fs/open.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/open.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 #include <linux/string.h>
   9 #include <linux/mm.h>
  10 #include <linux/file.h>
  11 #include <linux/fdtable.h>
  12 #include <linux/fsnotify.h>
  13 #include <linux/module.h>
  14 #include <linux/tty.h>
  15 #include <linux/namei.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/capability.h>
  18 #include <linux/securebits.h>
  19 #include <linux/security.h>
  20 #include <linux/mount.h>
  21 #include <linux/fcntl.h>
  22 #include <linux/slab.h>
  23 #include <linux/uaccess.h>
  24 #include <linux/fs.h>
  25 #include <linux/personality.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/syscalls.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/audit.h>
  30 #include <linux/falloc.h>
  31 #include <linux/fs_struct.h>
  32 #include <linux/ima.h>
  33 #include <linux/dnotify.h>
  34 #include <linux/compat.h>
  35 #include <linux/mnt_idmapping.h>
  36
  37 #include "internal.h"
  38
  39 int do_truncate(struct user_namespace *mnt_userns, struct dentry *dentry,
  40                 loff_t length, unsigned int time_attrs, struct file *filp)
  41 {
  42         int ret;
  43         struct iattr newattrs;
  44
  45         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  46         if (length < 0)
  47                 return -EINVAL;
  48
  49         newattrs.ia_size = length;
  50         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  51         if (filp) {
  52                 newattrs.ia_file = filp;
  53                 newattrs.ia_valid |= ATTR_FILE;
  54         }
  55
  56         /* Remove suid, sgid, and file capabilities on truncate too */
  57         ret = dentry_needs_remove_privs(dentry);
  58         if (ret < 0)
  59                 return ret;
  60         if (ret)
  61                 newattrs.ia_valid |= ret | ATTR_FORCE;
  62
  63         inode_lock(dentry->d_inode);
  64         /* Note any delegations or leases have already been broken: */
  65         ret = notify_change(mnt_userns, dentry, &newattrs, NULL);
  66         inode_unlock(dentry->d_inode);
  67         return ret;
  68 }
  69
  70 long vfs_truncate(const struct path *path, loff_t length)
  71 {
  72         struct user_namespace *mnt_userns;
  73         struct inode *inode;
  74         long error;
  75
  76         inode = path->dentry->d_inode;
  77
  78         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  79         if (S_ISDIR(inode->i_mode))
  80                 return -EISDIR;
  81         if (!S_ISREG(inode->i_mode))
  82                 return -EINVAL;
  83
  84         error = mnt_want_write(path->mnt);
  85         if (error)
  86                 goto out;
  87
  88         mnt_userns = mnt_user_ns(path->mnt);
  89         error = inode_permission(mnt_userns, inode, MAY_WRITE);
  90         if (error)
  91                 goto mnt_drop_write_and_out;
  92
  93         error = -EPERM;
  94         if (IS_APPEND(inode))
  95                 goto mnt_drop_write_and_out;
  96
  97         error = get_write_access(inode);
  98         if (error)
  99                 goto mnt_drop_write_and_out;
 100
 101         /*
 102          * Make sure that there are no leases.  get_write_access() protects
 103          * against the truncate racing with a lease-granting setlease().
 104          */
 105         error = break_lease(inode, O_WRONLY);
 106         if (error)
 107                 goto put_write_and_out;
 108
 109         error = security_path_truncate(path);
 110         if (!error)
 111                 error = do_truncate(mnt_userns, path->dentry, length, 0, NULL);
 112
 113 put_write_and_out:
 114         put_write_access(inode);
 115 mnt_drop_write_and_out:
 116         mnt_drop_write(path->mnt);
 117 out:
 118         return error;
 119 }
 120 EXPORT_SYMBOL_GPL(vfs_truncate);
 121
 122 long do_sys_truncate(const char __user *pathname, loff_t length)
 123 {
 124         unsigned int lookup_flags = LOOKUP_FOLLOW;
 125         struct path path;
 126         int error;
 127
 128         if (length < 0) /* sorry, but loff_t says... */
 129                 return -EINVAL;
 130
 131 retry:
 132         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 133         if (!error) {
 134                 error = vfs_truncate(&path, length);
 135                 path_put(&path);
 136         }
 137         if (retry_estale(error, lookup_flags)) {
 138                 lookup_flags |= LOOKUP_REVAL;
 139                 goto retry;
 140         }
 141         return error;
 142 }
 143
 144 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 145 {
 146         return do_sys_truncate(path, length);
 147 }
 148
 149 #ifdef CONFIG_COMPAT
 150 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 151 {
 152         return do_sys_truncate(path, length);
 153 }
 154 #endif
 155
 156 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 157 {
 158         struct inode *inode;
 159         struct dentry *dentry;
 160         struct fd f;
 161         int error;
 162
 163         error = -EINVAL;
 164         if (length < 0)
 165                 goto out;
 166         error = -EBADF;
 167         f = fdget(fd);
 168         if (!f.file)
 169                 goto out;
 170
 171         /* explicitly opened as large or we are on 64-bit box */
 172         if (f.file->f_flags & O_LARGEFILE)
 173                 small = 0;
 174
 175         dentry = f.file->f_path.dentry;
 176         inode = dentry->d_inode;
 177         error = -EINVAL;
 178         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 179                 goto out_putf;
 180
 181         error = -EINVAL;
 182         /* Cannot ftruncate over 2^31 bytes without large file support */
 183         if (small && length > MAX_NON_LFS)
 184                 goto out_putf;
 185
 186         error = -EPERM;
 187         /* Check IS_APPEND on real upper inode */
 188         if (IS_APPEND(file_inode(f.file)))
 189                 goto out_putf;
 190         sb_start_write(inode->i_sb);
 191         error = security_path_truncate(&f.file->f_path);
 192         if (!error)
 193                 error = do_truncate(file_mnt_user_ns(f.file), dentry, length,
 194                                     ATTR_MTIME | ATTR_CTIME, f.file);
 195         sb_end_write(inode->i_sb);
 196 out_putf:
 197         fdput(f);
 198 out:
 199         return error;
 200 }
 201
 202 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 203 {
 204         return do_sys_ftruncate(fd, length, 1);
 205 }
 206
 207 #ifdef CONFIG_COMPAT
 208 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 209 {
 210         return do_sys_ftruncate(fd, length, 1);
 211 }
 212 #endif
 213
 214 /* LFS versions of truncate are only needed on 32 bit machines */
 215 #if BITS_PER_LONG == 32
 216 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 217 {
 218         return do_sys_truncate(path, length);
 219 }
 220
 221 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 222 {
 223         return do_sys_ftruncate(fd, length, 0);
 224 }
 225 #endif /* BITS_PER_LONG == 32 */
 226
 227 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
 228 COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
 229                        compat_arg_u64_dual(length))
 230 {
 231         return ksys_truncate(pathname, compat_arg_u64_glue(length));
 232 }
 233 #endif
 234
 235 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
 236 COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
 237                        compat_arg_u64_dual(length))
 238 {
 239         return ksys_ftruncate(fd, compat_arg_u64_glue(length));
 240 }
 241 #endif
 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 int ksys_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 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 353 {
 354         return ksys_fallocate(fd, mode, offset, len);
 355 }
 356
 357 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
 358 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
 359                        compat_arg_u64_dual(len))
 360 {
 361         return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
 362                               compat_arg_u64_glue(len));
 363 }
 364 #endif
 365
 366 /*
 367  * access() needs to use the real uid/gid, not the effective uid/gid.
 368  * We do this by temporarily clearing all FS-related capabilities and
 369  * switching the fsuid/fsgid around to the real ones.
 370  */
 371 static const struct cred *access_override_creds(void)
 372 {
 373         const struct cred *old_cred;
 374         struct cred *override_cred;
 375
 376         override_cred = prepare_creds();
 377         if (!override_cred)
 378                 return NULL;
 379
 380         override_cred->fsuid = override_cred->uid;
 381         override_cred->fsgid = override_cred->gid;
 382
 383         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 384                 /* Clear the capabilities if we switch to a non-root user */
 385                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 386                 if (!uid_eq(override_cred->uid, root_uid))
 387                         cap_clear(override_cred->cap_effective);
 388                 else
 389                         override_cred->cap_effective =
 390                                 override_cred->cap_permitted;
 391         }
 392
 393         /*
 394          * The new set of credentials can *only* be used in
 395          * task-synchronous circumstances, and does not need
 396          * RCU freeing, unless somebody then takes a separate
 397          * reference to it.
 398          *
 399          * NOTE! This is _only_ true because this credential
 400          * is used purely for override_creds() that installs
 401          * it as the subjective cred. Other threads will be
 402          * accessing ->real_cred, not the subjective cred.
 403          *
 404          * If somebody _does_ make a copy of this (using the
 405          * 'get_current_cred()' function), that will clear the
 406          * non_rcu field, because now that other user may be
 407          * expecting RCU freeing. But normal thread-synchronous
 408          * cred accesses will keep things non-RCY.
 409          */
 410         override_cred->non_rcu = 1;
 411
 412         old_cred = override_creds(override_cred);
 413
 414         /* override_cred() gets its own ref */
 415         put_cred(override_cred);
 416
 417         return old_cred;
 418 }
 419
 420 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
 421 {
 422         struct path path;
 423         struct inode *inode;
 424         int res;
 425         unsigned int lookup_flags = LOOKUP_FOLLOW;
 426         const struct cred *old_cred = NULL;
 427
 428         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 429                 return -EINVAL;
 430
 431         if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
 432                 return -EINVAL;
 433
 434         if (flags & AT_SYMLINK_NOFOLLOW)
 435                 lookup_flags &= ~LOOKUP_FOLLOW;
 436         if (flags & AT_EMPTY_PATH)
 437                 lookup_flags |= LOOKUP_EMPTY;
 438
 439         if (!(flags & AT_EACCESS)) {
 440                 old_cred = access_override_creds();
 441                 if (!old_cred)
 442                         return -ENOMEM;
 443         }
 444
 445 retry:
 446         res = user_path_at(dfd, filename, lookup_flags, &path);
 447         if (res)
 448                 goto out;
 449
 450         inode = d_backing_inode(path.dentry);
 451
 452         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 453                 /*
 454                  * MAY_EXEC on regular files is denied if the fs is mounted
 455                  * with the "noexec" flag.
 456                  */
 457                 res = -EACCES;
 458                 if (path_noexec(&path))
 459                         goto out_path_release;
 460         }
 461
 462         res = inode_permission(mnt_user_ns(path.mnt), inode, mode | MAY_ACCESS);
 463         /* SuS v2 requires we report a read only fs too */
 464         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 465                 goto out_path_release;
 466         /*
 467          * This is a rare case where using __mnt_is_readonly()
 468          * is OK without a mnt_want/drop_write() pair.  Since
 469          * no actual write to the fs is performed here, we do
 470          * not need to telegraph to that to anyone.
 471          *
 472          * By doing this, we accept that this access is
 473          * inherently racy and know that the fs may change
 474          * state before we even see this result.
 475          */
 476         if (__mnt_is_readonly(path.mnt))
 477                 res = -EROFS;
 478
 479 out_path_release:
 480         path_put(&path);
 481         if (retry_estale(res, lookup_flags)) {
 482                 lookup_flags |= LOOKUP_REVAL;
 483                 goto retry;
 484         }
 485 out:
 486         if (old_cred)
 487                 revert_creds(old_cred);
 488
 489         return res;
 490 }
 491
 492 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 493 {
 494         return do_faccessat(dfd, filename, mode, 0);
 495 }
 496
 497 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
 498                 int, flags)
 499 {
 500         return do_faccessat(dfd, filename, mode, flags);
 501 }
 502
 503 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 504 {
 505         return do_faccessat(AT_FDCWD, filename, mode, 0);
 506 }
 507
 508 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 509 {
 510         struct path path;
 511         int error;
 512         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 513 retry:
 514         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 515         if (error)
 516                 goto out;
 517
 518         error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
 519         if (error)
 520                 goto dput_and_out;
 521
 522         set_fs_pwd(current->fs, &path);
 523
 524 dput_and_out:
 525         path_put(&path);
 526         if (retry_estale(error, lookup_flags)) {
 527                 lookup_flags |= LOOKUP_REVAL;
 528                 goto retry;
 529         }
 530 out:
 531         return error;
 532 }
 533
 534 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 535 {
 536         struct fd f = fdget_raw(fd);
 537         int error;
 538
 539         error = -EBADF;
 540         if (!f.file)
 541                 goto out;
 542
 543         error = -ENOTDIR;
 544         if (!d_can_lookup(f.file->f_path.dentry))
 545                 goto out_putf;
 546
 547         error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
 548         if (!error)
 549                 set_fs_pwd(current->fs, &f.file->f_path);
 550 out_putf:
 551         fdput(f);
 552 out:
 553         return error;
 554 }
 555
 556 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 557 {
 558         struct path path;
 559         int error;
 560         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 561 retry:
 562         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 563         if (error)
 564                 goto out;
 565
 566         error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
 567         if (error)
 568                 goto dput_and_out;
 569
 570         error = -EPERM;
 571         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 572                 goto dput_and_out;
 573         error = security_path_chroot(&path);
 574         if (error)
 575                 goto dput_and_out;
 576
 577         set_fs_root(current->fs, &path);
 578         error = 0;
 579 dput_and_out:
 580         path_put(&path);
 581         if (retry_estale(error, lookup_flags)) {
 582                 lookup_flags |= LOOKUP_REVAL;
 583                 goto retry;
 584         }
 585 out:
 586         return error;
 587 }
 588
 589 int chmod_common(const struct path *path, umode_t mode)
 590 {
 591         struct inode *inode = path->dentry->d_inode;
 592         struct inode *delegated_inode = NULL;
 593         struct iattr newattrs;
 594         int error;
 595
 596         error = mnt_want_write(path->mnt);
 597         if (error)
 598                 return error;
 599 retry_deleg:
 600         inode_lock(inode);
 601         error = security_path_chmod(path, mode);
 602         if (error)
 603                 goto out_unlock;
 604         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 605         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 606         error = notify_change(mnt_user_ns(path->mnt), path->dentry,
 607                               &newattrs, &delegated_inode);
 608 out_unlock:
 609         inode_unlock(inode);
 610         if (delegated_inode) {
 611                 error = break_deleg_wait(&delegated_inode);
 612                 if (!error)
 613                         goto retry_deleg;
 614         }
 615         mnt_drop_write(path->mnt);
 616         return error;
 617 }
 618
 619 int vfs_fchmod(struct file *file, umode_t mode)
 620 {
 621         audit_file(file);
 622         return chmod_common(&file->f_path, mode);
 623 }
 624
 625 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 626 {
 627         struct fd f = fdget(fd);
 628         int err = -EBADF;
 629
 630         if (f.file) {
 631                 err = vfs_fchmod(f.file, mode);
 632                 fdput(f);
 633         }
 634         return err;
 635 }
 636
 637 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
 638 {
 639         struct path path;
 640         int error;
 641         unsigned int lookup_flags = LOOKUP_FOLLOW;
 642 retry:
 643         error = user_path_at(dfd, filename, lookup_flags, &path);
 644         if (!error) {
 645                 error = chmod_common(&path, mode);
 646                 path_put(&path);
 647                 if (retry_estale(error, lookup_flags)) {
 648                         lookup_flags |= LOOKUP_REVAL;
 649                         goto retry;
 650                 }
 651         }
 652         return error;
 653 }
 654
 655 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 656                 umode_t, mode)
 657 {
 658         return do_fchmodat(dfd, filename, mode);
 659 }
 660
 661 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 662 {
 663         return do_fchmodat(AT_FDCWD, filename, mode);
 664 }
 665
 666 /**
 667  * setattr_vfsuid - check and set ia_fsuid attribute
 668  * @kuid: new inode owner
 669  *
 670  * Check whether @kuid is valid and if so generate and set vfsuid_t in
 671  * ia_vfsuid.
 672  *
 673  * Return: true if @kuid is valid, false if not.
 674  */
 675 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
 676 {
 677         if (!uid_valid(kuid))
 678                 return false;
 679         attr->ia_valid |= ATTR_UID;
 680         attr->ia_vfsuid = VFSUIDT_INIT(kuid);
 681         return true;
 682 }
 683
 684 /**
 685  * setattr_vfsgid - check and set ia_fsgid attribute
 686  * @kgid: new inode owner
 687  *
 688  * Check whether @kgid is valid and if so generate and set vfsgid_t in
 689  * ia_vfsgid.
 690  *
 691  * Return: true if @kgid is valid, false if not.
 692  */
 693 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
 694 {
 695         if (!gid_valid(kgid))
 696                 return false;
 697         attr->ia_valid |= ATTR_GID;
 698         attr->ia_vfsgid = VFSGIDT_INIT(kgid);
 699         return true;
 700 }
 701
 702 int chown_common(const struct path *path, uid_t user, gid_t group)
 703 {
 704         struct user_namespace *mnt_userns, *fs_userns;
 705         struct inode *inode = path->dentry->d_inode;
 706         struct inode *delegated_inode = NULL;
 707         int error;
 708         struct iattr newattrs;
 709         kuid_t uid;
 710         kgid_t gid;
 711
 712         uid = make_kuid(current_user_ns(), user);
 713         gid = make_kgid(current_user_ns(), group);
 714
 715         mnt_userns = mnt_user_ns(path->mnt);
 716         fs_userns = i_user_ns(inode);
 717
 718 retry_deleg:
 719         newattrs.ia_vfsuid = INVALID_VFSUID;
 720         newattrs.ia_vfsgid = INVALID_VFSGID;
 721         newattrs.ia_valid =  ATTR_CTIME;
 722         if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
 723                 return -EINVAL;
 724         if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
 725                 return -EINVAL;
 726         if (!S_ISDIR(inode->i_mode))
 727                 newattrs.ia_valid |=
 728                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 729         inode_lock(inode);
 730         /* Continue to send actual fs values, not the mount values. */
 731         error = security_path_chown(
 732                 path,
 733                 from_vfsuid(mnt_userns, fs_userns, newattrs.ia_vfsuid),
 734                 from_vfsgid(mnt_userns, fs_userns, newattrs.ia_vfsgid));
 735         if (!error)
 736                 error = notify_change(mnt_userns, path->dentry, &newattrs,
 737                                       &delegated_inode);
 738         inode_unlock(inode);
 739         if (delegated_inode) {
 740                 error = break_deleg_wait(&delegated_inode);
 741                 if (!error)
 742                         goto retry_deleg;
 743         }
 744         return error;
 745 }
 746
 747 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 748                 int flag)
 749 {
 750         struct path path;
 751         int error = -EINVAL;
 752         int lookup_flags;
 753
 754         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 755                 goto out;
 756
 757         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 758         if (flag & AT_EMPTY_PATH)
 759                 lookup_flags |= LOOKUP_EMPTY;
 760 retry:
 761         error = user_path_at(dfd, filename, lookup_flags, &path);
 762         if (error)
 763                 goto out;
 764         error = mnt_want_write(path.mnt);
 765         if (error)
 766                 goto out_release;
 767         error = chown_common(&path, user, group);
 768         mnt_drop_write(path.mnt);
 769 out_release:
 770         path_put(&path);
 771         if (retry_estale(error, lookup_flags)) {
 772                 lookup_flags |= LOOKUP_REVAL;
 773                 goto retry;
 774         }
 775 out:
 776         return error;
 777 }
 778
 779 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 780                 gid_t, group, int, flag)
 781 {
 782         return do_fchownat(dfd, filename, user, group, flag);
 783 }
 784
 785 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 786 {
 787         return do_fchownat(AT_FDCWD, filename, user, group, 0);
 788 }
 789
 790 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 791 {
 792         return do_fchownat(AT_FDCWD, filename, user, group,
 793                            AT_SYMLINK_NOFOLLOW);
 794 }
 795
 796 int vfs_fchown(struct file *file, uid_t user, gid_t group)
 797 {
 798         int error;
 799
 800         error = mnt_want_write_file(file);
 801         if (error)
 802                 return error;
 803         audit_file(file);
 804         error = chown_common(&file->f_path, user, group);
 805         mnt_drop_write_file(file);
 806         return error;
 807 }
 808
 809 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 810 {
 811         struct fd f = fdget(fd);
 812         int error = -EBADF;
 813
 814         if (f.file) {
 815                 error = vfs_fchown(f.file, user, group);
 816                 fdput(f);
 817         }
 818         return error;
 819 }
 820
 821 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 822 {
 823         return ksys_fchown(fd, user, group);
 824 }
 825
 826 static int do_dentry_open(struct file *f,
 827                           struct inode *inode,
 828                           int (*open)(struct inode *, struct file *))
 829 {
 830         static const struct file_operations empty_fops = {};
 831         int error;
 832
 833         path_get(&f->f_path);
 834         f->f_inode = inode;
 835         f->f_mapping = inode->i_mapping;
 836         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 837         f->f_sb_err = file_sample_sb_err(f);
 838
 839         if (unlikely(f->f_flags & O_PATH)) {
 840                 f->f_mode = FMODE_PATH | FMODE_OPENED;
 841                 f->f_op = &empty_fops;
 842                 return 0;
 843         }
 844
 845         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 846                 error = get_write_access(inode);
 847                 if (unlikely(error))
 848                         goto cleanup_file;
 849                 error = __mnt_want_write(f->f_path.mnt);
 850                 if (unlikely(error)) {
 851                         put_write_access(inode);
 852                         goto cleanup_file;
 853                 }
 854                 f->f_mode |= FMODE_WRITER;
 855         }
 856
 857         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 858         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 859                 f->f_mode |= FMODE_ATOMIC_POS;
 860
 861         f->f_op = fops_get(inode->i_fop);
 862         if (WARN_ON(!f->f_op)) {
 863                 error = -ENODEV;
 864                 goto cleanup_all;
 865         }
 866
 867         error = security_file_open(f);
 868         if (error)
 869                 goto cleanup_all;
 870
 871         error = break_lease(locks_inode(f), f->f_flags);
 872         if (error)
 873                 goto cleanup_all;
 874
 875         /* normally all 3 are set; ->open() can clear them if needed */
 876         f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 877         if (!open)
 878                 open = f->f_op->open;
 879         if (open) {
 880                 error = open(inode, f);
 881                 if (error)
 882                         goto cleanup_all;
 883         }
 884         f->f_mode |= FMODE_OPENED;
 885         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 886                 i_readcount_inc(inode);
 887         if ((f->f_mode & FMODE_READ) &&
 888              likely(f->f_op->read || f->f_op->read_iter))
 889                 f->f_mode |= FMODE_CAN_READ;
 890         if ((f->f_mode & FMODE_WRITE) &&
 891              likely(f->f_op->write || f->f_op->write_iter))
 892                 f->f_mode |= FMODE_CAN_WRITE;
 893         if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
 894                 f->f_mode &= ~FMODE_LSEEK;
 895         if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
 896                 f->f_mode |= FMODE_CAN_ODIRECT;
 897
 898         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 899         f->f_iocb_flags = iocb_flags(f);
 900
 901         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 902
 903         if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
 904                 return -EINVAL;
 905
 906         /*
 907          * XXX: Huge page cache doesn't support writing yet. Drop all page
 908          * cache for this file before processing writes.
 909          */
 910         if (f->f_mode & FMODE_WRITE) {
 911                 /*
 912                  * Paired with smp_mb() in collapse_file() to ensure nr_thps
 913                  * is up to date and the update to i_writecount by
 914                  * get_write_access() is visible. Ensures subsequent insertion
 915                  * of THPs into the page cache will fail.
 916                  */
 917                 smp_mb();
 918                 if (filemap_nr_thps(inode->i_mapping)) {
 919                         struct address_space *mapping = inode->i_mapping;
 920
 921                         filemap_invalidate_lock(inode->i_mapping);
 922                         /*
 923                          * unmap_mapping_range just need to be called once
 924                          * here, because the private pages is not need to be
 925                          * unmapped mapping (e.g. data segment of dynamic
 926                          * shared libraries here).
 927                          */
 928                         unmap_mapping_range(mapping, 0, 0, 0);
 929                         truncate_inode_pages(mapping, 0);
 930                         filemap_invalidate_unlock(inode->i_mapping);
 931                 }
 932         }
 933
 934         return 0;
 935
 936 cleanup_all:
 937         if (WARN_ON_ONCE(error > 0))
 938                 error = -EINVAL;
 939         fops_put(f->f_op);
 940         if (f->f_mode & FMODE_WRITER) {
 941                 put_write_access(inode);
 942                 __mnt_drop_write(f->f_path.mnt);
 943         }
 944 cleanup_file:
 945         path_put(&f->f_path);
 946         f->f_path.mnt = NULL;
 947         f->f_path.dentry = NULL;
 948         f->f_inode = NULL;
 949         return error;
 950 }
 951
 952 /**
 953  * finish_open - finish opening a file
 954  * @file: file pointer
 955  * @dentry: pointer to dentry
 956  * @open: open callback
 957  * @opened: state of open
 958  *
 959  * This can be used to finish opening a file passed to i_op->atomic_open().
 960  *
 961  * If the open callback is set to NULL, then the standard f_op->open()
 962  * filesystem callback is substituted.
 963  *
 964  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 965  * the return value of d_splice_alias(), then the caller needs to perform dput()
 966  * on it after finish_open().
 967  *
 968  * Returns zero on success or -errno if the open failed.
 969  */
 970 int finish_open(struct file *file, struct dentry *dentry,
 971                 int (*open)(struct inode *, struct file *))
 972 {
 973         BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
 974
 975         file->f_path.dentry = dentry;
 976         return do_dentry_open(file, d_backing_inode(dentry), open);
 977 }
 978 EXPORT_SYMBOL(finish_open);
 979
 980 /**
 981  * finish_no_open - finish ->atomic_open() without opening the file
 982  *
 983  * @file: file pointer
 984  * @dentry: dentry or NULL (as returned from ->lookup())
 985  *
 986  * This can be used to set the result of a successful lookup in ->atomic_open().
 987  *
 988  * NB: unlike finish_open() this function does consume the dentry reference and
 989  * the caller need not dput() it.
 990  *
 991  * Returns "0" which must be the return value of ->atomic_open() after having
 992  * called this function.
 993  */
 994 int finish_no_open(struct file *file, struct dentry *dentry)
 995 {
 996         file->f_path.dentry = dentry;
 997         return 0;
 998 }
 999 EXPORT_SYMBOL(finish_no_open);
1000
1001 char *file_path(struct file *filp, char *buf, int buflen)
1002 {
1003         return d_path(&filp->f_path, buf, buflen);
1004 }
1005 EXPORT_SYMBOL(file_path);
1006
1007 /**
1008  * vfs_open - open the file at the given path
1009  * @path: path to open
1010  * @file: newly allocated file with f_flag initialized
1011  * @cred: credentials to use
1012  */
1013 int vfs_open(const struct path *path, struct file *file)
1014 {
1015         file->f_path = *path;
1016         return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1017 }
1018
1019 struct file *dentry_open(const struct path *path, int flags,
1020                          const struct cred *cred)
1021 {
1022         int error;
1023         struct file *f;
1024
1025         validate_creds(cred);
1026
1027         /* We must always pass in a valid mount pointer. */
1028         BUG_ON(!path->mnt);
1029
1030         f = alloc_empty_file(flags, cred);
1031         if (!IS_ERR(f)) {
1032                 error = vfs_open(path, f);
1033                 if (error) {
1034                         fput(f);
1035                         f = ERR_PTR(error);
1036                 }
1037         }
1038         return f;
1039 }
1040 EXPORT_SYMBOL(dentry_open);
1041
1042 /**
1043  * dentry_create - Create and open a file
1044  * @path: path to create
1045  * @flags: O_ flags
1046  * @mode: mode bits for new file
1047  * @cred: credentials to use
1048  *
1049  * Caller must hold the parent directory's lock, and have prepared
1050  * a negative dentry, placed in @path->dentry, for the new file.
1051  *
1052  * Caller sets @path->mnt to the vfsmount of the filesystem where
1053  * the new file is to be created. The parent directory and the
1054  * negative dentry must reside on the same filesystem instance.
1055  *
1056  * On success, returns a "struct file *". Otherwise a ERR_PTR
1057  * is returned.
1058  */
1059 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1060                            const struct cred *cred)
1061 {
1062         struct file *f;
1063         int error;
1064
1065         validate_creds(cred);
1066         f = alloc_empty_file(flags, cred);
1067         if (IS_ERR(f))
1068                 return f;
1069
1070         error = vfs_create(mnt_user_ns(path->mnt),
1071                            d_inode(path->dentry->d_parent),
1072                            path->dentry, mode, true);
1073         if (!error)
1074                 error = vfs_open(path, f);
1075
1076         if (unlikely(error)) {
1077                 fput(f);
1078                 return ERR_PTR(error);
1079         }
1080         return f;
1081 }
1082 EXPORT_SYMBOL(dentry_create);
1083
1084 struct file *open_with_fake_path(const struct path *path, int flags,
1085                                 struct inode *inode, const struct cred *cred)
1086 {
1087         struct file *f = alloc_empty_file_noaccount(flags, cred);
1088         if (!IS_ERR(f)) {
1089                 int error;
1090
1091                 f->f_path = *path;
1092                 error = do_dentry_open(f, inode, NULL);
1093                 if (error) {
1094                         fput(f);
1095                         f = ERR_PTR(error);
1096                 }
1097         }
1098         return f;
1099 }
1100 EXPORT_SYMBOL(open_with_fake_path);
1101
1102 #define WILL_CREATE(flags)      (flags & (O_CREAT | __O_TMPFILE))
1103 #define O_PATH_FLAGS            (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1104
1105 inline struct open_how build_open_how(int flags, umode_t mode)
1106 {
1107         struct open_how how = {
1108                 .flags = flags & VALID_OPEN_FLAGS,
1109                 .mode = mode & S_IALLUGO,
1110         };
1111
1112         /* O_PATH beats everything else. */
1113         if (how.flags & O_PATH)
1114                 how.flags &= O_PATH_FLAGS;
1115         /* Modes should only be set for create-like flags. */
1116         if (!WILL_CREATE(how.flags))
1117                 how.mode = 0;
1118         return how;
1119 }
1120
1121 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1122 {
1123         u64 flags = how->flags;
1124         u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1125         int lookup_flags = 0;
1126         int acc_mode = ACC_MODE(flags);
1127
1128         BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1129                          "struct open_flags doesn't yet handle flags > 32 bits");
1130
1131         /*
1132          * Strip flags that either shouldn't be set by userspace like
1133          * FMODE_NONOTIFY or that aren't relevant in determining struct
1134          * open_flags like O_CLOEXEC.
1135          */
1136         flags &= ~strip;
1137
1138         /*
1139          * Older syscalls implicitly clear all of the invalid flags or argument
1140          * values before calling build_open_flags(), but openat2(2) checks all
1141          * of its arguments.
1142          */
1143         if (flags & ~VALID_OPEN_FLAGS)
1144                 return -EINVAL;
1145         if (how->resolve & ~VALID_RESOLVE_FLAGS)
1146                 return -EINVAL;
1147
1148         /* Scoping flags are mutually exclusive. */
1149         if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1150                 return -EINVAL;
1151
1152         /* Deal with the mode. */
1153         if (WILL_CREATE(flags)) {
1154                 if (how->mode & ~S_IALLUGO)
1155                         return -EINVAL;
1156                 op->mode = how->mode | S_IFREG;
1157         } else {
1158                 if (how->mode != 0)
1159                         return -EINVAL;
1160                 op->mode = 0;
1161         }
1162
1163         /*
1164          * In order to ensure programs get explicit errors when trying to use
1165          * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1166          * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1167          * have to require userspace to explicitly set it.
1168          */
1169         if (flags & __O_TMPFILE) {
1170                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1171                         return -EINVAL;
1172                 if (!(acc_mode & MAY_WRITE))
1173                         return -EINVAL;
1174         }
1175         if (flags & O_PATH) {
1176                 /* O_PATH only permits certain other flags to be set. */
1177                 if (flags & ~O_PATH_FLAGS)
1178                         return -EINVAL;
1179                 acc_mode = 0;
1180         }
1181
1182         /*
1183          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1184          * check for O_DSYNC if the need any syncing at all we enforce it's
1185          * always set instead of having to deal with possibly weird behaviour
1186          * for malicious applications setting only __O_SYNC.
1187          */
1188         if (flags & __O_SYNC)
1189                 flags |= O_DSYNC;
1190
1191         op->open_flag = flags;
1192
1193         /* O_TRUNC implies we need access checks for write permissions */
1194         if (flags & O_TRUNC)
1195                 acc_mode |= MAY_WRITE;
1196
1197         /* Allow the LSM permission hook to distinguish append
1198            access from general write access. */
1199         if (flags & O_APPEND)
1200                 acc_mode |= MAY_APPEND;
1201
1202         op->acc_mode = acc_mode;
1203
1204         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1205
1206         if (flags & O_CREAT) {
1207                 op->intent |= LOOKUP_CREATE;
1208                 if (flags & O_EXCL) {
1209                         op->intent |= LOOKUP_EXCL;
1210                         flags |= O_NOFOLLOW;
1211                 }
1212         }
1213
1214         if (flags & O_DIRECTORY)
1215                 lookup_flags |= LOOKUP_DIRECTORY;
1216         if (!(flags & O_NOFOLLOW))
1217                 lookup_flags |= LOOKUP_FOLLOW;
1218
1219         if (how->resolve & RESOLVE_NO_XDEV)
1220                 lookup_flags |= LOOKUP_NO_XDEV;
1221         if (how->resolve & RESOLVE_NO_MAGICLINKS)
1222                 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1223         if (how->resolve & RESOLVE_NO_SYMLINKS)
1224                 lookup_flags |= LOOKUP_NO_SYMLINKS;
1225         if (how->resolve & RESOLVE_BENEATH)
1226                 lookup_flags |= LOOKUP_BENEATH;
1227         if (how->resolve & RESOLVE_IN_ROOT)
1228                 lookup_flags |= LOOKUP_IN_ROOT;
1229         if (how->resolve & RESOLVE_CACHED) {
1230                 /* Don't bother even trying for create/truncate/tmpfile open */
1231                 if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1232                         return -EAGAIN;
1233                 lookup_flags |= LOOKUP_CACHED;
1234         }
1235
1236         op->lookup_flags = lookup_flags;
1237         return 0;
1238 }
1239
1240 /**
1241  * file_open_name - open file and return file pointer
1242  *
1243  * @name:       struct filename containing path to open
1244  * @flags:      open flags as per the open(2) second argument
1245  * @mode:       mode for the new file if O_CREAT is set, else ignored
1246  *
1247  * This is the helper to open a file from kernelspace if you really
1248  * have to.  But in generally you should not do this, so please move
1249  * along, nothing to see here..
1250  */
1251 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1252 {
1253         struct open_flags op;
1254         struct open_how how = build_open_how(flags, mode);
1255         int err = build_open_flags(&how, &op);
1256         if (err)
1257                 return ERR_PTR(err);
1258         return do_filp_open(AT_FDCWD, name, &op);
1259 }
1260
1261 /**
1262  * filp_open - open file and return file pointer
1263  *
1264  * @filename:   path to open
1265  * @flags:      open flags as per the open(2) second argument
1266  * @mode:       mode for the new file if O_CREAT is set, else ignored
1267  *
1268  * This is the helper to open a file from kernelspace if you really
1269  * have to.  But in generally you should not do this, so please move
1270  * along, nothing to see here..
1271  */
1272 struct file *filp_open(const char *filename, int flags, umode_t mode)
1273 {
1274         struct filename *name = getname_kernel(filename);
1275         struct file *file = ERR_CAST(name);
1276
1277         if (!IS_ERR(name)) {
1278                 file = file_open_name(name, flags, mode);
1279                 putname(name);
1280         }
1281         return file;
1282 }
1283 EXPORT_SYMBOL(filp_open);
1284
1285 struct file *file_open_root(const struct path *root,
1286                             const char *filename, int flags, umode_t mode)
1287 {
1288         struct open_flags op;
1289         struct open_how how = build_open_how(flags, mode);
1290         int err = build_open_flags(&how, &op);
1291         if (err)
1292                 return ERR_PTR(err);
1293         return do_file_open_root(root, filename, &op);
1294 }
1295 EXPORT_SYMBOL(file_open_root);
1296
1297 static long do_sys_openat2(int dfd, const char __user *filename,
1298                            struct open_how *how)
1299 {
1300         struct open_flags op;
1301         int fd = build_open_flags(how, &op);
1302         struct filename *tmp;
1303
1304         if (fd)
1305                 return fd;
1306
1307         tmp = getname(filename);
1308         if (IS_ERR(tmp))
1309                 return PTR_ERR(tmp);
1310
1311         fd = get_unused_fd_flags(how->flags);
1312         if (fd >= 0) {
1313                 struct file *f = do_filp_open(dfd, tmp, &op);
1314                 if (IS_ERR(f)) {
1315                         put_unused_fd(fd);
1316                         fd = PTR_ERR(f);
1317                 } else {
1318                         fsnotify_open(f);
1319                         fd_install(fd, f);
1320                 }
1321         }
1322         putname(tmp);
1323         return fd;
1324 }
1325
1326 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1327 {
1328         struct open_how how = build_open_how(flags, mode);
1329         return do_sys_openat2(dfd, filename, &how);
1330 }
1331
1332
1333 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1334 {
1335         if (force_o_largefile())
1336                 flags |= O_LARGEFILE;
1337         return do_sys_open(AT_FDCWD, filename, flags, mode);
1338 }
1339
1340 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1341                 umode_t, mode)
1342 {
1343         if (force_o_largefile())
1344                 flags |= O_LARGEFILE;
1345         return do_sys_open(dfd, filename, flags, mode);
1346 }
1347
1348 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1349                 struct open_how __user *, how, size_t, usize)
1350 {
1351         int err;
1352         struct open_how tmp;
1353
1354         BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1355         BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1356
1357         if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1358                 return -EINVAL;
1359
1360         err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1361         if (err)
1362                 return err;
1363
1364         audit_openat2_how(&tmp);
1365
1366         /* O_LARGEFILE is only allowed for non-O_PATH. */
1367         if (!(tmp.flags & O_PATH) && force_o_largefile())
1368                 tmp.flags |= O_LARGEFILE;
1369
1370         return do_sys_openat2(dfd, filename, &tmp);
1371 }
1372
1373 #ifdef CONFIG_COMPAT
1374 /*
1375  * Exactly like sys_open(), except that it doesn't set the
1376  * O_LARGEFILE flag.
1377  */
1378 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1379 {
1380         return do_sys_open(AT_FDCWD, filename, flags, mode);
1381 }
1382
1383 /*
1384  * Exactly like sys_openat(), except that it doesn't set the
1385  * O_LARGEFILE flag.
1386  */
1387 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1388 {
1389         return do_sys_open(dfd, filename, flags, mode);
1390 }
1391 #endif
1392
1393 #ifndef __alpha__
1394
1395 /*
1396  * For backward compatibility?  Maybe this should be moved
1397  * into arch/i386 instead?
1398  */
1399 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1400 {
1401         int flags = O_CREAT | O_WRONLY | O_TRUNC;
1402
1403         if (force_o_largefile())
1404                 flags |= O_LARGEFILE;
1405         return do_sys_open(AT_FDCWD, pathname, flags, mode);
1406 }
1407 #endif
1408
1409 /*
1410  * "id" is the POSIX thread ID. We use the
1411  * files pointer for this..
1412  */
1413 int filp_close(struct file *filp, fl_owner_t id)
1414 {
1415         int retval = 0;
1416
1417         if (!file_count(filp)) {
1418                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1419                 return 0;
1420         }
1421
1422         if (filp->f_op->flush)
1423                 retval = filp->f_op->flush(filp, id);
1424
1425         if (likely(!(filp->f_mode & FMODE_PATH))) {
1426                 dnotify_flush(filp, id);
1427                 locks_remove_posix(filp, id);
1428         }
1429         fput(filp);
1430         return retval;
1431 }
1432
1433 EXPORT_SYMBOL(filp_close);
1434
1435 /*
1436  * Careful here! We test whether the file pointer is NULL before
1437  * releasing the fd. This ensures that one clone task can't release
1438  * an fd while another clone is opening it.
1439  */
1440 SYSCALL_DEFINE1(close, unsigned int, fd)
1441 {
1442         int retval = close_fd(fd);
1443
1444         /* can't restart close syscall because file table entry was cleared */
1445         if (unlikely(retval == -ERESTARTSYS ||
1446                      retval == -ERESTARTNOINTR ||
1447                      retval == -ERESTARTNOHAND ||
1448                      retval == -ERESTART_RESTARTBLOCK))
1449                 retval = -EINTR;
1450
1451         return retval;
1452 }
1453
1454 /**
1455  * close_range() - Close all file descriptors in a given range.
1456  *
1457  * @fd:     starting file descriptor to close
1458  * @max_fd: last file descriptor to close
1459  * @flags:  reserved for future extensions
1460  *
1461  * This closes a range of file descriptors. All file descriptors
1462  * from @fd up to and including @max_fd are closed.
1463  * Currently, errors to close a given file descriptor are ignored.
1464  */
1465 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1466                 unsigned int, flags)
1467 {
1468         return __close_range(fd, max_fd, flags);
1469 }
1470
1471 /*
1472  * This routine simulates a hangup on the tty, to arrange that users
1473  * are given clean terminals at login time.
1474  */
1475 SYSCALL_DEFINE0(vhangup)
1476 {
1477         if (capable(CAP_SYS_TTY_CONFIG)) {
1478                 tty_vhangup_self();
1479                 return 0;
1480         }
1481         return -EPERM;
1482 }
1483
1484 /*
1485  * Called when an inode is about to be open.
1486  * We use this to disallow opening large files on 32bit systems if
1487  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1488  * on this flag in sys_open.
1489  */
1490 int generic_file_open(struct inode * inode, struct file * filp)
1491 {
1492         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1493                 return -EOVERFLOW;
1494         return 0;
1495 }
1496
1497 EXPORT_SYMBOL(generic_file_open);
1498
1499 /*
1500  * This is used by subsystems that don't want seekable
1501  * file descriptors. The function is not supposed to ever fail, the only
1502  * reason it returns an 'int' and not 'void' is so that it can be plugged
1503  * directly into file_operations structure.
1504  */
1505 int nonseekable_open(struct inode *inode, struct file *filp)
1506 {
1507         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1508         return 0;
1509 }
1510
1511 EXPORT_SYMBOL(nonseekable_open);
1512
1513 /*
1514  * stream_open is used by subsystems that want stream-like file descriptors.
1515  * Such file descriptors are not seekable and don't have notion of position
1516  * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1517  * Contrary to file descriptors of other regular files, .read() and .write()
1518  * can run simultaneously.
1519  *
1520  * stream_open never fails and is marked to return int so that it could be
1521  * directly used as file_operations.open .
1522  */
1523 int stream_open(struct inode *inode, struct file *filp)
1524 {
1525         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1526         filp->f_mode |= FMODE_STREAM;
1527         return 0;
1528 }
1529
1530 EXPORT_SYMBOL(stream_open);