-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Paul Howarth wrote:
Daniel J Walsh wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Paul Howarth wrote:
On Tue, 15 Jan 2008 14:53:18 -0500 Daniel J Walsh dwalsh@redhat.com wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Paul Howarth wrote:
Daniel J Walsh wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Paul Howarth wrote: > Daniel J Walsh wrote: >> -----BEGIN PGP SIGNED MESSAGE----- >> Hash: SHA1 >> >> Paul Howarth wrote: >>> Hi Dan, >>> >>> Daniel J Walsh wrote: >>>> -----BEGIN PGP SIGNED MESSAGE----- >>>> Hash: SHA1 >>>> >>>> Paul Howarth wrote: >>>>> Paul Howarth wrote: >>>>>> Since upgrading my mail server from Fedora 7 to Fedora 8, >>>>>> I've come across some problems with the sockets used for >>>>>> communication between >>>>>> sendmail and two of the "milter" plugins I'm using with it, >>>>>> namely milter-regex and spamass-milter. It's very likely >>>>>> that other milters >>>>>> will have similar issues. >>>>>> >>>>>> The sockets used are created when the milter starts, as >>>>>> follows: >>>>>> >>>>>> milter-regex: >>>>>> /var/spool/milter-regex/sock (var_spool_t, inherited from >>>>>> parent directory) >>>>>> >>>>>> spamass-milter: >>>>>> /var/run/spamass-milter/spamass-milter.sock >>>>>> (spamd_var_run_t, in policy) >>>>>> >>>>>> These are pretty well the upstream locations, though I'm >>>>>> open to moving the milter-regex socket from /var/spool >>>>>> to /var/run or elsewhere for consistency. >>>>>> >>>>>> Since moving to Fedora 8, I've had to add the following to >>>>>> local policy to get these milters working: >>>>>> >>>>>> allow sendmail_t spamd_var_run_t:dir { search getattr }; >>>>>> allow sendmail_t spamd_var_run_t:sock_file { getattr write }; >>>>>> allow sendmail_t var_spool_t:sock_file { getattr write }; >>>>>> allow sendmail_t initrc_t:unix_stream_socket { read write >>>>>> connectto }; >>>>>> >>>>>> The last of these is the strangest, and relates to Bug >>>>>> #425958 >>>>>> (https://bugzilla.redhat.com/show_bug.cgi?id=425958). Whilst >>>>>> the socket file itself has the context listed above, the >>>>>> unix domain socket that sendmail connects to is still >>>>>> initrc_t, as can be seen from the output of "netstat -lpZ": >>>>>> >>>>>> ... >>>>>> unix 2 [ ACC ] STREAM LISTENING 14142 >>>>>> 5853/spamass-milter system_u:system_r:initrc_t:s0 >>>>>> /var/run/spamass-milter/spamass-milter.sock >>>>>> unix 2 [ ACC ] STREAM LISTENING 13794 >>>>>> 5779/milter-regex system_u:system_r:initrc_t:s0 >>>>>> /var/spool/milter-regex/sock >>>>>> ... >>>>>> >>>>>> So, my questions are: >>>>>> >>>>>> 1. Why are the sockets still initrc_t? >>>>>> 2. Is this a kernel issue or a userspace issue that should be >>>>>> fixed in >>>>>> the milters? >>>>>> 3. Should there be a standard place for milter sockets to >>>>>> live, and if >>>>>> so, where? >>>>>> 4. How come this worked OK in Fedora 7 and previous releases? >>>>> Looking at the source code for these applications, I see that >>>>> both of >>>>> them use the smfi_setconn() function in the sendmail milter >>>>> library to >>>>> set up the sockets. It's therefore likely that this problem is >>>>> common to >>>>> all milter applications that use unix domain sockets. >>>>> >>>>> I'm now of the opinion that moving the directory locations >>>>> for these sockets is a bad idea - it would need corresponding >>>>> changes in people's >>>>> sendmail configuration files, which would lead to problems for >>>>> people >>>>> doing package updates, or installing from upstream sources. >>>>> Setting different context types for the directories (e.g. make >>>>> /var/spool/milter-regex spamd_var_run_t) would seem a better >>>>> option, along with policy tweaks to allow sendmail to do the >>>>> permissions checks >>>>> and write to the sockets). >>>>> >>>>> I'm still confused about the initrc_t sockets though. >>>>> >>>>> Paul. >>>>> >>>>> -- >>>>> fedora-selinux-list mailing list >>>>> fedora-selinux-list@redhat.com >>>>> https://www.redhat.com/mailman/listinfo/fedora-selinux-list >>>> Ok I will add this to the next update. >>> What exactly is "this"? The 4 "allow" rules mentioned above, the >>> context >>> type change for /var/spool/milter-regex mentioned later, both? >>> >>> Cheers, Paul. >>> >> Context change for /var/spool/milter-regex to spamd_var_run_t. >> sendmail >> can already use sockets in this directory. > So that includes the: > > allow sendmail_t initrc_t:unix_stream_socket { read write > connectto } > > ? > > Cheers, Paul. > Nope. I don't know what is running as initrc_t and I would bet this is a leaked file descriptor. Or at least a redirectiron of stdin/stdout.
I don't think it's a leaked file descriptor - that would be dontaudit-able, right? By not allowing communications with the initrc_t:unix_stream_socket, the milter fails to work:
==> /var/log/audit/audit.log <== type=AVC msg=audit(1200408212.783:142453): avc: denied { connectto } for pid=7805 comm="sendmail" path="/var/spool/milter-regex/sock" scontext=system_u:system_r:sendmail_t:s0 tcontext=system_u:system_r:initrc_t:s0 tclass=unix_stream_socket type=SYSCALL msg=audit(1200408212.783:142453): arch=40000003 syscall=102 success=no exit=-13 a0=3 a1=bfd9f600 a2=b7f79bd4 a3=0 items=0 ppid=7764 pid=7805 auid=0 uid=0 gid=0 euid=0 suid=0 fsuid=0 egid=51 sgid=51 fsgid=51 tty=(none) comm="sendmail" exe="/usr/sbin/sendmail.sendmail" subj=system_u:system_r:sendmail_t:s0 key=(null)
==> /var/log/maillog <== Jan 15 14:43:32 goalkeeper sendmail[7805]: NOQUEUE: connect from ard120.neoplus.adsl.tpnet.pl [83.26.189.120] Jan 15 14:43:32 goalkeeper sendmail[7805]: AUTH: available mech=CRAM-MD5 DIGEST-MD5, allowed mech=CRAM-MD5 DIGEST-MD5 LOGIN PLAIN Jan 15 14:43:32 goalkeeper sendmail[7805]: m0FEhW21007805: Milter (milter-regex): error connecting to filter: Permission denied Jan 15 14:43:32 goalkeeper sendmail[7805]: m0FEhW21007805: Milter (milter-regex): to error state Jan 15 14:43:32 goalkeeper sendmail[7805]: m0FEhW21007805: Milter: initialization failed, temp failing commands Jan 15 14:43:32 goalkeeper sendmail[7805]: m0FEhW21007805: SMTP MAIL command (pathrusim@zombanewmedia.com) from ard120.neoplus.adsl.tpnet.pl [83.26.189.120] tempfailed (due to previous checks)
The initrc_t type shows up in netstat but not in ls: # netstat -aZp | grep initrc tcp 0 0 goalkeeper.intra.:bacula-fd *:* LISTEN 5864/bacula-fd system_u:system_r:initrc_t:s0 udp 0 0 rbldns.intra.cit:domain *:* 5885/rbldnsd system_u:system_r:initrc_t:s0 unix 2 [ ACC ] STREAM LISTENING 14142 5853/spamass-milter system_u:system_r:initrc_t:s0 /var/run/spamass-milter/spamass-milter.sock unix 2 [ ACC ] STREAM LISTENING 13794 5779/milter-regex system_u:system_r:initrc_t:s0 /var/spool/milter-regex/sock unix 2 [ ] DGRAM 2150436 5779/milter-regex system_u:system_r:initrc_t:s0 unix 2 [ ] DGRAM 14141 5853/spamass-milter system_u:system_r:initrc_t:s0 # ls -lZ /var/run/spamass-milter/spamass-milter.sock /var/spool/milter-regex/sock srwxr-xr-x sa-milt sa-milt system_u:object_r:spamd_var_run_t:s0 /var/run/spamass-milter/spamass-milter.sock srw------- mregex mregex system_u:object_r:spamd_var_run_t:s0 /var/spool/milter-regex/sock
Paul.
Ok then I guess we need to label
chcon -t spamd_exec_t /usr/sbin/spamass-milter
And then build policy off of that.
Whilst that might result in a solution for spamass-milter, it's not going to help milter-regex or potentially any other milter (they're all likely to use the same libmilter [sendmail] API for setting up the sockets).
There seems to be something odd about sockets in general; the netstat output quoted above shows a couple of network-listening sockets with type initrc_t too, from a further two non-milter programs, namely bacula and rbldnsd. I also see the same issue with nasd and rpc.quotad. though I can also see a bunch of listening sockets with system_u:system_r:unconfined_t on my desktop.
Why might some of these apps transition to unconfined_t and others not?
And why does "ls" show a different type than "netstat"?
Paul.
ls is showing file context and netstat is showing processes.
Processes running as unconfined_t were started by unconfined_t without going through an initrc_exec_t type. So either you started these processes directly or the label of their start up script is wrong
ls -lZ /etc/init.d/*
restorecon -R -v /etc/init.d
Should fix.
I suspect that the stuff running in unconfined_t gets started as part of a Gnome session rather than via an initscript.
So we need to allow sendmail to read sockets setup by initrc_t?
Is it true to say (I think it is) that any process started via an initscript that doesn't transition to another domain (e.g. stuff that nobody has written policy for yet) will be in initrc_t?
If so, the following is currently needed.
Adding init_stream_connect_script(mailserver_delivery) init_rw_script_stream_sockets(mailserver_delivery)
Will allow all programs that deliver mail to read/write/connectto initrc_t unix_stream_sockets.
This looks right for now, though I'm tempted to hack together policy for my two milters at least. What I was thinking of was creating a milter_template along the lines of the apache_content_template that could be used as a starting point for milter applications (all of which will communicate with sendmail [and postfix too for that metter] in the same way), and then add on anything necessary for each individual milter (some of which would require nothing else, some would require database connectivity etc.).
Paul.
Sounds good.
All daemons that do not have policy and start from init run as initrc_t which is unconfined. The problem is that if they exec a domain that is confined, the transition will happen, and the execed domain may need to communicate back to the initrc_t domain. So allowing maildelivery to communicate with initrc_t seems to make sense.
inetd runs unknown domains as inetd_child_t which is also unconfined.