-
Notifications
You must be signed in to change notification settings - Fork 290
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
bpf: Handle raw tracepoint arguments #805
base: main
Are you sure you want to change the base?
Conversation
✅ Deploy Preview for aya-rs-docs ready!Built without sensitive environment variables
To edit notification comments on pull requests, go to your Netlify site configuration. |
@vadorovsky, this pull request is now in conflict and requires a rebase. |
299b3cb
to
567ebcf
Compare
567ebcf
to
685c783
Compare
685c783
to
96ae8cd
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM though I don't know much about tracepoint.
Reviewed 9 of 9 files at r1, all commit messages.
Reviewable status: all files reviewed, 7 unresolved discussions (waiting on @alessandrod and @vadorovsky)
bpf/aya-bpf/src/args.rs
line 333 at r1 (raw file):
impl RawTracepointArgs { pub fn new(args: *mut bpf_raw_tracepoint_args) -> Self {
doc comment?
bpf/aya-bpf/src/args.rs
line 344 at r1 (raw file):
/// about the structure of the `bpf_raw_tracepoint_args` type. The tracepoint arguments are /// represented as an array of `__u64` values (i.e., `__u64 args[0]`), and this method /// provides a way to access these arguments conveniently in Rust using `as_slice`.
here and everywhere, can you link to the specific method using [`path::to::as_slice`]? I'm struggling to understand what this is referring to.
EDIT: after some searching I see this is a wrapper in the generated bindings. it would be good to state that
bpf/aya-bpf/src/args.rs
line 346 at r1 (raw file):
/// provides a way to access these arguments conveniently in Rust using `as_slice`. /// /// However, the method does not check the total number of available arguments for a given
can you rephrase this to say that such a check is not possible? this may lead the reader to believe an improvement is possible
test/integration-ebpf/src/raw_tracepoint.rs
line 18 at r1 (raw file):
pub common_type: u16, pub common_flags: u8, _padding: u8,
i think an earlier version of this PR had this in a common crate, is there a reason you went away from that?
test/integration-test/src/tests/raw_tracepoint.rs
line 8 at r1 (raw file):
pub common_type: u16, pub common_flags: u8, _padding: u8,
how come this is needed manually?
test/integration-test/src/tests/raw_tracepoint.rs
line 20 at r1 (raw file):
prog.attach("sys_enter").unwrap(); let map: Array<_, SysEnterEvent> = Array::try_from(bpf.map_mut("RESULT").unwrap()).unwrap();
is the idea that this makes a syscall? how do you know that a syscall will have happened between here and the assertion below?
test/integration-test/src/tests/raw_tracepoint.rs
line 21 at r1 (raw file):
let map: Array<_, SysEnterEvent> = Array::try_from(bpf.map_mut("RESULT").unwrap()).unwrap(); let result = map.get(&0, 0).unwrap();
nit: destructure?
Previously, tamird (Tamir Duberstein) wrote…
There is always some syscall going on in the given moment, but actually, I think I should make it more reliable. I can try writing a PID of the current process into a static variable and then using some syscall here, in the user-space test, then explicitly expect it in the program and then write to a map. |
Previously, tamird (Tamir Duberstein) wrote…
Because BPF verifier is expecting all the memory used to be initialized. When Rust compiler adds extra bytes for alignment, those extra bytes are uninitialized, which often triggers verifier errors. See (ctrl+f "Alignment, padding and verifier errors"): https://deploy-preview-93--aya-rs.netlify.app/book/start/getting-data-to-user-space/#sharing-data |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewable status: all files reviewed, 7 unresolved discussions (waiting on @alessandrod and @vadorovsky)
test/integration-test/src/tests/raw_tracepoint.rs
line 8 at r1 (raw file):
Previously, vadorovsky wrote…
Because BPF verifier is expecting all the memory used to be initialized. When Rust compiler adds extra bytes for alignment, those extra bytes are uninitialized, which often triggers verifier errors. See (ctrl+f "Alignment, padding and verifier errors"):
https://deploy-preview-93--aya-rs.netlify.app/book/start/getting-data-to-user-space/#sharing-data
Ack, can you leave a small comment?
test/integration-test/src/tests/raw_tracepoint.rs
line 20 at r1 (raw file):
Previously, vadorovsky wrote…
There is always some syscall going on in the given moment, but actually, I think I should make it more reliable.
I can try writing a PID of the current process into a static variable and then using some syscall here, in the user-space test, then explicitly expect it in the program and then write to a map.
More reliable seems good.
@vadorovsky, this pull request is now in conflict and requires a rebase. |
96ae8cd
to
5e7787f
Compare
5e7787f
to
b9ad149
Compare
@vadorovsky, this pull request is now in conflict and requires a rebase. |
Provide an `arg()` method in `RawTracepointArgs` wrapper of `bpf_raw_tracepoint_args` and also in `RawTracepointContext`, so it's directly available in raw tracepoint programs. The methods and traits implemented here are unsafe. There is no way to reliably check the number of available arguments, so requesting a non-existing one leads to undefined behavior.
b9ad149
to
2ff8470
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewed 1 of 9 files at r1, 5 of 5 files at r2, 1 of 1 files at r3, 11 of 13 files at r4, all commit messages.
Reviewable status: 11 of 13 files reviewed, 4 unresolved discussions (waiting on @alessandrod and @vadorovsky)
Provide an
arg()
method inRawTracepointArgs
wrapper ofbpf_raw_tracepoint_args
and also inRawTracepointContext
, so it's directly available in raw tracepoint programs.The methods and traits implemented here are unsafe. There is no way to reliably check the number of available arguments, so requesting a non-existing one leads to undefined behavior.
This change is