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