kientzle edited this page Mar 29, 2012 · 5 revisions
Clone this wiki locally

How libarchive automatically determines the format of data.

Table of Contents


Libarchive uses an internal "bidding" process in which multiple modules inspect the incoming data. The technical LibarchiveInternals page describes the process in detail; this page walks through a simple example.

Keep in mind that the mechanism described here is completely handled within the library and is available to any program that uses libarchive. It also works with any input source, including pipes, sockets, and other data sources that cannot rewind the input.

Modules and Bidding

Libarchive is broken into separate modules for every format that it supports. Each such module includes a "bid" function that knows how to inspect the incoming data and provide a number that indicates how certain that module is that it can recognize that data.

Simplest example: A tar archive

When you use libarchive, you create the libarchive object and then invoke functions that register different modules. Suppose you use the following:

 struct archive *a = archive_read_new();

This registers two bidders with libarchive's read core: one bidder understands tar archives and the other understands cpio archives.

You can then open a file and prepare to read the archive contents:

 archive_read_open_filename(a, "input.tar", 10240);

This function is a shorthand for calling `archive_read_open1()` after registering an input function that reads blocks from a file. If you want to read data from some other data source, you can call the registration functions directly. (If you decide to roll your own, you might want to look at the source code for archive_read_open_filename for ideas.)

Inside `archive_read_open1()`, libarchive asks the input function for the first block and shows that block to each registered module. In this case, the cpio module will probably bid zero and the tar module will bid a higher value. So libarchive's read core will initialize the tar reader.

Now you can ask for data from the archive and those requests will be handled internally by the tar reader module:

 while (archive_read_next_header(a, &entry) == ARCHIVE_OK) {
    /* ... process next entry ... */

A more complex example: input.tar.gz.uu

Of course, tar archives are often compressed and sometimes processed even further. Consider a tar archive that has been compressed with `gzip` and then encoded with the `uuencode` utility. In order to read this, libarchive must first decode the uuencode wrapping, then uncompress the resulting data before it can evaluate the tar structure.

Besides _format_ modules that understand archive formats, libarchive also has _filter_ modules that understand particular compression or encoding formats. (Originally, the filter machinery was used only for decompression. It now supports many encoding formats and wrapper formats as well, but the term "compression" is still used by much of the code here.)

So let's start over and create a new archive reader:

 struct archive *a = archive_read_new();

This time, I used two convenience functions that register all formats and all filters that are supported by libarchive.

Now, when you open the archive using the same code as above, libarchive will first present the input to every registered filter module. For our example, the `uudecode` filter will recognize this input and bid the highest. Libarchive will then initialize the `uudecode` filter and ask for data from that filter. When libarchive asks for data from the uudecode filter, that filter will in turn ask the file input function for data from the file.

So now libarchive can take the output of the uudecode filter and again present that data to each registered filter module. In our example here, the gzip decompression module will win the next round of bidding. After initializing the gzip decompression module, libarchive will take the output of that module and again ask all filters to bid. This time, none of the filters will recognize the output, so all of the filters will bid zero. When libarchive cannot add any more filters to the processing list, it will ask format modules to bid.

When all of the bidding is complete, libarchive will end up with an internal list that looks like this:

  • File input function
  • UUdecode filter module
  • Gzip decompression module
  • Tar format module
When you ask for the next header, libarchive will pass that to the tar module, which will request data from the filter modules as necessary.

In this way, almost any combination of compression or encoding can be removed before the inner archive is evaluated. This allows libarchive to transparently process files such as compressed ISO images or RPM files (RPM files are really a simple encoding wrapper around a compressed cpio or xar archive).

By having the modules register with libarchive's read core, statically linked programs can select only the minimum set of functionality they need and not pay any overhead for formats or compression support that they don't require.