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