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 *pathname, struct statfs __user * buf)
 126 {
 127         struct path path;
 128         int error;
 129
 130         error = user_path(pathname, &path);
 131         if (!error) {
 132                 struct statfs tmp;
 133                 error = vfs_statfs_native(path.dentry, &tmp);
 134                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 135                         error = -EFAULT;
 136                 path_put(&path);
 137         }
 138         return error;
 139 }
 140
 141
 142 asmlinkage long sys_statfs64(const char __user *pathname, size_t sz, struct statfs64 __user *buf)
 143 {
 144         struct path path;
 145         long error;
 146
 147         if (sz != sizeof(*buf))
 148                 return -EINVAL;
 149         error = user_path(pathname, &path);
 150         if (!error) {
 151                 struct statfs64 tmp;
 152                 error = vfs_statfs64(path.dentry, &tmp);
 153                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 154                         error = -EFAULT;
 155                 path_put(&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 *pathname, loff_t length)
 227 {
 228         struct path path;
 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(pathname, &path);
 237         if (error)
 238                 goto out;
 239         inode = 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(path.mnt);
 251         if (error)
 252                 goto dput_and_out;
 253
 254         error = inode_permission(inode, 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(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(path.mnt);
 284 dput_and_out:
 285         path_put(&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 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 350 {
 351         return do_sys_truncate(path, length);
 352 }
 353 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 354 asmlinkage long SyS_truncate64(long path, loff_t length)
 355 {
 356         return SYSC_truncate64((const char __user *) path, length);
 357 }
 358 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 359 #endif
 360
 361 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 362 {
 363         long ret = do_sys_ftruncate(fd, length, 0);
 364         /* avoid REGPARM breakage on x86: */
 365         asmlinkage_protect(2, ret, fd, length);
 366         return ret;
 367 }
 368 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 369 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 370 {
 371         return SYSC_ftruncate64((unsigned int) fd, length);
 372 }
 373 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 374 #endif
 375 #endif /* BITS_PER_LONG == 32 */
 376
 377 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 378 {
 379         struct file *file;
 380         struct inode *inode;
 381         long ret = -EINVAL;
 382
 383         if (offset < 0 || len <= 0)
 384                 goto out;
 385
 386         /* Return error if mode is not supported */
 387         ret = -EOPNOTSUPP;
 388         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 389                 goto out;
 390
 391         ret = -EBADF;
 392         file = fget(fd);
 393         if (!file)
 394                 goto out;
 395         if (!(file->f_mode & FMODE_WRITE))
 396                 goto out_fput;
 397         /*
 398          * Revalidate the write permissions, in case security policy has
 399          * changed since the files were opened.
 400          */
 401         ret = security_file_permission(file, MAY_WRITE);
 402         if (ret)
 403                 goto out_fput;
 404
 405         inode = file->f_path.dentry->d_inode;
 406
 407         ret = -ESPIPE;
 408         if (S_ISFIFO(inode->i_mode))
 409                 goto out_fput;
 410
 411         ret = -ENODEV;
 412         /*
 413          * Let individual file system decide if it supports preallocation
 414          * for directories or not.
 415          */
 416         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 417                 goto out_fput;
 418
 419         ret = -EFBIG;
 420         /* Check for wrap through zero too */
 421         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 422                 goto out_fput;
 423
 424         if (inode->i_op && inode->i_op->fallocate)
 425                 ret = inode->i_op->fallocate(inode, mode, offset, len);
 426         else
 427                 ret = -EOPNOTSUPP;
 428
 429 out_fput:
 430         fput(file);
 431 out:
 432         return ret;
 433 }
 434 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 435 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 436 {
 437         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 438 }
 439 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 440 #endif
 441
 442 /*
 443  * access() needs to use the real uid/gid, not the effective uid/gid.
 444  * We do this by temporarily clearing all FS-related capabilities and
 445  * switching the fsuid/fsgid around to the real ones.
 446  */
 447 asmlinkage long sys_faccessat(int dfd, const char __user *filename, int mode)
 448 {
 449         struct path path;
 450         struct inode *inode;
 451         int old_fsuid, old_fsgid;
 452         kernel_cap_t uninitialized_var(old_cap);  /* !SECURE_NO_SETUID_FIXUP */
 453         int res;
 454
 455         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 456                 return -EINVAL;
 457
 458         old_fsuid = current->fsuid;
 459         old_fsgid = current->fsgid;
 460
 461         current->fsuid = current->uid;
 462         current->fsgid = current->gid;
 463
 464         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 465                 /*
 466                  * Clear the capabilities if we switch to a non-root user
 467                  */
 468 #ifndef CONFIG_SECURITY_FILE_CAPABILITIES
 469                 /*
 470                  * FIXME: There is a race here against sys_capset.  The
 471                  * capabilities can change yet we will restore the old
 472                  * value below.  We should hold task_capabilities_lock,
 473                  * but we cannot because user_path_at can sleep.
 474                  */
 475 #endif /* ndef CONFIG_SECURITY_FILE_CAPABILITIES */
 476                 if (current->uid)
 477                         old_cap = cap_set_effective(__cap_empty_set);
 478                 else
 479                         old_cap = cap_set_effective(current->cap_permitted);
 480         }
 481
 482         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 483         if (res)
 484                 goto out;
 485
 486         inode = path.dentry->d_inode;
 487
 488         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 489                 /*
 490                  * MAY_EXEC on regular files is denied if the fs is mounted
 491                  * with the "noexec" flag.
 492                  */
 493                 res = -EACCES;
 494                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 495                         goto out_path_release;
 496         }
 497
 498         res = inode_permission(inode, mode | MAY_ACCESS);
 499         /* SuS v2 requires we report a read only fs too */
 500         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 501                 goto out_path_release;
 502         /*
 503          * This is a rare case where using __mnt_is_readonly()
 504          * is OK without a mnt_want/drop_write() pair.  Since
 505          * no actual write to the fs is performed here, we do
 506          * not need to telegraph to that to anyone.
 507          *
 508          * By doing this, we accept that this access is
 509          * inherently racy and know that the fs may change
 510          * state before we even see this result.
 511          */
 512         if (__mnt_is_readonly(path.mnt))
 513                 res = -EROFS;
 514
 515 out_path_release:
 516         path_put(&path);
 517 out:
 518         current->fsuid = old_fsuid;
 519         current->fsgid = old_fsgid;
 520
 521         if (!issecure(SECURE_NO_SETUID_FIXUP))
 522                 cap_set_effective(old_cap);
 523
 524         return res;
 525 }
 526
 527 asmlinkage long sys_access(const char __user *filename, int mode)
 528 {
 529         return sys_faccessat(AT_FDCWD, filename, mode);
 530 }
 531
 532 asmlinkage long sys_chdir(const char __user * filename)
 533 {
 534         struct path path;
 535         int error;
 536
 537         error = user_path_dir(filename, &path);
 538         if (error)
 539                 goto out;
 540
 541         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 542         if (error)
 543                 goto dput_and_out;
 544
 545         set_fs_pwd(current->fs, &path);
 546
 547 dput_and_out:
 548         path_put(&path);
 549 out:
 550         return error;
 551 }
 552
 553 asmlinkage long sys_fchdir(unsigned int fd)
 554 {
 555         struct file *file;
 556         struct inode *inode;
 557         int error;
 558
 559         error = -EBADF;
 560         file = fget(fd);
 561         if (!file)
 562                 goto out;
 563
 564         inode = file->f_path.dentry->d_inode;
 565
 566         error = -ENOTDIR;
 567         if (!S_ISDIR(inode->i_mode))
 568                 goto out_putf;
 569
 570         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 571         if (!error)
 572                 set_fs_pwd(current->fs, &file->f_path);
 573 out_putf:
 574         fput(file);
 575 out:
 576         return error;
 577 }
 578
 579 asmlinkage long sys_chroot(const char __user * filename)
 580 {
 581         struct path path;
 582         int error;
 583
 584         error = user_path_dir(filename, &path);
 585         if (error)
 586                 goto out;
 587
 588         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 589         if (error)
 590                 goto dput_and_out;
 591
 592         error = -EPERM;
 593         if (!capable(CAP_SYS_CHROOT))
 594                 goto dput_and_out;
 595
 596         set_fs_root(current->fs, &path);
 597         error = 0;
 598 dput_and_out:
 599         path_put(&path);
 600 out:
 601         return error;
 602 }
 603
 604 asmlinkage long sys_fchmod(unsigned int fd, mode_t mode)
 605 {
 606         struct inode * inode;
 607         struct dentry * dentry;
 608         struct file * file;
 609         int err = -EBADF;
 610         struct iattr newattrs;
 611
 612         file = fget(fd);
 613         if (!file)
 614                 goto out;
 615
 616         dentry = file->f_path.dentry;
 617         inode = dentry->d_inode;
 618
 619         audit_inode(NULL, dentry);
 620
 621         err = mnt_want_write(file->f_path.mnt);
 622         if (err)
 623                 goto out_putf;
 624         mutex_lock(&inode->i_mutex);
 625         if (mode == (mode_t) -1)
 626                 mode = inode->i_mode;
 627         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 628         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 629         err = notify_change(dentry, &newattrs);
 630         mutex_unlock(&inode->i_mutex);
 631         mnt_drop_write(file->f_path.mnt);
 632 out_putf:
 633         fput(file);
 634 out:
 635         return err;
 636 }
 637
 638 asmlinkage long sys_fchmodat(int dfd, const char __user *filename,
 639                              mode_t mode)
 640 {
 641         struct path path;
 642         struct inode *inode;
 643         int error;
 644         struct iattr newattrs;
 645
 646         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 647         if (error)
 648                 goto out;
 649         inode = path.dentry->d_inode;
 650
 651         error = mnt_want_write(path.mnt);
 652         if (error)
 653                 goto dput_and_out;
 654         mutex_lock(&inode->i_mutex);
 655         if (mode == (mode_t) -1)
 656                 mode = inode->i_mode;
 657         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 658         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 659         error = notify_change(path.dentry, &newattrs);
 660         mutex_unlock(&inode->i_mutex);
 661         mnt_drop_write(path.mnt);
 662 dput_and_out:
 663         path_put(&path);
 664 out:
 665         return error;
 666 }
 667
 668 asmlinkage long sys_chmod(const char __user *filename, mode_t mode)
 669 {
 670         return sys_fchmodat(AT_FDCWD, filename, mode);
 671 }
 672
 673 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 674 {
 675         struct inode *inode = dentry->d_inode;
 676         int error;
 677         struct iattr newattrs;
 678
 679         newattrs.ia_valid =  ATTR_CTIME;
 680         if (user != (uid_t) -1) {
 681                 newattrs.ia_valid |= ATTR_UID;
 682                 newattrs.ia_uid = user;
 683         }
 684         if (group != (gid_t) -1) {
 685                 newattrs.ia_valid |= ATTR_GID;
 686                 newattrs.ia_gid = group;
 687         }
 688         if (!S_ISDIR(inode->i_mode))
 689                 newattrs.ia_valid |=
 690                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 691         mutex_lock(&inode->i_mutex);
 692         error = notify_change(dentry, &newattrs);
 693         mutex_unlock(&inode->i_mutex);
 694
 695         return error;
 696 }
 697
 698 asmlinkage long sys_chown(const char __user * filename, uid_t user, gid_t group)
 699 {
 700         struct path path;
 701         int error;
 702
 703         error = user_path(filename, &path);
 704         if (error)
 705                 goto out;
 706         error = mnt_want_write(path.mnt);
 707         if (error)
 708                 goto out_release;
 709         error = chown_common(path.dentry, user, group);
 710         mnt_drop_write(path.mnt);
 711 out_release:
 712         path_put(&path);
 713 out:
 714         return error;
 715 }
 716
 717 asmlinkage long sys_fchownat(int dfd, const char __user *filename, uid_t user,
 718                              gid_t group, int flag)
 719 {
 720         struct path path;
 721         int error = -EINVAL;
 722         int follow;
 723
 724         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 725                 goto out;
 726
 727         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 728         error = user_path_at(dfd, filename, follow, &path);
 729         if (error)
 730                 goto out;
 731         error = mnt_want_write(path.mnt);
 732         if (error)
 733                 goto out_release;
 734         error = chown_common(path.dentry, user, group);
 735         mnt_drop_write(path.mnt);
 736 out_release:
 737         path_put(&path);
 738 out:
 739         return error;
 740 }
 741
 742 asmlinkage long sys_lchown(const char __user * filename, uid_t user, gid_t group)
 743 {
 744         struct path path;
 745         int error;
 746
 747         error = user_lpath(filename, &path);
 748         if (error)
 749                 goto out;
 750         error = mnt_want_write(path.mnt);
 751         if (error)
 752                 goto out_release;
 753         error = chown_common(path.dentry, user, group);
 754         mnt_drop_write(path.mnt);
 755 out_release:
 756         path_put(&path);
 757 out:
 758         return error;
 759 }
 760
 761
 762 asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
 763 {
 764         struct file * file;
 765         int error = -EBADF;
 766         struct dentry * dentry;
 767
 768         file = fget(fd);
 769         if (!file)
 770                 goto out;
 771
 772         error = mnt_want_write(file->f_path.mnt);
 773         if (error)
 774                 goto out_fput;
 775         dentry = file->f_path.dentry;
 776         audit_inode(NULL, dentry);
 777         error = chown_common(dentry, user, group);
 778         mnt_drop_write(file->f_path.mnt);
 779 out_fput:
 780         fput(file);
 781 out:
 782         return error;
 783 }
 784
 785 /*
 786  * You have to be very careful that these write
 787  * counts get cleaned up in error cases and
 788  * upon __fput().  This should probably never
 789  * be called outside of __dentry_open().
 790  */
 791 static inline int __get_file_write_access(struct inode *inode,
 792                                           struct vfsmount *mnt)
 793 {
 794         int error;
 795         error = get_write_access(inode);
 796         if (error)
 797                 return error;
 798         /*
 799          * Do not take mount writer counts on
 800          * special files since no writes to
 801          * the mount itself will occur.
 802          */
 803         if (!special_file(inode->i_mode)) {
 804                 /*
 805                  * Balanced in __fput()
 806                  */
 807                 error = mnt_want_write(mnt);
 808                 if (error)
 809                         put_write_access(inode);
 810         }
 811         return error;
 812 }
 813
 814 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 815                                         int flags, struct file *f,
 816                                         int (*open)(struct inode *, struct file *))
 817 {
 818         struct inode *inode;
 819         int error;
 820
 821         f->f_flags = flags;
 822         f->f_mode = ((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 823                                 FMODE_PREAD | FMODE_PWRITE;
 824         inode = dentry->d_inode;
 825         if (f->f_mode & FMODE_WRITE) {
 826                 error = __get_file_write_access(inode, mnt);
 827                 if (error)
 828                         goto cleanup_file;
 829                 if (!special_file(inode->i_mode))
 830                         file_take_write(f);
 831         }
 832
 833         f->f_mapping = inode->i_mapping;
 834         f->f_path.dentry = dentry;
 835         f->f_path.mnt = mnt;
 836         f->f_pos = 0;
 837         f->f_op = fops_get(inode->i_fop);
 838         file_move(f, &inode->i_sb->s_files);
 839
 840         error = security_dentry_open(f);
 841         if (error)
 842                 goto cleanup_all;
 843
 844         if (!open && f->f_op)
 845                 open = f->f_op->open;
 846         if (open) {
 847                 error = open(inode, f);
 848                 if (error)
 849                         goto cleanup_all;
 850         }
 851
 852         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 853
 854         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 855
 856         /* NB: we're sure to have correct a_ops only after f_op->open */
 857         if (f->f_flags & O_DIRECT) {
 858                 if (!f->f_mapping->a_ops ||
 859                     ((!f->f_mapping->a_ops->direct_IO) &&
 860                     (!f->f_mapping->a_ops->get_xip_mem))) {
 861                         fput(f);
 862                         f = ERR_PTR(-EINVAL);
 863                 }
 864         }
 865
 866         return f;
 867
 868 cleanup_all:
 869         fops_put(f->f_op);
 870         if (f->f_mode & FMODE_WRITE) {
 871                 put_write_access(inode);
 872                 if (!special_file(inode->i_mode)) {
 873                         /*
 874                          * We don't consider this a real
 875                          * mnt_want/drop_write() pair
 876                          * because it all happenend right
 877                          * here, so just reset the state.
 878                          */
 879                         file_reset_write(f);
 880                         mnt_drop_write(mnt);
 881                 }
 882         }
 883         file_kill(f);
 884         f->f_path.dentry = NULL;
 885         f->f_path.mnt = NULL;
 886 cleanup_file:
 887         put_filp(f);
 888         dput(dentry);
 889         mntput(mnt);
 890         return ERR_PTR(error);
 891 }
 892
 893 /**
 894  * lookup_instantiate_filp - instantiates the open intent filp
 895  * @nd: pointer to nameidata
 896  * @dentry: pointer to dentry
 897  * @open: open callback
 898  *
 899  * Helper for filesystems that want to use lookup open intents and pass back
 900  * a fully instantiated struct file to the caller.
 901  * This function is meant to be called from within a filesystem's
 902  * lookup method.
 903  * Beware of calling it for non-regular files! Those ->open methods might block
 904  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 905  * leading to a deadlock, as nobody can open that fifo anymore, because
 906  * another process to open fifo will block on locked parent when doing lookup).
 907  * Note that in case of error, nd->intent.open.file is destroyed, but the
 908  * path information remains valid.
 909  * If the open callback is set to NULL, then the standard f_op->open()
 910  * filesystem callback is substituted.
 911  */
 912 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 913                 int (*open)(struct inode *, struct file *))
 914 {
 915         if (IS_ERR(nd->intent.open.file))
 916                 goto out;
 917         if (IS_ERR(dentry))
 918                 goto out_err;
 919         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 920                                              nd->intent.open.flags - 1,
 921                                              nd->intent.open.file,
 922                                              open);
 923 out:
 924         return nd->intent.open.file;
 925 out_err:
 926         release_open_intent(nd);
 927         nd->intent.open.file = (struct file *)dentry;
 928         goto out;
 929 }
 930 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 931
 932 /**
 933  * nameidata_to_filp - convert a nameidata to an open filp.
 934  * @nd: pointer to nameidata
 935  * @flags: open flags
 936  *
 937  * Note that this function destroys the original nameidata
 938  */
 939 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 940 {
 941         struct file *filp;
 942
 943         /* Pick up the filp from the open intent */
 944         filp = nd->intent.open.file;
 945         /* Has the filesystem initialised the file for us? */
 946         if (filp->f_path.dentry == NULL)
 947                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 948                                      NULL);
 949         else
 950                 path_put(&nd->path);
 951         return filp;
 952 }
 953
 954 /*
 955  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 956  * error.
 957  */
 958 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
 959 {
 960         int error;
 961         struct file *f;
 962
 963         /*
 964          * We must always pass in a valid mount pointer.   Historically
 965          * callers got away with not passing it, but we must enforce this at
 966          * the earliest possible point now to avoid strange problems deep in the
 967          * filesystem stack.
 968          */
 969         if (!mnt) {
 970                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 971                 dump_stack();
 972                 return ERR_PTR(-EINVAL);
 973         }
 974
 975         error = -ENFILE;
 976         f = get_empty_filp();
 977         if (f == NULL) {
 978                 dput(dentry);
 979                 mntput(mnt);
 980                 return ERR_PTR(error);
 981         }
 982
 983         return __dentry_open(dentry, mnt, flags, f, NULL);
 984 }
 985 EXPORT_SYMBOL(dentry_open);
 986
 987 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 988 {
 989         struct fdtable *fdt = files_fdtable(files);
 990         __FD_CLR(fd, fdt->open_fds);
 991         if (fd < files->next_fd)
 992                 files->next_fd = fd;
 993 }
 994
 995 void put_unused_fd(unsigned int fd)
 996 {
 997         struct files_struct *files = current->files;
 998         spin_lock(&files->file_lock);
 999         __put_unused_fd(files, fd);
1000         spin_unlock(&files->file_lock);
1001 }
1002
1003 EXPORT_SYMBOL(put_unused_fd);
1004
1005 /*
1006  * Install a file pointer in the fd array.
1007  *
1008  * The VFS is full of places where we drop the files lock between
1009  * setting the open_fds bitmap and installing the file in the file
1010  * array.  At any such point, we are vulnerable to a dup2() race
1011  * installing a file in the array before us.  We need to detect this and
1012  * fput() the struct file we are about to overwrite in this case.
1013  *
1014  * It should never happen - if we allow dup2() do it, _really_ bad things
1015  * will follow.
1016  */
1017
1018 void fd_install(unsigned int fd, struct file *file)
1019 {
1020         struct files_struct *files = current->files;
1021         struct fdtable *fdt;
1022         spin_lock(&files->file_lock);
1023         fdt = files_fdtable(files);
1024         BUG_ON(fdt->fd[fd] != NULL);
1025         rcu_assign_pointer(fdt->fd[fd], file);
1026         spin_unlock(&files->file_lock);
1027 }
1028
1029 EXPORT_SYMBOL(fd_install);
1030
1031 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1032 {
1033         char *tmp = getname(filename);
1034         int fd = PTR_ERR(tmp);
1035
1036         if (!IS_ERR(tmp)) {
1037                 fd = get_unused_fd_flags(flags);
1038                 if (fd >= 0) {
1039                         struct file *f = do_filp_open(dfd, tmp, flags, mode);
1040                         if (IS_ERR(f)) {
1041                                 put_unused_fd(fd);
1042                                 fd = PTR_ERR(f);
1043                         } else {
1044                                 fsnotify_open(f->f_path.dentry);
1045                                 fd_install(fd, f);
1046                         }
1047                 }
1048                 putname(tmp);
1049         }
1050         return fd;
1051 }
1052
1053 asmlinkage long sys_open(const char __user *filename, int flags, int mode)
1054 {
1055         long ret;
1056
1057         if (force_o_largefile())
1058                 flags |= O_LARGEFILE;
1059
1060         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1061         /* avoid REGPARM breakage on x86: */
1062         asmlinkage_protect(3, ret, filename, flags, mode);
1063         return ret;
1064 }
1065
1066 asmlinkage long sys_openat(int dfd, const char __user *filename, int flags,
1067                            int mode)
1068 {
1069         long ret;
1070
1071         if (force_o_largefile())
1072                 flags |= O_LARGEFILE;
1073
1074         ret = do_sys_open(dfd, filename, flags, mode);
1075         /* avoid REGPARM breakage on x86: */
1076         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1077         return ret;
1078 }
1079
1080 #ifndef __alpha__
1081
1082 /*
1083  * For backward compatibility?  Maybe this should be moved
1084  * into arch/i386 instead?
1085  */
1086 asmlinkage long sys_creat(const char __user * pathname, int mode)
1087 {
1088         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1089 }
1090
1091 #endif
1092
1093 /*
1094  * "id" is the POSIX thread ID. We use the
1095  * files pointer for this..
1096  */
1097 int filp_close(struct file *filp, fl_owner_t id)
1098 {
1099         int retval = 0;
1100
1101         if (!file_count(filp)) {
1102                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1103                 return 0;
1104         }
1105
1106         if (filp->f_op && filp->f_op->flush)
1107                 retval = filp->f_op->flush(filp, id);
1108
1109         dnotify_flush(filp, id);
1110         locks_remove_posix(filp, id);
1111         fput(filp);
1112         return retval;
1113 }
1114
1115 EXPORT_SYMBOL(filp_close);
1116
1117 /*
1118  * Careful here! We test whether the file pointer is NULL before
1119  * releasing the fd. This ensures that one clone task can't release
1120  * an fd while another clone is opening it.
1121  */
1122 asmlinkage long sys_close(unsigned int fd)
1123 {
1124         struct file * filp;
1125         struct files_struct *files = current->files;
1126         struct fdtable *fdt;
1127         int retval;
1128
1129         spin_lock(&files->file_lock);
1130         fdt = files_fdtable(files);
1131         if (fd >= fdt->max_fds)
1132                 goto out_unlock;
1133         filp = fdt->fd[fd];
1134         if (!filp)
1135                 goto out_unlock;
1136         rcu_assign_pointer(fdt->fd[fd], NULL);
1137         FD_CLR(fd, fdt->close_on_exec);
1138         __put_unused_fd(files, fd);
1139         spin_unlock(&files->file_lock);
1140         retval = filp_close(filp, files);
1141
1142         /* can't restart close syscall because file table entry was cleared */
1143         if (unlikely(retval == -ERESTARTSYS ||
1144                      retval == -ERESTARTNOINTR ||
1145                      retval == -ERESTARTNOHAND ||
1146                      retval == -ERESTART_RESTARTBLOCK))
1147                 retval = -EINTR;
1148
1149         return retval;
1150
1151 out_unlock:
1152         spin_unlock(&files->file_lock);
1153         return -EBADF;
1154 }
1155
1156 EXPORT_SYMBOL(sys_close);
1157
1158 /*
1159  * This routine simulates a hangup on the tty, to arrange that users
1160  * are given clean terminals at login time.
1161  */
1162 asmlinkage long sys_vhangup(void)
1163 {
1164         if (capable(CAP_SYS_TTY_CONFIG)) {
1165                 /* XXX: this needs locking */
1166                 tty_vhangup(current->signal->tty);
1167                 return 0;
1168         }
1169         return -EPERM;
1170 }
1171
1172 /*
1173  * Called when an inode is about to be open.
1174  * We use this to disallow opening large files on 32bit systems if
1175  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1176  * on this flag in sys_open.
1177  */
1178 int generic_file_open(struct inode * inode, struct file * filp)
1179 {
1180         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1181                 return -EOVERFLOW;
1182         return 0;
1183 }
1184
1185 EXPORT_SYMBOL(generic_file_open);
1186
1187 /*
1188  * This is used by subsystems that don't want seekable
1189  * file descriptors
1190  */
1191 int nonseekable_open(struct inode *inode, struct file *filp)
1192 {
1193         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1194         return 0;
1195 }
1196
1197 EXPORT_SYMBOL(nonseekable_open);