freebsd/share/man/man7/development.7

.\" Copyright (c) 1998, Matthew Dillon.  Terms and conditions are those of
.\" the BSD Copyright as specified in the file "/usr/src/COPYRIGHT" in
.\" the FreeBSD source tree.
.\"
.\" $FreeBSD$
.\"
.Dd December 21, 2002
.Dt DEVELOPMENT 7
.Os
.Sh NAME
.Nm development
.Nd introduction to development with the FreeBSD codebase
.Sh DESCRIPTION
This manual page describes how an ordinary sysop,
.Ux admin, or developer
can, without any special permission, obtain, maintain, and modify the
.Fx
codebase as well as how to maintaining a master build which can
then be exported to other machines in your network.
This manual page
is targeted to system operators, programmers, and developers.
.Pp
Please note that what is being described here is based on a complete
FreeBSD environment, not just the FreeBSD kernel.
The methods described
here are as applicable to production installations as it is to development
environments.
You need a good 12-17GB of disk space on one machine to make this work
conveniently.
.Sh SETTING UP THE ENVIRONMENT ON THE MASTER SERVER
Your master server should always run a stable, production version of the
.Fx
operating system.   This does not prevent you from doing -CURRENT
builds or development.  The last thing you want to do is to run an
unstable environment on your master server which could lead to a situation
where you lose the environment and/or cannot recover from a mistake.
.Pp
Create a huge partition called /FreeBSD.
8-12GB is recommended.
This partition will contain nearly all the development environment,
including the CVS tree, broken-out source, and possibly even object files.
You are going to export this partition to your other machines via a
READ-ONLY NFS export so do not mix it with other more security-sensitive
partitions.
.Pp
You have to make a choice in regards to
.Pa /usr/obj .
You can put
.Pa /usr/obj
in
.Pa /FreeBSD
or you can make
.Pa /usr/obj
its own partition.
I recommend making it a separate partition for several reasons.  First,
as a safety measure since this partition is written to a great deal.
Second, because you typically do not have to back it up.
Third, because it makes it far easier to mix and match the development
environments which are described later in this document.
I recommend a
.Pa /usr/obj
partition of at least 5GB.
.Pp
On the master server, use cvsup to automatically pull down and maintain
the
.Fx
CVS archive once a day.  The first pull will take a long time,
it is several gigabytes, but once you have it the daily syncs will be quite
small.
.Bd -literal -offset 4n
mkdir /FreeBSD/FreeBSD-CVS
rm -rf /home/ncvs
ln -s /FreeBSD/FreeBSD-CVS /home/ncvs
.Ed
.Pp
The cron job should look something like this (please randomize the time of
day!).
Note that you can use the cvsup file example directly from
/usr/share/examples without modification by supplying appropriate arguments
to cvsup.
.Bd -literal -offset 4n
33 6 * * *      /usr/local/bin/cvsup -g -r 20 -L 2 -h cvsup.freebsd.org /usr/share/examples/cvsup/cvs-supfile
.Ed
.Pp
Run the cvsup manually the first time to pull down the archive.  It could take
all day depending on how fast your connection is!
You will run all cvsup and cvs operations as root and you need to set
up a ~/.cvsrc (/root/.cvsrc) file, as shown below, for proper cvs operation.
Using ~/.cvsrc to specify cvs defaults is an excellent way
to "file and forget", but you should never forget that you put them in there.
.Bd -literal -offset 4n
# cvs -q
diff -u
update -Pd
checkout -P
.Ed
.Pp
Now use cvs to checkout a -STABLE source tree and a -CURRENT source tree,
as well as ports and docs, to create your initial source environment.
Keeping the broken-out source and ports in /FreeBSD allows you to export
it to other machines via read-only NFS.
This also means you only need to edit/maintain files in one place and all
your clients automatically pick up the changes.
.Bd -literal -offset 4n
mkdir /FreeBSD/FreeBSD-4.x
mkdir /FreeBSD/FreeBSD-current

cd /FreeBSD/FreeBSD-4.x
cvs -d /home/ncvs checkout -rRELENG_4 src

cd /FreeBSD/FreeBSD-current
cvs -d /home/ncvs checkout src
cvs -d /home/ncvs checkout ports
cvs -d /home/ncvs checkout doc
.Ed
.Pp
Now create a softlink for /usr/src and /usr/src2.
On the main server I always point /usr/src at -STABLE and /usr/src2 at
-CURRENT.  On client machines I usually do not have a /usr/src2 and I make
/usr/src point at whatever version of FreeBSD the client box is intended to
run.
.Bd -literal -offset 4n
cd /usr
rm -rf src src2
ln -s /FreeBSD/FreeBSD-4.x/src src	(could be -CURRENT on a client)
ln -s /FreeBSD/FreeBSD-current/src src2	(MASTER SERVER ONLY)
.Ed
.Pp
Now you have to make a choice for /usr/obj.
Well, hopefully you made it already and chose the partition method.  If you
chose poorly you probably intend to put it in /FreeBSD and, if so, this is
what you want to do:
.Bd -literal -offset 4n
(ONLY IF YOU MADE A POOR CHOICE AND PUT /usr/obj in /FreeBSD!)
mkdir /FreeBSD/obj
cd /usr
rm -rf obj
ln -s /FreeBSD/obj obj
.Ed
.Pp
Alternatively you may chose simply to leave /usr/obj in /usr.  If your
/usr is large enough this will work, but I do not recommend it for
safety reasons (/usr/obj is constantly being modified, /usr is not).
.Pp
Note that exporting /usr/obj via read-only NFS to your other boxes will
allow you to build on your main server and install from your other boxes.
If you also want to do builds on some or all of the clients you can simply
have /usr/obj be a local directory on those clients.
You should never export /usr/obj read-write, it will lead to all sorts of
problems and issues down the line and presents a security problem as well.
It is far easier to do builds on the master server and then only do installs
on the clients.
.Pp
I usually maintain my ports tree via CVS.
It is sitting right there in the master CVS archive and I've even told you
to check it out (see above).
With some fancy softlinks you can make the ports tree available both on your
master server and on all of your other machines.
Note that the ports tree exists only on the HEAD cvs branch, so its always
-CURRENT even on a -STABLE box.  This is what you do:
.Bd -literal -offset 4n
(THESE COMMANDS ON THE MASTER SERVER AND ON ALL CLIENTS)
cd /usr
rm -rf ports
ln -s /FreeBSD/FreeBSD-current/ports ports

cd /usr/ports   			(this pushes into the softlink)
rm -rf distfiles			(ON MASTER SERVER ONLY)
ln -s /usr/ports.distfiles distfiles	(ON MASTER SERVER ONLY)

mkdir /usr/ports.distfiles
mkdir /usr/ports.workdir
.Ed
.Pp
Since /usr/ports is softlinked into what will be read-only on all of your
clients, you have to tell the ports system to use a different working
directory to hold ports builds.
You want to add a line to your /etc/make.conf file on the master server
and on all your clients:
.Bd -literal -offset 4n
WRKDIRPREFIX=/usr/ports.workdir
.Ed
.Pp
You should try to make the directory you use for the ports working directory
as well as the directory used to hold distfiles consistent across all of your
machines.
If there isn't enough room in /usr/ports.distfiles and /usr/ports.workdir I
usually make those softlinks (since this is on /usr these are per-machine) to
where the distfiles and working space really are.
.Sh EXPORTING VIA NFS FROM THE MASTER SERVER
The master server needs to export /FreeBSD and /usr/obj via NFS so all the
rest of your machines can get at them.
I strongly recommend using a read-only export for both security and safety.
The environment I am describing in this manual page is designed primarily
around read-only NFS exports.
Your exports file on the master server should contain the following lines:
.Bd -literal -offset 4n
/FreeBSD -ro -alldirs -maproot=root: -network YOURLAN -mask YOURLANMASK
/usr/obj -ro -alldirs -maproot=root: -network YOURLAN -mask YOURLANMASK
.Ed
.Pp
Of course, NFS server operations must also be configured on that machine.
This is typically done via your /etc/rc.conf:
.Bd -literal -offset 4n
nfs_server_enable="YES"
nfs_server_flags="-u -t -n 4"
.Ed
.Sh THE CLIENT ENVIRONMENT
All of your client machines can import the development/build environment
directory simply by NFS mounting /FreeBSD and /usr/obj from the master
server.
A typical /etc/fstab entry on your client machines will be something like this:
.Bd -literal -offset 4n
masterserver:/FreeBSD     /FreeBSD        nfs     ro,bg    0       0
masterserver:/usr/obj     /usr/obj        nfs     ro,bg    0       0
.Ed
.Pp
And, of course, you should configure the client for NFS client operations
via /etc/rc.conf.
In particular, this will turn on nfsiod which will improve client-side NFS
performance:
.Bd -literal -offset 4n
nfs_client_enable="YES"
.Ed
.Pp
Each client should create softlinks for /usr/ports and /usr/src that point
into the NFS-mounted environment.
If a particular client is running -CURRENT, /usr/src
should be a softlink to /FreeBSD/FreeBSD-current/src.
If it is running -STABLE, /usr/src should be a softlink to
/FreeBSD/FreeBSD-4.x/src.  I do not usually create a /usr/src2 softlink on
clients, that is used as a convenient shortcut when working on the source
code on the master server only and could create massive confusion (of the
human variety) on a client.
.Bd -literal -offset 4n
(ON EACH CLIENT)
cd /usr
rm -rf ports src
ln -s /FreeBSD/FreeBSD-current/ports ports
ln -s /FreeBSD/FreeBSD-XXX/src src
.Ed
.Pp
Don't forget to create the working directories so you can build ports, as
previously described.
If these are not good locations, make them softlinks to the correct location.
Remember that /usr/ports/distfiles is exported by
the master server and is therefore going to point to the same place
(typically /usr/ports.distfiles) on every machine.
.Bd -literal -offset 4n
mkdir /usr/ports.distfiles
mkdir /usr/ports.workdir
.Ed
.Sh BUILDING KERNELS
Here is how you build a -STABLE kernel (on your main development box).
If you want to create a custom kernel, cp GENERIC to YOURKERNEL and then
edit it before configuring and building.
The kernel configuration file lives in /usr/src/sys/i386/conf/KERNELNAME.
.Bd -literal -offset 4n
cd /usr/src
make buildkernel KERNCONF=KERNELNAME
.Ed
.Pp
.Sy WARNING!
If you are familiar with the old config/cd/make method of building
a -STABLE kernel, note that the config method will put the build
environment in /usr/src/sys/compile/KERNELNAME instead of in /usr/obj.
.Pp
Building a -CURRENT kernel
.Bd -literal -offset 4n
cd /usr/src2		(on the master server)
make buildkernel KERNCONF=KERNELNAME
.Ed
.Sh INSTALLING KERNELS
Installing a -STABLE kernel (typically done on a client.
Only do this on your main development server if you want to install a new
kernel for your main development server):
.Bd -literal -offset 4n
cd /usr/src
make installkernel KERNCONF=KERNELNAME
.Ed
.Pp
If you are using the older config/cd/make build mechanism for stable, you
would install using:
.Bd -literal -offset 4n
cd /usr/src/sys/compile/KERNELNAME
make install
.Ed
.Pp
Installing a -CURRENT kernel (typically done only on a client)
.Bd -literal -offset 4n
(remember /usr/src is pointing to the client's specific environment)
cd /usr/src
make installkernel KERNCONF=KERNELNAME
.Ed
.Pp
.Sh BUILDING THE WORLD
This environment is designed such that you do all builds on the master server,
and then install from each client.
You can do builds on a client only if /usr/obj is local to that client.
Building the world is easy:
.Bd -literal -offset 4n
cd /usr/src
make buildworld
.Ed
.Pp
If you are on the master server you are running in a -STABLE environment, but
that does not prevent you from building the -CURRENT world.
Just cd into the appropriate source directory and you are set.  Do not
accidentally install it on your master server though!
.Bd -literal -offset 4n
cd /usr/src2
make buildworld
.Ed
.Sh INSTALLING THE WORLD
You can build on your main development server and install on clients.
The main development server must export /FreeBSD and /usr/obj via
read-only NFS to the clients.
.Pp
.Em NOTE!!!
If /usr/obj is a softlink on the master server, it must also be the EXACT
SAME softlink on each client.
If /usr/obj is a directory in /usr or a mount point on the master server,
then it must be (interchangeably) a directory in /usr or a mount point on
each client.
This is because the
absolute paths are expected to be the same when building the world as when
installing it, and you generally build it on your main development box
and install it from a client.
If you do not setup /usr/obj properly you will not be able to build on
machine and install on another.
.Bd -literal -offset 4n
(ON THE CLIENT)
(remember /usr/src is pointing to the client's specific environment)
cd /usr/src
make installworld
.Ed
.Pp
.Sy WARNING!
If builds work on the master server but installs do not work from the
clients, for example you try to install and the client complains that
the install tried to write into the read-only /usr/obj, then it is likely
that the /etc/make.conf file on the client does not match the one on the
master server closely enough and the install is trying to install something
that was not built.
.Sh DOING DEVELOPMENT ON A CLIENT (NOT JUST INSTALLING)
Developers often want to run buildkernel's or buildworld's on client
boxes simply to life-test the box.
You do this in the same manner that you buildkernel and buildworld on your
master server.
All you have to do is make sure that /usr/obj is pointing to local storage.
If you followed my advise and made /usr/obj its own partition on the master
server,
then it is typically going to be an NFS mount on the client.
Simply unmounting /usr/obj will leave you with a /usr/obj that is a
subdirectory in /usr which is typically local to the client.
You can then do builds to your heart's content!
.Sh MAINTAINING A LOCAL BRANCH
I have described how to maintain two versions of the source tree, a stable
version in /FreeBSD/FreeBSD-4.x and a current version
in /FreeBSD/FreeBSD-current.
There is absolutely nothing preventing you
from breaking out other versions of the source tree
into /FreeBSD/XXX.
In fact, my /FreeBSD partition also contains
.Ox ,
.Nx ,
and various flavors of Linux.
You may not necessarily be able to build non-FreeBSD operating systems on
your master server, but being able
to collect and manage source distributions from a central server is a very
useful thing to be able to do and you can certainly export to machines
which can build those other operating systems.
.Pp
Many developers choose to maintain a local branch of
.Fx
to test patches or build a custom distribution.
This can be done with CVS or another source code management system
(SubVersion, Perforce, BitKeeper) with its own repository.
Since the main 
.Fx
tree is based on CVS, the former is convenient.
.Pp
First, you need to modify your cvsup environment to avoid it modifying
the local changes you have committed to the repository.
It is important to remove the "delete" keyword from your supfile and to
add the CVSROOT subdirectory to your refuse file.
For more information, see
.Xr cvsup 1 .
.Pp
The
.Fx
version of CVS examines a custom environmental variable,
CVS_LOCAL_BRANCH_NUM, which specifies an integer to use when doing a cvs
tag/rtag.
Set this number to something high (say 1000) to avoid colliding
with potential future branches of the main repository.  For example,
branching a file with version 1.4 produces 1.4.1000.
Future commits to this branch will produce revisions 1.4.1000.1,
1.4.1000.2, etc.
.Pp
To fork your local branch, do:
.Bd -literal -offset 4n
cvs rtag -r RELENG_4 -b LOCAL_RELENG_4 src
.Ed
.Pp
After this, you can check out a copy from your local repository using the
new tag and begin making changes and committing them.
For more information on using cvs, see
.Xr cvs 1 .
.Pp
.Sy WARNING!
The cvsup utility may blow away changes made on a local branch in
some situations.
This has been reported to occur when the master CVS repository is
directly manipulated or an RCS file is changed.
At this point, cvsup notices that the client and server have entirely
different RCS files, so it does a full replace instead of trying to
send just deltas.
Ideally this situation should never arise, but in the real world it
happens all the time.
.Pp
While this is the only scenario where the problem should crop up,
there have been some suspicious-sounding reports of
CVS_LOCAL_BRANCH_NUM lossage that can't be explained by this alone.
Bottom line is, if you value your local branch then you
should back it up before every update.
.Sh UPDATING VIA CVS
The advantage of using cvsup to maintain an updated copy of the CVS
repository instead of using it to maintain source trees directly is that you
can then pick and choose when you bring your source tree (or pieces of your
source tree) up to date.
By using a cron job to maintain an updated CVS repository, you can update
your source tree at any time without any network cost as follows:
.Bd -literal -offset 4n
(on the main development server)
cd /usr/src
cvs -d /home/ncvs update
cd /usr/src2
cvs -d /home/ncvs update
cd /usr/ports
cvs -d /home/ncvs update
.Ed
.Pp
It is that simple, and since you are exporting the whole lot to your
clients, your clients have immediately visibility into the updated
source.
This is a good time to also remind you that most of the cvs operations
you do will be done as root, and that certain options are
required for CVS to operate properly on the
.Fx
repository.  For example,
.Fl Pd
is necessary when running "cvs update".
These options are typically placed in your ~/.cvsrc (as already described)
so you do not have to respecify them every time you run a CVS command.
Maintaining the CVS repository also gives you far more flexibility
in regards to breaking out multiple versions of the source tree.
It is a good idea to give your /FreeBSD partition a lot of space (I recommend
8-12GB) precisely for that reason.
If you can make it 15GB I would do it.
.Pp
I generally do not cvs update via a cron job.
This is because I generally want the source to not change out from under me
when I am developing code.
Instead I manually update the source every so often... when I feel it is
a good time.
My recommendation is to only keep the cvs repository synchronized via cron.
.Sh SEE ALSO
.Xr crontab 1 ,
.Xr crontab 5 ,
.Xr build 7 ,
.Xr firewall 7 ,
.Xr release 7 ,
.Xr tuning 7 ,
.Xr diskless 8
.Sh HISTORY
The
.Nm
manual page was originally written by
.An Matthew Dillon Aq dillon@FreeBSD.org
and first appeared
in
.Fx 5.0 ,
December 2002.