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         struct path path;
 429         struct inode *inode;
 430         int old_fsuid, old_fsgid;
 431         kernel_cap_t uninitialized_var(old_cap);  /* !SECURE_NO_SETUID_FIXUP */
 432         int res;
 433
 434         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 435                 return -EINVAL;
 436
 437         old_fsuid = current->fsuid;
 438         old_fsgid = current->fsgid;
 439
 440         current->fsuid = current->uid;
 441         current->fsgid = current->gid;
 442
 443         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 444                 /*
 445                  * Clear the capabilities if we switch to a non-root user
 446                  */
 447 #ifndef CONFIG_SECURITY_FILE_CAPABILITIES
 448                 /*
 449                  * FIXME: There is a race here against sys_capset.  The
 450                  * capabilities can change yet we will restore the old
 451                  * value below.  We should hold task_capabilities_lock,
 452                  * but we cannot because user_path_at can sleep.
 453                  */
 454 #endif /* ndef CONFIG_SECURITY_FILE_CAPABILITIES */
 455                 if (current->uid)
 456                         old_cap = cap_set_effective(__cap_empty_set);
 457                 else
 458                         old_cap = cap_set_effective(current->cap_permitted);
 459         }
 460
 461         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 462         if (res)
 463                 goto out;
 464
 465         inode = path.dentry->d_inode;
 466
 467         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 468                 /*
 469                  * MAY_EXEC on regular files is denied if the fs is mounted
 470                  * with the "noexec" flag.
 471                  */
 472                 res = -EACCES;
 473                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 474                         goto out_path_release;
 475         }
 476
 477         res = inode_permission(inode, mode | MAY_ACCESS);
 478         /* SuS v2 requires we report a read only fs too */
 479         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 480                 goto out_path_release;
 481         /*
 482          * This is a rare case where using __mnt_is_readonly()
 483          * is OK without a mnt_want/drop_write() pair.  Since
 484          * no actual write to the fs is performed here, we do
 485          * not need to telegraph to that to anyone.
 486          *
 487          * By doing this, we accept that this access is
 488          * inherently racy and know that the fs may change
 489          * state before we even see this result.
 490          */
 491         if (__mnt_is_readonly(path.mnt))
 492                 res = -EROFS;
 493
 494 out_path_release:
 495         path_put(&path);
 496 out:
 497         current->fsuid = old_fsuid;
 498         current->fsgid = old_fsgid;
 499
 500         if (!issecure(SECURE_NO_SETUID_FIXUP))
 501                 cap_set_effective(old_cap);
 502
 503         return res;
 504 }
 505
 506 asmlinkage long sys_access(const char __user *filename, int mode)
 507 {
 508         return sys_faccessat(AT_FDCWD, filename, mode);
 509 }
 510
 511 asmlinkage long sys_chdir(const char __user * filename)
 512 {
 513         struct path path;
 514         int error;
 515
 516         error = user_path_dir(filename, &path);
 517         if (error)
 518                 goto out;
 519
 520         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 521         if (error)
 522                 goto dput_and_out;
 523
 524         set_fs_pwd(current->fs, &path);
 525
 526 dput_and_out:
 527         path_put(&path);
 528 out:
 529         return error;
 530 }
 531
 532 asmlinkage long sys_fchdir(unsigned int fd)
 533 {
 534         struct file *file;
 535         struct inode *inode;
 536         int error;
 537
 538         error = -EBADF;
 539         file = fget(fd);
 540         if (!file)
 541                 goto out;
 542
 543         inode = file->f_path.dentry->d_inode;
 544
 545         error = -ENOTDIR;
 546         if (!S_ISDIR(inode->i_mode))
 547                 goto out_putf;
 548
 549         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 550         if (!error)
 551                 set_fs_pwd(current->fs, &file->f_path);
 552 out_putf:
 553         fput(file);
 554 out:
 555         return error;
 556 }
 557
 558 asmlinkage long sys_chroot(const char __user * filename)
 559 {
 560         struct path path;
 561         int error;
 562
 563         error = user_path_dir(filename, &path);
 564         if (error)
 565                 goto out;
 566
 567         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 568         if (error)
 569                 goto dput_and_out;
 570
 571         error = -EPERM;
 572         if (!capable(CAP_SYS_CHROOT))
 573                 goto dput_and_out;
 574
 575         set_fs_root(current->fs, &path);
 576         error = 0;
 577 dput_and_out:
 578         path_put(&path);
 579 out:
 580         return error;
 581 }
 582
 583 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 584 {
 585         struct inode * inode;
 586         struct dentry * dentry;
 587         struct file * file;
 588         int err = -EBADF;
 589         struct iattr newattrs;
 590
 591         file = fget(fd);
 592         if (!file)
 593                 goto out;
 594
 595         dentry = file->f_path.dentry;
 596         inode = dentry->d_inode;
 597
 598         audit_inode(NULL, dentry);
 599
 600         err = mnt_want_write(file->f_path.mnt);
 601         if (err)
 602                 goto out_putf;
 603         mutex_lock(&inode->i_mutex);
 604         if (mode == (mode_t) -1)
 605                 mode = inode->i_mode;
 606         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 607         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 608         err = notify_change(dentry, &newattrs);
 609         mutex_unlock(&inode->i_mutex);
 610         mnt_drop_write(file->f_path.mnt);
 611 out_putf:
 612         fput(file);
 613 out:
 614         return err;
 615 }
 616
 617 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 618                              mode_t mode)
 619 {
 620         struct path path;
 621         struct inode *inode;
 622         int error;
 623         struct iattr newattrs;
 624
 625         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 626         if (error)
 627                 goto out;
 628         inode = path.dentry->d_inode;
 629
 630         error = mnt_want_write(path.mnt);
 631         if (error)
 632                 goto dput_and_out;
 633         mutex_lock(&inode->i_mutex);
 634         if (mode == (mode_t) -1)
 635                 mode = inode->i_mode;
 636         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 637         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 638         error = notify_change(path.dentry, &newattrs);
 639         mutex_unlock(&inode->i_mutex);
 640         mnt_drop_write(path.mnt);
 641 dput_and_out:
 642         path_put(&path);
 643 out:
 644         return error;
 645 }
 646
 647 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 648 {
 649         return sys_fchmodat(AT_FDCWD, filename, mode);
 650 }
 651
 652 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 653 {
 654         struct inode *inode = dentry->d_inode;
 655         int error;
 656         struct iattr newattrs;
 657
 658         newattrs.ia_valid =  ATTR_CTIME;
 659         if (user != (uid_t) -1) {
 660                 newattrs.ia_valid |= ATTR_UID;
 661                 newattrs.ia_uid = user;
 662         }
 663         if (group != (gid_t) -1) {
 664                 newattrs.ia_valid |= ATTR_GID;
 665                 newattrs.ia_gid = group;
 666         }
 667         if (!S_ISDIR(inode->i_mode))
 668                 newattrs.ia_valid |=
 669                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 670         mutex_lock(&inode->i_mutex);
 671         error = notify_change(dentry, &newattrs);
 672         mutex_unlock(&inode->i_mutex);
 673
 674         return error;
 675 }
 676
 677 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 678 {
 679         struct path path;
 680         int error;
 681
 682         error = user_path(filename, &path);
 683         if (error)
 684                 goto out;
 685         error = mnt_want_write(path.mnt);
 686         if (error)
 687                 goto out_release;
 688         error = chown_common(path.dentry, user, group);
 689         mnt_drop_write(path.mnt);
 690 out_release:
 691         path_put(&path);
 692 out:
 693         return error;
 694 }
 695
 696 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 697                              gid_t group, int flag)
 698 {
 699         struct path path;
 700         int error = -EINVAL;
 701         int follow;
 702
 703         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 704                 goto out;
 705
 706         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 707         error = user_path_at(dfd, filename, follow, &path);
 708         if (error)
 709                 goto out;
 710         error = mnt_want_write(path.mnt);
 711         if (error)
 712                 goto out_release;
 713         error = chown_common(path.dentry, user, group);
 714         mnt_drop_write(path.mnt);
 715 out_release:
 716         path_put(&path);
 717 out:
 718         return error;
 719 }
 720
 721 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 722 {
 723         struct path path;
 724         int error;
 725
 726         error = user_lpath(filename, &path);
 727         if (error)
 728                 goto out;
 729         error = mnt_want_write(path.mnt);
 730         if (error)
 731                 goto out_release;
 732         error = chown_common(path.dentry, user, group);
 733         mnt_drop_write(path.mnt);
 734 out_release:
 735         path_put(&path);
 736 out:
 737         return error;
 738 }
 739
 740
 741 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 742 {
 743         struct file * file;
 744         int error = -EBADF;
 745         struct dentry * dentry;
 746
 747         file = fget(fd);
 748         if (!file)
 749                 goto out;
 750
 751         error = mnt_want_write(file->f_path.mnt);
 752         if (error)
 753                 goto out_fput;
 754         dentry = file->f_path.dentry;
 755         audit_inode(NULL, dentry);
 756         error = chown_common(dentry, user, group);
 757         mnt_drop_write(file->f_path.mnt);
 758 out_fput:
 759         fput(file);
 760 out:
 761         return error;
 762 }
 763
 764 /*
 765  * You have to be very careful that these write
 766  * counts get cleaned up in error cases and
 767  * upon __fput().  This should probably never
 768  * be called outside of __dentry_open().
 769  */
 770 static inline int __get_file_write_access(struct inode *inode,
 771                                           struct vfsmount *mnt)
 772 {
 773         int error;
 774         error = get_write_access(inode);
 775         if (error)
 776                 return error;
 777         /*
 778          * Do not take mount writer counts on
 779          * special files since no writes to
 780          * the mount itself will occur.
 781          */
 782         if (!special_file(inode->i_mode)) {
 783                 /*
 784                  * Balanced in __fput()
 785                  */
 786                 error = mnt_want_write(mnt);
 787                 if (error)
 788                         put_write_access(inode);
 789         }
 790         return error;
 791 }
 792
 793 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 794                                         int flags, struct file *f,
 795                                         int (*open)(struct inode *, struct file *))
 796 {
 797         struct inode *inode;
 798         int error;
 799
 800         f->f_flags = flags;
 801         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 802                                 FMODE_PREAD | FMODE_PWRITE;
 803         inode = dentry->d_inode;
 804         if (f->f_mode & FMODE_WRITE) {
 805                 error = __get_file_write_access(inode, mnt);
 806                 if (error)
 807                         goto cleanup_file;
 808                 if (!special_file(inode->i_mode))
 809                         file_take_write(f);
 810         }
 811
 812         f->f_mapping = inode->i_mapping;
 813         f->f_path.dentry = dentry;
 814         f->f_path.mnt = mnt;
 815         f->f_pos = 0;
 816         f->f_op = fops_get(inode->i_fop);
 817         file_move(f, &inode->i_sb->s_files);
 818
 819         error = security_dentry_open(f);
 820         if (error)
 821                 goto cleanup_all;
 822
 823         if (!open && f->f_op)
 824                 open = f->f_op->open;
 825         if (open) {
 826                 error = open(inode, f);
 827                 if (error)
 828                         goto cleanup_all;
 829         }
 830
 831         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 832
 833         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 834
 835         /* NB: we're sure to have correct a_ops only after f_op->open */
 836         if (f->f_flags & O_DIRECT) {
 837                 if (!f->f_mapping->a_ops ||
 838                     ((!f->f_mapping->a_ops->direct_IO) &&
 839                     (!f->f_mapping->a_ops->get_xip_mem))) {
 840                         fput(f);
 841                         f = ERR_PTR(-EINVAL);
 842                 }
 843         }
 844
 845         return f;
 846
 847 cleanup_all:
 848         fops_put(f->f_op);
 849         if (f->f_mode & FMODE_WRITE) {
 850                 put_write_access(inode);
 851                 if (!special_file(inode->i_mode)) {
 852                         /*
 853                          * We don't consider this a real
 854                          * mnt_want/drop_write() pair
 855                          * because it all happenend right
 856                          * here, so just reset the state.
 857                          */
 858                         file_reset_write(f);
 859                         mnt_drop_write(mnt);
 860                 }
 861         }
 862         file_kill(f);
 863         f->f_path.dentry = NULL;
 864         f->f_path.mnt = NULL;
 865 cleanup_file:
 866         put_filp(f);
 867         dput(dentry);
 868         mntput(mnt);
 869         return ERR_PTR(error);
 870 }
 871
 872 /**
 873  * lookup_instantiate_filp - instantiates the open intent filp
 874  * @nd: pointer to nameidata
 875  * @dentry: pointer to dentry
 876  * @open: open callback
 877  *
 878  * Helper for filesystems that want to use lookup open intents and pass back
 879  * a fully instantiated struct file to the caller.
 880  * This function is meant to be called from within a filesystem's
 881  * lookup method.
 882  * Beware of calling it for non-regular files! Those ->open methods might block
 883  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 884  * leading to a deadlock, as nobody can open that fifo anymore, because
 885  * another process to open fifo will block on locked parent when doing lookup).
 886  * Note that in case of error, nd->intent.open.file is destroyed, but the
 887  * path information remains valid.
 888  * If the open callback is set to NULL, then the standard f_op->open()
 889  * filesystem callback is substituted.
 890  */
 891 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 892                 int (*open)(struct inode *, struct file *))
 893 {
 894         if (IS_ERR(nd->intent.open.file))
 895                 goto out;
 896         if (IS_ERR(dentry))
 897                 goto out_err;
 898         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 899                                              nd->intent.open.flags - 1,
 900                                              nd->intent.open.file,
 901                                              open);
 902 out:
 903         return nd->intent.open.file;
 904 out_err:
 905         release_open_intent(nd);
 906         nd->intent.open.file = (struct file *)dentry;
 907         goto out;
 908 }
 909 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 910
 911 /**
 912  * nameidata_to_filp - convert a nameidata to an open filp.
 913  * @nd: pointer to nameidata
 914  * @flags: open flags
 915  *
 916  * Note that this function destroys the original nameidata
 917  */
 918 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 919 {
 920         struct file *filp;
 921
 922         /* Pick up the filp from the open intent */
 923         filp = nd->intent.open.file;
 924         /* Has the filesystem initialised the file for us? */
 925         if (filp->f_path.dentry == NULL)
 926                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 927                                      NULL);
 928         else
 929                 path_put(&nd->path);
 930         return filp;
 931 }
 932
 933 /*
 934  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 935  * error.
 936  */
 937 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 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);
 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);