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                 DQUOT_INIT(inode);
 275                 error = do_truncate(path.dentry, length, 0, NULL);
 276         }
 277
 278 put_write_and_out:
 279         put_write_access(inode);
 280 mnt_drop_write_and_out:
 281         mnt_drop_write(path.mnt);
 282 dput_and_out:
 283         path_put(&path);
 284 out:
 285         return error;
 286 }
 287
 288 SYSCALL_DEFINE2(truncate, const char __user *, path, unsigned long, length)
 289 {
 290         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 291         return do_sys_truncate(path, (long)length);
 292 }
 293
 294 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 295 {
 296         struct inode * inode;
 297         struct dentry *dentry;
 298         struct file * file;
 299         int error;
 300
 301         error = -EINVAL;
 302         if (length < 0)
 303                 goto out;
 304         error = -EBADF;
 305         file = fget(fd);
 306         if (!file)
 307                 goto out;
 308
 309         /* explicitly opened as large or we are on 64-bit box */
 310         if (file->f_flags & O_LARGEFILE)
 311                 small = 0;
 312
 313         dentry = file->f_path.dentry;
 314         inode = dentry->d_inode;
 315         error = -EINVAL;
 316         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 317                 goto out_putf;
 318
 319         error = -EINVAL;
 320         /* Cannot ftruncate over 2^31 bytes without large file support */
 321         if (small && length > MAX_NON_LFS)
 322                 goto out_putf;
 323
 324         error = -EPERM;
 325         if (IS_APPEND(inode))
 326                 goto out_putf;
 327
 328         error = locks_verify_truncate(inode, file, length);
 329         if (!error)
 330                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 331 out_putf:
 332         fput(file);
 333 out:
 334         return error;
 335 }
 336
 337 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 338 {
 339         long ret = do_sys_ftruncate(fd, length, 1);
 340         /* avoid REGPARM breakage on x86: */
 341         asmlinkage_protect(2, ret, fd, length);
 342         return ret;
 343 }
 344
 345 /* LFS versions of truncate are only needed on 32 bit machines */
 346 #if BITS_PER_LONG == 32
 347 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 348 {
 349         return do_sys_truncate(path, length);
 350 }
 351 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 352 asmlinkage long SyS_truncate64(long path, loff_t length)
 353 {
 354         return SYSC_truncate64((const char __user *) path, length);
 355 }
 356 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 357 #endif
 358
 359 SYSCALL_DEFINE(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 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 367 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 368 {
 369         return SYSC_ftruncate64((unsigned int) fd, length);
 370 }
 371 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 372 #endif
 373 #endif /* BITS_PER_LONG == 32 */
 374
 375 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 376 {
 377         struct file *file;
 378         struct inode *inode;
 379         long ret = -EINVAL;
 380
 381         if (offset < 0 || len <= 0)
 382                 goto out;
 383
 384         /* Return error if mode is not supported */
 385         ret = -EOPNOTSUPP;
 386         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 387                 goto out;
 388
 389         ret = -EBADF;
 390         file = fget(fd);
 391         if (!file)
 392                 goto out;
 393         if (!(file->f_mode & FMODE_WRITE))
 394                 goto out_fput;
 395         /*
 396          * Revalidate the write permissions, in case security policy has
 397          * changed since the files were opened.
 398          */
 399         ret = security_file_permission(file, MAY_WRITE);
 400         if (ret)
 401                 goto out_fput;
 402
 403         inode = file->f_path.dentry->d_inode;
 404
 405         ret = -ESPIPE;
 406         if (S_ISFIFO(inode->i_mode))
 407                 goto out_fput;
 408
 409         ret = -ENODEV;
 410         /*
 411          * Let individual file system decide if it supports preallocation
 412          * for directories or not.
 413          */
 414         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 415                 goto out_fput;
 416
 417         ret = -EFBIG;
 418         /* Check for wrap through zero too */
 419         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 420                 goto out_fput;
 421
 422         if (inode->i_op && inode->i_op->fallocate)
 423                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 424         else
 425                 ret = -EOPNOTSUPP;
 426
 427 out_fput:
 428         fput(file);
 429 out:
 430         return ret;
 431 }
 432 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 433 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 434 {
 435         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 436 }
 437 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 438 #endif
 439
 440 /*
 441  * access() needs to use the real uid/gid, not the effective uid/gid.
 442  * We do this by temporarily clearing all FS-related capabilities and
 443  * switching the fsuid/fsgid around to the real ones.
 444  */
 445 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 446 {
 447         struct path path;
 448         struct inode *inode;
 449         int old_fsuid, old_fsgid;
 450         kernel_cap_t uninitialized_var(old_cap);  /* !SECURE_NO_SETUID_FIXUP */
 451         int res;
 452
 453         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 454                 return -EINVAL;
 455
 456         old_fsuid = current->fsuid;
 457         old_fsgid = current->fsgid;
 458
 459         current->fsuid = current->uid;
 460         current->fsgid = current->gid;
 461
 462         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 463                 /*
 464                  * Clear the capabilities if we switch to a non-root user
 465                  */
 466 #ifndef CONFIG_SECURITY_FILE_CAPABILITIES
 467                 /*
 468                  * FIXME: There is a race here against sys_capset.  The
 469                  * capabilities can change yet we will restore the old
 470                  * value below.  We should hold task_capabilities_lock,
 471                  * but we cannot because user_path_at can sleep.
 472                  */
 473 #endif /* ndef CONFIG_SECURITY_FILE_CAPABILITIES */
 474                 if (current->uid)
 475                         old_cap = cap_set_effective(__cap_empty_set);
 476                 else
 477                         old_cap = cap_set_effective(current->cap_permitted);
 478         }
 479
 480         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 481         if (res)
 482                 goto out;
 483
 484         inode = path.dentry->d_inode;
 485
 486         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 487                 /*
 488                  * MAY_EXEC on regular files is denied if the fs is mounted
 489                  * with the "noexec" flag.
 490                  */
 491                 res = -EACCES;
 492                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 493                         goto out_path_release;
 494         }
 495
 496         res = inode_permission(inode, mode | MAY_ACCESS);
 497         /* SuS v2 requires we report a read only fs too */
 498         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 499                 goto out_path_release;
 500         /*
 501          * This is a rare case where using __mnt_is_readonly()
 502          * is OK without a mnt_want/drop_write() pair.  Since
 503          * no actual write to the fs is performed here, we do
 504          * not need to telegraph to that to anyone.
 505          *
 506          * By doing this, we accept that this access is
 507          * inherently racy and know that the fs may change
 508          * state before we even see this result.
 509          */
 510         if (__mnt_is_readonly(path.mnt))
 511                 res = -EROFS;
 512
 513 out_path_release:
 514         path_put(&path);
 515 out:
 516         current->fsuid = old_fsuid;
 517         current->fsgid = old_fsgid;
 518
 519         if (!issecure(SECURE_NO_SETUID_FIXUP))
 520                 cap_set_effective(old_cap);
 521
 522         return res;
 523 }
 524
 525 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 526 {
 527         return sys_faccessat(AT_FDCWD, filename, mode);
 528 }
 529
 530 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 531 {
 532         struct path path;
 533         int error;
 534
 535         error = user_path_dir(filename, &path);
 536         if (error)
 537                 goto out;
 538
 539         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 540         if (error)
 541                 goto dput_and_out;
 542
 543         set_fs_pwd(current->fs, &path);
 544
 545 dput_and_out:
 546         path_put(&path);
 547 out:
 548         return error;
 549 }
 550
 551 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 552 {
 553         struct file *file;
 554         struct inode *inode;
 555         int error;
 556
 557         error = -EBADF;
 558         file = fget(fd);
 559         if (!file)
 560                 goto out;
 561
 562         inode = file->f_path.dentry->d_inode;
 563
 564         error = -ENOTDIR;
 565         if (!S_ISDIR(inode->i_mode))
 566                 goto out_putf;
 567
 568         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 569         if (!error)
 570                 set_fs_pwd(current->fs, &file->f_path);
 571 out_putf:
 572         fput(file);
 573 out:
 574         return error;
 575 }
 576
 577 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 578 {
 579         struct path path;
 580         int error;
 581
 582         error = user_path_dir(filename, &path);
 583         if (error)
 584                 goto out;
 585
 586         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 587         if (error)
 588                 goto dput_and_out;
 589
 590         error = -EPERM;
 591         if (!capable(CAP_SYS_CHROOT))
 592                 goto dput_and_out;
 593
 594         set_fs_root(current->fs, &path);
 595         error = 0;
 596 dput_and_out:
 597         path_put(&path);
 598 out:
 599         return error;
 600 }
 601
 602 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 603 {
 604         struct inode * inode;
 605         struct dentry * dentry;
 606         struct file * file;
 607         int err = -EBADF;
 608         struct iattr newattrs;
 609
 610         file = fget(fd);
 611         if (!file)
 612                 goto out;
 613
 614         dentry = file->f_path.dentry;
 615         inode = dentry->d_inode;
 616
 617         audit_inode(NULL, dentry);
 618
 619         err = mnt_want_write(file->f_path.mnt);
 620         if (err)
 621                 goto out_putf;
 622         mutex_lock(&inode->i_mutex);
 623         if (mode == (mode_t) -1)
 624                 mode = inode->i_mode;
 625         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 626         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 627         err = notify_change(dentry, &newattrs);
 628         mutex_unlock(&inode->i_mutex);
 629         mnt_drop_write(file->f_path.mnt);
 630 out_putf:
 631         fput(file);
 632 out:
 633         return err;
 634 }
 635
 636 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 637                              mode_t mode)
 638 {
 639         struct path path;
 640         struct inode *inode;
 641         int error;
 642         struct iattr newattrs;
 643
 644         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 645         if (error)
 646                 goto out;
 647         inode = path.dentry->d_inode;
 648
 649         error = mnt_want_write(path.mnt);
 650         if (error)
 651                 goto dput_and_out;
 652         mutex_lock(&inode->i_mutex);
 653         if (mode == (mode_t) -1)
 654                 mode = inode->i_mode;
 655         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 656         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 657         error = notify_change(path.dentry, &newattrs);
 658         mutex_unlock(&inode->i_mutex);
 659         mnt_drop_write(path.mnt);
 660 dput_and_out:
 661         path_put(&path);
 662 out:
 663         return error;
 664 }
 665
 666 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 667 {
 668         return sys_fchmodat(AT_FDCWD, filename, mode);
 669 }
 670
 671 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 672 {
 673         struct inode *inode = dentry->d_inode;
 674         int error;
 675         struct iattr newattrs;
 676
 677         newattrs.ia_valid =  ATTR_CTIME;
 678         if (user != (uid_t) -1) {
 679                 newattrs.ia_valid |= ATTR_UID;
 680                 newattrs.ia_uid = user;
 681         }
 682         if (group != (gid_t) -1) {
 683                 newattrs.ia_valid |= ATTR_GID;
 684                 newattrs.ia_gid = group;
 685         }
 686         if (!S_ISDIR(inode->i_mode))
 687                 newattrs.ia_valid |=
 688                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 689         mutex_lock(&inode->i_mutex);
 690         error = notify_change(dentry, &newattrs);
 691         mutex_unlock(&inode->i_mutex);
 692
 693         return error;
 694 }
 695
 696 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 697 {
 698         struct path path;
 699         int error;
 700
 701         error = user_path(filename, &path);
 702         if (error)
 703                 goto out;
 704         error = mnt_want_write(path.mnt);
 705         if (error)
 706                 goto out_release;
 707         error = chown_common(path.dentry, user, group);
 708         mnt_drop_write(path.mnt);
 709 out_release:
 710         path_put(&path);
 711 out:
 712         return error;
 713 }
 714
 715 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 716                              gid_t group, int flag)
 717 {
 718         struct path path;
 719         int error = -EINVAL;
 720         int follow;
 721
 722         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 723                 goto out;
 724
 725         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 726         error = user_path_at(dfd, filename, follow, &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 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 741 {
 742         struct path path;
 743         int error;
 744
 745         error = user_lpath(filename, &path);
 746         if (error)
 747                 goto out;
 748         error = mnt_want_write(path.mnt);
 749         if (error)
 750                 goto out_release;
 751         error = chown_common(path.dentry, user, group);
 752         mnt_drop_write(path.mnt);
 753 out_release:
 754         path_put(&path);
 755 out:
 756         return error;
 757 }
 758
 759 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 760 {
 761         struct file * file;
 762         int error = -EBADF;
 763         struct dentry * dentry;
 764
 765         file = fget(fd);
 766         if (!file)
 767                 goto out;
 768
 769         error = mnt_want_write(file->f_path.mnt);
 770         if (error)
 771                 goto out_fput;
 772         dentry = file->f_path.dentry;
 773         audit_inode(NULL, dentry);
 774         error = chown_common(dentry, user, group);
 775         mnt_drop_write(file->f_path.mnt);
 776 out_fput:
 777         fput(file);
 778 out:
 779         return error;
 780 }
 781
 782 /*
 783  * You have to be very careful that these write
 784  * counts get cleaned up in error cases and
 785  * upon __fput().  This should probably never
 786  * be called outside of __dentry_open().
 787  */
 788 static inline int __get_file_write_access(struct inode *inode,
 789                                           struct vfsmount *mnt)
 790 {
 791         int error;
 792         error = get_write_access(inode);
 793         if (error)
 794                 return error;
 795         /*
 796          * Do not take mount writer counts on
 797          * special files since no writes to
 798          * the mount itself will occur.
 799          */
 800         if (!special_file(inode->i_mode)) {
 801                 /*
 802                  * Balanced in __fput()
 803                  */
 804                 error = mnt_want_write(mnt);
 805                 if (error)
 806                         put_write_access(inode);
 807         }
 808         return error;
 809 }
 810
 811 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 812                                         int flags, struct file *f,
 813                                         int (*open)(struct inode *, struct file *))
 814 {
 815         struct inode *inode;
 816         int error;
 817
 818         f->f_flags = flags;
 819         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 820                                 FMODE_PREAD | FMODE_PWRITE;
 821         inode = dentry->d_inode;
 822         if (f->f_mode & FMODE_WRITE) {
 823                 error = __get_file_write_access(inode, mnt);
 824                 if (error)
 825                         goto cleanup_file;
 826                 if (!special_file(inode->i_mode))
 827                         file_take_write(f);
 828         }
 829
 830         f->f_mapping = inode->i_mapping;
 831         f->f_path.dentry = dentry;
 832         f->f_path.mnt = mnt;
 833         f->f_pos = 0;
 834         f->f_op = fops_get(inode->i_fop);
 835         file_move(f, &inode->i_sb->s_files);
 836
 837         error = security_dentry_open(f);
 838         if (error)
 839                 goto cleanup_all;
 840
 841         if (!open && f->f_op)
 842                 open = f->f_op->open;
 843         if (open) {
 844                 error = open(inode, f);
 845                 if (error)
 846                         goto cleanup_all;
 847         }
 848
 849         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 850
 851         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 852
 853         /* NB: we're sure to have correct a_ops only after f_op->open */
 854         if (f->f_flags & O_DIRECT) {
 855                 if (!f->f_mapping->a_ops ||
 856                     ((!f->f_mapping->a_ops->direct_IO) &&
 857                     (!f->f_mapping->a_ops->get_xip_mem))) {
 858                         fput(f);
 859                         f = ERR_PTR(-EINVAL);
 860                 }
 861         }
 862
 863         return f;
 864
 865 cleanup_all:
 866         fops_put(f->f_op);
 867         if (f->f_mode & FMODE_WRITE) {
 868                 put_write_access(inode);
 869                 if (!special_file(inode->i_mode)) {
 870                         /*
 871                          * We don't consider this a real
 872                          * mnt_want/drop_write() pair
 873                          * because it all happenend right
 874                          * here, so just reset the state.
 875                          */
 876                         file_reset_write(f);
 877                         mnt_drop_write(mnt);
 878                 }
 879         }
 880         file_kill(f);
 881         f->f_path.dentry = NULL;
 882         f->f_path.mnt = NULL;
 883 cleanup_file:
 884         put_filp(f);
 885         dput(dentry);
 886         mntput(mnt);
 887         return ERR_PTR(error);
 888 }
 889
 890 /**
 891  * lookup_instantiate_filp - instantiates the open intent filp
 892  * @nd: pointer to nameidata
 893  * @dentry: pointer to dentry
 894  * @open: open callback
 895  *
 896  * Helper for filesystems that want to use lookup open intents and pass back
 897  * a fully instantiated struct file to the caller.
 898  * This function is meant to be called from within a filesystem's
 899  * lookup method.
 900  * Beware of calling it for non-regular files! Those ->open methods might block
 901  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 902  * leading to a deadlock, as nobody can open that fifo anymore, because
 903  * another process to open fifo will block on locked parent when doing lookup).
 904  * Note that in case of error, nd->intent.open.file is destroyed, but the
 905  * path information remains valid.
 906  * If the open callback is set to NULL, then the standard f_op->open()
 907  * filesystem callback is substituted.
 908  */
 909 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 910                 int (*open)(struct inode *, struct file *))
 911 {
 912         if (IS_ERR(nd->intent.open.file))
 913                 goto out;
 914         if (IS_ERR(dentry))
 915                 goto out_err;
 916         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 917                                              nd->intent.open.flags - 1,
 918                                              nd->intent.open.file,
 919                                              open);
 920 out:
 921         return nd->intent.open.file;
 922 out_err:
 923         release_open_intent(nd);
 924         nd->intent.open.file = (struct file *)dentry;
 925         goto out;
 926 }
 927 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 928
 929 /**
 930  * nameidata_to_filp - convert a nameidata to an open filp.
 931  * @nd: pointer to nameidata
 932  * @flags: open flags
 933  *
 934  * Note that this function destroys the original nameidata
 935  */
 936 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 937 {
 938         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);
 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 {
 957         int error;
 958         struct file *f;
 959
 960         /*
 961          * We must always pass in a valid mount pointer.   Historically
 962          * callers got away with not passing it, but we must enforce this at
 963          * the earliest possible point now to avoid strange problems deep in the
 964          * filesystem stack.
 965          */
 966         if (!mnt) {
 967                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 968                 dump_stack();
 969                 return ERR_PTR(-EINVAL);
 970         }
 971
 972         error = -ENFILE;
 973         f = get_empty_filp();
 974         if (f == NULL) {
 975                 dput(dentry);
 976                 mntput(mnt);
 977                 return ERR_PTR(error);
 978         }
 979
 980         return __dentry_open(dentry, mnt, flags, f, NULL);
 981 }
 982 EXPORT_SYMBOL(dentry_open);
 983
 984 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 985 {
 986         struct fdtable *fdt = files_fdtable(files);
 987         __FD_CLR(fd, fdt->open_fds);
 988         if (fd < files->next_fd)
 989                 files->next_fd = fd;
 990 }
 991
 992 void put_unused_fd(unsigned int fd)
 993 {
 994         struct files_struct *files = current->files;
 995         spin_lock(&files->file_lock);
 996         __put_unused_fd(files, fd);
 997         spin_unlock(&files->file_lock);
 998 }
 999
1000 EXPORT_SYMBOL(put_unused_fd);
1001
1002 /*
1003  * Install a file pointer in the fd array.
1004  *
1005  * The VFS is full of places where we drop the files lock between
1006  * setting the open_fds bitmap and installing the file in the file
1007  * array.  At any such point, we are vulnerable to a dup2() race
1008  * installing a file in the array before us.  We need to detect this and
1009  * fput() the struct file we are about to overwrite in this case.
1010  *
1011  * It should never happen - if we allow dup2() do it, _really_ bad things
1012  * will follow.
1013  */
1014
1015 void fd_install(unsigned int fd, struct file *file)
1016 {
1017         struct files_struct *files = current->files;
1018         struct fdtable *fdt;
1019         spin_lock(&files->file_lock);
1020         fdt = files_fdtable(files);
1021         BUG_ON(fdt->fd[fd] != NULL);
1022         rcu_assign_pointer(fdt->fd[fd], file);
1023         spin_unlock(&files->file_lock);
1024 }
1025
1026 EXPORT_SYMBOL(fd_install);
1027
1028 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1029 {
1030         char *tmp = getname(filename);
1031         int fd = PTR_ERR(tmp);
1032
1033         if (!IS_ERR(tmp)) {
1034                 fd = get_unused_fd_flags(flags);
1035                 if (fd >= 0) {
1036                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1037                         if (IS_ERR(f)) {
1038                                 put_unused_fd(fd);
1039                                 fd = PTR_ERR(f);
1040                         } else {
1041                                 fsnotify_open(f->f_path.dentry);
1042                                 fd_install(fd, f);
1043                         }
1044                 }
1045                 putname(tmp);
1046         }
1047         return fd;
1048 }
1049
1050 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1051 {
1052         long ret;
1053
1054         if (force_o_largefile())
1055                 flags |= O_LARGEFILE;
1056
1057         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1058         /* avoid REGPARM breakage on x86: */
1059         asmlinkage_protect(3, ret, filename, flags, mode);
1060         return ret;
1061 }
1062
1063 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1064                            int mode)
1065 {
1066         long ret;
1067
1068         if (force_o_largefile())
1069                 flags |= O_LARGEFILE;
1070
1071         ret = do_sys_open(dfd, filename, flags, mode);
1072         /* avoid REGPARM breakage on x86: */
1073         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1074         return ret;
1075 }
1076
1077 #ifndef __alpha__
1078
1079 /*
1080  * For backward compatibility?  Maybe this should be moved
1081  * into arch/i386 instead?
1082  */
1083 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1084 {
1085         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1086 }
1087
1088 #endif
1089
1090 /*
1091  * "id" is the POSIX thread ID. We use the
1092  * files pointer for this..
1093  */
1094 int filp_close(struct file *filp, fl_owner_t id)
1095 {
1096         int retval = 0;
1097
1098         if (!file_count(filp)) {
1099                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1100                 return 0;
1101         }
1102
1103         if (filp->f_op && filp->f_op->flush)
1104                 retval = filp->f_op->flush(filp, id);
1105
1106         dnotify_flush(filp, id);
1107         locks_remove_posix(filp, id);
1108         fput(filp);
1109         return retval;
1110 }
1111
1112 EXPORT_SYMBOL(filp_close);
1113
1114 /*
1115  * Careful here! We test whether the file pointer is NULL before
1116  * releasing the fd. This ensures that one clone task can't release
1117  * an fd while another clone is opening it.
1118  */
1119 SYSCALL_DEFINE1(close, unsigned int, fd)
1120 {
1121         struct file * filp;
1122         struct files_struct *files = current->files;
1123         struct fdtable *fdt;
1124         int retval;
1125
1126         spin_lock(&files->file_lock);
1127         fdt = files_fdtable(files);
1128         if (fd >= fdt->max_fds)
1129                 goto out_unlock;
1130         filp = fdt->fd[fd];
1131         if (!filp)
1132                 goto out_unlock;
1133         rcu_assign_pointer(fdt->fd[fd], NULL);
1134         FD_CLR(fd, fdt->close_on_exec);
1135         __put_unused_fd(files, fd);
1136         spin_unlock(&files->file_lock);
1137         retval = filp_close(filp, files);
1138
1139         /* can't restart close syscall because file table entry was cleared */
1140         if (unlikely(retval == -ERESTARTSYS ||
1141                      retval == -ERESTARTNOINTR ||
1142                      retval == -ERESTARTNOHAND ||
1143                      retval == -ERESTART_RESTARTBLOCK))
1144                 retval = -EINTR;
1145
1146         return retval;
1147
1148 out_unlock:
1149         spin_unlock(&files->file_lock);
1150         return -EBADF;
1151 }
1152 EXPORT_SYMBOL(sys_close);
1153
1154 /*
1155  * This routine simulates a hangup on the tty, to arrange that users
1156  * are given clean terminals at login time.
1157  */
1158 SYSCALL_DEFINE0(vhangup)
1159 {
1160         if (capable(CAP_SYS_TTY_CONFIG)) {
1161                 /* XXX: this needs locking */
1162                 tty_vhangup(current->signal->tty);
1163                 return 0;
1164         }
1165         return -EPERM;
1166 }
1167
1168 /*
1169  * Called when an inode is about to be open.
1170  * We use this to disallow opening large files on 32bit systems if
1171  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1172  * on this flag in sys_open.
1173  */
1174 int generic_file_open(struct inode * inode, struct file * filp)
1175 {
1176         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1177                 return -EOVERFLOW;
1178         return 0;
1179 }
1180
1181 EXPORT_SYMBOL(generic_file_open);
1182
1183 /*
1184  * This is used by subsystems that don't want seekable
1185  * file descriptors
1186  */
1187 int nonseekable_open(struct inode *inode, struct file *filp)
1188 {
1189         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1190         return 0;
1191 }
1192
1193 EXPORT_SYMBOL(nonseekable_open);