Allow for a leading '&' character to be present in the named transition
target strings to indicate that the transition should stack the current
profile with the specified profile.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
The parser was splitting up the namespace and profile name from named
transition targets only to rejoin it later when creating the binary
policy. This complicated the changes needed to support the stacking
identifier '&' in named transition targets.
To keep the stacking support simple, this patch keeps the entire named
transition target string intact from initial profile parsing to writing
out the binary.
All of these changes are straightforward except the hunk that removes
the namespace string addition to the vector in the process_dfa_entry()
function. After speaking with John, kernels with stacking have support
for consuming the namespace with the profile name.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
This patch separates the label parsing functionality from the program
termination and memory allocation duties of parse_label(). This will
ultimately help in creating simple helper functions that simply need to
check if a label contains a namespace.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
The add_named_transition function was written in a way that is difficult
to understand while attempting to read the function. This patch attempts
to clean it up.
First, this patch removes this confusing code flow issue:
if (!entry->ns) { ... }
if (entry->ns) { ... } else { ... }
It then unifies the way that the ns and nt_name strings of the cod_entry
struct are handled prior to calling add_entry_to_x_table() and/or
returning. ns and nt_name are now guaranteed to be NULL before
performing either of those actions.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
The copy_cod_entry() function was not copying the nt_name field of the
cod_entry struct.
This was discovered during code review and I'm not certain if it causes
any real world bugs.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
Instead of reusing opt_named_transition and be forced to reconstruct the
target path when is looks like ":odd:target", create simpler grammer
rules that have nothing to do with named transitions and namespaces.
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]
deny rules don't allow ix, Px, Ux etc. - only 'deny /foo x,' is allowed.
Acked-by: Seth Arnold <seth.arnold@canonical.com> for trunk and 2.10
Note: Seth mentioned in the mail that he doesn't like the 'deny x'
section too much, but we didn't find a better solution when discussing
it on IRC. Therefore I keep the patch unchanged, but will happily
review a follow-up patch if someone sends one ;-)
This test causes `make check` to fail but it is known bug so mark it as
a TODO test.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
This patch frees some leaked memory that occur when errors are
detected while adding variables to the parser's symbol table. While not
a significant issue currently due to the parser exiting on failures, as
the process of library-ifying the parser continues, these need to be
addressed. It also makes it easier to use tools like Address Sanitizer
on the parser against our test suite.
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
The first entry in the grouping_count array is never initialized to 0;
subsequent depths are. This patch initializes the whole array.
Issue found with valgrind.
Signed-off-by: Steve Beattie <steve@nxnw.org> (with improvement from Seth)
Acked-by: Seth Arnold <seth.arnold@canonical.com>
This patch fixes the unit test memory leaks found
by intrigeri using AddressSanitizer in the following email thread:
https://lists.ubuntu.com/archives/apparmor/2015-August/008491.html
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
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.
BugLink: http://bugs.launchpad.net/bugs/1534405
Patch -r 2952 switched over to using the library kernel interface, and
added a kernel_interface parameter to the dir_cb struct, that is
used to process directories.
Unfortunately kernel_interface parameter of the dir_cb struct is not being
properly initialized resulting in odd failures and sefaults when the parser
is processing directories.
Signed-off-by: John Johansen <john.johansen@canonical.com>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
This adds a basic support for parallel compiles. It uses a fork()/wait
model due to the parsers current dependence on global variables and
structures. It has been setup in a similar manner to how cilk handles
multithreading to make it easy to port to a managed thread model once
the parser removes the dependence on global compute structures in the
backend.
This patch adds two new command line flags
-j <n> or --jobs <n>
which follows the make syntax of specifying parallel jobs currently
defaults to -jauto
-j8 or --jobs=8 allows for 8 parallel jobs
-jauto or --jobs=auto sets the jobs to the # of cpus
-jx4 or --jobs=x4 sets the jobs to # of cpus * 4
-jx1 is equivalent to -jauto
Note: unlike make -j must be accompanied by an option
--max-jobs=<n>
allows setting hard cap on the number of jobs that can be specified
by --jobs. It defaults to the number of processors in the system * 8.
It supports the "auto" and "max" keywords, and using x<n> for a
multiple of the available cpus.
additionally the -d flag has been modified to take an optional parameter
and
--debug=jobs
will output debug information for the job control logic.
In light testing on one machine the job control logic provides a nice
performance boost. On an x86 test machine with 60 profiles in the
/etc/apparmor.d/ directory, for the command
time apparmor_parser -QT /etc/apparmor.d/
old (equiv of -j1):
real 0m10.968s
user 0m10.888s
sys 0m0.088s
ubuntu parallel load using xargs:
real 0m8.003s
user 0m21.680s
sys 0m0.216s
-j:
real 0m6.547s
user 0m17.900s
sys 0m0.132s
Signed-off-by: John Johansen <john.johansen@canonical.com>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
https://launchpad.net/bugs/1526085
Revno 2934 'Add fns to handle profile removal to the kernel interface'
introduced a regression in the parser's namespace support by causing the
--namespace-string option to be ignored. This resulted in the profile(s)
being loaded into the global namespace rather than the namespace
specified on the command line.
This patch fixes the bug by setting the Profile object's ns member, if
the --namespace-string option was specified, immediately after the
Profile object is allocated.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
2.9.x and 2.10 had some time stamp bugs around cache handling that
result in the cache getting a wrong time stamp, and then not getting
correctly updated when policy changes.
Force cache recompiles for these versions by bumping the parser abi
Signed-off-by: John Johansen <john.johansen@canonical.com>
Acked-by: Steve Beattie <steve@nxnw.org>
The parser is incorrectly screening off the bind flags on remount. The
following patch by Ash Wilson fixes this issue
BugLink: http://bugs.launchpad.net/bugs/1272028
Signed-off-by: Ash Wilson <smashwilson@gmail.com>
Signed-off-by: John Johansen <john.johansen@canonical.com>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
Drop the reference to the libapparmor policy_cache pseudo object when
the parser is done.
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
In recent commits, Tyler fixed some problems with the caching behavior
of the parser, as well as adjusting and improving the caching test
script to verify these behaviors.
In doing so, the test script adjusts the mtime of various
files and ensures that the written files have the expected mtime
timestamp. Unfortunately, the os.utime() function used to adjust mtime
in python 3.2 (as included in Ubuntu 12.04 LTS) does not update with
nanosecond precision, even though the timestamps returned by os.stat()
do have precision to nanoseconds. This causes the tests to fail when
running under python 3.2 with errors like the following:
======================================================================
FAIL: test_abstraction_newer_rewrites_cache (__main__.AAParserCachingTests)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/��PKGBUILDDIR��/parser/tst/testlib.py", line 50, in new_unittest_func
return unittest_func(self)
File "./caching.py", line 424, in test_abstraction_newer_rewrites_cache
self._set_mtime(self.abstraction, abstraction_mtime)
File "./caching.py", line 238, in _set_mtime
self.assertEquals(os.stat(path).st_mtime, mtime)
AssertionError: 1440337039.40212 != 1440337039.4021206
The following patch creates a new time stamp equality assertion
function that detects if it's running on python 3.2 or earlier, and
loosens the equality bounds when comparing the passed timestamps. On
python 3.3 and newer, where writing timestamps with nanosecond precision
is supported, the strict equality assertion is used.
(Note: I did not convert all time stamp comparisons, just ones where
the timestamp written and checked could be based on a timestamp
derived from os.stat().)
Reference: https://bugs.python.org/issue12904
Signed-off-by: Steve Beattie <steve@nxnw.org>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
The contents of the policy cache files varies based on kernel feature
support found in apparmorfs but the caching tests are mostly about
whether or not a cache file was generated and with the right timestamps.
This patch makes it so that the tests are not entirely skipped when
apparmorfs is not available. Instead, a flat features file will be used
in most cases and only the specific tests that require apparmorfs will
be skipped.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
This makes several improvements to the parser caching tests to verify
that the parser is properly consuming the mtime of profiles and
abstractions when dealing with the policy cache.
It introduces a simple abstraction file and tests the mtime handling by
changing the mtime on the profile, abstraction, apparmor_parser, and
cache file in various combinations to check the parser's behavior.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
This patch fixes a regression in setting the cache file's timestamp
handling that was introduced in r3079:
Set cache file tstamp to the mtime of most recent policy file tstamp
The previously used utimes(2) syscall requires a two element timeval
array. The first element in the array is the atime to be used and the
second element is the mtime. The equivalent of a one element timeval
array was being passed to it, resulting in garbage being used for the
mtime value. The utimes(2) syscall either failed, due to the invalid
input, or set mtime to an unexpected value. The return code wasn't being
checked so the failure went unknown.
This patch switches to utimensat(2) for a couple reasons. The UTIME_OMIT
special value allows us to preserve the inode's atime without calling
stat(2) to fetch the atime. It also allows for nanosecond precision
which better aligns with what stat(2) returns on the input profile and
abstraction files. That means that we can have the exact same mtime on
the input profile or abstraction file and the output cache file.
https://launchpad.net/bugs/1484178
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Seth Arnold <seth.arnold@canonical.com>
Acked-by: John Johansen <john.johansen@canonical.com>
"rcapparmor kill" results in a funny error message:
/lib/apparmor/rc.apparmor.functions: line 441: return: -v: invalid option
return: usage: return [n]
SLE12 includes a patch that prevents this error message, but also
prevents that $? is handed over correctly to rc_status. This means that
"rcapparmor kill" will happily display "done" even with a compiled-in
apparmor module that can't be unloaded.
This patch is the improved version - it adds a small helper function to
set $? (as handed over to aa_log_end_msg()) and then calls rc_status -v.
This means that "rcapparmor kill" now shows "failed" because it's
impossible to unload something that is compiled directly into the
kernel.
References: https://bugzilla.opensuse.org/show_bug.cgi?id=862170 (non-public)
Acked-by: Seth Arnold <seth.arnold@canonical.com> for 2.9 and trunk
Some of the newly added simple_tests contain lines like
profile foo@{FOO} { }
which are not supported by the tools because the '}' is in the same line,
while the tools expect \n as rule separator.
This patch changes those tests to
profile foo@{FOO} {
}
Acked-by: John Johansen <john.johansen@canonical.com>
Some of the include files added to simple_tests recently don't live in
one of the main include directories (includes/, includes-preamble/ or
include_tests/) which lets test-parser-simple-tests.py fail because
those files don't contain EXRESULT.
Instead of adding more exceptions to test-parser-simple-tests.py, this
patch adds DESCRIPTION and EXRESULT to those include files.
Acked-by: John Johansen <john.johansen@canonical.com>
