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 SYSCALL_DEFINE2(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 SYSCALL_DEFINE3(statfs64, const char __user *, pathname, size_t, sz, struct statfs64 __user *, buf)
 142 {
 143         struct path path;
 144         long error;
 145
 146         if (sz != sizeof(*buf))
 147                 return -EINVAL;
 148         error = user_path(pathname, &path);
 149         if (!error) {
 150                 struct statfs64 tmp;
 151                 error = vfs_statfs64(path.dentry, &tmp);
 152                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 153                         error = -EFAULT;
 154                 path_put(&path);
 155         }
 156         return error;
 157 }
 158
 159 SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct statfs __user *, buf)
 160 {
 161         struct file * file;
 162         struct statfs tmp;
 163         int error;
 164
 165         error = -EBADF;
 166         file = fget(fd);
 167         if (!file)
 168                 goto out;
 169         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 170         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 171                 error = -EFAULT;
 172         fput(file);
 173 out:
 174         return error;
 175 }
 176
 177 SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, size_t, sz, struct statfs64 __user *, buf)
 178 {
 179         struct file * file;
 180         struct statfs64 tmp;
 181         int error;
 182
 183         if (sz != sizeof(*buf))
 184                 return -EINVAL;
 185
 186         error = -EBADF;
 187         file = fget(fd);
 188         if (!file)
 189                 goto out;
 190         error = vfs_statfs64(file->f_path.dentry, &tmp);
 191         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 192                 error = -EFAULT;
 193         fput(file);
 194 out:
 195         return error;
 196 }
 197
 198 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 199         struct file *filp)
 200 {
 201         int err;
 202         struct iattr newattrs;
 203
 204         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 205         if (length < 0)
 206                 return -EINVAL;
 207
 208         newattrs.ia_size = length;
 209         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 210         if (filp) {
 211                 newattrs.ia_file = filp;
 212                 newattrs.ia_valid |= ATTR_FILE;
 213         }
 214
 215         /* Remove suid/sgid on truncate too */
 216         newattrs.ia_valid |= should_remove_suid(dentry);
 217
 218         mutex_lock(&dentry->d_inode->i_mutex);
 219         err = notify_change(dentry, &newattrs);
 220         mutex_unlock(&dentry->d_inode->i_mutex);
 221         return err;
 222 }
 223
 224 static long do_sys_truncate(const char __user *pathname, loff_t length)
 225 {
 226         struct path path;
 227         struct inode *inode;
 228         int error;
 229
 230         error = -EINVAL;
 231         if (length < 0) /* sorry, but loff_t says... */
 232                 goto out;
 233
 234         error = user_path(pathname, &path);
 235         if (error)
 236                 goto out;
 237         inode = path.dentry->d_inode;
 238
 239         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 240         error = -EISDIR;
 241         if (S_ISDIR(inode->i_mode))
 242                 goto dput_and_out;
 243
 244         error = -EINVAL;
 245         if (!S_ISREG(inode->i_mode))
 246                 goto dput_and_out;
 247
 248         error = mnt_want_write(path.mnt);
 249         if (error)
 250                 goto dput_and_out;
 251
 252         error = inode_permission(inode, MAY_WRITE);
 253         if (error)
 254                 goto mnt_drop_write_and_out;
 255
 256         error = -EPERM;
 257         if (IS_APPEND(inode))
 258                 goto mnt_drop_write_and_out;
 259
 260         error = get_write_access(inode);
 261         if (error)
 262                 goto mnt_drop_write_and_out;
 263
 264         /*
 265          * Make sure that there are no leases.  get_write_access() protects
 266          * against the truncate racing with a lease-granting setlease().
 267          */
 268         error = break_lease(inode, FMODE_WRITE);
 269         if (error)
 270                 goto put_write_and_out;
 271
 272         error = locks_verify_truncate(inode, NULL, length);
 273         if (!error)
 274                 error = security_path_truncate(&path, length, 0);
 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 SYSCALL_DEFINE2(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 = security_path_truncate(&file->f_path, length,
 333                                                ATTR_MTIME|ATTR_CTIME);
 334         if (!error)
 335                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 336 out_putf:
 337         fput(file);
 338 out:
 339         return error;
 340 }
 341
 342 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 343 {
 344         long ret = do_sys_ftruncate(fd, length, 1);
 345         /* avoid REGPARM breakage on x86: */
 346         asmlinkage_protect(2, ret, fd, length);
 347         return ret;
 348 }
 349
 350 /* LFS versions of truncate are only needed on 32 bit machines */
 351 #if BITS_PER_LONG == 32
 352 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 353 {
 354         return do_sys_truncate(path, length);
 355 }
 356 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 357 asmlinkage long SyS_truncate64(long path, loff_t length)
 358 {
 359         return SYSC_truncate64((const char __user *) path, length);
 360 }
 361 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 362 #endif
 363
 364 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 365 {
 366         long ret = do_sys_ftruncate(fd, length, 0);
 367         /* avoid REGPARM breakage on x86: */
 368         asmlinkage_protect(2, ret, fd, length);
 369         return ret;
 370 }
 371 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 372 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 373 {
 374         return SYSC_ftruncate64((unsigned int) fd, length);
 375 }
 376 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 377 #endif
 378 #endif /* BITS_PER_LONG == 32 */
 379
 380 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 381 {
 382         struct file *file;
 383         struct inode *inode;
 384         long ret = -EINVAL;
 385
 386         if (offset < 0 || len <= 0)
 387                 goto out;
 388
 389         /* Return error if mode is not supported */
 390         ret = -EOPNOTSUPP;
 391         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 392                 goto out;
 393
 394         ret = -EBADF;
 395         file = fget(fd);
 396         if (!file)
 397                 goto out;
 398         if (!(file->f_mode & FMODE_WRITE))
 399                 goto out_fput;
 400         /*
 401          * Revalidate the write permissions, in case security policy has
 402          * changed since the files were opened.
 403          */
 404         ret = security_file_permission(file, MAY_WRITE);
 405         if (ret)
 406                 goto out_fput;
 407
 408         inode = file->f_path.dentry->d_inode;
 409
 410         ret = -ESPIPE;
 411         if (S_ISFIFO(inode->i_mode))
 412                 goto out_fput;
 413
 414         ret = -ENODEV;
 415         /*
 416          * Let individual file system decide if it supports preallocation
 417          * for directories or not.
 418          */
 419         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 420                 goto out_fput;
 421
 422         ret = -EFBIG;
 423         /* Check for wrap through zero too */
 424         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 425                 goto out_fput;
 426
 427         if (inode->i_op->fallocate)
 428                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 429         else
 430                 ret = -EOPNOTSUPP;
 431
 432 out_fput:
 433         fput(file);
 434 out:
 435         return ret;
 436 }
 437 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 438 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 439 {
 440         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 441 }
 442 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 443 #endif
 444
 445 /*
 446  * access() needs to use the real uid/gid, not the effective uid/gid.
 447  * We do this by temporarily clearing all FS-related capabilities and
 448  * switching the fsuid/fsgid around to the real ones.
 449  */
 450 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 451 {
 452         const struct cred *old_cred;
 453         struct cred *override_cred;
 454         struct path path;
 455         struct inode *inode;
 456         int res;
 457
 458         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 459                 return -EINVAL;
 460
 461         override_cred = prepare_creds();
 462         if (!override_cred)
 463                 return -ENOMEM;
 464
 465         override_cred->fsuid = override_cred->uid;
 466         override_cred->fsgid = override_cred->gid;
 467
 468         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 469                 /* Clear the capabilities if we switch to a non-root user */
 470                 if (override_cred->uid)
 471                         cap_clear(override_cred->cap_effective);
 472                 else
 473                         override_cred->cap_effective =
 474                                 override_cred->cap_permitted;
 475         }
 476
 477         old_cred = override_creds(override_cred);
 478
 479         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 480         if (res)
 481                 goto out;
 482
 483         inode = path.dentry->d_inode;
 484
 485         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 486                 /*
 487                  * MAY_EXEC on regular files is denied if the fs is mounted
 488                  * with the "noexec" flag.
 489                  */
 490                 res = -EACCES;
 491                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 492                         goto out_path_release;
 493         }
 494
 495         res = inode_permission(inode, mode | MAY_ACCESS);
 496         /* SuS v2 requires we report a read only fs too */
 497         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 498                 goto out_path_release;
 499         /*
 500          * This is a rare case where using __mnt_is_readonly()
 501          * is OK without a mnt_want/drop_write() pair.  Since
 502          * no actual write to the fs is performed here, we do
 503          * not need to telegraph to that to anyone.
 504          *
 505          * By doing this, we accept that this access is
 506          * inherently racy and know that the fs may change
 507          * state before we even see this result.
 508          */
 509         if (__mnt_is_readonly(path.mnt))
 510                 res = -EROFS;
 511
 512 out_path_release:
 513         path_put(&path);
 514 out:
 515         revert_creds(old_cred);
 516         put_cred(override_cred);
 517         return res;
 518 }
 519
 520 asmlinkage long sys_access(const char __user *filename, int mode)
 521 {
 522         return sys_faccessat(AT_FDCWD, filename, mode);
 523 }
 524
 525 asmlinkage long sys_chdir(const char __user * filename)
 526 {
 527         struct path path;
 528         int error;
 529
 530         error = user_path_dir(filename, &path);
 531         if (error)
 532                 goto out;
 533
 534         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 535         if (error)
 536                 goto dput_and_out;
 537
 538         set_fs_pwd(current->fs, &path);
 539
 540 dput_and_out:
 541         path_put(&path);
 542 out:
 543         return error;
 544 }
 545
 546 asmlinkage long sys_fchdir(unsigned int fd)
 547 {
 548         struct file *file;
 549         struct inode *inode;
 550         int error;
 551
 552         error = -EBADF;
 553         file = fget(fd);
 554         if (!file)
 555                 goto out;
 556
 557         inode = file->f_path.dentry->d_inode;
 558
 559         error = -ENOTDIR;
 560         if (!S_ISDIR(inode->i_mode))
 561                 goto out_putf;
 562
 563         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 564         if (!error)
 565                 set_fs_pwd(current->fs, &file->f_path);
 566 out_putf:
 567         fput(file);
 568 out:
 569         return error;
 570 }
 571
 572 asmlinkage long sys_chroot(const char __user * filename)
 573 {
 574         struct path path;
 575         int error;
 576
 577         error = user_path_dir(filename, &path);
 578         if (error)
 579                 goto out;
 580
 581         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 582         if (error)
 583                 goto dput_and_out;
 584
 585         error = -EPERM;
 586         if (!capable(CAP_SYS_CHROOT))
 587                 goto dput_and_out;
 588
 589         set_fs_root(current->fs, &path);
 590         error = 0;
 591 dput_and_out:
 592         path_put(&path);
 593 out:
 594         return error;
 595 }
 596
 597 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 598 {
 599         struct inode * inode;
 600         struct dentry * dentry;
 601         struct file * file;
 602         int err = -EBADF;
 603         struct iattr newattrs;
 604
 605         file = fget(fd);
 606         if (!file)
 607                 goto out;
 608
 609         dentry = file->f_path.dentry;
 610         inode = dentry->d_inode;
 611
 612         audit_inode(NULL, dentry);
 613
 614         err = mnt_want_write(file->f_path.mnt);
 615         if (err)
 616                 goto out_putf;
 617         mutex_lock(&inode->i_mutex);
 618         if (mode == (mode_t) -1)
 619                 mode = inode->i_mode;
 620         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 621         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 622         err = notify_change(dentry, &newattrs);
 623         mutex_unlock(&inode->i_mutex);
 624         mnt_drop_write(file->f_path.mnt);
 625 out_putf:
 626         fput(file);
 627 out:
 628         return err;
 629 }
 630
 631 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 632                              mode_t mode)
 633 {
 634         struct path path;
 635         struct inode *inode;
 636         int error;
 637         struct iattr newattrs;
 638
 639         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 640         if (error)
 641                 goto out;
 642         inode = path.dentry->d_inode;
 643
 644         error = mnt_want_write(path.mnt);
 645         if (error)
 646                 goto dput_and_out;
 647         mutex_lock(&inode->i_mutex);
 648         if (mode == (mode_t) -1)
 649                 mode = inode->i_mode;
 650         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 651         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 652         error = notify_change(path.dentry, &newattrs);
 653         mutex_unlock(&inode->i_mutex);
 654         mnt_drop_write(path.mnt);
 655 dput_and_out:
 656         path_put(&path);
 657 out:
 658         return error;
 659 }
 660
 661 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 662 {
 663         return sys_fchmodat(AT_FDCWD, filename, mode);
 664 }
 665
 666 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 667 {
 668         struct inode *inode = dentry->d_inode;
 669         int error;
 670         struct iattr newattrs;
 671
 672         newattrs.ia_valid =  ATTR_CTIME;
 673         if (user != (uid_t) -1) {
 674                 newattrs.ia_valid |= ATTR_UID;
 675                 newattrs.ia_uid = user;
 676         }
 677         if (group != (gid_t) -1) {
 678                 newattrs.ia_valid |= ATTR_GID;
 679                 newattrs.ia_gid = group;
 680         }
 681         if (!S_ISDIR(inode->i_mode))
 682                 newattrs.ia_valid |=
 683                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 684         mutex_lock(&inode->i_mutex);
 685         error = notify_change(dentry, &newattrs);
 686         mutex_unlock(&inode->i_mutex);
 687
 688         return error;
 689 }
 690
 691 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 692 {
 693         struct path path;
 694         int error;
 695
 696         error = user_path(filename, &path);
 697         if (error)
 698                 goto out;
 699         error = mnt_want_write(path.mnt);
 700         if (error)
 701                 goto out_release;
 702         error = chown_common(path.dentry, user, group);
 703         mnt_drop_write(path.mnt);
 704 out_release:
 705         path_put(&path);
 706 out:
 707         return error;
 708 }
 709
 710 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 711                              gid_t group, int flag)
 712 {
 713         struct path path;
 714         int error = -EINVAL;
 715         int follow;
 716
 717         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 718                 goto out;
 719
 720         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 721         error = user_path_at(dfd, filename, follow, &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 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 736 {
 737         struct path path;
 738         int error;
 739
 740         error = user_lpath(filename, &path);
 741         if (error)
 742                 goto out;
 743         error = mnt_want_write(path.mnt);
 744         if (error)
 745                 goto out_release;
 746         error = chown_common(path.dentry, user, group);
 747         mnt_drop_write(path.mnt);
 748 out_release:
 749         path_put(&path);
 750 out:
 751         return error;
 752 }
 753
 754
 755 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 756 {
 757         struct file * file;
 758         int error = -EBADF;
 759         struct dentry * dentry;
 760
 761         file = fget(fd);
 762         if (!file)
 763                 goto out;
 764
 765         error = mnt_want_write(file->f_path.mnt);
 766         if (error)
 767                 goto out_fput;
 768         dentry = file->f_path.dentry;
 769         audit_inode(NULL, dentry);
 770         error = chown_common(dentry, user, group);
 771         mnt_drop_write(file->f_path.mnt);
 772 out_fput:
 773         fput(file);
 774 out:
 775         return error;
 776 }
 777
 778 /*
 779  * You have to be very careful that these write
 780  * counts get cleaned up in error cases and
 781  * upon __fput().  This should probably never
 782  * be called outside of __dentry_open().
 783  */
 784 static inline int __get_file_write_access(struct inode *inode,
 785                                           struct vfsmount *mnt)
 786 {
 787         int error;
 788         error = get_write_access(inode);
 789         if (error)
 790                 return error;
 791         /*
 792          * Do not take mount writer counts on
 793          * special files since no writes to
 794          * the mount itself will occur.
 795          */
 796         if (!special_file(inode->i_mode)) {
 797                 /*
 798                  * Balanced in __fput()
 799                  */
 800                 error = mnt_want_write(mnt);
 801                 if (error)
 802                         put_write_access(inode);
 803         }
 804         return error;
 805 }
 806
 807 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 808                                         int flags, struct file *f,
 809                                         int (*open)(struct inode *, struct file *),
 810                                         const struct cred *cred)
 811 {
 812         struct inode *inode;
 813         int error;
 814
 815         f->f_flags = flags;
 816         f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 817                                 FMODE_PREAD | FMODE_PWRITE;
 818         inode = dentry->d_inode;
 819         if (f->f_mode & FMODE_WRITE) {
 820                 error = __get_file_write_access(inode, mnt);
 821                 if (error)
 822                         goto cleanup_file;
 823                 if (!special_file(inode->i_mode))
 824                         file_take_write(f);
 825         }
 826
 827         f->f_mapping = inode->i_mapping;
 828         f->f_path.dentry = dentry;
 829         f->f_path.mnt = mnt;
 830         f->f_pos = 0;
 831         f->f_op = fops_get(inode->i_fop);
 832         file_move(f, &inode->i_sb->s_files);
 833
 834         error = security_dentry_open(f, cred);
 835         if (error)
 836                 goto cleanup_all;
 837
 838         if (!open && f->f_op)
 839                 open = f->f_op->open;
 840         if (open) {
 841                 error = open(inode, f);
 842                 if (error)
 843                         goto cleanup_all;
 844         }
 845
 846         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 847
 848         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 849
 850         /* NB: we're sure to have correct a_ops only after f_op->open */
 851         if (f->f_flags & O_DIRECT) {
 852                 if (!f->f_mapping->a_ops ||
 853                     ((!f->f_mapping->a_ops->direct_IO) &&
 854                     (!f->f_mapping->a_ops->get_xip_mem))) {
 855                         fput(f);
 856                         f = ERR_PTR(-EINVAL);
 857                 }
 858         }
 859
 860         return f;
 861
 862 cleanup_all:
 863         fops_put(f->f_op);
 864         if (f->f_mode & FMODE_WRITE) {
 865                 put_write_access(inode);
 866                 if (!special_file(inode->i_mode)) {
 867                         /*
 868                          * We don't consider this a real
 869                          * mnt_want/drop_write() pair
 870                          * because it all happenend right
 871                          * here, so just reset the state.
 872                          */
 873                         file_reset_write(f);
 874                         mnt_drop_write(mnt);
 875                 }
 876         }
 877         file_kill(f);
 878         f->f_path.dentry = NULL;
 879         f->f_path.mnt = NULL;
 880 cleanup_file:
 881         put_filp(f);
 882         dput(dentry);
 883         mntput(mnt);
 884         return ERR_PTR(error);
 885 }
 886
 887 /**
 888  * lookup_instantiate_filp - instantiates the open intent filp
 889  * @nd: pointer to nameidata
 890  * @dentry: pointer to dentry
 891  * @open: open callback
 892  *
 893  * Helper for filesystems that want to use lookup open intents and pass back
 894  * a fully instantiated struct file to the caller.
 895  * This function is meant to be called from within a filesystem's
 896  * lookup method.
 897  * Beware of calling it for non-regular files! Those ->open methods might block
 898  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 899  * leading to a deadlock, as nobody can open that fifo anymore, because
 900  * another process to open fifo will block on locked parent when doing lookup).
 901  * Note that in case of error, nd->intent.open.file is destroyed, but the
 902  * path information remains valid.
 903  * If the open callback is set to NULL, then the standard f_op->open()
 904  * filesystem callback is substituted.
 905  */
 906 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 907                 int (*open)(struct inode *, struct file *))
 908 {
 909         const struct cred *cred = current_cred();
 910
 911         if (IS_ERR(nd->intent.open.file))
 912                 goto out;
 913         if (IS_ERR(dentry))
 914                 goto out_err;
 915         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 916                                              nd->intent.open.flags - 1,
 917                                              nd->intent.open.file,
 918                                              open, cred);
 919 out:
 920         return nd->intent.open.file;
 921 out_err:
 922         release_open_intent(nd);
 923         nd->intent.open.file = (struct file *)dentry;
 924         goto out;
 925 }
 926 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 927
 928 /**
 929  * nameidata_to_filp - convert a nameidata to an open filp.
 930  * @nd: pointer to nameidata
 931  * @flags: open flags
 932  *
 933  * Note that this function destroys the original nameidata
 934  */
 935 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 936 {
 937         const struct cred *cred = current_cred();
 938         struct file *filp;
 939
 940         /* Pick up the filp from the open intent */
 941         filp = nd->intent.open.file;
 942         /* Has the filesystem initialised the file for us? */
 943         if (filp->f_path.dentry == NULL)
 944                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 945                                      NULL, cred);
 946         else
 947                 path_put(&nd->path);
 948         return filp;
 949 }
 950
 951 /*
 952  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 953  * error.
 954  */
 955 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 956                          const struct cred *cred)
 957 {
 958         int error;
 959         struct file *f;
 960
 961         /*
 962          * We must always pass in a valid mount pointer.   Historically
 963          * callers got away with not passing it, but we must enforce this at
 964          * the earliest possible point now to avoid strange problems deep in the
 965          * filesystem stack.
 966          */
 967         if (!mnt) {
 968                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 969                 dump_stack();
 970                 return ERR_PTR(-EINVAL);
 971         }
 972
 973         error = -ENFILE;
 974         f = get_empty_filp();
 975         if (f == NULL) {
 976                 dput(dentry);
 977                 mntput(mnt);
 978                 return ERR_PTR(error);
 979         }
 980
 981         return __dentry_open(dentry, mnt, flags, f, NULL, cred);
 982 }
 983 EXPORT_SYMBOL(dentry_open);
 984
 985 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 986 {
 987         struct fdtable *fdt = files_fdtable(files);
 988         __FD_CLR(fd, fdt->open_fds);
 989         if (fd < files->next_fd)
 990                 files->next_fd = fd;
 991 }
 992
 993 void put_unused_fd(unsigned int fd)
 994 {
 995         struct files_struct *files = current->files;
 996         spin_lock(&files->file_lock);
 997         __put_unused_fd(files, fd);
 998         spin_unlock(&files->file_lock);
 999 }
1000
1001 EXPORT_SYMBOL(put_unused_fd);
1002
1003 /*
1004  * Install a file pointer in the fd array.
1005  *
1006  * The VFS is full of places where we drop the files lock between
1007  * setting the open_fds bitmap and installing the file in the file
1008  * array.  At any such point, we are vulnerable to a dup2() race
1009  * installing a file in the array before us.  We need to detect this and
1010  * fput() the struct file we are about to overwrite in this case.
1011  *
1012  * It should never happen - if we allow dup2() do it, _really_ bad things
1013  * will follow.
1014  */
1015
1016 void fd_install(unsigned int fd, struct file *file)
1017 {
1018         struct files_struct *files = current->files;
1019         struct fdtable *fdt;
1020         spin_lock(&files->file_lock);
1021         fdt = files_fdtable(files);
1022         BUG_ON(fdt->fd[fd] != NULL);
1023         rcu_assign_pointer(fdt->fd[fd], file);
1024         spin_unlock(&files->file_lock);
1025 }
1026
1027 EXPORT_SYMBOL(fd_install);
1028
1029 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1030 {
1031         char *tmp = getname(filename);
1032         int fd = PTR_ERR(tmp);
1033
1034         if (!IS_ERR(tmp)) {
1035                 fd = get_unused_fd_flags(flags);
1036                 if (fd >= 0) {
1037                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1038                         if (IS_ERR(f)) {
1039                                 put_unused_fd(fd);
1040                                 fd = PTR_ERR(f);
1041                         } else {
1042                                 fsnotify_open(f->f_path.dentry);
1043                                 fd_install(fd, f);
1044                         }
1045                 }
1046                 putname(tmp);
1047         }
1048         return fd;
1049 }
1050
1051 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1052 {
1053         long ret;
1054
1055         if (force_o_largefile())
1056                 flags |= O_LARGEFILE;
1057
1058         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1059         /* avoid REGPARM breakage on x86: */
1060         asmlinkage_protect(3, ret, filename, flags, mode);
1061         return ret;
1062 }
1063
1064 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1065                            int mode)
1066 {
1067         long ret;
1068
1069         if (force_o_largefile())
1070                 flags |= O_LARGEFILE;
1071
1072         ret = do_sys_open(dfd, filename, flags, mode);
1073         /* avoid REGPARM breakage on x86: */
1074         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1075         return ret;
1076 }
1077
1078 #ifndef __alpha__
1079
1080 /*
1081  * For backward compatibility?  Maybe this should be moved
1082  * into arch/i386 instead?
1083  */
1084 asmlinkage long sys_creat(const char __user * pathname, int mode)
1085 {
1086         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1087 }
1088
1089 #endif
1090
1091 /*
1092  * "id" is the POSIX thread ID. We use the
1093  * files pointer for this..
1094  */
1095 int filp_close(struct file *filp, fl_owner_t id)
1096 {
1097         int retval = 0;
1098
1099         if (!file_count(filp)) {
1100                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1101                 return 0;
1102         }
1103
1104         if (filp->f_op && filp->f_op->flush)
1105                 retval = filp->f_op->flush(filp, id);
1106
1107         dnotify_flush(filp, id);
1108         locks_remove_posix(filp, id);
1109         fput(filp);
1110         return retval;
1111 }
1112
1113 EXPORT_SYMBOL(filp_close);
1114
1115 /*
1116  * Careful here! We test whether the file pointer is NULL before
1117  * releasing the fd. This ensures that one clone task can't release
1118  * an fd while another clone is opening it.
1119  */
1120 asmlinkage long sys_close(unsigned int fd)
1121 {
1122         struct file * filp;
1123         struct files_struct *files = current->files;
1124         struct fdtable *fdt;
1125         int retval;
1126
1127         spin_lock(&files->file_lock);
1128         fdt = files_fdtable(files);
1129         if (fd >= fdt->max_fds)
1130                 goto out_unlock;
1131         filp = fdt->fd[fd];
1132         if (!filp)
1133                 goto out_unlock;
1134         rcu_assign_pointer(fdt->fd[fd], NULL);
1135         FD_CLR(fd, fdt->close_on_exec);
1136         __put_unused_fd(files, fd);
1137         spin_unlock(&files->file_lock);
1138         retval = filp_close(filp, files);
1139
1140         /* can't restart close syscall because file table entry was cleared */
1141         if (unlikely(retval == -ERESTARTSYS ||
1142                      retval == -ERESTARTNOINTR ||
1143                      retval == -ERESTARTNOHAND ||
1144                      retval == -ERESTART_RESTARTBLOCK))
1145                 retval = -EINTR;
1146
1147         return retval;
1148
1149 out_unlock:
1150         spin_unlock(&files->file_lock);
1151         return -EBADF;
1152 }
1153
1154 EXPORT_SYMBOL(sys_close);
1155
1156 /*
1157  * This routine simulates a hangup on the tty, to arrange that users
1158  * are given clean terminals at login time.
1159  */
1160 asmlinkage long sys_vhangup(void)
1161 {
1162         if (capable(CAP_SYS_TTY_CONFIG)) {
1163                 tty_vhangup_self();
1164                 return 0;
1165         }
1166         return -EPERM;
1167 }
1168
1169 /*
1170  * Called when an inode is about to be open.
1171  * We use this to disallow opening large files on 32bit systems if
1172  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1173  * on this flag in sys_open.
1174  */
1175 int generic_file_open(struct inode * inode, struct file * filp)
1176 {
1177         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1178                 return -EOVERFLOW;
1179         return 0;
1180 }
1181
1182 EXPORT_SYMBOL(generic_file_open);
1183
1184 /*
1185  * This is used by subsystems that don't want seekable
1186  * file descriptors
1187  */
1188 int nonseekable_open(struct inode *inode, struct file *filp)
1189 {
1190         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1191         return 0;
1192 }
1193
1194 EXPORT_SYMBOL(nonseekable_open);