[PATCH v3 0/3] Probe host's ephemeral ports, rather than using RFC values
As discussed on our recent call, this implements basing which ports we consider "ephemeral" on probing the host's settings, rather than just assuming the RFC 6335 recommended values, which are not what Linux uses by default. I think this is more correct, but additionally using the Linux values means we consider more ports ephemeral, reducing kernel memory consumption for -t all -u all. Changes in v3: * Used in_port_t instead of plan uint16_t * Considered using sscanf() rather than strchr() + strtol(), but decided against it. I can never remember exactly what is and isn't accepted by scanf(), plus clang-tidy complained about it. Changes in v2: * Add missing close() for the sysctl file David Gibson (3): conf, fwd: Make ephemeral port logic more flexible conf, fwd: Don't attempt to forward port 0 fwd, conf: Probe host's ephemeral ports conf.c | 19 +++++++++++---- fwd.c | 74 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ fwd.h | 3 +++ util.h | 3 --- 4 files changed, 92 insertions(+), 7 deletions(-) -- 2.46.0
"Ephemeral" ports are those which the kernel may allocate as local
port numbers for outgoing connections or datagrams. Because of that,
they're generally not good choices for listening servers to bind to.
Thefore when using -t all, -u all or exclude-only ranges, we map only
non-ephemeral ports. Our logic for this is a bit rigid though: we
assume the ephemeral ports are always a fixed range at the top of the
port number space. We also assume PORT_EPHEMERAL_MIN is a multiple of
8, or we won't set the forward bitmap correctly.
Make the logic in conf.c more flexible, using a helper moved into
fwd.[ch], although we don't change which ports we consider ephemeral
(yet).
The new handling is undoubtedly more computationally expensive, but
since it's a once-off operation at start off, I don't think it really
matters.
Signed-off-by: David Gibson
When using -t all, -u all or exclude-only ranges, we'll attempt to forward
all non-ephemeral port numbers, including port 0. However, this won't work
as intended: bind() treats a zero port not as literal port 0, but as
"pick a port for me". Because of the special meaning of port 0, we mostly
outright exclude it in our handling.
Do the same for setting up forwards, not attempting to forward for port 0.
Signed-off-by: David Gibson
When we forward "all" ports (-t all or -u all), or use an exclude-only
range, we don't actually forward *all* ports - that wouln't leave local
ports to use for outgoing connections. Rather we forward all non-ephemeral
ports - those that won't be used for outgoing connections or datagrams.
Currently we assume the range of ephemeral ports is that recommended by
RFC 6335, 49152-65535. However, that's not the range used by default on
Linux, 32768-60999 but configurable with the net.ipv4.ip_local_port_range
sysctl.
We can't really know what range the guest will consider ephemeral, but if
it differs too much from the host it's likely to cause problems we can't
avoid anyway. So, using the host's ephemeral range is a better guess than
using the RFC 6335 range.
Therefore, add logic to probe the host's ephemeral range, falling back to
the RFC 6335 range if that fails. This has the bonus advantage of
reducing the number of ports bound by -t all -u all on most Linux machines
thereby reducing kernel memory usage. Specifically this reduces kernel
memory usage with -t all -u all from ~380MiB to ~289MiB.
Signed-off-by: David Gibson
On 29/08/2024 11:58, David Gibson wrote:
When we forward "all" ports (-t all or -u all), or use an exclude-only range, we don't actually forward *all* ports - that wouln't leave local ports to use for outgoing connections. Rather we forward all non-ephemeral ports - those that won't be used for outgoing connections or datagrams.
Currently we assume the range of ephemeral ports is that recommended by RFC 6335, 49152-65535. However, that's not the range used by default on Linux, 32768-60999 but configurable with the net.ipv4.ip_local_port_range sysctl.
We can't really know what range the guest will consider ephemeral, but if it differs too much from the host it's likely to cause problems we can't avoid anyway. So, using the host's ephemeral range is a better guess than using the RFC 6335 range.
Therefore, add logic to probe the host's ephemeral range, falling back to the RFC 6335 range if that fails. This has the bonus advantage of reducing the number of ports bound by -t all -u all on most Linux machines thereby reducing kernel memory usage. Specifically this reduces kernel memory usage with -t all -u all from ~380MiB to ~289MiB.
Signed-off-by: David Gibson
--- conf.c | 1 + fwd.c | 61 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++-- fwd.h | 1 + 3 files changed, 61 insertions(+), 2 deletions(-) diff --git a/conf.c b/conf.c index 3eb117ff..b2758864 100644 --- a/conf.c +++ b/conf.c @@ -1721,6 +1721,7 @@ void conf(struct ctx *c, int argc, char **argv) /* Inbound port options & DNS can be parsed now (after IPv4/IPv6 * settings) */ + fwd_probe_ephemeral(); udp_portmap_clear(); optind = 0; do { diff --git a/fwd.c b/fwd.c index 8fa312a7..a5050980 100644 --- a/fwd.c +++ b/fwd.c @@ -28,8 +28,65 @@ #include "flow_table.h"
/* Empheral port range: values from RFC 6335 */ -static const in_port_t fwd_ephemeral_min = (1 << 15) + (1 << 14); -static const in_port_t fwd_ephemeral_max = NUM_PORTS - 1; +static in_port_t fwd_ephemeral_min = (1 << 15) + (1 << 14); +static in_port_t fwd_ephemeral_max = NUM_PORTS - 1; + +#define PORT_RANGE_SYSCTL "/proc/sys/net/ipv4/ip_local_port_range" + +/** fwd_probe_ephemeral() - Determine what ports this host considers ephemeral + * + * Work out what ports the host thinks are emphemeral and record it for later + * use by fwd_port_is_ephemeral(). If we're unable to probe, assume the range + * recommended by RFC 6335. + */ +void fwd_probe_ephemeral(void) +{ + char *line, *tab, *end; + struct lineread lr; + long min, max; + ssize_t len; + int fd; + + fd = open(PORT_RANGE_SYSCTL, O_RDONLY | O_CLOEXEC); + if (fd < 0) { + warn_perror("Unable to open %s", PORT_RANGE_SYSCTL); + return; + } + + lineread_init(&lr, fd); + len = lineread_get(&lr, &line); + close(fd); + + if (len < 0) + goto parse_err; + + tab = strchr(line, '\t'); + if (!tab) + goto parse_err; + *tab = '\0'; + + errno = 0; + min = strtol(line, &end, 10); + if (*end || errno) + goto parse_err; + + errno = 0; + max = strtol(tab + 1, &end, 10); + if (*end || errno) + goto parse_err; + + if (min < 0 || min >= NUM_PORTS || + max < 0 || max >= NUM_PORTS) + goto parse_err; + + fwd_ephemeral_min = min; + fwd_ephemeral_max = max; + + return; + +parse_err: + warn("Unable to parse %s", PORT_RANGE_SYSCTL); +}
/** * fwd_port_is_ephemeral() - Is port number ephemeral? diff --git a/fwd.h b/fwd.h index 99dd66cf..3562f3ca 100644 --- a/fwd.h +++ b/fwd.h @@ -12,6 +12,7 @@ struct flowside; /* Number of ports for both TCP and UDP */ #define NUM_PORTS (1U << 16)
+void fwd_probe_ephemeral(void); bool fwd_port_is_ephemeral(in_port_t port);
enum fwd_ports_mode {
Reviewed-by: Laurent Vivier
On Thu, 29 Aug 2024 19:58:44 +1000
David Gibson
As discussed on our recent call, this implements basing which ports we consider "ephemeral" on probing the host's settings, rather than just assuming the RFC 6335 recommended values, which are not what Linux uses by default.
I think this is more correct, but additionally using the Linux values means we consider more ports ephemeral, reducing kernel memory consumption for -t all -u all.
Changes in v3: * Used in_port_t instead of plan uint16_t * Considered using sscanf() rather than strchr() + strtol(), but decided against it. I can never remember exactly what is and isn't accepted by scanf(), plus clang-tidy complained about it. Changes in v2: * Add missing close() for the sysctl file
Applied. -- Stefano
participants (3)
-
David Gibson
-
Laurent Vivier
-
Stefano Brivio