fs/open.c

   1 /*
   2  *  linux/fs/open.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/string.h>
   8 #include <linux/mm.h>
   9 #include <linux/file.h>
  10 #include <linux/fdtable.h>
  11 #include <linux/fsnotify.h>
  12 #include <linux/module.h>
  13 #include <linux/tty.h>
  14 #include <linux/namei.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/capability.h>
  17 #include <linux/securebits.h>
  18 #include <linux/security.h>
  19 #include <linux/mount.h>
  20 #include <linux/fcntl.h>
  21 #include <linux/slab.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/fs.h>
  24 #include <linux/personality.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/rcupdate.h>
  28 #include <linux/audit.h>
  29 #include <linux/falloc.h>
  30 #include <linux/fs_struct.h>
  31 #include <linux/ima.h>
  32 #include <linux/dnotify.h>
  33 #include <linux/compat.h>
  34
  35 #include "internal.h"
  36
  37 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  38         struct file *filp)
  39 {
  40         int ret;
  41         struct iattr newattrs;
  42
  43         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  44         if (length < 0)
  45                 return -EINVAL;
  46
  47         newattrs.ia_size = length;
  48         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  49         if (filp) {
  50                 newattrs.ia_file = filp;
  51                 newattrs.ia_valid |= ATTR_FILE;
  52         }
  53
  54         /* Remove suid, sgid, and file capabilities on truncate too */
  55         ret = dentry_needs_remove_privs(dentry);
  56         if (ret < 0)
  57                 return ret;
  58         if (ret)
  59                 newattrs.ia_valid |= ret | ATTR_FORCE;
  60
  61         inode_lock(dentry->d_inode);
  62         /* Note any delegations or leases have already been broken: */
  63         ret = notify_change(dentry, &newattrs, NULL);
  64         inode_unlock(dentry->d_inode);
  65         return ret;
  66 }
  67
  68 long vfs_truncate(const struct path *path, loff_t length)
  69 {
  70         struct inode *inode;
  71         struct dentry *upperdentry;
  72         long error;
  73
  74         inode = path->dentry->d_inode;
  75
  76         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  77         if (S_ISDIR(inode->i_mode))
  78                 return -EISDIR;
  79         if (!S_ISREG(inode->i_mode))
  80                 return -EINVAL;
  81
  82         error = mnt_want_write(path->mnt);
  83         if (error)
  84                 goto out;
  85
  86         error = inode_permission(inode, MAY_WRITE);
  87         if (error)
  88                 goto mnt_drop_write_and_out;
  89
  90         error = -EPERM;
  91         if (IS_APPEND(inode))
  92                 goto mnt_drop_write_and_out;
  93
  94         /*
  95          * If this is an overlayfs then do as if opening the file so we get
  96          * write access on the upper inode, not on the overlay inode.  For
  97          * non-overlay filesystems d_real() is an identity function.
  98          */
  99         upperdentry = d_real(path->dentry, NULL, O_WRONLY);
 100         error = PTR_ERR(upperdentry);
 101         if (IS_ERR(upperdentry))
 102                 goto mnt_drop_write_and_out;
 103
 104         error = get_write_access(upperdentry->d_inode);
 105         if (error)
 106                 goto mnt_drop_write_and_out;
 107
 108         /*
 109          * Make sure that there are no leases.  get_write_access() protects
 110          * against the truncate racing with a lease-granting setlease().
 111          */
 112         error = break_lease(inode, O_WRONLY);
 113         if (error)
 114                 goto put_write_and_out;
 115
 116         error = locks_verify_truncate(inode, NULL, length);
 117         if (!error)
 118                 error = security_path_truncate(path);
 119         if (!error)
 120                 error = do_truncate(path->dentry, length, 0, NULL);
 121
 122 put_write_and_out:
 123         put_write_access(upperdentry->d_inode);
 124 mnt_drop_write_and_out:
 125         mnt_drop_write(path->mnt);
 126 out:
 127         return error;
 128 }
 129 EXPORT_SYMBOL_GPL(vfs_truncate);
 130
 131 static long do_sys_truncate(const char __user *pathname, loff_t length)
 132 {
 133         unsigned int lookup_flags = LOOKUP_FOLLOW;
 134         struct path path;
 135         int error;
 136
 137         if (length < 0) /* sorry, but loff_t says... */
 138                 return -EINVAL;
 139
 140 retry:
 141         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 142         if (!error) {
 143                 error = vfs_truncate(&path, length);
 144                 path_put(&path);
 145         }
 146         if (retry_estale(error, lookup_flags)) {
 147                 lookup_flags |= LOOKUP_REVAL;
 148                 goto retry;
 149         }
 150         return error;
 151 }
 152
 153 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 154 {
 155         return do_sys_truncate(path, length);
 156 }
 157
 158 #ifdef CONFIG_COMPAT
 159 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 160 {
 161         return do_sys_truncate(path, length);
 162 }
 163 #endif
 164
 165 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 166 {
 167         struct inode *inode;
 168         struct dentry *dentry;
 169         struct fd f;
 170         int error;
 171
 172         error = -EINVAL;
 173         if (length < 0)
 174                 goto out;
 175         error = -EBADF;
 176         f = fdget(fd);
 177         if (!f.file)
 178                 goto out;
 179
 180         /* explicitly opened as large or we are on 64-bit box */
 181         if (f.file->f_flags & O_LARGEFILE)
 182                 small = 0;
 183
 184         dentry = f.file->f_path.dentry;
 185         inode = dentry->d_inode;
 186         error = -EINVAL;
 187         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 188                 goto out_putf;
 189
 190         error = -EINVAL;
 191         /* Cannot ftruncate over 2^31 bytes without large file support */
 192         if (small && length > MAX_NON_LFS)
 193                 goto out_putf;
 194
 195         error = -EPERM;
 196         if (IS_APPEND(inode))
 197                 goto out_putf;
 198
 199         sb_start_write(inode->i_sb);
 200         error = locks_verify_truncate(inode, f.file, length);
 201         if (!error)
 202                 error = security_path_truncate(&f.file->f_path);
 203         if (!error)
 204                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 205         sb_end_write(inode->i_sb);
 206 out_putf:
 207         fdput(f);
 208 out:
 209         return error;
 210 }
 211
 212 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 213 {
 214         return do_sys_ftruncate(fd, length, 1);
 215 }
 216
 217 #ifdef CONFIG_COMPAT
 218 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 219 {
 220         return do_sys_ftruncate(fd, length, 1);
 221 }
 222 #endif
 223
 224 /* LFS versions of truncate are only needed on 32 bit machines */
 225 #if BITS_PER_LONG == 32
 226 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 227 {
 228         return do_sys_truncate(path, length);
 229 }
 230
 231 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 232 {
 233         return do_sys_ftruncate(fd, length, 0);
 234 }
 235 #endif /* BITS_PER_LONG == 32 */
 236
 237
 238 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 239 {
 240         struct inode *inode = file_inode(file);
 241         long ret;
 242
 243         if (offset < 0 || len <= 0)
 244                 return -EINVAL;
 245
 246         /* Return error if mode is not supported */
 247         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 248                 return -EOPNOTSUPP;
 249
 250         /* Punch hole and zero range are mutually exclusive */
 251         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 252             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 253                 return -EOPNOTSUPP;
 254
 255         /* Punch hole must have keep size set */
 256         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 257             !(mode & FALLOC_FL_KEEP_SIZE))
 258                 return -EOPNOTSUPP;
 259
 260         /* Collapse range should only be used exclusively. */
 261         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 262             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 263                 return -EINVAL;
 264
 265         /* Insert range should only be used exclusively. */
 266         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 267             (mode & ~FALLOC_FL_INSERT_RANGE))
 268                 return -EINVAL;
 269
 270         /* Unshare range should only be used with allocate mode. */
 271         if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 272             (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 273                 return -EINVAL;
 274
 275         if (!(file->f_mode & FMODE_WRITE))
 276                 return -EBADF;
 277
 278         /*
 279          * We can only allow pure fallocate on append only files
 280          */
 281         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 282                 return -EPERM;
 283
 284         if (IS_IMMUTABLE(inode))
 285                 return -EPERM;
 286
 287         /*
 288          * We cannot allow any fallocate operation on an active swapfile
 289          */
 290         if (IS_SWAPFILE(inode))
 291                 return -ETXTBSY;
 292
 293         /*
 294          * Revalidate the write permissions, in case security policy has
 295          * changed since the files were opened.
 296          */
 297         ret = security_file_permission(file, MAY_WRITE);
 298         if (ret)
 299                 return ret;
 300
 301         if (S_ISFIFO(inode->i_mode))
 302                 return -ESPIPE;
 303
 304         if (S_ISDIR(inode->i_mode))
 305                 return -EISDIR;
 306
 307         if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 308                 return -ENODEV;
 309
 310         /* Check for wrap through zero too */
 311         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 312                 return -EFBIG;
 313
 314         if (!file->f_op->fallocate)
 315                 return -EOPNOTSUPP;
 316
 317         file_start_write(file);
 318         ret = file->f_op->fallocate(file, mode, offset, len);
 319
 320         /*
 321          * Create inotify and fanotify events.
 322          *
 323          * To keep the logic simple always create events if fallocate succeeds.
 324          * This implies that events are even created if the file size remains
 325          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 326          */
 327         if (ret == 0)
 328                 fsnotify_modify(file);
 329
 330         file_end_write(file);
 331         return ret;
 332 }
 333 EXPORT_SYMBOL_GPL(vfs_fallocate);
 334
 335 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 336 {
 337         struct fd f = fdget(fd);
 338         int error = -EBADF;
 339
 340         if (f.file) {
 341                 error = vfs_fallocate(f.file, mode, offset, len);
 342                 fdput(f);
 343         }
 344         return error;
 345 }
 346
 347 /*
 348  * access() needs to use the real uid/gid, not the effective uid/gid.
 349  * We do this by temporarily clearing all FS-related capabilities and
 350  * switching the fsuid/fsgid around to the real ones.
 351  */
 352 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 353 {
 354         const struct cred *old_cred;
 355         struct cred *override_cred;
 356         struct path path;
 357         struct inode *inode;
 358         int res;
 359         unsigned int lookup_flags = LOOKUP_FOLLOW;
 360
 361         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 362                 return -EINVAL;
 363
 364         override_cred = prepare_creds();
 365         if (!override_cred)
 366                 return -ENOMEM;
 367
 368         override_cred->fsuid = override_cred->uid;
 369         override_cred->fsgid = override_cred->gid;
 370
 371         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 372                 /* Clear the capabilities if we switch to a non-root user */
 373                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 374                 if (!uid_eq(override_cred->uid, root_uid))
 375                         cap_clear(override_cred->cap_effective);
 376                 else
 377                         override_cred->cap_effective =
 378                                 override_cred->cap_permitted;
 379         }
 380
 381         old_cred = override_creds(override_cred);
 382 retry:
 383         res = user_path_at(dfd, filename, lookup_flags, &path);
 384         if (res)
 385                 goto out;
 386
 387         inode = d_backing_inode(path.dentry);
 388
 389         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 390                 /*
 391                  * MAY_EXEC on regular files is denied if the fs is mounted
 392                  * with the "noexec" flag.
 393                  */
 394                 res = -EACCES;
 395                 if (path_noexec(&path))
 396                         goto out_path_release;
 397         }
 398
 399         res = inode_permission(inode, mode | MAY_ACCESS);
 400         /* SuS v2 requires we report a read only fs too */
 401         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 402                 goto out_path_release;
 403         /*
 404          * This is a rare case where using __mnt_is_readonly()
 405          * is OK without a mnt_want/drop_write() pair.  Since
 406          * no actual write to the fs is performed here, we do
 407          * not need to telegraph to that to anyone.
 408          *
 409          * By doing this, we accept that this access is
 410          * inherently racy and know that the fs may change
 411          * state before we even see this result.
 412          */
 413         if (__mnt_is_readonly(path.mnt))
 414                 res = -EROFS;
 415
 416 out_path_release:
 417         path_put(&path);
 418         if (retry_estale(res, lookup_flags)) {
 419                 lookup_flags |= LOOKUP_REVAL;
 420                 goto retry;
 421         }
 422 out:
 423         revert_creds(old_cred);
 424         put_cred(override_cred);
 425         return res;
 426 }
 427
 428 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 429 {
 430         return sys_faccessat(AT_FDCWD, filename, mode);
 431 }
 432
 433 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 434 {
 435         struct path path;
 436         int error;
 437         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 438 retry:
 439         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 440         if (error)
 441                 goto out;
 442
 443         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 444         if (error)
 445                 goto dput_and_out;
 446
 447         set_fs_pwd(current->fs, &path);
 448
 449 dput_and_out:
 450         path_put(&path);
 451         if (retry_estale(error, lookup_flags)) {
 452                 lookup_flags |= LOOKUP_REVAL;
 453                 goto retry;
 454         }
 455 out:
 456         return error;
 457 }
 458
 459 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 460 {
 461         struct fd f = fdget_raw(fd);
 462         struct inode *inode;
 463         int error = -EBADF;
 464
 465         error = -EBADF;
 466         if (!f.file)
 467                 goto out;
 468
 469         inode = file_inode(f.file);
 470
 471         error = -ENOTDIR;
 472         if (!S_ISDIR(inode->i_mode))
 473                 goto out_putf;
 474
 475         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 476         if (!error)
 477                 set_fs_pwd(current->fs, &f.file->f_path);
 478 out_putf:
 479         fdput(f);
 480 out:
 481         return error;
 482 }
 483
 484 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 485 {
 486         struct path path;
 487         int error;
 488         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 489 retry:
 490         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 491         if (error)
 492                 goto out;
 493
 494         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 495         if (error)
 496                 goto dput_and_out;
 497
 498         error = -EPERM;
 499         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 500                 goto dput_and_out;
 501         error = security_path_chroot(&path);
 502         if (error)
 503                 goto dput_and_out;
 504
 505         set_fs_root(current->fs, &path);
 506         error = 0;
 507 dput_and_out:
 508         path_put(&path);
 509         if (retry_estale(error, lookup_flags)) {
 510                 lookup_flags |= LOOKUP_REVAL;
 511                 goto retry;
 512         }
 513 out:
 514         return error;
 515 }
 516
 517 static int chmod_common(const struct path *path, umode_t mode)
 518 {
 519         struct inode *inode = path->dentry->d_inode;
 520         struct inode *delegated_inode = NULL;
 521         struct iattr newattrs;
 522         int error;
 523
 524         error = mnt_want_write(path->mnt);
 525         if (error)
 526                 return error;
 527 retry_deleg:
 528         inode_lock(inode);
 529         error = security_path_chmod(path, mode);
 530         if (error)
 531                 goto out_unlock;
 532         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 533         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 534         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 535 out_unlock:
 536         inode_unlock(inode);
 537         if (delegated_inode) {
 538                 error = break_deleg_wait(&delegated_inode);
 539                 if (!error)
 540                         goto retry_deleg;
 541         }
 542         mnt_drop_write(path->mnt);
 543         return error;
 544 }
 545
 546 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 547 {
 548         struct fd f = fdget(fd);
 549         int err = -EBADF;
 550
 551         if (f.file) {
 552                 audit_file(f.file);
 553                 err = chmod_common(&f.file->f_path, mode);
 554                 fdput(f);
 555         }
 556         return err;
 557 }
 558
 559 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 560 {
 561         struct path path;
 562         int error;
 563         unsigned int lookup_flags = LOOKUP_FOLLOW;
 564 retry:
 565         error = user_path_at(dfd, filename, lookup_flags, &path);
 566         if (!error) {
 567                 error = chmod_common(&path, mode);
 568                 path_put(&path);
 569                 if (retry_estale(error, lookup_flags)) {
 570                         lookup_flags |= LOOKUP_REVAL;
 571                         goto retry;
 572                 }
 573         }
 574         return error;
 575 }
 576
 577 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 578 {
 579         return sys_fchmodat(AT_FDCWD, filename, mode);
 580 }
 581
 582 static int chown_common(const struct path *path, uid_t user, gid_t group)
 583 {
 584         struct inode *inode = path->dentry->d_inode;
 585         struct inode *delegated_inode = NULL;
 586         int error;
 587         struct iattr newattrs;
 588         kuid_t uid;
 589         kgid_t gid;
 590
 591         uid = make_kuid(current_user_ns(), user);
 592         gid = make_kgid(current_user_ns(), group);
 593
 594 retry_deleg:
 595         newattrs.ia_valid =  ATTR_CTIME;
 596         if (user != (uid_t) -1) {
 597                 if (!uid_valid(uid))
 598                         return -EINVAL;
 599                 newattrs.ia_valid |= ATTR_UID;
 600                 newattrs.ia_uid = uid;
 601         }
 602         if (group != (gid_t) -1) {
 603                 if (!gid_valid(gid))
 604                         return -EINVAL;
 605                 newattrs.ia_valid |= ATTR_GID;
 606                 newattrs.ia_gid = gid;
 607         }
 608         if (!S_ISDIR(inode->i_mode))
 609                 newattrs.ia_valid |=
 610                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 611         inode_lock(inode);
 612         error = security_path_chown(path, uid, gid);
 613         if (!error)
 614                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 615         inode_unlock(inode);
 616         if (delegated_inode) {
 617                 error = break_deleg_wait(&delegated_inode);
 618                 if (!error)
 619                         goto retry_deleg;
 620         }
 621         return error;
 622 }
 623
 624 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 625                 gid_t, group, int, flag)
 626 {
 627         struct path path;
 628         int error = -EINVAL;
 629         int lookup_flags;
 630
 631         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 632                 goto out;
 633
 634         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 635         if (flag & AT_EMPTY_PATH)
 636                 lookup_flags |= LOOKUP_EMPTY;
 637 retry:
 638         error = user_path_at(dfd, filename, lookup_flags, &path);
 639         if (error)
 640                 goto out;
 641         error = mnt_want_write(path.mnt);
 642         if (error)
 643                 goto out_release;
 644         error = chown_common(&path, user, group);
 645         mnt_drop_write(path.mnt);
 646 out_release:
 647         path_put(&path);
 648         if (retry_estale(error, lookup_flags)) {
 649                 lookup_flags |= LOOKUP_REVAL;
 650                 goto retry;
 651         }
 652 out:
 653         return error;
 654 }
 655
 656 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 657 {
 658         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 659 }
 660
 661 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 662 {
 663         return sys_fchownat(AT_FDCWD, filename, user, group,
 664                             AT_SYMLINK_NOFOLLOW);
 665 }
 666
 667 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 668 {
 669         struct fd f = fdget(fd);
 670         int error = -EBADF;
 671
 672         if (!f.file)
 673                 goto out;
 674
 675         error = mnt_want_write_file(f.file);
 676         if (error)
 677                 goto out_fput;
 678         audit_file(f.file);
 679         error = chown_common(&f.file->f_path, user, group);
 680         mnt_drop_write_file(f.file);
 681 out_fput:
 682         fdput(f);
 683 out:
 684         return error;
 685 }
 686
 687 int open_check_o_direct(struct file *f)
 688 {
 689         /* NB: we're sure to have correct a_ops only after f_op->open */
 690         if (f->f_flags & O_DIRECT) {
 691                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 692                         return -EINVAL;
 693         }
 694         return 0;
 695 }
 696
 697 static int do_dentry_open(struct file *f,
 698                           struct inode *inode,
 699                           int (*open)(struct inode *, struct file *),
 700                           const struct cred *cred)
 701 {
 702         static const struct file_operations empty_fops = {};
 703         int error;
 704
 705         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 706                                 FMODE_PREAD | FMODE_PWRITE;
 707
 708         path_get(&f->f_path);
 709         f->f_inode = inode;
 710         f->f_mapping = inode->i_mapping;
 711
 712         if (unlikely(f->f_flags & O_PATH)) {
 713                 f->f_mode = FMODE_PATH;
 714                 f->f_op = &empty_fops;
 715                 return 0;
 716         }
 717
 718         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 719                 error = get_write_access(inode);
 720                 if (unlikely(error))
 721                         goto cleanup_file;
 722                 error = __mnt_want_write(f->f_path.mnt);
 723                 if (unlikely(error)) {
 724                         put_write_access(inode);
 725                         goto cleanup_file;
 726                 }
 727                 f->f_mode |= FMODE_WRITER;
 728         }
 729
 730         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 731         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 732                 f->f_mode |= FMODE_ATOMIC_POS;
 733
 734         f->f_op = fops_get(inode->i_fop);
 735         if (unlikely(WARN_ON(!f->f_op))) {
 736                 error = -ENODEV;
 737                 goto cleanup_all;
 738         }
 739
 740         error = security_file_open(f, cred);
 741         if (error)
 742                 goto cleanup_all;
 743
 744         error = break_lease(locks_inode(f), f->f_flags);
 745         if (error)
 746                 goto cleanup_all;
 747
 748         if (!open)
 749                 open = f->f_op->open;
 750         if (open) {
 751                 error = open(inode, f);
 752                 if (error)
 753                         goto cleanup_all;
 754         }
 755         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 756                 i_readcount_inc(inode);
 757         if ((f->f_mode & FMODE_READ) &&
 758              likely(f->f_op->read || f->f_op->read_iter))
 759                 f->f_mode |= FMODE_CAN_READ;
 760         if ((f->f_mode & FMODE_WRITE) &&
 761              likely(f->f_op->write || f->f_op->write_iter))
 762                 f->f_mode |= FMODE_CAN_WRITE;
 763
 764         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 765
 766         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 767
 768         return 0;
 769
 770 cleanup_all:
 771         fops_put(f->f_op);
 772         if (f->f_mode & FMODE_WRITER) {
 773                 put_write_access(inode);
 774                 __mnt_drop_write(f->f_path.mnt);
 775         }
 776 cleanup_file:
 777         path_put(&f->f_path);
 778         f->f_path.mnt = NULL;
 779         f->f_path.dentry = NULL;
 780         f->f_inode = NULL;
 781         return error;
 782 }
 783
 784 /**
 785  * finish_open - finish opening a file
 786  * @file: file pointer
 787  * @dentry: pointer to dentry
 788  * @open: open callback
 789  * @opened: state of open
 790  *
 791  * This can be used to finish opening a file passed to i_op->atomic_open().
 792  *
 793  * If the open callback is set to NULL, then the standard f_op->open()
 794  * filesystem callback is substituted.
 795  *
 796  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 797  * the return value of d_splice_alias(), then the caller needs to perform dput()
 798  * on it after finish_open().
 799  *
 800  * On successful return @file is a fully instantiated open file.  After this, if
 801  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 802  *
 803  * Returns zero on success or -errno if the open failed.
 804  */
 805 int finish_open(struct file *file, struct dentry *dentry,
 806                 int (*open)(struct inode *, struct file *),
 807                 int *opened)
 808 {
 809         int error;
 810         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 811
 812         file->f_path.dentry = dentry;
 813         error = do_dentry_open(file, d_backing_inode(dentry), open,
 814                                current_cred());
 815         if (!error)
 816                 *opened |= FILE_OPENED;
 817
 818         return error;
 819 }
 820 EXPORT_SYMBOL(finish_open);
 821
 822 /**
 823  * finish_no_open - finish ->atomic_open() without opening the file
 824  *
 825  * @file: file pointer
 826  * @dentry: dentry or NULL (as returned from ->lookup())
 827  *
 828  * This can be used to set the result of a successful lookup in ->atomic_open().
 829  *
 830  * NB: unlike finish_open() this function does consume the dentry reference and
 831  * the caller need not dput() it.
 832  *
 833  * Returns "1" which must be the return value of ->atomic_open() after having
 834  * called this function.
 835  */
 836 int finish_no_open(struct file *file, struct dentry *dentry)
 837 {
 838         file->f_path.dentry = dentry;
 839         return 1;
 840 }
 841 EXPORT_SYMBOL(finish_no_open);
 842
 843 char *file_path(struct file *filp, char *buf, int buflen)
 844 {
 845         return d_path(&filp->f_path, buf, buflen);
 846 }
 847 EXPORT_SYMBOL(file_path);
 848
 849 /**
 850  * vfs_open - open the file at the given path
 851  * @path: path to open
 852  * @file: newly allocated file with f_flag initialized
 853  * @cred: credentials to use
 854  */
 855 int vfs_open(const struct path *path, struct file *file,
 856              const struct cred *cred)
 857 {
 858         struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags);
 859
 860         if (IS_ERR(dentry))
 861                 return PTR_ERR(dentry);
 862
 863         file->f_path = *path;
 864         return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 865 }
 866
 867 struct file *dentry_open(const struct path *path, int flags,
 868                          const struct cred *cred)
 869 {
 870         int error;
 871         struct file *f;
 872
 873         validate_creds(cred);
 874
 875         /* We must always pass in a valid mount pointer. */
 876         BUG_ON(!path->mnt);
 877
 878         f = get_empty_filp();
 879         if (!IS_ERR(f)) {
 880                 f->f_flags = flags;
 881                 error = vfs_open(path, f, cred);
 882                 if (!error) {
 883                         /* from now on we need fput() to dispose of f */
 884                         error = open_check_o_direct(f);
 885                         if (error) {
 886                                 fput(f);
 887                                 f = ERR_PTR(error);
 888                         }
 889                 } else {
 890                         put_filp(f);
 891                         f = ERR_PTR(error);
 892                 }
 893         }
 894         return f;
 895 }
 896 EXPORT_SYMBOL(dentry_open);
 897
 898 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 899 {
 900         int lookup_flags = 0;
 901         int acc_mode = ACC_MODE(flags);
 902
 903         if (flags & (O_CREAT | __O_TMPFILE))
 904                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 905         else
 906                 op->mode = 0;
 907
 908         /* Must never be set by userspace */
 909         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 910
 911         /*
 912          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 913          * check for O_DSYNC if the need any syncing at all we enforce it's
 914          * always set instead of having to deal with possibly weird behaviour
 915          * for malicious applications setting only __O_SYNC.
 916          */
 917         if (flags & __O_SYNC)
 918                 flags |= O_DSYNC;
 919
 920         if (flags & __O_TMPFILE) {
 921                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 922                         return -EINVAL;
 923                 if (!(acc_mode & MAY_WRITE))
 924                         return -EINVAL;
 925         } else if (flags & O_PATH) {
 926                 /*
 927                  * If we have O_PATH in the open flag. Then we
 928                  * cannot have anything other than the below set of flags
 929                  */
 930                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 931                 acc_mode = 0;
 932         }
 933
 934         op->open_flag = flags;
 935
 936         /* O_TRUNC implies we need access checks for write permissions */
 937         if (flags & O_TRUNC)
 938                 acc_mode |= MAY_WRITE;
 939
 940         /* Allow the LSM permission hook to distinguish append
 941            access from general write access. */
 942         if (flags & O_APPEND)
 943                 acc_mode |= MAY_APPEND;
 944
 945         op->acc_mode = acc_mode;
 946
 947         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 948
 949         if (flags & O_CREAT) {
 950                 op->intent |= LOOKUP_CREATE;
 951                 if (flags & O_EXCL)
 952                         op->intent |= LOOKUP_EXCL;
 953         }
 954
 955         if (flags & O_DIRECTORY)
 956                 lookup_flags |= LOOKUP_DIRECTORY;
 957         if (!(flags & O_NOFOLLOW))
 958                 lookup_flags |= LOOKUP_FOLLOW;
 959         op->lookup_flags = lookup_flags;
 960         return 0;
 961 }
 962
 963 /**
 964  * file_open_name - open file and return file pointer
 965  *
 966  * @name:       struct filename containing path to open
 967  * @flags:      open flags as per the open(2) second argument
 968  * @mode:       mode for the new file if O_CREAT is set, else ignored
 969  *
 970  * This is the helper to open a file from kernelspace if you really
 971  * have to.  But in generally you should not do this, so please move
 972  * along, nothing to see here..
 973  */
 974 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 975 {
 976         struct open_flags op;
 977         int err = build_open_flags(flags, mode, &op);
 978         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 979 }
 980
 981 /**
 982  * filp_open - open file and return file pointer
 983  *
 984  * @filename:   path to open
 985  * @flags:      open flags as per the open(2) second argument
 986  * @mode:       mode for the new file if O_CREAT is set, else ignored
 987  *
 988  * This is the helper to open a file from kernelspace if you really
 989  * have to.  But in generally you should not do this, so please move
 990  * along, nothing to see here..
 991  */
 992 struct file *filp_open(const char *filename, int flags, umode_t mode)
 993 {
 994         struct filename *name = getname_kernel(filename);
 995         struct file *file = ERR_CAST(name);
 996
 997         if (!IS_ERR(name)) {
 998                 file = file_open_name(name, flags, mode);
 999                 putname(name);
1000         }
1001         return file;
1002 }
1003 EXPORT_SYMBOL(filp_open);
1004
1005 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1006                             const char *filename, int flags, umode_t mode)
1007 {
1008         struct open_flags op;
1009         int err = build_open_flags(flags, mode, &op);
1010         if (err)
1011                 return ERR_PTR(err);
1012         return do_file_open_root(dentry, mnt, filename, &op);
1013 }
1014 EXPORT_SYMBOL(file_open_root);
1015
1016 struct file *filp_clone_open(struct file *oldfile)
1017 {
1018         struct file *file;
1019         int retval;
1020
1021         file = get_empty_filp();
1022         if (IS_ERR(file))
1023                 return file;
1024
1025         file->f_flags = oldfile->f_flags;
1026         retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1027         if (retval) {
1028                 put_filp(file);
1029                 return ERR_PTR(retval);
1030         }
1031
1032         return file;
1033 }
1034 EXPORT_SYMBOL(filp_clone_open);
1035
1036 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1037 {
1038         struct open_flags op;
1039         int fd = build_open_flags(flags, mode, &op);
1040         struct filename *tmp;
1041
1042         if (fd)
1043                 return fd;
1044
1045         tmp = getname(filename);
1046         if (IS_ERR(tmp))
1047                 return PTR_ERR(tmp);
1048
1049         fd = get_unused_fd_flags(flags);
1050         if (fd >= 0) {
1051                 struct file *f = do_filp_open(dfd, tmp, &op);
1052                 if (IS_ERR(f)) {
1053                         put_unused_fd(fd);
1054                         fd = PTR_ERR(f);
1055                 } else {
1056                         fsnotify_open(f);
1057                         fd_install(fd, f);
1058                 }
1059         }
1060         putname(tmp);
1061         return fd;
1062 }
1063
1064 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1065 {
1066         if (force_o_largefile())
1067                 flags |= O_LARGEFILE;
1068
1069         return do_sys_open(AT_FDCWD, filename, flags, mode);
1070 }
1071
1072 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1073                 umode_t, mode)
1074 {
1075         if (force_o_largefile())
1076                 flags |= O_LARGEFILE;
1077
1078         return do_sys_open(dfd, filename, flags, mode);
1079 }
1080
1081 #ifndef __alpha__
1082
1083 /*
1084  * For backward compatibility?  Maybe this should be moved
1085  * into arch/i386 instead?
1086  */
1087 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1088 {
1089         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1090 }
1091
1092 #endif
1093
1094 /*
1095  * "id" is the POSIX thread ID. We use the
1096  * files pointer for this..
1097  */
1098 int filp_close(struct file *filp, fl_owner_t id)
1099 {
1100         int retval = 0;
1101
1102         if (!file_count(filp)) {
1103                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1104                 return 0;
1105         }
1106
1107         if (filp->f_op->flush)
1108                 retval = filp->f_op->flush(filp, id);
1109
1110         if (likely(!(filp->f_mode & FMODE_PATH))) {
1111                 dnotify_flush(filp, id);
1112                 locks_remove_posix(filp, id);
1113         }
1114         fput(filp);
1115         return retval;
1116 }
1117
1118 EXPORT_SYMBOL(filp_close);
1119
1120 /*
1121  * Careful here! We test whether the file pointer is NULL before
1122  * releasing the fd. This ensures that one clone task can't release
1123  * an fd while another clone is opening it.
1124  */
1125 SYSCALL_DEFINE1(close, unsigned int, fd)
1126 {
1127         int retval = __close_fd(current->files, fd);
1128
1129         /* can't restart close syscall because file table entry was cleared */
1130         if (unlikely(retval == -ERESTARTSYS ||
1131                      retval == -ERESTARTNOINTR ||
1132                      retval == -ERESTARTNOHAND ||
1133                      retval == -ERESTART_RESTARTBLOCK))
1134                 retval = -EINTR;
1135
1136         return retval;
1137 }
1138 EXPORT_SYMBOL(sys_close);
1139
1140 /*
1141  * This routine simulates a hangup on the tty, to arrange that users
1142  * are given clean terminals at login time.
1143  */
1144 SYSCALL_DEFINE0(vhangup)
1145 {
1146         if (capable(CAP_SYS_TTY_CONFIG)) {
1147                 tty_vhangup_self();
1148                 return 0;
1149         }
1150         return -EPERM;
1151 }
1152
1153 /*
1154  * Called when an inode is about to be open.
1155  * We use this to disallow opening large files on 32bit systems if
1156  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1157  * on this flag in sys_open.
1158  */
1159 int generic_file_open(struct inode * inode, struct file * filp)
1160 {
1161         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1162                 return -EOVERFLOW;
1163         return 0;
1164 }
1165
1166 EXPORT_SYMBOL(generic_file_open);
1167
1168 /*
1169  * This is used by subsystems that don't want seekable
1170  * file descriptors. The function is not supposed to ever fail, the only
1171  * reason it returns an 'int' and not 'void' is so that it can be plugged
1172  * directly into file_operations structure.
1173  */
1174 int nonseekable_open(struct inode *inode, struct file *filp)
1175 {
1176         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1177         return 0;
1178 }
1179
1180 EXPORT_SYMBOL(nonseekable_open);