Sakaki's EFI Install Guide/Installing the Gentoo Stage 3 Files

From Gentoo Wiki
Jump to: navigation, search


Our main goals in this section, which shadows Chapter 5 of the Gentoo handbook, will be to download and unpack the Gentoo 'stage 3' tarball, and to set up Portage's main configuration file (/etc/portage/make.conf).

We'll also present a brief backgrounder covering some key Portage topics, which may be skipped if desired.

Double-Checking the Date and Time

Before downloading anything, double-check that you have the correct time and date on the target machine (you should do, as you set it up earlier, but since having an incorrect date can cause problems later, it is best to make sure now). Check it with:

livecd ~ #date

Per the handbook, you should stick with UTC for now (the real timezone specification will come later in the install). If necessary, set the date and time, in MMDDhhmmYYYY format (Month, Day, hour, minute, year):

livecd ~ #date MMDDhhmmYYYY
Note
Substitute MMDDhhmmYYYY in the above with the correct date/time string. For example, to set the UTC date/time to 5:12pm on February 9th 2017, you would issue
livecd ~ #date 020917122017

Downloading, Verifying and Unpacking the Gentoo Stage 3 Tarball

The 'Stage 3' file is a tarball containing a populated directory structure from a basic Gentoo system. Unlike the minimal installation image however, the stage 3 system contains no kernel, only binaries and libraries essential to bootstrapping. As such, we will have to host it within a chroot from our existing (minimal install image) system, until we have recompiled all system files and libraries, built a fresh kernel, and the new system becomes self-hosting.

Note
You can use the system binaries and libraries shipped in the stage 3 as is, without rebuilding them, if you like - only the kernel build is mandatory. However, we give instructions here to rebuild everything, from the toolchain upwards, to achieve the same effect as an old 'stage 1' bootstrap. The parts that are optional will be flagged.

The stage 3 tarball is generally released on the same date as the minimal install image, and may be found (together with the usual contents and digest files) in the same autobuilds directory. As the amount of data involved here is small, pace the handbook we'll skip using the slightly awkward links browser / mirror selection process at this stage, and just grab the files directly with wget.

Change to the Gentoo filesystem root mountpoint:

livecd ~ #cd /mnt/gentoo

Now download the files. As before, substitute for YYYYMMDD in the below with the current release file date (open the link http://distfiles.gentoo.org/releases/amd64/autobuilds/latest-stage3-amd64.txt in a browser to determine the current name).

Note
Substitute the current release date for YYYYMMDD in the above commands.
Note
The autobuilds directory also contains a set of 'nomultilib' variants; these are not recommended for general use.

As before (when we downloaded the minimal install image), we have to go through a two-stage verification: first check the signature in the DIGESTS.asc file, and then check the digests in that file themselves. As this is the target machine (not the helper), we don't yet have the necessary Gentoo automated weekly release public key, whose fingerprint may be found on the Gentoo release engineering page. So let's fetch it now. Issue:

livecd /mnt/gentoo #gpg --keyserver pool.sks-keyservers.net --recv-key 2D182910
Note
As before, for this to work you must ensure you have enabled outbound access on your firewall for port 11371/tcp to allow HKP communication, along with the usual state-tracking input rule.
Alternatively, you can use the following command, to fetch the key over port 80 (which should be open on most firewalls):
livecd /mnt/gentoo #gpg --keyserver hkp://pool.sks-keyservers.net:80 --recv-key 2D182910
Note
If the above keyserver is unavailable for some reason, you should be able to use any other one, such as pgp.mit.edu for example.
Note
If the fingerprint has changed, substitute the correct value for 2D182910 in the above.

You should next verify that the key's full fingerprint matches that listed on the Release Engineering page:

livecd /mnt/gentoo #gpg --fingerprint 2D182910
pub   rsa4096 2009-08-25 [SC] [expires: 2019-08-22]
      13EB BDBE DE7A 1277 5DFD  B1BA BB57 2E0E 2D18 2910
uid           [ unknown] Gentoo Linux Release Engineering (Automated Weekly Release Key) <releng@gentoo.org>
Note
Although correct at the time of writing, the key ID you need to enter in the above command may differ from 2D182910, as may the fingerprint data shown. Always use the Release Engineering page data as your primary reference.

If all looks good, verify the digest file:

livecd /mnt/gentoo #gpg --verify stage3-amd64-*.tar.bz2.DIGESTS.asc
Note
As before, you can ignore gpg output such as:
gpg: WARNING: This key is not certified with a trusted signature!
gpg:          There is no indication that the signature belongs to the owner.

This is normal since you have just imported the public key, and not yet 'trusted' it.[1]

And assuming that worked (output reports 'Good signature'), next check the digests themselves; we'll use the SHA512 variants here:

livecd /mnt/gentoo #awk '/SHA512 HASH/{getline;print}' stage3-amd64-*.tar.bz2.DIGESTS.asc | sha512sum --check

If this outputs:

stage3-amd64-YYYYMMDD.tar.bz2: OK
stage3-amd64-YYYYMMDD.tar.bz2.CONTENTS: OK

then continue.

The last step in this stage is to unpack the tarball. Double check you are in the /mnt/gentoo directory, then issue:

livecd /mnt/gentoo #tar xvjpf stage3-amd64-*.tar.bz2 --xattrs --numeric-owner

As per the handbook, the options to tar are to extract, provide verbose output, decompress with bzip2 (j), preserve permissions and extract from a file, not standard input. Further, the --xattrs option ensures that extended attributes stored in the tarball are also copied, and the --numeric-owner option uses the numeric owner and group IDs from the tarball, even if they are different in the host system (they will not be here, since we are booted from an official minimal install image, but it doesn't hurt to specify this).

Check that the base system has unpacked OK:

livecd /mnt/gentoo #ls
bin   home   lost+found  proc  stage3-amd64-YYYYMMDD.tar.bz2              tmp
boot  lib    media       root  stage3-amd64-YYYYMMDD.tar.bz2.CONTENTS     usr
dev   lib32  mnt         run   stage3-amd64-YYYYMMDD.tar.bz2.DIGESTS.asc  var
etc   lib64  opt         sbin  sys

If you see a file structure similar to the above, then you can proceed. As we no longer need the stage files, they can be deleted now to save space:

livecd /mnt/gentoo #rm -v -f stage3-amd64-*

This structure looks, as it should, a lot like a normal Linux root directory, but it is positioned at /mnt/gentoo. Later, we'll bind in a few additional special directories from our running (minimal) system, and then chroot into this new base to continue with the installation.

Lastly, return back to root's home:

livecd /mnt/gentoo #cd ~

Before we continue, we'll take a brief detour to discuss some essential Gentoo / Portage background information and terminology. If you are an old hand with this, feel free to skip this material.

Gentoo, Portage, Ebuilds and emerge (Background Reading)

Gentoo is a source-based distribution, the heart of which is a powerful package manager called Portage. Portage itself has two main components:

  • the ebuild system, which performs the actual work of fetching, configuring, building and installing packages, and
  • the emerge tool, which provides a command line interface to invoke ebuilds, and also allows you to update the Portage tree (discussed below), resolve package dependencies, and other related tasks.
Note
If you're new to all of this, a useful introduction to Portage may be found in Chapter 1 of Part 2 of the Gentoo Handbook, and in this Wikipedia article. However, don't worry - you don't need to be an adept programmer to use Gentoo on a day-to-day basis! (In fact, if you'd like to skip over this background material now and continue with the next section of the install, just click here).

Package ebuilds are Bash shell scripts, or more accurately shell script fragments, that are sourced into a larger build system 'host' script. This host script provides a package management control flow that invokes a set of default 'hook' functions, which a particular package's ebuild may override if it needs to (these are covered in detail in the Gentoo Development Guide). The ebuild also must define a minimum set of variables to allow the whole process to operate successfully (for example, the URI from where a package's source tarball may be downloaded must be assigned to the SRC_URI variable).

Now, when you invoke an ebuild to install a particular (as yet uninstalled) package on your system (via emerge, for example, as described below), it will typically carry out the following tasks (inter alia):

  • check that the specified package can be installed (that is, that it isn't masked, or has an incompatible license requirement);
  • download the package's tarball (or other format source archive) from an upstream repository (or Gentoo mirror);
  • unpack the tarball in a temporary working area;
  • patch (and otherwise modify) the unpacked source if need be;
  • configure the source to prepare it for compilation on your machine;
  • compile / build the source, as a non-privileged user in the temporary work area;
  • run tests (if provided and required);
  • install the built package to a dummy filesystem root; and
  • copy ('merge') the package installation files from the dummy filesystem root to the real filesystem root (keeping a record of what gets done).

Up until the final file copy-over step (the 'merge' in emerge), all operations (even where the package's make install is invoked, for example) take place in a temporary staging area. This enables Portage to keep track of all the files installed by a particular package, limit the damage caused by failed compiles or installs, and facilitate simple removal of installed packages. Furthermore, for most of these tasks, Portage operates in a 'sandbox' mode, where attempts to write directly to the real root filesystem (rather than the temporary work area) are detected, and cause an error to be thrown (NB this is not intended as a security system per se, but it does help prevent accidental filesystem corruption).

Note
Portage will attempt to deal with build and runtime dependencies when emerging packages, and will automatically install such dependencies for you, by invoking their ebuilds.
Note
At this stage in the install, you won't be able to see the files referred to in the text below on your target PC, since the minimal install image has an empty /usr/portage directory, and the system we in the process of creating from the stage 3 tarball (whose root is currently at /mnt/gentoo) has no /usr/portage directory yet. This will be rectified in the next chapter.

Portage stores ebuilds in a hierarchical folder structure - the Portage tree (or repository), which by default is located under /usr/portage. The first tree level is the package category, which is used to organize packages into groups which have broadly similar functionality. So, for example, non-core development utilities are typically placed in the dev-util category (in folder/usr/portage/dev-util). The next tree level is the package name itself. To take a concrete example, the small utility diffstat (which, as its name suggests, displays a histogram of changes implied by a patch file, or other diff output), is located in the folder /usr/portage/dev-util/diffstat. Within that subdirectory we have the actual per-package content, specifically:

  • The ebuild files. Each supported version has a file of format <name>-<version>.ebuild. At the time of writing, there are two supported versions (1.60 and 1.61) of diffstat in the Portage tree, so the ebuilds are located at /usr/portage/dev-util/diffstat/diffstat-1.60.ebuild and /usr/portage/dev-util/diffstat/diffstat-1.61.ebuild. Portage supports a complex version numbering taxonomy which, for the most part, reflects upstream versioning (discussed further below), and most packages, like diffstat, will have multiple ebuild versions available at any given time.
  • Package metadata. This is stored in an xml-format text file (one per package), named metadata.xml. Its contents are described here, and can contain detailed package descriptions, email addresses for upstream maintainers, documentation about USE flags etc. diffstat's metadata file is at /usr/portage/dev-util/diffstat/metadata.xml.
  • A change log for the package's ebuild(s). This is a text file documenting what changes have been checked in to source control over time. The filename is ChangeLog, so diffstat's may be found at /usr/portage/dev-util/diffstat/ChangeLog.
  • A manifest file, which contains digests (SHA256, SHA512 and Whirlpool) and file sizes for the contents of the package directory and any referenced tarballs (and patches, if present). It is used to detect corruption and possible tampering during package download / installation. This manifest, which may optionally be digitally signed, is stored in the Manifest file; diffstat's therefore resides at /usr/portage/dev-util/diffstat/Manifest.
  • An optional files directory. This is used to hold patches and other small files that are supplementary to the main source tarball but referenced by one or more of the package's ebuilds. The directory may be absent if unused. As (at the time of writing) diffstat does not require patches, it has no files subdirectory either.
Note
Since the Portage tree, or repository, is nothing other than a set of files, it can easily be kept up to date with Gentoo's mirrored 'master copy' (and indeed by default this is done using rsync, whenever you issue an emerge --sync, for example).

A Simple ebuild (diffstat)

So what does an ebuild file actually look like, then? diffstat happens to be a good minimal example; here (at the time of writing) is what /usr/portage/dev-util/diffstat/diffstat-1.61.ebuild contains:

FILE /usr/portage/dev-util/diffstat/diffstat-1.61.ebuildA fairly minimal ebuild, relying on the default 'hook' functions and control flow
# Copyright 1999-2016 Gentoo Foundation
# Distributed under the terms of the GNU General Public License v2

EAPI=6

DESCRIPTION="Display a histogram of diff changes"
HOMEPAGE="http://invisible-island.net/diffstat/"
SRC_URI="ftp://invisible-island.net/diffstat/${P}.tgz"

LICENSE="HPND"
SLOT="0"
KEYWORDS="~alpha ~amd64 ~arm ~arm64 ~hppa ~ia64 ~mips ~ppc ~ppc64 ~s390 ~sh ~sparc ~x86 ~x86-fbsd ~amd64-linux ~x86-linux ~ppc-macos ~x86-macos ~sparc-solaris ~x86-solaris"
IUSE=""

Not a lot to see, is there? That's because diffstat uses a standard 'Autotools'-style build, without patches, so the default ebuild control flow (and invoked 'hook' functions) can do almost everything for us. Therefore, all that has to be done is:

  • to specify (via the EAPI variable) that the ebuild makes use of the most modern package manager functionality, including built-in default behaviours (version 6, at the time of writing).
  • to specify a brief DESCRIPTION, HOMEPAGE (both self-explanatory) and most importantly, SRC_URI; this last variable tells Portage the location from whence to download the package tarball, if it cannot find it in the Portage mirrors (the ${P} expands out to be the package name and version; a handy list of these special variables may be found here);
  • to specify the LICENSE (the relevant text may be found at /usr/src/portage/licenses/${LICENSE});
  • to specify that SLOTTING is not used by this ebuild (this is an advanced feature; see below for a brief overview); and
  • finally to list the architectures (KEYWORDS) for which this ebuild applies. Here, we can see that (at the time of writing) it is in testing (has a tilde) for all the architectures listed (alpha, amd64 etc.).
Note
At the time of writing, diffstat had no USE flags, hence IUSE="".

That's all that is needed in this case, because the default ebuild functions will automatically pull down the tarball, unpack it, issue a ./configure, issue a make, followed by a make install (to a dummy root), after which, the program file (plus manpage etc.) will be copied over ('merged') to the real filesystem (and any prior version's files safely unmerged immediately thereafter).

Note
In fact, the default ebuild flow will handle not just 'Autotools' packages, but also any package provided with a Makefile that can accept make and make install invocations, and respects the DESTDIR variable. By default, the ./configure step will be omitted if no configure file is found in the top-level source directory of the tarball after unpacking.

There are then two main ways to invoke the diffstat ebuild. The first (and more common way) is via emerge: typically, you would issue:

root #emerge --ask --verbose dev-util/diffstat
Note
This is just an example, this command is not part of the installation tutorial and you should not actually issue it at this point.
Note
It is also possible to tell emerge which specific version you want, for example, you could issue instead:
root #emerge --ask --verbose =dev-util/diffstat-1.60

This is an example of a qualified version atom, discussed below.

On the other side of the coin, it is possible to leave off the category qualifier when using emerge, but that's not recommended due to occasional ambiguities, where the same name occurs in multiple categories.

The second (lower level) way is invoke the ebuild directly; for example, you could issue:

root #cd /usr/portage/dev-util/diffstat/
root #ebuild diffstat-1.60.ebuild clean merge
Note
This is also just an example, these commands are not part of the installation tutorial and you should not actually issue them at this point.

which will clean Portage's temporary build directories, and then perform all the steps of the ebuild workflow, providing detailed output as it does so (you can also use the ebuild command to perform only certain steps, if you wish, and it can also create Manifest files; see the ebuild manpage for details).

Note
Unlike the emerge invocation, this will not add dev-util/diffstat to the @world set (see below for an explanation of what this means).

A More Complex ebuild (sign)

The diffstat example above is about as simple as a real-world ebuild gets!

However, one common additional requirement is the need to apply patches. To do this, an ebuild will typically override the default src_prepare ebuild 'hook' function (invoked by the standard ebuild flow after the source tarball has been successfully unpacked), and in the overridden version use the epatch utility function to apply patches held in the files directory.

Note
However, from EAPI 6 the default src_prepare function is no-longer a no-op, it will automatically apply any patches listed in the PATCHES array variable (and call eapply_user, to apply user patches)[2]. The ebuild we're about to look at however uses EAPI 5, so has to apply its required patches using epatch directly.

For example, consider the sign package, which provides a file signing and signature verification utility. It lives in the app-crypt category. Looking in its corresponding directory (/usr/portage/app-crypt/sign) we notice immediately that unlike diffstat, there is a files subdirectory, containing two patches (1.0.7-as-needed.patch and 1.0.7-openssl-0.9.8.patch).

Let's examine version 1.0.7 of the ebuild:

FILE /usr/portage/app-crypt/sign/sign-1.0.7.ebuildA slightly more complex ebuild, illustrating patching and conditional dependencies
# Copyright 1999-2017 Gentoo Foundation
# Distributed under the terms of the GNU General Public License v2

EAPI=5

inherit toolchain-funcs eutils

DESCRIPTION="File signing and signature verification utility"
HOMEPAGE="http://swapped.cc/sign/"
SRC_URI="http://swapped.cc/${PN}/files/${P}.tar.gz"

LICENSE="BZIP2"
SLOT="0"
KEYWORDS="amd64 ppc x86 ~amd64-linux ~x86-linux ~ppc-macos"
IUSE="libressl"

RDEPEND="
	!libressl? ( dev-libs/openssl:0= )
	libressl? ( dev-libs/libressl:0= )"
DEPEND="${RDEPEND}"

src_prepare() {
	epatch "${FILESDIR}"/${PV}-openssl-0.9.8.patch
	epatch "${FILESDIR}"/${PV}-as-needed.patch
	# remove -g from CFLAGS, it happens to break the build on ppc-macos
	sed -i -e 's/-g//' src/Makefile || die
}

src_compile() {
	emake CC="$(tc-getCC)"
}

src_install() {
	dobin ${PN}
	doman man/${PN}.1
	dodoc README
	dosym ${PN} /usr/bin/un${PN}
}

Most of this should be familiar enough from the diffstat example, but there are some new elements too. Specifically:

  • the inherit command is used to pull in two useful 'eclasses': eutils (which supplies the epatch function discussed shortly) and toolchain-funcs (which supplies tc-getCC, a function to return the name of the toolchain C compiler);
  • the SRC_URI makes use of the ${PN} variable, which expands out to the package name, without version (a full list of these convenience variables may be found here);
  • the IUSE definition is not blank: there is one optional USE flag here, libressl (which switches to that SSL library, as we shall discuss next; see below for a brief introduction to USE flags);
  • the RDEPEND variable specifies a set of runtime dependencies, and the DEPEND a set of build/install time dependencies, for the package. This is used by Portage to ensure that all prerequisites are also installed, when you ask to emerge app-crypt/sign. Notice that this build pulls in dev-libs/openssl by default, unless the libressl USE flag is specified, in which case dev-libs/libressl is pulled in instead;
  • the src_prepare 'hook' function (which by default is a no-op at EAPI 5) is overridden to perform two custom tasks:
    • to patch the source using the epatch utility, using patch files /usr/portage/app-crypt/sign/files/1.0.7-openssl-0.9.8.patch and /usr/portage/app-crypt/sign/files/1.0.7-as-needed.patch (${PV} expands to the package version, without name or revision tags). As described here, epatch intelligently attempts to apply patches using different -p levels etc.
    • to invoke a small sed script to modify the Makefile slightly.
  • the src_compile 'hook' function (which by default will simply invoke emake after some pre-preprocessing) is overridden to ensure that the C compiler is set correctly (the upstream Makefile not supporting environment-set CC values in this case);
  • the src_install 'hook' function (which by default will invoke emake DESTDIR="${D}" install and then further install any documentation specified via the DOCS array variable) is overridden to perform a manual install using some helper install functions; this is sometimes necessary if the upstream Makefile does not support the use of DESTDIR, or does not support the install target.
Note
Of course, ebuilds can be much more complex than either of the above two examples, but they should give you a basic idea of how the system works. For more details, I'd refer you to the Gentoo Development Guide "Ebuild Writing" section.

Ebuild Repositories (aka Overlays)

What if you want to modify an ebuild yourself, or add a new one? You could of course submit the ebuild to Gentoo using Bugzilla, but that only really applies to completed work you want to share. For work in progress, or private ebuilds, a different approach is required. You can't simply insert new entries into the /usr/portage tree, as they'll get overwritten next time you synchronize the Gentoo repository.

Instead, Portage supports the concept of ebuild repositories (historically known as "overlays") to address just this issue. These are simply additional collections of ebuilds and associated metadata, laid out in a similar filestructure to the main Portage tree, which Portage (by default, and as the historical name suggests) 'overlays' on the /usr/portage file structure. To illustrate, suppose you created a directory at, say, /tmp/myrepo, created the subfolders /tmp/myrepo/dev-util and /tmp/myrepo/dev-util/diffstat, then created an ebuild /tmp/myrepo/dev-util/diffstat/diffstat-1.60.ebuild (and manifest, /tmp/myrepo/dev-util/diffstat/Manifest), and then created the following (in /etc/portage/repos.conf/myrepo.conf, to inform Portage's plug-in sync system of its presence):

FILE /etc/portage/repos.conf/myrepo.confExample ebuild repository configuration file
[myrepo]
# Simple example 'overlaid' ebuild repository
location = /tmp/myrepo
priority = 100
auto-sync = no

Then, when referring to (or installing) diffstat, Portage would use your version, rather than the 'official' ebuild (however, if you had created an ebuild with a lower version number, say 1.57, then by default Portage would still use the higher numbered version, from the official /usr/portage 'underlay').

Note
There are actually a few more files you'd need to create in your overlay to make it functional (and you probably want to place it under source control, and not host the local copy under /tmp in any event!). See these instructions for further details.

We'll exploit this ability shortly, when we add the sakaki-tools ebuild repository (which will contain a number of useful tools used in this installation walk-through).

Portage's Configuration Files

Portage provides you, the user, with a great deal of flexibility. As such, it has many configuration options, specified via a set of files held in the /etc/portage directory (and subdirectories thereof). As our installation process is going to involve using Portage (via the command-line tool emerge) to download, then build and install up-to-date versions of all core system software, we first need to set up these configuration files appropriately.

The most important Portage configuration files you'll need to know about now are as follows (this is not complete - see this list for more information, and also the Portage manpage[3]):

Important
It is possible to have any of the below as subdirectories, rather than files, in which case the contents of the subdirectory will be parsed in alphabetical order.[4] Indeed, the subdirectory-based approach has now become the default on the Gentoo installation media (with the exception of /etc/portage/make.conf), so please bear that in mind when reading the table.
File in /etc/portage/ File Description
repos.conf Specifies site-specific repository configuration, including the mechanism and URI via which repositories should be synchronized.
make.conf Contains definitions of various important variables used by Portage. These variables tell the system, amongst other things:
  • the default licensing to accept (for example, accept only free software),
  • the system architecture to target (for example, the 'amd64' architecture - actually a generic reference to 64-bit processors, whether from AMD or Intel - used here),
  • some information about the system video card and input devices
  • whether to build the 'stable' versions of packages, or the latest, 'testing' version (happily, Gnome3 is present in the 'stable' branch of Gentoo, so we can use this);
  • the default system-wide USE flags (USE flags are Portage 'meta-instructions' to control the build process for packages; they are a core Portage concept and introduced in Chapter 2 of Part 2 of the Gentoo Handbook);
  • what URLs to use when syncing the Portage tree (and many source tarballs) (we'll want to point to local mirrors, to keep things fast),
  • what logging to perform during builds (we'll switch this on, since its very useful when things go wrong)
  • and many others.

To ease the problem of setting things up correctly for your particular use case (e.g., a headless server, or, as in our case, a GNOME desktop with systemd or OpenRC), Portage makes use of profiles. A profile specifies (inter alia) a set of default values for most of the variables in /etc/portage/make.conf, which will be used if the appropriate variable is not defined in the user's environment (checked first) or in the /etc/portage/make.conf file. (NB - so-called incremental variables, such as the one which holds the list of USE flags, are an exception to this masking approach, as they 'cascade' additively, from profile, through /etc/portage/make.conf, to the user's environment.)

It is also possible to specify overrides for certain elements of Portage's operation on a per-package basis, through the use of the following configuration files:

package.mask Versions of packages specified in package.mask are 'masked' - that is, blocked from installation (think of it as an installation blacklist). This is most commonly used to prevent Portage updating a package when there is some bug or incompatibility with the new release. It is also sometimes used to mask out everything in a large third-party ebuild repository, for security (with only the specific packages that are wanted then being allowed, by explicit citation in the package.unmask file (discussed next).
package.unmask This file overrides package.mask (think of it as an installation whitelist). It is sometimes used to allow 'activate' specific packages only from a large ebuild repository (which has been otherwise totally masked via package.mask, above).
package.use package.use contains a list of USE flags for individual packages. It comes in handy when specifying flags that have only localized meaning (e.g., suppressing the installation of Guest Additions in VirtualBox), or which you only want to turn on in very selective situations (such as the test flag, for example). You can also turn off USE flags for particular packages, by prefixing them with a minus sign ('-').
package.license The package.license file allows you to specify allowed licenses on a per-package basis. It's generally used where you have a restrictive licensing default (such as 'free software only', as we are going to set), but need to add some exceptions for a few cases.
package.accept_keywords The package.accept_keywords file primarily allows you to specify packages which should use the testing, rather than stable, software branch. It is best to keep the use of this to a minimum, to avoid dependency pollution, but it is sometimes necessary (for example, when using software for which no stabilized version yet exists in the tree).
Note
There are other things you can do with package.accept_keywords too, such as activating so-called 'live' (aka '9999') ebuilds, which track the tip of a branch in a version control system directly, but we will not utilise this in our tutorial.
env The env directory contains custom environment files that can be used to override default emerge behaviour, when cited for a given package in package.use (see below). For example, you could create a file called /etc/portage/env/no_build_parallelism.conf, and put in it MAKEOPTS="-j1". Then, you could apply this custom environment setting to any package that had a problem with this issue, as described next.
package.env The package.env file allows you to apply custom environment settings (as defined in /etc/portage/env/..., see above) to particular packages. For example, you could turn off build parallelism for a package by citing no_build_parallelism.conf against it, here.

Atoms, Packages, Categories, Versions, Sets and SLOTs

Finally for this background overview, there are a few Portage package mangement terms that are worth a brief recap:

  • As mentioned, a package refers to a homogeneous block of software which has a single provided ebuild per installable version, whether third-party (e.g., openvpn) or internal to Gentoo itself (e.g., gentoolkit).
  • Packages are grouped (as leaves of a tree) into categories, which describe broad classes of functionality. For example, openvpn is in the net-vpn category (along with other similar tools like tor and strongswan); gentoolkit is in the app-portage category (along with other Portage applications, like mirrorselect and elogviewer).
  • A package base atom simply refers to the name made up of the full category, followed by the package, without version information or other qualifiers. So for example net-misc/openvpn, app-portage/gentoolkit etc. You can find all the ebuilds in the currently sync'd tree for a given <category>/<packagename> base atom in the directory /usr/portage/<category>/<packagename> (so, for example, /usr/portage/dev-util/diffstat/), and find more information about that base atom online at https://packages.gentoo.org/package/<category>/<packagename> (so, for example, https://packages.gentoo.org/package/app-portage/gentoolkit). While it is often possible to drop the category name and simply use the package itself, it's generally safer to use the base atom, since two different packages of the same name may exist in different categories (e.g. axiom could refer to either dev-python/axiom, an object database over SQLite, or sci-mathematics/axiom, a computer algebra system).
  • It is generally possible to specify that a specific repository should be used to supply a package, by appending ::<reponame> to its atom. For example, emerge --ask --verbose dev-util/diffstat::myrepo would force Portage to install the diffstat package from the myrepo repository (and would fail if either that overlay was unknown, or if the dev-util/diffstat package was not present in it).
  • Any given package will normally be supported at multiple versions within Portage (one ebuild per version). Not all versions from the upstream tree may be present as ebuilds, only certain selected versions. The online package data referred to above will show what versions are available, on which architectures, and which are marked as 'stable', which are 'testing' (shown with a tilde ('~')), and which are masked (will not be installed by Portage, generally due to known problems or with the ebuild, or incompatibilities with other packages). You can fully qualify an atom by specifying its version as a suffix - generally, you take the base atom, then add a hyphen ('-'), then add a period-separated list of numbers (possibly finishing with a letter, and/or a revision suffix). So, for example, version 2.4.3 of openvpn would be written as net-vpn/openvpn-2.4.3; version 1.19.1 (r1) of wget as net-misc/wget-1.19.1-r1. Revisions are Gentoo ebuild specific, they do not relate to upstream versioning (one implication of which being, that different revisions of a particular version of a package will generally use the same upstream source tarball (although they may of course apply different patch sets etc.)).
  • When specifying atoms to Portage in certain places (such as configuration files, like /etc/portage/package.use), you can either specify base atoms (meaning apply the action to all ebuild versions), or a qualified version atom. You can qualify a versioned atom with:
    • A prefix ('>', '>=', '=', '<=', '<'], to restrict the action to particular versions relative to the stated variant (for example, if you appended ">=net-vpn/openvpn-2.4.3 inotify" to /etc/portage/package.use, you'd be telling Portage to apply the inotify use flag to any version of openvpn at or above 2.4.3.
    • A extended prefix: there are a number of these but the most important is '~', which is used to specify any revision of the base version specified. So, for example, ~app-portage/gentoolkit-0.3.3 would refer to app-portage/gentoolkit-0.3.3, app-portage/gentoolkit-0.3.3-r1, app-portage/gentoolkit-0.3.3-r2 etc. (where they exist, of course!)
    • A wildcard suffix ('*'). This can be used to match any version with the same string prefix. So for example, net-vpn/openvpn-2.4* would match (at the time of writing) net-vpn/openvpn-2.4.2-r1, net-vpn/openvpn-2.4.3, net-misc/openvpn-2.4.3-r1 etc.

For more information on atom naming, see the ebuild (5) manpage.[5]

  • A number of atoms may be grouped together into a set, so that operations (e.g. reinstallation) can be easily targeted at the whole group. Sets are special names and are prefixed by '@': some of these are pre-defined in Portage (for example, the @system set (containing vital system software packages, the contents of the stage 3 tarball plus other component dictated by your profile), or the dynamically populated @preserved-rebuild set (which holds a list of packages using libraries whose sonames have changed (during an upgrade or downgrade) but whose rebuild has not been triggered automatically). The @world set refers to all packages you explicitly requested be installed, and is contained in a file /var/lib/portage/world (note however that operations on the @world set will include the @system set, by default, not just what is in the /var/lib/portage/world file). You can even define your own sets if you like.
  • Portage also allows (subject to certain limitations) different versions of the same package to exist on a machine at the same time: we speak of them being installed in different SLOTs. We won't need to refer to the SLOT technology explicitly in this tutorial, but should you see a versioned atom with a colon ':' followed by some numbers and possibly other characters at the end, that's a SLOT reference. For example, with the x11-libs/gtk+ library, it is possible (at the time of writing) to have version 2.24.31-r1 and 3.22.15 installed in parallel, should you desire it (in SLOTs 2 and 3).[6] You might then see a reference to x11-libs/gtk+:3, which would refer to any version of gtk+ in SLOT 3 (which would, for example, cover version 3.22.16 as well).

That's about it for this sidebar on atoms and versioning, apart from one last point: unlike other Linux distributions, you'll see no reference to 'releases' of Gentoo itself - there's nothing similar to Ubuntu's "Xenial Xerus" or "Artful Aardvark", Debian's "Stretch" or "Buster", Fedora's "Heisenbug" or "v26" etc. That's because, once installed, Gentoo itself is essentially versionless - when you update your system (more on which later), all installed software updates to the latest supported versions (subject to restrictions imposed by the Gentoo developers and you yourself, through settings in /etc/portage/make.conf, /etc/portage/package.mask etc.).

The upside of this is that you can get access to the latest and (often) greatest versions of software as soon as new ebuilds get released into the tree. The downside is that (particularly on the 'testing' (rather than the 'stable') branch), sometimes updates fail to complete successfully, an occurrence that is very rare indeed when using binary distributed, release-based distributions such as Ubuntu.

Time to get back to the install!

Configuring /etc/portage/make.conf

Our first Portage configuration task is to ensure that the download / unpack / prepare / configure / compile / install / merge cycle (aka 'emerging') - which you'll see rather a lot of when running Gentoo - is as efficient as possible. That primarily means taking advantage of whatever parallelism your system can offer.

Important
Remember that we have not yet performed a chroot. As such, our vestigial system is still mounted at /mnt/gentoo. Therefore, our new system configuration files are at /mnt/gentoo/etc/portage, not /etc/portage, and so on. Confusingly, since the minimal install system is also a Gentoo system, there actually is a /etc/portage directory, but the files in there are not the ones you need to edit. Make sure you don't get mixed up! In what follows, if you are instructed to edit a file, its full path (including mountpoint prefix) will always be given, to avoid any ambiguity.

There are two main dimensions to this - the maximum number of concurrent Portage jobs that will be run at any one time, and the maximum number of parallel threads executed by the make process invoked by each ebuild itself.

As has been recommended, we'll set our number of concurrent jobs and parallel make threads to attempt, to be equal to the number of CPUs on the system, plus one.[7] We'll also prevent new jobs or compilations starting when the system load average hits or exceeds the number of CPUs.

The two variables we'll need to set here are EMERGE_DEFAULT_OPTS (for Portage job control) and MAKEOPTS (to pass options on to make). These are often defined in the make.conf file, but we want to allow the values to be set programmatically. Since Portage doesn't support fancy bash features like command substitution,[8] we'll set and export these variables in root's .bashrc instead (these will then override any conflicting values in the make.conf or profile, as explained earlier).

Note
Generally speaking, emerge is launched as the root user (superuser) in Gentoo. emerge usually drops its privilege level to run as the "portage" user when compiling.

Start up your favourite editor: in this tutorial we'll be assuming nano:

livecd ~ #nano -w /mnt/gentoo/root/.bashrc
Note
The -w option tells the nano editor not to auto-wrap long lines (auto-wrapping can really mess up config files!).

nano is a pretty simple editor to use: move around using the arrow keys, type to edit as you would in any text processing program, and exit with Ctrlx when done: you'll be prompted whether to save changes if you have modified the file. At this point, enter y and Enter to exit, saving changes, or n to exit without making changes. For some more information on the nano editor, see this Wiki entry.

Add the following text to the file:

FILE /mnt/gentoo/root/.bashrcSetting up MAKEOPTS and EMERGE_DEFAULT_OPTS
export NUMCPUS=$(nproc)
export NUMCPUSPLUSONE=$(( NUMCPUS + 1 ))
export MAKEOPTS="-j${NUMCPUSPLUSONE} -l${NUMCPUS}"
export EMERGE_DEFAULT_OPTS="--jobs=${NUMCPUSPLUSONE} --load-average=${NUMCPUS}"

Save and exit the nano editor.

Note
Should you experience problems with parallel make, and wish to fall back to a more conservative setting, you can do so globally by setting MAKEOPTS="-j1" in the above.

Next, we need to make sure that the .bashrc file is picked up by root's login shell, so copy across the default .bash_profile:

livecd ~ #cp -v /mnt/gentoo/etc/skel/.bash_profile /mnt/gentoo/root/

Next, on to the make.conf configuration file itself. The stage 3 tarball we extracted already contains a skeleton configuration. We'll open this file with nano (feel free to substitute your favourite alternative editor), delete the existing lines (in nano, Ctrlk can be used to quickly cut the current line), and enter our alternative configuration instead (see after for a line-by-line explanation). Issue:

livecd ~ #nano -w /mnt/gentoo/etc/portage/make.conf

Edit the file so it reads:

FILE /mnt/gentoo/etc/portage/make.confSetting up essential Portage variables
# Build setup as of <add current date>

# C and C++ compiler options for GCC.
CFLAGS="-march=native -O2 -pipe"
CXXFLAGS="${CFLAGS}"

# Note: MAKEOPTS and EMERGE_DEFAULT_OPTS are set in .bashrc

# Only free software, please.
ACCEPT_LICENSE="-* @FREE CC-Sampling-Plus-1.0"

# WARNING: Changing your CHOST is not something that should be done lightly.
# Please consult http://www.gentoo.org/doc/en/change-chost.xml before changing.
CHOST="x86_64-pc-linux-gnu"

# Use the 'stable' branch - 'testing' no longer required for Gnome 3.
ACCEPT_KEYWORDS="amd64"

# Additional USE flags in addition to those specified by the current profile.
USE=""
CPU_FLAGS_X86="mmx sse sse2"

# Important Portage directories.
PORTDIR="/usr/portage"
DISTDIR="${PORTDIR}/distfiles"
PKGDIR="${PORTDIR}/packages"

# Turn on logging - see http://gentoo-en.vfose.ru/wiki/Gentoo_maintenance.
PORTAGE_ELOG_CLASSES="info warn error log qa"
# Echo messages after emerge, also save to /var/log/portage/elog
PORTAGE_ELOG_SYSTEM="echo save"

# Ensure elogs saved in category subdirectories
FEATURES="split-elog"

# Settings for X11
VIDEO_CARDS="intel i965"
INPUT_DEVICES="evdev synaptics"
Note
Set VIDEO_CARDS and INPUT_DEVICES to appropriate values for your particular system in /etc/portage/make.conf. See table below for discussion.
Important
As discussed below, ensure that you only specify CPU_FLAGS_X86 flags that your CPU supports, otherwise compiled software may crash.

Save the file and exit nano.

Here is a brief summary of the shipped ('stage 3') values are, and what our version achieves:

Variable Value in Stage 3's make.conf Our Value Description
CFLAGS -O2 -pipe -march=native -O2 -pipe This variable is used to inform the GNU Compiler Collection (GCC) what switches to use when compiling source code. The defaults specify -O2, which sets the recommended level of optimization (producing smaller, faster code at the expense of a slightly longer compilation), and -pipe, which instructs the compiler to use pipes rather than temporary files where possible (speeding up compilation in exchange for additional memory requirements). We retain these, and add -march=native. This instructs the compiler to detect your CPU type automatically, and then produce code exploiting its idiosyncratic features, special instruction sets and so on. Setting -march=native implies that code produced will very likely not run on other CPUs: don't use it if you intend to compile packages for use on other machines!
CXXFLAGS ${CFLAGS} ${CFLAGS} This is the equivalent of CFLAGS for C++ code. We retain the default behaviour, which is to copy the CFLAGS.
ACCEPT_LICENSE absent -* @FREE CC-Sampling-Plus-1.0 This incremental variable controls which licenses are acceptable for packages used on your system, another nice feature of Portage. Here, we first of all disable any permitted licenses that may have been inherited from the profile (with -*) and then enable only the 'free' license metaset (with @FREE). (This is 'free' in the Free Software Foundation sense, so is relatively safe.[9])
We also allow CC-Sampling-Plus-1.0; essentially a free-use license, but one which is not currently included in @FREE; it may be viewed at /usr/portage/licenses/CC-Sampling-Plus-1.0. It is needed for some GNOME components.
CHOST x86_64-pc-linux-gnu x86_64-pc-linux-gnu The CHOST variable is very important. It is a dash-separated tuple of architecture-vendor-operating_system-C_library and is used to control the build process. The default value here (architecture: x86_64, vendor: pc, operating system: linux, C library: gnu) is fine for our purposes so we will not change it. In general, you should simply copy this value across from the original stage 3 make.conf file.
ACCEPT_KEYWORDS absent amd64 This variable instructs Portage which ebuild keywords it should accept. As Gnome 3 has now been stabilized, there is no need to use the 'testing' branch; (but should you wish to do so, use '~amd64' rather than 'amd64'; please note that a consequence of doing so is that you will receive very up-to-date versions of all the software on your system (good), and occasionally, you may encounter the odd problem when updating (due to conflicts or bugs that have not yet surfaced and been resolved) (not so good)). (For avoidance of doubt, amd64 covers both Intel and AMD processors with a 64-bit architecture.)
USE bindist "" As discussed above, use flags specify package features to Portage (and often, but not always, map directly to autoconf feature options 'under the cover'[10][11][12]). As we will be building packages for a personal machine, and not for binary redistribution, we omit the bindist flag; omitting it allows certain additional codecs etc. to be enabled.
CPU_FLAGS_X86 mmx sse sse2 mmx sse sse2
will set next chapter
This variable instructs Portage which processor-specific flags to use (specifying the availability of particular capabilities such as MMX, for example). It is now recommended to use this separate flag group (which is valid on amd64 also, despite the name), rather than place CPU flags directly into USE. We leave the default settings for now, but will use the app-portage/cpuinfo2cpuflags package to derive the appropriate optimized settings for us automatically (from /proc/cpuinfo), in the next chapter. (Note - these architecture flags should not be mixed up with the compiler-related CFLAGS and CXXFLAGS, although they appear somewhat similar. Generally, architecture use flags will set package-features (for example, in ffmpeg, enabling specific blocks of pre-written assembly code).)
PORTDIR /usr/portage /usr/portage This variable simply defines the location of the Portage tree. We leave it as-is.
DISTDIR ${PORTDIR}/distfiles ${PORTDIR}/distfiles This variable defines where Portage will store its source code tarballs. Again, we leave it as is.
PKGDIR ${PORTDIR}/packages ${PORTDIR}/packages This variable decides where binary packages will be stored, should you decide to download them (as an alternative to compiling from source), or to create your own (as a side-effect of compiling from source). We won't be doing either in this tutorial, but leave the setting as is anyhow. (Note - if you do create binaries for redistribution, you must set the bindist use flag.)
PORTAGE_ELOG_CLASSES absent info warn error log qa This variable tells Portage what kinds of ebuild messages you want logged. The flags given here switch on all messages; modify to suit your own requirements (see the Gentoo Handbook, part 3 chapter 1 for more details).
PORTAGE_ELOG_SYSTEM absent echo save This variable instructs Portage what to do with log messages - in this case echo them to the console after the emerge, and also save them (rather than pass them to a user-defined command, etc.) Note that you can also instruct Portage to save your full build logs if you wish: see swift's blog post here.
FEATURES absent split-elog As its name suggests, this incremental variable is used to turn on (or off) optional Portage features. In this case, we turn on split-elog, which ensures that the logs just discussed get saved in category subdirectories of /var/log/portage/elog; this makes them easier to navigate. If you want to add additional features, just append them to this variable, separated by a space.
VIDEO_CARDS absent intel i965 This variable is used to inform various packages which video card you have in your system (it is a USE_EXPAND variable). You can omit it, in which case modular support for all available systems is implied (as such, it's more efficient to specify it). The Panasonic CF-AX3 has modern integrated Intel graphics, as do many laptops, so we specify intel i965 here. If you have an nVidea card, and wish to use open-source drivers, you should specify nouveau instead here, for example; if an old ATI card from way before it was purchased by AMD, ati; if a pre-2015 ATI/AMD Radeon card, radeon; if you run a brand new Radeon R9 390 or RX 480, radeon amdgpu radeonsi, etc. (See these comments on the Gentoo wiki.)
Note that for a simple fallback driver, which should work on most systems (albeit with relatively low resolution and performance), you can also specify vesa here (however, the vesa driver currently does not work correctly when booted as a VirtualBox guest under EFI[13]). Another useful fallback value is fbdev, which specifies the simple x11-drivers/xf86-video-fbdev framebuffer device video driver.
INPUT_DEVICES absent evdev synaptics This variable instructs the X Window server (which we will be installing shortly) which input devices to support. It is also a USE_EXPAND variable. Most systems will require evdev; as the Panasonic CF-AX3 is a laptop with a touchpad, we also include synaptics (see these comments on the Gentoo wiki).

Next Steps

Now we have these options configured, we're ready to chroot into our 'stage 3' environment and start building! Click here to go to the next chapter, "Building the Gentoo Base System Minus Kernel".

Notes

  1. Information Security Stack Exchange: "Ways to sign gpg public key so it is trusted?"
  2. Gentoo Development Guide: "EAPI Usage and Description"
  3. Portage manpage
  4. Gentoo Wiki Archives: "TIP_New_/etc/portage_layout"
  5. ebuild (5) manpage
  6. Gentoo Development Guide: "Slotting"
  7. Preney, Paul. "Parallel Builds with Gentoo's Emerge"
  8. make.conf manpage
  9. Free Software Foundation: "What is free software?"
  10. Gentoo Development Guide: "The Basics of Autotools"
  11. Gentoo Development Guide: "Quickstart Ebuild Guide: Ebuild with USE Flags"
  12. Calcote, John. Autotools. No Starch Press, 2010. "Supporting Optional Features and Packages", p. 107 ff.
  13. Red Hat Bugzilla: Bug 742695: "EFI install in VirtualBox results in text install "
< Previous Home Next >