.ZIP File Format – Extra fields

Block layout

An extra field is a sequence of blocks, with no padding between them.

IDs 0 to 31 are reserved for PKWARE. The whole extra field is limited by its 16-bit length, and APPNOTE says the name, extra field and comment together should not generally exceed 65,535 bytes. A reader should skip IDs it does not know, using the data size, and check that each block fits.

Some writers put bytes in the extra field that are not blocks. Older Android zipalign padded with zeros, too few to be a block, and at least one program kept unrelated data there. The explorer treats fewer than 4 leftover bytes as padding, and a field whose first block cannot fit as unstructured data.

Header IDs

ZIP64, 0x0001

Described on the ZIP64 page. The fields present depend on which header fields hold sentinels.

NTFS, 0x000A

Three 8-byte file times in NTFS format, in units of 100 nanoseconds since 1 January 1601 UTC.

Patch descriptor, 0x000F

Marks a member as patch data. After the tag and size:

Flags bit 0 is for auto-detection, bit 1 means a self-patch, bits 4 and 5 are the action (0 none, 1 add, 2 delete, 3 patch), bits 8 and 9, 10 and 11, and 12 and 13 are the reaction to an absent, newer and unknown file (0 ask, 1 skip, 2 ignore, 3 fail). The rest are reserved.

PKWARE Unix, 0x000D

In the local header only.

Extended timestamp, 0x5455

Info-ZIP’s Unix times, also known as UT.

Times are signed 32-bit Unix seconds, UTC. In a central header, the flags still describe the local header but only the modification time is stored, so TSize is 5 or 1. A reader that decodes by flags alone reads off the end of a central block. Use TSize.

Info-ZIP Unix type 1, 0x5855

Obsolete, and to be read but not written. UX in the file. Ignored if 0x5455 or 0x7855 is also present.

Info-ZIP Unix type 2, 0x7855

Ux in the file. In a local header, TSize is 4, followed by UID (2) and GID (2). In a central header, TSize is 0 and there is no data: its presence only says the local header has the field.

Info-ZIP Unix type 3, 0x7875

ux in the file. It can hold larger owner IDs than 0x7855, whose values it overrides.

Unicode path, 0x7075

up in the file. Used to give a UTF-8 name while the header keeps an older encoding for old tools. See File names.

If the CRC-32 does not match the header’s name, the header was changed by a tool that did not update this block, so a reader must ignore it and use the header’s name. APPNOTE says these blocks should not be created when bit 11 is set.

Unicode comment, 0x6375

uc in the file. The same layout as the Unicode path, in a central header, with ComCRC32 computed over the central header’s comment.

WinZip AES, 0x9901

Marks a member as AE-1 or AE-2 encrypted. Present in both headers, with method set to 99.

See WinZip AES for how the data is protected.

FWKCS MD5, 0x4B46

JAR marker, 0xCAFE

Java’s JarOutputStream adds this block to the first entry it writes, with a data size of 0 and no data, as the first block in the extra field. It is left out if the entry already has it. Other entries do not get it.

Alignment and padding

Some formats need a member’s data to start at a multiple of some boundary, so that it can be used straight from the file, or in the case of Open Packaging so that it can grow later. Padding goes in an extra block in the local header.