Timo Beckers

bpf: grow verifier log buffer to 10MiB to prevent truncation

1

Just having seen a 4,6MB log from a bpf_host complexity issue, I started work on automatically growing the log buffer instead of hardcoding a size here, but hit a small...

init.sh: migrate alignchecker to use BTF instead of DWARF

22

With `cilium/ebpf`'s btf/ package now being public, there is no longer a reason to compile datapath ELFs with DWARF data and parsing it in userspace. We're compiling ELFs with `-g`...

With the recent addition of verifier error summaries, there are fewer reasons for disabling verifier output altogether. The existing behaviour is somewhat underspecified and underdocumented. Logging behaviour hinges on both...

errors: remove verifier truncation heuristic, rely on ENOSPC

1

Was getting strange results re-running prog loads with progressively larger buffers to fit verifier logs of large Cilium programs. The verifier always writes a NUL byte to the last byte...

Assemble a dummy BTF blob for probing StructOps maps

9

With https://github.com/cilium/ebpf/pull/321, an API for probing available map types in the kernel was added. However, a StructOps map requires a valid BTF blob to be specified in order to make...

sys: Implement fd metadata and leak tracing instrumentation

2

``` In an effort to improve the library's fd hygiene, this patch adds optional metadata to sys.FD to hold a name and a stack trace of where the fd was...

Kernel 5.16 saw the introduction of the bloom filter map type: https://lore.kernel.org/all/[email protected]/. This also included new `map_extra` field for maps that need more configuration than the existing map properties. In...

Over the past few months, while working on per-instruction metadata, we've identified the need for marshaling valid BTF blobs based on `btf.Type`s. A few use cases: - caller-provided `btf.Type`s for...

`ProgramInfo`'s `MapIDs()` bool returns false negatives

2

**Describe the bug** Currently, the `(*ProgramInfo).MapIDs()`'s bool return value is simply determined by `pi.ids != nil`, `pi.ids` being allocated when `NrMapIds > 0`. The `NrMapIds` field will always be zero...

Enable cleanup of tracefs entries (kprobe_events)

6

As raised by @etherandrius: https://cilium.slack.com/archives/CKC55JL8L/p1618310807065500. Currently, the group prefix for entries in `kprobe_events` is [hardcoded to `ebpf_`](https://github.com/cilium/ebpf/blob/dccbd5364e5b55091d649c1d2a1de0e1b2ff5522/link/kprobe.go#L166), resulting in identifiers like `ebpf_aabbccdd`. This might prove inflexible in some environments where...