The EPUB container (OCF)

An .epub file is a ZIP archive laid out by the Open Container Format (OCF), part of the EPUB specification. The same rules have applied, with small changes, since EPUB 2 (OCF 2.0.1, 2010); this page follows EPUB 3.3 (W3C, 2023) and notes where EPUB 2 differs.

ZIP rules

Each file is stored (method 0) or compressed with Deflate (method 8); no other method is allowed.
ZIP encryption is not allowed; EPUB has its own, in encryption.xml.
The archive may not be split or spanned over several files.
ZIP64 may be used, but only when the archive needs it.
File names are UTF-8.
The “version needed to extract” field is 10, 20 or 45 (1.0, 2.0 or 4.5).

The mimetype file

The first entry in the archive must be a file named mimetype, stored without compression, unencrypted and with no extra field, whose content is exactly the 20 ASCII bytes application/epub+zip, with no line end, padding or byte order mark. Because its position and length are fixed, the bytes from offset 30 of every valid EPUB read mimetypeapplication/epub+zip, and a program can recognise the format from them without unzipping.

00000000  50 4B 03 04 0A 00 00 00  00 00 xx xx xx xx xx xx   PK..............
00000010  xx xx 14 00 00 00 14 00  00 00 08 00 00 00 6D 69   ..............mi
00000020  6D 65 74 79 70 65 61 70  70 6C 69 63 61 74 69 6F   metypeapplicatio
00000030  6E 2F 65 70 75 62 2B 7A  69 70 50 4B 03 04 …      n/epub+zipPK..

Offset	Bytes	ZIP field	Value in an EPUB
`0`	`50 4B 03 04`	local file header signature	`PK\3\4`
`4`	`0A 00`	version needed	10 (often 20)
`6`	`00 00`	flags	0: no encryption and no data descriptor
`8`	`00 00`	method	0, stored
`10`	4 bytes	time and date	any
`14`	4 bytes	CRC-32	the CRC of the 20 bytes
`18`	`14 00 00 00`	compressed size	20
`22`	`14 00 00 00`	uncompressed size	20
`26`	`08 00`	name length	8
`28`	`00 00`	extra field length	0, required
`30`	`mimetype`	file name
`38`	`application/epub+zip`	file data
`58`	`50 4B 03 04`	next entry	usually `META-INF/container.xml`

File names and paths

Paths and names are case-sensitive, and the folder separator is /.
A file name is at most 255 bytes of UTF-8; a full path at most 65,535 bytes.
A name must not contain / " * : < > ? \ | or DEL (U+007F), C0 or C1 control characters, private-use characters or Unicode non-characters, and must not end with a full stop.
Two names in the same folder must not be the same after Unicode normalisation (NFC) and case folding.
Names should avoid spaces, which some older reading systems mishandle.

The META-INF folder

The root of the container must hold a folder META-INF. Its files are reserved; they are never listed in the package manifest and the book’s content must not link to them.

File	Status	Content
`container.xml`	Required	Where the package document is: see below.
`encryption.xml`	Optional; required when any file is encrypted or a font is obfuscated	Which files are encrypted and how: see encryption.
`signatures.xml`	Optional	XML Digital Signatures over files in the container.
`rights.xml`	Optional	Digital rights information, in a format the specification does not define (DRM systems such as Adobe’s put their licence here).
`metadata.xml`	Optional	Metadata about the container as a whole, in a format the specification does not define.
`manifest.xml`	Optional	A manifest of the container, in any format; kept for OpenDocument compatibility.

A reading system may also find files such as com.apple.ibooks.display-options.xml here, added by Apple and other vendors for their own settings; the specification does not define them.

container.xml

The file uses the namespace urn:oasis:names:tc:opendocument:xmlns:container.

Element	Attributes	Content
`container`	`version`: required, `1.0`	One `rootfiles`, then at most one `links`.
`rootfiles`		One or more `rootfile`.
`rootfile`	`full-path`: required, the path of a package document from the container root, with no leading `/`. `media-type`: required, `application/oebps-package+xml`.	Empty.
`links`		One or more `link`.
`link`	`href`: required. `rel`: required, the relationship. `media-type`: required when the target is inside the container.	Empty. Links resources about the whole container.

The first rootfile is the default rendition. Several rootfiles make a multiple-rendition publication (for example a fixed-layout and a reflowable version of one book), described by the separate EPUB Multiple-Rendition Publications note; most reading systems use only the first.

<?xml version="1.0" encoding="UTF-8"?>
<container version="1.0" xmlns="urn:oasis:names:tc:opendocument:xmlns:container">
  <rootfiles>
    <rootfile full-path="EPUB/package.opf" media-type="application/oebps-package+xml"/>
  </rootfiles>
</container>

encryption.xml

The root element is encryption in the same container namespace. It holds, in any order, EncryptedKey and EncryptedData elements from W3C XML Encryption (namespace http://www.w3.org/2001/04/xmlenc#), and may hold signatures.

Element	Meaning
`enc:EncryptedData`	One per encrypted file.
`enc:EncryptionMethod Algorithm="…"`	The algorithm, as a URI: for example `http://www.w3.org/2001/04/xmlenc#aes128-cbc`, or the font obfuscation URIs below.
`ds:KeyInfo`	Which key, often by name or by reference to an `EncryptedKey`.
`enc:CipherData` / `enc:CipherReference URI="…"`	The encrypted file, by its path in the container.
`enc:EncryptionProperties` / `enc:EncryptionProperty` / `Compression`	Optional, namespace `http://www.idpf.org/2016/encryption#compression`: `Method` 0 or 8 and `OriginalLength`, when the file was deflated before it was encrypted.

These files must never be encrypted: mimetype, the META-INF files listed above, and the package documents. Text-like files are deflated before they are encrypted; already compressed media (JPEG, MP3 and so on) should not be.

Font obfuscation

EPUB lets a publisher scramble embedded fonts slightly, so that a font cannot be pulled out of a book and installed as it is. It is not encryption: anyone with the book can undo it.

Take the publication’s unique identifier: the dc:identifier named by the package’s unique-identifier attribute.
Remove every space (U+0020), tab (U+0009), carriage return (U+000D) and line feed (U+000A).
Encode it as UTF-8 and take its SHA-1 hash: a 20-byte key.
XOR the first 1,040 bytes of the font file with the key repeated 52 times; leave the rest unchanged. A font shorter than 1,040 bytes is XORed whole.
Then deflate the result, if compressing, and store it in the ZIP.

Each obfuscated font has an EncryptedData entry with Algorithm="http://www.idpf.org/2008/embedding". The key is never written in the file. Reading systems undo it in the reverse order.

Adobe defined an earlier scheme, still found in many EPUB 2 books, with the algorithm URI http://ns.adobe.com/pdf/enc#RC: the key is the 16 bytes of the identifier read as a UUID (its hex digits, without urn:uuid: and hyphens), and only the first 1,024 bytes are XORed. It is not part of the EPUB specification.

Differences in EPUB 2

The mimetype file, the META-INF files and container.xml work the same way in EPUB 2 (OCF 2.0.1); the links element of container.xml and the compression property of encryption.xml were added later. An EPUB 2 and an EPUB 3 file are told apart only by the version attribute of the package document, not by the container.

.EPUB File Format – Container (OCF)