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