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