By default, 100000 characters are extracted when indexing the content. This default value can be changed by setting the index.mapping.attachment.indexed_chars setting. It can also be provided on a per document indexed using the _indexed_chars parameter. -1 can be set to extract all text, but note that all the text needs to be allowed to be represented in memory:

PUT /test/person/1
{
    "my_attachment" : {
        "_indexed_chars" : -1,
        "_content" : "... base64 encoded attachment ..."
    }
}