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/fdtable.h>
  11 #include <linux/quotaops.h>
  12 #include <linux/fsnotify.h>
  13 #include <linux/module.h>
  14 #include <linux/slab.h>
  15 #include <linux/tty.h>
  16 #include <linux/namei.h>
  17 #include <linux/backing-dev.h>
  18 #include <linux/capability.h>
  19 #include <linux/security.h>
  20 #include <linux/mount.h>
  21 #include <linux/vfs.h>
  22 #include <linux/fcntl.h>
  23 #include <asm/uaccess.h>
  24 #include <linux/fs.h>
  25 #include <linux/personality.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/syscalls.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/audit.h>
  30 #include <linux/falloc.h>
  31
  32 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  33 {
  34         int retval = -ENODEV;
  35
  36         if (dentry) {
  37                 retval = -ENOSYS;
  38                 if (dentry->d_sb->s_op->statfs) {
  39                         memset(buf, 0, sizeof(*buf));
  40                         retval = security_sb_statfs(dentry);
  41                         if (retval)
  42                                 return retval;
  43                         retval = dentry->d_sb->s_op->statfs(dentry, buf);
  44                         if (retval == 0 && buf->f_frsize == 0)
  45                                 buf->f_frsize = buf->f_bsize;
  46                 }
  47         }
  48         return retval;
  49 }
  50
  51 EXPORT_SYMBOL(vfs_statfs);
  52
  53 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
  54 {
  55         struct kstatfs st;
  56         int retval;
  57
  58         retval = vfs_statfs(dentry, &st);
  59         if (retval)
  60                 return retval;
  61
  62         if (sizeof(*buf) == sizeof(st))
  63                 memcpy(buf, &st, sizeof(st));
  64         else {
  65                 if (sizeof buf->f_blocks == 4) {
  66                         if ((st.f_blocks | st.f_bfree | st.f_bavail) &
  67                             0xffffffff00000000ULL)
  68                                 return -EOVERFLOW;
  69                         /*
  70                          * f_files and f_ffree may be -1; it's okay to stuff
  71                          * that into 32 bits
  72                          */
  73                         if (st.f_files != -1 &&
  74                             (st.f_files & 0xffffffff00000000ULL))
  75                                 return -EOVERFLOW;
  76                         if (st.f_ffree != -1 &&
  77                             (st.f_ffree & 0xffffffff00000000ULL))
  78                                 return -EOVERFLOW;
  79                 }
  80
  81                 buf->f_type = st.f_type;
  82                 buf->f_bsize = st.f_bsize;
  83                 buf->f_blocks = st.f_blocks;
  84                 buf->f_bfree = st.f_bfree;
  85                 buf->f_bavail = st.f_bavail;
  86                 buf->f_files = st.f_files;
  87                 buf->f_ffree = st.f_ffree;
  88                 buf->f_fsid = st.f_fsid;
  89                 buf->f_namelen = st.f_namelen;
  90                 buf->f_frsize = st.f_frsize;
  91                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  92         }
  93         return 0;
  94 }
  95
  96 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
  97 {
  98         struct kstatfs st;
  99         int retval;
 100
 101         retval = vfs_statfs(dentry, &st);
 102         if (retval)
 103                 return retval;
 104
 105         if (sizeof(*buf) == sizeof(st))
 106                 memcpy(buf, &st, sizeof(st));
 107         else {
 108                 buf->f_type = st.f_type;
 109                 buf->f_bsize = st.f_bsize;
 110                 buf->f_blocks = st.f_blocks;
 111                 buf->f_bfree = st.f_bfree;
 112                 buf->f_bavail = st.f_bavail;
 113                 buf->f_files = st.f_files;
 114                 buf->f_ffree = st.f_ffree;
 115                 buf->f_fsid = st.f_fsid;
 116                 buf->f_namelen = st.f_namelen;
 117                 buf->f_frsize = st.f_frsize;
 118                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 119         }
 120         return 0;
 121 }
 122
 123 asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
 124 {
 125         struct nameidata nd;
 126         int error;
 127
 128         error = user_path_walk(path, &nd);
 129         if (!error) {
 130                 struct statfs tmp;
 131                 error = vfs_statfs_native(nd.path.dentry, &tmp);
 132                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 133                         error = -EFAULT;
 134                 path_put(&nd.path);
 135         }
 136         return error;
 137 }
 138
 139
 140 asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
 141 {
 142         struct nameidata nd;
 143         long error;
 144
 145         if (sz != sizeof(*buf))
 146                 return -EINVAL;
 147         error = user_path_walk(path, &nd);
 148         if (!error) {
 149                 struct statfs64 tmp;
 150                 error = vfs_statfs64(nd.path.dentry, &tmp);
 151                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 152                         error = -EFAULT;
 153                 path_put(&nd.path);
 154         }
 155         return error;
 156 }
 157
 158
 159 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
 160 {
 161         struct file * file;
 162         struct statfs tmp;
 163         int error;
 164
 165         error = -EBADF;
 166         file = fget(fd);
 167         if (!file)
 168                 goto out;
 169         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 170         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 171                 error = -EFAULT;
 172         fput(file);
 173 out:
 174         return error;
 175 }
 176
 177 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
 178 {
 179         struct file * file;
 180         struct statfs64 tmp;
 181         int error;
 182
 183         if (sz != sizeof(*buf))
 184                 return -EINVAL;
 185
 186         error = -EBADF;
 187         file = fget(fd);
 188         if (!file)
 189                 goto out;
 190         error = vfs_statfs64(file->f_path.dentry, &tmp);
 191         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 192                 error = -EFAULT;
 193         fput(file);
 194 out:
 195         return error;
 196 }
 197
 198 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 199         struct file *filp)
 200 {
 201         int err;
 202         struct iattr newattrs;
 203
 204         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 205         if (length < 0)
 206                 return -EINVAL;
 207
 208         newattrs.ia_size = length;
 209         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 210         if (filp) {
 211                 newattrs.ia_file = filp;
 212                 newattrs.ia_valid |= ATTR_FILE;
 213         }
 214
 215         /* Remove suid/sgid on truncate too */
 216         newattrs.ia_valid |= should_remove_suid(dentry);
 217
 218         mutex_lock(&dentry->d_inode->i_mutex);
 219         err = notify_change(dentry, &newattrs);
 220         mutex_unlock(&dentry->d_inode->i_mutex);
 221         return err;
 222 }
 223
 224 static long do_sys_truncate(const char __user * path, loff_t length)
 225 {
 226         struct nameidata nd;
 227         struct inode * inode;
 228         int error;
 229
 230         error = -EINVAL;
 231         if (length < 0) /* sorry, but loff_t says... */
 232                 goto out;
 233
 234         error = user_path_walk(path, &nd);
 235         if (error)
 236                 goto out;
 237         inode = nd.path.dentry->d_inode;
 238
 239         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 240         error = -EISDIR;
 241         if (S_ISDIR(inode->i_mode))
 242                 goto dput_and_out;
 243
 244         error = -EINVAL;
 245         if (!S_ISREG(inode->i_mode))
 246                 goto dput_and_out;
 247
 248         error = mnt_want_write(nd.path.mnt);
 249         if (error)
 250                 goto dput_and_out;
 251
 252         error = vfs_permission(&nd, MAY_WRITE);
 253         if (error)
 254                 goto mnt_drop_write_and_out;
 255
 256         error = -EPERM;
 257         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 258                 goto mnt_drop_write_and_out;
 259
 260         error = get_write_access(inode);
 261         if (error)
 262                 goto mnt_drop_write_and_out;
 263
 264         /*
 265          * Make sure that there are no leases.  get_write_access() protects
 266          * against the truncate racing with a lease-granting setlease().
 267          */
 268         error = break_lease(inode, FMODE_WRITE);
 269         if (error)
 270                 goto put_write_and_out;
 271
 272         error = locks_verify_truncate(inode, NULL, length);
 273         if (!error) {
 274                 DQUOT_INIT(inode);
 275                 error = do_truncate(nd.path.dentry, length, 0, NULL);
 276         }
 277
 278 put_write_and_out:
 279         put_write_access(inode);
 280 mnt_drop_write_and_out:
 281         mnt_drop_write(nd.path.mnt);
 282 dput_and_out:
 283         path_put(&nd.path);
 284 out:
 285         return error;
 286 }
 287
 288 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 289 {
 290         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 291         return do_sys_truncate(path, (long)length);
 292 }
 293
 294 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 295 {
 296         struct inode * inode;
 297         struct dentry *dentry;
 298         struct file * file;
 299         int error;
 300
 301         error = -EINVAL;
 302         if (length < 0)
 303                 goto out;
 304         error = -EBADF;
 305         file = fget(fd);
 306         if (!file)
 307                 goto out;
 308
 309         /* explicitly opened as large or we are on 64-bit box */
 310         if (file->f_flags & O_LARGEFILE)
 311                 small = 0;
 312
 313         dentry = file->f_path.dentry;
 314         inode = dentry->d_inode;
 315         error = -EINVAL;
 316         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 317                 goto out_putf;
 318
 319         error = -EINVAL;
 320         /* Cannot ftruncate over 2^31 bytes without large file support */
 321         if (small && length > MAX_NON_LFS)
 322                 goto out_putf;
 323
 324         error = -EPERM;
 325         if (IS_APPEND(inode))
 326                 goto out_putf;
 327
 328         error = locks_verify_truncate(inode, file, length);
 329         if (!error)
 330                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 331 out_putf:
 332         fput(file);
 333 out:
 334         return error;
 335 }
 336
 337 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 338 {
 339         long ret = do_sys_ftruncate(fd, length, 1);
 340         /* avoid REGPARM breakage on x86: */
 341         asmlinkage_protect(2, ret, fd, length);
 342         return ret;
 343 }
 344
 345 /* LFS versions of truncate are only needed on 32 bit machines */
 346 #if BITS_PER_LONG == 32
 347 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 348 {
 349         return do_sys_truncate(path, length);
 350 }
 351
 352 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 353 {
 354         long ret = do_sys_ftruncate(fd, length, 0);
 355         /* avoid REGPARM breakage on x86: */
 356         asmlinkage_protect(2, ret, fd, length);
 357         return ret;
 358 }
 359 #endif
 360
 361 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 362 {
 363         struct file *file;
 364         struct inode *inode;
 365         long ret = -EINVAL;
 366
 367         if (offset < 0 || len <= 0)
 368                 goto out;
 369
 370         /* Return error if mode is not supported */
 371         ret = -EOPNOTSUPP;
 372         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 373                 goto out;
 374
 375         ret = -EBADF;
 376         file = fget(fd);
 377         if (!file)
 378                 goto out;
 379         if (!(file->f_mode & FMODE_WRITE))
 380                 goto out_fput;
 381         /*
 382          * Revalidate the write permissions, in case security policy has
 383          * changed since the files were opened.
 384          */
 385         ret = security_file_permission(file, MAY_WRITE);
 386         if (ret)
 387                 goto out_fput;
 388
 389         inode = file->f_path.dentry->d_inode;
 390
 391         ret = -ESPIPE;
 392         if (S_ISFIFO(inode->i_mode))
 393                 goto out_fput;
 394
 395         ret = -ENODEV;
 396         /*
 397          * Let individual file system decide if it supports preallocation
 398          * for directories or not.
 399          */
 400         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 401                 goto out_fput;
 402
 403         ret = -EFBIG;
 404         /* Check for wrap through zero too */
 405         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 406                 goto out_fput;
 407
 408         if (inode->i_op && inode->i_op->fallocate)
 409                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 410         else
 411                 ret = -EOPNOTSUPP;
 412
 413 out_fput:
 414         fput(file);
 415 out:
 416         return ret;
 417 }
 418
 419 /*
 420  * access() needs to use the real uid/gid, not the effective uid/gid.
 421  * We do this by temporarily clearing all FS-related capabilities and
 422  * switching the fsuid/fsgid around to the real ones.
 423  */
 424 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 425 {
 426         struct nameidata nd;
 427         int old_fsuid, old_fsgid;
 428         kernel_cap_t old_cap;
 429         int res;
 430
 431         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 432                 return -EINVAL;
 433
 434         old_fsuid = current->fsuid;
 435         old_fsgid = current->fsgid;
 436         old_cap = current->cap_effective;
 437
 438         current->fsuid = current->uid;
 439         current->fsgid = current->gid;
 440
 441         /*
 442          * Clear the capabilities if we switch to a non-root user
 443          *
 444          * FIXME: There is a race here against sys_capset.  The
 445          * capabilities can change yet we will restore the old
 446          * value below.  We should hold task_capabilities_lock,
 447          * but we cannot because user_path_walk can sleep.
 448          */
 449         if (current->uid)
 450                 cap_clear(current->cap_effective);
 451         else
 452                 current->cap_effective = current->cap_permitted;
 453
 454         res = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW|LOOKUP_ACCESS, &nd);
 455         if (res)
 456                 goto out;
 457
 458         res = vfs_permission(&nd, mode);
 459         /* SuS v2 requires we report a read only fs too */
 460         if(res || !(mode & S_IWOTH) ||
 461            special_file(nd.path.dentry->d_inode->i_mode))
 462                 goto out_path_release;
 463         /*
 464          * This is a rare case where using __mnt_is_readonly()
 465          * is OK without a mnt_want/drop_write() pair.  Since
 466          * no actual write to the fs is performed here, we do
 467          * not need to telegraph to that to anyone.
 468          *
 469          * By doing this, we accept that this access is
 470          * inherently racy and know that the fs may change
 471          * state before we even see this result.
 472          */
 473         if (__mnt_is_readonly(nd.path.mnt))
 474                 res = -EROFS;
 475
 476 out_path_release:
 477         path_put(&nd.path);
 478 out:
 479         current->fsuid = old_fsuid;
 480         current->fsgid = old_fsgid;
 481         current->cap_effective = old_cap;
 482
 483         return res;
 484 }
 485
 486 asmlinkage long sys_access(const char __user *filename, int mode)
 487 {
 488         return sys_faccessat(AT_FDCWD, filename, mode);
 489 }
 490
 491 asmlinkage long sys_chdir(const char __user * filename)
 492 {
 493         struct nameidata nd;
 494         int error;
 495
 496         error = __user_walk(filename,
 497                             LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
 498         if (error)
 499                 goto out;
 500
 501         error = vfs_permission(&nd, MAY_EXEC);
 502         if (error)
 503                 goto dput_and_out;
 504
 505         set_fs_pwd(current->fs, &nd.path);
 506
 507 dput_and_out:
 508         path_put(&nd.path);
 509 out:
 510         return error;
 511 }
 512
 513 asmlinkage long sys_fchdir(unsigned int fd)
 514 {
 515         struct file *file;
 516         struct inode *inode;
 517         int error;
 518
 519         error = -EBADF;
 520         file = fget(fd);
 521         if (!file)
 522                 goto out;
 523
 524         inode = file->f_path.dentry->d_inode;
 525
 526         error = -ENOTDIR;
 527         if (!S_ISDIR(inode->i_mode))
 528                 goto out_putf;
 529
 530         error = file_permission(file, MAY_EXEC);
 531         if (!error)
 532                 set_fs_pwd(current->fs, &file->f_path);
 533 out_putf:
 534         fput(file);
 535 out:
 536         return error;
 537 }
 538
 539 asmlinkage long sys_chroot(const char __user * filename)
 540 {
 541         struct nameidata nd;
 542         int error;
 543
 544         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY | LOOKUP_NOALT, &nd);
 545         if (error)
 546                 goto out;
 547
 548         error = vfs_permission(&nd, MAY_EXEC);
 549         if (error)
 550                 goto dput_and_out;
 551
 552         error = -EPERM;
 553         if (!capable(CAP_SYS_CHROOT))
 554                 goto dput_and_out;
 555
 556         set_fs_root(current->fs, &nd.path);
 557         set_fs_altroot();
 558         error = 0;
 559 dput_and_out:
 560         path_put(&nd.path);
 561 out:
 562         return error;
 563 }
 564
 565 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 566 {
 567         struct inode * inode;
 568         struct dentry * dentry;
 569         struct file * file;
 570         int err = -EBADF;
 571         struct iattr newattrs;
 572
 573         file = fget(fd);
 574         if (!file)
 575                 goto out;
 576
 577         dentry = file->f_path.dentry;
 578         inode = dentry->d_inode;
 579
 580         audit_inode(NULL, dentry);
 581
 582         err = mnt_want_write(file->f_path.mnt);
 583         if (err)
 584                 goto out_putf;
 585         err = -EPERM;
 586         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 587                 goto out_drop_write;
 588         mutex_lock(&inode->i_mutex);
 589         if (mode == (mode_t) -1)
 590                 mode = inode->i_mode;
 591         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 592         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 593         err = notify_change(dentry, &newattrs);
 594         mutex_unlock(&inode->i_mutex);
 595
 596 out_drop_write:
 597         mnt_drop_write(file->f_path.mnt);
 598 out_putf:
 599         fput(file);
 600 out:
 601         return err;
 602 }
 603
 604 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 605                              mode_t mode)
 606 {
 607         struct nameidata nd;
 608         struct inode * inode;
 609         int error;
 610         struct iattr newattrs;
 611
 612         error = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
 613         if (error)
 614                 goto out;
 615         inode = nd.path.dentry->d_inode;
 616
 617         error = mnt_want_write(nd.path.mnt);
 618         if (error)
 619                 goto dput_and_out;
 620
 621         error = -EPERM;
 622         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 623                 goto out_drop_write;
 624
 625         mutex_lock(&inode->i_mutex);
 626         if (mode == (mode_t) -1)
 627                 mode = inode->i_mode;
 628         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 629         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 630         error = notify_change(nd.path.dentry, &newattrs);
 631         mutex_unlock(&inode->i_mutex);
 632
 633 out_drop_write:
 634         mnt_drop_write(nd.path.mnt);
 635 dput_and_out:
 636         path_put(&nd.path);
 637 out:
 638         return error;
 639 }
 640
 641 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 642 {
 643         return sys_fchmodat(AT_FDCWD, filename, mode);
 644 }
 645
 646 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 647 {
 648         struct inode * inode;
 649         int error;
 650         struct iattr newattrs;
 651
 652         error = -ENOENT;
 653         if (!(inode = dentry->d_inode)) {
 654                 printk(KERN_ERR "chown_common: NULL inode\n");
 655                 goto out;
 656         }
 657         error = -EPERM;
 658         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 659                 goto out;
 660         newattrs.ia_valid =  ATTR_CTIME;
 661         if (user != (uid_t) -1) {
 662                 newattrs.ia_valid |= ATTR_UID;
 663                 newattrs.ia_uid = user;
 664         }
 665         if (group != (gid_t) -1) {
 666                 newattrs.ia_valid |= ATTR_GID;
 667                 newattrs.ia_gid = group;
 668         }
 669         if (!S_ISDIR(inode->i_mode))
 670                 newattrs.ia_valid |=
 671                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 672         mutex_lock(&inode->i_mutex);
 673         error = notify_change(dentry, &newattrs);
 674         mutex_unlock(&inode->i_mutex);
 675 out:
 676         return error;
 677 }
 678
 679 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 680 {
 681         struct nameidata nd;
 682         int error;
 683
 684         error = user_path_walk(filename, &nd);
 685         if (error)
 686                 goto out;
 687         error = mnt_want_write(nd.path.mnt);
 688         if (error)
 689                 goto out_release;
 690         error = chown_common(nd.path.dentry, user, group);
 691         mnt_drop_write(nd.path.mnt);
 692 out_release:
 693         path_put(&nd.path);
 694 out:
 695         return error;
 696 }
 697
 698 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 699                              gid_t group, int flag)
 700 {
 701         struct nameidata nd;
 702         int error = -EINVAL;
 703         int follow;
 704
 705         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 706                 goto out;
 707
 708         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 709         error = __user_walk_fd(dfd, filename, follow, &nd);
 710         if (error)
 711                 goto out;
 712         error = mnt_want_write(nd.path.mnt);
 713         if (error)
 714                 goto out_release;
 715         error = chown_common(nd.path.dentry, user, group);
 716         mnt_drop_write(nd.path.mnt);
 717 out_release:
 718         path_put(&nd.path);
 719 out:
 720         return error;
 721 }
 722
 723 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 724 {
 725         struct nameidata nd;
 726         int error;
 727
 728         error = user_path_walk_link(filename, &nd);
 729         if (error)
 730                 goto out;
 731         error = mnt_want_write(nd.path.mnt);
 732         if (error)
 733                 goto out_release;
 734         error = chown_common(nd.path.dentry, user, group);
 735         mnt_drop_write(nd.path.mnt);
 736 out_release:
 737         path_put(&nd.path);
 738 out:
 739         return error;
 740 }
 741
 742
 743 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 744 {
 745         struct file * file;
 746         int error = -EBADF;
 747         struct dentry * dentry;
 748
 749         file = fget(fd);
 750         if (!file)
 751                 goto out;
 752
 753         error = mnt_want_write(file->f_path.mnt);
 754         if (error)
 755                 goto out_fput;
 756         dentry = file->f_path.dentry;
 757         audit_inode(NULL, dentry);
 758         error = chown_common(dentry, user, group);
 759         mnt_drop_write(file->f_path.mnt);
 760 out_fput:
 761         fput(file);
 762 out:
 763         return error;
 764 }
 765
 766 /*
 767  * You have to be very careful that these write
 768  * counts get cleaned up in error cases and
 769  * upon __fput().  This should probably never
 770  * be called outside of __dentry_open().
 771  */
 772 static inline int __get_file_write_access(struct inode *inode,
 773                                           struct vfsmount *mnt)
 774 {
 775         int error;
 776         error = get_write_access(inode);
 777         if (error)
 778                 return error;
 779         /*
 780          * Do not take mount writer counts on
 781          * special files since no writes to
 782          * the mount itself will occur.
 783          */
 784         if (!special_file(inode->i_mode)) {
 785                 /*
 786                  * Balanced in __fput()
 787                  */
 788                 error = mnt_want_write(mnt);
 789                 if (error)
 790                         put_write_access(inode);
 791         }
 792         return error;
 793 }
 794
 795 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 796                                         int flags, struct file *f,
 797                                         int (*open)(struct inode *, struct file *))
 798 {
 799         struct inode *inode;
 800         int error;
 801
 802         f->f_flags = flags;
 803         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 804                                 FMODE_PREAD | FMODE_PWRITE;
 805         inode = dentry->d_inode;
 806         if (f->f_mode & FMODE_WRITE) {
 807                 error = __get_file_write_access(inode, mnt);
 808                 if (error)
 809                         goto cleanup_file;
 810                 if (!special_file(inode->i_mode))
 811                         file_take_write(f);
 812         }
 813
 814         f->f_mapping = inode->i_mapping;
 815         f->f_path.dentry = dentry;
 816         f->f_path.mnt = mnt;
 817         f->f_pos = 0;
 818         f->f_op = fops_get(inode->i_fop);
 819         file_move(f, &inode->i_sb->s_files);
 820
 821         error = security_dentry_open(f);
 822         if (error)
 823                 goto cleanup_all;
 824
 825         if (!open && f->f_op)
 826                 open = f->f_op->open;
 827         if (open) {
 828                 error = open(inode, f);
 829                 if (error)
 830                         goto cleanup_all;
 831         }
 832
 833         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 834
 835         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 836
 837         /* NB: we're sure to have correct a_ops only after f_op->open */
 838         if (f->f_flags & O_DIRECT) {
 839                 if (!f->f_mapping->a_ops ||
 840                     ((!f->f_mapping->a_ops->direct_IO) &&
 841                     (!f->f_mapping->a_ops->get_xip_mem))) {
 842                         fput(f);
 843                         f = ERR_PTR(-EINVAL);
 844                 }
 845         }
 846
 847         return f;
 848
 849 cleanup_all:
 850         fops_put(f->f_op);
 851         if (f->f_mode & FMODE_WRITE) {
 852                 put_write_access(inode);
 853                 if (!special_file(inode->i_mode)) {
 854                         /*
 855                          * We don't consider this a real
 856                          * mnt_want/drop_write() pair
 857                          * because it all happenend right
 858                          * here, so just reset the state.
 859                          */
 860                         file_reset_write(f);
 861                         mnt_drop_write(mnt);
 862                 }
 863         }
 864         file_kill(f);
 865         f->f_path.dentry = NULL;
 866         f->f_path.mnt = NULL;
 867 cleanup_file:
 868         put_filp(f);
 869         dput(dentry);
 870         mntput(mnt);
 871         return ERR_PTR(error);
 872 }
 873
 874 /**
 875  * lookup_instantiate_filp - instantiates the open intent filp
 876  * @nd: pointer to nameidata
 877  * @dentry: pointer to dentry
 878  * @open: open callback
 879  *
 880  * Helper for filesystems that want to use lookup open intents and pass back
 881  * a fully instantiated struct file to the caller.
 882  * This function is meant to be called from within a filesystem's
 883  * lookup method.
 884  * Beware of calling it for non-regular files! Those ->open methods might block
 885  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 886  * leading to a deadlock, as nobody can open that fifo anymore, because
 887  * another process to open fifo will block on locked parent when doing lookup).
 888  * Note that in case of error, nd->intent.open.file is destroyed, but the
 889  * path information remains valid.
 890  * If the open callback is set to NULL, then the standard f_op->open()
 891  * filesystem callback is substituted.
 892  */
 893 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 894                 int (*open)(struct inode *, struct file *))
 895 {
 896         if (IS_ERR(nd->intent.open.file))
 897                 goto out;
 898         if (IS_ERR(dentry))
 899                 goto out_err;
 900         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 901                                              nd->intent.open.flags - 1,
 902                                              nd->intent.open.file,
 903                                              open);
 904 out:
 905         return nd->intent.open.file;
 906 out_err:
 907         release_open_intent(nd);
 908         nd->intent.open.file = (struct file *)dentry;
 909         goto out;
 910 }
 911 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 912
 913 /**
 914  * nameidata_to_filp - convert a nameidata to an open filp.
 915  * @nd: pointer to nameidata
 916  * @flags: open flags
 917  *
 918  * Note that this function destroys the original nameidata
 919  */
 920 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 921 {
 922         struct file *filp;
 923
 924         /* Pick up the filp from the open intent */
 925         filp = nd->intent.open.file;
 926         /* Has the filesystem initialised the file for us? */
 927         if (filp->f_path.dentry == NULL)
 928                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 929                                      NULL);
 930         else
 931                 path_put(&nd->path);
 932         return filp;
 933 }
 934
 935 /*
 936  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 937  * error.
 938  */
 939 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 940 {
 941         int error;
 942         struct file *f;
 943
 944         /*
 945          * We must always pass in a valid mount pointer.   Historically
 946          * callers got away with not passing it, but we must enforce this at
 947          * the earliest possible point now to avoid strange problems deep in the
 948          * filesystem stack.
 949          */
 950         if (!mnt) {
 951                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 952                 dump_stack();
 953                 return ERR_PTR(-EINVAL);
 954         }
 955
 956         error = -ENFILE;
 957         f = get_empty_filp();
 958         if (f == NULL) {
 959                 dput(dentry);
 960                 mntput(mnt);
 961                 return ERR_PTR(error);
 962         }
 963
 964         return __dentry_open(dentry, mnt, flags, f, NULL);
 965 }
 966 EXPORT_SYMBOL(dentry_open);
 967
 968 /*
 969  * Find an empty file descriptor entry, and mark it busy.
 970  */
 971 int get_unused_fd_flags(int flags)
 972 {
 973         struct files_struct * files = current->files;
 974         int fd, error;
 975         struct fdtable *fdt;
 976
 977         error = -EMFILE;
 978         spin_lock(&files->file_lock);
 979
 980 repeat:
 981         fdt = files_fdtable(files);
 982         fd = find_next_zero_bit(fdt->open_fds->fds_bits, fdt->max_fds,
 983                                 files->next_fd);
 984
 985         /*
 986          * N.B. For clone tasks sharing a files structure, this test
 987          * will limit the total number of files that can be opened.
 988          */
 989         if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 990                 goto out;
 991
 992         /* Do we need to expand the fd array or fd set?  */
 993         error = expand_files(files, fd);
 994         if (error < 0)
 995                 goto out;
 996
 997         if (error) {
 998                 /*
 999                  * If we needed to expand the fs array we
1000                  * might have blocked - try again.
1001                  */
1002                 error = -EMFILE;
1003                 goto repeat;
1004         }
1005
1006         FD_SET(fd, fdt->open_fds);
1007         if (flags & O_CLOEXEC)
1008                 FD_SET(fd, fdt->close_on_exec);
1009         else
1010                 FD_CLR(fd, fdt->close_on_exec);
1011         files->next_fd = fd + 1;
1012 #if 1
1013         /* Sanity check */
1014         if (fdt->fd[fd] != NULL) {
1015                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
1016                 fdt->fd[fd] = NULL;
1017         }
1018 #endif
1019         error = fd;
1020
1021 out:
1022         spin_unlock(&files->file_lock);
1023         return error;
1024 }
1025
1026 int get_unused_fd(void)
1027 {
1028         return get_unused_fd_flags(0);
1029 }
1030
1031 EXPORT_SYMBOL(get_unused_fd);
1032
1033 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
1034 {
1035         struct fdtable *fdt = files_fdtable(files);
1036         __FD_CLR(fd, fdt->open_fds);
1037         if (fd < files->next_fd)
1038                 files->next_fd = fd;
1039 }
1040
1041 void put_unused_fd(unsigned int fd)
1042 {
1043         struct files_struct *files = current->files;
1044         spin_lock(&files->file_lock);
1045         __put_unused_fd(files, fd);
1046         spin_unlock(&files->file_lock);
1047 }
1048
1049 EXPORT_SYMBOL(put_unused_fd);
1050
1051 /*
1052  * Install a file pointer in the fd array.
1053  *
1054  * The VFS is full of places where we drop the files lock between
1055  * setting the open_fds bitmap and installing the file in the file
1056  * array.  At any such point, we are vulnerable to a dup2() race
1057  * installing a file in the array before us.  We need to detect this and
1058  * fput() the struct file we are about to overwrite in this case.
1059  *
1060  * It should never happen - if we allow dup2() do it, _really_ bad things
1061  * will follow.
1062  */
1063
1064 void fd_install(unsigned int fd, struct file *file)
1065 {
1066         struct files_struct *files = current->files;
1067         struct fdtable *fdt;
1068         spin_lock(&files->file_lock);
1069         fdt = files_fdtable(files);
1070         BUG_ON(fdt->fd[fd] != NULL);
1071         rcu_assign_pointer(fdt->fd[fd], file);
1072         spin_unlock(&files->file_lock);
1073 }
1074
1075 EXPORT_SYMBOL(fd_install);
1076
1077 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1078 {
1079         char *tmp = getname(filename);
1080         int fd = PTR_ERR(tmp);
1081
1082         if (!IS_ERR(tmp)) {
1083                 fd = get_unused_fd_flags(flags);
1084                 if (fd >= 0) {
1085                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1086                         if (IS_ERR(f)) {
1087                                 put_unused_fd(fd);
1088                                 fd = PTR_ERR(f);
1089                         } else {
1090                                 fsnotify_open(f->f_path.dentry);
1091                                 fd_install(fd, f);
1092                         }
1093                 }
1094                 putname(tmp);
1095         }
1096         return fd;
1097 }
1098
1099 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1100 {
1101         long ret;
1102
1103         if (force_o_largefile())
1104                 flags |= O_LARGEFILE;
1105
1106         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1107         /* avoid REGPARM breakage on x86: */
1108         asmlinkage_protect(3, ret, filename, flags, mode);
1109         return ret;
1110 }
1111
1112 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1113                            int mode)
1114 {
1115         long ret;
1116
1117         if (force_o_largefile())
1118                 flags |= O_LARGEFILE;
1119
1120         ret = do_sys_open(dfd, filename, flags, mode);
1121         /* avoid REGPARM breakage on x86: */
1122         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1123         return ret;
1124 }
1125
1126 #ifndef __alpha__
1127
1128 /*
1129  * For backward compatibility?  Maybe this should be moved
1130  * into arch/i386 instead?
1131  */
1132 asmlinkage long sys_creat(const char __user * pathname, int mode)
1133 {
1134         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1135 }
1136
1137 #endif
1138
1139 /*
1140  * "id" is the POSIX thread ID. We use the
1141  * files pointer for this..
1142  */
1143 int filp_close(struct file *filp, fl_owner_t id)
1144 {
1145         int retval = 0;
1146
1147         if (!file_count(filp)) {
1148                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1149                 return 0;
1150         }
1151
1152         if (filp->f_op && filp->f_op->flush)
1153                 retval = filp->f_op->flush(filp, id);
1154
1155         dnotify_flush(filp, id);
1156         locks_remove_posix(filp, id);
1157         fput(filp);
1158         return retval;
1159 }
1160
1161 EXPORT_SYMBOL(filp_close);
1162
1163 /*
1164  * Careful here! We test whether the file pointer is NULL before
1165  * releasing the fd. This ensures that one clone task can't release
1166  * an fd while another clone is opening it.
1167  */
1168 asmlinkage long sys_close(unsigned int fd)
1169 {
1170         struct file * filp;
1171         struct files_struct *files = current->files;
1172         struct fdtable *fdt;
1173         int retval;
1174
1175         spin_lock(&files->file_lock);
1176         fdt = files_fdtable(files);
1177         if (fd >= fdt->max_fds)
1178                 goto out_unlock;
1179         filp = fdt->fd[fd];
1180         if (!filp)
1181                 goto out_unlock;
1182         rcu_assign_pointer(fdt->fd[fd], NULL);
1183         FD_CLR(fd, fdt->close_on_exec);
1184         __put_unused_fd(files, fd);
1185         spin_unlock(&files->file_lock);
1186         retval = filp_close(filp, files);
1187
1188         /* can't restart close syscall because file table entry was cleared */
1189         if (unlikely(retval == -ERESTARTSYS ||
1190                      retval == -ERESTARTNOINTR ||
1191                      retval == -ERESTARTNOHAND ||
1192                      retval == -ERESTART_RESTARTBLOCK))
1193                 retval = -EINTR;
1194
1195         return retval;
1196
1197 out_unlock:
1198         spin_unlock(&files->file_lock);
1199         return -EBADF;
1200 }
1201
1202 EXPORT_SYMBOL(sys_close);
1203
1204 /*
1205  * This routine simulates a hangup on the tty, to arrange that users
1206  * are given clean terminals at login time.
1207  */
1208 asmlinkage long sys_vhangup(void)
1209 {
1210         if (capable(CAP_SYS_TTY_CONFIG)) {
1211                 /* XXX: this needs locking */
1212                 tty_vhangup(current->signal->tty);
1213                 return 0;
1214         }
1215         return -EPERM;
1216 }
1217
1218 /*
1219  * Called when an inode is about to be open.
1220  * We use this to disallow opening large files on 32bit systems if
1221  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1222  * on this flag in sys_open.
1223  */
1224 int generic_file_open(struct inode * inode, struct file * filp)
1225 {
1226         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1227                 return -EOVERFLOW;
1228         return 0;
1229 }
1230
1231 EXPORT_SYMBOL(generic_file_open);
1232
1233 /*
1234  * This is used by subsystems that don't want seekable
1235  * file descriptors
1236  */
1237 int nonseekable_open(struct inode *inode, struct file *filp)
1238 {
1239         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1240         return 0;
1241 }
1242
1243 EXPORT_SYMBOL(nonseekable_open);