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         /*
 377          * The new set of credentials can *only* be used in
 378          * task-synchronous circumstances, and does not need
 379          * RCU freeing, unless somebody then takes a separate
 380          * reference to it.
 381          *
 382          * NOTE! This is _only_ true because this credential
 383          * is used purely for override_creds() that installs
 384          * it as the subjective cred. Other threads will be
 385          * accessing ->real_cred, not the subjective cred.
 386          *
 387          * If somebody _does_ make a copy of this (using the
 388          * 'get_current_cred()' function), that will clear the
 389          * non_rcu field, because now that other user may be
 390          * expecting RCU freeing. But normal thread-synchronous
 391          * cred accesses will keep things non-RCY.
 392          */
 393         override_cred->non_rcu = 1;
 394
 395         old_cred = override_creds(override_cred);
 396 retry:
 397         res = user_path_at(dfd, filename, lookup_flags, &path);
 398         if (res)
 399                 goto out;
 400
 401         inode = d_backing_inode(path.dentry);
 402
 403         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 404                 /*
 405                  * MAY_EXEC on regular files is denied if the fs is mounted
 406                  * with the "noexec" flag.
 407                  */
 408                 res = -EACCES;
 409                 if (path_noexec(&path))
 410                         goto out_path_release;
 411         }
 412
 413         res = inode_permission(inode, mode | MAY_ACCESS);
 414         /* SuS v2 requires we report a read only fs too */
 415         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 416                 goto out_path_release;
 417         /*
 418          * This is a rare case where using __mnt_is_readonly()
 419          * is OK without a mnt_want/drop_write() pair.  Since
 420          * no actual write to the fs is performed here, we do
 421          * not need to telegraph to that to anyone.
 422          *
 423          * By doing this, we accept that this access is
 424          * inherently racy and know that the fs may change
 425          * state before we even see this result.
 426          */
 427         if (__mnt_is_readonly(path.mnt))
 428                 res = -EROFS;
 429
 430 out_path_release:
 431         path_put(&path);
 432         if (retry_estale(res, lookup_flags)) {
 433                 lookup_flags |= LOOKUP_REVAL;
 434                 goto retry;
 435         }
 436 out:
 437         revert_creds(old_cred);
 438         put_cred(override_cred);
 439         return res;
 440 }
 441
 442 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 443 {
 444         return do_faccessat(dfd, filename, mode);
 445 }
 446
 447 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 448 {
 449         return do_faccessat(AT_FDCWD, filename, mode);
 450 }
 451
 452 int ksys_chdir(const char __user *filename)
 453 {
 454         struct path path;
 455         int error;
 456         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 457 retry:
 458         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 459         if (error)
 460                 goto out;
 461
 462         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 463         if (error)
 464                 goto dput_and_out;
 465
 466         set_fs_pwd(current->fs, &path);
 467
 468 dput_and_out:
 469         path_put(&path);
 470         if (retry_estale(error, lookup_flags)) {
 471                 lookup_flags |= LOOKUP_REVAL;
 472                 goto retry;
 473         }
 474 out:
 475         return error;
 476 }
 477
 478 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 479 {
 480         return ksys_chdir(filename);
 481 }
 482
 483 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 484 {
 485         struct fd f = fdget_raw(fd);
 486         int error;
 487
 488         error = -EBADF;
 489         if (!f.file)
 490                 goto out;
 491
 492         error = -ENOTDIR;
 493         if (!d_can_lookup(f.file->f_path.dentry))
 494                 goto out_putf;
 495
 496         error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 497         if (!error)
 498                 set_fs_pwd(current->fs, &f.file->f_path);
 499 out_putf:
 500         fdput(f);
 501 out:
 502         return error;
 503 }
 504
 505 int ksys_chroot(const char __user *filename)
 506 {
 507         struct path path;
 508         int error;
 509         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 510 retry:
 511         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 512         if (error)
 513                 goto out;
 514
 515         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 516         if (error)
 517                 goto dput_and_out;
 518
 519         error = -EPERM;
 520         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 521                 goto dput_and_out;
 522         error = security_path_chroot(&path);
 523         if (error)
 524                 goto dput_and_out;
 525
 526         set_fs_root(current->fs, &path);
 527         error = 0;
 528 dput_and_out:
 529         path_put(&path);
 530         if (retry_estale(error, lookup_flags)) {
 531                 lookup_flags |= LOOKUP_REVAL;
 532                 goto retry;
 533         }
 534 out:
 535         return error;
 536 }
 537
 538 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 539 {
 540         return ksys_chroot(filename);
 541 }
 542
 543 static int chmod_common(const struct path *path, umode_t mode)
 544 {
 545         struct inode *inode = path->dentry->d_inode;
 546         struct inode *delegated_inode = NULL;
 547         struct iattr newattrs;
 548         int error;
 549
 550         error = mnt_want_write(path->mnt);
 551         if (error)
 552                 return error;
 553 retry_deleg:
 554         inode_lock(inode);
 555         error = security_path_chmod(path, mode);
 556         if (error)
 557                 goto out_unlock;
 558         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 559         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 560         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 561 out_unlock:
 562         inode_unlock(inode);
 563         if (delegated_inode) {
 564                 error = break_deleg_wait(&delegated_inode);
 565                 if (!error)
 566                         goto retry_deleg;
 567         }
 568         mnt_drop_write(path->mnt);
 569         return error;
 570 }
 571
 572 int ksys_fchmod(unsigned int fd, umode_t mode)
 573 {
 574         struct fd f = fdget(fd);
 575         int err = -EBADF;
 576
 577         if (f.file) {
 578                 audit_file(f.file);
 579                 err = chmod_common(&f.file->f_path, mode);
 580                 fdput(f);
 581         }
 582         return err;
 583 }
 584
 585 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 586 {
 587         return ksys_fchmod(fd, mode);
 588 }
 589
 590 int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
 591 {
 592         struct path path;
 593         int error;
 594         unsigned int lookup_flags = LOOKUP_FOLLOW;
 595 retry:
 596         error = user_path_at(dfd, filename, lookup_flags, &path);
 597         if (!error) {
 598                 error = chmod_common(&path, mode);
 599                 path_put(&path);
 600                 if (retry_estale(error, lookup_flags)) {
 601                         lookup_flags |= LOOKUP_REVAL;
 602                         goto retry;
 603                 }
 604         }
 605         return error;
 606 }
 607
 608 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 609                 umode_t, mode)
 610 {
 611         return do_fchmodat(dfd, filename, mode);
 612 }
 613
 614 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 615 {
 616         return do_fchmodat(AT_FDCWD, filename, mode);
 617 }
 618
 619 static int chown_common(const struct path *path, uid_t user, gid_t group)
 620 {
 621         struct inode *inode = path->dentry->d_inode;
 622         struct inode *delegated_inode = NULL;
 623         int error;
 624         struct iattr newattrs;
 625         kuid_t uid;
 626         kgid_t gid;
 627
 628         uid = make_kuid(current_user_ns(), user);
 629         gid = make_kgid(current_user_ns(), group);
 630
 631 retry_deleg:
 632         newattrs.ia_valid =  ATTR_CTIME;
 633         if (user != (uid_t) -1) {
 634                 if (!uid_valid(uid))
 635                         return -EINVAL;
 636                 newattrs.ia_valid |= ATTR_UID;
 637                 newattrs.ia_uid = uid;
 638         }
 639         if (group != (gid_t) -1) {
 640                 if (!gid_valid(gid))
 641                         return -EINVAL;
 642                 newattrs.ia_valid |= ATTR_GID;
 643                 newattrs.ia_gid = gid;
 644         }
 645         if (!S_ISDIR(inode->i_mode))
 646                 newattrs.ia_valid |=
 647                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 648         inode_lock(inode);
 649         error = security_path_chown(path, uid, gid);
 650         if (!error)
 651                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 652         inode_unlock(inode);
 653         if (delegated_inode) {
 654                 error = break_deleg_wait(&delegated_inode);
 655                 if (!error)
 656                         goto retry_deleg;
 657         }
 658         return error;
 659 }
 660
 661 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 662                 int flag)
 663 {
 664         struct path path;
 665         int error = -EINVAL;
 666         int lookup_flags;
 667
 668         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 669                 goto out;
 670
 671         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 672         if (flag & AT_EMPTY_PATH)
 673                 lookup_flags |= LOOKUP_EMPTY;
 674 retry:
 675         error = user_path_at(dfd, filename, lookup_flags, &path);
 676         if (error)
 677                 goto out;
 678         error = mnt_want_write(path.mnt);
 679         if (error)
 680                 goto out_release;
 681         error = chown_common(&path, user, group);
 682         mnt_drop_write(path.mnt);
 683 out_release:
 684         path_put(&path);
 685         if (retry_estale(error, lookup_flags)) {
 686                 lookup_flags |= LOOKUP_REVAL;
 687                 goto retry;
 688         }
 689 out:
 690         return error;
 691 }
 692
 693 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 694                 gid_t, group, int, flag)
 695 {
 696         return do_fchownat(dfd, filename, user, group, flag);
 697 }
 698
 699 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 700 {
 701         return do_fchownat(AT_FDCWD, filename, user, group, 0);
 702 }
 703
 704 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 705 {
 706         return do_fchownat(AT_FDCWD, filename, user, group,
 707                            AT_SYMLINK_NOFOLLOW);
 708 }
 709
 710 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 711 {
 712         struct fd f = fdget(fd);
 713         int error = -EBADF;
 714
 715         if (!f.file)
 716                 goto out;
 717
 718         error = mnt_want_write_file(f.file);
 719         if (error)
 720                 goto out_fput;
 721         audit_file(f.file);
 722         error = chown_common(&f.file->f_path, user, group);
 723         mnt_drop_write_file(f.file);
 724 out_fput:
 725         fdput(f);
 726 out:
 727         return error;
 728 }
 729
 730 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 731 {
 732         return ksys_fchown(fd, user, group);
 733 }
 734
 735 static int do_dentry_open(struct file *f,
 736                           struct inode *inode,
 737                           int (*open)(struct inode *, struct file *))
 738 {
 739         static const struct file_operations empty_fops = {};
 740         int error;
 741
 742         path_get(&f->f_path);
 743         f->f_inode = inode;
 744         f->f_mapping = inode->i_mapping;
 745
 746         /* Ensure that we skip any errors that predate opening of the file */
 747         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 748
 749         if (unlikely(f->f_flags & O_PATH)) {
 750                 f->f_mode = FMODE_PATH | FMODE_OPENED;
 751                 f->f_op = &empty_fops;
 752                 return 0;
 753         }
 754
 755         /* Any file opened for execve()/uselib() has to be a regular file. */
 756         if (unlikely(f->f_flags & FMODE_EXEC && !S_ISREG(inode->i_mode))) {
 757                 error = -EACCES;
 758                 goto cleanup_file;
 759         }
 760
 761         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 762                 error = get_write_access(inode);
 763                 if (unlikely(error))
 764                         goto cleanup_file;
 765                 error = __mnt_want_write(f->f_path.mnt);
 766                 if (unlikely(error)) {
 767                         put_write_access(inode);
 768                         goto cleanup_file;
 769                 }
 770                 f->f_mode |= FMODE_WRITER;
 771         }
 772
 773         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 774         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 775                 f->f_mode |= FMODE_ATOMIC_POS;
 776
 777         f->f_op = fops_get(inode->i_fop);
 778         if (unlikely(WARN_ON(!f->f_op))) {
 779                 error = -ENODEV;
 780                 goto cleanup_all;
 781         }
 782
 783         error = security_file_open(f);
 784         if (error)
 785                 goto cleanup_all;
 786
 787         error = break_lease(locks_inode(f), f->f_flags);
 788         if (error)
 789                 goto cleanup_all;
 790
 791         /* normally all 3 are set; ->open() can clear them if needed */
 792         f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 793         if (!open)
 794                 open = f->f_op->open;
 795         if (open) {
 796                 error = open(inode, f);
 797                 if (error)
 798                         goto cleanup_all;
 799         }
 800         f->f_mode |= FMODE_OPENED;
 801         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 802                 i_readcount_inc(inode);
 803         if ((f->f_mode & FMODE_READ) &&
 804              likely(f->f_op->read || f->f_op->read_iter))
 805                 f->f_mode |= FMODE_CAN_READ;
 806         if ((f->f_mode & FMODE_WRITE) &&
 807              likely(f->f_op->write || f->f_op->write_iter))
 808                 f->f_mode |= FMODE_CAN_WRITE;
 809
 810         f->f_write_hint = WRITE_LIFE_NOT_SET;
 811         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 812
 813         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 814
 815         /* NB: we're sure to have correct a_ops only after f_op->open */
 816         if (f->f_flags & O_DIRECT) {
 817                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 818                         return -EINVAL;
 819         }
 820         return 0;
 821
 822 cleanup_all:
 823         if (WARN_ON_ONCE(error > 0))
 824                 error = -EINVAL;
 825         fops_put(f->f_op);
 826         if (f->f_mode & FMODE_WRITER) {
 827                 put_write_access(inode);
 828                 __mnt_drop_write(f->f_path.mnt);
 829         }
 830 cleanup_file:
 831         path_put(&f->f_path);
 832         f->f_path.mnt = NULL;
 833         f->f_path.dentry = NULL;
 834         f->f_inode = NULL;
 835         return error;
 836 }
 837
 838 /**
 839  * finish_open - finish opening a file
 840  * @file: file pointer
 841  * @dentry: pointer to dentry
 842  * @open: open callback
 843  * @opened: state of open
 844  *
 845  * This can be used to finish opening a file passed to i_op->atomic_open().
 846  *
 847  * If the open callback is set to NULL, then the standard f_op->open()
 848  * filesystem callback is substituted.
 849  *
 850  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 851  * the return value of d_splice_alias(), then the caller needs to perform dput()
 852  * on it after finish_open().
 853  *
 854  * Returns zero on success or -errno if the open failed.
 855  */
 856 int finish_open(struct file *file, struct dentry *dentry,
 857                 int (*open)(struct inode *, struct file *))
 858 {
 859         BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
 860
 861         file->f_path.dentry = dentry;
 862         return do_dentry_open(file, d_backing_inode(dentry), open);
 863 }
 864 EXPORT_SYMBOL(finish_open);
 865
 866 /**
 867  * finish_no_open - finish ->atomic_open() without opening the file
 868  *
 869  * @file: file pointer
 870  * @dentry: dentry or NULL (as returned from ->lookup())
 871  *
 872  * This can be used to set the result of a successful lookup in ->atomic_open().
 873  *
 874  * NB: unlike finish_open() this function does consume the dentry reference and
 875  * the caller need not dput() it.
 876  *
 877  * Returns "0" which must be the return value of ->atomic_open() after having
 878  * called this function.
 879  */
 880 int finish_no_open(struct file *file, struct dentry *dentry)
 881 {
 882         file->f_path.dentry = dentry;
 883         return 0;
 884 }
 885 EXPORT_SYMBOL(finish_no_open);
 886
 887 char *file_path(struct file *filp, char *buf, int buflen)
 888 {
 889         return d_path(&filp->f_path, buf, buflen);
 890 }
 891 EXPORT_SYMBOL(file_path);
 892
 893 /**
 894  * vfs_open - open the file at the given path
 895  * @path: path to open
 896  * @file: newly allocated file with f_flag initialized
 897  * @cred: credentials to use
 898  */
 899 int vfs_open(const struct path *path, struct file *file)
 900 {
 901         file->f_path = *path;
 902         return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
 903 }
 904
 905 struct file *dentry_open(const struct path *path, int flags,
 906                          const struct cred *cred)
 907 {
 908         int error;
 909         struct file *f;
 910
 911         validate_creds(cred);
 912
 913         /* We must always pass in a valid mount pointer. */
 914         BUG_ON(!path->mnt);
 915
 916         f = alloc_empty_file(flags, cred);
 917         if (!IS_ERR(f)) {
 918                 error = vfs_open(path, f);
 919                 if (error) {
 920                         fput(f);
 921                         f = ERR_PTR(error);
 922                 }
 923         }
 924         return f;
 925 }
 926 EXPORT_SYMBOL(dentry_open);
 927
 928 struct file *open_with_fake_path(const struct path *path, int flags,
 929                                 struct inode *inode, const struct cred *cred)
 930 {
 931         struct file *f = alloc_empty_file_noaccount(flags, cred);
 932         if (!IS_ERR(f)) {
 933                 int error;
 934
 935                 f->f_path = *path;
 936                 error = do_dentry_open(f, inode, NULL);
 937                 if (error) {
 938                         fput(f);
 939                         f = ERR_PTR(error);
 940                 }
 941         }
 942         return f;
 943 }
 944 EXPORT_SYMBOL(open_with_fake_path);
 945
 946 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 947 {
 948         int lookup_flags = 0;
 949         int acc_mode = ACC_MODE(flags);
 950
 951         /*
 952          * Clear out all open flags we don't know about so that we don't report
 953          * them in fcntl(F_GETFD) or similar interfaces.
 954          */
 955         flags &= VALID_OPEN_FLAGS;
 956
 957         if (flags & (O_CREAT | __O_TMPFILE))
 958                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 959         else
 960                 op->mode = 0;
 961
 962         /* Must never be set by userspace */
 963         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 964
 965         /*
 966          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 967          * check for O_DSYNC if the need any syncing at all we enforce it's
 968          * always set instead of having to deal with possibly weird behaviour
 969          * for malicious applications setting only __O_SYNC.
 970          */
 971         if (flags & __O_SYNC)
 972                 flags |= O_DSYNC;
 973
 974         if (flags & __O_TMPFILE) {
 975                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 976                         return -EINVAL;
 977                 if (!(acc_mode & MAY_WRITE))
 978                         return -EINVAL;
 979         } else if (flags & O_PATH) {
 980                 /*
 981                  * If we have O_PATH in the open flag. Then we
 982                  * cannot have anything other than the below set of flags
 983                  */
 984                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 985                 acc_mode = 0;
 986         }
 987
 988         op->open_flag = flags;
 989
 990         /* O_TRUNC implies we need access checks for write permissions */
 991         if (flags & O_TRUNC)
 992                 acc_mode |= MAY_WRITE;
 993
 994         /* Allow the LSM permission hook to distinguish append
 995            access from general write access. */
 996         if (flags & O_APPEND)
 997                 acc_mode |= MAY_APPEND;
 998
 999         op->acc_mode = acc_mode;
1000
1001         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1002
1003         if (flags & O_CREAT) {
1004                 op->intent |= LOOKUP_CREATE;
1005                 if (flags & O_EXCL)
1006                         op->intent |= LOOKUP_EXCL;
1007         }
1008
1009         if (flags & O_DIRECTORY)
1010                 lookup_flags |= LOOKUP_DIRECTORY;
1011         if (!(flags & O_NOFOLLOW))
1012                 lookup_flags |= LOOKUP_FOLLOW;
1013         op->lookup_flags = lookup_flags;
1014         return 0;
1015 }
1016
1017 /**
1018  * file_open_name - open file and return file pointer
1019  *
1020  * @name:       struct filename containing path to open
1021  * @flags:      open flags as per the open(2) second argument
1022  * @mode:       mode for the new file if O_CREAT is set, else ignored
1023  *
1024  * This is the helper to open a file from kernelspace if you really
1025  * have to.  But in generally you should not do this, so please move
1026  * along, nothing to see here..
1027  */
1028 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1029 {
1030         struct open_flags op;
1031         int err = build_open_flags(flags, mode, &op);
1032         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
1033 }
1034
1035 /**
1036  * filp_open - open file and return file pointer
1037  *
1038  * @filename:   path to open
1039  * @flags:      open flags as per the open(2) second argument
1040  * @mode:       mode for the new file if O_CREAT is set, else ignored
1041  *
1042  * This is the helper to open a file from kernelspace if you really
1043  * have to.  But in generally you should not do this, so please move
1044  * along, nothing to see here..
1045  */
1046 struct file *filp_open(const char *filename, int flags, umode_t mode)
1047 {
1048         struct filename *name = getname_kernel(filename);
1049         struct file *file = ERR_CAST(name);
1050
1051         if (!IS_ERR(name)) {
1052                 file = file_open_name(name, flags, mode);
1053                 putname(name);
1054         }
1055         return file;
1056 }
1057 EXPORT_SYMBOL(filp_open);
1058
1059 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1060                             const char *filename, int flags, umode_t mode)
1061 {
1062         struct open_flags op;
1063         int err = build_open_flags(flags, mode, &op);
1064         if (err)
1065                 return ERR_PTR(err);
1066         return do_file_open_root(dentry, mnt, filename, &op);
1067 }
1068 EXPORT_SYMBOL(file_open_root);
1069
1070 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1071 {
1072         struct open_flags op;
1073         int fd = build_open_flags(flags, mode, &op);
1074         struct filename *tmp;
1075
1076         if (fd)
1077                 return fd;
1078
1079         tmp = getname(filename);
1080         if (IS_ERR(tmp))
1081                 return PTR_ERR(tmp);
1082
1083         fd = get_unused_fd_flags(flags);
1084         if (fd >= 0) {
1085                 struct file *f = do_filp_open(dfd, tmp, &op);
1086                 if (IS_ERR(f)) {
1087                         put_unused_fd(fd);
1088                         fd = PTR_ERR(f);
1089                 } else {
1090                         fsnotify_open(f);
1091                         fd_install(fd, f);
1092                 }
1093         }
1094         putname(tmp);
1095         return fd;
1096 }
1097
1098 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1099 {
1100         if (force_o_largefile())
1101                 flags |= O_LARGEFILE;
1102
1103         return do_sys_open(AT_FDCWD, filename, flags, mode);
1104 }
1105
1106 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1107                 umode_t, mode)
1108 {
1109         if (force_o_largefile())
1110                 flags |= O_LARGEFILE;
1111
1112         return do_sys_open(dfd, filename, flags, mode);
1113 }
1114
1115 #ifdef CONFIG_COMPAT
1116 /*
1117  * Exactly like sys_open(), except that it doesn't set the
1118  * O_LARGEFILE flag.
1119  */
1120 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1121 {
1122         return do_sys_open(AT_FDCWD, filename, flags, mode);
1123 }
1124
1125 /*
1126  * Exactly like sys_openat(), except that it doesn't set the
1127  * O_LARGEFILE flag.
1128  */
1129 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1130 {
1131         return do_sys_open(dfd, filename, flags, mode);
1132 }
1133 #endif
1134
1135 #ifndef __alpha__
1136
1137 /*
1138  * For backward compatibility?  Maybe this should be moved
1139  * into arch/i386 instead?
1140  */
1141 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1142 {
1143         return ksys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1144 }
1145
1146 #endif
1147
1148 /*
1149  * "id" is the POSIX thread ID. We use the
1150  * files pointer for this..
1151  */
1152 int filp_close(struct file *filp, fl_owner_t id)
1153 {
1154         int retval = 0;
1155
1156         if (!file_count(filp)) {
1157                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1158                 return 0;
1159         }
1160
1161         if (filp->f_op->flush)
1162                 retval = filp->f_op->flush(filp, id);
1163
1164         if (likely(!(filp->f_mode & FMODE_PATH))) {
1165                 dnotify_flush(filp, id);
1166                 locks_remove_posix(filp, id);
1167         }
1168         fput(filp);
1169         return retval;
1170 }
1171
1172 EXPORT_SYMBOL(filp_close);
1173
1174 /*
1175  * Careful here! We test whether the file pointer is NULL before
1176  * releasing the fd. This ensures that one clone task can't release
1177  * an fd while another clone is opening it.
1178  */
1179 SYSCALL_DEFINE1(close, unsigned int, fd)
1180 {
1181         int retval = __close_fd(current->files, fd);
1182
1183         /* can't restart close syscall because file table entry was cleared */
1184         if (unlikely(retval == -ERESTARTSYS ||
1185                      retval == -ERESTARTNOINTR ||
1186                      retval == -ERESTARTNOHAND ||
1187                      retval == -ERESTART_RESTARTBLOCK))
1188                 retval = -EINTR;
1189
1190         return retval;
1191 }
1192
1193 /*
1194  * This routine simulates a hangup on the tty, to arrange that users
1195  * are given clean terminals at login time.
1196  */
1197 SYSCALL_DEFINE0(vhangup)
1198 {
1199         if (capable(CAP_SYS_TTY_CONFIG)) {
1200                 tty_vhangup_self();
1201                 return 0;
1202         }
1203         return -EPERM;
1204 }
1205
1206 /*
1207  * Called when an inode is about to be open.
1208  * We use this to disallow opening large files on 32bit systems if
1209  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1210  * on this flag in sys_open.
1211  */
1212 int generic_file_open(struct inode * inode, struct file * filp)
1213 {
1214         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1215                 return -EOVERFLOW;
1216         return 0;
1217 }
1218
1219 EXPORT_SYMBOL(generic_file_open);
1220
1221 /*
1222  * This is used by subsystems that don't want seekable
1223  * file descriptors. The function is not supposed to ever fail, the only
1224  * reason it returns an 'int' and not 'void' is so that it can be plugged
1225  * directly into file_operations structure.
1226  */
1227 int nonseekable_open(struct inode *inode, struct file *filp)
1228 {
1229         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1230         return 0;
1231 }
1232
1233 EXPORT_SYMBOL(nonseekable_open);
1234
1235 /*
1236  * stream_open is used by subsystems that want stream-like file descriptors.
1237  * Such file descriptors are not seekable and don't have notion of position
1238  * (file.f_pos is always 0). Contrary to file descriptors of other regular
1239  * files, .read() and .write() can run simultaneously.
1240  *
1241  * stream_open never fails and is marked to return int so that it could be
1242  * directly used as file_operations.open .
1243  */
1244 int stream_open(struct inode *inode, struct file *filp)
1245 {
1246         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1247         filp->f_mode |= FMODE_STREAM;
1248         return 0;
1249 }
1250
1251 EXPORT_SYMBOL(stream_open);