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 asmlinkage long sys_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
 142 asmlinkage long sys_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
 161 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
 162 {
 163         struct file * file;
 164         struct statfs tmp;
 165         int error;
 166
 167         error = -EBADF;
 168         file = fget(fd);
 169         if (!file)
 170                 goto out;
 171         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 172         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 173                 error = -EFAULT;
 174         fput(file);
 175 out:
 176         return error;
 177 }
 178
 179 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
 180 {
 181         struct file * file;
 182         struct statfs64 tmp;
 183         int error;
 184
 185         if (sz != sizeof(*buf))
 186                 return -EINVAL;
 187
 188         error = -EBADF;
 189         file = fget(fd);
 190         if (!file)
 191                 goto out;
 192         error = vfs_statfs64(file->f_path.dentry, &tmp);
 193         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 194                 error = -EFAULT;
 195         fput(file);
 196 out:
 197         return error;
 198 }
 199
 200 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 201         struct file *filp)
 202 {
 203         int err;
 204         struct iattr newattrs;
 205
 206         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 207         if (length < 0)
 208                 return -EINVAL;
 209
 210         newattrs.ia_size = length;
 211         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 212         if (filp) {
 213                 newattrs.ia_file = filp;
 214                 newattrs.ia_valid |= ATTR_FILE;
 215         }
 216
 217         /* Remove suid/sgid on truncate too */
 218         newattrs.ia_valid |= should_remove_suid(dentry);
 219
 220         mutex_lock(&dentry->d_inode->i_mutex);
 221         err = notify_change(dentry, &newattrs);
 222         mutex_unlock(&dentry->d_inode->i_mutex);
 223         return err;
 224 }
 225
 226 static long do_sys_truncate(const char __user *pathname, loff_t length)
 227 {
 228         struct path path;
 229         struct inode *inode;
 230         int error;
 231
 232         error = -EINVAL;
 233         if (length < 0) /* sorry, but loff_t says... */
 234                 goto out;
 235
 236         error = user_path(pathname, &path);
 237         if (error)
 238                 goto out;
 239         inode = path.dentry->d_inode;
 240
 241         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 242         error = -EISDIR;
 243         if (S_ISDIR(inode->i_mode))
 244                 goto dput_and_out;
 245
 246         error = -EINVAL;
 247         if (!S_ISREG(inode->i_mode))
 248                 goto dput_and_out;
 249
 250         error = mnt_want_write(path.mnt);
 251         if (error)
 252                 goto dput_and_out;
 253
 254         error = inode_permission(inode, MAY_WRITE);
 255         if (error)
 256                 goto mnt_drop_write_and_out;
 257
 258         error = -EPERM;
 259         if (IS_APPEND(inode))
 260                 goto mnt_drop_write_and_out;
 261
 262         error = get_write_access(inode);
 263         if (error)
 264                 goto mnt_drop_write_and_out;
 265
 266         /*
 267          * Make sure that there are no leases.  get_write_access() protects
 268          * against the truncate racing with a lease-granting setlease().
 269          */
 270         error = break_lease(inode, FMODE_WRITE);
 271         if (error)
 272                 goto put_write_and_out;
 273
 274         error = locks_verify_truncate(inode, NULL, length);
 275         if (!error) {
 276                 DQUOT_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 asmlinkage long sys_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 = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 333 out_putf:
 334         fput(file);
 335 out:
 336         return error;
 337 }
 338
 339 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 340 {
 341         long ret = do_sys_ftruncate(fd, length, 1);
 342         /* avoid REGPARM breakage on x86: */
 343         asmlinkage_protect(2, ret, fd, length);
 344         return ret;
 345 }
 346
 347 /* LFS versions of truncate are only needed on 32 bit machines */
 348 #if BITS_PER_LONG == 32
 349 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 350 {
 351         return do_sys_truncate(path, length);
 352 }
 353
 354 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 355 {
 356         long ret = do_sys_ftruncate(fd, length, 0);
 357         /* avoid REGPARM breakage on x86: */
 358         asmlinkage_protect(2, ret, fd, length);
 359         return ret;
 360 }
 361 #endif
 362
 363 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 364 {
 365         struct file *file;
 366         struct inode *inode;
 367         long ret = -EINVAL;
 368
 369         if (offset < 0 || len <= 0)
 370                 goto out;
 371
 372         /* Return error if mode is not supported */
 373         ret = -EOPNOTSUPP;
 374         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 375                 goto out;
 376
 377         ret = -EBADF;
 378         file = fget(fd);
 379         if (!file)
 380                 goto out;
 381         if (!(file->f_mode & FMODE_WRITE))
 382                 goto out_fput;
 383         /*
 384          * Revalidate the write permissions, in case security policy has
 385          * changed since the files were opened.
 386          */
 387         ret = security_file_permission(file, MAY_WRITE);
 388         if (ret)
 389                 goto out_fput;
 390
 391         inode = file->f_path.dentry->d_inode;
 392
 393         ret = -ESPIPE;
 394         if (S_ISFIFO(inode->i_mode))
 395                 goto out_fput;
 396
 397         ret = -ENODEV;
 398         /*
 399          * Let individual file system decide if it supports preallocation
 400          * for directories or not.
 401          */
 402         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 403                 goto out_fput;
 404
 405         ret = -EFBIG;
 406         /* Check for wrap through zero too */
 407         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 408                 goto out_fput;
 409
 410         if (inode->i_op && inode->i_op->fallocate)
 411                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 412         else
 413                 ret = -EOPNOTSUPP;
 414
 415 out_fput:
 416         fput(file);
 417 out:
 418         return ret;
 419 }
 420
 421 /*
 422  * access() needs to use the real uid/gid, not the effective uid/gid.
 423  * We do this by temporarily clearing all FS-related capabilities and
 424  * switching the fsuid/fsgid around to the real ones.
 425  */
 426 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 427 {
 428         const struct cred *old_cred;
 429         struct cred *override_cred;
 430         struct path path;
 431         struct inode *inode;
 432         int res;
 433
 434         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 435                 return -EINVAL;
 436
 437         override_cred = prepare_creds();
 438         if (!override_cred)
 439                 return -ENOMEM;
 440
 441         override_cred->fsuid = override_cred->uid;
 442         override_cred->fsgid = override_cred->gid;
 443
 444         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 445                 /* Clear the capabilities if we switch to a non-root user */
 446                 if (override_cred->uid)
 447                         cap_clear(override_cred->cap_effective);
 448                 else
 449                         override_cred->cap_effective =
 450                                 override_cred->cap_permitted;
 451         }
 452
 453         old_cred = override_creds(override_cred);
 454
 455         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 456         if (res)
 457                 goto out;
 458
 459         inode = path.dentry->d_inode;
 460
 461         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 462                 /*
 463                  * MAY_EXEC on regular files is denied if the fs is mounted
 464                  * with the "noexec" flag.
 465                  */
 466                 res = -EACCES;
 467                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 468                         goto out_path_release;
 469         }
 470
 471         res = inode_permission(inode, mode | MAY_ACCESS);
 472         /* SuS v2 requires we report a read only fs too */
 473         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 474                 goto out_path_release;
 475         /*
 476          * This is a rare case where using __mnt_is_readonly()
 477          * is OK without a mnt_want/drop_write() pair.  Since
 478          * no actual write to the fs is performed here, we do
 479          * not need to telegraph to that to anyone.
 480          *
 481          * By doing this, we accept that this access is
 482          * inherently racy and know that the fs may change
 483          * state before we even see this result.
 484          */
 485         if (__mnt_is_readonly(path.mnt))
 486                 res = -EROFS;
 487
 488 out_path_release:
 489         path_put(&path);
 490 out:
 491         revert_creds(old_cred);
 492         put_cred(override_cred);
 493         return res;
 494 }
 495
 496 asmlinkage long sys_access(const char __user *filename, int mode)
 497 {
 498         return sys_faccessat(AT_FDCWD, filename, mode);
 499 }
 500
 501 asmlinkage long sys_chdir(const char __user * filename)
 502 {
 503         struct path path;
 504         int error;
 505
 506         error = user_path_dir(filename, &path);
 507         if (error)
 508                 goto out;
 509
 510         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 511         if (error)
 512                 goto dput_and_out;
 513
 514         set_fs_pwd(current->fs, &path);
 515
 516 dput_and_out:
 517         path_put(&path);
 518 out:
 519         return error;
 520 }
 521
 522 asmlinkage long sys_fchdir(unsigned int fd)
 523 {
 524         struct file *file;
 525         struct inode *inode;
 526         int error;
 527
 528         error = -EBADF;
 529         file = fget(fd);
 530         if (!file)
 531                 goto out;
 532
 533         inode = file->f_path.dentry->d_inode;
 534
 535         error = -ENOTDIR;
 536         if (!S_ISDIR(inode->i_mode))
 537                 goto out_putf;
 538
 539         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 540         if (!error)
 541                 set_fs_pwd(current->fs, &file->f_path);
 542 out_putf:
 543         fput(file);
 544 out:
 545         return error;
 546 }
 547
 548 asmlinkage long sys_chroot(const char __user * filename)
 549 {
 550         struct path path;
 551         int error;
 552
 553         error = user_path_dir(filename, &path);
 554         if (error)
 555                 goto out;
 556
 557         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 558         if (error)
 559                 goto dput_and_out;
 560
 561         error = -EPERM;
 562         if (!capable(CAP_SYS_CHROOT))
 563                 goto dput_and_out;
 564
 565         set_fs_root(current->fs, &path);
 566         error = 0;
 567 dput_and_out:
 568         path_put(&path);
 569 out:
 570         return error;
 571 }
 572
 573 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 574 {
 575         struct inode * inode;
 576         struct dentry * dentry;
 577         struct file * file;
 578         int err = -EBADF;
 579         struct iattr newattrs;
 580
 581         file = fget(fd);
 582         if (!file)
 583                 goto out;
 584
 585         dentry = file->f_path.dentry;
 586         inode = dentry->d_inode;
 587
 588         audit_inode(NULL, dentry);
 589
 590         err = mnt_want_write(file->f_path.mnt);
 591         if (err)
 592                 goto out_putf;
 593         mutex_lock(&inode->i_mutex);
 594         if (mode == (mode_t) -1)
 595                 mode = inode->i_mode;
 596         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 597         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 598         err = notify_change(dentry, &newattrs);
 599         mutex_unlock(&inode->i_mutex);
 600         mnt_drop_write(file->f_path.mnt);
 601 out_putf:
 602         fput(file);
 603 out:
 604         return err;
 605 }
 606
 607 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 608                              mode_t mode)
 609 {
 610         struct path path;
 611         struct inode *inode;
 612         int error;
 613         struct iattr newattrs;
 614
 615         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 616         if (error)
 617                 goto out;
 618         inode = path.dentry->d_inode;
 619
 620         error = mnt_want_write(path.mnt);
 621         if (error)
 622                 goto dput_and_out;
 623         mutex_lock(&inode->i_mutex);
 624         if (mode == (mode_t) -1)
 625                 mode = inode->i_mode;
 626         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 627         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 628         error = notify_change(path.dentry, &newattrs);
 629         mutex_unlock(&inode->i_mutex);
 630         mnt_drop_write(path.mnt);
 631 dput_and_out:
 632         path_put(&path);
 633 out:
 634         return error;
 635 }
 636
 637 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 638 {
 639         return sys_fchmodat(AT_FDCWD, filename, mode);
 640 }
 641
 642 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 643 {
 644         struct inode *inode = dentry->d_inode;
 645         int error;
 646         struct iattr newattrs;
 647
 648         newattrs.ia_valid =  ATTR_CTIME;
 649         if (user != (uid_t) -1) {
 650                 newattrs.ia_valid |= ATTR_UID;
 651                 newattrs.ia_uid = user;
 652         }
 653         if (group != (gid_t) -1) {
 654                 newattrs.ia_valid |= ATTR_GID;
 655                 newattrs.ia_gid = group;
 656         }
 657         if (!S_ISDIR(inode->i_mode))
 658                 newattrs.ia_valid |=
 659                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 660         mutex_lock(&inode->i_mutex);
 661         error = notify_change(dentry, &newattrs);
 662         mutex_unlock(&inode->i_mutex);
 663
 664         return error;
 665 }
 666
 667 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 668 {
 669         struct path path;
 670         int error;
 671
 672         error = user_path(filename, &path);
 673         if (error)
 674                 goto out;
 675         error = mnt_want_write(path.mnt);
 676         if (error)
 677                 goto out_release;
 678         error = chown_common(path.dentry, user, group);
 679         mnt_drop_write(path.mnt);
 680 out_release:
 681         path_put(&path);
 682 out:
 683         return error;
 684 }
 685
 686 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 687                              gid_t group, int flag)
 688 {
 689         struct path path;
 690         int error = -EINVAL;
 691         int follow;
 692
 693         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 694                 goto out;
 695
 696         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 697         error = user_path_at(dfd, filename, follow, &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 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 712 {
 713         struct path path;
 714         int error;
 715
 716         error = user_lpath(filename, &path);
 717         if (error)
 718                 goto out;
 719         error = mnt_want_write(path.mnt);
 720         if (error)
 721                 goto out_release;
 722         error = chown_common(path.dentry, user, group);
 723         mnt_drop_write(path.mnt);
 724 out_release:
 725         path_put(&path);
 726 out:
 727         return error;
 728 }
 729
 730
 731 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 732 {
 733         struct file * file;
 734         int error = -EBADF;
 735         struct dentry * dentry;
 736
 737         file = fget(fd);
 738         if (!file)
 739                 goto out;
 740
 741         error = mnt_want_write(file->f_path.mnt);
 742         if (error)
 743                 goto out_fput;
 744         dentry = file->f_path.dentry;
 745         audit_inode(NULL, dentry);
 746         error = chown_common(dentry, user, group);
 747         mnt_drop_write(file->f_path.mnt);
 748 out_fput:
 749         fput(file);
 750 out:
 751         return error;
 752 }
 753
 754 /*
 755  * You have to be very careful that these write
 756  * counts get cleaned up in error cases and
 757  * upon __fput().  This should probably never
 758  * be called outside of __dentry_open().
 759  */
 760 static inline int __get_file_write_access(struct inode *inode,
 761                                           struct vfsmount *mnt)
 762 {
 763         int error;
 764         error = get_write_access(inode);
 765         if (error)
 766                 return error;
 767         /*
 768          * Do not take mount writer counts on
 769          * special files since no writes to
 770          * the mount itself will occur.
 771          */
 772         if (!special_file(inode->i_mode)) {
 773                 /*
 774                  * Balanced in __fput()
 775                  */
 776                 error = mnt_want_write(mnt);
 777                 if (error)
 778                         put_write_access(inode);
 779         }
 780         return error;
 781 }
 782
 783 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 784                                         int flags, struct file *f,
 785                                         int (*open)(struct inode *, struct file *),
 786                                         const struct cred *cred)
 787 {
 788         struct inode *inode;
 789         int error;
 790
 791         f->f_flags = flags;
 792         f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 793                                 FMODE_PREAD | FMODE_PWRITE;
 794         inode = dentry->d_inode;
 795         if (f->f_mode & FMODE_WRITE) {
 796                 error = __get_file_write_access(inode, mnt);
 797                 if (error)
 798                         goto cleanup_file;
 799                 if (!special_file(inode->i_mode))
 800                         file_take_write(f);
 801         }
 802
 803         f->f_mapping = inode->i_mapping;
 804         f->f_path.dentry = dentry;
 805         f->f_path.mnt = mnt;
 806         f->f_pos = 0;
 807         f->f_op = fops_get(inode->i_fop);
 808         file_move(f, &inode->i_sb->s_files);
 809
 810         error = security_dentry_open(f, cred);
 811         if (error)
 812                 goto cleanup_all;
 813
 814         if (!open && f->f_op)
 815                 open = f->f_op->open;
 816         if (open) {
 817                 error = open(inode, f);
 818                 if (error)
 819                         goto cleanup_all;
 820         }
 821
 822         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 823
 824         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 825
 826         /* NB: we're sure to have correct a_ops only after f_op->open */
 827         if (f->f_flags & O_DIRECT) {
 828                 if (!f->f_mapping->a_ops ||
 829                     ((!f->f_mapping->a_ops->direct_IO) &&
 830                     (!f->f_mapping->a_ops->get_xip_mem))) {
 831                         fput(f);
 832                         f = ERR_PTR(-EINVAL);
 833                 }
 834         }
 835
 836         return f;
 837
 838 cleanup_all:
 839         fops_put(f->f_op);
 840         if (f->f_mode & FMODE_WRITE) {
 841                 put_write_access(inode);
 842                 if (!special_file(inode->i_mode)) {
 843                         /*
 844                          * We don't consider this a real
 845                          * mnt_want/drop_write() pair
 846                          * because it all happenend right
 847                          * here, so just reset the state.
 848                          */
 849                         file_reset_write(f);
 850                         mnt_drop_write(mnt);
 851                 }
 852         }
 853         file_kill(f);
 854         f->f_path.dentry = NULL;
 855         f->f_path.mnt = NULL;
 856 cleanup_file:
 857         put_filp(f);
 858         dput(dentry);
 859         mntput(mnt);
 860         return ERR_PTR(error);
 861 }
 862
 863 /**
 864  * lookup_instantiate_filp - instantiates the open intent filp
 865  * @nd: pointer to nameidata
 866  * @dentry: pointer to dentry
 867  * @open: open callback
 868  *
 869  * Helper for filesystems that want to use lookup open intents and pass back
 870  * a fully instantiated struct file to the caller.
 871  * This function is meant to be called from within a filesystem's
 872  * lookup method.
 873  * Beware of calling it for non-regular files! Those ->open methods might block
 874  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 875  * leading to a deadlock, as nobody can open that fifo anymore, because
 876  * another process to open fifo will block on locked parent when doing lookup).
 877  * Note that in case of error, nd->intent.open.file is destroyed, but the
 878  * path information remains valid.
 879  * If the open callback is set to NULL, then the standard f_op->open()
 880  * filesystem callback is substituted.
 881  */
 882 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 883                 int (*open)(struct inode *, struct file *))
 884 {
 885         const struct cred *cred = current_cred();
 886
 887         if (IS_ERR(nd->intent.open.file))
 888                 goto out;
 889         if (IS_ERR(dentry))
 890                 goto out_err;
 891         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 892                                              nd->intent.open.flags - 1,
 893                                              nd->intent.open.file,
 894                                              open, cred);
 895 out:
 896         return nd->intent.open.file;
 897 out_err:
 898         release_open_intent(nd);
 899         nd->intent.open.file = (struct file *)dentry;
 900         goto out;
 901 }
 902 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 903
 904 /**
 905  * nameidata_to_filp - convert a nameidata to an open filp.
 906  * @nd: pointer to nameidata
 907  * @flags: open flags
 908  *
 909  * Note that this function destroys the original nameidata
 910  */
 911 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 912 {
 913         const struct cred *cred = current_cred();
 914         struct file *filp;
 915
 916         /* Pick up the filp from the open intent */
 917         filp = nd->intent.open.file;
 918         /* Has the filesystem initialised the file for us? */
 919         if (filp->f_path.dentry == NULL)
 920                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 921                                      NULL, cred);
 922         else
 923                 path_put(&nd->path);
 924         return filp;
 925 }
 926
 927 /*
 928  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 929  * error.
 930  */
 931 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 932                          const struct cred *cred)
 933 {
 934         int error;
 935         struct file *f;
 936
 937         /*
 938          * We must always pass in a valid mount pointer.   Historically
 939          * callers got away with not passing it, but we must enforce this at
 940          * the earliest possible point now to avoid strange problems deep in the
 941          * filesystem stack.
 942          */
 943         if (!mnt) {
 944                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 945                 dump_stack();
 946                 return ERR_PTR(-EINVAL);
 947         }
 948
 949         error = -ENFILE;
 950         f = get_empty_filp();
 951         if (f == NULL) {
 952                 dput(dentry);
 953                 mntput(mnt);
 954                 return ERR_PTR(error);
 955         }
 956
 957         return __dentry_open(dentry, mnt, flags, f, NULL, cred);
 958 }
 959 EXPORT_SYMBOL(dentry_open);
 960
 961 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 962 {
 963         struct fdtable *fdt = files_fdtable(files);
 964         __FD_CLR(fd, fdt->open_fds);
 965         if (fd < files->next_fd)
 966                 files->next_fd = fd;
 967 }
 968
 969 void put_unused_fd(unsigned int fd)
 970 {
 971         struct files_struct *files = current->files;
 972         spin_lock(&files->file_lock);
 973         __put_unused_fd(files, fd);
 974         spin_unlock(&files->file_lock);
 975 }
 976
 977 EXPORT_SYMBOL(put_unused_fd);
 978
 979 /*
 980  * Install a file pointer in the fd array.
 981  *
 982  * The VFS is full of places where we drop the files lock between
 983  * setting the open_fds bitmap and installing the file in the file
 984  * array.  At any such point, we are vulnerable to a dup2() race
 985  * installing a file in the array before us.  We need to detect this and
 986  * fput() the struct file we are about to overwrite in this case.
 987  *
 988  * It should never happen - if we allow dup2() do it, _really_ bad things
 989  * will follow.
 990  */
 991
 992 void fd_install(unsigned int fd, struct file *file)
 993 {
 994         struct files_struct *files = current->files;
 995         struct fdtable *fdt;
 996         spin_lock(&files->file_lock);
 997         fdt = files_fdtable(files);
 998         BUG_ON(fdt->fd[fd] != NULL);
 999         rcu_assign_pointer(fdt->fd[fd], file);
1000         spin_unlock(&files->file_lock);
1001 }
1002
1003 EXPORT_SYMBOL(fd_install);
1004
1005 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1006 {
1007         char *tmp = getname(filename);
1008         int fd = PTR_ERR(tmp);
1009
1010         if (!IS_ERR(tmp)) {
1011                 fd = get_unused_fd_flags(flags);
1012                 if (fd >= 0) {
1013                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1014                         if (IS_ERR(f)) {
1015                                 put_unused_fd(fd);
1016                                 fd = PTR_ERR(f);
1017                         } else {
1018                                 fsnotify_open(f->f_path.dentry);
1019                                 fd_install(fd, f);
1020                         }
1021                 }
1022                 putname(tmp);
1023         }
1024         return fd;
1025 }
1026
1027 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1028 {
1029         long ret;
1030
1031         if (force_o_largefile())
1032                 flags |= O_LARGEFILE;
1033
1034         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1035         /* avoid REGPARM breakage on x86: */
1036         asmlinkage_protect(3, ret, filename, flags, mode);
1037         return ret;
1038 }
1039
1040 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1041                            int mode)
1042 {
1043         long ret;
1044
1045         if (force_o_largefile())
1046                 flags |= O_LARGEFILE;
1047
1048         ret = do_sys_open(dfd, filename, flags, mode);
1049         /* avoid REGPARM breakage on x86: */
1050         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1051         return ret;
1052 }
1053
1054 #ifndef __alpha__
1055
1056 /*
1057  * For backward compatibility?  Maybe this should be moved
1058  * into arch/i386 instead?
1059  */
1060 asmlinkage long sys_creat(const char __user * pathname, int mode)
1061 {
1062         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1063 }
1064
1065 #endif
1066
1067 /*
1068  * "id" is the POSIX thread ID. We use the
1069  * files pointer for this..
1070  */
1071 int filp_close(struct file *filp, fl_owner_t id)
1072 {
1073         int retval = 0;
1074
1075         if (!file_count(filp)) {
1076                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1077                 return 0;
1078         }
1079
1080         if (filp->f_op && filp->f_op->flush)
1081                 retval = filp->f_op->flush(filp, id);
1082
1083         dnotify_flush(filp, id);
1084         locks_remove_posix(filp, id);
1085         fput(filp);
1086         return retval;
1087 }
1088
1089 EXPORT_SYMBOL(filp_close);
1090
1091 /*
1092  * Careful here! We test whether the file pointer is NULL before
1093  * releasing the fd. This ensures that one clone task can't release
1094  * an fd while another clone is opening it.
1095  */
1096 asmlinkage long sys_close(unsigned int fd)
1097 {
1098         struct file * filp;
1099         struct files_struct *files = current->files;
1100         struct fdtable *fdt;
1101         int retval;
1102
1103         spin_lock(&files->file_lock);
1104         fdt = files_fdtable(files);
1105         if (fd >= fdt->max_fds)
1106                 goto out_unlock;
1107         filp = fdt->fd[fd];
1108         if (!filp)
1109                 goto out_unlock;
1110         rcu_assign_pointer(fdt->fd[fd], NULL);
1111         FD_CLR(fd, fdt->close_on_exec);
1112         __put_unused_fd(files, fd);
1113         spin_unlock(&files->file_lock);
1114         retval = filp_close(filp, files);
1115
1116         /* can't restart close syscall because file table entry was cleared */
1117         if (unlikely(retval == -ERESTARTSYS ||
1118                      retval == -ERESTARTNOINTR ||
1119                      retval == -ERESTARTNOHAND ||
1120                      retval == -ERESTART_RESTARTBLOCK))
1121                 retval = -EINTR;
1122
1123         return retval;
1124
1125 out_unlock:
1126         spin_unlock(&files->file_lock);
1127         return -EBADF;
1128 }
1129
1130 EXPORT_SYMBOL(sys_close);
1131
1132 /*
1133  * This routine simulates a hangup on the tty, to arrange that users
1134  * are given clean terminals at login time.
1135  */
1136 asmlinkage long sys_vhangup(void)
1137 {
1138         if (capable(CAP_SYS_TTY_CONFIG)) {
1139                 tty_vhangup_self();
1140                 return 0;
1141         }
1142         return -EPERM;
1143 }
1144
1145 /*
1146  * Called when an inode is about to be open.
1147  * We use this to disallow opening large files on 32bit systems if
1148  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1149  * on this flag in sys_open.
1150  */
1151 int generic_file_open(struct inode * inode, struct file * filp)
1152 {
1153         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1154                 return -EOVERFLOW;
1155         return 0;
1156 }
1157
1158 EXPORT_SYMBOL(generic_file_open);
1159
1160 /*
1161  * This is used by subsystems that don't want seekable
1162  * file descriptors
1163  */
1164 int nonseekable_open(struct inode *inode, struct file *filp)
1165 {
1166         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1167         return 0;
1168 }
1169
1170 EXPORT_SYMBOL(nonseekable_open);