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         struct inode *inode;
 464         int error = -EBADF;
 465
 466         error = -EBADF;
 467         if (!f.file)
 468                 goto out;
 469
 470         inode = file_inode(f.file);
 471
 472         error = -ENOTDIR;
 473         if (!S_ISDIR(inode->i_mode))
 474                 goto out_putf;
 475
 476         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 477         if (!error)
 478                 set_fs_pwd(current->fs, &f.file->f_path);
 479 out_putf:
 480         fdput(f);
 481 out:
 482         return error;
 483 }
 484
 485 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 486 {
 487         struct path path;
 488         int error;
 489         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 490 retry:
 491         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 492         if (error)
 493                 goto out;
 494
 495         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 496         if (error)
 497                 goto dput_and_out;
 498
 499         error = -EPERM;
 500         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 501                 goto dput_and_out;
 502         error = security_path_chroot(&path);
 503         if (error)
 504                 goto dput_and_out;
 505
 506         set_fs_root(current->fs, &path);
 507         error = 0;
 508 dput_and_out:
 509         path_put(&path);
 510         if (retry_estale(error, lookup_flags)) {
 511                 lookup_flags |= LOOKUP_REVAL;
 512                 goto retry;
 513         }
 514 out:
 515         return error;
 516 }
 517
 518 static int chmod_common(const struct path *path, umode_t mode)
 519 {
 520         struct inode *inode = path->dentry->d_inode;
 521         struct inode *delegated_inode = NULL;
 522         struct iattr newattrs;
 523         int error;
 524
 525         error = mnt_want_write(path->mnt);
 526         if (error)
 527                 return error;
 528 retry_deleg:
 529         inode_lock(inode);
 530         error = security_path_chmod(path, mode);
 531         if (error)
 532                 goto out_unlock;
 533         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 534         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 535         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 536 out_unlock:
 537         inode_unlock(inode);
 538         if (delegated_inode) {
 539                 error = break_deleg_wait(&delegated_inode);
 540                 if (!error)
 541                         goto retry_deleg;
 542         }
 543         mnt_drop_write(path->mnt);
 544         return error;
 545 }
 546
 547 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 548 {
 549         struct fd f = fdget(fd);
 550         int err = -EBADF;
 551
 552         if (f.file) {
 553                 audit_file(f.file);
 554                 err = chmod_common(&f.file->f_path, mode);
 555                 fdput(f);
 556         }
 557         return err;
 558 }
 559
 560 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 561 {
 562         struct path path;
 563         int error;
 564         unsigned int lookup_flags = LOOKUP_FOLLOW;
 565 retry:
 566         error = user_path_at(dfd, filename, lookup_flags, &path);
 567         if (!error) {
 568                 error = chmod_common(&path, mode);
 569                 path_put(&path);
 570                 if (retry_estale(error, lookup_flags)) {
 571                         lookup_flags |= LOOKUP_REVAL;
 572                         goto retry;
 573                 }
 574         }
 575         return error;
 576 }
 577
 578 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 579 {
 580         return sys_fchmodat(AT_FDCWD, filename, mode);
 581 }
 582
 583 static int chown_common(const struct path *path, uid_t user, gid_t group)
 584 {
 585         struct inode *inode = path->dentry->d_inode;
 586         struct inode *delegated_inode = NULL;
 587         int error;
 588         struct iattr newattrs;
 589         kuid_t uid;
 590         kgid_t gid;
 591
 592         uid = make_kuid(current_user_ns(), user);
 593         gid = make_kgid(current_user_ns(), group);
 594
 595 retry_deleg:
 596         newattrs.ia_valid =  ATTR_CTIME;
 597         if (user != (uid_t) -1) {
 598                 if (!uid_valid(uid))
 599                         return -EINVAL;
 600                 newattrs.ia_valid |= ATTR_UID;
 601                 newattrs.ia_uid = uid;
 602         }
 603         if (group != (gid_t) -1) {
 604                 if (!gid_valid(gid))
 605                         return -EINVAL;
 606                 newattrs.ia_valid |= ATTR_GID;
 607                 newattrs.ia_gid = gid;
 608         }
 609         if (!S_ISDIR(inode->i_mode))
 610                 newattrs.ia_valid |=
 611                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 612         inode_lock(inode);
 613         error = security_path_chown(path, uid, gid);
 614         if (!error)
 615                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 616         inode_unlock(inode);
 617         if (delegated_inode) {
 618                 error = break_deleg_wait(&delegated_inode);
 619                 if (!error)
 620                         goto retry_deleg;
 621         }
 622         return error;
 623 }
 624
 625 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 626                 gid_t, group, int, flag)
 627 {
 628         struct path path;
 629         int error = -EINVAL;
 630         int lookup_flags;
 631
 632         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 633                 goto out;
 634
 635         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 636         if (flag & AT_EMPTY_PATH)
 637                 lookup_flags |= LOOKUP_EMPTY;
 638 retry:
 639         error = user_path_at(dfd, filename, lookup_flags, &path);
 640         if (error)
 641                 goto out;
 642         error = mnt_want_write(path.mnt);
 643         if (error)
 644                 goto out_release;
 645         error = chown_common(&path, user, group);
 646         mnt_drop_write(path.mnt);
 647 out_release:
 648         path_put(&path);
 649         if (retry_estale(error, lookup_flags)) {
 650                 lookup_flags |= LOOKUP_REVAL;
 651                 goto retry;
 652         }
 653 out:
 654         return error;
 655 }
 656
 657 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 658 {
 659         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 660 }
 661
 662 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 663 {
 664         return sys_fchownat(AT_FDCWD, filename, user, group,
 665                             AT_SYMLINK_NOFOLLOW);
 666 }
 667
 668 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 669 {
 670         struct fd f = fdget(fd);
 671         int error = -EBADF;
 672
 673         if (!f.file)
 674                 goto out;
 675
 676         error = mnt_want_write_file(f.file);
 677         if (error)
 678                 goto out_fput;
 679         audit_file(f.file);
 680         error = chown_common(&f.file->f_path, user, group);
 681         mnt_drop_write_file(f.file);
 682 out_fput:
 683         fdput(f);
 684 out:
 685         return error;
 686 }
 687
 688 int open_check_o_direct(struct file *f)
 689 {
 690         /* NB: we're sure to have correct a_ops only after f_op->open */
 691         if (f->f_flags & O_DIRECT) {
 692                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 693                         return -EINVAL;
 694         }
 695         return 0;
 696 }
 697
 698 static int do_dentry_open(struct file *f,
 699                           struct inode *inode,
 700                           int (*open)(struct inode *, struct file *),
 701                           const struct cred *cred)
 702 {
 703         static const struct file_operations empty_fops = {};
 704         int error;
 705
 706         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 707                                 FMODE_PREAD | FMODE_PWRITE;
 708
 709         path_get(&f->f_path);
 710         f->f_inode = inode;
 711         f->f_mapping = inode->i_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_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 766
 767         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 768
 769         return 0;
 770
 771 cleanup_all:
 772         fops_put(f->f_op);
 773         if (f->f_mode & FMODE_WRITER) {
 774                 put_write_access(inode);
 775                 __mnt_drop_write(f->f_path.mnt);
 776         }
 777 cleanup_file:
 778         path_put(&f->f_path);
 779         f->f_path.mnt = NULL;
 780         f->f_path.dentry = NULL;
 781         f->f_inode = NULL;
 782         return error;
 783 }
 784
 785 /**
 786  * finish_open - finish opening a file
 787  * @file: file pointer
 788  * @dentry: pointer to dentry
 789  * @open: open callback
 790  * @opened: state of open
 791  *
 792  * This can be used to finish opening a file passed to i_op->atomic_open().
 793  *
 794  * If the open callback is set to NULL, then the standard f_op->open()
 795  * filesystem callback is substituted.
 796  *
 797  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 798  * the return value of d_splice_alias(), then the caller needs to perform dput()
 799  * on it after finish_open().
 800  *
 801  * On successful return @file is a fully instantiated open file.  After this, if
 802  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 803  *
 804  * Returns zero on success or -errno if the open failed.
 805  */
 806 int finish_open(struct file *file, struct dentry *dentry,
 807                 int (*open)(struct inode *, struct file *),
 808                 int *opened)
 809 {
 810         int error;
 811         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 812
 813         file->f_path.dentry = dentry;
 814         error = do_dentry_open(file, d_backing_inode(dentry), open,
 815                                current_cred());
 816         if (!error)
 817                 *opened |= FILE_OPENED;
 818
 819         return error;
 820 }
 821 EXPORT_SYMBOL(finish_open);
 822
 823 /**
 824  * finish_no_open - finish ->atomic_open() without opening the file
 825  *
 826  * @file: file pointer
 827  * @dentry: dentry or NULL (as returned from ->lookup())
 828  *
 829  * This can be used to set the result of a successful lookup in ->atomic_open().
 830  *
 831  * NB: unlike finish_open() this function does consume the dentry reference and
 832  * the caller need not dput() it.
 833  *
 834  * Returns "1" which must be the return value of ->atomic_open() after having
 835  * called this function.
 836  */
 837 int finish_no_open(struct file *file, struct dentry *dentry)
 838 {
 839         file->f_path.dentry = dentry;
 840         return 1;
 841 }
 842 EXPORT_SYMBOL(finish_no_open);
 843
 844 char *file_path(struct file *filp, char *buf, int buflen)
 845 {
 846         return d_path(&filp->f_path, buf, buflen);
 847 }
 848 EXPORT_SYMBOL(file_path);
 849
 850 /**
 851  * vfs_open - open the file at the given path
 852  * @path: path to open
 853  * @file: newly allocated file with f_flag initialized
 854  * @cred: credentials to use
 855  */
 856 int vfs_open(const struct path *path, struct file *file,
 857              const struct cred *cred)
 858 {
 859         struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags);
 860
 861         if (IS_ERR(dentry))
 862                 return PTR_ERR(dentry);
 863
 864         file->f_path = *path;
 865         return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 866 }
 867
 868 struct file *dentry_open(const struct path *path, int flags,
 869                          const struct cred *cred)
 870 {
 871         int error;
 872         struct file *f;
 873
 874         validate_creds(cred);
 875
 876         /* We must always pass in a valid mount pointer. */
 877         BUG_ON(!path->mnt);
 878
 879         f = get_empty_filp();
 880         if (!IS_ERR(f)) {
 881                 f->f_flags = flags;
 882                 error = vfs_open(path, f, cred);
 883                 if (!error) {
 884                         /* from now on we need fput() to dispose of f */
 885                         error = open_check_o_direct(f);
 886                         if (error) {
 887                                 fput(f);
 888                                 f = ERR_PTR(error);
 889                         }
 890                 } else {
 891                         put_filp(f);
 892                         f = ERR_PTR(error);
 893                 }
 894         }
 895         return f;
 896 }
 897 EXPORT_SYMBOL(dentry_open);
 898
 899 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 900 {
 901         int lookup_flags = 0;
 902         int acc_mode = ACC_MODE(flags);
 903
 904         /*
 905          * Clear out all open flags we don't know about so that we don't report
 906          * them in fcntl(F_GETFD) or similar interfaces.
 907          */
 908         flags &= VALID_OPEN_FLAGS;
 909
 910         if (flags & (O_CREAT | __O_TMPFILE))
 911                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 912         else
 913                 op->mode = 0;
 914
 915         /* Must never be set by userspace */
 916         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 917
 918         /*
 919          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 920          * check for O_DSYNC if the need any syncing at all we enforce it's
 921          * always set instead of having to deal with possibly weird behaviour
 922          * for malicious applications setting only __O_SYNC.
 923          */
 924         if (flags & __O_SYNC)
 925                 flags |= O_DSYNC;
 926
 927         if (flags & __O_TMPFILE) {
 928                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 929                         return -EINVAL;
 930                 if (!(acc_mode & MAY_WRITE))
 931                         return -EINVAL;
 932         } else if (flags & O_PATH) {
 933                 /*
 934                  * If we have O_PATH in the open flag. Then we
 935                  * cannot have anything other than the below set of flags
 936                  */
 937                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 938                 acc_mode = 0;
 939         }
 940
 941         op->open_flag = flags;
 942
 943         /* O_TRUNC implies we need access checks for write permissions */
 944         if (flags & O_TRUNC)
 945                 acc_mode |= MAY_WRITE;
 946
 947         /* Allow the LSM permission hook to distinguish append
 948            access from general write access. */
 949         if (flags & O_APPEND)
 950                 acc_mode |= MAY_APPEND;
 951
 952         op->acc_mode = acc_mode;
 953
 954         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 955
 956         if (flags & O_CREAT) {
 957                 op->intent |= LOOKUP_CREATE;
 958                 if (flags & O_EXCL)
 959                         op->intent |= LOOKUP_EXCL;
 960         }
 961
 962         if (flags & O_DIRECTORY)
 963                 lookup_flags |= LOOKUP_DIRECTORY;
 964         if (!(flags & O_NOFOLLOW))
 965                 lookup_flags |= LOOKUP_FOLLOW;
 966         op->lookup_flags = lookup_flags;
 967         return 0;
 968 }
 969
 970 /**
 971  * file_open_name - open file and return file pointer
 972  *
 973  * @name:       struct filename containing path to open
 974  * @flags:      open flags as per the open(2) second argument
 975  * @mode:       mode for the new file if O_CREAT is set, else ignored
 976  *
 977  * This is the helper to open a file from kernelspace if you really
 978  * have to.  But in generally you should not do this, so please move
 979  * along, nothing to see here..
 980  */
 981 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 982 {
 983         struct open_flags op;
 984         int err = build_open_flags(flags, mode, &op);
 985         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 986 }
 987
 988 /**
 989  * filp_open - open file and return file pointer
 990  *
 991  * @filename:   path to open
 992  * @flags:      open flags as per the open(2) second argument
 993  * @mode:       mode for the new file if O_CREAT is set, else ignored
 994  *
 995  * This is the helper to open a file from kernelspace if you really
 996  * have to.  But in generally you should not do this, so please move
 997  * along, nothing to see here..
 998  */
 999 struct file *filp_open(const char *filename, int flags, umode_t mode)
1000 {
1001         struct filename *name = getname_kernel(filename);
1002         struct file *file = ERR_CAST(name);
1003
1004         if (!IS_ERR(name)) {
1005                 file = file_open_name(name, flags, mode);
1006                 putname(name);
1007         }
1008         return file;
1009 }
1010 EXPORT_SYMBOL(filp_open);
1011
1012 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1013                             const char *filename, int flags, umode_t mode)
1014 {
1015         struct open_flags op;
1016         int err = build_open_flags(flags, mode, &op);
1017         if (err)
1018                 return ERR_PTR(err);
1019         return do_file_open_root(dentry, mnt, filename, &op);
1020 }
1021 EXPORT_SYMBOL(file_open_root);
1022
1023 struct file *filp_clone_open(struct file *oldfile)
1024 {
1025         struct file *file;
1026         int retval;
1027
1028         file = get_empty_filp();
1029         if (IS_ERR(file))
1030                 return file;
1031
1032         file->f_flags = oldfile->f_flags;
1033         retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1034         if (retval) {
1035                 put_filp(file);
1036                 return ERR_PTR(retval);
1037         }
1038
1039         return file;
1040 }
1041 EXPORT_SYMBOL(filp_clone_open);
1042
1043 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1044 {
1045         struct open_flags op;
1046         int fd = build_open_flags(flags, mode, &op);
1047         struct filename *tmp;
1048
1049         if (fd)
1050                 return fd;
1051
1052         tmp = getname(filename);
1053         if (IS_ERR(tmp))
1054                 return PTR_ERR(tmp);
1055
1056         fd = get_unused_fd_flags(flags);
1057         if (fd >= 0) {
1058                 struct file *f = do_filp_open(dfd, tmp, &op);
1059                 if (IS_ERR(f)) {
1060                         put_unused_fd(fd);
1061                         fd = PTR_ERR(f);
1062                 } else {
1063                         fsnotify_open(f);
1064                         fd_install(fd, f);
1065                 }
1066         }
1067         putname(tmp);
1068         return fd;
1069 }
1070
1071 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1072 {
1073         if (force_o_largefile())
1074                 flags |= O_LARGEFILE;
1075
1076         return do_sys_open(AT_FDCWD, filename, flags, mode);
1077 }
1078
1079 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1080                 umode_t, mode)
1081 {
1082         if (force_o_largefile())
1083                 flags |= O_LARGEFILE;
1084
1085         return do_sys_open(dfd, filename, flags, mode);
1086 }
1087
1088 #ifdef CONFIG_COMPAT
1089 /*
1090  * Exactly like sys_open(), except that it doesn't set the
1091  * O_LARGEFILE flag.
1092  */
1093 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1094 {
1095         return do_sys_open(AT_FDCWD, filename, flags, mode);
1096 }
1097
1098 /*
1099  * Exactly like sys_openat(), except that it doesn't set the
1100  * O_LARGEFILE flag.
1101  */
1102 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1103 {
1104         return do_sys_open(dfd, filename, flags, mode);
1105 }
1106 #endif
1107
1108 #ifndef __alpha__
1109
1110 /*
1111  * For backward compatibility?  Maybe this should be moved
1112  * into arch/i386 instead?
1113  */
1114 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1115 {
1116         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1117 }
1118
1119 #endif
1120
1121 /*
1122  * "id" is the POSIX thread ID. We use the
1123  * files pointer for this..
1124  */
1125 int filp_close(struct file *filp, fl_owner_t id)
1126 {
1127         int retval = 0;
1128
1129         if (!file_count(filp)) {
1130                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1131                 return 0;
1132         }
1133
1134         if (filp->f_op->flush)
1135                 retval = filp->f_op->flush(filp, id);
1136
1137         if (likely(!(filp->f_mode & FMODE_PATH))) {
1138                 dnotify_flush(filp, id);
1139                 locks_remove_posix(filp, id);
1140         }
1141         fput(filp);
1142         return retval;
1143 }
1144
1145 EXPORT_SYMBOL(filp_close);
1146
1147 /*
1148  * Careful here! We test whether the file pointer is NULL before
1149  * releasing the fd. This ensures that one clone task can't release
1150  * an fd while another clone is opening it.
1151  */
1152 SYSCALL_DEFINE1(close, unsigned int, fd)
1153 {
1154         int retval = __close_fd(current->files, fd);
1155
1156         /* can't restart close syscall because file table entry was cleared */
1157         if (unlikely(retval == -ERESTARTSYS ||
1158                      retval == -ERESTARTNOINTR ||
1159                      retval == -ERESTARTNOHAND ||
1160                      retval == -ERESTART_RESTARTBLOCK))
1161                 retval = -EINTR;
1162
1163         return retval;
1164 }
1165 EXPORT_SYMBOL(sys_close);
1166
1167 /*
1168  * This routine simulates a hangup on the tty, to arrange that users
1169  * are given clean terminals at login time.
1170  */
1171 SYSCALL_DEFINE0(vhangup)
1172 {
1173         if (capable(CAP_SYS_TTY_CONFIG)) {
1174                 tty_vhangup_self();
1175                 return 0;
1176         }
1177         return -EPERM;
1178 }
1179
1180 /*
1181  * Called when an inode is about to be open.
1182  * We use this to disallow opening large files on 32bit systems if
1183  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1184  * on this flag in sys_open.
1185  */
1186 int generic_file_open(struct inode * inode, struct file * filp)
1187 {
1188         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1189                 return -EOVERFLOW;
1190         return 0;
1191 }
1192
1193 EXPORT_SYMBOL(generic_file_open);
1194
1195 /*
1196  * This is used by subsystems that don't want seekable
1197  * file descriptors. The function is not supposed to ever fail, the only
1198  * reason it returns an 'int' and not 'void' is so that it can be plugged
1199  * directly into file_operations structure.
1200  */
1201 int nonseekable_open(struct inode *inode, struct file *filp)
1202 {
1203         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1204         return 0;
1205 }
1206
1207 EXPORT_SYMBOL(nonseekable_open);