On 9/4/25 10:28 AM, Amery Hung wrote:
On Fri, Aug 29, 2025 at 11:22 AM Martin KaFai Lau <martin.lau@xxxxxxxxx> wrote:
On 8/28/25 6:39 AM, Nimrod Oren wrote:
I'm currently working on a series that converts the xdp_native program
to use dynptr for accessing header data. If accepted, it should provide
better performance, since dynptr can access without copying the data.
The bpf_xdp_adjust_tail is aware of xdp_buff_has_frags. Is there a reason that
bpf_xdp_adjust_head cannot handle frags also?
I am not aware of reasons that would stop this.
Are you suggesting another way to pop headers? E.g., use
bpf_xdp_adjust_head() to shrink the first frag from the front and call
bpf_xdp_store_bytes() to move the remaining headers
bpf_xdp_pull_data is useful on its own, nothing change there. On top of that,
bpf_xdp_adjust_head() should be useful also for bpf prog that does not care
about the linear/frag layout and stay with the bpf_dynptr helpers (or the
bpf_xdp_{store,load}_bytes you mentioned). Also, if I read it correctly,
bpf_xdp_adjust_head can increase the head of multi buf xdp but not shrinking it
while the bpf_xdp_adjust_tail can do both. It could be a surprise to use. The
adjust_head support can be a followup though. I think some of your work in this
series is pretty close to having adjust_head support also, so I was wondering if
there is reason that adjust_head cannot be supported.
