Date: Thu, 9 Oct 2025 13:11:11 GMT From: Michael Tuexen <tuexen@FreeBSD.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-branches@FreeBSD.org Subject: git: 4774687d4087 - stable/15 - tcp lro: use the flowid only when it has hash properties Message-ID: <202510091311.599DBBpX041879@gitrepo.freebsd.org>
next in thread | raw e-mail | index | archive | help
The branch stable/15 has been updated by tuexen: URL: https://cgit.FreeBSD.org/src/commit/?id=4774687d4087cf40e46939f29473f7cb8c269ad5 commit 4774687d4087cf40e46939f29473f7cb8c269ad5 Author: Michael Tuexen <tuexen@FreeBSD.org> AuthorDate: 2025-10-09 12:37:11 +0000 Commit: Michael Tuexen <tuexen@FreeBSD.org> CommitDate: 2025-10-09 12:53:09 +0000 tcp lro: use the flowid only when it has hash properties When a packet is provided to LRO using tcp_lro_queue_mbuf(), a sequence number is computed based on the m_pkthdr.flowid provided by he driver. The implicit assumption is that the m_pkthdr.flowid has hash properties. The recent use of tcp_lro_queue_mbuf() in iflib exposed a bug in at least one driver (igc) , which * reports always that is uses M_HASHTYPE_OPAQUE. * sets in some cases m_pkthdr.flowid not consistently for packets belonging to the same TCP connection. This results in severe performance problems for the base TCP stack, since it handles the packets in the wrong sequence, although they were received in the correct sequence. To protect against such misbehaving drivers, just take the m_pkthdr.flowid only into account, if it has hash properties. The performance problems were observed by gallatin@ and analyzed together with rrs@. Reviewed by: gallatin Tested by: gallatin Sponsored by: Netflix, Inc. Differential Revision: https://reviews.freebsd.org/D52989 (cherry picked from commit 73f56fa542ecf6ea6819c9e17fc60962eeb726d8) --- sys/netinet/tcp_lro.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/sys/netinet/tcp_lro.c b/sys/netinet/tcp_lro.c index 64efa4bf060f..9b5baf115855 100644 --- a/sys/netinet/tcp_lro.c +++ b/sys/netinet/tcp_lro.c @@ -1475,10 +1475,11 @@ tcp_lro_queue_mbuf(struct lro_ctrl *lc, struct mbuf *mb) } /* create sequence number */ - lc->lro_mbuf_data[lc->lro_mbuf_count].seq = - (((uint64_t)M_HASHTYPE_GET(mb)) << 56) | - (((uint64_t)mb->m_pkthdr.flowid) << 24) | - ((uint64_t)lc->lro_mbuf_count); + lc->lro_mbuf_data[lc->lro_mbuf_count].seq = lc->lro_mbuf_count; + if (M_HASHTYPE_ISHASH(mb)) + lc->lro_mbuf_data[lc->lro_mbuf_count].seq |= + (((uint64_t)M_HASHTYPE_GET(mb)) << 56) | + (((uint64_t)mb->m_pkthdr.flowid) << 24); /* enter mbuf */ lc->lro_mbuf_data[lc->lro_mbuf_count].mb = mb;
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202510091311.599DBBpX041879>