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_valid =  ATTR_CTIME;
 720         if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
 721                 return -EINVAL;
 722         if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
 723                 return -EINVAL;
 724         if (!S_ISDIR(inode->i_mode))
 725                 newattrs.ia_valid |=
 726                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 727         inode_lock(inode);
 728         /* Continue to send actual fs values, not the mount values. */
 729         error = security_path_chown(
 730                 path,
 731                 from_vfsuid(mnt_userns, fs_userns, newattrs.ia_vfsuid),
 732                 from_vfsgid(mnt_userns, fs_userns, newattrs.ia_vfsgid));
 733         if (!error)
 734                 error = notify_change(mnt_userns, path->dentry, &newattrs,
 735                                       &delegated_inode);
 736         inode_unlock(inode);
 737         if (delegated_inode) {
 738                 error = break_deleg_wait(&delegated_inode);
 739                 if (!error)
 740                         goto retry_deleg;
 741         }
 742         return error;
 743 }
 744
 745 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 746                 int flag)
 747 {
 748         struct path path;
 749         int error = -EINVAL;
 750         int lookup_flags;
 751
 752         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 753                 goto out;
 754
 755         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 756         if (flag & AT_EMPTY_PATH)
 757                 lookup_flags |= LOOKUP_EMPTY;
 758 retry:
 759         error = user_path_at(dfd, filename, lookup_flags, &path);
 760         if (error)
 761                 goto out;
 762         error = mnt_want_write(path.mnt);
 763         if (error)
 764                 goto out_release;
 765         error = chown_common(&path, user, group);
 766         mnt_drop_write(path.mnt);
 767 out_release:
 768         path_put(&path);
 769         if (retry_estale(error, lookup_flags)) {
 770                 lookup_flags |= LOOKUP_REVAL;
 771                 goto retry;
 772         }
 773 out:
 774         return error;
 775 }
 776
 777 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 778                 gid_t, group, int, flag)
 779 {
 780         return do_fchownat(dfd, filename, user, group, flag);
 781 }
 782
 783 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 784 {
 785         return do_fchownat(AT_FDCWD, filename, user, group, 0);
 786 }
 787
 788 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 789 {
 790         return do_fchownat(AT_FDCWD, filename, user, group,
 791                            AT_SYMLINK_NOFOLLOW);
 792 }
 793
 794 int vfs_fchown(struct file *file, uid_t user, gid_t group)
 795 {
 796         int error;
 797
 798         error = mnt_want_write_file(file);
 799         if (error)
 800                 return error;
 801         audit_file(file);
 802         error = chown_common(&file->f_path, user, group);
 803         mnt_drop_write_file(file);
 804         return error;
 805 }
 806
 807 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 808 {
 809         struct fd f = fdget(fd);
 810         int error = -EBADF;
 811
 812         if (f.file) {
 813                 error = vfs_fchown(f.file, user, group);
 814                 fdput(f);
 815         }
 816         return error;
 817 }
 818
 819 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 820 {
 821         return ksys_fchown(fd, user, group);
 822 }
 823
 824 static int do_dentry_open(struct file *f,
 825                           struct inode *inode,
 826                           int (*open)(struct inode *, struct file *))
 827 {
 828         static const struct file_operations empty_fops = {};
 829         int error;
 830
 831         path_get(&f->f_path);
 832         f->f_inode = inode;
 833         f->f_mapping = inode->i_mapping;
 834         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 835         f->f_sb_err = file_sample_sb_err(f);
 836
 837         if (unlikely(f->f_flags & O_PATH)) {
 838                 f->f_mode = FMODE_PATH | FMODE_OPENED;
 839                 f->f_op = &empty_fops;
 840                 return 0;
 841         }
 842
 843         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 844                 error = get_write_access(inode);
 845                 if (unlikely(error))
 846                         goto cleanup_file;
 847                 error = __mnt_want_write(f->f_path.mnt);
 848                 if (unlikely(error)) {
 849                         put_write_access(inode);
 850                         goto cleanup_file;
 851                 }
 852                 f->f_mode |= FMODE_WRITER;
 853         }
 854
 855         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 856         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 857                 f->f_mode |= FMODE_ATOMIC_POS;
 858
 859         f->f_op = fops_get(inode->i_fop);
 860         if (WARN_ON(!f->f_op)) {
 861                 error = -ENODEV;
 862                 goto cleanup_all;
 863         }
 864
 865         error = security_file_open(f);
 866         if (error)
 867                 goto cleanup_all;
 868
 869         error = break_lease(locks_inode(f), f->f_flags);
 870         if (error)
 871                 goto cleanup_all;
 872
 873         /* normally all 3 are set; ->open() can clear them if needed */
 874         f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 875         if (!open)
 876                 open = f->f_op->open;
 877         if (open) {
 878                 error = open(inode, f);
 879                 if (error)
 880                         goto cleanup_all;
 881         }
 882         f->f_mode |= FMODE_OPENED;
 883         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 884                 i_readcount_inc(inode);
 885         if ((f->f_mode & FMODE_READ) &&
 886              likely(f->f_op->read || f->f_op->read_iter))
 887                 f->f_mode |= FMODE_CAN_READ;
 888         if ((f->f_mode & FMODE_WRITE) &&
 889              likely(f->f_op->write || f->f_op->write_iter))
 890                 f->f_mode |= FMODE_CAN_WRITE;
 891         if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
 892                 f->f_mode &= ~FMODE_LSEEK;
 893         if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
 894                 f->f_mode |= FMODE_CAN_ODIRECT;
 895
 896         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 897         f->f_iocb_flags = iocb_flags(f);
 898
 899         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 900
 901         if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
 902                 return -EINVAL;
 903
 904         /*
 905          * XXX: Huge page cache doesn't support writing yet. Drop all page
 906          * cache for this file before processing writes.
 907          */
 908         if (f->f_mode & FMODE_WRITE) {
 909                 /*
 910                  * Paired with smp_mb() in collapse_file() to ensure nr_thps
 911                  * is up to date and the update to i_writecount by
 912                  * get_write_access() is visible. Ensures subsequent insertion
 913                  * of THPs into the page cache will fail.
 914                  */
 915                 smp_mb();
 916                 if (filemap_nr_thps(inode->i_mapping)) {
 917                         struct address_space *mapping = inode->i_mapping;
 918
 919                         filemap_invalidate_lock(inode->i_mapping);
 920                         /*
 921                          * unmap_mapping_range just need to be called once
 922                          * here, because the private pages is not need to be
 923                          * unmapped mapping (e.g. data segment of dynamic
 924                          * shared libraries here).
 925                          */
 926                         unmap_mapping_range(mapping, 0, 0, 0);
 927                         truncate_inode_pages(mapping, 0);
 928                         filemap_invalidate_unlock(inode->i_mapping);
 929                 }
 930         }
 931
 932         return 0;
 933
 934 cleanup_all:
 935         if (WARN_ON_ONCE(error > 0))
 936                 error = -EINVAL;
 937         fops_put(f->f_op);
 938         if (f->f_mode & FMODE_WRITER) {
 939                 put_write_access(inode);
 940                 __mnt_drop_write(f->f_path.mnt);
 941         }
 942 cleanup_file:
 943         path_put(&f->f_path);
 944         f->f_path.mnt = NULL;
 945         f->f_path.dentry = NULL;
 946         f->f_inode = NULL;
 947         return error;
 948 }
 949
 950 /**
 951  * finish_open - finish opening a file
 952  * @file: file pointer
 953  * @dentry: pointer to dentry
 954  * @open: open callback
 955  * @opened: state of open
 956  *
 957  * This can be used to finish opening a file passed to i_op->atomic_open().
 958  *
 959  * If the open callback is set to NULL, then the standard f_op->open()
 960  * filesystem callback is substituted.
 961  *
 962  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 963  * the return value of d_splice_alias(), then the caller needs to perform dput()
 964  * on it after finish_open().
 965  *
 966  * Returns zero on success or -errno if the open failed.
 967  */
 968 int finish_open(struct file *file, struct dentry *dentry,
 969                 int (*open)(struct inode *, struct file *))
 970 {
 971         BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
 972
 973         file->f_path.dentry = dentry;
 974         return do_dentry_open(file, d_backing_inode(dentry), open);
 975 }
 976 EXPORT_SYMBOL(finish_open);
 977
 978 /**
 979  * finish_no_open - finish ->atomic_open() without opening the file
 980  *
 981  * @file: file pointer
 982  * @dentry: dentry or NULL (as returned from ->lookup())
 983  *
 984  * This can be used to set the result of a successful lookup in ->atomic_open().
 985  *
 986  * NB: unlike finish_open() this function does consume the dentry reference and
 987  * the caller need not dput() it.
 988  *
 989  * Returns "0" which must be the return value of ->atomic_open() after having
 990  * called this function.
 991  */
 992 int finish_no_open(struct file *file, struct dentry *dentry)
 993 {
 994         file->f_path.dentry = dentry;
 995         return 0;
 996 }
 997 EXPORT_SYMBOL(finish_no_open);
 998
 999 char *file_path(struct file *filp, char *buf, int buflen)
1000 {
1001         return d_path(&filp->f_path, buf, buflen);
1002 }
1003 EXPORT_SYMBOL(file_path);
1004
1005 /**
1006  * vfs_open - open the file at the given path
1007  * @path: path to open
1008  * @file: newly allocated file with f_flag initialized
1009  * @cred: credentials to use
1010  */
1011 int vfs_open(const struct path *path, struct file *file)
1012 {
1013         file->f_path = *path;
1014         return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1015 }
1016
1017 struct file *dentry_open(const struct path *path, int flags,
1018                          const struct cred *cred)
1019 {
1020         int error;
1021         struct file *f;
1022
1023         validate_creds(cred);
1024
1025         /* We must always pass in a valid mount pointer. */
1026         BUG_ON(!path->mnt);
1027
1028         f = alloc_empty_file(flags, cred);
1029         if (!IS_ERR(f)) {
1030                 error = vfs_open(path, f);
1031                 if (error) {
1032                         fput(f);
1033                         f = ERR_PTR(error);
1034                 }
1035         }
1036         return f;
1037 }
1038 EXPORT_SYMBOL(dentry_open);
1039
1040 /**
1041  * dentry_create - Create and open a file
1042  * @path: path to create
1043  * @flags: O_ flags
1044  * @mode: mode bits for new file
1045  * @cred: credentials to use
1046  *
1047  * Caller must hold the parent directory's lock, and have prepared
1048  * a negative dentry, placed in @path->dentry, for the new file.
1049  *
1050  * Caller sets @path->mnt to the vfsmount of the filesystem where
1051  * the new file is to be created. The parent directory and the
1052  * negative dentry must reside on the same filesystem instance.
1053  *
1054  * On success, returns a "struct file *". Otherwise a ERR_PTR
1055  * is returned.
1056  */
1057 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1058                            const struct cred *cred)
1059 {
1060         struct file *f;
1061         int error;
1062
1063         validate_creds(cred);
1064         f = alloc_empty_file(flags, cred);
1065         if (IS_ERR(f))
1066                 return f;
1067
1068         error = vfs_create(mnt_user_ns(path->mnt),
1069                            d_inode(path->dentry->d_parent),
1070                            path->dentry, mode, true);
1071         if (!error)
1072                 error = vfs_open(path, f);
1073
1074         if (unlikely(error)) {
1075                 fput(f);
1076                 return ERR_PTR(error);
1077         }
1078         return f;
1079 }
1080 EXPORT_SYMBOL(dentry_create);
1081
1082 struct file *open_with_fake_path(const struct path *path, int flags,
1083                                 struct inode *inode, const struct cred *cred)
1084 {
1085         struct file *f = alloc_empty_file_noaccount(flags, cred);
1086         if (!IS_ERR(f)) {
1087                 int error;
1088
1089                 f->f_path = *path;
1090                 error = do_dentry_open(f, inode, NULL);
1091                 if (error) {
1092                         fput(f);
1093                         f = ERR_PTR(error);
1094                 }
1095         }
1096         return f;
1097 }
1098 EXPORT_SYMBOL(open_with_fake_path);
1099
1100 #define WILL_CREATE(flags)      (flags & (O_CREAT | __O_TMPFILE))
1101 #define O_PATH_FLAGS            (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1102
1103 inline struct open_how build_open_how(int flags, umode_t mode)
1104 {
1105         struct open_how how = {
1106                 .flags = flags & VALID_OPEN_FLAGS,
1107                 .mode = mode & S_IALLUGO,
1108         };
1109
1110         /* O_PATH beats everything else. */
1111         if (how.flags & O_PATH)
1112                 how.flags &= O_PATH_FLAGS;
1113         /* Modes should only be set for create-like flags. */
1114         if (!WILL_CREATE(how.flags))
1115                 how.mode = 0;
1116         return how;
1117 }
1118
1119 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1120 {
1121         u64 flags = how->flags;
1122         u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1123         int lookup_flags = 0;
1124         int acc_mode = ACC_MODE(flags);
1125
1126         BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1127                          "struct open_flags doesn't yet handle flags > 32 bits");
1128
1129         /*
1130          * Strip flags that either shouldn't be set by userspace like
1131          * FMODE_NONOTIFY or that aren't relevant in determining struct
1132          * open_flags like O_CLOEXEC.
1133          */
1134         flags &= ~strip;
1135
1136         /*
1137          * Older syscalls implicitly clear all of the invalid flags or argument
1138          * values before calling build_open_flags(), but openat2(2) checks all
1139          * of its arguments.
1140          */
1141         if (flags & ~VALID_OPEN_FLAGS)
1142                 return -EINVAL;
1143         if (how->resolve & ~VALID_RESOLVE_FLAGS)
1144                 return -EINVAL;
1145
1146         /* Scoping flags are mutually exclusive. */
1147         if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1148                 return -EINVAL;
1149
1150         /* Deal with the mode. */
1151         if (WILL_CREATE(flags)) {
1152                 if (how->mode & ~S_IALLUGO)
1153                         return -EINVAL;
1154                 op->mode = how->mode | S_IFREG;
1155         } else {
1156                 if (how->mode != 0)
1157                         return -EINVAL;
1158                 op->mode = 0;
1159         }
1160
1161         /*
1162          * In order to ensure programs get explicit errors when trying to use
1163          * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1164          * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1165          * have to require userspace to explicitly set it.
1166          */
1167         if (flags & __O_TMPFILE) {
1168                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1169                         return -EINVAL;
1170                 if (!(acc_mode & MAY_WRITE))
1171                         return -EINVAL;
1172         }
1173         if (flags & O_PATH) {
1174                 /* O_PATH only permits certain other flags to be set. */
1175                 if (flags & ~O_PATH_FLAGS)
1176                         return -EINVAL;
1177                 acc_mode = 0;
1178         }
1179
1180         /*
1181          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
1182          * check for O_DSYNC if the need any syncing at all we enforce it's
1183          * always set instead of having to deal with possibly weird behaviour
1184          * for malicious applications setting only __O_SYNC.
1185          */
1186         if (flags & __O_SYNC)
1187                 flags |= O_DSYNC;
1188
1189         op->open_flag = flags;
1190
1191         /* O_TRUNC implies we need access checks for write permissions */
1192         if (flags & O_TRUNC)
1193                 acc_mode |= MAY_WRITE;
1194
1195         /* Allow the LSM permission hook to distinguish append
1196            access from general write access. */
1197         if (flags & O_APPEND)
1198                 acc_mode |= MAY_APPEND;
1199
1200         op->acc_mode = acc_mode;
1201
1202         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1203
1204         if (flags & O_CREAT) {
1205                 op->intent |= LOOKUP_CREATE;
1206                 if (flags & O_EXCL) {
1207                         op->intent |= LOOKUP_EXCL;
1208                         flags |= O_NOFOLLOW;
1209                 }
1210         }
1211
1212         if (flags & O_DIRECTORY)
1213                 lookup_flags |= LOOKUP_DIRECTORY;
1214         if (!(flags & O_NOFOLLOW))
1215                 lookup_flags |= LOOKUP_FOLLOW;
1216
1217         if (how->resolve & RESOLVE_NO_XDEV)
1218                 lookup_flags |= LOOKUP_NO_XDEV;
1219         if (how->resolve & RESOLVE_NO_MAGICLINKS)
1220                 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1221         if (how->resolve & RESOLVE_NO_SYMLINKS)
1222                 lookup_flags |= LOOKUP_NO_SYMLINKS;
1223         if (how->resolve & RESOLVE_BENEATH)
1224                 lookup_flags |= LOOKUP_BENEATH;
1225         if (how->resolve & RESOLVE_IN_ROOT)
1226                 lookup_flags |= LOOKUP_IN_ROOT;
1227         if (how->resolve & RESOLVE_CACHED) {
1228                 /* Don't bother even trying for create/truncate/tmpfile open */
1229                 if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1230                         return -EAGAIN;
1231                 lookup_flags |= LOOKUP_CACHED;
1232         }
1233
1234         op->lookup_flags = lookup_flags;
1235         return 0;
1236 }
1237
1238 /**
1239  * file_open_name - open file and return file pointer
1240  *
1241  * @name:       struct filename containing path to open
1242  * @flags:      open flags as per the open(2) second argument
1243  * @mode:       mode for the new file if O_CREAT is set, else ignored
1244  *
1245  * This is the helper to open a file from kernelspace if you really
1246  * have to.  But in generally you should not do this, so please move
1247  * along, nothing to see here..
1248  */
1249 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1250 {
1251         struct open_flags op;
1252         struct open_how how = build_open_how(flags, mode);
1253         int err = build_open_flags(&how, &op);
1254         if (err)
1255                 return ERR_PTR(err);
1256         return do_filp_open(AT_FDCWD, name, &op);
1257 }
1258
1259 /**
1260  * filp_open - open file and return file pointer
1261  *
1262  * @filename:   path to open
1263  * @flags:      open flags as per the open(2) second argument
1264  * @mode:       mode for the new file if O_CREAT is set, else ignored
1265  *
1266  * This is the helper to open a file from kernelspace if you really
1267  * have to.  But in generally you should not do this, so please move
1268  * along, nothing to see here..
1269  */
1270 struct file *filp_open(const char *filename, int flags, umode_t mode)
1271 {
1272         struct filename *name = getname_kernel(filename);
1273         struct file *file = ERR_CAST(name);
1274
1275         if (!IS_ERR(name)) {
1276                 file = file_open_name(name, flags, mode);
1277                 putname(name);
1278         }
1279         return file;
1280 }
1281 EXPORT_SYMBOL(filp_open);
1282
1283 struct file *file_open_root(const struct path *root,
1284                             const char *filename, int flags, umode_t mode)
1285 {
1286         struct open_flags op;
1287         struct open_how how = build_open_how(flags, mode);
1288         int err = build_open_flags(&how, &op);
1289         if (err)
1290                 return ERR_PTR(err);
1291         return do_file_open_root(root, filename, &op);
1292 }
1293 EXPORT_SYMBOL(file_open_root);
1294
1295 static long do_sys_openat2(int dfd, const char __user *filename,
1296                            struct open_how *how)
1297 {
1298         struct open_flags op;
1299         int fd = build_open_flags(how, &op);
1300         struct filename *tmp;
1301
1302         if (fd)
1303                 return fd;
1304
1305         tmp = getname(filename);
1306         if (IS_ERR(tmp))
1307                 return PTR_ERR(tmp);
1308
1309         fd = get_unused_fd_flags(how->flags);
1310         if (fd >= 0) {
1311                 struct file *f = do_filp_open(dfd, tmp, &op);
1312                 if (IS_ERR(f)) {
1313                         put_unused_fd(fd);
1314                         fd = PTR_ERR(f);
1315                 } else {
1316                         fsnotify_open(f);
1317                         fd_install(fd, f);
1318                 }
1319         }
1320         putname(tmp);
1321         return fd;
1322 }
1323
1324 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1325 {
1326         struct open_how how = build_open_how(flags, mode);
1327         return do_sys_openat2(dfd, filename, &how);
1328 }
1329
1330
1331 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1332 {
1333         if (force_o_largefile())
1334                 flags |= O_LARGEFILE;
1335         return do_sys_open(AT_FDCWD, filename, flags, mode);
1336 }
1337
1338 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1339                 umode_t, mode)
1340 {
1341         if (force_o_largefile())
1342                 flags |= O_LARGEFILE;
1343         return do_sys_open(dfd, filename, flags, mode);
1344 }
1345
1346 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1347                 struct open_how __user *, how, size_t, usize)
1348 {
1349         int err;
1350         struct open_how tmp;
1351
1352         BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1353         BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1354
1355         if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1356                 return -EINVAL;
1357
1358         err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1359         if (err)
1360                 return err;
1361
1362         audit_openat2_how(&tmp);
1363
1364         /* O_LARGEFILE is only allowed for non-O_PATH. */
1365         if (!(tmp.flags & O_PATH) && force_o_largefile())
1366                 tmp.flags |= O_LARGEFILE;
1367
1368         return do_sys_openat2(dfd, filename, &tmp);
1369 }
1370
1371 #ifdef CONFIG_COMPAT
1372 /*
1373  * Exactly like sys_open(), except that it doesn't set the
1374  * O_LARGEFILE flag.
1375  */
1376 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1377 {
1378         return do_sys_open(AT_FDCWD, filename, flags, mode);
1379 }
1380
1381 /*
1382  * Exactly like sys_openat(), except that it doesn't set the
1383  * O_LARGEFILE flag.
1384  */
1385 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1386 {
1387         return do_sys_open(dfd, filename, flags, mode);
1388 }
1389 #endif
1390
1391 #ifndef __alpha__
1392
1393 /*
1394  * For backward compatibility?  Maybe this should be moved
1395  * into arch/i386 instead?
1396  */
1397 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1398 {
1399         int flags = O_CREAT | O_WRONLY | O_TRUNC;
1400
1401         if (force_o_largefile())
1402                 flags |= O_LARGEFILE;
1403         return do_sys_open(AT_FDCWD, pathname, flags, mode);
1404 }
1405 #endif
1406
1407 /*
1408  * "id" is the POSIX thread ID. We use the
1409  * files pointer for this..
1410  */
1411 int filp_close(struct file *filp, fl_owner_t id)
1412 {
1413         int retval = 0;
1414
1415         if (!file_count(filp)) {
1416                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1417                 return 0;
1418         }
1419
1420         if (filp->f_op->flush)
1421                 retval = filp->f_op->flush(filp, id);
1422
1423         if (likely(!(filp->f_mode & FMODE_PATH))) {
1424                 dnotify_flush(filp, id);
1425                 locks_remove_posix(filp, id);
1426         }
1427         fput(filp);
1428         return retval;
1429 }
1430
1431 EXPORT_SYMBOL(filp_close);
1432
1433 /*
1434  * Careful here! We test whether the file pointer is NULL before
1435  * releasing the fd. This ensures that one clone task can't release
1436  * an fd while another clone is opening it.
1437  */
1438 SYSCALL_DEFINE1(close, unsigned int, fd)
1439 {
1440         int retval = close_fd(fd);
1441
1442         /* can't restart close syscall because file table entry was cleared */
1443         if (unlikely(retval == -ERESTARTSYS ||
1444                      retval == -ERESTARTNOINTR ||
1445                      retval == -ERESTARTNOHAND ||
1446                      retval == -ERESTART_RESTARTBLOCK))
1447                 retval = -EINTR;
1448
1449         return retval;
1450 }
1451
1452 /**
1453  * close_range() - Close all file descriptors in a given range.
1454  *
1455  * @fd:     starting file descriptor to close
1456  * @max_fd: last file descriptor to close
1457  * @flags:  reserved for future extensions
1458  *
1459  * This closes a range of file descriptors. All file descriptors
1460  * from @fd up to and including @max_fd are closed.
1461  * Currently, errors to close a given file descriptor are ignored.
1462  */
1463 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1464                 unsigned int, flags)
1465 {
1466         return __close_range(fd, max_fd, flags);
1467 }
1468
1469 /*
1470  * This routine simulates a hangup on the tty, to arrange that users
1471  * are given clean terminals at login time.
1472  */
1473 SYSCALL_DEFINE0(vhangup)
1474 {
1475         if (capable(CAP_SYS_TTY_CONFIG)) {
1476                 tty_vhangup_self();
1477                 return 0;
1478         }
1479         return -EPERM;
1480 }
1481
1482 /*
1483  * Called when an inode is about to be open.
1484  * We use this to disallow opening large files on 32bit systems if
1485  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1486  * on this flag in sys_open.
1487  */
1488 int generic_file_open(struct inode * inode, struct file * filp)
1489 {
1490         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1491                 return -EOVERFLOW;
1492         return 0;
1493 }
1494
1495 EXPORT_SYMBOL(generic_file_open);
1496
1497 /*
1498  * This is used by subsystems that don't want seekable
1499  * file descriptors. The function is not supposed to ever fail, the only
1500  * reason it returns an 'int' and not 'void' is so that it can be plugged
1501  * directly into file_operations structure.
1502  */
1503 int nonseekable_open(struct inode *inode, struct file *filp)
1504 {
1505         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1506         return 0;
1507 }
1508
1509 EXPORT_SYMBOL(nonseekable_open);
1510
1511 /*
1512  * stream_open is used by subsystems that want stream-like file descriptors.
1513  * Such file descriptors are not seekable and don't have notion of position
1514  * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1515  * Contrary to file descriptors of other regular files, .read() and .write()
1516  * can run simultaneously.
1517  *
1518  * stream_open never fails and is marked to return int so that it could be
1519  * directly used as file_operations.open .
1520  */
1521 int stream_open(struct inode *inode, struct file *filp)
1522 {
1523         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1524         filp->f_mode |= FMODE_STREAM;
1525         return 0;
1526 }
1527
1528 EXPORT_SYMBOL(stream_open);