| Message ID | 20200909181556.2945496-1-ncardwell@google.com |
|---|---|
| Headers | show |
| Series | tcp: increase flexibility of EBPF congestion control initialization | expand |
On Wed, Sep 09, 2020 at 02:15:52PM -0400, Neal Cardwell wrote: > This patch series reorganizes TCP congestion control initialization so that if > EBPF code called by tcp_init_transfer() sets the congestion control algorithm > by calling setsockopt(TCP_CONGESTION) then the TCP stack initializes the > congestion control module immediately, instead of having tcp_init_transfer() > later initialize the congestion control module. > > This increases flexibility for the EBPF code that runs at connection > establishment time, and simplifies the code. > > This has the following benefits: > > (1) This allows CC module customizations made by the EBPF called in > tcp_init_transfer() to persist, and not be wiped out by a later > call to tcp_init_congestion_control() in tcp_init_transfer(). > > (2) Does not flip the order of EBPF and CC init, to avoid causing bugs > for existing code upstream that depends on the current order. > > (3) Does not cause 2 initializations for for CC in the case where the > EBPF called in tcp_init_transfer() wants to set the CC to a new CC > algorithm. > > (4) Allows follow-on simplifications to the code in net/core/filter.c > and net/ipv4/tcp_cong.c, which currently both have some complexity > to special-case CC initialization to avoid double CC > initialization if EBPF sets the CC. Thanks for this work. Only have one nit in patch 3 for consideration. Acked-by: Martin KaFai Lau <kafai@fb.com>
On Wed, Sep 9, 2020 at 8:36 PM Martin KaFai Lau <kafai@fb.com> wrote: > > On Wed, Sep 09, 2020 at 02:15:52PM -0400, Neal Cardwell wrote: > > This patch series reorganizes TCP congestion control initialization so that if > > EBPF code called by tcp_init_transfer() sets the congestion control algorithm > > by calling setsockopt(TCP_CONGESTION) then the TCP stack initializes the > > congestion control module immediately, instead of having tcp_init_transfer() > > later initialize the congestion control module. > > > > This increases flexibility for the EBPF code that runs at connection > > establishment time, and simplifies the code. > > > > This has the following benefits: > > > > (1) This allows CC module customizations made by the EBPF called in > > tcp_init_transfer() to persist, and not be wiped out by a later > > call to tcp_init_congestion_control() in tcp_init_transfer(). > > > > (2) Does not flip the order of EBPF and CC init, to avoid causing bugs > > for existing code upstream that depends on the current order. > > > > (3) Does not cause 2 initializations for for CC in the case where the > > EBPF called in tcp_init_transfer() wants to set the CC to a new CC > > algorithm. > > > > (4) Allows follow-on simplifications to the code in net/core/filter.c > > and net/ipv4/tcp_cong.c, which currently both have some complexity > > to special-case CC initialization to avoid double CC > > initialization if EBPF sets the CC. > Thanks for this work. Only have one nit in patch 3 for consideration. > > Acked-by: Martin KaFai Lau <kafai@fb.com> Thanks for the review! I like your suggestion in patch 3 to further simplify the code. Do you mind submitting your idea for a follow-on clean-up/refactor as a separate follow-on commit? thanks, neal
On Wed, Sep 09, 2020 at 11:21:50PM -0400, Neal Cardwell wrote: > On Wed, Sep 9, 2020 at 8:36 PM Martin KaFai Lau <kafai@fb.com> wrote: > > > > On Wed, Sep 09, 2020 at 02:15:52PM -0400, Neal Cardwell wrote: > > > This patch series reorganizes TCP congestion control initialization so that if > > > EBPF code called by tcp_init_transfer() sets the congestion control algorithm > > > by calling setsockopt(TCP_CONGESTION) then the TCP stack initializes the > > > congestion control module immediately, instead of having tcp_init_transfer() > > > later initialize the congestion control module. > > > > > > This increases flexibility for the EBPF code that runs at connection > > > establishment time, and simplifies the code. > > > > > > This has the following benefits: > > > > > > (1) This allows CC module customizations made by the EBPF called in > > > tcp_init_transfer() to persist, and not be wiped out by a later > > > call to tcp_init_congestion_control() in tcp_init_transfer(). > > > > > > (2) Does not flip the order of EBPF and CC init, to avoid causing bugs > > > for existing code upstream that depends on the current order. > > > > > > (3) Does not cause 2 initializations for for CC in the case where the > > > EBPF called in tcp_init_transfer() wants to set the CC to a new CC > > > algorithm. > > > > > > (4) Allows follow-on simplifications to the code in net/core/filter.c > > > and net/ipv4/tcp_cong.c, which currently both have some complexity > > > to special-case CC initialization to avoid double CC > > > initialization if EBPF sets the CC. > > Thanks for this work. Only have one nit in patch 3 for consideration. > > > > Acked-by: Martin KaFai Lau <kafai@fb.com> > > Thanks for the review! I like your suggestion in patch 3 to further > simplify the code. Do you mind submitting your idea for a follow-on > clean-up/refactor as a separate follow-on commit? Sure. will do.
On Wed, Sep 9, 2020 at 8:24 PM Neal Cardwell <ncardwell@google.com> wrote: > > On Wed, Sep 9, 2020 at 8:36 PM Martin KaFai Lau <kafai@fb.com> wrote: > > > > On Wed, Sep 09, 2020 at 02:15:52PM -0400, Neal Cardwell wrote: > > > This patch series reorganizes TCP congestion control initialization so that if > > > EBPF code called by tcp_init_transfer() sets the congestion control algorithm > > > by calling setsockopt(TCP_CONGESTION) then the TCP stack initializes the > > > congestion control module immediately, instead of having tcp_init_transfer() > > > later initialize the congestion control module. > > > > > > This increases flexibility for the EBPF code that runs at connection > > > establishment time, and simplifies the code. > > > > > > This has the following benefits: > > > > > > (1) This allows CC module customizations made by the EBPF called in > > > tcp_init_transfer() to persist, and not be wiped out by a later > > > call to tcp_init_congestion_control() in tcp_init_transfer(). > > > > > > (2) Does not flip the order of EBPF and CC init, to avoid causing bugs > > > for existing code upstream that depends on the current order. > > > > > > (3) Does not cause 2 initializations for for CC in the case where the > > > EBPF called in tcp_init_transfer() wants to set the CC to a new CC > > > algorithm. > > > > > > (4) Allows follow-on simplifications to the code in net/core/filter.c > > > and net/ipv4/tcp_cong.c, which currently both have some complexity > > > to special-case CC initialization to avoid double CC > > > initialization if EBPF sets the CC. > > Thanks for this work. Only have one nit in patch 3 for consideration. > > > > Acked-by: Martin KaFai Lau <kafai@fb.com> > > Thanks for the review! I like your suggestion in patch 3 to further > simplify the code. Do you mind submitting your idea for a follow-on > clean-up/refactor as a separate follow-on commit? I think it's better to be folded into this set. It needs rebase to bpf-next anyway.
OK, thanks everyone. I will rebase the series onto bpf-next and include a patch with Martin's suggested follow-on cleanup. Will resubmit ASAP. thanks, neal On Thu, Sep 10, 2020 at 1:31 AM Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote: > > On Wed, Sep 9, 2020 at 8:24 PM Neal Cardwell <ncardwell@google.com> wrote: > > > > On Wed, Sep 9, 2020 at 8:36 PM Martin KaFai Lau <kafai@fb.com> wrote: > > > > > > On Wed, Sep 09, 2020 at 02:15:52PM -0400, Neal Cardwell wrote: > > > > This patch series reorganizes TCP congestion control initialization so that if > > > > EBPF code called by tcp_init_transfer() sets the congestion control algorithm > > > > by calling setsockopt(TCP_CONGESTION) then the TCP stack initializes the > > > > congestion control module immediately, instead of having tcp_init_transfer() > > > > later initialize the congestion control module. > > > > > > > > This increases flexibility for the EBPF code that runs at connection > > > > establishment time, and simplifies the code. > > > > > > > > This has the following benefits: > > > > > > > > (1) This allows CC module customizations made by the EBPF called in > > > > tcp_init_transfer() to persist, and not be wiped out by a later > > > > call to tcp_init_congestion_control() in tcp_init_transfer(). > > > > > > > > (2) Does not flip the order of EBPF and CC init, to avoid causing bugs > > > > for existing code upstream that depends on the current order. > > > > > > > > (3) Does not cause 2 initializations for for CC in the case where the > > > > EBPF called in tcp_init_transfer() wants to set the CC to a new CC > > > > algorithm. > > > > > > > > (4) Allows follow-on simplifications to the code in net/core/filter.c > > > > and net/ipv4/tcp_cong.c, which currently both have some complexity > > > > to special-case CC initialization to avoid double CC > > > > initialization if EBPF sets the CC. > > > Thanks for this work. Only have one nit in patch 3 for consideration. > > > > > > Acked-by: Martin KaFai Lau <kafai@fb.com> > > > > Thanks for the review! I like your suggestion in patch 3 to further > > simplify the code. Do you mind submitting your idea for a follow-on > > clean-up/refactor as a separate follow-on commit? > > I think it's better to be folded into this set. > It needs rebase to bpf-next anyway.
This patch series reorganizes TCP congestion control initialization so that if EBPF code called by tcp_init_transfer() sets the congestion control algorithm by calling setsockopt(TCP_CONGESTION) then the TCP stack initializes the congestion control module immediately, instead of having tcp_init_transfer() later initialize the congestion control module. This increases flexibility for the EBPF code that runs at connection establishment time, and simplifies the code. This has the following benefits: (1) This allows CC module customizations made by the EBPF called in tcp_init_transfer() to persist, and not be wiped out by a later call to tcp_init_congestion_control() in tcp_init_transfer(). (2) Does not flip the order of EBPF and CC init, to avoid causing bugs for existing code upstream that depends on the current order. (3) Does not cause 2 initializations for for CC in the case where the EBPF called in tcp_init_transfer() wants to set the CC to a new CC algorithm. (4) Allows follow-on simplifications to the code in net/core/filter.c and net/ipv4/tcp_cong.c, which currently both have some complexity to special-case CC initialization to avoid double CC initialization if EBPF sets the CC. Neal Cardwell (4): tcp: only init congestion control if not initialized already tcp: simplify EBPF TCP_CONGESTION to always init CC tcp: simplify tcp_set_congestion_control(): always reinitialize tcp: simplify _bpf_setsockopt(): remove flags argument include/net/inet_connection_sock.h | 3 ++- include/net/tcp.h | 2 +- net/core/filter.c | 18 ++++-------------- net/ipv4/tcp.c | 3 ++- net/ipv4/tcp_cong.c | 14 ++++---------- net/ipv4/tcp_input.c | 4 +++- 6 files changed, 16 insertions(+), 28 deletions(-)