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