When pasta periodically scans bound ports and binds them on the other side in order to forward traffic, we bind UDP ports for corresponding TCP port numbers, too, to support protocols and applications such as iperf3 which use UDP port numbers matching the ones used by the TCP data connection. If we scan UDP ports in order to bind UDP ports, we skip detection of the UDP ports we already bound ourselves, to avoid looping back our own ports. Same with scanning and binding TCP ports. But if we scan for TCP ports in order to bind UDP ports, we need to skip bound TCP ports too, otherwise, as David pointed out: - we find a bound TCP port on side A, and bind the corresponding TCP and UDP ports on side B - at the next periodic scan, we find that UDP port bound on side B, and we bind the corresponding UDP port on side A - at this point, we unbind that UDP port on side B: we would otherwise loop back our own port. To fix this, we need to avoid binding UDP ports that we already bound, on the other side, as a consequence of finding a corresponding bound TCP port. Reproducing this issue is straightforward: ./pasta -- iperf3 -s # Wait one second, then from another terminal: iperf3 -c ::1 -u Reported-by: Akihiro Suda <akihiro.suda.cz(a)hco.ntt.co.jp> Analysed-by: David Gibson <david(a)gibson.dropbear.id.au> Fixes: 457ff122e33c ("udp,pasta: Periodically scan for ports to automatically forward") Signed-off-by: Stefano Brivio <sbrivio(a)redhat.com> --- v2: Include changes for udp.c port_fwd.c | 27 +++++++++++++++++++-------- port_fwd.h | 3 ++- udp.c | 4 ++-- util.c | 21 +++++++++++++++++++++ util.h | 1 + 5 files changed, 45 insertions(+), 11 deletions(-) diff --git a/port_fwd.c b/port_fwd.c index eac8d2f..91b021b 100644 --- a/port_fwd.c +++ b/port_fwd.c @@ -86,22 +86,33 @@ void port_fwd_scan_tcp(struct port_fwd *fwd, const struct port_fwd *rev) * port_fwd_scan_tcp() - Scan /proc to update TCP forwarding map * @fwd: Forwarding information to update * @rev: Forwarding information for the reverse direction - * @tcp: Corresponding TCP forwarding information + * @tcp_fwd: Corresponding TCP forwarding information + * @tcp_rev: TCP forwarding information for the reverse direction */ void port_fwd_scan_udp(struct port_fwd *fwd, const struct port_fwd *rev, - const struct port_fwd *tcp) + const struct port_fwd *tcp_fwd, + const struct port_fwd *tcp_rev) { + uint8_t exclude[PORT_BITMAP_SIZE]; + + bitmap_sum(exclude, PORT_BITMAP_SIZE, rev->map, tcp_rev->map); + memset(fwd->map, 0, PORT_BITMAP_SIZE); - procfs_scan_listen(fwd->scan4, UDP_LISTEN, fwd->map, rev->map); - procfs_scan_listen(fwd->scan6, UDP_LISTEN, fwd->map, rev->map); + procfs_scan_listen(fwd->scan4, UDP_LISTEN, fwd->map, exclude); + procfs_scan_listen(fwd->scan6, UDP_LISTEN, fwd->map, exclude); /* Also forward UDP ports with the same numbers as bound TCP ports. * This is useful for a handful of protocols (e.g. iperf3) where a TCP * control port is used to set up transfers on a corresponding UDP * port. + * + * This means we need to skip numbers of TCP ports bound on the other + * side, too. Otherwise, we would detect corresponding UDP ports as + * bound and try to forward them from the opposite side, but it's + * already us handling them. */ - procfs_scan_listen(tcp->scan4, TCP_LISTEN, fwd->map, rev->map); - procfs_scan_listen(tcp->scan6, TCP_LISTEN, fwd->map, rev->map); + procfs_scan_listen(tcp_fwd->scan4, TCP_LISTEN, fwd->map, exclude); + procfs_scan_listen(tcp_fwd->scan6, TCP_LISTEN, fwd->map, exclude); } /** @@ -126,7 +137,7 @@ void port_fwd_init(struct ctx *c) c->udp.fwd_in.f.scan4 = open_in_ns(c, "/proc/net/udp", flags); c->udp.fwd_in.f.scan6 = open_in_ns(c, "/proc/net/udp6", flags); port_fwd_scan_udp(&c->udp.fwd_in.f, &c->udp.fwd_out.f, - &c->tcp.fwd_in); + &c->tcp.fwd_in, &c->tcp.fwd_out); } if (c->tcp.fwd_out.mode == FWD_AUTO) { c->tcp.fwd_out.scan4 = open("/proc/net/tcp", flags); @@ -137,6 +148,6 @@ void port_fwd_init(struct ctx *c) c->udp.fwd_out.f.scan4 = open("/proc/net/udp", flags); c->udp.fwd_out.f.scan6 = open("/proc/net/udp6", flags); port_fwd_scan_udp(&c->udp.fwd_out.f, &c->udp.fwd_in.f, - &c->tcp.fwd_out); + &c->tcp.fwd_out, &c->tcp.fwd_in); } } diff --git a/port_fwd.h b/port_fwd.h index 8a823d8..f6bf1b5 100644 --- a/port_fwd.h +++ b/port_fwd.h @@ -37,7 +37,8 @@ struct port_fwd { void port_fwd_scan_tcp(struct port_fwd *fwd, const struct port_fwd *rev); void port_fwd_scan_udp(struct port_fwd *fwd, const struct port_fwd *rev, - const struct port_fwd *tcp); + const struct port_fwd *tcp_fwd, + const struct port_fwd *tcp_rev); void port_fwd_init(struct ctx *c); #endif /* PORT_FWD_H */ diff --git a/udp.c b/udp.c index cc1ea9c..1f8c306 100644 --- a/udp.c +++ b/udp.c @@ -1260,13 +1260,13 @@ void udp_timer(struct ctx *c, const struct timespec *ts) if (c->mode == MODE_PASTA) { if (c->udp.fwd_out.f.mode == FWD_AUTO) { port_fwd_scan_udp(&c->udp.fwd_out.f, &c->udp.fwd_in.f, - &c->tcp.fwd_out); + &c->tcp.fwd_out, &c->tcp.fwd_in); NS_CALL(udp_port_rebind_outbound, c); } if (c->udp.fwd_in.f.mode == FWD_AUTO) { port_fwd_scan_udp(&c->udp.fwd_in.f, &c->udp.fwd_out.f, - &c->tcp.fwd_in); + &c->tcp.fwd_in, &c->tcp.fwd_out); udp_port_rebind(c, false); } } diff --git a/util.c b/util.c index c38ab7e..d296fd9 100644 --- a/util.c +++ b/util.c @@ -325,6 +325,27 @@ int bitmap_isset(const uint8_t *map, int bit) return !!(*word & BITMAP_BIT(bit)); } +/** + * bitmap_sum() - Sum (logic or) of two bitmaps + * @dst: Pointer to result bitmap + * @size: Size of bitmaps, in bytes + * @a: First operand + * @b: Second operand + */ +void bitmap_sum(uint8_t *dst, size_t size, const uint8_t *a, const uint8_t *b) +{ + unsigned long *dw = (unsigned long *)dst; + unsigned long *aw = (unsigned long *)a; + unsigned long *bw = (unsigned long *)b; + size_t i; + + for (i = 0; i < size / sizeof(long); i++, dw++, aw++, bw++) + *dw = *aw | *bw; + + for (i = size / sizeof(long) * sizeof(long); i < size; i++) + dst[i] = a[i] | b[i]; +} + /* * ns_enter() - Enter configured user (unless already joined) and network ns * @c: Execution context diff --git a/util.h b/util.h index 78a8fb2..1f1b06a 100644 --- a/util.h +++ b/util.h @@ -216,6 +216,7 @@ int timespec_diff_ms(const struct timespec *a, const struct timespec *b); void bitmap_set(uint8_t *map, int bit); void bitmap_clear(uint8_t *map, int bit); int bitmap_isset(const uint8_t *map, int bit); +void bitmap_sum(uint8_t *dst, size_t size, const uint8_t *a, const uint8_t *b); char *line_read(char *buf, size_t len, int fd); void ns_enter(const struct ctx *c); bool ns_is_init(void); -- 2.39.2
On Tue, Nov 21, 2023 at 07:01:52PM +0100, Stefano Brivio wrote:When pasta periodically scans bound ports and binds them on the other side in order to forward traffic, we bind UDP ports for corresponding TCP port numbers, too, to support protocols and applications such as iperf3 which use UDP port numbers matching the ones used by the TCP data connection. If we scan UDP ports in order to bind UDP ports, we skip detection of the UDP ports we already bound ourselves, to avoid looping back our own ports. Same with scanning and binding TCP ports. But if we scan for TCP ports in order to bind UDP ports, we need to skip bound TCP ports too, otherwise, as David pointed out: - we find a bound TCP port on side A, and bind the corresponding TCP and UDP ports on side B - at the next periodic scan, we find that UDP port bound on side B, and we bind the corresponding UDP port on side A - at this point, we unbind that UDP port on side B: we would otherwise loop back our own port. To fix this, we need to avoid binding UDP ports that we already bound, on the other side, as a consequence of finding a corresponding bound TCP port. Reproducing this issue is straightforward: ./pasta -- iperf3 -s # Wait one second, then from another terminal: iperf3 -c ::1 -u Reported-by: Akihiro Suda <akihiro.suda.cz(a)hco.ntt.co.jp> Analysed-by: David Gibson <david(a)gibson.dropbear.id.au> Fixes: 457ff122e33c ("udp,pasta: Periodically scan for ports to automatically forward") Signed-off-by: Stefano Brivio <sbrivio(a)redhat.com>LGTM, except for one nit: [...]+/** + * bitmap_sum() - Sum (logic or) of two bitmapsI don't like the name bitmap_sum() since this isn't really an addable object in the usual sense. I'd prefer either bitmap_or() (thinking of the bits as bits) or bitmap_union() (thinking of the bitmaps as sets).+ * @dst: Pointer to result bitmap + * @size: Size of bitmaps, in bytes + * @a: First operand + * @b: Second operand + */ +void bitmap_sum(uint8_t *dst, size_t size, const uint8_t *a, const uint8_t *b) +{ + unsigned long *dw = (unsigned long *)dst; + unsigned long *aw = (unsigned long *)a; + unsigned long *bw = (unsigned long *)b; + size_t i; + + for (i = 0; i < size / sizeof(long); i++, dw++, aw++, bw++) + *dw = *aw | *bw; + + for (i = size / sizeof(long) * sizeof(long); i < size; i++) + dst[i] = a[i] | b[i]; +} + /* * ns_enter() - Enter configured user (unless already joined) and network ns * @c: Execution context diff --git a/util.h b/util.h index 78a8fb2..1f1b06a 100644 --- a/util.h +++ b/util.h @@ -216,6 +216,7 @@ int timespec_diff_ms(const struct timespec *a, const struct timespec *b); void bitmap_set(uint8_t *map, int bit); void bitmap_clear(uint8_t *map, int bit); int bitmap_isset(const uint8_t *map, int bit); +void bitmap_sum(uint8_t *dst, size_t size, const uint8_t *a, const uint8_t *b); char *line_read(char *buf, size_t len, int fd); void ns_enter(const struct ctx *c); bool ns_is_init(void);-- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson
On Wed, 22 Nov 2023 10:58:24 +1100 David Gibson <david(a)gibson.dropbear.id.au> wrote:On Tue, Nov 21, 2023 at 07:01:52PM +0100, Stefano Brivio wrote:Hmm, yes, I went with "sum" in the sense of logical addition -- not that it really "sums" the bitmaps. But then...When pasta periodically scans bound ports and binds them on the other side in order to forward traffic, we bind UDP ports for corresponding TCP port numbers, too, to support protocols and applications such as iperf3 which use UDP port numbers matching the ones used by the TCP data connection. If we scan UDP ports in order to bind UDP ports, we skip detection of the UDP ports we already bound ourselves, to avoid looping back our own ports. Same with scanning and binding TCP ports. But if we scan for TCP ports in order to bind UDP ports, we need to skip bound TCP ports too, otherwise, as David pointed out: - we find a bound TCP port on side A, and bind the corresponding TCP and UDP ports on side B - at the next periodic scan, we find that UDP port bound on side B, and we bind the corresponding UDP port on side A - at this point, we unbind that UDP port on side B: we would otherwise loop back our own port. To fix this, we need to avoid binding UDP ports that we already bound, on the other side, as a consequence of finding a corresponding bound TCP port. Reproducing this issue is straightforward: ./pasta -- iperf3 -s # Wait one second, then from another terminal: iperf3 -c ::1 -u Reported-by: Akihiro Suda <akihiro.suda.cz(a)hco.ntt.co.jp> Analysed-by: David Gibson <david(a)gibson.dropbear.id.au> Fixes: 457ff122e33c ("udp,pasta: Periodically scan for ports to automatically forward") Signed-off-by: Stefano Brivio <sbrivio(a)redhat.com>LGTM, except for one nit: [...]+/** + * bitmap_sum() - Sum (logic or) of two bitmapsI don't like the name bitmap_sum() since this isn't really an addable object in the usual sense.I'd prefer either bitmap_or() (thinking of the bits as bits) or bitmap_union() (thinking of the bitmaps as sets).I could just use "_or" like the Linux kernel does. Respinning. -- Stefano