2021-05-24 00:11:58 +03:00
|
|
|
|
// Package rootfs extracts all layers of a Docker container image to a single
|
|
|
|
|
|
// tarball. It will go trough all layers in order and copy every file to the
|
|
|
|
|
|
// destination archive.
|
|
|
|
|
|
//
|
2021-05-24 00:11:58 +03:00
|
|
|
|
// It will also reasonably process those files.
|
2021-05-24 00:11:58 +03:00
|
|
|
|
//
|
|
|
|
|
|
// == Non-directory will be copied only once ==
|
|
|
|
|
|
// A non-directory will be copied only once, only from within it's past
|
|
|
|
|
|
// occurrence. I.e. if file /a/b was found in layers 0 and 2, only the file
|
|
|
|
|
|
// from layer 2 will be used.
|
|
|
|
|
|
// Directories will always be copied, even if there are duplicates. This is
|
|
|
|
|
|
// to avoid a situation like this:
|
|
|
|
|
|
// layer0:
|
|
|
|
|
|
// - ./dir/
|
|
|
|
|
|
// - ./dir/file
|
|
|
|
|
|
// layer1:
|
|
|
|
|
|
// - ./dir/
|
|
|
|
|
|
// - ./dir/file
|
|
|
|
|
|
// In theory, the directory from layer 1 takes precedence, so a tarball like
|
|
|
|
|
|
// this could be created:
|
|
|
|
|
|
// - ./dir/ (from layer1)
|
|
|
|
|
|
// - ./dir/file1 (from layer1)
|
|
|
|
|
|
// However, imagine the following:
|
|
|
|
|
|
// layer0:
|
|
|
|
|
|
// - ./dir/
|
|
|
|
|
|
// - ./dir/file1
|
|
|
|
|
|
// layer1:
|
|
|
|
|
|
// - ./dir/
|
|
|
|
|
|
// Then the resulting tarball would have:
|
|
|
|
|
|
// - ./dir/file1 (from layer1)
|
|
|
|
|
|
// - ./dir/ (from layer0)
|
|
|
|
|
|
// Which would mean `untar` would try to untar a file to a directory which
|
|
|
|
|
|
// was not yet created. Therefore directories will be copied to the resulting
|
|
|
|
|
|
// tar in the order they appear in the layers.
|
|
|
|
|
|
//
|
|
|
|
|
|
// == Special files: .dockerenv ==
|
|
|
|
|
|
//
|
|
|
|
|
|
// .dockernv is present in all docker containers, and is likely to remain
|
|
|
|
|
|
// such. So if you do `docker export <container>`, the resulting tarball will
|
|
|
|
|
|
// have this file. rootfs will not add it. You are welcome to append one
|
|
|
|
|
|
// yourself.
|
|
|
|
|
|
//
|
|
|
|
|
|
// == Special files: opaque files and dirs (.wh.*) ==
|
|
|
|
|
|
//
|
|
|
|
|
|
// From mount.aufs(8)[1]:
|
|
|
|
|
|
//
|
|
|
|
|
|
// The whiteout is for hiding files on lower branches. Also it is applied to
|
|
|
|
|
|
// stop readdir going lower branches. The latter case is called ‘opaque
|
|
|
|
|
|
// directory.’ Any whiteout is an empty file, it means whiteout is just an
|
|
|
|
|
|
// mark. In the case of hiding lower files, the name of whiteout is
|
|
|
|
|
|
// ‘.wh.<filename>.’ And in the case of stopping readdir, the name is
|
|
|
|
|
|
// ‘.wh..wh..opq’. All whiteouts are hardlinked, including ‘<writable branch
|
|
|
|
|
|
// top dir>/.wh..wh.aufs`.
|
|
|
|
|
|
//
|
|
|
|
|
|
// My interpretation:
|
2021-05-24 00:11:58 +03:00
|
|
|
|
// - a file/hardlink called `.wh..wh..opq` means that directory contents from
|
|
|
|
|
|
// the layers below the mentioned file should be ignored. Higher layers may add
|
2021-05-24 00:11:58 +03:00
|
|
|
|
// files on top.
|
2021-05-24 00:11:58 +03:00
|
|
|
|
// * Ambiguity: should the directory from the lower layers be removed? I am
|
|
|
|
|
|
// assuming yes, but this assumptions is baseless.
|
|
|
|
|
|
// - if file/hardlink `.wh.([^/]+)` is found, $1 should be deleted from the
|
|
|
|
|
|
// current and lower layers.
|
|
|
|
|
|
//
|
|
|
|
|
|
// Note: these may be regular files in practice. So this implementation will
|
|
|
|
|
|
// match either.
|
2021-05-24 00:11:58 +03:00
|
|
|
|
//
|
|
|
|
|
|
// == Tar format ==
|
|
|
|
|
|
//
|
|
|
|
|
|
// Since we do care about long filenames and large file sizes (>8GB), we are
|
|
|
|
|
|
// using "classic" GNU Tar. However, at least NetBSD pax is known to have
|
|
|
|
|
|
// problems reading it[2].
|
|
|
|
|
|
//
|
|
|
|
|
|
// [1]: https://manpages.debian.org/unstable/aufs-tools/mount.aufs.8.en.html
|
|
|
|
|
|
// [2]: https://mgorny.pl/articles/portability-of-tar-features.html
|
|
|
|
|
|
package rootfs
|
