151 {
1152         int retval = 0;
1153
1154         if (!file_count(filp)) {
1155                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1156                 return 0;
1157         }
1158
1159         if (filp->f_op && filp->f_op->flush)
1160                 retval = filp->f_op->flush(filp, id);
1161
1162         dnotify_flush(filp, id);
1163         locks_remove_posix(filp, id);
1164         fput(filp);
1165         return retval;
1166 }
1167
1168 EXPORT_SYMBOL(filp_close);
1169
1170 /*
1171  * Careful here! We test whether the file pointer is NULL before
1172  * releasing the fd. This ensures that one clone task can't release
1173  * an fd while another clone is opening it.
1174  */
1175 asmlinkage long sys_close(unsigned int fd)
1176 {
1177         struct file * filp;
1178         struct files_struct *files = current->files;
1179         struct fdtable *fdt;
1180         int retval;
1181
1182         spin_lock(&files->file_lock);
1183         fdt = files_fdtable(files);
1184         if (fd >= fdt->max_fds)
1185                 goto out_unlock;
1186         filp = fdt->fd[fd];
1187         if (!filp)
1188                 goto out_unlock;
1189         rcu_assign_pointer(fdt->fd[fd], NULL);
1190         FD_CLR(fd, fdt->close_on_exec);
1191         __put_unused_fd(files, fd);
1192         spin_unlock(&files->file_lock);
1193         retval = filp_close(filp, files);
1194
1195         /* can't restart close syscall because file table entry was cleared */
1196         if (unlikely(retval == -ERESTARTSYS ||
1197                      retval == -ERESTARTNOINTR ||
1198                      retval == -ERESTARTNOHAND ||
1199                      retval == -ERESTART_RESTARTBLOCK))
1200                 retval = -EINTR;
1201
1202         return retval;
1203
1204 out_unlock:
1205         spin_unlock(&files->file_lock);
1206         return -EBADF;
1207 }
1208
1209 EXPORT_SYMBOL(sys_close);
1210
1211 /*
1212  * This routine simulates a hangup on the tty, to arrange that users
1213  * are given clean terminals at login time.
1214  */
1215 asmlinkage long sys_vhangup(void)
1216 {
1217         if (capable(CAP_SYS_TTY_CONFIG)) {
1218                 /* XXX: this needs locking */
1219                 tty_vhangup(current->signal->tty);
1220                 return 0;
1221         }
1222         return -EPERM;
1223 }
1224
1225 /*
1226  * Called when an inode is about to be open.
1227  * We use this to disallow opening large files on 32bit systems if
1228  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1229  * on this flag in sys_open.
1230  */
1231 int generic_file_open(struct inode * inode, struct file * filp)
1232 {
1233         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1234                 return -EOVERFLOW;
1235         return 0;
1236 }
1237
1238 EXPORT_SYMBOL(generic_file_open);
1239
1240 /*
1241  * This is used by subsystems that don't want seekable
1242  * file descriptors
1243  */
1244 int nonseekable_open(struct inode *inode, struct file *filp)
1245 {
1246         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1247         return 0;
1248 }
1249
1250 EXPORT_SYMBOL(nonseekable_open);