Re: [PATCH net] ipvs: reject negative sync_refresh_period values

From: Julian Anastasov

Date: Mon Nov 02 2026 - 09:16:58 EST



Hello,

On Mon, 2 Nov 2026, Yingjie Wang wrote:

> net.ipv4.vs.sync_refresh_period is documented as an unsigned number of
> seconds. proc_dointvec_jiffies() accepts a signed integer and converts
> it to jiffies before storing it in an unsigned int field. Consequently,
> writing -1 stores UINT_MAX - HZ + 1 jiffies and bypasses the finite
> bound applied to ordinary positive writes.
>
> This changes the behavior of ip_vs_sync_conn_needed(). With
> sync_threshold="0 1" (threshold 0, period 1) and
> sync_refresh_period=-1, only the first of twelve packets requests a
> sync message; the zero-refresh control requests all twelve.
>
> Store the value in seconds, as documented, and convert it at the single
> consumer. Use proc_douintvec_minmax() to reject negative input. Cap the
> value at INT_MAX / HZ to preserve the prior jiffies range and avoid
> overflow in secs_to_jiffies().
>
> The issue was reproduced on an affected 7.0 distribution kernel in a
> temporary network namespace. A patched target kernel was built with W=1
> and exercised in QEMU; -1 is now rejected with EINVAL, zero and positive
> values retain their behavior, and the HZ=1000 boundary accepts 2147483
> while rejecting 2147484.
>
> Fixes: 749c42b620a9 ("ipvs: reduce sync rate with time thresholds")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Yingjie Wang <1075151112@xxxxxx>
> ---
> include/net/ip_vs.h | 7 ++++---
> net/netfilter/ipvs/ip_vs_ctl.c | 8 ++++++--
> 2 files changed, 10 insertions(+), 5 deletions(-)
>
> diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
> index 9a83bff32569..66f26c3d1f75 100644
> --- a/include/net/ip_vs.h
> +++ b/include/net/ip_vs.h
> @@ -19,6 +19,7 @@
>
> #include <linux/compiler.h>
> #include <linux/timer.h>
> +#include <linux/jiffies.h>

This is not needed. The remaining part looks ok. Prefer
to use "nf" tag for IPVS patches.

> #include <linux/bug.h>
>
> #include <net/checksum.h>
> @@ -1267,7 +1268,7 @@ struct netns_ipvs {
> int sysctl_sloppy_sctp;
> int sysctl_expire_quiescent_template;
> int sysctl_sync_threshold[2];
> - unsigned int sysctl_sync_refresh_period;
> + unsigned int sysctl_sync_refresh_period; /* seconds */
> int sysctl_sync_retries;
> int sysctl_nat_icmp_send;
> int sysctl_pmtu_disc;
> @@ -1335,7 +1336,7 @@ struct netns_ipvs {
> #define DEFAULT_SYNC_VER 1
> #define DEFAULT_SLOPPY_TCP 0
> #define DEFAULT_SLOPPY_SCTP 0
> -#define DEFAULT_SYNC_REFRESH_PERIOD (0U * HZ)
> +#define DEFAULT_SYNC_REFRESH_PERIOD 0U
> #define DEFAULT_SYNC_RETRIES 0
> #define IPVS_SYNC_WAKEUP_RATE 8
> #define IPVS_SYNC_QLEN_MAX (IPVS_SYNC_WAKEUP_RATE * 4)
> @@ -1363,7 +1364,7 @@ static inline int sysctl_sync_period(struct netns_ipvs *ipvs)
>
> static inline unsigned int sysctl_sync_refresh_period(struct netns_ipvs *ipvs)
> {
> - return READ_ONCE(ipvs->sysctl_sync_refresh_period);
> + return secs_to_jiffies(READ_ONCE(ipvs->sysctl_sync_refresh_period));
> }
>
> static inline int sysctl_sync_retries(struct netns_ipvs *ipvs)
> diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
> index ce9fe20d6106..82726b6b5e3a 100644
> --- a/net/netfilter/ipvs/ip_vs_ctl.c
> +++ b/net/netfilter/ipvs/ip_vs_ctl.c
> @@ -2369,6 +2369,8 @@ static int ip_vs_zero_all(struct netns_ipvs *ipvs)
>
> #ifdef CONFIG_SYSCTL
>
> +static unsigned int sync_refresh_period_max = INT_MAX / HZ;
> +
> static int
> proc_do_conn_max(const struct ctl_table *table, int write,
> void *buffer, size_t *lenp, loff_t *ppos)
> @@ -2824,9 +2826,11 @@ static struct ctl_table vs_vars[] = {
> },
> {
> .procname = "sync_refresh_period",
> - .maxlen = sizeof(int),
> + .maxlen = sizeof(unsigned int),
> .mode = 0644,
> - .proc_handler = proc_dointvec_jiffies,
> + .proc_handler = proc_douintvec_minmax,

It is sad that there is no proc_dointvec_jiffies_minmax

> + .extra1 = SYSCTL_ZERO,
> + .extra2 = &sync_refresh_period_max,
> },
> {
> .procname = "sync_retries",
>
> base-commit: cb284914e749388eef402938bdb3115fe1f6d337
> --
> 2.43.0

Regards

--
Julian Anastasov <ja@xxxxxx>