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, 0);
 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         /*
 383          * The new set of credentials can *only* be used in
 384          * task-synchronous circumstances, and does not need
 385          * RCU freeing, unless somebody then takes a separate
 386          * reference to it.
 387          *
 388          * NOTE! This is _only_ true because this credential
 389          * is used purely for override_creds() that installs
 390          * it as the subjective cred. Other threads will be
 391          * accessing ->real_cred, not the subjective cred.
 392          *
 393          * If somebody _does_ make a copy of this (using the
 394          * 'get_current_cred()' function), that will clear the
 395          * non_rcu field, because now that other user may be
 396          * expecting RCU freeing. But normal thread-synchronous
 397          * cred accesses will keep things non-RCY.
 398          */
 399         override_cred->non_rcu = 1;
 400
 401         old_cred = override_creds(override_cred);
 402 retry:
 403         res = user_path_at(dfd, filename, lookup_flags, &path);
 404         if (res)
 405                 goto out;
 406
 407         inode = d_backing_inode(path.dentry);
 408
 409         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 410                 /*
 411                  * MAY_EXEC on regular files is denied if the fs is mounted
 412                  * with the "noexec" flag.
 413                  */
 414                 res = -EACCES;
 415                 if (path_noexec(&path))
 416                         goto out_path_release;
 417         }
 418
 419         res = inode_permission(inode, mode | MAY_ACCESS);
 420         /* SuS v2 requires we report a read only fs too */
 421         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 422                 goto out_path_release;
 423         /*
 424          * This is a rare case where using __mnt_is_readonly()
 425          * is OK without a mnt_want/drop_write() pair.  Since
 426          * no actual write to the fs is performed here, we do
 427          * not need to telegraph to that to anyone.
 428          *
 429          * By doing this, we accept that this access is
 430          * inherently racy and know that the fs may change
 431          * state before we even see this result.
 432          */
 433         if (__mnt_is_readonly(path.mnt))
 434                 res = -EROFS;
 435
 436 out_path_release:
 437         path_put(&path);
 438         if (retry_estale(res, lookup_flags)) {
 439                 lookup_flags |= LOOKUP_REVAL;
 440                 goto retry;
 441         }
 442 out:
 443         revert_creds(old_cred);
 444         put_cred(override_cred);
 445         return res;
 446 }
 447
 448 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 449 {
 450         return sys_faccessat(AT_FDCWD, filename, mode);
 451 }
 452
 453 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 454 {
 455         struct path path;
 456         int error;
 457         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 458 retry:
 459         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 460         if (error)
 461                 goto out;
 462
 463         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 464         if (error)
 465                 goto dput_and_out;
 466
 467         set_fs_pwd(current->fs, &path);
 468
 469 dput_and_out:
 470         path_put(&path);
 471         if (retry_estale(error, lookup_flags)) {
 472                 lookup_flags |= LOOKUP_REVAL;
 473                 goto retry;
 474         }
 475 out:
 476         return error;
 477 }
 478
 479 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 480 {
 481         struct fd f = fdget_raw(fd);
 482         int error;
 483
 484         error = -EBADF;
 485         if (!f.file)
 486                 goto out;
 487
 488         error = -ENOTDIR;
 489         if (!d_can_lookup(f.file->f_path.dentry))
 490                 goto out_putf;
 491
 492         error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
 493         if (!error)
 494                 set_fs_pwd(current->fs, &f.file->f_path);
 495 out_putf:
 496         fdput(f);
 497 out:
 498         return error;
 499 }
 500
 501 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 502 {
 503         struct path path;
 504         int error;
 505         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 506 retry:
 507         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 508         if (error)
 509                 goto out;
 510
 511         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 512         if (error)
 513                 goto dput_and_out;
 514
 515         error = -EPERM;
 516         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 517                 goto dput_and_out;
 518         error = security_path_chroot(&path);
 519         if (error)
 520                 goto dput_and_out;
 521
 522         set_fs_root(current->fs, &path);
 523         error = 0;
 524 dput_and_out:
 525         path_put(&path);
 526         if (retry_estale(error, lookup_flags)) {
 527                 lookup_flags |= LOOKUP_REVAL;
 528                 goto retry;
 529         }
 530 out:
 531         return error;
 532 }
 533
 534 static int chmod_common(const struct path *path, umode_t mode)
 535 {
 536         struct inode *inode = path->dentry->d_inode;
 537         struct inode *delegated_inode = NULL;
 538         struct iattr newattrs;
 539         int error;
 540
 541         error = mnt_want_write(path->mnt);
 542         if (error)
 543                 return error;
 544 retry_deleg:
 545         inode_lock(inode);
 546         error = security_path_chmod(path, mode);
 547         if (error)
 548                 goto out_unlock;
 549         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 550         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 551         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 552 out_unlock:
 553         inode_unlock(inode);
 554         if (delegated_inode) {
 555                 error = break_deleg_wait(&delegated_inode);
 556                 if (!error)
 557                         goto retry_deleg;
 558         }
 559         mnt_drop_write(path->mnt);
 560         return error;
 561 }
 562
 563 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 564 {
 565         struct fd f = fdget(fd);
 566         int err = -EBADF;
 567
 568         if (f.file) {
 569                 audit_file(f.file);
 570                 err = chmod_common(&f.file->f_path, mode);
 571                 fdput(f);
 572         }
 573         return err;
 574 }
 575
 576 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 577 {
 578         struct path path;
 579         int error;
 580         unsigned int lookup_flags = LOOKUP_FOLLOW;
 581 retry:
 582         error = user_path_at(dfd, filename, lookup_flags, &path);
 583         if (!error) {
 584                 error = chmod_common(&path, mode);
 585                 path_put(&path);
 586                 if (retry_estale(error, lookup_flags)) {
 587                         lookup_flags |= LOOKUP_REVAL;
 588                         goto retry;
 589                 }
 590         }
 591         return error;
 592 }
 593
 594 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 595 {
 596         return sys_fchmodat(AT_FDCWD, filename, mode);
 597 }
 598
 599 static int chown_common(const struct path *path, uid_t user, gid_t group)
 600 {
 601         struct inode *inode = path->dentry->d_inode;
 602         struct inode *delegated_inode = NULL;
 603         int error;
 604         struct iattr newattrs;
 605         kuid_t uid;
 606         kgid_t gid;
 607
 608         uid = make_kuid(current_user_ns(), user);
 609         gid = make_kgid(current_user_ns(), group);
 610
 611 retry_deleg:
 612         newattrs.ia_valid =  ATTR_CTIME;
 613         if (user != (uid_t) -1) {
 614                 if (!uid_valid(uid))
 615                         return -EINVAL;
 616                 newattrs.ia_valid |= ATTR_UID;
 617                 newattrs.ia_uid = uid;
 618         }
 619         if (group != (gid_t) -1) {
 620                 if (!gid_valid(gid))
 621                         return -EINVAL;
 622                 newattrs.ia_valid |= ATTR_GID;
 623                 newattrs.ia_gid = gid;
 624         }
 625         if (!S_ISDIR(inode->i_mode))
 626                 newattrs.ia_valid |=
 627                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 628         inode_lock(inode);
 629         error = security_path_chown(path, uid, gid);
 630         if (!error)
 631                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 632         inode_unlock(inode);
 633         if (delegated_inode) {
 634                 error = break_deleg_wait(&delegated_inode);
 635                 if (!error)
 636                         goto retry_deleg;
 637         }
 638         return error;
 639 }
 640
 641 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 642                 gid_t, group, int, flag)
 643 {
 644         struct path path;
 645         int error = -EINVAL;
 646         int lookup_flags;
 647
 648         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 649                 goto out;
 650
 651         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 652         if (flag & AT_EMPTY_PATH)
 653                 lookup_flags |= LOOKUP_EMPTY;
 654 retry:
 655         error = user_path_at(dfd, filename, lookup_flags, &path);
 656         if (error)
 657                 goto out;
 658         error = mnt_want_write(path.mnt);
 659         if (error)
 660                 goto out_release;
 661         error = chown_common(&path, user, group);
 662         mnt_drop_write(path.mnt);
 663 out_release:
 664         path_put(&path);
 665         if (retry_estale(error, lookup_flags)) {
 666                 lookup_flags |= LOOKUP_REVAL;
 667                 goto retry;
 668         }
 669 out:
 670         return error;
 671 }
 672
 673 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 674 {
 675         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 676 }
 677
 678 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 679 {
 680         return sys_fchownat(AT_FDCWD, filename, user, group,
 681                             AT_SYMLINK_NOFOLLOW);
 682 }
 683
 684 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 685 {
 686         struct fd f = fdget(fd);
 687         int error = -EBADF;
 688
 689         if (!f.file)
 690                 goto out;
 691
 692         error = mnt_want_write_file_path(f.file);
 693         if (error)
 694                 goto out_fput;
 695         audit_file(f.file);
 696         error = chown_common(&f.file->f_path, user, group);
 697         mnt_drop_write_file_path(f.file);
 698 out_fput:
 699         fdput(f);
 700 out:
 701         return error;
 702 }
 703
 704 int open_check_o_direct(struct file *f)
 705 {
 706         /* NB: we're sure to have correct a_ops only after f_op->open */
 707         if (f->f_flags & O_DIRECT) {
 708                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 709                         return -EINVAL;
 710         }
 711         return 0;
 712 }
 713
 714 static int do_dentry_open(struct file *f,
 715                           struct inode *inode,
 716                           int (*open)(struct inode *, struct file *),
 717                           const struct cred *cred)
 718 {
 719         static const struct file_operations empty_fops = {};
 720         int error;
 721
 722         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 723                                 FMODE_PREAD | FMODE_PWRITE;
 724
 725         path_get(&f->f_path);
 726         f->f_inode = inode;
 727         f->f_mapping = inode->i_mapping;
 728
 729         /* Ensure that we skip any errors that predate opening of the file */
 730         f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
 731
 732         if (unlikely(f->f_flags & O_PATH)) {
 733                 f->f_mode = FMODE_PATH;
 734                 f->f_op = &empty_fops;
 735                 return 0;
 736         }
 737
 738         /* Any file opened for execve()/uselib() has to be a regular file. */
 739         if (unlikely(f->f_flags & FMODE_EXEC && !S_ISREG(inode->i_mode))) {
 740                 error = -EACCES;
 741                 goto cleanup_file;
 742         }
 743
 744         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 745                 error = get_write_access(inode);
 746                 if (unlikely(error))
 747                         goto cleanup_file;
 748                 error = __mnt_want_write(f->f_path.mnt);
 749                 if (unlikely(error)) {
 750                         put_write_access(inode);
 751                         goto cleanup_file;
 752                 }
 753                 f->f_mode |= FMODE_WRITER;
 754         }
 755
 756         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 757         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 758                 f->f_mode |= FMODE_ATOMIC_POS;
 759
 760         f->f_op = fops_get(inode->i_fop);
 761         if (unlikely(WARN_ON(!f->f_op))) {
 762                 error = -ENODEV;
 763                 goto cleanup_all;
 764         }
 765
 766         error = security_file_open(f, cred);
 767         if (error)
 768                 goto cleanup_all;
 769
 770         error = break_lease(locks_inode(f), f->f_flags);
 771         if (error)
 772                 goto cleanup_all;
 773
 774         if (!open)
 775                 open = f->f_op->open;
 776         if (open) {
 777                 error = open(inode, f);
 778                 if (error)
 779                         goto cleanup_all;
 780         }
 781         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 782                 i_readcount_inc(inode);
 783         if ((f->f_mode & FMODE_READ) &&
 784              likely(f->f_op->read || f->f_op->read_iter))
 785                 f->f_mode |= FMODE_CAN_READ;
 786         if ((f->f_mode & FMODE_WRITE) &&
 787              likely(f->f_op->write || f->f_op->write_iter))
 788                 f->f_mode |= FMODE_CAN_WRITE;
 789
 790         f->f_write_hint = WRITE_LIFE_NOT_SET;
 791         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 792
 793         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 794
 795         return 0;
 796
 797 cleanup_all:
 798         fops_put(f->f_op);
 799         if (f->f_mode & FMODE_WRITER) {
 800                 put_write_access(inode);
 801                 __mnt_drop_write(f->f_path.mnt);
 802         }
 803 cleanup_file:
 804         path_put(&f->f_path);
 805         f->f_path.mnt = NULL;
 806         f->f_path.dentry = NULL;
 807         f->f_inode = NULL;
 808         return error;
 809 }
 810
 811 /**
 812  * finish_open - finish opening a file
 813  * @file: file pointer
 814  * @dentry: pointer to dentry
 815  * @open: open callback
 816  * @opened: state of open
 817  *
 818  * This can be used to finish opening a file passed to i_op->atomic_open().
 819  *
 820  * If the open callback is set to NULL, then the standard f_op->open()
 821  * filesystem callback is substituted.
 822  *
 823  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 824  * the return value of d_splice_alias(), then the caller needs to perform dput()
 825  * on it after finish_open().
 826  *
 827  * Returns zero on success or -errno if the open failed.
 828  */
 829 int finish_open(struct file *file, struct dentry *dentry,
 830                 int (*open)(struct inode *, struct file *),
 831                 int *opened)
 832 {
 833         int error;
 834         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 835
 836         file->f_path.dentry = dentry;
 837         error = do_dentry_open(file, d_backing_inode(dentry), open,
 838                                current_cred());
 839         if (!error)
 840                 *opened |= FILE_OPENED;
 841
 842         return error;
 843 }
 844 EXPORT_SYMBOL(finish_open);
 845
 846 /**
 847  * finish_no_open - finish ->atomic_open() without opening the file
 848  *
 849  * @file: file pointer
 850  * @dentry: dentry or NULL (as returned from ->lookup())
 851  *
 852  * This can be used to set the result of a successful lookup in ->atomic_open().
 853  *
 854  * NB: unlike finish_open() this function does consume the dentry reference and
 855  * the caller need not dput() it.
 856  *
 857  * Returns "1" which must be the return value of ->atomic_open() after having
 858  * called this function.
 859  */
 860 int finish_no_open(struct file *file, struct dentry *dentry)
 861 {
 862         file->f_path.dentry = dentry;
 863         return 1;
 864 }
 865 EXPORT_SYMBOL(finish_no_open);
 866
 867 char *file_path(struct file *filp, char *buf, int buflen)
 868 {
 869         return d_path(&filp->f_path, buf, buflen);
 870 }
 871 EXPORT_SYMBOL(file_path);
 872
 873 /**
 874  * vfs_open - open the file at the given path
 875  * @path: path to open
 876  * @file: newly allocated file with f_flag initialized
 877  * @cred: credentials to use
 878  */
 879 int vfs_open(const struct path *path, struct file *file,
 880              const struct cred *cred)
 881 {
 882         struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags, 0);
 883
 884         if (IS_ERR(dentry))
 885                 return PTR_ERR(dentry);
 886
 887         file->f_path = *path;
 888         return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 889 }
 890
 891 struct file *dentry_open(const struct path *path, int flags,
 892                          const struct cred *cred)
 893 {
 894         int error;
 895         struct file *f;
 896
 897         validate_creds(cred);
 898
 899         /* We must always pass in a valid mount pointer. */
 900         BUG_ON(!path->mnt);
 901
 902         f = get_empty_filp();
 903         if (!IS_ERR(f)) {
 904                 f->f_flags = flags;
 905                 error = vfs_open(path, f, cred);
 906                 if (!error) {
 907                         /* from now on we need fput() to dispose of f */
 908                         error = open_check_o_direct(f);
 909                         if (error) {
 910                                 fput(f);
 911                                 f = ERR_PTR(error);
 912                         }
 913                 } else {
 914                         put_filp(f);
 915                         f = ERR_PTR(error);
 916                 }
 917         }
 918         return f;
 919 }
 920 EXPORT_SYMBOL(dentry_open);
 921
 922 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 923 {
 924         int lookup_flags = 0;
 925         int acc_mode = ACC_MODE(flags);
 926
 927         /*
 928          * Clear out all open flags we don't know about so that we don't report
 929          * them in fcntl(F_GETFD) or similar interfaces.
 930          */
 931         flags &= VALID_OPEN_FLAGS;
 932
 933         if (flags & (O_CREAT | __O_TMPFILE))
 934                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 935         else
 936                 op->mode = 0;
 937
 938         /* Must never be set by userspace */
 939         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 940
 941         /*
 942          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 943          * check for O_DSYNC if the need any syncing at all we enforce it's
 944          * always set instead of having to deal with possibly weird behaviour
 945          * for malicious applications setting only __O_SYNC.
 946          */
 947         if (flags & __O_SYNC)
 948                 flags |= O_DSYNC;
 949
 950         if (flags & __O_TMPFILE) {
 951                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 952                         return -EINVAL;
 953                 if (!(acc_mode & MAY_WRITE))
 954                         return -EINVAL;
 955         } else if (flags & O_PATH) {
 956                 /*
 957                  * If we have O_PATH in the open flag. Then we
 958                  * cannot have anything other than the below set of flags
 959                  */
 960                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 961                 acc_mode = 0;
 962         }
 963
 964         op->open_flag = flags;
 965
 966         /* O_TRUNC implies we need access checks for write permissions */
 967         if (flags & O_TRUNC)
 968                 acc_mode |= MAY_WRITE;
 969
 970         /* Allow the LSM permission hook to distinguish append
 971            access from general write access. */
 972         if (flags & O_APPEND)
 973                 acc_mode |= MAY_APPEND;
 974
 975         op->acc_mode = acc_mode;
 976
 977         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 978
 979         if (flags & O_CREAT) {
 980                 op->intent |= LOOKUP_CREATE;
 981                 if (flags & O_EXCL)
 982                         op->intent |= LOOKUP_EXCL;
 983         }
 984
 985         if (flags & O_DIRECTORY)
 986                 lookup_flags |= LOOKUP_DIRECTORY;
 987         if (!(flags & O_NOFOLLOW))
 988                 lookup_flags |= LOOKUP_FOLLOW;
 989         op->lookup_flags = lookup_flags;
 990         return 0;
 991 }
 992
 993 /**
 994  * file_open_name - open file and return file pointer
 995  *
 996  * @name:       struct filename containing path to open
 997  * @flags:      open flags as per the open(2) second argument
 998  * @mode:       mode for the new file if O_CREAT is set, else ignored
 999  *
1000  * This is the helper to open a file from kernelspace if you really
1001  * have to.  But in generally you should not do this, so please move
1002  * along, nothing to see here..
1003  */
1004 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1005 {
1006         struct open_flags op;
1007         int err = build_open_flags(flags, mode, &op);
1008         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
1009 }
1010
1011 /**
1012  * filp_open - open file and return file pointer
1013  *
1014  * @filename:   path to open
1015  * @flags:      open flags as per the open(2) second argument
1016  * @mode:       mode for the new file if O_CREAT is set, else ignored
1017  *
1018  * This is the helper to open a file from kernelspace if you really
1019  * have to.  But in generally you should not do this, so please move
1020  * along, nothing to see here..
1021  */
1022 struct file *filp_open(const char *filename, int flags, umode_t mode)
1023 {
1024         struct filename *name = getname_kernel(filename);
1025         struct file *file = ERR_CAST(name);
1026
1027         if (!IS_ERR(name)) {
1028                 file = file_open_name(name, flags, mode);
1029                 putname(name);
1030         }
1031         return file;
1032 }
1033 EXPORT_SYMBOL(filp_open);
1034
1035 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1036                             const char *filename, int flags, umode_t mode)
1037 {
1038         struct open_flags op;
1039         int err = build_open_flags(flags, mode, &op);
1040         if (err)
1041                 return ERR_PTR(err);
1042         return do_file_open_root(dentry, mnt, filename, &op);
1043 }
1044 EXPORT_SYMBOL(file_open_root);
1045
1046 struct file *filp_clone_open(struct file *oldfile)
1047 {
1048         struct file *file;
1049         int retval;
1050
1051         file = get_empty_filp();
1052         if (IS_ERR(file))
1053                 return file;
1054
1055         file->f_flags = oldfile->f_flags;
1056         retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1057         if (retval) {
1058                 put_filp(file);
1059                 return ERR_PTR(retval);
1060         }
1061
1062         return file;
1063 }
1064 EXPORT_SYMBOL(filp_clone_open);
1065
1066 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1067 {
1068         struct open_flags op;
1069         int fd = build_open_flags(flags, mode, &op);
1070         struct filename *tmp;
1071
1072         if (fd)
1073                 return fd;
1074
1075         tmp = getname(filename);
1076         if (IS_ERR(tmp))
1077                 return PTR_ERR(tmp);
1078
1079         fd = get_unused_fd_flags(flags);
1080         if (fd >= 0) {
1081                 struct file *f = do_filp_open(dfd, tmp, &op);
1082                 if (IS_ERR(f)) {
1083                         put_unused_fd(fd);
1084                         fd = PTR_ERR(f);
1085                 } else {
1086                         fsnotify_open(f);
1087                         fd_install(fd, f);
1088                 }
1089         }
1090         putname(tmp);
1091         return fd;
1092 }
1093
1094 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1095 {
1096         if (force_o_largefile())
1097                 flags |= O_LARGEFILE;
1098
1099         return do_sys_open(AT_FDCWD, filename, flags, mode);
1100 }
1101
1102 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1103                 umode_t, mode)
1104 {
1105         if (force_o_largefile())
1106                 flags |= O_LARGEFILE;
1107
1108         return do_sys_open(dfd, filename, flags, mode);
1109 }
1110
1111 #ifdef CONFIG_COMPAT
1112 /*
1113  * Exactly like sys_open(), except that it doesn't set the
1114  * O_LARGEFILE flag.
1115  */
1116 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1117 {
1118         return do_sys_open(AT_FDCWD, filename, flags, mode);
1119 }
1120
1121 /*
1122  * Exactly like sys_openat(), except that it doesn't set the
1123  * O_LARGEFILE flag.
1124  */
1125 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1126 {
1127         return do_sys_open(dfd, filename, flags, mode);
1128 }
1129 #endif
1130
1131 #ifndef __alpha__
1132
1133 /*
1134  * For backward compatibility?  Maybe this should be moved
1135  * into arch/i386 instead?
1136  */
1137 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1138 {
1139         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1140 }
1141
1142 #endif
1143
1144 /*
1145  * "id" is the POSIX thread ID. We use the
1146  * files pointer for this..
1147  */
1148 int filp_close(struct file *filp, fl_owner_t id)
1149 {
1150         int retval = 0;
1151
1152         if (!file_count(filp)) {
1153                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1154                 return 0;
1155         }
1156
1157         if (filp->f_op->flush)
1158                 retval = filp->f_op->flush(filp, id);
1159
1160         if (likely(!(filp->f_mode & FMODE_PATH))) {
1161                 dnotify_flush(filp, id);
1162                 locks_remove_posix(filp, id);
1163         }
1164         fput(filp);
1165         return retval;
1166 }
1167
1168 EXPORT_SYMBOL(filp_close);
1169
1170 /*
1171  * Careful here! We test whether the file pointer is NULL before
1172  * releasing the fd. This ensures that one clone task can't release
1173  * an fd while another clone is opening it.
1174  */
1175 SYSCALL_DEFINE1(close, unsigned int, fd)
1176 {
1177         int retval = __close_fd(current->files, fd);
1178
1179         /* can't restart close syscall because file table entry was cleared */
1180         if (unlikely(retval == -ERESTARTSYS ||
1181                      retval == -ERESTARTNOINTR ||
1182                      retval == -ERESTARTNOHAND ||
1183                      retval == -ERESTART_RESTARTBLOCK))
1184                 retval = -EINTR;
1185
1186         return retval;
1187 }
1188 EXPORT_SYMBOL(sys_close);
1189
1190 /*
1191  * This routine simulates a hangup on the tty, to arrange that users
1192  * are given clean terminals at login time.
1193  */
1194 SYSCALL_DEFINE0(vhangup)
1195 {
1196         if (capable(CAP_SYS_TTY_CONFIG)) {
1197                 tty_vhangup_self();
1198                 return 0;
1199         }
1200         return -EPERM;
1201 }
1202
1203 /*
1204  * Called when an inode is about to be open.
1205  * We use this to disallow opening large files on 32bit systems if
1206  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1207  * on this flag in sys_open.
1208  */
1209 int generic_file_open(struct inode * inode, struct file * filp)
1210 {
1211         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1212                 return -EOVERFLOW;
1213         return 0;
1214 }
1215
1216 EXPORT_SYMBOL(generic_file_open);
1217
1218 /*
1219  * This is used by subsystems that don't want seekable
1220  * file descriptors. The function is not supposed to ever fail, the only
1221  * reason it returns an 'int' and not 'void' is so that it can be plugged
1222  * directly into file_operations structure.
1223  */
1224 int nonseekable_open(struct inode *inode, struct file *filp)
1225 {
1226         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1227         return 0;
1228 }
1229
1230 EXPORT_SYMBOL(nonseekable_open);
1231
1232 /*
1233  * stream_open is used by subsystems that want stream-like file descriptors.
1234  * Such file descriptors are not seekable and don't have notion of position
1235  * (file.f_pos is always 0). Contrary to file descriptors of other regular
1236  * files, .read() and .write() can run simultaneously.
1237  *
1238  * stream_open never fails and is marked to return int so that it could be
1239  * directly used as file_operations.open .
1240  */
1241 int stream_open(struct inode *inode, struct file *filp)
1242 {
1243         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1244         filp->f_mode |= FMODE_STREAM;
1245         return 0;
1246 }
1247
1248 EXPORT_SYMBOL(stream_open);