In the environ regression test, when the exec() of the child process
fails, we don't report FAIL to stdout, so the regression tests consider
it an error rather than a failure and abort, short-circuiting the
test script.
This commit fixes this by emitting the FAIL message when the result
from the wait() syscall indicates the child process did not succeed.
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
exec_stack picked up a fix to address a semantic change introduced in
4.8 kernels.
However older kernels don't need the extra permission and the exec_stack
test is the only test we currently have that caught the semantic change.
Keep exec_stack to the minimum set of permissions needed for a given
kernel. Which allows us to use exec_stack as a test to detect the
semantic change showing up in unexpected place until we have a test
specifically designed for this.
Signed-off-by: John Johansen <john.johansen@canonical.com>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
https://launchpad.net/bugs/1628745
The following upstream kernel commit changed the semantics of the exec
permission check in the 4.8 kernel:
commit 9f834ec18defc369d73ccf9e87a2790bfa05bf46
Author: Linus Torvalds <torvalds@linux-foundation.org>
Date: Mon Aug 22 16:41:46 2016 -0700
binfmt_elf: switch to new creds when switching to new mm
That change means that the target profile of an exec transition must
have permission to map the binary being executed. This patch fixes
regression test failures while the exec_stack.sh test is running against
4.8 and newer kernels by granting mapping permission to the target
profile.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
profile_A//&:ns1://unconfined (mixed)
this is confusing and can even break some trusted helpers. The unconfined
profile has been special cased and now will report enforce when stacking
with unconfined
profile_A//&:ns1://unconfined (enforce)
This patch fixes the regression tests to work with this change
Signed-off-by: John Johansen <john.johansen@canonical.com>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
Bug: https://launchpad.net/bugs/1521400
The onexec.sh test has periodically exhibited unexplicable failures that
are possibly due to race conditions when onexec.sh is verifying the
/proc/PID/attr/{current,exec} values of the process under test. This
patch attempts to solve the flaky test failures by removing the need for
IPC to coordinate between the test script and the test program.
The old onexec test program is removed and the transition test program
is used instead. This allows for the test script to tell the transition
test program what its current and exec procattr labels should be via
command line options.
Since IPC is no longer needed, the signal:ALL allow rule can be dropped
from the test profile. A new allow rule is needed to grant reading of
/proc/*/attr/{current,exec} since transition must verify the contents of
these files.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
Add optional command line parameters to the transition test program that
can be used to verify a certain label and/or mode that should be found
in /proc/self/attr/exec.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
Add logic to the at_secure.sh test script to verifies that the parser is
new enough to support change_profile exec modes and determine what the
kernel's support for change_profile exec modes before verifying that
AT_SECURE is set correctly after various exec transitions.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
The gen_change_profile() function must be changed to allow the extra
condition in change_profiles rules.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
Quote $@ so that the for loop doesn't iterate on the space-delimited
version of the rule(s) under test. This allows more complex rules such
as "change_profile foo -> bar," to be tested where, before this patch,
only "change_profile," could be tested.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
The AT_SECURE value in the kernel's per-process auxiliary vector is what
signals to libc that the process' environment should be scrubbed. This
new set of regression tests checks the AT_SECURE value after performing
the various types of exec transitions that AppArmor supports (file rules
with different exec access modes and change_profile rules).
Different versions of the kernel handle AT_SECURE differently with
respect to change_profile rules. This change in behavior was introduced
in the AppArmor profile stacking kernel support and the tests are
conditionalized to account for this change.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
Extend the transition test program to allow for changing to a new
profile. This change will be useful in test scripts that need to test
operations across profile stacks and/or profile changes.
The calls to aa_stack_onexec() and aa_stack_profile() are build-time
conditionalized on whether or not the libapparmor being used has
implemented those functions.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
This test will soon be made to do more than just stack a new profile.
It will be extended to allow for changing to a new profile and,
therefore, should be renamed.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
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 $() operator only captures stdout, it never matched
in the string comparison. This is easily solved by redirecting
its stderr to stdout.
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
The stacking tests worked fine when using in-tree programs and libraries
but the tests unexpectedly failed when USE_SYSTEM=1 was specified. This
patch makes use of the addimage:$test argument to mkprofile.pl to
generate the correct file permissions needed to use the system binaries.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
I configured the stacking test binary to only be built when libapparmor
2.11 is present. The versioning of the 2.11 Beta 1 release (2.10.95)
causes that check to fail and the stacking tests to not be used.
This patch adjusts the libapparmor version check to be aware of the 2.11
Beta 1 versioning.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Christian Boltz <apparmor@cboltz.de>
This adds support to the profile generator script for change_profile
rules, giving the ability to write the 3 factor version of the rule
(e.g. "change_profile /t -> A_PROFILE") which was significantly more
difficult using straight raw rules, which is why we don't have any 3
factor rule tests.
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
Now that the onexec test program notices that it failed to send SIGSTOP
to itself, causing a whole bunch of tests to be detected as failing,
grant the ability to send and receive signals to the onexec tests.
(The onexec tests are not tests intended to verify signal mediation.)
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
The onexec test was ignoring errors from the kill() call, so it didn't
notice when it had failed to send SIGSTOP to itself.
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
Based on a patch by John Johansen <john.johansen@canonical.com>
(I converted the check to look for the process directory in /proc
rather than sending signal 0 to the task, as John had done in a patch
sent to me, to prevent failures in signal delivery from blocking the
check from working correctly.)
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
Based on a patch by John Johansen <john.johansen@canonical.com>
Add more details to the checks in the regression tests onexec tests,
to make debugging failures easier. Also, use more local variables
to indicate what and how many arguments are expected to the onexec
check_* functions.
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
The stacking test binary links against libapparmor for
aa_stack_profile() and aa_stack_onexec(), which will be present in 2.11.
This means that regression test builds using the system libapparmor
should not build the stacking test binary unless the libapparmor 2.11 or
newer is present.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
Policy namespaces are not well supported in older parsers and kernels.
This is a case where the kernel support doesn't seem to be working.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
Stacking is a complex feature and, in order to sufficiently test all
aspects of stacking, a relatively complex test program is needed.
This patch adds a program that can call
aa_stack_onexec()/aa_stack_profile(), perform file IO on a given file
path, verify that the current confinement context is what it is expected
to be, and/or execute itself or another program.
The confinement context verification can handle stacked labels with any
ordering.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
Some tidying up is needed in order to reuse do_open(). This patch
eliminates the chance of returning 0 due to errno being not set. It also
adjusts the file string to be const.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
The idea is that the $test profile grants $file access and the
$othertest profile grants $subfile access. Both profiles grant
$stacktest access. The tests verify that after changing to the stacked
$othertest//&$test profile, only $stacktest can be accessed.
Similar tests are also added for stacking with a namespaced profile.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
The kernel patches that implement AppArmor profile stacking made changes
that allow the the backed for change_profile to detect if the target
profile does not exist prior to checking if the current profile allows
the change_profile.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
https://launchpad.net/bugs/1540666
Reuse the new parse_label() function to initialize named_transition
structs so that transition targets, when used with change_profile, are
properly seperated into a profile namespace and profile name.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
Bug: https://launchpad.net/bugs/1379874
https://launchpad.net/bugs/1544387
Don't split namespaces from profile names using YACC grammar. Instead,
treat the entire string as a label in the grammer. The label can then be
split into a namespace and a profile name using the new parse_label()
function.
This fixes a bug that caused the profile keyword to not be used with a
label containing a namespace in the profile declaration.
Fixing this bug uncovered a bad parser test case at
simple_tests/profile/profile_ns_ok1.sd. The test case mistakenly
included two definitions of the :foo:unattached profile despite being
marked as expected to pass. I've adjusted the name of one of the
profiles to :foo:unattached2.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
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>
[cboltz: Add 'unspec' to the network domain keywords of the utils]
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>
Remove the Perl aa-exec implementation, move the aa-exec(8) man page to
binutils/, and point the regression test to the C based aa-exec in
binutils/.
Note that the new C aa-exec does not implement the --file option which
was present in the Perl aa-exec. It encouraged running programs as root,
since root privileges were required to load the specified profile.
All other features of the Perl aa-exec are present in the C aa-exec.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
Add regression tests for the --profile, --namespace, and --immediate
options of aa-exec.
A new variable is added to uservars.inc to point to the in-tree or
system aa-exec depending on the presence of the USE_SYSTEM=1 make
variable at build time.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
In testing against the 4.1 kernel, the syscall_sysctl testcase started
failing even in the unconfined case. What the test program does is
attempt to adjust the kernel.threads-max sysctl to be slightly larger
and see if the operation succeeds by reading the value back out. It
also attempts to save the original value and restore it. The test
was failing because (in VMs at least) the default value chosen by
the kernel for the kernel.threads-max setting was high enough that
attempts to increase it would be ignored (likely to prevent too much
use of kernel memory by threads), helpfully without any message being
report to dmesg. Thus, the initial read of the current value would
succeed, the write of that value + 1024 would appear to succeed,
but then reading the value back out and comparing it to the expected
value would fail, as it would still be the original value, not the
expected new value.
This patch attempts to address this by first attempting to raise
the value, and if that does not appear to work, to then attempt
to lower it. It also refactors the code a bit by creating helper
functions to perform the actual sysctl(2) calls to make the code a
bit easier to read.
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
The local defines in the link_subset test collide and result in build
warnings. Replace the defines with a naming that won't collide and
makes it clear a local define for the test is being used.
Signed-off-by: John Johansen <john.johansen@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
1. The test is using the wrong defines: It is using the defines from the
parser for the packed dfa permissions. This set of permissions is not
meant to be exposed to the outside world
2. The kernel is using the wrong mapping function for the permissions
in the file class. This results in partially exposing the packed
permissions, but even then it doesn't fully line up with the packed
permissions, and is not correct for several of the potential permissions.
Attached is a patch that fixes the test, and moves the two tests that
fail due to the kernel to xpass.
Signed-off-by: John Johansen <john.johansen@canonical.com>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
In the commit "Rev 3169: regression tests: have
ptrace use PTRACE_GETREGSET by default", I created
some ifdef magic to use the per arch general purpose
register data structures for various architectures,
including arm64. Unfortunately, in the upstream glibc commit
7d05a8168b
<bits/ptrace.h> is no longer included in the arm64 specific user.h,
which defined the structure as 'struct user_pt_regs'; instead user.h
was converted to define 'struct user_regs_struct'. Because of this, the
ptrace test fails to compile on arm64 when glibc is 2.20 or newer.
This patch adjusts the ptrace test to use the newer structure on arm64
if it's detected that a newer glibc is detected and reverts to using
the older one for older glibcs. It also adds an error when compiling
on architectures that haven't been incorporated yet.
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: John Johansen <john.johansen@canonical.com>
Bug: https://bugs.launchpad.net/apparmor/+bug/1470985
The ptrace regression test fails to compile on the arm64 platform,
because it uses PTRACE_GETREGS and not the newer PTRACE_GETREGSET
interface for getting access to arch-specific register information[0].
However, fixing it is complicated by the fact that the struct name
for for the general purpose registers is not named consistently
across architectures. This patch attempts to address those issues,
and compiles at least on i386, amd64, arm64, arm (armhf), ppc64,
and ppc64el. The test is verified to continue to function correctly
on i386 and amd64.
[0] https://sourceware.org/ml/archer/2010-q3/msg00193.html
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: John Johansen <john.johansen@canonical.com>
The aa_policy_cache_new() and aa_policy_cache_remove() functions are
changed to accept a dirfd parameter.
The cache dirfd (by default, /etc/apparmor.d/cache) is opened earlier in
aa_policy_cache_new(). Previously, the directory wasn't accessed until
later in the following call chain:
aa_policy_cache_new() -> init_cache_features() -> create_cache()
Because of this change, the logic to create the cache dir must be moved
from create_cache() to aa_policy_cache_new().
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
The most common case when creating an aa_kernel_interface object will be
to do so while using the current kernel's feature set for the
kernel_features parameter. Rather than have callers instantiate their
own aa_features object in this situation, aa_kernel_interface_new()
should do it for them if they specify NULL for the kernel_features
parameter.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
The most common case when creating an aa_policy_cache object will be to
do so while using the current kernel's feature set for the
kernel_features parameter. Rather than have callers instantiate their
own aa_features object in this situation, aa_policy_cache_new() should
do it for them if they specify NULL for the kernel_features parameter.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
This patch changes the aa_policy_cache_new() prototype and gets rid of
aa_policy_cache_is_valid() and aa_policy_cache_create().
The create bool of aa_policy_cache_new() is replaced with a 16 bit
unsigned int used to specify the maximum number of caches that should be
present in the specified cache directory. If the number is exceeded, the
old cache directories are reaped. The definition of "old" is private to
libapparmor and only 1 cache directory is currently supported. However,
that will change in the near future and multiple cache directories will
be supported.
If 0 is specified for the max_caches parameter, no new caches can be
created and only an existing, valid cache can be used. An error is
returned if no valid caches exist in that case.
If UINT16_MAX is specified, an unlimited amount of caches can be created
and reaping is disabled.
This means that 0 to (2^16)-2, or infinite, caches will be supported in
the future.
This change allows for the parser to continue to support the
--skip-bad-cache (by passing 0 for max_caches) and the --write-cache
option (by passing 1 or more for max_caches) without confusing
libapparmor users with the aa_policy_cache_{is_valid,create}()
functions.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
Rename require_features to require_kernel_features and
have_features to kernel_features
to indicate they are tests for kernel features, as now there are tests
for parser features and in the future there might be library features
as well.
Signed-off-by: John Johansen <john.johansen@canonical.com>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
has not been updated. The issue is that the regression tests detect the
kernel features set and generate policy that the parser may not be able
to compile.
Augment the regressions tests with a couple simple functions to test what
is supported by the parser, and update the test conditionals to use them.
Signed-off-by: John Johansen <john.johansen@canonical.com>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
A number of simple query tests based on read and write perms of files
and directories.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
The test program was querying its own profile. Adjust the profile
generation so that a separate profile is generated and have query_label
query the separate profile.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
This patch modifies the socketpair.c test to verify the return value of
aa_getpeercon() based upon the expected label and expected mode lengths.
The test had to be changed slightly so that the returned mode, from
aa_getpeercon(), was preserved. It was being overwritten with the
special NO_MODE value.
This change helps to make sure that future changes to the code behind
aa_getpeercon() does not unintentionally change the function's return
value.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
