November 2024 - dev

[PATCH 0/6] ndp: Unsolicited RAs
by David Gibson 12 Nov '24

12 Nov '24

The default route a guest obtains via NDP has an expiry. We set this to the maximum allowed, 65535s (about 18 hours). We missed, however, that after that expiry, the guest won't send a new Router Solicitation. Instead it expects an unsolicited Router Advertisement to have come at some point before the expiry. This means that on an IPv6 setup, the default will disappear after 18 hours. Correct this by sending unsolicited Router Advertisements as required by RFC 4861. Along the way we make a number of small cleanups to the NDP code. Link: https://github.com/kubevirt/kubevirt/issues/13191 David Gibson (6): ndp: Remove redundant update to addr_seen ndp: Add ndp_send() helper ndp: Split out helpers for sending specific NDP message types ndp: Use struct assignment in preference to memcpy() for IPv6 addresses ndp: Make route lifetime a #define ndp: Send unsolicited Router Advertisements ip.h | 9 +++ ndp.c | 210 ++++++++++++++++++++++++++++++++++++-------------------- ndp.h | 7 +- passt.c | 3 + 4 files changed, 152 insertions(+), 77 deletions(-) -- 2.47.0

1 6

[PATCH v10 00/12] Add vhost-user support to passt. (part 3)
by Laurent Vivier 11 Nov '24

11 Nov '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 v10: - rebase v9 on top of my changes - remove last 4 patches from v9 as 'tcp: Pass TCP header and payload separately to tcp_fill_headers[46]()' introduces a regression in "make check" for me. - addressed comments from David - I tried to cleanup iov management, but I was not able to remove the update of the first iov to point to the header or to the data - upd_vu_hdrlen()/tcp_vu_hdrlen() includes now the size of the virtio-net header - I didn't address comments from Stefano. - I didn't fix the bug with seek_offset_cap v9: [David Gibson] - Rebased on current main - Conflicts with v4/v6 buffer merge addressed - Conflicts with TCP options construction rework addressed - Added several cleanup patches on top - Added a number of IOV and buffer cleanups on top - The aim is that these should allow more sharing of logic between the vhost-user and non-vhost-user pathes, although they've only minimally accomplished that so far. v8: - remove iov_size() from vu_collect_one_frame() - move vu_packet_check_range() to vu_common.c - fix UDP when dlen is 0. v7: - rebase - use vu_collect_one_frame() to do vu_collect() (collect multiple frame) - add vhost-user tests from Stefano v6: - rebase - extract 3 patches from "vhost-user: add vhost-user": passt: rename tap_sock_init() to tap_backend_init() tcp: Export headers functions udp: Prepare udp.c to be shared with vhost-user - introduce new functions vu_collect_one_frame(), vu_collect(), vu_set_vnethdr(), vu_flush(), vu_send_single() to be called from tcp_vu.c, udp_vu.c and ICMP/DHCP where vhost-user code was duplicated. 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 David Gibson (4): tcp: Use only netinet/tcp.h instead of linux/tcp.h tcp_vu: Share more header construction between IPv4 and IPv6 paths tcp: Move tcp_l2_buf_fill_headers() to tcp_buf.c tcp: Adjust iov_len before filling headers Laurent Vivier (7): packet: replace struct desc by struct iovec vhost-user: introduce virtio API vhost-user: introduce vhost-user API udp: Prepare udp.c to be shared with vhost-user tcp: Export headers functions passt: rename tap_sock_init() to tap_backend_init() vhost-user: add vhost-user Stefano Brivio (1): test: Add tests for passt in vhost-user mode Makefile | 9 +- conf.c | 21 +- epoll_type.h | 4 + iov.c | 1 - isolation.c | 17 +- packet.c | 91 ++-- packet.h | 22 +- passt.1 | 10 +- passt.c | 11 +- passt.h | 7 + pcap.c | 1 - tap.c | 129 ++++-- tap.h | 7 +- tcp.c | 76 +--- tcp_buf.c | 41 +- tcp_internal.h | 19 +- tcp_vu.c | 498 +++++++++++++++++++++ tcp_vu.h | 12 + test/lib/perf_report | 15 + test/lib/setup | 77 +++- test/lib/setup_ugly | 2 +- test/passt_vu | 1 + test/passt_vu_in_ns | 1 + test/perf/passt_vu_tcp | 211 +++++++++ test/perf/passt_vu_udp | 159 +++++++ test/run | 25 ++ test/two_guests_vu | 1 + udp.c | 85 ++-- udp_internal.h | 34 ++ udp_vu.c | 336 ++++++++++++++ udp_vu.h | 13 + util.h | 8 + vhost_user.c | 981 +++++++++++++++++++++++++++++++++++++++++ vhost_user.h | 206 +++++++++ virtio.c | 660 +++++++++++++++++++++++++++ virtio.h | 184 ++++++++ vu_common.c | 285 ++++++++++++ vu_common.h | 60 +++ 38 files changed, 4116 insertions(+), 204 deletions(-) create mode 100644 tcp_vu.c create mode 100644 tcp_vu.h create mode 120000 test/passt_vu create mode 120000 test/passt_vu_in_ns create mode 100644 test/perf/passt_vu_tcp create mode 100644 test/perf/passt_vu_udp create mode 120000 test/two_guests_vu 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 -- 2.47.0

3 16

[PATCH 0/2] test: Fixes for prefixlen handling
by David Gibson 08 Nov '24

08 Nov '24

Jon hit some problems when attempting to run the testsuite with an IPv6 prefix length of 40, rather than 64. Here are a couple of patches that should fix that, along with a couple of adjacent oddities. David Gibson (2): test: Don't require 64-bit prefixes in perf tests test: Improve test for NDP assigned prefix test/passt/ndp | 4 ++-- test/pasta/ndp | 4 ++-- test/perf/pasta_tcp | 2 +- test/perf/pasta_udp | 2 +- 4 files changed, 6 insertions(+), 6 deletions(-) -- 2.47.0

2 4

[PATCH v2 0/4] Avoid running cppcheck on system headers
by David Gibson 08 Nov '24

08 Nov '24

It turns out cppcheck has inbuilt knowledge of the C library, and isn't typically given the system headers. Avoiding doing so reduces the runtime to less than half of what it currently is. For non-obvious reasons, this change also exposes some new warnings. Some are real, one is a cppcheck bug. Fix and/or workaround these then make the change to the cppcheck options. This is based on my earlier series with clangd configuration and fixes. v2: * Dropped patches already merged * Rebased on Stefano's series of lint fixes * Add a missing #include in 1/4 and 2/4 * Adjust 3/4 not to die if close_range() is unavailable David Gibson (4): log: Only check for FALLOC_FL_COLLAPSE_RANGE availability at runtime linux_dep: Move close_range() conditional handling to linux_dep.h linux_dep: Fix CLOSE_RANGE_UNSHARE availability handling cppcheck: Don't check the system headers Makefile | 15 +-------------- linux_dep.h | 22 ++++++++++++++++++++++ log.c | 10 ++-------- util.c | 17 +++++++++++++++-- util.h | 19 ------------------- 5 files changed, 40 insertions(+), 43 deletions(-) -- 2.47.0

2 5

[PATCH v2 0/1] RFC: IOV tail helpers
by David Gibson 08 Nov '24

08 Nov '24

As mentioned here's my draft patch adding a helper to pull headers off an IOV, checking for contiguity, alignment and so forth. It's based around the idea of an "iov tail", that is an iov from which you've already consumed (in some sense) some data from the beginning. It uses an explicit offset to track this, so we don't have to copy an entire iovec array in order to trim a bit off it. But, it's built so that if you pull enough off the front to remove some entire buffers, it won't repeatedly have to step through them working out later offsets. It may be somewhat overengineered, let me know what you think of the interface. v2: * Substantial changes to comments and names, to make the purpose clearer David Gibson (1): iov: iov tail helpers iov.c | 93 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ iov.h | 76 ++++++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 169 insertions(+) -- 2.47.0

2 2

[PATCH 0/1] RFC: IOV tail helpers
by David Gibson 08 Nov '24

08 Nov '24

As mentioned here's my draft patch adding a helper to pull headers off an IOV, checking for contiguity, alignment and so forth. It's based around the idea of an "iov tail", that is an iov from which you've already consumed (in some sense) some data from the beginning. It uses an explicit offset to track this, so we don't have to copy an entire iovec array in order to trim a bit off it. But, it's built so that if you pull enough off the front to remove some entire buffers, it won't repeatedly have to step through them working out later offsets. It may be somewhat overengineered, let me know what you think of the interface. David Gibson (1): iov: iov tail helpers iov.c | 83 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ iov.h | 24 +++++++++++++++++ 2 files changed, 107 insertions(+) -- 2.47.0

2 5

[PATCH 0/2] Small fixes for nstool
by David Gibson 08 Nov '24

08 Nov '24

While working on the exeter tests, I noticed a couple of things missing in nstool. These make sense standalone, even if we don't have an urgent need for them without the exeter tests. David Gibson (2): test: Rename propagating signal handler test: Make nstool hold robust against interruptions to control clients test/nstool.c | 40 ++++++++++++++++++++++++++++------------ 1 file changed, 28 insertions(+), 12 deletions(-) -- 2.47.0

2 6

[PATCH 0/6] Fix some more static checker warnings
by Stefano Brivio 08 Nov '24

08 Nov '24

Stefano Brivio (6): dhcpv6: Use for loop instead of goto to avoid false positive cppcheck warning dhcpv6: Turn some option headers pointers to const tap: Cast TAP_BUF_BYTES - ETH_MAX_MTU to ssize_t, not TAP_BUF_BYTES util: Define small and big thresholds for socket buffers as unsigned long long passt: Use NOLINT clang-tidy block instead of NOLINTNEXTLINE tap, tcp, util: Add some missing SOCK_CLOEXEC flags dhcpv6.c | 51 +++++++++++++++++++++++---------------------------- passt.c | 3 ++- tap.c | 7 ++++--- tcp.c | 2 +- util.c | 3 ++- util.h | 9 ++++++--- 6 files changed, 38 insertions(+), 37 deletions(-) -- 2.43.0

2 12

[PATCH 0/8] Avoid running cppcheck on system headers
by David Gibson 08 Nov '24

08 Nov '24

It turns out cppcheck has inbuilt knowledge of the C library, and isn't typically given the system headers. Avoiding doing so reduces the runtime to less than half of what it currently is. For non-obvious reasons, this change also exposes some new warnings. Some are real, one is a cppcheck bug. Fix and/or workaround these then make the change to the cppcheck options. This is based on my earlier series with clangd configuration and fixes. David Gibson (8): linux_dep: Generalise tcp_info.h to handling Linux extension compatibility log: Only check for FALLOC_FL_COLLAPSE_RANGE availability at runtime linux_dep: Move close_range() conditional handling to linux_dep.h linux_dep: Fix CLOSE_RANGE_UNSHARE availability handling ndp: Use const pointer for ndp_ns packet udp: Don't dereference uflow before NULL check in udp_reply_sock_handler() util: Work around cppcheck bug 6936 cppcheck: Don't check the system headers Makefile | 17 ++--------------- tcp_info.h => linux_dep.h | 32 ++++++++++++++++++++++++++++---- log.c | 9 +-------- ndp.c | 4 ++-- tcp.c | 2 +- udp.c | 5 +++-- util.h | 29 ++++++++++------------------- 7 files changed, 47 insertions(+), 51 deletions(-) rename tcp_info.h => linux_dep.h (85%) -- 2.47.0

2 17

[PATCH 00/12] Minor fixups for or inspired by clangd and related tools
by David Gibson 07 Nov '24

07 Nov '24

I've been experimenting with Zed and clangd recently. Currently it generates an enormous number of largely spurious errors and warnings on the passt code base. Mostly that's due to its default configurations not suiting us. This series adds some configuration that addresses a number of those warnings, though there remain many more for now. Some of the warnings also look reasonable, so I have a grab bag of fixes or workarounds for some of those two. David Gibson (12): clang: Add .clang-format file Makefile: Simplify exclusion of qrap from static checks clang: Move clang-tidy configuration from Makefile to .clang-tidy arch: Avoid explicit access to 'environ' flow: Correct type of flowside_at_sidx() netlink: RTA_PAYLOAD() returns int, not size_t Makefile: Move NETNS_RUN_DIR definition to C code seccomp: Simplify handling of AUDIT_ARCH Makefile: Use -DARCH for qrap only Makefile: Don't attempt to auto-detect stack size clang: Add rudimentary clangd configuration util: Remove unused ffsl() function .clang-format | 126 ++++++++++++++++++++++++++++++++++++++++++++++ .clang-tidy | 93 ++++++++++++++++++++++++++++++++++ .clangd | 3 ++ Makefile | 136 +++----------------------------------------------- arch.c | 2 +- conf.c | 2 + flow_table.h | 2 +- netlink.c | 4 +- seccomp.sh | 14 +++++- util.h | 5 +- 10 files changed, 247 insertions(+), 140 deletions(-) create mode 100644 .clang-format create mode 100644 .clang-tidy create mode 100644 .clangd -- 2.47.0

2 16

2024

2023

2022

2021

dev November 2024