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.path.dentry, &tmp);
 131                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 132                         error = -EFAULT;
 133                 path_put(&nd.path);
 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.path.dentry, &tmp);
 150                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 151                         error = -EFAULT;
 152                 path_put(&nd.path);
 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.path.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         error = get_write_access(inode);
 260         if (error)
 261                 goto dput_and_out;
 262
 263         /*
 264          * Make sure that there are no leases.  get_write_access() protects
 265          * against the truncate racing with a lease-granting setlease().
 266          */
 267         error = break_lease(inode, FMODE_WRITE);
 268         if (error)
 269                 goto put_write_and_out;
 270
 271         error = locks_verify_truncate(inode, NULL, length);
 272         if (!error) {
 273                 DQUOT_INIT(inode);
 274                 error = do_truncate(nd.path.dentry, length, 0, NULL);
 275         }
 276
 277 put_write_and_out:
 278         put_write_access(inode);
 279 dput_and_out:
 280         path_put(&nd.path);
 281 out:
 282         return error;
 283 }
 284
 285 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 286 {
 287         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 288         return do_sys_truncate(path, (long)length);
 289 }
 290
 291 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 292 {
 293         struct inode * inode;
 294         struct dentry *dentry;
 295         struct file * file;
 296         int error;
 297
 298         error = -EINVAL;
 299         if (length < 0)
 300                 goto out;
 301         error = -EBADF;
 302         file = fget(fd);
 303         if (!file)
 304                 goto out;
 305
 306         /* explicitly opened as large or we are on 64-bit box */
 307         if (file->f_flags & O_LARGEFILE)
 308                 small = 0;
 309
 310         dentry = file->f_path.dentry;
 311         inode = dentry->d_inode;
 312         error = -EINVAL;
 313         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 314                 goto out_putf;
 315
 316         error = -EINVAL;
 317         /* Cannot ftruncate over 2^31 bytes without large file support */
 318         if (small && length > MAX_NON_LFS)
 319                 goto out_putf;
 320
 321         error = -EPERM;
 322         if (IS_APPEND(inode))
 323                 goto out_putf;
 324
 325         error = locks_verify_truncate(inode, file, length);
 326         if (!error)
 327                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 328 out_putf:
 329         fput(file);
 330 out:
 331         return error;
 332 }
 333
 334 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 335 {
 336         long ret = do_sys_ftruncate(fd, length, 1);
 337         /* avoid REGPARM breakage on x86: */
 338         asmlinkage_protect(2, ret, fd, length);
 339         return ret;
 340 }
 341
 342 /* LFS versions of truncate are only needed on 32 bit machines */
 343 #if BITS_PER_LONG == 32
 344 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 345 {
 346         return do_sys_truncate(path, length);
 347 }
 348
 349 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 350 {
 351         long ret = do_sys_ftruncate(fd, length, 0);
 352         /* avoid REGPARM breakage on x86: */
 353         asmlinkage_protect(2, ret, fd, length);
 354         return ret;
 355 }
 356 #endif
 357
 358 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 359 {
 360         struct file *file;
 361         struct inode *inode;
 362         long ret = -EINVAL;
 363
 364         if (offset < 0 || len <= 0)
 365                 goto out;
 366
 367         /* Return error if mode is not supported */
 368         ret = -EOPNOTSUPP;
 369         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 370                 goto out;
 371
 372         ret = -EBADF;
 373         file = fget(fd);
 374         if (!file)
 375                 goto out;
 376         if (!(file->f_mode & FMODE_WRITE))
 377                 goto out_fput;
 378         /*
 379          * Revalidate the write permissions, in case security policy has
 380          * changed since the files were opened.
 381          */
 382         ret = security_file_permission(file, MAY_WRITE);
 383         if (ret)
 384                 goto out_fput;
 385
 386         inode = file->f_path.dentry->d_inode;
 387
 388         ret = -ESPIPE;
 389         if (S_ISFIFO(inode->i_mode))
 390                 goto out_fput;
 391
 392         ret = -ENODEV;
 393         /*
 394          * Let individual file system decide if it supports preallocation
 395          * for directories or not.
 396          */
 397         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 398                 goto out_fput;
 399
 400         ret = -EFBIG;
 401         /* Check for wrap through zero too */
 402         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 403                 goto out_fput;
 404
 405         if (inode->i_op && inode->i_op->fallocate)
 406                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 407         else
 408                 ret = -EOPNOTSUPP;
 409
 410 out_fput:
 411         fput(file);
 412 out:
 413         return ret;
 414 }
 415
 416 /*
 417  * access() needs to use the real uid/gid, not the effective uid/gid.
 418  * We do this by temporarily clearing all FS-related capabilities and
 419  * switching the fsuid/fsgid around to the real ones.
 420  */
 421 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 422 {
 423         struct nameidata nd;
 424         int old_fsuid, old_fsgid;
 425         kernel_cap_t old_cap;
 426         int res;
 427
 428         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 429                 return -EINVAL;
 430
 431         old_fsuid = current->fsuid;
 432         old_fsgid = current->fsgid;
 433         old_cap = current->cap_effective;
 434
 435         current->fsuid = current->uid;
 436         current->fsgid = current->gid;
 437
 438         /*
 439          * Clear the capabilities if we switch to a non-root user
 440          *
 441          * FIXME: There is a race here against sys_capset.  The
 442          * capabilities can change yet we will restore the old
 443          * value below.  We should hold task_capabilities_lock,
 444          * but we cannot because user_path_walk can sleep.
 445          */
 446         if (current->uid)
 447                 cap_clear(current->cap_effective);
 448         else
 449                 current->cap_effective = current->cap_permitted;
 450
 451         res = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 452         if (res)
 453                 goto out;
 454
 455         res = vfs_permission(&nd, mode);
 456         /* SuS v2 requires we report a read only fs too */
 457         if(res || !(mode & S_IWOTH) ||
 458            special_file(nd.path.dentry->d_inode->i_mode))
 459                 goto out_path_release;
 460
 461         if(IS_RDONLY(nd.path.dentry->d_inode))
 462                 res = -EROFS;
 463
 464 out_path_release:
 465         path_put(&nd.path);
 466 out:
 467         current->fsuid = old_fsuid;
 468         current->fsgid = old_fsgid;
 469         current->cap_effective = old_cap;
 470
 471         return res;
 472 }
 473
 474 asmlinkage long sys_access(const char __user *filename, int mode)
 475 {
 476         return sys_faccessat(AT_FDCWD, filename, mode);
 477 }
 478
 479 asmlinkage long sys_chdir(const char __user * filename)
 480 {
 481         struct nameidata nd;
 482         int error;
 483
 484         error = __user_walk(filename,
 485                             LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 486         if (error)
 487                 goto out;
 488
 489         error = vfs_permission(&nd, MAY_EXEC);
 490         if (error)
 491                 goto dput_and_out;
 492
 493         set_fs_pwd(current->fs, &nd.path);
 494
 495 dput_and_out:
 496         path_put(&nd.path);
 497 out:
 498         return error;
 499 }
 500
 501 asmlinkage long sys_fchdir(unsigned int fd)
 502 {
 503         struct file *file;
 504         struct inode *inode;
 505         int error;
 506
 507         error = -EBADF;
 508         file = fget(fd);
 509         if (!file)
 510                 goto out;
 511
 512         inode = file->f_path.dentry->d_inode;
 513
 514         error = -ENOTDIR;
 515         if (!S_ISDIR(inode->i_mode))
 516                 goto out_putf;
 517
 518         error = file_permission(file, MAY_EXEC);
 519         if (!error)
 520                 set_fs_pwd(current->fs, &file->f_path);
 521 out_putf:
 522         fput(file);
 523 out:
 524         return error;
 525 }
 526
 527 asmlinkage long sys_chroot(const char __user * filename)
 528 {
 529         struct nameidata nd;
 530         int error;
 531
 532         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 533         if (error)
 534                 goto out;
 535
 536         error = vfs_permission(&nd, MAY_EXEC);
 537         if (error)
 538                 goto dput_and_out;
 539
 540         error = -EPERM;
 541         if (!capable(CAP_SYS_CHROOT))
 542                 goto dput_and_out;
 543
 544         set_fs_root(current->fs, &nd.path);
 545         set_fs_altroot();
 546         error = 0;
 547 dput_and_out:
 548         path_put(&nd.path);
 549 out:
 550         return error;
 551 }
 552
 553 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 554 {
 555         struct inode * inode;
 556         struct dentry * dentry;
 557         struct file * file;
 558         int err = -EBADF;
 559         struct iattr newattrs;
 560
 561         file = fget(fd);
 562         if (!file)
 563                 goto out;
 564
 565         dentry = file->f_path.dentry;
 566         inode = dentry->d_inode;
 567
 568         audit_inode(NULL, dentry);
 569
 570         err = -EROFS;
 571         if (IS_RDONLY(inode))
 572                 goto out_putf;
 573         err = -EPERM;
 574         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 575                 goto out_putf;
 576         mutex_lock(&inode->i_mutex);
 577         if (mode == (mode_t) -1)
 578                 mode = inode->i_mode;
 579         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 580         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 581         err = notify_change(dentry, &newattrs);
 582         mutex_unlock(&inode->i_mutex);
 583
 584 out_putf:
 585         fput(file);
 586 out:
 587         return err;
 588 }
 589
 590 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 591                              mode_t mode)
 592 {
 593         struct nameidata nd;
 594         struct inode * inode;
 595         int error;
 596         struct iattr newattrs;
 597
 598         error = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
 599         if (error)
 600                 goto out;
 601         inode = nd.path.dentry->d_inode;
 602
 603         error = -EROFS;
 604         if (IS_RDONLY(inode))
 605                 goto dput_and_out;
 606
 607         error = -EPERM;
 608         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 609                 goto dput_and_out;
 610
 611         mutex_lock(&inode->i_mutex);
 612         if (mode == (mode_t) -1)
 613                 mode = inode->i_mode;
 614         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 615         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 616         error = notify_change(nd.path.dentry, &newattrs);
 617         mutex_unlock(&inode->i_mutex);
 618
 619 dput_and_out:
 620         path_put(&nd.path);
 621 out:
 622         return error;
 623 }
 624
 625 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 626 {
 627         return sys_fchmodat(AT_FDCWD, filename, mode);
 628 }
 629
 630 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 631 {
 632         struct inode * inode;
 633         int error;
 634         struct iattr newattrs;
 635
 636         error = -ENOENT;
 637         if (!(inode = dentry->d_inode)) {
 638                 printk(KERN_ERR "chown_common: NULL inode\n");
 639                 goto out;
 640         }
 641         error = -EROFS;
 642         if (IS_RDONLY(inode))
 643                 goto out;
 644         error = -EPERM;
 645         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 646                 goto out;
 647         newattrs.ia_valid =  ATTR_CTIME;
 648         if (user != (uid_t) -1) {
 649                 newattrs.ia_valid |= ATTR_UID;
 650                 newattrs.ia_uid = user;
 651         }
 652         if (group != (gid_t) -1) {
 653                 newattrs.ia_valid |= ATTR_GID;
 654                 newattrs.ia_gid = group;
 655         }
 656         if (!S_ISDIR(inode->i_mode))
 657                 newattrs.ia_valid |=
 658                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 659         mutex_lock(&inode->i_mutex);
 660         error = notify_change(dentry, &newattrs);
 661         mutex_unlock(&inode->i_mutex);
 662 out:
 663         return error;
 664 }
 665
 666 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 667 {
 668         struct nameidata nd;
 669         int error;
 670
 671         error = user_path_walk(filename, &nd);
 672         if (error)
 673                 goto out;
 674         error = chown_common(nd.path.dentry, user, group);
 675         path_put(&nd.path);
 676 out:
 677         return error;
 678 }
 679
 680 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 681                              gid_t group, int flag)
 682 {
 683         struct nameidata nd;
 684         int error = -EINVAL;
 685         int follow;
 686
 687         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 688                 goto out;
 689
 690         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 691         error = __user_walk_fd(dfd, filename, follow, &nd);
 692         if (error)
 693                 goto out;
 694         error = chown_common(nd.path.dentry, user, group);
 695         path_put(&nd.path);
 696 out:
 697         return error;
 698 }
 699
 700 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 701 {
 702         struct nameidata nd;
 703         int error;
 704
 705         error = user_path_walk_link(filename, &nd);
 706         if (error)
 707                 goto out;
 708         error = chown_common(nd.path.dentry, user, group);
 709         path_put(&nd.path);
 710 out:
 711         return error;
 712 }
 713
 714
 715 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 716 {
 717         struct file * file;
 718         int error = -EBADF;
 719         struct dentry * dentry;
 720
 721         file = fget(fd);
 722         if (!file)
 723                 goto out;
 724
 725         dentry = file->f_path.dentry;
 726         audit_inode(NULL, dentry);
 727         error = chown_common(dentry, user, group);
 728         fput(file);
 729 out:
 730         return error;
 731 }
 732
 733 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 734                                         int flags, struct file *f,
 735                                         int (*open)(struct inode *, struct file *))
 736 {
 737         struct inode *inode;
 738         int error;
 739
 740         f->f_flags = flags;
 741         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 742                                 FMODE_PREAD | FMODE_PWRITE;
 743         inode = dentry->d_inode;
 744         if (f->f_mode & FMODE_WRITE) {
 745                 error = get_write_access(inode);
 746                 if (error)
 747                         goto cleanup_file;
 748         }
 749
 750         f->f_mapping = inode->i_mapping;
 751         f->f_path.dentry = dentry;
 752         f->f_path.mnt = mnt;
 753         f->f_pos = 0;
 754         f->f_op = fops_get(inode->i_fop);
 755         file_move(f, &inode->i_sb->s_files);
 756
 757         error = security_dentry_open(f);
 758         if (error)
 759                 goto cleanup_all;
 760
 761         if (!open && f->f_op)
 762                 open = f->f_op->open;
 763         if (open) {
 764                 error = open(inode, f);
 765                 if (error)
 766                         goto cleanup_all;
 767         }
 768
 769         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 770
 771         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 772
 773         /* NB: we're sure to have correct a_ops only after f_op->open */
 774         if (f->f_flags & O_DIRECT) {
 775                 if (!f->f_mapping->a_ops ||
 776                     ((!f->f_mapping->a_ops->direct_IO) &&
 777                     (!f->f_mapping->a_ops->get_xip_page))) {
 778                         fput(f);
 779                         f = ERR_PTR(-EINVAL);
 780                 }
 781         }
 782
 783         return f;
 784
 785 cleanup_all:
 786         fops_put(f->f_op);
 787         if (f->f_mode & FMODE_WRITE)
 788                 put_write_access(inode);
 789         file_kill(f);
 790         f->f_path.dentry = NULL;
 791         f->f_path.mnt = NULL;
 792 cleanup_file:
 793         put_filp(f);
 794         dput(dentry);
 795         mntput(mnt);
 796         return ERR_PTR(error);
 797 }
 798
 799 /*
 800  * Note that while the flag value (low two bits) for sys_open means:
 801  *      00 - read-only
 802  *      01 - write-only
 803  *      10 - read-write
 804  *      11 - special
 805  * it is changed into
 806  *      00 - no permissions needed
 807  *      01 - read-permission
 808  *      10 - write-permission
 809  *      11 - read-write
 810  * for the internal routines (ie open_namei()/follow_link() etc). 00 is
 811  * used by symlinks.
 812  */
 813 static struct file *do_filp_open(int dfd, const char *filename, int flags,
 814                                  int mode)
 815 {
 816         int namei_flags, error;
 817         struct nameidata nd;
 818
 819         namei_flags = flags;
 820         if ((namei_flags+1) & O_ACCMODE)
 821                 namei_flags++;
 822
 823         error = open_namei(dfd, filename, namei_flags, mode, &nd);
 824         if (!error)
 825                 return nameidata_to_filp(&nd, flags);
 826
 827         return ERR_PTR(error);
 828 }
 829
 830 struct file *filp_open(const char *filename, int flags, int mode)
 831 {
 832         return do_filp_open(AT_FDCWD, filename, flags, mode);
 833 }
 834 EXPORT_SYMBOL(filp_open);
 835
 836 /**
 837  * lookup_instantiate_filp - instantiates the open intent filp
 838  * @nd: pointer to nameidata
 839  * @dentry: pointer to dentry
 840  * @open: open callback
 841  *
 842  * Helper for filesystems that want to use lookup open intents and pass back
 843  * a fully instantiated struct file to the caller.
 844  * This function is meant to be called from within a filesystem's
 845  * lookup method.
 846  * Beware of calling it for non-regular files! Those ->open methods might block
 847  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 848  * leading to a deadlock, as nobody can open that fifo anymore, because
 849  * another process to open fifo will block on locked parent when doing lookup).
 850  * Note that in case of error, nd->intent.open.file is destroyed, but the
 851  * path information remains valid.
 852  * If the open callback is set to NULL, then the standard f_op->open()
 853  * filesystem callback is substituted.
 854  */
 855 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 856                 int (*open)(struct inode *, struct file *))
 857 {
 858         if (IS_ERR(nd->intent.open.file))
 859                 goto out;
 860         if (IS_ERR(dentry))
 861                 goto out_err;
 862         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 863                                              nd->intent.open.flags - 1,
 864                                              nd->intent.open.file,
 865                                              open);
 866 out:
 867         return nd->intent.open.file;
 868 out_err:
 869         release_open_intent(nd);
 870         nd->intent.open.file = (struct file *)dentry;
 871         goto out;
 872 }
 873 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 874
 875 /**
 876  * nameidata_to_filp - convert a nameidata to an open filp.
 877  * @nd: pointer to nameidata
 878  * @flags: open flags
 879  *
 880  * Note that this function destroys the original nameidata
 881  */
 882 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 883 {
 884         struct file *filp;
 885
 886         /* Pick up the filp from the open intent */
 887         filp = nd->intent.open.file;
 888         /* Has the filesystem initialised the file for us? */
 889         if (filp->f_path.dentry == NULL)
 890                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 891                                      NULL);
 892         else
 893                 path_put(&nd->path);
 894         return filp;
 895 }
 896
 897 /*
 898  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 899  * error.
 900  */
 901 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 902 {
 903         int error;
 904         struct file *f;
 905
 906         /*
 907          * We must always pass in a valid mount pointer.   Historically
 908          * callers got away with not passing it, but we must enforce this at
 909          * the earliest possible point now to avoid strange problems deep in the
 910          * filesystem stack.
 911          */
 912         if (!mnt) {
 913                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 914                 dump_stack();
 915                 return ERR_PTR(-EINVAL);
 916         }
 917
 918         error = -ENFILE;
 919         f = get_empty_filp();
 920         if (f == NULL) {
 921                 dput(dentry);
 922                 mntput(mnt);
 923                 return ERR_PTR(error);
 924         }
 925
 926         return __dentry_open(dentry, mnt, flags, f, NULL);
 927 }
 928 EXPORT_SYMBOL(dentry_open);
 929
 930 /*
 931  * Find an empty file descriptor entry, and mark it busy.
 932  */
 933 int get_unused_fd_flags(int flags)
 934 {
 935         struct files_struct * files = current->files;
 936         int fd, error;
 937         struct fdtable *fdt;
 938
 939         error = -EMFILE;
 940         spin_lock(&files->file_lock);
 941
 942 repeat:
 943         fdt = files_fdtable(files);
 944         fd = find_next_zero_bit(fdt->open_fds->fds_bits, fdt->max_fds,
 945                                 files->next_fd);
 946
 947         /*
 948          * N.B. For clone tasks sharing a files structure, this test
 949          * will limit the total number of files that can be opened.
 950          */
 951         if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 952                 goto out;
 953
 954         /* Do we need to expand the fd array or fd set?  */
 955         error = expand_files(files, fd);
 956         if (error < 0)
 957                 goto out;
 958
 959         if (error) {
 960                 /*
 961                  * If we needed to expand the fs array we
 962                  * might have blocked - try again.
 963                  */
 964                 error = -EMFILE;
 965                 goto repeat;
 966         }
 967
 968         FD_SET(fd, fdt->open_fds);
 969         if (flags & O_CLOEXEC)
 970                 FD_SET(fd, fdt->close_on_exec);
 971         else
 972                 FD_CLR(fd, fdt->close_on_exec);
 973         files->next_fd = fd + 1;
 974 #if 1
 975         /* Sanity check */
 976         if (fdt->fd[fd] != NULL) {
 977                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
 978                 fdt->fd[fd] = NULL;
 979         }
 980 #endif
 981         error = fd;
 982
 983 out:
 984         spin_unlock(&files->file_lock);
 985         return error;
 986 }
 987
 988 int get_unused_fd(void)
 989 {
 990         return get_unused_fd_flags(0);
 991 }
 992
 993 EXPORT_SYMBOL(get_unused_fd);
 994
 995 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 996 {
 997         struct fdtable *fdt = files_fdtable(files);
 998         __FD_CLR(fd, fdt->open_fds);
 999         if (fd < files->next_fd)
1000                 files->next_fd = fd;
1001 }
1002
1003 void put_unused_fd(unsigned int fd)
1004 {
1005         struct files_struct *files = current->files;
1006         spin_lock(&files->file_lock);
1007         __put_unused_fd(files, fd);
1008         spin_unlock(&files->file_lock);
1009 }
1010
1011 EXPORT_SYMBOL(put_unused_fd);
1012
1013 /*
1014  * Install a file pointer in the fd array.
1015  *
1016  * The VFS is full of places where we drop the files lock between
1017  * setting the open_fds bitmap and installing the file in the file
1018  * array.  At any such point, we are vulnerable to a dup2() race
1019  * installing a file in the array before us.  We need to detect this and
1020  * fput() the struct file we are about to overwrite in this case.
1021  *
1022  * It should never happen - if we allow dup2() do it, _really_ bad things
1023  * will follow.
1024  */
1025
1026 void fd_install(unsigned int fd, struct file *file)
1027 {
1028         struct files_struct *files = current->files;
1029         struct fdtable *fdt;
1030         spin_lock(&files->file_lock);
1031         fdt = files_fdtable(files);
1032         BUG_ON(fdt->fd[fd] != NULL);
1033         rcu_assign_pointer(fdt->fd[fd], file);
1034         spin_unlock(&files->file_lock);
1035 }
1036
1037 EXPORT_SYMBOL(fd_install);
1038
1039 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1040 {
1041         char *tmp = getname(filename);
1042         int fd = PTR_ERR(tmp);
1043
1044         if (!IS_ERR(tmp)) {
1045                 fd = get_unused_fd_flags(flags);
1046                 if (fd >= 0) {
1047                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1048                         if (IS_ERR(f)) {
1049                                 put_unused_fd(fd);
1050                                 fd = PTR_ERR(f);
1051                         } else {
1052                                 fsnotify_open(f->f_path.dentry);
1053                                 fd_install(fd, f);
1054                         }
1055                 }
1056                 putname(tmp);
1057         }
1058         return fd;
1059 }
1060
1061 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1062 {
1063         long ret;
1064
1065         if (force_o_largefile())
1066                 flags |= O_LARGEFILE;
1067
1068         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1069         /* avoid REGPARM breakage on x86: */
1070         asmlinkage_protect(3, ret, filename, flags, mode);
1071         return ret;
1072 }
1073
1074 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1075                            int mode)
1076 {
1077         long ret;
1078
1079         if (force_o_largefile())
1080                 flags |= O_LARGEFILE;
1081
1082         ret = do_sys_open(dfd, filename, flags, mode);
1083         /* avoid REGPARM breakage on x86: */
1084         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1085         return ret;
1086 }
1087
1088 #ifndef __alpha__
1089
1090 /*
1091  * For backward compatibility?  Maybe this should be moved
1092  * into arch/i386 instead?
1093  */
1094 asmlinkage long sys_creat(const char __user * pathname, int mode)
1095 {
1096         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1097 }
1098
1099 #endif
1100
1101 /*
1102  * "id" is the POSIX thread ID. We use the
1103  * files pointer for this..
1104  */
1105 int filp_close(struct file *filp, fl_owner_t id)
1106 {
1107         int retval = 0;
1108
1109         if (!file_count(filp)) {
1110                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1111                 return 0;
1112         }
1113
1114         if (filp->f_op && filp->f_op->flush)
1115                 retval = filp->f_op->flush(filp, id);
1116
1117         dnotify_flush(filp, id);
1118         locks_remove_posix(filp, id);
1119         fput(filp);
1120         return retval;
1121 }
1122
1123 EXPORT_SYMBOL(filp_close);
1124
1125 /*
1126  * Careful here! We test whether the file pointer is NULL before
1127  * releasing the fd. This ensures that one clone task can't release
1128  * an fd while another clone is opening it.
1129  */
1130 asmlinkage long sys_close(unsigned int fd)
1131 {
1132         struct file * filp;
1133         struct files_struct *files = current->files;
1134         struct fdtable *fdt;
1135         int retval;
1136
1137         spin_lock(&files->file_lock);
1138         fdt = files_fdtable(files);
1139         if (fd >= fdt->max_fds)
1140                 goto out_unlock;
1141         filp = fdt->fd[fd];
1142         if (!filp)
1143                 goto out_unlock;
1144         rcu_assign_pointer(fdt->fd[fd], NULL);
1145         FD_CLR(fd, fdt->close_on_exec);
1146         __put_unused_fd(files, fd);
1147         spin_unlock(&files->file_lock);
1148         retval = filp_close(filp, files);
1149
1150         /* can't restart close syscall because file table entry was cleared */
1151         if (unlikely(retval == -ERESTARTSYS ||
1152                      retval == -ERESTARTNOINTR ||
1153                      retval == -ERESTARTNOHAND ||
1154                      retval == -ERESTART_RESTARTBLOCK))
1155                 retval = -EINTR;
1156
1157         return retval;
1158
1159 out_unlock:
1160         spin_unlock(&files->file_lock);
1161         return -EBADF;
1162 }
1163
1164 EXPORT_SYMBOL(sys_close);
1165
1166 /*
1167  * This routine simulates a hangup on the tty, to arrange that users
1168  * are given clean terminals at login time.
1169  */
1170 asmlinkage long sys_vhangup(void)
1171 {
1172         if (capable(CAP_SYS_TTY_CONFIG)) {
1173                 /* XXX: this needs locking */
1174                 tty_vhangup(current->signal->tty);
1175                 return 0;
1176         }
1177         return -EPERM;
1178 }
1179
1180 /*
1181  * Called when an inode is about to be open.
1182  * We use this to disallow opening large files on 32bit systems if
1183  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1184  * on this flag in sys_open.
1185  */
1186 int generic_file_open(struct inode * inode, struct file * filp)
1187 {
1188         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1189                 return -EOVERFLOW;
1190         return 0;
1191 }
1192
1193 EXPORT_SYMBOL(generic_file_open);
1194
1195 /*
1196  * This is used by subsystems that don't want seekable
1197  * file descriptors
1198  */
1199 int nonseekable_open(struct inode *inode, struct file *filp)
1200 {
1201         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1202         return 0;
1203 }
1204
1205 EXPORT_SYMBOL(nonseekable_open);