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, unsigned int time_attrs,
 198         struct file *filp)
 199 {
 200         int err;
 201         struct iattr newattrs;
 202
 203         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 204         if (length < 0)
 205                 return -EINVAL;
 206
 207         newattrs.ia_size = length;
 208         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 209         if (filp) {
 210                 newattrs.ia_file = filp;
 211                 newattrs.ia_valid |= ATTR_FILE;
 212         }
 213
 214         down(&dentry->d_inode->i_sem);
 215         err = notify_change(dentry, &newattrs);
 216         up(&dentry->d_inode->i_sem);
 217         return err;
 218 }
 219
 220 static long do_sys_truncate(const char __user * path, loff_t length)
 221 {
 222         struct nameidata nd;
 223         struct inode * inode;
 224         int error;
 225
 226         error = -EINVAL;
 227         if (length < 0) /* sorry, but loff_t says... */
 228                 goto out;
 229
 230         error = user_path_walk(path, &nd);
 231         if (error)
 232                 goto out;
 233         inode = nd.dentry->d_inode;
 234
 235         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 236         error = -EISDIR;
 237         if (S_ISDIR(inode->i_mode))
 238                 goto dput_and_out;
 239
 240         error = -EINVAL;
 241         if (!S_ISREG(inode->i_mode))
 242                 goto dput_and_out;
 243
 244         error = vfs_permission(&nd, MAY_WRITE);
 245         if (error)
 246                 goto dput_and_out;
 247
 248         error = -EROFS;
 249         if (IS_RDONLY(inode))
 250                 goto dput_and_out;
 251
 252         error = -EPERM;
 253         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 254                 goto dput_and_out;
 255
 256         /*
 257          * Make sure that there are no leases.
 258          */
 259         error = break_lease(inode, FMODE_WRITE);
 260         if (error)
 261                 goto dput_and_out;
 262
 263         error = get_write_access(inode);
 264         if (error)
 265                 goto dput_and_out;
 266
 267         error = locks_verify_truncate(inode, NULL, length);
 268         if (!error) {
 269                 DQUOT_INIT(inode);
 270                 error = do_truncate(nd.dentry, length, 0, NULL);
 271         }
 272         put_write_access(inode);
 273
 274 dput_and_out:
 275         path_release(&nd);
 276 out:
 277         return error;
 278 }
 279
 280 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 281 {
 282         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 283         return do_sys_truncate(path, (long)length);
 284 }
 285
 286 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 287 {
 288         struct inode * inode;
 289         struct dentry *dentry;
 290         struct file * file;
 291         int error;
 292
 293         error = -EINVAL;
 294         if (length < 0)
 295                 goto out;
 296         error = -EBADF;
 297         file = fget(fd);
 298         if (!file)
 299                 goto out;
 300
 301         /* explicitly opened as large or we are on 64-bit box */
 302         if (file->f_flags & O_LARGEFILE)
 303                 small = 0;
 304
 305         dentry = file->f_dentry;
 306         inode = dentry->d_inode;
 307         error = -EINVAL;
 308         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 309                 goto out_putf;
 310
 311         error = -EINVAL;
 312         /* Cannot ftruncate over 2^31 bytes without large file support */
 313         if (small && length > MAX_NON_LFS)
 314                 goto out_putf;
 315
 316         error = -EPERM;
 317         if (IS_APPEND(inode))
 318                 goto out_putf;
 319
 320         error = locks_verify_truncate(inode, file, length);
 321         if (!error)
 322                 error = do_truncate(dentry, length, 0, file);
 323 out_putf:
 324         fput(file);
 325 out:
 326         return error;
 327 }
 328
 329 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 330 {
 331         return do_sys_ftruncate(fd, length, 1);
 332 }
 333
 334 /* LFS versions of truncate are only needed on 32 bit machines */
 335 #if BITS_PER_LONG == 32
 336 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 337 {
 338         return do_sys_truncate(path, length);
 339 }
 340
 341 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 342 {
 343         return do_sys_ftruncate(fd, length, 0);
 344 }
 345 #endif
 346
 347 #ifdef __ARCH_WANT_SYS_UTIME
 348
 349 /*
 350  * sys_utime() can be implemented in user-level using sys_utimes().
 351  * Is this for backwards compatibility?  If so, why not move it
 352  * into the appropriate arch directory (for those architectures that
 353  * need it).
 354  */
 355
 356 /* If times==NULL, set access and modification to current time,
 357  * must be owner or have write permission.
 358  * Else, update from *times, must be owner or super user.
 359  */
 360 asmlinkage long sys_utime(char __user * filename, struct utimbuf __user * times)
 361 {
 362         int error;
 363         struct nameidata nd;
 364         struct inode * inode;
 365         struct iattr newattrs;
 366
 367         error = user_path_walk(filename, &nd);
 368         if (error)
 369                 goto out;
 370         inode = nd.dentry->d_inode;
 371
 372         error = -EROFS;
 373         if (IS_RDONLY(inode))
 374                 goto dput_and_out;
 375
 376         /* Don't worry, the checks are done in inode_change_ok() */
 377         newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
 378         if (times) {
 379                 error = -EPERM;
 380                 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 381                         goto dput_and_out;
 382
 383                 error = get_user(newattrs.ia_atime.tv_sec, &times->actime);
 384                 newattrs.ia_atime.tv_nsec = 0;
 385                 if (!error)
 386                         error = get_user(newattrs.ia_mtime.tv_sec, &times->modtime);
 387                 newattrs.ia_mtime.tv_nsec = 0;
 388                 if (error)
 389                         goto dput_and_out;
 390
 391                 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
 392         } else {
 393                 error = -EACCES;
 394                 if (IS_IMMUTABLE(inode))
 395                         goto dput_and_out;
 396
 397                 if (current->fsuid != inode->i_uid &&
 398                     (error = vfs_permission(&nd, MAY_WRITE)) != 0)
 399                         goto dput_and_out;
 400         }
 401         down(&inode->i_sem);
 402         error = notify_change(nd.dentry, &newattrs);
 403         up(&inode->i_sem);
 404 dput_and_out:
 405         path_release(&nd);
 406 out:
 407         return error;
 408 }
 409
 410 #endif
 411
 412 /* If times==NULL, set access and modification to current time,
 413  * must be owner or have write permission.
 414  * Else, update from *times, must be owner or super user.
 415  */
 416 long do_utimes(char __user * filename, struct timeval * times)
 417 {
 418         int error;
 419         struct nameidata nd;
 420         struct inode * inode;
 421         struct iattr newattrs;
 422
 423         error = user_path_walk(filename, &nd);
 424
 425         if (error)
 426                 goto out;
 427         inode = nd.dentry->d_inode;
 428
 429         error = -EROFS;
 430         if (IS_RDONLY(inode))
 431                 goto dput_and_out;
 432
 433         /* Don't worry, the checks are done in inode_change_ok() */
 434         newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
 435         if (times) {
 436                 error = -EPERM;
 437                 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 438                         goto dput_and_out;
 439
 440                 newattrs.ia_atime.tv_sec = times[0].tv_sec;
 441                 newattrs.ia_atime.tv_nsec = times[0].tv_usec * 1000;
 442                 newattrs.ia_mtime.tv_sec = times[1].tv_sec;
 443                 newattrs.ia_mtime.tv_nsec = times[1].tv_usec * 1000;
 444                 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
 445         } else {
 446                 error = -EACCES;
 447                 if (IS_IMMUTABLE(inode))
 448                         goto dput_and_out;
 449
 450                 if (current->fsuid != inode->i_uid &&
 451                     (error = vfs_permission(&nd, MAY_WRITE)) != 0)
 452                         goto dput_and_out;
 453         }
 454         down(&inode->i_sem);
 455         error = notify_change(nd.dentry, &newattrs);
 456         up(&inode->i_sem);
 457 dput_and_out:
 458         path_release(&nd);
 459 out:
 460         return error;
 461 }
 462
 463 asmlinkage long sys_utimes(char __user * filename, struct timeval __user * utimes)
 464 {
 465         struct timeval times[2];
 466
 467         if (utimes && copy_from_user(&times, utimes, sizeof(times)))
 468                 return -EFAULT;
 469         return do_utimes(filename, utimes ? times : NULL);
 470 }
 471
 472
 473 /*
 474  * access() needs to use the real uid/gid, not the effective uid/gid.
 475  * We do this by temporarily clearing all FS-related capabilities and
 476  * switching the fsuid/fsgid around to the real ones.
 477  */
 478 asmlinkage long sys_access(const char __user * filename, int mode)
 479 {
 480         struct nameidata nd;
 481         int old_fsuid, old_fsgid;
 482         kernel_cap_t old_cap;
 483         int res;
 484
 485         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 486                 return -EINVAL;
 487
 488         old_fsuid = current->fsuid;
 489         old_fsgid = current->fsgid;
 490         old_cap = current->cap_effective;
 491
 492         current->fsuid = current->uid;
 493         current->fsgid = current->gid;
 494
 495         /*
 496          * Clear the capabilities if we switch to a non-root user
 497          *
 498          * FIXME: There is a race here against sys_capset.  The
 499          * capabilities can change yet we will restore the old
 500          * value below.  We should hold task_capabilities_lock,
 501          * but we cannot because user_path_walk can sleep.
 502          */
 503         if (current->uid)
 504                 cap_clear(current->cap_effective);
 505         else
 506                 current->cap_effective = current->cap_permitted;
 507
 508         res = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 509         if (!res) {
 510                 res = vfs_permission(&nd, mode);
 511                 /* SuS v2 requires we report a read only fs too */
 512                 if(!res && (mode & S_IWOTH) && IS_RDONLY(nd.dentry->d_inode)
 513                    && !special_file(nd.dentry->d_inode->i_mode))
 514                         res = -EROFS;
 515                 path_release(&nd);
 516         }
 517
 518         current->fsuid = old_fsuid;
 519         current->fsgid = old_fsgid;
 520         current->cap_effective = old_cap;
 521
 522         return res;
 523 }
 524
 525 asmlinkage long sys_chdir(const char __user * filename)
 526 {
 527         struct nameidata nd;
 528         int error;
 529
 530         error = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &nd);
 531         if (error)
 532                 goto out;
 533
 534         error = vfs_permission(&nd, MAY_EXEC);
 535         if (error)
 536                 goto dput_and_out;
 537
 538         set_fs_pwd(current->fs, nd.mnt, nd.dentry);
 539
 540 dput_and_out:
 541         path_release(&nd);
 542 out:
 543         return error;
 544 }
 545
 546 asmlinkage long sys_fchdir(unsigned int fd)
 547 {
 548         struct file *file;
 549         struct dentry *dentry;
 550         struct inode *inode;
 551         struct vfsmount *mnt;
 552         int error;
 553
 554         error = -EBADF;
 555         file = fget(fd);
 556         if (!file)
 557                 goto out;
 558
 559         dentry = file->f_dentry;
 560         mnt = file->f_vfsmnt;
 561         inode = dentry->d_inode;
 562
 563         error = -ENOTDIR;
 564         if (!S_ISDIR(inode->i_mode))
 565                 goto out_putf;
 566
 567         error = file_permission(file, MAY_EXEC);
 568         if (!error)
 569                 set_fs_pwd(current->fs, mnt, dentry);
 570 out_putf:
 571         fput(file);
 572 out:
 573         return error;
 574 }
 575
 576 asmlinkage long sys_chroot(const char __user * filename)
 577 {
 578         struct nameidata nd;
 579         int error;
 580
 581         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 582         if (error)
 583                 goto out;
 584
 585         error = vfs_permission(&nd, MAY_EXEC);
 586         if (error)
 587                 goto dput_and_out;
 588
 589         error = -EPERM;
 590         if (!capable(CAP_SYS_CHROOT))
 591                 goto dput_and_out;
 592
 593         set_fs_root(current->fs, nd.mnt, nd.dentry);
 594         set_fs_altroot();
 595         error = 0;
 596 dput_and_out:
 597         path_release(&nd);
 598 out:
 599         return error;
 600 }
 601
 602 asmlinkage long sys_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_dentry;
 615         inode = dentry->d_inode;
 616
 617         err = -EROFS;
 618         if (IS_RDONLY(inode))
 619                 goto out_putf;
 620         err = -EPERM;
 621         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 622                 goto out_putf;
 623         down(&inode->i_sem);
 624         if (mode == (mode_t) -1)
 625                 mode = inode->i_mode;
 626         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 627         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 628         err = notify_change(dentry, &newattrs);
 629         up(&inode->i_sem);
 630
 631 out_putf:
 632         fput(file);
 633 out:
 634         return err;
 635 }
 636
 637 asmlinkage long sys_chmod(const char __user * filename, mode_t mode)
 638 {
 639         struct nameidata nd;
 640         struct inode * inode;
 641         int error;
 642         struct iattr newattrs;
 643
 644         error = user_path_walk(filename, &nd);
 645         if (error)
 646                 goto out;
 647         inode = nd.dentry->d_inode;
 648
 649         error = -EROFS;
 650         if (IS_RDONLY(inode))
 651                 goto dput_and_out;
 652
 653         error = -EPERM;
 654         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 655                 goto dput_and_out;
 656
 657         down(&inode->i_sem);
 658         if (mode == (mode_t) -1)
 659                 mode = inode->i_mode;
 660         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 661         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 662         error = notify_change(nd.dentry, &newattrs);
 663         up(&inode->i_sem);
 664
 665 dput_and_out:
 666         path_release(&nd);
 667 out:
 668         return error;
 669 }
 670
 671 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 672 {
 673         struct inode * inode;
 674         int error;
 675         struct iattr newattrs;
 676
 677         error = -ENOENT;
 678         if (!(inode = dentry->d_inode)) {
 679                 printk(KERN_ERR "chown_common: NULL inode\n");
 680                 goto out;
 681         }
 682         error = -EROFS;
 683         if (IS_RDONLY(inode))
 684                 goto out;
 685         error = -EPERM;
 686         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 687                 goto out;
 688         newattrs.ia_valid =  ATTR_CTIME;
 689         if (user != (uid_t) -1) {
 690                 newattrs.ia_valid |= ATTR_UID;
 691                 newattrs.ia_uid = user;
 692         }
 693         if (group != (gid_t) -1) {
 694                 newattrs.ia_valid |= ATTR_GID;
 695                 newattrs.ia_gid = group;
 696         }
 697         if (!S_ISDIR(inode->i_mode))
 698                 newattrs.ia_valid |= ATTR_KILL_SUID|ATTR_KILL_SGID;
 699         down(&inode->i_sem);
 700         error = notify_change(dentry, &newattrs);
 701         up(&inode->i_sem);
 702 out:
 703         return error;
 704 }
 705
 706 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 707 {
 708         struct nameidata nd;
 709         int error;
 710
 711         error = user_path_walk(filename, &nd);
 712         if (!error) {
 713                 error = chown_common(nd.dentry, user, group);
 714                 path_release(&nd);
 715         }
 716         return error;
 717 }
 718
 719 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 720 {
 721         struct nameidata nd;
 722         int error;
 723
 724         error = user_path_walk_link(filename, &nd);
 725         if (!error) {
 726                 error = chown_common(nd.dentry, user, group);
 727                 path_release(&nd);
 728         }
 729         return error;
 730 }
 731
 732
 733 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 734 {
 735         struct file * file;
 736         int error = -EBADF;
 737
 738         file = fget(fd);
 739         if (file) {
 740                 error = chown_common(file->f_dentry, user, group);
 741                 fput(file);
 742         }
 743         return error;
 744 }
 745
 746 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 747                                         int flags, struct file *f,
 748                                         int (*open)(struct inode *, struct file *))
 749 {
 750         struct inode *inode;
 751         int error;
 752
 753         f->f_flags = flags;
 754         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 755                                 FMODE_PREAD | FMODE_PWRITE;
 756         inode = dentry->d_inode;
 757         if (f->f_mode & FMODE_WRITE) {
 758                 error = get_write_access(inode);
 759                 if (error)
 760                         goto cleanup_file;
 761         }
 762
 763         f->f_mapping = inode->i_mapping;
 764         f->f_dentry = dentry;
 765         f->f_vfsmnt = mnt;
 766         f->f_pos = 0;
 767         f->f_op = fops_get(inode->i_fop);
 768         file_move(f, &inode->i_sb->s_files);
 769
 770         if (!open && f->f_op)
 771                 open = f->f_op->open;
 772         if (open) {
 773                 error = open(inode, f);
 774                 if (error)
 775                         goto cleanup_all;
 776         }
 777
 778         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 779
 780         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 781
 782         /* NB: we're sure to have correct a_ops only after f_op->open */
 783         if (f->f_flags & O_DIRECT) {
 784                 if (!f->f_mapping->a_ops ||
 785                     ((!f->f_mapping->a_ops->direct_IO) &&
 786                     (!f->f_mapping->a_ops->get_xip_page))) {
 787                         fput(f);
 788                         f = ERR_PTR(-EINVAL);
 789                 }
 790         }
 791
 792         return f;
 793
 794 cleanup_all:
 795         fops_put(f->f_op);
 796         if (f->f_mode & FMODE_WRITE)
 797                 put_write_access(inode);
 798         file_kill(f);
 799         f->f_dentry = NULL;
 800         f->f_vfsmnt = NULL;
 801 cleanup_file:
 802         put_filp(f);
 803         dput(dentry);
 804         mntput(mnt);
 805         return ERR_PTR(error);
 806 }
 807
 808 /*
 809  * Note that while the flag value (low two bits) for sys_open means:
 810  *      00 - read-only
 811  *      01 - write-only
 812  *      10 - read-write
 813  *      11 - special
 814  * it is changed into
 815  *      00 - no permissions needed
 816  *      01 - read-permission
 817  *      10 - write-permission
 818  *      11 - read-write
 819  * for the internal routines (ie open_namei()/follow_link() etc). 00 is
 820  * used by symlinks.
 821  */
 822 struct file *filp_open(const char * filename, int flags, int mode)
 823 {
 824         int namei_flags, error;
 825         struct nameidata nd;
 826
 827         namei_flags = flags;
 828         if ((namei_flags+1) & O_ACCMODE)
 829                 namei_flags++;
 830
 831         error = open_namei(filename, namei_flags, mode, &nd);
 832         if (!error)
 833                 return nameidata_to_filp(&nd, flags);
 834
 835         return ERR_PTR(error);
 836 }
 837 EXPORT_SYMBOL(filp_open);
 838
 839 /**
 840  * lookup_instantiate_filp - instantiates the open intent filp
 841  * @nd: pointer to nameidata
 842  * @dentry: pointer to dentry
 843  * @open: open callback
 844  *
 845  * Helper for filesystems that want to use lookup open intents and pass back
 846  * a fully instantiated struct file to the caller.
 847  * This function is meant to be called from within a filesystem's
 848  * lookup method.
 849  * Note that in case of error, nd->intent.open.file is destroyed, but the
 850  * path information remains valid.
 851  * If the open callback is set to NULL, then the standard f_op->open()
 852  * filesystem callback is substituted.
 853  */
 854 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 855                 int (*open)(struct inode *, struct file *))
 856 {
 857         if (IS_ERR(nd->intent.open.file))
 858                 goto out;
 859         if (IS_ERR(dentry))
 860                 goto out_err;
 861         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->mnt),
 862                                              nd->intent.open.flags - 1,
 863                                              nd->intent.open.file,
 864                                              open);
 865 out:
 866         return nd->intent.open.file;
 867 out_err:
 868         release_open_intent(nd);
 869         nd->intent.open.file = (struct file *)dentry;
 870         goto out;
 871 }
 872 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 873
 874 /**
 875  * nameidata_to_filp - convert a nameidata to an open filp.
 876  * @nd: pointer to nameidata
 877  * @flags: open flags
 878  *
 879  * Note that this function destroys the original nameidata
 880  */
 881 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 882 {
 883         struct file *filp;
 884
 885         /* Pick up the filp from the open intent */
 886         filp = nd->intent.open.file;
 887         /* Has the filesystem initialised the file for us? */
 888         if (filp->f_dentry == NULL)
 889                 filp = __dentry_open(nd->dentry, nd->mnt, flags, filp, NULL);
 890         else
 891                 path_release(nd);
 892         return filp;
 893 }
 894
 895 /*
 896  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 897  * error.
 898  */
 899 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 900 {
 901         int error;
 902         struct file *f;
 903
 904         error = -ENFILE;
 905         f = get_empty_filp();
 906         if (f == NULL) {
 907                 dput(dentry);
 908                 mntput(mnt);
 909                 return ERR_PTR(error);
 910         }
 911
 912         return __dentry_open(dentry, mnt, flags, f, NULL);
 913 }
 914 EXPORT_SYMBOL(dentry_open);
 915
 916 /*
 917  * Find an empty file descriptor entry, and mark it busy.
 918  */
 919 int get_unused_fd(void)
 920 {
 921         struct files_struct * files = current->files;
 922         int fd, error;
 923         struct fdtable *fdt;
 924
 925         error = -EMFILE;
 926         spin_lock(&files->file_lock);
 927
 928 repeat:
 929         fdt = files_fdtable(files);
 930         fd = find_next_zero_bit(fdt->open_fds->fds_bits,
 931                                 fdt->max_fdset,
 932                                 fdt->next_fd);
 933
 934         /*
 935          * N.B. For clone tasks sharing a files structure, this test
 936          * will limit the total number of files that can be opened.
 937          */
 938         if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 939                 goto out;
 940
 941         /* Do we need to expand the fd array or fd set?  */
 942         error = expand_files(files, fd);
 943         if (error < 0)
 944                 goto out;
 945
 946         if (error) {
 947                 /*
 948                  * If we needed to expand the fs array we
 949                  * might have blocked - try again.
 950                  */
 951                 error = -EMFILE;
 952                 goto repeat;
 953         }
 954
 955         FD_SET(fd, fdt->open_fds);
 956         FD_CLR(fd, fdt->close_on_exec);
 957         fdt->next_fd = fd + 1;
 958 #if 1
 959         /* Sanity check */
 960         if (fdt->fd[fd] != NULL) {
 961                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
 962                 fdt->fd[fd] = NULL;
 963         }
 964 #endif
 965         error = fd;
 966
 967 out:
 968         spin_unlock(&files->file_lock);
 969         return error;
 970 }
 971
 972 EXPORT_SYMBOL(get_unused_fd);
 973
 974 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 975 {
 976         struct fdtable *fdt = files_fdtable(files);
 977         __FD_CLR(fd, fdt->open_fds);
 978         if (fd < fdt->next_fd)
 979                 fdt->next_fd = fd;
 980 }
 981
 982 void fastcall put_unused_fd(unsigned int fd)
 983 {
 984         struct files_struct *files = current->files;
 985         spin_lock(&files->file_lock);
 986         __put_unused_fd(files, fd);
 987         spin_unlock(&files->file_lock);
 988 }
 989
 990 EXPORT_SYMBOL(put_unused_fd);
 991
 992 /*
 993  * Install a file pointer in the fd array.
 994  *
 995  * The VFS is full of places where we drop the files lock between
 996  * setting the open_fds bitmap and installing the file in the file
 997  * array.  At any such point, we are vulnerable to a dup2() race
 998  * installing a file in the array before us.  We need to detect this and
 999  * fput() the struct file we are about to overwrite in this case.
1000  *
1001  * It should never happen - if we allow dup2() do it, _really_ bad things
1002  * will follow.
1003  */
1004
1005 void fastcall fd_install(unsigned int fd, struct file * file)
1006 {
1007         struct files_struct *files = current->files;
1008         struct fdtable *fdt;
1009         spin_lock(&files->file_lock);
1010         fdt = files_fdtable(files);
1011         BUG_ON(fdt->fd[fd] != NULL);
1012         rcu_assign_pointer(fdt->fd[fd], file);
1013         spin_unlock(&files->file_lock);
1014 }
1015
1016 EXPORT_SYMBOL(fd_install);
1017
1018 long do_sys_open(const char __user *filename, int flags, int mode)
1019 {
1020         char *tmp = getname(filename);
1021         int fd = PTR_ERR(tmp);
1022
1023         if (!IS_ERR(tmp)) {
1024                 fd = get_unused_fd();
1025                 if (fd >= 0) {
1026                         struct file *f = filp_open(tmp, flags, mode);
1027                         if (IS_ERR(f)) {
1028                                 put_unused_fd(fd);
1029                                 fd = PTR_ERR(f);
1030                         } else {
1031                                 fsnotify_open(f->f_dentry);
1032                                 fd_install(fd, f);
1033                         }
1034                 }
1035                 putname(tmp);
1036         }
1037         return fd;
1038 }
1039
1040 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1041 {
1042         if (force_o_largefile())
1043                 flags |= O_LARGEFILE;
1044
1045         return do_sys_open(filename, flags, mode);
1046 }
1047 EXPORT_SYMBOL_GPL(sys_open);
1048
1049 #ifndef __alpha__
1050
1051 /*
1052  * For backward compatibility?  Maybe this should be moved
1053  * into arch/i386 instead?
1054  */
1055 asmlinkage long sys_creat(const char __user * pathname, int mode)
1056 {
1057         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1058 }
1059
1060 #endif
1061
1062 /*
1063  * "id" is the POSIX thread ID. We use the
1064  * files pointer for this..
1065  */
1066 int filp_close(struct file *filp, fl_owner_t id)
1067 {
1068         int retval = 0;
1069
1070         if (!file_count(filp)) {
1071                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1072                 return 0;
1073         }
1074
1075         if (filp->f_op && filp->f_op->flush)
1076                 retval = filp->f_op->flush(filp);
1077
1078         dnotify_flush(filp, id);
1079         locks_remove_posix(filp, id);
1080         fput(filp);
1081         return retval;
1082 }
1083
1084 EXPORT_SYMBOL(filp_close);
1085
1086 /*
1087  * Careful here! We test whether the file pointer is NULL before
1088  * releasing the fd. This ensures that one clone task can't release
1089  * an fd while another clone is opening it.
1090  */
1091 asmlinkage long sys_close(unsigned int fd)
1092 {
1093         struct file * filp;
1094         struct files_struct *files = current->files;
1095         struct fdtable *fdt;
1096
1097         spin_lock(&files->file_lock);
1098         fdt = files_fdtable(files);
1099         if (fd >= fdt->max_fds)
1100                 goto out_unlock;
1101         filp = fdt->fd[fd];
1102         if (!filp)
1103                 goto out_unlock;
1104         rcu_assign_pointer(fdt->fd[fd], NULL);
1105         FD_CLR(fd, fdt->close_on_exec);
1106         __put_unused_fd(files, fd);
1107         spin_unlock(&files->file_lock);
1108         return filp_close(filp, files);
1109
1110 out_unlock:
1111         spin_unlock(&files->file_lock);
1112         return -EBADF;
1113 }
1114
1115 EXPORT_SYMBOL(sys_close);
1116
1117 /*
1118  * This routine simulates a hangup on the tty, to arrange that users
1119  * are given clean terminals at login time.
1120  */
1121 asmlinkage long sys_vhangup(void)
1122 {
1123         if (capable(CAP_SYS_TTY_CONFIG)) {
1124                 tty_vhangup(current->signal->tty);
1125                 return 0;
1126         }
1127         return -EPERM;
1128 }
1129
1130 /*
1131  * Called when an inode is about to be open.
1132  * We use this to disallow opening large files on 32bit systems if
1133  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1134  * on this flag in sys_open.
1135  */
1136 int generic_file_open(struct inode * inode, struct file * filp)
1137 {
1138         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1139                 return -EFBIG;
1140         return 0;
1141 }
1142
1143 EXPORT_SYMBOL(generic_file_open);
1144
1145 /*
1146  * This is used by subsystems that don't want seekable
1147  * file descriptors
1148  */
1149 int nonseekable_open(struct inode *inode, struct file *filp)
1150 {
1151         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1152         return 0;
1153 }
1154
1155 EXPORT_SYMBOL(nonseekable_open);