libbpf-tools/profile: Add support for PID-namespacing #5152
Conversation
ekyooo
requested review from
drzaeus77,
goldshtn,
yonghong-song,
4ast,
brendangregg and
davemarchevsky
as code owners
ekyooo
force-pushed
the
profile_pidns
branch
from
4f415e1 to
b0a1119
Compare
libbpf-tools/profile.c
Outdated
| @@ -595,6 +610,8 @@ int main(int argc, char **argv) | |||
| env.perf_max_stack_depth * sizeof(unsigned long)); | |||
| bpf_map__set_max_entries(obj->maps.stackmap, env.stack_storage_size); | |||
|
|
|||
| set_pidns(obj); | |||
There was a problem hiding this comment.
You can catch the returned value and show a warning message in case PID namespace does not exist?
libbpf-tools/profile.c
Outdated
| if (kernel_at_least(5, 7)) | ||
| return -EPERM; |
There was a problem hiding this comment.
The problem with fixed version check is that they will not work on kernel where a specific patch could have been backported.
Is there a way to rather check for the existence of a given bpf helper?
There was a problem hiding this comment.
In bcc/tools, we have a few cases (including profile) checking kernel version for bpf_get_ns_current_pid_tgid() availability. But for libbpf-tools, I think we can do better.
See trace_helpers.c. There are a few probe_*() functions which tries to detect whether a particular feature (prog_type, map_type) is supported or not. I think we can have another probe function to check whether bpf_get_ns_current_pid_tgid() is supported or not.
There was a problem hiding this comment.
Thank you for good feedback.
I have applied the opinion.
@eiffel-fl
Could you please check if it's what you want?
Thank you.
libbpf-tools/profile.c
Outdated
| @@ -595,6 +631,10 @@ int main(int argc, char **argv) | |||
| env.perf_max_stack_depth * sizeof(unsigned long)); | |||
| bpf_map__set_max_entries(obj->maps.stackmap, env.stack_storage_size); | |||
|
|
|||
| err = set_pidns(obj); | |||
| if (err) | |||
There was a problem hiding this comment.
Let us do 'err && env.verbose' here. We want to print out the below information under verbose mode.
There was a problem hiding this comment.
it's modified. thank you.
libbpf-tools/profile.c
Outdated
| if (kernel_at_least(5, 7)) | ||
| return -EPERM; |
There was a problem hiding this comment.
In bcc/tools, we have a few cases (including profile) checking kernel version for bpf_get_ns_current_pid_tgid() availability. But for libbpf-tools, I think we can do better.
See trace_helpers.c. There are a few probe_*() functions which tries to detect whether a particular feature (prog_type, map_type) is supported or not. I think we can have another probe function to check whether bpf_get_ns_current_pid_tgid() is supported or not.
Add PID translation for nested PID namespace environments, such as container.
…he kernel Support PID namespace mapping only in kernels where bpf_get_ns_current_pid_tgid is available. Add a utility function to check if bpf_get_ns_current_pid_tgid is available in the target kernel. Use this function in the profile to determine support for PID namespace mapping.
| { .code = BPF_JMP | BPF_EXIT }, | ||
| }; | ||
|
|
||
| insn_cnt = sizeof(insns) / sizeof(struct bpf_insn); |
There was a problem hiding this comment.
| insn_cnt = sizeof(insns) / sizeof(struct bpf_insn); | |
| insn_cnt = sizeof(insns) / sizeof(insns[0]); |
This way, you do not have problem if one day we need to change the type.
| struct bpf_pidns_info ns = {}; | ||
|
|
||
| if (use_pidns && !bpf_get_ns_current_pid_tgid(pidns_dev, pidns_ino, &ns, | ||
| sizeof(struct bpf_pidns_info))) { |
There was a problem hiding this comment.
Use ns for sizeof instead of the type name?
| @@ -1248,6 +1248,28 @@ bool probe_ringbuf() | |||
| return true; | |||
| } | |||
|
|
|||
| bool probe_bpf_ns_current_pid_tgid(void) | |||
There was a problem hiding this comment.
I would first add a commit introducing this helper, then using it in another commit in profile.c.
What do you think?
Support PID namespace translation as memleak.py does.