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, unsigned long, length)
 294 {
 295         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 296         return do_sys_truncate(path, (long)length);
 297 }
 298
 299 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 300 {
 301         struct inode * inode;
 302         struct dentry *dentry;
 303         struct file * file;
 304         int error;
 305
 306         error = -EINVAL;
 307         if (length < 0)
 308                 goto out;
 309         error = -EBADF;
 310         file = fget(fd);
 311         if (!file)
 312                 goto out;
 313
 314         /* explicitly opened as large or we are on 64-bit box */
 315         if (file->f_flags & O_LARGEFILE)
 316                 small = 0;
 317
 318         dentry = file->f_path.dentry;
 319         inode = dentry->d_inode;
 320         error = -EINVAL;
 321         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 322                 goto out_putf;
 323
 324         error = -EINVAL;
 325         /* Cannot ftruncate over 2^31 bytes without large file support */
 326         if (small && length > MAX_NON_LFS)
 327                 goto out_putf;
 328
 329         error = -EPERM;
 330         if (IS_APPEND(inode))
 331                 goto out_putf;
 332
 333         error = locks_verify_truncate(inode, file, length);
 334         if (!error)
 335                 error = security_path_truncate(&file->f_path, length,
 336                                                ATTR_MTIME|ATTR_CTIME);
 337         if (!error)
 338                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 339 out_putf:
 340         fput(file);
 341 out:
 342         return error;
 343 }
 344
 345 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 346 {
 347         long ret = do_sys_ftruncate(fd, length, 1);
 348         /* avoid REGPARM breakage on x86: */
 349         asmlinkage_protect(2, ret, fd, length);
 350         return ret;
 351 }
 352
 353 /* LFS versions of truncate are only needed on 32 bit machines */
 354 #if BITS_PER_LONG == 32
 355 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 356 {
 357         return do_sys_truncate(path, length);
 358 }
 359 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 360 asmlinkage long SyS_truncate64(long path, loff_t length)
 361 {
 362         return SYSC_truncate64((const char __user *) path, length);
 363 }
 364 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 365 #endif
 366
 367 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 368 {
 369         long ret = do_sys_ftruncate(fd, length, 0);
 370         /* avoid REGPARM breakage on x86: */
 371         asmlinkage_protect(2, ret, fd, length);
 372         return ret;
 373 }
 374 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 375 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 376 {
 377         return SYSC_ftruncate64((unsigned int) fd, length);
 378 }
 379 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 380 #endif
 381 #endif /* BITS_PER_LONG == 32 */
 382
 383
 384 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 385 {
 386         struct inode *inode = file->f_path.dentry->d_inode;
 387         long ret;
 388
 389         if (offset < 0 || len <= 0)
 390                 return -EINVAL;
 391
 392         /* Return error if mode is not supported */
 393         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 394                 return -EOPNOTSUPP;
 395
 396         if (!(file->f_mode & FMODE_WRITE))
 397                 return -EBADF;
 398         /*
 399          * Revalidate the write permissions, in case security policy has
 400          * changed since the files were opened.
 401          */
 402         ret = security_file_permission(file, MAY_WRITE);
 403         if (ret)
 404                 return ret;
 405
 406         if (S_ISFIFO(inode->i_mode))
 407                 return -ESPIPE;
 408
 409         /*
 410          * Let individual file system decide if it supports preallocation
 411          * for directories or not.
 412          */
 413         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 414                 return -ENODEV;
 415
 416         /* Check for wrap through zero too */
 417         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 418                 return -EFBIG;
 419
 420         if (!inode->i_op->fallocate)
 421                 return -EOPNOTSUPP;
 422
 423         return inode->i_op->fallocate(inode, mode, offset, len);
 424 }
 425
 426 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 427 {
 428         struct file *file;
 429         int error = -EBADF;
 430
 431         file = fget(fd);
 432         if (file) {
 433                 error = do_fallocate(file, mode, offset, len);
 434                 fput(file);
 435         }
 436
 437         return error;
 438 }
 439
 440 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 441 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 442 {
 443         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 444 }
 445 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 446 #endif
 447
 448 /*
 449  * access() needs to use the real uid/gid, not the effective uid/gid.
 450  * We do this by temporarily clearing all FS-related capabilities and
 451  * switching the fsuid/fsgid around to the real ones.
 452  */
 453 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 454 {
 455         const struct cred *old_cred;
 456         struct cred *override_cred;
 457         struct path path;
 458         struct inode *inode;
 459         int res;
 460
 461         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 462                 return -EINVAL;
 463
 464         override_cred = prepare_creds();
 465         if (!override_cred)
 466                 return -ENOMEM;
 467
 468         override_cred->fsuid = override_cred->uid;
 469         override_cred->fsgid = override_cred->gid;
 470
 471         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 472                 /* Clear the capabilities if we switch to a non-root user */
 473                 if (override_cred->uid)
 474                         cap_clear(override_cred->cap_effective);
 475                 else
 476                         override_cred->cap_effective =
 477                                 override_cred->cap_permitted;
 478         }
 479
 480         old_cred = override_creds(override_cred);
 481
 482         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 483         if (res)
 484                 goto out;
 485
 486         inode = path.dentry->d_inode;
 487
 488         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 489                 /*
 490                  * MAY_EXEC on regular files is denied if the fs is mounted
 491                  * with the "noexec" flag.
 492                  */
 493                 res = -EACCES;
 494                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 495                         goto out_path_release;
 496         }
 497
 498         res = inode_permission(inode, mode | MAY_ACCESS);
 499         /* SuS v2 requires we report a read only fs too */
 500         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 501                 goto out_path_release;
 502         /*
 503          * This is a rare case where using __mnt_is_readonly()
 504          * is OK without a mnt_want/drop_write() pair.  Since
 505          * no actual write to the fs is performed here, we do
 506          * not need to telegraph to that to anyone.
 507          *
 508          * By doing this, we accept that this access is
 509          * inherently racy and know that the fs may change
 510          * state before we even see this result.
 511          */
 512         if (__mnt_is_readonly(path.mnt))
 513                 res = -EROFS;
 514
 515 out_path_release:
 516         path_put(&path);
 517 out:
 518         revert_creds(old_cred);
 519         put_cred(override_cred);
 520         return res;
 521 }
 522
 523 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 524 {
 525         return sys_faccessat(AT_FDCWD, filename, mode);
 526 }
 527
 528 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 529 {
 530         struct path path;
 531         int error;
 532
 533         error = user_path_dir(filename, &path);
 534         if (error)
 535                 goto out;
 536
 537         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 538         if (error)
 539                 goto dput_and_out;
 540
 541         set_fs_pwd(current->fs, &path);
 542
 543 dput_and_out:
 544         path_put(&path);
 545 out:
 546         return error;
 547 }
 548
 549 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 550 {
 551         struct file *file;
 552         struct inode *inode;
 553         int error;
 554
 555         error = -EBADF;
 556         file = fget(fd);
 557         if (!file)
 558                 goto out;
 559
 560         inode = file->f_path.dentry->d_inode;
 561
 562         error = -ENOTDIR;
 563         if (!S_ISDIR(inode->i_mode))
 564                 goto out_putf;
 565
 566         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 567         if (!error)
 568                 set_fs_pwd(current->fs, &file->f_path);
 569 out_putf:
 570         fput(file);
 571 out:
 572         return error;
 573 }
 574
 575 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 576 {
 577         struct path path;
 578         int error;
 579
 580         error = user_path_dir(filename, &path);
 581         if (error)
 582                 goto out;
 583
 584         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 585         if (error)
 586                 goto dput_and_out;
 587
 588         error = -EPERM;
 589         if (!capable(CAP_SYS_CHROOT))
 590                 goto dput_and_out;
 591
 592         set_fs_root(current->fs, &path);
 593         error = 0;
 594 dput_and_out:
 595         path_put(&path);
 596 out:
 597         return error;
 598 }
 599
 600 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 601 {
 602         struct inode * inode;
 603         struct dentry * dentry;
 604         struct file * file;
 605         int err = -EBADF;
 606         struct iattr newattrs;
 607
 608         file = fget(fd);
 609         if (!file)
 610                 goto out;
 611
 612         dentry = file->f_path.dentry;
 613         inode = dentry->d_inode;
 614
 615         audit_inode(NULL, dentry);
 616
 617         err = mnt_want_write_file(file);
 618         if (err)
 619                 goto out_putf;
 620         mutex_lock(&inode->i_mutex);
 621         if (mode == (mode_t) -1)
 622                 mode = inode->i_mode;
 623         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 624         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 625         err = notify_change(dentry, &newattrs);
 626         mutex_unlock(&inode->i_mutex);
 627         mnt_drop_write(file->f_path.mnt);
 628 out_putf:
 629         fput(file);
 630 out:
 631         return err;
 632 }
 633
 634 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 635 {
 636         struct path path;
 637         struct inode *inode;
 638         int error;
 639         struct iattr newattrs;
 640
 641         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 642         if (error)
 643                 goto out;
 644         inode = path.dentry->d_inode;
 645
 646         error = mnt_want_write(path.mnt);
 647         if (error)
 648                 goto dput_and_out;
 649         mutex_lock(&inode->i_mutex);
 650         if (mode == (mode_t) -1)
 651                 mode = inode->i_mode;
 652         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 653         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 654         error = notify_change(path.dentry, &newattrs);
 655         mutex_unlock(&inode->i_mutex);
 656         mnt_drop_write(path.mnt);
 657 dput_and_out:
 658         path_put(&path);
 659 out:
 660         return error;
 661 }
 662
 663 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 664 {
 665         return sys_fchmodat(AT_FDCWD, filename, mode);
 666 }
 667
 668 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 669 {
 670         struct inode *inode = dentry->d_inode;
 671         int error;
 672         struct iattr newattrs;
 673
 674         newattrs.ia_valid =  ATTR_CTIME;
 675         if (user != (uid_t) -1) {
 676                 newattrs.ia_valid |= ATTR_UID;
 677                 newattrs.ia_uid = user;
 678         }
 679         if (group != (gid_t) -1) {
 680                 newattrs.ia_valid |= ATTR_GID;
 681                 newattrs.ia_gid = group;
 682         }
 683         if (!S_ISDIR(inode->i_mode))
 684                 newattrs.ia_valid |=
 685                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 686         mutex_lock(&inode->i_mutex);
 687         error = notify_change(dentry, &newattrs);
 688         mutex_unlock(&inode->i_mutex);
 689
 690         return error;
 691 }
 692
 693 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 694 {
 695         struct path path;
 696         int error;
 697
 698         error = user_path(filename, &path);
 699         if (error)
 700                 goto out;
 701         error = mnt_want_write(path.mnt);
 702         if (error)
 703                 goto out_release;
 704         error = chown_common(path.dentry, user, group);
 705         mnt_drop_write(path.mnt);
 706 out_release:
 707         path_put(&path);
 708 out:
 709         return error;
 710 }
 711
 712 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 713                 gid_t, group, int, flag)
 714 {
 715         struct path path;
 716         int error = -EINVAL;
 717         int follow;
 718
 719         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 720                 goto out;
 721
 722         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 723         error = user_path_at(dfd, filename, follow, &path);
 724         if (error)
 725                 goto out;
 726         error = mnt_want_write(path.mnt);
 727         if (error)
 728                 goto out_release;
 729         error = chown_common(path.dentry, user, group);
 730         mnt_drop_write(path.mnt);
 731 out_release:
 732         path_put(&path);
 733 out:
 734         return error;
 735 }
 736
 737 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 738 {
 739         struct path path;
 740         int error;
 741
 742         error = user_lpath(filename, &path);
 743         if (error)
 744                 goto out;
 745         error = mnt_want_write(path.mnt);
 746         if (error)
 747                 goto out_release;
 748         error = chown_common(path.dentry, user, group);
 749         mnt_drop_write(path.mnt);
 750 out_release:
 751         path_put(&path);
 752 out:
 753         return error;
 754 }
 755
 756 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 757 {
 758         struct file * file;
 759         int error = -EBADF;
 760         struct dentry * dentry;
 761
 762         file = fget(fd);
 763         if (!file)
 764                 goto out;
 765
 766         error = mnt_want_write_file(file);
 767         if (error)
 768                 goto out_fput;
 769         dentry = file->f_path.dentry;
 770         audit_inode(NULL, dentry);
 771         error = chown_common(dentry, user, group);
 772         mnt_drop_write(file->f_path.mnt);
 773 out_fput:
 774         fput(file);
 775 out:
 776         return error;
 777 }
 778
 779 /*
 780  * You have to be very careful that these write
 781  * counts get cleaned up in error cases and
 782  * upon __fput().  This should probably never
 783  * be called outside of __dentry_open().
 784  */
 785 static inline int __get_file_write_access(struct inode *inode,
 786                                           struct vfsmount *mnt)
 787 {
 788         int error;
 789         error = get_write_access(inode);
 790         if (error)
 791                 return error;
 792         /*
 793          * Do not take mount writer counts on
 794          * special files since no writes to
 795          * the mount itself will occur.
 796          */
 797         if (!special_file(inode->i_mode)) {
 798                 /*
 799                  * Balanced in __fput()
 800                  */
 801                 error = mnt_want_write(mnt);
 802                 if (error)
 803                         put_write_access(inode);
 804         }
 805         return error;
 806 }
 807
 808 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 809                                         int flags, struct file *f,
 810                                         int (*open)(struct inode *, struct file *),
 811                                         const struct cred *cred)
 812 {
 813         struct inode *inode;
 814         int error;
 815
 816         f->f_flags = flags;
 817         f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 818                                 FMODE_PREAD | FMODE_PWRITE;
 819         inode = dentry->d_inode;
 820         if (f->f_mode & FMODE_WRITE) {
 821                 error = __get_file_write_access(inode, mnt);
 822                 if (error)
 823                         goto cleanup_file;
 824                 if (!special_file(inode->i_mode))
 825                         file_take_write(f);
 826         }
 827
 828         f->f_mapping = inode->i_mapping;
 829         f->f_path.dentry = dentry;
 830         f->f_path.mnt = mnt;
 831         f->f_pos = 0;
 832         f->f_op = fops_get(inode->i_fop);
 833         file_move(f, &inode->i_sb->s_files);
 834
 835         error = security_dentry_open(f, cred);
 836         if (error)
 837                 goto cleanup_all;
 838
 839         if (!open && f->f_op)
 840                 open = f->f_op->open;
 841         if (open) {
 842                 error = open(inode, f);
 843                 if (error)
 844                         goto cleanup_all;
 845         }
 846
 847         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 848
 849         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 850
 851         /* NB: we're sure to have correct a_ops only after f_op->open */
 852         if (f->f_flags & O_DIRECT) {
 853                 if (!f->f_mapping->a_ops ||
 854                     ((!f->f_mapping->a_ops->direct_IO) &&
 855                     (!f->f_mapping->a_ops->get_xip_mem))) {
 856                         fput(f);
 857                         f = ERR_PTR(-EINVAL);
 858                 }
 859         }
 860
 861         return f;
 862
 863 cleanup_all:
 864         fops_put(f->f_op);
 865         if (f->f_mode & FMODE_WRITE) {
 866                 put_write_access(inode);
 867                 if (!special_file(inode->i_mode)) {
 868                         /*
 869                          * We don't consider this a real
 870                          * mnt_want/drop_write() pair
 871                          * because it all happenend right
 872                          * here, so just reset the state.
 873                          */
 874                         file_reset_write(f);
 875                         mnt_drop_write(mnt);
 876                 }
 877         }
 878         file_kill(f);
 879         f->f_path.dentry = NULL;
 880         f->f_path.mnt = NULL;
 881 cleanup_file:
 882         put_filp(f);
 883         dput(dentry);
 884         mntput(mnt);
 885         return ERR_PTR(error);
 886 }
 887
 888 /**
 889  * lookup_instantiate_filp - instantiates the open intent filp
 890  * @nd: pointer to nameidata
 891  * @dentry: pointer to dentry
 892  * @open: open callback
 893  *
 894  * Helper for filesystems that want to use lookup open intents and pass back
 895  * a fully instantiated struct file to the caller.
 896  * This function is meant to be called from within a filesystem's
 897  * lookup method.
 898  * Beware of calling it for non-regular files! Those ->open methods might block
 899  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 900  * leading to a deadlock, as nobody can open that fifo anymore, because
 901  * another process to open fifo will block on locked parent when doing lookup).
 902  * Note that in case of error, nd->intent.open.file is destroyed, but the
 903  * path information remains valid.
 904  * If the open callback is set to NULL, then the standard f_op->open()
 905  * filesystem callback is substituted.
 906  */
 907 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 908                 int (*open)(struct inode *, struct file *))
 909 {
 910         const struct cred *cred = current_cred();
 911
 912         if (IS_ERR(nd->intent.open.file))
 913                 goto out;
 914         if (IS_ERR(dentry))
 915                 goto out_err;
 916         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 917                                              nd->intent.open.flags - 1,
 918                                              nd->intent.open.file,
 919                                              open, cred);
 920 out:
 921         return nd->intent.open.file;
 922 out_err:
 923         release_open_intent(nd);
 924         nd->intent.open.file = (struct file *)dentry;
 925         goto out;
 926 }
 927 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 928
 929 /**
 930  * nameidata_to_filp - convert a nameidata to an open filp.
 931  * @nd: pointer to nameidata
 932  * @flags: open flags
 933  *
 934  * Note that this function destroys the original nameidata
 935  */
 936 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 937 {
 938         const struct cred *cred = current_cred();
 939         struct file *filp;
 940
 941         /* Pick up the filp from the open intent */
 942         filp = nd->intent.open.file;
 943         /* Has the filesystem initialised the file for us? */
 944         if (filp->f_path.dentry == NULL)
 945                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 946                                      NULL, cred);
 947         else
 948                 path_put(&nd->path);
 949         return filp;
 950 }
 951
 952 /*
 953  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 954  * error.
 955  */
 956 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 957                          const struct cred *cred)
 958 {
 959         int error;
 960         struct file *f;
 961
 962         validate_creds(cred);
 963
 964         /*
 965          * We must always pass in a valid mount pointer.   Historically
 966          * callers got away with not passing it, but we must enforce this at
 967          * the earliest possible point now to avoid strange problems deep in the
 968          * filesystem stack.
 969          */
 970         if (!mnt) {
 971                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 972                 dump_stack();
 973                 return ERR_PTR(-EINVAL);
 974         }
 975
 976         error = -ENFILE;
 977         f = get_empty_filp();
 978         if (f == NULL) {
 979                 dput(dentry);
 980                 mntput(mnt);
 981                 return ERR_PTR(error);
 982         }
 983
 984         return __dentry_open(dentry, mnt, flags, f, NULL, cred);
 985 }
 986 EXPORT_SYMBOL(dentry_open);
 987
 988 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 989 {
 990         struct fdtable *fdt = files_fdtable(files);
 991         __FD_CLR(fd, fdt->open_fds);
 992         if (fd < files->next_fd)
 993                 files->next_fd = fd;
 994 }
 995
 996 void put_unused_fd(unsigned int fd)
 997 {
 998         struct files_struct *files = current->files;
 999         spin_lock(&files->file_lock);
1000         __put_unused_fd(files, fd);
1001         spin_unlock(&files->file_lock);
1002 }
1003
1004 EXPORT_SYMBOL(put_unused_fd);
1005
1006 /*
1007  * Install a file pointer in the fd array.
1008  *
1009  * The VFS is full of places where we drop the files lock between
1010  * setting the open_fds bitmap and installing the file in the file
1011  * array.  At any such point, we are vulnerable to a dup2() race
1012  * installing a file in the array before us.  We need to detect this and
1013  * fput() the struct file we are about to overwrite in this case.
1014  *
1015  * It should never happen - if we allow dup2() do it, _really_ bad things
1016  * will follow.
1017  */
1018
1019 void fd_install(unsigned int fd, struct file *file)
1020 {
1021         struct files_struct *files = current->files;
1022         struct fdtable *fdt;
1023         spin_lock(&files->file_lock);
1024         fdt = files_fdtable(files);
1025         BUG_ON(fdt->fd[fd] != NULL);
1026         rcu_assign_pointer(fdt->fd[fd], file);
1027         spin_unlock(&files->file_lock);
1028 }
1029
1030 EXPORT_SYMBOL(fd_install);
1031
1032 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1033 {
1034         char *tmp = getname(filename);
1035         int fd = PTR_ERR(tmp);
1036
1037         if (!IS_ERR(tmp)) {
1038                 fd = get_unused_fd_flags(flags);
1039                 if (fd >= 0) {
1040                         struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
1041                         if (IS_ERR(f)) {
1042                                 put_unused_fd(fd);
1043                                 fd = PTR_ERR(f);
1044                         } else {
1045                                 fsnotify_open(f->f_path.dentry);
1046                                 fd_install(fd, f);
1047                         }
1048                 }
1049                 putname(tmp);
1050         }
1051         return fd;
1052 }
1053
1054 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1055 {
1056         long ret;
1057
1058         if (force_o_largefile())
1059                 flags |= O_LARGEFILE;
1060
1061         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1062         /* avoid REGPARM breakage on x86: */
1063         asmlinkage_protect(3, ret, filename, flags, mode);
1064         return ret;
1065 }
1066
1067 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1068                 int, mode)
1069 {
1070         long ret;
1071
1072         if (force_o_largefile())
1073                 flags |= O_LARGEFILE;
1074
1075         ret = do_sys_open(dfd, filename, flags, mode);
1076         /* avoid REGPARM breakage on x86: */
1077         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1078         return ret;
1079 }
1080
1081 #ifndef __alpha__
1082
1083 /*
1084  * For backward compatibility?  Maybe this should be moved
1085  * into arch/i386 instead?
1086  */
1087 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1088 {
1089         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1090 }
1091
1092 #endif
1093
1094 /*
1095  * "id" is the POSIX thread ID. We use the
1096  * files pointer for this..
1097  */
1098 int filp_close(struct file *filp, fl_owner_t id)
1099 {
1100         int retval = 0;
1101
1102         if (!file_count(filp)) {
1103                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1104                 return 0;
1105         }
1106
1107         if (filp->f_op && filp->f_op->flush)
1108                 retval = filp->f_op->flush(filp, id);
1109
1110         dnotify_flush(filp, id);
1111         locks_remove_posix(filp, id);
1112         fput(filp);
1113         return retval;
1114 }
1115
1116 EXPORT_SYMBOL(filp_close);
1117
1118 /*
1119  * Careful here! We test whether the file pointer is NULL before
1120  * releasing the fd. This ensures that one clone task can't release
1121  * an fd while another clone is opening it.
1122  */
1123 SYSCALL_DEFINE1(close, unsigned int, fd)
1124 {
1125         struct file * filp;
1126         struct files_struct *files = current->files;
1127         struct fdtable *fdt;
1128         int retval;
1129
1130         spin_lock(&files->file_lock);
1131         fdt = files_fdtable(files);
1132         if (fd >= fdt->max_fds)
1133                 goto out_unlock;
1134         filp = fdt->fd[fd];
1135         if (!filp)
1136                 goto out_unlock;
1137         rcu_assign_pointer(fdt->fd[fd], NULL);
1138         FD_CLR(fd, fdt->close_on_exec);
1139         __put_unused_fd(files, fd);
1140         spin_unlock(&files->file_lock);
1141         retval = filp_close(filp, files);
1142
1143         /* can't restart close syscall because file table entry was cleared */
1144         if (unlikely(retval == -ERESTARTSYS ||
1145                      retval == -ERESTARTNOINTR ||
1146                      retval == -ERESTARTNOHAND ||
1147                      retval == -ERESTART_RESTARTBLOCK))
1148                 retval = -EINTR;
1149
1150         return retval;
1151
1152 out_unlock:
1153         spin_unlock(&files->file_lock);
1154         return -EBADF;
1155 }
1156 EXPORT_SYMBOL(sys_close);
1157
1158 /*
1159  * This routine simulates a hangup on the tty, to arrange that users
1160  * are given clean terminals at login time.
1161  */
1162 SYSCALL_DEFINE0(vhangup)
1163 {
1164         if (capable(CAP_SYS_TTY_CONFIG)) {
1165                 tty_vhangup_self();
1166                 return 0;
1167         }
1168         return -EPERM;
1169 }
1170
1171 /*
1172  * Called when an inode is about to be open.
1173  * We use this to disallow opening large files on 32bit systems if
1174  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1175  * on this flag in sys_open.
1176  */
1177 int generic_file_open(struct inode * inode, struct file * filp)
1178 {
1179         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1180                 return -EOVERFLOW;
1181         return 0;
1182 }
1183
1184 EXPORT_SYMBOL(generic_file_open);
1185
1186 /*
1187  * This is used by subsystems that don't want seekable
1188  * file descriptors
1189  */
1190 int nonseekable_open(struct inode *inode, struct file *filp)
1191 {
1192         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1193         return 0;
1194 }
1195
1196 EXPORT_SYMBOL(nonseekable_open);