Re: [PATCH bpf-next v8 1/3] bpf: remove extra lock_sock for TCP_ZEROCOPY_RECEIVE

On Thu, Jan 14, 2021 at 8:05 PM Stanislav Fomichev [off-list ref] wrote:

On Thu, Jan 14, 2021 at 7:57 PM Alexei Starovoitov
[off-list ref] wrote:

quoted

On Thu, Jan 14, 2021 at 7:40 PM Stanislav Fomichev [off-list ref] wrote:

quoted

On Thu, Jan 14, 2021 at 7:27 PM Alexei Starovoitov
[off-list ref] wrote:

quoted

On Wed, Jan 13, 2021 at 1:33 PM Stanislav Fomichev [off-list ref] wrote:

quoted

Add custom implementation of getsockopt hook for TCP_ZEROCOPY_RECEIVE.
We skip generic hooks for TCP_ZEROCOPY_RECEIVE and have a custom
call in do_tcp_getsockopt using the on-stack data. This removes
3% overhead for locking/unlocking the socket.

Without this patch:
     3.38%     0.07%  tcp_mmap  [kernel.kallsyms]  [k] __cgroup_bpf_run_filter_getsockopt
            |
             --3.30%--__cgroup_bpf_run_filter_getsockopt
                       |
                        --0.81%--__kmalloc

With the patch applied:
     0.52%     0.12%  tcp_mmap  [kernel.kallsyms]  [k] __cgroup_bpf_run_filter_getsockopt_kern

Signed-off-by: Stanislav Fomichev <redacted>
Cc: Martin KaFai Lau <redacted>
Cc: Song Liu <redacted>
Cc: Eric Dumazet <edumazet@google.com>
Acked-by: Martin KaFai Lau <redacted>

Few issues in this patch and the patch 2 doesn't apply:
Switched to a new branch 'tmp'
Applying: bpf: Remove extra lock_sock for TCP_ZEROCOPY_RECEIVE
.git/rebase-apply/patch:295: trailing whitespace.
#endif
.git/rebase-apply/patch:306: trailing whitespace.
union tcp_word_hdr {
.git/rebase-apply/patch:309: trailing whitespace.
};
.git/rebase-apply/patch:311: trailing whitespace.
#define tcp_flag_word(tp) ( ((union tcp_word_hdr *)(tp))->words [3])
.git/rebase-apply/patch:313: trailing whitespace.
enum {
warning: squelched 1 whitespace error
warning: 6 lines add whitespace errors.
Applying: bpf: Try to avoid kzalloc in cgroup/{s,g}etsockopt
error: patch failed: kernel/bpf/cgroup.c:1390
error: kernel/bpf/cgroup.c: patch does not apply
Patch failed at 0002 bpf: Try to avoid kzalloc in cgroup/{s,g}etsockopt

Sorry, I mentioned in the cover letter that the series requires
4be34f3d0731 ("bpf: Don't leak memory in bpf getsockopt when optlen == 0")
which is only in the bpf tree. No sure when bpf & bpf-next merge.
Or are you trying to apply on top of that?

hmm. It will take a while to wait for the trees to converge.
Ok. I've cherry-picked that bpf commit and applied 3 patches on top,
but the test failed to build:

progs/sockopt_sk.c:60:47: error: use of undeclared identifier
'TCP_ZEROCOPY_RECEIVE'
        if (ctx->level == SOL_TCP && ctx->optname == TCP_ZEROCOPY_RECEIVE) {
                                                     ^
progs/sockopt_sk.c:66:16: error: invalid application of 'sizeof' to an
incomplete type 'struct tcp_zerocopy_receive'
                if (optval + sizeof(struct tcp_zerocopy_receive) > optval_end)

Looks like copied uapi/tcp.h into tools/ wasn't in the include path.

Interesting, let me try to understand where it comes on my system
because it did work even without this uapi/tcp.h so I might
have messed something up. Thank you!

You probably have a newer glibc. Mine is old. I think our CI doesn't
use glibc and
is probably missing the newest tcp.h as well.