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