From: Jeff Xu <jeffxu@google.com>
To: "Mickaël Salaün" <mic@digikod.net>
Cc: jeffxu@chromium.org, jorgelo@chromium.org, keescook@chromium.org,
linux-security-module@vger.kernel.org, groeck@chromium.org
Subject: Re: [PATCH v2 1/1] selftests/landlock: skip ptrace_test according to YAMA
Date: Thu, 15 Dec 2022 12:34:08 -0800 [thread overview]
Message-ID: <CALmYWFurtUhVqibcPyBXF=pcWRDtDfe3HxMJRGKe29dEAOtVLg@mail.gmail.com> (raw)
In-Reply-To: <43c925e3-ce72-3ba1-15cf-6fbb34c485f1@digikod.net>
Hi Mickaël
Thanks for reviewing.
On Thu, Dec 15, 2022 at 10:34 AM Mickaël Salaün <mic@digikod.net> wrote:
>
> This is much better! We can tailor a bit more the tests though.
>
> On 13/12/2022 19:58, jeffxu@chromium.org wrote:
> > From: Jeff Xu <jeffxu@google.com>
> >
> > Add check for yama setting for ptrace_test.
> >
> > Signed-off-by: Jeff Xu <jeffxu@google.com>
> > ---
> > .../testing/selftests/landlock/ptrace_test.c | 34 +++++++++++++++++++
> > 1 file changed, 34 insertions(+)
> >
> > diff --git a/tools/testing/selftests/landlock/ptrace_test.c b/tools/testing/selftests/landlock/ptrace_test.c
> > index c28ef98ff3ac..8565a25a9587 100644
> > --- a/tools/testing/selftests/landlock/ptrace_test.c
> > +++ b/tools/testing/selftests/landlock/ptrace_test.c
> > @@ -60,6 +60,24 @@ static int test_ptrace_read(const pid_t pid)
> > return 0;
> > }
> >
> > +static int get_ptrace_scope(void)
>
> Please rename to get_yama_ptrace_scope().
>
Done.
> > +{
> > + int ret = -1;
> > + char buf[2];
> > + int fd = open("/proc/sys/kernel/yama/ptrace_scope", O_RDONLY);
> > +
> > + if (fd < 0)
> > + return 0;
> > +
> > + if (read(fd, &buf, 1) < 0)
> > + return -1;
> > +
> > + buf[1] = '\0';
>
> You can replace that with `char buf[2] = {};`
>
Done.
The Compiler seems to be getting a lot smarter :) Thanks.
>
> > + ret = atoi(buf);
> > + close(fd);
> > + return ret;
> > +}
> > +
> > /* clang-format off */
> > FIXTURE(hierarchy) {};
> > /* clang-format on */
> > @@ -69,6 +87,7 @@ FIXTURE_VARIANT(hierarchy)
> > const bool domain_both;
> > const bool domain_parent;
> > const bool domain_child;
> > + const int yama_value;
>
> Please rename to yama_ptrace_scope_max and remove the extra space.
>
why _max ? yama_ptrace_scope_current is more proporate ?
FYI: This is the current sysctl setting.
> > };
> >
> > /*
> > @@ -93,6 +112,7 @@ FIXTURE_VARIANT_ADD(hierarchy, allow_without_domain) {
> > .domain_both = false,
> > .domain_parent = false,
> > .domain_child = false,
> > + .yama_value = 0,
> > };
> >
> > /*
> > @@ -110,6 +130,7 @@ FIXTURE_VARIANT_ADD(hierarchy, allow_with_one_domain) {
> > .domain_both = false,
> > .domain_parent = false,
> > .domain_child = true,
> > + .yama_value = 1,
> > };
> >
> > /*
> > @@ -126,6 +147,7 @@ FIXTURE_VARIANT_ADD(hierarchy, deny_with_parent_domain) {
> > .domain_both = false,
> > .domain_parent = true,
> > .domain_child = false,
> > + .yama_value = 0,
> > };
> >
> > /*
> > @@ -143,6 +165,7 @@ FIXTURE_VARIANT_ADD(hierarchy, deny_with_sibling_domain) {
> > .domain_both = false,
> > .domain_parent = true,
> > .domain_child = true,
> > + .yama_value = 2,
> > };
> >
> > /*
> > @@ -160,6 +183,7 @@ FIXTURE_VARIANT_ADD(hierarchy, allow_sibling_domain) {
> > .domain_both = true,
> > .domain_parent = false,
> > .domain_child = false,
> > + .yama_value = 0,
> > };
> >
> > /*
> > @@ -178,6 +202,7 @@ FIXTURE_VARIANT_ADD(hierarchy, allow_with_nested_domain) {
> > .domain_both = true,
> > .domain_parent = false,
> > .domain_child = true,
> > + .yama_value = 1,
> > };
> >
> > /*
> > @@ -196,6 +221,7 @@ FIXTURE_VARIANT_ADD(hierarchy, deny_with_nested_and_parent_domain) {
> > .domain_both = true,
> > .domain_parent = true,
> > .domain_child = false,
> > + .yama_value = 0,
> > };
> >
> > /*
> > @@ -216,6 +242,7 @@ FIXTURE_VARIANT_ADD(hierarchy, deny_with_forked_domain) {
> > .domain_both = true,
> > .domain_parent = true,
> > .domain_child = true,
> > + .yama_value = 2,
> > };
> >
> > FIXTURE_SETUP(hierarchy)
> > @@ -232,9 +259,16 @@ TEST_F(hierarchy, trace)
> > pid_t child, parent;
> > int status, err_proc_read;
> > int pipe_child[2], pipe_parent[2];
> > + int yama;
>
> Please rename to yama_ptrace_scope.
>
Done.
>
> > char buf_parent;
> > long ret;
> >
> > + yama = get_ptrace_scope();
> > + ASSERT_LE(0, yama);
> > +
> > + if (variant->yama_value < yama)
>
> if (yama_ptrace_scope >= 3)
>
> > + SKIP(return, "unsupported yama value %d", yama);
>
> "Yama forbids any ptrace use (scope 3)"
>
>
why comparing with 3? the test will skip particular hierarchy,
according to current
sysctl yama_ptrace setting.
For example: when kernel.yama.ptrace_scope = 1 the result will be like:
localhost /usr/local/bin # ./ptrace_test
TAP version 13
1..8
# Starting 8 tests from 9 test cases.
# RUN hierarchy.allow_without_domain.trace ...
# SKIP unsupported yama value 1
# OK hierarchy.allow_without_domain.trace
ok 1 # SKIP unsupported yama value 1
# RUN hierarchy.allow_with_one_domain.trace ...
# OK hierarchy.allow_with_one_domain.trace
ok 2 hierarchy.allow_with_one_domain.trace
# RUN hierarchy.deny_with_parent_domain.trace ...
# SKIP unsupported yama value 1
# OK hierarchy.deny_with_parent_domain.trace
ok 3 # SKIP unsupported yama value 1
# RUN hierarchy.deny_with_sibling_domain.trace ...
# OK hierarchy.deny_with_sibling_domain.trace
ok 4 hierarchy.deny_with_sibling_domain.trace
# RUN hierarchy.allow_sibling_domain.trace ...
# SKIP unsupported yama value 1
# OK hierarchy.allow_sibling_domain.trace
ok 5 # SKIP unsupported yama value 1
# RUN hierarchy.allow_with_nested_domain.trace ...
# OK hierarchy.allow_with_nested_domain.trace
ok 6 hierarchy.allow_with_nested_domain.trace
# RUN hierarchy.deny_with_nested_and_parent_domain.trace ...
# SKIP unsupported yama value 1
# OK hierarchy.deny_with_nested_and_parent_domain.trace
ok 7 # SKIP unsupported yama value 1
# RUN hierarchy.deny_with_forked_domain.trace ...
# OK hierarchy.deny_with_forked_domain.trace
ok 8 hierarchy.deny_with_forked_domain.trace
# PASSED: 8 / 8 tests passed.
# Totals: pass:4 fail:0 xfail:0 xpass:0 skip:4 error:0
when yama.ptrace_scope is 2:
localhost /usr/local/bin # sysctl kernel.yama.ptrace_scope=2
kernel.yama.ptrace_scope = 2
localhost /usr/local/bin # ./ptrace_test
TAP version 13
1..8
# Starting 8 tests from 9 test cases.
# RUN hierarchy.allow_without_domain.trace ...
# SKIP unsupported yama value 2
# OK hierarchy.allow_without_domain.trace
ok 1 # SKIP unsupported yama value 2
# RUN hierarchy.allow_with_one_domain.trace ...
# SKIP unsupported yama value 2
# OK hierarchy.allow_with_one_domain.trace
ok 2 # SKIP unsupported yama value 2
# RUN hierarchy.deny_with_parent_domain.trace ...
# SKIP unsupported yama value 2
# OK hierarchy.deny_with_parent_domain.trace
ok 3 # SKIP unsupported yama value 2
# RUN hierarchy.deny_with_sibling_domain.trace ...
# OK hierarchy.deny_with_sibling_domain.trace
ok 4 hierarchy.deny_with_sibling_domain.trace
# RUN hierarchy.allow_sibling_domain.trace ...
# SKIP unsupported yama value 2
# OK hierarchy.allow_sibling_domain.trace
ok 5 # SKIP unsupported yama value 2
# RUN hierarchy.allow_with_nested_domain.trace ...
# SKIP unsupported yama value 2
# OK hierarchy.allow_with_nested_domain.trace
ok 6 # SKIP unsupported yama value 2
# RUN hierarchy.deny_with_nested_and_parent_domain.trace ...
# SKIP unsupported yama value 2
# OK hierarchy.deny_with_nested_and_parent_domain.trace
ok 7 # SKIP unsupported yama value 2
# RUN hierarchy.deny_with_forked_domain.trace ...
# OK hierarchy.deny_with_forked_domain.trace
ok 8 hierarchy.deny_with_forked_domain.trace
# PASSED: 8 / 8 tests passed.
# Totals: pass:2 fail:0 xfail:0 xpass:0 skip:6 error:0
> This check skips the whole test, whereas the issues with Yama are about:
> - a child process tracing its parent;
> - the PTRACE_TRACEME case.
>
> I think the main remaining parts to change is the `if
> (variant->domain_*` checks to extend with the yama_ptrace_scope_max check.
>
> However, it is useful to highlight that a test didn't fully cover
> Landlock checks. I think the best approach is to call SKIP() at the end
> of TEST_F(hierarchy, trace) if yama_ptrace_scope >= 1 . This way, we
> test as much as possible (Landlock and Yama) and we mark the "tampered"
> tests as skipped.
>
I believe the test case should not have a lot of branches and logic
(if/else), which makes
the test case more complex and harder to read. By that reason, SKIP()
is better at beginning
of the testcase.
Another reason is resource cleanup. When SKIP() is not at the
beginning of tests,
the cleanup logic can get much more complicated because there are more
combinations of resource cleanup to
to be dealt with, after SKIP().
Therefore I do not believe in "tests as much as possible" in a single
test, I would rather have a
dedicated test for the situation.
Thoughts ?
>
> > +
> > /*
> > * Removes all effective and permitted capabilities to not interfere
> > * with cap_ptrace_access_check() in case of PTRACE_MODE_FSCREDS.
next prev parent reply other threads:[~2022-12-15 20:34 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-12-13 18:58 [PATCH v2 0/1] selftests/landlock: Fix selftest ptrace_test run fail jeffxu
2022-12-13 18:58 ` [PATCH v2 1/1] selftests/landlock: skip ptrace_test according to YAMA jeffxu
2022-12-15 18:34 ` Mickaël Salaün
2022-12-15 20:34 ` Jeff Xu [this message]
2022-12-15 20:42 ` Jeff Xu
2022-12-16 10:15 ` Mickaël Salaün
2022-12-16 15:18 ` Jeff Xu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CALmYWFurtUhVqibcPyBXF=pcWRDtDfe3HxMJRGKe29dEAOtVLg@mail.gmail.com' \
--to=jeffxu@google.com \
--cc=groeck@chromium.org \
--cc=jeffxu@chromium.org \
--cc=jorgelo@chromium.org \
--cc=keescook@chromium.org \
--cc=linux-security-module@vger.kernel.org \
--cc=mic@digikod.net \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).