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
  36 #include "internal.h"
  37
  38 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  39         struct file *filp)
  40 {
  41         int ret;
  42         struct iattr newattrs;
  43
  44         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  45         if (length < 0)
  46                 return -EINVAL;
  47
  48         newattrs.ia_size = length;
  49         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  50         if (filp) {
  51                 newattrs.ia_file = filp;
  52                 newattrs.ia_valid |= ATTR_FILE;
  53         }
  54
  55         /* Remove suid, sgid, and file capabilities on truncate too */
  56         ret = dentry_needs_remove_privs(dentry);
  57         if (ret < 0)
  58                 return ret;
  59         if (ret)
  60                 newattrs.ia_valid |= ret | ATTR_FORCE;
  61
  62         inode_lock(dentry->d_inode);
  63         /* Note any delegations or leases have already been broken: */
  64         ret = notify_change(dentry, &newattrs, NULL);
  65         inode_unlock(dentry->d_inode);
  66         return ret;
  67 }
  68
  69 long vfs_truncate(const struct path *path, loff_t length)
  70 {
  71         struct inode *inode;
  72         long error;
  73
  74         inode = path->dentry->d_inode;
  75
  76         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  77         if (S_ISDIR(inode->i_mode))
  78                 return -EISDIR;
  79         if (!S_ISREG(inode->i_mode))
  80                 return -EINVAL;
  81
  82         error = mnt_want_write(path->mnt);
  83         if (error)
  84                 goto out;
  85
  86         error = inode_permission(inode, MAY_WRITE);
  87         if (error)
  88                 goto mnt_drop_write_and_out;
  89
  90         error = -EPERM;
  91         if (IS_APPEND(inode))
  92                 goto mnt_drop_write_and_out;
  93
  94         error = get_write_access(inode);
  95         if (error)
  96                 goto mnt_drop_write_and_out;
  97
  98         /*
  99          * Make sure that there are no leases.  get_write_access() protects
 100          * against the truncate racing with a lease-granting setlease().
 101          */
 102         error = break_lease(inode, O_WRONLY);
 103         if (error)
 104                 goto put_write_and_out;
 105
 106         error = locks_verify_truncate(inode, NULL, length);
 107         if (!error)
 108                 error = security_path_truncate(path);
 109         if (!error)
 110                 error = do_truncate(path->dentry, length, 0, NULL);
 111
 112 put_write_and_out:
 113         put_write_access(inode);
 114 mnt_drop_write_and_out:
 115         mnt_drop_write(path->mnt);
 116 out:
 117         return error;
 118 }
 119 EXPORT_SYMBOL_GPL(vfs_truncate);
 120
 121 long do_sys_truncate(const char __user *pathname, loff_t length)
 122 {
 123         unsigned int lookup_flags = LOOKUP_FOLLOW;
 124         struct path path;
 125         int error;
 126
 127         if (length < 0) /* sorry, but loff_t says... */
 128                 return -EINVAL;
 129
 130 retry:
 131         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 132         if (!error) {
 133                 error = vfs_truncate(&path, length);
 134                 path_put(&path);
 135         }
 136         if (retry_estale(error, lookup_flags)) {
 137                 lookup_flags |= LOOKUP_REVAL;
 138                 goto retry;
 139         }
 140         return error;
 141 }
 142
 143 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 144 {
 145         return do_sys_truncate(path, length);
 146 }
 147
 148 #ifdef CONFIG_COMPAT
 149 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 150 {
 151         return do_sys_truncate(path, length);
 152 }
 153 #endif
 154
 155 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 156 {
 157         struct inode *inode;
 158         struct dentry *dentry;
 159         struct fd f;
 160         int error;
 161
 162         error = -EINVAL;
 163         if (length < 0)
 164                 goto out;
 165         error = -EBADF;
 166         f = fdget(fd);
 167         if (!f.file)
 168                 goto out;
 169
 170         /* explicitly opened as large or we are on 64-bit box */
 171         if (f.file->f_flags & O_LARGEFILE)
 172                 small = 0;
 173
 174         dentry = f.file->f_path.dentry;
 175         inode = dentry->d_inode;
 176         error = -EINVAL;
 177         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 178                 goto out_putf;
 179
 180         error = -EINVAL;
 181         /* Cannot ftruncate over 2^31 bytes without large file support */
 182         if (small && length > MAX_NON_LFS)
 183                 goto out_putf;
 184
 185         error = -EPERM;
 186         /* Check IS_APPEND on real upper inode */
 187         if (IS_APPEND(file_inode(f.file)))
 188                 goto out_putf;
 189
 190         sb_start_write(inode->i_sb);
 191         error = locks_verify_truncate(inode, f.file, length);
 192         if (!error)
 193                 error = security_path_truncate(&f.file->f_path);
 194         if (!error)
 195                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 196         sb_end_write(inode->i_sb);
 197 out_putf:
 198         fdput(f);
 199 out:
 200         return error;
 201 }
 202
 203 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 204 {
 205         return do_sys_ftruncate(fd, length, 1);
 206 }
 207
 208 #ifdef CONFIG_COMPAT
 209 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 210 {
 211         return do_sys_ftruncate(fd, length, 1);
 212 }
 213 #endif
 214
 215 /* LFS versions of truncate are only needed on 32 bit machines */
 216 #if BITS_PER_LONG == 32
 217 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 218 {
 219         return do_sys_truncate(path, length);
 220 }
 221
 222 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 223 {
 224         return do_sys_ftruncate(fd, length, 0);
 225 }
 226 #endif /* BITS_PER_LONG == 32 */
 227
 228
 229 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 230 {
 231         struct inode *inode = file_inode(file);
 232         long ret;
 233
 234         if (offset < 0 || len <= 0)
 235                 return -EINVAL;
 236
 237         /* Return error if mode is not supported */
 238         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 239                 return -EOPNOTSUPP;
 240
 241         /* Punch hole and zero range are mutually exclusive */
 242         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 243             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 244                 return -EOPNOTSUPP;
 245
 246         /* Punch hole must have keep size set */
 247         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 248             !(mode & FALLOC_FL_KEEP_SIZE))
 249                 return -EOPNOTSUPP;
 250
 251         /* Collapse range should only be used exclusively. */
 252         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 253             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 254                 return -EINVAL;
 255
 256         /* Insert range should only be used exclusively. */
 257         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 258             (mode & ~FALLOC_FL_INSERT_RANGE))
 259                 return -EINVAL;
 260
 261         /* Unshare range should only be used with allocate mode. */
 262         if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 263             (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 264                 return -EINVAL;
 265
 266         if (!(file->f_mode & FMODE_WRITE))
 267                 return -EBADF;
 268
 269         /*
 270          * We can only allow pure fallocate on append only files
 271          */
 272         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 273                 return -EPERM;
 274
 275         if (IS_IMMUTABLE(inode))
 276                 return -EPERM;
 277
 278         /*
 279          * We cannot allow any fallocate operation on an active swapfile
 280          */
 281         if (IS_SWAPFILE(inode))
 282                 return -ETXTBSY;
 283
 284         /*
 285          * Revalidate the write permissions, in case security policy has
 286          * changed since the files were opened.
 287          */
 288         ret = security_file_permission(file, MAY_WRITE);
 289         if (ret)
 290                 return ret;
 291
 292         if (S_ISFIFO(inode->i_mode))
 293                 return -ESPIPE;
 294
 295         if (S_ISDIR(inode->i_mode))
 296                 return -EISDIR;
 297
 298         if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 299                 return -ENODEV;
 300
 301         /* Check for wrap through zero too */
 302         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 303                 return -EFBIG;
 304
 305         if (!file->f_op->fallocate)
 306                 return -EOPNOTSUPP;
 307
 308         file_start_write(file);
 309         ret = file->f_op->fallocate(file, mode, offset, len);
 310
 311         /*
 312          * Create inotify and fanotify events.
 313          *
 314          * To keep the logic simple always create events if fallocate succeeds.
 315          * This implies that events are even created if the file size remains
 316          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 317          */
 318         if (ret == 0)
 319                 fsnotify_modify(file);
 320
 321         file_end_write(file);
 322         return ret;
 323 }
 324 EXPORT_SYMBOL_GPL(vfs_fallocate);
 325
 326 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
 327 {
 328         struct fd f = fdget(fd);
 329         int error = -EBADF;
 330
 331         if (f.file) {
 332                 error = vfs_fallocate(f.file, mode, offset, len);
 333                 fdput(f);
 334         }
 335         return error;
 336 }
 337
 338 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 339 {
 340         return ksys_fallocate(fd, mode, offset, len);
 341 }
 342
 343 /*
 344  * access() needs to use the real uid/gid, not the effective uid/gid.
 345  * We do this by temporarily clearing all FS-related capabilities and
 346  * switching the fsuid/fsgid around to the real ones.
 347  */
 348 long do_faccessat(int dfd, const char __user *filename, int mode)
 349 {
 350         const struct cred *old_cred;
 351         struct cred *override_cred;
 352         struct path path;
 353         struct inode *inode;
 354         int res;
 355         unsigned int lookup_flags = LOOKUP_FOLLOW;
 356
 357         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 358                 return -EINVAL;
 359
 360         override_cred = prepare_creds();
 361         if (!override_cred)
 362                 return -ENOMEM;
 363
 364         override_cred->fsuid = override_cred->uid;
 365         override_cred->fsgid = override_cred->gid;
 366
 367         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 368                 /* Clear the capabilities if we switch to a non-root user */
 369                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 370                 if (!uid_eq(override_cred->uid, root_uid))
 371                         cap_clear(override_cred->cap_effective);
 372                 else
 373                         override_cred->cap_effective =
 374                                 override_cred->cap_permitted;
 375         }
 376
 377         old_cred = override_creds(override_cred);
 378 retry:
 379         res = user_path_at(dfd, filename, lookup_flags, &path);
 380         if (res)
 381                 goto out;
 382
 383         inode = d_backing_inode(path.dentry);
 384
 385         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 386                 /*
 387                  * MAY_EXEC on regular files is denied if the fs is mounted
 388                  * with the "noexec" flag.
 389                  */
 390                 res = -EACCES;
 391                 if (path_noexec(&path))
 392                         goto out_path_release;
 393         }
 394
 395         res = inode_permission(inode, mode | MAY_ACCESS);
 396         /* SuS v2 requires we report a read only fs too */
 397         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 398                 goto out_path_release;
 399         /*
 400          * This is a rare case where using __mnt_is_readonly()
 401          * is OK without a mnt_want/drop_write() pair.  Since
 402          * no actual write to the fs is performed here, we do
 403          * not need to telegraph to that to anyone.
 404          *
 405          * By doing this, we accept that this access is
 406          * inherently racy and know that the fs may change
 407          * state before we even see this result.
 408          */
 409         if (__mnt_is_readonly(path.mnt))
 410                 res = -EROFS;
 411
 412 out_path_release:
 413         path_put(&path);
 414         if (retry_estale(res, lookup_flags)) {
 415                 lookup_flags |= LOOKUP_REVAL;
 416                 goto retry;
 417         }
 418 out:
 419         revert_creds(old_cred);
 420         put_cred(override_cred);
 421         return res;
 422 }
 423
 424 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 425 {
 426         return do_faccessat(dfd, filename, mode);
 427 }
 428
 429 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 430 {
 431         return do_faccessat(AT_FDCWD, filename, mode);
 432 }
 433
 434 int ksys_chdir(const char __user *filename)
 435 {
 436         struct path path;
 437         int error;
 438         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 439 retry:
 440         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 441         if (error)
 442                 goto out;
 443
 444         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 445         if (error)
 446                 goto dput_and_out;
 447
 448         set_fs_pwd(current->fs, &path);
 449
 450 dput_and_out:
 451         path_put(&path);
 452         if (retry_estale(error, lookup_flags)) {
 453                 lookup_flags |= LOOKUP_REVAL;
 454                 goto retry;
 455         }
 456 out:
 457         return error;
 458 }
 459
 460 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 461 {
 462         return ksys_chdir(filename);
 463 }
 464
 465 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 466 {
 467         struct fd f = fdget_raw(fd);
 468         int error;
 469
 470         error = -EBADF;
 471         if (!f.file)
 472                 goto out;
 473
 474         error = -ENOTDIR;
 475         if (!d_can_lookup(f.file->f_path.dentry))
 476                 goto out_putf;
 477
 478         error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 479         if (!error)
 480                 set_fs_pwd(current->fs, &f.file->f_path);
 481 out_putf:
 482         fdput(f);
 483 out:
 484         return error;
 485 }
 486
 487 int ksys_chroot(const char __user *filename)
 488 {
 489         struct path path;
 490         int error;
 491         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 492 retry:
 493         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 494         if (error)
 495                 goto out;
 496
 497         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 498         if (error)
 499                 goto dput_and_out;
 500
 501         error = -EPERM;
 502         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 503                 goto dput_and_out;
 504         error = security_path_chroot(&path);
 505         if (error)
 506                 goto dput_and_out;
 507
 508         set_fs_root(current->fs, &path);
 509         error = 0;
 510 dput_and_out:
 511         path_put(&path);
 512         if (retry_estale(error, lookup_flags)) {
 513                 lookup_flags |= LOOKUP_REVAL;
 514                 goto retry;
 515         }
 516 out:
 517         return error;
 518 }
 519
 520 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 521 {
 522         return ksys_chroot(filename);
 523 }
 524
 525 static int chmod_common(const struct path *path, umode_t mode)
 526 {
 527         struct inode *inode = path->dentry->d_inode;
 528         struct inode *delegated_inode = NULL;
 529         struct iattr newattrs;
 530         int error;
 531
 532         error = mnt_want_write(path->mnt);
 533         if (error)
 534                 return error;
 535 retry_deleg:
 536         inode_lock(inode);
 537         error = security_path_chmod(path, mode);
 538         if (error)
 539                 goto out_unlock;
 540         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 541         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 542         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 543 out_unlock:
 544         inode_unlock(inode);
 545         if (delegated_inode) {
 546                 error = break_deleg_wait(&delegated_inode);
 547                 if (!error)
 548                         goto retry_deleg;
 549         }
 550         mnt_drop_write(path->mnt);
 551         return error;
 552 }
 553
 554 int ksys_fchmod(unsigned int fd, umode_t mode)
 555 {
 556         struct fd f = fdget(fd);
 557         int err = -EBADF;
 558
 559         if (f.file) {
 560                 audit_file(f.file);
 561                 err = chmod_common(&f.file->f_path, mode);
 562                 fdput(f);
 563         }
 564         return err;
 565 }
 566
 567 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 568 {
 569         return ksys_fchmod(fd, mode);
 570 }
 571
 572 int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
 573 {
 574         struct path path;
 575         int error;
 576         unsigned int lookup_flags = LOOKUP_FOLLOW;
 577 retry:
 578         error = user_path_at(dfd, filename, lookup_flags, &path);
 579         if (!error) {
 580                 error = chmod_common(&path, mode);
 581                 path_put(&path);
 582                 if (retry_estale(error, lookup_flags)) {
 583                         lookup_flags |= LOOKUP_REVAL;
 584                         goto retry;
 585                 }
 586         }
 587         return error;
 588 }
 589
 590 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
 591                 umode_t, mode)
 592 {
 593         return do_fchmodat(dfd, filename, mode);
 594 }
 595
 596 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 597 {
 598         return do_fchmodat(AT_FDCWD, filename, mode);
 599 }
 600
 601 static int chown_common(const struct path *path, uid_t user, gid_t group)
 602 {
 603         struct inode *inode = path->dentry->d_inode;
 604         struct inode *delegated_inode = NULL;
 605         int error;
 606         struct iattr newattrs;
 607         kuid_t uid;
 608         kgid_t gid;
 609
 610         uid = make_kuid(current_user_ns(), user);
 611         gid = make_kgid(current_user_ns(), group);
 612
 613 retry_deleg:
 614         newattrs.ia_valid =  ATTR_CTIME;
 615         if (user != (uid_t) -1) {
 616                 if (!uid_valid(uid))
 617                         return -EINVAL;
 618                 newattrs.ia_valid |= ATTR_UID;
 619                 newattrs.ia_uid = uid;
 620         }
 621         if (group != (gid_t) -1) {
 622                 if (!gid_valid(gid))
 623                         return -EINVAL;
 624                 newattrs.ia_valid |= ATTR_GID;
 625                 newattrs.ia_gid = gid;
 626         }
 627         if (!S_ISDIR(inode->i_mode))
 628                 newattrs.ia_valid |=
 629                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 630         inode_lock(inode);
 631         error = security_path_chown(path, uid, gid);
 632         if (!error)
 633                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 634         inode_unlock(inode);
 635         if (delegated_inode) {
 636                 error = break_deleg_wait(&delegated_inode);
 637                 if (!error)
 638                         goto retry_deleg;
 639         }
 640         return error;
 641 }
 642
 643 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
 644                 int flag)
 645 {
 646         struct path path;
 647         int error = -EINVAL;
 648         int lookup_flags;
 649
 650         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 651                 goto out;
 652
 653         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 654         if (flag & AT_EMPTY_PATH)
 655                 lookup_flags |= LOOKUP_EMPTY;
 656 retry:
 657         error = user_path_at(dfd, filename, lookup_flags, &path);
 658         if (error)
 659                 goto out;
 660         error = mnt_want_write(path.mnt);
 661         if (error)
 662                 goto out_release;
 663         error = chown_common(&path, user, group);
 664         mnt_drop_write(path.mnt);
 665 out_release:
 666         path_put(&path);
 667         if (retry_estale(error, lookup_flags)) {
 668                 lookup_flags |= LOOKUP_REVAL;
 669                 goto retry;
 670         }
 671 out:
 672         return error;
 673 }
 674
 675 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 676                 gid_t, group, int, flag)
 677 {
 678         return do_fchownat(dfd, filename, user, group, flag);
 679 }
 680
 681 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 682 {
 683         return do_fchownat(AT_FDCWD, filename, user, group, 0);
 684 }
 685
 686 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 687 {
 688         return do_fchownat(AT_FDCWD, filename, user, group,
 689                            AT_SYMLINK_NOFOLLOW);
 690 }
 691
 692 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
 693 {
 694         struct fd f = fdget(fd);
 695         int error = -EBADF;
 696
 697         if (!f.file)
 698                 goto out;
 699
 700         error = mnt_want_write_file(f.file);
 701         if (error)
 702                 goto out_fput;
 703         audit_file(f.file);
 704         error = chown_common(&f.file->f_path, user, group);
 705         mnt_drop_write_file(f.file);
 706 out_fput:
 707         fdput(f);
 708 out:
 709         return error;
 710 }
 711
 712 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 713 {
 714         return ksys_fchown(fd, user, group);
 715 }
 716
 717 static int do_dentry_open(struct file *f,
 718                           struct inode *inode,
 719                           int (*open)(struct inode *, struct file *))
 720 {
 721         static const struct file_operations empty_fops = {};
 722         int error;
 723
 724         path_get(&f->f_path);
 725         f->f_inode = inode;
 726         f->f_mapping = inode->i_mapping;
 727
 728         /* Ensure that we skip any errors that predate opening of the file */
 729         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 730
 731         if (unlikely(f->f_flags & O_PATH)) {
 732                 f->f_mode = FMODE_PATH | FMODE_OPENED;
 733                 f->f_op = &empty_fops;
 734                 return 0;
 735         }
 736
 737         /* Any file opened for execve()/uselib() has to be a regular file. */
 738         if (unlikely(f->f_flags & FMODE_EXEC && !S_ISREG(inode->i_mode))) {
 739                 error = -EACCES;
 740                 goto cleanup_file;
 741         }
 742
 743         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 744                 error = get_write_access(inode);
 745                 if (unlikely(error))
 746                         goto cleanup_file;
 747                 error = __mnt_want_write(f->f_path.mnt);
 748                 if (unlikely(error)) {
 749                         put_write_access(inode);
 750                         goto cleanup_file;
 751                 }
 752                 f->f_mode |= FMODE_WRITER;
 753         }
 754
 755         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 756         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 757                 f->f_mode |= FMODE_ATOMIC_POS;
 758
 759         f->f_op = fops_get(inode->i_fop);
 760         if (unlikely(WARN_ON(!f->f_op))) {
 761                 error = -ENODEV;
 762                 goto cleanup_all;
 763         }
 764
 765         error = security_file_open(f);
 766         if (error)
 767                 goto cleanup_all;
 768
 769         error = break_lease(locks_inode(f), f->f_flags);
 770         if (error)
 771                 goto cleanup_all;
 772
 773         /* normally all 3 are set; ->open() can clear them if needed */
 774         f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
 775         if (!open)
 776                 open = f->f_op->open;
 777         if (open) {
 778                 error = open(inode, f);
 779                 if (error)
 780                         goto cleanup_all;
 781         }
 782         f->f_mode |= FMODE_OPENED;
 783         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 784                 i_readcount_inc(inode);
 785         if ((f->f_mode & FMODE_READ) &&
 786              likely(f->f_op->read || f->f_op->read_iter))
 787                 f->f_mode |= FMODE_CAN_READ;
 788         if ((f->f_mode & FMODE_WRITE) &&
 789              likely(f->f_op->write || f->f_op->write_iter))
 790                 f->f_mode |= FMODE_CAN_WRITE;
 791
 792         f->f_write_hint = WRITE_LIFE_NOT_SET;
 793         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 794
 795         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 796
 797         /* NB: we're sure to have correct a_ops only after f_op->open */
 798         if (f->f_flags & O_DIRECT) {
 799                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 800                         return -EINVAL;
 801         }
 802         return 0;
 803
 804 cleanup_all:
 805         if (WARN_ON_ONCE(error > 0))
 806                 error = -EINVAL;
 807         fops_put(f->f_op);
 808         if (f->f_mode & FMODE_WRITER) {
 809                 put_write_access(inode);
 810                 __mnt_drop_write(f->f_path.mnt);
 811         }
 812 cleanup_file:
 813         path_put(&f->f_path);
 814         f->f_path.mnt = NULL;
 815         f->f_path.dentry = NULL;
 816         f->f_inode = NULL;
 817         return error;
 818 }
 819
 820 /**
 821  * finish_open - finish opening a file
 822  * @file: file pointer
 823  * @dentry: pointer to dentry
 824  * @open: open callback
 825  * @opened: state of open
 826  *
 827  * This can be used to finish opening a file passed to i_op->atomic_open().
 828  *
 829  * If the open callback is set to NULL, then the standard f_op->open()
 830  * filesystem callback is substituted.
 831  *
 832  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 833  * the return value of d_splice_alias(), then the caller needs to perform dput()
 834  * on it after finish_open().
 835  *
 836  * On successful return @file is a fully instantiated open file.  After this, if
 837  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 838  *
 839  * Returns zero on success or -errno if the open failed.
 840  */
 841 int finish_open(struct file *file, struct dentry *dentry,
 842                 int (*open)(struct inode *, struct file *))
 843 {
 844         BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
 845
 846         file->f_path.dentry = dentry;
 847         return do_dentry_open(file, d_backing_inode(dentry), open);
 848 }
 849 EXPORT_SYMBOL(finish_open);
 850
 851 /**
 852  * finish_no_open - finish ->atomic_open() without opening the file
 853  *
 854  * @file: file pointer
 855  * @dentry: dentry or NULL (as returned from ->lookup())
 856  *
 857  * This can be used to set the result of a successful lookup in ->atomic_open().
 858  *
 859  * NB: unlike finish_open() this function does consume the dentry reference and
 860  * the caller need not dput() it.
 861  *
 862  * Returns "0" which must be the return value of ->atomic_open() after having
 863  * called this function.
 864  */
 865 int finish_no_open(struct file *file, struct dentry *dentry)
 866 {
 867         file->f_path.dentry = dentry;
 868         return 0;
 869 }
 870 EXPORT_SYMBOL(finish_no_open);
 871
 872 char *file_path(struct file *filp, char *buf, int buflen)
 873 {
 874         return d_path(&filp->f_path, buf, buflen);
 875 }
 876 EXPORT_SYMBOL(file_path);
 877
 878 /**
 879  * vfs_open - open the file at the given path
 880  * @path: path to open
 881  * @file: newly allocated file with f_flag initialized
 882  * @cred: credentials to use
 883  */
 884 int vfs_open(const struct path *path, struct file *file)
 885 {
 886         file->f_path = *path;
 887         return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
 888 }
 889
 890 struct file *dentry_open(const struct path *path, int flags,
 891                          const struct cred *cred)
 892 {
 893         int error;
 894         struct file *f;
 895
 896         validate_creds(cred);
 897
 898         /* We must always pass in a valid mount pointer. */
 899         BUG_ON(!path->mnt);
 900
 901         f = alloc_empty_file(flags, cred);
 902         if (!IS_ERR(f)) {
 903                 error = vfs_open(path, f);
 904                 if (error) {
 905                         fput(f);
 906                         f = ERR_PTR(error);
 907                 }
 908         }
 909         return f;
 910 }
 911 EXPORT_SYMBOL(dentry_open);
 912
 913 struct file *open_with_fake_path(const struct path *path, int flags,
 914                                 struct inode *inode, const struct cred *cred)
 915 {
 916         struct file *f = alloc_empty_file_noaccount(flags, cred);
 917         if (!IS_ERR(f)) {
 918                 int error;
 919
 920                 f->f_path = *path;
 921                 error = do_dentry_open(f, inode, NULL);
 922                 if (error) {
 923                         fput(f);
 924                         f = ERR_PTR(error);
 925                 }
 926         }
 927         return f;
 928 }
 929 EXPORT_SYMBOL(open_with_fake_path);
 930
 931 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 932 {
 933         int lookup_flags = 0;
 934         int acc_mode = ACC_MODE(flags);
 935
 936         /*
 937          * Clear out all open flags we don't know about so that we don't report
 938          * them in fcntl(F_GETFD) or similar interfaces.
 939          */
 940         flags &= VALID_OPEN_FLAGS;
 941
 942         if (flags & (O_CREAT | __O_TMPFILE))
 943                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 944         else
 945                 op->mode = 0;
 946
 947         /* Must never be set by userspace */
 948         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 949
 950         /*
 951          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 952          * check for O_DSYNC if the need any syncing at all we enforce it's
 953          * always set instead of having to deal with possibly weird behaviour
 954          * for malicious applications setting only __O_SYNC.
 955          */
 956         if (flags & __O_SYNC)
 957                 flags |= O_DSYNC;
 958
 959         if (flags & __O_TMPFILE) {
 960                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 961                         return -EINVAL;
 962                 if (!(acc_mode & MAY_WRITE))
 963                         return -EINVAL;
 964         } else if (flags & O_PATH) {
 965                 /*
 966                  * If we have O_PATH in the open flag. Then we
 967                  * cannot have anything other than the below set of flags
 968                  */
 969                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 970                 acc_mode = 0;
 971         }
 972
 973         op->open_flag = flags;
 974
 975         /* O_TRUNC implies we need access checks for write permissions */
 976         if (flags & O_TRUNC)
 977                 acc_mode |= MAY_WRITE;
 978
 979         /* Allow the LSM permission hook to distinguish append
 980            access from general write access. */
 981         if (flags & O_APPEND)
 982                 acc_mode |= MAY_APPEND;
 983
 984         op->acc_mode = acc_mode;
 985
 986         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 987
 988         if (flags & O_CREAT) {
 989                 op->intent |= LOOKUP_CREATE;
 990                 if (flags & O_EXCL)
 991                         op->intent |= LOOKUP_EXCL;
 992         }
 993
 994         if (flags & O_DIRECTORY)
 995                 lookup_flags |= LOOKUP_DIRECTORY;
 996         if (!(flags & O_NOFOLLOW))
 997                 lookup_flags |= LOOKUP_FOLLOW;
 998         op->lookup_flags = lookup_flags;
 999         return 0;
1000 }
1001
1002 /**
1003  * file_open_name - open file and return file pointer
1004  *
1005  * @name:       struct filename containing path to open
1006  * @flags:      open flags as per the open(2) second argument
1007  * @mode:       mode for the new file if O_CREAT is set, else ignored
1008  *
1009  * This is the helper to open a file from kernelspace if you really
1010  * have to.  But in generally you should not do this, so please move
1011  * along, nothing to see here..
1012  */
1013 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1014 {
1015         struct open_flags op;
1016         int err = build_open_flags(flags, mode, &op);
1017         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
1018 }
1019
1020 /**
1021  * filp_open - open file and return file pointer
1022  *
1023  * @filename:   path to open
1024  * @flags:      open flags as per the open(2) second argument
1025  * @mode:       mode for the new file if O_CREAT is set, else ignored
1026  *
1027  * This is the helper to open a file from kernelspace if you really
1028  * have to.  But in generally you should not do this, so please move
1029  * along, nothing to see here..
1030  */
1031 struct file *filp_open(const char *filename, int flags, umode_t mode)
1032 {
1033         struct filename *name = getname_kernel(filename);
1034         struct file *file = ERR_CAST(name);
1035
1036         if (!IS_ERR(name)) {
1037                 file = file_open_name(name, flags, mode);
1038                 putname(name);
1039         }
1040         return file;
1041 }
1042 EXPORT_SYMBOL(filp_open);
1043
1044 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1045                             const char *filename, int flags, umode_t mode)
1046 {
1047         struct open_flags op;
1048         int err = build_open_flags(flags, mode, &op);
1049         if (err)
1050                 return ERR_PTR(err);
1051         return do_file_open_root(dentry, mnt, filename, &op);
1052 }
1053 EXPORT_SYMBOL(file_open_root);
1054
1055 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1056 {
1057         struct open_flags op;
1058         int fd = build_open_flags(flags, mode, &op);
1059         struct filename *tmp;
1060
1061         if (fd)
1062                 return fd;
1063
1064         tmp = getname(filename);
1065         if (IS_ERR(tmp))
1066                 return PTR_ERR(tmp);
1067
1068         fd = get_unused_fd_flags(flags);
1069         if (fd >= 0) {
1070                 struct file *f = do_filp_open(dfd, tmp, &op);
1071                 if (IS_ERR(f)) {
1072                         put_unused_fd(fd);
1073                         fd = PTR_ERR(f);
1074                 } else {
1075                         fsnotify_open(f);
1076                         fd_install(fd, f);
1077                 }
1078         }
1079         putname(tmp);
1080         return fd;
1081 }
1082
1083 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1084 {
1085         if (force_o_largefile())
1086                 flags |= O_LARGEFILE;
1087
1088         return do_sys_open(AT_FDCWD, filename, flags, mode);
1089 }
1090
1091 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1092                 umode_t, mode)
1093 {
1094         if (force_o_largefile())
1095                 flags |= O_LARGEFILE;
1096
1097         return do_sys_open(dfd, filename, flags, mode);
1098 }
1099
1100 #ifdef CONFIG_COMPAT
1101 /*
1102  * Exactly like sys_open(), except that it doesn't set the
1103  * O_LARGEFILE flag.
1104  */
1105 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1106 {
1107         return do_sys_open(AT_FDCWD, filename, flags, mode);
1108 }
1109
1110 /*
1111  * Exactly like sys_openat(), except that it doesn't set the
1112  * O_LARGEFILE flag.
1113  */
1114 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1115 {
1116         return do_sys_open(dfd, filename, flags, mode);
1117 }
1118 #endif
1119
1120 #ifndef __alpha__
1121
1122 /*
1123  * For backward compatibility?  Maybe this should be moved
1124  * into arch/i386 instead?
1125  */
1126 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1127 {
1128         return ksys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1129 }
1130
1131 #endif
1132
1133 /*
1134  * "id" is the POSIX thread ID. We use the
1135  * files pointer for this..
1136  */
1137 int filp_close(struct file *filp, fl_owner_t id)
1138 {
1139         int retval = 0;
1140
1141         if (!file_count(filp)) {
1142                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1143                 return 0;
1144         }
1145
1146         if (filp->f_op->flush)
1147                 retval = filp->f_op->flush(filp, id);
1148
1149         if (likely(!(filp->f_mode & FMODE_PATH))) {
1150                 dnotify_flush(filp, id);
1151                 locks_remove_posix(filp, id);
1152         }
1153         fput(filp);
1154         return retval;
1155 }
1156
1157 EXPORT_SYMBOL(filp_close);
1158
1159 /*
1160  * Careful here! We test whether the file pointer is NULL before
1161  * releasing the fd. This ensures that one clone task can't release
1162  * an fd while another clone is opening it.
1163  */
1164 SYSCALL_DEFINE1(close, unsigned int, fd)
1165 {
1166         int retval = __close_fd(current->files, fd);
1167
1168         /* can't restart close syscall because file table entry was cleared */
1169         if (unlikely(retval == -ERESTARTSYS ||
1170                      retval == -ERESTARTNOINTR ||
1171                      retval == -ERESTARTNOHAND ||
1172                      retval == -ERESTART_RESTARTBLOCK))
1173                 retval = -EINTR;
1174
1175         return retval;
1176 }
1177
1178 /*
1179  * This routine simulates a hangup on the tty, to arrange that users
1180  * are given clean terminals at login time.
1181  */
1182 SYSCALL_DEFINE0(vhangup)
1183 {
1184         if (capable(CAP_SYS_TTY_CONFIG)) {
1185                 tty_vhangup_self();
1186                 return 0;
1187         }
1188         return -EPERM;
1189 }
1190
1191 /*
1192  * Called when an inode is about to be open.
1193  * We use this to disallow opening large files on 32bit systems if
1194  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1195  * on this flag in sys_open.
1196  */
1197 int generic_file_open(struct inode * inode, struct file * filp)
1198 {
1199         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1200                 return -EOVERFLOW;
1201         return 0;
1202 }
1203
1204 EXPORT_SYMBOL(generic_file_open);
1205
1206 /*
1207  * This is used by subsystems that don't want seekable
1208  * file descriptors. The function is not supposed to ever fail, the only
1209  * reason it returns an 'int' and not 'void' is so that it can be plugged
1210  * directly into file_operations structure.
1211  */
1212 int nonseekable_open(struct inode *inode, struct file *filp)
1213 {
1214         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1215         return 0;
1216 }
1217
1218 EXPORT_SYMBOL(nonseekable_open);
1219
1220 /*
1221  * stream_open is used by subsystems that want stream-like file descriptors.
1222  * Such file descriptors are not seekable and don't have notion of position
1223  * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1224  * Contrary to file descriptors of other regular files, .read() and .write()
1225  * can run simultaneously.
1226  *
1227  * stream_open never fails and is marked to return int so that it could be
1228  * directly used as file_operations.open .
1229  */
1230 int stream_open(struct inode *inode, struct file *filp)
1231 {
1232         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1233         filp->f_mode |= FMODE_STREAM;
1234         return 0;
1235 }
1236
1237 EXPORT_SYMBOL(stream_open);