fs/open.c

   1 /*
   2  *  linux/fs/open.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/string.h>
   8 #include <linux/mm.h>
   9 #include <linux/file.h>
  10 #include <linux/fdtable.h>
  11 #include <linux/fsnotify.h>
  12 #include <linux/module.h>
  13 #include <linux/tty.h>
  14 #include <linux/namei.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/capability.h>
  17 #include <linux/securebits.h>
  18 #include <linux/security.h>
  19 #include <linux/mount.h>
  20 #include <linux/fcntl.h>
  21 #include <linux/slab.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/fs.h>
  24 #include <linux/personality.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/rcupdate.h>
  28 #include <linux/audit.h>
  29 #include <linux/falloc.h>
  30 #include <linux/fs_struct.h>
  31 #include <linux/ima.h>
  32 #include <linux/dnotify.h>
  33 #include <linux/compat.h>
  34
  35 #include "internal.h"
  36
  37 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  38         struct file *filp)
  39 {
  40         int ret;
  41         struct iattr newattrs;
  42
  43         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  44         if (length < 0)
  45                 return -EINVAL;
  46
  47         newattrs.ia_size = length;
  48         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  49         if (filp) {
  50                 newattrs.ia_file = filp;
  51                 newattrs.ia_valid |= ATTR_FILE;
  52         }
  53
  54         /* Remove suid, sgid, and file capabilities on truncate too */
  55         ret = dentry_needs_remove_privs(dentry);
  56         if (ret < 0)
  57                 return ret;
  58         if (ret)
  59                 newattrs.ia_valid |= ret | ATTR_FORCE;
  60
  61         inode_lock(dentry->d_inode);
  62         /* Note any delegations or leases have already been broken: */
  63         ret = notify_change(dentry, &newattrs, NULL);
  64         inode_unlock(dentry->d_inode);
  65         return ret;
  66 }
  67
  68 long vfs_truncate(const struct path *path, loff_t length)
  69 {
  70         struct inode *inode;
  71         struct dentry *upperdentry;
  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         /*
  95          * If this is an overlayfs then do as if opening the file so we get
  96          * write access on the upper inode, not on the overlay inode.  For
  97          * non-overlay filesystems d_real() is an identity function.
  98          */
  99         upperdentry = d_real(path->dentry, NULL, O_WRONLY);
 100         error = PTR_ERR(upperdentry);
 101         if (IS_ERR(upperdentry))
 102                 goto mnt_drop_write_and_out;
 103
 104         error = get_write_access(upperdentry->d_inode);
 105         if (error)
 106                 goto mnt_drop_write_and_out;
 107
 108         /*
 109          * Make sure that there are no leases.  get_write_access() protects
 110          * against the truncate racing with a lease-granting setlease().
 111          */
 112         error = break_lease(inode, O_WRONLY);
 113         if (error)
 114                 goto put_write_and_out;
 115
 116         error = locks_verify_truncate(inode, NULL, length);
 117         if (!error)
 118                 error = security_path_truncate(path);
 119         if (!error)
 120                 error = do_truncate(path->dentry, length, 0, NULL);
 121
 122 put_write_and_out:
 123         put_write_access(upperdentry->d_inode);
 124 mnt_drop_write_and_out:
 125         mnt_drop_write(path->mnt);
 126 out:
 127         return error;
 128 }
 129 EXPORT_SYMBOL_GPL(vfs_truncate);
 130
 131 static long do_sys_truncate(const char __user *pathname, loff_t length)
 132 {
 133         unsigned int lookup_flags = LOOKUP_FOLLOW;
 134         struct path path;
 135         int error;
 136
 137         if (length < 0) /* sorry, but loff_t says... */
 138                 return -EINVAL;
 139
 140 retry:
 141         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 142         if (!error) {
 143                 error = vfs_truncate(&path, length);
 144                 path_put(&path);
 145         }
 146         if (retry_estale(error, lookup_flags)) {
 147                 lookup_flags |= LOOKUP_REVAL;
 148                 goto retry;
 149         }
 150         return error;
 151 }
 152
 153 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 154 {
 155         return do_sys_truncate(path, length);
 156 }
 157
 158 #ifdef CONFIG_COMPAT
 159 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 160 {
 161         return do_sys_truncate(path, length);
 162 }
 163 #endif
 164
 165 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 166 {
 167         struct inode *inode;
 168         struct dentry *dentry;
 169         struct fd f;
 170         int error;
 171
 172         error = -EINVAL;
 173         if (length < 0)
 174                 goto out;
 175         error = -EBADF;
 176         f = fdget(fd);
 177         if (!f.file)
 178                 goto out;
 179
 180         /* explicitly opened as large or we are on 64-bit box */
 181         if (f.file->f_flags & O_LARGEFILE)
 182                 small = 0;
 183
 184         dentry = f.file->f_path.dentry;
 185         inode = dentry->d_inode;
 186         error = -EINVAL;
 187         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 188                 goto out_putf;
 189
 190         error = -EINVAL;
 191         /* Cannot ftruncate over 2^31 bytes without large file support */
 192         if (small && length > MAX_NON_LFS)
 193                 goto out_putf;
 194
 195         error = -EPERM;
 196         /* Check IS_APPEND on real upper inode */
 197         if (IS_APPEND(file_inode(f.file)))
 198                 goto out_putf;
 199
 200         sb_start_write(inode->i_sb);
 201         error = locks_verify_truncate(inode, f.file, length);
 202         if (!error)
 203                 error = security_path_truncate(&f.file->f_path);
 204         if (!error)
 205                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 206         sb_end_write(inode->i_sb);
 207 out_putf:
 208         fdput(f);
 209 out:
 210         return error;
 211 }
 212
 213 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 214 {
 215         return do_sys_ftruncate(fd, length, 1);
 216 }
 217
 218 #ifdef CONFIG_COMPAT
 219 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 220 {
 221         return do_sys_ftruncate(fd, length, 1);
 222 }
 223 #endif
 224
 225 /* LFS versions of truncate are only needed on 32 bit machines */
 226 #if BITS_PER_LONG == 32
 227 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 228 {
 229         return do_sys_truncate(path, length);
 230 }
 231
 232 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 233 {
 234         return do_sys_ftruncate(fd, length, 0);
 235 }
 236 #endif /* BITS_PER_LONG == 32 */
 237
 238
 239 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 240 {
 241         struct inode *inode = file_inode(file);
 242         long ret;
 243
 244         if (offset < 0 || len <= 0)
 245                 return -EINVAL;
 246
 247         /* Return error if mode is not supported */
 248         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 249                 return -EOPNOTSUPP;
 250
 251         /* Punch hole and zero range are mutually exclusive */
 252         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 253             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 254                 return -EOPNOTSUPP;
 255
 256         /* Punch hole must have keep size set */
 257         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 258             !(mode & FALLOC_FL_KEEP_SIZE))
 259                 return -EOPNOTSUPP;
 260
 261         /* Collapse range should only be used exclusively. */
 262         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 263             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 264                 return -EINVAL;
 265
 266         /* Insert range should only be used exclusively. */
 267         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 268             (mode & ~FALLOC_FL_INSERT_RANGE))
 269                 return -EINVAL;
 270
 271         /* Unshare range should only be used with allocate mode. */
 272         if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 273             (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 274                 return -EINVAL;
 275
 276         if (!(file->f_mode & FMODE_WRITE))
 277                 return -EBADF;
 278
 279         /*
 280          * We can only allow pure fallocate on append only files
 281          */
 282         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 283                 return -EPERM;
 284
 285         if (IS_IMMUTABLE(inode))
 286                 return -EPERM;
 287
 288         /*
 289          * We cannot allow any fallocate operation on an active swapfile
 290          */
 291         if (IS_SWAPFILE(inode))
 292                 return -ETXTBSY;
 293
 294         /*
 295          * Revalidate the write permissions, in case security policy has
 296          * changed since the files were opened.
 297          */
 298         ret = security_file_permission(file, MAY_WRITE);
 299         if (ret)
 300                 return ret;
 301
 302         if (S_ISFIFO(inode->i_mode))
 303                 return -ESPIPE;
 304
 305         if (S_ISDIR(inode->i_mode))
 306                 return -EISDIR;
 307
 308         if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 309                 return -ENODEV;
 310
 311         /* Check for wrap through zero too */
 312         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 313                 return -EFBIG;
 314
 315         if (!file->f_op->fallocate)
 316                 return -EOPNOTSUPP;
 317
 318         file_start_write(file);
 319         ret = file->f_op->fallocate(file, mode, offset, len);
 320
 321         /*
 322          * Create inotify and fanotify events.
 323          *
 324          * To keep the logic simple always create events if fallocate succeeds.
 325          * This implies that events are even created if the file size remains
 326          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 327          */
 328         if (ret == 0)
 329                 fsnotify_modify(file);
 330
 331         file_end_write(file);
 332         return ret;
 333 }
 334 EXPORT_SYMBOL_GPL(vfs_fallocate);
 335
 336 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 337 {
 338         struct fd f = fdget(fd);
 339         int error = -EBADF;
 340
 341         if (f.file) {
 342                 error = vfs_fallocate(f.file, mode, offset, len);
 343                 fdput(f);
 344         }
 345         return error;
 346 }
 347
 348 /*
 349  * access() needs to use the real uid/gid, not the effective uid/gid.
 350  * We do this by temporarily clearing all FS-related capabilities and
 351  * switching the fsuid/fsgid around to the real ones.
 352  */
 353 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 354 {
 355         const struct cred *old_cred;
 356         struct cred *override_cred;
 357         struct path path;
 358         struct inode *inode;
 359         int res;
 360         unsigned int lookup_flags = LOOKUP_FOLLOW;
 361
 362         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 363                 return -EINVAL;
 364
 365         override_cred = prepare_creds();
 366         if (!override_cred)
 367                 return -ENOMEM;
 368
 369         override_cred->fsuid = override_cred->uid;
 370         override_cred->fsgid = override_cred->gid;
 371
 372         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 373                 /* Clear the capabilities if we switch to a non-root user */
 374                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 375                 if (!uid_eq(override_cred->uid, root_uid))
 376                         cap_clear(override_cred->cap_effective);
 377                 else
 378                         override_cred->cap_effective =
 379                                 override_cred->cap_permitted;
 380         }
 381
 382         old_cred = override_creds(override_cred);
 383 retry:
 384         res = user_path_at(dfd, filename, lookup_flags, &path);
 385         if (res)
 386                 goto out;
 387
 388         inode = d_backing_inode(path.dentry);
 389
 390         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 391                 /*
 392                  * MAY_EXEC on regular files is denied if the fs is mounted
 393                  * with the "noexec" flag.
 394                  */
 395                 res = -EACCES;
 396                 if (path_noexec(&path))
 397                         goto out_path_release;
 398         }
 399
 400         res = inode_permission(inode, mode | MAY_ACCESS);
 401         /* SuS v2 requires we report a read only fs too */
 402         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 403                 goto out_path_release;
 404         /*
 405          * This is a rare case where using __mnt_is_readonly()
 406          * is OK without a mnt_want/drop_write() pair.  Since
 407          * no actual write to the fs is performed here, we do
 408          * not need to telegraph to that to anyone.
 409          *
 410          * By doing this, we accept that this access is
 411          * inherently racy and know that the fs may change
 412          * state before we even see this result.
 413          */
 414         if (__mnt_is_readonly(path.mnt))
 415                 res = -EROFS;
 416
 417 out_path_release:
 418         path_put(&path);
 419         if (retry_estale(res, lookup_flags)) {
 420                 lookup_flags |= LOOKUP_REVAL;
 421                 goto retry;
 422         }
 423 out:
 424         revert_creds(old_cred);
 425         put_cred(override_cred);
 426         return res;
 427 }
 428
 429 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 430 {
 431         return sys_faccessat(AT_FDCWD, filename, mode);
 432 }
 433
 434 SYSCALL_DEFINE1(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(fchdir, unsigned int, fd)
 461 {
 462         struct fd f = fdget_raw(fd);
 463         int error;
 464
 465         error = -EBADF;
 466         if (!f.file)
 467                 goto out;
 468
 469         error = -ENOTDIR;
 470         if (!d_can_lookup(f.file->f_path.dentry))
 471                 goto out_putf;
 472
 473         error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 474         if (!error)
 475                 set_fs_pwd(current->fs, &f.file->f_path);
 476 out_putf:
 477         fdput(f);
 478 out:
 479         return error;
 480 }
 481
 482 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 483 {
 484         struct path path;
 485         int error;
 486         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 487 retry:
 488         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 489         if (error)
 490                 goto out;
 491
 492         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 493         if (error)
 494                 goto dput_and_out;
 495
 496         error = -EPERM;
 497         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 498                 goto dput_and_out;
 499         error = security_path_chroot(&path);
 500         if (error)
 501                 goto dput_and_out;
 502
 503         set_fs_root(current->fs, &path);
 504         error = 0;
 505 dput_and_out:
 506         path_put(&path);
 507         if (retry_estale(error, lookup_flags)) {
 508                 lookup_flags |= LOOKUP_REVAL;
 509                 goto retry;
 510         }
 511 out:
 512         return error;
 513 }
 514
 515 static int chmod_common(const struct path *path, umode_t mode)
 516 {
 517         struct inode *inode = path->dentry->d_inode;
 518         struct inode *delegated_inode = NULL;
 519         struct iattr newattrs;
 520         int error;
 521
 522         error = mnt_want_write(path->mnt);
 523         if (error)
 524                 return error;
 525 retry_deleg:
 526         inode_lock(inode);
 527         error = security_path_chmod(path, mode);
 528         if (error)
 529                 goto out_unlock;
 530         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 531         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 532         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 533 out_unlock:
 534         inode_unlock(inode);
 535         if (delegated_inode) {
 536                 error = break_deleg_wait(&delegated_inode);
 537                 if (!error)
 538                         goto retry_deleg;
 539         }
 540         mnt_drop_write(path->mnt);
 541         return error;
 542 }
 543
 544 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 545 {
 546         struct fd f = fdget(fd);
 547         int err = -EBADF;
 548
 549         if (f.file) {
 550                 audit_file(f.file);
 551                 err = chmod_common(&f.file->f_path, mode);
 552                 fdput(f);
 553         }
 554         return err;
 555 }
 556
 557 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 558 {
 559         struct path path;
 560         int error;
 561         unsigned int lookup_flags = LOOKUP_FOLLOW;
 562 retry:
 563         error = user_path_at(dfd, filename, lookup_flags, &path);
 564         if (!error) {
 565                 error = chmod_common(&path, mode);
 566                 path_put(&path);
 567                 if (retry_estale(error, lookup_flags)) {
 568                         lookup_flags |= LOOKUP_REVAL;
 569                         goto retry;
 570                 }
 571         }
 572         return error;
 573 }
 574
 575 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 576 {
 577         return sys_fchmodat(AT_FDCWD, filename, mode);
 578 }
 579
 580 static int chown_common(const struct path *path, uid_t user, gid_t group)
 581 {
 582         struct inode *inode = path->dentry->d_inode;
 583         struct inode *delegated_inode = NULL;
 584         int error;
 585         struct iattr newattrs;
 586         kuid_t uid;
 587         kgid_t gid;
 588
 589         uid = make_kuid(current_user_ns(), user);
 590         gid = make_kgid(current_user_ns(), group);
 591
 592 retry_deleg:
 593         newattrs.ia_valid =  ATTR_CTIME;
 594         if (user != (uid_t) -1) {
 595                 if (!uid_valid(uid))
 596                         return -EINVAL;
 597                 newattrs.ia_valid |= ATTR_UID;
 598                 newattrs.ia_uid = uid;
 599         }
 600         if (group != (gid_t) -1) {
 601                 if (!gid_valid(gid))
 602                         return -EINVAL;
 603                 newattrs.ia_valid |= ATTR_GID;
 604                 newattrs.ia_gid = gid;
 605         }
 606         if (!S_ISDIR(inode->i_mode))
 607                 newattrs.ia_valid |=
 608                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 609         inode_lock(inode);
 610         error = security_path_chown(path, uid, gid);
 611         if (!error)
 612                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 613         inode_unlock(inode);
 614         if (delegated_inode) {
 615                 error = break_deleg_wait(&delegated_inode);
 616                 if (!error)
 617                         goto retry_deleg;
 618         }
 619         return error;
 620 }
 621
 622 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 623                 gid_t, group, int, flag)
 624 {
 625         struct path path;
 626         int error = -EINVAL;
 627         int lookup_flags;
 628
 629         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 630                 goto out;
 631
 632         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 633         if (flag & AT_EMPTY_PATH)
 634                 lookup_flags |= LOOKUP_EMPTY;
 635 retry:
 636         error = user_path_at(dfd, filename, lookup_flags, &path);
 637         if (error)
 638                 goto out;
 639         error = mnt_want_write(path.mnt);
 640         if (error)
 641                 goto out_release;
 642         error = chown_common(&path, user, group);
 643         mnt_drop_write(path.mnt);
 644 out_release:
 645         path_put(&path);
 646         if (retry_estale(error, lookup_flags)) {
 647                 lookup_flags |= LOOKUP_REVAL;
 648                 goto retry;
 649         }
 650 out:
 651         return error;
 652 }
 653
 654 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 655 {
 656         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 657 }
 658
 659 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 660 {
 661         return sys_fchownat(AT_FDCWD, filename, user, group,
 662                             AT_SYMLINK_NOFOLLOW);
 663 }
 664
 665 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 666 {
 667         struct fd f = fdget(fd);
 668         int error = -EBADF;
 669
 670         if (!f.file)
 671                 goto out;
 672
 673         error = mnt_want_write_file(f.file);
 674         if (error)
 675                 goto out_fput;
 676         audit_file(f.file);
 677         error = chown_common(&f.file->f_path, user, group);
 678         mnt_drop_write_file(f.file);
 679 out_fput:
 680         fdput(f);
 681 out:
 682         return error;
 683 }
 684
 685 int open_check_o_direct(struct file *f)
 686 {
 687         /* NB: we're sure to have correct a_ops only after f_op->open */
 688         if (f->f_flags & O_DIRECT) {
 689                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 690                         return -EINVAL;
 691         }
 692         return 0;
 693 }
 694
 695 static int do_dentry_open(struct file *f,
 696                           struct inode *inode,
 697                           int (*open)(struct inode *, struct file *),
 698                           const struct cred *cred)
 699 {
 700         static const struct file_operations empty_fops = {};
 701         int error;
 702
 703         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 704                                 FMODE_PREAD | FMODE_PWRITE;
 705
 706         path_get(&f->f_path);
 707         f->f_inode = inode;
 708         f->f_mapping = inode->i_mapping;
 709
 710         /* Ensure that we skip any errors that predate opening of the file */
 711         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 712
 713         if (unlikely(f->f_flags & O_PATH)) {
 714                 f->f_mode = FMODE_PATH;
 715                 f->f_op = &empty_fops;
 716                 return 0;
 717         }
 718
 719         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 720                 error = get_write_access(inode);
 721                 if (unlikely(error))
 722                         goto cleanup_file;
 723                 error = __mnt_want_write(f->f_path.mnt);
 724                 if (unlikely(error)) {
 725                         put_write_access(inode);
 726                         goto cleanup_file;
 727                 }
 728                 f->f_mode |= FMODE_WRITER;
 729         }
 730
 731         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 732         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 733                 f->f_mode |= FMODE_ATOMIC_POS;
 734
 735         f->f_op = fops_get(inode->i_fop);
 736         if (unlikely(WARN_ON(!f->f_op))) {
 737                 error = -ENODEV;
 738                 goto cleanup_all;
 739         }
 740
 741         error = security_file_open(f, cred);
 742         if (error)
 743                 goto cleanup_all;
 744
 745         error = break_lease(locks_inode(f), f->f_flags);
 746         if (error)
 747                 goto cleanup_all;
 748
 749         if (!open)
 750                 open = f->f_op->open;
 751         if (open) {
 752                 error = open(inode, f);
 753                 if (error)
 754                         goto cleanup_all;
 755         }
 756         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 757                 i_readcount_inc(inode);
 758         if ((f->f_mode & FMODE_READ) &&
 759              likely(f->f_op->read || f->f_op->read_iter))
 760                 f->f_mode |= FMODE_CAN_READ;
 761         if ((f->f_mode & FMODE_WRITE) &&
 762              likely(f->f_op->write || f->f_op->write_iter))
 763                 f->f_mode |= FMODE_CAN_WRITE;
 764
 765         f->f_write_hint = WRITE_LIFE_NOT_SET;
 766         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 767
 768         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 769
 770         return 0;
 771
 772 cleanup_all:
 773         fops_put(f->f_op);
 774         if (f->f_mode & FMODE_WRITER) {
 775                 put_write_access(inode);
 776                 __mnt_drop_write(f->f_path.mnt);
 777         }
 778 cleanup_file:
 779         path_put(&f->f_path);
 780         f->f_path.mnt = NULL;
 781         f->f_path.dentry = NULL;
 782         f->f_inode = NULL;
 783         return error;
 784 }
 785
 786 /**
 787  * finish_open - finish opening a file
 788  * @file: file pointer
 789  * @dentry: pointer to dentry
 790  * @open: open callback
 791  * @opened: state of open
 792  *
 793  * This can be used to finish opening a file passed to i_op->atomic_open().
 794  *
 795  * If the open callback is set to NULL, then the standard f_op->open()
 796  * filesystem callback is substituted.
 797  *
 798  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 799  * the return value of d_splice_alias(), then the caller needs to perform dput()
 800  * on it after finish_open().
 801  *
 802  * On successful return @file is a fully instantiated open file.  After this, if
 803  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 804  *
 805  * Returns zero on success or -errno if the open failed.
 806  */
 807 int finish_open(struct file *file, struct dentry *dentry,
 808                 int (*open)(struct inode *, struct file *),
 809                 int *opened)
 810 {
 811         int error;
 812         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 813
 814         file->f_path.dentry = dentry;
 815         error = do_dentry_open(file, d_backing_inode(dentry), open,
 816                                current_cred());
 817         if (!error)
 818                 *opened |= FILE_OPENED;
 819
 820         return error;
 821 }
 822 EXPORT_SYMBOL(finish_open);
 823
 824 /**
 825  * finish_no_open - finish ->atomic_open() without opening the file
 826  *
 827  * @file: file pointer
 828  * @dentry: dentry or NULL (as returned from ->lookup())
 829  *
 830  * This can be used to set the result of a successful lookup in ->atomic_open().
 831  *
 832  * NB: unlike finish_open() this function does consume the dentry reference and
 833  * the caller need not dput() it.
 834  *
 835  * Returns "1" which must be the return value of ->atomic_open() after having
 836  * called this function.
 837  */
 838 int finish_no_open(struct file *file, struct dentry *dentry)
 839 {
 840         file->f_path.dentry = dentry;
 841         return 1;
 842 }
 843 EXPORT_SYMBOL(finish_no_open);
 844
 845 char *file_path(struct file *filp, char *buf, int buflen)
 846 {
 847         return d_path(&filp->f_path, buf, buflen);
 848 }
 849 EXPORT_SYMBOL(file_path);
 850
 851 /**
 852  * vfs_open - open the file at the given path
 853  * @path: path to open
 854  * @file: newly allocated file with f_flag initialized
 855  * @cred: credentials to use
 856  */
 857 int vfs_open(const struct path *path, struct file *file,
 858              const struct cred *cred)
 859 {
 860         struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags);
 861
 862         if (IS_ERR(dentry))
 863                 return PTR_ERR(dentry);
 864
 865         file->f_path = *path;
 866         return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 867 }
 868
 869 struct file *dentry_open(const struct path *path, int flags,
 870                          const struct cred *cred)
 871 {
 872         int error;
 873         struct file *f;
 874
 875         validate_creds(cred);
 876
 877         /* We must always pass in a valid mount pointer. */
 878         BUG_ON(!path->mnt);
 879
 880         f = get_empty_filp();
 881         if (!IS_ERR(f)) {
 882                 f->f_flags = flags;
 883                 error = vfs_open(path, f, cred);
 884                 if (!error) {
 885                         /* from now on we need fput() to dispose of f */
 886                         error = open_check_o_direct(f);
 887                         if (error) {
 888                                 fput(f);
 889                                 f = ERR_PTR(error);
 890                         }
 891                 } else {
 892                         put_filp(f);
 893                         f = ERR_PTR(error);
 894                 }
 895         }
 896         return f;
 897 }
 898 EXPORT_SYMBOL(dentry_open);
 899
 900 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 901 {
 902         int lookup_flags = 0;
 903         int acc_mode = ACC_MODE(flags);
 904
 905         /*
 906          * Clear out all open flags we don't know about so that we don't report
 907          * them in fcntl(F_GETFD) or similar interfaces.
 908          */
 909         flags &= VALID_OPEN_FLAGS;
 910
 911         if (flags & (O_CREAT | __O_TMPFILE))
 912                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 913         else
 914                 op->mode = 0;
 915
 916         /* Must never be set by userspace */
 917         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 918
 919         /*
 920          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 921          * check for O_DSYNC if the need any syncing at all we enforce it's
 922          * always set instead of having to deal with possibly weird behaviour
 923          * for malicious applications setting only __O_SYNC.
 924          */
 925         if (flags & __O_SYNC)
 926                 flags |= O_DSYNC;
 927
 928         if (flags & __O_TMPFILE) {
 929                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 930                         return -EINVAL;
 931                 if (!(acc_mode & MAY_WRITE))
 932                         return -EINVAL;
 933         } else if (flags & O_PATH) {
 934                 /*
 935                  * If we have O_PATH in the open flag. Then we
 936                  * cannot have anything other than the below set of flags
 937                  */
 938                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 939                 acc_mode = 0;
 940         }
 941
 942         op->open_flag = flags;
 943
 944         /* O_TRUNC implies we need access checks for write permissions */
 945         if (flags & O_TRUNC)
 946                 acc_mode |= MAY_WRITE;
 947
 948         /* Allow the LSM permission hook to distinguish append
 949            access from general write access. */
 950         if (flags & O_APPEND)
 951                 acc_mode |= MAY_APPEND;
 952
 953         op->acc_mode = acc_mode;
 954
 955         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 956
 957         if (flags & O_CREAT) {
 958                 op->intent |= LOOKUP_CREATE;
 959                 if (flags & O_EXCL)
 960                         op->intent |= LOOKUP_EXCL;
 961         }
 962
 963         if (flags & O_DIRECTORY)
 964                 lookup_flags |= LOOKUP_DIRECTORY;
 965         if (!(flags & O_NOFOLLOW))
 966                 lookup_flags |= LOOKUP_FOLLOW;
 967         op->lookup_flags = lookup_flags;
 968         return 0;
 969 }
 970
 971 /**
 972  * file_open_name - open file and return file pointer
 973  *
 974  * @name:       struct filename containing path to open
 975  * @flags:      open flags as per the open(2) second argument
 976  * @mode:       mode for the new file if O_CREAT is set, else ignored
 977  *
 978  * This is the helper to open a file from kernelspace if you really
 979  * have to.  But in generally you should not do this, so please move
 980  * along, nothing to see here..
 981  */
 982 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 983 {
 984         struct open_flags op;
 985         int err = build_open_flags(flags, mode, &op);
 986         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 987 }
 988
 989 /**
 990  * filp_open - open file and return file pointer
 991  *
 992  * @filename:   path to open
 993  * @flags:      open flags as per the open(2) second argument
 994  * @mode:       mode for the new file if O_CREAT is set, else ignored
 995  *
 996  * This is the helper to open a file from kernelspace if you really
 997  * have to.  But in generally you should not do this, so please move
 998  * along, nothing to see here..
 999  */
1000 struct file *filp_open(const char *filename, int flags, umode_t mode)
1001 {
1002         struct filename *name = getname_kernel(filename);
1003         struct file *file = ERR_CAST(name);
1004
1005         if (!IS_ERR(name)) {
1006                 file = file_open_name(name, flags, mode);
1007                 putname(name);
1008         }
1009         return file;
1010 }
1011 EXPORT_SYMBOL(filp_open);
1012
1013 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1014                             const char *filename, int flags, umode_t mode)
1015 {
1016         struct open_flags op;
1017         int err = build_open_flags(flags, mode, &op);
1018         if (err)
1019                 return ERR_PTR(err);
1020         return do_file_open_root(dentry, mnt, filename, &op);
1021 }
1022 EXPORT_SYMBOL(file_open_root);
1023
1024 struct file *filp_clone_open(struct file *oldfile)
1025 {
1026         struct file *file;
1027         int retval;
1028
1029         file = get_empty_filp();
1030         if (IS_ERR(file))
1031                 return file;
1032
1033         file->f_flags = oldfile->f_flags;
1034         retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1035         if (retval) {
1036                 put_filp(file);
1037                 return ERR_PTR(retval);
1038         }
1039
1040         return file;
1041 }
1042 EXPORT_SYMBOL(filp_clone_open);
1043
1044 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1045 {
1046         struct open_flags op;
1047         int fd = build_open_flags(flags, mode, &op);
1048         struct filename *tmp;
1049
1050         if (fd)
1051                 return fd;
1052
1053         tmp = getname(filename);
1054         if (IS_ERR(tmp))
1055                 return PTR_ERR(tmp);
1056
1057         fd = get_unused_fd_flags(flags);
1058         if (fd >= 0) {
1059                 struct file *f = do_filp_open(dfd, tmp, &op);
1060                 if (IS_ERR(f)) {
1061                         put_unused_fd(fd);
1062                         fd = PTR_ERR(f);
1063                 } else {
1064                         fsnotify_open(f);
1065                         fd_install(fd, f);
1066                 }
1067         }
1068         putname(tmp);
1069         return fd;
1070 }
1071
1072 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1073 {
1074         if (force_o_largefile())
1075                 flags |= O_LARGEFILE;
1076
1077         return do_sys_open(AT_FDCWD, filename, flags, mode);
1078 }
1079
1080 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1081                 umode_t, mode)
1082 {
1083         if (force_o_largefile())
1084                 flags |= O_LARGEFILE;
1085
1086         return do_sys_open(dfd, filename, flags, mode);
1087 }
1088
1089 #ifdef CONFIG_COMPAT
1090 /*
1091  * Exactly like sys_open(), except that it doesn't set the
1092  * O_LARGEFILE flag.
1093  */
1094 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1095 {
1096         return do_sys_open(AT_FDCWD, filename, flags, mode);
1097 }
1098
1099 /*
1100  * Exactly like sys_openat(), except that it doesn't set the
1101  * O_LARGEFILE flag.
1102  */
1103 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1104 {
1105         return do_sys_open(dfd, filename, flags, mode);
1106 }
1107 #endif
1108
1109 #ifndef __alpha__
1110
1111 /*
1112  * For backward compatibility?  Maybe this should be moved
1113  * into arch/i386 instead?
1114  */
1115 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1116 {
1117         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1118 }
1119
1120 #endif
1121
1122 /*
1123  * "id" is the POSIX thread ID. We use the
1124  * files pointer for this..
1125  */
1126 int filp_close(struct file *filp, fl_owner_t id)
1127 {
1128         int retval = 0;
1129
1130         if (!file_count(filp)) {
1131                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1132                 return 0;
1133         }
1134
1135         if (filp->f_op->flush)
1136                 retval = filp->f_op->flush(filp, id);
1137
1138         if (likely(!(filp->f_mode & FMODE_PATH))) {
1139                 dnotify_flush(filp, id);
1140                 locks_remove_posix(filp, id);
1141         }
1142         fput(filp);
1143         return retval;
1144 }
1145
1146 EXPORT_SYMBOL(filp_close);
1147
1148 /*
1149  * Careful here! We test whether the file pointer is NULL before
1150  * releasing the fd. This ensures that one clone task can't release
1151  * an fd while another clone is opening it.
1152  */
1153 SYSCALL_DEFINE1(close, unsigned int, fd)
1154 {
1155         int retval = __close_fd(current->files, fd);
1156
1157         /* can't restart close syscall because file table entry was cleared */
1158         if (unlikely(retval == -ERESTARTSYS ||
1159                      retval == -ERESTARTNOINTR ||
1160                      retval == -ERESTARTNOHAND ||
1161                      retval == -ERESTART_RESTARTBLOCK))
1162                 retval = -EINTR;
1163
1164         return retval;
1165 }
1166 EXPORT_SYMBOL(sys_close);
1167
1168 /*
1169  * This routine simulates a hangup on the tty, to arrange that users
1170  * are given clean terminals at login time.
1171  */
1172 SYSCALL_DEFINE0(vhangup)
1173 {
1174         if (capable(CAP_SYS_TTY_CONFIG)) {
1175                 tty_vhangup_self();
1176                 return 0;
1177         }
1178         return -EPERM;
1179 }
1180
1181 /*
1182  * Called when an inode is about to be open.
1183  * We use this to disallow opening large files on 32bit systems if
1184  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1185  * on this flag in sys_open.
1186  */
1187 int generic_file_open(struct inode * inode, struct file * filp)
1188 {
1189         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1190                 return -EOVERFLOW;
1191         return 0;
1192 }
1193
1194 EXPORT_SYMBOL(generic_file_open);
1195
1196 /*
1197  * This is used by subsystems that don't want seekable
1198  * file descriptors. The function is not supposed to ever fail, the only
1199  * reason it returns an 'int' and not 'void' is so that it can be plugged
1200  * directly into file_operations structure.
1201  */
1202 int nonseekable_open(struct inode *inode, struct file *filp)
1203 {
1204         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1205         return 0;
1206 }
1207
1208 EXPORT_SYMBOL(nonseekable_open);