- dev - passt

[PATCH v22 0/2] State migration
by Stefano Brivio 14 Feb '25

14 Feb '25

Not many changes compared to v21: - Drop Bash-only == from tests, restore functionality with dash - Rebase: this is now based on top of "tcp: Keep updating window and checking for socket data after FIN from guest" and "tcp, tcp_splice: Don't set SO_SNDBUF and SO_RCVBUF to maximum values" Still missing: - tests with half-closed socket - complete error handling with (maybe?) tcpi_state == 0 as indication that we failed to dump a given flow and we just want to restore the next one Stefano Brivio (2): migrate: Migrate TCP flows test: Add migration tests contrib/selinux/passt.te | 4 +- flow.c | 260 ++++++++++ flow.h | 8 + migrate.c | 19 + passt.c | 6 +- repair.c | 1 - tcp.c | 918 ++++++++++++++++++++++++++++++++++++ tcp_conn.h | 103 ++++ test/lib/layout | 55 ++- test/lib/setup | 138 +++++- test/lib/test | 48 ++ test/migrate/basic | 59 +++ test/migrate/bidirectional | 64 +++ test/migrate/iperf3_bidir6 | 58 +++ test/migrate/iperf3_in4 | 50 ++ test/migrate/iperf3_in6 | 58 +++ test/migrate/iperf3_out4 | 50 ++ test/migrate/iperf3_out6 | 58 +++ test/migrate/rampstream_in | 10 +- test/migrate/rampstream_out | 6 +- test/run | 34 +- 21 files changed, 1990 insertions(+), 17 deletions(-) create mode 100644 test/migrate/basic create mode 100644 test/migrate/bidirectional create mode 100644 test/migrate/iperf3_bidir6 create mode 100644 test/migrate/iperf3_in4 create mode 100644 test/migrate/iperf3_in6 create mode 100644 test/migrate/iperf3_out4 create mode 100644 test/migrate/iperf3_out6 -- 2.43.0

1 2

[PATCH] tcp: Keep updating window and checking for socket data after FIN from guest
by Stefano Brivio 14 Feb '25

14 Feb '25

Once we get a FIN segment from the container/guest, we enter something resembling CLOSE_WAIT (from the perspective of the peer), but that doesn't mean that we should stop processing window updates from the guest and checking for socket data if the guest acknowledges something. If we don't do that, we can very easily run into a situation where we send a burst of data to the tap, get a zero window update, along with a FIN segment, because the flow is meant to be unidirectional, and now the connection will be stuck forever, because we'll ignore updates. Reproducer, server: $ pasta --config-net -t 9999 -- sh -c 'echo DONE | socat TCP-LISTEN:9997,shut-down STDIO' and client: $ ./test/rampstream send 50000 | socat -u STDIN TCP:$LOCAL_ADDR:9997 2025/02/13 09:14:45 socat[2997126] E write(5, 0x55f5dbf47000, 8192): Broken pipe while at it, update the message string for the third passive close state (which we see in this case): it's CLOSE_WAIT, not LAST_ACK. Signed-off-by: Stefano Brivio <sbrivio(a)redhat.com> --- tcp.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/tcp.c b/tcp.c index a1d6c53..16d01f6 100644 --- a/tcp.c +++ b/tcp.c @@ -338,7 +338,7 @@ static const char *tcp_state_str[] __attribute((__unused__)) = { "SYN_RCVD", /* approximately maps to TAP_SYN_ACK_SENT */ /* Passive close: */ - "CLOSE_WAIT", "CLOSE_WAIT", "LAST_ACK", "LAST_ACK", "LAST_ACK", + "CLOSE_WAIT", "CLOSE_WAIT", "CLOSE_WAIT", "LAST_ACK", "LAST_ACK", /* Active close (+5): */ "CLOSING", "FIN_WAIT_1", "FIN_WAIT_1", "FIN_WAIT_2", "TIME_WAIT", }; @@ -1968,6 +1968,8 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af, /* Established connections not accepting data from tap */ if (conn->events & TAP_FIN_RCVD) { tcp_update_seqack_from_tap(c, conn, ntohl(th->ack_seq)); + tcp_tap_window_update(conn, ntohs(th->window)); + tcp_data_from_sock(c, conn); if (conn->events & SOCK_FIN_RCVD && conn->seq_ack_from_tap == conn->seq_to_tap) -- 2.43.0

2 1

[PATCH] udp: create and send ICMPv4 to local peer when applicable
by Jon Maloy 13 Feb '25

13 Feb '25

When a local peer sends a UDP message to a non-existing port on an existing remote host, that host will return an ICMP message containing the error code ICMP_PORT_UNREACH, plus the header and the first eight bytes of the original message. If the sender socket has been connected, it uses this message to issue a "Connection Refused" event to the user. Until now, we have only read such events from the externally facing socket, but we don't forward them back to the local sender because we cannot read the ICMP message directly to user space. Because of this, the local peer will hang and wait for a response that never arrives. We now fix this for IPv4 by recreating and forwarding a correct ICMP message back to the internal sender. We synthesize the message based on the information in the extended error structure, plus the returned part of the original message body. Note that for the sake of completeness, we even produce ICMP messages for other error codes. We have noticed that at least ICMP_PROT_UNREACH is propagated as an error event back to the user. Signed-off-by: Jon Maloy <jmaloy(a)redhat.com> --- tap.c | 4 +-- tap.h | 2 ++ udp.c | 71 ++++++++++++++++++++++++++++++++++++++++++-------- udp_internal.h | 2 +- udp_vu.c | 4 +-- 5 files changed, 67 insertions(+), 16 deletions(-) diff --git a/tap.c b/tap.c index cd32a90..f83aac5 100644 --- a/tap.c +++ b/tap.c @@ -142,8 +142,8 @@ static void *tap_push_l2h(const struct ctx *c, void *buf, uint16_t proto) * * Return: pointer at which to write the packet's payload */ -static void *tap_push_ip4h(struct iphdr *ip4h, struct in_addr src, - struct in_addr dst, size_t l4len, uint8_t proto) +void *tap_push_ip4h(struct iphdr *ip4h, struct in_addr src, + struct in_addr dst, size_t l4len, uint8_t proto) { uint16_t l3len = l4len + sizeof(*ip4h); diff --git a/tap.h b/tap.h index dfbd8b9..3ba00c1 100644 --- a/tap.h +++ b/tap.h @@ -44,6 +44,8 @@ static inline void tap_hdr_update(struct tap_hdr *thdr, size_t l2len) thdr->vnet_len = htonl(l2len); } +void *tap_push_ip4h(struct iphdr *ip4h, struct in_addr src, + struct in_addr dst, size_t l4len, uint8_t proto); void tap_udp4_send(const struct ctx *c, struct in_addr src, in_port_t sport, struct in_addr dst, in_port_t dport, const void *in, size_t dlen); diff --git a/udp.c b/udp.c index 923cc38..4b800b6 100644 --- a/udp.c +++ b/udp.c @@ -87,6 +87,7 @@ #include <netinet/in.h> #include <netinet/ip.h> #include <netinet/udp.h> +#include <netinet/ip_icmp.h> #include <stdint.h> #include <stddef.h> #include <string.h> @@ -402,25 +403,70 @@ static void udp_tap_prepare(const struct mmsghdr *mmh, (*tap_iov)[UDP_IOV_PAYLOAD].iov_len = l4len; } +/** + * udp_send_conn_fail_icmp4() - Construct and send ICMP to local peer + * @c: Execution context + * @ee: Extended error descriptor + * @ref: epoll reference + * @iov: First bytes (max 8) of original UDP message body + */ +static void udp_send_conn_fail_icmp4(const struct ctx *c, + const struct sock_extended_err *ee, + union epoll_ref ref, + struct iovec *iov) +{ + flow_sidx_t tosidx = flow_sidx_opposite(ref.flowside); + const struct flowside *toside = flowside_at_sidx(tosidx); + struct in_addr oaddr = toside->oaddr.v4mapped.a4; + struct in_addr eaddr = toside->eaddr.v4mapped.a4; + struct iov_tail data = IOV_TAIL(iov, 1, 0); + struct { + struct icmphdr icmp4h; + struct iphdr ip4h; + struct udphdr uh; + char data[8]; + } msg; + size_t udplen = sizeof(msg.uh) + iov->iov_len; + size_t msglen = sizeof(msg.icmp4h) + sizeof(msg.ip4h) + udplen; + + memset(&msg, 0, sizeof(msg)); + msg.icmp4h.type = ee->ee_type; + msg.icmp4h.code = ee->ee_code; + tap_push_ip4h(&msg.ip4h, eaddr, oaddr, udplen, IPPROTO_UDP); + msg.uh.source = htons(toside->eport); + msg.uh.dest = htons(toside->oport); + msg.uh.len = htons(udplen); + csum_udp4(&msg.uh, oaddr, eaddr, &data); + memcpy(&msg.data, iov->iov_base, iov->iov_len); + tap_icmp4_send(c, oaddr, eaddr, &msg, msglen); +} + /** * udp_sock_recverr() - Receive and clear an error from a socket - * @s: Socket to receive from + * @c: Execution context + * @ref: epoll reference * * Return: 1 if error received and processed, 0 if no more errors in queue, < 0 * if there was an error reading the queue * * #syscalls recvmsg */ -static int udp_sock_recverr(int s) +static int udp_sock_recverr(const struct ctx *c, union epoll_ref ref) { const struct sock_extended_err *ee; const struct cmsghdr *hdr; char buf[CMSG_SPACE(sizeof(*ee))]; + char udp_data[8]; + int s = ref.fd; + struct iovec iov = { + .iov_base = udp_data, + .iov_len = sizeof(udp_data) + }; struct msghdr mh = { .msg_name = NULL, .msg_namelen = 0, - .msg_iov = NULL, - .msg_iovlen = 0, + .msg_iov = &iov, + .msg_iovlen = 1, .msg_control = buf, .msg_controllen = sizeof(buf), }; @@ -450,8 +496,10 @@ static int udp_sock_recverr(int s) } ee = (const struct sock_extended_err *)CMSG_DATA(hdr); - - /* TODO: When possible propagate and otherwise handle errors */ + if (ee->ee_type == ICMP_DEST_UNREACH) { + iov.iov_len = rc; + udp_send_conn_fail_icmp4(c, ee, ref, &iov); + } debug("%s error on UDP socket %i: %s", str_ee_origin(ee), s, strerror_(ee->ee_errno)); @@ -461,15 +509,16 @@ static int udp_sock_recverr(int s) /** * udp_sock_errs() - Process errors on a socket * @c: Execution context - * @s: Socket to receive from + * @ref: epoll reference * @events: epoll events bitmap * * Return: Number of errors handled, or < 0 if we have an unrecoverable error */ -int udp_sock_errs(const struct ctx *c, int s, uint32_t events) +int udp_sock_errs(const struct ctx *c, union epoll_ref ref, uint32_t events) { unsigned n_err = 0; socklen_t errlen; + int s = ref.fd; int rc, err; ASSERT(!c->no_udp); @@ -478,7 +527,7 @@ int udp_sock_errs(const struct ctx *c, int s, uint32_t events) return 0; /* Nothing to do */ /* Empty the error queue */ - while ((rc = udp_sock_recverr(s)) > 0) + while ((rc = udp_sock_recverr(c, ref)) > 0) n_err += rc; if (rc < 0) @@ -558,7 +607,7 @@ static void udp_buf_listen_sock_handler(const struct ctx *c, const socklen_t sasize = sizeof(udp_meta[0].s_in); int n, i; - if (udp_sock_errs(c, ref.fd, events) < 0) { + if (udp_sock_errs(c, ref, events) < 0) { err("UDP: Unrecoverable error on listening socket:" " (%s port %hu)", pif_name(ref.udp.pif), ref.udp.port); /* FIXME: what now? close/re-open socket? */ @@ -661,7 +710,7 @@ static void udp_buf_reply_sock_handler(const struct ctx *c, union epoll_ref ref, from_s = uflow->s[ref.flowside.sidei]; - if (udp_sock_errs(c, from_s, events) < 0) { + if (udp_sock_errs(c, ref, events) < 0) { flow_err(uflow, "Unrecoverable error on reply socket"); flow_err_details(uflow); udp_flow_close(c, uflow); diff --git a/udp_internal.h b/udp_internal.h index cc80e30..3b081f5 100644 --- a/udp_internal.h +++ b/udp_internal.h @@ -30,5 +30,5 @@ size_t udp_update_hdr4(struct iphdr *ip4h, struct udp_payload_t *bp, size_t udp_update_hdr6(struct ipv6hdr *ip6h, struct udp_payload_t *bp, const struct flowside *toside, size_t dlen, bool no_udp_csum); -int udp_sock_errs(const struct ctx *c, int s, uint32_t events); +int udp_sock_errs(const struct ctx *c, union epoll_ref ref, uint32_t events); #endif /* UDP_INTERNAL_H */ diff --git a/udp_vu.c b/udp_vu.c index 4123510..c26a223 100644 --- a/udp_vu.c +++ b/udp_vu.c @@ -227,7 +227,7 @@ void udp_vu_listen_sock_handler(const struct ctx *c, union epoll_ref ref, struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE]; int i; - if (udp_sock_errs(c, ref.fd, events) < 0) { + if (udp_sock_errs(c, ref, events) < 0) { err("UDP: Unrecoverable error on listening socket:" " (%s port %hu)", pif_name(ref.udp.pif), ref.udp.port); return; @@ -302,7 +302,7 @@ void udp_vu_reply_sock_handler(const struct ctx *c, union epoll_ref ref, ASSERT(!c->no_udp); - if (udp_sock_errs(c, from_s, events) < 0) { + if (udp_sock_errs(c, ref, events) < 0) { flow_err(uflow, "Unrecoverable error on reply socket"); flow_err_details(uflow); udp_flow_close(c, uflow); -- 2.48.0

3 7

[PATCH v21 0/5] State migration
by David Gibson 13 Feb '25

13 Feb '25

As discussed on IRC, I think we finally squashed, or at least worked around, the bugs with restoring the sendq. David Gibson (3): fixup: TCP_REPAIR_WINDOW before send unsent fixup?: Don't fail whole migration for errors restoring sndq debug Stefano Brivio (2): migrate: Migrate TCP flows test: Add migration tests contrib/selinux/passt.te | 4 +- flow.c | 260 ++++++++++ flow.h | 8 + migrate.c | 19 + passt.c | 6 +- repair.c | 1 - tcp.c | 919 ++++++++++++++++++++++++++++++++++++ tcp_conn.h | 103 ++++ test/lib/layout | 55 ++- test/lib/setup | 138 +++++- test/lib/test | 48 ++ test/migrate/basic | 59 +++ test/migrate/bidirectional | 64 +++ test/migrate/iperf3_bidir6 | 58 +++ test/migrate/iperf3_in4 | 50 ++ test/migrate/iperf3_in6 | 58 +++ test/migrate/iperf3_out4 | 50 ++ test/migrate/iperf3_out6 | 58 +++ test/migrate/rampstream_in | 10 +- test/migrate/rampstream_out | 6 +- test/run | 46 +- 21 files changed, 1991 insertions(+), 29 deletions(-) create mode 100644 test/migrate/basic create mode 100644 test/migrate/bidirectional create mode 100644 test/migrate/iperf3_bidir6 create mode 100644 test/migrate/iperf3_in4 create mode 100644 test/migrate/iperf3_in6 create mode 100644 test/migrate/iperf3_out4 create mode 100644 test/migrate/iperf3_out6 -- 2.48.1

1 5

[PATCH v20 0/5] State migration
by David Gibson 13 Feb '25

13 Feb '25

Has some temporary hacks for debugging problems with migrate/iperf3_bidir6. Has a couple of tentative fixes for possible bugs, but it's proving difficult to characterize the errors. David Gibson (3): fixup?: loop on EGAIN in repair queue depause fixup: TCP_REPAIR_WINDOW before send unsent Stefano Brivio (2): migrate: Migrate TCP flows test: Add migration tests contrib/selinux/passt.te | 4 +- flow.c | 260 ++++++++++ flow.h | 8 + migrate.c | 19 + passt.c | 6 +- repair.c | 1 - tcp.c | 915 ++++++++++++++++++++++++++++++++++++ tcp_conn.h | 103 ++++ test/lib/layout | 55 ++- test/lib/setup | 138 +++++- test/lib/test | 48 ++ test/migrate/basic | 59 +++ test/migrate/bidirectional | 64 +++ test/migrate/iperf3_bidir6 | 58 +++ test/migrate/iperf3_in4 | 50 ++ test/migrate/iperf3_in6 | 58 +++ test/migrate/iperf3_out4 | 50 ++ test/migrate/iperf3_out6 | 58 +++ test/migrate/rampstream_in | 10 +- test/migrate/rampstream_out | 6 +- test/run | 46 +- 21 files changed, 1987 insertions(+), 29 deletions(-) create mode 100644 test/migrate/basic create mode 100644 test/migrate/bidirectional create mode 100644 test/migrate/iperf3_bidir6 create mode 100644 test/migrate/iperf3_in4 create mode 100644 test/migrate/iperf3_in6 create mode 100644 test/migrate/iperf3_out4 create mode 100644 test/migrate/iperf3_out6 -- 2.48.1

1 5

[PATCH] selinux: Add rules needed to run tests
by Stefano Brivio 13 Feb '25

13 Feb '25

...other than being convenient, they might be reasonably representative of typical stand-alone usage. Signed-off-by: Stefano Brivio <sbrivio(a)redhat.com> --- contrib/selinux/passt.te | 4 ++++ contrib/selinux/pasta.te | 14 ++++++++++++-- 2 files changed, 16 insertions(+), 2 deletions(-) diff --git a/contrib/selinux/passt.te b/contrib/selinux/passt.te index c6cea34..6e7a4cb 100644 --- a/contrib/selinux/passt.te +++ b/contrib/selinux/passt.te @@ -20,6 +20,7 @@ require { type fs_t; type tmp_t; type user_tmp_t; + type user_home_t; type tmpfs_t; type root_t; @@ -80,6 +81,9 @@ allow passt_t root_t:dir mounton; allow passt_t tmp_t:dir { add_name mounton remove_name write }; allow passt_t tmpfs_t:filesystem mount; allow passt_t fs_t:filesystem unmount; +allow passt_t user_home_t:dir search; +allow passt_t user_tmp_t:fifo_file append; +allow passt_t user_tmp_t:file map; manage_files_pattern(passt_t, user_tmp_t, user_tmp_t) files_pid_filetrans(passt_t, user_tmp_t, file) diff --git a/contrib/selinux/pasta.te b/contrib/selinux/pasta.te index d0ff0cc..89c8043 100644 --- a/contrib/selinux/pasta.te +++ b/contrib/selinux/pasta.te @@ -18,6 +18,7 @@ require { type bin_t; type user_home_t; type user_home_dir_t; + type user_tmp_t; type fs_t; type tmp_t; type tmpfs_t; @@ -56,8 +57,10 @@ require { attribute port_type; type port_t; type http_port_t; + type http_cache_port_t; type ssh_port_t; type reserved_port_t; + type unreserved_port_t; type dns_port_t; type dhcpc_port_t; type chronyd_port_t; @@ -122,8 +125,8 @@ domain_auto_trans(pasta_t, ping_exec_t, ping_t); allow pasta_t nsfs_t:file { open read }; -allow pasta_t user_home_t:dir getattr; -allow pasta_t user_home_t:file { open read getattr setattr }; +allow pasta_t user_home_t:dir { getattr search }; +allow pasta_t user_home_t:file { open read getattr setattr execute execute_no_trans map}; allow pasta_t user_home_dir_t:dir { search getattr open add_name read write }; allow pasta_t user_home_dir_t:file { create open read write }; allow pasta_t tmp_t:dir { add_name mounton remove_name write }; @@ -133,6 +136,11 @@ allow pasta_t root_t:dir mounton; manage_files_pattern(pasta_t, pasta_pid_t, pasta_pid_t) files_pid_filetrans(pasta_t, pasta_pid_t, file) +allow pasta_t user_tmp_t:dir { add_name remove_name search write }; +allow pasta_t user_tmp_t:fifo_file append; +allow pasta_t user_tmp_t:file { create open write }; +allow pasta_t user_tmp_t:sock_file { create unlink }; + allow pasta_t console_device_t:chr_file { open write getattr ioctl }; allow pasta_t user_devpts_t:chr_file { getattr read write ioctl }; logging_send_syslog_msg(pasta_t) @@ -160,6 +168,8 @@ allow pasta_t self:udp_socket create_stream_socket_perms; allow pasta_t reserved_port_t:udp_socket name_bind; allow pasta_t llmnr_port_t:tcp_socket name_bind; allow pasta_t llmnr_port_t:udp_socket name_bind; +allow pasta_t http_cache_port_t:tcp_socket { name_bind name_connect }; +allow pasta_t unreserved_port_t:udp_socket name_bind; corenet_udp_sendrecv_generic_node(pasta_t) corenet_udp_bind_generic_node(pasta_t) allow pasta_t node_t:icmp_socket { name_bind node_bind }; -- 2.43.0

1 0

[PATCH v19 0/2] A-bit-less-draft state migration
by Stefano Brivio 13 Feb '25

13 Feb '25

Compared to v18: - add save and restore of permitted SACK and timestamps - return ENOBUFS if pending queues are too large, not ENOMEM - print SO_PEEK_OFF state from v18 9/9 - fix headers of rampstream_in and rampstream_out tests - filter out non-established TCP flows - fix SELinux rule for queue ioctl(): it's getattr, not ioctl Still missing: - reasonable error handling, and "empty flows" if target expects them but we fail to fetch them - tests with half-closed sockets Stefano Brivio (2): migrate: Migrate TCP flows test: Add migration tests contrib/selinux/passt.te | 4 +- flow.c | 260 ++++++++++ flow.h | 8 + migrate.c | 19 + passt.c | 6 +- repair.c | 1 - tcp.c | 916 ++++++++++++++++++++++++++++++++++++ tcp_conn.h | 103 ++++ test/lib/layout | 55 ++- test/lib/setup | 138 +++++- test/lib/test | 48 ++ test/migrate/basic | 59 +++ test/migrate/bidirectional | 64 +++ test/migrate/iperf3_bidir6 | 58 +++ test/migrate/iperf3_in4 | 50 ++ test/migrate/iperf3_in6 | 58 +++ test/migrate/iperf3_out4 | 50 ++ test/migrate/iperf3_out6 | 58 +++ test/migrate/rampstream_in | 10 +- test/migrate/rampstream_out | 6 +- test/run | 29 ++ 21 files changed, 1984 insertions(+), 16 deletions(-) create mode 100644 test/migrate/basic create mode 100644 test/migrate/bidirectional create mode 100644 test/migrate/iperf3_bidir6 create mode 100644 test/migrate/iperf3_in4 create mode 100644 test/migrate/iperf3_in6 create mode 100644 test/migrate/iperf3_out4 create mode 100644 test/migrate/iperf3_out6 -- 2.43.0

1 2

[PATCH v18 0/9] State migration (kinda draft again)
by David Gibson 12 Feb '25

12 Feb '25

More debugging of bugs found with the rampstream test today. * Send queue transfer bug I spotted by inspection a nasty bug which would mean we never properly transfer the send queue (in repair mode we read it into the the wrong buffer, then transferred the right one). I think we only didn't hit this because in each of the cases I've seen the send queue has been empty. I think that makes sense: with high speed local transfers, there's probably more than enough time between when the guest is stopped and when we dump the queue for the sndbuf to drain completely to the peer. * Queue transfer error checking I've made the error checking when extracting and reloading the queue a bit more robust. * rampstream corruption bug I found the cause of the stream corruption bug. I don't think repair mode fully supports SO_PEEK_OFF semantics, but apparently it shares enough code with the normal recv() path that the peek offset applied, which meant (I think) we skipped the "already sent" portion of the rcv queue when dumping it. I've fixed this by disabling SO_PEEK_OFF (setting it to -1) before migration on the source. This needs some fixing to deal correctly with the case of a failed migration which resumes on the source. * rampstream unexpected EOF bug Unfortunately that wasn't the only bug. With the peek offset fixed, I no longer get stream corruption, but I still get unexpected EOFs on the rampstream_in test (at least with 64M [rw]mem_max). The receiving rampstream is getting an EOF because passt is sending an RST to the guest 14.1750: Flow 0 (TCP connection): TCP reset at tcp_sock_handler:2270 That happens because we get an EPOLLERR on the socket at some point after migration. From some earlier debugging hacks, I think that's an ECONNRESET specifically, but I haven't debugged further because I was focused on the corruption bug. David Gibson (3): migrate: Migrate guest observed addresses rampstream: Add utility to test for corruption of data streams debug Stefano Brivio (6): migrate: Skeleton of live migration logic Add interfaces and configuration bits for passt-repair vhost_user: Make source quit after reporting migration state tcp: Get bound address for connected inbound sockets too migrate: Migrate TCP flows test: Add migration tests Makefile | 14 +- conf.c | 43 +- contrib/selinux/passt.te | 2 +- epoll_type.h | 6 +- flow.c | 259 +++++++++- flow.h | 8 + flow_table.h | 6 +- migrate.c | 309 ++++++++++++ migrate.h | 51 ++ passt.1 | 11 + passt.c | 21 +- passt.h | 15 + repair.c | 218 +++++++++ repair.h | 16 + tap.c | 65 +-- tcp.c | 921 +++++++++++++++++++++++++++++++++++- tcp_conn.h | 99 ++++ test/.gitignore | 1 + test/Makefile | 5 +- test/lib/layout | 55 ++- test/lib/setup | 140 +++++- test/lib/test | 48 ++ test/migrate/basic | 59 +++ test/migrate/bidirectional | 64 +++ test/migrate/iperf3_bidir6 | 58 +++ test/migrate/iperf3_in4 | 50 ++ test/migrate/iperf3_in6 | 58 +++ test/migrate/iperf3_out4 | 50 ++ test/migrate/iperf3_out6 | 58 +++ test/migrate/rampstream_in | 60 +++ test/migrate/rampstream_out | 55 +++ test/passt.mbuto | 5 +- test/rampstream-check.sh | 3 + test/rampstream.c | 143 ++++++ test/run | 29 ++ util.c | 62 +++ util.h | 30 ++ vhost_user.c | 67 +-- virtio.h | 4 - vu_common.c | 49 +- vu_common.h | 2 +- 41 files changed, 3014 insertions(+), 205 deletions(-) create mode 100644 migrate.c create mode 100644 migrate.h create mode 100644 repair.c create mode 100644 repair.h create mode 100644 test/migrate/basic create mode 100644 test/migrate/bidirectional create mode 100644 test/migrate/iperf3_bidir6 create mode 100644 test/migrate/iperf3_in4 create mode 100644 test/migrate/iperf3_in6 create mode 100644 test/migrate/iperf3_out4 create mode 100644 test/migrate/iperf3_out6 create mode 100644 test/migrate/rampstream_in create mode 100644 test/migrate/rampstream_out create mode 100755 test/rampstream-check.sh create mode 100644 test/rampstream.c -- 2.48.1

2 10

[PATCH] tcp_vu: Fix off-by one in header count array adjustment
by Stefano Brivio 12 Feb '25

12 Feb '25

From: Laurent Vivier <lvivier(a)redhat.com> Not entirely clear to me why, but Laurent proposed this patch to fix an issue were we would end up with a zero buf_cnt in tcp_vu_data_from_sock() and all sorts of weirdnesses. Reported-by: David Gibson <david(a)gibson.dropbear.id.au> [sbrivio: commit message, albeit not really descriptive] Signed-off-by: Stefano Brivio <sbrivio(a)redhat.com> --- tcp_vu.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tcp_vu.c b/tcp_vu.c index fad7065..0622f17 100644 --- a/tcp_vu.c +++ b/tcp_vu.c @@ -261,7 +261,7 @@ static ssize_t tcp_vu_sock_recv(const struct ctx *c, len -= iov->iov_len; } /* adjust head count */ - while (head_cnt > 0 && head[head_cnt - 1] > i) + while (head_cnt > 0 && head[head_cnt - 1] >= i) head_cnt--; /* mark end of array */ head[head_cnt] = i; -- 2.43.0

2 2

[PATCH] tcp: Don't discard window information on keep-alive segments
by Stefano Brivio 12 Feb '25

12 Feb '25

It looks like a detail, but it's critical if we're dealing with somebody, such as near-future self, using TCP_REPAIR to migrate TCP connections in the guest or container. The last packet sent from the 'source' process/guest/container typically reports a small window, or zero, because the guest/container hadn't been draining it for a while. The next packet, appearing as the target sets TCP_REPAIR_OFF on the migrated socket, is a keep-alive (also called "window probe" in CRIU or TCP_REPAIR-related code), and it comes with an updated window value, reflecting the pre-migration "regular" value. If we ignore it, it might take a while/forever before we realise we can actually restart sending. Fixes: 238c69f9af45 ("tcp: Acknowledge keep-alive segments, ignore them for the rest") Signed-off-by: Stefano Brivio <sbrivio(a)redhat.com> --- tcp.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/tcp.c b/tcp.c index af6bd95..2addf4a 100644 --- a/tcp.c +++ b/tcp.c @@ -1664,8 +1664,10 @@ static int tcp_data_from_tap(const struct ctx *c, struct tcp_tap_conn *conn, tcp_send_flag(c, conn, ACK); tcp_timer_ctl(c, conn); - if (p->count == 1) + if (p->count == 1) { + tcp_tap_window_update(conn, ntohs(th->window)); return 1; + } continue; } -- 2.43.0

2 2