= Rust U4Pak
Mathias Panzenböck grosser.meister.morti@gmx.net; L. Sprengel l.sprengel@pm.me
v1.4.0
:icons: font
:sectnums:
:toc: left
:toclevels: 3
image:https://img.shields.io/github/actions/workflow/status/panzi/rust-u4pak/rust.yml?branch=main["Build status", link=https://github.com/panzi/rust-u4pak/actions/workflows/rust.yml]
image:https://img.shields.io/github/v/release/panzi/rust-u4pak["Release",link="https://github.com/panzi/rust-u4pak/releases"]
image:https://img.shields.io/github/license/panzi/rust-u4pak?cachebust=1["License",link="https://github.com/panzi/rust-u4pak/blob/main/LICENSE.txt"]
More or less re-implementation of https://github.com/panzi/u4pak[Python U4Pak]
for fun, ease of use (standalone binary), and speed (multi-threading).
This is a tool to pack, unpack, check, and list the contents of Unreal Engine 4
packages. Under Linux it can also be used to read-only FUSE-mount archives. Note
that only a limited number of pak versions are supported.
.Supported Versions
|====
|Pak Version |Unreal Engine Version |Reading |Writing
|1 |< 4.0 |✔️ |✔️
|2 |4.0 - 4.2 |✔️ |✔️
|3 |4.3 - 4.15 |✔️ |✔️
|4 |4.16 - 4.19 |✔️ |❌
|5 |4.20 |✔️ |❌
|6 |- |✔️ |❌
|7 |4.21 |✔️ |❌
|8 |4.22 - 4.24 |✔️ |❌
|9 |4.25 |✔️ |❌
|10 |- |✔️ |❌
|11 |4.26 - 4.27 |✔️ |❌
|====
.Supported Compression
|====
|Name |Supported
|None |✔️
|https://zlib.net[Zlib] |✔️
|https://www.gnu.org/software/gzip/[Gzip] |❌
|https://lz4.github.io/lz4/[LZ4] |❌
|====
Instead of passing arguments you can also put the arguments in a file with the
extension .u4pak and pass the path to that instead. This is useful for Windows
users that aren't used to a terminal. You can even associate the extension with
u4pak.exe so that it will be automatically opened with it when you double click
it. File paths in a .u4pak file are relative to the directory containing the
file. The syntax of these files is not shell syntax. If you don't have any white
space, double quotes ("
), or hash marks (#
) in your file names you don't have
to worry about anything. #
is used to start a comment line (only if it doesn't
touch any non-white space on it's left) and "
is used to quote arguments
containing white space, #
, or "
. In order to write a "
in a quoted argument
you simply need to double it, meaning an argument that contains nothing but a
single "
is written as """"
. Newlines are ignored like any other white space.
An example .u4pak file whould be:
# This is packing my project:
pack
--version=4
--mount-point=../../..
"C:\Users\Alice\My Documents\U4Project\NewArchive.pak"
":rename=/:C:\Users\Alice\My Documents\U4Project\Some Files"
":zlib,rename=/:Some Other Files"
If u4pak.exe is run by double clicking or by dropping a .u4pak file on it it
won't immediately close the terminal window, but will instead ask you to press
ENTER. It does this so you have a chance to read the output. Since I don't use
Windows (I cross compile on Linux and test with Wine) I could not test this
particular feature. If it doesn't work please report a bug. In order to force
the "Press ENTER to continue..." message you can pass the argument
--pause-on-exit=always
(Windows-only).
== Usage
u4pak [--pause-on-exit=<always|never|auto>] [SUBCOMMAND]
Or:
u4pak "C:\Path\to\arguments.u4pak"
== Sub-Commands
|====
| Sub-Command |Description
| check | Check consistency of a package
| help | Prints general help message or the help of the given subcommand(s)
| info | Show summarized information of a package
| list | List content of a package
| mount | Mount package as read-only filesystem (Linux-only)
| pack | Create a new package
| unpack | Unpack content of a package
|====
For help to the various sub-commands run u4pak help SUBCOMMAND
.
== File Format
Byte order is little endian and the character encoding of file names seems to be
ASCII (or ISO-8859-1/UTF-8 that coincidentally only uses ASCII compatiple
characters).
Offsets and sizes seem to be 64bit or at least unsigned 32bit integers. If
interpreted as 32bit integers all sizes (except the size of file names) and offsets
are followed by another 32bit integer of the value 0, which makes me guess these
are 64bit values. Also some values exceed the range of signed 32bit integers, so
they have to be at least unsigned 32bit integers.
This information was reverse engineered from the Elemental
https://wiki.unrealengine.com/Linux_Demos[Demo] for Linux (which contains a 2.5
GB .pak file), the https://www.techpowerup.com/download/unreal-engine-4-five-tech-demos/[Unreal Engine 4 - Five Tech Demos],
version 7 was reverse engineered from the https://store.steampowered.com/app/813630/Supraland/[Supraland Demo],
and the Conan Exiles variant was reverse egineered from the
https://steamcommunity.com/sharedfiles/filedetails/?id=1765361591[SandstormFix_EXP workshop item].
Reverse engineering was done by poking around the pak files with a hex editor,
no kind of decompilation was done by me.
Versions >= 8 where reversed from an empty unreal engine project.
.Basic layout
- Data Records
- Index
Index Header
Index Records
- Footer
In order to parse a file you need to read the footer first. The footer contains
an offset pointer to the start of the index records. The index records then
contain offset pointers to the data records.
[[encoded-record]]
=== Encoded Record
[subs="quotes"]
Offset Size Type Description
0 4 uint32_t bitfield containing record information
0-5 : Compression block size
6-21 : Compression blocks count
22 : Encrypted
23-28: Compression method
29 : Size 32-bit safe?
30 : Uncompressed size 32-bit safe?
31 : Offset 32-bit safe?
if offset 32 bit
4 4 uint32_t offset
else
4 8 uint64_t offset
end
if uncompressed size 32 bit
? 4 uint32_t uncompressed size
else
? 8 uint64_t uncompressed size
end
if compression method != 0x00
if size 32 bit
? 4 uint32_t size
else
? 4 uint32_t size
end
end
if compression block count > 0 && (encrypted || compression block count != 1)
for in 0..compression block count_
? 4 uint32_t block size
end
end
=== Record
NOTE: This structure, while still present in version >= 10 is not used anymore by default. See xref:encoded-record[Encoded Record] for the current record information.
[subs="quotes"]
Offset Size Type Description
0 8 uint64_t offset
8 8 uint64_t size (N)
16 8 uint64_t uncompressed size
24 4 uint32_t compression method:
0x00 ... none
0x01 ... zlib
0x10 ... bias memory
0x20 ... bias speed
if version <= 1
28 8 uint64_t timestamp
end
? 20 uint8_t[20] data sha1 hash
if version >= 3
_if compressionmethod != 0x00
?+20 4 uint32_t block count (M)
?+24 M*16 CB[M] compression blocks
end
? 1 uint8_t is encrypted
?+1 4 uint32_t The uncompressed size of each compression block.
end The last block can be smaller, of course.
if variant == "Conan Exiles"
? 4 uint32_t Unknown field. For Conan Exiles index record only
seen it to have the value 0.
end
=== Compression Block (CB)
[horizontal]
Size:: 16 bytes
[subs="quotes"]
Offset Size Type Description
0 8 uint64_t compressed data block start offset.
version <= 4: offset is absolute to the file
version 7: offset is relative to the offset
field in the corresponding Record
8 8 uint64_t compressed data block end offset.
There may or may not be a gap between blocks.
version <= 4: offset is absolute to the file
version 7: offset is relative to the offset
field in the corresponding Record
=== Data Record
[subs="quotes"]
Offset Size Type Description
0 ? Record file metadata (offset field is 0, N = compressed_size)
if variant == "Conan Exiles"
? 20 ? Unknown. Maybe another SHA-1 sum of something?
The first 4 bytes have values other than the extra
4 bytes in the index record, which is why I didn't
put those into the general record structure.
_else if version >= 4 and compressionmethod != 0x00
? 4 uint32_t Unknown.
end
? N uint8_t[N] file data
[NOTE]
Starting with version 4 there is an additional 4 bytes in the repeated
data record copy (the record that precedes the actual file date, not the
record in the index). I don't know what that is. It is not always the same value.
E.g. it is the same for some files, but different for others. The first 2 bytes
seem to be always the same (0x78 0x9c
in a v4 and a v7 pak that I saw), so maybe
its 2 16 bit fields?
This is why I've deactivated packing for versions > 3.
[[fdi]]
=== Full Directory Index
The data structure is a map<DirectoryName, map<FileName, offset>>.
[subs="normal"]
Offset Size Type Description
0 4 uint32_t directory count (D)
for i in 0..D
? 4 int32_t directory name size (DS)
For some games a negative value means it's a UTF-16
string in 2 -S bytes.
?+4 DS char[DS] directory name (includes terminating null byte)
?+4+DS 4 uint32_t file count (F)
for j in 0..F
? 4 int32_t file name size (FS)
For some games a negative value means it's a UTF-16
string in 2 -S bytes.
?+4 FS char[FS] file name (includes terminating null byte)
?+4+FS 4 uint32_t offset in xref:directory-info[encoded entry info]
end
end
=== Index Record
NOTE: This structure, while still present in version >= 10 is not used anymore by default. See xref:fdi[Full Directory Index] for record information.
[subs="quotes"]
Offset Size Type Description
0 4 int32_t file name size (S)
For some games a negative value means it's a UTF-16
string in 2 * -S bytes.
4 S char[S] file name (includes terminating null byte)
4+S ? Record file metadata
if variant == "Conan Exiles"
? 4 ? Unknown. Only saw all 0 so far.
end
=== Index
[subs="normal,callouts"]
.Version >= 10
Offset Size Type Description
0 4 int32_t mount point size (S)
For some games a negative value means it's a UTF-16
string in 2 * -S bytes.
4 S char[S] mount point (includes terminating null byte)
S+4 4 int32_t entry count
S+8 8 uint64_t path hash seed <1>
S+16 4 uint32_t has path hash index
if has path index != 0
S+20 8 int64_t path hash index offset
S+28 8 int64_t path hash index size
S+36 20 uint8_t[20] path hash index hash
end
? 4 uint32_t has full directory index
if has full directory index != 0
?+4 8 int64_t full directory index offset
?+12 8 full directory index size
?+20 20 uint8_t[20] full directory index hash
end
? 4 int32_t encoded entry info size (P)
?+4 P uint8_t[P] encoded entry info [[directory-info]]
?+P 4 uint32_t file count, probably unused / 0 (N)
?+P+4 ? IndexRecord[N] records
<1> Needs clarification
[subs="quotes"]
.Legacy (Version < 10)
----
*Offset Size Type Description*
0 4 int32_t mount point size (S)
For some games a negative value means it's a UTF-16
string in 2 * -S bytes.
4 S char[S] mount point (includes terminating null byte)
S+4 4 uint32_t record count (N)
S+8 ? IndexRecord[N] records
----
=== Footer
[horizontal]
.Size and index features
|====
|Versions |Size |Index Encryption |Encryption Key Guid |Compression method name |Frozen Index
|v1 - v3 |44 bytes |❌ |❌ |❌ |❌
|v4 - v6 | 45 bytes |✔️ |❌ |❌ |❌
|v7 | 65 bytes |✔️ |✔️ |❌ |❌
|v8 | 193 bytes |✔️ |✔️ |✔️(Max. 4) |❌
|v9 | 226 bytes |✔️ |✔️ |✔️(Max. 5) |✔️
|v10 - v11 | 225 bytes |✔️ |✔️ |✔️(Max. 5) |❌
|====
[subs="quotes"]
----
*Offset Size Type Description*
_if version >= 7_
0 20 uint8_t[20] encryption key Guid
_end_
_if version >= 4_
? 1 uint8_t encrypted index
_end_
? 4 uint32_t magic: 0x5A6F12E1
?+4 4 uint32_t version: 1 - 11
?+8 8 uint64_t index offset
?+16 8 uint64_t index size
?+24 20 uint8_t[20] index sha1 hash
_if version == 9_
?+44 1 uint8_t frozen index
_end_
_if version == 8_
?+44 128 uint8_t[128] compression methods (4 * 32 chars)
_else if version > 8_
? 160 uint8_t[160] compression methods (5 * 32 chars)
_end_
----
== Related Projects
* https://github.com/panzi/fezpak[fezpak]: pack, unpack, list and mount FEZ .pak archives
* https://github.com/panzi/psypkg[psypkg]: pack, unpack, list and mount Psychonauts .pkg archives
* https://github.com/panzi/bgebf[bgebf]: unpack, list and mount Beyond Good and Evil .bf archives
* https://github.com/panzi/unvpk[unvpk]: extract, list, check and mount Valve .vpk archives (C++)
* https://github.com/panzi/rust-vpk[rust-vpk]: Rust rewrite of the above (Rust)
* https://github.com/panzi/t2fbq[t2fbq]: unpack, list and mount Trine 2 .fbq archives
* https://github.com/panzi/u4pak[u4pak]: old Python version of this program
== MPLv2 License
Rust U4Pak - pack, unpack, check, list and mount Unreal Engine 4 packages
Copyright (C) {localyear} Mathias Panzenböck, L. Sprengel
This Source Code Form is subject to the terms of the Mozilla Public
License, v. 2.0. If a copy of the MPL was not distributed with this
file, You can obtain one at https://mozilla.org/MPL/2.0/.