[PATCH v2 0/2] vhost-user, dhcp: Fix iPXE network boot over vhost-user
iPXE network boot over vhost-user was broken because passt
unconditionally skipped TCP/UDP checksum computation, relying on
VIRTIO_NET_HDR_F_DATA_VALID to tell the guest the checksums were
valid. Linux guests happened to work because their virtio-net driver
honours DATA_VALID regardless of feature negotiation, but iPXE
verifies checksums strictly and never negotiates VIRTIO_NET_F_GUEST_CSUM.
This series handles correctly VIRTIO_NET_F_GUEST_CSUM feature to fix that
and adds a minimal --dhcp-boot option that populates the BOOTP/DHCP 'file'
field, providing just enough for testing iPXE UDP and TCP support.
This can be tested as following:
- Create an ipxe file configuration
cat > boot-alpine.ipxe <
According to the virtio-net specification, when the VIRTIO_NET_F_GUEST_CSUM
is negotiated, the device can set VIRTIO_NET_HDR_F_DATA_VALID in the
virtio-net header to indicate that packet checksums have been validated,
allowing the guest to skip verification. Without this feature, the device
must provide fully checksummed packets.
The vhost-user TCP and UDP paths were unconditionally skipping checksum
computation, regardless of whether GUEST_CSUM was negotiated. This
went undetected with Linux guests because Linux's virtio-net driver
honours VIRTIO_NET_HDR_F_DATA_VALID regardless of whether
VIRTIO_NET_F_GUEST_CSUM was negotiated, marking such packets as
CHECKSUM_UNNECESSARY and skipping verification.
iPXE, however, does not negotiate GUEST_CSUM, ignores the DATA_VALID
flag entirely, and always verifies checksums. This caused TCP
connections to fail: the SYN-ACK had a zero TCP checksum, iPXE rejected
it, and the connection timed out in SYN_RCVD.
Adding --pcap happened to mask the bug, because the pcap code path
forces checksum computation to ensure correct captures.
Offer VIRTIO_NET_F_GUEST_CSUM in the device features, and only skip
checksum computation when the guest has actually negotiated it. When
GUEST_CSUM is not negotiated, always compute valid checksums as required
by the specification.
We keep setting VIRTIO_NET_HDR_F_DATA_VALID unconditionally in
VU_HEADER: when GUEST_CSUM is negotiated, the flag lets the guest skip
checksum verification; when it is not, the spec says the guest should
ignore the flags field, so setting it is harmless.
Signed-off-by: Laurent Vivier
Add a --dhcp-boot option that populates the 'file' field in DHCP reply
messages with the given filename.
Using --dhcp-boot together with --no-dhcp is rejected at startup.
Signed-off-by: Laurent Vivier
Hi Stefano, could you apply this one too? Thanks, Laurent On 4/16/26 18:21, Laurent Vivier wrote:
According to the virtio-net specification, when the VIRTIO_NET_F_GUEST_CSUM is negotiated, the device can set VIRTIO_NET_HDR_F_DATA_VALID in the virtio-net header to indicate that packet checksums have been validated, allowing the guest to skip verification. Without this feature, the device must provide fully checksummed packets.
The vhost-user TCP and UDP paths were unconditionally skipping checksum computation, regardless of whether GUEST_CSUM was negotiated. This went undetected with Linux guests because Linux's virtio-net driver honours VIRTIO_NET_HDR_F_DATA_VALID regardless of whether VIRTIO_NET_F_GUEST_CSUM was negotiated, marking such packets as CHECKSUM_UNNECESSARY and skipping verification.
iPXE, however, does not negotiate GUEST_CSUM, ignores the DATA_VALID flag entirely, and always verifies checksums. This caused TCP connections to fail: the SYN-ACK had a zero TCP checksum, iPXE rejected it, and the connection timed out in SYN_RCVD.
Adding --pcap happened to mask the bug, because the pcap code path forces checksum computation to ensure correct captures.
Offer VIRTIO_NET_F_GUEST_CSUM in the device features, and only skip checksum computation when the guest has actually negotiated it. When GUEST_CSUM is not negotiated, always compute valid checksums as required by the specification.
We keep setting VIRTIO_NET_HDR_F_DATA_VALID unconditionally in VU_HEADER: when GUEST_CSUM is negotiated, the flag lets the guest skip checksum verification; when it is not, the spec says the guest should ignore the flags field, so setting it is harmless.
Signed-off-by: Laurent Vivier
Reviewed-by: David Gibson --- tcp_vu.c | 8 ++++++-- udp_vu.c | 7 ++++--- vhost_user.c | 1 + 3 files changed, 11 insertions(+), 5 deletions(-) diff --git a/tcp_vu.c b/tcp_vu.c index 7f7e43860b10..3da14a4942d6 100644 --- a/tcp_vu.c +++ b/tcp_vu.c @@ -124,6 +124,7 @@ int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags) struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE]; size_t optlen, hdrlen, iov_cnt, iov_used; struct vu_virtq_element flags_elem[2]; + uint32_t csum_flags = IP4_CSUM; struct iovec flags_iov[64]; int elem_cnt, dup_elem_cnt = 0; struct tcp_syn_opts opts; @@ -135,6 +136,9 @@ int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags) uint32_t seq; int ret;
+ if (*c->pcap || !vu_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) + csum_flags |= TCP_CSUM; + hdrlen = tcp_vu_hdrlen(CONN_V6(conn));
elem_cnt = vu_collect(vdev, vq, &flags_elem[0], 1, @@ -173,7 +177,7 @@ int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags) iov_from_buf(payload.iov, payload.cnt, payload.off, &opts, optlen); tcp_fill_headers(c, conn, &eh, CONN_V4(conn) ? &ip4h : NULL, CONN_V6(conn) ? &ip6h : NULL, &th, &payload, - optlen, IP4_CSUM | (*c->pcap ? TCP_CSUM : 0), seq); + optlen, csum_flags, seq);
vu_pad(flags_elem[0].in_sg, iov_cnt, hdrlen + optlen);
@@ -519,7 +523,7 @@ int tcp_vu_data_from_sock(const struct ctx *c, struct tcp_tap_conn *conn)
hdrlen = tcp_vu_hdrlen(v6); check = IP4_CSUM; - if (*c->pcap) + if (*c->pcap || !vu_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) check |= TCP_CSUM; for (i = 0, previous_dlen = -1; i < frame_cnt; i++) { struct iovec *iov = &iov_vu[frame[i].idx_iovec]; diff --git a/udp_vu.c b/udp_vu.c index 8cf50ca1c38f..54d824b5530c 100644 --- a/udp_vu.c +++ b/udp_vu.c @@ -233,12 +233,13 @@ void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx) if (iov_cnt > 0) { struct iov_tail data = IOV_TAIL(iov_vu, iov_cnt, VNET_HLEN); udp_vu_prepare(c, &data, toside, dlen); - if (*c->pcap) { + if (!vu_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM) || + *c->pcap) udp_vu_csum(toside, &data, dlen); + vu_pad(iov_vu, iov_cnt, hdrlen + dlen); + if (*c->pcap) pcap_iov(iov_vu, iov_cnt, VNET_HLEN, hdrlen + dlen - VNET_HLEN); - } - vu_pad(iov_vu, iov_cnt, hdrlen + dlen); vu_flush(vdev, vq, elem, elem_used, hdrlen + dlen); vu_queue_notify(vdev, vq); } diff --git a/vhost_user.c b/vhost_user.c index f062badd3311..a1259c2624c0 100644 --- a/vhost_user.c +++ b/vhost_user.c @@ -322,6 +322,7 @@ static bool vu_get_features_exec(struct vu_dev *vdev, { uint64_t features = 1ULL << VIRTIO_F_VERSION_1 | + 1ULL << VIRTIO_NET_F_GUEST_CSUM | 1ULL << VIRTIO_NET_F_MRG_RXBUF | 1ULL << VHOST_F_LOG_ALL | 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
On Tue, 26 May 2026 14:58:16 +0200
Laurent Vivier
Hi Stefano,
could you apply this one too?
Wait, sorry, I skipped the whole series as I thought Anshu was anyway working on a more general approach for 2/2. I should just apply 1/2 instead, right? -- Stefano
On 5/26/26 15:48, Stefano Brivio wrote:
On Tue, 26 May 2026 14:58:16 +0200 Laurent Vivier
wrote: Hi Stefano,
could you apply this one too?
Wait, sorry, I skipped the whole series as I thought Anshu was anyway working on a more general approach for 2/2.
I should just apply 1/2 instead, right?
Yes, Anshu's series cover 2/2 but patch 1/2 fixes a problem between iPXE and passt vhost-user implementation. Thanks, Laurent
On Tue, 26 May 2026 16:00:55 +0200
Laurent Vivier
On 5/26/26 15:48, Stefano Brivio wrote:
On Tue, 26 May 2026 14:58:16 +0200 Laurent Vivier
wrote: Hi Stefano,
could you apply this one too?
Wait, sorry, I skipped the whole series as I thought Anshu was anyway working on a more general approach for 2/2.
I should just apply 1/2 instead, right?
Yes, Anshu's series cover 2/2 but patch 1/2 fixes a problem between iPXE and passt vhost-user implementation.
...so I tried applying this one but it has quite a few conflicts with your own series that I just applied. Do you happen to have a rebased version at hand? Otherwise I'll go ahead and solve the conflicts myself, they're not complicated. -- Stefano
On 5/26/26 16:48, Stefano Brivio wrote:
On Tue, 26 May 2026 16:00:55 +0200 Laurent Vivier
wrote: On 5/26/26 15:48, Stefano Brivio wrote:
On Tue, 26 May 2026 14:58:16 +0200 Laurent Vivier
wrote: Hi Stefano,
could you apply this one too?
Wait, sorry, I skipped the whole series as I thought Anshu was anyway working on a more general approach for 2/2.
I should just apply 1/2 instead, right?
Yes, Anshu's series cover 2/2 but patch 1/2 fixes a problem between iPXE and passt vhost-user implementation.
...so I tried applying this one but it has quite a few conflicts with your own series that I just applied.
Do you happen to have a rebased version at hand? Otherwise I'll go ahead and solve the conflicts myself, they're not complicated.
No, I don't have something already rebased. Laurent
On Tue, 26 May 2026 16:50:30 +0200
Laurent Vivier
On 5/26/26 16:48, Stefano Brivio wrote:
On Tue, 26 May 2026 16:00:55 +0200 Laurent Vivier
wrote: On 5/26/26 15:48, Stefano Brivio wrote:
On Tue, 26 May 2026 14:58:16 +0200 Laurent Vivier
wrote: Hi Stefano,
could you apply this one too?
Wait, sorry, I skipped the whole series as I thought Anshu was anyway working on a more general approach for 2/2.
I should just apply 1/2 instead, right?
Yes, Anshu's series cover 2/2 but patch 1/2 fixes a problem between iPXE and passt vhost-user implementation.
...so I tried applying this one but it has quite a few conflicts with your own series that I just applied.
Do you happen to have a rebased version at hand? Otherwise I'll go ahead and solve the conflicts myself, they're not complicated.
No, I don't have something already rebased.
Applying now. Due to commit dec66c02b5e4 ("udp: Pass iov_tail to udp_update_hdr4()/udp_update_hdr6()"), merged meanwhile, this part in udp_vu_sock_to_tap(): if (!vu_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM) || *c->pcap) udp_vu_csum(toside, &data, dlen); became, in udp_vu_prepare(): no_csum = vu_has_feature(c->vdev, VIRTIO_NET_F_GUEST_CSUM) && !*c->pcap; [...] udp_update_hdr4(&iph, &uh, payload, toside, dlen, no_csum); [...] udp_update_hdr6(&ip6h, &uh, payload, toside, dlen, no_csum); -- Stefano
participants (2)
-
Laurent Vivier
-
Stefano Brivio