aya: Implement RingBuf

This implements the userspace binding for RingBuf.

Instead of streaming the samples as heap buffers, the process_ring
function takes a callback to which we pass the event's byte region,
roughly following [libbpf]'s API design. This avoids a copy and allows
marking the consumer pointer in a timely manner.

[libbpf]: https://github.com/libbpf/libbpf/blob/master/src/ringbuf.c

Co-authored-by: William Findlay <william@williamfindlay.com>

aya/src/bpf.rs

@@ -10,6 +10,7 @@ use std::{
 use aya_obj::{
     btf::{BtfFeatures, BtfRelocationError},
     generated::{BPF_F_SLEEPABLE, BPF_F_XDP_HAS_FRAGS},
+    maps::InvalidMapTypeError,
     relocation::BpfRelocationError,
     BpfSectionKind, Features,
 };
@@ -369,12 +370,16 @@ impl<'a> BpfLoader<'a> {
                 continue;
             }
 
+            let map_type: bpf_map_type =
+                obj.map_type()
+                    .try_into()
+                    .map_err(|InvalidMapTypeError { map_type }| {
+                        BpfError::MapError(MapError::InvalidMapType { map_type })
+                    })?;
             match self.max_entries.get(name.as_str()) {
                 Some(size) => obj.set_max_entries(*size),
                 None => {
-                    if obj.map_type() == BPF_MAP_TYPE_PERF_EVENT_ARRAY as u32
-                        && obj.max_entries() == 0
-                    {
+                    if map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && obj.max_entries() == 0 {
                         obj.set_max_entries(
                             possible_cpus()
                                 .map_err(|error| BpfError::FileError {
@@ -386,6 +391,12 @@ impl<'a> BpfLoader<'a> {
                     }
                 }
             }
+            if map_type == BPF_MAP_TYPE_RINGBUF {
+                obj.set_max_entries(adjust_to_page_size(
+                    crate::util::page_size() as u32,
+                    obj.max_entries(),
+                ))
+            }
             let mut map = MapData {
                 obj,
                 fd: None,
@@ -627,6 +638,7 @@ fn parse_map(data: (String, MapData)) -> Result<(String, Map), BpfError> {
         BPF_MAP_TYPE_PERCPU_HASH => Ok(Map::PerCpuHashMap(map)),
         BPF_MAP_TYPE_LRU_PERCPU_HASH => Ok(Map::PerCpuLruHashMap(map)),
         BPF_MAP_TYPE_PERF_EVENT_ARRAY => Ok(Map::PerfEventArray(map)),
+        BPF_MAP_TYPE_RINGBUF => Ok(Map::RingBuf(map)),
         BPF_MAP_TYPE_SOCKHASH => Ok(Map::SockHash(map)),
         BPF_MAP_TYPE_SOCKMAP => Ok(Map::SockMap(map)),
         BPF_MAP_TYPE_BLOOM_FILTER => Ok(Map::BloomFilter(map)),
@@ -642,6 +654,47 @@ fn parse_map(data: (String, MapData)) -> Result<(String, Map), BpfError> {
     Ok((name, map))
 }
 
+// Adjusts the byte size of a RingBuf map to match a power-of-two multiple of the page size.
+//
+// This mirrors the logic used by libbpf.
+// See https://github.com/libbpf/libbpf/blob/ec6f716eda43fd0f4b865ddcebe0ce8cb56bf445/src/libbpf.c#L2461-L2463
+const fn adjust_to_page_size(page_size: u32, byte_size: u32) -> u32 {
+    if byte_size == 0 {
+        return 0;
+    }
+    let quotient = byte_size / page_size;
+    let remainder = byte_size % page_size;
+    if remainder == 0 && quotient.is_power_of_two() {
+        return byte_size;
+    }
+    let mut pages_needed = quotient;
+    if remainder > 0 {
+        pages_needed += 1
+    }
+    pages_needed.next_power_of_two() * page_size
+}
+
+#[cfg(test)]
+mod tests {
+    use super::adjust_to_page_size;
+
+    #[test]
+    fn test_adjust_to_page_size() {
+        const PAGE_SIZE: u32 = 4096;
+        [
+            (0, 0),
+            (4096, 1),
+            (4096, 4095),
+            (4096, 4096),
+            (8192, 4097),
+            (8192, 8192),
+            (16384, 8193),
+        ]
+        .into_iter()
+        .for_each(|(exp, input)| assert_eq!(exp, adjust_to_page_size(PAGE_SIZE, input)))
+    }
+}
+
 impl<'a> Default for BpfLoader<'a> {
     fn default() -> Self {
         BpfLoader::new()

aya/src/maps/mod.rs

@@ -68,6 +68,7 @@ pub mod hash_map;
 pub mod lpm_trie;
 pub mod perf;
 pub mod queue;
+pub mod ring_buf;
 pub mod sock;
 pub mod stack;
 pub mod stack_trace;
@@ -81,6 +82,7 @@ pub use lpm_trie::LpmTrie;
 pub use perf::AsyncPerfEventArray;
 pub use perf::PerfEventArray;
 pub use queue::Queue;
+pub use ring_buf::RingBuf;
 pub use sock::{SockHash, SockMap};
 pub use stack::Stack;
 pub use stack_trace::StackTraceMap;
@@ -247,6 +249,8 @@ pub enum Map {
     PerCpuLruHashMap(MapData),
     /// A [`PerfEventArray`] map
     PerfEventArray(MapData),
+    /// A [`RingBuf`] map
+    RingBuf(MapData),
     /// A [`SockMap`] map
     SockMap(MapData),
     /// A [`SockHash`] map
@@ -275,6 +279,7 @@ impl Map {
             Map::PerCpuHashMap(map) => map.obj.map_type(),
             Map::PerCpuLruHashMap(map) => map.obj.map_type(),
             Map::PerfEventArray(map) => map.obj.map_type(),
+            Map::RingBuf(map) => map.obj.map_type(),
             Map::SockHash(map) => map.obj.map_type(),
             Map::SockMap(map) => map.obj.map_type(),
             Map::BloomFilter(map) => map.obj.map_type(),
@@ -336,6 +341,7 @@ impl_try_from_map!(
     SockMap from Map::SockMap,
     PerfEventArray from Map::PerfEventArray,
     StackTraceMap from Map::StackTraceMap,
+    RingBuf from Map::RingBuf,
 );
 
 #[cfg(feature = "async")]