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                 error = security_path_truncate(&path, length, 0);
 277         if (!error) {
 278                 DQUOT_INIT(inode);
 279                 error = do_truncate(path.dentry, length, 0, NULL);
 280         }
 281
 282 put_write_and_out:
 283         put_write_access(inode);
 284 mnt_drop_write_and_out:
 285         mnt_drop_write(path.mnt);
 286 dput_and_out:
 287         path_put(&path);
 288 out:
 289         return error;
 290 }
 291
 292 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 293 {
 294         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 295         return do_sys_truncate(path, (long)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 asmlinkage long sys_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 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 355 {
 356         return do_sys_truncate(path, length);
 357 }
 358
 359 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 360 {
 361         long ret = do_sys_ftruncate(fd, length, 0);
 362         /* avoid REGPARM breakage on x86: */
 363         asmlinkage_protect(2, ret, fd, length);
 364         return ret;
 365 }
 366 #endif
 367
 368 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 369 {
 370         struct file *file;
 371         struct inode *inode;
 372         long ret = -EINVAL;
 373
 374         if (offset < 0 || len <= 0)
 375                 goto out;
 376
 377         /* Return error if mode is not supported */
 378         ret = -EOPNOTSUPP;
 379         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 380                 goto out;
 381
 382         ret = -EBADF;
 383         file = fget(fd);
 384         if (!file)
 385                 goto out;
 386         if (!(file->f_mode & FMODE_WRITE))
 387                 goto out_fput;
 388         /*
 389          * Revalidate the write permissions, in case security policy has
 390          * changed since the files were opened.
 391          */
 392         ret = security_file_permission(file, MAY_WRITE);
 393         if (ret)
 394                 goto out_fput;
 395
 396         inode = file->f_path.dentry->d_inode;
 397
 398         ret = -ESPIPE;
 399         if (S_ISFIFO(inode->i_mode))
 400                 goto out_fput;
 401
 402         ret = -ENODEV;
 403         /*
 404          * Let individual file system decide if it supports preallocation
 405          * for directories or not.
 406          */
 407         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 408                 goto out_fput;
 409
 410         ret = -EFBIG;
 411         /* Check for wrap through zero too */
 412         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 413                 goto out_fput;
 414
 415         if (inode->i_op && inode->i_op->fallocate)
 416                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 417         else
 418                 ret = -EOPNOTSUPP;
 419
 420 out_fput:
 421         fput(file);
 422 out:
 423         return ret;
 424 }
 425
 426 /*
 427  * access() needs to use the real uid/gid, not the effective uid/gid.
 428  * We do this by temporarily clearing all FS-related capabilities and
 429  * switching the fsuid/fsgid around to the real ones.
 430  */
 431 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 432 {
 433         const struct cred *old_cred;
 434         struct cred *override_cred;
 435         struct path path;
 436         struct inode *inode;
 437         int res;
 438
 439         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 440                 return -EINVAL;
 441
 442         override_cred = prepare_creds();
 443         if (!override_cred)
 444                 return -ENOMEM;
 445
 446         override_cred->fsuid = override_cred->uid;
 447         override_cred->fsgid = override_cred->gid;
 448
 449         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 450                 /* Clear the capabilities if we switch to a non-root user */
 451                 if (override_cred->uid)
 452                         cap_clear(override_cred->cap_effective);
 453                 else
 454                         override_cred->cap_effective =
 455                                 override_cred->cap_permitted;
 456         }
 457
 458         old_cred = override_creds(override_cred);
 459
 460         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 461         if (res)
 462                 goto out;
 463
 464         inode = path.dentry->d_inode;
 465
 466         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 467                 /*
 468                  * MAY_EXEC on regular files is denied if the fs is mounted
 469                  * with the "noexec" flag.
 470                  */
 471                 res = -EACCES;
 472                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 473                         goto out_path_release;
 474         }
 475
 476         res = inode_permission(inode, mode | MAY_ACCESS);
 477         /* SuS v2 requires we report a read only fs too */
 478         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 479                 goto out_path_release;
 480         /*
 481          * This is a rare case where using __mnt_is_readonly()
 482          * is OK without a mnt_want/drop_write() pair.  Since
 483          * no actual write to the fs is performed here, we do
 484          * not need to telegraph to that to anyone.
 485          *
 486          * By doing this, we accept that this access is
 487          * inherently racy and know that the fs may change
 488          * state before we even see this result.
 489          */
 490         if (__mnt_is_readonly(path.mnt))
 491                 res = -EROFS;
 492
 493 out_path_release:
 494         path_put(&path);
 495 out:
 496         revert_creds(old_cred);
 497         put_cred(override_cred);
 498         return res;
 499 }
 500
 501 asmlinkage long sys_access(const char __user *filename, int mode)
 502 {
 503         return sys_faccessat(AT_FDCWD, filename, mode);
 504 }
 505
 506 asmlinkage long sys_chdir(const char __user * filename)
 507 {
 508         struct path path;
 509         int error;
 510
 511         error = user_path_dir(filename, &path);
 512         if (error)
 513                 goto out;
 514
 515         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 516         if (error)
 517                 goto dput_and_out;
 518
 519         set_fs_pwd(current->fs, &path);
 520
 521 dput_and_out:
 522         path_put(&path);
 523 out:
 524         return error;
 525 }
 526
 527 asmlinkage long sys_fchdir(unsigned int fd)
 528 {
 529         struct file *file;
 530         struct inode *inode;
 531         int error;
 532
 533         error = -EBADF;
 534         file = fget(fd);
 535         if (!file)
 536                 goto out;
 537
 538         inode = file->f_path.dentry->d_inode;
 539
 540         error = -ENOTDIR;
 541         if (!S_ISDIR(inode->i_mode))
 542                 goto out_putf;
 543
 544         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 545         if (!error)
 546                 set_fs_pwd(current->fs, &file->f_path);
 547 out_putf:
 548         fput(file);
 549 out:
 550         return error;
 551 }
 552
 553 asmlinkage long sys_chroot(const char __user * filename)
 554 {
 555         struct path path;
 556         int error;
 557
 558         error = user_path_dir(filename, &path);
 559         if (error)
 560                 goto out;
 561
 562         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 563         if (error)
 564                 goto dput_and_out;
 565
 566         error = -EPERM;
 567         if (!capable(CAP_SYS_CHROOT))
 568                 goto dput_and_out;
 569
 570         set_fs_root(current->fs, &path);
 571         error = 0;
 572 dput_and_out:
 573         path_put(&path);
 574 out:
 575         return error;
 576 }
 577
 578 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 579 {
 580         struct inode * inode;
 581         struct dentry * dentry;
 582         struct file * file;
 583         int err = -EBADF;
 584         struct iattr newattrs;
 585
 586         file = fget(fd);
 587         if (!file)
 588                 goto out;
 589
 590         dentry = file->f_path.dentry;
 591         inode = dentry->d_inode;
 592
 593         audit_inode(NULL, dentry);
 594
 595         err = mnt_want_write(file->f_path.mnt);
 596         if (err)
 597                 goto out_putf;
 598         mutex_lock(&inode->i_mutex);
 599         if (mode == (mode_t) -1)
 600                 mode = inode->i_mode;
 601         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 602         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 603         err = notify_change(dentry, &newattrs);
 604         mutex_unlock(&inode->i_mutex);
 605         mnt_drop_write(file->f_path.mnt);
 606 out_putf:
 607         fput(file);
 608 out:
 609         return err;
 610 }
 611
 612 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 613                              mode_t mode)
 614 {
 615         struct path path;
 616         struct inode *inode;
 617         int error;
 618         struct iattr newattrs;
 619
 620         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 621         if (error)
 622                 goto out;
 623         inode = path.dentry->d_inode;
 624
 625         error = mnt_want_write(path.mnt);
 626         if (error)
 627                 goto dput_and_out;
 628         mutex_lock(&inode->i_mutex);
 629         if (mode == (mode_t) -1)
 630                 mode = inode->i_mode;
 631         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 632         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 633         error = notify_change(path.dentry, &newattrs);
 634         mutex_unlock(&inode->i_mutex);
 635         mnt_drop_write(path.mnt);
 636 dput_and_out:
 637         path_put(&path);
 638 out:
 639         return error;
 640 }
 641
 642 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 643 {
 644         return sys_fchmodat(AT_FDCWD, filename, mode);
 645 }
 646
 647 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 648 {
 649         struct inode *inode = dentry->d_inode;
 650         int error;
 651         struct iattr newattrs;
 652
 653         newattrs.ia_valid =  ATTR_CTIME;
 654         if (user != (uid_t) -1) {
 655                 newattrs.ia_valid |= ATTR_UID;
 656                 newattrs.ia_uid = user;
 657         }
 658         if (group != (gid_t) -1) {
 659                 newattrs.ia_valid |= ATTR_GID;
 660                 newattrs.ia_gid = group;
 661         }
 662         if (!S_ISDIR(inode->i_mode))
 663                 newattrs.ia_valid |=
 664                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 665         mutex_lock(&inode->i_mutex);
 666         error = notify_change(dentry, &newattrs);
 667         mutex_unlock(&inode->i_mutex);
 668
 669         return error;
 670 }
 671
 672 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 673 {
 674         struct path path;
 675         int error;
 676
 677         error = user_path(filename, &path);
 678         if (error)
 679                 goto out;
 680         error = mnt_want_write(path.mnt);
 681         if (error)
 682                 goto out_release;
 683         error = chown_common(path.dentry, user, group);
 684         mnt_drop_write(path.mnt);
 685 out_release:
 686         path_put(&path);
 687 out:
 688         return error;
 689 }
 690
 691 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 692                              gid_t group, int flag)
 693 {
 694         struct path path;
 695         int error = -EINVAL;
 696         int follow;
 697
 698         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 699                 goto out;
 700
 701         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 702         error = user_path_at(dfd, filename, follow, &path);
 703         if (error)
 704                 goto out;
 705         error = mnt_want_write(path.mnt);
 706         if (error)
 707                 goto out_release;
 708         error = chown_common(path.dentry, user, group);
 709         mnt_drop_write(path.mnt);
 710 out_release:
 711         path_put(&path);
 712 out:
 713         return error;
 714 }
 715
 716 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 717 {
 718         struct path path;
 719         int error;
 720
 721         error = user_lpath(filename, &path);
 722         if (error)
 723                 goto out;
 724         error = mnt_want_write(path.mnt);
 725         if (error)
 726                 goto out_release;
 727         error = chown_common(path.dentry, user, group);
 728         mnt_drop_write(path.mnt);
 729 out_release:
 730         path_put(&path);
 731 out:
 732         return error;
 733 }
 734
 735
 736 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 737 {
 738         struct file * file;
 739         int error = -EBADF;
 740         struct dentry * dentry;
 741
 742         file = fget(fd);
 743         if (!file)
 744                 goto out;
 745
 746         error = mnt_want_write(file->f_path.mnt);
 747         if (error)
 748                 goto out_fput;
 749         dentry = file->f_path.dentry;
 750         audit_inode(NULL, dentry);
 751         error = chown_common(dentry, user, group);
 752         mnt_drop_write(file->f_path.mnt);
 753 out_fput:
 754         fput(file);
 755 out:
 756         return error;
 757 }
 758
 759 /*
 760  * You have to be very careful that these write
 761  * counts get cleaned up in error cases and
 762  * upon __fput().  This should probably never
 763  * be called outside of __dentry_open().
 764  */
 765 static inline int __get_file_write_access(struct inode *inode,
 766                                           struct vfsmount *mnt)
 767 {
 768         int error;
 769         error = get_write_access(inode);
 770         if (error)
 771                 return error;
 772         /*
 773          * Do not take mount writer counts on
 774          * special files since no writes to
 775          * the mount itself will occur.
 776          */
 777         if (!special_file(inode->i_mode)) {
 778                 /*
 779                  * Balanced in __fput()
 780                  */
 781                 error = mnt_want_write(mnt);
 782                 if (error)
 783                         put_write_access(inode);
 784         }
 785         return error;
 786 }
 787
 788 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 789                                         int flags, struct file *f,
 790                                         int (*open)(struct inode *, struct file *),
 791                                         const struct cred *cred)
 792 {
 793         struct inode *inode;
 794         int error;
 795
 796         f->f_flags = flags;
 797         f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 798                                 FMODE_PREAD | FMODE_PWRITE;
 799         inode = dentry->d_inode;
 800         if (f->f_mode & FMODE_WRITE) {
 801                 error = __get_file_write_access(inode, mnt);
 802                 if (error)
 803                         goto cleanup_file;
 804                 if (!special_file(inode->i_mode))
 805                         file_take_write(f);
 806         }
 807
 808         f->f_mapping = inode->i_mapping;
 809         f->f_path.dentry = dentry;
 810         f->f_path.mnt = mnt;
 811         f->f_pos = 0;
 812         f->f_op = fops_get(inode->i_fop);
 813         file_move(f, &inode->i_sb->s_files);
 814
 815         error = security_dentry_open(f, cred);
 816         if (error)
 817                 goto cleanup_all;
 818
 819         if (!open && f->f_op)
 820                 open = f->f_op->open;
 821         if (open) {
 822                 error = open(inode, f);
 823                 if (error)
 824                         goto cleanup_all;
 825         }
 826
 827         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 828
 829         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 830
 831         /* NB: we're sure to have correct a_ops only after f_op->open */
 832         if (f->f_flags & O_DIRECT) {
 833                 if (!f->f_mapping->a_ops ||
 834                     ((!f->f_mapping->a_ops->direct_IO) &&
 835                     (!f->f_mapping->a_ops->get_xip_mem))) {
 836                         fput(f);
 837                         f = ERR_PTR(-EINVAL);
 838                 }
 839         }
 840
 841         return f;
 842
 843 cleanup_all:
 844         fops_put(f->f_op);
 845         if (f->f_mode & FMODE_WRITE) {
 846                 put_write_access(inode);
 847                 if (!special_file(inode->i_mode)) {
 848                         /*
 849                          * We don't consider this a real
 850                          * mnt_want/drop_write() pair
 851                          * because it all happenend right
 852                          * here, so just reset the state.
 853                          */
 854                         file_reset_write(f);
 855                         mnt_drop_write(mnt);
 856                 }
 857         }
 858         file_kill(f);
 859         f->f_path.dentry = NULL;
 860         f->f_path.mnt = NULL;
 861 cleanup_file:
 862         put_filp(f);
 863         dput(dentry);
 864         mntput(mnt);
 865         return ERR_PTR(error);
 866 }
 867
 868 /**
 869  * lookup_instantiate_filp - instantiates the open intent filp
 870  * @nd: pointer to nameidata
 871  * @dentry: pointer to dentry
 872  * @open: open callback
 873  *
 874  * Helper for filesystems that want to use lookup open intents and pass back
 875  * a fully instantiated struct file to the caller.
 876  * This function is meant to be called from within a filesystem's
 877  * lookup method.
 878  * Beware of calling it for non-regular files! Those ->open methods might block
 879  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 880  * leading to a deadlock, as nobody can open that fifo anymore, because
 881  * another process to open fifo will block on locked parent when doing lookup).
 882  * Note that in case of error, nd->intent.open.file is destroyed, but the
 883  * path information remains valid.
 884  * If the open callback is set to NULL, then the standard f_op->open()
 885  * filesystem callback is substituted.
 886  */
 887 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 888                 int (*open)(struct inode *, struct file *))
 889 {
 890         const struct cred *cred = current_cred();
 891
 892         if (IS_ERR(nd->intent.open.file))
 893                 goto out;
 894         if (IS_ERR(dentry))
 895                 goto out_err;
 896         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 897                                              nd->intent.open.flags - 1,
 898                                              nd->intent.open.file,
 899                                              open, cred);
 900 out:
 901         return nd->intent.open.file;
 902 out_err:
 903         release_open_intent(nd);
 904         nd->intent.open.file = (struct file *)dentry;
 905         goto out;
 906 }
 907 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 908
 909 /**
 910  * nameidata_to_filp - convert a nameidata to an open filp.
 911  * @nd: pointer to nameidata
 912  * @flags: open flags
 913  *
 914  * Note that this function destroys the original nameidata
 915  */
 916 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 917 {
 918         const struct cred *cred = current_cred();
 919         struct file *filp;
 920
 921         /* Pick up the filp from the open intent */
 922         filp = nd->intent.open.file;
 923         /* Has the filesystem initialised the file for us? */
 924         if (filp->f_path.dentry == NULL)
 925                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 926                                      NULL, cred);
 927         else
 928                 path_put(&nd->path);
 929         return filp;
 930 }
 931
 932 /*
 933  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 934  * error.
 935  */
 936 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 937                          const struct cred *cred)
 938 {
 939         int error;
 940         struct file *f;
 941
 942         /*
 943          * We must always pass in a valid mount pointer.   Historically
 944          * callers got away with not passing it, but we must enforce this at
 945          * the earliest possible point now to avoid strange problems deep in the
 946          * filesystem stack.
 947          */
 948         if (!mnt) {
 949                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 950                 dump_stack();
 951                 return ERR_PTR(-EINVAL);
 952         }
 953
 954         error = -ENFILE;
 955         f = get_empty_filp();
 956         if (f == NULL) {
 957                 dput(dentry);
 958                 mntput(mnt);
 959                 return ERR_PTR(error);
 960         }
 961
 962         return __dentry_open(dentry, mnt, flags, f, NULL, cred);
 963 }
 964 EXPORT_SYMBOL(dentry_open);
 965
 966 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 967 {
 968         struct fdtable *fdt = files_fdtable(files);
 969         __FD_CLR(fd, fdt->open_fds);
 970         if (fd < files->next_fd)
 971                 files->next_fd = fd;
 972 }
 973
 974 void put_unused_fd(unsigned int fd)
 975 {
 976         struct files_struct *files = current->files;
 977         spin_lock(&files->file_lock);
 978         __put_unused_fd(files, fd);
 979         spin_unlock(&files->file_lock);
 980 }
 981
 982 EXPORT_SYMBOL(put_unused_fd);
 983
 984 /*
 985  * Install a file pointer in the fd array.
 986  *
 987  * The VFS is full of places where we drop the files lock between
 988  * setting the open_fds bitmap and installing the file in the file
 989  * array.  At any such point, we are vulnerable to a dup2() race
 990  * installing a file in the array before us.  We need to detect this and
 991  * fput() the struct file we are about to overwrite in this case.
 992  *
 993  * It should never happen - if we allow dup2() do it, _really_ bad things
 994  * will follow.
 995  */
 996
 997 void fd_install(unsigned int fd, struct file *file)
 998 {
 999         struct files_struct *files = current->files;
1000         struct fdtable *fdt;
1001         spin_lock(&files->file_lock);
1002         fdt = files_fdtable(files);
1003         BUG_ON(fdt->fd[fd] != NULL);
1004         rcu_assign_pointer(fdt->fd[fd], file);
1005         spin_unlock(&files->file_lock);
1006 }
1007
1008 EXPORT_SYMBOL(fd_install);
1009
1010 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1011 {
1012         char *tmp = getname(filename);
1013         int fd = PTR_ERR(tmp);
1014
1015         if (!IS_ERR(tmp)) {
1016                 fd = get_unused_fd_flags(flags);
1017                 if (fd >= 0) {
1018                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1019                         if (IS_ERR(f)) {
1020                                 put_unused_fd(fd);
1021                                 fd = PTR_ERR(f);
1022                         } else {
1023                                 fsnotify_open(f->f_path.dentry);
1024                                 fd_install(fd, f);
1025                         }
1026                 }
1027                 putname(tmp);
1028         }
1029         return fd;
1030 }
1031
1032 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1033 {
1034         long ret;
1035
1036         if (force_o_largefile())
1037                 flags |= O_LARGEFILE;
1038
1039         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1040         /* avoid REGPARM breakage on x86: */
1041         asmlinkage_protect(3, ret, filename, flags, mode);
1042         return ret;
1043 }
1044
1045 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1046                            int mode)
1047 {
1048         long ret;
1049
1050         if (force_o_largefile())
1051                 flags |= O_LARGEFILE;
1052
1053         ret = do_sys_open(dfd, filename, flags, mode);
1054         /* avoid REGPARM breakage on x86: */
1055         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1056         return ret;
1057 }
1058
1059 #ifndef __alpha__
1060
1061 /*
1062  * For backward compatibility?  Maybe this should be moved
1063  * into arch/i386 instead?
1064  */
1065 asmlinkage long sys_creat(const char __user * pathname, int mode)
1066 {
1067         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1068 }
1069
1070 #endif
1071
1072 /*
1073  * "id" is the POSIX thread ID. We use the
1074  * files pointer for this..
1075  */
1076 int filp_close(struct file *filp, fl_owner_t id)
1077 {
1078         int retval = 0;
1079
1080         if (!file_count(filp)) {
1081                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1082                 return 0;
1083         }
1084
1085         if (filp->f_op && filp->f_op->flush)
1086                 retval = filp->f_op->flush(filp, id);
1087
1088         dnotify_flush(filp, id);
1089         locks_remove_posix(filp, id);
1090         fput(filp);
1091         return retval;
1092 }
1093
1094 EXPORT_SYMBOL(filp_close);
1095
1096 /*
1097  * Careful here! We test whether the file pointer is NULL before
1098  * releasing the fd. This ensures that one clone task can't release
1099  * an fd while another clone is opening it.
1100  */
1101 asmlinkage long sys_close(unsigned int fd)
1102 {
1103         struct file * filp;
1104         struct files_struct *files = current->files;
1105         struct fdtable *fdt;
1106         int retval;
1107
1108         spin_lock(&files->file_lock);
1109         fdt = files_fdtable(files);
1110         if (fd >= fdt->max_fds)
1111                 goto out_unlock;
1112         filp = fdt->fd[fd];
1113         if (!filp)
1114                 goto out_unlock;
1115         rcu_assign_pointer(fdt->fd[fd], NULL);
1116         FD_CLR(fd, fdt->close_on_exec);
1117         __put_unused_fd(files, fd);
1118         spin_unlock(&files->file_lock);
1119         retval = filp_close(filp, files);
1120
1121         /* can't restart close syscall because file table entry was cleared */
1122         if (unlikely(retval == -ERESTARTSYS ||
1123                      retval == -ERESTARTNOINTR ||
1124                      retval == -ERESTARTNOHAND ||
1125                      retval == -ERESTART_RESTARTBLOCK))
1126                 retval = -EINTR;
1127
1128         return retval;
1129
1130 out_unlock:
1131         spin_unlock(&files->file_lock);
1132         return -EBADF;
1133 }
1134
1135 EXPORT_SYMBOL(sys_close);
1136
1137 /*
1138  * This routine simulates a hangup on the tty, to arrange that users
1139  * are given clean terminals at login time.
1140  */
1141 asmlinkage long sys_vhangup(void)
1142 {
1143         if (capable(CAP_SYS_TTY_CONFIG)) {
1144                 tty_vhangup_self();
1145                 return 0;
1146         }
1147         return -EPERM;
1148 }
1149
1150 /*
1151  * Called when an inode is about to be open.
1152  * We use this to disallow opening large files on 32bit systems if
1153  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1154  * on this flag in sys_open.
1155  */
1156 int generic_file_open(struct inode * inode, struct file * filp)
1157 {
1158         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1159                 return -EOVERFLOW;
1160         return 0;
1161 }
1162
1163 EXPORT_SYMBOL(generic_file_open);
1164
1165 /*
1166  * This is used by subsystems that don't want seekable
1167  * file descriptors
1168  */
1169 int nonseekable_open(struct inode *inode, struct file *filp)
1170 {
1171         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1172         return 0;
1173 }
1174
1175 EXPORT_SYMBOL(nonseekable_open);