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/dnotify.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
  24 #define special_file(m) (S_ISCHR(m)||S_ISBLK(m)||S_ISFIFO(m)||S_ISSOCK(m))
  25
  26 int vfs_statfs(struct super_block *sb, struct kstatfs *buf)
  27 {
  28         int retval = -ENODEV;
  29
  30         if (sb) {
  31                 retval = -ENOSYS;
  32                 if (sb->s_op->statfs) {
  33                         memset(buf, 0, sizeof(*buf));
  34                         retval = security_sb_statfs(sb);
  35                         if (retval)
  36                                 return retval;
  37                         retval = sb->s_op->statfs(sb, buf);
  38                         if (retval == 0 && buf->f_frsize == 0)
  39                                 buf->f_frsize = buf->f_bsize;
  40                 }
  41         }
  42         return retval;
  43 }
  44
  45 static int vfs_statfs_native(struct super_block *sb, struct statfs *buf)
  46 {
  47         struct kstatfs st;
  48         int retval;
  49
  50         retval = vfs_statfs(sb, &st);
  51         if (retval)
  52                 return retval;
  53
  54         if (sizeof(*buf) == sizeof(st))
  55                 memcpy(buf, &st, sizeof(st));
  56         else {
  57                 if (sizeof buf->f_blocks == 4) {
  58                         if ((st.f_blocks | st.f_bfree |
  59                              st.f_bavail | st.f_files | st.f_ffree) &
  60                             0xffffffff00000000ULL)
  61                                 return -EOVERFLOW;
  62                 }
  63
  64                 buf->f_type = st.f_type;
  65                 buf->f_bsize = st.f_bsize;
  66                 buf->f_blocks = st.f_blocks;
  67                 buf->f_bfree = st.f_bfree;
  68                 buf->f_bavail = st.f_bavail;
  69                 buf->f_files = st.f_files;
  70                 buf->f_ffree = st.f_ffree;
  71                 buf->f_fsid = st.f_fsid;
  72                 buf->f_namelen = st.f_namelen;
  73                 buf->f_frsize = st.f_frsize;
  74                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  75         }
  76         return 0;
  77 }
  78
  79 static int vfs_statfs64(struct super_block *sb, struct statfs64 *buf)
  80 {
  81         struct kstatfs st;
  82         int retval;
  83
  84         retval = vfs_statfs(sb, &st);
  85         if (retval)
  86                 return retval;
  87
  88         if (sizeof(*buf) == sizeof(st))
  89                 memcpy(buf, &st, sizeof(st));
  90         else {
  91                 buf->f_type = st.f_type;
  92                 buf->f_bsize = st.f_bsize;
  93                 buf->f_blocks = st.f_blocks;
  94                 buf->f_bfree = st.f_bfree;
  95                 buf->f_bavail = st.f_bavail;
  96                 buf->f_files = st.f_files;
  97                 buf->f_ffree = st.f_ffree;
  98                 buf->f_fsid = st.f_fsid;
  99                 buf->f_namelen = st.f_namelen;
 100                 buf->f_frsize = st.f_frsize;
 101                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 102         }
 103         return 0;
 104 }
 105
 106 asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
 107 {
 108         struct nameidata nd;
 109         int error;
 110
 111         error = user_path_walk(path, &nd);
 112         if (!error) {
 113                 struct statfs tmp;
 114                 error = vfs_statfs_native(nd.dentry->d_inode->i_sb, &tmp);
 115                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 116                         error = -EFAULT;
 117                 path_release(&nd);
 118         }
 119         return error;
 120 }
 121
 122
 123 asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
 124 {
 125         struct nameidata nd;
 126         long error;
 127
 128         if (sz != sizeof(*buf))
 129                 return -EINVAL;
 130         error = user_path_walk(path, &nd);
 131         if (!error) {
 132                 struct statfs64 tmp;
 133                 error = vfs_statfs64(nd.dentry->d_inode->i_sb, &tmp);
 134                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 135                         error = -EFAULT;
 136                 path_release(&nd);
 137         }
 138         return error;
 139 }
 140
 141
 142 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
 143 {
 144         struct file * file;
 145         struct statfs tmp;
 146         int error;
 147
 148         error = -EBADF;
 149         file = fget(fd);
 150         if (!file)
 151                 goto out;
 152         error = vfs_statfs_native(file->f_dentry->d_inode->i_sb, &tmp);
 153         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 154                 error = -EFAULT;
 155         fput(file);
 156 out:
 157         return error;
 158 }
 159
 160 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
 161 {
 162         struct file * file;
 163         struct statfs64 tmp;
 164         int error;
 165
 166         if (sz != sizeof(*buf))
 167                 return -EINVAL;
 168
 169         error = -EBADF;
 170         file = fget(fd);
 171         if (!file)
 172                 goto out;
 173         error = vfs_statfs64(file->f_dentry->d_inode->i_sb, &tmp);
 174         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 175                 error = -EFAULT;
 176         fput(file);
 177 out:
 178         return error;
 179 }
 180
 181 int do_truncate(struct dentry *dentry, loff_t length)
 182 {
 183         int err;
 184         struct iattr newattrs;
 185
 186         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 187         if (length < 0)
 188                 return -EINVAL;
 189
 190         newattrs.ia_size = length;
 191         newattrs.ia_valid = ATTR_SIZE | ATTR_CTIME;
 192         down(&dentry->d_inode->i_sem);
 193         err = notify_change(dentry, &newattrs);
 194         up(&dentry->d_inode->i_sem);
 195         return err;
 196 }
 197
 198 static inline long do_sys_truncate(const char __user * path, loff_t length)
 199 {
 200         struct nameidata nd;
 201         struct inode * inode;
 202         int error;
 203
 204         error = -EINVAL;
 205         if (length < 0) /* sorry, but loff_t says... */
 206                 goto out;
 207
 208         error = user_path_walk(path, &nd);
 209         if (error)
 210                 goto out;
 211         inode = nd.dentry->d_inode;
 212
 213         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 214         error = -EISDIR;
 215         if (S_ISDIR(inode->i_mode))
 216                 goto dput_and_out;
 217
 218         error = -EINVAL;
 219         if (!S_ISREG(inode->i_mode))
 220                 goto dput_and_out;
 221
 222         error = permission(inode,MAY_WRITE,&nd);
 223         if (error)
 224                 goto dput_and_out;
 225
 226         error = -EROFS;
 227         if (IS_RDONLY(inode))
 228                 goto dput_and_out;
 229
 230         error = -EPERM;
 231         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 232                 goto dput_and_out;
 233
 234         /*
 235          * Make sure that there are no leases.
 236          */
 237         error = break_lease(inode, FMODE_WRITE);
 238         if (error)
 239                 goto dput_and_out;
 240
 241         error = get_write_access(inode);
 242         if (error)
 243                 goto dput_and_out;
 244
 245         error = locks_verify_truncate(inode, NULL, length);
 246         if (!error) {
 247                 DQUOT_INIT(inode);
 248                 error = do_truncate(nd.dentry, length);
 249         }
 250         put_write_access(inode);
 251
 252 dput_and_out:
 253         path_release(&nd);
 254 out:
 255         return error;
 256 }
 257
 258 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 259 {
 260         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 261         return do_sys_truncate(path, (long)length);
 262 }
 263
 264 static inline long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 265 {
 266         struct inode * inode;
 267         struct dentry *dentry;
 268         struct file * file;
 269         int error;
 270
 271         error = -EINVAL;
 272         if (length < 0)
 273                 goto out;
 274         error = -EBADF;
 275         file = fget(fd);
 276         if (!file)
 277                 goto out;
 278
 279         /* explicitly opened as large or we are on 64-bit box */
 280         if (file->f_flags & O_LARGEFILE)
 281                 small = 0;
 282
 283         dentry = file->f_dentry;
 284         inode = dentry->d_inode;
 285         error = -EINVAL;
 286         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 287                 goto out_putf;
 288
 289         error = -EINVAL;
 290         /* Cannot ftruncate over 2^31 bytes without large file support */
 291         if (small && length > MAX_NON_LFS)
 292                 goto out_putf;
 293
 294         error = -EPERM;
 295         if (IS_APPEND(inode))
 296                 goto out_putf;
 297
 298         error = locks_verify_truncate(inode, file, length);
 299         if (!error)
 300                 error = do_truncate(dentry, length);
 301 out_putf:
 302         fput(file);
 303 out:
 304         return error;
 305 }
 306
 307 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 308 {
 309         return do_sys_ftruncate(fd, length, 1);
 310 }
 311
 312 /* LFS versions of truncate are only needed on 32 bit machines */
 313 #if BITS_PER_LONG == 32
 314 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 315 {
 316         return do_sys_truncate(path, length);
 317 }
 318
 319 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 320 {
 321         return do_sys_ftruncate(fd, length, 0);
 322 }
 323 #endif
 324
 325 #if !(defined(__alpha__) || defined(__ia64__))
 326
 327 /*
 328  * sys_utime() can be implemented in user-level using sys_utimes().
 329  * Is this for backwards compatibility?  If so, why not move it
 330  * into the appropriate arch directory (for those architectures that
 331  * need it).
 332  */
 333
 334 /* If times==NULL, set access and modification to current time,
 335  * must be owner or have write permission.
 336  * Else, update from *times, must be owner or super user.
 337  */
 338 asmlinkage long sys_utime(char __user * filename, struct utimbuf __user * times)
 339 {
 340         int error;
 341         struct nameidata nd;
 342         struct inode * inode;
 343         struct iattr newattrs;
 344
 345         error = user_path_walk(filename, &nd);
 346         if (error)
 347                 goto out;
 348         inode = nd.dentry->d_inode;
 349
 350         error = -EROFS;
 351         if (IS_RDONLY(inode))
 352                 goto dput_and_out;
 353
 354         /* Don't worry, the checks are done in inode_change_ok() */
 355         newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
 356         if (times) {
 357                 error = get_user(newattrs.ia_atime.tv_sec, &times->actime);
 358                 newattrs.ia_atime.tv_nsec = 0;
 359                 if (!error)
 360                         error = get_user(newattrs.ia_mtime.tv_sec, &times->modtime);
 361                 newattrs.ia_mtime.tv_nsec = 0;
 362                 if (error)
 363                         goto dput_and_out;
 364
 365                 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
 366         } else {
 367                 if (current->fsuid != inode->i_uid &&
 368                     (error = permission(inode,MAY_WRITE,&nd)) != 0)
 369                         goto dput_and_out;
 370         }
 371         down(&inode->i_sem);
 372         error = notify_change(nd.dentry, &newattrs);
 373         up(&inode->i_sem);
 374 dput_and_out:
 375         path_release(&nd);
 376 out:
 377         return error;
 378 }
 379
 380 #endif
 381
 382 /* If times==NULL, set access and modification to current time,
 383  * must be owner or have write permission.
 384  * Else, update from *times, must be owner or super user.
 385  */
 386 long do_utimes(char __user * filename, struct timeval * times)
 387 {
 388         int error;
 389         struct nameidata nd;
 390         struct inode * inode;
 391         struct iattr newattrs;
 392
 393         error = user_path_walk(filename, &nd);
 394
 395         if (error)
 396                 goto out;
 397         inode = nd.dentry->d_inode;
 398
 399         error = -EROFS;
 400         if (IS_RDONLY(inode))
 401                 goto dput_and_out;
 402
 403         /* Don't worry, the checks are done in inode_change_ok() */
 404         newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
 405         if (times) {
 406                 newattrs.ia_atime.tv_sec = times[0].tv_sec;
 407                 newattrs.ia_atime.tv_nsec = times[0].tv_usec * 1000;
 408                 newattrs.ia_mtime.tv_sec = times[1].tv_sec;
 409                 newattrs.ia_mtime.tv_nsec = times[1].tv_usec * 1000;
 410                 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
 411         } else {
 412                 if (current->fsuid != inode->i_uid &&
 413                     (error = permission(inode,MAY_WRITE,&nd)) != 0)
 414                         goto dput_and_out;
 415         }
 416         down(&inode->i_sem);
 417         error = notify_change(nd.dentry, &newattrs);
 418         up(&inode->i_sem);
 419 dput_and_out:
 420         path_release(&nd);
 421 out:
 422         return error;
 423 }
 424
 425 asmlinkage long sys_utimes(char __user * filename, struct timeval __user * utimes)
 426 {
 427         struct timeval times[2];
 428
 429         if (utimes && copy_from_user(&times, utimes, sizeof(times)))
 430                 return -EFAULT;
 431         return do_utimes(filename, utimes ? times : NULL);
 432 }
 433
 434
 435 /*
 436  * access() needs to use the real uid/gid, not the effective uid/gid.
 437  * We do this by temporarily clearing all FS-related capabilities and
 438  * switching the fsuid/fsgid around to the real ones.
 439  */
 440 asmlinkage long sys_access(const char __user * filename, int mode)
 441 {
 442         struct nameidata nd;
 443         int old_fsuid, old_fsgid;
 444         kernel_cap_t old_cap;
 445         int res;
 446
 447         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 448                 return -EINVAL;
 449
 450         old_fsuid = current->fsuid;
 451         old_fsgid = current->fsgid;
 452         old_cap = current->cap_effective;
 453
 454         current->fsuid = current->uid;
 455         current->fsgid = current->gid;
 456
 457         /*
 458          * Clear the capabilities if we switch to a non-root user
 459          *
 460          * FIXME: There is a race here against sys_capset.  The
 461          * capabilities can change yet we will restore the old
 462          * value below.  We should hold task_capabilities_lock,
 463          * but we cannot because user_path_walk can sleep.
 464          */
 465         if (current->uid)
 466                 cap_clear(current->cap_effective);
 467         else
 468                 current->cap_effective = current->cap_permitted;
 469
 470         res = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 471         if (!res) {
 472                 res = permission(nd.dentry->d_inode, mode, &nd);
 473                 /* SuS v2 requires we report a read only fs too */
 474                 if(!res && (mode & S_IWOTH) && IS_RDONLY(nd.dentry->d_inode)
 475                    && !special_file(nd.dentry->d_inode->i_mode))
 476                         res = -EROFS;
 477                 path_release(&nd);
 478         }
 479
 480         current->fsuid = old_fsuid;
 481         current->fsgid = old_fsgid;
 482         current->cap_effective = old_cap;
 483
 484         return res;
 485 }
 486
 487 asmlinkage long sys_chdir(const char __user * filename)
 488 {
 489         struct nameidata nd;
 490         int error;
 491
 492         error = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &nd);
 493         if (error)
 494                 goto out;
 495
 496         error = permission(nd.dentry->d_inode,MAY_EXEC,&nd);
 497         if (error)
 498                 goto dput_and_out;
 499
 500         set_fs_pwd(current->fs, nd.mnt, nd.dentry);
 501
 502 dput_and_out:
 503         path_release(&nd);
 504 out:
 505         return error;
 506 }
 507
 508 asmlinkage long sys_fchdir(unsigned int fd)
 509 {
 510         struct file *file;
 511         struct dentry *dentry;
 512         struct inode *inode;
 513         struct vfsmount *mnt;
 514         int error;
 515
 516         error = -EBADF;
 517         file = fget(fd);
 518         if (!file)
 519                 goto out;
 520
 521         dentry = file->f_dentry;
 522         mnt = file->f_vfsmnt;
 523         inode = dentry->d_inode;
 524
 525         error = -ENOTDIR;
 526         if (!S_ISDIR(inode->i_mode))
 527                 goto out_putf;
 528
 529         error = permission(inode, MAY_EXEC, NULL);
 530         if (!error)
 531                 set_fs_pwd(current->fs, mnt, dentry);
 532 out_putf:
 533         fput(file);
 534 out:
 535         return error;
 536 }
 537
 538 asmlinkage long sys_chroot(const char __user * filename)
 539 {
 540         struct nameidata nd;
 541         int error;
 542
 543         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 544         if (error)
 545                 goto out;
 546
 547         error = permission(nd.dentry->d_inode,MAY_EXEC,&nd);
 548         if (error)
 549                 goto dput_and_out;
 550
 551         error = -EPERM;
 552         if (!capable(CAP_SYS_CHROOT))
 553                 goto dput_and_out;
 554
 555         set_fs_root(current->fs, nd.mnt, nd.dentry);
 556         set_fs_altroot();
 557         error = 0;
 558 dput_and_out:
 559         path_release(&nd);
 560 out:
 561         return error;
 562 }
 563
 564 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 565 {
 566         struct inode * inode;
 567         struct dentry * dentry;
 568         struct file * file;
 569         int err = -EBADF;
 570         struct iattr newattrs;
 571
 572         file = fget(fd);
 573         if (!file)
 574                 goto out;
 575
 576         dentry = file->f_dentry;
 577         inode = dentry->d_inode;
 578
 579         err = -EROFS;
 580         if (IS_RDONLY(inode))
 581                 goto out_putf;
 582         err = -EPERM;
 583         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 584                 goto out_putf;
 585         down(&inode->i_sem);
 586         if (mode == (mode_t) -1)
 587                 mode = inode->i_mode;
 588         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 589         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 590         err = notify_change(dentry, &newattrs);
 591         up(&inode->i_sem);
 592
 593 out_putf:
 594         fput(file);
 595 out:
 596         return err;
 597 }
 598
 599 asmlinkage long sys_chmod(const char __user * filename, mode_t mode)
 600 {
 601         struct nameidata nd;
 602         struct inode * inode;
 603         int error;
 604         struct iattr newattrs;
 605
 606         error = user_path_walk(filename, &nd);
 607         if (error)
 608                 goto out;
 609         inode = nd.dentry->d_inode;
 610
 611         error = -EROFS;
 612         if (IS_RDONLY(inode))
 613                 goto dput_and_out;
 614
 615         error = -EPERM;
 616         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 617                 goto dput_and_out;
 618
 619         down(&inode->i_sem);
 620         if (mode == (mode_t) -1)
 621                 mode = inode->i_mode;
 622         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 623         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 624         error = notify_change(nd.dentry, &newattrs);
 625         up(&inode->i_sem);
 626
 627 dput_and_out:
 628         path_release(&nd);
 629 out:
 630         return error;
 631 }
 632
 633 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 634 {
 635         struct inode * inode;
 636         int error;
 637         struct iattr newattrs;
 638
 639         error = -ENOENT;
 640         if (!(inode = dentry->d_inode)) {
 641                 printk(KERN_ERR "chown_common: NULL inode\n");
 642                 goto out;
 643         }
 644         error = -EROFS;
 645         if (IS_RDONLY(inode))
 646                 goto out;
 647         error = -EPERM;
 648         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 649                 goto out;
 650         newattrs.ia_valid =  ATTR_CTIME;
 651         if (user != (uid_t) -1) {
 652                 newattrs.ia_valid |= ATTR_UID;
 653                 newattrs.ia_uid = user;
 654         }
 655         if (group != (gid_t) -1) {
 656                 newattrs.ia_valid |= ATTR_GID;
 657                 newattrs.ia_gid = group;
 658         }
 659         if (!S_ISDIR(inode->i_mode))
 660                 newattrs.ia_valid |= ATTR_KILL_SUID|ATTR_KILL_SGID;
 661         down(&inode->i_sem);
 662         error = notify_change(dentry, &newattrs);
 663         up(&inode->i_sem);
 664 out:
 665         return error;
 666 }
 667
 668 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 669 {
 670         struct nameidata nd;
 671         int error;
 672
 673         error = user_path_walk(filename, &nd);
 674         if (!error) {
 675                 error = chown_common(nd.dentry, user, group);
 676                 path_release(&nd);
 677         }
 678         return error;
 679 }
 680
 681 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 682 {
 683         struct nameidata nd;
 684         int error;
 685
 686         error = user_path_walk_link(filename, &nd);
 687         if (!error) {
 688                 error = chown_common(nd.dentry, user, group);
 689                 path_release(&nd);
 690         }
 691         return error;
 692 }
 693
 694
 695 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 696 {
 697         struct file * file;
 698         int error = -EBADF;
 699
 700         file = fget(fd);
 701         if (file) {
 702                 error = chown_common(file->f_dentry, user, group);
 703                 fput(file);
 704         }
 705         return error;
 706 }
 707
 708 /*
 709  * Note that while the flag value (low two bits) for sys_open means:
 710  *      00 - read-only
 711  *      01 - write-only
 712  *      10 - read-write
 713  *      11 - special
 714  * it is changed into
 715  *      00 - no permissions needed
 716  *      01 - read-permission
 717  *      10 - write-permission
 718  *      11 - read-write
 719  * for the internal routines (ie open_namei()/follow_link() etc). 00 is
 720  * used by symlinks.
 721  */
 722 struct file *filp_open(const char * filename, int flags, int mode)
 723 {
 724         int namei_flags, error;
 725         struct nameidata nd;
 726
 727         namei_flags = flags;
 728         if ((namei_flags+1) & O_ACCMODE)
 729                 namei_flags++;
 730         if (namei_flags & O_TRUNC)
 731                 namei_flags |= 2;
 732
 733         error = open_namei(filename, namei_flags, mode, &nd);
 734         if (!error)
 735                 return dentry_open(nd.dentry, nd.mnt, flags);
 736
 737         return ERR_PTR(error);
 738 }
 739
 740 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 741 {
 742         struct file * f;
 743         struct inode *inode;
 744         int error;
 745
 746         error = -ENFILE;
 747         f = get_empty_filp();
 748         if (!f)
 749                 goto cleanup_dentry;
 750         f->f_flags = flags;
 751         f->f_mode = (flags+1) & O_ACCMODE;
 752         inode = dentry->d_inode;
 753         if (f->f_mode & FMODE_WRITE) {
 754                 error = get_write_access(inode);
 755                 if (error)
 756                         goto cleanup_file;
 757         }
 758
 759         file_ra_state_init(&f->f_ra, inode->i_mapping);
 760         f->f_dentry = dentry;
 761         f->f_vfsmnt = mnt;
 762         f->f_pos = 0;
 763         f->f_op = fops_get(inode->i_fop);
 764         file_move(f, &inode->i_sb->s_files);
 765
 766         if (f->f_op && f->f_op->open) {
 767                 error = f->f_op->open(inode,f);
 768                 if (error)
 769                         goto cleanup_all;
 770         }
 771         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 772
 773         /* NB: we're sure to have correct a_ops only after f_op->open */
 774         if (f->f_flags & O_DIRECT) {
 775                 if (!inode->i_mapping || !inode->i_mapping->a_ops ||
 776                         !inode->i_mapping->a_ops->direct_IO) {
 777                                 fput(f);
 778                                 f = ERR_PTR(-EINVAL);
 779                 }
 780         }
 781
 782         return f;
 783
 784 cleanup_all:
 785         fops_put(f->f_op);
 786         if (f->f_mode & FMODE_WRITE)
 787                 put_write_access(inode);
 788         file_kill(f);
 789         f->f_dentry = NULL;
 790         f->f_vfsmnt = NULL;
 791 cleanup_file:
 792         put_filp(f);
 793 cleanup_dentry:
 794         dput(dentry);
 795         mntput(mnt);
 796         return ERR_PTR(error);
 797 }
 798
 799 /*
 800  * Find an empty file descriptor entry, and mark it busy.
 801  */
 802 int get_unused_fd(void)
 803 {
 804         struct files_struct * files = current->files;
 805         int fd, error;
 806
 807         error = -EMFILE;
 808         spin_lock(&files->file_lock);
 809
 810 repeat:
 811         fd = find_next_zero_bit(files->open_fds->fds_bits,
 812                                 files->max_fdset,
 813                                 files->next_fd);
 814
 815         /*
 816          * N.B. For clone tasks sharing a files structure, this test
 817          * will limit the total number of files that can be opened.
 818          */
 819         if (fd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
 820                 goto out;
 821
 822         /* Do we need to expand the fdset array? */
 823         if (fd >= files->max_fdset) {
 824                 error = expand_fdset(files, fd);
 825                 if (!error) {
 826                         error = -EMFILE;
 827                         goto repeat;
 828                 }
 829                 goto out;
 830         }
 831
 832         /*
 833          * Check whether we need to expand the fd array.
 834          */
 835         if (fd >= files->max_fds) {
 836                 error = expand_fd_array(files, fd);
 837                 if (!error) {
 838                         error = -EMFILE;
 839                         goto repeat;
 840                 }
 841                 goto out;
 842         }
 843
 844         FD_SET(fd, files->open_fds);
 845         FD_CLR(fd, files->close_on_exec);
 846         files->next_fd = fd + 1;
 847 #if 1
 848         /* Sanity check */
 849         if (files->fd[fd] != NULL) {
 850                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
 851                 files->fd[fd] = NULL;
 852         }
 853 #endif
 854         error = fd;
 855
 856 out:
 857         spin_unlock(&files->file_lock);
 858         return error;
 859 }
 860
 861 static inline void __put_unused_fd(struct files_struct *files, unsigned int fd)
 862 {
 863         __FD_CLR(fd, files->open_fds);
 864         if (fd < files->next_fd)
 865                 files->next_fd = fd;
 866 }
 867
 868 void put_unused_fd(unsigned int fd)
 869 {
 870         struct files_struct *files = current->files;
 871         spin_lock(&files->file_lock);
 872         __put_unused_fd(files, fd);
 873         spin_unlock(&files->file_lock);
 874 }
 875
 876 /*
 877  * Install a file pointer in the fd array.
 878  *
 879  * The VFS is full of places where we drop the files lock between
 880  * setting the open_fds bitmap and installing the file in the file
 881  * array.  At any such point, we are vulnerable to a dup2() race
 882  * installing a file in the array before us.  We need to detect this and
 883  * fput() the struct file we are about to overwrite in this case.
 884  *
 885  * It should never happen - if we allow dup2() do it, _really_ bad things
 886  * will follow.
 887  */
 888
 889 void fd_install(unsigned int fd, struct file * file)
 890 {
 891         struct files_struct *files = current->files;
 892         spin_lock(&files->file_lock);
 893         if (unlikely(files->fd[fd] != NULL))
 894                 BUG();
 895         files->fd[fd] = file;
 896         spin_unlock(&files->file_lock);
 897 }
 898
 899 asmlinkage long sys_open(const char __user * filename, int flags, int mode)
 900 {
 901         char * tmp;
 902         int fd, error;
 903
 904 #if BITS_PER_LONG != 32
 905         flags |= O_LARGEFILE;
 906 #endif
 907         tmp = getname(filename);
 908         fd = PTR_ERR(tmp);
 909         if (!IS_ERR(tmp)) {
 910                 fd = get_unused_fd();
 911                 if (fd >= 0) {
 912                         struct file *f = filp_open(tmp, flags, mode);
 913                         error = PTR_ERR(f);
 914                         if (IS_ERR(f))
 915                                 goto out_error;
 916                         fd_install(fd, f);
 917                 }
 918 out:
 919                 putname(tmp);
 920         }
 921         return fd;
 922
 923 out_error:
 924         put_unused_fd(fd);
 925         fd = error;
 926         goto out;
 927 }
 928
 929 #ifndef __alpha__
 930
 931 /*
 932  * For backward compatibility?  Maybe this should be moved
 933  * into arch/i386 instead?
 934  */
 935 asmlinkage long sys_creat(const char __user * pathname, int mode)
 936 {
 937         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
 938 }
 939
 940 #endif
 941
 942 /*
 943  * "id" is the POSIX thread ID. We use the
 944  * files pointer for this..
 945  */
 946 int filp_close(struct file *filp, fl_owner_t id)
 947 {
 948         int retval;
 949
 950         if (!file_count(filp)) {
 951                 printk(KERN_ERR "VFS: Close: file count is 0\n");
 952                 return 0;
 953         }
 954         retval = 0;
 955         if (filp->f_op && filp->f_op->flush)
 956                 retval = filp->f_op->flush(filp);
 957         dnotify_flush(filp, id);
 958         locks_remove_posix(filp, id);
 959         fput(filp);
 960         return retval;
 961 }
 962
 963 /*
 964  * Careful here! We test whether the file pointer is NULL before
 965  * releasing the fd. This ensures that one clone task can't release
 966  * an fd while another clone is opening it.
 967  */
 968 asmlinkage long sys_close(unsigned int fd)
 969 {
 970         struct file * filp;
 971         struct files_struct *files = current->files;
 972
 973         spin_lock(&files->file_lock);
 974         if (fd >= files->max_fds)
 975                 goto out_unlock;
 976         filp = files->fd[fd];
 977         if (!filp)
 978                 goto out_unlock;
 979         files->fd[fd] = NULL;
 980         FD_CLR(fd, files->close_on_exec);
 981         __put_unused_fd(files, fd);
 982         spin_unlock(&files->file_lock);
 983         return filp_close(filp, files);
 984
 985 out_unlock:
 986         spin_unlock(&files->file_lock);
 987         return -EBADF;
 988 }
 989
 990 /*
 991  * This routine simulates a hangup on the tty, to arrange that users
 992  * are given clean terminals at login time.
 993  */
 994 asmlinkage long sys_vhangup(void)
 995 {
 996         if (capable(CAP_SYS_TTY_CONFIG)) {
 997                 tty_vhangup(current->tty);
 998                 return 0;
 999         }
1000         return -EPERM;
1001 }
1002
1003 /*
1004  * Called when an inode is about to be open.
1005  * We use this to disallow opening large files on 32bit systems if
1006  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1007  * on this flag in sys_open.
1008  */
1009 int generic_file_open(struct inode * inode, struct file * filp)
1010 {
1011         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1012                 return -EFBIG;
1013         return 0;
1014 }
1015
1016 EXPORT_SYMBOL(generic_file_open);