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/securebits.h>
  20 #include <linux/security.h>
  21 #include <linux/mount.h>
  22 #include <linux/vfs.h>
  23 #include <linux/fcntl.h>
  24 #include <asm/uaccess.h>
  25 #include <linux/fs.h>
  26 #include <linux/personality.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/rcupdate.h>
  30 #include <linux/audit.h>
  31 #include <linux/falloc.h>
  32
  33 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  34 {
  35         int retval = -ENODEV;
  36
  37         if (dentry) {
  38                 retval = -ENOSYS;
  39                 if (dentry->d_sb->s_op->statfs) {
  40                         memset(buf, 0, sizeof(*buf));
  41                         retval = security_sb_statfs(dentry);
  42                         if (retval)
  43                                 return retval;
  44                         retval = dentry->d_sb->s_op->statfs(dentry, buf);
  45                         if (retval == 0 && buf->f_frsize == 0)
  46                                 buf->f_frsize = buf->f_bsize;
  47                 }
  48         }
  49         return retval;
  50 }
  51
  52 EXPORT_SYMBOL(vfs_statfs);
  53
  54 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
  55 {
  56         struct kstatfs st;
  57         int retval;
  58
  59         retval = vfs_statfs(dentry, &st);
  60         if (retval)
  61                 return retval;
  62
  63         if (sizeof(*buf) == sizeof(st))
  64                 memcpy(buf, &st, sizeof(st));
  65         else {
  66                 if (sizeof buf->f_blocks == 4) {
  67                         if ((st.f_blocks | st.f_bfree | st.f_bavail |
  68                              st.f_bsize | st.f_frsize) &
  69                             0xffffffff00000000ULL)
  70                                 return -EOVERFLOW;
  71                         /*
  72                          * f_files and f_ffree may be -1; it's okay to stuff
  73                          * that into 32 bits
  74                          */
  75                         if (st.f_files != -1 &&
  76                             (st.f_files & 0xffffffff00000000ULL))
  77                                 return -EOVERFLOW;
  78                         if (st.f_ffree != -1 &&
  79                             (st.f_ffree & 0xffffffff00000000ULL))
  80                                 return -EOVERFLOW;
  81                 }
  82
  83                 buf->f_type = st.f_type;
  84                 buf->f_bsize = st.f_bsize;
  85                 buf->f_blocks = st.f_blocks;
  86                 buf->f_bfree = st.f_bfree;
  87                 buf->f_bavail = st.f_bavail;
  88                 buf->f_files = st.f_files;
  89                 buf->f_ffree = st.f_ffree;
  90                 buf->f_fsid = st.f_fsid;
  91                 buf->f_namelen = st.f_namelen;
  92                 buf->f_frsize = st.f_frsize;
  93                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  94         }
  95         return 0;
  96 }
  97
  98 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
  99 {
 100         struct kstatfs st;
 101         int retval;
 102
 103         retval = vfs_statfs(dentry, &st);
 104         if (retval)
 105                 return retval;
 106
 107         if (sizeof(*buf) == sizeof(st))
 108                 memcpy(buf, &st, sizeof(st));
 109         else {
 110                 buf->f_type = st.f_type;
 111                 buf->f_bsize = st.f_bsize;
 112                 buf->f_blocks = st.f_blocks;
 113                 buf->f_bfree = st.f_bfree;
 114                 buf->f_bavail = st.f_bavail;
 115                 buf->f_files = st.f_files;
 116                 buf->f_ffree = st.f_ffree;
 117                 buf->f_fsid = st.f_fsid;
 118                 buf->f_namelen = st.f_namelen;
 119                 buf->f_frsize = st.f_frsize;
 120                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 121         }
 122         return 0;
 123 }
 124
 125 asmlinkage long sys_statfs(const char __user * path, struct statfs __user * buf)
 126 {
 127         struct nameidata nd;
 128         int error;
 129
 130         error = user_path_walk(path, &nd);
 131         if (!error) {
 132                 struct statfs tmp;
 133                 error = vfs_statfs_native(nd.path.dentry, &tmp);
 134                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 135                         error = -EFAULT;
 136                 path_put(&nd.path);
 137         }
 138         return error;
 139 }
 140
 141
 142 asmlinkage long sys_statfs64(const char __user *path, size_t sz, struct statfs64 __user *buf)
 143 {
 144         struct nameidata nd;
 145         long error;
 146
 147         if (sz != sizeof(*buf))
 148                 return -EINVAL;
 149         error = user_path_walk(path, &nd);
 150         if (!error) {
 151                 struct statfs64 tmp;
 152                 error = vfs_statfs64(nd.path.dentry, &tmp);
 153                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 154                         error = -EFAULT;
 155                 path_put(&nd.path);
 156         }
 157         return error;
 158 }
 159
 160
 161 asmlinkage long sys_fstatfs(unsigned int fd, struct statfs __user * buf)
 162 {
 163         struct file * file;
 164         struct statfs tmp;
 165         int error;
 166
 167         error = -EBADF;
 168         file = fget(fd);
 169         if (!file)
 170                 goto out;
 171         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 172         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 173                 error = -EFAULT;
 174         fput(file);
 175 out:
 176         return error;
 177 }
 178
 179 asmlinkage long sys_fstatfs64(unsigned int fd, size_t sz, struct statfs64 __user *buf)
 180 {
 181         struct file * file;
 182         struct statfs64 tmp;
 183         int error;
 184
 185         if (sz != sizeof(*buf))
 186                 return -EINVAL;
 187
 188         error = -EBADF;
 189         file = fget(fd);
 190         if (!file)
 191                 goto out;
 192         error = vfs_statfs64(file->f_path.dentry, &tmp);
 193         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 194                 error = -EFAULT;
 195         fput(file);
 196 out:
 197         return error;
 198 }
 199
 200 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 201         struct file *filp)
 202 {
 203         int err;
 204         struct iattr newattrs;
 205
 206         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 207         if (length < 0)
 208                 return -EINVAL;
 209
 210         newattrs.ia_size = length;
 211         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 212         if (filp) {
 213                 newattrs.ia_file = filp;
 214                 newattrs.ia_valid |= ATTR_FILE;
 215         }
 216
 217         /* Remove suid/sgid on truncate too */
 218         newattrs.ia_valid |= should_remove_suid(dentry);
 219
 220         mutex_lock(&dentry->d_inode->i_mutex);
 221         err = notify_change(dentry, &newattrs);
 222         mutex_unlock(&dentry->d_inode->i_mutex);
 223         return err;
 224 }
 225
 226 static long do_sys_truncate(const char __user * path, loff_t length)
 227 {
 228         struct nameidata nd;
 229         struct inode * inode;
 230         int error;
 231
 232         error = -EINVAL;
 233         if (length < 0) /* sorry, but loff_t says... */
 234                 goto out;
 235
 236         error = user_path_walk(path, &nd);
 237         if (error)
 238                 goto out;
 239         inode = nd.path.dentry->d_inode;
 240
 241         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 242         error = -EISDIR;
 243         if (S_ISDIR(inode->i_mode))
 244                 goto dput_and_out;
 245
 246         error = -EINVAL;
 247         if (!S_ISREG(inode->i_mode))
 248                 goto dput_and_out;
 249
 250         error = mnt_want_write(nd.path.mnt);
 251         if (error)
 252                 goto dput_and_out;
 253
 254         error = vfs_permission(&nd, MAY_WRITE);
 255         if (error)
 256                 goto mnt_drop_write_and_out;
 257
 258         error = -EPERM;
 259         if (IS_APPEND(inode))
 260                 goto mnt_drop_write_and_out;
 261
 262         error = get_write_access(inode);
 263         if (error)
 264                 goto mnt_drop_write_and_out;
 265
 266         /*
 267          * Make sure that there are no leases.  get_write_access() protects
 268          * against the truncate racing with a lease-granting setlease().
 269          */
 270         error = break_lease(inode, FMODE_WRITE);
 271         if (error)
 272                 goto put_write_and_out;
 273
 274         error = locks_verify_truncate(inode, NULL, length);
 275         if (!error) {
 276                 DQUOT_INIT(inode);
 277                 error = do_truncate(nd.path.dentry, length, 0, NULL);
 278         }
 279
 280 put_write_and_out:
 281         put_write_access(inode);
 282 mnt_drop_write_and_out:
 283         mnt_drop_write(nd.path.mnt);
 284 dput_and_out:
 285         path_put(&nd.path);
 286 out:
 287         return error;
 288 }
 289
 290 asmlinkage long sys_truncate(const char __user * path, unsigned long length)
 291 {
 292         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 293         return do_sys_truncate(path, (long)length);
 294 }
 295
 296 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 297 {
 298         struct inode * inode;
 299         struct dentry *dentry;
 300         struct file * file;
 301         int error;
 302
 303         error = -EINVAL;
 304         if (length < 0)
 305                 goto out;
 306         error = -EBADF;
 307         file = fget(fd);
 308         if (!file)
 309                 goto out;
 310
 311         /* explicitly opened as large or we are on 64-bit box */
 312         if (file->f_flags & O_LARGEFILE)
 313                 small = 0;
 314
 315         dentry = file->f_path.dentry;
 316         inode = dentry->d_inode;
 317         error = -EINVAL;
 318         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 319                 goto out_putf;
 320
 321         error = -EINVAL;
 322         /* Cannot ftruncate over 2^31 bytes without large file support */
 323         if (small && length > MAX_NON_LFS)
 324                 goto out_putf;
 325
 326         error = -EPERM;
 327         if (IS_APPEND(inode))
 328                 goto out_putf;
 329
 330         error = locks_verify_truncate(inode, file, length);
 331         if (!error)
 332                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 333 out_putf:
 334         fput(file);
 335 out:
 336         return error;
 337 }
 338
 339 asmlinkage long sys_ftruncate(unsigned int fd, unsigned long length)
 340 {
 341         long ret = do_sys_ftruncate(fd, length, 1);
 342         /* avoid REGPARM breakage on x86: */
 343         asmlinkage_protect(2, ret, fd, length);
 344         return ret;
 345 }
 346
 347 /* LFS versions of truncate are only needed on 32 bit machines */
 348 #if BITS_PER_LONG == 32
 349 asmlinkage long sys_truncate64(const char __user * path, loff_t length)
 350 {
 351         return do_sys_truncate(path, length);
 352 }
 353
 354 asmlinkage long sys_ftruncate64(unsigned int fd, loff_t length)
 355 {
 356         long ret = do_sys_ftruncate(fd, length, 0);
 357         /* avoid REGPARM breakage on x86: */
 358         asmlinkage_protect(2, ret, fd, length);
 359         return ret;
 360 }
 361 #endif
 362
 363 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 364 {
 365         struct file *file;
 366         struct inode *inode;
 367         long ret = -EINVAL;
 368
 369         if (offset < 0 || len <= 0)
 370                 goto out;
 371
 372         /* Return error if mode is not supported */
 373         ret = -EOPNOTSUPP;
 374         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 375                 goto out;
 376
 377         ret = -EBADF;
 378         file = fget(fd);
 379         if (!file)
 380                 goto out;
 381         if (!(file->f_mode & FMODE_WRITE))
 382                 goto out_fput;
 383         /*
 384          * Revalidate the write permissions, in case security policy has
 385          * changed since the files were opened.
 386          */
 387         ret = security_file_permission(file, MAY_WRITE);
 388         if (ret)
 389                 goto out_fput;
 390
 391         inode = file->f_path.dentry->d_inode;
 392
 393         ret = -ESPIPE;
 394         if (S_ISFIFO(inode->i_mode))
 395                 goto out_fput;
 396
 397         ret = -ENODEV;
 398         /*
 399          * Let individual file system decide if it supports preallocation
 400          * for directories or not.
 401          */
 402         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 403                 goto out_fput;
 404
 405         ret = -EFBIG;
 406         /* Check for wrap through zero too */
 407         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 408                 goto out_fput;
 409
 410         if (inode->i_op && inode->i_op->fallocate)
 411                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 412         else
 413                 ret = -EOPNOTSUPP;
 414
 415 out_fput:
 416         fput(file);
 417 out:
 418         return ret;
 419 }
 420
 421 /*
 422  * access() needs to use the real uid/gid, not the effective uid/gid.
 423  * We do this by temporarily clearing all FS-related capabilities and
 424  * switching the fsuid/fsgid around to the real ones.
 425  */
 426 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 427 {
 428         struct nameidata nd;
 429         int old_fsuid, old_fsgid;
 430         kernel_cap_t uninitialized_var(old_cap);  /* !SECURE_NO_SETUID_FIXUP */
 431         int res;
 432
 433         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 434                 return -EINVAL;
 435
 436         old_fsuid = current->fsuid;
 437         old_fsgid = current->fsgid;
 438
 439         current->fsuid = current->uid;
 440         current->fsgid = current->gid;
 441
 442         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 443                 /*
 444                  * Clear the capabilities if we switch to a non-root user
 445                  */
 446 #ifndef CONFIG_SECURITY_FILE_CAPABILITIES
 447                 /*
 448                  * FIXME: There is a race here against sys_capset.  The
 449                  * capabilities can change yet we will restore the old
 450                  * value below.  We should hold task_capabilities_lock,
 451                  * but we cannot because user_path_walk can sleep.
 452                  */
 453 #endif /* ndef CONFIG_SECURITY_FILE_CAPABILITIES */
 454                 if (current->uid)
 455                         old_cap = cap_set_effective(__cap_empty_set);
 456                 else
 457                         old_cap = cap_set_effective(current->cap_permitted);
 458         }
 459
 460         res = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
 461         if (res)
 462                 goto out;
 463
 464         res = vfs_permission(&nd, mode | MAY_ACCESS);
 465         /* SuS v2 requires we report a read only fs too */
 466         if(res || !(mode & S_IWOTH) ||
 467            special_file(nd.path.dentry->d_inode->i_mode))
 468                 goto out_path_release;
 469         /*
 470          * This is a rare case where using __mnt_is_readonly()
 471          * is OK without a mnt_want/drop_write() pair.  Since
 472          * no actual write to the fs is performed here, we do
 473          * not need to telegraph to that to anyone.
 474          *
 475          * By doing this, we accept that this access is
 476          * inherently racy and know that the fs may change
 477          * state before we even see this result.
 478          */
 479         if (__mnt_is_readonly(nd.path.mnt))
 480                 res = -EROFS;
 481
 482 out_path_release:
 483         path_put(&nd.path);
 484 out:
 485         current->fsuid = old_fsuid;
 486         current->fsgid = old_fsgid;
 487
 488         if (!issecure(SECURE_NO_SETUID_FIXUP))
 489                 cap_set_effective(old_cap);
 490
 491         return res;
 492 }
 493
 494 asmlinkage long sys_access(const char __user *filename, int mode)
 495 {
 496         return sys_faccessat(AT_FDCWD, filename, mode);
 497 }
 498
 499 asmlinkage long sys_chdir(const char __user * filename)
 500 {
 501         struct nameidata nd;
 502         int error;
 503
 504         error = __user_walk(filename, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &nd);
 505         if (error)
 506                 goto out;
 507
 508         error = vfs_permission(&nd, MAY_EXEC | MAY_ACCESS);
 509         if (error)
 510                 goto dput_and_out;
 511
 512         set_fs_pwd(current->fs, &nd.path);
 513
 514 dput_and_out:
 515         path_put(&nd.path);
 516 out:
 517         return error;
 518 }
 519
 520 asmlinkage long sys_fchdir(unsigned int fd)
 521 {
 522         struct file *file;
 523         struct inode *inode;
 524         int error;
 525
 526         error = -EBADF;
 527         file = fget(fd);
 528         if (!file)
 529                 goto out;
 530
 531         inode = file->f_path.dentry->d_inode;
 532
 533         error = -ENOTDIR;
 534         if (!S_ISDIR(inode->i_mode))
 535                 goto out_putf;
 536
 537         error = file_permission(file, MAY_EXEC | MAY_ACCESS);
 538         if (!error)
 539                 set_fs_pwd(current->fs, &file->f_path);
 540 out_putf:
 541         fput(file);
 542 out:
 543         return error;
 544 }
 545
 546 asmlinkage long sys_chroot(const char __user * filename)
 547 {
 548         struct nameidata nd;
 549         int error;
 550
 551         error = __user_walk(filename, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &nd);
 552         if (error)
 553                 goto out;
 554
 555         error = vfs_permission(&nd, MAY_EXEC | MAY_ACCESS);
 556         if (error)
 557                 goto dput_and_out;
 558
 559         error = -EPERM;
 560         if (!capable(CAP_SYS_CHROOT))
 561                 goto dput_and_out;
 562
 563         set_fs_root(current->fs, &nd.path);
 564         error = 0;
 565 dput_and_out:
 566         path_put(&nd.path);
 567 out:
 568         return error;
 569 }
 570
 571 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 572 {
 573         struct inode * inode;
 574         struct dentry * dentry;
 575         struct file * file;
 576         int err = -EBADF;
 577         struct iattr newattrs;
 578
 579         file = fget(fd);
 580         if (!file)
 581                 goto out;
 582
 583         dentry = file->f_path.dentry;
 584         inode = dentry->d_inode;
 585
 586         audit_inode(NULL, dentry);
 587
 588         err = mnt_want_write(file->f_path.mnt);
 589         if (err)
 590                 goto out_putf;
 591         err = -EPERM;
 592         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 593                 goto out_drop_write;
 594         mutex_lock(&inode->i_mutex);
 595         if (mode == (mode_t) -1)
 596                 mode = inode->i_mode;
 597         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 598         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 599         err = notify_change(dentry, &newattrs);
 600         mutex_unlock(&inode->i_mutex);
 601
 602 out_drop_write:
 603         mnt_drop_write(file->f_path.mnt);
 604 out_putf:
 605         fput(file);
 606 out:
 607         return err;
 608 }
 609
 610 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 611                              mode_t mode)
 612 {
 613         struct nameidata nd;
 614         struct inode * inode;
 615         int error;
 616         struct iattr newattrs;
 617
 618         error = __user_walk_fd(dfd, filename, LOOKUP_FOLLOW, &nd);
 619         if (error)
 620                 goto out;
 621         inode = nd.path.dentry->d_inode;
 622
 623         error = mnt_want_write(nd.path.mnt);
 624         if (error)
 625                 goto dput_and_out;
 626
 627         error = -EPERM;
 628         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 629                 goto out_drop_write;
 630
 631         mutex_lock(&inode->i_mutex);
 632         if (mode == (mode_t) -1)
 633                 mode = inode->i_mode;
 634         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 635         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 636         error = notify_change(nd.path.dentry, &newattrs);
 637         mutex_unlock(&inode->i_mutex);
 638
 639 out_drop_write:
 640         mnt_drop_write(nd.path.mnt);
 641 dput_and_out:
 642         path_put(&nd.path);
 643 out:
 644         return error;
 645 }
 646
 647 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 648 {
 649         return sys_fchmodat(AT_FDCWD, filename, mode);
 650 }
 651
 652 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 653 {
 654         struct inode * inode;
 655         int error;
 656         struct iattr newattrs;
 657
 658         error = -ENOENT;
 659         if (!(inode = dentry->d_inode)) {
 660                 printk(KERN_ERR "chown_common: NULL inode\n");
 661                 goto out;
 662         }
 663         error = -EPERM;
 664         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 665                 goto out;
 666         newattrs.ia_valid =  ATTR_CTIME;
 667         if (user != (uid_t) -1) {
 668                 newattrs.ia_valid |= ATTR_UID;
 669                 newattrs.ia_uid = user;
 670         }
 671         if (group != (gid_t) -1) {
 672                 newattrs.ia_valid |= ATTR_GID;
 673                 newattrs.ia_gid = group;
 674         }
 675         if (!S_ISDIR(inode->i_mode))
 676                 newattrs.ia_valid |=
 677                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 678         mutex_lock(&inode->i_mutex);
 679         error = notify_change(dentry, &newattrs);
 680         mutex_unlock(&inode->i_mutex);
 681 out:
 682         return error;
 683 }
 684
 685 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 686 {
 687         struct nameidata nd;
 688         int error;
 689
 690         error = user_path_walk(filename, &nd);
 691         if (error)
 692                 goto out;
 693         error = mnt_want_write(nd.path.mnt);
 694         if (error)
 695                 goto out_release;
 696         error = chown_common(nd.path.dentry, user, group);
 697         mnt_drop_write(nd.path.mnt);
 698 out_release:
 699         path_put(&nd.path);
 700 out:
 701         return error;
 702 }
 703
 704 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 705                              gid_t group, int flag)
 706 {
 707         struct nameidata nd;
 708         int error = -EINVAL;
 709         int follow;
 710
 711         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 712                 goto out;
 713
 714         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 715         error = __user_walk_fd(dfd, filename, follow, &nd);
 716         if (error)
 717                 goto out;
 718         error = mnt_want_write(nd.path.mnt);
 719         if (error)
 720                 goto out_release;
 721         error = chown_common(nd.path.dentry, user, group);
 722         mnt_drop_write(nd.path.mnt);
 723 out_release:
 724         path_put(&nd.path);
 725 out:
 726         return error;
 727 }
 728
 729 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 730 {
 731         struct nameidata nd;
 732         int error;
 733
 734         error = user_path_walk_link(filename, &nd);
 735         if (error)
 736                 goto out;
 737         error = mnt_want_write(nd.path.mnt);
 738         if (error)
 739                 goto out_release;
 740         error = chown_common(nd.path.dentry, user, group);
 741         mnt_drop_write(nd.path.mnt);
 742 out_release:
 743         path_put(&nd.path);
 744 out:
 745         return error;
 746 }
 747
 748
 749 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 750 {
 751         struct file * file;
 752         int error = -EBADF;
 753         struct dentry * dentry;
 754
 755         file = fget(fd);
 756         if (!file)
 757                 goto out;
 758
 759         error = mnt_want_write(file->f_path.mnt);
 760         if (error)
 761                 goto out_fput;
 762         dentry = file->f_path.dentry;
 763         audit_inode(NULL, dentry);
 764         error = chown_common(dentry, user, group);
 765         mnt_drop_write(file->f_path.mnt);
 766 out_fput:
 767         fput(file);
 768 out:
 769         return error;
 770 }
 771
 772 /*
 773  * You have to be very careful that these write
 774  * counts get cleaned up in error cases and
 775  * upon __fput().  This should probably never
 776  * be called outside of __dentry_open().
 777  */
 778 static inline int __get_file_write_access(struct inode *inode,
 779                                           struct vfsmount *mnt)
 780 {
 781         int error;
 782         error = get_write_access(inode);
 783         if (error)
 784                 return error;
 785         /*
 786          * Do not take mount writer counts on
 787          * special files since no writes to
 788          * the mount itself will occur.
 789          */
 790         if (!special_file(inode->i_mode)) {
 791                 /*
 792                  * Balanced in __fput()
 793                  */
 794                 error = mnt_want_write(mnt);
 795                 if (error)
 796                         put_write_access(inode);
 797         }
 798         return error;
 799 }
 800
 801 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 802                                         int flags, struct file *f,
 803                                         int (*open)(struct inode *, struct file *))
 804 {
 805         struct inode *inode;
 806         int error;
 807
 808         f->f_flags = flags;
 809         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 810                                 FMODE_PREAD | FMODE_PWRITE;
 811         inode = dentry->d_inode;
 812         if (f->f_mode & FMODE_WRITE) {
 813                 error = __get_file_write_access(inode, mnt);
 814                 if (error)
 815                         goto cleanup_file;
 816                 if (!special_file(inode->i_mode))
 817                         file_take_write(f);
 818         }
 819
 820         f->f_mapping = inode->i_mapping;
 821         f->f_path.dentry = dentry;
 822         f->f_path.mnt = mnt;
 823         f->f_pos = 0;
 824         f->f_op = fops_get(inode->i_fop);
 825         file_move(f, &inode->i_sb->s_files);
 826
 827         error = security_dentry_open(f);
 828         if (error)
 829                 goto cleanup_all;
 830
 831         if (!open && f->f_op)
 832                 open = f->f_op->open;
 833         if (open) {
 834                 error = open(inode, f);
 835                 if (error)
 836                         goto cleanup_all;
 837         }
 838
 839         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 840
 841         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 842
 843         /* NB: we're sure to have correct a_ops only after f_op->open */
 844         if (f->f_flags & O_DIRECT) {
 845                 if (!f->f_mapping->a_ops ||
 846                     ((!f->f_mapping->a_ops->direct_IO) &&
 847                     (!f->f_mapping->a_ops->get_xip_mem))) {
 848                         fput(f);
 849                         f = ERR_PTR(-EINVAL);
 850                 }
 851         }
 852
 853         return f;
 854
 855 cleanup_all:
 856         fops_put(f->f_op);
 857         if (f->f_mode & FMODE_WRITE) {
 858                 put_write_access(inode);
 859                 if (!special_file(inode->i_mode)) {
 860                         /*
 861                          * We don't consider this a real
 862                          * mnt_want/drop_write() pair
 863                          * because it all happenend right
 864                          * here, so just reset the state.
 865                          */
 866                         file_reset_write(f);
 867                         mnt_drop_write(mnt);
 868                 }
 869         }
 870         file_kill(f);
 871         f->f_path.dentry = NULL;
 872         f->f_path.mnt = NULL;
 873 cleanup_file:
 874         put_filp(f);
 875         dput(dentry);
 876         mntput(mnt);
 877         return ERR_PTR(error);
 878 }
 879
 880 /**
 881  * lookup_instantiate_filp - instantiates the open intent filp
 882  * @nd: pointer to nameidata
 883  * @dentry: pointer to dentry
 884  * @open: open callback
 885  *
 886  * Helper for filesystems that want to use lookup open intents and pass back
 887  * a fully instantiated struct file to the caller.
 888  * This function is meant to be called from within a filesystem's
 889  * lookup method.
 890  * Beware of calling it for non-regular files! Those ->open methods might block
 891  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 892  * leading to a deadlock, as nobody can open that fifo anymore, because
 893  * another process to open fifo will block on locked parent when doing lookup).
 894  * Note that in case of error, nd->intent.open.file is destroyed, but the
 895  * path information remains valid.
 896  * If the open callback is set to NULL, then the standard f_op->open()
 897  * filesystem callback is substituted.
 898  */
 899 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 900                 int (*open)(struct inode *, struct file *))
 901 {
 902         if (IS_ERR(nd->intent.open.file))
 903                 goto out;
 904         if (IS_ERR(dentry))
 905                 goto out_err;
 906         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 907                                              nd->intent.open.flags - 1,
 908                                              nd->intent.open.file,
 909                                              open);
 910 out:
 911         return nd->intent.open.file;
 912 out_err:
 913         release_open_intent(nd);
 914         nd->intent.open.file = (struct file *)dentry;
 915         goto out;
 916 }
 917 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 918
 919 /**
 920  * nameidata_to_filp - convert a nameidata to an open filp.
 921  * @nd: pointer to nameidata
 922  * @flags: open flags
 923  *
 924  * Note that this function destroys the original nameidata
 925  */
 926 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 927 {
 928         struct file *filp;
 929
 930         /* Pick up the filp from the open intent */
 931         filp = nd->intent.open.file;
 932         /* Has the filesystem initialised the file for us? */
 933         if (filp->f_path.dentry == NULL)
 934                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 935                                      NULL);
 936         else
 937                 path_put(&nd->path);
 938         return filp;
 939 }
 940
 941 /*
 942  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 943  * error.
 944  */
 945 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 946 {
 947         int error;
 948         struct file *f;
 949
 950         /*
 951          * We must always pass in a valid mount pointer.   Historically
 952          * callers got away with not passing it, but we must enforce this at
 953          * the earliest possible point now to avoid strange problems deep in the
 954          * filesystem stack.
 955          */
 956         if (!mnt) {
 957                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 958                 dump_stack();
 959                 return ERR_PTR(-EINVAL);
 960         }
 961
 962         error = -ENFILE;
 963         f = get_empty_filp();
 964         if (f == NULL) {
 965                 dput(dentry);
 966                 mntput(mnt);
 967                 return ERR_PTR(error);
 968         }
 969
 970         return __dentry_open(dentry, mnt, flags, f, NULL);
 971 }
 972 EXPORT_SYMBOL(dentry_open);
 973
 974 /*
 975  * Find an empty file descriptor entry, and mark it busy.
 976  */
 977 int get_unused_fd_flags(int flags)
 978 {
 979         struct files_struct * files = current->files;
 980         int fd, error;
 981         struct fdtable *fdt;
 982
 983         error = -EMFILE;
 984         spin_lock(&files->file_lock);
 985
 986 repeat:
 987         fdt = files_fdtable(files);
 988         fd = find_next_zero_bit(fdt->open_fds->fds_bits, fdt->max_fds,
 989                                 files->next_fd);
 990
 991         /*
 992          * N.B. For clone tasks sharing a files structure, this test
 993          * will limit the total number of files that can be opened.
 994          */
 995         if (fd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 996                 goto out;
 997
 998         /* Do we need to expand the fd array or fd set?  */
 999         error = expand_files(files, fd);
1000         if (error < 0)
1001                 goto out;
1002
1003         if (error) {
1004                 /*
1005                  * If we needed to expand the fs array we
1006                  * might have blocked - try again.
1007                  */
1008                 error = -EMFILE;
1009                 goto repeat;
1010         }
1011
1012         FD_SET(fd, fdt->open_fds);
1013         if (flags & O_CLOEXEC)
1014                 FD_SET(fd, fdt->close_on_exec);
1015         else
1016                 FD_CLR(fd, fdt->close_on_exec);
1017         files->next_fd = fd + 1;
1018 #if 1
1019         /* Sanity check */
1020         if (fdt->fd[fd] != NULL) {
1021                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
1022                 fdt->fd[fd] = NULL;
1023         }
1024 #endif
1025         error = fd;
1026
1027 out:
1028         spin_unlock(&files->file_lock);
1029         return error;
1030 }
1031
1032 int get_unused_fd(void)
1033 {
1034         return get_unused_fd_flags(0);
1035 }
1036
1037 EXPORT_SYMBOL(get_unused_fd);
1038
1039 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
1040 {
1041         struct fdtable *fdt = files_fdtable(files);
1042         __FD_CLR(fd, fdt->open_fds);
1043         if (fd < files->next_fd)
1044                 files->next_fd = fd;
1045 }
1046
1047 void put_unused_fd(unsigned int fd)
1048 {
1049         struct files_struct *files = current->files;
1050         spin_lock(&files->file_lock);
1051         __put_unused_fd(files, fd);
1052         spin_unlock(&files->file_lock);
1053 }
1054
1055 EXPORT_SYMBOL(put_unused_fd);
1056
1057 /*
1058  * Install a file pointer in the fd array.
1059  *
1060  * The VFS is full of places where we drop the files lock between
1061  * setting the open_fds bitmap and installing the file in the file
1062  * array.  At any such point, we are vulnerable to a dup2() race
1063  * installing a file in the array before us.  We need to detect this and
1064  * fput() the struct file we are about to overwrite in this case.
1065  *
1066  * It should never happen - if we allow dup2() do it, _really_ bad things
1067  * will follow.
1068  */
1069
1070 void fd_install(unsigned int fd, struct file *file)
1071 {
1072         struct files_struct *files = current->files;
1073         struct fdtable *fdt;
1074         spin_lock(&files->file_lock);
1075         fdt = files_fdtable(files);
1076         BUG_ON(fdt->fd[fd] != NULL);
1077         rcu_assign_pointer(fdt->fd[fd], file);
1078         spin_unlock(&files->file_lock);
1079 }
1080
1081 EXPORT_SYMBOL(fd_install);
1082
1083 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1084 {
1085         char *tmp = getname(filename);
1086         int fd = PTR_ERR(tmp);
1087
1088         if (!IS_ERR(tmp)) {
1089                 fd = get_unused_fd_flags(flags);
1090                 if (fd >= 0) {
1091                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1092                         if (IS_ERR(f)) {
1093                                 put_unused_fd(fd);
1094                                 fd = PTR_ERR(f);
1095                         } else {
1096                                 fsnotify_open(f->f_path.dentry);
1097                                 fd_install(fd, f);
1098                         }
1099                 }
1100                 putname(tmp);
1101         }
1102         return fd;
1103 }
1104
1105 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1106 {
1107         long ret;
1108
1109         if (force_o_largefile())
1110                 flags |= O_LARGEFILE;
1111
1112         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1113         /* avoid REGPARM breakage on x86: */
1114         asmlinkage_protect(3, ret, filename, flags, mode);
1115         return ret;
1116 }
1117
1118 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1119                            int mode)
1120 {
1121         long ret;
1122
1123         if (force_o_largefile())
1124                 flags |= O_LARGEFILE;
1125
1126         ret = do_sys_open(dfd, filename, flags, mode);
1127         /* avoid REGPARM breakage on x86: */
1128         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1129         return ret;
1130 }
1131
1132 #ifndef __alpha__
1133
1134 /*
1135  * For backward compatibility?  Maybe this should be moved
1136  * into arch/i386 instead?
1137  */
1138 asmlinkage long sys_creat(const char __user * pathname, int mode)
1139 {
1140         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1141 }
1142
1143 #endif
1144
1145 /*
1146  * "id" is the POSIX thread ID. We use the
1147  * files pointer for this..
1148  */
1149 int filp_close(struct file *filp, fl_owner_t id)
1150 {
1151         int retval = 0;
1152
1153         if (!file_count(filp)) {
1154                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1155                 return 0;
1156         }
1157
1158         if (filp->f_op && filp->f_op->flush)
1159                 retval = filp->f_op->flush(filp, id);
1160
1161         dnotify_flush(filp, id);
1162         locks_remove_posix(filp, id);
1163         fput(filp);
1164         return retval;
1165 }
1166
1167 EXPORT_SYMBOL(filp_close);
1168
1169 /*
1170  * Careful here! We test whether the file pointer is NULL before
1171  * releasing the fd. This ensures that one clone task can't release
1172  * an fd while another clone is opening it.
1173  */
1174 asmlinkage long sys_close(unsigned int fd)
1175 {
1176         struct file * filp;
1177         struct files_struct *files = current->files;
1178         struct fdtable *fdt;
1179         int retval;
1180
1181         spin_lock(&files->file_lock);
1182         fdt = files_fdtable(files);
1183         if (fd >= fdt->max_fds)
1184                 goto out_unlock;
1185         filp = fdt->fd[fd];
1186         if (!filp)
1187                 goto out_unlock;
1188         rcu_assign_pointer(fdt->fd[fd], NULL);
1189         FD_CLR(fd, fdt->close_on_exec);
1190         __put_unused_fd(files, fd);
1191         spin_unlock(&files->file_lock);
1192         retval = filp_close(filp, files);
1193
1194         /* can't restart close syscall because file table entry was cleared */
1195         if (unlikely(retval == -ERESTARTSYS ||
1196                      retval == -ERESTARTNOINTR ||
1197                      retval == -ERESTARTNOHAND ||
1198                      retval == -ERESTART_RESTARTBLOCK))
1199                 retval = -EINTR;
1200
1201         return retval;
1202
1203 out_unlock:
1204         spin_unlock(&files->file_lock);
1205         return -EBADF;
1206 }
1207
1208 EXPORT_SYMBOL(sys_close);
1209
1210 /*
1211  * This routine simulates a hangup on the tty, to arrange that users
1212  * are given clean terminals at login time.
1213  */
1214 asmlinkage long sys_vhangup(void)
1215 {
1216         if (capable(CAP_SYS_TTY_CONFIG)) {
1217                 /* XXX: this needs locking */
1218                 tty_vhangup(current->signal->tty);
1219                 return 0;
1220         }
1221         return -EPERM;
1222 }
1223
1224 /*
1225  * Called when an inode is about to be open.
1226  * We use this to disallow opening large files on 32bit systems if
1227  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1228  * on this flag in sys_open.
1229  */
1230 int generic_file_open(struct inode * inode, struct file * filp)
1231 {
1232         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1233                 return -EOVERFLOW;
1234         return 0;
1235 }
1236
1237 EXPORT_SYMBOL(generic_file_open);
1238
1239 /*
1240  * This is used by subsystems that don't want seekable
1241  * file descriptors
1242  */
1243 int nonseekable_open(struct inode *inode, struct file *filp)
1244 {
1245         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1246         return 0;
1247 }
1248
1249 EXPORT_SYMBOL(nonseekable_open);