Fix ringbuffer ring size reporting

[usamasaqib]

The data pages of a ring buffer are mapped twice in a contiguous virtual region to allow easy wrap around when writing data. The code returns the ring size as the capacity of this mmaped region. However, this is wrong since the kernel double maps the underlying pages.

This PR fixes this by reporting the ring size as half the capacity of the underlying mmap region.

[brycekahle]

I think you need to fix TestRingbufReader to accommodate this change.

[florianl]

Here we mmap() memory, that is later reported via (rr *ringReader) size() int {} and is publicly accessable via (r *Reader) BufferSize() int {}

ebpf/ringbuf/ring.go

Line 27 in c959fd7

prod, err := unix.Mmap(mapFD, (int64)(os.Getpagesize()), os.Getpagesize()+2*size, unix.PROT_READ, unix.MAP_SHARED)

I'm not yet sure about the proposed change. Maybe updating the documentation to (r *Reader) BufferSize() int {} could help, as an alternative?

[brycekahle]

As the person who added BufferSize in #1167, the intent was to return the total amount of data the user could write into the ring buffer. We use this primarily to determine ring buffer utilization in tandem with Remaining.

[usamasaqib]

Here we mmap() memory, that is later reported via (rr *ringReader) size() int {} and is publicly accessable via (r *Reader) BufferSize() int {}

ebpf/ringbuf/ring.go

Line 27 in c959fd7

prod, err := unix.Mmap(mapFD, (int64)(os.Getpagesize()), os.Getpagesize()+2*size, unix.PROT_READ, unix.MAP_SHARED)

I'm not yet sure about the proposed change. Maybe updating the documentation to (r *Reader) BufferSize() int {} could help, as an alternative?

Hey @florianl, in the mmap call we set the size as PAGE_SIZE + 2*size, where size is the max entries specified by the user. The call doubles it because of the unique memory layout of the buffer discussed above. However, the capacity infact is still size.

As Bryce mentioned, BufferSize is used as the total amount of data the buffer can hold, so it should be adjusted according to this.

[lmb]

Thanks all!