[FAQ Index] [To Section 9 - Migrating to OpenBSD] [To Section 11 - The X Window System]
On OpenBSD, users who are in the wheel group are allowed to use su(1) to become root. Otherwise, the user will receive an error.
If you are creating new users with adduser(8), you can add them to the wheel group by answering "wheel" at the "Invite user into other groups:" prompt. Existing users must be added to the wheel group by hand. Here is an example of a /etc/group entry which has had the user ericj added to the wheel group:
If you want to give access to superuser privileges without adding users to the wheel group, use sudo(8).
To duplicate your filesystem use dump(8) and restore(8). For example, to duplicate everything under directory SRC to directory DST, do a:
# cd /SRC; dump 0f - . | (cd /DST; restore -rf - )
dump is designed to give you plenty of backup capabilities, and it may be an overkill if you just want to duplicate a part of a (or an entire) filesystem. The command tar(1) may be faster for this operation. The format looks very similar:
# cd /SRC; tar cf - . | (cd /DST; tar xpf - )
The main files a system administrator should concentrate on are /etc/rc.conf (for guidance), /etc/rc.conf.local (for changes), /etc/rc.local and /etc/rc.shutdown. To get a look of how the rc(8) procedure works, here is the flow:
After the kernel is booted, /etc/rc is started:
Most daemons and services that come with OpenBSD are controlled on boot by variables defined in the /etc/rc.conf configuration file. Take a look at the /etc/rc.conf file. You'll see lines similar to this:
ftpd_flags=NO # for non-inetd use: ftpd_flags=""
A line like this shows that ftpd(8) is not to start up with the system (at least not as a daemon via rc(8); ftpd is often run out of inetd(8), see Anonymous FTP FAQ to read more about this). Each line has a comment showing you the flags for common usage of that daemon or service. This doesn't mean that you must run that daemon or service with those flags. Read the relevant manual page to see how you can have that daemon or service start up in any way you like.
We strongly suggest you do not alter /etc/rc.conf itself. Instead, create or edit the file /etc/rc.conf.local, copy just the lines you need to change from /etc/rc.conf and adjust them as you like. This makes future upgrades easier -- all the changes are in the one file that isn't touched during upgrade.
For example, here is the default line pertaining to httpd(8).
httpd_flags=NO # for normal use: "" (or "-DSSL" after reading ssl(8))
Here you can see that starting up httpd normally requires no flags, so a line like httpd_flags="" added to /etc/rc.conf.local will do the job. But to start httpd with SSL enabled (refer to the SSL FAQ or ssl(8)), you would start with a line like httpd_flags="-DSSL", though you may need to add other parameters for other reasons.
For other daemons which you might install on the system via packages or other ways, you could use the /etc/rc.local file. For example, I've installed a daemon which lies at /usr/local/sbin/daemonx. I want it to start at boot time. I could put an entry into /etc/rc.local like this:
if [ -x /usr/local/sbin/daemonx ]; then echo 'Starting daemonx'; /usr/local/sbin/daemonx fi
(If the daemon does not automatically detach on startup, remember to add a "&" at the end of the command line.)
From now on, this daemon will be started at boot. You will be able to see any errors on boot, a normal boot with no errors would show a line like this:
These scripts, one per daemon, are invoked by rc. The order for system daemons is coded into rc, and the order for add-in packages is managed by the pkg_scripts environment variable, which would be set in /etc/rc.conf.local. Note that simply placing a script in this directory does not cause it to be run on boot; the name of the script must be specified in the pkg_scripts variable to start on boot.
The starting of system scripts is determined by entries in the /etc/rc.conf.local file. For example, /etc/rc.d/httpd does not start httpd(8) unless /etc/rc.conf or /etc/rc.conf.local contains a line defining the "httpd_flags" variable. To help make sure your system will come up as expected on the next boot, the rc.d scripts will not run their daemon if the appropriate variable is not defined. You can, of course, manually invoke /usr/sbin/httpd directly with whatever options you wish, if you wish to run the program manually.
Note that rather than having each script in rc.d managing the entire startup, shutdown, reload, restart, and check operations, most rc.d scripts can be reduced to specifying a very few variables, and invoking the rc.subr script, which manages most of the standard way of doing these tasks.
For example, our above daemonx application could be started with a /etc/rc.d file consisting of:
and adding the daemon name to the pkg_scripts variable in /etc/rc.conf.local.#!/bin/sh daemon="/usr/local/sbin/daemonx" . /etc/rc.d/rc.subr rc_cmd $1
/etc/rc.shutdown is a script that is run at shutdown. Anything you want done before the system shuts down should be added to this file. If you have apm, you can also set "powerdown=YES", which will give you the equivalent of "shutdown -p".
# grep relay-domains /etc/mail/sendmail.cf
The output may look something like this:
If this file doesn't exist, create it. You will need to enter the hosts who are sending mail remotely with the following syntax:
.domain.com #Allow relaying for/to any host in domain.com sub.domain.com #Allow relaying for/to sub.domain.com and any host in that domain 10.2 #Allow relaying from all hosts in the IP net 10.2.*.*
Don't forget send a 'HangUP' signal to sendmail (a signal which causes most daemons to re-read their configuration file):
# kill -HUP `head -1 /var/run/sendmail.pid`
By default, Sendmail uses DNS for name resolution, not the /etc/hosts file. The behavior can be changed through the use of the /etc/mail/service.switch file.
If you wish to query the hosts file before DNS servers, create a /etc/mail/service.switch file which contains the following line:
hosts files dns
If you wish to query ONLY the hosts file, use the following:
Send Sendmail a HUP signal:
# kill -HUP `head -1 /var/run/sendmail.pid`
and the changes will take effect.
OpenBSD ships with an SSL-ready httpd and RSA libraries. For use with httpd(8), you must first have a certificate created. This will be kept in /etc/ssl/ with the corresponding key in /etc/ssl/private/. The steps shown here are taken in part from the ssl(8) man page. Refer to it for further information. This FAQ entry only outlines how to create an RSA certificate for web servers, not a DSA server certificate. To find out how to do so, please refer to the ssl(8) man page.
To start off, you need to create your server key and certificate using OpenSSL:
# openssl genrsa -out /etc/ssl/private/server.key 2048
Or, if you wish the key to be encrypted with a passphrase that you will have to type in when starting servers
# openssl genrsa -aes256 -out /etc/ssl/private/server.key 2048
The next step is to generate a Certificate Signing Request which is used to get a Certifying Authority (CA) to sign your certificate. To do this use the command:
# openssl req -new -key /etc/ssl/private/server.key -out /etc/ssl/private/server.csr
This server.csr file can then be given to Certifying Authority who will sign the key. One such CA is Thawte Certification which you can reach at http://www.thawte.com/.
If you cannot afford this, or just want to sign the certificate yourself, you can use the following.
# openssl x509 -sha256 -req -days 365 -in /etc/ssl/private/server.csr \ -signkey /etc/ssl/private/server.key -out /etc/ssl/server.crt
With /etc/ssl/server.crt and /etc/ssl/private/server.key in place, you should be able to start httpd(8) with the -DSSL flag (see the section about rc(8) in this faq), enabling https transactions with your machine on port 443.
If you edit /etc/passwd directly, your changes will be lost. OpenBSD generates /etc/passwd dynamically with pwd_mkdb(8). The main password file in OpenBSD is /etc/master.passwd. According to pwd_mkdb(8),
FILES /etc/master.passwd current password file /etc/passwd a 6th Edition-style password file /etc/pwd.db insecure password database file /etc/pwd.db.tmp temporary file /etc/spwd.db secure password database file /etc/spwd.db.tmp temporary file
In a traditional Unix password file, such as /etc/passwd, everything including the user's encrypted password is available to anyone on the system (and is a prime target for programs such as Crack). 4.4BSD introduced the master.passwd file, which has an extended format (with additional options beyond those provided by /etc/passwd) and is only readable by root. For faster access to data, the library calls which access this data normally read /etc/pwd.db and /etc/spwd.db.
OpenBSD does come with a tool with which you should edit your password file. It is called vipw(8). Vipw will use vi (or your favourite editor defined per $EDITOR) to edit /etc/master.passwd. After you are done editing, it will re-create /etc/passwd, /etc/pwd.db, and /etc/spwd.db as per your changes. Vipw also takes care of locking these files, so that if anyone else attempts to change them at the same time, they will be denied access.
OpenBSD provides two commands for easily adding users to the system:
The easiest way to add a user in OpenBSD is to use the adduser(8) script. You can configure adduser(8) by editing /etc/adduser.conf. adduser(8) allows for consistency checks on /etc/passwd, /etc/group, and shell databases. It will create the entries and $HOME directories for you. It can even send a message to the user welcoming them. Here is an example user, testuser, being added to a system. He/she will be given the $HOME directory /home/testuser, made a member of the group guest, and given the shell /bin/ksh.
# adduser Use option ``-silent'' if you don't want to see all warnings and questions. Reading /etc/shells Check /etc/master.passwd Check /etc/group Ok, let's go. Don't worry about mistakes. There will be a chance later to correct any input. Enter username : testuser Enter full name : Test FAQ User Enter shell csh ksh nologin sh [ksh]: ksh Uid : Enter Login group testuser [testuser]: guest Login group is ``guest''. Invite testuser into other groups: guest no [no]: no Login class authpf daemon default staff [default]: Enter Enter password : Type password, then Enter Enter password again : Type password, then Enter Name: testuser Password: **** Fullname: Test FAQ User Uid: 1002 Gid: 31 (guest) Groups: guest Login Class: default HOME: /home/testuser Shell: /bin/ksh OK? (y/n) [y]: y Added user ``testuser'' Copy files from /etc/skel to /home/testuser Add another user? (y/n) [y]: n Goodbye!
To delete users you should use the rmuser(8) utility. This will remove all existence of a user. It will remove any crontab(1) entries, their $HOME dir (if it is owned by the user), and their mail. Of course it will also remove their /etc/passwd and /etc/group entries. Next is an example of removing the user that was added above. Notice you are prompted for the name, and whether or not to remove the user's home directory.
# rmuser Enter login name for user to remove: testuser Matching password entry: testuser:$2a$07$ZWnBOsbqMJ.ducQBfsTKUe3PL97Ve1AHWJ0A4uLamniLNXLeYrEie:1002 :31::0:0:Test FAQ User:/home/testuser:/bin/ksh Is this the entry you wish to remove? y Remove user's home directory (/home/testuser)? y Updating password file, updating databases, done. Updating group file: done. Removing user's home directory (/home/testuser): done.
These tools are less interactive than the adduser(8) command, which makes them easier to use in scripts.
The full set of tools is:
The command /usr/sbin/user is just a frontend to the rest of the /usr/sbin/user* commands. Therefore, the following commands can be added by using user add or useradd, which form you chose doesn't change the results at all. Remember, since user(8) is not interactive, the easiest way to add users is with adduser(8).
useradd(8) is less daunting to use if you know the default settings beforehand. These settings are located in usermgmt.conf(5) and can be viewed by doing:
$ user add -D group users base_dir /home skel_dir /etc/skel shell /bin/csh inactive 0 expire Null (unset) range 1000..60000
These defaults will be used unless you specify alternatives with the command line options. For example, we want the user to be added to the group guest, not users. One more little hurdle with adding users, is that passwords must be specified on the command line. Importantly, the passwords must be encrypted, so you need to use the encrypt(1) utility. For example: OpenBSD's passwords by default use the Blowfish algorithm for 6 rounds. Here is an example to create an encrypted password to give to useradd(8).
$ encrypt -p -b 6 Enter string: $2a$06$YOdOZM3.4m6MObBXjeZtBOWArqC2.uRJZXUkOghbieIvSWXVJRzlq
Now that we have an encrypted password, we are ready to add the user. We will add the same user with the same specifications as the user we added above, via adduser(8).
# user add -p '$2a$06$YOdOZM3.4m6MObBXjeZtBOWArqC2.uRJZXUkOghbieIvSWXVJRzlq' -u 1002 \ -s /bin/ksh -c "Test FAQ User" -m -g guest testuser
Note: Make sure to use ' ' (single quotes) around the password string, not " " (double quotes) as the shell will interpret these before sending it to user(8). In addition to that, make sure you specify the -m option if you want the user's home directory created and the files from /etc/skel copied over.
To see that the user was created correctly, we can use many different utilities. Below are a few commands you can use to quickly check that everything was created correctly.
$ ls -la /home total 14 drwxr-xr-x 5 root wheel 512 May 12 14:29 . drwxr-xr-x 15 root wheel 512 Apr 25 20:52 .. drwxr-xr-x 24 ericj wheel 2560 May 12 13:38 ericj drwxr-xr-x 2 testuser guest 512 May 12 14:28 testuser $ id testuser uid=1002(testuser) gid=31(guest) groups=31(guest) $ finger testuser Login: testuser Name: Test FAQ User Directory: /home/testuser Shell: /bin/ksh Last login Sat Apr 22 16:05 (EDT) on ttyC2 No Mail. No Plan.
In addition to these commands, user(8) provides its own utility to show user characteristics, called userinfo(8).
$ userinfo testuser login testuser passwd * uid 1002 groups guest change Wed Dec 31 19:00:00 1969 class gecos Test FAQ User dir /home/testuser shell /bin/ksh expire Wed Dec 31 19:00:00 1969
To remove users with the user(8) hierarchy of commands, you will use userdel(8). This is a very simple, yet usable command. To remove the user created in the last example, simply:
# userdel -r testuser
Notice the -r option, which must be specified if you want the users home directory to be deleted as well. Alternatively, you can specify -p and not -r and this will lock the user's account, but not remove any information.
There are a few ways to do this, but a very common way to do such is to add "/usr/bin/false" into "/etc/shells". Then when you set a users shell to "/usr/bin/false", they will not be able log in interactively, but will be able to use ftp capabilities. You may also want to restrict access by Confining users to their home directory in ftpd.
Quotas are used to limit user's space that they have available to them on your disk drives. It can be very helpful in situations where you have limited resources. Quotas can be set by user and/or by group.
The first step to setting up quotas is to make sure that "option QUOTA" is in your Kernel Configuration. This option is in the GENERIC kernel. After this, you need to mark in /etc/fstab the filesystems which will have quotas enabled. The keywords userquota and groupquota should be used to mark each filesystem that you will be using quotas on. By default, the files quota.user and quota.group will be created at the root of that filesystem to hold the quota information. This default can be overridden by specifying the file name with the quota option in /etc/fstab, such as "userquota=/var/quotas/quota.user". Here is an example /etc/fstab that has one filesystem with userquotas enabled, and the quota file in a non-standard location:
/dev/wd0a / ffs rw,userquota=/var/quotas/quota.user 1 1
Now it's time to set the user's quotas. To do so you use the utility edquota(8). A simple use is just "edquota <user>". edquota(8) will use vi(1) to edit the quotas unless the environmental variable EDITOR is set to a different editor. For example:
# edquota ericj
This will give you output similar to this:
Quotas for user ericj: /: KBytes in use: 62, limits (soft = 0, hard = 0) inodes in use: 25, limits (soft = 0, hard = 0)
To add limits, edit it to give results like this:
Quotas for user ericj: /: KBytes in use: 62, limits (soft = 1000, hard = 1050) inodes in use: 25, limits (soft = 0, hard = 0)
Note that the quota allocation is in 1k blocks. In this case, the softlimit is set to 1000k, and the hardlimit is set to 1050k. A softlimit is a limit where the user is just warned when they cross it and have until their grace period is up to get their disk usage below their limit. Grace periods can be set by using the -t option on edquota(8). After the grace period is over the softlimit is handled as a hardlimit. This usually results in an allocation failure.
Now that the quotas are set, you need to turn the quotas on. To do this use quotaon(8). For example:
# quotaon -a
This will go through /etc/fstab to turn on the filesystems with quota options. Now that quotas are up and running, you can view them using quota(1). Using a command of "quota <user>" will give that user's information. When called with no arguments, the quota(1) command will give your quota statistics. For example:
# quota ericj
Will result in output similar to this:
Disk quotas for user ericj (uid 1001): Filesystem blocks quota limit grace files quota limit grace / 62 1000 1050 27 0 0
By default quotas set in /etc/fstab will be started on boot. To turn them off use
# quotaoff -a
Anonymous FTP allows users without accounts to access files on your computer via the File Transfer Protocol. This will give an overview of setting up the anonymous FTP server, and its logging, etc.
To start off, you need to have an ftp account on your system. This account should not have a usable password. Here we will set the login directory to /home/ftp, but you can put it wherever you want. When using anonymous ftp, the ftp daemon will chroot itself to the home directory of the ftp user. To read up more on that, read the ftpd(8) and chroot(2) man pages. Here is an example of adding the ftp user. I will do this using adduser(8). We also need to add /usr/bin/false to our /etc/shells, this is the "shell" that we will be giving to the ftp user. This won't allow them to login, even though we will give them an empty password. To do this you can simply do
After this, you are ready to add the ftp user:# echo /usr/bin/false >> /etc/shells
# adduser Use option ``-silent'' if you don't want to see all warnings and questions. Reading /etc/shells Check /etc/master.passwd Check /etc/group Ok, let's go. Don't worry about mistakes. There will be a chance later to correct any input. Enter username : ftp Enter full name : anonymous ftp Enter shell csh false ksh nologin sh [ksh]: false Uid : Enter Login group ftp [ftp]: Enter Login group is ``ftp''. Invite ftp into other groups: guest no [no]: Enter Login class authpf daemon default staff [default]: Enter Enter password : Enter Set the password so that user cannot logon? (y/n) [n]: y Name: ftp Password: **** Fullname: anonymous ftp Uid: 1002 Gid: 1002 (ftp) Groups: ftp Login Class: default HOME: /home/ftp Shell: /usr/bin/false OK? (y/n) [y]: Enter Added user ``ftp'' Copy files from /etc/skel to /home/ftp Add another user? (y/n) [y]: n Goodbye!
Along with the user, this created the directory /home/ftp (though you can use other locations as well, of course). There are some changes that we will have to make to get it ready for anonymous ftp. Again these changes are explained in the ftpd(8) man page.
You do not need to make a /home/ftp/usr or /home/ftp/bin directory.
Note that all these directories should be owned by ''root''. Here is a listing of what the directories should look like after their creation.
# pwd /home # ls -laR ftp total 5 dr-xr-xr-x 5 root ftp 512 Jul 6 11:33 . drwxr-xr-x 7 root wheel 512 Jul 6 10:58 .. dr-x--x--x 2 root ftp 512 Jul 6 11:34 etc dr-xr-xr-x 2 root ftp 512 Jul 6 11:33 pub ftp/etc: total 43 dr-x--x--x 2 root ftp 512 Jul 6 11:34 . dr-xr-xr-x 5 root ftp 512 Jul 6 11:33 .. -r--r--r-- 1 root ftp 316 Jul 6 11:34 group -r--r--r-- 1 root ftp 40960 Jul 6 11:34 pwd.db ftp/pub: total 2 dr-xr-xr-x 2 root ftp 512 Jul 6 11:33 . dr-xr-xr-x 5 root ftp 512 Jul 6 11:33 ..
You can choose to start ftpd either by inetd(8) or from the rc scripts. Generally, running ftpd as a daemon from the rc scripts is recommended and is what we will show here; a sample line for running it from inetd(8) is commented out in the default /etc/inetd.conf.
Here is an example line to add to /etc/rc.conf.local to start ftpd(8) at boot.
The options used here are:ftpd_flags="-llUSA"
It is not necessary to give startup options to ftpd to allow anonymous connections; the preceding steps of creating the 'ftp' user is sufficient. However, to stop anonymous connections it isn't necessary to undo everything, just restart ftpd with the -n option included. Anonymous connections are then disabled.
By default, when logging in by ftp, users can change to any directory on the filesystem that they have access to. This may not be desirable in some cases. It is possible to restrict what users may see through ftp sessions by chrooting them to their home directory.
If you only wish to allow chrooted ftp logins, use the -A option to ftpd(8).
If you wish to apply them more finely, OpenBSD's login capability infrastructure and ftpd(8) together make this easy.
Users in a login class with the ftp-chroot variable set are automatically chrooted. Additionally, you can add a username to the file /etc/ftpchroot to chroot those usernames. A user only needs to be listed in one of these locations.
Even with OpenBSD, bugs happen. Some bugs may lead to reliability issues (i.e., something may cause the system to stop functioning as desired). Other bugs may lead to security vulnerabilities (which may allow others to "use" your computer in unintended and unauthorized ways). When a critical bug is found, the fix will be committed to the -current source tree, and patches will be released for the supported releases of OpenBSD. These patches appear on the errata web page, and are separated into "common" errata that impact all platforms, and errata that impact only one or more, but not all, platforms.
Note, however, that patches aren't released for new features or additional hardware support for OpenBSD, and are only published for important reliability fixes or security problems that should be addressed right away on impacted systems (which is often NOT all systems, depending on their purpose).
There are three ways to update your system with patched code:
All patches posted to the errata web page are patches directly against the indicated release's source tree. Patches against the latest CVS tree might also include other changes that wouldn't be wanted on a release system. This is important: If you have installed a snapshot, checked out the source trees at the time you obtained that snapshot and attempt to patch it using a published patch, you may well find the patch doesn't apply, as that code may have changed.
Patches for the OpenBSD Operating System are distributed as "Unified diffs", which are text files that hold differences to the original source code. They are NOT distributed in binary form. This means that to apply a patch you must have the source code from the RELEASE version of your system available. In general, it is advisable to acquire the entire source tree before applying a patch. If you are running a release from an official CDROM, the source trees are available as files on disk 3, these can also be downloaded from the FTP servers. We will assume you have the entire tree checked out.
For our example here, we will look at patch 001 for OpenBSD 3.6 dealing with the st(4) driver, which handles tape drives. Without this patch, recovering data from backups was quite difficult. People using a tape drive needed this patch, however those without a tape drive may have had no particular need to install it. Let's look at the patch:
As you will note, the top of the patch includes brief instructions on applying it. We will assume you have put this patch into the /usr/src directory, in which case, the following steps are used:# more 001_st.patch Apply by doing: cd /usr/src patch -p0 < 001_st.patch Rebuild your kernel. Index: sys/scsi/st.c =================================================================== RCS file: /cvs/src/sys/scsi/st.c,v retrieving revision 1.41 retrieving revision 18.104.22.168 diff -u -p -r1.41 -r22.214.171.124 --- sys/scsi/st.c 1 Aug 2004 23:01:06 -0000 1.41 +++ sys/scsi/st.c 2 Nov 2004 01:05:50 -0000 126.96.36.199 @@ -1815,7 +1815,7 @@ st_interpret_sense(xs) u_int8_t skey = sense->flags & SSD_KEY; int32_t info; - if (((sense->flags & SDEV_OPEN) == 0) || + if (((sc_link->flags & SDEV_OPEN) == 0) || (serr != 0x70 && serr != 0x71)) return (EJUSTRETURN); /* let the generic code handle it */
Note the "Hunk #1 succeeded" message above. This indicates the patch was applied successfully. Many patches are more complex than this one, and will involve multiple hunks and multiple files, in which case, you should verify that all hunks succeeded on all files. If they did not, it normally means your source tree was different in some way from the release source tree the patch was created from, or you didn't follow instructions carefully, or your patch was mangled. Patches are very sensitive to "white space" -- copying and pasting from your browser will often change tab characters into spaces or otherwise alter the white space of a file, making it not apply.# cd /usr/src # patch -p0 < 001_st.patch Hmm... Looks like a unified diff to me... The text leading up to this was: -------------------------- |Apply by doing: | cd /usr/src | patch -p0 < 001_st.patch | |Rebuild your kernel. | |Index: sys/scsi/st.c |=================================================================== |RCS file: /cvs/src/sys/scsi/st.c,v |retrieving revision 1.41 |retrieving revision 188.8.131.52 |diff -u -p -r1.41 -r184.108.40.206 |--- sys/scsi/st.c 1 Aug 2004 23:01:06 -0000 1.41 |+++ sys/scsi/st.c 2 Nov 2004 01:05:50 -0000 220.127.116.11 -------------------------- Patching file sys/scsi/st.c using Plan A... Hunk #1 succeeded at 1815. <-- Look for this message! done
At this point, you can build and install the new kernel, then reboot the system as normal.
Not all patches are for the kernel. In some cases, you will have to rebuild individual utilities. At other times, will require recompiling all utilities statically linked to a patched library. Follow the guidance in the header of the patch, and if uncertain, rebuild the entire system.
Patches that are irrelevant to your particular system need not be applied -- usually. For example, if you did not have a tape drive on your system, you would not benefit from the above patch. However, patches are assumed to be applied "in order" -- it is possible that a later patch is dependent upon an earlier one. Be aware of this if you elect to "pick and choose" which patches you apply, and if in doubt, apply them all, in order.
In OpenBSD, the Apache httpd(8) and nginx(8) web servers have been chroot(2)ed by default. While this is a tremendous boost to security, it can create issues, if you are not prepared.
Put bluntly, chroot(2)ing web servers is something not done by default in most other operating systems. Many applications and system configurations will not work in a chroot(2) without some customization. Further, it must be remembered that security and convenience are often not compatible goals.
First, we install the wwwcount package. We configure it and test it, and we find it doesn't seem to work, we get an Apache message saying "Internal Server Error". First step is to stop and restart Apache with the -u switch to verify that the problem is the chroot(2)ing, and not the system configuration.
After doing this, we see the counter works properly, at least after we change the ownership on a directory so that Apache (and the CGIs it runs) can write to the files it keeps. So, we definitely have a chroot problem, so we stop and restart Apache again, using the default chrooting:# apachectl stop /usr/sbin/apachectl stop: httpd stopped # httpd -u
# apachectl stop /usr/sbin/apachectl stop: httpd stopped # httpd
A good starting point would be to assume wwwcount uses some libraries and other files it can't get to in the chroot. We can use the ldd(1) command to find out the dynamic object dependencies that the CGI needs:
Ok, here is a problem, two files that are not available in the chroot(2) environment. So, we copy them over:# cd /var/www/cgi-bin/ # ldd Count.cgi Count.cgi: Start End Type Open Ref GrpRef Name 1c000000 3c007000 exe 1 0 0 /var/www/cgi-bin/Count.cgi 0c085000 2c0be000 rlib 0 1 0 /usr/lib/libc.so.57.0 08713000 08713000 rtld 0 1 0 /usr/libexec/ld.so
and try the counter again.# mkdir -p /var/www/usr/lib /var/www/usr/libexec # cp /usr/lib/libc.so.57.0 /var/www/usr/lib # cp /usr/libexec/ld.so /var/www/usr/libexec
Well, now the program is running at least, and giving us error messages directly: "Unable to open config file for reading". Progress, but not done yet. The configuration file is normally in /var/www/wwwcount/conf, but within the chroot environment, that would seem to be /wwwcount/conf. Our options are to either recompile the program to make it work where the files are now, or move the data files. As we installed from a package, we'll just move the data file. In order to use the same config either chroot(2)ed or not, we'll use a symbolic link:
Note that the symbolic link is crafted to work within the chroot. Again, we test... and we find we have yet another issue. Now wwwcount is complaining that it can't find the "strip image" files it uses to display messages. After a bit of searching, we find those are stored in /usr/local/lib/wwwcount, so we have to copy those into the chroot, as well.# mkdir -p /var/www/var/www # cd /var/www/var/www # ln -s ../../wwwcount wwwcount
we test again... and it works!# tar cf - /usr/local/lib/wwwcount | (cd /var/www; tar xpf - )
Note that we have copied over only files that are absolutely required for operation. In general, only the minimum files needed to run an application should be copied into the chroot.
Not every application can or should be chroot(2)ed.
The goal is a secure web server, chroot(2)ing is just a tool to accomplish this, it is not the goal itself. Remember, the starting configuration of the OpenBSD chroot(2)ed Apache is where the user the httpd(8) program is running as can not run any programs, can not alter any files, and can not assume another user's identity. Loosen these restrictions, you have lessened your security, chroot or no chroot.
Some applications are pretty simple, and chroot(2)ing them makes sense. Others are very complex, and are either not worth the effort of forcing them into a chroot(2), or by the time you copy enough of the system into the chroot, you have lost the benefit of the chroot(2) environment. For example, the OpenWebMail program requires the ability to read and write to the mail directory, the user's home directory, and must be able to work as any user on the system. Attempting to push it into a chroot would be completely pointless, as you would end up disabling all the benefits of chroot(2)ing. Even with an application as simple as the above counter, it must write to disk (to keep track of its counters), so some benefit of the chroot(2) is lost.
Any application which has to assume root privileges to operate is pointless to attempt to chroot(2), as root can generally escape a chroot(2).
Do not forget, if the chrooting process for your application is too difficult, you may not upgrade or update the system as often as you should. This could end up making your system LESS secure than a more maintainable system with the chroot feature deactivated.
The default shell for root on OpenBSD is ksh.
A traditional Unix guideline is to only use statically compiled shells for root, because if your system comes up in single user mode, non-root partitions won't be mounted and dynamically linked shells won't be able to access libraries located in the /usr partition. This isn't actually a significant issue for OpenBSD, as the system will prompt you for a shell when it comes up in single user mode, and the default is sh. The three standard shells in OpenBSD (csh, sh and ksh) are all statically linked, and thus usable in single user mode.
Users comfortable with bash, often used on Linux systems, will probably find ksh very familiar. Ksh(1) provides most of the commonly used features in bash, including tab completion, command line editing and history via the arrow keys, and CTRL-A/CTRL-E to jump to beginning/end of the command line. If other features of bash are desired, bash itself can be loaded via either packages or ports.
The command prompt of ksh can easily be changed to something providing more information than the default "$ " by setting the PS1 variable. For example, inserting the following line:
in your /etc/profile produces the following command prompt:export PS1='$PWD $ '
See the file /etc/ksh.kshrc, which includes many useful features and examples, and may be invoked in your user's .profile./home/nick $
OpenBSD's ksh(1) has been enhanced with a number of "special characters" for the primary prompt string, PS1, similar to those used in bash. For example:
\e - Insert an ASCII escape character.(see the ksh(1) man page for more details, and many, many more special characters! Also note the "$" character has special meaning inside double quotes, so handle it carefully)
\h - The hostname, minus domain name.
\H - The full hostname, including domain name.
\n - Insert a newline character.
\t - The current time, in 24-hour HH:MM:SS format.
\u - The current user's username.
\w - The current working directory. $HOME is abbreviated as `~'.
\W - The basename of the current working directory.
\$ - Displays "#" for root users, "$" for non-root users.
One could use the following command:
to give an overly verbose but somewhat useful prompt.export PS1="\n\u@\H\n\w \\$ "
OpenBSD can be used for both servers and clients of databases containing user credentials, group information and other network-related data.
Of course, you could use various directory services on OpenBSD. But YP is the only one that can be accessed directly using standard C-library functions like getpwent(3), getgrent(3), gethostbyname(3) and so on. Thus, if you keep your data in a YP database, you do not need to copy it to local configuration files like master.passwd(5) before you can use it, for example to authenticate system users.
YP is a directory service compatible with Sun Microsystems NIS (Network Information System). See yp(8) for an overview of the available manual pages. Be careful, some operating systems contain directory services bearing similar names but all the same being incompatible, for example NIS+.
To use other directory services except YP, you either need to populate local configuration files from the directory, or you need a YP frontend to the directory. For example, you can use the sysutils/login_ldap port when you choose the former, while the ypldap(8) daemon provides the latter.
For some applications, simply synchronizing a small number of configuration files among a group of machines using tools like cron(8), scp(1) or rsync (available from ports) constitutes an easy and robust alternative to a full-blown directory service.
For compatibility reasons, all security features built into the OpenBSD implementation of YP are switched off by default. Even when they are all switched on, the NIS protocol is still inherently insecure for two reasons: All data, including sensitive data like password hashes, is transmitted unencrypted across the network, and neither the client nor the server can reliably verify each other's identity.
Thus, before setting up any YP server, you should consider whether these inherent security flaws are acceptable in your context. In particular, YP is inadequate if potential attackers have physical access to your network. Anybody gaining root access to any computer connected to your network segments carrying YP traffic can bind your YP domain and retrieve its data. In some cases, passing YP traffic through SSL or IPSec tunnels might be an option, or you might consider combining YP with kerberos(8) authentication.
A YP server serves a group of clients called a "domain". You should first select a domain name; it can be an arbitrary string and need not be related in any way to DNS domain names. Choosing a random name like "Eepoo5vi" can marginally improve security, though the effect is mostly in security by obscurity. In case you need to maintain several distinct YP domains, it's probably better to choose descriptive names like "sales", "marketing" and "research" in order to forestall system administration errors caused by obscurity. Also note that some versions of SunOS require using the host's DNS domain name, so your choice might be restricted in a network including such hosts.
Use the domainname(1) utility to set the domain name, and put it into the file defaultdomain(5) to have it automatically set at system startup time.
echo "puffynet" > /etc/defaultdomain domainname `cat /etc/defaultdomain`
Initialise the YP server using the interactive command
At this point, it is not necessary to specify slave servers yet. To add slave servers, you can rerun ypinit(8) later, using the -u option. Setting up at least one slave server for each domain is useful to avoid service interruptions, should the master server ever go down or lose network connectivity, in particular since client processes trying to access YP maps block indefinitely until they receive the requested information. Thus, YP service interruptions typically render the client hosts completely unusable until YP is back to service.ypinit -m
Decide where to store the source files to generate your YP maps from. Keeping the server configuration separate from the served configuration helps to control which information will be served and which won't, so the default /etc often isn't the best choice.
The only inconvenience caused by changing the source directory is that you will not be able to add, remove and modify users and groups in the YP domain using utilities like user(8) and group(8). Instead, you will have to edit the configuration files with a text editor.
To define the source directory, edit the file /var/yp/`domainname`/Makefile and change the DIR variable, e.g.
Consider customizing other variables in /var/yp/`domainname`/Makefile. See Makefile.yp(8) for details.
For example, even in case you use the default source directory /etc, you do not usually need all accounts and groups existing on the server on all your client hosts. In particular, not serving the root account and thus keeping root's password hash confidential is often beneficial to security. Review the values of MINUID, MAXUID, MINGID and MAXGID and adjust them to you needs.
If all your YP clients run OpenBSD or FreeBSD, exclude the encrypted passwords from the passwd maps by setting UNSECURE="" in /var/yp/`domainname`/Makefile.
The former practice of editing the template file /var/yp/Makefile.yp is no longer recommended. Changes to that file affect all domains initialized after the change, but do not affect domains initialized before the change, so this is error-prone either way: You both risk that the intended changes do not take effect, and you risk to forget about them and have them affect other domains later which they were never intended for.
Create the source directory and populate it with the configuration files you need. See Makefile.yp(8) to learn which YP maps require which source files. For the format of the individual configuration files, refer to passwd(5), group(5), hosts(5) and so on, and look at the examples in /etc.
Create the initial version of your YP maps using the commands
Do not worry about error messages from yppush(8) right now. The YP server is not yet running.cd /var/yp make
YP uses rpc(3) (remote procedure calls) to communicate with clients, so it is necessary to enable portmap(8). To do so, edit rc.conf.local(8) and set portmap_flags="". This will start the portmapper on next boot. You can avoid rebooting by also starting it manually:
echo 'portmap_flags=""' >> /etc/rc.conf.local portmap
Consider using either the securenet(5) or the ypserv.acl(5) security feature of the YP server daemon. But be aware that both of these only provide IP based access control. Thus, they only help as long as potential attackers have neither physical access to the hardware of the network segments carrying your YP traffic nor root access to any host connected to those network segments.
Finally, start the YP server daemon:
It will automatically be restarted at boot time as long as the directory /var/yp/`domainname` continues to exist.ypserv
To test the new server, consider making it its own client, following the instructions in the first part of the next section. In case you don't want the server to use its own maps, you can disable the client part after the test with the following commands:
pkill ypbind rm -rf /var/yp/binding
If you wish to allow users to change their passwords from client machines, then you must enable yppasswdd(8):
In case you left the source directory at the default /etc, just use yppasswdd_flags="".echo 'yppasswdd_flags="-d /etc/yp/src/puffynet"' >> /etc/rc.conf.local rpc.yppasswdd
Remember that each time you change a file sourced by a YP map, you must regenerate your YP maps.
This updates all database files in /var/yp/`domainname`, with one exception: The file ypservers.db, listing all YP master and slave servers associated with the domain, is created directly from ypinit -m and modified exclusively by ypinit -u. In case you accidentally delete it, run ypinit -u to recreate it from scratch.cd /var/yp make
Like on the server, you must set the domain name and enable the portmapper:
echo "puffynet" > /etc/defaultdomain domainname `cat /etc/defaultdomain` echo 'portmap_flags=""' >> /etc/rc.conf.local portmap
It is recommended to provide a list of YP servers in the configuration file /etc/yp/`domainname`. Otherwise, the YP client daemon will use network broadcasts to find YP servers for its domain. Explicitly specifying the servers is both more robust and marginally less open to attack. If you have not set up any slave servers, just put the host name of the master server into /etc/yp/`domainname`.
The YP client daemon is called ypbind(8). Starting it manually will create the directory /var/yp/binding, such that it will be automatically restarted at boot time.
If all went well you should be able to query the YP server using ypcat(1) and see your passwd map returned.
Other useful tools for debugging your YP setup include ypmatch(1) and yptest(8).ypcat passwd bob:*:5001:5000:Bob Nuggets:/home/bob:/usr/local/bin/zsh ...
For a list of standard YP maps and their standard usage, see Makefile.yp(8). The most common use cases include:
If you want to include all user accounts from the YP domain, append the default YP marker to the master password file and rebuild the password database:
For details on selective inclusion and exclusion of user accounts, see passwd(5). To test whether inclusion actually works, use the id(1) utility.echo '+:*::::::::' >> /etc/master.passwd pwd_mkdb -p /etc/master.passwd
If you want to include all groups from the YP domain, append the default YP marker to the group file:
For details on selective group inclusion, see group(5).echo '+:*::' >> /etc/group
If you are distributing hostnames via YP, you should now review resolv.conf(5) and check that the name service lookup order is correct. Most users will require a line like this:
lookup file yp bind
To use one of the extended character sets, the LC_CTYPE environment variable must be set to the name of a supported locale. LC_CTYPE will only affect the character set available to applications. It will not change the language used for application messages.
The list of supported locales can be obtained by running the command:
The LC_CTYPE environment variable can be set in one of the following ways:locale -a
to ~/.xsession before starting the window manager (see the section on customizing X for details). This example enables the Unicode (UTF-8) character set and will also cause applications such as xterm(1) to enable UTF-8 mode by default.export LC_CTYPE="en_US.UTF-8"
If logging in via the text console add a line such as
to ~/.profile. Note that the text console supports ASCII and ISO8859-1 only. It does not support UTF-8.export LC_CTYPE="en_US.ISO8859-1"
Few utilities in the base system support UTF-8 at this time. Most will use ASCII in the UTF-8 locale. However, many programs from the ports collection do support UTF-8.
UTF-8 can also be used with specific applications only by starting those applications in uxterm(1). This works even if the login session uses a non-UTF-8 locale.
When logging into remote systems with ssh(1), the LC_CTYPE environment variable is not propagated and will need to be manually set to the same value used by the local terminal.
The language used for application messages can be changed by setting the LC_MESSAGES environment variable to the name of a supported locale. This can be done in the same way as described for LC_CTYPE above. Both LC_MESSAGES and LC_CTYPE should be set to the same value.
Few utilities in the base system support languages other than English at this time. However, many programs from the ports collection support localized messages in various languages. They will fall back to English if the desired language is not available.
[FAQ Index] [To Section 9 - Migrating to OpenBSD] [To Section 11 - The X Window System]