Packet format

e-Amusement uses XML for its application layer payloads. This XML is either verbatim, or in a custom packed binary format.

The XML format

Each tag that contains a value has a __type attribute that identifies what type it is. Array types have a __count attribute indicating how many items are in the array. Binary blobs additionally have a __size attribute indicating their length (this is notably not present on strings, however).

It is perhaps simpler to illustrate with an example, so:

<?xml version='1.0' encoding='UTF-8'?>
<call model="KFC:J:A:A:2019020600" srcid="1000" tag="b0312077">
    <eventlog method="write">
        <retrycnt __type="u32" />
        <data>
            <eventid __type="str">G_CARDED</eventid>
            <eventorder __type="s32">5</eventorder>
            <pcbtime __type="u64">1639669516779</pcbtime>
            <gamesession __type="s64">1</gamesession>
            <strdata1 __type="str" />
            <strdata2 __type="str" />
            <numdata1 __type="s64">1</numdata1>
            <numdata2 __type="s64" />
            <locationid __type="str">ea</locationid>
        </data>
    </eventlog>
</call>

Arrays are encoded by concatenating every value together, with spaces between them. Data types that have multiple values, are serialized similarly.

Therefore, an element storing an array of 3u8 ([(1, 2, 3), (4, 5, 6)]) would look like this

<demo __type="3u8" __count="2">1 2 3 4 5 6</demo>

Besides this, this is otherwise a rather standard XML.

Packed binary overview

Many packets, rather than using a string-based XML format, use a custom binary packed format instead. While it can be a little confusing, remembering that this is encoding an XML tree can make it easier to parse.

To start with, let's take a look at the overall structure of the packets.

Every packet starts with the magic byte 0xA0. Following this is the content byte, the encoding byte, and then the 2's compliment of the encoding byte.

Possible values for the content byte are:

Source code details

Not totally cleaned these up yet, but the general concept of how packets are parsed can be seen fairly clearly. At a high level, we have a single function that validates the header, parses out the schema, then goes to read the body of the packet, if we're expecting it. The arguments to parse_packet_header will make more sense in a moment.

parse_packet_header has a lot of things going on, so I'm just pulling out a few important snippets here.

We first read out four bytes from the start of the packet, and convert that to an integer; nothing especially magic here. The next block however is potentially not the first that you might have expected to see. Based on the two flags passed into the function arguments, we are going to subtract a value from this header. Specifically, the first byte we subtract is always 0xa0, then the second byte are those C value in the table above.

Finally, we mask out the first two bytes, and assert that they're both null. That is, they are exactly equal to the value we subtracted from them. Of note here is that the caller to this function "decides" what sort of packet it is expecting.

We can also see the check for ~E here. If that check passes, we return the E byte, otherwise we're going to error.

The encoding flag indicates the encoding for all string types in the packet (more on those later). Possible values are:

Data is assumed by default to be in ISO 8859 encoding. That is, for encodings 0x00 and 0x40, no transformation is performed on the binary data to produce readable text.

ASCII encoding is true 7-bit ASCII, with the 8th bit always set to 0. This is validated.

Source code details

The full table for these values can be found in libavs.

A second table exists just before this on in the source, responsible for the <?xml version='1.0' encoding='??'?> line in XML files.

This is indexed using the following function, which maps the above encoding IDs to 1, 2, 3, 4 and 5 respectively.

char* xml_get_encoding_name(uint encoding_id) {
    return ENCODING_NAME_TABLE[((encoding_id & 0xe0) >> 5) * 4];
}

While validating ~E isn't technically required, it acts as a useful assertion that the packet being parsed is valid.

The packet schema header

Following the 4 byte header, is a 4 byte integer containing the length of the next part of the header (this is technically made redundant as this structure is also terminated).

This part of the header defines the schema that the main payload uses.

A tag definition follows one of the following three formats:

• Compressed names:

  0	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
  Type	nlen	Tag name
  Attributes and children															FE

• Full names, short length:

  0	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
  Type	0x40-0x64	Tag name
  Attributes and children															FE

• Full names, long length:

  0	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
  Type	0x80-0x8f	0x00-0xff	Tag name
  Attributes and children															FE

The encoding of structure names varies depending on the packet content byte. If the content flag indicated we have a full string, we first need to check if the value of the first byte exceeds 0x7f. If it does, we need to read an additional byte. In the single byte case, we subtract 0x3f¹ to get our real length. In the two byte case we subtract 0x7fbf². In the latter case, the maximum allowed length is 0x1000.
¹ simplified from (length & ~0x40) + 0x01
² simplified from (length & ~0x8000) + 0x41

If we are instead parsing packed names, then the names are encoded as densely packed 6 bit values. The length prefix (nlen) determines the length of the final unpacked string. The acceptable alphabet is 0123456789:ABCDEFGHIJKLMNOPQRSTUVWXYZ_abcdefghijklmnopqrstuvwxyz, and the packed values are indecies within this alphabet. The maximum length for a name in this mode is 36 bytes (0x24).

The children can be a combination of either attribute names, or child tags. Attribute names are represented by the byte 0x2E followed by a length prefixed name as defined above. Child tags follow the above format. Type 0x2E must therefore be considered reserved as a possible structure type. As they carry special meaning in text-bsaed XML encoding, attribute names beginning with __ are disallowed.

Attributes (type 0x2E) represent a string attribute. Any other attribute must be defined as a child tag. Is it notable that 0 children is allowable, which is how the majority of values are encoded.

All valid IDs, and their respective type, are listed in the following table. The bucket column here will be used later when unpacking the main data, so we need not worry about it for now, but be warned it exists and is possibly the least fun part of this format.

Strings should be encoded and decoded according to the encoding specified in the packet header. Null termination is optional, however should be stripped during decoding.

All of these IDs are & 0x3F. Any value can be turned into an array by setting the 7^th bit high (| 0x40). Arrays of this form, in the data section, will be an aligned size: u32 immediately followed by size bytes' worth of (unaligned!) values of the unmasked type. Despite being a u32, the maximum length allowed is 0xffffff.

Source code details

The full table for these values can be found in libavs. This table contains the names of every tag, along with additional information such as how many bytes that data type requires, and which parsing function should be used.

Note about the array type:

While I'm not totally sure, I have a suspicion this type is used internally as a pseudo-type. Trying to identify its function as a parsable type has some obvious blockers:

All of the types have convenient printf-using helper functions that are used to emit them when serializing XML. All except one.

If we have a look inside the function that populates node sizes (libavs-win32.dll:0x1000cf00), it has an explicit case, however is the same fallback as the default case.

In the same function, however, we can find a second (technically first) check for the array type.

This seems to suggest that internally arrays are represented as a normal node, with the array type, however when serializing it's converted into the array types we're used to (well, will be after the next sections) by masking 0x40 onto the contained type.

Also of interest from this snippet is the fact that void, bin, str, and attr cannot be arrays. void and attr make sense, however str and bin are more interesting. I suspect this is because konami want to be able to preallocate the memory, which wouldn't be possible with these variable length structures.

The data section

This is where all the actual packet data is. For the most part, parsing this is the easy part. We traverse our schema, and read values out of the packet according to the value indicated in the schema. Unfortunately, konami decided all data should be aligned very specifically, and that gaps left during alignment should be backfilled later. This makes both reading and writing somewhat more complicated, however the system can be fairly easily understood.

Firstly, we divide the payload up into 4 byte chunks. Each chunk can be allocated to either store individual bytes, shorts, or ints (these are the buckets in the table above). When reading or writing a value, we first check if a chunk allocated to the desired type's bucket is available and has free/as-yet-unread space within it. If so, we will store/read our data to/from there. If there is no such chunk, we claim the next unclaimed chunk for our bucket.

For example, imagine we write the sequence byte, int, byte, short, byte, int, short. The final output should look like:

While this might seem a silly system compared to just not aligning values, it is at least possible to intuit that it helps reduce wasted space. It should be noted that any variable-length structure, such as a string or an array, claims all chunks it encroaches on for the int bucket, disallowing the storage of bytes or shorts within them.

class Packer:
    def __init__(self, offset=0):
        self._word_cursor = offset
        self._short_cursor = offset
        self._byte_cursor = offset
        self._boundary = offset % 4

    def _next_block(self):
        self._word_cursor += 4
        return self._word_cursor - 4

    def request_allocation(self, size):
        if size == 0:
            return self._word_cursor
        elif size == 1:
            if self._byte_cursor % 4 == self._boundary:
                self._byte_cursor = self._next_block() + 1
            else:
                self._byte_cursor += 1
            return self._byte_cursor - 1
        elif size == 2:
            if self._short_cursor % 4 == self._boundary:
                self._short_cursor = self._next_block() + 2
            else:
                self._short_cursor += 2
            return self._short_cursor - 2
        else:
            old_cursor = self._word_cursor
            for _ in range(math.ceil(size / 4)):
                self._word_cursor += 4
            return old_cursor

    def notify_skipped(self, no_bytes):
        for _ in range(math.ceil(no_bytes / 4)):
            self.request_allocation(4)