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 #define CREATE_TRACE_POINTS
  38 #include <trace/events/fs.h>
  39
  40 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  41         struct file *filp)
  42 {
  43         int ret;
  44         struct iattr newattrs;
  45
  46         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  47         if (length < 0)
  48                 return -EINVAL;
  49
  50         newattrs.ia_size = length;
  51         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  52         if (filp) {
  53                 newattrs.ia_file = filp;
  54                 newattrs.ia_valid |= ATTR_FILE;
  55         }
  56
  57         /* Remove suid, sgid, and file capabilities on truncate too */
  58         ret = dentry_needs_remove_privs(dentry);
  59         if (ret < 0)
  60                 return ret;
  61         if (ret)
  62                 newattrs.ia_valid |= ret | ATTR_FORCE;
  63
  64         inode_lock(dentry->d_inode);
  65         /* Note any delegations or leases have already been broken: */
  66         ret = notify_change(dentry, &newattrs, NULL);
  67         inode_unlock(dentry->d_inode);
  68         return ret;
  69 }
  70 EXPORT_SYMBOL_GPL(do_truncate);
  71
  72 long vfs_truncate(const struct path *path, loff_t length)
  73 {
  74         struct inode *inode;
  75         struct dentry *upperdentry;
  76         long error;
  77
  78         inode = path->dentry->d_inode;
  79
  80         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  81         if (S_ISDIR(inode->i_mode))
  82                 return -EISDIR;
  83         if (!S_ISREG(inode->i_mode))
  84                 return -EINVAL;
  85
  86         error = mnt_want_write(path->mnt);
  87         if (error)
  88                 goto out;
  89
  90         error = inode_permission(inode, MAY_WRITE);
  91         if (error)
  92                 goto mnt_drop_write_and_out;
  93
  94         error = -EPERM;
  95         if (IS_APPEND(inode))
  96                 goto mnt_drop_write_and_out;
  97
  98         /*
  99          * If this is an overlayfs then do as if opening the file so we get
 100          * write access on the upper inode, not on the overlay inode.  For
 101          * non-overlay filesystems d_real() is an identity function.
 102          */
 103         upperdentry = d_real(path->dentry, NULL, O_WRONLY);
 104         error = PTR_ERR(upperdentry);
 105         if (IS_ERR(upperdentry))
 106                 goto mnt_drop_write_and_out;
 107
 108         error = get_write_access(upperdentry->d_inode);
 109         if (error)
 110                 goto mnt_drop_write_and_out;
 111
 112         /*
 113          * Make sure that there are no leases.  get_write_access() protects
 114          * against the truncate racing with a lease-granting setlease().
 115          */
 116         error = break_lease(inode, O_WRONLY);
 117         if (error)
 118                 goto put_write_and_out;
 119
 120         error = locks_verify_truncate(inode, NULL, length);
 121         if (!error)
 122                 error = security_path_truncate(path);
 123         if (!error)
 124                 error = do_truncate(path->dentry, length, 0, NULL);
 125
 126 put_write_and_out:
 127         put_write_access(upperdentry->d_inode);
 128 mnt_drop_write_and_out:
 129         mnt_drop_write(path->mnt);
 130 out:
 131         return error;
 132 }
 133 EXPORT_SYMBOL_GPL(vfs_truncate);
 134
 135 static long do_sys_truncate(const char __user *pathname, loff_t length)
 136 {
 137         unsigned int lookup_flags = LOOKUP_FOLLOW;
 138         struct path path;
 139         int error;
 140
 141         if (length < 0) /* sorry, but loff_t says... */
 142                 return -EINVAL;
 143
 144 retry:
 145         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 146         if (!error) {
 147                 error = vfs_truncate(&path, length);
 148                 path_put(&path);
 149         }
 150         if (retry_estale(error, lookup_flags)) {
 151                 lookup_flags |= LOOKUP_REVAL;
 152                 goto retry;
 153         }
 154         return error;
 155 }
 156
 157 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 158 {
 159         return do_sys_truncate(path, length);
 160 }
 161
 162 #ifdef CONFIG_COMPAT
 163 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 164 {
 165         return do_sys_truncate(path, length);
 166 }
 167 #endif
 168
 169 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 170 {
 171         struct inode *inode;
 172         struct dentry *dentry;
 173         struct fd f;
 174         int error;
 175
 176         error = -EINVAL;
 177         if (length < 0)
 178                 goto out;
 179         error = -EBADF;
 180         f = fdget(fd);
 181         if (!f.file)
 182                 goto out;
 183
 184         /* explicitly opened as large or we are on 64-bit box */
 185         if (f.file->f_flags & O_LARGEFILE)
 186                 small = 0;
 187
 188         dentry = f.file->f_path.dentry;
 189         inode = dentry->d_inode;
 190         error = -EINVAL;
 191         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 192                 goto out_putf;
 193
 194         error = -EINVAL;
 195         /* Cannot ftruncate over 2^31 bytes without large file support */
 196         if (small && length > MAX_NON_LFS)
 197                 goto out_putf;
 198
 199         error = -EPERM;
 200         if (IS_APPEND(inode))
 201                 goto out_putf;
 202
 203         sb_start_write(inode->i_sb);
 204         error = locks_verify_truncate(inode, f.file, length);
 205         if (!error)
 206                 error = security_path_truncate(&f.file->f_path);
 207         if (!error)
 208                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 209         sb_end_write(inode->i_sb);
 210 out_putf:
 211         fdput(f);
 212 out:
 213         return error;
 214 }
 215
 216 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 217 {
 218         return do_sys_ftruncate(fd, length, 1);
 219 }
 220
 221 #ifdef CONFIG_COMPAT
 222 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 223 {
 224         return do_sys_ftruncate(fd, length, 1);
 225 }
 226 #endif
 227
 228 /* LFS versions of truncate are only needed on 32 bit machines */
 229 #if BITS_PER_LONG == 32
 230 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 231 {
 232         return do_sys_truncate(path, length);
 233 }
 234
 235 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 236 {
 237         return do_sys_ftruncate(fd, length, 0);
 238 }
 239 #endif /* BITS_PER_LONG == 32 */
 240
 241
 242 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 243 {
 244         struct inode *inode = file_inode(file);
 245         long ret;
 246
 247         if (offset < 0 || len <= 0)
 248                 return -EINVAL;
 249
 250         /* Return error if mode is not supported */
 251         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 252                 return -EOPNOTSUPP;
 253
 254         /* Punch hole and zero range are mutually exclusive */
 255         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 256             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 257                 return -EOPNOTSUPP;
 258
 259         /* Punch hole must have keep size set */
 260         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 261             !(mode & FALLOC_FL_KEEP_SIZE))
 262                 return -EOPNOTSUPP;
 263
 264         /* Collapse range should only be used exclusively. */
 265         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 266             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 267                 return -EINVAL;
 268
 269         /* Insert range should only be used exclusively. */
 270         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 271             (mode & ~FALLOC_FL_INSERT_RANGE))
 272                 return -EINVAL;
 273
 274         /* Unshare range should only be used with allocate mode. */
 275         if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 276             (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 277                 return -EINVAL;
 278
 279         if (!(file->f_mode & FMODE_WRITE))
 280                 return -EBADF;
 281
 282         /*
 283          * We can only allow pure fallocate on append only files
 284          */
 285         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 286                 return -EPERM;
 287
 288         if (IS_IMMUTABLE(inode))
 289                 return -EPERM;
 290
 291         /*
 292          * We cannot allow any fallocate operation on an active swapfile
 293          */
 294         if (IS_SWAPFILE(inode))
 295                 return -ETXTBSY;
 296
 297         /*
 298          * Revalidate the write permissions, in case security policy has
 299          * changed since the files were opened.
 300          */
 301         ret = security_file_permission(file, MAY_WRITE);
 302         if (ret)
 303                 return ret;
 304
 305         if (S_ISFIFO(inode->i_mode))
 306                 return -ESPIPE;
 307
 308         /*
 309          * Let individual file system decide if it supports preallocation
 310          * for directories or not.
 311          */
 312         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
 313             !S_ISBLK(inode->i_mode))
 314                 return -ENODEV;
 315
 316         /* Check for wrap through zero too */
 317         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 318                 return -EFBIG;
 319
 320         if (!file->f_op->fallocate)
 321                 return -EOPNOTSUPP;
 322
 323         sb_start_write(inode->i_sb);
 324         ret = file->f_op->fallocate(file, mode, offset, len);
 325
 326         /*
 327          * Create inotify and fanotify events.
 328          *
 329          * To keep the logic simple always create events if fallocate succeeds.
 330          * This implies that events are even created if the file size remains
 331          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 332          */
 333         if (ret == 0)
 334                 fsnotify_modify(file);
 335
 336         sb_end_write(inode->i_sb);
 337         return ret;
 338 }
 339 EXPORT_SYMBOL_GPL(vfs_fallocate);
 340
 341 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 342 {
 343         struct fd f = fdget(fd);
 344         int error = -EBADF;
 345
 346         if (f.file) {
 347                 error = vfs_fallocate(f.file, mode, offset, len);
 348                 fdput(f);
 349         }
 350         return error;
 351 }
 352
 353 /*
 354  * access() needs to use the real uid/gid, not the effective uid/gid.
 355  * We do this by temporarily clearing all FS-related capabilities and
 356  * switching the fsuid/fsgid around to the real ones.
 357  */
 358 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 359 {
 360         const struct cred *old_cred;
 361         struct cred *override_cred;
 362         struct path path;
 363         struct inode *inode;
 364         int res;
 365         unsigned int lookup_flags = LOOKUP_FOLLOW;
 366
 367         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 368                 return -EINVAL;
 369
 370         override_cred = prepare_creds();
 371         if (!override_cred)
 372                 return -ENOMEM;
 373
 374         override_cred->fsuid = override_cred->uid;
 375         override_cred->fsgid = override_cred->gid;
 376
 377         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 378                 /* Clear the capabilities if we switch to a non-root user */
 379                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 380                 if (!uid_eq(override_cred->uid, root_uid))
 381                         cap_clear(override_cred->cap_effective);
 382                 else
 383                         override_cred->cap_effective =
 384                                 override_cred->cap_permitted;
 385         }
 386
 387         old_cred = override_creds(override_cred);
 388 retry:
 389         res = user_path_at(dfd, filename, lookup_flags, &path);
 390         if (res)
 391                 goto out;
 392
 393         inode = d_backing_inode(path.dentry);
 394
 395         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 396                 /*
 397                  * MAY_EXEC on regular files is denied if the fs is mounted
 398                  * with the "noexec" flag.
 399                  */
 400                 res = -EACCES;
 401                 if (path_noexec(&path))
 402                         goto out_path_release;
 403         }
 404
 405         res = inode_permission(inode, mode | MAY_ACCESS);
 406         /* SuS v2 requires we report a read only fs too */
 407         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 408                 goto out_path_release;
 409         /*
 410          * This is a rare case where using __mnt_is_readonly()
 411          * is OK without a mnt_want/drop_write() pair.  Since
 412          * no actual write to the fs is performed here, we do
 413          * not need to telegraph to that to anyone.
 414          *
 415          * By doing this, we accept that this access is
 416          * inherently racy and know that the fs may change
 417          * state before we even see this result.
 418          */
 419         if (__mnt_is_readonly(path.mnt))
 420                 res = -EROFS;
 421
 422 out_path_release:
 423         path_put(&path);
 424         if (retry_estale(res, lookup_flags)) {
 425                 lookup_flags |= LOOKUP_REVAL;
 426                 goto retry;
 427         }
 428 out:
 429         revert_creds(old_cred);
 430         put_cred(override_cred);
 431         return res;
 432 }
 433
 434 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 435 {
 436         return sys_faccessat(AT_FDCWD, filename, mode);
 437 }
 438
 439 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 440 {
 441         struct path path;
 442         int error;
 443         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 444 retry:
 445         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 446         if (error)
 447                 goto out;
 448
 449         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 450         if (error)
 451                 goto dput_and_out;
 452
 453         set_fs_pwd(current->fs, &path);
 454
 455 dput_and_out:
 456         path_put(&path);
 457         if (retry_estale(error, lookup_flags)) {
 458                 lookup_flags |= LOOKUP_REVAL;
 459                 goto retry;
 460         }
 461 out:
 462         return error;
 463 }
 464
 465 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 466 {
 467         struct fd f = fdget_raw(fd);
 468         struct inode *inode;
 469         int error = -EBADF;
 470
 471         error = -EBADF;
 472         if (!f.file)
 473                 goto out;
 474
 475         inode = file_inode(f.file);
 476
 477         error = -ENOTDIR;
 478         if (!S_ISDIR(inode->i_mode))
 479                 goto out_putf;
 480
 481         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 482         if (!error)
 483                 set_fs_pwd(current->fs, &f.file->f_path);
 484 out_putf:
 485         fdput(f);
 486 out:
 487         return error;
 488 }
 489
 490 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 491 {
 492         struct path path;
 493         int error;
 494         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 495 retry:
 496         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 497         if (error)
 498                 goto out;
 499
 500         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 501         if (error)
 502                 goto dput_and_out;
 503
 504         error = -EPERM;
 505         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 506                 goto dput_and_out;
 507         error = security_path_chroot(&path);
 508         if (error)
 509                 goto dput_and_out;
 510
 511         set_fs_root(current->fs, &path);
 512         error = 0;
 513 dput_and_out:
 514         path_put(&path);
 515         if (retry_estale(error, lookup_flags)) {
 516                 lookup_flags |= LOOKUP_REVAL;
 517                 goto retry;
 518         }
 519 out:
 520         return error;
 521 }
 522
 523 static int chmod_common(const struct path *path, umode_t mode)
 524 {
 525         struct inode *inode = path->dentry->d_inode;
 526         struct inode *delegated_inode = NULL;
 527         struct iattr newattrs;
 528         int error;
 529
 530         error = mnt_want_write(path->mnt);
 531         if (error)
 532                 return error;
 533 retry_deleg:
 534         inode_lock(inode);
 535         error = security_path_chmod(path, mode);
 536         if (error)
 537                 goto out_unlock;
 538         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 539         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 540         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 541 out_unlock:
 542         inode_unlock(inode);
 543         if (delegated_inode) {
 544                 error = break_deleg_wait(&delegated_inode);
 545                 if (!error)
 546                         goto retry_deleg;
 547         }
 548         mnt_drop_write(path->mnt);
 549         return error;
 550 }
 551
 552 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 553 {
 554         struct fd f = fdget(fd);
 555         int err = -EBADF;
 556
 557         if (f.file) {
 558                 audit_file(f.file);
 559                 err = chmod_common(&f.file->f_path, mode);
 560                 fdput(f);
 561         }
 562         return err;
 563 }
 564
 565 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 566 {
 567         struct path path;
 568         int error;
 569         unsigned int lookup_flags = LOOKUP_FOLLOW;
 570 retry:
 571         error = user_path_at(dfd, filename, lookup_flags, &path);
 572         if (!error) {
 573                 error = chmod_common(&path, mode);
 574                 path_put(&path);
 575                 if (retry_estale(error, lookup_flags)) {
 576                         lookup_flags |= LOOKUP_REVAL;
 577                         goto retry;
 578                 }
 579         }
 580         return error;
 581 }
 582
 583 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 584 {
 585         return sys_fchmodat(AT_FDCWD, filename, mode);
 586 }
 587
 588 static int chown_common(const struct path *path, uid_t user, gid_t group)
 589 {
 590         struct inode *inode = path->dentry->d_inode;
 591         struct inode *delegated_inode = NULL;
 592         int error;
 593         struct iattr newattrs;
 594         kuid_t uid;
 595         kgid_t gid;
 596
 597         uid = make_kuid(current_user_ns(), user);
 598         gid = make_kgid(current_user_ns(), group);
 599
 600 retry_deleg:
 601         newattrs.ia_valid =  ATTR_CTIME;
 602         if (user != (uid_t) -1) {
 603                 if (!uid_valid(uid))
 604                         return -EINVAL;
 605                 newattrs.ia_valid |= ATTR_UID;
 606                 newattrs.ia_uid = uid;
 607         }
 608         if (group != (gid_t) -1) {
 609                 if (!gid_valid(gid))
 610                         return -EINVAL;
 611                 newattrs.ia_valid |= ATTR_GID;
 612                 newattrs.ia_gid = gid;
 613         }
 614         if (!S_ISDIR(inode->i_mode))
 615                 newattrs.ia_valid |=
 616                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 617         inode_lock(inode);
 618         error = security_path_chown(path, uid, gid);
 619         if (!error)
 620                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 621         inode_unlock(inode);
 622         if (delegated_inode) {
 623                 error = break_deleg_wait(&delegated_inode);
 624                 if (!error)
 625                         goto retry_deleg;
 626         }
 627         return error;
 628 }
 629
 630 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 631                 gid_t, group, int, flag)
 632 {
 633         struct path path;
 634         int error = -EINVAL;
 635         int lookup_flags;
 636
 637         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 638                 goto out;
 639
 640         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 641         if (flag & AT_EMPTY_PATH)
 642                 lookup_flags |= LOOKUP_EMPTY;
 643 retry:
 644         error = user_path_at(dfd, filename, lookup_flags, &path);
 645         if (error)
 646                 goto out;
 647         error = mnt_want_write(path.mnt);
 648         if (error)
 649                 goto out_release;
 650         error = chown_common(&path, user, group);
 651         mnt_drop_write(path.mnt);
 652 out_release:
 653         path_put(&path);
 654         if (retry_estale(error, lookup_flags)) {
 655                 lookup_flags |= LOOKUP_REVAL;
 656                 goto retry;
 657         }
 658 out:
 659         return error;
 660 }
 661
 662 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 663 {
 664         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 665 }
 666
 667 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 668 {
 669         return sys_fchownat(AT_FDCWD, filename, user, group,
 670                             AT_SYMLINK_NOFOLLOW);
 671 }
 672
 673 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 674 {
 675         struct fd f = fdget(fd);
 676         int error = -EBADF;
 677
 678         if (!f.file)
 679                 goto out;
 680
 681         error = mnt_want_write_file(f.file);
 682         if (error)
 683                 goto out_fput;
 684         audit_file(f.file);
 685         error = chown_common(&f.file->f_path, user, group);
 686         mnt_drop_write_file(f.file);
 687 out_fput:
 688         fdput(f);
 689 out:
 690         return error;
 691 }
 692
 693 int open_check_o_direct(struct file *f)
 694 {
 695         /* NB: we're sure to have correct a_ops only after f_op->open */
 696         if (f->f_flags & O_DIRECT) {
 697                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 698                         return -EINVAL;
 699         }
 700         return 0;
 701 }
 702 EXPORT_SYMBOL_GPL(open_check_o_direct);
 703
 704 static int do_dentry_open(struct file *f,
 705                           struct inode *inode,
 706                           int (*open)(struct inode *, struct file *),
 707                           const struct cred *cred)
 708 {
 709         static const struct file_operations empty_fops = {};
 710         int error;
 711
 712         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 713                                 FMODE_PREAD | FMODE_PWRITE;
 714
 715         path_get(&f->f_path);
 716         f->f_inode = inode;
 717         f->f_mapping = inode->i_mapping;
 718
 719         if (unlikely(f->f_flags & O_PATH)) {
 720                 f->f_mode = FMODE_PATH;
 721                 f->f_op = &empty_fops;
 722                 return 0;
 723         }
 724
 725         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 726                 error = get_write_access(inode);
 727                 if (unlikely(error))
 728                         goto cleanup_file;
 729                 error = __mnt_want_write(f->f_path.mnt);
 730                 if (unlikely(error)) {
 731                         put_write_access(inode);
 732                         goto cleanup_file;
 733                 }
 734                 f->f_mode |= FMODE_WRITER;
 735         }
 736
 737         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 738         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 739                 f->f_mode |= FMODE_ATOMIC_POS;
 740
 741         f->f_op = fops_get(inode->i_fop);
 742         if (unlikely(WARN_ON(!f->f_op))) {
 743                 error = -ENODEV;
 744                 goto cleanup_all;
 745         }
 746
 747         error = security_file_open(f, cred);
 748         if (error)
 749                 goto cleanup_all;
 750
 751         error = break_lease(locks_inode(f), f->f_flags);
 752         if (error)
 753                 goto cleanup_all;
 754
 755         if (!open)
 756                 open = f->f_op->open;
 757         if (open) {
 758                 error = open(inode, f);
 759                 if (error)
 760                         goto cleanup_all;
 761         }
 762         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 763                 i_readcount_inc(inode);
 764         if ((f->f_mode & FMODE_READ) &&
 765              likely(f->f_op->read || f->f_op->read_iter))
 766                 f->f_mode |= FMODE_CAN_READ;
 767         if ((f->f_mode & FMODE_WRITE) &&
 768              likely(f->f_op->write || f->f_op->write_iter))
 769                 f->f_mode |= FMODE_CAN_WRITE;
 770
 771         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 772
 773         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 774
 775         return 0;
 776
 777 cleanup_all:
 778         fops_put(f->f_op);
 779         if (f->f_mode & FMODE_WRITER) {
 780                 put_write_access(inode);
 781                 __mnt_drop_write(f->f_path.mnt);
 782         }
 783 cleanup_file:
 784         path_put(&f->f_path);
 785         f->f_path.mnt = NULL;
 786         f->f_path.dentry = NULL;
 787         f->f_inode = NULL;
 788         return error;
 789 }
 790
 791 /**
 792  * finish_open - finish opening a file
 793  * @file: file pointer
 794  * @dentry: pointer to dentry
 795  * @open: open callback
 796  * @opened: state of open
 797  *
 798  * This can be used to finish opening a file passed to i_op->atomic_open().
 799  *
 800  * If the open callback is set to NULL, then the standard f_op->open()
 801  * filesystem callback is substituted.
 802  *
 803  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 804  * the return value of d_splice_alias(), then the caller needs to perform dput()
 805  * on it after finish_open().
 806  *
 807  * On successful return @file is a fully instantiated open file.  After this, if
 808  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 809  *
 810  * Returns zero on success or -errno if the open failed.
 811  */
 812 int finish_open(struct file *file, struct dentry *dentry,
 813                 int (*open)(struct inode *, struct file *),
 814                 int *opened)
 815 {
 816         int error;
 817         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 818
 819         file->f_path.dentry = dentry;
 820         error = do_dentry_open(file, d_backing_inode(dentry), open,
 821                                current_cred());
 822         if (!error)
 823                 *opened |= FILE_OPENED;
 824
 825         return error;
 826 }
 827 EXPORT_SYMBOL(finish_open);
 828
 829 /**
 830  * finish_no_open - finish ->atomic_open() without opening the file
 831  *
 832  * @file: file pointer
 833  * @dentry: dentry or NULL (as returned from ->lookup())
 834  *
 835  * This can be used to set the result of a successful lookup in ->atomic_open().
 836  *
 837  * NB: unlike finish_open() this function does consume the dentry reference and
 838  * the caller need not dput() it.
 839  *
 840  * Returns "1" which must be the return value of ->atomic_open() after having
 841  * called this function.
 842  */
 843 int finish_no_open(struct file *file, struct dentry *dentry)
 844 {
 845         file->f_path.dentry = dentry;
 846         return 1;
 847 }
 848 EXPORT_SYMBOL(finish_no_open);
 849
 850 char *file_path(struct file *filp, char *buf, int buflen)
 851 {
 852         return d_path(&filp->f_path, buf, buflen);
 853 }
 854 EXPORT_SYMBOL(file_path);
 855
 856 /**
 857  * vfs_open - open the file at the given path
 858  * @path: path to open
 859  * @file: newly allocated file with f_flag initialized
 860  * @cred: credentials to use
 861  */
 862 int vfs_open(const struct path *path, struct file *file,
 863              const struct cred *cred)
 864 {
 865         struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags);
 866
 867         if (IS_ERR(dentry))
 868                 return PTR_ERR(dentry);
 869
 870         file->f_path = *path;
 871         return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 872 }
 873
 874 struct file *dentry_open(const struct path *path, int flags,
 875                          const struct cred *cred)
 876 {
 877         int error;
 878         struct file *f;
 879
 880         validate_creds(cred);
 881
 882         /* We must always pass in a valid mount pointer. */
 883         BUG_ON(!path->mnt);
 884
 885         f = get_empty_filp();
 886         if (!IS_ERR(f)) {
 887                 f->f_flags = flags;
 888                 error = vfs_open(path, f, cred);
 889                 if (!error) {
 890                         /* from now on we need fput() to dispose of f */
 891                         error = open_check_o_direct(f);
 892                         if (error) {
 893                                 fput(f);
 894                                 f = ERR_PTR(error);
 895                         }
 896                 } else {
 897                         put_filp(f);
 898                         f = ERR_PTR(error);
 899                 }
 900         }
 901         return f;
 902 }
 903 EXPORT_SYMBOL(dentry_open);
 904
 905 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 906 {
 907         int lookup_flags = 0;
 908         int acc_mode = ACC_MODE(flags);
 909
 910         if (flags & (O_CREAT | __O_TMPFILE))
 911                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 912         else
 913                 op->mode = 0;
 914
 915         /* Must never be set by userspace */
 916         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 917
 918         /*
 919          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 920          * check for O_DSYNC if the need any syncing at all we enforce it's
 921          * always set instead of having to deal with possibly weird behaviour
 922          * for malicious applications setting only __O_SYNC.
 923          */
 924         if (flags & __O_SYNC)
 925                 flags |= O_DSYNC;
 926
 927         if (flags & __O_TMPFILE) {
 928                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 929                         return -EINVAL;
 930                 if (!(acc_mode & MAY_WRITE))
 931                         return -EINVAL;
 932         } else if (flags & O_PATH) {
 933                 /*
 934                  * If we have O_PATH in the open flag. Then we
 935                  * cannot have anything other than the below set of flags
 936                  */
 937                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 938                 acc_mode = 0;
 939         }
 940
 941         op->open_flag = flags;
 942
 943         /* O_TRUNC implies we need access checks for write permissions */
 944         if (flags & O_TRUNC)
 945                 acc_mode |= MAY_WRITE;
 946
 947         /* Allow the LSM permission hook to distinguish append
 948            access from general write access. */
 949         if (flags & O_APPEND)
 950                 acc_mode |= MAY_APPEND;
 951
 952         op->acc_mode = acc_mode;
 953
 954         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 955
 956         if (flags & O_CREAT) {
 957                 op->intent |= LOOKUP_CREATE;
 958                 if (flags & O_EXCL)
 959                         op->intent |= LOOKUP_EXCL;
 960         }
 961
 962         if (flags & O_DIRECTORY)
 963                 lookup_flags |= LOOKUP_DIRECTORY;
 964         if (!(flags & O_NOFOLLOW))
 965                 lookup_flags |= LOOKUP_FOLLOW;
 966         op->lookup_flags = lookup_flags;
 967         return 0;
 968 }
 969
 970 /**
 971  * file_open_name - open file and return file pointer
 972  *
 973  * @name:       struct filename containing path to open
 974  * @flags:      open flags as per the open(2) second argument
 975  * @mode:       mode for the new file if O_CREAT is set, else ignored
 976  *
 977  * This is the helper to open a file from kernelspace if you really
 978  * have to.  But in generally you should not do this, so please move
 979  * along, nothing to see here..
 980  */
 981 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 982 {
 983         struct open_flags op;
 984         int err = build_open_flags(flags, mode, &op);
 985         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 986 }
 987
 988 /**
 989  * filp_open - open file and return file pointer
 990  *
 991  * @filename:   path to open
 992  * @flags:      open flags as per the open(2) second argument
 993  * @mode:       mode for the new file if O_CREAT is set, else ignored
 994  *
 995  * This is the helper to open a file from kernelspace if you really
 996  * have to.  But in generally you should not do this, so please move
 997  * along, nothing to see here..
 998  */
 999 struct file *filp_open(const char *filename, int flags, umode_t mode)
1000 {
1001         struct filename *name = getname_kernel(filename);
1002         struct file *file = ERR_CAST(name);
1003
1004         if (!IS_ERR(name)) {
1005                 file = file_open_name(name, flags, mode);
1006                 putname(name);
1007         }
1008         return file;
1009 }
1010 EXPORT_SYMBOL(filp_open);
1011
1012 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1013                             const char *filename, int flags, umode_t mode)
1014 {
1015         struct open_flags op;
1016         int err = build_open_flags(flags, mode, &op);
1017         if (err)
1018                 return ERR_PTR(err);
1019         return do_file_open_root(dentry, mnt, filename, &op);
1020 }
1021 EXPORT_SYMBOL(file_open_root);
1022
1023 struct file *filp_clone_open(struct file *oldfile)
1024 {
1025         struct file *file;
1026         int retval;
1027
1028         file = get_empty_filp();
1029         if (IS_ERR(file))
1030                 return file;
1031
1032         file->f_flags = oldfile->f_flags;
1033         retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1034         if (retval) {
1035                 put_filp(file);
1036                 return ERR_PTR(retval);
1037         }
1038
1039         return file;
1040 }
1041 EXPORT_SYMBOL(filp_clone_open);
1042
1043 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1044 {
1045         struct open_flags op;
1046         int fd = build_open_flags(flags, mode, &op);
1047         struct filename *tmp;
1048
1049         if (fd)
1050                 return fd;
1051
1052         tmp = getname(filename);
1053         if (IS_ERR(tmp))
1054                 return PTR_ERR(tmp);
1055
1056         fd = get_unused_fd_flags(flags);
1057         if (fd >= 0) {
1058                 struct file *f = do_filp_open(dfd, tmp, &op);
1059                 if (IS_ERR(f)) {
1060                         put_unused_fd(fd);
1061                         fd = PTR_ERR(f);
1062                 } else {
1063                         fsnotify_open(f);
1064                         fd_install(fd, f);
1065                         trace_do_sys_open(tmp->name, flags, mode);
1066                 }
1067         }
1068         putname(tmp);
1069         return fd;
1070 }
1071
1072 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1073 {
1074         if (force_o_largefile())
1075                 flags |= O_LARGEFILE;
1076
1077         return do_sys_open(AT_FDCWD, filename, flags, mode);
1078 }
1079
1080 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1081                 umode_t, mode)
1082 {
1083         if (force_o_largefile())
1084                 flags |= O_LARGEFILE;
1085
1086         return do_sys_open(dfd, filename, flags, mode);
1087 }
1088
1089 #ifndef __alpha__
1090
1091 /*
1092  * For backward compatibility?  Maybe this should be moved
1093  * into arch/i386 instead?
1094  */
1095 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1096 {
1097         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1098 }
1099
1100 #endif
1101
1102 /*
1103  * "id" is the POSIX thread ID. We use the
1104  * files pointer for this..
1105  */
1106 int filp_close(struct file *filp, fl_owner_t id)
1107 {
1108         int retval = 0;
1109
1110         if (!file_count(filp)) {
1111                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1112                 return 0;
1113         }
1114
1115         if (filp->f_op->flush)
1116                 retval = filp->f_op->flush(filp, id);
1117
1118         if (likely(!(filp->f_mode & FMODE_PATH))) {
1119                 dnotify_flush(filp, id);
1120                 locks_remove_posix(filp, id);
1121         }
1122         fput(filp);
1123         return retval;
1124 }
1125
1126 EXPORT_SYMBOL(filp_close);
1127
1128 /*
1129  * Careful here! We test whether the file pointer is NULL before
1130  * releasing the fd. This ensures that one clone task can't release
1131  * an fd while another clone is opening it.
1132  */
1133 SYSCALL_DEFINE1(close, unsigned int, fd)
1134 {
1135         int retval = __close_fd(current->files, fd);
1136
1137         /* can't restart close syscall because file table entry was cleared */
1138         if (unlikely(retval == -ERESTARTSYS ||
1139                      retval == -ERESTARTNOINTR ||
1140                      retval == -ERESTARTNOHAND ||
1141                      retval == -ERESTART_RESTARTBLOCK))
1142                 retval = -EINTR;
1143
1144         return retval;
1145 }
1146 EXPORT_SYMBOL(sys_close);
1147
1148 /*
1149  * This routine simulates a hangup on the tty, to arrange that users
1150  * are given clean terminals at login time.
1151  */
1152 SYSCALL_DEFINE0(vhangup)
1153 {
1154         if (capable(CAP_SYS_TTY_CONFIG)) {
1155                 tty_vhangup_self();
1156                 return 0;
1157         }
1158         return -EPERM;
1159 }
1160
1161 /*
1162  * Called when an inode is about to be open.
1163  * We use this to disallow opening large files on 32bit systems if
1164  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1165  * on this flag in sys_open.
1166  */
1167 int generic_file_open(struct inode * inode, struct file * filp)
1168 {
1169         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1170                 return -EOVERFLOW;
1171         return 0;
1172 }
1173
1174 EXPORT_SYMBOL(generic_file_open);
1175
1176 /*
1177  * This is used by subsystems that don't want seekable
1178  * file descriptors. The function is not supposed to ever fail, the only
1179  * reason it returns an 'int' and not 'void' is so that it can be plugged
1180  * directly into file_operations structure.
1181  */
1182 int nonseekable_open(struct inode *inode, struct file *filp)
1183 {
1184         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1185         return 0;
1186 }
1187
1188 EXPORT_SYMBOL(nonseekable_open);