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         struct dentry *upperdentry;
  72         long error;
  73
  74         inode = path->dentry->d_inode;
  75
  76         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  77         if (S_ISDIR(inode->i_mode))
  78                 return -EISDIR;
  79         if (!S_ISREG(inode->i_mode))
  80                 return -EINVAL;
  81
  82         error = mnt_want_write(path->mnt);
  83         if (error)
  84                 goto out;
  85
  86         error = inode_permission(inode, MAY_WRITE);
  87         if (error)
  88                 goto mnt_drop_write_and_out;
  89
  90         error = -EPERM;
  91         if (IS_APPEND(inode))
  92                 goto mnt_drop_write_and_out;
  93
  94         /*
  95          * If this is an overlayfs then do as if opening the file so we get
  96          * write access on the upper inode, not on the overlay inode.  For
  97          * non-overlay filesystems d_real() is an identity function.
  98          */
  99         upperdentry = d_real(path->dentry, NULL, O_WRONLY, 0);
 100         error = PTR_ERR(upperdentry);
 101         if (IS_ERR(upperdentry))
 102                 goto mnt_drop_write_and_out;
 103
 104         error = get_write_access(upperdentry->d_inode);
 105         if (error)
 106                 goto mnt_drop_write_and_out;
 107
 108         /*
 109          * Make sure that there are no leases.  get_write_access() protects
 110          * against the truncate racing with a lease-granting setlease().
 111          */
 112         error = break_lease(inode, O_WRONLY);
 113         if (error)
 114                 goto put_write_and_out;
 115
 116         error = locks_verify_truncate(inode, NULL, length);
 117         if (!error)
 118                 error = security_path_truncate(path);
 119         if (!error)
 120                 error = do_truncate(path->dentry, length, 0, NULL);
 121
 122 put_write_and_out:
 123         put_write_access(upperdentry->d_inode);
 124 mnt_drop_write_and_out:
 125         mnt_drop_write(path->mnt);
 126 out:
 127         return error;
 128 }
 129 EXPORT_SYMBOL_GPL(vfs_truncate);
 130
 131 long do_sys_truncate(const char __user *pathname, loff_t length)
 132 {
 133         unsigned int lookup_flags = LOOKUP_FOLLOW;
 134         struct path path;
 135         int error;
 136
 137         if (length < 0) /* sorry, but loff_t says... */
 138                 return -EINVAL;
 139
 140 retry:
 141         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 142         if (!error) {
 143                 error = vfs_truncate(&path, length);
 144                 path_put(&path);
 145         }
 146         if (retry_estale(error, lookup_flags)) {
 147                 lookup_flags |= LOOKUP_REVAL;
 148                 goto retry;
 149         }
 150         return error;
 151 }
 152
 153 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 154 {
 155         return do_sys_truncate(path, length);
 156 }
 157
 158 #ifdef CONFIG_COMPAT
 159 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 160 {
 161         return do_sys_truncate(path, length);
 162 }
 163 #endif
 164
 165 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 166 {
 167         struct inode *inode;
 168         struct dentry *dentry;
 169         struct fd f;
 170         int error;
 171
 172         error = -EINVAL;
 173         if (length < 0)
 174                 goto out;
 175         error = -EBADF;
 176         f = fdget(fd);
 177         if (!f.file)
 178                 goto out;
 179
 180         /* explicitly opened as large or we are on 64-bit box */
 181         if (f.file->f_flags & O_LARGEFILE)
 182                 small = 0;
 183
 184         dentry = f.file->f_path.dentry;
 185         inode = dentry->d_inode;
 186         error = -EINVAL;
 187         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 188                 goto out_putf;
 189
 190         error = -EINVAL;
 191         /* Cannot ftruncate over 2^31 bytes without large file support */
 192         if (small && length > MAX_NON_LFS)
 193                 goto out_putf;
 194
 195         error = -EPERM;
 196         /* Check IS_APPEND on real upper inode */
 197         if (IS_APPEND(file_inode(f.file)))
 198                 goto out_putf;
 199
 200         sb_start_write(inode->i_sb);
 201         error = locks_verify_truncate(inode, f.file, length);
 202         if (!error)
 203                 error = security_path_truncate(&f.file->f_path);
 204         if (!error)
 205                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 206         sb_end_write(inode->i_sb);
 207 out_putf:
 208         fdput(f);
 209 out:
 210         return error;
 211 }
 212
 213 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 214 {
 215         return do_sys_ftruncate(fd, length, 1);
 216 }
 217
 218 #ifdef CONFIG_COMPAT
 219 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 220 {
 221         return do_sys_ftruncate(fd, length, 1);
 222 }
 223 #endif
 224
 225 /* LFS versions of truncate are only needed on 32 bit machines */
 226 #if BITS_PER_LONG == 32
 227 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 228 {
 229         return do_sys_truncate(path, length);
 230 }
 231
 232 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 233 {
 234         return do_sys_ftruncate(fd, length, 0);
 235 }
 236 #endif /* BITS_PER_LONG == 32 */
 237
 238
 239 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 240 {
 241         struct inode *inode = file_inode(file);
 242         long ret;
 243
 244         if (offset < 0 || len <= 0)
 245                 return -EINVAL;
 246
 247         /* Return error if mode is not supported */
 248         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 249                 return -EOPNOTSUPP;
 250
 251         /* Punch hole and zero range are mutually exclusive */
 252         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 253             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 254                 return -EOPNOTSUPP;
 255
 256         /* Punch hole must have keep size set */
 257         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 258             !(mode & FALLOC_FL_KEEP_SIZE))
 259                 return -EOPNOTSUPP;
 260
 261         /* Collapse range should only be used exclusively. */
 262         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 263             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 264                 return -EINVAL;
 265
 266         /* Insert range should only be used exclusively. */
 267         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 268             (mode & ~FALLOC_FL_INSERT_RANGE))
 269                 return -EINVAL;
 270
 271         /* Unshare range should only be used with allocate mode. */
 272         if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 273             (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 274                 return -EINVAL;
 275
 276         if (!(file->f_mode & FMODE_WRITE))
 277                 return -EBADF;
 278
 279         /*
 280          * We can only allow pure fallocate on append only files
 281          */
 282         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 283                 return -EPERM;
 284
 285         if (IS_IMMUTABLE(inode))
 286                 return -EPERM;
 287
 288         /*
 289          * We cannot allow any fallocate operation on an active swapfile
 290          */
 291         if (IS_SWAPFILE(inode))
 292                 return -ETXTBSY;
 293
 294         /*
 295          * Revalidate the write permissions, in case security policy has
 296          * changed since the files were opened.
 297          */
 298         ret = security_file_permission(file, MAY_WRITE);
 299         if (ret)
 300                 return ret;
 301
 302         if (S_ISFIFO(inode->i_mode))
 303                 return -ESPIPE;
 304
 305         if (S_ISDIR(inode->i_mode))
 306                 return -EISDIR;
 307
 308         if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 309                 return -ENODEV;
 310
 311         /* Check for wrap through zero too */
 312         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 313                 return -EFBIG;
 314
 315         if (!file->f_op->fallocate)
 316                 return -EOPNOTSUPP;
 317
 318         file_start_write(file);
 319         ret = file->f_op->fallocate(file, mode, offset, len);
 320
 321         /*
 322          * Create inotify and fanotify events.
 323          *
 324          * To keep the logic simple always create events if fallocate succeeds.
 325          * This implies that events are even created if the file size remains
 326          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 327          */
 328         if (ret == 0)
 329                 fsnotify_modify(file);
 330
 331         file_end_write(file);
 332         return ret;
 333 }
 334 EXPORT_SYMBOL_GPL(vfs_fallocate);
 335
 336 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 337 {
 338         struct fd f = fdget(fd);
 339         int error = -EBADF;
 340
 341         if (f.file) {
 342                 error = vfs_fallocate(f.file, mode, offset, len);
 343                 fdput(f);
 344         }
 345         return error;
 346 }
 347
 348 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 349 {
 350         return ksys_fallocate(fd, mode, offset, len);
 351 }
 352
 353 /*
 354  * access() needs to use the real uid/gid, not the effective uid/gid.
 355  * We do this by temporarily clearing all FS-related capabilities and
 356  * switching the fsuid/fsgid around to the real ones.
 357  */
 358 long do_faccessat(int dfd, const char __user *filename, int mode)
 359 {
 360         const struct cred *old_cred;
 361         struct cred *override_cred;
 362         struct path path;
 363         struct inode *inode;
 364         int res;
 365         unsigned int lookup_flags = LOOKUP_FOLLOW;
 366
 367         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 368                 return -EINVAL;
 369
 370         override_cred = prepare_creds();
 371         if (!override_cred)
 372                 return -ENOMEM;
 373
 374         override_cred->fsuid = override_cred->uid;
 375         override_cred->fsgid = override_cred->gid;
 376
 377         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 378                 /* Clear the capabilities if we switch to a non-root user */
 379                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 380                 if (!uid_eq(override_cred->uid, root_uid))
 381                         cap_clear(override_cred->cap_effective);
 382                 else
 383                         override_cred->cap_effective =
 384                                 override_cred->cap_permitted;
 385         }
 386
 387         old_cred = override_creds(override_cred);
 388 retry:
 389         res = user_path_at(dfd, filename, lookup_flags, &path);
 390         if (res)
 391                 goto out;
 392
 393         inode = d_backing_inode(path.dentry);
 394
 395         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 396                 /*
 397                  * MAY_EXEC on regular files is denied if the fs is mounted
 398                  * with the "noexec" flag.
 399                  */
 400                 res = -EACCES;
 401                 if (path_noexec(&path))
 402                         goto out_path_release;
 403         }
 404
 405         res = inode_permission(inode, mode | MAY_ACCESS);
 406         /* SuS v2 requires we report a read only fs too */
 407         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 408                 goto out_path_release;
 409         /*
 410          * This is a rare case where using __mnt_is_readonly()
 411          * is OK without a mnt_want/drop_write() pair.  Since
 412          * no actual write to the fs is performed here, we do
 413          * not need to telegraph to that to anyone.
 414          *
 415          * By doing this, we accept that this access is
 416          * inherently racy and know that the fs may change
 417          * state before we even see this result.
 418          */
 419         if (__mnt_is_readonly(path.mnt))
 420                 res = -EROFS;
 421
 422 out_path_release:
 423         path_put(&path);
 424         if (retry_estale(res, lookup_flags)) {
 425                 lookup_flags |= LOOKUP_REVAL;
 426                 goto retry;
 427         }
 428 out:
 429         revert_creds(old_cred);
 430         put_cred(override_cred);
 431         return res;
 432 }
 433
 434 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 435 {
 436         return do_faccessat(dfd, filename, mode);
 437 }
 438
 439 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 440 {
 441         return do_faccessat(AT_FDCWD, filename, mode);
 442 }
 443
 444 int ksys_chdir(const char __user *filename)
 445 {
 446         struct path path;
 447         int error;
 448         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 449 retry:
 450         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 451         if (error)
 452                 goto out;
 453
 454         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 455         if (error)
 456                 goto dput_and_out;
 457
 458         set_fs_pwd(current->fs, &path);
 459
 460 dput_and_out:
 461         path_put(&path);
 462         if (retry_estale(error, lookup_flags)) {
 463                 lookup_flags |= LOOKUP_REVAL;
 464                 goto retry;
 465         }
 466 out:
 467         return error;
 468 }
 469
 470 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 471 {
 472         return ksys_chdir(filename);
 473 }
 474
 475 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 476 {
 477         struct fd f = fdget_raw(fd);
 478         int error;
 479
 480         error = -EBADF;
 481         if (!f.file)
 482                 goto out;
 483
 484         error = -ENOTDIR;
 485         if (!d_can_lookup(f.file->f_path.dentry))
 486                 goto out_putf;
 487
 488         error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 489         if (!error)
 490                 set_fs_pwd(current->fs, &f.file->f_path);
 491 out_putf:
 492         fdput(f);
 493 out:
 494         return error;
 495 }
 496
 497 int ksys_chroot(const char __user *filename)
 498 {
 499         struct path path;
 500         int error;
 501         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 502 retry:
 503         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 504         if (error)
 505                 goto out;
 506
 507         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 508         if (error)
 509                 goto dput_and_out;
 510
 511         error = -EPERM;
 512         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 513                 goto dput_and_out;
 514         error = security_path_chroot(&path);
 515         if (error)
 516                 goto dput_and_out;
 517
 518         set_fs_root(current->fs, &path);
 519         error = 0;
 520 dput_and_out:
 521         path_put(&path);
 522         if (retry_estale(error, lookup_flags)) {
 523                 lookup_flags |= LOOKUP_REVAL;
 524                 goto retry;
 525         }
 526 out:
 527         return error;
 528 }
 529
 530 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 531 {
 532         return ksys_chroot(filename);
 533 }
 534
 535 static int chmod_common(const struct path *path, umode_t mode)
 536 {
 537         struct inode *inode = path->dentry->d_inode;
 538         struct inode *delegated_inode = NULL;
 539         struct iattr newattrs;
 540         int error;
 541
 542         error = mnt_want_write(path->mnt);
 543         if (error)
 544                 return error;
 545 retry_deleg:
 546         inode_lock(inode);
 547         error = security_path_chmod(path, mode);
 548         if (error)
 549                 goto out_unlock;
 550         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 551         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 552         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 553 out_unlock:
 554         inode_unlock(inode);
 555         if (delegated_inode) {
 556                 error = break_deleg_wait(&delegated_inode);
 557                 if (!error)
 558                         goto retry_deleg;
 559         }
 560         mnt_drop_write(path->mnt);
 561         return error;
 562 }
 563
 564 int ksys_fchmod(unsigned int fd, umode_t mode)
 565 {
 566         struct fd f = fdget(fd);
 567         int err = -EBADF;
 568
 569         if (f.file) {
 570                 audit_file(f.file);
 571                 err = chmod_common(&f.file->f_path, mode);
 572                 fdput(f);
 573         }
 574         return err;
 575 }
 576
 577 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 578 {
 579         return ksys_fchmod(fd, mode);
 580 }
 581
 582 int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
 583 {
 584         struct path path;
 585         int error;
 586         unsigned int lookup_flags = LOOKUP_FOLLOW;
 587 retry:
 588         error = user_path_at(dfd, filename, lookup_flags, &path);
 589         if (!error) {
 590                 error = chmod_common(&path, mode);
 591                 path_put(&path);
 592                 if (retry_estale(error, lookup_flags)) {
 593                         lookup_flags |= LOOKUP_REVAL;
 594                         goto retry;
 595                 }
 596         }
 597         return error;
 598 }
 599
 600 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 601                 umode_t, mode)
 602 {
 603         return do_fchmodat(dfd, filename, mode);
 604 }
 605
 606 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 607 {
 608         return do_fchmodat(AT_FDCWD, filename, mode);
 609 }
 610
 611 static int chown_common(const struct path *path, uid_t user, gid_t group)
 612 {
 613         struct inode *inode = path->dentry->d_inode;
 614         struct inode *delegated_inode = NULL;
 615         int error;
 616         struct iattr newattrs;
 617         kuid_t uid;
 618         kgid_t gid;
 619
 620         uid = make_kuid(current_user_ns(), user);
 621         gid = make_kgid(current_user_ns(), group);
 622
 623 retry_deleg:
 624         newattrs.ia_valid =  ATTR_CTIME;
 625         if (user != (uid_t) -1) {
 626                 if (!uid_valid(uid))
 627                         return -EINVAL;
 628                 newattrs.ia_valid |= ATTR_UID;
 629                 newattrs.ia_uid = uid;
 630         }
 631         if (group != (gid_t) -1) {
 632                 if (!gid_valid(gid))
 633                         return -EINVAL;
 634                 newattrs.ia_valid |= ATTR_GID;
 635                 newattrs.ia_gid = gid;
 636         }
 637         if (!S_ISDIR(inode->i_mode))
 638                 newattrs.ia_valid |=
 639                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 640         inode_lock(inode);
 641         error = security_path_chown(path, uid, gid);
 642         if (!error)
 643                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 644         inode_unlock(inode);
 645         if (delegated_inode) {
 646                 error = break_deleg_wait(&delegated_inode);
 647                 if (!error)
 648                         goto retry_deleg;
 649         }
 650         return error;
 651 }
 652
 653 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 654                 int flag)
 655 {
 656         struct path path;
 657         int error = -EINVAL;
 658         int lookup_flags;
 659
 660         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 661                 goto out;
 662
 663         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 664         if (flag & AT_EMPTY_PATH)
 665                 lookup_flags |= LOOKUP_EMPTY;
 666 retry:
 667         error = user_path_at(dfd, filename, lookup_flags, &path);
 668         if (error)
 669                 goto out;
 670         error = mnt_want_write(path.mnt);
 671         if (error)
 672                 goto out_release;
 673         error = chown_common(&path, user, group);
 674         mnt_drop_write(path.mnt);
 675 out_release:
 676         path_put(&path);
 677         if (retry_estale(error, lookup_flags)) {
 678                 lookup_flags |= LOOKUP_REVAL;
 679                 goto retry;
 680         }
 681 out:
 682         return error;
 683 }
 684
 685 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 686                 gid_t, group, int, flag)
 687 {
 688         return do_fchownat(dfd, filename, user, group, flag);
 689 }
 690
 691 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 692 {
 693         return do_fchownat(AT_FDCWD, filename, user, group, 0);
 694 }
 695
 696 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 697 {
 698         return do_fchownat(AT_FDCWD, filename, user, group,
 699                            AT_SYMLINK_NOFOLLOW);
 700 }
 701
 702 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 703 {
 704         struct fd f = fdget(fd);
 705         int error = -EBADF;
 706
 707         if (!f.file)
 708                 goto out;
 709
 710         error = mnt_want_write_file_path(f.file);
 711         if (error)
 712                 goto out_fput;
 713         audit_file(f.file);
 714         error = chown_common(&f.file->f_path, user, group);
 715         mnt_drop_write_file_path(f.file);
 716 out_fput:
 717         fdput(f);
 718 out:
 719         return error;
 720 }
 721
 722 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 723 {
 724         return ksys_fchown(fd, user, group);
 725 }
 726
 727 int open_check_o_direct(struct file *f)
 728 {
 729         /* NB: we're sure to have correct a_ops only after f_op->open */
 730         if (f->f_flags & O_DIRECT) {
 731                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 732                         return -EINVAL;
 733         }
 734         return 0;
 735 }
 736
 737 static int do_dentry_open(struct file *f,
 738                           struct inode *inode,
 739                           int (*open)(struct inode *, struct file *),
 740                           const struct cred *cred)
 741 {
 742         static const struct file_operations empty_fops = {};
 743         int error;
 744
 745         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 746                                 FMODE_PREAD | FMODE_PWRITE;
 747
 748         path_get(&f->f_path);
 749         f->f_inode = inode;
 750         f->f_mapping = inode->i_mapping;
 751
 752         /* Ensure that we skip any errors that predate opening of the file */
 753         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 754
 755         if (unlikely(f->f_flags & O_PATH)) {
 756                 f->f_mode = FMODE_PATH;
 757                 f->f_op = &empty_fops;
 758                 return 0;
 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, cred);
 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         if (!open)
 792                 open = f->f_op->open;
 793         if (open) {
 794                 error = open(inode, f);
 795                 if (error)
 796                         goto cleanup_all;
 797         }
 798         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 799                 i_readcount_inc(inode);
 800         if ((f->f_mode & FMODE_READ) &&
 801              likely(f->f_op->read || f->f_op->read_iter))
 802                 f->f_mode |= FMODE_CAN_READ;
 803         if ((f->f_mode & FMODE_WRITE) &&
 804              likely(f->f_op->write || f->f_op->write_iter))
 805                 f->f_mode |= FMODE_CAN_WRITE;
 806
 807         f->f_write_hint = WRITE_LIFE_NOT_SET;
 808         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 809
 810         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 811
 812         return 0;
 813
 814 cleanup_all:
 815         fops_put(f->f_op);
 816         if (f->f_mode & FMODE_WRITER) {
 817                 put_write_access(inode);
 818                 __mnt_drop_write(f->f_path.mnt);
 819         }
 820 cleanup_file:
 821         path_put(&f->f_path);
 822         f->f_path.mnt = NULL;
 823         f->f_path.dentry = NULL;
 824         f->f_inode = NULL;
 825         return error;
 826 }
 827
 828 /**
 829  * finish_open - finish opening a file
 830  * @file: file pointer
 831  * @dentry: pointer to dentry
 832  * @open: open callback
 833  * @opened: state of open
 834  *
 835  * This can be used to finish opening a file passed to i_op->atomic_open().
 836  *
 837  * If the open callback is set to NULL, then the standard f_op->open()
 838  * filesystem callback is substituted.
 839  *
 840  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 841  * the return value of d_splice_alias(), then the caller needs to perform dput()
 842  * on it after finish_open().
 843  *
 844  * On successful return @file is a fully instantiated open file.  After this, if
 845  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 846  *
 847  * Returns zero on success or -errno if the open failed.
 848  */
 849 int finish_open(struct file *file, struct dentry *dentry,
 850                 int (*open)(struct inode *, struct file *),
 851                 int *opened)
 852 {
 853         int error;
 854         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 855
 856         file->f_path.dentry = dentry;
 857         error = do_dentry_open(file, d_backing_inode(dentry), open,
 858                                current_cred());
 859         if (!error)
 860                 *opened |= FILE_OPENED;
 861
 862         return error;
 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 "1" 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 1;
 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              const struct cred *cred)
 901 {
 902         struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags, 0);
 903
 904         if (IS_ERR(dentry))
 905                 return PTR_ERR(dentry);
 906
 907         file->f_path = *path;
 908         return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 909 }
 910
 911 struct file *dentry_open(const struct path *path, int flags,
 912                          const struct cred *cred)
 913 {
 914         int error;
 915         struct file *f;
 916
 917         validate_creds(cred);
 918
 919         /* We must always pass in a valid mount pointer. */
 920         BUG_ON(!path->mnt);
 921
 922         f = get_empty_filp();
 923         if (!IS_ERR(f)) {
 924                 f->f_flags = flags;
 925                 error = vfs_open(path, f, cred);
 926                 if (!error) {
 927                         /* from now on we need fput() to dispose of f */
 928                         error = open_check_o_direct(f);
 929                         if (error) {
 930                                 fput(f);
 931                                 f = ERR_PTR(error);
 932                         }
 933                 } else {
 934                         put_filp(f);
 935                         f = ERR_PTR(error);
 936                 }
 937         }
 938         return f;
 939 }
 940 EXPORT_SYMBOL(dentry_open);
 941
 942 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 943 {
 944         int lookup_flags = 0;
 945         int acc_mode = ACC_MODE(flags);
 946
 947         /*
 948          * Clear out all open flags we don't know about so that we don't report
 949          * them in fcntl(F_GETFD) or similar interfaces.
 950          */
 951         flags &= VALID_OPEN_FLAGS;
 952
 953         if (flags & (O_CREAT | __O_TMPFILE))
 954                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 955         else
 956                 op->mode = 0;
 957
 958         /* Must never be set by userspace */
 959         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 960
 961         /*
 962          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 963          * check for O_DSYNC if the need any syncing at all we enforce it's
 964          * always set instead of having to deal with possibly weird behaviour
 965          * for malicious applications setting only __O_SYNC.
 966          */
 967         if (flags & __O_SYNC)
 968                 flags |= O_DSYNC;
 969
 970         if (flags & __O_TMPFILE) {
 971                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 972                         return -EINVAL;
 973                 if (!(acc_mode & MAY_WRITE))
 974                         return -EINVAL;
 975         } else if (flags & O_PATH) {
 976                 /*
 977                  * If we have O_PATH in the open flag. Then we
 978                  * cannot have anything other than the below set of flags
 979                  */
 980                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 981                 acc_mode = 0;
 982         }
 983
 984         op->open_flag = flags;
 985
 986         /* O_TRUNC implies we need access checks for write permissions */
 987         if (flags & O_TRUNC)
 988                 acc_mode |= MAY_WRITE;
 989
 990         /* Allow the LSM permission hook to distinguish append
 991            access from general write access. */
 992         if (flags & O_APPEND)
 993                 acc_mode |= MAY_APPEND;
 994
 995         op->acc_mode = acc_mode;
 996
 997         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 998
 999         if (flags & O_CREAT) {
1000                 op->intent |= LOOKUP_CREATE;
1001                 if (flags & O_EXCL)
1002                         op->intent |= LOOKUP_EXCL;
1003         }
1004
1005         if (flags & O_DIRECTORY)
1006                 lookup_flags |= LOOKUP_DIRECTORY;
1007         if (!(flags & O_NOFOLLOW))
1008                 lookup_flags |= LOOKUP_FOLLOW;
1009         op->lookup_flags = lookup_flags;
1010         return 0;
1011 }
1012
1013 /**
1014  * file_open_name - open file and return file pointer
1015  *
1016  * @name:       struct filename containing 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 *file_open_name(struct filename *name, int flags, umode_t mode)
1025 {
1026         struct open_flags op;
1027         int err = build_open_flags(flags, mode, &op);
1028         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
1029 }
1030
1031 /**
1032  * filp_open - open file and return file pointer
1033  *
1034  * @filename:   path to open
1035  * @flags:      open flags as per the open(2) second argument
1036  * @mode:       mode for the new file if O_CREAT is set, else ignored
1037  *
1038  * This is the helper to open a file from kernelspace if you really
1039  * have to.  But in generally you should not do this, so please move
1040  * along, nothing to see here..
1041  */
1042 struct file *filp_open(const char *filename, int flags, umode_t mode)
1043 {
1044         struct filename *name = getname_kernel(filename);
1045         struct file *file = ERR_CAST(name);
1046
1047         if (!IS_ERR(name)) {
1048                 file = file_open_name(name, flags, mode);
1049                 putname(name);
1050         }
1051         return file;
1052 }
1053 EXPORT_SYMBOL(filp_open);
1054
1055 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1056                             const char *filename, int flags, umode_t mode)
1057 {
1058         struct open_flags op;
1059         int err = build_open_flags(flags, mode, &op);
1060         if (err)
1061                 return ERR_PTR(err);
1062         return do_file_open_root(dentry, mnt, filename, &op);
1063 }
1064 EXPORT_SYMBOL(file_open_root);
1065
1066 struct file *filp_clone_open(struct file *oldfile)
1067 {
1068         struct file *file;
1069         int retval;
1070
1071         file = get_empty_filp();
1072         if (IS_ERR(file))
1073                 return file;
1074
1075         file->f_flags = oldfile->f_flags;
1076         retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1077         if (retval) {
1078                 put_filp(file);
1079                 return ERR_PTR(retval);
1080         }
1081
1082         return file;
1083 }
1084 EXPORT_SYMBOL(filp_clone_open);
1085
1086 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1087 {
1088         struct open_flags op;
1089         int fd = build_open_flags(flags, mode, &op);
1090         struct filename *tmp;
1091
1092         if (fd)
1093                 return fd;
1094
1095         tmp = getname(filename);
1096         if (IS_ERR(tmp))
1097                 return PTR_ERR(tmp);
1098
1099         fd = get_unused_fd_flags(flags);
1100         if (fd >= 0) {
1101                 struct file *f = do_filp_open(dfd, tmp, &op);
1102                 if (IS_ERR(f)) {
1103                         put_unused_fd(fd);
1104                         fd = PTR_ERR(f);
1105                 } else {
1106                         fsnotify_open(f);
1107                         fd_install(fd, f);
1108                 }
1109         }
1110         putname(tmp);
1111         return fd;
1112 }
1113
1114 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1115 {
1116         if (force_o_largefile())
1117                 flags |= O_LARGEFILE;
1118
1119         return do_sys_open(AT_FDCWD, filename, flags, mode);
1120 }
1121
1122 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1123                 umode_t, mode)
1124 {
1125         if (force_o_largefile())
1126                 flags |= O_LARGEFILE;
1127
1128         return do_sys_open(dfd, filename, flags, mode);
1129 }
1130
1131 #ifdef CONFIG_COMPAT
1132 /*
1133  * Exactly like sys_open(), except that it doesn't set the
1134  * O_LARGEFILE flag.
1135  */
1136 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1137 {
1138         return do_sys_open(AT_FDCWD, filename, flags, mode);
1139 }
1140
1141 /*
1142  * Exactly like sys_openat(), except that it doesn't set the
1143  * O_LARGEFILE flag.
1144  */
1145 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1146 {
1147         return do_sys_open(dfd, filename, flags, mode);
1148 }
1149 #endif
1150
1151 #ifndef __alpha__
1152
1153 /*
1154  * For backward compatibility?  Maybe this should be moved
1155  * into arch/i386 instead?
1156  */
1157 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1158 {
1159         return ksys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1160 }
1161
1162 #endif
1163
1164 /*
1165  * "id" is the POSIX thread ID. We use the
1166  * files pointer for this..
1167  */
1168 int filp_close(struct file *filp, fl_owner_t id)
1169 {
1170         int retval = 0;
1171
1172         if (!file_count(filp)) {
1173                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1174                 return 0;
1175         }
1176
1177         if (filp->f_op->flush)
1178                 retval = filp->f_op->flush(filp, id);
1179
1180         if (likely(!(filp->f_mode & FMODE_PATH))) {
1181                 dnotify_flush(filp, id);
1182                 locks_remove_posix(filp, id);
1183         }
1184         fput(filp);
1185         return retval;
1186 }
1187
1188 EXPORT_SYMBOL(filp_close);
1189
1190 /*
1191  * Careful here! We test whether the file pointer is NULL before
1192  * releasing the fd. This ensures that one clone task can't release
1193  * an fd while another clone is opening it.
1194  */
1195 SYSCALL_DEFINE1(close, unsigned int, fd)
1196 {
1197         int retval = __close_fd(current->files, fd);
1198
1199         /* can't restart close syscall because file table entry was cleared */
1200         if (unlikely(retval == -ERESTARTSYS ||
1201                      retval == -ERESTARTNOINTR ||
1202                      retval == -ERESTARTNOHAND ||
1203                      retval == -ERESTART_RESTARTBLOCK))
1204                 retval = -EINTR;
1205
1206         return retval;
1207 }
1208
1209 /*
1210  * This routine simulates a hangup on the tty, to arrange that users
1211  * are given clean terminals at login time.
1212  */
1213 SYSCALL_DEFINE0(vhangup)
1214 {
1215         if (capable(CAP_SYS_TTY_CONFIG)) {
1216                 tty_vhangup_self();
1217                 return 0;
1218         }
1219         return -EPERM;
1220 }
1221
1222 /*
1223  * Called when an inode is about to be open.
1224  * We use this to disallow opening large files on 32bit systems if
1225  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1226  * on this flag in sys_open.
1227  */
1228 int generic_file_open(struct inode * inode, struct file * filp)
1229 {
1230         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1231                 return -EOVERFLOW;
1232         return 0;
1233 }
1234
1235 EXPORT_SYMBOL(generic_file_open);
1236
1237 /*
1238  * This is used by subsystems that don't want seekable
1239  * file descriptors. The function is not supposed to ever fail, the only
1240  * reason it returns an 'int' and not 'void' is so that it can be plugged
1241  * directly into file_operations structure.
1242  */
1243 int nonseekable_open(struct inode *inode, struct file *filp)
1244 {
1245         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1246         return 0;
1247 }
1248
1249 EXPORT_SYMBOL(nonseekable_open);