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 #include <linux/fs_struct.h>
  33
  34 int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  35 {
  36         int retval = -ENODEV;
  37
  38         if (dentry) {
  39                 retval = -ENOSYS;
  40                 if (dentry->d_sb->s_op->statfs) {
  41                         memset(buf, 0, sizeof(*buf));
  42                         retval = security_sb_statfs(dentry);
  43                         if (retval)
  44                                 return retval;
  45                         retval = dentry->d_sb->s_op->statfs(dentry, buf);
  46                         if (retval == 0 && buf->f_frsize == 0)
  47                                 buf->f_frsize = buf->f_bsize;
  48                 }
  49         }
  50         return retval;
  51 }
  52
  53 EXPORT_SYMBOL(vfs_statfs);
  54
  55 static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
  56 {
  57         struct kstatfs st;
  58         int retval;
  59
  60         retval = vfs_statfs(dentry, &st);
  61         if (retval)
  62                 return retval;
  63
  64         if (sizeof(*buf) == sizeof(st))
  65                 memcpy(buf, &st, sizeof(st));
  66         else {
  67                 if (sizeof buf->f_blocks == 4) {
  68                         if ((st.f_blocks | st.f_bfree | st.f_bavail |
  69                              st.f_bsize | st.f_frsize) &
  70                             0xffffffff00000000ULL)
  71                                 return -EOVERFLOW;
  72                         /*
  73                          * f_files and f_ffree may be -1; it's okay to stuff
  74                          * that into 32 bits
  75                          */
  76                         if (st.f_files != -1 &&
  77                             (st.f_files & 0xffffffff00000000ULL))
  78                                 return -EOVERFLOW;
  79                         if (st.f_ffree != -1 &&
  80                             (st.f_ffree & 0xffffffff00000000ULL))
  81                                 return -EOVERFLOW;
  82                 }
  83
  84                 buf->f_type = st.f_type;
  85                 buf->f_bsize = st.f_bsize;
  86                 buf->f_blocks = st.f_blocks;
  87                 buf->f_bfree = st.f_bfree;
  88                 buf->f_bavail = st.f_bavail;
  89                 buf->f_files = st.f_files;
  90                 buf->f_ffree = st.f_ffree;
  91                 buf->f_fsid = st.f_fsid;
  92                 buf->f_namelen = st.f_namelen;
  93                 buf->f_frsize = st.f_frsize;
  94                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
  95         }
  96         return 0;
  97 }
  98
  99 static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
 100 {
 101         struct kstatfs st;
 102         int retval;
 103
 104         retval = vfs_statfs(dentry, &st);
 105         if (retval)
 106                 return retval;
 107
 108         if (sizeof(*buf) == sizeof(st))
 109                 memcpy(buf, &st, sizeof(st));
 110         else {
 111                 buf->f_type = st.f_type;
 112                 buf->f_bsize = st.f_bsize;
 113                 buf->f_blocks = st.f_blocks;
 114                 buf->f_bfree = st.f_bfree;
 115                 buf->f_bavail = st.f_bavail;
 116                 buf->f_files = st.f_files;
 117                 buf->f_ffree = st.f_ffree;
 118                 buf->f_fsid = st.f_fsid;
 119                 buf->f_namelen = st.f_namelen;
 120                 buf->f_frsize = st.f_frsize;
 121                 memset(buf->f_spare, 0, sizeof(buf->f_spare));
 122         }
 123         return 0;
 124 }
 125
 126 SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct statfs __user *, buf)
 127 {
 128         struct path path;
 129         int error;
 130
 131         error = user_path(pathname, &path);
 132         if (!error) {
 133                 struct statfs tmp;
 134                 error = vfs_statfs_native(path.dentry, &tmp);
 135                 if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 136                         error = -EFAULT;
 137                 path_put(&path);
 138         }
 139         return error;
 140 }
 141
 142 SYSCALL_DEFINE3(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 SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct statfs __user *, buf)
 161 {
 162         struct file * file;
 163         struct statfs tmp;
 164         int error;
 165
 166         error = -EBADF;
 167         file = fget(fd);
 168         if (!file)
 169                 goto out;
 170         error = vfs_statfs_native(file->f_path.dentry, &tmp);
 171         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 172                 error = -EFAULT;
 173         fput(file);
 174 out:
 175         return error;
 176 }
 177
 178 SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, size_t, sz, struct statfs64 __user *, buf)
 179 {
 180         struct file * file;
 181         struct statfs64 tmp;
 182         int error;
 183
 184         if (sz != sizeof(*buf))
 185                 return -EINVAL;
 186
 187         error = -EBADF;
 188         file = fget(fd);
 189         if (!file)
 190                 goto out;
 191         error = vfs_statfs64(file->f_path.dentry, &tmp);
 192         if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
 193                 error = -EFAULT;
 194         fput(file);
 195 out:
 196         return error;
 197 }
 198
 199 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 200         struct file *filp)
 201 {
 202         int err;
 203         struct iattr newattrs;
 204
 205         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
 206         if (length < 0)
 207                 return -EINVAL;
 208
 209         newattrs.ia_size = length;
 210         newattrs.ia_valid = ATTR_SIZE | time_attrs;
 211         if (filp) {
 212                 newattrs.ia_file = filp;
 213                 newattrs.ia_valid |= ATTR_FILE;
 214         }
 215
 216         /* Remove suid/sgid on truncate too */
 217         newattrs.ia_valid |= should_remove_suid(dentry);
 218
 219         mutex_lock(&dentry->d_inode->i_mutex);
 220         err = notify_change(dentry, &newattrs);
 221         mutex_unlock(&dentry->d_inode->i_mutex);
 222         return err;
 223 }
 224
 225 static long do_sys_truncate(const char __user *pathname, loff_t length)
 226 {
 227         struct path path;
 228         struct inode *inode;
 229         int error;
 230
 231         error = -EINVAL;
 232         if (length < 0) /* sorry, but loff_t says... */
 233                 goto out;
 234
 235         error = user_path(pathname, &path);
 236         if (error)
 237                 goto out;
 238         inode = path.dentry->d_inode;
 239
 240         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
 241         error = -EISDIR;
 242         if (S_ISDIR(inode->i_mode))
 243                 goto dput_and_out;
 244
 245         error = -EINVAL;
 246         if (!S_ISREG(inode->i_mode))
 247                 goto dput_and_out;
 248
 249         error = mnt_want_write(path.mnt);
 250         if (error)
 251                 goto dput_and_out;
 252
 253         error = inode_permission(inode, MAY_WRITE);
 254         if (error)
 255                 goto mnt_drop_write_and_out;
 256
 257         error = -EPERM;
 258         if (IS_APPEND(inode))
 259                 goto mnt_drop_write_and_out;
 260
 261         error = get_write_access(inode);
 262         if (error)
 263                 goto mnt_drop_write_and_out;
 264
 265         /*
 266          * Make sure that there are no leases.  get_write_access() protects
 267          * against the truncate racing with a lease-granting setlease().
 268          */
 269         error = break_lease(inode, FMODE_WRITE);
 270         if (error)
 271                 goto put_write_and_out;
 272
 273         error = locks_verify_truncate(inode, NULL, length);
 274         if (!error)
 275                 error = security_path_truncate(&path, length, 0);
 276         if (!error) {
 277                 vfs_dq_init(inode);
 278                 error = do_truncate(path.dentry, length, 0, NULL);
 279         }
 280
 281 put_write_and_out:
 282         put_write_access(inode);
 283 mnt_drop_write_and_out:
 284         mnt_drop_write(path.mnt);
 285 dput_and_out:
 286         path_put(&path);
 287 out:
 288         return error;
 289 }
 290
 291 SYSCALL_DEFINE2(truncate, const char __user *, path, unsigned long, length)
 292 {
 293         /* on 32-bit boxen it will cut the range 2^31--2^32-1 off */
 294         return do_sys_truncate(path, (long)length);
 295 }
 296
 297 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 298 {
 299         struct inode * inode;
 300         struct dentry *dentry;
 301         struct file * file;
 302         int error;
 303
 304         error = -EINVAL;
 305         if (length < 0)
 306                 goto out;
 307         error = -EBADF;
 308         file = fget(fd);
 309         if (!file)
 310                 goto out;
 311
 312         /* explicitly opened as large or we are on 64-bit box */
 313         if (file->f_flags & O_LARGEFILE)
 314                 small = 0;
 315
 316         dentry = file->f_path.dentry;
 317         inode = dentry->d_inode;
 318         error = -EINVAL;
 319         if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 320                 goto out_putf;
 321
 322         error = -EINVAL;
 323         /* Cannot ftruncate over 2^31 bytes without large file support */
 324         if (small && length > MAX_NON_LFS)
 325                 goto out_putf;
 326
 327         error = -EPERM;
 328         if (IS_APPEND(inode))
 329                 goto out_putf;
 330
 331         error = locks_verify_truncate(inode, file, length);
 332         if (!error)
 333                 error = security_path_truncate(&file->f_path, length,
 334                                                ATTR_MTIME|ATTR_CTIME);
 335         if (!error)
 336                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
 337 out_putf:
 338         fput(file);
 339 out:
 340         return error;
 341 }
 342
 343 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 344 {
 345         long ret = do_sys_ftruncate(fd, length, 1);
 346         /* avoid REGPARM breakage on x86: */
 347         asmlinkage_protect(2, ret, fd, length);
 348         return ret;
 349 }
 350
 351 /* LFS versions of truncate are only needed on 32 bit machines */
 352 #if BITS_PER_LONG == 32
 353 SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
 354 {
 355         return do_sys_truncate(path, length);
 356 }
 357 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 358 asmlinkage long SyS_truncate64(long path, loff_t length)
 359 {
 360         return SYSC_truncate64((const char __user *) path, length);
 361 }
 362 SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
 363 #endif
 364
 365 SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
 366 {
 367         long ret = do_sys_ftruncate(fd, length, 0);
 368         /* avoid REGPARM breakage on x86: */
 369         asmlinkage_protect(2, ret, fd, length);
 370         return ret;
 371 }
 372 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 373 asmlinkage long SyS_ftruncate64(long fd, loff_t length)
 374 {
 375         return SYSC_ftruncate64((unsigned int) fd, length);
 376 }
 377 SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
 378 #endif
 379 #endif /* BITS_PER_LONG == 32 */
 380
 381
 382 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 383 {
 384         struct inode *inode = file->f_path.dentry->d_inode;
 385         long ret;
 386
 387         if (offset < 0 || len <= 0)
 388                 return -EINVAL;
 389
 390         /* Return error if mode is not supported */
 391         if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
 392                 return -EOPNOTSUPP;
 393
 394         if (!(file->f_mode & FMODE_WRITE))
 395                 return -EBADF;
 396         /*
 397          * Revalidate the write permissions, in case security policy has
 398          * changed since the files were opened.
 399          */
 400         ret = security_file_permission(file, MAY_WRITE);
 401         if (ret)
 402                 return ret;
 403
 404         if (S_ISFIFO(inode->i_mode))
 405                 return -ESPIPE;
 406
 407         /*
 408          * Let individual file system decide if it supports preallocation
 409          * for directories or not.
 410          */
 411         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 412                 return -ENODEV;
 413
 414         /* Check for wrap through zero too */
 415         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 416                 return -EFBIG;
 417
 418         if (!inode->i_op->fallocate)
 419                 return -EOPNOTSUPP;
 420
 421         return inode->i_op->fallocate(inode, mode, offset, len);
 422 }
 423
 424 SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
 425 {
 426         struct file *file;
 427         int error = -EBADF;
 428
 429         file = fget(fd);
 430         if (file) {
 431                 error = do_fallocate(file, mode, offset, len);
 432                 fput(file);
 433         }
 434
 435         return error;
 436 }
 437
 438 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 439 asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
 440 {
 441         return SYSC_fallocate((int)fd, (int)mode, offset, len);
 442 }
 443 SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
 444 #endif
 445
 446 /*
 447  * access() needs to use the real uid/gid, not the effective uid/gid.
 448  * We do this by temporarily clearing all FS-related capabilities and
 449  * switching the fsuid/fsgid around to the real ones.
 450  */
 451 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 452 {
 453         const struct cred *old_cred;
 454         struct cred *override_cred;
 455         struct path path;
 456         struct inode *inode;
 457         int res;
 458
 459         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 460                 return -EINVAL;
 461
 462         override_cred = prepare_creds();
 463         if (!override_cred)
 464                 return -ENOMEM;
 465
 466         override_cred->fsuid = override_cred->uid;
 467         override_cred->fsgid = override_cred->gid;
 468
 469         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 470                 /* Clear the capabilities if we switch to a non-root user */
 471                 if (override_cred->uid)
 472                         cap_clear(override_cred->cap_effective);
 473                 else
 474                         override_cred->cap_effective =
 475                                 override_cred->cap_permitted;
 476         }
 477
 478         old_cred = override_creds(override_cred);
 479
 480         res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 481         if (res)
 482                 goto out;
 483
 484         inode = path.dentry->d_inode;
 485
 486         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 487                 /*
 488                  * MAY_EXEC on regular files is denied if the fs is mounted
 489                  * with the "noexec" flag.
 490                  */
 491                 res = -EACCES;
 492                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 493                         goto out_path_release;
 494         }
 495
 496         res = inode_permission(inode, mode | MAY_ACCESS);
 497         /* SuS v2 requires we report a read only fs too */
 498         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 499                 goto out_path_release;
 500         /*
 501          * This is a rare case where using __mnt_is_readonly()
 502          * is OK without a mnt_want/drop_write() pair.  Since
 503          * no actual write to the fs is performed here, we do
 504          * not need to telegraph to that to anyone.
 505          *
 506          * By doing this, we accept that this access is
 507          * inherently racy and know that the fs may change
 508          * state before we even see this result.
 509          */
 510         if (__mnt_is_readonly(path.mnt))
 511                 res = -EROFS;
 512
 513 out_path_release:
 514         path_put(&path);
 515 out:
 516         revert_creds(old_cred);
 517         put_cred(override_cred);
 518         return res;
 519 }
 520
 521 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 522 {
 523         return sys_faccessat(AT_FDCWD, filename, mode);
 524 }
 525
 526 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 527 {
 528         struct path path;
 529         int error;
 530
 531         error = user_path_dir(filename, &path);
 532         if (error)
 533                 goto out;
 534
 535         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 536         if (error)
 537                 goto dput_and_out;
 538
 539         set_fs_pwd(current->fs, &path);
 540
 541 dput_and_out:
 542         path_put(&path);
 543 out:
 544         return error;
 545 }
 546
 547 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 548 {
 549         struct file *file;
 550         struct inode *inode;
 551         int error;
 552
 553         error = -EBADF;
 554         file = fget(fd);
 555         if (!file)
 556                 goto out;
 557
 558         inode = file->f_path.dentry->d_inode;
 559
 560         error = -ENOTDIR;
 561         if (!S_ISDIR(inode->i_mode))
 562                 goto out_putf;
 563
 564         error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
 565         if (!error)
 566                 set_fs_pwd(current->fs, &file->f_path);
 567 out_putf:
 568         fput(file);
 569 out:
 570         return error;
 571 }
 572
 573 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 574 {
 575         struct path path;
 576         int error;
 577
 578         error = user_path_dir(filename, &path);
 579         if (error)
 580                 goto out;
 581
 582         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
 583         if (error)
 584                 goto dput_and_out;
 585
 586         error = -EPERM;
 587         if (!capable(CAP_SYS_CHROOT))
 588                 goto dput_and_out;
 589
 590         set_fs_root(current->fs, &path);
 591         error = 0;
 592 dput_and_out:
 593         path_put(&path);
 594 out:
 595         return error;
 596 }
 597
 598 SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
 599 {
 600         struct inode * inode;
 601         struct dentry * dentry;
 602         struct file * file;
 603         int err = -EBADF;
 604         struct iattr newattrs;
 605
 606         file = fget(fd);
 607         if (!file)
 608                 goto out;
 609
 610         dentry = file->f_path.dentry;
 611         inode = dentry->d_inode;
 612
 613         audit_inode(NULL, dentry);
 614
 615         err = mnt_want_write_file(file);
 616         if (err)
 617                 goto out_putf;
 618         mutex_lock(&inode->i_mutex);
 619         if (mode == (mode_t) -1)
 620                 mode = inode->i_mode;
 621         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 622         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 623         err = notify_change(dentry, &newattrs);
 624         mutex_unlock(&inode->i_mutex);
 625         mnt_drop_write(file->f_path.mnt);
 626 out_putf:
 627         fput(file);
 628 out:
 629         return err;
 630 }
 631
 632 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
 633 {
 634         struct path path;
 635         struct inode *inode;
 636         int error;
 637         struct iattr newattrs;
 638
 639         error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
 640         if (error)
 641                 goto out;
 642         inode = path.dentry->d_inode;
 643
 644         error = mnt_want_write(path.mnt);
 645         if (error)
 646                 goto dput_and_out;
 647         mutex_lock(&inode->i_mutex);
 648         if (mode == (mode_t) -1)
 649                 mode = inode->i_mode;
 650         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 651         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 652         error = notify_change(path.dentry, &newattrs);
 653         mutex_unlock(&inode->i_mutex);
 654         mnt_drop_write(path.mnt);
 655 dput_and_out:
 656         path_put(&path);
 657 out:
 658         return error;
 659 }
 660
 661 SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
 662 {
 663         return sys_fchmodat(AT_FDCWD, filename, mode);
 664 }
 665
 666 static int chown_common(struct dentry * dentry, uid_t user, gid_t group)
 667 {
 668         struct inode *inode = dentry->d_inode;
 669         int error;
 670         struct iattr newattrs;
 671
 672         newattrs.ia_valid =  ATTR_CTIME;
 673         if (user != (uid_t) -1) {
 674                 newattrs.ia_valid |= ATTR_UID;
 675                 newattrs.ia_uid = user;
 676         }
 677         if (group != (gid_t) -1) {
 678                 newattrs.ia_valid |= ATTR_GID;
 679                 newattrs.ia_gid = group;
 680         }
 681         if (!S_ISDIR(inode->i_mode))
 682                 newattrs.ia_valid |=
 683                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 684         mutex_lock(&inode->i_mutex);
 685         error = notify_change(dentry, &newattrs);
 686         mutex_unlock(&inode->i_mutex);
 687
 688         return error;
 689 }
 690
 691 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 692 {
 693         struct path path;
 694         int error;
 695
 696         error = user_path(filename, &path);
 697         if (error)
 698                 goto out;
 699         error = mnt_want_write(path.mnt);
 700         if (error)
 701                 goto out_release;
 702         error = chown_common(path.dentry, user, group);
 703         mnt_drop_write(path.mnt);
 704 out_release:
 705         path_put(&path);
 706 out:
 707         return error;
 708 }
 709
 710 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 711                 gid_t, group, int, flag)
 712 {
 713         struct path path;
 714         int error = -EINVAL;
 715         int follow;
 716
 717         if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
 718                 goto out;
 719
 720         follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 721         error = user_path_at(dfd, filename, follow, &path);
 722         if (error)
 723                 goto out;
 724         error = mnt_want_write(path.mnt);
 725         if (error)
 726                 goto out_release;
 727         error = chown_common(path.dentry, user, group);
 728         mnt_drop_write(path.mnt);
 729 out_release:
 730         path_put(&path);
 731 out:
 732         return error;
 733 }
 734
 735 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 736 {
 737         struct path path;
 738         int error;
 739
 740         error = user_lpath(filename, &path);
 741         if (error)
 742                 goto out;
 743         error = mnt_want_write(path.mnt);
 744         if (error)
 745                 goto out_release;
 746         error = chown_common(path.dentry, user, group);
 747         mnt_drop_write(path.mnt);
 748 out_release:
 749         path_put(&path);
 750 out:
 751         return error;
 752 }
 753
 754 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 755 {
 756         struct file * file;
 757         int error = -EBADF;
 758         struct dentry * dentry;
 759
 760         file = fget(fd);
 761         if (!file)
 762                 goto out;
 763
 764         error = mnt_want_write_file(file);
 765         if (error)
 766                 goto out_fput;
 767         dentry = file->f_path.dentry;
 768         audit_inode(NULL, dentry);
 769         error = chown_common(dentry, user, group);
 770         mnt_drop_write(file->f_path.mnt);
 771 out_fput:
 772         fput(file);
 773 out:
 774         return error;
 775 }
 776
 777 /*
 778  * You have to be very careful that these write
 779  * counts get cleaned up in error cases and
 780  * upon __fput().  This should probably never
 781  * be called outside of __dentry_open().
 782  */
 783 static inline int __get_file_write_access(struct inode *inode,
 784                                           struct vfsmount *mnt)
 785 {
 786         int error;
 787         error = get_write_access(inode);
 788         if (error)
 789                 return error;
 790         /*
 791          * Do not take mount writer counts on
 792          * special files since no writes to
 793          * the mount itself will occur.
 794          */
 795         if (!special_file(inode->i_mode)) {
 796                 /*
 797                  * Balanced in __fput()
 798                  */
 799                 error = mnt_want_write(mnt);
 800                 if (error)
 801                         put_write_access(inode);
 802         }
 803         return error;
 804 }
 805
 806 static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
 807                                         int flags, struct file *f,
 808                                         int (*open)(struct inode *, struct file *),
 809                                         const struct cred *cred)
 810 {
 811         struct inode *inode;
 812         int error;
 813
 814         f->f_flags = flags;
 815         f->f_mode = (__force fmode_t)((flags+1) & O_ACCMODE) | FMODE_LSEEK |
 816                                 FMODE_PREAD | FMODE_PWRITE;
 817         inode = dentry->d_inode;
 818         if (f->f_mode & FMODE_WRITE) {
 819                 error = __get_file_write_access(inode, mnt);
 820                 if (error)
 821                         goto cleanup_file;
 822                 if (!special_file(inode->i_mode))
 823                         file_take_write(f);
 824         }
 825
 826         f->f_mapping = inode->i_mapping;
 827         f->f_path.dentry = dentry;
 828         f->f_path.mnt = mnt;
 829         f->f_pos = 0;
 830         f->f_op = fops_get(inode->i_fop);
 831         file_move(f, &inode->i_sb->s_files);
 832
 833         error = security_dentry_open(f, cred);
 834         if (error)
 835                 goto cleanup_all;
 836
 837         if (!open && f->f_op)
 838                 open = f->f_op->open;
 839         if (open) {
 840                 error = open(inode, f);
 841                 if (error)
 842                         goto cleanup_all;
 843         }
 844
 845         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 846
 847         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 848
 849         /* NB: we're sure to have correct a_ops only after f_op->open */
 850         if (f->f_flags & O_DIRECT) {
 851                 if (!f->f_mapping->a_ops ||
 852                     ((!f->f_mapping->a_ops->direct_IO) &&
 853                     (!f->f_mapping->a_ops->get_xip_mem))) {
 854                         fput(f);
 855                         f = ERR_PTR(-EINVAL);
 856                 }
 857         }
 858
 859         return f;
 860
 861 cleanup_all:
 862         fops_put(f->f_op);
 863         if (f->f_mode & FMODE_WRITE) {
 864                 put_write_access(inode);
 865                 if (!special_file(inode->i_mode)) {
 866                         /*
 867                          * We don't consider this a real
 868                          * mnt_want/drop_write() pair
 869                          * because it all happenend right
 870                          * here, so just reset the state.
 871                          */
 872                         file_reset_write(f);
 873                         mnt_drop_write(mnt);
 874                 }
 875         }
 876         file_kill(f);
 877         f->f_path.dentry = NULL;
 878         f->f_path.mnt = NULL;
 879 cleanup_file:
 880         put_filp(f);
 881         dput(dentry);
 882         mntput(mnt);
 883         return ERR_PTR(error);
 884 }
 885
 886 /**
 887  * lookup_instantiate_filp - instantiates the open intent filp
 888  * @nd: pointer to nameidata
 889  * @dentry: pointer to dentry
 890  * @open: open callback
 891  *
 892  * Helper for filesystems that want to use lookup open intents and pass back
 893  * a fully instantiated struct file to the caller.
 894  * This function is meant to be called from within a filesystem's
 895  * lookup method.
 896  * Beware of calling it for non-regular files! Those ->open methods might block
 897  * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
 898  * leading to a deadlock, as nobody can open that fifo anymore, because
 899  * another process to open fifo will block on locked parent when doing lookup).
 900  * Note that in case of error, nd->intent.open.file is destroyed, but the
 901  * path information remains valid.
 902  * If the open callback is set to NULL, then the standard f_op->open()
 903  * filesystem callback is substituted.
 904  */
 905 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
 906                 int (*open)(struct inode *, struct file *))
 907 {
 908         const struct cred *cred = current_cred();
 909
 910         if (IS_ERR(nd->intent.open.file))
 911                 goto out;
 912         if (IS_ERR(dentry))
 913                 goto out_err;
 914         nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
 915                                              nd->intent.open.flags - 1,
 916                                              nd->intent.open.file,
 917                                              open, cred);
 918 out:
 919         return nd->intent.open.file;
 920 out_err:
 921         release_open_intent(nd);
 922         nd->intent.open.file = (struct file *)dentry;
 923         goto out;
 924 }
 925 EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
 926
 927 /**
 928  * nameidata_to_filp - convert a nameidata to an open filp.
 929  * @nd: pointer to nameidata
 930  * @flags: open flags
 931  *
 932  * Note that this function destroys the original nameidata
 933  */
 934 struct file *nameidata_to_filp(struct nameidata *nd, int flags)
 935 {
 936         const struct cred *cred = current_cred();
 937         struct file *filp;
 938
 939         /* Pick up the filp from the open intent */
 940         filp = nd->intent.open.file;
 941         /* Has the filesystem initialised the file for us? */
 942         if (filp->f_path.dentry == NULL)
 943                 filp = __dentry_open(nd->path.dentry, nd->path.mnt, flags, filp,
 944                                      NULL, cred);
 945         else
 946                 path_put(&nd->path);
 947         return filp;
 948 }
 949
 950 /*
 951  * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
 952  * error.
 953  */
 954 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 955                          const struct cred *cred)
 956 {
 957         int error;
 958         struct file *f;
 959
 960         /*
 961          * We must always pass in a valid mount pointer.   Historically
 962          * callers got away with not passing it, but we must enforce this at
 963          * the earliest possible point now to avoid strange problems deep in the
 964          * filesystem stack.
 965          */
 966         if (!mnt) {
 967                 printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
 968                 dump_stack();
 969                 return ERR_PTR(-EINVAL);
 970         }
 971
 972         error = -ENFILE;
 973         f = get_empty_filp();
 974         if (f == NULL) {
 975                 dput(dentry);
 976                 mntput(mnt);
 977                 return ERR_PTR(error);
 978         }
 979
 980         return __dentry_open(dentry, mnt, flags, f, NULL, cred);
 981 }
 982 EXPORT_SYMBOL(dentry_open);
 983
 984 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 985 {
 986         struct fdtable *fdt = files_fdtable(files);
 987         __FD_CLR(fd, fdt->open_fds);
 988         if (fd < files->next_fd)
 989                 files->next_fd = fd;
 990 }
 991
 992 void put_unused_fd(unsigned int fd)
 993 {
 994         struct files_struct *files = current->files;
 995         spin_lock(&files->file_lock);
 996         __put_unused_fd(files, fd);
 997         spin_unlock(&files->file_lock);
 998 }
 999
1000 EXPORT_SYMBOL(put_unused_fd);
1001
1002 /*
1003  * Install a file pointer in the fd array.
1004  *
1005  * The VFS is full of places where we drop the files lock between
1006  * setting the open_fds bitmap and installing the file in the file
1007  * array.  At any such point, we are vulnerable to a dup2() race
1008  * installing a file in the array before us.  We need to detect this and
1009  * fput() the struct file we are about to overwrite in this case.
1010  *
1011  * It should never happen - if we allow dup2() do it, _really_ bad things
1012  * will follow.
1013  */
1014
1015 void fd_install(unsigned int fd, struct file *file)
1016 {
1017         struct files_struct *files = current->files;
1018         struct fdtable *fdt;
1019         spin_lock(&files->file_lock);
1020         fdt = files_fdtable(files);
1021         BUG_ON(fdt->fd[fd] != NULL);
1022         rcu_assign_pointer(fdt->fd[fd], file);
1023         spin_unlock(&files->file_lock);
1024 }
1025
1026 EXPORT_SYMBOL(fd_install);
1027
1028 long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1029 {
1030         char *tmp = getname(filename);
1031         int fd = PTR_ERR(tmp);
1032
1033         if (!IS_ERR(tmp)) {
1034                 fd = get_unused_fd_flags(flags);
1035                 if (fd >= 0) {
1036                         struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
1037                         if (IS_ERR(f)) {
1038                                 put_unused_fd(fd);
1039                                 fd = PTR_ERR(f);
1040                         } else {
1041                                 fsnotify_open(f->f_path.dentry);
1042                                 fd_install(fd, f);
1043                         }
1044                 }
1045                 putname(tmp);
1046         }
1047         return fd;
1048 }
1049
1050 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1051 {
1052         long ret;
1053
1054         if (force_o_largefile())
1055                 flags |= O_LARGEFILE;
1056
1057         ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1058         /* avoid REGPARM breakage on x86: */
1059         asmlinkage_protect(3, ret, filename, flags, mode);
1060         return ret;
1061 }
1062
1063 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1064                 int, mode)
1065 {
1066         long ret;
1067
1068         if (force_o_largefile())
1069                 flags |= O_LARGEFILE;
1070
1071         ret = do_sys_open(dfd, filename, flags, mode);
1072         /* avoid REGPARM breakage on x86: */
1073         asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1074         return ret;
1075 }
1076
1077 #ifndef __alpha__
1078
1079 /*
1080  * For backward compatibility?  Maybe this should be moved
1081  * into arch/i386 instead?
1082  */
1083 SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1084 {
1085         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1086 }
1087
1088 #endif
1089
1090 /*
1091  * "id" is the POSIX thread ID. We use the
1092  * files pointer for this..
1093  */
1094 int filp_close(struct file *filp, fl_owner_t id)
1095 {
1096         int retval = 0;
1097
1098         if (!file_count(filp)) {
1099                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1100                 return 0;
1101         }
1102
1103         if (filp->f_op && filp->f_op->flush)
1104                 retval = filp->f_op->flush(filp, id);
1105
1106         dnotify_flush(filp, id);
1107         locks_remove_posix(filp, id);
1108         fput(filp);
1109         return retval;
1110 }
1111
1112 EXPORT_SYMBOL(filp_close);
1113
1114 /*
1115  * Careful here! We test whether the file pointer is NULL before
1116  * releasing the fd. This ensures that one clone task can't release
1117  * an fd while another clone is opening it.
1118  */
1119 SYSCALL_DEFINE1(close, unsigned int, fd)
1120 {
1121         struct file * filp;
1122         struct files_struct *files = current->files;
1123         struct fdtable *fdt;
1124         int retval;
1125
1126         spin_lock(&files->file_lock);
1127         fdt = files_fdtable(files);
1128         if (fd >= fdt->max_fds)
1129                 goto out_unlock;
1130         filp = fdt->fd[fd];
1131         if (!filp)
1132                 goto out_unlock;
1133         rcu_assign_pointer(fdt->fd[fd], NULL);
1134         FD_CLR(fd, fdt->close_on_exec);
1135         __put_unused_fd(files, fd);
1136         spin_unlock(&files->file_lock);
1137         retval = filp_close(filp, files);
1138
1139         /* can't restart close syscall because file table entry was cleared */
1140         if (unlikely(retval == -ERESTARTSYS ||
1141                      retval == -ERESTARTNOINTR ||
1142                      retval == -ERESTARTNOHAND ||
1143                      retval == -ERESTART_RESTARTBLOCK))
1144                 retval = -EINTR;
1145
1146         return retval;
1147
1148 out_unlock:
1149         spin_unlock(&files->file_lock);
1150         return -EBADF;
1151 }
1152 EXPORT_SYMBOL(sys_close);
1153
1154 /*
1155  * This routine simulates a hangup on the tty, to arrange that users
1156  * are given clean terminals at login time.
1157  */
1158 SYSCALL_DEFINE0(vhangup)
1159 {
1160         if (capable(CAP_SYS_TTY_CONFIG)) {
1161                 tty_vhangup_self();
1162                 return 0;
1163         }
1164         return -EPERM;
1165 }
1166
1167 /*
1168  * Called when an inode is about to be open.
1169  * We use this to disallow opening large files on 32bit systems if
1170  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1171  * on this flag in sys_open.
1172  */
1173 int generic_file_open(struct inode * inode, struct file * filp)
1174 {
1175         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1176                 return -EOVERFLOW;
1177         return 0;
1178 }
1179
1180 EXPORT_SYMBOL(generic_file_open);
1181
1182 /*
1183  * This is used by subsystems that don't want seekable
1184  * file descriptors
1185  */
1186 int nonseekable_open(struct inode *inode, struct file *filp)
1187 {
1188         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1189         return 0;
1190 }
1191
1192 EXPORT_SYMBOL(nonseekable_open);