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