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/fsnotify.h>
  12 #include <linux/module.h>
  13 #include <linux/tty.h>
  14 #include <linux/namei.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/capability.h>
  17 #include <linux/securebits.h>
  18 #include <linux/security.h>
  19 #include <linux/mount.h>
  20 #include <linux/fcntl.h>
  21 #include <linux/slab.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/fs.h>
  24 #include <linux/personality.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/rcupdate.h>
  28 #include <linux/audit.h>
  29 #include <linux/falloc.h>
  30 #include <linux/fs_struct.h>
  31 #include <linux/ima.h>
  32 #include <linux/dnotify.h>
  33 #include <linux/compat.h>
  34
  35 #include "internal.h"
  36
  37 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  38         struct file *filp)
  39 {
  40         int ret;
  41         struct iattr newattrs;
  42
  43         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  44         if (length < 0)
  45                 return -EINVAL;
  46
  47         newattrs.ia_size = length;
  48         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  49         if (filp) {
  50                 newattrs.ia_file = filp;
  51                 newattrs.ia_valid |= ATTR_FILE;
  52         }
  53
  54         /* Remove suid, sgid, and file capabilities on truncate too */
  55         ret = dentry_needs_remove_privs(dentry);
  56         if (ret < 0)
  57                 return ret;
  58         if (ret)
  59                 newattrs.ia_valid |= ret | ATTR_FORCE;
  60
  61         inode_lock(dentry->d_inode);
  62         /* Note any delegations or leases have already been broken: */
  63         ret = notify_change(dentry, &newattrs, NULL);
  64         inode_unlock(dentry->d_inode);
  65         return ret;
  66 }
  67
  68 long vfs_truncate(const struct path *path, loff_t length)
  69 {
  70         struct inode *inode;
  71         long error;
  72
  73         inode = path->dentry->d_inode;
  74
  75         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  76         if (S_ISDIR(inode->i_mode))
  77                 return -EISDIR;
  78         if (!S_ISREG(inode->i_mode))
  79                 return -EINVAL;
  80
  81         error = mnt_want_write(path->mnt);
  82         if (error)
  83                 goto out;
  84
  85         error = inode_permission(inode, MAY_WRITE);
  86         if (error)
  87                 goto mnt_drop_write_and_out;
  88
  89         error = -EPERM;
  90         if (IS_APPEND(inode))
  91                 goto mnt_drop_write_and_out;
  92
  93         error = get_write_access(inode);
  94         if (error)
  95                 goto mnt_drop_write_and_out;
  96
  97         /*
  98          * Make sure that there are no leases.  get_write_access() protects
  99          * against the truncate racing with a lease-granting setlease().
 100          */
 101         error = break_lease(inode, O_WRONLY);
 102         if (error)
 103                 goto put_write_and_out;
 104
 105         error = locks_verify_truncate(inode, NULL, length);
 106         if (!error)
 107                 error = security_path_truncate(path);
 108         if (!error)
 109                 error = do_truncate(path->dentry, length, 0, NULL);
 110
 111 put_write_and_out:
 112         put_write_access(inode);
 113 mnt_drop_write_and_out:
 114         mnt_drop_write(path->mnt);
 115 out:
 116         return error;
 117 }
 118 EXPORT_SYMBOL_GPL(vfs_truncate);
 119
 120 long do_sys_truncate(const char __user *pathname, loff_t length)
 121 {
 122         unsigned int lookup_flags = LOOKUP_FOLLOW;
 123         struct path path;
 124         int error;
 125
 126         if (length < 0) /* sorry, but loff_t says... */
 127                 return -EINVAL;
 128
 129 retry:
 130         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 131         if (!error) {
 132                 error = vfs_truncate(&path, length);
 133                 path_put(&path);
 134         }
 135         if (retry_estale(error, lookup_flags)) {
 136                 lookup_flags |= LOOKUP_REVAL;
 137                 goto retry;
 138         }
 139         return error;
 140 }
 141
 142 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 143 {
 144         return do_sys_truncate(path, length);
 145 }
 146
 147 #ifdef CONFIG_COMPAT
 148 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 149 {
 150         return do_sys_truncate(path, length);
 151 }
 152 #endif
 153
 154 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 155 {
 156         struct inode *inode;
 157         struct dentry *dentry;
 158         struct fd f;
 159         int error;
 160
 161         error = -EINVAL;
 162         if (length < 0)
 163                 goto out;
 164         error = -EBADF;
 165         f = fdget(fd);
 166         if (!f.file)
 167                 goto out;
 168
 169         /* explicitly opened as large or we are on 64-bit box */
 170         if (f.file->f_flags & O_LARGEFILE)
 171                 small = 0;
 172
 173         dentry = f.file->f_path.dentry;
 174         inode = dentry->d_inode;
 175         error = -EINVAL;
 176         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 177                 goto out_putf;
 178
 179         error = -EINVAL;
 180         /* Cannot ftruncate over 2^31 bytes without large file support */
 181         if (small && length > MAX_NON_LFS)
 182                 goto out_putf;
 183
 184         error = -EPERM;
 185         /* Check IS_APPEND on real upper inode */
 186         if (IS_APPEND(file_inode(f.file)))
 187                 goto out_putf;
 188
 189         sb_start_write(inode->i_sb);
 190         error = locks_verify_truncate(inode, f.file, length);
 191         if (!error)
 192                 error = security_path_truncate(&f.file->f_path);
 193         if (!error)
 194                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 195         sb_end_write(inode->i_sb);
 196 out_putf:
 197         fdput(f);
 198 out:
 199         return error;
 200 }
 201
 202 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 203 {
 204         return do_sys_ftruncate(fd, length, 1);
 205 }
 206
 207 #ifdef CONFIG_COMPAT
 208 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 209 {
 210         return do_sys_ftruncate(fd, length, 1);
 211 }
 212 #endif
 213
 214 /* LFS versions of truncate are only needed on 32 bit machines */
 215 #if BITS_PER_LONG == 32
 216 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 217 {
 218         return do_sys_truncate(path, length);
 219 }
 220
 221 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 222 {
 223         return do_sys_ftruncate(fd, length, 0);
 224 }
 225 #endif /* BITS_PER_LONG == 32 */
 226
 227
 228 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 229 {
 230         struct inode *inode = file_inode(file);
 231         long ret;
 232
 233         if (offset < 0 || len <= 0)
 234                 return -EINVAL;
 235
 236         /* Return error if mode is not supported */
 237         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 238                 return -EOPNOTSUPP;
 239
 240         /* Punch hole and zero range are mutually exclusive */
 241         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 242             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 243                 return -EOPNOTSUPP;
 244
 245         /* Punch hole must have keep size set */
 246         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 247             !(mode & FALLOC_FL_KEEP_SIZE))
 248                 return -EOPNOTSUPP;
 249
 250         /* Collapse range should only be used exclusively. */
 251         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 252             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 253                 return -EINVAL;
 254
 255         /* Insert range should only be used exclusively. */
 256         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 257             (mode & ~FALLOC_FL_INSERT_RANGE))
 258                 return -EINVAL;
 259
 260         /* Unshare range should only be used with allocate mode. */
 261         if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 262             (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 263                 return -EINVAL;
 264
 265         if (!(file->f_mode & FMODE_WRITE))
 266                 return -EBADF;
 267
 268         /*
 269          * We can only allow pure fallocate on append only files
 270          */
 271         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 272                 return -EPERM;
 273
 274         if (IS_IMMUTABLE(inode))
 275                 return -EPERM;
 276
 277         /*
 278          * We cannot allow any fallocate operation on an active swapfile
 279          */
 280         if (IS_SWAPFILE(inode))
 281                 return -ETXTBSY;
 282
 283         /*
 284          * Revalidate the write permissions, in case security policy has
 285          * changed since the files were opened.
 286          */
 287         ret = security_file_permission(file, MAY_WRITE);
 288         if (ret)
 289                 return ret;
 290
 291         if (S_ISFIFO(inode->i_mode))
 292                 return -ESPIPE;
 293
 294         if (S_ISDIR(inode->i_mode))
 295                 return -EISDIR;
 296
 297         if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 298                 return -ENODEV;
 299
 300         /* Check for wrap through zero too */
 301         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 302                 return -EFBIG;
 303
 304         if (!file->f_op->fallocate)
 305                 return -EOPNOTSUPP;
 306
 307         file_start_write(file);
 308         ret = file->f_op->fallocate(file, mode, offset, len);
 309
 310         /*
 311          * Create inotify and fanotify events.
 312          *
 313          * To keep the logic simple always create events if fallocate succeeds.
 314          * This implies that events are even created if the file size remains
 315          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 316          */
 317         if (ret == 0)
 318                 fsnotify_modify(file);
 319
 320         file_end_write(file);
 321         return ret;
 322 }
 323 EXPORT_SYMBOL_GPL(vfs_fallocate);
 324
 325 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 326 {
 327         struct fd f = fdget(fd);
 328         int error = -EBADF;
 329
 330         if (f.file) {
 331                 error = vfs_fallocate(f.file, mode, offset, len);
 332                 fdput(f);
 333         }
 334         return error;
 335 }
 336
 337 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 338 {
 339         return ksys_fallocate(fd, mode, offset, len);
 340 }
 341
 342 /*
 343  * access() needs to use the real uid/gid, not the effective uid/gid.
 344  * We do this by temporarily clearing all FS-related capabilities and
 345  * switching the fsuid/fsgid around to the real ones.
 346  */
 347 long do_faccessat(int dfd, const char __user *filename, int mode)
 348 {
 349         const struct cred *old_cred;
 350         struct cred *override_cred;
 351         struct path path;
 352         struct inode *inode;
 353         int res;
 354         unsigned int lookup_flags = LOOKUP_FOLLOW;
 355
 356         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 357                 return -EINVAL;
 358
 359         override_cred = prepare_creds();
 360         if (!override_cred)
 361                 return -ENOMEM;
 362
 363         override_cred->fsuid = override_cred->uid;
 364         override_cred->fsgid = override_cred->gid;
 365
 366         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 367                 /* Clear the capabilities if we switch to a non-root user */
 368                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 369                 if (!uid_eq(override_cred->uid, root_uid))
 370                         cap_clear(override_cred->cap_effective);
 371                 else
 372                         override_cred->cap_effective =
 373                                 override_cred->cap_permitted;
 374         }
 375
 376         old_cred = override_creds(override_cred);
 377 retry:
 378         res = user_path_at(dfd, filename, lookup_flags, &path);
 379         if (res)
 380                 goto out;
 381
 382         inode = d_backing_inode(path.dentry);
 383
 384         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 385                 /*
 386                  * MAY_EXEC on regular files is denied if the fs is mounted
 387                  * with the "noexec" flag.
 388                  */
 389                 res = -EACCES;
 390                 if (path_noexec(&path))
 391                         goto out_path_release;
 392         }
 393
 394         res = inode_permission(inode, mode | MAY_ACCESS);
 395         /* SuS v2 requires we report a read only fs too */
 396         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 397                 goto out_path_release;
 398         /*
 399          * This is a rare case where using __mnt_is_readonly()
 400          * is OK without a mnt_want/drop_write() pair.  Since
 401          * no actual write to the fs is performed here, we do
 402          * not need to telegraph to that to anyone.
 403          *
 404          * By doing this, we accept that this access is
 405          * inherently racy and know that the fs may change
 406          * state before we even see this result.
 407          */
 408         if (__mnt_is_readonly(path.mnt))
 409                 res = -EROFS;
 410
 411 out_path_release:
 412         path_put(&path);
 413         if (retry_estale(res, lookup_flags)) {
 414                 lookup_flags |= LOOKUP_REVAL;
 415                 goto retry;
 416         }
 417 out:
 418         revert_creds(old_cred);
 419         put_cred(override_cred);
 420         return res;
 421 }
 422
 423 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 424 {
 425         return do_faccessat(dfd, filename, mode);
 426 }
 427
 428 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 429 {
 430         return do_faccessat(AT_FDCWD, filename, mode);
 431 }
 432
 433 int ksys_chdir(const char __user *filename)
 434 {
 435         struct path path;
 436         int error;
 437         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 438 retry:
 439         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 440         if (error)
 441                 goto out;
 442
 443         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 444         if (error)
 445                 goto dput_and_out;
 446
 447         set_fs_pwd(current->fs, &path);
 448
 449 dput_and_out:
 450         path_put(&path);
 451         if (retry_estale(error, lookup_flags)) {
 452                 lookup_flags |= LOOKUP_REVAL;
 453                 goto retry;
 454         }
 455 out:
 456         return error;
 457 }
 458
 459 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 460 {
 461         return ksys_chdir(filename);
 462 }
 463
 464 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 465 {
 466         struct fd f = fdget_raw(fd);
 467         int error;
 468
 469         error = -EBADF;
 470         if (!f.file)
 471                 goto out;
 472
 473         error = -ENOTDIR;
 474         if (!d_can_lookup(f.file->f_path.dentry))
 475                 goto out_putf;
 476
 477         error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 478         if (!error)
 479                 set_fs_pwd(current->fs, &f.file->f_path);
 480 out_putf:
 481         fdput(f);
 482 out:
 483         return error;
 484 }
 485
 486 int ksys_chroot(const char __user *filename)
 487 {
 488         struct path path;
 489         int error;
 490         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 491 retry:
 492         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 493         if (error)
 494                 goto out;
 495
 496         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 497         if (error)
 498                 goto dput_and_out;
 499
 500         error = -EPERM;
 501         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 502                 goto dput_and_out;
 503         error = security_path_chroot(&path);
 504         if (error)
 505                 goto dput_and_out;
 506
 507         set_fs_root(current->fs, &path);
 508         error = 0;
 509 dput_and_out:
 510         path_put(&path);
 511         if (retry_estale(error, lookup_flags)) {
 512                 lookup_flags |= LOOKUP_REVAL;
 513                 goto retry;
 514         }
 515 out:
 516         return error;
 517 }
 518
 519 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 520 {
 521         return ksys_chroot(filename);
 522 }
 523
 524 static int chmod_common(const struct path *path, umode_t mode)
 525 {
 526         struct inode *inode = path->dentry->d_inode;
 527         struct inode *delegated_inode = NULL;
 528         struct iattr newattrs;
 529         int error;
 530
 531         error = mnt_want_write(path->mnt);
 532         if (error)
 533                 return error;
 534 retry_deleg:
 535         inode_lock(inode);
 536         error = security_path_chmod(path, mode);
 537         if (error)
 538                 goto out_unlock;
 539         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 540         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 541         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 542 out_unlock:
 543         inode_unlock(inode);
 544         if (delegated_inode) {
 545                 error = break_deleg_wait(&delegated_inode);
 546                 if (!error)
 547                         goto retry_deleg;
 548         }
 549         mnt_drop_write(path->mnt);
 550         return error;
 551 }
 552
 553 int ksys_fchmod(unsigned int fd, umode_t mode)
 554 {
 555         struct fd f = fdget(fd);
 556         int err = -EBADF;
 557
 558         if (f.file) {
 559                 audit_file(f.file);
 560                 err = chmod_common(&f.file->f_path, mode);
 561                 fdput(f);
 562         }
 563         return err;
 564 }
 565
 566 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 567 {
 568         return ksys_fchmod(fd, mode);
 569 }
 570
 571 int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
 572 {
 573         struct path path;
 574         int error;
 575         unsigned int lookup_flags = LOOKUP_FOLLOW;
 576 retry:
 577         error = user_path_at(dfd, filename, lookup_flags, &path);
 578         if (!error) {
 579                 error = chmod_common(&path, mode);
 580                 path_put(&path);
 581                 if (retry_estale(error, lookup_flags)) {
 582                         lookup_flags |= LOOKUP_REVAL;
 583                         goto retry;
 584                 }
 585         }
 586         return error;
 587 }
 588
 589 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 590                 umode_t, mode)
 591 {
 592         return do_fchmodat(dfd, filename, mode);
 593 }
 594
 595 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 596 {
 597         return do_fchmodat(AT_FDCWD, filename, mode);
 598 }
 599
 600 static int chown_common(const struct path *path, uid_t user, gid_t group)
 601 {
 602         struct inode *inode = path->dentry->d_inode;
 603         struct inode *delegated_inode = NULL;
 604         int error;
 605         struct iattr newattrs;
 606         kuid_t uid;
 607         kgid_t gid;
 608
 609         uid = make_kuid(current_user_ns(), user);
 610         gid = make_kgid(current_user_ns(), group);
 611
 612 retry_deleg:
 613         newattrs.ia_valid =  ATTR_CTIME;
 614         if (user != (uid_t) -1) {
 615                 if (!uid_valid(uid))
 616                         return -EINVAL;
 617                 newattrs.ia_valid |= ATTR_UID;
 618                 newattrs.ia_uid = uid;
 619         }
 620         if (group != (gid_t) -1) {
 621                 if (!gid_valid(gid))
 622                         return -EINVAL;
 623                 newattrs.ia_valid |= ATTR_GID;
 624                 newattrs.ia_gid = gid;
 625         }
 626         if (!S_ISDIR(inode->i_mode))
 627                 newattrs.ia_valid |=
 628                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 629         inode_lock(inode);
 630         error = security_path_chown(path, uid, gid);
 631         if (!error)
 632                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 633         inode_unlock(inode);
 634         if (delegated_inode) {
 635                 error = break_deleg_wait(&delegated_inode);
 636                 if (!error)
 637                         goto retry_deleg;
 638         }
 639         return error;
 640 }
 641
 642 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 643                 int flag)
 644 {
 645         struct path path;
 646         int error = -EINVAL;
 647         int lookup_flags;
 648
 649         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 650                 goto out;
 651
 652         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 653         if (flag & AT_EMPTY_PATH)
 654                 lookup_flags |= LOOKUP_EMPTY;
 655 retry:
 656         error = user_path_at(dfd, filename, lookup_flags, &path);
 657         if (error)
 658                 goto out;
 659         error = mnt_want_write(path.mnt);
 660         if (error)
 661                 goto out_release;
 662         error = chown_common(&path, user, group);
 663         mnt_drop_write(path.mnt);
 664 out_release:
 665         path_put(&path);
 666         if (retry_estale(error, lookup_flags)) {
 667                 lookup_flags |= LOOKUP_REVAL;
 668                 goto retry;
 669         }
 670 out:
 671         return error;
 672 }
 673
 674 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 675                 gid_t, group, int, flag)
 676 {
 677         return do_fchownat(dfd, filename, user, group, flag);
 678 }
 679
 680 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 681 {
 682         return do_fchownat(AT_FDCWD, filename, user, group, 0);
 683 }
 684
 685 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 686 {
 687         return do_fchownat(AT_FDCWD, filename, user, group,
 688                            AT_SYMLINK_NOFOLLOW);
 689 }
 690
 691 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 692 {
 693         struct fd f = fdget(fd);
 694         int error = -EBADF;
 695
 696         if (!f.file)
 697                 goto out;
 698
 699         error = mnt_want_write_file(f.file);
 700         if (error)
 701                 goto out_fput;
 702         audit_file(f.file);
 703         error = chown_common(&f.file->f_path, user, group);
 704         mnt_drop_write_file(f.file);
 705 out_fput:
 706         fdput(f);
 707 out:
 708         return error;
 709 }
 710
 711 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 712 {
 713         return ksys_fchown(fd, user, group);
 714 }
 715
 716 static int do_dentry_open(struct file *f,
 717                           struct inode *inode,
 718                           int (*open)(struct inode *, struct file *))
 719 {
 720         static const struct file_operations empty_fops = {};
 721         int error;
 722
 723         path_get(&f->f_path);
 724         f->f_inode = inode;
 725         f->f_mapping = inode->i_mapping;
 726
 727         /* Ensure that we skip any errors that predate opening of the file */
 728         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 729
 730         if (unlikely(f->f_flags & O_PATH)) {
 731                 f->f_mode = FMODE_PATH | FMODE_OPENED;
 732                 f->f_op = &empty_fops;
 733                 return 0;
 734         }
 735
 736         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 737                 error = get_write_access(inode);
 738                 if (unlikely(error))
 739                         goto cleanup_file;
 740                 error = __mnt_want_write(f->f_path.mnt);
 741                 if (unlikely(error)) {
 742                         put_write_access(inode);
 743                         goto cleanup_file;
 744                 }
 745                 f->f_mode |= FMODE_WRITER;
 746         }
 747
 748         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 749         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 750                 f->f_mode |= FMODE_ATOMIC_POS;
 751
 752         f->f_op = fops_get(inode->i_fop);
 753         if (unlikely(WARN_ON(!f->f_op))) {
 754                 error = -ENODEV;
 755                 goto cleanup_all;
 756         }
 757
 758         error = security_file_open(f);
 759         if (error)
 760                 goto cleanup_all;
 761
 762         error = break_lease(locks_inode(f), f->f_flags);
 763         if (error)
 764                 goto cleanup_all;
 765
 766         /* normally all 3 are set; ->open() can clear them if needed */
 767         f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 768         if (!open)
 769                 open = f->f_op->open;
 770         if (open) {
 771                 error = open(inode, f);
 772                 if (error)
 773                         goto cleanup_all;
 774         }
 775         f->f_mode |= FMODE_OPENED;
 776         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 777                 i_readcount_inc(inode);
 778         if ((f->f_mode & FMODE_READ) &&
 779              likely(f->f_op->read || f->f_op->read_iter))
 780                 f->f_mode |= FMODE_CAN_READ;
 781         if ((f->f_mode & FMODE_WRITE) &&
 782              likely(f->f_op->write || f->f_op->write_iter))
 783                 f->f_mode |= FMODE_CAN_WRITE;
 784
 785         f->f_write_hint = WRITE_LIFE_NOT_SET;
 786         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 787
 788         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 789
 790         /* NB: we're sure to have correct a_ops only after f_op->open */
 791         if (f->f_flags & O_DIRECT) {
 792                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 793                         return -EINVAL;
 794         }
 795         return 0;
 796
 797 cleanup_all:
 798         if (WARN_ON_ONCE(error > 0))
 799                 error = -EINVAL;
 800         fops_put(f->f_op);
 801         if (f->f_mode & FMODE_WRITER) {
 802                 put_write_access(inode);
 803                 __mnt_drop_write(f->f_path.mnt);
 804         }
 805 cleanup_file:
 806         path_put(&f->f_path);
 807         f->f_path.mnt = NULL;
 808         f->f_path.dentry = NULL;
 809         f->f_inode = NULL;
 810         return error;
 811 }
 812
 813 /**
 814  * finish_open - finish opening a file
 815  * @file: file pointer
 816  * @dentry: pointer to dentry
 817  * @open: open callback
 818  * @opened: state of open
 819  *
 820  * This can be used to finish opening a file passed to i_op->atomic_open().
 821  *
 822  * If the open callback is set to NULL, then the standard f_op->open()
 823  * filesystem callback is substituted.
 824  *
 825  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 826  * the return value of d_splice_alias(), then the caller needs to perform dput()
 827  * on it after finish_open().
 828  *
 829  * On successful return @file is a fully instantiated open file.  After this, if
 830  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 831  *
 832  * Returns zero on success or -errno if the open failed.
 833  */
 834 int finish_open(struct file *file, struct dentry *dentry,
 835                 int (*open)(struct inode *, struct file *))
 836 {
 837         BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
 838
 839         file->f_path.dentry = dentry;
 840         return do_dentry_open(file, d_backing_inode(dentry), open);
 841 }
 842 EXPORT_SYMBOL(finish_open);
 843
 844 /**
 845  * finish_no_open - finish ->atomic_open() without opening the file
 846  *
 847  * @file: file pointer
 848  * @dentry: dentry or NULL (as returned from ->lookup())
 849  *
 850  * This can be used to set the result of a successful lookup in ->atomic_open().
 851  *
 852  * NB: unlike finish_open() this function does consume the dentry reference and
 853  * the caller need not dput() it.
 854  *
 855  * Returns "0" which must be the return value of ->atomic_open() after having
 856  * called this function.
 857  */
 858 int finish_no_open(struct file *file, struct dentry *dentry)
 859 {
 860         file->f_path.dentry = dentry;
 861         return 0;
 862 }
 863 EXPORT_SYMBOL(finish_no_open);
 864
 865 char *file_path(struct file *filp, char *buf, int buflen)
 866 {
 867         return d_path(&filp->f_path, buf, buflen);
 868 }
 869 EXPORT_SYMBOL(file_path);
 870
 871 /**
 872  * vfs_open - open the file at the given path
 873  * @path: path to open
 874  * @file: newly allocated file with f_flag initialized
 875  * @cred: credentials to use
 876  */
 877 int vfs_open(const struct path *path, struct file *file)
 878 {
 879         file->f_path = *path;
 880         return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
 881 }
 882
 883 struct file *dentry_open(const struct path *path, int flags,
 884                          const struct cred *cred)
 885 {
 886         int error;
 887         struct file *f;
 888
 889         validate_creds(cred);
 890
 891         /* We must always pass in a valid mount pointer. */
 892         BUG_ON(!path->mnt);
 893
 894         f = alloc_empty_file(flags, cred);
 895         if (!IS_ERR(f)) {
 896                 error = vfs_open(path, f);
 897                 if (error) {
 898                         fput(f);
 899                         f = ERR_PTR(error);
 900                 }
 901         }
 902         return f;
 903 }
 904 EXPORT_SYMBOL(dentry_open);
 905
 906 struct file *open_with_fake_path(const struct path *path, int flags,
 907                                 struct inode *inode, const struct cred *cred)
 908 {
 909         struct file *f = alloc_empty_file_noaccount(flags, cred);
 910         if (!IS_ERR(f)) {
 911                 int error;
 912
 913                 f->f_path = *path;
 914                 error = do_dentry_open(f, inode, NULL);
 915                 if (error) {
 916                         fput(f);
 917                         f = ERR_PTR(error);
 918                 }
 919         }
 920         return f;
 921 }
 922 EXPORT_SYMBOL(open_with_fake_path);
 923
 924 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 925 {
 926         int lookup_flags = 0;
 927         int acc_mode = ACC_MODE(flags);
 928
 929         /*
 930          * Clear out all open flags we don't know about so that we don't report
 931          * them in fcntl(F_GETFD) or similar interfaces.
 932          */
 933         flags &= VALID_OPEN_FLAGS;
 934
 935         if (flags & (O_CREAT | __O_TMPFILE))
 936                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 937         else
 938                 op->mode = 0;
 939
 940         /* Must never be set by userspace */
 941         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 942
 943         /*
 944          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 945          * check for O_DSYNC if the need any syncing at all we enforce it's
 946          * always set instead of having to deal with possibly weird behaviour
 947          * for malicious applications setting only __O_SYNC.
 948          */
 949         if (flags & __O_SYNC)
 950                 flags |= O_DSYNC;
 951
 952         if (flags & __O_TMPFILE) {
 953                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 954                         return -EINVAL;
 955                 if (!(acc_mode & MAY_WRITE))
 956                         return -EINVAL;
 957         } else if (flags & O_PATH) {
 958                 /*
 959                  * If we have O_PATH in the open flag. Then we
 960                  * cannot have anything other than the below set of flags
 961                  */
 962                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 963                 acc_mode = 0;
 964         }
 965
 966         op->open_flag = flags;
 967
 968         /* O_TRUNC implies we need access checks for write permissions */
 969         if (flags & O_TRUNC)
 970                 acc_mode |= MAY_WRITE;
 971
 972         /* Allow the LSM permission hook to distinguish append
 973            access from general write access. */
 974         if (flags & O_APPEND)
 975                 acc_mode |= MAY_APPEND;
 976
 977         op->acc_mode = acc_mode;
 978
 979         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 980
 981         if (flags & O_CREAT) {
 982                 op->intent |= LOOKUP_CREATE;
 983                 if (flags & O_EXCL)
 984                         op->intent |= LOOKUP_EXCL;
 985         }
 986
 987         if (flags & O_DIRECTORY)
 988                 lookup_flags |= LOOKUP_DIRECTORY;
 989         if (!(flags & O_NOFOLLOW))
 990                 lookup_flags |= LOOKUP_FOLLOW;
 991         op->lookup_flags = lookup_flags;
 992         return 0;
 993 }
 994
 995 /**
 996  * file_open_name - open file and return file pointer
 997  *
 998  * @name:       struct filename containing path to open
 999  * @flags:      open flags as per the open(2) second argument
1000  * @mode:       mode for the new file if O_CREAT is set, else ignored
1001  *
1002  * This is the helper to open a file from kernelspace if you really
1003  * have to.  But in generally you should not do this, so please move
1004  * along, nothing to see here..
1005  */
1006 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1007 {
1008         struct open_flags op;
1009         int err = build_open_flags(flags, mode, &op);
1010         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
1011 }
1012
1013 /**
1014  * filp_open - open file and return file pointer
1015  *
1016  * @filename:   path to open
1017  * @flags:      open flags as per the open(2) second argument
1018  * @mode:       mode for the new file if O_CREAT is set, else ignored
1019  *
1020  * This is the helper to open a file from kernelspace if you really
1021  * have to.  But in generally you should not do this, so please move
1022  * along, nothing to see here..
1023  */
1024 struct file *filp_open(const char *filename, int flags, umode_t mode)
1025 {
1026         struct filename *name = getname_kernel(filename);
1027         struct file *file = ERR_CAST(name);
1028
1029         if (!IS_ERR(name)) {
1030                 file = file_open_name(name, flags, mode);
1031                 putname(name);
1032         }
1033         return file;
1034 }
1035 EXPORT_SYMBOL(filp_open);
1036
1037 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1038                             const char *filename, int flags, umode_t mode)
1039 {
1040         struct open_flags op;
1041         int err = build_open_flags(flags, mode, &op);
1042         if (err)
1043                 return ERR_PTR(err);
1044         return do_file_open_root(dentry, mnt, filename, &op);
1045 }
1046 EXPORT_SYMBOL(file_open_root);
1047
1048 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1049 {
1050         struct open_flags op;
1051         int fd = build_open_flags(flags, mode, &op);
1052         struct filename *tmp;
1053
1054         if (fd)
1055                 return fd;
1056
1057         tmp = getname(filename);
1058         if (IS_ERR(tmp))
1059                 return PTR_ERR(tmp);
1060
1061         fd = get_unused_fd_flags(flags);
1062         if (fd >= 0) {
1063                 struct file *f = do_filp_open(dfd, tmp, &op);
1064                 if (IS_ERR(f)) {
1065                         put_unused_fd(fd);
1066                         fd = PTR_ERR(f);
1067                 } else {
1068                         fsnotify_open(f);
1069                         fd_install(fd, f);
1070                 }
1071         }
1072         putname(tmp);
1073         return fd;
1074 }
1075
1076 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1077 {
1078         if (force_o_largefile())
1079                 flags |= O_LARGEFILE;
1080
1081         return do_sys_open(AT_FDCWD, filename, flags, mode);
1082 }
1083
1084 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1085                 umode_t, mode)
1086 {
1087         if (force_o_largefile())
1088                 flags |= O_LARGEFILE;
1089
1090         return do_sys_open(dfd, filename, flags, mode);
1091 }
1092
1093 #ifdef CONFIG_COMPAT
1094 /*
1095  * Exactly like sys_open(), except that it doesn't set the
1096  * O_LARGEFILE flag.
1097  */
1098 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1099 {
1100         return do_sys_open(AT_FDCWD, filename, flags, mode);
1101 }
1102
1103 /*
1104  * Exactly like sys_openat(), except that it doesn't set the
1105  * O_LARGEFILE flag.
1106  */
1107 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1108 {
1109         return do_sys_open(dfd, filename, flags, mode);
1110 }
1111 #endif
1112
1113 #ifndef __alpha__
1114
1115 /*
1116  * For backward compatibility?  Maybe this should be moved
1117  * into arch/i386 instead?
1118  */
1119 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1120 {
1121         return ksys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1122 }
1123
1124 #endif
1125
1126 /*
1127  * "id" is the POSIX thread ID. We use the
1128  * files pointer for this..
1129  */
1130 int filp_close(struct file *filp, fl_owner_t id)
1131 {
1132         int retval = 0;
1133
1134         if (!file_count(filp)) {
1135                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1136                 return 0;
1137         }
1138
1139         if (filp->f_op->flush)
1140                 retval = filp->f_op->flush(filp, id);
1141
1142         if (likely(!(filp->f_mode & FMODE_PATH))) {
1143                 dnotify_flush(filp, id);
1144                 locks_remove_posix(filp, id);
1145         }
1146         fput(filp);
1147         return retval;
1148 }
1149
1150 EXPORT_SYMBOL(filp_close);
1151
1152 /*
1153  * Careful here! We test whether the file pointer is NULL before
1154  * releasing the fd. This ensures that one clone task can't release
1155  * an fd while another clone is opening it.
1156  */
1157 SYSCALL_DEFINE1(close, unsigned int, fd)
1158 {
1159         int retval = __close_fd(current->files, fd);
1160
1161         /* can't restart close syscall because file table entry was cleared */
1162         if (unlikely(retval == -ERESTARTSYS ||
1163                      retval == -ERESTARTNOINTR ||
1164                      retval == -ERESTARTNOHAND ||
1165                      retval == -ERESTART_RESTARTBLOCK))
1166                 retval = -EINTR;
1167
1168         return retval;
1169 }
1170
1171 /*
1172  * This routine simulates a hangup on the tty, to arrange that users
1173  * are given clean terminals at login time.
1174  */
1175 SYSCALL_DEFINE0(vhangup)
1176 {
1177         if (capable(CAP_SYS_TTY_CONFIG)) {
1178                 tty_vhangup_self();
1179                 return 0;
1180         }
1181         return -EPERM;
1182 }
1183
1184 /*
1185  * Called when an inode is about to be open.
1186  * We use this to disallow opening large files on 32bit systems if
1187  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1188  * on this flag in sys_open.
1189  */
1190 int generic_file_open(struct inode * inode, struct file * filp)
1191 {
1192         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1193                 return -EOVERFLOW;
1194         return 0;
1195 }
1196
1197 EXPORT_SYMBOL(generic_file_open);
1198
1199 /*
1200  * This is used by subsystems that don't want seekable
1201  * file descriptors. The function is not supposed to ever fail, the only
1202  * reason it returns an 'int' and not 'void' is so that it can be plugged
1203  * directly into file_operations structure.
1204  */
1205 int nonseekable_open(struct inode *inode, struct file *filp)
1206 {
1207         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1208         return 0;
1209 }
1210
1211 EXPORT_SYMBOL(nonseekable_open);