| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Tue, 13 Oct 2020 14:56:12 -0700
From: Andrii Nakryiko <andrii.nakryiko@...il.com>
To: Alexei Starovoitov <alexei.starovoitov@...il.com>
Cc: Jiri Olsa <jolsa@...hat.com>, Andrii Nakryiko <andrii@...nel.org>,
"David S. Miller" <davem@...emloft.net>,
Daniel Borkmann <daniel@...earbox.net>,
John Fastabend <john.fastabend@...il.com>,
Network Development <netdev@...r.kernel.org>,
bpf <bpf@...r.kernel.org>, Kernel Team <kernel-team@...com>
Subject: Re: [PATCH v2 bpf-next 3/4] selftests/bpf: Add profiler test
On Tue, Oct 13, 2020 at 2:03 PM Alexei Starovoitov
<alexei.starovoitov@...il.com> wrote:
>
> On Tue, Oct 13, 2020 at 12:56 PM Jiri Olsa <jolsa@...hat.com> wrote:
> >
> > On Thu, Oct 08, 2020 at 06:12:39PM -0700, Alexei Starovoitov wrote:
> >
> > SNIP
> >
> > > +
> > > +#ifdef UNROLL
> > > +#pragma unroll
> > > +#endif
> > > + for (int i = 0; i < MAX_CGROUPS_PATH_DEPTH; i++) {
> > > + filepart_length =
> > > + bpf_probe_read_str(payload, MAX_PATH, BPF_CORE_READ(cgroup_node, name));
> > > + if (!cgroup_node)
> > > + return payload;
> > > + if (cgroup_node == cgroup_root_node)
> > > + *root_pos = payload - payload_start;
> > > + if (filepart_length <= MAX_PATH) {
> > > + barrier_var(filepart_length);
> > > + payload += filepart_length;
> > > + }
> > > + cgroup_node = BPF_CORE_READ(cgroup_node, parent);
> > > + }
> > > + return payload;
> > > +}
> > > +
> > > +static ino_t get_inode_from_kernfs(struct kernfs_node* node)
> > > +{
> > > + struct kernfs_node___52* node52 = (void*)node;
> > > +
> > > + if (bpf_core_field_exists(node52->id.ino)) {
> > > + barrier_var(node52);
> > > + return BPF_CORE_READ(node52, id.ino);
> > > + } else {
> > > + barrier_var(node);
> > > + return (u64)BPF_CORE_READ(node, id);
> > > + }
> > > +}
> > > +
> > > +int pids_cgrp_id = 1;
> >
> >
> > hi,
> > I'm getting compilation failure with this:
> >
> > CLNG-LLC [test_maps] profiler2.o
> > In file included from progs/profiler2.c:6:
> > progs/profiler.inc.h:246:5: error: redefinition of 'pids_cgrp_id' as different kind of symbol
> > int pids_cgrp_id = 1;
> > ^
> > /home/jolsa/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:14531:2: note: previous definition is here
> > pids_cgrp_id = 11,
>
> Interesting.
> You probably have CONFIG_CGROUP_PIDS in your .config?
> I don't and bpf CI doesn't have it either, so this issue wasn't spotted earlier.
>
> I can hard code 11, of course, but
> but maybe Andrii has a cool way to use co-re to deal with this?
Cool or not, but we do have a way to deal with it. :)
> I think
> "extern bool CONFIG_CGROUP_PIDS __kconfig"
> won't work.
> A good opportunity to try to use bpf_core_enum_value_exists() ?
There are fews parts here.
First, we can't rely that vmlinux.h has pids_cgrp_id enum defined, so
we need to define our own. But in such a way that it doesn't collide
with enum cgroup_subsys_id in vmlinux.h. ___suffix for the rescue
here:
enum cgroup_subsys_id___local {
pids_cgrp_id___local = 1, /* anything but zero */
};
If value is zero, built-in will complain. Need to check with Yonghong
on why that is. For all things CO-RE, ___local suffix is going to be
ignored.
Second, detecting if the kernel even has this pids_cgrp_id defined.
That could be done with __kconfig as you mentioned, actually (but need
__weak as well). But one can also use bpf_core_enum_value_exists as
well. So it's either:
extern bool CONFIG_CGROUP_PIDS __kconfig __weak;
...
if (ENABLE_CGROUP_V1_RESOLVER && CONFIG_CGROUP_PIDS) {
...
}
Or a bit more verbose way with relos:
if (ENABLE_CGROUP_V1_RESOLVER &&
bpf_core_enum_value_exists(enum cgroup_subsys_id___local,
pids_cgrp_id___local)) {
...
}
Third, actually getting the value of enum. This one would be
impossible without CO-RE reloc, but that's exactly what
bpf_core_enum_value() exists for:
int cgrp_id = bpf_core_enum_value(enum cgroup_subsys_id___local,
pids_cgrp_id___local);
I'd go with Kconfig + bpf_core_enum_value(), as it's shorter and
nicer. This compiles and works with my Kconfig, but I haven't checked
with CONFIG_CGROUP_PIDS defined.
diff --git a/tools/testing/selftests/bpf/progs/profiler.inc.h
b/tools/testing/selftests/bpf/progs/profiler.inc.h
index 00578311a423..79b8d2860a5c 100644
--- a/tools/testing/selftests/bpf/progs/profiler.inc.h
+++ b/tools/testing/selftests/bpf/progs/profiler.inc.h
@@ -243,7 +243,11 @@ static ino_t get_inode_from_kernfs(struct
kernfs_node* node)
}
}
-int pids_cgrp_id = 1;
+extern bool CONFIG_CGROUP_PIDS __kconfig __weak;
+
+enum cgroup_subsys_id___local {
+ pids_cgrp_id___local = 1, /* anything but zero */
+};
static INLINE void* populate_cgroup_info(struct cgroup_data_t* cgroup_data,
struct task_struct* task,
@@ -253,7 +257,9 @@ static INLINE void* populate_cgroup_info(struct
cgroup_data_t* cgroup_data,
BPF_CORE_READ(task, nsproxy, cgroup_ns, root_cset,
dfl_cgrp, kn);
struct kernfs_node* proc_kernfs = BPF_CORE_READ(task, cgroups,
dfl_cgrp, kn);
- if (ENABLE_CGROUP_V1_RESOLVER) {
+ if (ENABLE_CGROUP_V1_RESOLVER && CONFIG_CGROUP_PIDS) {
+ int cgrp_id = bpf_core_enum_value(enum
cgroup_subsys_id___local, pids_cgrp_id___local);
+
#ifdef UNROLL
#pragma unroll
#endif
@@ -262,7 +268,7 @@ static INLINE void* populate_cgroup_info(struct
cgroup_data_t* cgroup_data,
BPF_CORE_READ(task, cgroups, subsys[i]);
if (subsys != NULL) {
int subsys_id = BPF_CORE_READ(subsys, ss, id);
- if (subsys_id == pids_cgrp_id) {
+ if (subsys_id == cgrp_id) {
proc_kernfs =
BPF_CORE_READ(subsys, cgroup, kn);
root_kernfs =
BPF_CORE_READ(subsys, ss, root, kf_root, kn);
break;
Powered by blists - more mailing lists