.ZIP File Format – ZIP64

The sentinels: when each field moves

A field too small for its value is set to all ones, 0xFFFF or 0xFFFFFFFF, and the real value goes in a ZIP64 field. A value of exactly 0xFFFFFFFF also has to move, because in that field it can only mean “see ZIP64”.

Each field moves on its own, so a member can have a ZIP64 offset and 32-bit sizes. A reader treats a value as moved only if the field holds its sentinel. A sentinel with no matching ZIP64 data is a damaged archive. The explorer reports that case, and uses a sentinel in the end record only when a ZIP64 end record is present.

ZIP64 end of central directory record

Comes after the central directory, and any digital signature, and before the locator.

ZIP64 end of central directory locator

A fixed 20-byte record immediately before the end record. Finding it there is how a reader knows the archive uses ZIP64.

The offset counts from the start of the disk it refers to, so an archive with data in front can need shifting here as well. The locator must be on the same disk as the end record.

The 0x0001 extra field

The order of the fields is fixed, and each appears only if the matching field in the record is set to its sentinel. So the field’s position depends on which earlier fields are present.

Note that the order here is original size then compressed size, the opposite of their order in the headers.

In a local header, the block must hold both sizes whenever it is used, even if only one of them overflows. If the central directory is encrypted and bit 13 is set, the original size in the local header is stored as zero.

The presence of this block on a member with a data descriptor tells a reader to expect the descriptor’s 64-bit sizes. Writers that do not yet know the sizes when they write the local header differ in what they put in it, so a reader should take the sizes from the central directory where it has one.

Reading ZIP64

1. Find the end record. Look 20 bytes before it for the locator signature.
2. If the locator is present, read the ZIP64 end record at its offset. Check that it lies before the locator and carries the 50 4B 06 06 signature.
3. Take each value from the end record unless it holds a sentinel, in which case take the 64-bit value from the ZIP64 record.
4. Walk the central directory. For each header, read its 0x0001 block for each field that holds a sentinel, in the fixed order.

Writing ZIP64

• Use ZIP64 fields for the values that need them. They are also allowed for smaller values, and APPNOTE permits a writer to use them regardless of a file’s size.
• Write the ZIP64 end record and locator whenever any end record field holds a sentinel.
• Set the version needed to at least 45 on every member using ZIP64.
• Keep total values and per-disk counts consistent, and write the locator’s disk count as 1 for a single file.