query: Extract tracepoint & k(ret)probe from perf_event links #1213
Conversation
| /// A tracepoint event. | ||
| Tracepoint { | ||
| /// The tracepoint name. | ||
| name: Option<CString>, |
There was a problem hiding this comment.
the other link type that has strings to return is RawTracepointLinkInfo and it returns String but I think it's saner to return a CString/OsString` and let the user decode if needed
Jongy
force-pushed
the
add-link-perf-event
branch
from
1e610b0 to
a451904
Compare
Jongy
force-pushed
the
add-link-perf-event
branch
from
a451904 to
2c37073
Compare
| /// TODO: Add support for `BPF_PERF_EVENT_EVENT`, `BPF_PERF_EVENT_UPROBE` | ||
| /// `BPF_PERF_EVENT_URETPROBE` |
There was a problem hiding this comment.
I don't think we should be using doc comments for TODOs; please add proper documentation instead.
| /// TODO: Add support for `BPF_PERF_EVENT_EVENT`, `BPF_PERF_EVENT_UPROBE` | |
| /// `BPF_PERF_EVENT_URETPROBE` | |
| // TODO: Add support for `BPF_PERF_EVENT_EVENT`, `BPF_PERF_EVENT_UPROBE` | |
| // `BPF_PERF_EVENT_URETPROBE` |
|
|
||
| print!(" {probe_type}"); | ||
| if *addr != 0 { | ||
| print!(" addr={addr:x}"); |
There was a problem hiding this comment.
Should this be
| print!(" addr={addr:x}"); | |
| print!(" addr={addr:#x}"); |
?
| // Handle two-phase call for perf event string data if needed (this mimics the | ||
| // behavior of bpftool). |
There was a problem hiding this comment.
Nit: Comment (and naming of need_second_call) seems a bit bogus: there is no first call in scope. It could conceptually have been called from anywhere.
There was a problem hiding this comment.
(the first call is from the gen_info_impl macro, but I'll rename to something clearer :) )
| .cookie | ||
| }; | ||
| let name = (tp_name != 0).then(|| unsafe { | ||
| CStr::from_ptr(tp_name as *const c_char).to_owned() |
There was a problem hiding this comment.
Isn't the typical contract that the kernel would set the actual name length? Or is this not the case here?
There was a problem hiding this comment.
It actually has the length - but I followed bpftool here, which just prints the string. If I have to guess - the returned length is the full one and the provided name is the null-terminated string (truncated), but i'll make sure and comment.
I'll try. There are no tests for link infos from |
heavily inspired by matching bpftool code