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/quotaops.h>
  11 #include <linux/fsnotify.h>
  12 #include <linux/module.h>
  13 #include <linux/slab.h>
  14 #include <linux/tty.h>
  15 #include <linux/namei.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/capability.h>
  18 #include <linux/security.h>
  19 #include <linux/mount.h>
  20 #include <linux/vfs.h>
  21 #include <linux/fcntl.h>
  22 #include <asm/uaccess.h>
  23 #include <linux/fs.h>
  24 #include <linux/personality.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/rcupdate.h>
  28 #include <linux/audit.h>
  29 #include <linux/falloc.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_path.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_path.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         /* Remove suid/sgid on truncate too */
 215         newattrs.ia_valid |= should_remove_suid(dentry);
 216
 217         mutex_lock(&dentry->d_inode->i_mutex);
 218         err = notify_change(dentry, &newattrs);
 219         mutex_unlock(&dentry->d_inode->i_mutex);
 220         return err;
 221 }
 222
 223 static long do_sys_truncate(const char __user * path, loff_t length)
 224 {
 225         struct nameidata nd;
 226         struct inode * inode;
 227         int error;
 228
 229         error = -EINVAL;
 230         if (length < 0) /* sorry, but loff_t says... */
 231                 goto out;
 232
 233         error = user_path_walk(path, &nd);
 234         if (error)
 235                 goto out;
 236         inode = nd.dentry->d_inode;
 237
 238         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 239         error = -EISDIR;
 240         if (S_ISDIR(inode->i_mode))
 241                 goto dput_and_out;
 242
 243         error = -EINVAL;
 244         if (!S_ISREG(inode->i_mode))
 245                 goto dput_and_out;
 246
 247         error = vfs_permission(&nd, MAY_WRITE);
 248         if (error)
 249                 goto dput_and_out;
 250
 251         error = -EROFS;
 252         if (IS_RDONLY(inode))
 253                 goto dput_and_out;
 254
 255         error = -EPERM;
 256         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 257                 goto dput_and_out;
 258
 259         /*
 260          * Make sure that there are no leases.
 261          */
 262         error = break_lease(inode, FMODE_WRITE);
 263         if (error)
 264                 goto dput_and_out;
 265
 266         error = get_write_access(inode);
 267         if (error)
 268                 goto dput_and_out;
 269
 270         error = locks_verify_truncate(inode, NULL, length);
 271         if (!error) {
 272                 DQUOT_INIT(inode);
 273                 error = do_truncate(nd.dentry, length, 0, NULL);
 274         }
 275         put_write_access(inode);
 276
 277 dput_and_out:
 278         path_release(&nd);
 279 out:
 280         return error;
 281 }
 282
 283 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 284 {
 285         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 286         return do_sys_truncate(path, (long)length);
 287 }
 288
 289 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 290 {
 291         struct inode * inode;
 292         struct dentry *dentry;
 293         struct file * file;
 294         int error;
 295
 296         error = -EINVAL;
 297         if (length < 0)
 298                 goto out;
 299         error = -EBADF;
 300         file = fget(fd);
 301         if (!file)
 302                 goto out;
 303
 304         /* explicitly opened as large or we are on 64-bit box */
 305         if (file->f_flags & O_LARGEFILE)
 306                 small = 0;
 307
 308         dentry = file->f_path.dentry;
 309         inode = dentry->d_inode;
 310         error = -EINVAL;
 311         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 312                 goto out_putf;
 313
 314         error = -EINVAL;
 315         /* Cannot ftruncate over 2^31 bytes without large file support */
 316         if (small && length > MAX_NON_LFS)
 317                 goto out_putf;
 318
 319         error = -EPERM;
 320         if (IS_APPEND(inode))
 321                 goto out_putf;
 322
 323         error = locks_verify_truncate(inode, file, length);
 324         if (!error)
 325                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 326 out_putf:
 327         fput(file);
 328 out:
 329         return error;
 330 }
 331
 332 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 333 {
 334         long ret = do_sys_ftruncate(fd, length, 1);
 335         /* avoid REGPARM breakage on x86: */
 336         prevent_tail_call(ret);
 337         return ret;
 338 }
 339
 340 /* LFS versions of truncate are only needed on 32 bit machines */
 341 #if BITS_PER_LONG == 32
 342 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 343 {
 344         return do_sys_truncate(path, length);
 345 }
 346
 347 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 348 {
 349         long ret = do_sys_ftruncate(fd, length, 0);
 350         /* avoid REGPARM breakage on x86: */
 351         prevent_tail_call(ret);
 352         return ret;
 353 }
 354 #endif
 355
 356 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 357 {
 358         struct file *file;
 359         struct inode *inode;
 360         long ret = -EINVAL;
 361
 362         if (offset < 0 || len <= 0)
 363                 goto out;
 364
 365         /* Return error if mode is not supported */
 366         ret = -EOPNOTSUPP;
 367         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 368                 goto out;
 369
 370         ret = -EBADF;
 371         file = fget(fd);
 372         if (!file)
 373                 goto out;
 374         if (!(file->f_mode & FMODE_WRITE))
 375                 goto out_fput;
 376         /*
 377          * Revalidate the write permissions, in case security policy has
 378          * changed since the files were opened.
 379          */
 380         ret = security_file_permission(file, MAY_WRITE);
 381         if (ret)
 382                 goto out_fput;
 383
 384         inode = file->f_path.dentry->d_inode;
 385
 386         ret = -ESPIPE;
 387         if (S_ISFIFO(inode->i_mode))
 388                 goto out_fput;
 389
 390         ret = -ENODEV;
 391         /*
 392          * Let individual file system decide if it supports preallocation
 393          * for directories or not.
 394          */
 395         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 396                 goto out_fput;
 397
 398         ret = -EFBIG;
 399         /* Check for wrap through zero too */
 400         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 401                 goto out_fput;
 402
 403         if (inode->i_op && inode->i_op->fallocate)
 404                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 405         else
 406                 ret = -EOPNOTSUPP;
 407
 408 out_fput:
 409         fput(file);
 410 out:
 411         return ret;
 412 }
 413
 414 /*
 415  * access() needs to use the real uid/gid, not the effective uid/gid.
 416  * We do this by temporarily clearing all FS-related capabilities and
 417  * switching the fsuid/fsgid around to the real ones.
 418  */
 419 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 420 {
 421         struct nameidata nd;
 422         int old_fsuid, old_fsgid;
 423         kernel_cap_t old_cap;
 424         int res;
 425
 426         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 427                 return -EINVAL;
 428
 429         old_fsuid = current->fsuid;
 430         old_fsgid = current->fsgid;
 431         old_cap = current->cap_effective;
 432
 433         current->fsuid = current->uid;
 434         current->fsgid = current->gid;
 435
 436         /*
 437          * Clear the capabilities if we switch to a non-root user
 438          *
 439          * FIXME: There is a race here against sys_capset.  The
 440          * capabilities can change yet we will restore the old
 441          * value below.  We should hold task_capabilities_lock,
 442          * but we cannot because user_path_walk can sleep.
 443          */
 444         if (current->uid)
 445                 cap_clear(current->cap_effective);
 446         else
 447                 current->cap_effective = current->cap_permitted;
 448
 449         res = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 450         if (res)
 451                 goto out;
 452
 453         res = vfs_permission(&nd, mode);
 454         /* SuS v2 requires we report a read only fs too */
 455         if(res || !(mode & S_IWOTH) ||
 456            special_file(nd.dentry->d_inode->i_mode))
 457                 goto out_path_release;
 458
 459         if(IS_RDONLY(nd.dentry->d_inode))
 460                 res = -EROFS;
 461
 462 out_path_release:
 463         path_release(&nd);
 464 out:
 465         current->fsuid = old_fsuid;
 466         current->fsgid = old_fsgid;
 467         current->cap_effective = old_cap;
 468
 469         return res;
 470 }
 471
 472 asmlinkage long sys_access(const char __user *filename, int mode)
 473 {
 474         return sys_faccessat(AT_FDCWD, filename, mode);
 475 }
 476
 477 asmlinkage long sys_chdir(const char __user * filename)
 478 {
 479         struct nameidata nd;
 480         int error;
 481
 482         error = __user_walk(filename,
 483                             LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 484         if (error)
 485                 goto out;
 486
 487         error = vfs_permission(&nd, MAY_EXEC);
 488         if (error)
 489                 goto dput_and_out;
 490
 491         set_fs_pwd(current->fs, nd.mnt, nd.dentry);
 492
 493 dput_and_out:
 494         path_release(&nd);
 495 out:
 496         return error;
 497 }
 498
 499 asmlinkage long sys_fchdir(unsigned int fd)
 500 {
 501         struct file *file;
 502         struct dentry *dentry;
 503         struct inode *inode;
 504         struct vfsmount *mnt;
 505         int error;
 506
 507         error = -EBADF;
 508         file = fget(fd);
 509         if (!file)
 510                 goto out;
 511
 512         dentry = file->f_path.dentry;
 513         mnt = file->f_path.mnt;
 514         inode = dentry->d_inode;
 515
 516         error = -ENOTDIR;
 517         if (!S_ISDIR(inode->i_mode))
 518                 goto out_putf;
 519
 520         error = file_permission(file, MAY_EXEC);
 521         if (!error)
 522                 set_fs_pwd(current->fs, mnt, dentry);
 523 out_putf:
 524         fput(file);
 525 out:
 526         return error;
 527 }
 528
 529 asmlinkage long sys_chroot(const char __user * filename)
 530 {
 531         struct nameidata nd;
 532         int error;
 533
 534         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 535         if (error)
 536                 goto out;
 537
 538         error = vfs_permission(&nd, MAY_EXEC);
 539         if (error)
 540                 goto dput_and_out;
 541
 542         error = -EPERM;
 543         if (!capable(CAP_SYS_CHROOT))
 544                 goto dput_and_out;
 545
 546         set_fs_root(current->fs, nd.mnt, nd.dentry);
 547         set_fs_altroot();
 548         error = 0;
 549 dput_and_out:
 550         path_release(&nd);
 551 out:
 552         return error;
 553 }
 554
 555 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 556 {
 557         struct inode * inode;
 558         struct dentry * dentry;
 559         struct file * file;
 560         int err = -EBADF;
 561         struct iattr newattrs;
 562
 563         file = fget(fd);
 564         if (!file)
 565                 goto out;
 566
 567         dentry = file->f_path.dentry;
 568         inode = dentry->d_inode;
 569
 570         audit_inode(NULL, inode);
 571
 572         err = -EROFS;
 573         if (IS_RDONLY(inode))
 574                 goto out_putf;
 575         err = -EPERM;
 576         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 577                 goto out_putf;
 578         mutex_lock(&inode->i_mutex);
 579         if (mode == (mode_t) -1)
 580                 mode = inode->i_mode;
 581         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 582         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 583         err = notify_change(dentry, &newattrs);
 584         mutex_unlock(&inode->i_mutex);
 585
 586 out_putf:
 587         fput(file);
 588 out:
 589         return err;
 590 }
 591
 592 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 593                              mode_t mode)
 594 {
 595         struct nameidata nd;
 596         struct inode * inode;
 597         int error;
 598         struct iattr newattrs;
 599
 600         error = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
 601         if (error)
 602                 goto out;
 603         inode = nd.dentry->d_inode;
 604
 605         error = -EROFS;
 606         if (IS_RDONLY(inode))
 607                 goto dput_and_out;
 608
 609         error = -EPERM;
 610         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 611                 goto dput_and_out;
 612
 613         mutex_lock(&inode->i_mutex);
 614         if (mode == (mode_t) -1)
 615                 mode = inode->i_mode;
 616         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 617         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 618         error = notify_change(nd.dentry, &newattrs);
 619         mutex_unlock(&inode->i_mutex);
 620
 621 dput_and_out:
 622         path_release(&nd);
 623 out:
 624         return error;
 625 }
 626
 627 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 628 {
 629         return sys_fchmodat(AT_FDCWD, filename, mode);
 630 }
 631
 632 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 633 {
 634         struct inode * inode;
 635         int error;
 636         struct iattr newattrs;
 637
 638         error = -ENOENT;
 639         if (!(inode = dentry->d_inode)) {
 640                 printk(KERN_ERR "chown_common: NULL inode\n");
 641                 goto out;
 642         }
 643         error = -EROFS;
 644         if (IS_RDONLY(inode))
 645                 goto out;
 646         error = -EPERM;
 647         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 648                 goto out;
 649         newattrs.ia_valid =  ATTR_CTIME;
 650         if (user != (uid_t) -1) {
 651                 newattrs.ia_valid |= ATTR_UID;
 652                 newattrs.ia_uid = user;
 653         }
 654         if (group != (gid_t) -1) {
 655                 newattrs.ia_valid |= ATTR_GID;
 656                 newattrs.ia_gid = group;
 657         }
 658         if (!S_ISDIR(inode->i_mode))
 659                 newattrs.ia_valid |= ATTR_KILL_SUID|ATTR_KILL_SGID;
 660         mutex_lock(&inode->i_mutex);
 661         error = notify_change(dentry, &newattrs);
 662         mutex_unlock(&inode->i_mutex);
 663 out:
 664         return error;
 665 }
 666
 667 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 668 {
 669         struct nameidata nd;
 670         int error;
 671
 672         error = user_path_walk(filename, &nd);
 673         if (error)
 674                 goto out;
 675         error = chown_common(nd.dentry, user, group);
 676         path_release(&nd);
 677 out:
 678         return error;
 679 }
 680
 681 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 682                              gid_t group, int flag)
 683 {
 684         struct nameidata nd;
 685         int error = -EINVAL;
 686         int follow;
 687
 688         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 689                 goto out;
 690
 691         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 692         error = __user_walk_fd(dfd, filename, follow, &nd);
 693         if (error)
 694                 goto out;
 695         error = chown_common(nd.dentry, user, group);
 696         path_release(&nd);
 697 out:
 698         return error;
 699 }
 700
 701 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 702 {
 703         struct nameidata nd;
 704         int error;
 705
 706         error = user_path_walk_link(filename, &nd);
 707         if (error)
 708                 goto out;
 709         error = chown_common(nd.dentry, user, group);
 710         path_release(&nd);
 711 out:
 712         return error;
 713 }
 714
 715
 716 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 717 {
 718         struct file * file;
 719         int error = -EBADF;
 720         struct dentry * dentry;
 721
 722         file = fget(fd);
 723         if (!file)
 724                 goto out;
 725
 726         dentry = file->f_path.dentry;
 727         audit_inode(NULL, dentry->d_inode);
 728         error = chown_common(dentry, user, group);
 729         fput(file);
 730 out:
 731         return error;
 732 }
 733
 734 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 735                                         int flags, struct file *f,
 736                                         int (*open)(struct inode *, struct file *))
 737 {
 738         struct inode *inode;
 739         int error;
 740
 741         f->f_flags = flags;
 742         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 743                                 FMODE_PREAD | FMODE_PWRITE;
 744         inode = dentry->d_inode;
 745         if (f->f_mode & FMODE_WRITE) {
 746                 error = get_write_access(inode);
 747                 if (error)
 748                         goto cleanup_file;
 749         }
 750
 751         f->f_mapping = inode->i_mapping;
 752         f->f_path.dentry = dentry;
 753         f->f_path.mnt = mnt;
 754         f->f_pos = 0;
 755         f->f_op = fops_get(inode->i_fop);
 756         file_move(f, &inode->i_sb->s_files);
 757
 758         if (!open && f->f_op)
 759                 open = f->f_op->open;
 760         if (open) {
 761                 error = open(inode, f);
 762                 if (error)
 763                         goto cleanup_all;
 764         }
 765
 766         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 767
 768         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 769
 770         /* NB: we're sure to have correct a_ops only after f_op->open */
 771         if (f->f_flags & O_DIRECT) {
 772                 if (!f->f_mapping->a_ops ||
 773                     ((!f->f_mapping->a_ops->direct_IO) &&
 774                     (!f->f_mapping->a_ops->get_xip_page))) {
 775                         fput(f);
 776                         f = ERR_PTR(-EINVAL);
 777                 }
 778         }
 779
 780         return f;
 781
 782 cleanup_all:
 783         fops_put(f->f_op);
 784         if (f->f_mode & FMODE_WRITE)
 785                 put_write_access(inode);
 786         file_kill(f);
 787         f->f_path.dentry = NULL;
 788         f->f_path.mnt = NULL;
 789 cleanup_file:
 790         put_filp(f);
 791         dput(dentry);
 792         mntput(mnt);
 793         return ERR_PTR(error);
 794 }
 795
 796 /*
 797  * Note that while the flag value (low two bits) for sys_open means:
 798  *      00 - read-only
 799  *      01 - write-only
 800  *      10 - read-write
 801  *      11 - special
 802  * it is changed into
 803  *      00 - no permissions needed
 804  *      01 - read-permission
 805  *      10 - write-permission
 806  *      11 - read-write
 807  * for the internal routines (ie open_namei()/follow_link() etc). 00 is
 808  * used by symlinks.
 809  */
 810 static struct file *do_filp_open(int dfd, const char *filename, int flags,
 811                                  int mode)
 812 {
 813         int namei_flags, error;
 814         struct nameidata nd;
 815
 816         namei_flags = flags;
 817         if ((namei_flags+1) & O_ACCMODE)
 818                 namei_flags++;
 819
 820         error = open_namei(dfd, filename, namei_flags, mode, &nd);
 821         if (!error)
 822                 return nameidata_to_filp(&nd, flags);
 823
 824         return ERR_PTR(error);
 825 }
 826
 827 struct file *filp_open(const char *filename, int flags, int mode)
 828 {
 829         return do_filp_open(AT_FDCWD, filename, flags, mode);
 830 }
 831 EXPORT_SYMBOL(filp_open);
 832
 833 /**
 834  * lookup_instantiate_filp - instantiates the open intent filp
 835  * @nd: pointer to nameidata
 836  * @dentry: pointer to dentry
 837  * @open: open callback
 838  *
 839  * Helper for filesystems that want to use lookup open intents and pass back
 840  * a fully instantiated struct file to the caller.
 841  * This function is meant to be called from within a filesystem's
 842  * lookup method.
 843  * Beware of calling it for non-regular files! Those ->open methods might block
 844  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 845  * leading to a deadlock, as nobody can open that fifo anymore, because
 846  * another process to open fifo will block on locked parent when doing lookup).
 847  * Note that in case of error, nd->intent.open.file is destroyed, but the
 848  * path information remains valid.
 849  * If the open callback is set to NULL, then the standard f_op->open()
 850  * filesystem callback is substituted.
 851  */
 852 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 853                 int (*open)(struct inode *, struct file *))
 854 {
 855         if (IS_ERR(nd->intent.open.file))
 856                 goto out;
 857         if (IS_ERR(dentry))
 858                 goto out_err;
 859         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->mnt),
 860                                              nd->intent.open.flags - 1,
 861                                              nd->intent.open.file,
 862                                              open);
 863 out:
 864         return nd->intent.open.file;
 865 out_err:
 866         release_open_intent(nd);
 867         nd->intent.open.file = (struct file *)dentry;
 868         goto out;
 869 }
 870 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 871
 872 /**
 873  * nameidata_to_filp - convert a nameidata to an open filp.
 874  * @nd: pointer to nameidata
 875  * @flags: open flags
 876  *
 877  * Note that this function destroys the original nameidata
 878  */
 879 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 880 {
 881         struct file *filp;
 882
 883         /* Pick up the filp from the open intent */
 884         filp = nd->intent.open.file;
 885         /* Has the filesystem initialised the file for us? */
 886         if (filp->f_path.dentry == NULL)
 887                 filp = __dentry_open(nd->dentry, nd->mnt, flags, filp, NULL);
 888         else
 889                 path_release(nd);
 890         return filp;
 891 }
 892
 893 /*
 894  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 895  * error.
 896  */
 897 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 898 {
 899         int error;
 900         struct file *f;
 901
 902         error = -ENFILE;
 903         f = get_empty_filp();
 904         if (f == NULL) {
 905                 dput(dentry);
 906                 mntput(mnt);
 907                 return ERR_PTR(error);
 908         }
 909
 910         return __dentry_open(dentry, mnt, flags, f, NULL);
 911 }
 912 EXPORT_SYMBOL(dentry_open);
 913
 914 /*
 915  * Find an empty file descriptor entry, and mark it busy.
 916  */
 917 int get_unused_fd_flags(int flags)
 918 {
 919         struct files_struct * files = current->files;
 920         int fd, error;
 921         struct fdtable *fdt;
 922
 923         error = -EMFILE;
 924         spin_lock(&files->file_lock);
 925
 926 repeat:
 927         fdt = files_fdtable(files);
 928         fd = find_next_zero_bit(fdt->open_fds->fds_bits, fdt->max_fds,
 929                                 files->next_fd);
 930
 931         /*
 932          * N.B. For clone tasks sharing a files structure, this test
 933          * will limit the total number of files that can be opened.
 934          */
 935         if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 936                 goto out;
 937
 938         /* Do we need to expand the fd array or fd set?  */
 939         error = expand_files(files, fd);
 940         if (error < 0)
 941                 goto out;
 942
 943         if (error) {
 944                 /*
 945                  * If we needed to expand the fs array we
 946                  * might have blocked - try again.
 947                  */
 948                 error = -EMFILE;
 949                 goto repeat;
 950         }
 951
 952         FD_SET(fd, fdt->open_fds);
 953         if (flags & O_CLOEXEC)
 954                 FD_SET(fd, fdt->close_on_exec);
 955         else
 956                 FD_CLR(fd, fdt->close_on_exec);
 957         files->next_fd = fd + 1;
 958 #if 1
 959         /* Sanity check */
 960         if (fdt->fd[fd] != NULL) {
 961                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
 962                 fdt->fd[fd] = NULL;
 963         }
 964 #endif
 965         error = fd;
 966
 967 out:
 968         spin_unlock(&files->file_lock);
 969         return error;
 970 }
 971
 972 int get_unused_fd(void)
 973 {
 974         return get_unused_fd_flags(0);
 975 }
 976
 977 EXPORT_SYMBOL(get_unused_fd);
 978
 979 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 980 {
 981         struct fdtable *fdt = files_fdtable(files);
 982         __FD_CLR(fd, fdt->open_fds);
 983         if (fd < files->next_fd)
 984                 files->next_fd = fd;
 985 }
 986
 987 void fastcall put_unused_fd(unsigned int fd)
 988 {
 989         struct files_struct *files = current->files;
 990         spin_lock(&files->file_lock);
 991         __put_unused_fd(files, fd);
 992         spin_unlock(&files->file_lock);
 993 }
 994
 995 EXPORT_SYMBOL(put_unused_fd);
 996
 997 /*
 998  * Install a file pointer in the fd array.
 999  *
1000  * The VFS is full of places where we drop the files lock between
1001  * setting the open_fds bitmap and installing the file in the file
1002  * array.  At any such point, we are vulnerable to a dup2() race
1003  * installing a file in the array before us.  We need to detect this and
1004  * fput() the struct file we are about to overwrite in this case.
1005  *
1006  * It should never happen - if we allow dup2() do it, _really_ bad things
1007  * will follow.
1008  */
1009
1010 void fastcall fd_install(unsigned int fd, struct file * file)
1011 {
1012         struct files_struct *files = current->files;
1013         struct fdtable *fdt;
1014         spin_lock(&files->file_lock);
1015         fdt = files_fdtable(files);
1016         BUG_ON(fdt->fd[fd] != NULL);
1017         rcu_assign_pointer(fdt->fd[fd], file);
1018         spin_unlock(&files->file_lock);
1019 }
1020
1021 EXPORT_SYMBOL(fd_install);
1022
1023 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1024 {
1025         char *tmp = getname(filename);
1026         int fd = PTR_ERR(tmp);
1027
1028         if (!IS_ERR(tmp)) {
1029                 fd = get_unused_fd_flags(flags);
1030                 if (fd >= 0) {
1031                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1032                         if (IS_ERR(f)) {
1033                                 put_unused_fd(fd);
1034                                 fd = PTR_ERR(f);
1035                         } else {
1036                                 fsnotify_open(f->f_path.dentry);
1037                                 fd_install(fd, f);
1038                         }
1039                 }
1040                 putname(tmp);
1041         }
1042         return fd;
1043 }
1044
1045 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1046 {
1047         long ret;
1048
1049         if (force_o_largefile())
1050                 flags |= O_LARGEFILE;
1051
1052         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1053         /* avoid REGPARM breakage on x86: */
1054         prevent_tail_call(ret);
1055         return ret;
1056 }
1057 EXPORT_SYMBOL_GPL(sys_open);
1058
1059 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1060                            int mode)
1061 {
1062         long ret;
1063
1064         if (force_o_largefile())
1065                 flags |= O_LARGEFILE;
1066
1067         ret = do_sys_open(dfd, filename, flags, mode);
1068         /* avoid REGPARM breakage on x86: */
1069         prevent_tail_call(ret);
1070         return ret;
1071 }
1072
1073 #ifndef __alpha__
1074
1075 /*
1076  * For backward compatibility?  Maybe this should be moved
1077  * into arch/i386 instead?
1078  */
1079 asmlinkage long sys_creat(const char __user * pathname, int mode)
1080 {
1081         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1082 }
1083
1084 #endif
1085
1086 /*
1087  * "id" is the POSIX thread ID. We use the
1088  * files pointer for this..
1089  */
1090 int filp_close(struct file *filp, fl_owner_t id)
1091 {
1092         int retval = 0;
1093
1094         if (!file_count(filp)) {
1095                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1096                 return 0;
1097         }
1098
1099         if (filp->f_op && filp->f_op->flush)
1100                 retval = filp->f_op->flush(filp, id);
1101
1102         dnotify_flush(filp, id);
1103         locks_remove_posix(filp, id);
1104         fput(filp);
1105         return retval;
1106 }
1107
1108 EXPORT_SYMBOL(filp_close);
1109
1110 /*
1111  * Careful here! We test whether the file pointer is NULL before
1112  * releasing the fd. This ensures that one clone task can't release
1113  * an fd while another clone is opening it.
1114  */
1115 asmlinkage long sys_close(unsigned int fd)
1116 {
1117         struct file * filp;
1118         struct files_struct *files = current->files;
1119         struct fdtable *fdt;
1120         int retval;
1121
1122         spin_lock(&files->file_lock);
1123         fdt = files_fdtable(files);
1124         if (fd >= fdt->max_fds)
1125                 goto out_unlock;
1126         filp = fdt->fd[fd];
1127         if (!filp)
1128                 goto out_unlock;
1129         rcu_assign_pointer(fdt->fd[fd], NULL);
1130         FD_CLR(fd, fdt->close_on_exec);
1131         __put_unused_fd(files, fd);
1132         spin_unlock(&files->file_lock);
1133         retval = filp_close(filp, files);
1134
1135         /* can't restart close syscall because file table entry was cleared */
1136         if (unlikely(retval == -ERESTARTSYS ||
1137                      retval == -ERESTARTNOINTR ||
1138                      retval == -ERESTARTNOHAND ||
1139                      retval == -ERESTART_RESTARTBLOCK))
1140                 retval = -EINTR;
1141
1142         return retval;
1143
1144 out_unlock:
1145         spin_unlock(&files->file_lock);
1146         return -EBADF;
1147 }
1148
1149 EXPORT_SYMBOL(sys_close);
1150
1151 /*
1152  * This routine simulates a hangup on the tty, to arrange that users
1153  * are given clean terminals at login time.
1154  */
1155 asmlinkage long sys_vhangup(void)
1156 {
1157         if (capable(CAP_SYS_TTY_CONFIG)) {
1158                 /* XXX: this needs locking */
1159                 tty_vhangup(current->signal->tty);
1160                 return 0;
1161         }
1162         return -EPERM;
1163 }
1164
1165 /*
1166  * Called when an inode is about to be open.
1167  * We use this to disallow opening large files on 32bit systems if
1168  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1169  * on this flag in sys_open.
1170  */
1171 int generic_file_open(struct inode * inode, struct file * filp)
1172 {
1173         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1174                 return -EFBIG;
1175         return 0;
1176 }
1177
1178 EXPORT_SYMBOL(generic_file_open);
1179
1180 /*
1181  * This is used by subsystems that don't want seekable
1182  * file descriptors
1183  */
1184 int nonseekable_open(struct inode *inode, struct file *filp)
1185 {
1186         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1187         return 0;
1188 }
1189
1190 EXPORT_SYMBOL(nonseekable_open);