fs/open.c

   1 /*
   2  *  linux/fs/open.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/string.h>
   8 #include <linux/mm.h>
   9 #include <linux/file.h>
  10 #include <linux/smp_lock.h>
  11 #include <linux/quotaops.h>
  12 #include <linux/fsnotify.h>
  13 #include <linux/module.h>
  14 #include <linux/slab.h>
  15 #include <linux/tty.h>
  16 #include <linux/namei.h>
  17 #include <linux/backing-dev.h>
  18 #include <linux/capability.h>
  19 #include <linux/security.h>
  20 #include <linux/mount.h>
  21 #include <linux/vfs.h>
  22 #include <linux/fcntl.h>
  23 #include <asm/uaccess.h>
  24 #include <linux/fs.h>
  25 #include <linux/personality.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/syscalls.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/audit.h>
  30
  31 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  32 {
  33         int retval = -ENODEV;
  34
  35         if (dentry) {
  36                 retval = -ENOSYS;
  37                 if (dentry->d_sb->s_op->statfs) {
  38                         memset(buf, 0, sizeof(*buf));
  39                         retval = security_sb_statfs(dentry);
  40                         if (retval)
  41                                 return retval;
  42                         retval = dentry->d_sb->s_op->statfs(dentry, 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 dentry *dentry, struct statfs *buf)
  53 {
  54         struct kstatfs st;
  55         int retval;
  56
  57         retval = vfs_statfs(dentry, &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 dentry *dentry, struct statfs64 *buf)
  96 {
  97         struct kstatfs st;
  98         int retval;
  99
 100         retval = vfs_statfs(dentry, &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, &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, &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, &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, &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         mutex_lock(&dentry->d_inode->i_mutex);
 215         err = notify_change(dentry, &newattrs);
 216         mutex_unlock(&dentry->d_inode->i_mutex);
 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, ATTR_MTIME|ATTR_CTIME, 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         long ret = do_sys_ftruncate(fd, length, 1);
 332         /* avoid REGPARM breakage on x86: */
 333         prevent_tail_call(ret);
 334         return ret;
 335 }
 336
 337 /* LFS versions of truncate are only needed on 32 bit machines */
 338 #if BITS_PER_LONG == 32
 339 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 340 {
 341         return do_sys_truncate(path, length);
 342 }
 343
 344 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 345 {
 346         long ret = do_sys_ftruncate(fd, length, 0);
 347         /* avoid REGPARM breakage on x86: */
 348         prevent_tail_call(ret);
 349         return ret;
 350 }
 351 #endif
 352
 353 /*
 354  * access() needs to use the real uid/gid, not the effective uid/gid.
 355  * We do this by temporarily clearing all FS-related capabilities and
 356  * switching the fsuid/fsgid around to the real ones.
 357  */
 358 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 359 {
 360         struct nameidata nd;
 361         int old_fsuid, old_fsgid;
 362         kernel_cap_t old_cap;
 363         int res;
 364
 365         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 366                 return -EINVAL;
 367
 368         old_fsuid = current->fsuid;
 369         old_fsgid = current->fsgid;
 370         old_cap = current->cap_effective;
 371
 372         current->fsuid = current->uid;
 373         current->fsgid = current->gid;
 374
 375         /*
 376          * Clear the capabilities if we switch to a non-root user
 377          *
 378          * FIXME: There is a race here against sys_capset.  The
 379          * capabilities can change yet we will restore the old
 380          * value below.  We should hold task_capabilities_lock,
 381          * but we cannot because user_path_walk can sleep.
 382          */
 383         if (current->uid)
 384                 cap_clear(current->cap_effective);
 385         else
 386                 current->cap_effective = current->cap_permitted;
 387
 388         res = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 389         if (res)
 390                 goto out;
 391
 392         res = vfs_permission(&nd, mode);
 393         /* SuS v2 requires we report a read only fs too */
 394         if(res || !(mode & S_IWOTH) ||
 395            special_file(nd.dentry->d_inode->i_mode))
 396                 goto out_path_release;
 397
 398         if(IS_RDONLY(nd.dentry->d_inode))
 399                 res = -EROFS;
 400
 401 out_path_release:
 402         path_release(&nd);
 403 out:
 404         current->fsuid = old_fsuid;
 405         current->fsgid = old_fsgid;
 406         current->cap_effective = old_cap;
 407
 408         return res;
 409 }
 410
 411 asmlinkage long sys_access(const char __user *filename, int mode)
 412 {
 413         return sys_faccessat(AT_FDCWD, filename, mode);
 414 }
 415
 416 asmlinkage long sys_chdir(const char __user * filename)
 417 {
 418         struct nameidata nd;
 419         int error;
 420
 421         error = __user_walk(filename,
 422                             LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 423         if (error)
 424                 goto out;
 425
 426         error = vfs_permission(&nd, MAY_EXEC);
 427         if (error)
 428                 goto dput_and_out;
 429
 430         set_fs_pwd(current->fs, nd.mnt, nd.dentry);
 431
 432 dput_and_out:
 433         path_release(&nd);
 434 out:
 435         return error;
 436 }
 437
 438 asmlinkage long sys_fchdir(unsigned int fd)
 439 {
 440         struct file *file;
 441         struct dentry *dentry;
 442         struct inode *inode;
 443         struct vfsmount *mnt;
 444         int error;
 445
 446         error = -EBADF;
 447         file = fget(fd);
 448         if (!file)
 449                 goto out;
 450
 451         dentry = file->f_dentry;
 452         mnt = file->f_vfsmnt;
 453         inode = dentry->d_inode;
 454
 455         error = -ENOTDIR;
 456         if (!S_ISDIR(inode->i_mode))
 457                 goto out_putf;
 458
 459         error = file_permission(file, MAY_EXEC);
 460         if (!error)
 461                 set_fs_pwd(current->fs, mnt, dentry);
 462 out_putf:
 463         fput(file);
 464 out:
 465         return error;
 466 }
 467
 468 asmlinkage long sys_chroot(const char __user * filename)
 469 {
 470         struct nameidata nd;
 471         int error;
 472
 473         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 474         if (error)
 475                 goto out;
 476
 477         error = vfs_permission(&nd, MAY_EXEC);
 478         if (error)
 479                 goto dput_and_out;
 480
 481         error = -EPERM;
 482         if (!capable(CAP_SYS_CHROOT))
 483                 goto dput_and_out;
 484
 485         set_fs_root(current->fs, nd.mnt, nd.dentry);
 486         set_fs_altroot();
 487         error = 0;
 488 dput_and_out:
 489         path_release(&nd);
 490 out:
 491         return error;
 492 }
 493
 494 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 495 {
 496         struct inode * inode;
 497         struct dentry * dentry;
 498         struct file * file;
 499         int err = -EBADF;
 500         struct iattr newattrs;
 501
 502         file = fget(fd);
 503         if (!file)
 504                 goto out;
 505
 506         dentry = file->f_dentry;
 507         inode = dentry->d_inode;
 508
 509         audit_inode(NULL, inode);
 510
 511         err = -EROFS;
 512         if (IS_RDONLY(inode))
 513                 goto out_putf;
 514         err = -EPERM;
 515         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 516                 goto out_putf;
 517         mutex_lock(&inode->i_mutex);
 518         if (mode == (mode_t) -1)
 519                 mode = inode->i_mode;
 520         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 521         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 522         err = notify_change(dentry, &newattrs);
 523         mutex_unlock(&inode->i_mutex);
 524
 525 out_putf:
 526         fput(file);
 527 out:
 528         return err;
 529 }
 530
 531 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 532                              mode_t mode)
 533 {
 534         struct nameidata nd;
 535         struct inode * inode;
 536         int error;
 537         struct iattr newattrs;
 538
 539         error = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
 540         if (error)
 541                 goto out;
 542         inode = nd.dentry->d_inode;
 543
 544         error = -EROFS;
 545         if (IS_RDONLY(inode))
 546                 goto dput_and_out;
 547
 548         error = -EPERM;
 549         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 550                 goto dput_and_out;
 551
 552         mutex_lock(&inode->i_mutex);
 553         if (mode == (mode_t) -1)
 554                 mode = inode->i_mode;
 555         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 556         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 557         error = notify_change(nd.dentry, &newattrs);
 558         mutex_unlock(&inode->i_mutex);
 559
 560 dput_and_out:
 561         path_release(&nd);
 562 out:
 563         return error;
 564 }
 565
 566 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 567 {
 568         return sys_fchmodat(AT_FDCWD, filename, mode);
 569 }
 570
 571 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 572 {
 573         struct inode * inode;
 574         int error;
 575         struct iattr newattrs;
 576
 577         error = -ENOENT;
 578         if (!(inode = dentry->d_inode)) {
 579                 printk(KERN_ERR "chown_common: NULL inode\n");
 580                 goto out;
 581         }
 582         error = -EROFS;
 583         if (IS_RDONLY(inode))
 584                 goto out;
 585         error = -EPERM;
 586         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 587                 goto out;
 588         newattrs.ia_valid =  ATTR_CTIME;
 589         if (user != (uid_t) -1) {
 590                 newattrs.ia_valid |= ATTR_UID;
 591                 newattrs.ia_uid = user;
 592         }
 593         if (group != (gid_t) -1) {
 594                 newattrs.ia_valid |= ATTR_GID;
 595                 newattrs.ia_gid = group;
 596         }
 597         if (!S_ISDIR(inode->i_mode))
 598                 newattrs.ia_valid |= ATTR_KILL_SUID|ATTR_KILL_SGID;
 599         mutex_lock(&inode->i_mutex);
 600         error = notify_change(dentry, &newattrs);
 601         mutex_unlock(&inode->i_mutex);
 602 out:
 603         return error;
 604 }
 605
 606 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 607 {
 608         struct nameidata nd;
 609         int error;
 610
 611         error = user_path_walk(filename, &nd);
 612         if (error)
 613                 goto out;
 614         error = chown_common(nd.dentry, user, group);
 615         path_release(&nd);
 616 out:
 617         return error;
 618 }
 619
 620 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 621                              gid_t group, int flag)
 622 {
 623         struct nameidata nd;
 624         int error = -EINVAL;
 625         int follow;
 626
 627         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 628                 goto out;
 629
 630         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 631         error = __user_walk_fd(dfd, filename, follow, &nd);
 632         if (error)
 633                 goto out;
 634         error = chown_common(nd.dentry, user, group);
 635         path_release(&nd);
 636 out:
 637         return error;
 638 }
 639
 640 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 641 {
 642         struct nameidata nd;
 643         int error;
 644
 645         error = user_path_walk_link(filename, &nd);
 646         if (error)
 647                 goto out;
 648         error = chown_common(nd.dentry, user, group);
 649         path_release(&nd);
 650 out:
 651         return error;
 652 }
 653
 654
 655 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 656 {
 657         struct file * file;
 658         int error = -EBADF;
 659         struct dentry * dentry;
 660
 661         file = fget(fd);
 662         if (!file)
 663                 goto out;
 664
 665         dentry = file->f_dentry;
 666         audit_inode(NULL, dentry->d_inode);
 667         error = chown_common(dentry, user, group);
 668         fput(file);
 669 out:
 670         return error;
 671 }
 672
 673 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 674                                         int flags, struct file *f,
 675                                         int (*open)(struct inode *, struct file *))
 676 {
 677         struct inode *inode;
 678         int error;
 679
 680         f->f_flags = flags;
 681         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 682                                 FMODE_PREAD | FMODE_PWRITE;
 683         inode = dentry->d_inode;
 684         if (f->f_mode & FMODE_WRITE) {
 685                 error = get_write_access(inode);
 686                 if (error)
 687                         goto cleanup_file;
 688         }
 689
 690         f->f_mapping = inode->i_mapping;
 691         f->f_dentry = dentry;
 692         f->f_vfsmnt = mnt;
 693         f->f_pos = 0;
 694         f->f_op = fops_get(inode->i_fop);
 695         file_move(f, &inode->i_sb->s_files);
 696
 697         if (!open && f->f_op)
 698                 open = f->f_op->open;
 699         if (open) {
 700                 error = open(inode, f);
 701                 if (error)
 702                         goto cleanup_all;
 703         }
 704
 705         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 706
 707         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 708
 709         /* NB: we're sure to have correct a_ops only after f_op->open */
 710         if (f->f_flags & O_DIRECT) {
 711                 if (!f->f_mapping->a_ops ||
 712                     ((!f->f_mapping->a_ops->direct_IO) &&
 713                     (!f->f_mapping->a_ops->get_xip_page))) {
 714                         fput(f);
 715                         f = ERR_PTR(-EINVAL);
 716                 }
 717         }
 718
 719         return f;
 720
 721 cleanup_all:
 722         fops_put(f->f_op);
 723         if (f->f_mode & FMODE_WRITE)
 724                 put_write_access(inode);
 725         file_kill(f);
 726         f->f_dentry = NULL;
 727         f->f_vfsmnt = NULL;
 728 cleanup_file:
 729         put_filp(f);
 730         dput(dentry);
 731         mntput(mnt);
 732         return ERR_PTR(error);
 733 }
 734
 735 /*
 736  * Note that while the flag value (low two bits) for sys_open means:
 737  *      00 - read-only
 738  *      01 - write-only
 739  *      10 - read-write
 740  *      11 - special
 741  * it is changed into
 742  *      00 - no permissions needed
 743  *      01 - read-permission
 744  *      10 - write-permission
 745  *      11 - read-write
 746  * for the internal routines (ie open_namei()/follow_link() etc). 00 is
 747  * used by symlinks.
 748  */
 749 static struct file *do_filp_open(int dfd, const char *filename, int flags,
 750                                  int mode)
 751 {
 752         int namei_flags, error;
 753         struct nameidata nd;
 754
 755         namei_flags = flags;
 756         if ((namei_flags+1) & O_ACCMODE)
 757                 namei_flags++;
 758
 759         error = open_namei(dfd, filename, namei_flags, mode, &nd);
 760         if (!error)
 761                 return nameidata_to_filp(&nd, flags);
 762
 763         return ERR_PTR(error);
 764 }
 765
 766 struct file *filp_open(const char *filename, int flags, int mode)
 767 {
 768         return do_filp_open(AT_FDCWD, filename, flags, mode);
 769 }
 770 EXPORT_SYMBOL(filp_open);
 771
 772 /**
 773  * lookup_instantiate_filp - instantiates the open intent filp
 774  * @nd: pointer to nameidata
 775  * @dentry: pointer to dentry
 776  * @open: open callback
 777  *
 778  * Helper for filesystems that want to use lookup open intents and pass back
 779  * a fully instantiated struct file to the caller.
 780  * This function is meant to be called from within a filesystem's
 781  * lookup method.
 782  * Beware of calling it for non-regular files! Those ->open methods might block
 783  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 784  * leading to a deadlock, as nobody can open that fifo anymore, because
 785  * another process to open fifo will block on locked parent when doing lookup).
 786  * Note that in case of error, nd->intent.open.file is destroyed, but the
 787  * path information remains valid.
 788  * If the open callback is set to NULL, then the standard f_op->open()
 789  * filesystem callback is substituted.
 790  */
 791 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 792                 int (*open)(struct inode *, struct file *))
 793 {
 794         if (IS_ERR(nd->intent.open.file))
 795                 goto out;
 796         if (IS_ERR(dentry))
 797                 goto out_err;
 798         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->mnt),
 799                                              nd->intent.open.flags - 1,
 800                                              nd->intent.open.file,
 801                                              open);
 802 out:
 803         return nd->intent.open.file;
 804 out_err:
 805         release_open_intent(nd);
 806         nd->intent.open.file = (struct file *)dentry;
 807         goto out;
 808 }
 809 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 810
 811 /**
 812  * nameidata_to_filp - convert a nameidata to an open filp.
 813  * @nd: pointer to nameidata
 814  * @flags: open flags
 815  *
 816  * Note that this function destroys the original nameidata
 817  */
 818 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 819 {
 820         struct file *filp;
 821
 822         /* Pick up the filp from the open intent */
 823         filp = nd->intent.open.file;
 824         /* Has the filesystem initialised the file for us? */
 825         if (filp->f_dentry == NULL)
 826                 filp = __dentry_open(nd->dentry, nd->mnt, flags, filp, NULL);
 827         else
 828                 path_release(nd);
 829         return filp;
 830 }
 831
 832 /*
 833  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 834  * error.
 835  */
 836 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 837 {
 838         int error;
 839         struct file *f;
 840
 841         error = -ENFILE;
 842         f = get_empty_filp();
 843         if (f == NULL) {
 844                 dput(dentry);
 845                 mntput(mnt);
 846                 return ERR_PTR(error);
 847         }
 848
 849         return __dentry_open(dentry, mnt, flags, f, NULL);
 850 }
 851 EXPORT_SYMBOL(dentry_open);
 852
 853 /*
 854  * Find an empty file descriptor entry, and mark it busy.
 855  */
 856 int get_unused_fd(void)
 857 {
 858         struct files_struct * files = current->files;
 859         int fd, error;
 860         struct fdtable *fdt;
 861
 862         error = -EMFILE;
 863         spin_lock(&files->file_lock);
 864
 865 repeat:
 866         fdt = files_fdtable(files);
 867         fd = find_next_zero_bit(fdt->open_fds->fds_bits,
 868                                 fdt->max_fdset,
 869                                 files->next_fd);
 870
 871         /*
 872          * N.B. For clone tasks sharing a files structure, this test
 873          * will limit the total number of files that can be opened.
 874          */
 875         if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 876                 goto out;
 877
 878         /* Do we need to expand the fd array or fd set?  */
 879         error = expand_files(files, fd);
 880         if (error < 0)
 881                 goto out;
 882
 883         if (error) {
 884                 /*
 885                  * If we needed to expand the fs array we
 886                  * might have blocked - try again.
 887                  */
 888                 error = -EMFILE;
 889                 goto repeat;
 890         }
 891
 892         FD_SET(fd, fdt->open_fds);
 893         FD_CLR(fd, fdt->close_on_exec);
 894         files->next_fd = fd + 1;
 895 #if 1
 896         /* Sanity check */
 897         if (fdt->fd[fd] != NULL) {
 898                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
 899                 fdt->fd[fd] = NULL;
 900         }
 901 #endif
 902         error = fd;
 903
 904 out:
 905         spin_unlock(&files->file_lock);
 906         return error;
 907 }
 908
 909 EXPORT_SYMBOL(get_unused_fd);
 910
 911 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 912 {
 913         struct fdtable *fdt = files_fdtable(files);
 914         __FD_CLR(fd, fdt->open_fds);
 915         if (fd < files->next_fd)
 916                 files->next_fd = fd;
 917 }
 918
 919 void fastcall put_unused_fd(unsigned int fd)
 920 {
 921         struct files_struct *files = current->files;
 922         spin_lock(&files->file_lock);
 923         __put_unused_fd(files, fd);
 924         spin_unlock(&files->file_lock);
 925 }
 926
 927 EXPORT_SYMBOL(put_unused_fd);
 928
 929 /*
 930  * Install a file pointer in the fd array.
 931  *
 932  * The VFS is full of places where we drop the files lock between
 933  * setting the open_fds bitmap and installing the file in the file
 934  * array.  At any such point, we are vulnerable to a dup2() race
 935  * installing a file in the array before us.  We need to detect this and
 936  * fput() the struct file we are about to overwrite in this case.
 937  *
 938  * It should never happen - if we allow dup2() do it, _really_ bad things
 939  * will follow.
 940  */
 941
 942 void fastcall fd_install(unsigned int fd, struct file * file)
 943 {
 944         struct files_struct *files = current->files;
 945         struct fdtable *fdt;
 946         spin_lock(&files->file_lock);
 947         fdt = files_fdtable(files);
 948         BUG_ON(fdt->fd[fd] != NULL);
 949         rcu_assign_pointer(fdt->fd[fd], file);
 950         spin_unlock(&files->file_lock);
 951 }
 952
 953 EXPORT_SYMBOL(fd_install);
 954
 955 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
 956 {
 957         char *tmp = getname(filename);
 958         int fd = PTR_ERR(tmp);
 959
 960         if (!IS_ERR(tmp)) {
 961                 fd = get_unused_fd();
 962                 if (fd >= 0) {
 963                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
 964                         if (IS_ERR(f)) {
 965                                 put_unused_fd(fd);
 966                                 fd = PTR_ERR(f);
 967                         } else {
 968                                 fsnotify_open(f->f_dentry);
 969                                 fd_install(fd, f);
 970                         }
 971                 }
 972                 putname(tmp);
 973         }
 974         return fd;
 975 }
 976
 977 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
 978 {
 979         long ret;
 980
 981         if (force_o_largefile())
 982                 flags |= O_LARGEFILE;
 983
 984         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
 985         /* avoid REGPARM breakage on x86: */
 986         prevent_tail_call(ret);
 987         return ret;
 988 }
 989 EXPORT_SYMBOL_GPL(sys_open);
 990
 991 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
 992                            int mode)
 993 {
 994         long ret;
 995
 996         if (force_o_largefile())
 997                 flags |= O_LARGEFILE;
 998
 999         ret = do_sys_open(dfd, filename, flags, mode);
1000         /* avoid REGPARM breakage on x86: */
1001         prevent_tail_call(ret);
1002         return ret;
1003 }
1004
1005 #ifndef __alpha__
1006
1007 /*
1008  * For backward compatibility?  Maybe this should be moved
1009  * into arch/i386 instead?
1010  */
1011 asmlinkage long sys_creat(const char __user * pathname, int mode)
1012 {
1013         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1014 }
1015
1016 #endif
1017
1018 /*
1019  * "id" is the POSIX thread ID. We use the
1020  * files pointer for this..
1021  */
1022 int filp_close(struct file *filp, fl_owner_t id)
1023 {
1024         int retval = 0;
1025
1026         if (!file_count(filp)) {
1027                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1028                 return 0;
1029         }
1030
1031         if (filp->f_op && filp->f_op->flush)
1032                 retval = filp->f_op->flush(filp, id);
1033
1034         dnotify_flush(filp, id);
1035         locks_remove_posix(filp, id);
1036         fput(filp);
1037         return retval;
1038 }
1039
1040 EXPORT_SYMBOL(filp_close);
1041
1042 /*
1043  * Careful here! We test whether the file pointer is NULL before
1044  * releasing the fd. This ensures that one clone task can't release
1045  * an fd while another clone is opening it.
1046  */
1047 asmlinkage long sys_close(unsigned int fd)
1048 {
1049         struct file * filp;
1050         struct files_struct *files = current->files;
1051         struct fdtable *fdt;
1052         int retval;
1053
1054         spin_lock(&files->file_lock);
1055         fdt = files_fdtable(files);
1056         if (fd >= fdt->max_fds)
1057                 goto out_unlock;
1058         filp = fdt->fd[fd];
1059         if (!filp)
1060                 goto out_unlock;
1061         rcu_assign_pointer(fdt->fd[fd], NULL);
1062         FD_CLR(fd, fdt->close_on_exec);
1063         __put_unused_fd(files, fd);
1064         spin_unlock(&files->file_lock);
1065         retval = filp_close(filp, files);
1066
1067         /* can't restart close syscall because file table entry was cleared */
1068         if (unlikely(retval == -ERESTARTSYS ||
1069                      retval == -ERESTARTNOINTR ||
1070                      retval == -ERESTARTNOHAND ||
1071                      retval == -ERESTART_RESTARTBLOCK))
1072                 retval = -EINTR;
1073
1074         return retval;
1075
1076 out_unlock:
1077         spin_unlock(&files->file_lock);
1078         return -EBADF;
1079 }
1080
1081 EXPORT_SYMBOL(sys_close);
1082
1083 /*
1084  * This routine simulates a hangup on the tty, to arrange that users
1085  * are given clean terminals at login time.
1086  */
1087 asmlinkage long sys_vhangup(void)
1088 {
1089         if (capable(CAP_SYS_TTY_CONFIG)) {
1090                 tty_vhangup(current->signal->tty);
1091                 return 0;
1092         }
1093         return -EPERM;
1094 }
1095
1096 /*
1097  * Called when an inode is about to be open.
1098  * We use this to disallow opening large files on 32bit systems if
1099  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1100  * on this flag in sys_open.
1101  */
1102 int generic_file_open(struct inode * inode, struct file * filp)
1103 {
1104         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1105                 return -EFBIG;
1106         return 0;
1107 }
1108
1109 EXPORT_SYMBOL(generic_file_open);
1110
1111 /*
1112  * This is used by subsystems that don't want seekable
1113  * file descriptors
1114  */
1115 int nonseekable_open(struct inode *inode, struct file *filp)
1116 {
1117         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1118         return 0;
1119 }
1120
1121 EXPORT_SYMBOL(nonseekable_open);