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, 0);
 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         /* Check IS_APPEND on real upper inode */
 201         if (IS_APPEND(file_inode(f.file)))
 202                 goto out_putf;
 203
 204         sb_start_write(inode->i_sb);
 205         error = locks_verify_truncate(inode, f.file, length);
 206         if (!error)
 207                 error = security_path_truncate(&f.file->f_path);
 208         if (!error)
 209                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 210         sb_end_write(inode->i_sb);
 211 out_putf:
 212         fdput(f);
 213 out:
 214         return error;
 215 }
 216
 217 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 218 {
 219         return do_sys_ftruncate(fd, length, 1);
 220 }
 221
 222 #ifdef CONFIG_COMPAT
 223 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 224 {
 225         return do_sys_ftruncate(fd, length, 1);
 226 }
 227 #endif
 228
 229 /* LFS versions of truncate are only needed on 32 bit machines */
 230 #if BITS_PER_LONG == 32
 231 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 232 {
 233         return do_sys_truncate(path, length);
 234 }
 235
 236 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 237 {
 238         return do_sys_ftruncate(fd, length, 0);
 239 }
 240 #endif /* BITS_PER_LONG == 32 */
 241
 242
 243 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 244 {
 245         struct inode *inode = file_inode(file);
 246         long ret;
 247
 248         if (offset < 0 || len <= 0)
 249                 return -EINVAL;
 250
 251         /* Return error if mode is not supported */
 252         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 253                 return -EOPNOTSUPP;
 254
 255         /* Punch hole and zero range are mutually exclusive */
 256         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 257             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 258                 return -EOPNOTSUPP;
 259
 260         /* Punch hole must have keep size set */
 261         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 262             !(mode & FALLOC_FL_KEEP_SIZE))
 263                 return -EOPNOTSUPP;
 264
 265         /* Collapse range should only be used exclusively. */
 266         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 267             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 268                 return -EINVAL;
 269
 270         /* Insert range should only be used exclusively. */
 271         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 272             (mode & ~FALLOC_FL_INSERT_RANGE))
 273                 return -EINVAL;
 274
 275         /* Unshare range should only be used with allocate mode. */
 276         if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 277             (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 278                 return -EINVAL;
 279
 280         if (!(file->f_mode & FMODE_WRITE))
 281                 return -EBADF;
 282
 283         /*
 284          * We can only allow pure fallocate on append only files
 285          */
 286         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 287                 return -EPERM;
 288
 289         if (IS_IMMUTABLE(inode))
 290                 return -EPERM;
 291
 292         /*
 293          * We cannot allow any fallocate operation on an active swapfile
 294          */
 295         if (IS_SWAPFILE(inode))
 296                 return -ETXTBSY;
 297
 298         /*
 299          * Revalidate the write permissions, in case security policy has
 300          * changed since the files were opened.
 301          */
 302         ret = security_file_permission(file, MAY_WRITE);
 303         if (ret)
 304                 return ret;
 305
 306         if (S_ISFIFO(inode->i_mode))
 307                 return -ESPIPE;
 308
 309         if (S_ISDIR(inode->i_mode))
 310                 return -EISDIR;
 311
 312         if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 313                 return -ENODEV;
 314
 315         /* Check for wrap through zero too */
 316         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 317                 return -EFBIG;
 318
 319         if (!file->f_op->fallocate)
 320                 return -EOPNOTSUPP;
 321
 322         file_start_write(file);
 323         ret = file->f_op->fallocate(file, mode, offset, len);
 324
 325         /*
 326          * Create inotify and fanotify events.
 327          *
 328          * To keep the logic simple always create events if fallocate succeeds.
 329          * This implies that events are even created if the file size remains
 330          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 331          */
 332         if (ret == 0)
 333                 fsnotify_modify(file);
 334
 335         file_end_write(file);
 336         return ret;
 337 }
 338 EXPORT_SYMBOL_GPL(vfs_fallocate);
 339
 340 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 341 {
 342         struct fd f = fdget(fd);
 343         int error = -EBADF;
 344
 345         if (f.file) {
 346                 error = vfs_fallocate(f.file, mode, offset, len);
 347                 fdput(f);
 348         }
 349         return error;
 350 }
 351
 352 /*
 353  * access() needs to use the real uid/gid, not the effective uid/gid.
 354  * We do this by temporarily clearing all FS-related capabilities and
 355  * switching the fsuid/fsgid around to the real ones.
 356  */
 357 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 358 {
 359         const struct cred *old_cred;
 360         struct cred *override_cred;
 361         struct path path;
 362         struct inode *inode;
 363         int res;
 364         unsigned int lookup_flags = LOOKUP_FOLLOW;
 365
 366         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 367                 return -EINVAL;
 368
 369         override_cred = prepare_creds();
 370         if (!override_cred)
 371                 return -ENOMEM;
 372
 373         override_cred->fsuid = override_cred->uid;
 374         override_cred->fsgid = override_cred->gid;
 375
 376         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 377                 /* Clear the capabilities if we switch to a non-root user */
 378                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 379                 if (!uid_eq(override_cred->uid, root_uid))
 380                         cap_clear(override_cred->cap_effective);
 381                 else
 382                         override_cred->cap_effective =
 383                                 override_cred->cap_permitted;
 384         }
 385
 386         old_cred = override_creds(override_cred);
 387 retry:
 388         res = user_path_at(dfd, filename, lookup_flags, &path);
 389         if (res)
 390                 goto out;
 391
 392         inode = d_backing_inode(path.dentry);
 393
 394         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 395                 /*
 396                  * MAY_EXEC on regular files is denied if the fs is mounted
 397                  * with the "noexec" flag.
 398                  */
 399                 res = -EACCES;
 400                 if (path_noexec(&path))
 401                         goto out_path_release;
 402         }
 403
 404         res = inode_permission(inode, mode | MAY_ACCESS);
 405         /* SuS v2 requires we report a read only fs too */
 406         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 407                 goto out_path_release;
 408         /*
 409          * This is a rare case where using __mnt_is_readonly()
 410          * is OK without a mnt_want/drop_write() pair.  Since
 411          * no actual write to the fs is performed here, we do
 412          * not need to telegraph to that to anyone.
 413          *
 414          * By doing this, we accept that this access is
 415          * inherently racy and know that the fs may change
 416          * state before we even see this result.
 417          */
 418         if (__mnt_is_readonly(path.mnt))
 419                 res = -EROFS;
 420
 421 out_path_release:
 422         path_put(&path);
 423         if (retry_estale(res, lookup_flags)) {
 424                 lookup_flags |= LOOKUP_REVAL;
 425                 goto retry;
 426         }
 427 out:
 428         revert_creds(old_cred);
 429         put_cred(override_cred);
 430         return res;
 431 }
 432
 433 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 434 {
 435         return sys_faccessat(AT_FDCWD, filename, mode);
 436 }
 437
 438 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 439 {
 440         struct path path;
 441         int error;
 442         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 443 retry:
 444         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 445         if (error)
 446                 goto out;
 447
 448         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 449         if (error)
 450                 goto dput_and_out;
 451
 452         set_fs_pwd(current->fs, &path);
 453
 454 dput_and_out:
 455         path_put(&path);
 456         if (retry_estale(error, lookup_flags)) {
 457                 lookup_flags |= LOOKUP_REVAL;
 458                 goto retry;
 459         }
 460 out:
 461         return error;
 462 }
 463
 464 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 465 {
 466         struct fd f = fdget_raw(fd);
 467         int error;
 468
 469         error = -EBADF;
 470         if (!f.file)
 471                 goto out;
 472
 473         error = -ENOTDIR;
 474         if (!d_can_lookup(f.file->f_path.dentry))
 475                 goto out_putf;
 476
 477         error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 478         if (!error)
 479                 set_fs_pwd(current->fs, &f.file->f_path);
 480 out_putf:
 481         fdput(f);
 482 out:
 483         return error;
 484 }
 485
 486 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 487 {
 488         struct path path;
 489         int error;
 490         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 491 retry:
 492         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 493         if (error)
 494                 goto out;
 495
 496         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 497         if (error)
 498                 goto dput_and_out;
 499
 500         error = -EPERM;
 501         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 502                 goto dput_and_out;
 503         error = security_path_chroot(&path);
 504         if (error)
 505                 goto dput_and_out;
 506
 507         set_fs_root(current->fs, &path);
 508         error = 0;
 509 dput_and_out:
 510         path_put(&path);
 511         if (retry_estale(error, lookup_flags)) {
 512                 lookup_flags |= LOOKUP_REVAL;
 513                 goto retry;
 514         }
 515 out:
 516         return error;
 517 }
 518
 519 static int chmod_common(const struct path *path, umode_t mode)
 520 {
 521         struct inode *inode = path->dentry->d_inode;
 522         struct inode *delegated_inode = NULL;
 523         struct iattr newattrs;
 524         int error;
 525
 526         error = mnt_want_write(path->mnt);
 527         if (error)
 528                 return error;
 529 retry_deleg:
 530         inode_lock(inode);
 531         error = security_path_chmod(path, mode);
 532         if (error)
 533                 goto out_unlock;
 534         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 535         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 536         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 537 out_unlock:
 538         inode_unlock(inode);
 539         if (delegated_inode) {
 540                 error = break_deleg_wait(&delegated_inode);
 541                 if (!error)
 542                         goto retry_deleg;
 543         }
 544         mnt_drop_write(path->mnt);
 545         return error;
 546 }
 547
 548 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 549 {
 550         struct fd f = fdget(fd);
 551         int err = -EBADF;
 552
 553         if (f.file) {
 554                 audit_file(f.file);
 555                 err = chmod_common(&f.file->f_path, mode);
 556                 fdput(f);
 557         }
 558         return err;
 559 }
 560
 561 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 562 {
 563         struct path path;
 564         int error;
 565         unsigned int lookup_flags = LOOKUP_FOLLOW;
 566 retry:
 567         error = user_path_at(dfd, filename, lookup_flags, &path);
 568         if (!error) {
 569                 error = chmod_common(&path, mode);
 570                 path_put(&path);
 571                 if (retry_estale(error, lookup_flags)) {
 572                         lookup_flags |= LOOKUP_REVAL;
 573                         goto retry;
 574                 }
 575         }
 576         return error;
 577 }
 578
 579 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 580 {
 581         return sys_fchmodat(AT_FDCWD, filename, mode);
 582 }
 583
 584 static int chown_common(const struct path *path, uid_t user, gid_t group)
 585 {
 586         struct inode *inode = path->dentry->d_inode;
 587         struct inode *delegated_inode = NULL;
 588         int error;
 589         struct iattr newattrs;
 590         kuid_t uid;
 591         kgid_t gid;
 592
 593         uid = make_kuid(current_user_ns(), user);
 594         gid = make_kgid(current_user_ns(), group);
 595
 596 retry_deleg:
 597         newattrs.ia_valid =  ATTR_CTIME;
 598         if (user != (uid_t) -1) {
 599                 if (!uid_valid(uid))
 600                         return -EINVAL;
 601                 newattrs.ia_valid |= ATTR_UID;
 602                 newattrs.ia_uid = uid;
 603         }
 604         if (group != (gid_t) -1) {
 605                 if (!gid_valid(gid))
 606                         return -EINVAL;
 607                 newattrs.ia_valid |= ATTR_GID;
 608                 newattrs.ia_gid = gid;
 609         }
 610         if (!S_ISDIR(inode->i_mode))
 611                 newattrs.ia_valid |=
 612                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 613         inode_lock(inode);
 614         error = security_path_chown(path, uid, gid);
 615         if (!error)
 616                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 617         inode_unlock(inode);
 618         if (delegated_inode) {
 619                 error = break_deleg_wait(&delegated_inode);
 620                 if (!error)
 621                         goto retry_deleg;
 622         }
 623         return error;
 624 }
 625
 626 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 627                 gid_t, group, int, flag)
 628 {
 629         struct path path;
 630         int error = -EINVAL;
 631         int lookup_flags;
 632
 633         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 634                 goto out;
 635
 636         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 637         if (flag & AT_EMPTY_PATH)
 638                 lookup_flags |= LOOKUP_EMPTY;
 639 retry:
 640         error = user_path_at(dfd, filename, lookup_flags, &path);
 641         if (error)
 642                 goto out;
 643         error = mnt_want_write(path.mnt);
 644         if (error)
 645                 goto out_release;
 646         error = chown_common(&path, user, group);
 647         mnt_drop_write(path.mnt);
 648 out_release:
 649         path_put(&path);
 650         if (retry_estale(error, lookup_flags)) {
 651                 lookup_flags |= LOOKUP_REVAL;
 652                 goto retry;
 653         }
 654 out:
 655         return error;
 656 }
 657
 658 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 659 {
 660         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 661 }
 662
 663 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 664 {
 665         return sys_fchownat(AT_FDCWD, filename, user, group,
 666                             AT_SYMLINK_NOFOLLOW);
 667 }
 668
 669 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 670 {
 671         struct fd f = fdget(fd);
 672         int error = -EBADF;
 673
 674         if (!f.file)
 675                 goto out;
 676
 677         error = mnt_want_write_file_path(f.file);
 678         if (error)
 679                 goto out_fput;
 680         audit_file(f.file);
 681         error = chown_common(&f.file->f_path, user, group);
 682         mnt_drop_write_file_path(f.file);
 683 out_fput:
 684         fdput(f);
 685 out:
 686         return error;
 687 }
 688
 689 int open_check_o_direct(struct file *f)
 690 {
 691         /* NB: we're sure to have correct a_ops only after f_op->open */
 692         if (f->f_flags & O_DIRECT) {
 693                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 694                         return -EINVAL;
 695         }
 696         return 0;
 697 }
 698 EXPORT_SYMBOL_GPL(open_check_o_direct);
 699
 700 static int do_dentry_open(struct file *f,
 701                           struct inode *inode,
 702                           int (*open)(struct inode *, struct file *),
 703                           const struct cred *cred)
 704 {
 705         static const struct file_operations empty_fops = {};
 706         int error;
 707
 708         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 709                                 FMODE_PREAD | FMODE_PWRITE;
 710
 711         path_get(&f->f_path);
 712         f->f_inode = inode;
 713         f->f_mapping = inode->i_mapping;
 714
 715         /* Ensure that we skip any errors that predate opening of the file */
 716         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 717
 718         if (unlikely(f->f_flags & O_PATH)) {
 719                 f->f_mode = FMODE_PATH;
 720                 f->f_op = &empty_fops;
 721                 return 0;
 722         }
 723
 724         /* Any file opened for execve()/uselib() has to be a regular file. */
 725         if (unlikely(f->f_flags & FMODE_EXEC && !S_ISREG(inode->i_mode))) {
 726                 error = -EACCES;
 727                 goto cleanup_file;
 728         }
 729
 730         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 731                 error = get_write_access(inode);
 732                 if (unlikely(error))
 733                         goto cleanup_file;
 734                 error = __mnt_want_write(f->f_path.mnt);
 735                 if (unlikely(error)) {
 736                         put_write_access(inode);
 737                         goto cleanup_file;
 738                 }
 739                 f->f_mode |= FMODE_WRITER;
 740         }
 741
 742         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 743         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 744                 f->f_mode |= FMODE_ATOMIC_POS;
 745
 746         f->f_op = fops_get(inode->i_fop);
 747         if (unlikely(WARN_ON(!f->f_op))) {
 748                 error = -ENODEV;
 749                 goto cleanup_all;
 750         }
 751
 752         error = security_file_open(f, cred);
 753         if (error)
 754                 goto cleanup_all;
 755
 756         error = break_lease(locks_inode(f), f->f_flags);
 757         if (error)
 758                 goto cleanup_all;
 759
 760         if (!open)
 761                 open = f->f_op->open;
 762         if (open) {
 763                 error = open(inode, f);
 764                 if (error)
 765                         goto cleanup_all;
 766         }
 767         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 768                 i_readcount_inc(inode);
 769         if ((f->f_mode & FMODE_READ) &&
 770              likely(f->f_op->read || f->f_op->read_iter))
 771                 f->f_mode |= FMODE_CAN_READ;
 772         if ((f->f_mode & FMODE_WRITE) &&
 773              likely(f->f_op->write || f->f_op->write_iter))
 774                 f->f_mode |= FMODE_CAN_WRITE;
 775
 776         f->f_write_hint = WRITE_LIFE_NOT_SET;
 777         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 778
 779         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 780
 781         return 0;
 782
 783 cleanup_all:
 784         fops_put(f->f_op);
 785         if (f->f_mode & FMODE_WRITER) {
 786                 put_write_access(inode);
 787                 __mnt_drop_write(f->f_path.mnt);
 788         }
 789 cleanup_file:
 790         path_put(&f->f_path);
 791         f->f_path.mnt = NULL;
 792         f->f_path.dentry = NULL;
 793         f->f_inode = NULL;
 794         return error;
 795 }
 796
 797 /**
 798  * finish_open - finish opening a file
 799  * @file: file pointer
 800  * @dentry: pointer to dentry
 801  * @open: open callback
 802  * @opened: state of open
 803  *
 804  * This can be used to finish opening a file passed to i_op->atomic_open().
 805  *
 806  * If the open callback is set to NULL, then the standard f_op->open()
 807  * filesystem callback is substituted.
 808  *
 809  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 810  * the return value of d_splice_alias(), then the caller needs to perform dput()
 811  * on it after finish_open().
 812  *
 813  * On successful return @file is a fully instantiated open file.  After this, if
 814  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 815  *
 816  * Returns zero on success or -errno if the open failed.
 817  */
 818 int finish_open(struct file *file, struct dentry *dentry,
 819                 int (*open)(struct inode *, struct file *),
 820                 int *opened)
 821 {
 822         int error;
 823         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 824
 825         file->f_path.dentry = dentry;
 826         error = do_dentry_open(file, d_backing_inode(dentry), open,
 827                                current_cred());
 828         if (!error)
 829                 *opened |= FILE_OPENED;
 830
 831         return error;
 832 }
 833 EXPORT_SYMBOL(finish_open);
 834
 835 /**
 836  * finish_no_open - finish ->atomic_open() without opening the file
 837  *
 838  * @file: file pointer
 839  * @dentry: dentry or NULL (as returned from ->lookup())
 840  *
 841  * This can be used to set the result of a successful lookup in ->atomic_open().
 842  *
 843  * NB: unlike finish_open() this function does consume the dentry reference and
 844  * the caller need not dput() it.
 845  *
 846  * Returns "1" which must be the return value of ->atomic_open() after having
 847  * called this function.
 848  */
 849 int finish_no_open(struct file *file, struct dentry *dentry)
 850 {
 851         file->f_path.dentry = dentry;
 852         return 1;
 853 }
 854 EXPORT_SYMBOL(finish_no_open);
 855
 856 char *file_path(struct file *filp, char *buf, int buflen)
 857 {
 858         return d_path(&filp->f_path, buf, buflen);
 859 }
 860 EXPORT_SYMBOL(file_path);
 861
 862 /**
 863  * vfs_open - open the file at the given path
 864  * @path: path to open
 865  * @file: newly allocated file with f_flag initialized
 866  * @cred: credentials to use
 867  */
 868 int vfs_open(const struct path *path, struct file *file,
 869              const struct cred *cred)
 870 {
 871         struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags, 0);
 872
 873         if (IS_ERR(dentry))
 874                 return PTR_ERR(dentry);
 875
 876         file->f_path = *path;
 877         return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 878 }
 879
 880 struct file *dentry_open(const struct path *path, int flags,
 881                          const struct cred *cred)
 882 {
 883         int error;
 884         struct file *f;
 885
 886         validate_creds(cred);
 887
 888         /* We must always pass in a valid mount pointer. */
 889         BUG_ON(!path->mnt);
 890
 891         f = get_empty_filp();
 892         if (!IS_ERR(f)) {
 893                 f->f_flags = flags;
 894                 error = vfs_open(path, f, cred);
 895                 if (!error) {
 896                         /* from now on we need fput() to dispose of f */
 897                         error = open_check_o_direct(f);
 898                         if (error) {
 899                                 fput(f);
 900                                 f = ERR_PTR(error);
 901                         }
 902                 } else {
 903                         put_filp(f);
 904                         f = ERR_PTR(error);
 905                 }
 906         }
 907         return f;
 908 }
 909 EXPORT_SYMBOL(dentry_open);
 910
 911 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 912 {
 913         int lookup_flags = 0;
 914         int acc_mode = ACC_MODE(flags);
 915
 916         /*
 917          * Clear out all open flags we don't know about so that we don't report
 918          * them in fcntl(F_GETFD) or similar interfaces.
 919          */
 920         flags &= VALID_OPEN_FLAGS;
 921
 922         if (flags & (O_CREAT | __O_TMPFILE))
 923                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 924         else
 925                 op->mode = 0;
 926
 927         /* Must never be set by userspace */
 928         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 929
 930         /*
 931          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 932          * check for O_DSYNC if the need any syncing at all we enforce it's
 933          * always set instead of having to deal with possibly weird behaviour
 934          * for malicious applications setting only __O_SYNC.
 935          */
 936         if (flags & __O_SYNC)
 937                 flags |= O_DSYNC;
 938
 939         if (flags & __O_TMPFILE) {
 940                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 941                         return -EINVAL;
 942                 if (!(acc_mode & MAY_WRITE))
 943                         return -EINVAL;
 944         } else if (flags & O_PATH) {
 945                 /*
 946                  * If we have O_PATH in the open flag. Then we
 947                  * cannot have anything other than the below set of flags
 948                  */
 949                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 950                 acc_mode = 0;
 951         }
 952
 953         op->open_flag = flags;
 954
 955         /* O_TRUNC implies we need access checks for write permissions */
 956         if (flags & O_TRUNC)
 957                 acc_mode |= MAY_WRITE;
 958
 959         /* Allow the LSM permission hook to distinguish append
 960            access from general write access. */
 961         if (flags & O_APPEND)
 962                 acc_mode |= MAY_APPEND;
 963
 964         op->acc_mode = acc_mode;
 965
 966         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 967
 968         if (flags & O_CREAT) {
 969                 op->intent |= LOOKUP_CREATE;
 970                 if (flags & O_EXCL)
 971                         op->intent |= LOOKUP_EXCL;
 972         }
 973
 974         if (flags & O_DIRECTORY)
 975                 lookup_flags |= LOOKUP_DIRECTORY;
 976         if (!(flags & O_NOFOLLOW))
 977                 lookup_flags |= LOOKUP_FOLLOW;
 978         op->lookup_flags = lookup_flags;
 979         return 0;
 980 }
 981
 982 /**
 983  * file_open_name - open file and return file pointer
 984  *
 985  * @name:       struct filename containing path to open
 986  * @flags:      open flags as per the open(2) second argument
 987  * @mode:       mode for the new file if O_CREAT is set, else ignored
 988  *
 989  * This is the helper to open a file from kernelspace if you really
 990  * have to.  But in generally you should not do this, so please move
 991  * along, nothing to see here..
 992  */
 993 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 994 {
 995         struct open_flags op;
 996         int err = build_open_flags(flags, mode, &op);
 997         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 998 }
 999
1000 /**
1001  * filp_open - open file and return file pointer
1002  *
1003  * @filename:   path to open
1004  * @flags:      open flags as per the open(2) second argument
1005  * @mode:       mode for the new file if O_CREAT is set, else ignored
1006  *
1007  * This is the helper to open a file from kernelspace if you really
1008  * have to.  But in generally you should not do this, so please move
1009  * along, nothing to see here..
1010  */
1011 struct file *filp_open(const char *filename, int flags, umode_t mode)
1012 {
1013         struct filename *name = getname_kernel(filename);
1014         struct file *file = ERR_CAST(name);
1015
1016         if (!IS_ERR(name)) {
1017                 file = file_open_name(name, flags, mode);
1018                 putname(name);
1019         }
1020         return file;
1021 }
1022 EXPORT_SYMBOL(filp_open);
1023
1024 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1025                             const char *filename, int flags, umode_t mode)
1026 {
1027         struct open_flags op;
1028         int err = build_open_flags(flags, mode, &op);
1029         if (err)
1030                 return ERR_PTR(err);
1031         return do_file_open_root(dentry, mnt, filename, &op);
1032 }
1033 EXPORT_SYMBOL(file_open_root);
1034
1035 struct file *filp_clone_open(struct file *oldfile)
1036 {
1037         struct file *file;
1038         int retval;
1039
1040         file = get_empty_filp();
1041         if (IS_ERR(file))
1042                 return file;
1043
1044         file->f_flags = oldfile->f_flags;
1045         retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1046         if (retval) {
1047                 put_filp(file);
1048                 return ERR_PTR(retval);
1049         }
1050
1051         return file;
1052 }
1053 EXPORT_SYMBOL(filp_clone_open);
1054
1055 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1056 {
1057         struct open_flags op;
1058         int fd = build_open_flags(flags, mode, &op);
1059         struct filename *tmp;
1060
1061         if (fd)
1062                 return fd;
1063
1064         tmp = getname(filename);
1065         if (IS_ERR(tmp))
1066                 return PTR_ERR(tmp);
1067
1068         fd = get_unused_fd_flags(flags);
1069         if (fd >= 0) {
1070                 struct file *f = do_filp_open(dfd, tmp, &op);
1071                 if (IS_ERR(f)) {
1072                         put_unused_fd(fd);
1073                         fd = PTR_ERR(f);
1074                 } else {
1075                         fsnotify_open(f);
1076                         fd_install(fd, f);
1077                         trace_do_sys_open(tmp->name, flags, mode);
1078                 }
1079         }
1080         putname(tmp);
1081         return fd;
1082 }
1083
1084 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1085 {
1086         if (force_o_largefile())
1087                 flags |= O_LARGEFILE;
1088
1089         return do_sys_open(AT_FDCWD, filename, flags, mode);
1090 }
1091
1092 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1093                 umode_t, mode)
1094 {
1095         if (force_o_largefile())
1096                 flags |= O_LARGEFILE;
1097
1098         return do_sys_open(dfd, filename, flags, mode);
1099 }
1100
1101 #ifdef CONFIG_COMPAT
1102 /*
1103  * Exactly like sys_open(), except that it doesn't set the
1104  * O_LARGEFILE flag.
1105  */
1106 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1107 {
1108         return do_sys_open(AT_FDCWD, filename, flags, mode);
1109 }
1110
1111 /*
1112  * Exactly like sys_openat(), except that it doesn't set the
1113  * O_LARGEFILE flag.
1114  */
1115 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1116 {
1117         return do_sys_open(dfd, filename, flags, mode);
1118 }
1119 #endif
1120
1121 #ifndef __alpha__
1122
1123 /*
1124  * For backward compatibility?  Maybe this should be moved
1125  * into arch/i386 instead?
1126  */
1127 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1128 {
1129         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1130 }
1131
1132 #endif
1133
1134 /*
1135  * "id" is the POSIX thread ID. We use the
1136  * files pointer for this..
1137  */
1138 int filp_close(struct file *filp, fl_owner_t id)
1139 {
1140         int retval = 0;
1141
1142         if (!file_count(filp)) {
1143                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1144                 return 0;
1145         }
1146
1147         if (filp->f_op->flush)
1148                 retval = filp->f_op->flush(filp, id);
1149
1150         if (likely(!(filp->f_mode & FMODE_PATH))) {
1151                 dnotify_flush(filp, id);
1152                 locks_remove_posix(filp, id);
1153         }
1154         fput(filp);
1155         return retval;
1156 }
1157
1158 EXPORT_SYMBOL(filp_close);
1159
1160 /*
1161  * Careful here! We test whether the file pointer is NULL before
1162  * releasing the fd. This ensures that one clone task can't release
1163  * an fd while another clone is opening it.
1164  */
1165 SYSCALL_DEFINE1(close, unsigned int, fd)
1166 {
1167         int retval = __close_fd(current->files, fd);
1168
1169         /* can't restart close syscall because file table entry was cleared */
1170         if (unlikely(retval == -ERESTARTSYS ||
1171                      retval == -ERESTARTNOINTR ||
1172                      retval == -ERESTARTNOHAND ||
1173                      retval == -ERESTART_RESTARTBLOCK))
1174                 retval = -EINTR;
1175
1176         return retval;
1177 }
1178 EXPORT_SYMBOL(sys_close);
1179
1180 /*
1181  * This routine simulates a hangup on the tty, to arrange that users
1182  * are given clean terminals at login time.
1183  */
1184 SYSCALL_DEFINE0(vhangup)
1185 {
1186         if (capable(CAP_SYS_TTY_CONFIG)) {
1187                 tty_vhangup_self();
1188                 return 0;
1189         }
1190         return -EPERM;
1191 }
1192
1193 /*
1194  * Called when an inode is about to be open.
1195  * We use this to disallow opening large files on 32bit systems if
1196  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1197  * on this flag in sys_open.
1198  */
1199 int generic_file_open(struct inode * inode, struct file * filp)
1200 {
1201         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1202                 return -EOVERFLOW;
1203         return 0;
1204 }
1205
1206 EXPORT_SYMBOL(generic_file_open);
1207
1208 /*
1209  * This is used by subsystems that don't want seekable
1210  * file descriptors. The function is not supposed to ever fail, the only
1211  * reason it returns an 'int' and not 'void' is so that it can be plugged
1212  * directly into file_operations structure.
1213  */
1214 int nonseekable_open(struct inode *inode, struct file *filp)
1215 {
1216         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1217         return 0;
1218 }
1219
1220 EXPORT_SYMBOL(nonseekable_open);
1221
1222 /*
1223  * stream_open is used by subsystems that want stream-like file descriptors.
1224  * Such file descriptors are not seekable and don't have notion of position
1225  * (file.f_pos is always 0). Contrary to file descriptors of other regular
1226  * files, .read() and .write() can run simultaneously.
1227  *
1228  * stream_open never fails and is marked to return int so that it could be
1229  * directly used as file_operations.open .
1230  */
1231 int stream_open(struct inode *inode, struct file *filp)
1232 {
1233         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1234         filp->f_mode |= FMODE_STREAM;
1235         return 0;
1236 }
1237
1238 EXPORT_SYMBOL(stream_open);