| Message ID | 20230505060818.60037-3-zhoufeng.zf@bytedance.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 49e0263ab40f15a11cefa397bb17a0204505cead |
| Headers | show |
| Series | Introduce a new kfunc of bpf_task_under_cgroup | expand |
On Thu, May 4, 2023 at 11:08 PM Feng zhou <zhoufeng.zf@bytedance.com> wrote: > > From: Feng Zhou <zhoufeng.zf@bytedance.com> > > test_progs: > Tests new kfunc bpf_task_under_cgroup(). > > The bpf program saves the new task's pid within a given cgroup to > the remote_pid, which is convenient for the user-mode program to > verify the test correctness. > > The user-mode program creates its own mount namespace, and mounts the > cgroupsv2 hierarchy in there, call the fork syscall, then check if > remote_pid and local_pid are unequal. > > Signed-off-by: Feng Zhou <zhoufeng.zf@bytedance.com> > Acked-by: Yonghong Song <yhs@fb.com> > --- Hi Feng, I have a comment about the methodology of the test, but the patch looks ok to me. Why do we have to test via a tracing program? I think what we need is just a task and a cgroup. Since we have the kfunc bpf_task_from_pid() and bpf_cgroup_from_id(), we can write a syscall program which takes a pid and a cgroup id as input and get the task and cgroup objects directly in the program. I like testing via a syscall program because it doesn't depend on the newtask tracepoint and it should be simpler. But I'm ok with the current version of the patch, just have some thoughts. Hao
在 2023/5/5 15:13, Hao Luo 写道: > On Thu, May 4, 2023 at 11:08 PM Feng zhou <zhoufeng.zf@bytedance.com> wrote: >> >> From: Feng Zhou <zhoufeng.zf@bytedance.com> >> >> test_progs: >> Tests new kfunc bpf_task_under_cgroup(). >> >> The bpf program saves the new task's pid within a given cgroup to >> the remote_pid, which is convenient for the user-mode program to >> verify the test correctness. >> >> The user-mode program creates its own mount namespace, and mounts the >> cgroupsv2 hierarchy in there, call the fork syscall, then check if >> remote_pid and local_pid are unequal. >> >> Signed-off-by: Feng Zhou <zhoufeng.zf@bytedance.com> >> Acked-by: Yonghong Song <yhs@fb.com> >> --- > > Hi Feng, > > I have a comment about the methodology of the test, but the patch > looks ok to me. Why do we have to test via a tracing program? I think > what we need is just a task and a cgroup. Since we have the kfunc > bpf_task_from_pid() and bpf_cgroup_from_id(), we can write a syscall > program which takes a pid and a cgroup id as input and get the task > and cgroup objects directly in the program. > > I like testing via a syscall program because it doesn't depend on the > newtask tracepoint and it should be simpler. But I'm ok with the > current version of the patch, just have some thoughts. > > Hao Yes, your method is also very good. The reason why I did this is because of Song's suggestion before, hope that the parameter of the hook point will have a task, so I chose this to test.
On 5/5/23 12:24 AM, Feng Zhou wrote: > 在 2023/5/5 15:13, Hao Luo 写道: >> On Thu, May 4, 2023 at 11:08 PM Feng zhou <zhoufeng.zf@bytedance.com> >> wrote: >>> >>> From: Feng Zhou <zhoufeng.zf@bytedance.com> >>> >>> test_progs: >>> Tests new kfunc bpf_task_under_cgroup(). >>> >>> The bpf program saves the new task's pid within a given cgroup to >>> the remote_pid, which is convenient for the user-mode program to >>> verify the test correctness. >>> >>> The user-mode program creates its own mount namespace, and mounts the >>> cgroupsv2 hierarchy in there, call the fork syscall, then check if >>> remote_pid and local_pid are unequal. >>> >>> Signed-off-by: Feng Zhou <zhoufeng.zf@bytedance.com> >>> Acked-by: Yonghong Song <yhs@fb.com> >>> --- >> >> Hi Feng, >> >> I have a comment about the methodology of the test, but the patch >> looks ok to me. Why do we have to test via a tracing program? I think >> what we need is just a task and a cgroup. Since we have the kfunc >> bpf_task_from_pid() and bpf_cgroup_from_id(), we can write a syscall >> program which takes a pid and a cgroup id as input and get the task >> and cgroup objects directly in the program. >> >> I like testing via a syscall program because it doesn't depend on the >> newtask tracepoint and it should be simpler. But I'm ok with the >> current version of the patch, just have some thoughts. >> >> Hao > > Yes, your method is also very good. The reason why I did this is because > of Song's suggestion before, hope that the parameter of the hook point > will have a task, so I chose this to test. The motivation of this patch is: Trace sched related functions, such as enqueue_task_fair, it is necessary to specify a task instead of the current task which within a given cgroup. So I think it is okay to have a test related to sched.
diff --git a/tools/testing/selftests/bpf/DENYLIST.s390x b/tools/testing/selftests/bpf/DENYLIST.s390x index c7463f3ec3c0..5061d9e24c16 100644 --- a/tools/testing/selftests/bpf/DENYLIST.s390x +++ b/tools/testing/selftests/bpf/DENYLIST.s390x @@ -26,3 +26,4 @@ user_ringbuf # failed to find kernel BTF type ID of verif_stats # trace_vprintk__open_and_load unexpected error: -9 (?) xdp_bonding # failed to auto-attach program 'trace_on_entry': -524 (trampoline) xdp_metadata # JIT does not support calling kernel function (kfunc) +test_task_under_cgroup # JIT does not support calling kernel function (kfunc) diff --git a/tools/testing/selftests/bpf/prog_tests/task_under_cgroup.c b/tools/testing/selftests/bpf/prog_tests/task_under_cgroup.c new file mode 100644 index 000000000000..4224727fb364 --- /dev/null +++ b/tools/testing/selftests/bpf/prog_tests/task_under_cgroup.c @@ -0,0 +1,53 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2023 Bytedance */ + +#include <sys/syscall.h> +#include <test_progs.h> +#include <cgroup_helpers.h> +#include "test_task_under_cgroup.skel.h" + +#define FOO "/foo" + +void test_task_under_cgroup(void) +{ + struct test_task_under_cgroup *skel; + int ret, foo; + pid_t pid; + + foo = test__join_cgroup(FOO); + if (!ASSERT_OK(foo < 0, "cgroup_join_foo")) + return; + + skel = test_task_under_cgroup__open(); + if (!ASSERT_OK_PTR(skel, "test_task_under_cgroup__open")) + goto cleanup; + + skel->rodata->local_pid = getpid(); + skel->bss->remote_pid = getpid(); + skel->rodata->cgid = get_cgroup_id(FOO); + + ret = test_task_under_cgroup__load(skel); + if (!ASSERT_OK(ret, "test_task_under_cgroup__load")) + goto cleanup; + + ret = test_task_under_cgroup__attach(skel); + if (!ASSERT_OK(ret, "test_task_under_cgroup__attach")) + goto cleanup; + + pid = fork(); + if (pid == 0) + exit(0); + + ret = (pid == -1); + if (ASSERT_OK(ret, "fork process")) + wait(NULL); + + test_task_under_cgroup__detach(skel); + + ASSERT_NEQ(skel->bss->remote_pid, skel->rodata->local_pid, + "test task_under_cgroup"); + +cleanup: + test_task_under_cgroup__destroy(skel); + close(foo); +} diff --git a/tools/testing/selftests/bpf/progs/test_task_under_cgroup.c b/tools/testing/selftests/bpf/progs/test_task_under_cgroup.c new file mode 100644 index 000000000000..56cdc0a553f0 --- /dev/null +++ b/tools/testing/selftests/bpf/progs/test_task_under_cgroup.c @@ -0,0 +1,51 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2023 Bytedance */ + +#include <vmlinux.h> +#include <bpf/bpf_tracing.h> +#include <bpf/bpf_helpers.h> + +#include "bpf_misc.h" + +struct cgroup *bpf_cgroup_from_id(u64 cgid) __ksym; +long bpf_task_under_cgroup(struct task_struct *task, struct cgroup *ancestor) __ksym; +void bpf_cgroup_release(struct cgroup *p) __ksym; +struct task_struct *bpf_task_acquire(struct task_struct *p) __ksym; +void bpf_task_release(struct task_struct *p) __ksym; + +const volatile int local_pid; +const volatile __u64 cgid; +int remote_pid; + +SEC("tp_btf/task_newtask") +int BPF_PROG(handle__task_newtask, struct task_struct *task, u64 clone_flags) +{ + struct cgroup *cgrp = NULL; + struct task_struct *acquired; + + if (local_pid != (bpf_get_current_pid_tgid() >> 32)) + return 0; + + acquired = bpf_task_acquire(task); + if (!acquired) + return 0; + + if (local_pid == acquired->tgid) + goto out; + + cgrp = bpf_cgroup_from_id(cgid); + if (!cgrp) + goto out; + + if (bpf_task_under_cgroup(acquired, cgrp)) + remote_pid = acquired->tgid; + +out: + if (cgrp) + bpf_cgroup_release(cgrp); + bpf_task_release(acquired); + + return 0; +} + +char _license[] SEC("license") = "GPL";