On 8/16/23 3:39 PM, David Vernet wrote: > On Wed, Aug 16, 2023 at 03:10:23PM -0400, David Marchevsky wrote: >> On 8/16/23 1:44 PM, David Vernet wrote: >>> On Wed, Aug 16, 2023 at 09:58:13AM -0700, Dave Marchevsky wrote: >>>> This patch adds selftests that exercise kfunc flavor relocation >>>> functionality added in the previous patch. The actual kfunc defined in >>>> kernel/bpf/helpers.c is >>>> >>>> struct task_struct *bpf_task_acquire(struct task_struct *p) >>>> >>>> The following relocation behaviors are checked: >>>> >>>> struct task_struct *bpf_task_acquire___one(struct task_struct *name) >>>> * Should succeed despite differing param name >>>> >>>> struct task_struct *bpf_task_acquire___two(struct task_struct *p, void *ctx) >>>> * Should fail because there is no two-param bpf_task_acquire >>>> >>>> struct task_struct *bpf_task_acquire___three(void *ctx) >>>> * Should fail because, despite vmlinux's bpf_task_acquire having one param, >>>> the types don't match >>>> >>>> Changelog: >>>> v1 -> v2: https://lore.kernel.org/bpf/20230811201346.3240403-2-davemarchevsky@xxxxxx/ >>>> * Change comment on bpf_task_acquire___two to more accurately reflect >>>> that it fails in same codepath as bpf_task_acquire___three, and to >>>> not mention dead code elimination as thats an implementation detail >>>> (Yonghong) >>>> >>>> Signed-off-by: Dave Marchevsky <davemarchevsky@xxxxxx> >>>> --- >>>> .../selftests/bpf/prog_tests/task_kfunc.c | 1 + >>>> .../selftests/bpf/progs/task_kfunc_success.c | 37 +++++++++++++++++++ >>>> 2 files changed, 38 insertions(+) >>>> >>>> diff --git a/tools/testing/selftests/bpf/prog_tests/task_kfunc.c b/tools/testing/selftests/bpf/prog_tests/task_kfunc.c >>>> index 740d5f644b40..99abb0350154 100644 >>>> --- a/tools/testing/selftests/bpf/prog_tests/task_kfunc.c >>>> +++ b/tools/testing/selftests/bpf/prog_tests/task_kfunc.c >>>> @@ -79,6 +79,7 @@ static const char * const success_tests[] = { >>>> "test_task_from_pid_current", >>>> "test_task_from_pid_invalid", >>>> "task_kfunc_acquire_trusted_walked", >>>> + "test_task_kfunc_flavor_relo", >>>> }; >>>> >>>> void test_task_kfunc(void) >>>> diff --git a/tools/testing/selftests/bpf/progs/task_kfunc_success.c b/tools/testing/selftests/bpf/progs/task_kfunc_success.c >>>> index b09371bba204..ffbe3ff72639 100644 >>>> --- a/tools/testing/selftests/bpf/progs/task_kfunc_success.c >>>> +++ b/tools/testing/selftests/bpf/progs/task_kfunc_success.c >>> >>> Do you think it's worth it to also add a failure case for if there's no >>> correct bpf_taks_acquire___one(), to verify e.g. that we can't resolve >>> bpf_task_acquire___three(void *ctx) __ksym __weak? >>> >> >> IIUC you're asking about whether it's possible to fail loading the program >> entirely if _none_ of the three variants resolve successfully? If so, I >> sent out a response to another email in this round of your comments that should >> address it. > > Sorry, that was unclear in the way I worded it. I understand that the > program will still load if none of the variants resolve succesfully. I > was asking whether we should add a test that verifies that the wrong > variant won't be resolved if a correct one isn't present. Maybe that's > overkill? Seems prudent to add just in case, though. Something like > this: > > SEC("tp_btf/task_newtask") > int BPF_PROG(test_task_kfunc_flavor_relo_not_found, > struct task_struct *task, u64 clone_flags) > { > /* Neither symbol should resolve successfully. */ > if (bpf_ksym_exists(bpf_task_acquire___two)) > err = 1; > else if (bpf_ksym_exists(bpf_task_acquire___three)) > err = 2; > > return 0; > } > I was leaning towards pushing back here, but agree with you after digging and seeing: * weak symbols aren't discussed in the C99 standard at all and are an ELF specification concept * my previous bullet point isn't really relevant to what you're saying here as you're talking about the linkage process more generally * Then I started digging in the C99 standard and realized that even if there was something in there that would allow me to say "well by definition I don't need to test for this", would be too obscure and I should just add the test >> >>>> @@ -18,6 +18,13 @@ int err, pid; >>>> */ >>>> >>>> struct task_struct *bpf_task_acquire(struct task_struct *p) __ksym __weak; >>>> + >>>> +struct task_struct *bpf_task_acquire___one(struct task_struct *task) __ksym __weak; >>>> +/* The two-param bpf_task_acquire doesn't exist */ >>>> +struct task_struct *bpf_task_acquire___two(struct task_struct *p, void *ctx) __ksym __weak; >>>> +/* Incorrect type for first param */ >>>> +struct task_struct *bpf_task_acquire___three(void *ctx) __ksym __weak; >>>> + >>>> void invalid_kfunc(void) __ksym __weak; >>>> void bpf_testmod_test_mod_kfunc(int i) __ksym __weak; >>>> >>>> @@ -55,6 +62,36 @@ static int test_acquire_release(struct task_struct *task) >>>> return 0; >>>> } >>>> >>>> +SEC("tp_btf/task_newtask") >>>> +int BPF_PROG(test_task_kfunc_flavor_relo, struct task_struct *task, u64 clone_flags) >>>> +{ >>>> + struct task_struct *acquired = NULL; >>>> + int fake_ctx = 42; >>>> + >>>> + if (bpf_ksym_exists(bpf_task_acquire___one)) { >>>> + acquired = bpf_task_acquire___one(task); >>>> + } else if (bpf_ksym_exists(bpf_task_acquire___two)) { >>>> + /* Here, bpf_object__resolve_ksym_func_btf_id's find_ksym_btf_id >>>> + * call will find vmlinux's bpf_task_acquire, but subsequent >>>> + * bpf_core_types_are_compat will fail >>>> + */ >>>> + acquired = bpf_task_acquire___two(task, &fake_ctx); >>>> + err = 3; >>>> + return 0; >>>> + } else if (bpf_ksym_exists(bpf_task_acquire___three)) { >>>> + /* bpf_core_types_are_compat will fail similarly to above case */ >>>> + acquired = bpf_task_acquire___three(&fake_ctx); >>>> + err = 4; >>>> + return 0; >>>> + } >>>> + >>>> + if (acquired) >>>> + bpf_task_release(acquired); >>> >>> Might be slightly simpler to do the release + return immediately in the >>> bpf_task_acquire___one branch, and then to just do the following here >>> without the if / else: >>> >>> err = 5; >>> return 0; >>> >>> What do you think? >>> >> >> Eh, I like this form more because it's easier to visually distinguish that the >> bpf_task_acquire___one case above is not a 'failure' case and should >> successfully resolve, whereas the other two bail out early. >> >>>> + else >>>> + err = 5; >>>> + return 0; >>>> +} >>>> + >>>> SEC("tp_btf/task_newtask") >>>> int BPF_PROG(test_task_acquire_release_argument, struct task_struct *task, u64 clone_flags) >>>> { >>>> -- >>>> 2.34.1 >>>> >>>>
