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, and file capabilities on truncate too */
  55         ret = dentry_needs_remove_privs(dentry);
  56         if (ret < 0)
  57                 return ret;
  58         if (ret)
  59                 newattrs.ia_valid |= ret | ATTR_FORCE;
  60
  61         inode_lock(dentry->d_inode);
  62         /* Note any delegations or leases have already been broken: */
  63         ret = notify_change(dentry, &newattrs, NULL);
  64         inode_unlock(dentry->d_inode);
  65         return ret;
  66 }
  67
  68 long vfs_truncate(const struct path *path, loff_t length)
  69 {
  70         struct inode *inode;
  71         struct dentry *upperdentry;
  72         long error;
  73
  74         inode = path->dentry->d_inode;
  75
  76         /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
  77         if (S_ISDIR(inode->i_mode))
  78                 return -EISDIR;
  79         if (!S_ISREG(inode->i_mode))
  80                 return -EINVAL;
  81
  82         error = mnt_want_write(path->mnt);
  83         if (error)
  84                 goto out;
  85
  86         error = inode_permission(inode, MAY_WRITE);
  87         if (error)
  88                 goto mnt_drop_write_and_out;
  89
  90         error = -EPERM;
  91         if (IS_APPEND(inode))
  92                 goto mnt_drop_write_and_out;
  93
  94         /*
  95          * If this is an overlayfs then do as if opening the file so we get
  96          * write access on the upper inode, not on the overlay inode.  For
  97          * non-overlay filesystems d_real() is an identity function.
  98          */
  99         upperdentry = d_real(path->dentry, NULL, O_WRONLY);
 100         error = PTR_ERR(upperdentry);
 101         if (IS_ERR(upperdentry))
 102                 goto mnt_drop_write_and_out;
 103
 104         error = get_write_access(upperdentry->d_inode);
 105         if (error)
 106                 goto mnt_drop_write_and_out;
 107
 108         /*
 109          * Make sure that there are no leases.  get_write_access() protects
 110          * against the truncate racing with a lease-granting setlease().
 111          */
 112         error = break_lease(inode, O_WRONLY);
 113         if (error)
 114                 goto put_write_and_out;
 115
 116         error = locks_verify_truncate(inode, NULL, length);
 117         if (!error)
 118                 error = security_path_truncate(path);
 119         if (!error)
 120                 error = do_truncate(path->dentry, length, 0, NULL);
 121
 122 put_write_and_out:
 123         put_write_access(upperdentry->d_inode);
 124 mnt_drop_write_and_out:
 125         mnt_drop_write(path->mnt);
 126 out:
 127         return error;
 128 }
 129 EXPORT_SYMBOL_GPL(vfs_truncate);
 130
 131 static long do_sys_truncate(const char __user *pathname, loff_t length)
 132 {
 133         unsigned int lookup_flags = LOOKUP_FOLLOW;
 134         struct path path;
 135         int error;
 136
 137         if (length < 0) /* sorry, but loff_t says... */
 138                 return -EINVAL;
 139
 140 retry:
 141         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 142         if (!error) {
 143                 error = vfs_truncate(&path, length);
 144                 path_put(&path);
 145         }
 146         if (retry_estale(error, lookup_flags)) {
 147                 lookup_flags |= LOOKUP_REVAL;
 148                 goto retry;
 149         }
 150         return error;
 151 }
 152
 153 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
 154 {
 155         return do_sys_truncate(path, length);
 156 }
 157
 158 #ifdef CONFIG_COMPAT
 159 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
 160 {
 161         return do_sys_truncate(path, length);
 162 }
 163 #endif
 164
 165 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
 166 {
 167         struct inode *inode;
 168         struct dentry *dentry;
 169         struct fd f;
 170         int error;
 171
 172         error = -EINVAL;
 173         if (length < 0)
 174                 goto out;
 175         error = -EBADF;
 176         f = fdget(fd);
 177         if (!f.file)
 178                 goto out;
 179
 180         /* explicitly opened as large or we are on 64-bit box */
 181         if (f.file->f_flags & O_LARGEFILE)
 182                 small = 0;
 183
 184         dentry = f.file->f_path.dentry;
 185         inode = dentry->d_inode;
 186         error = -EINVAL;
 187         if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
 188                 goto out_putf;
 189
 190         error = -EINVAL;
 191         /* Cannot ftruncate over 2^31 bytes without large file support */
 192         if (small && length > MAX_NON_LFS)
 193                 goto out_putf;
 194
 195         error = -EPERM;
 196         if (IS_APPEND(inode))
 197                 goto out_putf;
 198
 199         sb_start_write(inode->i_sb);
 200         error = locks_verify_truncate(inode, f.file, length);
 201         if (!error)
 202                 error = security_path_truncate(&f.file->f_path);
 203         if (!error)
 204                 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
 205         sb_end_write(inode->i_sb);
 206 out_putf:
 207         fdput(f);
 208 out:
 209         return error;
 210 }
 211
 212 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
 213 {
 214         return do_sys_ftruncate(fd, length, 1);
 215 }
 216
 217 #ifdef CONFIG_COMPAT
 218 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
 219 {
 220         return do_sys_ftruncate(fd, length, 1);
 221 }
 222 #endif
 223
 224 /* LFS versions of truncate are only needed on 32 bit machines */
 225 #if BITS_PER_LONG == 32
 226 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
 227 {
 228         return do_sys_truncate(path, length);
 229 }
 230
 231 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
 232 {
 233         return do_sys_ftruncate(fd, length, 0);
 234 }
 235 #endif /* BITS_PER_LONG == 32 */
 236
 237
 238 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 239 {
 240         struct inode *inode = file_inode(file);
 241         long ret;
 242
 243         if (offset < 0 || len <= 0)
 244                 return -EINVAL;
 245
 246         /* Return error if mode is not supported */
 247         if (mode & ~FALLOC_FL_SUPPORTED_MASK)
 248                 return -EOPNOTSUPP;
 249
 250         /* Punch hole and zero range are mutually exclusive */
 251         if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
 252             (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
 253                 return -EOPNOTSUPP;
 254
 255         /* Punch hole must have keep size set */
 256         if ((mode & FALLOC_FL_PUNCH_HOLE) &&
 257             !(mode & FALLOC_FL_KEEP_SIZE))
 258                 return -EOPNOTSUPP;
 259
 260         /* Collapse range should only be used exclusively. */
 261         if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
 262             (mode & ~FALLOC_FL_COLLAPSE_RANGE))
 263                 return -EINVAL;
 264
 265         /* Insert range should only be used exclusively. */
 266         if ((mode & FALLOC_FL_INSERT_RANGE) &&
 267             (mode & ~FALLOC_FL_INSERT_RANGE))
 268                 return -EINVAL;
 269
 270         /* Unshare range should only be used with allocate mode. */
 271         if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
 272             (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
 273                 return -EINVAL;
 274
 275         if (!(file->f_mode & FMODE_WRITE))
 276                 return -EBADF;
 277
 278         /*
 279          * We can only allow pure fallocate on append only files
 280          */
 281         if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
 282                 return -EPERM;
 283
 284         if (IS_IMMUTABLE(inode))
 285                 return -EPERM;
 286
 287         /*
 288          * We cannot allow any fallocate operation on an active swapfile
 289          */
 290         if (IS_SWAPFILE(inode))
 291                 return -ETXTBSY;
 292
 293         /*
 294          * Revalidate the write permissions, in case security policy has
 295          * changed since the files were opened.
 296          */
 297         ret = security_file_permission(file, MAY_WRITE);
 298         if (ret)
 299                 return ret;
 300
 301         if (S_ISFIFO(inode->i_mode))
 302                 return -ESPIPE;
 303
 304         /*
 305          * Let individual file system decide if it supports preallocation
 306          * for directories or not.
 307          */
 308         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
 309             !S_ISBLK(inode->i_mode))
 310                 return -ENODEV;
 311
 312         /* Check for wrap through zero too */
 313         if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
 314                 return -EFBIG;
 315
 316         if (!file->f_op->fallocate)
 317                 return -EOPNOTSUPP;
 318
 319         sb_start_write(inode->i_sb);
 320         ret = file->f_op->fallocate(file, mode, offset, len);
 321
 322         /*
 323          * Create inotify and fanotify events.
 324          *
 325          * To keep the logic simple always create events if fallocate succeeds.
 326          * This implies that events are even created if the file size remains
 327          * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
 328          */
 329         if (ret == 0)
 330                 fsnotify_modify(file);
 331
 332         sb_end_write(inode->i_sb);
 333         return ret;
 334 }
 335 EXPORT_SYMBOL_GPL(vfs_fallocate);
 336
 337 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
 338 {
 339         struct fd f = fdget(fd);
 340         int error = -EBADF;
 341
 342         if (f.file) {
 343                 error = vfs_fallocate(f.file, mode, offset, len);
 344                 fdput(f);
 345         }
 346         return error;
 347 }
 348
 349 /*
 350  * access() needs to use the real uid/gid, not the effective uid/gid.
 351  * We do this by temporarily clearing all FS-related capabilities and
 352  * switching the fsuid/fsgid around to the real ones.
 353  */
 354 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 355 {
 356         const struct cred *old_cred;
 357         struct cred *override_cred;
 358         struct path path;
 359         struct inode *inode;
 360         int res;
 361         unsigned int lookup_flags = LOOKUP_FOLLOW;
 362
 363         if (mode & ~S_IRWXO)    /* where's F_OK, X_OK, W_OK, R_OK? */
 364                 return -EINVAL;
 365
 366         override_cred = prepare_creds();
 367         if (!override_cred)
 368                 return -ENOMEM;
 369
 370         override_cred->fsuid = override_cred->uid;
 371         override_cred->fsgid = override_cred->gid;
 372
 373         if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 374                 /* Clear the capabilities if we switch to a non-root user */
 375                 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
 376                 if (!uid_eq(override_cred->uid, root_uid))
 377                         cap_clear(override_cred->cap_effective);
 378                 else
 379                         override_cred->cap_effective =
 380                                 override_cred->cap_permitted;
 381         }
 382
 383         old_cred = override_creds(override_cred);
 384 retry:
 385         res = user_path_at(dfd, filename, lookup_flags, &path);
 386         if (res)
 387                 goto out;
 388
 389         inode = d_backing_inode(path.dentry);
 390
 391         if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
 392                 /*
 393                  * MAY_EXEC on regular files is denied if the fs is mounted
 394                  * with the "noexec" flag.
 395                  */
 396                 res = -EACCES;
 397                 if (path_noexec(&path))
 398                         goto out_path_release;
 399         }
 400
 401         res = inode_permission(inode, mode | MAY_ACCESS);
 402         /* SuS v2 requires we report a read only fs too */
 403         if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
 404                 goto out_path_release;
 405         /*
 406          * This is a rare case where using __mnt_is_readonly()
 407          * is OK without a mnt_want/drop_write() pair.  Since
 408          * no actual write to the fs is performed here, we do
 409          * not need to telegraph to that to anyone.
 410          *
 411          * By doing this, we accept that this access is
 412          * inherently racy and know that the fs may change
 413          * state before we even see this result.
 414          */
 415         if (__mnt_is_readonly(path.mnt))
 416                 res = -EROFS;
 417
 418 out_path_release:
 419         path_put(&path);
 420         if (retry_estale(res, lookup_flags)) {
 421                 lookup_flags |= LOOKUP_REVAL;
 422                 goto retry;
 423         }
 424 out:
 425         revert_creds(old_cred);
 426         put_cred(override_cred);
 427         return res;
 428 }
 429
 430 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
 431 {
 432         return sys_faccessat(AT_FDCWD, filename, mode);
 433 }
 434
 435 SYSCALL_DEFINE1(chdir, const char __user *, filename)
 436 {
 437         struct path path;
 438         int error;
 439         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 440 retry:
 441         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 442         if (error)
 443                 goto out;
 444
 445         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 446         if (error)
 447                 goto dput_and_out;
 448
 449         set_fs_pwd(current->fs, &path);
 450
 451 dput_and_out:
 452         path_put(&path);
 453         if (retry_estale(error, lookup_flags)) {
 454                 lookup_flags |= LOOKUP_REVAL;
 455                 goto retry;
 456         }
 457 out:
 458         return error;
 459 }
 460
 461 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
 462 {
 463         struct fd f = fdget_raw(fd);
 464         struct inode *inode;
 465         int error = -EBADF;
 466
 467         error = -EBADF;
 468         if (!f.file)
 469                 goto out;
 470
 471         inode = file_inode(f.file);
 472
 473         error = -ENOTDIR;
 474         if (!S_ISDIR(inode->i_mode))
 475                 goto out_putf;
 476
 477         error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
 478         if (!error)
 479                 set_fs_pwd(current->fs, &f.file->f_path);
 480 out_putf:
 481         fdput(f);
 482 out:
 483         return error;
 484 }
 485
 486 SYSCALL_DEFINE1(chroot, const char __user *, filename)
 487 {
 488         struct path path;
 489         int error;
 490         unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 491 retry:
 492         error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
 493         if (error)
 494                 goto out;
 495
 496         error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
 497         if (error)
 498                 goto dput_and_out;
 499
 500         error = -EPERM;
 501         if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
 502                 goto dput_and_out;
 503         error = security_path_chroot(&path);
 504         if (error)
 505                 goto dput_and_out;
 506
 507         set_fs_root(current->fs, &path);
 508         error = 0;
 509 dput_and_out:
 510         path_put(&path);
 511         if (retry_estale(error, lookup_flags)) {
 512                 lookup_flags |= LOOKUP_REVAL;
 513                 goto retry;
 514         }
 515 out:
 516         return error;
 517 }
 518
 519 static int chmod_common(const struct path *path, umode_t mode)
 520 {
 521         struct inode *inode = path->dentry->d_inode;
 522         struct inode *delegated_inode = NULL;
 523         struct iattr newattrs;
 524         int error;
 525
 526         error = mnt_want_write(path->mnt);
 527         if (error)
 528                 return error;
 529 retry_deleg:
 530         inode_lock(inode);
 531         error = security_path_chmod(path, mode);
 532         if (error)
 533                 goto out_unlock;
 534         newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 535         newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
 536         error = notify_change(path->dentry, &newattrs, &delegated_inode);
 537 out_unlock:
 538         inode_unlock(inode);
 539         if (delegated_inode) {
 540                 error = break_deleg_wait(&delegated_inode);
 541                 if (!error)
 542                         goto retry_deleg;
 543         }
 544         mnt_drop_write(path->mnt);
 545         return error;
 546 }
 547
 548 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
 549 {
 550         struct fd f = fdget(fd);
 551         int err = -EBADF;
 552
 553         if (f.file) {
 554                 audit_file(f.file);
 555                 err = chmod_common(&f.file->f_path, mode);
 556                 fdput(f);
 557         }
 558         return err;
 559 }
 560
 561 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
 562 {
 563         struct path path;
 564         int error;
 565         unsigned int lookup_flags = LOOKUP_FOLLOW;
 566 retry:
 567         error = user_path_at(dfd, filename, lookup_flags, &path);
 568         if (!error) {
 569                 error = chmod_common(&path, mode);
 570                 path_put(&path);
 571                 if (retry_estale(error, lookup_flags)) {
 572                         lookup_flags |= LOOKUP_REVAL;
 573                         goto retry;
 574                 }
 575         }
 576         return error;
 577 }
 578
 579 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
 580 {
 581         return sys_fchmodat(AT_FDCWD, filename, mode);
 582 }
 583
 584 static int chown_common(const struct path *path, uid_t user, gid_t group)
 585 {
 586         struct inode *inode = path->dentry->d_inode;
 587         struct inode *delegated_inode = NULL;
 588         int error;
 589         struct iattr newattrs;
 590         kuid_t uid;
 591         kgid_t gid;
 592
 593         uid = make_kuid(current_user_ns(), user);
 594         gid = make_kgid(current_user_ns(), group);
 595
 596 retry_deleg:
 597         newattrs.ia_valid =  ATTR_CTIME;
 598         if (user != (uid_t) -1) {
 599                 if (!uid_valid(uid))
 600                         return -EINVAL;
 601                 newattrs.ia_valid |= ATTR_UID;
 602                 newattrs.ia_uid = uid;
 603         }
 604         if (group != (gid_t) -1) {
 605                 if (!gid_valid(gid))
 606                         return -EINVAL;
 607                 newattrs.ia_valid |= ATTR_GID;
 608                 newattrs.ia_gid = gid;
 609         }
 610         if (!S_ISDIR(inode->i_mode))
 611                 newattrs.ia_valid |=
 612                         ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 613         inode_lock(inode);
 614         error = security_path_chown(path, uid, gid);
 615         if (!error)
 616                 error = notify_change(path->dentry, &newattrs, &delegated_inode);
 617         inode_unlock(inode);
 618         if (delegated_inode) {
 619                 error = break_deleg_wait(&delegated_inode);
 620                 if (!error)
 621                         goto retry_deleg;
 622         }
 623         return error;
 624 }
 625
 626 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
 627                 gid_t, group, int, flag)
 628 {
 629         struct path path;
 630         int error = -EINVAL;
 631         int lookup_flags;
 632
 633         if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 634                 goto out;
 635
 636         lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
 637         if (flag & AT_EMPTY_PATH)
 638                 lookup_flags |= LOOKUP_EMPTY;
 639 retry:
 640         error = user_path_at(dfd, filename, lookup_flags, &path);
 641         if (error)
 642                 goto out;
 643         error = mnt_want_write(path.mnt);
 644         if (error)
 645                 goto out_release;
 646         error = chown_common(&path, user, group);
 647         mnt_drop_write(path.mnt);
 648 out_release:
 649         path_put(&path);
 650         if (retry_estale(error, lookup_flags)) {
 651                 lookup_flags |= LOOKUP_REVAL;
 652                 goto retry;
 653         }
 654 out:
 655         return error;
 656 }
 657
 658 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
 659 {
 660         return sys_fchownat(AT_FDCWD, filename, user, group, 0);
 661 }
 662
 663 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
 664 {
 665         return sys_fchownat(AT_FDCWD, filename, user, group,
 666                             AT_SYMLINK_NOFOLLOW);
 667 }
 668
 669 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
 670 {
 671         struct fd f = fdget(fd);
 672         int error = -EBADF;
 673
 674         if (!f.file)
 675                 goto out;
 676
 677         error = mnt_want_write_file(f.file);
 678         if (error)
 679                 goto out_fput;
 680         audit_file(f.file);
 681         error = chown_common(&f.file->f_path, user, group);
 682         mnt_drop_write_file(f.file);
 683 out_fput:
 684         fdput(f);
 685 out:
 686         return error;
 687 }
 688
 689 int open_check_o_direct(struct file *f)
 690 {
 691         /* NB: we're sure to have correct a_ops only after f_op->open */
 692         if (f->f_flags & O_DIRECT) {
 693                 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
 694                         return -EINVAL;
 695         }
 696         return 0;
 697 }
 698
 699 static int do_dentry_open(struct file *f,
 700                           struct inode *inode,
 701                           int (*open)(struct inode *, struct file *),
 702                           const struct cred *cred)
 703 {
 704         static const struct file_operations empty_fops = {};
 705         int error;
 706
 707         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
 708                                 FMODE_PREAD | FMODE_PWRITE;
 709
 710         path_get(&f->f_path);
 711         f->f_inode = inode;
 712         f->f_mapping = inode->i_mapping;
 713
 714         if (unlikely(f->f_flags & O_PATH)) {
 715                 f->f_mode = FMODE_PATH;
 716                 f->f_op = &empty_fops;
 717                 return 0;
 718         }
 719
 720         if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
 721                 error = get_write_access(inode);
 722                 if (unlikely(error))
 723                         goto cleanup_file;
 724                 error = __mnt_want_write(f->f_path.mnt);
 725                 if (unlikely(error)) {
 726                         put_write_access(inode);
 727                         goto cleanup_file;
 728                 }
 729                 f->f_mode |= FMODE_WRITER;
 730         }
 731
 732         /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
 733         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
 734                 f->f_mode |= FMODE_ATOMIC_POS;
 735
 736         f->f_op = fops_get(inode->i_fop);
 737         if (unlikely(WARN_ON(!f->f_op))) {
 738                 error = -ENODEV;
 739                 goto cleanup_all;
 740         }
 741
 742         error = security_file_open(f, cred);
 743         if (error)
 744                 goto cleanup_all;
 745
 746         error = break_lease(locks_inode(f), f->f_flags);
 747         if (error)
 748                 goto cleanup_all;
 749
 750         if (!open)
 751                 open = f->f_op->open;
 752         if (open) {
 753                 error = open(inode, f);
 754                 if (error)
 755                         goto cleanup_all;
 756         }
 757         if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 758                 i_readcount_inc(inode);
 759         if ((f->f_mode & FMODE_READ) &&
 760              likely(f->f_op->read || f->f_op->read_iter))
 761                 f->f_mode |= FMODE_CAN_READ;
 762         if ((f->f_mode & FMODE_WRITE) &&
 763              likely(f->f_op->write || f->f_op->write_iter))
 764                 f->f_mode |= FMODE_CAN_WRITE;
 765
 766         f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
 767
 768         file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
 769
 770         return 0;
 771
 772 cleanup_all:
 773         fops_put(f->f_op);
 774         if (f->f_mode & FMODE_WRITER) {
 775                 put_write_access(inode);
 776                 __mnt_drop_write(f->f_path.mnt);
 777         }
 778 cleanup_file:
 779         path_put(&f->f_path);
 780         f->f_path.mnt = NULL;
 781         f->f_path.dentry = NULL;
 782         f->f_inode = NULL;
 783         return error;
 784 }
 785
 786 /**
 787  * finish_open - finish opening a file
 788  * @file: file pointer
 789  * @dentry: pointer to dentry
 790  * @open: open callback
 791  * @opened: state of open
 792  *
 793  * This can be used to finish opening a file passed to i_op->atomic_open().
 794  *
 795  * If the open callback is set to NULL, then the standard f_op->open()
 796  * filesystem callback is substituted.
 797  *
 798  * NB: the dentry reference is _not_ consumed.  If, for example, the dentry is
 799  * the return value of d_splice_alias(), then the caller needs to perform dput()
 800  * on it after finish_open().
 801  *
 802  * On successful return @file is a fully instantiated open file.  After this, if
 803  * an error occurs in ->atomic_open(), it needs to clean up with fput().
 804  *
 805  * Returns zero on success or -errno if the open failed.
 806  */
 807 int finish_open(struct file *file, struct dentry *dentry,
 808                 int (*open)(struct inode *, struct file *),
 809                 int *opened)
 810 {
 811         int error;
 812         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 813
 814         file->f_path.dentry = dentry;
 815         error = do_dentry_open(file, d_backing_inode(dentry), open,
 816                                current_cred());
 817         if (!error)
 818                 *opened |= FILE_OPENED;
 819
 820         return error;
 821 }
 822 EXPORT_SYMBOL(finish_open);
 823
 824 /**
 825  * finish_no_open - finish ->atomic_open() without opening the file
 826  *
 827  * @file: file pointer
 828  * @dentry: dentry or NULL (as returned from ->lookup())
 829  *
 830  * This can be used to set the result of a successful lookup in ->atomic_open().
 831  *
 832  * NB: unlike finish_open() this function does consume the dentry reference and
 833  * the caller need not dput() it.
 834  *
 835  * Returns "1" which must be the return value of ->atomic_open() after having
 836  * called this function.
 837  */
 838 int finish_no_open(struct file *file, struct dentry *dentry)
 839 {
 840         file->f_path.dentry = dentry;
 841         return 1;
 842 }
 843 EXPORT_SYMBOL(finish_no_open);
 844
 845 char *file_path(struct file *filp, char *buf, int buflen)
 846 {
 847         return d_path(&filp->f_path, buf, buflen);
 848 }
 849 EXPORT_SYMBOL(file_path);
 850
 851 /**
 852  * vfs_open - open the file at the given path
 853  * @path: path to open
 854  * @file: newly allocated file with f_flag initialized
 855  * @cred: credentials to use
 856  */
 857 int vfs_open(const struct path *path, struct file *file,
 858              const struct cred *cred)
 859 {
 860         struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags);
 861
 862         if (IS_ERR(dentry))
 863                 return PTR_ERR(dentry);
 864
 865         file->f_path = *path;
 866         return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
 867 }
 868
 869 struct file *dentry_open(const struct path *path, int flags,
 870                          const struct cred *cred)
 871 {
 872         int error;
 873         struct file *f;
 874
 875         validate_creds(cred);
 876
 877         /* We must always pass in a valid mount pointer. */
 878         BUG_ON(!path->mnt);
 879
 880         f = get_empty_filp();
 881         if (!IS_ERR(f)) {
 882                 f->f_flags = flags;
 883                 error = vfs_open(path, f, cred);
 884                 if (!error) {
 885                         /* from now on we need fput() to dispose of f */
 886                         error = open_check_o_direct(f);
 887                         if (error) {
 888                                 fput(f);
 889                                 f = ERR_PTR(error);
 890                         }
 891                 } else {
 892                         put_filp(f);
 893                         f = ERR_PTR(error);
 894                 }
 895         }
 896         return f;
 897 }
 898 EXPORT_SYMBOL(dentry_open);
 899
 900 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
 901 {
 902         int lookup_flags = 0;
 903         int acc_mode = ACC_MODE(flags);
 904
 905         /*
 906          * Clear out all open flags we don't know about so that we don't report
 907          * them in fcntl(F_GETFD) or similar interfaces.
 908          */
 909         flags &= VALID_OPEN_FLAGS;
 910
 911         if (flags & (O_CREAT | __O_TMPFILE))
 912                 op->mode = (mode & S_IALLUGO) | S_IFREG;
 913         else
 914                 op->mode = 0;
 915
 916         /* Must never be set by userspace */
 917         flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
 918
 919         /*
 920          * O_SYNC is implemented as __O_SYNC|O_DSYNC.  As many places only
 921          * check for O_DSYNC if the need any syncing at all we enforce it's
 922          * always set instead of having to deal with possibly weird behaviour
 923          * for malicious applications setting only __O_SYNC.
 924          */
 925         if (flags & __O_SYNC)
 926                 flags |= O_DSYNC;
 927
 928         if (flags & __O_TMPFILE) {
 929                 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
 930                         return -EINVAL;
 931                 if (!(acc_mode & MAY_WRITE))
 932                         return -EINVAL;
 933         } else if (flags & O_PATH) {
 934                 /*
 935                  * If we have O_PATH in the open flag. Then we
 936                  * cannot have anything other than the below set of flags
 937                  */
 938                 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
 939                 acc_mode = 0;
 940         }
 941
 942         op->open_flag = flags;
 943
 944         /* O_TRUNC implies we need access checks for write permissions */
 945         if (flags & O_TRUNC)
 946                 acc_mode |= MAY_WRITE;
 947
 948         /* Allow the LSM permission hook to distinguish append
 949            access from general write access. */
 950         if (flags & O_APPEND)
 951                 acc_mode |= MAY_APPEND;
 952
 953         op->acc_mode = acc_mode;
 954
 955         op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
 956
 957         if (flags & O_CREAT) {
 958                 op->intent |= LOOKUP_CREATE;
 959                 if (flags & O_EXCL)
 960                         op->intent |= LOOKUP_EXCL;
 961         }
 962
 963         if (flags & O_DIRECTORY)
 964                 lookup_flags |= LOOKUP_DIRECTORY;
 965         if (!(flags & O_NOFOLLOW))
 966                 lookup_flags |= LOOKUP_FOLLOW;
 967         op->lookup_flags = lookup_flags;
 968         return 0;
 969 }
 970
 971 /**
 972  * file_open_name - open file and return file pointer
 973  *
 974  * @name:       struct filename containing path to open
 975  * @flags:      open flags as per the open(2) second argument
 976  * @mode:       mode for the new file if O_CREAT is set, else ignored
 977  *
 978  * This is the helper to open a file from kernelspace if you really
 979  * have to.  But in generally you should not do this, so please move
 980  * along, nothing to see here..
 981  */
 982 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
 983 {
 984         struct open_flags op;
 985         int err = build_open_flags(flags, mode, &op);
 986         return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
 987 }
 988
 989 /**
 990  * filp_open - open file and return file pointer
 991  *
 992  * @filename:   path to open
 993  * @flags:      open flags as per the open(2) second argument
 994  * @mode:       mode for the new file if O_CREAT is set, else ignored
 995  *
 996  * This is the helper to open a file from kernelspace if you really
 997  * have to.  But in generally you should not do this, so please move
 998  * along, nothing to see here..
 999  */
1000 struct file *filp_open(const char *filename, int flags, umode_t mode)
1001 {
1002         struct filename *name = getname_kernel(filename);
1003         struct file *file = ERR_CAST(name);
1004
1005         if (!IS_ERR(name)) {
1006                 file = file_open_name(name, flags, mode);
1007                 putname(name);
1008         }
1009         return file;
1010 }
1011 EXPORT_SYMBOL(filp_open);
1012
1013 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1014                             const char *filename, int flags, umode_t mode)
1015 {
1016         struct open_flags op;
1017         int err = build_open_flags(flags, mode, &op);
1018         if (err)
1019                 return ERR_PTR(err);
1020         return do_file_open_root(dentry, mnt, filename, &op);
1021 }
1022 EXPORT_SYMBOL(file_open_root);
1023
1024 struct file *filp_clone_open(struct file *oldfile)
1025 {
1026         struct file *file;
1027         int retval;
1028
1029         file = get_empty_filp();
1030         if (IS_ERR(file))
1031                 return file;
1032
1033         file->f_flags = oldfile->f_flags;
1034         retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1035         if (retval) {
1036                 put_filp(file);
1037                 return ERR_PTR(retval);
1038         }
1039
1040         return file;
1041 }
1042 EXPORT_SYMBOL(filp_clone_open);
1043
1044 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1045 {
1046         struct open_flags op;
1047         int fd = build_open_flags(flags, mode, &op);
1048         struct filename *tmp;
1049
1050         if (fd)
1051                 return fd;
1052
1053         tmp = getname(filename);
1054         if (IS_ERR(tmp))
1055                 return PTR_ERR(tmp);
1056
1057         fd = get_unused_fd_flags(flags);
1058         if (fd >= 0) {
1059                 struct file *f = do_filp_open(dfd, tmp, &op);
1060                 if (IS_ERR(f)) {
1061                         put_unused_fd(fd);
1062                         fd = PTR_ERR(f);
1063                 } else {
1064                         fsnotify_open(f);
1065                         fd_install(fd, f);
1066                 }
1067         }
1068         putname(tmp);
1069         return fd;
1070 }
1071
1072 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1073 {
1074         if (force_o_largefile())
1075                 flags |= O_LARGEFILE;
1076
1077         return do_sys_open(AT_FDCWD, filename, flags, mode);
1078 }
1079
1080 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1081                 umode_t, mode)
1082 {
1083         if (force_o_largefile())
1084                 flags |= O_LARGEFILE;
1085
1086         return do_sys_open(dfd, filename, flags, mode);
1087 }
1088
1089 #ifndef __alpha__
1090
1091 /*
1092  * For backward compatibility?  Maybe this should be moved
1093  * into arch/i386 instead?
1094  */
1095 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1096 {
1097         return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1098 }
1099
1100 #endif
1101
1102 /*
1103  * "id" is the POSIX thread ID. We use the
1104  * files pointer for this..
1105  */
1106 int filp_close(struct file *filp, fl_owner_t id)
1107 {
1108         int retval = 0;
1109
1110         if (!file_count(filp)) {
1111                 printk(KERN_ERR "VFS: Close: file count is 0\n");
1112                 return 0;
1113         }
1114
1115         if (filp->f_op->flush)
1116                 retval = filp->f_op->flush(filp, id);
1117
1118         if (likely(!(filp->f_mode & FMODE_PATH))) {
1119                 dnotify_flush(filp, id);
1120                 locks_remove_posix(filp, id);
1121         }
1122         fput(filp);
1123         return retval;
1124 }
1125
1126 EXPORT_SYMBOL(filp_close);
1127
1128 /*
1129  * Careful here! We test whether the file pointer is NULL before
1130  * releasing the fd. This ensures that one clone task can't release
1131  * an fd while another clone is opening it.
1132  */
1133 SYSCALL_DEFINE1(close, unsigned int, fd)
1134 {
1135         int retval = __close_fd(current->files, fd);
1136
1137         /* can't restart close syscall because file table entry was cleared */
1138         if (unlikely(retval == -ERESTARTSYS ||
1139                      retval == -ERESTARTNOINTR ||
1140                      retval == -ERESTARTNOHAND ||
1141                      retval == -ERESTART_RESTARTBLOCK))
1142                 retval = -EINTR;
1143
1144         return retval;
1145 }
1146 EXPORT_SYMBOL(sys_close);
1147
1148 /*
1149  * This routine simulates a hangup on the tty, to arrange that users
1150  * are given clean terminals at login time.
1151  */
1152 SYSCALL_DEFINE0(vhangup)
1153 {
1154         if (capable(CAP_SYS_TTY_CONFIG)) {
1155                 tty_vhangup_self();
1156                 return 0;
1157         }
1158         return -EPERM;
1159 }
1160
1161 /*
1162  * Called when an inode is about to be open.
1163  * We use this to disallow opening large files on 32bit systems if
1164  * the caller didn't specify O_LARGEFILE.  On 64bit systems we force
1165  * on this flag in sys_open.
1166  */
1167 int generic_file_open(struct inode * inode, struct file * filp)
1168 {
1169         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1170                 return -EOVERFLOW;
1171         return 0;
1172 }
1173
1174 EXPORT_SYMBOL(generic_file_open);
1175
1176 /*
1177  * This is used by subsystems that don't want seekable
1178  * file descriptors. The function is not supposed to ever fail, the only
1179  * reason it returns an 'int' and not 'void' is so that it can be plugged
1180  * directly into file_operations structure.
1181  */
1182 int nonseekable_open(struct inode *inode, struct file *filp)
1183 {
1184         filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1185         return 0;
1186 }
1187
1188 EXPORT_SYMBOL(nonseekable_open);