fs/open.c

   1 /*
   2  *  linux/fs/open.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/string.h>
   8 #include <linux/mm.h>
   9 #include <linux/file.h>
  10 #include <linux/fdtable.h>
  11 #include <linux/fsnotify.h>
  12 #include <linux/module.h>
  13 #include <linux/tty.h>
  14 #include <linux/namei.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/capability.h>
  17 #include <linux/securebits.h>
  18 #include <linux/security.h>
  19 #include <linux/mount.h>
  20 #include <linux/fcntl.h>
  21 #include <linux/slab.h>
  22 #include <asm/uaccess.h>
  23 #include <linux/fs.h>
  24 #include <linux/personality.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/rcupdate.h>
  28 #include <linux/audit.h>
  29 #include <linux/falloc.h>
  30 #include <linux/fs_struct.h>
  31 #include <linux/ima.h>
  32 #include <linux/dnotify.h>
  33 #include <linux/compat.h>
  34
  35 #include "internal.h"
  36
  37 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
  38         struct file *filp)
  39 {
  40         int ret;
  41         struct iattr newattrs;
  42
  43         /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
  44         if (length < 0)
  45                 return -EINVAL;
  46
  47         newattrs.ia_size = length;
  48         newattrs.ia_valid = ATTR_SIZE | time_attrs;
  49         if (filp) {
  50                 newattrs.ia_file = filp;
  51                 newattrs.ia_valid |= ATTR_FILE;
  52         }
  53
  54         /* Remove suid/sgid on truncate too */
  55         ret = should_remove_suid(dentry);
  56         if (ret)
  57                 newattrs.ia_valid |= ret | ATTR_FORCE;
  58
  59         mutex_lock(&dentry->d_inode->i_mutex);
  60         /* Note any delegations or leases have already been broken: */
  61         ret = notify_change(dentry, &newattrs, NULL);
  62         mutex_unlock(&dentry->d_inode->i_mutex);
  63         return ret;
  64 }
  65
  66 long vfs_truncate(struct path *path, loff_t length)
  67 {
  68         struct inode *inode;
  69         long error;
  70
  71         inode = path->dentry->d_inode;
  72
  73         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  74         if (S_ISDIR(inode->i_mode))
  75                 return -EISDIR;
  76         if (!S_ISREG(inode->i_mode))
  77                 return -EINVAL;
  78
  79         error = mnt_want_write(path->mnt);
  80         if (error)
  81                 goto out;
  82
  83         error = inode_permission(inode, MAY_WRITE);
  84         if (error)
  85                 goto mnt_drop_write_and_out;
  86
  87         error = -EPERM;
  88         if (IS_APPEND(inode))
  89                 goto mnt_drop_write_and_out;
  90
  91         error = get_write_access(inode);
  92         if (error)
  93                 goto mnt_drop_write_and_out;
  94
  95         /*
  96          * Make sure that there are no leases.  get_write_access() protects
  97          * against the truncate racing with a lease-granting setlease().
  98          */
  99         error = break_lease(inode, O_WRONLY);
 100         if (error)
 101                 goto put_write_and_out;
 102
 103         error = locks_verify_truncate(inode, NULL, length);
 104         if (!error)
 105                 error = security_path_truncate(path);
 106         if (!error)
 107                 error = do_truncate(path->dentry, length, 0, NULL);
 108
 109 put_write_and_out:
 110         put_write_access(inode);
 111 mnt_drop_write_and_out:
 112         mnt_drop_write(path->mnt);
 113 out:
 114         return error;
 115 }
 116 EXPORT_SYMBOL_GPL(vfs_truncate);
 117
 118 static long do_sys_truncate(const char __user *pathname, loff_t length)
 119 {
 120         unsigned int lookup_flags = LOOKUP_FOLLOW;
 121         struct path path;
 122         int error;
 123
 124         if (length < 0) /* sorry, but loff_t says... */
 125                 return -EINVAL;
 126
 127 retry:
 128         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 129         if (!error) {
 130                 error = vfs_truncate(&path, length);
 131                 path_put(&path);
 132         }
 133         if (retry_estale(error, lookup_flags)) {
 134                 lookup_flags |= LOOKUP_REVAL;
 135                 goto retry;
 136         }
 137         return error;
 138 }
 139
 140 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 141 {
 142         return do_sys_truncate(path, length);
 143 }
 144
 145 #ifdef CONFIG_COMPAT
 146 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 147 {
 148         return do_sys_truncate(path, length);
 149 }
 150 #endif
 151
 152 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 153 {
 154         struct inode *inode;
 155         struct dentry *dentry;
 156         struct fd f;
 157         int error;
 158
 159         error = -EINVAL;
 160         if (length < 0)
 161                 goto out;
 162         error = -EBADF;
 163         f = fdget(fd);
 164         if (!f.file)
 165                 goto out;
 166
 167         /* explicitly opened as large or we are on 64-bit box */
 168         if (f.file->f_flags & O_LARGEFILE)
 169                 small = 0;
 170
 171         dentry = f.file->f_path.dentry;
 172         inode = dentry->d_inode;
 173         error = -EINVAL;
 174         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 175                 goto out_putf;
 176
 177         error = -EINVAL;
 178         /* Cannot ftruncate over 2^31 bytes without large file support */
 179         if (small && length > MAX_NON_LFS)
 180                 goto out_putf;
 181
 182         error = -EPERM;
 183         if (IS_APPEND(inode))
 184                 goto out_putf;
 185
 186         sb_start_write(inode->i_sb);
 187         error = locks_verify_truncate(inode, f.file, length);
 188         if (!error)
 189                 error = security_path_truncate(&f.file->f_path);
 190         if (!error)
 191                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 192         sb_end_write(inode->i_sb);
 193 out_putf:
 194         fdput(f);
 195 out:
 196         return error;
 197 }
 198
 199 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 200 {
 201         return do_sys_ftruncate(fd, length, 1);
 202 }
 203
 204 #ifdef CONFIG_COMPAT
 205 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 206 {
 207         return do_sys_ftruncate(fd, length, 1);
 208 }
 209 #endif
 210
 211 /* LFS versions of truncate are only needed on 32 bit machines */
 212 #if BITS_PER_LONG == 32
 213 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 214 {
 215         return do_sys_truncate(path, length);
 216 }
 217
 218 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 219 {
 220         return do_sys_ftruncate(fd, length, 0);
 221 }
 222 #endif /* BITS_PER_LONG == 32 */
 223
 224
 225 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 226 {
 227         struct inode *inode = file_inode(file);
 228         long ret;
 229
 230         if (offset < 0 || len <= 0)
 231                 return -EINVAL;
 232
 233         /* Return error if mode is not supported */
 234         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
 235                      FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
 236                 return -EOPNOTSUPP;
 237
 238         /* Punch hole and zero range are mutually exclusive */
 239         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 240             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 241                 return -EOPNOTSUPP;
 242
 243         /* Punch hole must have keep size set */
 244         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 245             !(mode & FALLOC_FL_KEEP_SIZE))
 246                 return -EOPNOTSUPP;
 247
 248         /* Collapse range should only be used exclusively. */
 249         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 250             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 251                 return -EINVAL;
 252
 253         if (!(file->f_mode & FMODE_WRITE))
 254                 return -EBADF;
 255
 256         /*
 257          * We can only allow pure fallocate on append only files
 258          */
 259         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 260                 return -EPERM;
 261
 262         if (IS_IMMUTABLE(inode))
 263                 return -EPERM;
 264
 265         /*
 266          * We 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
 313 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 314 {
 315         struct fd f = fdget(fd);
 316         int error = -EBADF;
 317
 318         if (f.file) {
 319                 error = do_fallocate(f.file, mode, offset, len);
 320                 fdput(f);
 321         }
 322         return error;
 323 }
 324
 325 /*
 326  * access() needs to use the real uid/gid, not the effective uid/gid.
 327  * We do this by temporarily clearing all FS-related capabilities and
 328  * switching the fsuid/fsgid around to the real ones.
 329  */
 330 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 331 {
 332         const struct cred *old_cred;
 333         struct cred *override_cred;
 334         struct path path;
 335         struct inode *inode;
 336         int res;
 337         unsigned int lookup_flags = LOOKUP_FOLLOW;
 338
 339         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 340                 return -EINVAL;
 341
 342         override_cred = prepare_creds();
 343         if (!override_cred)
 344                 return -ENOMEM;
 345
 346         override_cred->fsuid = override_cred->uid;
 347         override_cred->fsgid = override_cred->gid;
 348
 349         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 350                 /* Clear the capabilities if we switch to a non-root user */
 351                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 352                 if (!uid_eq(override_cred->uid, root_uid))
 353                         cap_clear(override_cred->cap_effective);
 354                 else
 355                         override_cred->cap_effective =
 356                                 override_cred->cap_permitted;
 357         }
 358
 359         old_cred = override_creds(override_cred);
 360 retry:
 361         res = user_path_at(dfd, filename, lookup_flags, &path);
 362         if (res)
 363                 goto out;
 364
 365         inode = path.dentry->d_inode;
 366
 367         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 368                 /*
 369                  * MAY_EXEC on regular files is denied if the fs is mounted
 370                  * with the "noexec" flag.
 371                  */
 372                 res = -EACCES;
 373                 if (path.mnt->mnt_flags & MNT_NOEXEC)
 374                         goto out_path_release;
 375         }
 376
 377         res = inode_permission(inode, mode | MAY_ACCESS);
 378         /* SuS v2 requires we report a read only fs too */
 379         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 380                 goto out_path_release;
 381         /*
 382          * This is a rare case where using __mnt_is_readonly()
 383          * is OK without a mnt_want/drop_write() pair.  Since
 384          * no actual write to the fs is performed here, we do
 385          * not need to telegraph to that to anyone.
 386          *
 387          * By doing this, we accept that this access is
 388          * inherently racy and know that the fs may change
 389          * state before we even see this result.
 390          */
 391         if (__mnt_is_readonly(path.mnt))
 392                 res = -EROFS;
 393
 394 out_path_release:
 395         path_put(&path);
 396         if (retry_estale(res, lookup_flags)) {
 397                 lookup_flags |= LOOKUP_REVAL;
 398                 goto retry;
 399         }
 400 out:
 401         revert_creds(old_cred);
 402         put_cred(override_cred);
 403         return res;
 404 }
 405
 406 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 407 {
 408         return sys_faccessat(AT_FDCWD, filename, mode);
 409 }
 410
 411 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 412 {
 413         struct path path;
 414         int error;
 415         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 416 retry:
 417         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 418         if (error)
 419                 goto out;
 420
 421         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 422         if (error)
 423                 goto dput_and_out;
 424
 425         set_fs_pwd(current->fs, &path);
 426
 427 dput_and_out:
 428         path_put(&path);
 429         if (retry_estale(error, lookup_flags)) {
 430                 lookup_flags |= LOOKUP_REVAL;
 431                 goto retry;
 432         }
 433 out:
 434         return error;
 435 }
 436
 437 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 438 {
 439         struct fd f = fdget_raw(fd);
 440         struct inode *inode;
 441         int error = -EBADF;
 442
 443         error = -EBADF;
 444         if (!f.file)
 445                 goto out;
 446
 447         inode = file_inode(f.file);
 448
 449         error = -ENOTDIR;
 450         if (!S_ISDIR(inode->i_mode))
 451                 goto out_putf;
 452
 453         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 454         if (!error)
 455                 set_fs_pwd(current->fs, &f.file->f_path);
 456 out_putf:
 457         fdput(f);
 458 out:
 459         return error;
 460 }
 461
 462 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 463 {
 464         struct path path;
 465         int error;
 466         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 467 retry:
 468         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 469         if (error)
 470                 goto out;
 471
 472         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 473         if (error)
 474                 goto dput_and_out;
 475
 476         error = -EPERM;
 477         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 478                 goto dput_and_out;
 479         error = security_path_chroot(&path);
 480         if (error)
 481                 goto dput_and_out;
 482
 483         set_fs_root(current->fs, &path);
 484         error = 0;
 485 dput_and_out:
 486         path_put(&path);
 487         if (retry_estale(error, lookup_flags)) {
 488                 lookup_flags |= LOOKUP_REVAL;
 489                 goto retry;
 490         }
 491 out:
 492         return error;
 493 }
 494
 495 static int chmod_common(struct path *path, umode_t mode)
 496 {
 497         struct inode *inode = path->dentry->d_inode;
 498         struct inode *delegated_inode = NULL;
 499         struct iattr newattrs;
 500         int error;
 501
 502         error = mnt_want_write(path->mnt);
 503         if (error)
 504                 return error;
 505 retry_deleg:
 506         mutex_lock(&inode->i_mutex);
 507         error = security_path_chmod(path, mode);
 508         if (error)
 509                 goto out_unlock;
 510         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 511         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 512         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 513 out_unlock:
 514         mutex_unlock(&inode->i_mutex);
 515         if (delegated_inode) {
 516                 error = break_deleg_wait(&delegated_inode);
 517                 if (!error)
 518                         goto retry_deleg;
 519         }
 520         mnt_drop_write(path->mnt);
 521         return error;
 522 }
 523
 524 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 525 {
 526         struct fd f = fdget(fd);
 527         int err = -EBADF;
 528
 529         if (f.file) {
 530                 audit_file(f.file);
 531                 err = chmod_common(&f.file->f_path, mode);
 532                 fdput(f);
 533         }
 534         return err;
 535 }
 536
 537 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 538 {
 539         struct path path;
 540         int error;
 541         unsigned int lookup_flags = LOOKUP_FOLLOW;
 542 retry:
 543         error = user_path_at(dfd, filename, lookup_flags, &path);
 544         if (!error) {
 545                 error = chmod_common(&path, mode);
 546                 path_put(&path);
 547                 if (retry_estale(error, lookup_flags)) {
 548                         lookup_flags |= LOOKUP_REVAL;
 549                         goto retry;
 550                 }
 551         }
 552         return error;
 553 }
 554
 555 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 556 {
 557         return sys_fchmodat(AT_FDCWD, filename, mode);
 558 }
 559
 560 static int chown_common(struct path *path, uid_t user, gid_t group)
 561 {
 562         struct inode *inode = path->dentry->d_inode;
 563         struct inode *delegated_inode = NULL;
 564         int error;
 565         struct iattr newattrs;
 566         kuid_t uid;
 567         kgid_t gid;
 568
 569         uid = make_kuid(current_user_ns(), user);
 570         gid = make_kgid(current_user_ns(), group);
 571
 572         newattrs.ia_valid =  ATTR_CTIME;
 573         if (user != (uid_t) -1) {
 574                 if (!uid_valid(uid))
 575                         return -EINVAL;
 576                 newattrs.ia_valid |= ATTR_UID;
 577                 newattrs.ia_uid = uid;
 578         }
 579         if (group != (gid_t) -1) {
 580                 if (!gid_valid(gid))
 581                         return -EINVAL;
 582                 newattrs.ia_valid |= ATTR_GID;
 583                 newattrs.ia_gid = gid;
 584         }
 585         if (!S_ISDIR(inode->i_mode))
 586                 newattrs.ia_valid |=
 587                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 588 retry_deleg:
 589         mutex_lock(&inode->i_mutex);
 590         error = security_path_chown(path, uid, gid);
 591         if (!error)
 592                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 593         mutex_unlock(&inode->i_mutex);
 594         if (delegated_inode) {
 595                 error = break_deleg_wait(&delegated_inode);
 596                 if (!error)
 597                         goto retry_deleg;
 598         }
 599         return error;
 600 }
 601
 602 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 603                 gid_t, group, int, flag)
 604 {
 605         struct path path;
 606         int error = -EINVAL;
 607         int lookup_flags;
 608
 609         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 610                 goto out;
 611
 612         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 613         if (flag & AT_EMPTY_PATH)
 614                 lookup_flags |= LOOKUP_EMPTY;
 615 retry:
 616         error = user_path_at(dfd, filename, lookup_flags, &path);
 617         if (error)
 618                 goto out;
 619         error = mnt_want_write(path.mnt);
 620         if (error)
 621                 goto out_release;
 622         error = chown_common(&path, user, group);
 623         mnt_drop_write(path.mnt);
 624 out_release:
 625         path_put(&path);
 626         if (retry_estale(error, lookup_flags)) {
 627                 lookup_flags |= LOOKUP_REVAL;
 628                 goto retry;
 629         }
 630 out:
 631         return error;
 632 }
 633
 634 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 635 {
 636         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 637 }
 638
 639 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 640 {
 641         return sys_fchownat(AT_FDCWD, filename, user, group,
 642                             AT_SYMLINK_NOFOLLOW);
 643 }
 644
 645 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 646 {
 647         struct fd f = fdget(fd);
 648         int error = -EBADF;
 649
 650         if (!f.file)
 651                 goto out;
 652
 653         error = mnt_want_write_file(f.file);
 654         if (error)
 655                 goto out_fput;
 656         audit_file(f.file);
 657         error = chown_common(&f.file->f_path, user, group);
 658         mnt_drop_write_file(f.file);
 659 out_fput:
 660         fdput(f);
 661 out:
 662         return error;
 663 }
 664
 665 int open_check_o_direct(struct file *f)
 666 {
 667         /* NB: we're sure to have correct a_ops only after f_op->open */
 668         if (f->f_flags & O_DIRECT) {
 669                 if (!f->f_mapping->a_ops ||
 670                     ((!f->f_mapping->a_ops->direct_IO) &&
 671                     (!f->f_mapping->a_ops->get_xip_mem))) {
 672                         return -EINVAL;
 673                 }
 674         }
 675         return 0;
 676 }
 677
 678 static int do_dentry_open(struct file *f,
 679                           int (*open)(struct inode *, struct file *),
 680                           const struct cred *cred)
 681 {
 682         static const struct file_operations empty_fops = {};
 683         struct inode *inode;
 684         int error;
 685
 686         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 687                                 FMODE_PREAD | FMODE_PWRITE;
 688
 689         path_get(&f->f_path);
 690         inode = f->f_inode = f->f_path.dentry->d_inode;
 691         f->f_mapping = inode->i_mapping;
 692
 693         if (unlikely(f->f_flags & O_PATH)) {
 694                 f->f_mode = FMODE_PATH;
 695                 f->f_op = &empty_fops;
 696                 return 0;
 697         }
 698
 699         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 700                 error = get_write_access(inode);
 701                 if (unlikely(error))
 702                         goto cleanup_file;
 703                 error = __mnt_want_write(f->f_path.mnt);
 704                 if (unlikely(error)) {
 705                         put_write_access(inode);
 706                         goto cleanup_file;
 707                 }
 708                 f->f_mode |= FMODE_WRITER;
 709         }
 710
 711         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 712         if (S_ISREG(inode->i_mode))
 713                 f->f_mode |= FMODE_ATOMIC_POS;
 714
 715         f->f_op = fops_get(inode->i_fop);
 716         if (unlikely(WARN_ON(!f->f_op))) {
 717                 error = -ENODEV;
 718                 goto cleanup_all;
 719         }
 720
 721         error = security_file_open(f, cred);
 722         if (error)
 723                 goto cleanup_all;
 724
 725         error = break_lease(inode, f->f_flags);
 726         if (error)
 727                 goto cleanup_all;
 728
 729         if (!open)
 730                 open = f->f_op->open;
 731         if (open) {
 732                 error = open(inode, f);
 733                 if (error)
 734                         goto cleanup_all;
 735         }
 736         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 737                 i_readcount_inc(inode);
 738         if ((f->f_mode & FMODE_READ) &&
 739              likely(f->f_op->read || f->f_op->aio_read || f->f_op->read_iter))
 740                 f->f_mode |= FMODE_CAN_READ;
 741         if ((f->f_mode & FMODE_WRITE) &&
 742              likely(f->f_op->write || f->f_op->aio_write || f->f_op->write_iter))
 743                 f->f_mode |= FMODE_CAN_WRITE;
 744
 745         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 746
 747         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 748
 749         return 0;
 750
 751 cleanup_all:
 752         fops_put(f->f_op);
 753         if (f->f_mode & FMODE_WRITER) {
 754                 put_write_access(inode);
 755                 __mnt_drop_write(f->f_path.mnt);
 756         }
 757 cleanup_file:
 758         path_put(&f->f_path);
 759         f->f_path.mnt = NULL;
 760         f->f_path.dentry = NULL;
 761         f->f_inode = NULL;
 762         return error;
 763 }
 764
 765 /**
 766  * finish_open - finish opening a file
 767  * @file: file pointer
 768  * @dentry: pointer to dentry
 769  * @open: open callback
 770  * @opened: state of open
 771  *
 772  * This can be used to finish opening a file passed to i_op->atomic_open().
 773  *
 774  * If the open callback is set to NULL, then the standard f_op->open()
 775  * filesystem callback is substituted.
 776  *
 777  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 778  * the return value of d_splice_alias(), then the caller needs to perform dput()
 779  * on it after finish_open().
 780  *
 781  * On successful return @file is a fully instantiated open file.  After this, if
 782  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 783  *
 784  * Returns zero on success or -errno if the open failed.
 785  */
 786 int finish_open(struct file *file, struct dentry *dentry,
 787                 int (*open)(struct inode *, struct file *),
 788                 int *opened)
 789 {
 790         int error;
 791         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 792
 793         file->f_path.dentry = dentry;
 794         error = do_dentry_open(file, open, current_cred());
 795         if (!error)
 796                 *opened |= FILE_OPENED;
 797
 798         return error;
 799 }
 800 EXPORT_SYMBOL(finish_open);
 801
 802 /**
 803  * finish_no_open - finish ->atomic_open() without opening the file
 804  *
 805  * @file: file pointer
 806  * @dentry: dentry or NULL (as returned from ->lookup())
 807  *
 808  * This can be used to set the result of a successful lookup in ->atomic_open().
 809  *
 810  * NB: unlike finish_open() this function does consume the dentry reference and
 811  * the caller need not dput() it.
 812  *
 813  * Returns "1" which must be the return value of ->atomic_open() after having
 814  * called this function.
 815  */
 816 int finish_no_open(struct file *file, struct dentry *dentry)
 817 {
 818         file->f_path.dentry = dentry;
 819         return 1;
 820 }
 821 EXPORT_SYMBOL(finish_no_open);
 822
 823 struct file *dentry_open(const struct path *path, int flags,
 824                          const struct cred *cred)
 825 {
 826         int error;
 827         struct file *f;
 828
 829         validate_creds(cred);
 830
 831         /* We must always pass in a valid mount pointer. */
 832         BUG_ON(!path->mnt);
 833
 834         f = get_empty_filp();
 835         if (!IS_ERR(f)) {
 836                 f->f_flags = flags;
 837                 error = vfs_open(path, f, cred);
 838                 if (!error) {
 839                         /* from now on we need fput() to dispose of f */
 840                         error = open_check_o_direct(f);
 841                         if (error) {
 842                                 fput(f);
 843                                 f = ERR_PTR(error);
 844                         }
 845                 } else {
 846                         put_filp(f);
 847                         f = ERR_PTR(error);
 848                 }
 849         }
 850         return f;
 851 }
 852 EXPORT_SYMBOL(dentry_open);
 853
 854 /**
 855  * vfs_open - open the file at the given path
 856  * @path: path to open
 857  * @filp: newly allocated file with f_flag initialized
 858  * @cred: credentials to use
 859  */
 860 int vfs_open(const struct path *path, struct file *filp,
 861              const struct cred *cred)
 862 {
 863         struct inode *inode = path->dentry->d_inode;
 864
 865         if (inode->i_op->dentry_open)
 866                 return inode->i_op->dentry_open(path->dentry, filp, cred);
 867         else {
 868                 filp->f_path = *path;
 869                 return do_dentry_open(filp, NULL, cred);
 870         }
 871 }
 872 EXPORT_SYMBOL(vfs_open);
 873
 874 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 875 {
 876         int lookup_flags = 0;
 877         int acc_mode;
 878
 879         if (flags & (O_CREAT | __O_TMPFILE))
 880                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 881         else
 882                 op->mode = 0;
 883
 884         /* Must never be set by userspace */
 885         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 886
 887         /*
 888          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 889          * check for O_DSYNC if the need any syncing at all we enforce it's
 890          * always set instead of having to deal with possibly weird behaviour
 891          * for malicious applications setting only __O_SYNC.
 892          */
 893         if (flags & __O_SYNC)
 894                 flags |= O_DSYNC;
 895
 896         if (flags & __O_TMPFILE) {
 897                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 898                         return -EINVAL;
 899                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 900                 if (!(acc_mode & MAY_WRITE))
 901                         return -EINVAL;
 902         } else if (flags & O_PATH) {
 903                 /*
 904                  * If we have O_PATH in the open flag. Then we
 905                  * cannot have anything other than the below set of flags
 906                  */
 907                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 908                 acc_mode = 0;
 909         } else {
 910                 acc_mode = MAY_OPEN | ACC_MODE(flags);
 911         }
 912
 913         op->open_flag = flags;
 914
 915         /* O_TRUNC implies we need access checks for write permissions */
 916         if (flags & O_TRUNC)
 917                 acc_mode |= MAY_WRITE;
 918
 919         /* Allow the LSM permission hook to distinguish append
 920            access from general write access. */
 921         if (flags & O_APPEND)
 922                 acc_mode |= MAY_APPEND;
 923
 924         op->acc_mode = acc_mode;
 925
 926         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 927
 928         if (flags & O_CREAT) {
 929                 op->intent |= LOOKUP_CREATE;
 930                 if (flags & O_EXCL)
 931                         op->intent |= LOOKUP_EXCL;
 932         }
 933
 934         if (flags & O_DIRECTORY)
 935                 lookup_flags |= LOOKUP_DIRECTORY;
 936         if (!(flags & O_NOFOLLOW))
 937                 lookup_flags |= LOOKUP_FOLLOW;
 938         op->lookup_flags = lookup_flags;
 939         return 0;
 940 }
 941
 942 /**
 943  * file_open_name - open file and return file pointer
 944  *
 945  * @name:       struct filename containing path to open
 946  * @flags:      open flags as per the open(2) second argument
 947  * @mode:       mode for the new file if O_CREAT is set, else ignored
 948  *
 949  * This is the helper to open a file from kernelspace if you really
 950  * have to.  But in generally you should not do this, so please move
 951  * along, nothing to see here..
 952  */
 953 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 954 {
 955         struct open_flags op;
 956         int err = build_open_flags(flags, mode, &op);
 957         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 958 }
 959
 960 /**
 961  * filp_open - open file and return file pointer
 962  *
 963  * @filename:   path to open
 964  * @flags:      open flags as per the open(2) second argument
 965  * @mode:       mode for the new file if O_CREAT is set, else ignored
 966  *
 967  * This is the helper to open a file from kernelspace if you really
 968  * have to.  But in generally you should not do this, so please move
 969  * along, nothing to see here..
 970  */
 971 struct file *filp_open(const char *filename, int flags, umode_t mode)
 972 {
 973         struct filename name = {.name = filename};
 974         return file_open_name(&name, flags, mode);
 975 }
 976 EXPORT_SYMBOL(filp_open);
 977
 978 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
 979                             const char *filename, int flags)
 980 {
 981         struct open_flags op;
 982         int err = build_open_flags(flags, 0, &op);
 983         if (err)
 984                 return ERR_PTR(err);
 985         if (flags & O_CREAT)
 986                 return ERR_PTR(-EINVAL);
 987         if (!filename && (flags & O_DIRECTORY))
 988                 if (!dentry->d_inode->i_op->lookup)
 989                         return ERR_PTR(-ENOTDIR);
 990         return do_file_open_root(dentry, mnt, filename, &op);
 991 }
 992 EXPORT_SYMBOL(file_open_root);
 993
 994 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
 995 {
 996         struct open_flags op;
 997         int fd = build_open_flags(flags, mode, &op);
 998         struct filename *tmp;
 999
1000         if (fd)
1001                 return fd;
1002
1003         tmp = getname(filename);
1004         if (IS_ERR(tmp))
1005                 return PTR_ERR(tmp);
1006
1007         fd = get_unused_fd_flags(flags);
1008         if (fd >= 0) {
1009                 struct file *f = do_filp_open(dfd, tmp, &op);
1010                 if (IS_ERR(f)) {
1011                         put_unused_fd(fd);
1012                         fd = PTR_ERR(f);
1013                 } else {
1014                         fsnotify_open(f);
1015                         fd_install(fd, f);
1016                 }
1017         }
1018         putname(tmp);
1019         return fd;
1020 }
1021
1022 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1023 {
1024         if (force_o_largefile())
1025                 flags |= O_LARGEFILE;
1026
1027         return do_sys_open(AT_FDCWD, filename, flags, mode);
1028 }
1029
1030 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1031                 umode_t, mode)
1032 {
1033         if (force_o_largefile())
1034                 flags |= O_LARGEFILE;
1035
1036         return do_sys_open(dfd, filename, flags, mode);
1037 }
1038
1039 #ifndef __alpha__
1040
1041 /*
1042  * For backward compatibility?  Maybe this should be moved
1043  * into arch/i386 instead?
1044  */
1045 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1046 {
1047         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1048 }
1049
1050 #endif
1051
1052 /*
1053  * "id" is the POSIX thread ID. We use the
1054  * files pointer for this..
1055  */
1056 int filp_close(struct file *filp, fl_owner_t id)
1057 {
1058         int retval = 0;
1059
1060         if (!file_count(filp)) {
1061                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1062                 return 0;
1063         }
1064
1065         if (filp->f_op->flush)
1066                 retval = filp->f_op->flush(filp, id);
1067
1068         if (likely(!(filp->f_mode & FMODE_PATH))) {
1069                 dnotify_flush(filp, id);
1070                 locks_remove_posix(filp, id);
1071         }
1072         fput(filp);
1073         return retval;
1074 }
1075
1076 EXPORT_SYMBOL(filp_close);
1077
1078 /*
1079  * Careful here! We test whether the file pointer is NULL before
1080  * releasing the fd. This ensures that one clone task can't release
1081  * an fd while another clone is opening it.
1082  */
1083 SYSCALL_DEFINE1(close, unsigned int, fd)
1084 {
1085         int retval = __close_fd(current->files, fd);
1086
1087         /* can't restart close syscall because file table entry was cleared */
1088         if (unlikely(retval == -ERESTARTSYS ||
1089                      retval == -ERESTARTNOINTR ||
1090                      retval == -ERESTARTNOHAND ||
1091                      retval == -ERESTART_RESTARTBLOCK))
1092                 retval = -EINTR;
1093
1094         return retval;
1095 }
1096 EXPORT_SYMBOL(sys_close);
1097
1098 /*
1099  * This routine simulates a hangup on the tty, to arrange that users
1100  * are given clean terminals at login time.
1101  */
1102 SYSCALL_DEFINE0(vhangup)
1103 {
1104         if (capable(CAP_SYS_TTY_CONFIG)) {
1105                 tty_vhangup_self();
1106                 return 0;
1107         }
1108         return -EPERM;
1109 }
1110
1111 /*
1112  * Called when an inode is about to be open.
1113  * We use this to disallow opening large files on 32bit systems if
1114  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1115  * on this flag in sys_open.
1116  */
1117 int generic_file_open(struct inode * inode, struct file * filp)
1118 {
1119         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1120                 return -EOVERFLOW;
1121         return 0;
1122 }
1123
1124 EXPORT_SYMBOL(generic_file_open);
1125
1126 /*
1127  * This is used by subsystems that don't want seekable
1128  * file descriptors. The function is not supposed to ever fail, the only
1129  * reason it returns an 'int' and not 'void' is so that it can be plugged
1130  * directly into file_operations structure.
1131  */
1132 int nonseekable_open(struct inode *inode, struct file *filp)
1133 {
1134         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1135         return 0;
1136 }
1137
1138 EXPORT_SYMBOL(nonseekable_open);