- dev - passt

[PATCH v3 0/5] Fixes for Fedora 43
by David Gibson 11 Jan '26

11 Jan '26

While away, I updated my laptop to Fedora 43. Naturally a bunch of things in the passt tests broke. First there were the usual breakages due to a new static checker version (also one real bug). Then several other problems. Changes in v3: * Fixed a really stupid mistake in applying the change to 2/5 Changes in v2: * Trivial whitespace fix from Laurent * Used more natural array iteration in 2/5 (suggested by Laurent) * Added Laurent's R-b tags David Gibson (5): util: Be more defensive about buffer overruns in read_file() migrate: Don't use terminator element for versions[] array treewide: Don't rely on terminator records in ip[46].dns arrays test: Handle Operating System Command escapes in terminal output test: Include sshd-auth in mbuto guest image conf.c | 8 ++++++-- dhcp.c | 4 +++- dhcpv6.c | 4 +++- migrate.c | 9 ++++----- ndp.c | 4 +++- passt.h | 4 ++-- test/lib/term | 4 ++-- test/passt.mbuto | 6 ++++++ util.c | 2 +- 9 files changed, 30 insertions(+), 15 deletions(-) -- 2.52.0

2 7

[PATCH v2] udp_vu: Discard datagrams when RX virtqueue is not usable
by Laurent Vivier 11 Jan '26

11 Jan '26

During vhost-user device initialization, UDP datagrams may arrive on listening sockets before the guest has enabled the RX virtqueue. When this happens, udp_vu_sock_recv() returns 0 without consuming the datagram from the socket. The caller, udp_sock_fwd(), uses a while loop with udp_peek_addr() to process pending datagrams. Since the datagram remains in the socket buffer, udp_peek_addr() keeps returning data available, causing a busy loop with 100% CPU usage. To avoid that, we need to discard the data when the virtqueue is not ready. udp_buf_sock_to_tap() actually does the same as it reads data with udp_sock_recv() and if fd_tap is not initialized tap_send_frames() drops them. Fixes: 28997fcb29b5 ("vhost-user: add vhost-user") Link: https://bugs.passt.top/show_bug.cgi?id=185 Signed-off-by: Laurent Vivier <lvivier(a)redhat.com> --- Notes: v2: - move recvmsg() from udp_vu_sock_to_tap() to udp_vu_sock_recv() udp_vu.c | 32 ++++++++++++++++++++++---------- 1 file changed, 22 insertions(+), 10 deletions(-) diff --git a/udp_vu.c b/udp_vu.c index c30dcf97698f..3774d538a2d0 100644 --- a/udp_vu.c +++ b/udp_vu.c @@ -65,7 +65,8 @@ static size_t udp_vu_hdrlen(bool v6) * @v6: Set for IPv6 connections * @dlen: Size of received data (output) * - * Return: number of iov entries used to store the datagram + * Return: number of iov entries used to store the datagram, 0 if the datagram + * was discarded because the virtqueue is not ready, -1 on error */ static int udp_vu_sock_recv(const struct ctx *c, struct vu_virtq *vq, int s, bool v6, ssize_t *dlen) @@ -77,6 +78,15 @@ static int udp_vu_sock_recv(const struct ctx *c, struct vu_virtq *vq, int s, ASSERT(!c->no_udp); + if (!vu_queue_enabled(vq) || !vu_queue_started(vq)) { + debug("Got UDP packet, but RX virtqueue not usable yet"); + + if (recvmsg(s, &msg, MSG_DONTWAIT) < 0) + debug_perror("Failed to discard datagram"); + + return 0; + } + /* compute L2 header length */ hdrlen = udp_vu_hdrlen(v6); @@ -87,7 +97,7 @@ static int udp_vu_sock_recv(const struct ctx *c, struct vu_virtq *vq, int s, sizeof(struct virtio_net_hdr_mrg_rxbuf), NULL); if (iov_cnt == 0) - return 0; + return -1; /* reserve space for the headers */ ASSERT(iov_vu[0].iov_len >= MAX(hdrlen, ETH_ZLEN)); @@ -101,7 +111,7 @@ static int udp_vu_sock_recv(const struct ctx *c, struct vu_virtq *vq, int s, *dlen = recvmsg(s, &msg, 0); if (*dlen < 0) { vu_queue_rewind(vq, iov_cnt); - return 0; + return -1; } /* restore the pointer to the headers address */ @@ -216,15 +226,17 @@ void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx) int iov_used; iov_used = udp_vu_sock_recv(c, vq, s, v6, &dlen); - if (iov_used <= 0) + if (iov_used < 0) break; - udp_vu_prepare(c, toside, dlen); - if (*c->pcap) { - udp_vu_csum(toside, iov_used); - pcap_iov(iov_vu, iov_used, - sizeof(struct virtio_net_hdr_mrg_rxbuf)); + if (iov_used > 0) { + udp_vu_prepare(c, toside, dlen); + if (*c->pcap) { + udp_vu_csum(toside, iov_used); + pcap_iov(iov_vu, iov_used, + sizeof(struct virtio_net_hdr_mrg_rxbuf)); + } + vu_flush(vdev, vq, elem, iov_used); } - vu_flush(vdev, vq, elem, iov_used); } } -- 2.52.0

3 4

[PATCH v2 0/3] Clean ups to epoll references
by David Gibson 11 Jan '26

11 Jan '26

A handful of simple cleanups to epoll references. Changes in v2: * Removed bogus comment line left over from an early draft David Gibson (3): tcp: Remove unused tcp_epoll_ref epoll_ctl: Add missing description for flowside field of epoll_ref fwd, tcp, udp: Consolidate epoll refs for listening sockets epoll_ctl.h | 7 +++---- fwd.h | 13 +++++++++++++ tcp.c | 8 ++++---- tcp.h | 24 ------------------------ udp.c | 6 +++--- udp.h | 15 --------------- 6 files changed, 23 insertions(+), 50 deletions(-) -- 2.52.0

3 7

[PATCH v2 0/5] flow: Introduce flow_epoll_set() to centralize epoll operations
by Laurent Vivier 09 Jan '26

09 Jan '26

Currently, each protocol handler (TCP, TCP splice, ICMP/ping, UDP) has its own code to add or modify file descriptors in epoll. This leads to duplicated boilerplate across icmp.c, tcp.c, tcp_splice.c, and udp_flow.c, with each setting up epoll_ref unions and calling epoll_ctl() with flow-type-specific details. This series introduces flow_epoll_set() in flow.c to handle epoll operations for all flow types in a unified way. The function provides a simpler interface where the caller passes the command, type, events, and file descriptor, and flow_epoll_set() handles the epoll_ref setup and epoll_ctl() call. The API is: int flow_epoll_set(int command, enum epoll_type type, const struct flow_common *f, uint32_t events, int fd, unsigned int sidei); This centralized approach will be essential for queue pair migration in the upcoming multithreading work, where flows need to be moved between different epoll instances owned by different threads. Preparatory patches: - Patch 1 removes dead timer update code in tcp_epoll_ctl() - Patch 2 removes unneeded epoll_ref indirection in udp_flow - Patch 3 refactors udp_flow_sock() to assign the socket internally - Patch 4 refactors tcp_splice_conn_epoll_events() to per-side computation Core patch: - Patch 5 introduces flow_epoll_set() used by all protocol handlers Changes in v2: - Added patch 1: Remove dead timer update in tcp_epoll_ctl() since flow table compaction was eliminated - Added patch 2: Remove unnecessary epoll_ref indirection in udp_flow.c - Dropped v1 patch 6 (compute events inside flow_epoll_set()) - Dropped v1 patch 7 (retrieve fd from flow structure) - Added epoll_ctl() command parameter (EPOLL_CTL_ADD/EPOLL_CTL_MOD) - Callers must set epollfd via flow_epollid_set() before calling flow_epoll_set() - Removed FLOW_EPOLL_TIMER_ADD/FLOW_EPOLL_TIMER_MOD constants - Reduced series from 7 to 5 patches Laurent Vivier (5): tcp: remove timer update in tcp_epoll_ctl() udp_flow: remove unneeded epoll_ref indirection udp_flow: Assign socket to flow inside udp_flow_sock() tcp_splice: Refactor tcp_splice_conn_epoll_events() to per-side computation flow: Introduce flow_epoll_set() to centralize epoll operations flow.c | 33 +++++++++++++++ flow.h | 3 ++ icmp.c | 11 ++--- tcp.c | 72 ++++++++++++++------------------ tcp_splice.c | 115 ++++++++++++++++++++++++--------------------------- udp_flow.c | 19 +++------ 6 files changed, 131 insertions(+), 122 deletions(-) -- 2.52.0

2 9

[PATCH] tcp: remove timer update in tcp_epoll_ctl()
by Laurent Vivier 09 Jan '26

09 Jan '26

Timer update in tcp_epoll_ctl() has been added by bb708111833e ("treewide: Packet abstraction with mandatory boundary checks") because epoll_ref stores "conn - tc" that can change in tcp_table_compact(). But since e2e8219f13b8 ("flow, tcp: Consolidate flow pointer<->index helpers") we use FLOW_IDX() and since 8981a720aac4 ("flow: Avoid moving flow entries to compact table") flow table doesn't use a compaction procedure so FLOW_IDX() never changes. Updating the timer event is now a no-op, remove it from tcp_epoll_ctl(). Fixes: 8981a720aac4 ("flow: Avoid moving flow entries to compact table") Signed-off-by: Laurent Vivier <lvivier(a)redhat.com> --- tcp.c | 12 ------------ 1 file changed, 12 deletions(-) diff --git a/tcp.c b/tcp.c index e7fa85f346b2..5141cdc7e839 100644 --- a/tcp.c +++ b/tcp.c @@ -552,18 +552,6 @@ static int tcp_epoll_ctl(const struct ctx *c, struct tcp_tap_conn *conn) flow_epollid_set(&conn->f, EPOLLFD_ID_DEFAULT); - if (conn->timer != -1) { - union epoll_ref ref_t = { .type = EPOLL_TYPE_TCP_TIMER, - .fd = conn->timer, - .flow = FLOW_IDX(conn) }; - struct epoll_event ev_t = { .data.u64 = ref_t.u64, - .events = EPOLLIN | EPOLLET }; - - if (epoll_ctl(flow_epollfd(&conn->f), EPOLL_CTL_MOD, - conn->timer, &ev_t)) - return -errno; - } - return 0; } -- 2.52.0

2 1

[PATCH] apparmor: Allow reading TCP RTO sysctl parameters
by Stefano Brivio 08 Jan '26

08 Jan '26

From: Martin Pitt <mpitt(a)redhat.com> Since commits 3dde0e07804e ("tcp: Update data retransmission timeout") and 1a834879a2f7 ("tcp: Clamp the retry timeout") from 2025-12-02, passt reads additional TCP-related sysctl parameters from /proc to configure retransmission timeout behavior: - /proc/sys/net/ipv4/tcp_syn_retries - /proc/sys/net/ipv4/tcp_syn_linear_timeouts - /proc/sys/net/ipv4/tcp_rto_max_ms These are read by tcp_get_rto_params() during initialization. Adjust the AppArmor profile accordingly. Link: https://bugzilla.redhat.com/show_bug.cgi?id=2426863 Link: https://github.com/cockpit-project/bots/pull/8568 Fixes: 3dde0e07804e ("tcp: Update data retransmission timeout") Fixes: 1a834879a2f7 ("tcp: Clamp the retry timeout") Signed-off-by: Martin Pitt <mpitt(a)redhat.com> [sbrivio: Minor formatting change, changed commit references] Signed-off-by: Stefano Brivio <sbrivio(a)redhat.com> --- Martin, I took the liberty to add a newline around the lines you're adding so that we need a single comment and everything is aligned. I also changed the format of commit references to what we typically use, which is inspired by the preferred format for kernel submissions, I used this one-liner: git show -s --abbrev=12 --pretty=format:"Fixes: %h (\"%s\")" "${SHA}" which actually belongs to CONTRIBUTING.md, cf. page 1241 of my to-do book. contrib/apparmor/abstractions/passt | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/contrib/apparmor/abstractions/passt b/contrib/apparmor/abstractions/passt index 43fd63f..25b2ea8 100644 --- a/contrib/apparmor/abstractions/passt +++ b/contrib/apparmor/abstractions/passt @@ -36,6 +36,10 @@ @{PROC}/sys/net/ipv4/ip_local_port_range r, # fwd_probe_ephemeral() + @{PROC}/sys/net/ipv4/tcp_syn_retries r, # tcp_get_rto_params(), tcp.c + @{PROC}/sys/net/ipv4/tcp_syn_linear_timeouts r, + @{PROC}/sys/net/ipv4/tcp_rto_max_ms r, + network netlink raw, # nl_sock_init_do(), netlink.c network inet stream, # tcp.c -- 2.43.0

1 0

[PATCH v3 01/14] inany: Extend inany_ntop() to treat NULL as a fully unspecified address
by David Gibson 08 Jan '26

08 Jan '26

In a number of places we're using a convention that a NULL inany represents a dual-stack unspecified address, that is one which includes both 0.0.0.0 and ::. Extend inany_ntop() to handle that convention, representing it as "*". Signed-off-by: David Gibson <david(a)gibson.dropbear.id.au> --- inany.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/inany.c b/inany.c index 7680439d..87a4d8b6 100644 --- a/inany.c +++ b/inany.c @@ -22,7 +22,7 @@ const union inany_addr inany_loopback4 = INANY_INIT4(IN4ADDR_LOOPBACK_INIT); const union inany_addr inany_any4 = INANY_INIT4(IN4ADDR_ANY_INIT); /** inany_ntop - Convert an IPv[46] address to text format - * @src: IPv[46] address + * @src: IPv[46] address (NULL for unspecified) * @dst: output buffer, minimum INANY_ADDRSTRLEN bytes * @size: size of buffer at @dst * @@ -30,9 +30,12 @@ const union inany_addr inany_any4 = INANY_INIT4(IN4ADDR_ANY_INIT); */ const char *inany_ntop(const union inany_addr *src, char *dst, socklen_t size) { - const struct in_addr *v4 = inany_v4(src); + const struct in_addr *v4; + + if (!src) + return strncpy(dst, "*", size); - if (v4) + if ((v4 = inany_v4(src))) return inet_ntop(AF_INET, v4, dst, size); return inet_ntop(AF_INET6, &src->a6, dst, size); -- 2.52.0

2 1

[PATCH] udp_flow: remove unneeded epoll_ref indirection
by Laurent Vivier 08 Jan '26

08 Jan '26

The fref union was used to convert flow_sidx_t to uint32_t for assignment to ref.data. This is unnecessary since epoll_ref already contains a flowside member of type flow_sidx_t, so we can assign directly. This aligns with how icmp.c and other callers assign flow_sidx_t to epoll_ref. Signed-off-by: Laurent Vivier <lvivier(a)redhat.com> --- udp_flow.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) diff --git a/udp_flow.c b/udp_flow.c index 8907f2f72741..0ba788060db7 100644 --- a/udp_flow.c +++ b/udp_flow.c @@ -74,10 +74,6 @@ static int udp_flow_sock(const struct ctx *c, { const struct flowside *side = &uflow->f.side[sidei]; uint8_t pif = uflow->f.pif[sidei]; - union { - flow_sidx_t sidx; - uint32_t data; - } fref = { .sidx = FLOW_SIDX(uflow, sidei) }; union epoll_ref ref; int rc; int s; @@ -89,7 +85,7 @@ static int udp_flow_sock(const struct ctx *c, } ref.type = EPOLL_TYPE_UDP; - ref.data = fref.data; + ref.flowside = FLOW_SIDX(uflow, sidei); ref.fd = s; flow_epollid_set(&uflow->f, EPOLLFD_ID_DEFAULT); -- 2.52.0

2 1

[PATCH v2 0/5] Fixes for Fedora 43
by David Gibson 07 Jan '26

07 Jan '26

While away, I updated my laptop to Fedora 43. Naturally a bunch of things in the passt tests broke. First there were the usual breakages due to a new static checker version (also one real bug). Then several other problems. Changes in v2: * Trivial whitespace fix from Laurent * Used more natural array iteration in 2/5 (suggested by Laurent) * Added Laurent's R-b tags David Gibson (5): util: Be more defensive about buffer overruns in read_file() migrate: Don't use terminator element for versions[] array treewide: Don't rely on terminator records in ip[46].dns arrays test: Handle Operating System Command escapes in terminal output test: Include sshd-auth in mbuto guest image conf.c | 8 ++++++-- dhcp.c | 4 +++- dhcpv6.c | 4 +++- migrate.c | 8 ++++---- ndp.c | 4 +++- passt.h | 4 ++-- test/lib/term | 4 ++-- test/passt.mbuto | 6 ++++++ util.c | 2 +- 9 files changed, 30 insertions(+), 14 deletions(-) -- 2.52.0

1 4

[PATCH v3 1/5] util: Be more defensive about buffer overruns in read_file()
by David Gibson 07 Jan '26

07 Jan '26

clang-21.1.7 complains about read_file(), thinking that total_read might come to exceed buf_size, leading to an out of bounds access at the end of the function. In fact, the semantics of read()'s return mean this can't ever happen. But we already have to check for the total_read == buf_size case, so it's basically free to change it to >= and suppress the error. Signed-off-by: David Gibson <david(a)gibson.dropbear.id.au> Reviewed-by: Laurent Vivier <lvivier(a)redhat.com> --- util.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/util.c b/util.c index 27303950..a48f727c 100644 --- a/util.c +++ b/util.c @@ -715,7 +715,7 @@ static ssize_t read_file(const char *path, char *buf, size_t buf_size) close(fd); - if (total_read == buf_size) { + if (total_read >= buf_size) { buf[buf_size - 1] = '\0'; return -ENOBUFS; } -- 2.52.0

1 0