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