- dev - passt

[PATCH] tcp: unify payload and flags l2 frames array
by Jon Maloy 23 Sep '24

23 Sep '24

In order to reduce memory and code footprint, we merge the array for l2 flag frames into the one for payload frames. This change also ensures that no flag message will be sent out over the l2 media bypassing already queued payload messages. Signed-off-by: Jon Maloy <jmaloy(a)redhat.com> --- tcp.c | 3 +-- tcp_buf.c | 70 ++++++++++++------------------------------------------- 2 files changed, 16 insertions(+), 57 deletions(-) diff --git a/tcp.c b/tcp.c index 12e864f..dd060d0 100644 --- a/tcp.c +++ b/tcp.c @@ -868,7 +868,6 @@ bool tcp_flow_defer(const struct tcp_tap_conn *conn) /* cppcheck-suppress [constParameterPointer, unmatchedSuppression] */ void tcp_defer_handler(struct ctx *c) { - tcp_flags_flush(c); tcp_payload_flush(c); } @@ -1148,7 +1147,7 @@ static void tcp_update_seqack_from_tap(const struct ctx *c, * 1 otherwise */ int tcp_prepare_flags(const struct ctx *c, struct tcp_tap_conn *conn, - int flags, struct tcphdr *th, char *data, + int flags, struct tcphdr *th, uint8_t *data, size_t *optlen) { struct tcp_info tinfo = { 0 }; diff --git a/tcp_buf.c b/tcp_buf.c index 625c29c..58636d1 100644 --- a/tcp_buf.c +++ b/tcp_buf.c @@ -33,6 +33,7 @@ #include "tcp_internal.h" #include "tcp_buf.h" +#define PAYLOAD_FLAGS htons(0x5010) /* doff = 5, ack = 1 */ #define TCP_FRAMES_MEM 128 #define TCP_FRAMES \ (c->mode == MODE_PASTA ? 1 : TCP_FRAMES_MEM) @@ -52,21 +53,6 @@ struct tcp_payload_t { } __attribute__ ((packed, aligned(__alignof__(unsigned int)))); #endif -/** - * struct tcp_flags_t - TCP header and data to send zero-length - * segments (flags) - * @th: TCP header - * @opts TCP options - */ -struct tcp_flags_t { - struct tcphdr th; - char opts[OPT_MSS_LEN + OPT_WS_LEN + 1]; -#ifdef __AVX2__ -} __attribute__ ((packed, aligned(32))); -#else -} __attribute__ ((packed, aligned(__alignof__(unsigned int)))); -#endif - /* Ethernet header for IPv4 and IPv6 frames */ static struct ethhdr tcp4_eth_src; static struct ethhdr tcp6_eth_src; @@ -87,22 +73,10 @@ static_assert(MSS6 <= sizeof(tcp_payload[0].data), "MSS6 is greater than 65516") static struct tcp_tap_conn *tcp_frame_conns[TCP_FRAMES_MEM]; static unsigned int tcp_payload_used; -static struct tap_hdr tcp_flags_tap_hdr[TCP_FRAMES_MEM]; -/* IPv4 headers for TCP segment without payload */ -static struct iphdr tcp4_flags_ip[TCP_FRAMES_MEM]; -/* TCP segments without payload for IPv4 frames */ -static struct tcp_flags_t tcp_flags[TCP_FRAMES_MEM]; - -static unsigned int tcp_flags_used; - -/* IPv6 headers for TCP segment without payload */ -static struct ipv6hdr tcp6_flags_ip[TCP_FRAMES_MEM]; - /* recvmsg()/sendmsg() data for tap */ static struct iovec iov_sock [TCP_FRAMES_MEM + 1]; static struct iovec tcp_l2_iov[TCP_FRAMES_MEM][TCP_NUM_IOVS]; -static struct iovec tcp_l2_flags_iov[TCP_FRAMES_MEM][TCP_NUM_IOVS]; /** * tcp_update_l2_buf() - Update Ethernet header buffers with addresses @@ -135,13 +109,6 @@ void tcp_sock_iov_init(const struct ctx *c) tcp_payload[i].th.ack = 1; } - for (i = 0; i < ARRAY_SIZE(tcp_flags); i++) { - tcp6_flags_ip[i] = ip6; - tcp4_flags_ip[i] = iph; - tcp_flags[i].th.doff = sizeof(struct tcphdr) / 4; - tcp_flags[i].th.ack = 1; - } - for (i = 0; i < TCP_FRAMES_MEM; i++) { struct iovec *iov = tcp_l2_iov[i]; @@ -149,14 +116,6 @@ void tcp_sock_iov_init(const struct ctx *c) iov[TCP_IOV_ETH].iov_len = sizeof(struct ethhdr); iov[TCP_IOV_PAYLOAD].iov_base = &tcp_payload[i]; } - - for (i = 0; i < TCP_FRAMES_MEM; i++) { - struct iovec *iov = tcp_l2_flags_iov[i]; - - iov[TCP_IOV_TAP] = tap_hdr_iov(c, &tcp_flags_tap_hdr[i]); - iov[TCP_IOV_ETH].iov_len = sizeof(struct ethhdr); - iov[TCP_IOV_PAYLOAD].iov_base = &tcp_flags[i]; - } } /** @@ -165,9 +124,7 @@ void tcp_sock_iov_init(const struct ctx *c) */ void tcp_flags_flush(const struct ctx *c) { - tap_send_frames(c, &tcp_l2_flags_iov[0][0], TCP_NUM_IOVS, - tcp_flags_used); - tcp_flags_used = 0; + tcp_payload_flush(c); } /** @@ -225,37 +182,35 @@ void tcp_payload_flush(const struct ctx *c) */ int tcp_buf_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags) { - struct tcp_flags_t *payload; + struct tcp_payload_t *payload; struct iovec *iov; size_t optlen; size_t l4len; uint32_t seq; int ret; - iov = tcp_l2_flags_iov[tcp_flags_used]; + iov = tcp_l2_iov[tcp_payload_used]; if (CONN_V4(conn)) { - iov[TCP_IOV_IP] = IOV_OF_LVALUE(tcp4_flags_ip[tcp_flags_used]); + iov[TCP_IOV_IP] = IOV_OF_LVALUE(tcp4_payload_ip[tcp_payload_used]); iov[TCP_IOV_ETH].iov_base = &tcp4_eth_src; } else { - iov[TCP_IOV_IP] = IOV_OF_LVALUE(tcp6_flags_ip[tcp_flags_used]); + iov[TCP_IOV_IP] = IOV_OF_LVALUE(tcp6_payload_ip[tcp_payload_used]); iov[TCP_IOV_ETH].iov_base = &tcp6_eth_src; } payload = iov[TCP_IOV_PAYLOAD].iov_base; seq = conn->seq_to_tap; ret = tcp_prepare_flags(c, conn, flags, &payload->th, - payload->opts, &optlen); + payload->data, &optlen); if (ret <= 0) return ret; - tcp_flags_used++; + tcp_payload_used++; l4len = tcp_l2_buf_fill_headers(conn, iov, optlen, NULL, seq, false); iov[TCP_IOV_PAYLOAD].iov_len = l4len; - if (flags & DUP_ACK) { - struct iovec *dup_iov; + struct iovec *dup_iov = tcp_l2_iov[tcp_payload_used++]; - dup_iov = tcp_l2_flags_iov[tcp_flags_used++]; memcpy(dup_iov[TCP_IOV_TAP].iov_base, iov[TCP_IOV_TAP].iov_base, iov[TCP_IOV_TAP].iov_len); dup_iov[TCP_IOV_ETH].iov_base = iov[TCP_IOV_ETH].iov_base; @@ -265,7 +220,7 @@ int tcp_buf_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags) dup_iov[TCP_IOV_PAYLOAD].iov_len = l4len; } - if (tcp_flags_used > TCP_FRAMES_MEM - 2) + if (tcp_payload_used > TCP_FRAMES_MEM - 2) tcp_flags_flush(c); return 0; @@ -282,8 +237,10 @@ int tcp_buf_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags) static void tcp_data_to_tap(const struct ctx *c, struct tcp_tap_conn *conn, ssize_t dlen, int no_csum, uint32_t seq) { + struct tcp_payload_t *payload; const uint16_t *check = NULL; struct iovec *iov; + uint16_t *flags; size_t l4len; conn->seq_to_tap = seq + dlen; @@ -302,6 +259,9 @@ static void tcp_data_to_tap(const struct ctx *c, struct tcp_tap_conn *conn, iov[TCP_IOV_IP] = IOV_OF_LVALUE(tcp6_payload_ip[tcp_payload_used]); iov[TCP_IOV_ETH].iov_base = &tcp6_eth_src; } + payload = iov[TCP_IOV_PAYLOAD].iov_base; + flags = &payload->th.window - 1; + *(flags) = PAYLOAD_FLAGS; l4len = tcp_l2_buf_fill_headers(conn, iov, dlen, check, seq, false); iov[TCP_IOV_PAYLOAD].iov_len = l4len; if (++tcp_payload_used > TCP_FRAMES_MEM - 1) -- 2.45.2

3 2

[PATCH v3 0/2] util: Fix some problems in write_remainder()
by David Gibson 19 Sep '24

19 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

3 6

[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 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 0/2] 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

[PATCH v4 0/4] Add vhost-user support to passt. (part 3)
by Laurent Vivier 12 Sep '24

12 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 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 | 15 +- 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 | 656 +++++++++++++++++++++++++ tcp_vu.h | 12 + udp.c | 76 +-- udp.h | 8 +- udp_internal.h | 34 ++ udp_vu.c | 386 +++++++++++++++ udp_vu.h | 13 + util.h | 8 + vhost_user.c | 1267 ++++++++++++++++++++++++++++++++++++++++++++++++ vhost_user.h | 203 ++++++++ virtio.c | 659 +++++++++++++++++++++++++ virtio.h | 185 +++++++ vu_common.c | 35 ++ vu_common.h | 34 ++ 31 files changed, 3801 insertions(+), 139 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

2 14