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 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  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 EXPORT_SYMBOL(vfs_statfs);
  46
  47 static int vfs_statfs_native(struct super_block *sb, struct statfs *buf)
  48 {
  49         struct kstatfs st;
  50         int retval;
  51
  52         retval = vfs_statfs(sb, &st);
  53         if (retval)
  54                 return retval;
  55
  56         if (sizeof(*buf) == sizeof(st))
  57                 memcpy(buf, &st, sizeof(st));
  58         else {
  59                 if (sizeof buf->f_blocks == 4) {
  60                         if ((st.f_blocks | st.f_bfree |
  61                              st.f_bavail | st.f_files | st.f_ffree) &
  62                             0xffffffff00000000ULL)
  63                                 return -EOVERFLOW;
  64                 }
  65
  66                 buf->f_type = st.f_type;
  67                 buf->f_bsize = st.f_bsize;
  68                 buf->f_blocks = st.f_blocks;
  69                 buf->f_bfree = st.f_bfree;
  70                 buf->f_bavail = st.f_bavail;
  71                 buf->f_files = st.f_files;
  72                 buf->f_ffree = st.f_ffree;
  73                 buf->f_fsid = st.f_fsid;
  74                 buf->f_namelen = st.f_namelen;
  75                 buf->f_frsize = st.f_frsize;
  76                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  77         }
  78         return 0;
  79 }
  80
  81 static int vfs_statfs64(struct super_block *sb, struct statfs64 *buf)
  82 {
  83         struct kstatfs st;
  84         int retval;
  85
  86         retval = vfs_statfs(sb, &st);
  87         if (retval)
  88                 return retval;
  89
  90         if (sizeof(*buf) == sizeof(st))
  91                 memcpy(buf, &st, sizeof(st));
  92         else {
  93                 buf->f_type = st.f_type;
  94                 buf->f_bsize = st.f_bsize;
  95                 buf->f_blocks = st.f_blocks;
  96                 buf->f_bfree = st.f_bfree;
  97                 buf->f_bavail = st.f_bavail;
  98                 buf->f_files = st.f_files;
  99                 buf->f_ffree = st.f_ffree;
 100                 buf->f_fsid = st.f_fsid;
 101                 buf->f_namelen = st.f_namelen;
 102                 buf->f_frsize = st.f_frsize;
 103                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 104         }
 105         return 0;
 106 }
 107
 108 asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
 109 {
 110         struct nameidata nd;
 111         int error;
 112
 113         error = user_path_walk(path, &nd);
 114         if (!error) {
 115                 struct statfs tmp;
 116                 error = vfs_statfs_native(nd.dentry->d_inode->i_sb, &tmp);
 117                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 118                         error = -EFAULT;
 119                 path_release(&nd);
 120         }
 121         return error;
 122 }
 123
 124
 125 asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
 126 {
 127         struct nameidata nd;
 128         long error;
 129
 130         if (sz != sizeof(*buf))
 131                 return -EINVAL;
 132         error = user_path_walk(path, &nd);
 133         if (!error) {
 134                 struct statfs64 tmp;
 135                 error = vfs_statfs64(nd.dentry->d_inode->i_sb, &tmp);
 136                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 137                         error = -EFAULT;
 138                 path_release(&nd);
 139         }
 140         return error;
 141 }
 142
 143
 144 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
 145 {
 146         struct file * file;
 147         struct statfs tmp;
 148         int error;
 149
 150         error = -EBADF;
 151         file = fget(fd);
 152         if (!file)
 153                 goto out;
 154         error = vfs_statfs_native(file->f_dentry->d_inode->i_sb, &tmp);
 155         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 156                 error = -EFAULT;
 157         fput(file);
 158 out:
 159         return error;
 160 }
 161
 162 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
 163 {
 164         struct file * file;
 165         struct statfs64 tmp;
 166         int error;
 167
 168         if (sz != sizeof(*buf))
 169                 return -EINVAL;
 170
 171         error = -EBADF;
 172         file = fget(fd);
 173         if (!file)
 174                 goto out;
 175         error = vfs_statfs64(file->f_dentry->d_inode->i_sb, &tmp);
 176         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 177                 error = -EFAULT;
 178         fput(file);
 179 out:
 180         return error;
 181 }
 182
 183 int do_truncate(struct dentry *dentry, loff_t length)
 184 {
 185         int err;
 186         struct iattr newattrs;
 187
 188         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 189         if (length < 0)
 190                 return -EINVAL;
 191
 192         newattrs.ia_size = length;
 193         newattrs.ia_valid = ATTR_SIZE | ATTR_CTIME;
 194         down(&dentry->d_inode->i_sem);
 195         err = notify_change(dentry, &newattrs);
 196         up(&dentry->d_inode->i_sem);
 197         return err;
 198 }
 199
 200 static inline long do_sys_truncate(const char __user * path, loff_t length)
 201 {
 202         struct nameidata nd;
 203         struct inode * inode;
 204         int error;
 205
 206         error = -EINVAL;
 207         if (length < 0) /* sorry, but loff_t says... */
 208                 goto out;
 209
 210         error = user_path_walk(path, &nd);
 211         if (error)
 212                 goto out;
 213         inode = nd.dentry->d_inode;
 214
 215         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 216         error = -EISDIR;
 217         if (S_ISDIR(inode->i_mode))
 218                 goto dput_and_out;
 219
 220         error = -EINVAL;
 221         if (!S_ISREG(inode->i_mode))
 222                 goto dput_and_out;
 223
 224         error = permission(inode,MAY_WRITE,&nd);
 225         if (error)
 226                 goto dput_and_out;
 227
 228         error = -EROFS;
 229         if (IS_RDONLY(inode))
 230                 goto dput_and_out;
 231
 232         error = -EPERM;
 233         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 234                 goto dput_and_out;
 235
 236         /*
 237          * Make sure that there are no leases.
 238          */
 239         error = break_lease(inode, FMODE_WRITE);
 240         if (error)
 241                 goto dput_and_out;
 242
 243         error = get_write_access(inode);
 244         if (error)
 245                 goto dput_and_out;
 246
 247         error = locks_verify_truncate(inode, NULL, length);
 248         if (!error) {
 249                 DQUOT_INIT(inode);
 250                 error = do_truncate(nd.dentry, length);
 251         }
 252         put_write_access(inode);
 253
 254 dput_and_out:
 255         path_release(&nd);
 256 out:
 257         return error;
 258 }
 259
 260 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 261 {
 262         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 263         return do_sys_truncate(path, (long)length);
 264 }
 265
 266 static inline long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 267 {
 268         struct inode * inode;
 269         struct dentry *dentry;
 270         struct file * file;
 271         int error;
 272
 273         error = -EINVAL;
 274         if (length < 0)
 275                 goto out;
 276         error = -EBADF;
 277         file = fget(fd);
 278         if (!file)
 279                 goto out;
 280
 281         /* explicitly opened as large or we are on 64-bit box */
 282         if (file->f_flags & O_LARGEFILE)
 283                 small = 0;
 284
 285         dentry = file->f_dentry;
 286         inode = dentry->d_inode;
 287         error = -EINVAL;
 288         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 289                 goto out_putf;
 290
 291         error = -EINVAL;
 292         /* Cannot ftruncate over 2^31 bytes without large file support */
 293         if (small && length > MAX_NON_LFS)
 294                 goto out_putf;
 295
 296         error = -EPERM;
 297         if (IS_APPEND(inode))
 298                 goto out_putf;
 299
 300         error = locks_verify_truncate(inode, file, length);
 301         if (!error)
 302                 error = do_truncate(dentry, length);
 303 out_putf:
 304         fput(file);
 305 out:
 306         return error;
 307 }
 308
 309 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 310 {
 311         return do_sys_ftruncate(fd, length, 1);
 312 }
 313
 314 /* LFS versions of truncate are only needed on 32 bit machines */
 315 #if BITS_PER_LONG == 32
 316 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 317 {
 318         return do_sys_truncate(path, length);
 319 }
 320
 321 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 322 {
 323         return do_sys_ftruncate(fd, length, 0);
 324 }
 325 #endif
 326
 327 #if !(defined(__alpha__) || defined(__ia64__))
 328
 329 /*
 330  * sys_utime() can be implemented in user-level using sys_utimes().
 331  * Is this for backwards compatibility?  If so, why not move it
 332  * into the appropriate arch directory (for those architectures that
 333  * need it).
 334  */
 335
 336 /* If times==NULL, set access and modification to current time,
 337  * must be owner or have write permission.
 338  * Else, update from *times, must be owner or super user.
 339  */
 340 asmlinkage long sys_utime(char __user * filename, struct utimbuf __user * times)
 341 {
 342         int error;
 343         struct nameidata nd;
 344         struct inode * inode;
 345         struct iattr newattrs;
 346
 347         error = user_path_walk(filename, &nd);
 348         if (error)
 349                 goto out;
 350         inode = nd.dentry->d_inode;
 351
 352         error = -EROFS;
 353         if (IS_RDONLY(inode))
 354                 goto dput_and_out;
 355
 356         /* Don't worry, the checks are done in inode_change_ok() */
 357         newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
 358         if (times) {
 359                 error = -EPERM;
 360                 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 361                         goto dput_and_out;
 362
 363                 error = get_user(newattrs.ia_atime.tv_sec, &times->actime);
 364                 newattrs.ia_atime.tv_nsec = 0;
 365                 if (!error)
 366                         error = get_user(newattrs.ia_mtime.tv_sec, &times->modtime);
 367                 newattrs.ia_mtime.tv_nsec = 0;
 368                 if (error)
 369                         goto dput_and_out;
 370
 371                 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
 372         } else {
 373                 error = -EACCES;
 374                 if (IS_IMMUTABLE(inode))
 375                         goto dput_and_out;
 376
 377                 if (current->fsuid != inode->i_uid &&
 378                     (error = permission(inode,MAY_WRITE,&nd)) != 0)
 379                         goto dput_and_out;
 380         }
 381         down(&inode->i_sem);
 382         error = notify_change(nd.dentry, &newattrs);
 383         up(&inode->i_sem);
 384 dput_and_out:
 385         path_release(&nd);
 386 out:
 387         return error;
 388 }
 389
 390 #endif
 391
 392 /* If times==NULL, set access and modification to current time,
 393  * must be owner or have write permission.
 394  * Else, update from *times, must be owner or super user.
 395  */
 396 long do_utimes(char __user * filename, struct timeval * times)
 397 {
 398         int error;
 399         struct nameidata nd;
 400         struct inode * inode;
 401         struct iattr newattrs;
 402
 403         error = user_path_walk(filename, &nd);
 404
 405         if (error)
 406                 goto out;
 407         inode = nd.dentry->d_inode;
 408
 409         error = -EROFS;
 410         if (IS_RDONLY(inode))
 411                 goto dput_and_out;
 412
 413         /* Don't worry, the checks are done in inode_change_ok() */
 414         newattrs.ia_valid = ATTR_CTIME | ATTR_MTIME | ATTR_ATIME;
 415         if (times) {
 416                 error = -EPERM;
 417                 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 418                         goto dput_and_out;
 419
 420                 newattrs.ia_atime.tv_sec = times[0].tv_sec;
 421                 newattrs.ia_atime.tv_nsec = times[0].tv_usec * 1000;
 422                 newattrs.ia_mtime.tv_sec = times[1].tv_sec;
 423                 newattrs.ia_mtime.tv_nsec = times[1].tv_usec * 1000;
 424                 newattrs.ia_valid |= ATTR_ATIME_SET | ATTR_MTIME_SET;
 425         } else {
 426                 error = -EACCES;
 427                 if (IS_IMMUTABLE(inode))
 428                         goto dput_and_out;
 429
 430                 if (current->fsuid != inode->i_uid &&
 431                     (error = permission(inode,MAY_WRITE,&nd)) != 0)
 432                         goto dput_and_out;
 433         }
 434         down(&inode->i_sem);
 435         error = notify_change(nd.dentry, &newattrs);
 436         up(&inode->i_sem);
 437 dput_and_out:
 438         path_release(&nd);
 439 out:
 440         return error;
 441 }
 442
 443 asmlinkage long sys_utimes(char __user * filename, struct timeval __user * utimes)
 444 {
 445         struct timeval times[2];
 446
 447         if (utimes && copy_from_user(&times, utimes, sizeof(times)))
 448                 return -EFAULT;
 449         return do_utimes(filename, utimes ? times : NULL);
 450 }
 451
 452
 453 /*
 454  * access() needs to use the real uid/gid, not the effective uid/gid.
 455  * We do this by temporarily clearing all FS-related capabilities and
 456  * switching the fsuid/fsgid around to the real ones.
 457  */
 458 asmlinkage long sys_access(const char __user * filename, int mode)
 459 {
 460         struct nameidata nd;
 461         int old_fsuid, old_fsgid;
 462         kernel_cap_t old_cap;
 463         int res;
 464
 465         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 466                 return -EINVAL;
 467
 468         old_fsuid = current->fsuid;
 469         old_fsgid = current->fsgid;
 470         old_cap = current->cap_effective;
 471
 472         current->fsuid = current->uid;
 473         current->fsgid = current->gid;
 474
 475         /*
 476          * Clear the capabilities if we switch to a non-root user
 477          *
 478          * FIXME: There is a race here against sys_capset.  The
 479          * capabilities can change yet we will restore the old
 480          * value below.  We should hold task_capabilities_lock,
 481          * but we cannot because user_path_walk can sleep.
 482          */
 483         if (current->uid)
 484                 cap_clear(current->cap_effective);
 485         else
 486                 current->cap_effective = current->cap_permitted;
 487
 488         res = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 489         if (!res) {
 490                 res = permission(nd.dentry->d_inode, mode, &nd);
 491                 /* SuS v2 requires we report a read only fs too */
 492                 if(!res && (mode & S_IWOTH) && IS_RDONLY(nd.dentry->d_inode)
 493                    && !special_file(nd.dentry->d_inode->i_mode))
 494                         res = -EROFS;
 495                 path_release(&nd);
 496         }
 497
 498         current->fsuid = old_fsuid;
 499         current->fsgid = old_fsgid;
 500         current->cap_effective = old_cap;
 501
 502         return res;
 503 }
 504
 505 asmlinkage long sys_chdir(const char __user * filename)
 506 {
 507         struct nameidata nd;
 508         int error;
 509
 510         error = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &nd);
 511         if (error)
 512                 goto out;
 513
 514         error = permission(nd.dentry->d_inode,MAY_EXEC,&nd);
 515         if (error)
 516                 goto dput_and_out;
 517
 518         set_fs_pwd(current->fs, nd.mnt, nd.dentry);
 519
 520 dput_and_out:
 521         path_release(&nd);
 522 out:
 523         return error;
 524 }
 525
 526 asmlinkage long sys_fchdir(unsigned int fd)
 527 {
 528         struct file *file;
 529         struct dentry *dentry;
 530         struct inode *inode;
 531         struct vfsmount *mnt;
 532         int error;
 533
 534         error = -EBADF;
 535         file = fget(fd);
 536         if (!file)
 537                 goto out;
 538
 539         dentry = file->f_dentry;
 540         mnt = file->f_vfsmnt;
 541         inode = dentry->d_inode;
 542
 543         error = -ENOTDIR;
 544         if (!S_ISDIR(inode->i_mode))
 545                 goto out_putf;
 546
 547         error = permission(inode, MAY_EXEC, NULL);
 548         if (!error)
 549                 set_fs_pwd(current->fs, mnt, dentry);
 550 out_putf:
 551         fput(file);
 552 out:
 553         return error;
 554 }
 555
 556 asmlinkage long sys_chroot(const char __user * filename)
 557 {
 558         struct nameidata nd;
 559         int error;
 560
 561         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 562         if (error)
 563                 goto out;
 564
 565         error = permission(nd.dentry->d_inode,MAY_EXEC,&nd);
 566         if (error)
 567                 goto dput_and_out;
 568
 569         error = -EPERM;
 570         if (!capable(CAP_SYS_CHROOT))
 571                 goto dput_and_out;
 572
 573         set_fs_root(current->fs, nd.mnt, nd.dentry);
 574         set_fs_altroot();
 575         error = 0;
 576 dput_and_out:
 577         path_release(&nd);
 578 out:
 579         return error;
 580 }
 581
 582 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 583 {
 584         struct inode * inode;
 585         struct dentry * dentry;
 586         struct file * file;
 587         int err = -EBADF;
 588         struct iattr newattrs;
 589
 590         file = fget(fd);
 591         if (!file)
 592                 goto out;
 593
 594         dentry = file->f_dentry;
 595         inode = dentry->d_inode;
 596
 597         err = -EROFS;
 598         if (IS_RDONLY(inode))
 599                 goto out_putf;
 600         err = -EPERM;
 601         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 602                 goto out_putf;
 603         down(&inode->i_sem);
 604         if (mode == (mode_t) -1)
 605                 mode = inode->i_mode;
 606         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 607         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 608         err = notify_change(dentry, &newattrs);
 609         up(&inode->i_sem);
 610
 611 out_putf:
 612         fput(file);
 613 out:
 614         return err;
 615 }
 616
 617 asmlinkage long sys_chmod(const char __user * filename, mode_t mode)
 618 {
 619         struct nameidata nd;
 620         struct inode * inode;
 621         int error;
 622         struct iattr newattrs;
 623
 624         error = user_path_walk(filename, &nd);
 625         if (error)
 626                 goto out;
 627         inode = nd.dentry->d_inode;
 628
 629         error = -EROFS;
 630         if (IS_RDONLY(inode))
 631                 goto dput_and_out;
 632
 633         error = -EPERM;
 634         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 635                 goto dput_and_out;
 636
 637         down(&inode->i_sem);
 638         if (mode == (mode_t) -1)
 639                 mode = inode->i_mode;
 640         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 641         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 642         error = notify_change(nd.dentry, &newattrs);
 643         up(&inode->i_sem);
 644
 645 dput_and_out:
 646         path_release(&nd);
 647 out:
 648         return error;
 649 }
 650
 651 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 652 {
 653         struct inode * inode;
 654         int error;
 655         struct iattr newattrs;
 656
 657         error = -ENOENT;
 658         if (!(inode = dentry->d_inode)) {
 659                 printk(KERN_ERR "chown_common: NULL inode\n");
 660                 goto out;
 661         }
 662         error = -EROFS;
 663         if (IS_RDONLY(inode))
 664                 goto out;
 665         error = -EPERM;
 666         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 667                 goto out;
 668         newattrs.ia_valid =  ATTR_CTIME;
 669         if (user != (uid_t) -1) {
 670                 newattrs.ia_valid |= ATTR_UID;
 671                 newattrs.ia_uid = user;
 672         }
 673         if (group != (gid_t) -1) {
 674                 newattrs.ia_valid |= ATTR_GID;
 675                 newattrs.ia_gid = group;
 676         }
 677         if (!S_ISDIR(inode->i_mode))
 678                 newattrs.ia_valid |= ATTR_KILL_SUID|ATTR_KILL_SGID;
 679         down(&inode->i_sem);
 680         error = notify_change(dentry, &newattrs);
 681         up(&inode->i_sem);
 682 out:
 683         return error;
 684 }
 685
 686 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 687 {
 688         struct nameidata nd;
 689         int error;
 690
 691         error = user_path_walk(filename, &nd);
 692         if (!error) {
 693                 error = chown_common(nd.dentry, user, group);
 694                 path_release(&nd);
 695         }
 696         return error;
 697 }
 698
 699 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 700 {
 701         struct nameidata nd;
 702         int error;
 703
 704         error = user_path_walk_link(filename, &nd);
 705         if (!error) {
 706                 error = chown_common(nd.dentry, user, group);
 707                 path_release(&nd);
 708         }
 709         return error;
 710 }
 711
 712
 713 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 714 {
 715         struct file * file;
 716         int error = -EBADF;
 717
 718         file = fget(fd);
 719         if (file) {
 720                 error = chown_common(file->f_dentry, user, group);
 721                 fput(file);
 722         }
 723         return error;
 724 }
 725
 726 /*
 727  * Note that while the flag value (low two bits) for sys_open means:
 728  *      00 - read-only
 729  *      01 - write-only
 730  *      10 - read-write
 731  *      11 - special
 732  * it is changed into
 733  *      00 - no permissions needed
 734  *      01 - read-permission
 735  *      10 - write-permission
 736  *      11 - read-write
 737  * for the internal routines (ie open_namei()/follow_link() etc). 00 is
 738  * used by symlinks.
 739  */
 740 struct file *filp_open(const char * filename, int flags, int mode)
 741 {
 742         int namei_flags, error;
 743         struct nameidata nd;
 744
 745         namei_flags = flags;
 746         if ((namei_flags+1) & O_ACCMODE)
 747                 namei_flags++;
 748         if (namei_flags & O_TRUNC)
 749                 namei_flags |= 2;
 750
 751         error = open_namei(filename, namei_flags, mode, &nd);
 752         if (!error)
 753                 return dentry_open(nd.dentry, nd.mnt, flags);
 754
 755         return ERR_PTR(error);
 756 }
 757
 758 EXPORT_SYMBOL(filp_open);
 759
 760 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 761 {
 762         struct file * f;
 763         struct inode *inode;
 764         int error;
 765
 766         error = -ENFILE;
 767         f = get_empty_filp();
 768         if (!f)
 769                 goto cleanup_dentry;
 770         f->f_flags = flags;
 771         f->f_mode = (flags+1) & O_ACCMODE;
 772         inode = dentry->d_inode;
 773         if (f->f_mode & FMODE_WRITE) {
 774                 error = get_write_access(inode);
 775                 if (error)
 776                         goto cleanup_file;
 777         }
 778
 779         f->f_mapping = inode->i_mapping;
 780         file_ra_state_init(&f->f_ra, f->f_mapping);
 781         f->f_dentry = dentry;
 782         f->f_vfsmnt = mnt;
 783         f->f_pos = 0;
 784         f->f_op = fops_get(inode->i_fop);
 785         file_move(f, &inode->i_sb->s_files);
 786
 787         if (f->f_op && f->f_op->open) {
 788                 error = f->f_op->open(inode,f);
 789                 if (error)
 790                         goto cleanup_all;
 791         }
 792         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 793
 794         /* NB: we're sure to have correct a_ops only after f_op->open */
 795         if (f->f_flags & O_DIRECT) {
 796                 if (!f->f_mapping || !f->f_mapping->a_ops ||
 797                         !f->f_mapping->a_ops->direct_IO) {
 798                                 fput(f);
 799                                 f = ERR_PTR(-EINVAL);
 800                 }
 801         }
 802
 803         return f;
 804
 805 cleanup_all:
 806         fops_put(f->f_op);
 807         if (f->f_mode & FMODE_WRITE)
 808                 put_write_access(inode);
 809         file_kill(f);
 810         f->f_dentry = NULL;
 811         f->f_vfsmnt = NULL;
 812 cleanup_file:
 813         put_filp(f);
 814 cleanup_dentry:
 815         dput(dentry);
 816         mntput(mnt);
 817         return ERR_PTR(error);
 818 }
 819
 820 EXPORT_SYMBOL(dentry_open);
 821
 822 /*
 823  * Find an empty file descriptor entry, and mark it busy.
 824  */
 825 int get_unused_fd(void)
 826 {
 827         struct files_struct * files = current->files;
 828         int fd, error;
 829
 830         error = -EMFILE;
 831         spin_lock(&files->file_lock);
 832
 833 repeat:
 834         fd = find_next_zero_bit(files->open_fds->fds_bits,
 835                                 files->max_fdset,
 836                                 files->next_fd);
 837
 838         /*
 839          * N.B. For clone tasks sharing a files structure, this test
 840          * will limit the total number of files that can be opened.
 841          */
 842         if (fd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
 843                 goto out;
 844
 845         /* Do we need to expand the fdset array? */
 846         if (fd >= files->max_fdset) {
 847                 error = expand_fdset(files, fd);
 848                 if (!error) {
 849                         error = -EMFILE;
 850                         goto repeat;
 851                 }
 852                 goto out;
 853         }
 854
 855         /*
 856          * Check whether we need to expand the fd array.
 857          */
 858         if (fd >= files->max_fds) {
 859                 error = expand_fd_array(files, fd);
 860                 if (!error) {
 861                         error = -EMFILE;
 862                         goto repeat;
 863                 }
 864                 goto out;
 865         }
 866
 867         FD_SET(fd, files->open_fds);
 868         FD_CLR(fd, files->close_on_exec);
 869         files->next_fd = fd + 1;
 870 #if 1
 871         /* Sanity check */
 872         if (files->fd[fd] != NULL) {
 873                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
 874                 files->fd[fd] = NULL;
 875         }
 876 #endif
 877         error = fd;
 878
 879 out:
 880         spin_unlock(&files->file_lock);
 881         return error;
 882 }
 883
 884 EXPORT_SYMBOL(get_unused_fd);
 885
 886 static inline void __put_unused_fd(struct files_struct *files, unsigned int fd)
 887 {
 888         __FD_CLR(fd, files->open_fds);
 889         if (fd < files->next_fd)
 890                 files->next_fd = fd;
 891 }
 892
 893 void fastcall put_unused_fd(unsigned int fd)
 894 {
 895         struct files_struct *files = current->files;
 896         spin_lock(&files->file_lock);
 897         __put_unused_fd(files, fd);
 898         spin_unlock(&files->file_lock);
 899 }
 900
 901 EXPORT_SYMBOL(put_unused_fd);
 902
 903 /*
 904  * Install a file pointer in the fd array.
 905  *
 906  * The VFS is full of places where we drop the files lock between
 907  * setting the open_fds bitmap and installing the file in the file
 908  * array.  At any such point, we are vulnerable to a dup2() race
 909  * installing a file in the array before us.  We need to detect this and
 910  * fput() the struct file we are about to overwrite in this case.
 911  *
 912  * It should never happen - if we allow dup2() do it, _really_ bad things
 913  * will follow.
 914  */
 915
 916 void fastcall fd_install(unsigned int fd, struct file * file)
 917 {
 918         struct files_struct *files = current->files;
 919         spin_lock(&files->file_lock);
 920         if (unlikely(files->fd[fd] != NULL))
 921                 BUG();
 922         files->fd[fd] = file;
 923         spin_unlock(&files->file_lock);
 924 }
 925
 926 EXPORT_SYMBOL(fd_install);
 927
 928 asmlinkage long sys_open(const char __user * filename, int flags, int mode)
 929 {
 930         char * tmp;
 931         int fd, error;
 932
 933 #if BITS_PER_LONG != 32
 934         flags |= O_LARGEFILE;
 935 #endif
 936         tmp = getname(filename);
 937         fd = PTR_ERR(tmp);
 938         if (!IS_ERR(tmp)) {
 939                 fd = get_unused_fd();
 940                 if (fd >= 0) {
 941                         struct file *f = filp_open(tmp, flags, mode);
 942                         error = PTR_ERR(f);
 943                         if (IS_ERR(f))
 944                                 goto out_error;
 945                         fd_install(fd, f);
 946                 }
 947 out:
 948                 putname(tmp);
 949         }
 950         return fd;
 951
 952 out_error:
 953         put_unused_fd(fd);
 954         fd = error;
 955         goto out;
 956 }
 957 EXPORT_SYMBOL_GPL(sys_open);
 958
 959 #ifndef __alpha__
 960
 961 /*
 962  * For backward compatibility?  Maybe this should be moved
 963  * into arch/i386 instead?
 964  */
 965 asmlinkage long sys_creat(const char __user * pathname, int mode)
 966 {
 967         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
 968 }
 969
 970 #endif
 971
 972 /*
 973  * "id" is the POSIX thread ID. We use the
 974  * files pointer for this..
 975  */
 976 int filp_close(struct file *filp, fl_owner_t id)
 977 {
 978         int retval;
 979
 980         /* Report and clear outstanding errors */
 981         retval = filp->f_error;
 982         if (retval)
 983                 filp->f_error = 0;
 984
 985         if (!file_count(filp)) {
 986                 printk(KERN_ERR "VFS: Close: file count is 0\n");
 987                 return retval;
 988         }
 989
 990         if (filp->f_op && filp->f_op->flush) {
 991                 int err = filp->f_op->flush(filp);
 992                 if (!retval)
 993                         retval = err;
 994         }
 995
 996         dnotify_flush(filp, id);
 997         locks_remove_posix(filp, id);
 998         fput(filp);
 999         return retval;
1000 }
1001
1002 EXPORT_SYMBOL(filp_close);
1003
1004 /*
1005  * Careful here! We test whether the file pointer is NULL before
1006  * releasing the fd. This ensures that one clone task can't release
1007  * an fd while another clone is opening it.
1008  */
1009 asmlinkage long sys_close(unsigned int fd)
1010 {
1011         struct file * filp;
1012         struct files_struct *files = current->files;
1013
1014         spin_lock(&files->file_lock);
1015         if (fd >= files->max_fds)
1016                 goto out_unlock;
1017         filp = files->fd[fd];
1018         if (!filp)
1019                 goto out_unlock;
1020         files->fd[fd] = NULL;
1021         FD_CLR(fd, files->close_on_exec);
1022         __put_unused_fd(files, fd);
1023         spin_unlock(&files->file_lock);
1024         return filp_close(filp, files);
1025
1026 out_unlock:
1027         spin_unlock(&files->file_lock);
1028         return -EBADF;
1029 }
1030
1031 EXPORT_SYMBOL(sys_close);
1032
1033 /*
1034  * This routine simulates a hangup on the tty, to arrange that users
1035  * are given clean terminals at login time.
1036  */
1037 asmlinkage long sys_vhangup(void)
1038 {
1039         if (capable(CAP_SYS_TTY_CONFIG)) {
1040                 tty_vhangup(current->tty);
1041                 return 0;
1042         }
1043         return -EPERM;
1044 }
1045
1046 /*
1047  * Called when an inode is about to be open.
1048  * We use this to disallow opening large files on 32bit systems if
1049  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1050  * on this flag in sys_open.
1051  */
1052 int generic_file_open(struct inode * inode, struct file * filp)
1053 {
1054         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1055                 return -EFBIG;
1056         return 0;
1057 }
1058
1059 EXPORT_SYMBOL(generic_file_open);