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 #include <linux/fs_struct.h>
  33
  34 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  35 {
  36         int retval = -ENODEV;
  37
  38         if (dentry) {
  39                 retval = -ENOSYS;
  40                 if (dentry->d_sb->s_op->statfs) {
  41                         memset(buf, 0, sizeof(*buf));
  42                         retval = security_sb_statfs(dentry);
  43                         if (retval)
  44                                 return retval;
  45                         retval = dentry->d_sb->s_op->statfs(dentry, buf);
  46                         if (retval == 0 && buf->f_frsize == 0)
  47                                 buf->f_frsize = buf->f_bsize;
  48                 }
  49         }
  50         return retval;
  51 }
  52
  53 EXPORT_SYMBOL(vfs_statfs);
  54
  55 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
  56 {
  57         struct kstatfs st;
  58         int retval;
  59
  60         retval = vfs_statfs(dentry, &st);
  61         if (retval)
  62                 return retval;
  63
  64         if (sizeof(*buf) == sizeof(st))
  65                 memcpy(buf, &st, sizeof(st));
  66         else {
  67                 if (sizeof buf->f_blocks == 4) {
  68                         if ((st.f_blocks | st.f_bfree | st.f_bavail |
  69                              st.f_bsize | st.f_frsize) &
  70                             0xffffffff00000000ULL)
  71                                 return -EOVERFLOW;
  72                         /*
  73                          * f_files and f_ffree may be -1; it's okay to stuff
  74                          * that into 32 bits
  75                          */
  76                         if (st.f_files != -1 &&
  77                             (st.f_files & 0xffffffff00000000ULL))
  78                                 return -EOVERFLOW;
  79                         if (st.f_ffree != -1 &&
  80                             (st.f_ffree & 0xffffffff00000000ULL))
  81                                 return -EOVERFLOW;
  82                 }
  83
  84                 buf->f_type = st.f_type;
  85                 buf->f_bsize = st.f_bsize;
  86                 buf->f_blocks = st.f_blocks;
  87                 buf->f_bfree = st.f_bfree;
  88                 buf->f_bavail = st.f_bavail;
  89                 buf->f_files = st.f_files;
  90                 buf->f_ffree = st.f_ffree;
  91                 buf->f_fsid = st.f_fsid;
  92                 buf->f_namelen = st.f_namelen;
  93                 buf->f_frsize = st.f_frsize;
  94                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  95         }
  96         return 0;
  97 }
  98
  99 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
 100 {
 101         struct kstatfs st;
 102         int retval;
 103
 104         retval = vfs_statfs(dentry, &st);
 105         if (retval)
 106                 return retval;
 107
 108         if (sizeof(*buf) == sizeof(st))
 109                 memcpy(buf, &st, sizeof(st));
 110         else {
 111                 buf->f_type = st.f_type;
 112                 buf->f_bsize = st.f_bsize;
 113                 buf->f_blocks = st.f_blocks;
 114                 buf->f_bfree = st.f_bfree;
 115                 buf->f_bavail = st.f_bavail;
 116                 buf->f_files = st.f_files;
 117                 buf->f_ffree = st.f_ffree;
 118                 buf->f_fsid = st.f_fsid;
 119                 buf->f_namelen = st.f_namelen;
 120                 buf->f_frsize = st.f_frsize;
 121                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 122         }
 123         return 0;
 124 }
 125
 126 SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct statfs __user *, buf)
 127 {
 128         struct path path;
 129         int error;
 130
 131         error = user_path(pathname, &path);
 132         if (!error) {
 133                 struct statfs tmp;
 134                 error = vfs_statfs_native(path.dentry, &tmp);
 135                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 136                         error = -EFAULT;
 137                 path_put(&path);
 138         }
 139         return error;
 140 }
 141
 142 SYSCALL_DEFINE3(statfs64, const char __user *, pathname, size_t, sz, struct statfs64 __user *, buf)
 143 {
 144         struct path path;
 145         long error;
 146
 147         if (sz != sizeof(*buf))
 148                 return -EINVAL;
 149         error = user_path(pathname, &path);
 150         if (!error) {
 151                 struct statfs64 tmp;
 152                 error = vfs_statfs64(path.dentry, &tmp);
 153                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 154                         error = -EFAULT;
 155                 path_put(&path);
 156         }
 157         return error;
 158 }
 159
 160 SYSCALL_DEFINE2(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 SYSCALL_DEFINE3(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 ret;
 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         ret = should_remove_suid(dentry);
 218         if (ret)
 219                 newattrs.ia_valid |= ret | ATTR_FORCE;
 220
 221         mutex_lock(&dentry->d_inode->i_mutex);
 222         ret = notify_change(dentry, &newattrs);
 223         mutex_unlock(&dentry->d_inode->i_mutex);
 224         return ret;
 225 }
 226
 227 static long do_sys_truncate(const char __user *pathname, loff_t length)
 228 {
 229         struct path path;
 230         struct inode *inode;
 231         int error;
 232
 233         error = -EINVAL;
 234         if (length < 0) /* sorry, but loff_t says... */
 235                 goto out;
 236
 237         error = user_path(pathname, &path);
 238         if (error)
 239                 goto out;
 240         inode = path.dentry->d_inode;
 241
 242         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 243         error = -EISDIR;
 244         if (S_ISDIR(inode->i_mode))
 245                 goto dput_and_out;
 246
 247         error = -EINVAL;
 248         if (!S_ISREG(inode->i_mode))
 249                 goto dput_and_out;
 250
 251         error = mnt_want_write(path.mnt);
 252         if (error)
 253                 goto dput_and_out;
 254
 255         error = inode_permission(inode, MAY_WRITE);
 256         if (error)
 257                 goto mnt_drop_write_and_out;
 258
 259         error = -EPERM;
 260         if (IS_APPEND(inode))
 261                 goto mnt_drop_write_and_out;
 262
 263         error = get_write_access(inode);
 264         if (error)
 265                 goto mnt_drop_write_and_out;
 266
 267         /*
 268          * Make sure that there are no leases.  get_write_access() protects
 269          * against the truncate racing with a lease-granting setlease().
 270          */
 271         error = break_lease(inode, FMODE_WRITE);
 272         if (error)
 273                 goto put_write_and_out;
 274
 275         error = locks_verify_truncate(inode, NULL, length);
 276         if (!error)
 277                 error = security_path_truncate(&path, length, 0);
 278         if (!error) {
 279                 vfs_dq_init(inode);
 280                 error = do_truncate(path.dentry, length, 0, NULL);
 281         }
 282
 283 put_write_and_out:
 284         put_write_access(inode);
 285 mnt_drop_write_and_out:
 286         mnt_drop_write(path.mnt);
 287 dput_and_out:
 288         path_put(&path);
 289 out:
 290         return error;
 291 }
 292
 293 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 294 {
 295         return do_sys_truncate(path, length);
 296 }
 297
 298 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 299 {
 300         struct inode * inode;
 301         struct dentry *dentry;
 302         struct file * file;
 303         int error;
 304
 305         error = -EINVAL;
 306         if (length < 0)
 307                 goto out;
 308         error = -EBADF;
 309         file = fget(fd);
 310         if (!file)
 311                 goto out;
 312
 313         /* explicitly opened as large or we are on 64-bit box */
 314         if (file->f_flags & O_LARGEFILE)
 315                 small = 0;
 316
 317         dentry = file->f_path.dentry;
 318         inode = dentry->d_inode;
 319         error = -EINVAL;
 320         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 321                 goto out_putf;
 322
 323         error = -EINVAL;
 324         /* Cannot ftruncate over 2^31 bytes without large file support */
 325         if (small && length > MAX_NON_LFS)
 326                 goto out_putf;
 327
 328         error = -EPERM;
 329         if (IS_APPEND(inode))
 330                 goto out_putf;
 331
 332         error = locks_verify_truncate(inode, file, length);
 333         if (!error)
 334                 error = security_path_truncate(&file->f_path, length,
 335                                                ATTR_MTIME|ATTR_CTIME);
 336         if (!error)
 337                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 338 out_putf:
 339         fput(file);
 340 out:
 341         return error;
 342 }
 343
 344 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 345 {
 346         long ret = do_sys_ftruncate(fd, length, 1);
 347         /* avoid REGPARM breakage on x86: */
 348         asmlinkage_protect(2, ret, fd, length);
 349         return ret;
 350 }
 351
 352 /* LFS versions of truncate are only needed on 32 bit machines */
 353 #if BITS_PER_LONG == 32
 354 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 355 {
 356         return do_sys_truncate(path, length);
 357 }
 358 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 359 asmlinkage long SyS_truncate64(long path, loff_t length)
 360 {
 361         return SYSC_truncate64((const char __user *) path, length);
 362 }
 363 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 364 #endif
 365
 366 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 367 {
 368         long ret = do_sys_ftruncate(fd, length, 0);
 369         /* avoid REGPARM breakage on x86: */
 370         asmlinkage_protect(2, ret, fd, length);
 371         return ret;
 372 }
 373 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 374 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 375 {
 376         return SYSC_ftruncate64((unsigned int) fd, length);
 377 }
 378 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 379 #endif
 380 #endif /* BITS_PER_LONG == 32 */
 381
 382
 383 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 384 {
 385         struct inode *inode = file->f_path.dentry->d_inode;
 386         long ret;
 387
 388         if (offset < 0 || len <= 0)
 389                 return -EINVAL;
 390
 391         /* Return error if mode is not supported */
 392         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 393                 return -EOPNOTSUPP;
 394
 395         if (!(file->f_mode & FMODE_WRITE))
 396                 return -EBADF;
 397         /*
 398          * Revalidate the write permissions, in case security policy has
 399          * changed since the files were opened.
 400          */
 401         ret = security_file_permission(file, MAY_WRITE);
 402         if (ret)
 403                 return ret;
 404
 405         if (S_ISFIFO(inode->i_mode))
 406                 return -ESPIPE;
 407
 408         /*
 409          * Let individual file system decide if it supports preallocation
 410          * for directories or not.
 411          */
 412         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 413                 return -ENODEV;
 414
 415         /* Check for wrap through zero too */
 416         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 417                 return -EFBIG;
 418
 419         if (!inode->i_op->fallocate)
 420                 return -EOPNOTSUPP;
 421
 422         return inode->i_op->fallocate(inode, mode, offset, len);
 423 }
 424
 425 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 426 {
 427         struct file *file;
 428         int error = -EBADF;
 429
 430         file = fget(fd);
 431         if (file) {
 432                 error = do_fallocate(file, mode, offset, len);
 433                 fput(file);
 434         }
 435
 436         return error;
 437 }
 438
 439 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 440 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 441 {
 442         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 443 }
 444 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 445 #endif
 446
 447 /*
 448  * access() needs to use the real uid/gid, not the effective uid/gid.
 449  * We do this by temporarily clearing all FS-related capabilities and
 450  * switching the fsuid/fsgid around to the real ones.
 451  */
 452 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 453 {
 454         const struct cred *old_cred;
 455         struct cred *override_cred;
 456         struct path path;
 457         struct inode *inode;
 458         int res;
 459
 460         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 461                 return -EINVAL;
 462
 463         override_cred = prepare_creds();
 464         if (!override_cred)
 465                 return -ENOMEM;
 466
 467         override_cred->fsuid = override_cred->uid;
 468         override_cred->fsgid = override_cred->gid;
 469
 470         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 471                 /* Clear the capabilities if we switch to a non-root user */
 472                 if (override_cred->uid)
 473                         cap_clear(override_cred->cap_effective);
 474                 else
 475                         override_cred->cap_effective =
 476                                 override_cred->cap_permitted;
 477         }
 478
 479         old_cred = override_creds(override_cred);
 480
 481         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 482         if (res)
 483                 goto out;
 484
 485         inode = path.dentry->d_inode;
 486
 487         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 488                 /*
 489                  * MAY_EXEC on regular files is denied if the fs is mounted
 490                  * with the "noexec" flag.
 491                  */
 492                 res = -EACCES;
 493                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 494                         goto out_path_release;
 495         }
 496
 497         res = inode_permission(inode, mode | MAY_ACCESS);
 498         /* SuS v2 requires we report a read only fs too */
 499         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 500                 goto out_path_release;
 501         /*
 502          * This is a rare case where using __mnt_is_readonly()
 503          * is OK without a mnt_want/drop_write() pair.  Since
 504          * no actual write to the fs is performed here, we do
 505          * not need to telegraph to that to anyone.
 506          *
 507          * By doing this, we accept that this access is
 508          * inherently racy and know that the fs may change
 509          * state before we even see this result.
 510          */
 511         if (__mnt_is_readonly(path.mnt))
 512                 res = -EROFS;
 513
 514 out_path_release:
 515         path_put(&path);
 516 out:
 517         revert_creds(old_cred);
 518         put_cred(override_cred);
 519         return res;
 520 }
 521
 522 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 523 {
 524         return sys_faccessat(AT_FDCWD, filename, mode);
 525 }
 526
 527 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 528 {
 529         struct path path;
 530         int error;
 531
 532         error = user_path_dir(filename, &path);
 533         if (error)
 534                 goto out;
 535
 536         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 537         if (error)
 538                 goto dput_and_out;
 539
 540         set_fs_pwd(current->fs, &path);
 541
 542 dput_and_out:
 543         path_put(&path);
 544 out:
 545         return error;
 546 }
 547
 548 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 549 {
 550         struct file *file;
 551         struct inode *inode;
 552         int error;
 553
 554         error = -EBADF;
 555         file = fget(fd);
 556         if (!file)
 557                 goto out;
 558
 559         inode = file->f_path.dentry->d_inode;
 560
 561         error = -ENOTDIR;
 562         if (!S_ISDIR(inode->i_mode))
 563                 goto out_putf;
 564
 565         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 566         if (!error)
 567                 set_fs_pwd(current->fs, &file->f_path);
 568 out_putf:
 569         fput(file);
 570 out:
 571         return error;
 572 }
 573
 574 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 575 {
 576         struct path path;
 577         int error;
 578
 579         error = user_path_dir(filename, &path);
 580         if (error)
 581                 goto out;
 582
 583         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 584         if (error)
 585                 goto dput_and_out;
 586
 587         error = -EPERM;
 588         if (!capable(CAP_SYS_CHROOT))
 589                 goto dput_and_out;
 590
 591         set_fs_root(current->fs, &path);
 592         error = 0;
 593 dput_and_out:
 594         path_put(&path);
 595 out:
 596         return error;
 597 }
 598
 599 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 600 {
 601         struct inode * inode;
 602         struct dentry * dentry;
 603         struct file * file;
 604         int err = -EBADF;
 605         struct iattr newattrs;
 606
 607         file = fget(fd);
 608         if (!file)
 609                 goto out;
 610
 611         dentry = file->f_path.dentry;
 612         inode = dentry->d_inode;
 613
 614         audit_inode(NULL, dentry);
 615
 616         err = mnt_want_write_file(file);
 617         if (err)
 618                 goto out_putf;
 619         mutex_lock(&inode->i_mutex);
 620         if (mode == (mode_t) -1)
 621                 mode = inode->i_mode;
 622         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 623         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 624         err = notify_change(dentry, &newattrs);
 625         mutex_unlock(&inode->i_mutex);
 626         mnt_drop_write(file->f_path.mnt);
 627 out_putf:
 628         fput(file);
 629 out:
 630         return err;
 631 }
 632
 633 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 634 {
 635         struct path path;
 636         struct inode *inode;
 637         int error;
 638         struct iattr newattrs;
 639
 640         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 641         if (error)
 642                 goto out;
 643         inode = path.dentry->d_inode;
 644
 645         error = mnt_want_write(path.mnt);
 646         if (error)
 647                 goto dput_and_out;
 648         mutex_lock(&inode->i_mutex);
 649         if (mode == (mode_t) -1)
 650                 mode = inode->i_mode;
 651         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 652         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 653         error = notify_change(path.dentry, &newattrs);
 654         mutex_unlock(&inode->i_mutex);
 655         mnt_drop_write(path.mnt);
 656 dput_and_out:
 657         path_put(&path);
 658 out:
 659         return error;
 660 }
 661
 662 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 663 {
 664         return sys_fchmodat(AT_FDCWD, filename, mode);
 665 }
 666
 667 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 668 {
 669         struct inode *inode = dentry->d_inode;
 670         int error;
 671         struct iattr newattrs;
 672
 673         newattrs.ia_valid =  ATTR_CTIME;
 674         if (user != (uid_t) -1) {
 675                 newattrs.ia_valid |= ATTR_UID;
 676                 newattrs.ia_uid = user;
 677         }
 678         if (group != (gid_t) -1) {
 679                 newattrs.ia_valid |= ATTR_GID;
 680                 newattrs.ia_gid = group;
 681         }
 682         if (!S_ISDIR(inode->i_mode))
 683                 newattrs.ia_valid |=
 684                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 685         mutex_lock(&inode->i_mutex);
 686         error = notify_change(dentry, &newattrs);
 687         mutex_unlock(&inode->i_mutex);
 688
 689         return error;
 690 }
 691
 692 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 693 {
 694         struct path path;
 695         int error;
 696
 697         error = user_path(filename, &path);
 698         if (error)
 699                 goto out;
 700         error = mnt_want_write(path.mnt);
 701         if (error)
 702                 goto out_release;
 703         error = chown_common(path.dentry, user, group);
 704         mnt_drop_write(path.mnt);
 705 out_release:
 706         path_put(&path);
 707 out:
 708         return error;
 709 }
 710
 711 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 712                 gid_t, group, int, flag)
 713 {
 714         struct path path;
 715         int error = -EINVAL;
 716         int follow;
 717
 718         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 719                 goto out;
 720
 721         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 722         error = user_path_at(dfd, filename, follow, &path);
 723         if (error)
 724                 goto out;
 725         error = mnt_want_write(path.mnt);
 726         if (error)
 727                 goto out_release;
 728         error = chown_common(path.dentry, user, group);
 729         mnt_drop_write(path.mnt);
 730 out_release:
 731         path_put(&path);
 732 out:
 733         return error;
 734 }
 735
 736 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 737 {
 738         struct path path;
 739         int error;
 740
 741         error = user_lpath(filename, &path);
 742         if (error)
 743                 goto out;
 744         error = mnt_want_write(path.mnt);
 745         if (error)
 746                 goto out_release;
 747         error = chown_common(path.dentry, user, group);
 748         mnt_drop_write(path.mnt);
 749 out_release:
 750         path_put(&path);
 751 out:
 752         return error;
 753 }
 754
 755 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 756 {
 757         struct file * file;
 758         int error = -EBADF;
 759         struct dentry * dentry;
 760
 761         file = fget(fd);
 762         if (!file)
 763                 goto out;
 764
 765         error = mnt_want_write_file(file);
 766         if (error)
 767                 goto out_fput;
 768         dentry = file->f_path.dentry;
 769         audit_inode(NULL, dentry);
 770         error = chown_common(dentry, user, group);
 771         mnt_drop_write(file->f_path.mnt);
 772 out_fput:
 773         fput(file);
 774 out:
 775         return error;
 776 }
 777
 778 /*
 779  * You have to be very careful that these write
 780  * counts get cleaned up in error cases and
 781  * upon __fput().  This should probably never
 782  * be called outside of __dentry_open().
 783  */
 784 static inline int __get_file_write_access(struct inode *inode,
 785                                           struct vfsmount *mnt)
 786 {
 787         int error;
 788         error = get_write_access(inode);
 789         if (error)
 790                 return error;
 791         /*
 792          * Do not take mount writer counts on
 793          * special files since no writes to
 794          * the mount itself will occur.
 795          */
 796         if (!special_file(inode->i_mode)) {
 797                 /*
 798                  * Balanced in __fput()
 799                  */
 800                 error = mnt_want_write(mnt);
 801                 if (error)
 802                         put_write_access(inode);
 803         }
 804         return error;
 805 }
 806
 807 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 808                                         int flags, struct file *f,
 809                                         int (*open)(struct inode *, struct file *),
 810                                         const struct cred *cred)
 811 {
 812         struct inode *inode;
 813         int error;
 814
 815         f->f_flags = flags;
 816         f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 817                                 FMODE_PREAD | FMODE_PWRITE;
 818         inode = dentry->d_inode;
 819         if (f->f_mode & FMODE_WRITE) {
 820                 error = __get_file_write_access(inode, mnt);
 821                 if (error)
 822                         goto cleanup_file;
 823                 if (!special_file(inode->i_mode))
 824                         file_take_write(f);
 825         }
 826
 827         f->f_mapping = inode->i_mapping;
 828         f->f_path.dentry = dentry;
 829         f->f_path.mnt = mnt;
 830         f->f_pos = 0;
 831         f->f_op = fops_get(inode->i_fop);
 832         file_move(f, &inode->i_sb->s_files);
 833
 834         error = security_dentry_open(f, cred);
 835         if (error)
 836                 goto cleanup_all;
 837
 838         if (!open && f->f_op)
 839                 open = f->f_op->open;
 840         if (open) {
 841                 error = open(inode, f);
 842                 if (error)
 843                         goto cleanup_all;
 844         }
 845
 846         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 847
 848         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 849
 850         /* NB: we're sure to have correct a_ops only after f_op->open */
 851         if (f->f_flags & O_DIRECT) {
 852                 if (!f->f_mapping->a_ops ||
 853                     ((!f->f_mapping->a_ops->direct_IO) &&
 854                     (!f->f_mapping->a_ops->get_xip_mem))) {
 855                         fput(f);
 856                         f = ERR_PTR(-EINVAL);
 857                 }
 858         }
 859
 860         return f;
 861
 862 cleanup_all:
 863         fops_put(f->f_op);
 864         if (f->f_mode & FMODE_WRITE) {
 865                 put_write_access(inode);
 866                 if (!special_file(inode->i_mode)) {
 867                         /*
 868                          * We don't consider this a real
 869                          * mnt_want/drop_write() pair
 870                          * because it all happenend right
 871                          * here, so just reset the state.
 872                          */
 873                         file_reset_write(f);
 874                         mnt_drop_write(mnt);
 875                 }
 876         }
 877         file_kill(f);
 878         f->f_path.dentry = NULL;
 879         f->f_path.mnt = NULL;
 880 cleanup_file:
 881         put_filp(f);
 882         dput(dentry);
 883         mntput(mnt);
 884         return ERR_PTR(error);
 885 }
 886
 887 /**
 888  * lookup_instantiate_filp - instantiates the open intent filp
 889  * @nd: pointer to nameidata
 890  * @dentry: pointer to dentry
 891  * @open: open callback
 892  *
 893  * Helper for filesystems that want to use lookup open intents and pass back
 894  * a fully instantiated struct file to the caller.
 895  * This function is meant to be called from within a filesystem's
 896  * lookup method.
 897  * Beware of calling it for non-regular files! Those ->open methods might block
 898  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 899  * leading to a deadlock, as nobody can open that fifo anymore, because
 900  * another process to open fifo will block on locked parent when doing lookup).
 901  * Note that in case of error, nd->intent.open.file is destroyed, but the
 902  * path information remains valid.
 903  * If the open callback is set to NULL, then the standard f_op->open()
 904  * filesystem callback is substituted.
 905  */
 906 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 907                 int (*open)(struct inode *, struct file *))
 908 {
 909         const struct cred *cred = current_cred();
 910
 911         if (IS_ERR(nd->intent.open.file))
 912                 goto out;
 913         if (IS_ERR(dentry))
 914                 goto out_err;
 915         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 916                                              nd->intent.open.flags - 1,
 917                                              nd->intent.open.file,
 918                                              open, cred);
 919 out:
 920         return nd->intent.open.file;
 921 out_err:
 922         release_open_intent(nd);
 923         nd->intent.open.file = (struct file *)dentry;
 924         goto out;
 925 }
 926 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 927
 928 /**
 929  * nameidata_to_filp - convert a nameidata to an open filp.
 930  * @nd: pointer to nameidata
 931  * @flags: open flags
 932  *
 933  * Note that this function destroys the original nameidata
 934  */
 935 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 936 {
 937         const struct cred *cred = current_cred();
 938         struct file *filp;
 939
 940         /* Pick up the filp from the open intent */
 941         filp = nd->intent.open.file;
 942         /* Has the filesystem initialised the file for us? */
 943         if (filp->f_path.dentry == NULL)
 944                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 945                                      NULL, cred);
 946         else
 947                 path_put(&nd->path);
 948         return filp;
 949 }
 950
 951 /*
 952  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 953  * error.
 954  */
 955 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 956                          const struct cred *cred)
 957 {
 958         int error;
 959         struct file *f;
 960
 961         validate_creds(cred);
 962
 963         /*
 964          * We must always pass in a valid mount pointer.   Historically
 965          * callers got away with not passing it, but we must enforce this at
 966          * the earliest possible point now to avoid strange problems deep in the
 967          * filesystem stack.
 968          */
 969         if (!mnt) {
 970                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 971                 dump_stack();
 972                 return ERR_PTR(-EINVAL);
 973         }
 974
 975         error = -ENFILE;
 976         f = get_empty_filp();
 977         if (f == NULL) {
 978                 dput(dentry);
 979                 mntput(mnt);
 980                 return ERR_PTR(error);
 981         }
 982
 983         return __dentry_open(dentry, mnt, flags, f, NULL, cred);
 984 }
 985 EXPORT_SYMBOL(dentry_open);
 986
 987 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 988 {
 989         struct fdtable *fdt = files_fdtable(files);
 990         __FD_CLR(fd, fdt->open_fds);
 991         if (fd < files->next_fd)
 992                 files->next_fd = fd;
 993 }
 994
 995 void put_unused_fd(unsigned int fd)
 996 {
 997         struct files_struct *files = current->files;
 998         spin_lock(&files->file_lock);
 999         __put_unused_fd(files, fd);
1000         spin_unlock(&files->file_lock);
1001 }
1002
1003 EXPORT_SYMBOL(put_unused_fd);
1004
1005 /*
1006  * Install a file pointer in the fd array.
1007  *
1008  * The VFS is full of places where we drop the files lock between
1009  * setting the open_fds bitmap and installing the file in the file
1010  * array.  At any such point, we are vulnerable to a dup2() race
1011  * installing a file in the array before us.  We need to detect this and
1012  * fput() the struct file we are about to overwrite in this case.
1013  *
1014  * It should never happen - if we allow dup2() do it, _really_ bad things
1015  * will follow.
1016  */
1017
1018 void fd_install(unsigned int fd, struct file *file)
1019 {
1020         struct files_struct *files = current->files;
1021         struct fdtable *fdt;
1022         spin_lock(&files->file_lock);
1023         fdt = files_fdtable(files);
1024         BUG_ON(fdt->fd[fd] != NULL);
1025         rcu_assign_pointer(fdt->fd[fd], file);
1026         spin_unlock(&files->file_lock);
1027 }
1028
1029 EXPORT_SYMBOL(fd_install);
1030
1031 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1032 {
1033         char *tmp = getname(filename);
1034         int fd = PTR_ERR(tmp);
1035
1036         if (!IS_ERR(tmp)) {
1037                 fd = get_unused_fd_flags(flags);
1038                 if (fd >= 0) {
1039                         struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
1040                         if (IS_ERR(f)) {
1041                                 put_unused_fd(fd);
1042                                 fd = PTR_ERR(f);
1043                         } else {
1044                                 fsnotify_open(f->f_path.dentry);
1045                                 fd_install(fd, f);
1046                         }
1047                 }
1048                 putname(tmp);
1049         }
1050         return fd;
1051 }
1052
1053 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1054 {
1055         long ret;
1056
1057         if (force_o_largefile())
1058                 flags |= O_LARGEFILE;
1059
1060         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1061         /* avoid REGPARM breakage on x86: */
1062         asmlinkage_protect(3, ret, filename, flags, mode);
1063         return ret;
1064 }
1065
1066 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1067                 int, mode)
1068 {
1069         long ret;
1070
1071         if (force_o_largefile())
1072                 flags |= O_LARGEFILE;
1073
1074         ret = do_sys_open(dfd, filename, flags, mode);
1075         /* avoid REGPARM breakage on x86: */
1076         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1077         return ret;
1078 }
1079
1080 #ifndef __alpha__
1081
1082 /*
1083  * For backward compatibility?  Maybe this should be moved
1084  * into arch/i386 instead?
1085  */
1086 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1087 {
1088         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1089 }
1090
1091 #endif
1092
1093 /*
1094  * "id" is the POSIX thread ID. We use the
1095  * files pointer for this..
1096  */
1097 int filp_close(struct file *filp, fl_owner_t id)
1098 {
1099         int retval = 0;
1100
1101         if (!file_count(filp)) {
1102                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1103                 return 0;
1104         }
1105
1106         if (filp->f_op && filp->f_op->flush)
1107                 retval = filp->f_op->flush(filp, id);
1108
1109         dnotify_flush(filp, id);
1110         locks_remove_posix(filp, id);
1111         fput(filp);
1112         return retval;
1113 }
1114
1115 EXPORT_SYMBOL(filp_close);
1116
1117 /*
1118  * Careful here! We test whether the file pointer is NULL before
1119  * releasing the fd. This ensures that one clone task can't release
1120  * an fd while another clone is opening it.
1121  */
1122 SYSCALL_DEFINE1(close, unsigned int, fd)
1123 {
1124         struct file * filp;
1125         struct files_struct *files = current->files;
1126         struct fdtable *fdt;
1127         int retval;
1128
1129         spin_lock(&files->file_lock);
1130         fdt = files_fdtable(files);
1131         if (fd >= fdt->max_fds)
1132                 goto out_unlock;
1133         filp = fdt->fd[fd];
1134         if (!filp)
1135                 goto out_unlock;
1136         rcu_assign_pointer(fdt->fd[fd], NULL);
1137         FD_CLR(fd, fdt->close_on_exec);
1138         __put_unused_fd(files, fd);
1139         spin_unlock(&files->file_lock);
1140         retval = filp_close(filp, files);
1141
1142         /* can't restart close syscall because file table entry was cleared */
1143         if (unlikely(retval == -ERESTARTSYS ||
1144                      retval == -ERESTARTNOINTR ||
1145                      retval == -ERESTARTNOHAND ||
1146                      retval == -ERESTART_RESTARTBLOCK))
1147                 retval = -EINTR;
1148
1149         return retval;
1150
1151 out_unlock:
1152         spin_unlock(&files->file_lock);
1153         return -EBADF;
1154 }
1155 EXPORT_SYMBOL(sys_close);
1156
1157 /*
1158  * This routine simulates a hangup on the tty, to arrange that users
1159  * are given clean terminals at login time.
1160  */
1161 SYSCALL_DEFINE0(vhangup)
1162 {
1163         if (capable(CAP_SYS_TTY_CONFIG)) {
1164                 tty_vhangup_self();
1165                 return 0;
1166         }
1167         return -EPERM;
1168 }
1169
1170 /*
1171  * Called when an inode is about to be open.
1172  * We use this to disallow opening large files on 32bit systems if
1173  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1174  * on this flag in sys_open.
1175  */
1176 int generic_file_open(struct inode * inode, struct file * filp)
1177 {
1178         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1179                 return -EOVERFLOW;
1180         return 0;
1181 }
1182
1183 EXPORT_SYMBOL(generic_file_open);
1184
1185 /*
1186  * This is used by subsystems that don't want seekable
1187  * file descriptors
1188  */
1189 int nonseekable_open(struct inode *inode, struct file *filp)
1190 {
1191         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1192         return 0;
1193 }
1194
1195 EXPORT_SYMBOL(nonseekable_open);