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/utime.h>
  10 #include <linux/file.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/quotaops.h>
  13 #include <linux/fsnotify.h>
  14 #include <linux/module.h>
  15 #include <linux/slab.h>
  16 #include <linux/tty.h>
  17 #include <linux/namei.h>
  18 #include <linux/backing-dev.h>
  19 #include <linux/security.h>
  20 #include <linux/mount.h>
  21 #include <linux/vfs.h>
  22 #include <asm/uaccess.h>
  23 #include <linux/fs.h>
  24 #include <linux/personality.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/rcupdate.h>
  28
  29 #include <asm/unistd.h>
  30
  31 int vfs_statfs(struct super_block *sb, struct kstatfs *buf)
  32 {
  33         int retval = -ENODEV;
  34
  35         if (sb) {
  36                 retval = -ENOSYS;
  37                 if (sb->s_op->statfs) {
  38                         memset(buf, 0, sizeof(*buf));
  39                         retval = security_sb_statfs(sb);
  40                         if (retval)
  41                                 return retval;
  42                         retval = sb->s_op->statfs(sb, buf);
  43                         if (retval == 0 && buf->f_frsize == 0)
  44                                 buf->f_frsize = buf->f_bsize;
  45                 }
  46         }
  47         return retval;
  48 }
  49
  50 EXPORT_SYMBOL(vfs_statfs);
  51
  52 static int vfs_statfs_native(struct super_block *sb, struct statfs *buf)
  53 {
  54         struct kstatfs st;
  55         int retval;
  56
  57         retval = vfs_statfs(sb, &st);
  58         if (retval)
  59                 return retval;
  60
  61         if (sizeof(*buf) == sizeof(st))
  62                 memcpy(buf, &st, sizeof(st));
  63         else {
  64                 if (sizeof buf->f_blocks == 4) {
  65                         if ((st.f_blocks | st.f_bfree | st.f_bavail) &
  66                             0xffffffff00000000ULL)
  67                                 return -EOVERFLOW;
  68                         /*
  69                          * f_files and f_ffree may be -1; it's okay to stuff
  70                          * that into 32 bits
  71                          */
  72                         if (st.f_files != -1 &&
  73                             (st.f_files & 0xffffffff00000000ULL))
  74                                 return -EOVERFLOW;
  75                         if (st.f_ffree != -1 &&
  76                             (st.f_ffree & 0xffffffff00000000ULL))
  77                                 return -EOVERFLOW;
  78                 }
  79
  80                 buf->f_type = st.f_type;
  81                 buf->f_bsize = st.f_bsize;
  82                 buf->f_blocks = st.f_blocks;
  83                 buf->f_bfree = st.f_bfree;
  84                 buf->f_bavail = st.f_bavail;
  85                 buf->f_files = st.f_files;
  86                 buf->f_ffree = st.f_ffree;
  87                 buf->f_fsid = st.f_fsid;
  88                 buf->f_namelen = st.f_namelen;
  89                 buf->f_frsize = st.f_frsize;
  90                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  91         }
  92         return 0;
  93 }
  94
  95 static int vfs_statfs64(struct super_block *sb, struct statfs64 *buf)
  96 {
  97         struct kstatfs st;
  98         int retval;
  99
 100         retval = vfs_statfs(sb, &st);
 101         if (retval)
 102                 return retval;
 103
 104         if (sizeof(*buf) == sizeof(st))
 105                 memcpy(buf, &st, sizeof(st));
 106         else {
 107                 buf->f_type = st.f_type;
 108                 buf->f_bsize = st.f_bsize;
 109                 buf->f_blocks = st.f_blocks;
 110                 buf->f_bfree = st.f_bfree;
 111                 buf->f_bavail = st.f_bavail;
 112                 buf->f_files = st.f_files;
 113                 buf->f_ffree = st.f_ffree;
 114                 buf->f_fsid = st.f_fsid;
 115                 buf->f_namelen = st.f_namelen;
 116                 buf->f_frsize = st.f_frsize;
 117                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 118         }
 119         return 0;
 120 }
 121
 122 asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
 123 {
 124         struct nameidata nd;
 125         int error;
 126
 127         error = user_path_walk(path, &nd);
 128         if (!error) {
 129                 struct statfs tmp;
 130                 error = vfs_statfs_native(nd.dentry->d_inode->i_sb, &tmp);
 131                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 132                         error = -EFAULT;
 133                 path_release(&nd);
 134         }
 135         return error;
 136 }
 137
 138
 139 asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
 140 {
 141         struct nameidata nd;
 142         long error;
 143
 144         if (sz != sizeof(*buf))
 145                 return -EINVAL;
 146         error = user_path_walk(path, &nd);
 147         if (!error) {
 148                 struct statfs64 tmp;
 149                 error = vfs_statfs64(nd.dentry->d_inode->i_sb, &tmp);
 150                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 151                         error = -EFAULT;
 152                 path_release(&nd);
 153         }
 154         return error;
 155 }
 156
 157
 158 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
 159 {
 160         struct file * file;
 161         struct statfs tmp;
 162         int error;
 163
 164         error = -EBADF;
 165         file = fget(fd);
 166         if (!file)
 167                 goto out;
 168         error = vfs_statfs_native(file->f_dentry->d_inode->i_sb, &tmp);
 169         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 170                 error = -EFAULT;
 171         fput(file);
 172 out:
 173         return error;
 174 }
 175
 176 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
 177 {
 178         struct file * file;
 179         struct statfs64 tmp;
 180         int error;
 181
 182         if (sz != sizeof(*buf))
 183                 return -EINVAL;
 184
 185         error = -EBADF;
 186         file = fget(fd);
 187         if (!file)
 188                 goto out;
 189         error = vfs_statfs64(file->f_dentry->d_inode->i_sb, &tmp);
 190         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 191                 error = -EFAULT;
 192         fput(file);
 193 out:
 194         return error;
 195 }
 196
 197 int do_truncate(struct dentry *dentry, loff_t length, struct file *filp)
 198 {
 199         int err;
 200         struct iattr newattrs;
 201
 202         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 203         if (length < 0)
 204                 return -EINVAL;
 205
 206         newattrs.ia_size = length;
 207         newattrs.ia_valid = ATTR_SIZE | ATTR_CTIME;
 208         if (filp) {
 209                 newattrs.ia_file = filp;
 210                 newattrs.ia_valid |= ATTR_FILE;
 211         }
 212
 213         down(&dentry->d_inode->i_sem);
 214         err = notify_change(dentry, &newattrs);
 215         up(&dentry->d_inode->i_sem);
 216         return err;
 217 }
 218
 219 static inline long do_sys_truncate(const char __user * path, loff_t length)
 220 {
 221         struct nameidata nd;
 222         struct inode * inode;
 223         int error;
 224
 225         error = -EINVAL;
 226         if (length < 0) /* sorry, but loff_t says... */
 227                 goto out;
 228
 229         error = user_path_walk(path, &nd);
 230         if (error)
 231                 goto out;
 232         inode = nd.dentry->d_inode;
 233
 234         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 235         error = -EISDIR;
 236         if (S_ISDIR(inode->i_mode))
 237                 goto dput_and_out;
 238
 239         error = -EINVAL;
 240         if (!S_ISREG(inode->i_mode))
 241                 goto dput_and_out;
 242
 243         error = vfs_permission(&nd, MAY_WRITE);
 244         if (error)
 245                 goto dput_and_out;
 246
 247         error = -EROFS;
 248         if (IS_RDONLY(inode))
 249                 goto dput_and_out;
 250
 251         error = -EPERM;
 252         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 253                 goto dput_and_out;
 254
 255         /*
 256          * Make sure that there are no leases.
 257          */
 258         error = break_lease(inode, FMODE_WRITE);
 259         if (error)
 260                 goto dput_and_out;
 261
 262         error = get_write_access(inode);
 263         if (error)
 264                 goto dput_and_out;
 265
 266         error = locks_verify_truncate(inode, NULL, length);
 267         if (!error) {
 268                 DQUOT_INIT(inode);
 269                 error = do_truncate(nd.dentry, length, NULL);
 270         }
 271         put_write_access(inode);
 272
 273 dput_and_out:
 274         path_release(&nd);
 275 out:
 276         return error;
 277 }
 278
 279 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 280 {
 281         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 282         return do_sys_truncate(path, (long)length);
 283 }
 284
 285 static inline long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 286 {
 287         struct inode * inode;
 288         struct dentry *dentry;
 289         struct file * file;
 290         int error;
 291
 292         error = -EINVAL;
 293         if (length < 0)
 294                 goto out;
 295         error = -EBADF;
 296         file = fget(fd);
 297         if (!file)
 298                 goto out;
 299
 300         /* explicitly opened as large or we are on 64-bit box */
 301         if (file->f_flags & O_LARGEFILE)
 302                 small = 0;
 303
 304         dentry = file->f_dentry;
 305         inode = dentry->d_inode;
 306         error = -EINVAL;
 307         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 308                 goto out_putf;
 309
 310         error = -EINVAL;
 311         /* Cannot ftruncate over 2^31 bytes without large file support */
 312         if (small && length > MAX_NON_LFS)
 313                 goto out_putf;
 314
 315         error = -EPERM;
 316         if (IS_APPEND(inode))
 317                 goto out_putf;
 318
 319         error = locks_verify_truncate(inode, file, length);
 320         if (!error)
 321                 error = do_truncate(dentry, length, file);
 322 out_putf:
 323         fput(file);
 324 out:
 325         return error;
 326 }
 327
 328 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 329 {
 330         return do_sys_ftruncate(fd, length, 1);
 331 }
 332
 333 /* LFS versions of truncate are only needed on 32 bit machines */
 334 #if BITS_PER_LONG == 32
 335 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 336 {
 337         return do_sys_truncate(path, length);
 338 }
 339
 340 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 341 {
 342         return do_sys_ftruncate(fd, length, 0);
 343 }
 344 #endif
 345
 346 #ifdef __ARCH_WANT_SYS_UTIME
 347
 348 /*
 349  * sys_utime() can be implemented in user-level using sys_utimes().
 350  * Is this for backwards compatibility?  If so, why not move it
 351  * into the appropriate arch directory (for those architectures that
 352  * need it).
 353  */
 354
 355 /* If times==NULL, set access and modification to current time,
 356  * must be owner or have write permission.
 357  * Else, update from *times, must be owner or super user.
 358  */
 359 asmlinkage long sys_utime(char __user * filename, struct utimbuf __user * times)
 360 {
 361         int error;
 362         struct nameidata nd;
 363         struct inode * inode;
 364         struct iattr newattrs;
 365
 366         error = user_path_walk(filename, &nd);
 367         if (error)
 368                 goto out;
 369         inode = nd.dentry->d_inode;
 370
 371         error = -EROFS;
 372         if (IS_RDONLY(inode))
 373                 goto dput_and_out;
 374
 375         /* Don't worry, the checks are done in inode_change_ok() */
 376         newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
 377         if (times) {
 378                 error = -EPERM;
 379                 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 380                         goto dput_and_out;
 381
 382                 error = get_user(newattrs.ia_atime.tv_sec, &times->actime);
 383                 newattrs.ia_atime.tv_nsec = 0;
 384                 if (!error)
 385                         error = get_user(newattrs.ia_mtime.tv_sec, &times->modtime);
 386                 newattrs.ia_mtime.tv_nsec = 0;
 387                 if (error)
 388                         goto dput_and_out;
 389
 390                 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
 391         } else {
 392                 error = -EACCES;
 393                 if (IS_IMMUTABLE(inode))
 394                         goto dput_and_out;
 395
 396                 if (current->fsuid != inode->i_uid &&
 397                     (error = vfs_permission(&nd, MAY_WRITE)) != 0)
 398                         goto dput_and_out;
 399         }
 400         down(&inode->i_sem);
 401         error = notify_change(nd.dentry, &newattrs);
 402         up(&inode->i_sem);
 403 dput_and_out:
 404         path_release(&nd);
 405 out:
 406         return error;
 407 }
 408
 409 #endif
 410
 411 /* If times==NULL, set access and modification to current time,
 412  * must be owner or have write permission.
 413  * Else, update from *times, must be owner or super user.
 414  */
 415 long do_utimes(char __user * filename, struct timeval * times)
 416 {
 417         int error;
 418         struct nameidata nd;
 419         struct inode * inode;
 420         struct iattr newattrs;
 421
 422         error = user_path_walk(filename, &nd);
 423
 424         if (error)
 425                 goto out;
 426         inode = nd.dentry->d_inode;
 427
 428         error = -EROFS;
 429         if (IS_RDONLY(inode))
 430                 goto dput_and_out;
 431
 432         /* Don't worry, the checks are done in inode_change_ok() */
 433         newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
 434         if (times) {
 435                 error = -EPERM;
 436                 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 437                         goto dput_and_out;
 438
 439                 newattrs.ia_atime.tv_sec = times[0].tv_sec;
 440                 newattrs.ia_atime.tv_nsec = times[0].tv_usec * 1000;
 441                 newattrs.ia_mtime.tv_sec = times[1].tv_sec;
 442                 newattrs.ia_mtime.tv_nsec = times[1].tv_usec * 1000;
 443                 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
 444         } else {
 445                 error = -EACCES;
 446                 if (IS_IMMUTABLE(inode))
 447                         goto dput_and_out;
 448
 449                 if (current->fsuid != inode->i_uid &&
 450                     (error = vfs_permission(&nd, MAY_WRITE)) != 0)
 451                         goto dput_and_out;
 452         }
 453         down(&inode->i_sem);
 454         error = notify_change(nd.dentry, &newattrs);
 455         up(&inode->i_sem);
 456 dput_and_out:
 457         path_release(&nd);
 458 out:
 459         return error;
 460 }
 461
 462 asmlinkage long sys_utimes(char __user * filename, struct timeval __user * utimes)
 463 {
 464         struct timeval times[2];
 465
 466         if (utimes && copy_from_user(&times, utimes, sizeof(times)))
 467                 return -EFAULT;
 468         return do_utimes(filename, utimes ? times : NULL);
 469 }
 470
 471
 472 /*
 473  * access() needs to use the real uid/gid, not the effective uid/gid.
 474  * We do this by temporarily clearing all FS-related capabilities and
 475  * switching the fsuid/fsgid around to the real ones.
 476  */
 477 asmlinkage long sys_access(const char __user * filename, int mode)
 478 {
 479         struct nameidata nd;
 480         int old_fsuid, old_fsgid;
 481         kernel_cap_t old_cap;
 482         int res;
 483
 484         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 485                 return -EINVAL;
 486
 487         old_fsuid = current->fsuid;
 488         old_fsgid = current->fsgid;
 489         old_cap = current->cap_effective;
 490
 491         current->fsuid = current->uid;
 492         current->fsgid = current->gid;
 493
 494         /*
 495          * Clear the capabilities if we switch to a non-root user
 496          *
 497          * FIXME: There is a race here against sys_capset.  The
 498          * capabilities can change yet we will restore the old
 499          * value below.  We should hold task_capabilities_lock,
 500          * but we cannot because user_path_walk can sleep.
 501          */
 502         if (current->uid)
 503                 cap_clear(current->cap_effective);
 504         else
 505                 current->cap_effective = current->cap_permitted;
 506
 507         res = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 508         if (!res) {
 509                 res = vfs_permission(&nd, mode);
 510                 /* SuS v2 requires we report a read only fs too */
 511                 if(!res && (mode & S_IWOTH) && IS_RDONLY(nd.dentry->d_inode)
 512                    && !special_file(nd.dentry->d_inode->i_mode))
 513                         res = -EROFS;
 514                 path_release(&nd);
 515         }
 516
 517         current->fsuid = old_fsuid;
 518         current->fsgid = old_fsgid;
 519         current->cap_effective = old_cap;
 520
 521         return res;
 522 }
 523
 524 asmlinkage long sys_chdir(const char __user * filename)
 525 {
 526         struct nameidata nd;
 527         int error;
 528
 529         error = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &nd);
 530         if (error)
 531                 goto out;
 532
 533         error = vfs_permission(&nd, MAY_EXEC);
 534         if (error)
 535                 goto dput_and_out;
 536
 537         set_fs_pwd(current->fs, nd.mnt, nd.dentry);
 538
 539 dput_and_out:
 540         path_release(&nd);
 541 out:
 542         return error;
 543 }
 544
 545 asmlinkage long sys_fchdir(unsigned int fd)
 546 {
 547         struct file *file;
 548         struct dentry *dentry;
 549         struct inode *inode;
 550         struct vfsmount *mnt;
 551         int error;
 552
 553         error = -EBADF;
 554         file = fget(fd);
 555         if (!file)
 556                 goto out;
 557
 558         dentry = file->f_dentry;
 559         mnt = file->f_vfsmnt;
 560         inode = dentry->d_inode;
 561
 562         error = -ENOTDIR;
 563         if (!S_ISDIR(inode->i_mode))
 564                 goto out_putf;
 565
 566         error = file_permission(file, MAY_EXEC);
 567         if (!error)
 568                 set_fs_pwd(current->fs, mnt, dentry);
 569 out_putf:
 570         fput(file);
 571 out:
 572         return error;
 573 }
 574
 575 asmlinkage long sys_chroot(const char __user * filename)
 576 {
 577         struct nameidata nd;
 578         int error;
 579
 580         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 581         if (error)
 582                 goto out;
 583
 584         error = vfs_permission(&nd, MAY_EXEC);
 585         if (error)
 586                 goto dput_and_out;
 587
 588         error = -EPERM;
 589         if (!capable(CAP_SYS_CHROOT))
 590                 goto dput_and_out;
 591
 592         set_fs_root(current->fs, nd.mnt, nd.dentry);
 593         set_fs_altroot();
 594         error = 0;
 595 dput_and_out:
 596         path_release(&nd);
 597 out:
 598         return error;
 599 }
 600
 601 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 602 {
 603         struct inode * inode;
 604         struct dentry * dentry;
 605         struct file * file;
 606         int err = -EBADF;
 607         struct iattr newattrs;
 608
 609         file = fget(fd);
 610         if (!file)
 611                 goto out;
 612
 613         dentry = file->f_dentry;
 614         inode = dentry->d_inode;
 615
 616         err = -EROFS;
 617         if (IS_RDONLY(inode))
 618                 goto out_putf;
 619         err = -EPERM;
 620         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 621                 goto out_putf;
 622         down(&inode->i_sem);
 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         up(&inode->i_sem);
 629
 630 out_putf:
 631         fput(file);
 632 out:
 633         return err;
 634 }
 635
 636 asmlinkage long sys_chmod(const char __user * filename, mode_t mode)
 637 {
 638         struct nameidata nd;
 639         struct inode * inode;
 640         int error;
 641         struct iattr newattrs;
 642
 643         error = user_path_walk(filename, &nd);
 644         if (error)
 645                 goto out;
 646         inode = nd.dentry->d_inode;
 647
 648         error = -EROFS;
 649         if (IS_RDONLY(inode))
 650                 goto dput_and_out;
 651
 652         error = -EPERM;
 653         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 654                 goto dput_and_out;
 655
 656         down(&inode->i_sem);
 657         if (mode == (mode_t) -1)
 658                 mode = inode->i_mode;
 659         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 660         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 661         error = notify_change(nd.dentry, &newattrs);
 662         up(&inode->i_sem);
 663
 664 dput_and_out:
 665         path_release(&nd);
 666 out:
 667         return error;
 668 }
 669
 670 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 671 {
 672         struct inode * inode;
 673         int error;
 674         struct iattr newattrs;
 675
 676         error = -ENOENT;
 677         if (!(inode = dentry->d_inode)) {
 678                 printk(KERN_ERR "chown_common: NULL inode\n");
 679                 goto out;
 680         }
 681         error = -EROFS;
 682         if (IS_RDONLY(inode))
 683                 goto out;
 684         error = -EPERM;
 685         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 686                 goto out;
 687         newattrs.ia_valid =  ATTR_CTIME;
 688         if (user != (uid_t) -1) {
 689                 newattrs.ia_valid |= ATTR_UID;
 690                 newattrs.ia_uid = user;
 691         }
 692         if (group != (gid_t) -1) {
 693                 newattrs.ia_valid |= ATTR_GID;
 694                 newattrs.ia_gid = group;
 695         }
 696         if (!S_ISDIR(inode->i_mode))
 697                 newattrs.ia_valid |= ATTR_KILL_SUID|ATTR_KILL_SGID;
 698         down(&inode->i_sem);
 699         error = notify_change(dentry, &newattrs);
 700         up(&inode->i_sem);
 701 out:
 702         return error;
 703 }
 704
 705 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 706 {
 707         struct nameidata nd;
 708         int error;
 709
 710         error = user_path_walk(filename, &nd);
 711         if (!error) {
 712                 error = chown_common(nd.dentry, user, group);
 713                 path_release(&nd);
 714         }
 715         return error;
 716 }
 717
 718 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 719 {
 720         struct nameidata nd;
 721         int error;
 722
 723         error = user_path_walk_link(filename, &nd);
 724         if (!error) {
 725                 error = chown_common(nd.dentry, user, group);
 726                 path_release(&nd);
 727         }
 728         return error;
 729 }
 730
 731
 732 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 733 {
 734         struct file * file;
 735         int error = -EBADF;
 736
 737         file = fget(fd);
 738         if (file) {
 739                 error = chown_common(file->f_dentry, user, group);
 740                 fput(file);
 741         }
 742         return error;
 743 }
 744
 745 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 746                                         int flags, struct file *f,
 747                                         int (*open)(struct inode *, struct file *))
 748 {
 749         struct inode *inode;
 750         int error;
 751
 752         f->f_flags = flags;
 753         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 754                                 FMODE_PREAD | FMODE_PWRITE;
 755         inode = dentry->d_inode;
 756         if (f->f_mode & FMODE_WRITE) {
 757                 error = get_write_access(inode);
 758                 if (error)
 759                         goto cleanup_file;
 760         }
 761
 762         f->f_mapping = inode->i_mapping;
 763         f->f_dentry = dentry;
 764         f->f_vfsmnt = mnt;
 765         f->f_pos = 0;
 766         f->f_op = fops_get(inode->i_fop);
 767         file_move(f, &inode->i_sb->s_files);
 768
 769         if (!open && f->f_op)
 770                 open = f->f_op->open;
 771         if (open) {
 772                 error = open(inode, f);
 773                 if (error)
 774                         goto cleanup_all;
 775         }
 776
 777         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 778
 779         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 780
 781         /* NB: we're sure to have correct a_ops only after f_op->open */
 782         if (f->f_flags & O_DIRECT) {
 783                 if (!f->f_mapping->a_ops ||
 784                     ((!f->f_mapping->a_ops->direct_IO) &&
 785                     (!f->f_mapping->a_ops->get_xip_page))) {
 786                         fput(f);
 787                         f = ERR_PTR(-EINVAL);
 788                 }
 789         }
 790
 791         return f;
 792
 793 cleanup_all:
 794         fops_put(f->f_op);
 795         if (f->f_mode & FMODE_WRITE)
 796                 put_write_access(inode);
 797         file_kill(f);
 798         f->f_dentry = NULL;
 799         f->f_vfsmnt = NULL;
 800 cleanup_file:
 801         put_filp(f);
 802         dput(dentry);
 803         mntput(mnt);
 804         return ERR_PTR(error);
 805 }
 806
 807 /*
 808  * Note that while the flag value (low two bits) for sys_open means:
 809  *      00 - read-only
 810  *      01 - write-only
 811  *      10 - read-write
 812  *      11 - special
 813  * it is changed into
 814  *      00 - no permissions needed
 815  *      01 - read-permission
 816  *      10 - write-permission
 817  *      11 - read-write
 818  * for the internal routines (ie open_namei()/follow_link() etc). 00 is
 819  * used by symlinks.
 820  */
 821 struct file *filp_open(const char * filename, int flags, int mode)
 822 {
 823         int namei_flags, error;
 824         struct nameidata nd;
 825
 826         namei_flags = flags;
 827         if ((namei_flags+1) & O_ACCMODE)
 828                 namei_flags++;
 829
 830         error = open_namei(filename, namei_flags, mode, &nd);
 831         if (!error)
 832                 return nameidata_to_filp(&nd, flags);
 833
 834         return ERR_PTR(error);
 835 }
 836 EXPORT_SYMBOL(filp_open);
 837
 838 /**
 839  * lookup_instantiate_filp - instantiates the open intent filp
 840  * @nd: pointer to nameidata
 841  * @dentry: pointer to dentry
 842  * @open: open callback
 843  *
 844  * Helper for filesystems that want to use lookup open intents and pass back
 845  * a fully instantiated struct file to the caller.
 846  * This function is meant to be called from within a filesystem's
 847  * lookup method.
 848  * Note that in case of error, nd->intent.open.file is destroyed, but the
 849  * path information remains valid.
 850  * If the open callback is set to NULL, then the standard f_op->open()
 851  * filesystem callback is substituted.
 852  */
 853 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 854                 int (*open)(struct inode *, struct file *))
 855 {
 856         if (IS_ERR(nd->intent.open.file))
 857                 goto out;
 858         if (IS_ERR(dentry))
 859                 goto out_err;
 860         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->mnt),
 861                                              nd->intent.open.flags - 1,
 862                                              nd->intent.open.file,
 863                                              open);
 864 out:
 865         return nd->intent.open.file;
 866 out_err:
 867         release_open_intent(nd);
 868         nd->intent.open.file = (struct file *)dentry;
 869         goto out;
 870 }
 871 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 872
 873 /**
 874  * nameidata_to_filp - convert a nameidata to an open filp.
 875  * @nd: pointer to nameidata
 876  * @flags: open flags
 877  *
 878  * Note that this function destroys the original nameidata
 879  */
 880 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 881 {
 882         struct file *filp;
 883
 884         /* Pick up the filp from the open intent */
 885         filp = nd->intent.open.file;
 886         /* Has the filesystem initialised the file for us? */
 887         if (filp->f_dentry == NULL)
 888                 filp = __dentry_open(nd->dentry, nd->mnt, flags, filp, NULL);
 889         else
 890                 path_release(nd);
 891         return filp;
 892 }
 893
 894 /*
 895  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 896  * error.
 897  */
 898 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 899 {
 900         int error;
 901         struct file *f;
 902
 903         error = -ENFILE;
 904         f = get_empty_filp();
 905         if (f == NULL) {
 906                 dput(dentry);
 907                 mntput(mnt);
 908                 return ERR_PTR(error);
 909         }
 910
 911         return __dentry_open(dentry, mnt, flags, f, NULL);
 912 }
 913 EXPORT_SYMBOL(dentry_open);
 914
 915 /*
 916  * Find an empty file descriptor entry, and mark it busy.
 917  */
 918 int get_unused_fd(void)
 919 {
 920         struct files_struct * files = current->files;
 921         int fd, error;
 922         struct fdtable *fdt;
 923
 924         error = -EMFILE;
 925         spin_lock(&files->file_lock);
 926
 927 repeat:
 928         fdt = files_fdtable(files);
 929         fd = find_next_zero_bit(fdt->open_fds->fds_bits,
 930                                 fdt->max_fdset,
 931                                 fdt->next_fd);
 932
 933         /*
 934          * N.B. For clone tasks sharing a files structure, this test
 935          * will limit the total number of files that can be opened.
 936          */
 937         if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 938                 goto out;
 939
 940         /* Do we need to expand the fd array or fd set?  */
 941         error = expand_files(files, fd);
 942         if (error < 0)
 943                 goto out;
 944
 945         if (error) {
 946                 /*
 947                  * If we needed to expand the fs array we
 948                  * might have blocked - try again.
 949                  */
 950                 error = -EMFILE;
 951                 goto repeat;
 952         }
 953
 954         FD_SET(fd, fdt->open_fds);
 955         FD_CLR(fd, fdt->close_on_exec);
 956         fdt->next_fd = fd + 1;
 957 #if 1
 958         /* Sanity check */
 959         if (fdt->fd[fd] != NULL) {
 960                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
 961                 fdt->fd[fd] = NULL;
 962         }
 963 #endif
 964         error = fd;
 965
 966 out:
 967         spin_unlock(&files->file_lock);
 968         return error;
 969 }
 970
 971 EXPORT_SYMBOL(get_unused_fd);
 972
 973 static inline void __put_unused_fd(struct files_struct *files, unsigned int fd)
 974 {
 975         struct fdtable *fdt = files_fdtable(files);
 976         __FD_CLR(fd, fdt->open_fds);
 977         if (fd < fdt->next_fd)
 978                 fdt->next_fd = fd;
 979 }
 980
 981 void fastcall put_unused_fd(unsigned int fd)
 982 {
 983         struct files_struct *files = current->files;
 984         spin_lock(&files->file_lock);
 985         __put_unused_fd(files, fd);
 986         spin_unlock(&files->file_lock);
 987 }
 988
 989 EXPORT_SYMBOL(put_unused_fd);
 990
 991 /*
 992  * Install a file pointer in the fd array.
 993  *
 994  * The VFS is full of places where we drop the files lock between
 995  * setting the open_fds bitmap and installing the file in the file
 996  * array.  At any such point, we are vulnerable to a dup2() race
 997  * installing a file in the array before us.  We need to detect this and
 998  * fput() the struct file we are about to overwrite in this case.
 999  *
1000  * It should never happen - if we allow dup2() do it, _really_ bad things
1001  * will follow.
1002  */
1003
1004 void fastcall fd_install(unsigned int fd, struct file * file)
1005 {
1006         struct files_struct *files = current->files;
1007         struct fdtable *fdt;
1008         spin_lock(&files->file_lock);
1009         fdt = files_fdtable(files);
1010         BUG_ON(fdt->fd[fd] != NULL);
1011         rcu_assign_pointer(fdt->fd[fd], file);
1012         spin_unlock(&files->file_lock);
1013 }
1014
1015 EXPORT_SYMBOL(fd_install);
1016
1017 long do_sys_open(const char __user *filename, int flags, int mode)
1018 {
1019         char *tmp = getname(filename);
1020         int fd = PTR_ERR(tmp);
1021
1022         if (!IS_ERR(tmp)) {
1023                 fd = get_unused_fd();
1024                 if (fd >= 0) {
1025                         struct file *f = filp_open(tmp, flags, mode);
1026                         if (IS_ERR(f)) {
1027                                 put_unused_fd(fd);
1028                                 fd = PTR_ERR(f);
1029                         } else {
1030                                 fsnotify_open(f->f_dentry);
1031                                 fd_install(fd, f);
1032                         }
1033                 }
1034                 putname(tmp);
1035         }
1036         return fd;
1037 }
1038
1039 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1040 {
1041         if (force_o_largefile())
1042                 flags |= O_LARGEFILE;
1043
1044         return do_sys_open(filename, flags, mode);
1045 }
1046 EXPORT_SYMBOL_GPL(sys_open);
1047
1048 #ifndef __alpha__
1049
1050 /*
1051  * For backward compatibility?  Maybe this should be moved
1052  * into arch/i386 instead?
1053  */
1054 asmlinkage long sys_creat(const char __user * pathname, int mode)
1055 {
1056         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1057 }
1058
1059 #endif
1060
1061 /*
1062  * "id" is the POSIX thread ID. We use the
1063  * files pointer for this..
1064  */
1065 int filp_close(struct file *filp, fl_owner_t id)
1066 {
1067         int retval = 0;
1068
1069         if (!file_count(filp)) {
1070                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1071                 return 0;
1072         }
1073
1074         if (filp->f_op && filp->f_op->flush)
1075                 retval = filp->f_op->flush(filp);
1076
1077         dnotify_flush(filp, id);
1078         locks_remove_posix(filp, id);
1079         fput(filp);
1080         return retval;
1081 }
1082
1083 EXPORT_SYMBOL(filp_close);
1084
1085 /*
1086  * Careful here! We test whether the file pointer is NULL before
1087  * releasing the fd. This ensures that one clone task can't release
1088  * an fd while another clone is opening it.
1089  */
1090 asmlinkage long sys_close(unsigned int fd)
1091 {
1092         struct file * filp;
1093         struct files_struct *files = current->files;
1094         struct fdtable *fdt;
1095
1096         spin_lock(&files->file_lock);
1097         fdt = files_fdtable(files);
1098         if (fd >= fdt->max_fds)
1099                 goto out_unlock;
1100         filp = fdt->fd[fd];
1101         if (!filp)
1102                 goto out_unlock;
1103         rcu_assign_pointer(fdt->fd[fd], NULL);
1104         FD_CLR(fd, fdt->close_on_exec);
1105         __put_unused_fd(files, fd);
1106         spin_unlock(&files->file_lock);
1107         return filp_close(filp, files);
1108
1109 out_unlock:
1110         spin_unlock(&files->file_lock);
1111         return -EBADF;
1112 }
1113
1114 EXPORT_SYMBOL(sys_close);
1115
1116 /*
1117  * This routine simulates a hangup on the tty, to arrange that users
1118  * are given clean terminals at login time.
1119  */
1120 asmlinkage long sys_vhangup(void)
1121 {
1122         if (capable(CAP_SYS_TTY_CONFIG)) {
1123                 tty_vhangup(current->signal->tty);
1124                 return 0;
1125         }
1126         return -EPERM;
1127 }
1128
1129 /*
1130  * Called when an inode is about to be open.
1131  * We use this to disallow opening large files on 32bit systems if
1132  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1133  * on this flag in sys_open.
1134  */
1135 int generic_file_open(struct inode * inode, struct file * filp)
1136 {
1137         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1138                 return -EFBIG;
1139         return 0;
1140 }
1141
1142 EXPORT_SYMBOL(generic_file_open);
1143
1144 /*
1145  * This is used by subsystems that don't want seekable
1146  * file descriptors
1147  */
1148 int nonseekable_open(struct inode *inode, struct file *filp)
1149 {
1150         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1151         return 0;
1152 }
1153
1154 EXPORT_SYMBOL(nonseekable_open);