September 2024 - dev

[PATCH v5 0/4] Add vhost-user support to passt. (part 3)
by Laurent Vivier 19 Sep '24

19 Sep '24

This series of patches adds vhost-user support to passt and then allows passt to connect to QEMU network backend using virtqueue rather than a socket. With QEMU, rather than using to connect: -netdev stream,id=s,server=off,addr.type=unix,addr.path=/tmp/passt_1.socket we will use: -chardev socket,id=chr0,path=/tmp/passt_1.socket -netdev vhost-user,id=netdev0,chardev=chr0 -device virtio-net,netdev=netdev0 -object memory-backend-memfd,id=memfd0,share=on,size=$RAMSIZE -numa node,memdev=memfd0 The memory backend is needed to share data between passt and QEMU. Performance comparison between "-netdev stream" and "-netdev vhost-user": $ iperf3 -c localhost -p 10001 -t 60 -6 -u -b 50G socket: [ 5] 0.00-60.05 sec 95.6 GBytes 13.7 Gbits/sec 0.017 ms 6998988/10132413 (69%) receiver vhost-user: [ 5] 0.00-60.04 sec 237 GBytes 33.9 Gbits/sec 0.006 ms 53673/7813770 (0.69%) receiver $ iperf3 -c localhost -p 10001 -t 60 -4 -u -b 50G socket: [ 5] 0.00-60.05 sec 98.9 GBytes 14.1 Gbits/sec 0.018 ms 6260735/9501832 (66%) receiver vhost-user: [ 5] 0.00-60.05 sec 235 GBytes 33.7 Gbits/sec 0.008 ms 37581/7752699 (0.48%) receiver $ iperf3 -c localhost -p 10001 -t 60 -6 socket: [ 5] 0.00-60.00 sec 17.3 GBytes 2.48 Gbits/sec 0 sender [ 5] 0.00-60.06 sec 17.3 GBytes 2.48 Gbits/sec receiver vhost-user: [ 5] 0.00-60.00 sec 191 GBytes 27.4 Gbits/sec 0 sender [ 5] 0.00-60.05 sec 191 GBytes 27.3 Gbits/sec receiver $ iperf3 -c localhost -p 10001 -t 60 -4 socket: [ 5] 0.00-60.00 sec 15.6 GBytes 2.24 Gbits/sec 0 sender [ 5] 0.00-60.06 sec 15.6 GBytes 2.24 Gbits/sec receiver vhost-user: [ 5] 0.00-60.00 sec 189 GBytes 27.1 Gbits/sec 0 sender [ 5] 0.00-60.04 sec 189 GBytes 27.0 Gbits/sec receiver v5: - rebase on top of 2024_09_06.6b38f07 - rework udp_vu.c as ref.udp.v6 has been removed and we need to know if we receive IPv4 or IPv6 frame when we prepare the guest buffers for recvmsg() - remove vnet->hdrlen as the size is always the same with virtio-net v1 - address comments from David and Stefano v4: - rebase on top of 2024_08_21.1d6142f (rebasing on top of 620e19a1b48a ("udp: Merge udp[46]_mh_recv arrays") introduces a regression in the measure of the latency with UDP because I think I don't replace correctly ref.udp.v6 that is removed by this commit) - Addressed most of the comments from David and Stefano (I didn't want to postpone this version to next week, so I'll address the remaining comments in the next version). v3: - rebase on top of flow table - update tcp_vu.c to look like udp_vu.c (recv()/prepare()/send_frame()) - address comments from Stefano and David on version 2 v2: - remove PATCH 4 - rewrite PATCH 2 and 3 to follow passt coding style - move some code from PATCH 3 to PATCH 4 (previously PATCH 5) - partially addressed David's comment on PATCH 5 Laurent Vivier (4): packet: replace struct desc by struct iovec vhost-user: introduce virtio API vhost-user: introduce vhost-user API vhost-user: add vhost-user Makefile | 6 +- checksum.c | 1 - conf.c | 23 +- epoll_type.h | 4 + iov.c | 1 - isolation.c | 17 +- packet.c | 91 ++-- packet.h | 22 +- passt.1 | 10 +- passt.c | 26 +- passt.h | 6 + pcap.c | 1 - tap.c | 111 ++++- tap.h | 5 +- tcp.c | 31 +- tcp_buf.c | 8 +- tcp_internal.h | 3 +- tcp_vu.c | 647 +++++++++++++++++++++++++ tcp_vu.h | 12 + udp.c | 78 +-- udp.h | 8 +- udp_internal.h | 34 ++ udp_vu.c | 397 +++++++++++++++ udp_vu.h | 13 + util.h | 8 + vhost_user.c | 1247 ++++++++++++++++++++++++++++++++++++++++++++++++ vhost_user.h | 209 ++++++++ virtio.c | 659 +++++++++++++++++++++++++ virtio.h | 184 +++++++ vu_common.c | 36 ++ vu_common.h | 34 ++ 31 files changed, 3792 insertions(+), 140 deletions(-) create mode 100644 tcp_vu.c create mode 100644 tcp_vu.h create mode 100644 udp_internal.h create mode 100644 udp_vu.c create mode 100644 udp_vu.h create mode 100644 vhost_user.c create mode 100644 vhost_user.h create mode 100644 virtio.c create mode 100644 virtio.h create mode 100644 vu_common.c create mode 100644 vu_common.h -- 2.46.0

3 7

[PATCH v3 0/2] Allow UDP and TCP checksum to be disabled
by Laurent Vivier 18 Sep '24

18 Sep '24

With vhost-user we can disable the checksum of UDP and TCP. Add a generic parameter for each of them to disable the checksum. v3: align parameters description with tab s/O/0/ v2: s/UPD/UDP/ Add David's R-b in PATCH 2 Laurent Vivier (2): udp: Allow checksum to be disabled tcp: Allow checksum to be disabled tcp.c | 52 ++++++++++++++++++++++++++------------------ tcp_buf.c | 8 ++++--- tcp_internal.h | 3 ++- udp.c | 58 ++++++++++++++++++++++++++++++++++---------------- 4 files changed, 78 insertions(+), 43 deletions(-) -- 2.46.0

2 3

[PATCH v3 0/2] util: Fix some problems in write_remainder()
by David Gibson 18 Sep '24

18 Sep '24

With Markus Armbruster's help I re-examined the Coverity warning we've had in this function for a while. While I still don't understand some steps in the report it's giving, I did spot some real problems in the vicinity. It turns out fixing those also fixes the Coverity warning, so here we go. Changes since v2: * Handle EINTR within write_all_buf() (suggested by Markus Armbruster) David Gibson (2): util: Add helper to write() all of a buffer util: Remove possible quadratic behaviour from write_remainder() pcap.c | 3 +-- util.c | 50 +++++++++++++++++++++++++++++++++++++++++--------- util.h | 1 + 3 files changed, 43 insertions(+), 11 deletions(-) -- 2.46.0

2 3

[PATCH 0/4] Some small TCP cleanups
by David Gibson 18 Sep '24

18 Sep '24

These are the preliminary cleanups from the EPOLLET rework, which we've decided to shelve for now. Changes since epoll rework series: * Separated from later patches * Tweaked to remove a messy else-after-goto David Gibson (4): tcp: Make some extra functions private tcp: Clean up tcpi_snd_wnd probing tcp: Simplify ifdef logic in tcp_update_seqack_wnd() tcp: Make tcp_update_seqack_wnd()s force_seq parameter explicitly boolean tcp.c | 107 +++++++++++++++++++++++++++++++++---------------- tcp.h | 13 +++--- tcp_buf.c | 12 +++--- tcp_buf.h | 6 +-- tcp_internal.h | 6 +-- 5 files changed, 90 insertions(+), 54 deletions(-) -- 2.46.0

2 5

[PATCH v2 0/2] util: Fix some problems in write_remainder()
by David Gibson 18 Sep '24

18 Sep '24

With Markus Armbruster's help I re-examined the Coverity warning we've had in this function for a while. While I still don't understand some steps in the report it's giving, I did spot some real problems in the vicinity. It turns out fixing those also fixes the Coverity warning, so here we go. David Gibson (2): util: Add helper to write() all of a buffer util: Remove possible quadratic behaviour from write_remainder() pcap.c | 3 +-- util.c | 45 ++++++++++++++++++++++++++++++++++++--------- util.h | 1 + 3 files changed, 38 insertions(+), 11 deletions(-) -- 2.46.0

2 6

[PATCH v2 0/4] mbuto: Make it work on Arch Linux
by Lukasz Gut 18 Sep '24

18 Sep '24

Hi Stefano, This set of patches makes mbuto work also on Arch Linux. I have changed PATCH 4/4 based on your comments. Thank you for pointing out the correct recepient list. Cheers, Lukasz

1 4

[PATCH v2 0/2] Allow UDP and TCP checksum to be disabled
by Laurent Vivier 18 Sep '24

18 Sep '24

With vhost-user we can disable the checksum of UDP and TCP. Add a generic parameter for each of them to disable the checksum. v2: s/UPD/UDP/ Add David's R-b in PATCH 2 Laurent Vivier (2): udp: Allow checksum to be disabled tcp: Allow checksum to be disabled tcp.c | 24 +++++++++++++++++------- tcp_buf.c | 8 +++++--- tcp_internal.h | 3 ++- udp.c | 41 +++++++++++++++++++++++++++++++---------- 4 files changed, 55 insertions(+), 21 deletions(-) -- 2.46.0

3 6

[PATCH v2 00/10] RFC: Clean up TCP epoll mask handling
by David Gibson 18 Sep '24

18 Sep '24

This is a draft series culminating in greatly simplifying the handling of the epoll event mask for TCP sockets. Doing that requires two major preliminaries: 1) We also need to alter event mask handling for the tap interface, to track and report when it's ready to accept more data. We need this to "unstick" connections where we have pending data on the socket, but weren't able to forward it because we ran out of tap interface buffer space. 2) We need to alter handling of ack sequence numbers to the tap device. Without this, the event mask changes expose a fragility where depending on the precise order things are called we could update the ack pointer without actually sending an ack to the tap interface. This is not ready to go yet. For one thing it tanks some of the throughput numbers for reasons I haven't yet diagnosed. It could also certainly do with another set of eyes looking critically over the event logic changes. In addition, the fragility of the ack-to-tap handling remains. The series fixes the specific problem, but it remains non-obviously unsafe to call tcp_update_seqack_from_tap() if a packet isn't going to be immediately and unconditionally sent to the guest. I've been trying to figure out how to make that more robust without introducing additional TCP_INFO calls, but my brain's seizing up a bit at this point. Patches 1..4/10 are preliminary cleanups which should be safe, though. Feel free to apply as many of those ones as you're happy with. David Gibson (10): tcp: Make some extra functions private tcp: Clean up tcpi_snd_wnd probing tcp: Simplify ifdef logic in tcp_update_seqack_wnd() tcp: Make tcp_update_seqack_wnd()s force_seq parameter explicitly boolean tcp: On socket EPOLLOUT, send new ACK to tap immediately tap: Re-introduce EPOLLET for tap connections tap: Keep track of whether there might be space in the tap buffers tcp: Keep track of connections blocked due to a full tap interface tcp: Move deferred handling functions later in tcp.c tcp: Simplify epoll event mask management tap.c | 58 ++++++++--- tap.h | 1 + tcp.c | 265 ++++++++++++++++++++++++++++--------------------- tcp.h | 13 +-- tcp_buf.c | 15 +-- tcp_buf.h | 6 +- tcp_conn.h | 1 + tcp_internal.h | 6 +- 8 files changed, 222 insertions(+), 143 deletions(-) -- 2.46.0

2 14

[PATCH v2 0/2] tcp: unify IPv4 and IPv6 tap queues
by Jon Maloy 18 Sep '24

18 Sep '24

This should save us some memory and code. --- v2: - Setting pointers to pre-set IP and MAC headers on the fly instead of copying them. - Merged patch #2 and #3 from v1 Jon Maloy (2): tcp: set ip and eth headers in l2 tap queues on the fly tcp: unify l2 TCPv4 and TCPv6 queues and structures tcp.c | 6 +- tcp_buf.c | 257 ++++++++++++++++++------------------------------------ tcp_buf.h | 3 +- 3 files changed, 88 insertions(+), 178 deletions(-) -- 2.45.2

3 6

[PATCH 0/2] Allow UDP and TCP checksum to be disabled
by Laurent Vivier 17 Sep '24

17 Sep '24

With vhost-user we can disable the checksum of UDP and TCP. Add a generic parameter for each of them to disable the checksum. Laurent Vivier (2): udp: Allow checksum to be disabled tcp: Allow checksum to be disabled tcp.c | 24 +++++++++++++++++------- tcp_buf.c | 8 +++++--- tcp_internal.h | 3 ++- udp.c | 41 +++++++++++++++++++++++++++++++---------- 4 files changed, 55 insertions(+), 21 deletions(-) -- 2.46.0

2 4

2024

2023

2022

2021

dev September 2024