This is needed for winbindd (since samba 4.4.x), but smbd could also need it.
References: https://bugzilla.opensuse.org/show_bug.cgi?id=990006
Acked-by: Seth Arnold <seth.arnold@canonical.com> for trunk, 2.10 and 2.9.
This behaviour makes sense (for example to force the confined program to
use a fallback path), but is probably surprising for users, so we should
document it.
References: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=826218#37
Acked-by: John Johansen <john.johansen@canonical.com> for trunk, 2.10 and 2.9
Thanks to reading the wrong directory in read_inactive_profiles()
(profile_dir instead of extra_profile_dir), aa-genprof never asked about
using a profile from the extra_profile_dir.
Sounds like an easy fix, right? ;-)
After fixing this (last chunk), several other errors popped up, one
after the other:
- get_profile() missed a required parameter in a serialize_profile() call
- when saving the profile, it was written to extra_profile_dir, not to
profile_dir where it (as a now-active profile) should be. This is
fixed by removing the filename from existing_profiles{} so that it can
pick up the default name.
- CMD_FINISHED (when asking if the extra profile should be used or a new
one) behaved exactly like CMD_CREATE_PROFILE, but this is surprising
for the user. Remove it to avoid confusion.
- displaying the extra profile was only implemented in YaST mode
- get_pager() returned None, not an actual pager. Since we have 'less'
hardcoded at several places, also return it in get_pager()
Finally, also remove CMD_FINISHED from the get_profile() test in
test-translations.py.
(test-translations.py is only in trunk, therefore this part of the patch
is obviously trunk-only.)
Acked-by: Seth Arnold <seth.arnold@canonical.com> for trunk
Acked-by: John Johansen <john.johansen@canonical.com> for trunk + a 50% ACK for 2.10 and 2.9
Acked-by: Kshitij Gupta <kgupta8592@gmail.com> for trunk, 2.10 and 2.9
Those events are actually network events, so ideally we should map them
as such. Unfortunately this requires bigger changes, so here is a hotfix
that ignores those events and thus avoids crashing aa-logprof.
References: https://bugs.launchpad.net/apparmor/+bug/1577051https://bugs.launchpad.net/apparmor/+bug/1582374
Acked-by: Seth Arnold <seth.arnold@canonical.com> for trunk, 2.10 and 2.9
The latest iputils merged ping and ping6 into a single binary that does
both IPv4 and IPv6 pings (by default, it really does both).
This means we need to allow network inet6 raw in the ping profile.
References: https://bugzilla.opensuse.org/show_bug.cgi?id=980596
(contains more details and example output)
Acked-by: Steve Beattie <steve@nxnw.org> for trunk, 2.10 and 2.9
From: Simon McVittie <simon.mcvittie@collabora.co.uk>
Date: Wed, 4 May 2016 13:48:36 +0100
Subject: dbus-session-strict: allow access to the user bus socket
If dbus is configured with --enable-user-bus (for example in the
dbus-user-session package in Debian and its derivatives), and the user
session is started with systemd, then the "dbus-daemon --session" will be
started by "systemd --user" and listen on $XDG_RUNTIME_DIR/bus. Similarly,
on systems where dbus-daemon has been replaced with kdbus, the
bridge/proxy used to provide compatibility with the traditional D-Bus
protocol listens on that same socket.
In practice, $XDG_RUNTIME_DIR is /run/user/$uid on all systemd systems,
where $uid represents the numeric uid. I have not used /{var/,}run here,
because systemd does not support configurations where /var/run and /run
are distinct; in practice, /var/run is a symbolic link.
Based on a patch by Sjoerd Simons, which originally used the historical
path /run/user/*/dbus/user_bus_socket. That path was popularized by the
user-session-units git repository, but has never been used in a released
version of dbus and should be considered unsupported.
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
From: Simon McVittie <simon.mcvittie@collabora.co.uk>
Date: Wed, 11 May 2016 13:52:56 +0100
Subject: syscall_sysctl test: correctly skip if CONFIG_SYSCTL_SYSCALL=n
This test attempts to auto-skip the sysctl() part if that syscall
was not compiled into the current kernel, via
CONFIG_SYSCTL_SYSCALL=n. Unfortunately, this didn't actually work,
for two reasons:
* Because "${test} ro" wasn't in "&&", "||", a pipeline or an "if",
and it had nonzero exit status, the trap on ERR was triggered,
causing execution of the error_handler() shell function, which
aborts the test with a failed status. The rules for ERR are the
same as for "set -e", so we can circumvent it in the same ways.
* Because sysctl_syscall.c prints its diagnostic message to stderr,
but the $() operator only captures stdout, it never matched
in the string comparison. This is easily solved by redirecting
its stderr to stdout.
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
Some people have the full hostname in their syslog messages, so
libapparmor needs to accept hostnames that contain dots.
References: https://bugs.launchpad.net/apparmor/+bug/1453300 comments
#1 and #2 (the log samples reported by scrx in #apparmor)
Acked-by: Seth Arnold <seth.arnold@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
for trunk, 2.10 and 2.9.
https://launchpad.net/bugs/1569316
When Ubuntu made the jump from network-manager 1.0.4 to 1.1.93, the
dnsmasq process spawned from network-manager started hitting a
disconnected path denial:
audit: type=1400 audit(1460463960.943:31702): apparmor="ALLOWED"
operation="connect" info="Failed name lookup - disconnected path"
error=-13 profile="/usr/sbin/dnsmasq"
name="run/dbus/system_bus_socket" pid=3448 comm="dnsmasq"
requested_mask="wr" denied_mask="wr" fsuid=65534 ouid=0
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Christian Boltz <apparmor@cboltz.de>
Since the latest openSUSE Tumbleweed update (dovecot 2.2.21 -> 2.2.22),
dovecot/auth writes to /var/run/dovecot/stats-user.
Acked-by: Seth Arnold <seth.arnold@canonical.com> for trunk, 2.10 and 2.9.
acmetool is an alternative client for Let's Encrypt.
(https://github.com/hlandau/acme/)
It stores the certificates etc. in the following directory layout:
/var/lib/acme/live/<domain> -> ../certs/<hash>
/var/lib/acme/certs/<hash>/cert
/var/lib/acme/certs/<hash>/chain
/var/lib/acme/certs/<hash>/privkey -> ../../keys/<hash>/privkey
/var/lib/acme/certs/<hash>/url
/var/lib/acme/certs/<hash>/fullchain
/var/lib/acme/keys/<hash>/privkey
This patch adds the needed permissions to the ssl_certs and ssl_keys
abstractions so that the certificates can be used.
Acked-by: Seth Arnold <seth.arnold@canonical.com> for trunk, 2.10 and 2.9.
In /etc/nscd.conf there is an option allowing to restart nscd after a
certain time. However, this requires reading /proc/self/cmdline -
otherwise nscd will disable paranoia mode.
References: https://bugzilla.opensuse.org/show_bug.cgi?id=971790
Acked-By: Jamie Strandboge <jamie@canonical.com> for trunk, 2.10 and 2.9
Since 2.9 r2978, test-aa.py fails thanks to a missing import of
'var_transform'. This patch adds the missing import.
Acked-by: Seth Arnold <seth.arnold@canonical.com>
write_prof_data[hat] is correct (it only contains one profile, see bug 1528139),
write_prof_data[profile][hat] is not and returns an empty (sub)hasher.
This affects RE_PROFILE_START and RE_PROFILE_BARE_FILE_ENTRY.
Acked-by: Kshitij Gupta <kgupta8592@gmail.com> for trunk, 2.9 and 2.10
dovecot-lda needs to read and write /tmp/dovecot.lda.*.
It also needs to be able to execute sendmail to send sieve vacation
mails.
For now, I'm using a child profile for sendmail to avoid introducing a
new profile with possible regressions. This child profile is based on
the usr.sbin.sendmail profile in extras and should cover both postfix'
and sendmail's sendmail.
I also mixed in some bits that were needed for (postfix) sendmail on my
servers, and dropped some rules that were obsolete (directory rules not
ending with a /) or covered by an abstraction.
In the future, we might want to provide a stand-alone profile for
sendmail (based on this child profile) and change the rule in the
dovecot-lda profile to Px.
References: https://bugzilla.opensuse.org/show_bug.cgi?id=954959https://bugzilla.opensuse.org/show_bug.cgi?id=954958
Acked-by: Seth Arnold <seth.arnold@canonical.com> for trunk, 2.10 and 2.9.
https://launchpad.net/bugs/1546455
Don't filter out AF_UNSPEC from the list of valid protocol families so
that the parser will accept rules such as 'network unspec,'.
There are certain syscalls, such as socket(2), where the LSM hooks are
called before the protocol family is validated. In these cases, AppArmor
was emitting denials even though socket(2) will eventually fail. There
may be cases where AF_UNSPEC sockets are accepted and we need to make
sure that we're mediating those appropriately.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Suggested-by: Steve Beattie <steve@nxnw.org>
Acked-by: John Johansen <john.johansen@canonical.com>
If a profile file contains multiple profiles, aa-mergeprof crashes on
saving in write_profile() because the second profile in the file is not
listed in 'changed'. (This happens only if the second profile didn't
change.)
This patch first checks if 'changed' contains the profile before
pop()ing it.
Reproducer: copy utils/test/cleanprof_test.in to your profile directory
and run aa-mergeprof utils/test/cleanprof_test.out. Then just press
's' to save the profile.
Acked-by: Kshitij Gupta <kgupta8592@gmail.com> for trunk, 2.10 and 2.9
If autodep() is called with a pname starting with / (which can happen
for (N)amed exec depending on the user input), this pname is mapped to
bin_name.
This might look like a good idea, however if the given pname doesn't
exist as file on-disk, autodep() returns None instead of a (mostly
empty) profile. (Reproducer: choose (N)amed, enter "/foo/bar")
Further down the road, this results in two things:
a) the None result gets written as empty profile file (with only a "Last
modified" line)
b) a crash if someone chooses to add an abstraction to the None, because
None doesn't support the delete_duplicates() method for obvious
reasons ;-)
Unfortunately this patch also introduces a regression - aa-logprof now
fails to follow the exec and doesn't ask about the log events for the
exec target anymore. However this doesn't really matter because of a) -
asking and saving to /dev/null vs. not asking isn't a real difference
;-)
Actually the patch slightly improves things - it creates a profile for
the exec target, but only with the depmod() defaults (abstractions/base)
and always in complain mode.
I'd prefer a patch that also creates a complete profile for the exec
target, but that isn't as easy as fixing the issues mentioned above and
therefore is something for a future fix. To avoid we forget it, I opened
https://bugs.launchpad.net/apparmor/+bug/1545155
Note: 2.9 "only" writes an empty file and doesn't crash - but writing
an empty profile is still an improvement.
Acked-by: Kshitij Gupta <kgupta8592@gmail.com> for trunk, 2.10 and 2.9
Most probably-file log events can also be network events. Therefore
check for request_mask in all events, not only file_perm, file_inherit
and (from the latest bugreport) file_receive.
References: https://bugs.launchpad.net/apparmor/+bug/1540562
Acked-by: Kshitij Gupta <kgupta8592@gmail.com> for trunk, 2.10 and 2.9.
If reading /dev/urandom failed, the corresponding file descriptor was
leaked through the error path.
Coverity CID #56012
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
I suspect that the incorrect description of EPERM was copied from
the aa_change_hat man page, where it is possible to see EPERM if the
application is not confined by AppArmor.
This patch corrects the description by documenting that the only
possible way to see EPERM is if a confined application has the
no_new_privs bit set.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Reported-by: Seth Arnold <seth.arnold@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
It is possible that file descriptors will be revalidated after an
aa_change_profile() but there is a lot of complexity involved that
doesn't need to be spelled out in the man page. Instead, mention that
revalidation is possible but the only way to ensure that file
descriptors are not passed on is to close them.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Reported-by: Seth Arnold <seth.arnold@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
The statement was meant to convey the difference between aa_change_hat()
and aa_change_profile(). Unfortunately, it read as if there was
something preventing a program from using aa_change_profile() twice to
move from profile A to profile B and back to profile A, even if profiles
A and B contained the necessary rules.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Reported-by: Seth Arnold <seth.arnold@canonical.com>
Acked-by: Christian Boltz <apparmor@cboltz.de>
When hitting an unknown line while parsing a profile, it's a good idea
to include that line in the error message ;-)
Note: 2.9 would print a literal \n because it doesn't have apparmor.fail,
so it will get a slightly different patch with spaces instead of \n.
Acked-by: Steve Beattie <steve@nxnw.org> for trunk, 2.10 and 2.9.
Merge from trunk revision 3342
bug: https://bugs.launchpad.net/bugs/1531325
This patch defines the arch specific registers struct for s390 for the
ptrace regression test.
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: John Johansen <john.johansen@canonical.com>
'!' is a reserved symbol and needs to be escaped in AARE.
Note: aare.py only exists in trunk, therefore this part is trunk-only.
Acked-by: Seth Arnold <seth.arnold@canonical.com> for trunk, 2.10 and 2.9 as needed.
The capnames list missed a comma, which lead to the funny
"mac_overridesyslog" capability name.
__debug_capabilities() seems to be the only user of capnames, which
might explain why this bug wasn't noticed earlier.
Acked-by: Seth Arnold <seth.arnold@canonical.com> for trunk, 2.10 and 2.9.
If parse_event_for_tree() raises an AppArmorException (for example
because of an invalid/unknown request_mask), catch it in read_log() and
re-raise it together with the log line causing the Exception.
Acked-by: Steve Beattie <steve@nxnw.org> for trunk, 2.10 and 2.9.
Note: 2.9 can't handle \n in exception messages, therefore I'm using spaces.
handle_children() has some special code for handling link events with
denied_mask = 'l'. Unfortunately this special code depends on a regex
that matches the old, obsolete log format - in a not really parsed
format ("^from .* to .*$").
The result was that aa-logprof did not ask about events containing 'l'
in denied_mask.
Fortunately the fix is easy - delete the code with the special handling
for 'l' events, and the remaining code that handles other file
permissions will handle it :-)
References: Bugreport by pfak on IRC
Testcase (with hand-tuned log event):
aa-logprof -f <( echo 'Jan 7 03:11:24 mail kernel: [191223.562261] type=1400 audit(1452136284.727:344): apparmor="ALLOWED" operation="link" profile="/usr/sbin/smbd" name="/foo" pid=10262 comm=616D617669736420286368362D3130 requested_mask="l" denied_mask="l" fsuid=110 ouid=110 target="/bar"')
should ask to add '/foo l,' to the profile.
Acked-by: Seth Arnold <seth.arnold@canonical.com> for trunk, 2.10 and 2.9.
allow read on /run/systemd/resolve/resolv.conf for systems using networkd
(LP: #1529074)
Signed-Off-By: Jamie Strandboge <jamie@canonical.com>
Acked-by: Christian Boltz <apparmor@cboltz.de>
r2637 added support for parsing unix rules, but forgot to add write
support. The result was that a profile lost its unix rules when it was
saved.
This patch adds the write_unix_rules() and write_unix() functions (based
on the write_pivot_root() and write_pivot_root_rules() functions) and
makes sure they get called at the right place.
The cleanprof testcase gets an unix rule added to ensure it's not
deleted when writing the profile. (Note that minitools_test.py is not
part of the default "make check", however I always run it.)
References: https://bugs.launchpad.net/apparmor/+bug/1522938https://bugzilla.opensuse.org/show_bug.cgi?id=954104
Acked-by: Tyler Hicks <tyhicks@canonical.com> for trunk, 2.10 and 2.9.
'change_hat' events have the target profile in 'name2', not in 'name'
(which is None and therefore causes a crash when checking if it contains
'//')
Also add the log event causing this crash to the libapparmor testsuite.
References: https://bugs.launchpad.net/apparmor/+bug/1523297
Acked-by: John Johansen <john.johansen@canonical.com> for trunk, 2.10 and 2.9.
Parsing variables was broken in several ways:
- empty quotes (representing an intentionally empty value) were lost,
causing parser failures
- items consisting of only one letter were lost due to a bug in RE_VARS
- RE_VARS didn't start with ^, which means leading garbage (= syntax
errors) was ignored
- trailing garbage was also ignored
This patch fixes those issues in separate_vars() and changes
var_transform() to write out empty quotes (instead of nothing) for empty
values.
Also add some tests for separate_vars() with empty quotes and adjust
several tests with invalid syntax to expect an AppArmorException.
var_transform() gets some tests added.
Finally, remove 3 testcases from the "fails to raise an exception" list
in test-parser-simple-tests.py.
Acked-by: John Johansen <john.johansen@canonical.com> for trunk and 2.9
(which also implies 2.10)
Note: 2.9 doesn't have test-parser-simple-tests.py, therefore it won't
get that part of the patch.
Creating a file is in theory covered by the 'a' permission, however
discussion on IRC brought up that depending on the open flags it might
not be enough (real-world example: creating the apache pid file).
Therefore change the mapping to 'w' permissions. That might allow more
than needed in some cases, but makes sure the profile always works.
Acked-by: Kshitij Gupta <kgupta8592@gmail.com> for 2.9, 2.10 and trunk
After switching to winbindd as test profile, comments about the ntpd
profile don't make sense anymore ;-)
The patch also includes a whitespace fix.
Acked-by: Kshitij Gupta <kgupta8592@gmail.com>
Acked-by: Steve Beattie <steve@nxnw.org>
Acked-by: John Johansen <john.johansen@canonical.com> for 2.9
Change minitools_test.py to use the winbind instead of the ntpd profile
for testing. The tests broke because the ntpd profile has the
attach_disconnected flag set now, and therefore didn't match the
expected flags anymore.
Also replace the usage of filecmp.cmp() in the cleanprof test with
reading the file and using assertEqual - this has the advantage that we
get a full diff instead of just "files differ".
Acked-by: Kshitij Gupta <kgupta8592@gmail.com>
Acked-by: John Johansen <john.johansen@canonical.com> for 2.9
This allows to run minitools_test.py as non-root user.
Also add a check that only creates the force-complain directory if it
doesn't exist yet.
Note: With this patch applied, there are still 4 failing tests, probably
caused by changes in the profiles that are used in the tests.
Acked-by: Kshitij Gupta <kgupta8592@gmail.com>
Acked-by: John Johansen <john.johansen@canonical.com> for 2.9
For some reasons, it needs to do that to find readable, writeable and
executable files.
See also https://bugzilla.opensuse.org/show_bug.cgi?id=945592
Acked-by: Seth Arnold <seth.arnold@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com> for 2.9
Change aa-notify parse_message() to also honor complain mode log events.
This affects both modes - desktop notifications and the summary report.
Acked-by: Steve Beattie <steve@nxnw.org>
Acked-by: John Johansen <john.johansen@canonical.com> for 2.9
aa-complain is part of the enforce/complain/disable triple. Therefore
I expect it to actually load a profile in complain mode.
To do this, it has to delete the 'disable' symlink, but set_complain()
in aa.py didn't do this (and therefore kept the profile disabled).
Acked-by: Kshitij Gupta <kgupta8592@gmail.com>
Acked-by: John Johansen <john.johansen@canonical.com> for 2.9
