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 = mnt_want_write(nd.path.mnt);
 248         if (error)
 249                 goto dput_and_out;
 250
 251         error = vfs_permission(&nd, MAY_WRITE);
 252         if (error)
 253                 goto mnt_drop_write_and_out;
 254
 255         error = -EPERM;
 256         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 257                 goto mnt_drop_write_and_out;
 258
 259         error = get_write_access(inode);
 260         if (error)
 261                 goto mnt_drop_write_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 mnt_drop_write_and_out:
 280         mnt_drop_write(nd.path.mnt);
 281 dput_and_out:
 282         path_put(&nd.path);
 283 out:
 284         return error;
 285 }
 286
 287 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 288 {
 289         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 290         return do_sys_truncate(path, (long)length);
 291 }
 292
 293 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 294 {
 295         struct inode * inode;
 296         struct dentry *dentry;
 297         struct file * file;
 298         int error;
 299
 300         error = -EINVAL;
 301         if (length < 0)
 302                 goto out;
 303         error = -EBADF;
 304         file = fget(fd);
 305         if (!file)
 306                 goto out;
 307
 308         /* explicitly opened as large or we are on 64-bit box */
 309         if (file->f_flags & O_LARGEFILE)
 310                 small = 0;
 311
 312         dentry = file->f_path.dentry;
 313         inode = dentry->d_inode;
 314         error = -EINVAL;
 315         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 316                 goto out_putf;
 317
 318         error = -EINVAL;
 319         /* Cannot ftruncate over 2^31 bytes without large file support */
 320         if (small && length > MAX_NON_LFS)
 321                 goto out_putf;
 322
 323         error = -EPERM;
 324         if (IS_APPEND(inode))
 325                 goto out_putf;
 326
 327         error = locks_verify_truncate(inode, file, length);
 328         if (!error)
 329                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 330 out_putf:
 331         fput(file);
 332 out:
 333         return error;
 334 }
 335
 336 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 337 {
 338         long ret = do_sys_ftruncate(fd, length, 1);
 339         /* avoid REGPARM breakage on x86: */
 340         asmlinkage_protect(2, ret, fd, length);
 341         return ret;
 342 }
 343
 344 /* LFS versions of truncate are only needed on 32 bit machines */
 345 #if BITS_PER_LONG == 32
 346 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 347 {
 348         return do_sys_truncate(path, length);
 349 }
 350
 351 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 352 {
 353         long ret = do_sys_ftruncate(fd, length, 0);
 354         /* avoid REGPARM breakage on x86: */
 355         asmlinkage_protect(2, ret, fd, length);
 356         return ret;
 357 }
 358 #endif
 359
 360 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 361 {
 362         struct file *file;
 363         struct inode *inode;
 364         long ret = -EINVAL;
 365
 366         if (offset < 0 || len <= 0)
 367                 goto out;
 368
 369         /* Return error if mode is not supported */
 370         ret = -EOPNOTSUPP;
 371         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 372                 goto out;
 373
 374         ret = -EBADF;
 375         file = fget(fd);
 376         if (!file)
 377                 goto out;
 378         if (!(file->f_mode & FMODE_WRITE))
 379                 goto out_fput;
 380         /*
 381          * Revalidate the write permissions, in case security policy has
 382          * changed since the files were opened.
 383          */
 384         ret = security_file_permission(file, MAY_WRITE);
 385         if (ret)
 386                 goto out_fput;
 387
 388         inode = file->f_path.dentry->d_inode;
 389
 390         ret = -ESPIPE;
 391         if (S_ISFIFO(inode->i_mode))
 392                 goto out_fput;
 393
 394         ret = -ENODEV;
 395         /*
 396          * Let individual file system decide if it supports preallocation
 397          * for directories or not.
 398          */
 399         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 400                 goto out_fput;
 401
 402         ret = -EFBIG;
 403         /* Check for wrap through zero too */
 404         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 405                 goto out_fput;
 406
 407         if (inode->i_op && inode->i_op->fallocate)
 408                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 409         else
 410                 ret = -EOPNOTSUPP;
 411
 412 out_fput:
 413         fput(file);
 414 out:
 415         return ret;
 416 }
 417
 418 /*
 419  * access() needs to use the real uid/gid, not the effective uid/gid.
 420  * We do this by temporarily clearing all FS-related capabilities and
 421  * switching the fsuid/fsgid around to the real ones.
 422  */
 423 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 424 {
 425         struct nameidata nd;
 426         int old_fsuid, old_fsgid;
 427         kernel_cap_t old_cap;
 428         int res;
 429
 430         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 431                 return -EINVAL;
 432
 433         old_fsuid = current->fsuid;
 434         old_fsgid = current->fsgid;
 435         old_cap = current->cap_effective;
 436
 437         current->fsuid = current->uid;
 438         current->fsgid = current->gid;
 439
 440         /*
 441          * Clear the capabilities if we switch to a non-root user
 442          *
 443          * FIXME: There is a race here against sys_capset.  The
 444          * capabilities can change yet we will restore the old
 445          * value below.  We should hold task_capabilities_lock,
 446          * but we cannot because user_path_walk can sleep.
 447          */
 448         if (current->uid)
 449                 cap_clear(current->cap_effective);
 450         else
 451                 current->cap_effective = current->cap_permitted;
 452
 453         res = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 454         if (res)
 455                 goto out;
 456
 457         res = vfs_permission(&nd, mode);
 458         /* SuS v2 requires we report a read only fs too */
 459         if(res || !(mode & S_IWOTH) ||
 460            special_file(nd.path.dentry->d_inode->i_mode))
 461                 goto out_path_release;
 462         /*
 463          * This is a rare case where using __mnt_is_readonly()
 464          * is OK without a mnt_want/drop_write() pair.  Since
 465          * no actual write to the fs is performed here, we do
 466          * not need to telegraph to that to anyone.
 467          *
 468          * By doing this, we accept that this access is
 469          * inherently racy and know that the fs may change
 470          * state before we even see this result.
 471          */
 472         if (__mnt_is_readonly(nd.path.mnt))
 473                 res = -EROFS;
 474
 475 out_path_release:
 476         path_put(&nd.path);
 477 out:
 478         current->fsuid = old_fsuid;
 479         current->fsgid = old_fsgid;
 480         current->cap_effective = old_cap;
 481
 482         return res;
 483 }
 484
 485 asmlinkage long sys_access(const char __user *filename, int mode)
 486 {
 487         return sys_faccessat(AT_FDCWD, filename, mode);
 488 }
 489
 490 asmlinkage long sys_chdir(const char __user * filename)
 491 {
 492         struct nameidata nd;
 493         int error;
 494
 495         error = __user_walk(filename,
 496                             LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 497         if (error)
 498                 goto out;
 499
 500         error = vfs_permission(&nd, MAY_EXEC);
 501         if (error)
 502                 goto dput_and_out;
 503
 504         set_fs_pwd(current->fs, &nd.path);
 505
 506 dput_and_out:
 507         path_put(&nd.path);
 508 out:
 509         return error;
 510 }
 511
 512 asmlinkage long sys_fchdir(unsigned int fd)
 513 {
 514         struct file *file;
 515         struct inode *inode;
 516         int error;
 517
 518         error = -EBADF;
 519         file = fget(fd);
 520         if (!file)
 521                 goto out;
 522
 523         inode = file->f_path.dentry->d_inode;
 524
 525         error = -ENOTDIR;
 526         if (!S_ISDIR(inode->i_mode))
 527                 goto out_putf;
 528
 529         error = file_permission(file, MAY_EXEC);
 530         if (!error)
 531                 set_fs_pwd(current->fs, &file->f_path);
 532 out_putf:
 533         fput(file);
 534 out:
 535         return error;
 536 }
 537
 538 asmlinkage long sys_chroot(const char __user * filename)
 539 {
 540         struct nameidata nd;
 541         int error;
 542
 543         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 544         if (error)
 545                 goto out;
 546
 547         error = vfs_permission(&nd, MAY_EXEC);
 548         if (error)
 549                 goto dput_and_out;
 550
 551         error = -EPERM;
 552         if (!capable(CAP_SYS_CHROOT))
 553                 goto dput_and_out;
 554
 555         set_fs_root(current->fs, &nd.path);
 556         set_fs_altroot();
 557         error = 0;
 558 dput_and_out:
 559         path_put(&nd.path);
 560 out:
 561         return error;
 562 }
 563
 564 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 565 {
 566         struct inode * inode;
 567         struct dentry * dentry;
 568         struct file * file;
 569         int err = -EBADF;
 570         struct iattr newattrs;
 571
 572         file = fget(fd);
 573         if (!file)
 574                 goto out;
 575
 576         dentry = file->f_path.dentry;
 577         inode = dentry->d_inode;
 578
 579         audit_inode(NULL, dentry);
 580
 581         err = mnt_want_write(file->f_path.mnt);
 582         if (err)
 583                 goto out_putf;
 584         err = -EPERM;
 585         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 586                 goto out_drop_write;
 587         mutex_lock(&inode->i_mutex);
 588         if (mode == (mode_t) -1)
 589                 mode = inode->i_mode;
 590         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 591         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 592         err = notify_change(dentry, &newattrs);
 593         mutex_unlock(&inode->i_mutex);
 594
 595 out_drop_write:
 596         mnt_drop_write(file->f_path.mnt);
 597 out_putf:
 598         fput(file);
 599 out:
 600         return err;
 601 }
 602
 603 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 604                              mode_t mode)
 605 {
 606         struct nameidata nd;
 607         struct inode * inode;
 608         int error;
 609         struct iattr newattrs;
 610
 611         error = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
 612         if (error)
 613                 goto out;
 614         inode = nd.path.dentry->d_inode;
 615
 616         error = mnt_want_write(nd.path.mnt);
 617         if (error)
 618                 goto dput_and_out;
 619
 620         error = -EPERM;
 621         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 622                 goto out_drop_write;
 623
 624         mutex_lock(&inode->i_mutex);
 625         if (mode == (mode_t) -1)
 626                 mode = inode->i_mode;
 627         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 628         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 629         error = notify_change(nd.path.dentry, &newattrs);
 630         mutex_unlock(&inode->i_mutex);
 631
 632 out_drop_write:
 633         mnt_drop_write(nd.path.mnt);
 634 dput_and_out:
 635         path_put(&nd.path);
 636 out:
 637         return error;
 638 }
 639
 640 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 641 {
 642         return sys_fchmodat(AT_FDCWD, filename, mode);
 643 }
 644
 645 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 646 {
 647         struct inode * inode;
 648         int error;
 649         struct iattr newattrs;
 650
 651         error = -ENOENT;
 652         if (!(inode = dentry->d_inode)) {
 653                 printk(KERN_ERR "chown_common: NULL inode\n");
 654                 goto out;
 655         }
 656         error = -EPERM;
 657         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 658                 goto out;
 659         newattrs.ia_valid =  ATTR_CTIME;
 660         if (user != (uid_t) -1) {
 661                 newattrs.ia_valid |= ATTR_UID;
 662                 newattrs.ia_uid = user;
 663         }
 664         if (group != (gid_t) -1) {
 665                 newattrs.ia_valid |= ATTR_GID;
 666                 newattrs.ia_gid = group;
 667         }
 668         if (!S_ISDIR(inode->i_mode))
 669                 newattrs.ia_valid |=
 670                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 671         mutex_lock(&inode->i_mutex);
 672         error = notify_change(dentry, &newattrs);
 673         mutex_unlock(&inode->i_mutex);
 674 out:
 675         return error;
 676 }
 677
 678 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 679 {
 680         struct nameidata nd;
 681         int error;
 682
 683         error = user_path_walk(filename, &nd);
 684         if (error)
 685                 goto out;
 686         error = mnt_want_write(nd.path.mnt);
 687         if (error)
 688                 goto out_release;
 689         error = chown_common(nd.path.dentry, user, group);
 690         mnt_drop_write(nd.path.mnt);
 691 out_release:
 692         path_put(&nd.path);
 693 out:
 694         return error;
 695 }
 696
 697 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 698                              gid_t group, int flag)
 699 {
 700         struct nameidata nd;
 701         int error = -EINVAL;
 702         int follow;
 703
 704         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 705                 goto out;
 706
 707         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 708         error = __user_walk_fd(dfd, filename, follow, &nd);
 709         if (error)
 710                 goto out;
 711         error = mnt_want_write(nd.path.mnt);
 712         if (error)
 713                 goto out_release;
 714         error = chown_common(nd.path.dentry, user, group);
 715         mnt_drop_write(nd.path.mnt);
 716 out_release:
 717         path_put(&nd.path);
 718 out:
 719         return error;
 720 }
 721
 722 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 723 {
 724         struct nameidata nd;
 725         int error;
 726
 727         error = user_path_walk_link(filename, &nd);
 728         if (error)
 729                 goto out;
 730         error = mnt_want_write(nd.path.mnt);
 731         if (error)
 732                 goto out_release;
 733         error = chown_common(nd.path.dentry, user, group);
 734         mnt_drop_write(nd.path.mnt);
 735 out_release:
 736         path_put(&nd.path);
 737 out:
 738         return error;
 739 }
 740
 741
 742 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 743 {
 744         struct file * file;
 745         int error = -EBADF;
 746         struct dentry * dentry;
 747
 748         file = fget(fd);
 749         if (!file)
 750                 goto out;
 751
 752         error = mnt_want_write(file->f_path.mnt);
 753         if (error)
 754                 goto out_fput;
 755         dentry = file->f_path.dentry;
 756         audit_inode(NULL, dentry);
 757         error = chown_common(dentry, user, group);
 758         mnt_drop_write(file->f_path.mnt);
 759 out_fput:
 760         fput(file);
 761 out:
 762         return error;
 763 }
 764
 765 /*
 766  * You have to be very careful that these write
 767  * counts get cleaned up in error cases and
 768  * upon __fput().  This should probably never
 769  * be called outside of __dentry_open().
 770  */
 771 static inline int __get_file_write_access(struct inode *inode,
 772                                           struct vfsmount *mnt)
 773 {
 774         int error;
 775         error = get_write_access(inode);
 776         if (error)
 777                 return error;
 778         /*
 779          * Do not take mount writer counts on
 780          * special files since no writes to
 781          * the mount itself will occur.
 782          */
 783         if (!special_file(inode->i_mode)) {
 784                 /*
 785                  * Balanced in __fput()
 786                  */
 787                 error = mnt_want_write(mnt);
 788                 if (error)
 789                         put_write_access(inode);
 790         }
 791         return error;
 792 }
 793
 794 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 795                                         int flags, struct file *f,
 796                                         int (*open)(struct inode *, struct file *))
 797 {
 798         struct inode *inode;
 799         int error;
 800
 801         f->f_flags = flags;
 802         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 803                                 FMODE_PREAD | FMODE_PWRITE;
 804         inode = dentry->d_inode;
 805         if (f->f_mode & FMODE_WRITE) {
 806                 error = __get_file_write_access(inode, mnt);
 807                 if (error)
 808                         goto cleanup_file;
 809                 if (!special_file(inode->i_mode))
 810                         file_take_write(f);
 811         }
 812
 813         f->f_mapping = inode->i_mapping;
 814         f->f_path.dentry = dentry;
 815         f->f_path.mnt = mnt;
 816         f->f_pos = 0;
 817         f->f_op = fops_get(inode->i_fop);
 818         file_move(f, &inode->i_sb->s_files);
 819
 820         error = security_dentry_open(f);
 821         if (error)
 822                 goto cleanup_all;
 823
 824         if (!open && f->f_op)
 825                 open = f->f_op->open;
 826         if (open) {
 827                 error = open(inode, f);
 828                 if (error)
 829                         goto cleanup_all;
 830         }
 831
 832         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 833
 834         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 835
 836         /* NB: we're sure to have correct a_ops only after f_op->open */
 837         if (f->f_flags & O_DIRECT) {
 838                 if (!f->f_mapping->a_ops ||
 839                     ((!f->f_mapping->a_ops->direct_IO) &&
 840                     (!f->f_mapping->a_ops->get_xip_mem))) {
 841                         fput(f);
 842                         f = ERR_PTR(-EINVAL);
 843                 }
 844         }
 845
 846         return f;
 847
 848 cleanup_all:
 849         fops_put(f->f_op);
 850         if (f->f_mode & FMODE_WRITE) {
 851                 put_write_access(inode);
 852                 if (!special_file(inode->i_mode)) {
 853                         /*
 854                          * We don't consider this a real
 855                          * mnt_want/drop_write() pair
 856                          * because it all happenend right
 857                          * here, so just reset the state.
 858                          */
 859                         file_reset_write(f);
 860                         mnt_drop_write(mnt);
 861                 }
 862         }
 863         file_kill(f);
 864         f->f_path.dentry = NULL;
 865         f->f_path.mnt = NULL;
 866 cleanup_file:
 867         put_filp(f);
 868         dput(dentry);
 869         mntput(mnt);
 870         return ERR_PTR(error);
 871 }
 872
 873 /**
 874  * lookup_instantiate_filp - instantiates the open intent filp
 875  * @nd: pointer to nameidata
 876  * @dentry: pointer to dentry
 877  * @open: open callback
 878  *
 879  * Helper for filesystems that want to use lookup open intents and pass back
 880  * a fully instantiated struct file to the caller.
 881  * This function is meant to be called from within a filesystem's
 882  * lookup method.
 883  * Beware of calling it for non-regular files! Those ->open methods might block
 884  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 885  * leading to a deadlock, as nobody can open that fifo anymore, because
 886  * another process to open fifo will block on locked parent when doing lookup).
 887  * Note that in case of error, nd->intent.open.file is destroyed, but the
 888  * path information remains valid.
 889  * If the open callback is set to NULL, then the standard f_op->open()
 890  * filesystem callback is substituted.
 891  */
 892 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 893                 int (*open)(struct inode *, struct file *))
 894 {
 895         if (IS_ERR(nd->intent.open.file))
 896                 goto out;
 897         if (IS_ERR(dentry))
 898                 goto out_err;
 899         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 900                                              nd->intent.open.flags - 1,
 901                                              nd->intent.open.file,
 902                                              open);
 903 out:
 904         return nd->intent.open.file;
 905 out_err:
 906         release_open_intent(nd);
 907         nd->intent.open.file = (struct file *)dentry;
 908         goto out;
 909 }
 910 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 911
 912 /**
 913  * nameidata_to_filp - convert a nameidata to an open filp.
 914  * @nd: pointer to nameidata
 915  * @flags: open flags
 916  *
 917  * Note that this function destroys the original nameidata
 918  */
 919 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 920 {
 921         struct file *filp;
 922
 923         /* Pick up the filp from the open intent */
 924         filp = nd->intent.open.file;
 925         /* Has the filesystem initialised the file for us? */
 926         if (filp->f_path.dentry == NULL)
 927                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 928                                      NULL);
 929         else
 930                 path_put(&nd->path);
 931         return filp;
 932 }
 933
 934 /*
 935  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 936  * error.
 937  */
 938 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 939 {
 940         int error;
 941         struct file *f;
 942
 943         /*
 944          * We must always pass in a valid mount pointer.   Historically
 945          * callers got away with not passing it, but we must enforce this at
 946          * the earliest possible point now to avoid strange problems deep in the
 947          * filesystem stack.
 948          */
 949         if (!mnt) {
 950                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 951                 dump_stack();
 952                 return ERR_PTR(-EINVAL);
 953         }
 954
 955         error = -ENFILE;
 956         f = get_empty_filp();
 957         if (f == NULL) {
 958                 dput(dentry);
 959                 mntput(mnt);
 960                 return ERR_PTR(error);
 961         }
 962
 963         return __dentry_open(dentry, mnt, flags, f, NULL);
 964 }
 965 EXPORT_SYMBOL(dentry_open);
 966
 967 /*
 968  * Find an empty file descriptor entry, and mark it busy.
 969  */
 970 int get_unused_fd_flags(int flags)
 971 {
 972         struct files_struct * files = current->files;
 973         int fd, error;
 974         struct fdtable *fdt;
 975
 976         error = -EMFILE;
 977         spin_lock(&files->file_lock);
 978
 979 repeat:
 980         fdt = files_fdtable(files);
 981         fd = find_next_zero_bit(fdt->open_fds->fds_bits, fdt->max_fds,
 982                                 files->next_fd);
 983
 984         /*
 985          * N.B. For clone tasks sharing a files structure, this test
 986          * will limit the total number of files that can be opened.
 987          */
 988         if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 989                 goto out;
 990
 991         /* Do we need to expand the fd array or fd set?  */
 992         error = expand_files(files, fd);
 993         if (error < 0)
 994                 goto out;
 995
 996         if (error) {
 997                 /*
 998                  * If we needed to expand the fs array we
 999                  * might have blocked - try again.
1000                  */
1001                 error = -EMFILE;
1002                 goto repeat;
1003         }
1004
1005         FD_SET(fd, fdt->open_fds);
1006         if (flags & O_CLOEXEC)
1007                 FD_SET(fd, fdt->close_on_exec);
1008         else
1009                 FD_CLR(fd, fdt->close_on_exec);
1010         files->next_fd = fd + 1;
1011 #if 1
1012         /* Sanity check */
1013         if (fdt->fd[fd] != NULL) {
1014                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
1015                 fdt->fd[fd] = NULL;
1016         }
1017 #endif
1018         error = fd;
1019
1020 out:
1021         spin_unlock(&files->file_lock);
1022         return error;
1023 }
1024
1025 int get_unused_fd(void)
1026 {
1027         return get_unused_fd_flags(0);
1028 }
1029
1030 EXPORT_SYMBOL(get_unused_fd);
1031
1032 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
1033 {
1034         struct fdtable *fdt = files_fdtable(files);
1035         __FD_CLR(fd, fdt->open_fds);
1036         if (fd < files->next_fd)
1037                 files->next_fd = fd;
1038 }
1039
1040 void put_unused_fd(unsigned int fd)
1041 {
1042         struct files_struct *files = current->files;
1043         spin_lock(&files->file_lock);
1044         __put_unused_fd(files, fd);
1045         spin_unlock(&files->file_lock);
1046 }
1047
1048 EXPORT_SYMBOL(put_unused_fd);
1049
1050 /*
1051  * Install a file pointer in the fd array.
1052  *
1053  * The VFS is full of places where we drop the files lock between
1054  * setting the open_fds bitmap and installing the file in the file
1055  * array.  At any such point, we are vulnerable to a dup2() race
1056  * installing a file in the array before us.  We need to detect this and
1057  * fput() the struct file we are about to overwrite in this case.
1058  *
1059  * It should never happen - if we allow dup2() do it, _really_ bad things
1060  * will follow.
1061  */
1062
1063 void fd_install(unsigned int fd, struct file *file)
1064 {
1065         struct files_struct *files = current->files;
1066         struct fdtable *fdt;
1067         spin_lock(&files->file_lock);
1068         fdt = files_fdtable(files);
1069         BUG_ON(fdt->fd[fd] != NULL);
1070         rcu_assign_pointer(fdt->fd[fd], file);
1071         spin_unlock(&files->file_lock);
1072 }
1073
1074 EXPORT_SYMBOL(fd_install);
1075
1076 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1077 {
1078         char *tmp = getname(filename);
1079         int fd = PTR_ERR(tmp);
1080
1081         if (!IS_ERR(tmp)) {
1082                 fd = get_unused_fd_flags(flags);
1083                 if (fd >= 0) {
1084                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1085                         if (IS_ERR(f)) {
1086                                 put_unused_fd(fd);
1087                                 fd = PTR_ERR(f);
1088                         } else {
1089                                 fsnotify_open(f->f_path.dentry);
1090                                 fd_install(fd, f);
1091                         }
1092                 }
1093                 putname(tmp);
1094         }
1095         return fd;
1096 }
1097
1098 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1099 {
1100         long ret;
1101
1102         if (force_o_largefile())
1103                 flags |= O_LARGEFILE;
1104
1105         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1106         /* avoid REGPARM breakage on x86: */
1107         asmlinkage_protect(3, ret, filename, flags, mode);
1108         return ret;
1109 }
1110
1111 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1112                            int mode)
1113 {
1114         long ret;
1115
1116         if (force_o_largefile())
1117                 flags |= O_LARGEFILE;
1118
1119         ret = do_sys_open(dfd, filename, flags, mode);
1120         /* avoid REGPARM breakage on x86: */
1121         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1122         return ret;
1123 }
1124
1125 #ifndef __alpha__
1126
1127 /*
1128  * For backward compatibility?  Maybe this should be moved
1129  * into arch/i386 instead?
1130  */
1131 asmlinkage long sys_creat(const char __user * pathname, int mode)
1132 {
1133         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1134 }
1135
1136 #endif
1137
1138 /*
1139  * "id" is the POSIX thread ID. We use the
1140  * files pointer for this..
1141  */
1142 int filp_close(struct file *filp, fl_owner_t id)
1143 {
1144         int retval = 0;
1145
1146         if (!file_count(filp)) {
1147                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1148                 return 0;
1149         }
1150
1151         if (filp->f_op && filp->f_op->flush)
1152                 retval = filp->f_op->flush(filp, id);
1153
1154         dnotify_flush(filp, id);
1155         locks_remove_posix(filp, id);
1156         fput(filp);
1157         return retval;
1158 }
1159
1160 EXPORT_SYMBOL(filp_close);
1161
1162 /*
1163  * Careful here! We test whether the file pointer is NULL before
1164  * releasing the fd. This ensures that one clone task can't release
1165  * an fd while another clone is opening it.
1166  */
1167 asmlinkage long sys_close(unsigned int fd)
1168 {
1169         struct file * filp;
1170         struct files_struct *files = current->files;
1171         struct fdtable *fdt;
1172         int retval;
1173
1174         spin_lock(&files->file_lock);
1175         fdt = files_fdtable(files);
1176         if (fd >= fdt->max_fds)
1177                 goto out_unlock;
1178         filp = fdt->fd[fd];
1179         if (!filp)
1180                 goto out_unlock;
1181         rcu_assign_pointer(fdt->fd[fd], NULL);
1182         FD_CLR(fd, fdt->close_on_exec);
1183         __put_unused_fd(files, fd);
1184         spin_unlock(&files->file_lock);
1185         retval = filp_close(filp, files);
1186
1187         /* can't restart close syscall because file table entry was cleared */
1188         if (unlikely(retval == -ERESTARTSYS ||
1189                      retval == -ERESTARTNOINTR ||
1190                      retval == -ERESTARTNOHAND ||
1191                      retval == -ERESTART_RESTARTBLOCK))
1192                 retval = -EINTR;
1193
1194         return retval;
1195
1196 out_unlock:
1197         spin_unlock(&files->file_lock);
1198         return -EBADF;
1199 }
1200
1201 EXPORT_SYMBOL(sys_close);
1202
1203 /*
1204  * This routine simulates a hangup on the tty, to arrange that users
1205  * are given clean terminals at login time.
1206  */
1207 asmlinkage long sys_vhangup(void)
1208 {
1209         if (capable(CAP_SYS_TTY_CONFIG)) {
1210                 /* XXX: this needs locking */
1211                 tty_vhangup(current->signal->tty);
1212                 return 0;
1213         }
1214         return -EPERM;
1215 }
1216
1217 /*
1218  * Called when an inode is about to be open.
1219  * We use this to disallow opening large files on 32bit systems if
1220  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1221  * on this flag in sys_open.
1222  */
1223 int generic_file_open(struct inode * inode, struct file * filp)
1224 {
1225         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1226                 return -EOVERFLOW;
1227         return 0;
1228 }
1229
1230 EXPORT_SYMBOL(generic_file_open);
1231
1232 /*
1233  * This is used by subsystems that don't want seekable
1234  * file descriptors
1235  */
1236 int nonseekable_open(struct inode *inode, struct file *filp)
1237 {
1238         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1239         return 0;
1240 }
1241
1242 EXPORT_SYMBOL(nonseekable_open);