CRIU can influence the PID of the threads it wants to create.
CRIU uses /proc/sys/kernel/ns_last_pidto tell the kernel which
PID it wants for the next clone().
So it has to write to that file. This feels like a problematic as
it opens up the container writing to all sysctl_kernel_t.
Using new label container_t will just write to
sysctl_kernel_ns_last_pid_t instad writing to more generic
sysctl_kernel_t files.
This is the update I have made based on suggestions for the previous
patches to add a udev_run interface. This adds the new domain udevadm_t
which is entered from /usr/bin/udevadm.
It seems to meet the needs that I have, but there are some things to
note that are probably important.
1) There are a few systemd services that use udevadm during startup.
I have granted the permisssions that I need based on denials I was
seeing during startup (the machine would fail to start without the
permisions).
2) In the udev.fc file there are other binaries that I don't have on a
RHEL7 box that maybe should also be labeled udevadm_exec_t.
e.g. /usr/bin/udevinfo and /usr/bin/udevsend
But as I don't have those binaries to test, I have not updated the
type of that binary.
3) There are some places that call udev_domtrans that maybe should now
be using udevadm_domtrans - rpm.te, hal.te, hotplug.te. Again,
these are not things that I am using in my current situation and am
unable to test the interactions to know if the change is correct.
Other than that, I think this was a good suggestion to split udevadm
into a different domain.
Only change for v4 is to use stream_connect_pattern as suggested.
Signed-off-by: Dave Sugar <dsugar@tresys.com>
I'm seeing a denial for udev to map /etc/udev/hwdb.bin.
This creates and uses a new interface to allow the needed
permission for udev.
type=AVC msg=audit(1551886176.948:642): avc: denied { map } for pid=5187 comm="systemd-udevd" path="/etc/udev/hwdb.bin" dev="dm-1" ino=6509618 scontext=system_u:system_r:udev_t:s0-s0:c0.c1023 tcontext=system_u:object_r:systemd_hwdb_t:s0 tclass=file permissive=1
Updated from previous to create a new interface.
Signed-off-by: Dave Sugar <dsugar@tresys.com>
I have a case where I'm labeling media with my own types to control
access. But that is requiring that I relabel from iso9660_t to my
own type. This interface allows that relabel.
type=AVC msg=audit(1551621984.372:919): avc: denied { relabelfrom } for pid=9717 comm="mount" scontext=staff_u:staff_r:mymedia_sudo_t:s0-s0:c0.c1023 tcontext=system_u:object_r:iso9660_t:s0 tclass=filesystem permissive=0
Signed-off-by: Dave Sugar <dsugar@tresys.com>
AIDE has a compile time option WITH_MMAP which allows AIDE to
map files during scanning. RHEL7 has set this option in the
aide rpm they distribute.
Changes made to add a tunable to enable permissions allowing
aide to map files that it needs. I have set the default to
false as this seems perfered (in my mind).
Signed-off-by: Dave Sugar <dsugar@tresys.com>
Currently freshclam can only be started from cron or init. This adds
the option of starting from a different process and optionally
transitioning or staying in the callers domain.
Signed-off-by: Dave Sugar <dsugar@tresys.com>
When trying to remove files_read_non_auth_files(restorecond_t), the
following AVC denial occurs:
type=AVC msg=audit(1550921968.443:654): avc: denied { open } for
pid=281 comm="restorecond"
path="/etc/selinux/refpolicy/contexts/customizable_types" dev="vda1"
ino=928006 scontext=system_u:system_r:restorecond_t
tcontext=system_u:object_r:default_context_t tclass=file
permissive=1
type=AVC msg=audit(1550921968.443:654): avc: denied { read } for
pid=281 comm="restorecond" name="customizable_types" dev="vda1"
ino=928006 scontext=system_u:system_r:restorecond_t
tcontext=system_u:object_r:default_context_t tclass=file
permissive=1
As /etc/selinux/${SELINUXTYPE}/contexts/customizable_types is needed by
restorecond, allow this access.
Since systemd 241, systemd-journald is using kill(pid, 0) in order to
find dead processes and reduce its cache. The relevant commit is
91714a7f42
("journald: periodically drop cache for all dead PIDs"). This commit
added a call to pid_is_unwaited(c->pid), which is a function implemented in
https://github.com/systemd/systemd/blob/v241/src/basic/process-util.c#L936 :
bool pid_is_unwaited(pid_t pid) {
/* Checks whether a PID is still valid at all, including a zombie */
if (pid < 0)
return false;
if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */
return true;
if (pid == getpid_cached())
return true;
if (kill(pid, 0) >= 0)
return true;
return errno != ESRCH;
}
This new code triggers the following AVC denials:
type=AVC msg=audit(1550911933.606:332): avc: denied { signull }
for pid=224 comm="systemd-journal"
scontext=system_u:system_r:syslogd_t
tcontext=system_u:system_r:auditd_t tclass=process permissive=1
type=AVC msg=audit(1550911933.606:333): avc: denied { signull }
for pid=224 comm="systemd-journal"
scontext=system_u:system_r:syslogd_t
tcontext=system_u:system_r:dhcpc_t tclass=process permissive=1
type=AVC msg=audit(1550911933.606:334): avc: denied { signull }
for pid=224 comm="systemd-journal"
scontext=system_u:system_r:syslogd_t
tcontext=system_u:system_r:sshd_t tclass=process permissive=1
I'm seeing a bunch of denials for various processes (some refpolicy
domains, some my own application domains) attempting to access
/etc/pki. They seem to be working OK even with the denial. The
tunable authlogin_nsswitch_use_ldap controls access to cert_t
(for domains that are part of nsswitch_domain attribute). Use this
new interface when that tunable is off to quiet the denials.
Signed-off-by: Dave Sugar <dsugar@tresys.com>
When calling hostnamectl to set the hostname it needs sys_admin
capability to actually set the hostname.
Feb 13 11:47:14 localhost.localdomain systemd-hostnamed[7221]: Failed to set host name: Operation not permitted
type=AVC msg=audit(1550058524.656:1988): avc: denied { sys_admin } for pid=7873 comm="systemd-hostnam" capability=21 scontext=system_u:system_r:systemd_hostnamed_t:s0 tcontext=system_u:system_r:systemd_hostnamed_t:s0 tclass=capability permissive=0
Signed-off-by: Dave Sugar <dsugar@tresys.com>
init (systemd) needs to read /etc/hostname during boot
to retreive the hostname to apply to the system.
Feb 06 18:37:06 localhost.localdomain kernel: type=1400 audit(1549478223.842:3): avc: denied { read } for pid=1 comm="systemd" name="hostname" dev="dm-1" ino=1262975 scontext=system_u:system_r:init_t:s0 tcontext=system_u:object_r:net_conf_t:s0 tclass=file permissive=0
Signed-off-by: Dave Sugar <dsugar@tresys.com>
Stubby is a DNS resolver that encrypts DNS queries and transmits them to
a resolver in a TLS channel. It therefore requires less permissions than
a traditionnal DNS resolver such as named or unbound (provided by module
"bind").
cf. https://dnsprivacy.org/wiki/display/DP/DNS+Privacy+Daemon+-+Stubby
This program is packaged for Arch Linux, Debian, etc.
DNS-over-TLS uses TCP port 853, which does not seem to conflict with
existing ports. Label it like other DNS ports.
init_dbus_chat(stubby_t) is required on systemd-based distributions
because stubby's service uses DynamicUser=yes [1]. Without this
statement, the following denials are reported by dbus:
type=USER_AVC msg=audit(1550007165.936:257): pid=274 uid=81
auid=4294967295 ses=4294967295 subj=system_u:system_r:system_dbusd_t
msg='avc: denied { send_msg } for msgtype=method_call
interface=org.freedesktop.DBus member=Hello
dest=org.freedesktop.DBus spid=649
scontext=system_u:system_r:stubby_t
tcontext=system_u:system_r:system_dbusd_t tclass=dbus permissive=1
exe="/usr/bin/dbus-daemon" sauid=81 hostname=? addr=? terminal=?'
type=USER_AVC msg=audit(1550007165.939:258): pid=274 uid=81
auid=4294967295 ses=4294967295 subj=system_u:system_r:system_dbusd_t
msg='avc: denied { send_msg } for msgtype=method_call
interface=org.freedesktop.systemd1.Manager
member=LookupDynamicUserByUID dest=org.freedesktop.systemd1 spid=649
tpid=1 scontext=system_u:system_r:stubby_t
tcontext=system_u:system_r:init_t tclass=dbus permissive=1
exe="/usr/bin/dbus-daemon" sauid=81 hostname=? addr=? terminal=?'
type=USER_AVC msg=audit(1550007165.939:259): pid=274 uid=81
auid=4294967295 ses=4294967295 subj=system_u:system_r:system_dbusd_t
msg='avc: denied { send_msg } for msgtype=method_return dest=:1.39
spid=1 tpid=649 scontext=system_u:system_r:init_t
tcontext=system_u:system_r:stubby_t tclass=dbus permissive=1
exe="/usr/bin/dbus-daemon" sauid=81 hostname=? addr=? terminal=?'
[1] https://github.com/getdnsapi/stubby/blob/v0.2.5/systemd/stubby.service#L8
systemd-modules-load is used to pre-load kernal modules as the system comes up.
It was running initc_t which didn't have permissions to actually load kernel
modules. This change sets up a new domain for this service and grants permission
necessary to load kernel modules.
Feb 05 03:38:20 4c4c4544-0052-5410-8043-b1c04f503232 kernel: type=1400 audit(1549337898.886:10): avc: denied { read } for pid=4257 comm="systemd-modules" name="fuse.ko.xz" dev="dm-1" ino=2390271 scontext=system_u:system_r:initrc_t:s0 tcontext=system_u:object_r:modules_object_t:s0 tclass=file permissive=1
Feb 05 03:38:20 4c4c4544-0052-5410-8043-b1c04f503232 kernel: type=1400 audit(1549337898.886:11): avc: denied { open } for pid=4257 comm="systemd-modules" path="/usr/lib/modules/3.10.0-957.1.3.el7.x86_64/kernel/fs/fuse/fuse.ko.xz" dev="dm-1" ino=2390271 scontext=system_u:system_r:initrc_t:s0 tcontext=system_u:object_r:modules_object_t:s0 tclass=file permissive=1
Signed-off-by: Dave Sugar <dsugar@tresys.com>
There are several nacl binaries that need labels.
Put an ifdef debian for some chromium paths.
Git policy misses chromium_role() lines, were they in another patch that was
submitted at the same time?
I don't know what this is for but doesn't seem harmful to allow it:
type=PROCTITLE msg=audit(28/01/19 19:31:42.361:3218) : proctitle=/bin/bash /usr/bin/google-chrome
type=SYSCALL msg=audit(28/01/19 19:31:42.361:3218) : arch=x86_64 syscall=openat success=yes exit=3 a0=0xffffff9c a1=0x563328f7b590 a2=O_WRONLY|O_CREAT|O_TRUNC a3=0x1b6 items=0 ppid=5158 pid=5166 auid=test uid=test gid=test euid=test suid=test fsuid=test egid=test sgid=test fsgid=test tty=pts7 ses=232 comm=google-chrome exe=/bin/bash subj=user_u:user_r:chromium_t:s0 key=(null)
type=AVC msg=audit(28/01/19 19:31:42.361:3218) : avc: granted { associate } for pid=5166 comm=google-chrome name=63 scontext=user_u:object_r:chromium_t:s0 tcontext=system_u:object_r:proc_t:s0 tclass=filesystem
type=AVC msg=audit(28/01/19 19:31:42.361:3218) : avc: granted { create } for pid=5166 comm=google-chrome name=63 scontext=user_u:user_r:chromium_t:s0 tcontext=user_u:object_r:chromium_t:s0 tclass=file
type=AVC msg=audit(28/01/19 19:31:42.361:3218) : avc: granted { add_name } for pid=5166 comm=google-chrome name=63 scontext=user_u:user_r:chromium_t:s0 tcontext=user_u:user_r:chromium_t:s0 tclass=dir
Allow domain_use_interactive_fds() for running via ssh -X.
Allow managing xdg data, cache, and config.
Allow reading public data from apt and dpkg, probably from lsb_release or some
other shell script.
How does the whold naclhelper thing work anyway? I'm nervous about process
share access involving chromium_sandbox_t, is that really what we want?
Added lots of other stuff like searching cgroup dirs etc.
Allow transition to dpkg_t with nnp, Dominick seems to imply this shouldn't
be necessary.
Lots of little stuff for system_cronjob_t.
Other minor trivial changes that should be obvious.
I think the old timesync labelling wasn't working anyway due to -- for a
directory name.
A couple of patches for devicekit calling dmidecode (this is part of replacing
some kmem access that was discussed on this list and rejected as a misfeature
in Debian DMI related code ages ago).
The rest should be obvious.
Chris PeBenito
Lukas Vrabec
Chris PeBenito
Sugar, David
Nicolas Iooss
Russell Coker
Alexander Miroshnichenko
Jason Zaman
Alexander Miroshnichenko