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_truncate2(struct vfsmount *mnt, struct dentry *dentry, loff_t length,
  38                 unsigned int time_attrs, 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_change2(mnt, dentry, &newattrs, NULL);
  64         inode_unlock(dentry->d_inode);
  65         return ret;
  66 }
  67 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  68         struct file *filp)
  69 {
  70         return do_truncate2(NULL, dentry, length, time_attrs, filp);
  71 }
  72
  73 long vfs_truncate(const struct path *path, loff_t length)
  74 {
  75         struct inode *inode;
  76         struct vfsmount *mnt;
  77         struct dentry *upperdentry;
  78         long error;
  79
  80         inode = path->dentry->d_inode;
  81         mnt = path->mnt;
  82
  83         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  84         if (S_ISDIR(inode->i_mode))
  85                 return -EISDIR;
  86         if (!S_ISREG(inode->i_mode))
  87                 return -EINVAL;
  88
  89         error = mnt_want_write(path->mnt);
  90         if (error)
  91                 goto out;
  92
  93         error = inode_permission2(mnt, inode, MAY_WRITE);
  94         if (error)
  95                 goto mnt_drop_write_and_out;
  96
  97         error = -EPERM;
  98         if (IS_APPEND(inode))
  99                 goto mnt_drop_write_and_out;
 100
 101         /*
 102          * If this is an overlayfs then do as if opening the file so we get
 103          * write access on the upper inode, not on the overlay inode.  For
 104          * non-overlay filesystems d_real() is an identity function.
 105          */
 106         upperdentry = d_real(path->dentry, NULL, O_WRONLY, 0);
 107         error = PTR_ERR(upperdentry);
 108         if (IS_ERR(upperdentry))
 109                 goto mnt_drop_write_and_out;
 110
 111         error = get_write_access(upperdentry->d_inode);
 112         if (error)
 113                 goto mnt_drop_write_and_out;
 114
 115         /*
 116          * Make sure that there are no leases.  get_write_access() protects
 117          * against the truncate racing with a lease-granting setlease().
 118          */
 119         error = break_lease(inode, O_WRONLY);
 120         if (error)
 121                 goto put_write_and_out;
 122
 123         error = locks_verify_truncate(inode, NULL, length);
 124         if (!error)
 125                 error = security_path_truncate(path);
 126         if (!error)
 127                 error = do_truncate2(mnt, path->dentry, length, 0, NULL);
 128
 129 put_write_and_out:
 130         put_write_access(upperdentry->d_inode);
 131 mnt_drop_write_and_out:
 132         mnt_drop_write(path->mnt);
 133 out:
 134         return error;
 135 }
 136 EXPORT_SYMBOL_GPL(vfs_truncate);
 137
 138 static long do_sys_truncate(const char __user *pathname, loff_t length)
 139 {
 140         unsigned int lookup_flags = LOOKUP_FOLLOW;
 141         struct path path;
 142         int error;
 143
 144         if (length < 0) /* sorry, but loff_t says... */
 145                 return -EINVAL;
 146
 147 retry:
 148         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 149         if (!error) {
 150                 error = vfs_truncate(&path, length);
 151                 path_put(&path);
 152         }
 153         if (retry_estale(error, lookup_flags)) {
 154                 lookup_flags |= LOOKUP_REVAL;
 155                 goto retry;
 156         }
 157         return error;
 158 }
 159
 160 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 161 {
 162         return do_sys_truncate(path, length);
 163 }
 164
 165 #ifdef CONFIG_COMPAT
 166 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 167 {
 168         return do_sys_truncate(path, length);
 169 }
 170 #endif
 171
 172 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 173 {
 174         struct inode *inode;
 175         struct dentry *dentry;
 176         struct vfsmount *mnt;
 177         struct fd f;
 178         int error;
 179
 180         error = -EINVAL;
 181         if (length < 0)
 182                 goto out;
 183         error = -EBADF;
 184         f = fdget(fd);
 185         if (!f.file)
 186                 goto out;
 187
 188         /* explicitly opened as large or we are on 64-bit box */
 189         if (f.file->f_flags & O_LARGEFILE)
 190                 small = 0;
 191
 192         dentry = f.file->f_path.dentry;
 193         mnt = f.file->f_path.mnt;
 194         inode = dentry->d_inode;
 195         error = -EINVAL;
 196         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 197                 goto out_putf;
 198
 199         error = -EINVAL;
 200         /* Cannot ftruncate over 2^31 bytes without large file support */
 201         if (small && length > MAX_NON_LFS)
 202                 goto out_putf;
 203
 204         error = -EPERM;
 205         /* Check IS_APPEND on real upper inode */
 206         if (IS_APPEND(file_inode(f.file)))
 207                 goto out_putf;
 208
 209         sb_start_write(inode->i_sb);
 210         error = locks_verify_truncate(inode, f.file, length);
 211         if (!error)
 212                 error = security_path_truncate(&f.file->f_path);
 213         if (!error)
 214                 error = do_truncate2(mnt, dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 215         sb_end_write(inode->i_sb);
 216 out_putf:
 217         fdput(f);
 218 out:
 219         return error;
 220 }
 221
 222 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 223 {
 224         return do_sys_ftruncate(fd, length, 1);
 225 }
 226
 227 #ifdef CONFIG_COMPAT
 228 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 229 {
 230         return do_sys_ftruncate(fd, length, 1);
 231 }
 232 #endif
 233
 234 /* LFS versions of truncate are only needed on 32 bit machines */
 235 #if BITS_PER_LONG == 32
 236 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 237 {
 238         return do_sys_truncate(path, length);
 239 }
 240
 241 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 242 {
 243         return do_sys_ftruncate(fd, length, 0);
 244 }
 245 #endif /* BITS_PER_LONG == 32 */
 246
 247
 248 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 249 {
 250         struct inode *inode = file_inode(file);
 251         long ret;
 252
 253         if (offset < 0 || len <= 0)
 254                 return -EINVAL;
 255
 256         /* Return error if mode is not supported */
 257         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 258                 return -EOPNOTSUPP;
 259
 260         /* Punch hole and zero range are mutually exclusive */
 261         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 262             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 263                 return -EOPNOTSUPP;
 264
 265         /* Punch hole must have keep size set */
 266         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 267             !(mode & FALLOC_FL_KEEP_SIZE))
 268                 return -EOPNOTSUPP;
 269
 270         /* Collapse range should only be used exclusively. */
 271         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 272             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 273                 return -EINVAL;
 274
 275         /* Insert range should only be used exclusively. */
 276         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 277             (mode & ~FALLOC_FL_INSERT_RANGE))
 278                 return -EINVAL;
 279
 280         /* Unshare range should only be used with allocate mode. */
 281         if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 282             (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 283                 return -EINVAL;
 284
 285         if (!(file->f_mode & FMODE_WRITE))
 286                 return -EBADF;
 287
 288         /*
 289          * We can only allow pure fallocate on append only files
 290          */
 291         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 292                 return -EPERM;
 293
 294         if (IS_IMMUTABLE(inode))
 295                 return -EPERM;
 296
 297         /*
 298          * We cannot allow any fallocate operation on an active swapfile
 299          */
 300         if (IS_SWAPFILE(inode))
 301                 return -ETXTBSY;
 302
 303         /*
 304          * Revalidate the write permissions, in case security policy has
 305          * changed since the files were opened.
 306          */
 307         ret = security_file_permission(file, MAY_WRITE);
 308         if (ret)
 309                 return ret;
 310
 311         if (S_ISFIFO(inode->i_mode))
 312                 return -ESPIPE;
 313
 314         if (S_ISDIR(inode->i_mode))
 315                 return -EISDIR;
 316
 317         if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 318                 return -ENODEV;
 319
 320         /* Check for wrap through zero too */
 321         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 322                 return -EFBIG;
 323
 324         if (!file->f_op->fallocate)
 325                 return -EOPNOTSUPP;
 326
 327         file_start_write(file);
 328         ret = file->f_op->fallocate(file, mode, offset, len);
 329
 330         /*
 331          * Create inotify and fanotify events.
 332          *
 333          * To keep the logic simple always create events if fallocate succeeds.
 334          * This implies that events are even created if the file size remains
 335          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 336          */
 337         if (ret == 0)
 338                 fsnotify_modify(file);
 339
 340         file_end_write(file);
 341         return ret;
 342 }
 343 EXPORT_SYMBOL_GPL(vfs_fallocate);
 344
 345 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 346 {
 347         struct fd f = fdget(fd);
 348         int error = -EBADF;
 349
 350         if (f.file) {
 351                 error = vfs_fallocate(f.file, mode, offset, len);
 352                 fdput(f);
 353         }
 354         return error;
 355 }
 356
 357 /*
 358  * access() needs to use the real uid/gid, not the effective uid/gid.
 359  * We do this by temporarily clearing all FS-related capabilities and
 360  * switching the fsuid/fsgid around to the real ones.
 361  */
 362 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 363 {
 364         const struct cred *old_cred;
 365         struct cred *override_cred;
 366         struct path path;
 367         struct inode *inode;
 368         struct vfsmount *mnt;
 369         int res;
 370         unsigned int lookup_flags = LOOKUP_FOLLOW;
 371
 372         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 373                 return -EINVAL;
 374
 375         override_cred = prepare_creds();
 376         if (!override_cred)
 377                 return -ENOMEM;
 378
 379         override_cred->fsuid = override_cred->uid;
 380         override_cred->fsgid = override_cred->gid;
 381
 382         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 383                 /* Clear the capabilities if we switch to a non-root user */
 384                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 385                 if (!uid_eq(override_cred->uid, root_uid))
 386                         cap_clear(override_cred->cap_effective);
 387                 else
 388                         override_cred->cap_effective =
 389                                 override_cred->cap_permitted;
 390         }
 391
 392         old_cred = override_creds(override_cred);
 393 retry:
 394         res = user_path_at(dfd, filename, lookup_flags, &path);
 395         if (res)
 396                 goto out;
 397
 398         inode = d_backing_inode(path.dentry);
 399         mnt = path.mnt;
 400
 401         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 402                 /*
 403                  * MAY_EXEC on regular files is denied if the fs is mounted
 404                  * with the "noexec" flag.
 405                  */
 406                 res = -EACCES;
 407                 if (path_noexec(&path))
 408                         goto out_path_release;
 409         }
 410
 411         res = inode_permission2(mnt, inode, mode | MAY_ACCESS);
 412         /* SuS v2 requires we report a read only fs too */
 413         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 414                 goto out_path_release;
 415         /*
 416          * This is a rare case where using __mnt_is_readonly()
 417          * is OK without a mnt_want/drop_write() pair.  Since
 418          * no actual write to the fs is performed here, we do
 419          * not need to telegraph to that to anyone.
 420          *
 421          * By doing this, we accept that this access is
 422          * inherently racy and know that the fs may change
 423          * state before we even see this result.
 424          */
 425         if (__mnt_is_readonly(path.mnt))
 426                 res = -EROFS;
 427
 428 out_path_release:
 429         path_put(&path);
 430         if (retry_estale(res, lookup_flags)) {
 431                 lookup_flags |= LOOKUP_REVAL;
 432                 goto retry;
 433         }
 434 out:
 435         revert_creds(old_cred);
 436         put_cred(override_cred);
 437         return res;
 438 }
 439
 440 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 441 {
 442         return sys_faccessat(AT_FDCWD, filename, mode);
 443 }
 444
 445 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 446 {
 447         struct path path;
 448         int error;
 449         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 450 retry:
 451         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 452         if (error)
 453                 goto out;
 454
 455         error = inode_permission2(path.mnt, path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 456         if (error)
 457                 goto dput_and_out;
 458
 459         set_fs_pwd(current->fs, &path);
 460
 461 dput_and_out:
 462         path_put(&path);
 463         if (retry_estale(error, lookup_flags)) {
 464                 lookup_flags |= LOOKUP_REVAL;
 465                 goto retry;
 466         }
 467 out:
 468         return error;
 469 }
 470
 471 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 472 {
 473         struct fd f = fdget_raw(fd);
 474         int error;
 475
 476         error = -EBADF;
 477         if (!f.file)
 478                 goto out;
 479
 480         error = -ENOTDIR;
 481         if (!d_can_lookup(f.file->f_path.dentry))
 482                 goto out_putf;
 483
 484         error = inode_permission2(f.file->f_path.mnt, file_inode(f.file),
 485                                 MAY_EXEC | MAY_CHDIR);
 486         if (!error)
 487                 set_fs_pwd(current->fs, &f.file->f_path);
 488 out_putf:
 489         fdput(f);
 490 out:
 491         return error;
 492 }
 493
 494 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 495 {
 496         struct path path;
 497         int error;
 498         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 499 retry:
 500         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 501         if (error)
 502                 goto out;
 503
 504         error = inode_permission2(path.mnt, path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 505         if (error)
 506                 goto dput_and_out;
 507
 508         error = -EPERM;
 509         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 510                 goto dput_and_out;
 511         error = security_path_chroot(&path);
 512         if (error)
 513                 goto dput_and_out;
 514
 515         set_fs_root(current->fs, &path);
 516         error = 0;
 517 dput_and_out:
 518         path_put(&path);
 519         if (retry_estale(error, lookup_flags)) {
 520                 lookup_flags |= LOOKUP_REVAL;
 521                 goto retry;
 522         }
 523 out:
 524         return error;
 525 }
 526
 527 static int chmod_common(const struct path *path, umode_t mode)
 528 {
 529         struct inode *inode = path->dentry->d_inode;
 530         struct inode *delegated_inode = NULL;
 531         struct iattr newattrs;
 532         int error;
 533
 534         error = mnt_want_write(path->mnt);
 535         if (error)
 536                 return error;
 537 retry_deleg:
 538         inode_lock(inode);
 539         error = security_path_chmod(path, mode);
 540         if (error)
 541                 goto out_unlock;
 542         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 543         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 544         error = notify_change2(path->mnt, path->dentry, &newattrs, &delegated_inode);
 545 out_unlock:
 546         inode_unlock(inode);
 547         if (delegated_inode) {
 548                 error = break_deleg_wait(&delegated_inode);
 549                 if (!error)
 550                         goto retry_deleg;
 551         }
 552         mnt_drop_write(path->mnt);
 553         return error;
 554 }
 555
 556 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 557 {
 558         struct fd f = fdget(fd);
 559         int err = -EBADF;
 560
 561         if (f.file) {
 562                 audit_file(f.file);
 563                 err = chmod_common(&f.file->f_path, mode);
 564                 fdput(f);
 565         }
 566         return err;
 567 }
 568
 569 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 570 {
 571         struct path path;
 572         int error;
 573         unsigned int lookup_flags = LOOKUP_FOLLOW;
 574 retry:
 575         error = user_path_at(dfd, filename, lookup_flags, &path);
 576         if (!error) {
 577                 error = chmod_common(&path, mode);
 578                 path_put(&path);
 579                 if (retry_estale(error, lookup_flags)) {
 580                         lookup_flags |= LOOKUP_REVAL;
 581                         goto retry;
 582                 }
 583         }
 584         return error;
 585 }
 586
 587 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 588 {
 589         return sys_fchmodat(AT_FDCWD, filename, mode);
 590 }
 591
 592 static int chown_common(const struct path *path, uid_t user, gid_t group)
 593 {
 594         struct inode *inode = path->dentry->d_inode;
 595         struct inode *delegated_inode = NULL;
 596         int error;
 597         struct iattr newattrs;
 598         kuid_t uid;
 599         kgid_t gid;
 600
 601         uid = make_kuid(current_user_ns(), user);
 602         gid = make_kgid(current_user_ns(), group);
 603
 604 retry_deleg:
 605         newattrs.ia_valid =  ATTR_CTIME;
 606         if (user != (uid_t) -1) {
 607                 if (!uid_valid(uid))
 608                         return -EINVAL;
 609                 newattrs.ia_valid |= ATTR_UID;
 610                 newattrs.ia_uid = uid;
 611         }
 612         if (group != (gid_t) -1) {
 613                 if (!gid_valid(gid))
 614                         return -EINVAL;
 615                 newattrs.ia_valid |= ATTR_GID;
 616                 newattrs.ia_gid = gid;
 617         }
 618         if (!S_ISDIR(inode->i_mode))
 619                 newattrs.ia_valid |=
 620                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 621         inode_lock(inode);
 622         error = security_path_chown(path, uid, gid);
 623         if (!error)
 624                 error = notify_change2(path->mnt, path->dentry, &newattrs, &delegated_inode);
 625         inode_unlock(inode);
 626         if (delegated_inode) {
 627                 error = break_deleg_wait(&delegated_inode);
 628                 if (!error)
 629                         goto retry_deleg;
 630         }
 631         return error;
 632 }
 633
 634 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 635                 gid_t, group, int, flag)
 636 {
 637         struct path path;
 638         int error = -EINVAL;
 639         int lookup_flags;
 640
 641         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 642                 goto out;
 643
 644         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 645         if (flag & AT_EMPTY_PATH)
 646                 lookup_flags |= LOOKUP_EMPTY;
 647 retry:
 648         error = user_path_at(dfd, filename, lookup_flags, &path);
 649         if (error)
 650                 goto out;
 651         error = mnt_want_write(path.mnt);
 652         if (error)
 653                 goto out_release;
 654         error = chown_common(&path, user, group);
 655         mnt_drop_write(path.mnt);
 656 out_release:
 657         path_put(&path);
 658         if (retry_estale(error, lookup_flags)) {
 659                 lookup_flags |= LOOKUP_REVAL;
 660                 goto retry;
 661         }
 662 out:
 663         return error;
 664 }
 665
 666 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 667 {
 668         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 669 }
 670
 671 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 672 {
 673         return sys_fchownat(AT_FDCWD, filename, user, group,
 674                             AT_SYMLINK_NOFOLLOW);
 675 }
 676
 677 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 678 {
 679         struct fd f = fdget(fd);
 680         int error = -EBADF;
 681
 682         if (!f.file)
 683                 goto out;
 684
 685         error = mnt_want_write_file_path(f.file);
 686         if (error)
 687                 goto out_fput;
 688         audit_file(f.file);
 689         error = chown_common(&f.file->f_path, user, group);
 690         mnt_drop_write_file_path(f.file);
 691 out_fput:
 692         fdput(f);
 693 out:
 694         return error;
 695 }
 696
 697 int open_check_o_direct(struct file *f)
 698 {
 699         /* NB: we're sure to have correct a_ops only after f_op->open */
 700         if (f->f_flags & O_DIRECT) {
 701                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 702                         return -EINVAL;
 703         }
 704         return 0;
 705 }
 706
 707 static int do_dentry_open(struct file *f,
 708                           struct inode *inode,
 709                           int (*open)(struct inode *, struct file *),
 710                           const struct cred *cred)
 711 {
 712         static const struct file_operations empty_fops = {};
 713         int error;
 714
 715         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 716                                 FMODE_PREAD | FMODE_PWRITE;
 717
 718         path_get(&f->f_path);
 719         f->f_inode = inode;
 720         f->f_mapping = inode->i_mapping;
 721
 722         /* Ensure that we skip any errors that predate opening of the file */
 723         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 724
 725         if (unlikely(f->f_flags & O_PATH)) {
 726                 f->f_mode = FMODE_PATH;
 727                 f->f_op = &empty_fops;
 728                 return 0;
 729         }
 730
 731         /* Any file opened for execve()/uselib() has to be a regular file. */
 732         if (unlikely(f->f_flags & FMODE_EXEC && !S_ISREG(inode->i_mode))) {
 733                 error = -EACCES;
 734                 goto cleanup_file;
 735         }
 736
 737         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 738                 error = get_write_access(inode);
 739                 if (unlikely(error))
 740                         goto cleanup_file;
 741                 error = __mnt_want_write(f->f_path.mnt);
 742                 if (unlikely(error)) {
 743                         put_write_access(inode);
 744                         goto cleanup_file;
 745                 }
 746                 f->f_mode |= FMODE_WRITER;
 747         }
 748
 749         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 750         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 751                 f->f_mode |= FMODE_ATOMIC_POS;
 752
 753         f->f_op = fops_get(inode->i_fop);
 754         if (unlikely(WARN_ON(!f->f_op))) {
 755                 error = -ENODEV;
 756                 goto cleanup_all;
 757         }
 758
 759         error = security_file_open(f, cred);
 760         if (error)
 761                 goto cleanup_all;
 762
 763         error = break_lease(locks_inode(f), f->f_flags);
 764         if (error)
 765                 goto cleanup_all;
 766
 767         if (!open)
 768                 open = f->f_op->open;
 769         if (open) {
 770                 error = open(inode, f);
 771                 if (error)
 772                         goto cleanup_all;
 773         }
 774         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 775                 i_readcount_inc(inode);
 776         if ((f->f_mode & FMODE_READ) &&
 777              likely(f->f_op->read || f->f_op->read_iter))
 778                 f->f_mode |= FMODE_CAN_READ;
 779         if ((f->f_mode & FMODE_WRITE) &&
 780              likely(f->f_op->write || f->f_op->write_iter))
 781                 f->f_mode |= FMODE_CAN_WRITE;
 782
 783         f->f_write_hint = WRITE_LIFE_NOT_SET;
 784         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 785
 786         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 787
 788         return 0;
 789
 790 cleanup_all:
 791         fops_put(f->f_op);
 792         if (f->f_mode & FMODE_WRITER) {
 793                 put_write_access(inode);
 794                 __mnt_drop_write(f->f_path.mnt);
 795         }
 796 cleanup_file:
 797         path_put(&f->f_path);
 798         f->f_path.mnt = NULL;
 799         f->f_path.dentry = NULL;
 800         f->f_inode = NULL;
 801         return error;
 802 }
 803
 804 /**
 805  * finish_open - finish opening a file
 806  * @file: file pointer
 807  * @dentry: pointer to dentry
 808  * @open: open callback
 809  * @opened: state of open
 810  *
 811  * This can be used to finish opening a file passed to i_op->atomic_open().
 812  *
 813  * If the open callback is set to NULL, then the standard f_op->open()
 814  * filesystem callback is substituted.
 815  *
 816  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 817  * the return value of d_splice_alias(), then the caller needs to perform dput()
 818  * on it after finish_open().
 819  *
 820  * On successful return @file is a fully instantiated open file.  After this, if
 821  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 822  *
 823  * Returns zero on success or -errno if the open failed.
 824  */
 825 int finish_open(struct file *file, struct dentry *dentry,
 826                 int (*open)(struct inode *, struct file *),
 827                 int *opened)
 828 {
 829         int error;
 830         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 831
 832         file->f_path.dentry = dentry;
 833         error = do_dentry_open(file, d_backing_inode(dentry), open,
 834                                current_cred());
 835         if (!error)
 836                 *opened |= FILE_OPENED;
 837
 838         return error;
 839 }
 840 EXPORT_SYMBOL(finish_open);
 841
 842 /**
 843  * finish_no_open - finish ->atomic_open() without opening the file
 844  *
 845  * @file: file pointer
 846  * @dentry: dentry or NULL (as returned from ->lookup())
 847  *
 848  * This can be used to set the result of a successful lookup in ->atomic_open().
 849  *
 850  * NB: unlike finish_open() this function does consume the dentry reference and
 851  * the caller need not dput() it.
 852  *
 853  * Returns "1" which must be the return value of ->atomic_open() after having
 854  * called this function.
 855  */
 856 int finish_no_open(struct file *file, struct dentry *dentry)
 857 {
 858         file->f_path.dentry = dentry;
 859         return 1;
 860 }
 861 EXPORT_SYMBOL(finish_no_open);
 862
 863 char *file_path(struct file *filp, char *buf, int buflen)
 864 {
 865         return d_path(&filp->f_path, buf, buflen);
 866 }
 867 EXPORT_SYMBOL(file_path);
 868
 869 /**
 870  * vfs_open - open the file at the given path
 871  * @path: path to open
 872  * @file: newly allocated file with f_flag initialized
 873  * @cred: credentials to use
 874  */
 875 int vfs_open(const struct path *path, struct file *file,
 876              const struct cred *cred)
 877 {
 878         struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags, 0);
 879
 880         if (IS_ERR(dentry))
 881                 return PTR_ERR(dentry);
 882
 883         file->f_path = *path;
 884         return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 885 }
 886
 887 struct file *dentry_open(const struct path *path, int flags,
 888                          const struct cred *cred)
 889 {
 890         int error;
 891         struct file *f;
 892
 893         validate_creds(cred);
 894
 895         /* We must always pass in a valid mount pointer. */
 896         BUG_ON(!path->mnt);
 897
 898         f = get_empty_filp();
 899         if (!IS_ERR(f)) {
 900                 f->f_flags = flags;
 901                 error = vfs_open(path, f, cred);
 902                 if (!error) {
 903                         /* from now on we need fput() to dispose of f */
 904                         error = open_check_o_direct(f);
 905                         if (error) {
 906                                 fput(f);
 907                                 f = ERR_PTR(error);
 908                         }
 909                 } else {
 910                         put_filp(f);
 911                         f = ERR_PTR(error);
 912                 }
 913         }
 914         return f;
 915 }
 916 EXPORT_SYMBOL(dentry_open);
 917
 918 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 919 {
 920         int lookup_flags = 0;
 921         int acc_mode = ACC_MODE(flags);
 922
 923         /*
 924          * Clear out all open flags we don't know about so that we don't report
 925          * them in fcntl(F_GETFD) or similar interfaces.
 926          */
 927         flags &= VALID_OPEN_FLAGS;
 928
 929         if (flags & (O_CREAT | __O_TMPFILE))
 930                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 931         else
 932                 op->mode = 0;
 933
 934         /* Must never be set by userspace */
 935         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 936
 937         /*
 938          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 939          * check for O_DSYNC if the need any syncing at all we enforce it's
 940          * always set instead of having to deal with possibly weird behaviour
 941          * for malicious applications setting only __O_SYNC.
 942          */
 943         if (flags & __O_SYNC)
 944                 flags |= O_DSYNC;
 945
 946         if (flags & __O_TMPFILE) {
 947                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 948                         return -EINVAL;
 949                 if (!(acc_mode & MAY_WRITE))
 950                         return -EINVAL;
 951         } else if (flags & O_PATH) {
 952                 /*
 953                  * If we have O_PATH in the open flag. Then we
 954                  * cannot have anything other than the below set of flags
 955                  */
 956                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 957                 acc_mode = 0;
 958         }
 959
 960         op->open_flag = flags;
 961
 962         /* O_TRUNC implies we need access checks for write permissions */
 963         if (flags & O_TRUNC)
 964                 acc_mode |= MAY_WRITE;
 965
 966         /* Allow the LSM permission hook to distinguish append
 967            access from general write access. */
 968         if (flags & O_APPEND)
 969                 acc_mode |= MAY_APPEND;
 970
 971         op->acc_mode = acc_mode;
 972
 973         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 974
 975         if (flags & O_CREAT) {
 976                 op->intent |= LOOKUP_CREATE;
 977                 if (flags & O_EXCL)
 978                         op->intent |= LOOKUP_EXCL;
 979         }
 980
 981         if (flags & O_DIRECTORY)
 982                 lookup_flags |= LOOKUP_DIRECTORY;
 983         if (!(flags & O_NOFOLLOW))
 984                 lookup_flags |= LOOKUP_FOLLOW;
 985         op->lookup_flags = lookup_flags;
 986         return 0;
 987 }
 988
 989 /**
 990  * file_open_name - open file and return file pointer
 991  *
 992  * @name:       struct filename containing path to open
 993  * @flags:      open flags as per the open(2) second argument
 994  * @mode:       mode for the new file if O_CREAT is set, else ignored
 995  *
 996  * This is the helper to open a file from kernelspace if you really
 997  * have to.  But in generally you should not do this, so please move
 998  * along, nothing to see here..
 999  */
1000 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1001 {
1002         struct open_flags op;
1003         int err = build_open_flags(flags, mode, &op);
1004         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
1005 }
1006
1007 /**
1008  * filp_open - open file and return file pointer
1009  *
1010  * @filename:   path to open
1011  * @flags:      open flags as per the open(2) second argument
1012  * @mode:       mode for the new file if O_CREAT is set, else ignored
1013  *
1014  * This is the helper to open a file from kernelspace if you really
1015  * have to.  But in generally you should not do this, so please move
1016  * along, nothing to see here..
1017  */
1018 struct file *filp_open(const char *filename, int flags, umode_t mode)
1019 {
1020         struct filename *name = getname_kernel(filename);
1021         struct file *file = ERR_CAST(name);
1022
1023         if (!IS_ERR(name)) {
1024                 file = file_open_name(name, flags, mode);
1025                 putname(name);
1026         }
1027         return file;
1028 }
1029 EXPORT_SYMBOL(filp_open);
1030
1031 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1032                             const char *filename, int flags, umode_t mode)
1033 {
1034         struct open_flags op;
1035         int err = build_open_flags(flags, mode, &op);
1036         if (err)
1037                 return ERR_PTR(err);
1038         return do_file_open_root(dentry, mnt, filename, &op);
1039 }
1040 EXPORT_SYMBOL(file_open_root);
1041
1042 struct file *filp_clone_open(struct file *oldfile)
1043 {
1044         struct file *file;
1045         int retval;
1046
1047         file = get_empty_filp();
1048         if (IS_ERR(file))
1049                 return file;
1050
1051         file->f_flags = oldfile->f_flags;
1052         retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1053         if (retval) {
1054                 put_filp(file);
1055                 return ERR_PTR(retval);
1056         }
1057
1058         return file;
1059 }
1060 EXPORT_SYMBOL(filp_clone_open);
1061
1062 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1063 {
1064         struct open_flags op;
1065         int fd = build_open_flags(flags, mode, &op);
1066         struct filename *tmp;
1067
1068         if (fd)
1069                 return fd;
1070
1071         tmp = getname(filename);
1072         if (IS_ERR(tmp))
1073                 return PTR_ERR(tmp);
1074
1075         fd = get_unused_fd_flags(flags);
1076         if (fd >= 0) {
1077                 struct file *f = do_filp_open(dfd, tmp, &op);
1078                 if (IS_ERR(f)) {
1079                         put_unused_fd(fd);
1080                         fd = PTR_ERR(f);
1081                 } else {
1082                         fsnotify_open(f);
1083                         fd_install(fd, f);
1084                 }
1085         }
1086         putname(tmp);
1087         return fd;
1088 }
1089
1090 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1091 {
1092         if (force_o_largefile())
1093                 flags |= O_LARGEFILE;
1094
1095         return do_sys_open(AT_FDCWD, filename, flags, mode);
1096 }
1097
1098 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1099                 umode_t, mode)
1100 {
1101         if (force_o_largefile())
1102                 flags |= O_LARGEFILE;
1103
1104         return do_sys_open(dfd, filename, flags, mode);
1105 }
1106
1107 #ifdef CONFIG_COMPAT
1108 /*
1109  * Exactly like sys_open(), except that it doesn't set the
1110  * O_LARGEFILE flag.
1111  */
1112 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1113 {
1114         return do_sys_open(AT_FDCWD, filename, flags, mode);
1115 }
1116
1117 /*
1118  * Exactly like sys_openat(), except that it doesn't set the
1119  * O_LARGEFILE flag.
1120  */
1121 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1122 {
1123         return do_sys_open(dfd, filename, flags, mode);
1124 }
1125 #endif
1126
1127 #ifndef __alpha__
1128
1129 /*
1130  * For backward compatibility?  Maybe this should be moved
1131  * into arch/i386 instead?
1132  */
1133 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1134 {
1135         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1136 }
1137
1138 #endif
1139
1140 /*
1141  * "id" is the POSIX thread ID. We use the
1142  * files pointer for this..
1143  */
1144 int filp_close(struct file *filp, fl_owner_t id)
1145 {
1146         int retval = 0;
1147
1148         if (!file_count(filp)) {
1149                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1150                 return 0;
1151         }
1152
1153         if (filp->f_op->flush)
1154                 retval = filp->f_op->flush(filp, id);
1155
1156         if (likely(!(filp->f_mode & FMODE_PATH))) {
1157                 dnotify_flush(filp, id);
1158                 locks_remove_posix(filp, id);
1159         }
1160         fput(filp);
1161         return retval;
1162 }
1163
1164 EXPORT_SYMBOL(filp_close);
1165
1166 /*
1167  * Careful here! We test whether the file pointer is NULL before
1168  * releasing the fd. This ensures that one clone task can't release
1169  * an fd while another clone is opening it.
1170  */
1171 SYSCALL_DEFINE1(close, unsigned int, fd)
1172 {
1173         int retval = __close_fd(current->files, fd);
1174
1175         /* can't restart close syscall because file table entry was cleared */
1176         if (unlikely(retval == -ERESTARTSYS ||
1177                      retval == -ERESTARTNOINTR ||
1178                      retval == -ERESTARTNOHAND ||
1179                      retval == -ERESTART_RESTARTBLOCK))
1180                 retval = -EINTR;
1181
1182         return retval;
1183 }
1184 EXPORT_SYMBOL(sys_close);
1185
1186 /*
1187  * This routine simulates a hangup on the tty, to arrange that users
1188  * are given clean terminals at login time.
1189  */
1190 SYSCALL_DEFINE0(vhangup)
1191 {
1192         if (capable(CAP_SYS_TTY_CONFIG)) {
1193                 tty_vhangup_self();
1194                 return 0;
1195         }
1196         return -EPERM;
1197 }
1198
1199 /*
1200  * Called when an inode is about to be open.
1201  * We use this to disallow opening large files on 32bit systems if
1202  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1203  * on this flag in sys_open.
1204  */
1205 int generic_file_open(struct inode * inode, struct file * filp)
1206 {
1207         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1208                 return -EOVERFLOW;
1209         return 0;
1210 }
1211
1212 EXPORT_SYMBOL(generic_file_open);
1213
1214 /*
1215  * This is used by subsystems that don't want seekable
1216  * file descriptors. The function is not supposed to ever fail, the only
1217  * reason it returns an 'int' and not 'void' is so that it can be plugged
1218  * directly into file_operations structure.
1219  */
1220 int nonseekable_open(struct inode *inode, struct file *filp)
1221 {
1222         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1223         return 0;
1224 }
1225
1226 EXPORT_SYMBOL(nonseekable_open);