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                              st.f_bsize | st.f_frsize) &
  69                             0xffffffff00000000ULL)
  70                                 return -EOVERFLOW;
  71                         /*
  72                          * f_files and f_ffree may be -1; it's okay to stuff
  73                          * that into 32 bits
  74                          */
  75                         if (st.f_files != -1 &&
  76                             (st.f_files & 0xffffffff00000000ULL))
  77                                 return -EOVERFLOW;
  78                         if (st.f_ffree != -1 &&
  79                             (st.f_ffree & 0xffffffff00000000ULL))
  80                                 return -EOVERFLOW;
  81                 }
  82
  83                 buf->f_type = st.f_type;
  84                 buf->f_bsize = st.f_bsize;
  85                 buf->f_blocks = st.f_blocks;
  86                 buf->f_bfree = st.f_bfree;
  87                 buf->f_bavail = st.f_bavail;
  88                 buf->f_files = st.f_files;
  89                 buf->f_ffree = st.f_ffree;
  90                 buf->f_fsid = st.f_fsid;
  91                 buf->f_namelen = st.f_namelen;
  92                 buf->f_frsize = st.f_frsize;
  93                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  94         }
  95         return 0;
  96 }
  97
  98 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
  99 {
 100         struct kstatfs st;
 101         int retval;
 102
 103         retval = vfs_statfs(dentry, &st);
 104         if (retval)
 105                 return retval;
 106
 107         if (sizeof(*buf) == sizeof(st))
 108                 memcpy(buf, &st, sizeof(st));
 109         else {
 110                 buf->f_type = st.f_type;
 111                 buf->f_bsize = st.f_bsize;
 112                 buf->f_blocks = st.f_blocks;
 113                 buf->f_bfree = st.f_bfree;
 114                 buf->f_bavail = st.f_bavail;
 115                 buf->f_files = st.f_files;
 116                 buf->f_ffree = st.f_ffree;
 117                 buf->f_fsid = st.f_fsid;
 118                 buf->f_namelen = st.f_namelen;
 119                 buf->f_frsize = st.f_frsize;
 120                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 121         }
 122         return 0;
 123 }
 124
 125 SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct statfs __user *, buf)
 126 {
 127         struct path path;
 128         int error;
 129
 130         error = user_path(pathname, &path);
 131         if (!error) {
 132                 struct statfs tmp;
 133                 error = vfs_statfs_native(path.dentry, &tmp);
 134                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 135                         error = -EFAULT;
 136                 path_put(&path);
 137         }
 138         return error;
 139 }
 140
 141 SYSCALL_DEFINE3(statfs64, const char __user *, pathname, size_t, sz, struct statfs64 __user *, buf)
 142 {
 143         struct path path;
 144         long error;
 145
 146         if (sz != sizeof(*buf))
 147                 return -EINVAL;
 148         error = user_path(pathname, &path);
 149         if (!error) {
 150                 struct statfs64 tmp;
 151                 error = vfs_statfs64(path.dentry, &tmp);
 152                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 153                         error = -EFAULT;
 154                 path_put(&path);
 155         }
 156         return error;
 157 }
 158
 159 SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct statfs __user *, buf)
 160 {
 161         struct file * file;
 162         struct statfs tmp;
 163         int error;
 164
 165         error = -EBADF;
 166         file = fget(fd);
 167         if (!file)
 168                 goto out;
 169         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 170         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 171                 error = -EFAULT;
 172         fput(file);
 173 out:
 174         return error;
 175 }
 176
 177 SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, size_t, sz, struct statfs64 __user *, buf)
 178 {
 179         struct file * file;
 180         struct statfs64 tmp;
 181         int error;
 182
 183         if (sz != sizeof(*buf))
 184                 return -EINVAL;
 185
 186         error = -EBADF;
 187         file = fget(fd);
 188         if (!file)
 189                 goto out;
 190         error = vfs_statfs64(file->f_path.dentry, &tmp);
 191         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 192                 error = -EFAULT;
 193         fput(file);
 194 out:
 195         return error;
 196 }
 197
 198 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 199         struct file *filp)
 200 {
 201         int err;
 202         struct iattr newattrs;
 203
 204         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 205         if (length < 0)
 206                 return -EINVAL;
 207
 208         newattrs.ia_size = length;
 209         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 210         if (filp) {
 211                 newattrs.ia_file = filp;
 212                 newattrs.ia_valid |= ATTR_FILE;
 213         }
 214
 215         /* Remove suid/sgid on truncate too */
 216         newattrs.ia_valid |= should_remove_suid(dentry);
 217
 218         mutex_lock(&dentry->d_inode->i_mutex);
 219         err = notify_change(dentry, &newattrs);
 220         mutex_unlock(&dentry->d_inode->i_mutex);
 221         return err;
 222 }
 223
 224 static long do_sys_truncate(const char __user *pathname, loff_t length)
 225 {
 226         struct path path;
 227         struct inode *inode;
 228         int error;
 229
 230         error = -EINVAL;
 231         if (length < 0) /* sorry, but loff_t says... */
 232                 goto out;
 233
 234         error = user_path(pathname, &path);
 235         if (error)
 236                 goto out;
 237         inode = path.dentry->d_inode;
 238
 239         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 240         error = -EISDIR;
 241         if (S_ISDIR(inode->i_mode))
 242                 goto dput_and_out;
 243
 244         error = -EINVAL;
 245         if (!S_ISREG(inode->i_mode))
 246                 goto dput_and_out;
 247
 248         error = mnt_want_write(path.mnt);
 249         if (error)
 250                 goto dput_and_out;
 251
 252         error = inode_permission(inode, MAY_WRITE);
 253         if (error)
 254                 goto mnt_drop_write_and_out;
 255
 256         error = -EPERM;
 257         if (IS_APPEND(inode))
 258                 goto mnt_drop_write_and_out;
 259
 260         error = get_write_access(inode);
 261         if (error)
 262                 goto mnt_drop_write_and_out;
 263
 264         /*
 265          * Make sure that there are no leases.  get_write_access() protects
 266          * against the truncate racing with a lease-granting setlease().
 267          */
 268         error = break_lease(inode, FMODE_WRITE);
 269         if (error)
 270                 goto put_write_and_out;
 271
 272         error = locks_verify_truncate(inode, NULL, length);
 273         if (!error)
 274                 error = security_path_truncate(&path, length, 0);
 275         if (!error) {
 276                 vfs_dq_init(inode);
 277                 error = do_truncate(path.dentry, length, 0, NULL);
 278         }
 279
 280 put_write_and_out:
 281         put_write_access(inode);
 282 mnt_drop_write_and_out:
 283         mnt_drop_write(path.mnt);
 284 dput_and_out:
 285         path_put(&path);
 286 out:
 287         return error;
 288 }
 289
 290 SYSCALL_DEFINE2(truncate, const char __user *, path, unsigned long, length)
 291 {
 292         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 293         return do_sys_truncate(path, (long)length);
 294 }
 295
 296 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 297 {
 298         struct inode * inode;
 299         struct dentry *dentry;
 300         struct file * file;
 301         int error;
 302
 303         error = -EINVAL;
 304         if (length < 0)
 305                 goto out;
 306         error = -EBADF;
 307         file = fget(fd);
 308         if (!file)
 309                 goto out;
 310
 311         /* explicitly opened as large or we are on 64-bit box */
 312         if (file->f_flags & O_LARGEFILE)
 313                 small = 0;
 314
 315         dentry = file->f_path.dentry;
 316         inode = dentry->d_inode;
 317         error = -EINVAL;
 318         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 319                 goto out_putf;
 320
 321         error = -EINVAL;
 322         /* Cannot ftruncate over 2^31 bytes without large file support */
 323         if (small && length > MAX_NON_LFS)
 324                 goto out_putf;
 325
 326         error = -EPERM;
 327         if (IS_APPEND(inode))
 328                 goto out_putf;
 329
 330         error = locks_verify_truncate(inode, file, length);
 331         if (!error)
 332                 error = security_path_truncate(&file->f_path, length,
 333                                                ATTR_MTIME|ATTR_CTIME);
 334         if (!error)
 335                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 336 out_putf:
 337         fput(file);
 338 out:
 339         return error;
 340 }
 341
 342 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 343 {
 344         long ret = do_sys_ftruncate(fd, length, 1);
 345         /* avoid REGPARM breakage on x86: */
 346         asmlinkage_protect(2, ret, fd, length);
 347         return ret;
 348 }
 349
 350 /* LFS versions of truncate are only needed on 32 bit machines */
 351 #if BITS_PER_LONG == 32
 352 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 353 {
 354         return do_sys_truncate(path, length);
 355 }
 356 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 357 asmlinkage long SyS_truncate64(long path, loff_t length)
 358 {
 359         return SYSC_truncate64((const char __user *) path, length);
 360 }
 361 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 362 #endif
 363
 364 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 365 {
 366         long ret = do_sys_ftruncate(fd, length, 0);
 367         /* avoid REGPARM breakage on x86: */
 368         asmlinkage_protect(2, ret, fd, length);
 369         return ret;
 370 }
 371 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 372 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 373 {
 374         return SYSC_ftruncate64((unsigned int) fd, length);
 375 }
 376 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 377 #endif
 378 #endif /* BITS_PER_LONG == 32 */
 379
 380 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 381 {
 382         struct file *file;
 383         struct inode *inode;
 384         long ret = -EINVAL;
 385
 386         if (offset < 0 || len <= 0)
 387                 goto out;
 388
 389         /* Return error if mode is not supported */
 390         ret = -EOPNOTSUPP;
 391         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 392                 goto out;
 393
 394         ret = -EBADF;
 395         file = fget(fd);
 396         if (!file)
 397                 goto out;
 398         if (!(file->f_mode & FMODE_WRITE))
 399                 goto out_fput;
 400         /*
 401          * Revalidate the write permissions, in case security policy has
 402          * changed since the files were opened.
 403          */
 404         ret = security_file_permission(file, MAY_WRITE);
 405         if (ret)
 406                 goto out_fput;
 407
 408         inode = file->f_path.dentry->d_inode;
 409
 410         ret = -ESPIPE;
 411         if (S_ISFIFO(inode->i_mode))
 412                 goto out_fput;
 413
 414         ret = -ENODEV;
 415         /*
 416          * Let individual file system decide if it supports preallocation
 417          * for directories or not.
 418          */
 419         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 420                 goto out_fput;
 421
 422         ret = -EFBIG;
 423         /* Check for wrap through zero too */
 424         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 425                 goto out_fput;
 426
 427         if (inode->i_op->fallocate)
 428                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 429         else
 430                 ret = -EOPNOTSUPP;
 431
 432 out_fput:
 433         fput(file);
 434 out:
 435         return ret;
 436 }
 437 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 438 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 439 {
 440         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 441 }
 442 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 443 #endif
 444
 445 /*
 446  * access() needs to use the real uid/gid, not the effective uid/gid.
 447  * We do this by temporarily clearing all FS-related capabilities and
 448  * switching the fsuid/fsgid around to the real ones.
 449  */
 450 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 451 {
 452         const struct cred *old_cred;
 453         struct cred *override_cred;
 454         struct path path;
 455         struct inode *inode;
 456         int res;
 457
 458         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 459                 return -EINVAL;
 460
 461         override_cred = prepare_creds();
 462         if (!override_cred)
 463                 return -ENOMEM;
 464
 465         override_cred->fsuid = override_cred->uid;
 466         override_cred->fsgid = override_cred->gid;
 467
 468         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 469                 /* Clear the capabilities if we switch to a non-root user */
 470                 if (override_cred->uid)
 471                         cap_clear(override_cred->cap_effective);
 472                 else
 473                         override_cred->cap_effective =
 474                                 override_cred->cap_permitted;
 475         }
 476
 477         old_cred = override_creds(override_cred);
 478
 479         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 480         if (res)
 481                 goto out;
 482
 483         inode = path.dentry->d_inode;
 484
 485         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 486                 /*
 487                  * MAY_EXEC on regular files is denied if the fs is mounted
 488                  * with the "noexec" flag.
 489                  */
 490                 res = -EACCES;
 491                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 492                         goto out_path_release;
 493         }
 494
 495         res = inode_permission(inode, mode | MAY_ACCESS);
 496         /* SuS v2 requires we report a read only fs too */
 497         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 498                 goto out_path_release;
 499         /*
 500          * This is a rare case where using __mnt_is_readonly()
 501          * is OK without a mnt_want/drop_write() pair.  Since
 502          * no actual write to the fs is performed here, we do
 503          * not need to telegraph to that to anyone.
 504          *
 505          * By doing this, we accept that this access is
 506          * inherently racy and know that the fs may change
 507          * state before we even see this result.
 508          */
 509         if (__mnt_is_readonly(path.mnt))
 510                 res = -EROFS;
 511
 512 out_path_release:
 513         path_put(&path);
 514 out:
 515         revert_creds(old_cred);
 516         put_cred(override_cred);
 517         return res;
 518 }
 519
 520 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 521 {
 522         return sys_faccessat(AT_FDCWD, filename, mode);
 523 }
 524
 525 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 526 {
 527         struct path path;
 528         int error;
 529
 530         error = user_path_dir(filename, &path);
 531         if (error)
 532                 goto out;
 533
 534         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 535         if (error)
 536                 goto dput_and_out;
 537
 538         set_fs_pwd(current->fs, &path);
 539
 540 dput_and_out:
 541         path_put(&path);
 542 out:
 543         return error;
 544 }
 545
 546 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 547 {
 548         struct file *file;
 549         struct inode *inode;
 550         int error;
 551
 552         error = -EBADF;
 553         file = fget(fd);
 554         if (!file)
 555                 goto out;
 556
 557         inode = file->f_path.dentry->d_inode;
 558
 559         error = -ENOTDIR;
 560         if (!S_ISDIR(inode->i_mode))
 561                 goto out_putf;
 562
 563         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 564         if (!error)
 565                 set_fs_pwd(current->fs, &file->f_path);
 566 out_putf:
 567         fput(file);
 568 out:
 569         return error;
 570 }
 571
 572 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 573 {
 574         struct path path;
 575         int error;
 576
 577         error = user_path_dir(filename, &path);
 578         if (error)
 579                 goto out;
 580
 581         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 582         if (error)
 583                 goto dput_and_out;
 584
 585         error = -EPERM;
 586         if (!capable(CAP_SYS_CHROOT))
 587                 goto dput_and_out;
 588
 589         set_fs_root(current->fs, &path);
 590         error = 0;
 591 dput_and_out:
 592         path_put(&path);
 593 out:
 594         return error;
 595 }
 596
 597 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 598 {
 599         struct inode * inode;
 600         struct dentry * dentry;
 601         struct file * file;
 602         int err = -EBADF;
 603         struct iattr newattrs;
 604
 605         file = fget(fd);
 606         if (!file)
 607                 goto out;
 608
 609         dentry = file->f_path.dentry;
 610         inode = dentry->d_inode;
 611
 612         audit_inode(NULL, dentry);
 613
 614         err = mnt_want_write(file->f_path.mnt);
 615         if (err)
 616                 goto out_putf;
 617         mutex_lock(&inode->i_mutex);
 618         if (mode == (mode_t) -1)
 619                 mode = inode->i_mode;
 620         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 621         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 622         err = notify_change(dentry, &newattrs);
 623         mutex_unlock(&inode->i_mutex);
 624         mnt_drop_write(file->f_path.mnt);
 625 out_putf:
 626         fput(file);
 627 out:
 628         return err;
 629 }
 630
 631 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 632 {
 633         struct path path;
 634         struct inode *inode;
 635         int error;
 636         struct iattr newattrs;
 637
 638         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 639         if (error)
 640                 goto out;
 641         inode = path.dentry->d_inode;
 642
 643         error = mnt_want_write(path.mnt);
 644         if (error)
 645                 goto dput_and_out;
 646         mutex_lock(&inode->i_mutex);
 647         if (mode == (mode_t) -1)
 648                 mode = inode->i_mode;
 649         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 650         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 651         error = notify_change(path.dentry, &newattrs);
 652         mutex_unlock(&inode->i_mutex);
 653         mnt_drop_write(path.mnt);
 654 dput_and_out:
 655         path_put(&path);
 656 out:
 657         return error;
 658 }
 659
 660 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 661 {
 662         return sys_fchmodat(AT_FDCWD, filename, mode);
 663 }
 664
 665 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 666 {
 667         struct inode *inode = dentry->d_inode;
 668         int error;
 669         struct iattr newattrs;
 670
 671         newattrs.ia_valid =  ATTR_CTIME;
 672         if (user != (uid_t) -1) {
 673                 newattrs.ia_valid |= ATTR_UID;
 674                 newattrs.ia_uid = user;
 675         }
 676         if (group != (gid_t) -1) {
 677                 newattrs.ia_valid |= ATTR_GID;
 678                 newattrs.ia_gid = group;
 679         }
 680         if (!S_ISDIR(inode->i_mode))
 681                 newattrs.ia_valid |=
 682                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 683         mutex_lock(&inode->i_mutex);
 684         error = notify_change(dentry, &newattrs);
 685         mutex_unlock(&inode->i_mutex);
 686
 687         return error;
 688 }
 689
 690 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 691 {
 692         struct path path;
 693         int error;
 694
 695         error = user_path(filename, &path);
 696         if (error)
 697                 goto out;
 698         error = mnt_want_write(path.mnt);
 699         if (error)
 700                 goto out_release;
 701         error = chown_common(path.dentry, user, group);
 702         mnt_drop_write(path.mnt);
 703 out_release:
 704         path_put(&path);
 705 out:
 706         return error;
 707 }
 708
 709 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 710                 gid_t, group, int, flag)
 711 {
 712         struct path path;
 713         int error = -EINVAL;
 714         int follow;
 715
 716         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 717                 goto out;
 718
 719         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 720         error = user_path_at(dfd, filename, follow, &path);
 721         if (error)
 722                 goto out;
 723         error = mnt_want_write(path.mnt);
 724         if (error)
 725                 goto out_release;
 726         error = chown_common(path.dentry, user, group);
 727         mnt_drop_write(path.mnt);
 728 out_release:
 729         path_put(&path);
 730 out:
 731         return error;
 732 }
 733
 734 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 735 {
 736         struct path path;
 737         int error;
 738
 739         error = user_lpath(filename, &path);
 740         if (error)
 741                 goto out;
 742         error = mnt_want_write(path.mnt);
 743         if (error)
 744                 goto out_release;
 745         error = chown_common(path.dentry, user, group);
 746         mnt_drop_write(path.mnt);
 747 out_release:
 748         path_put(&path);
 749 out:
 750         return error;
 751 }
 752
 753 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 754 {
 755         struct file * file;
 756         int error = -EBADF;
 757         struct dentry * dentry;
 758
 759         file = fget(fd);
 760         if (!file)
 761                 goto out;
 762
 763         error = mnt_want_write(file->f_path.mnt);
 764         if (error)
 765                 goto out_fput;
 766         dentry = file->f_path.dentry;
 767         audit_inode(NULL, dentry);
 768         error = chown_common(dentry, user, group);
 769         mnt_drop_write(file->f_path.mnt);
 770 out_fput:
 771         fput(file);
 772 out:
 773         return error;
 774 }
 775
 776 /*
 777  * You have to be very careful that these write
 778  * counts get cleaned up in error cases and
 779  * upon __fput().  This should probably never
 780  * be called outside of __dentry_open().
 781  */
 782 static inline int __get_file_write_access(struct inode *inode,
 783                                           struct vfsmount *mnt)
 784 {
 785         int error;
 786         error = get_write_access(inode);
 787         if (error)
 788                 return error;
 789         /*
 790          * Do not take mount writer counts on
 791          * special files since no writes to
 792          * the mount itself will occur.
 793          */
 794         if (!special_file(inode->i_mode)) {
 795                 /*
 796                  * Balanced in __fput()
 797                  */
 798                 error = mnt_want_write(mnt);
 799                 if (error)
 800                         put_write_access(inode);
 801         }
 802         return error;
 803 }
 804
 805 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 806                                         int flags, struct file *f,
 807                                         int (*open)(struct inode *, struct file *),
 808                                         const struct cred *cred)
 809 {
 810         struct inode *inode;
 811         int error;
 812
 813         f->f_flags = flags;
 814         f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 815                                 FMODE_PREAD | FMODE_PWRITE;
 816         inode = dentry->d_inode;
 817         if (f->f_mode & FMODE_WRITE) {
 818                 error = __get_file_write_access(inode, mnt);
 819                 if (error)
 820                         goto cleanup_file;
 821                 if (!special_file(inode->i_mode))
 822                         file_take_write(f);
 823         }
 824
 825         f->f_mapping = inode->i_mapping;
 826         f->f_path.dentry = dentry;
 827         f->f_path.mnt = mnt;
 828         f->f_pos = 0;
 829         f->f_op = fops_get(inode->i_fop);
 830         file_move(f, &inode->i_sb->s_files);
 831
 832         error = security_dentry_open(f, cred);
 833         if (error)
 834                 goto cleanup_all;
 835
 836         if (!open && f->f_op)
 837                 open = f->f_op->open;
 838         if (open) {
 839                 error = open(inode, f);
 840                 if (error)
 841                         goto cleanup_all;
 842         }
 843
 844         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 845
 846         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 847
 848         /* NB: we're sure to have correct a_ops only after f_op->open */
 849         if (f->f_flags & O_DIRECT) {
 850                 if (!f->f_mapping->a_ops ||
 851                     ((!f->f_mapping->a_ops->direct_IO) &&
 852                     (!f->f_mapping->a_ops->get_xip_mem))) {
 853                         fput(f);
 854                         f = ERR_PTR(-EINVAL);
 855                 }
 856         }
 857
 858         return f;
 859
 860 cleanup_all:
 861         fops_put(f->f_op);
 862         if (f->f_mode & FMODE_WRITE) {
 863                 put_write_access(inode);
 864                 if (!special_file(inode->i_mode)) {
 865                         /*
 866                          * We don't consider this a real
 867                          * mnt_want/drop_write() pair
 868                          * because it all happenend right
 869                          * here, so just reset the state.
 870                          */
 871                         file_reset_write(f);
 872                         mnt_drop_write(mnt);
 873                 }
 874         }
 875         file_kill(f);
 876         f->f_path.dentry = NULL;
 877         f->f_path.mnt = NULL;
 878 cleanup_file:
 879         put_filp(f);
 880         dput(dentry);
 881         mntput(mnt);
 882         return ERR_PTR(error);
 883 }
 884
 885 /**
 886  * lookup_instantiate_filp - instantiates the open intent filp
 887  * @nd: pointer to nameidata
 888  * @dentry: pointer to dentry
 889  * @open: open callback
 890  *
 891  * Helper for filesystems that want to use lookup open intents and pass back
 892  * a fully instantiated struct file to the caller.
 893  * This function is meant to be called from within a filesystem's
 894  * lookup method.
 895  * Beware of calling it for non-regular files! Those ->open methods might block
 896  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 897  * leading to a deadlock, as nobody can open that fifo anymore, because
 898  * another process to open fifo will block on locked parent when doing lookup).
 899  * Note that in case of error, nd->intent.open.file is destroyed, but the
 900  * path information remains valid.
 901  * If the open callback is set to NULL, then the standard f_op->open()
 902  * filesystem callback is substituted.
 903  */
 904 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 905                 int (*open)(struct inode *, struct file *))
 906 {
 907         const struct cred *cred = current_cred();
 908
 909         if (IS_ERR(nd->intent.open.file))
 910                 goto out;
 911         if (IS_ERR(dentry))
 912                 goto out_err;
 913         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 914                                              nd->intent.open.flags - 1,
 915                                              nd->intent.open.file,
 916                                              open, cred);
 917 out:
 918         return nd->intent.open.file;
 919 out_err:
 920         release_open_intent(nd);
 921         nd->intent.open.file = (struct file *)dentry;
 922         goto out;
 923 }
 924 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 925
 926 /**
 927  * nameidata_to_filp - convert a nameidata to an open filp.
 928  * @nd: pointer to nameidata
 929  * @flags: open flags
 930  *
 931  * Note that this function destroys the original nameidata
 932  */
 933 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 934 {
 935         const struct cred *cred = current_cred();
 936         struct file *filp;
 937
 938         /* Pick up the filp from the open intent */
 939         filp = nd->intent.open.file;
 940         /* Has the filesystem initialised the file for us? */
 941         if (filp->f_path.dentry == NULL)
 942                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 943                                      NULL, cred);
 944         else
 945                 path_put(&nd->path);
 946         return filp;
 947 }
 948
 949 /*
 950  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 951  * error.
 952  */
 953 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 954                          const struct cred *cred)
 955 {
 956         int error;
 957         struct file *f;
 958
 959         /*
 960          * We must always pass in a valid mount pointer.   Historically
 961          * callers got away with not passing it, but we must enforce this at
 962          * the earliest possible point now to avoid strange problems deep in the
 963          * filesystem stack.
 964          */
 965         if (!mnt) {
 966                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 967                 dump_stack();
 968                 return ERR_PTR(-EINVAL);
 969         }
 970
 971         error = -ENFILE;
 972         f = get_empty_filp();
 973         if (f == NULL) {
 974                 dput(dentry);
 975                 mntput(mnt);
 976                 return ERR_PTR(error);
 977         }
 978
 979         return __dentry_open(dentry, mnt, flags, f, NULL, cred);
 980 }
 981 EXPORT_SYMBOL(dentry_open);
 982
 983 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 984 {
 985         struct fdtable *fdt = files_fdtable(files);
 986         __FD_CLR(fd, fdt->open_fds);
 987         if (fd < files->next_fd)
 988                 files->next_fd = fd;
 989 }
 990
 991 void put_unused_fd(unsigned int fd)
 992 {
 993         struct files_struct *files = current->files;
 994         spin_lock(&files->file_lock);
 995         __put_unused_fd(files, fd);
 996         spin_unlock(&files->file_lock);
 997 }
 998
 999 EXPORT_SYMBOL(put_unused_fd);
1000
1001 /*
1002  * Install a file pointer in the fd array.
1003  *
1004  * The VFS is full of places where we drop the files lock between
1005  * setting the open_fds bitmap and installing the file in the file
1006  * array.  At any such point, we are vulnerable to a dup2() race
1007  * installing a file in the array before us.  We need to detect this and
1008  * fput() the struct file we are about to overwrite in this case.
1009  *
1010  * It should never happen - if we allow dup2() do it, _really_ bad things
1011  * will follow.
1012  */
1013
1014 void fd_install(unsigned int fd, struct file *file)
1015 {
1016         struct files_struct *files = current->files;
1017         struct fdtable *fdt;
1018         spin_lock(&files->file_lock);
1019         fdt = files_fdtable(files);
1020         BUG_ON(fdt->fd[fd] != NULL);
1021         rcu_assign_pointer(fdt->fd[fd], file);
1022         spin_unlock(&files->file_lock);
1023 }
1024
1025 EXPORT_SYMBOL(fd_install);
1026
1027 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1028 {
1029         char *tmp = getname(filename);
1030         int fd = PTR_ERR(tmp);
1031
1032         if (!IS_ERR(tmp)) {
1033                 fd = get_unused_fd_flags(flags);
1034                 if (fd >= 0) {
1035                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1036                         if (IS_ERR(f)) {
1037                                 put_unused_fd(fd);
1038                                 fd = PTR_ERR(f);
1039                         } else {
1040                                 fsnotify_open(f->f_path.dentry);
1041                                 fd_install(fd, f);
1042                         }
1043                 }
1044                 putname(tmp);
1045         }
1046         return fd;
1047 }
1048
1049 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1050 {
1051         long ret;
1052
1053         if (force_o_largefile())
1054                 flags |= O_LARGEFILE;
1055
1056         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1057         /* avoid REGPARM breakage on x86: */
1058         asmlinkage_protect(3, ret, filename, flags, mode);
1059         return ret;
1060 }
1061
1062 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1063                 int, mode)
1064 {
1065         long ret;
1066
1067         if (force_o_largefile())
1068                 flags |= O_LARGEFILE;
1069
1070         ret = do_sys_open(dfd, filename, flags, mode);
1071         /* avoid REGPARM breakage on x86: */
1072         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1073         return ret;
1074 }
1075
1076 #ifndef __alpha__
1077
1078 /*
1079  * For backward compatibility?  Maybe this should be moved
1080  * into arch/i386 instead?
1081  */
1082 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1083 {
1084         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1085 }
1086
1087 #endif
1088
1089 /*
1090  * "id" is the POSIX thread ID. We use the
1091  * files pointer for this..
1092  */
1093 int filp_close(struct file *filp, fl_owner_t id)
1094 {
1095         int retval = 0;
1096
1097         if (!file_count(filp)) {
1098                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1099                 return 0;
1100         }
1101
1102         if (filp->f_op && filp->f_op->flush)
1103                 retval = filp->f_op->flush(filp, id);
1104
1105         dnotify_flush(filp, id);
1106         locks_remove_posix(filp, id);
1107         fput(filp);
1108         return retval;
1109 }
1110
1111 EXPORT_SYMBOL(filp_close);
1112
1113 /*
1114  * Careful here! We test whether the file pointer is NULL before
1115  * releasing the fd. This ensures that one clone task can't release
1116  * an fd while another clone is opening it.
1117  */
1118 SYSCALL_DEFINE1(close, unsigned int, fd)
1119 {
1120         struct file * filp;
1121         struct files_struct *files = current->files;
1122         struct fdtable *fdt;
1123         int retval;
1124
1125         spin_lock(&files->file_lock);
1126         fdt = files_fdtable(files);
1127         if (fd >= fdt->max_fds)
1128                 goto out_unlock;
1129         filp = fdt->fd[fd];
1130         if (!filp)
1131                 goto out_unlock;
1132         rcu_assign_pointer(fdt->fd[fd], NULL);
1133         FD_CLR(fd, fdt->close_on_exec);
1134         __put_unused_fd(files, fd);
1135         spin_unlock(&files->file_lock);
1136         retval = filp_close(filp, files);
1137
1138         /* can't restart close syscall because file table entry was cleared */
1139         if (unlikely(retval == -ERESTARTSYS ||
1140                      retval == -ERESTARTNOINTR ||
1141                      retval == -ERESTARTNOHAND ||
1142                      retval == -ERESTART_RESTARTBLOCK))
1143                 retval = -EINTR;
1144
1145         return retval;
1146
1147 out_unlock:
1148         spin_unlock(&files->file_lock);
1149         return -EBADF;
1150 }
1151 EXPORT_SYMBOL(sys_close);
1152
1153 /*
1154  * This routine simulates a hangup on the tty, to arrange that users
1155  * are given clean terminals at login time.
1156  */
1157 SYSCALL_DEFINE0(vhangup)
1158 {
1159         if (capable(CAP_SYS_TTY_CONFIG)) {
1160                 tty_vhangup_self();
1161                 return 0;
1162         }
1163         return -EPERM;
1164 }
1165
1166 /*
1167  * Called when an inode is about to be open.
1168  * We use this to disallow opening large files on 32bit systems if
1169  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1170  * on this flag in sys_open.
1171  */
1172 int generic_file_open(struct inode * inode, struct file * filp)
1173 {
1174         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1175                 return -EOVERFLOW;
1176         return 0;
1177 }
1178
1179 EXPORT_SYMBOL(generic_file_open);
1180
1181 /*
1182  * This is used by subsystems that don't want seekable
1183  * file descriptors
1184  */
1185 int nonseekable_open(struct inode *inode, struct file *filp)
1186 {
1187         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1188         return 0;
1189 }
1190
1191 EXPORT_SYMBOL(nonseekable_open);