mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH net-next v3 0/3] tcp/mptcp: prevent blackhole timeout arithmetic wraparound
@ 2026-11-16 16:46 Yingjie Wang
  2026-11-16 16:46 ` [PATCH net-next v3 1/3] tcp: avoid wrapping Fast Open blackhole timeout Yingjie Wang
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Yingjie Wang @ 2026-11-16 16:46 UTC (permalink / raw)
  To: netdev
  Cc: mptcp, linux-kernel, corbet, skhan, rdunlap, linux-doc, edumazet,
	ncardwell, gang.yan, kuniyu, davem, kuba, pabeni, horms, matttbe,
	martineau, geliang, ycheng, weiwan

The TCP Fast Open and MPTCP blackhole handlers multiply writable timeout
sysctl values before adding the result to a jiffies timestamp. The
multiplication is performed in 32-bit unsigned arithmetic, so a large
value can wrap into a short delay even on 64-bit kernels.

With HZ=1000 and the maximum multiplier of 64, any timeout above 67108
seconds wraps; 67109 seconds becomes 8,704 jiffies (8.7 seconds) instead
of about 50 days. Patches 1 and 2 compute the delay in u64 and cap it at
MAX_JIFFY_OFFSET before adding it to the saved timestamp. Patch 1 also
limits the TCP sysctl to one day.

Patch 3 applies the same one-day limit to net.mptcp.blackhole_timeout,
following Eric's comment that the two sysctls should behave consistently
and Matthieu's agreement.

Both limits reject values that were previously accepted, so the series
targets net-next and is not intended for stable.

Changes since v2:

- Target net-next and drop Cc: stable, as requested by Eric. Keep the
  Fixes tags.
- Move version history and earlier test results to the cover letter.
  The commit messages describe the problem, the fix, and how the final
  code was validated.
- Use 67109 seconds as the example and reproducer timeout. It is inside
  the one-day limit and still wraps at the 64-fold backoff with HZ=1000.
- Mention in the changelogs that writes above the limit now fail.
- Add patch 3, limiting net.mptcp.blackhole_timeout to one day.
- Patches 1 and 2 have the same code as in v2. Gang Yan's Reviewed-by is
  kept on patch 2.

v2: https://lore.kernel.org/r/cover.1794672619.git.x-professor@outlook.com
v1: https://lore.kernel.org/r/cover.1794325665.git.x-professor@outlook.com

Testing on net-next 7ccc5e05043b, x86-64, HZ=1000. The loopback
reproducers induce 12 TFO or MPTCP blackholes and then open a probe
connection:

- Timeout 67109 seconds, probe after 10 seconds: the unpatched kernel
  retried active TFO (1.056 s, TCPFastOpenActiveFail +1) and MP_CAPABLE
  (3.053 s, MPCapableSYNTXDisabled +0). With a 5 second wait it still
  skipped both, matching the wrapped 8.7 second interval. The v3 kernel
  skipped both after 10 seconds (0.000 s; +0 and +1).
- Timeout 6912213 seconds, probe after 1.2 seconds: the unpatched kernel
  retried both. On a test kernel with patches 1 and 2 applied and only
  the TCP input limit removed, both remained disabled.
- On the v3 kernel, both sysctls accept 0, 1, 3600, 86399 and 86400 and
  reject -1, 86401, 6912213 and INT_MAX with EINVAL, leaving the previous
  value unchanged, in the initial and in a new network namespace.
  Defaults are unchanged. With 86400 second timeouts, the reproducers
  still disabled active TFO and MPTCP.
- The affected objects build with W=1 on x86-64 and i386 at each patch.

The 32-bit behaviour of the MAX_JIFFY_OFFSET cap was checked by i386
builds and arithmetic, not at runtime. The TCP and MPTCP selftests were
not run locally; none of them sets these sysctls.

The issues were found using a static analysis tool.

Yingjie Wang (3):
  tcp: avoid wrapping Fast Open blackhole timeout
  mptcp: avoid wrapping blackhole timeout
  mptcp: limit blackhole_timeout to one day

 Documentation/networking/ip-sysctl.rst    |  1 +
 Documentation/networking/mptcp-sysctl.rst |  1 +
 net/ipv4/sysctl_net_ipv4.c                |  1 +
 net/ipv4/tcp_fastopen.c                   | 12 +++++++++---
 net/mptcp/ctrl.c                          | 14 +++++++++++---
 5 files changed, 23 insertions(+), 6 deletions(-)


base-commit: 7ccc5e05043b1bb84326b19a2b05938d8f7a2ae6
-- 
2.43.0


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH net-next v3 1/3] tcp: avoid wrapping Fast Open blackhole timeout
  2026-11-16 16:46 [PATCH net-next v3 0/3] tcp/mptcp: prevent blackhole timeout arithmetic wraparound Yingjie Wang
@ 2026-11-16 16:46 ` Yingjie Wang
  2026-11-16 16:46 ` [PATCH net-next v3 2/3] mptcp: avoid wrapping " Yingjie Wang
  2026-11-16 16:46 ` [PATCH net-next v3 3/3] mptcp: limit blackhole_timeout to one day Yingjie Wang
  2 siblings, 0 replies; 4+ messages in thread
From: Yingjie Wang @ 2026-11-16 16:46 UTC (permalink / raw)
  To: netdev
  Cc: mptcp, linux-kernel, corbet, skhan, rdunlap, linux-doc, edumazet,
	ncardwell, gang.yan, kuniyu, davem, kuba, pabeni, horms, matttbe,
	martineau, geliang, ycheng, weiwan

tcp_fastopen_active_should_disable() calculates the active-disable
interval as multiplier * tfo_bh_timeout * HZ. Since all operands are
32-bit, a large tcp_fastopen_blackhole_timeout_sec can wrap the delay,
even on 64-bit kernels, and re-enable active Fast Open too soon.

With HZ=1000 and the maximum multiplier of 64, any timeout above 67108
seconds wraps. For example, 67109 seconds should produce 4,294,976,000
jiffies (about 50 days), but the calculation truncates it to 8,704
jiffies (8.7 seconds).

This issue was found using a static analysis tool. It was confirmed with
a loopback reproducer that induces 12 Fast Open blackholes and opens
another connection 10 seconds later: with a 67109 second timeout, the
unpatched kernel retried active Fast Open. The reproducer needs neither
kernel modifications nor error injection.

Compute the interval in u64 and cap it at MAX_JIFFY_OFFSET before adding
it to the saved timestamp. On 32-bit kernels, the cap keeps the interval
within the range that time_before() can compare, which limits the longest
backoff there to about 12.4 days at HZ=1000.

Also limit the writable initial timeout to one day, so the exponential
backoff does not start from an impractically long period. Writes above
86400 seconds, which were previously accepted, now fail with -EINVAL and
leave the current value unchanged. The wider calculation is still needed
below that limit, as the 67109 second example shows.

With this patch, the same reproducer kept active Fast Open disabled, and
the sysctl accepts 86400 and rejects 86401. The TCP selftests were not
run; none of them sets this sysctl.

Fixes: cf1ef3f0719b ("net/tcp_fastopen: Disable active side TFO in certain scenarios")
Signed-off-by: Yingjie Wang <x-professor@outlook.com>
---
 Documentation/networking/ip-sysctl.rst |  1 +
 net/ipv4/sysctl_net_ipv4.c             |  1 +
 net/ipv4/tcp_fastopen.c                | 12 +++++++++---
 3 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/Documentation/networking/ip-sysctl.rst b/Documentation/networking/ip-sysctl.rst
index 0b7bfbd98a92..dc33f82a8997 100644
--- a/Documentation/networking/ip-sysctl.rst
+++ b/Documentation/networking/ip-sysctl.rst
@@ -1055,6 +1055,7 @@ tcp_fastopen_blackhole_timeout_sec - INTEGER
 	get detected right after Fastopen is re-enabled and will reset to
 	initial value when the blackhole issue goes away.
 	0 to disable the blackhole detection.
+	The maximum value is one day (86400 seconds).
 
 	By default, it is set to 0 (feature is disabled).
 
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 9df93ad219db..fb0f3124836d 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -1152,6 +1152,7 @@ static const struct ctl_table ipv4_net_table[] = {
 		.mode		= 0644,
 		.proc_handler	= proc_tfo_blackhole_detect_timeout,
 		.extra1		= SYSCTL_ZERO,
+		.extra2		= &one_day_secs,
 	},
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 	{
diff --git a/net/ipv4/tcp_fastopen.c b/net/ipv4/tcp_fastopen.c
index 7f7eb3cc1c04..883f0f1e7e07 100644
--- a/net/ipv4/tcp_fastopen.c
+++ b/net/ipv4/tcp_fastopen.c
@@ -613,6 +613,7 @@ bool tcp_fastopen_active_should_disable(struct sock *sk)
 	unsigned long timeout;
 	int tfo_da_times;
 	int multiplier;
+	u64 delay;
 
 	if (!tfo_bh_timeout)
 		return false;
@@ -624,12 +625,17 @@ bool tcp_fastopen_active_should_disable(struct sock *sk)
 	/* Paired with smp_mb__before_atomic() in tcp_fastopen_active_disable() */
 	smp_rmb();
 
-	/* Limit timeout to max: 2^6 * initial timeout */
+	/* Limit the backoff multiplier to 64. */
 	multiplier = 1 << min(tfo_da_times - 1, 6);
+	delay = (u64)multiplier * tfo_bh_timeout * HZ;
 
-	/* Paired with the WRITE_ONCE() in tcp_fastopen_active_disable(). */
+	/*
+	 * Cap the delay for safe jiffies comparisons on 32-bit systems.
+	 * The timestamp read is paired with WRITE_ONCE() in
+	 * tcp_fastopen_active_disable().
+	 */
 	timeout = READ_ONCE(sock_net(sk)->ipv4.tfo_active_disable_stamp) +
-		  multiplier * tfo_bh_timeout * HZ;
+		  min(delay, MAX_JIFFY_OFFSET);
 	if (time_before(jiffies, timeout))
 		return true;
 
-- 
2.43.0


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH net-next v3 2/3] mptcp: avoid wrapping blackhole timeout
  2026-11-16 16:46 [PATCH net-next v3 0/3] tcp/mptcp: prevent blackhole timeout arithmetic wraparound Yingjie Wang
  2026-11-16 16:46 ` [PATCH net-next v3 1/3] tcp: avoid wrapping Fast Open blackhole timeout Yingjie Wang
@ 2026-11-16 16:46 ` Yingjie Wang
  2026-11-16 16:46 ` [PATCH net-next v3 3/3] mptcp: limit blackhole_timeout to one day Yingjie Wang
  2 siblings, 0 replies; 4+ messages in thread
From: Yingjie Wang @ 2026-11-16 16:46 UTC (permalink / raw)
  To: netdev
  Cc: mptcp, linux-kernel, corbet, skhan, rdunlap, linux-doc, edumazet,
	ncardwell, gang.yan, kuniyu, davem, kuba, pabeni, horms, matttbe,
	martineau, geliang, ycheng, weiwan, Gang Yan

mptcp_active_should_disable() calculates the active-disable interval as
multiplier * blackhole_timeout * HZ. Since all operands are 32-bit, a
large blackhole_timeout can wrap the delay, even on 64-bit kernels, and
make active MPTCP available again while the blackhole is still present.

With HZ=1000 and the maximum multiplier of 64, any timeout above 67108
seconds wraps. For example, 67109 seconds should produce 4,294,976,000
jiffies (about 50 days), but the calculation truncates it to 8,704
jiffies (8.7 seconds).

This issue was found using a static analysis tool. It was confirmed with
a loopback reproducer that induces 12 MPTCP blackholes and opens another
connection 10 seconds later: with a 67109 second timeout, the unpatched
kernel retried MP_CAPABLE. The reproducer needs neither kernel
modifications nor error injection.

Compute the interval in u64 and cap it at MAX_JIFFY_OFFSET before adding
it to the saved timestamp. On 32-bit kernels, the cap keeps the interval
within the range that time_before() can compare. The sysctl input range
is unchanged; values whose interval exceeds the cap are shortened to it.

With this patch, the same reproducer fell back to TCP without retrying
MP_CAPABLE. The MPTCP selftests were not run; none of them sets this
sysctl.

Fixes: 27069e7cb3d1 ("mptcp: disable active MPTCP in case of blackhole")
Reviewed-by: Gang Yan <yangang@kylinos.cn>
Signed-off-by: Yingjie Wang <x-professor@outlook.com>
---
 net/mptcp/ctrl.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/net/mptcp/ctrl.c b/net/mptcp/ctrl.c
index 63c5747f0f63..0b10e94e7f43 100644
--- a/net/mptcp/ctrl.c
+++ b/net/mptcp/ctrl.c
@@ -487,6 +487,7 @@ bool mptcp_active_should_disable(struct sock *ssk)
 	unsigned long timeout;
 	int disable_times;
 	int multiplier;
+	u64 delay;
 
 	pernet = mptcp_get_pernet(net);
 	blackhole_timeout = READ_ONCE(pernet->blackhole_timeout);
@@ -501,12 +502,17 @@ bool mptcp_active_should_disable(struct sock *ssk)
 	/* Paired with smp_mb__before_atomic() in mptcp_active_disable() */
 	smp_rmb();
 
-	/* Limit timeout to max: 2^6 * initial timeout */
+	/* Limit the backoff multiplier to 64. */
 	multiplier = 1 << min(disable_times - 1, 6);
+	delay = (u64)multiplier * blackhole_timeout * HZ;
 
-	/* Paired with the WRITE_ONCE() in mptcp_active_disable(). */
+	/*
+	 * Cap the delay for safe jiffies comparisons on 32-bit systems.
+	 * The timestamp read is paired with WRITE_ONCE() in
+	 * mptcp_active_disable().
+	 */
 	timeout = READ_ONCE(pernet->active_disable_stamp) +
-		  multiplier * blackhole_timeout * HZ;
+		  min(delay, MAX_JIFFY_OFFSET);
 
 	return time_before(jiffies, timeout);
 }
-- 
2.43.0


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH net-next v3 3/3] mptcp: limit blackhole_timeout to one day
  2026-11-16 16:46 [PATCH net-next v3 0/3] tcp/mptcp: prevent blackhole timeout arithmetic wraparound Yingjie Wang
  2026-11-16 16:46 ` [PATCH net-next v3 1/3] tcp: avoid wrapping Fast Open blackhole timeout Yingjie Wang
  2026-11-16 16:46 ` [PATCH net-next v3 2/3] mptcp: avoid wrapping " Yingjie Wang
@ 2026-11-16 16:46 ` Yingjie Wang
  2 siblings, 0 replies; 4+ messages in thread
From: Yingjie Wang @ 2026-11-16 16:46 UTC (permalink / raw)
  To: netdev
  Cc: mptcp, linux-kernel, corbet, skhan, rdunlap, linux-doc, edumazet,
	ncardwell, gang.yan, kuniyu, davem, kuba, pabeni, horms, matttbe,
	martineau, geliang, ycheng, weiwan

net.mptcp.blackhole_timeout accepts any non-negative int as the initial
period for disabling active MPTCP after a blackhole, and the period can
then grow 64-fold. A multi-day initial period is not useful, and on
32-bit kernels, values above about 1,073,741 seconds at HZ=1000 are
shortened by the MAX_JIFFY_OFFSET cap even on the first blackhole.

Limit the writable value to one day, matching
net.ipv4.tcp_fastopen_blackhole_timeout_sec, and document the limit. The
default of 3600 seconds is unchanged.

Writes above 86400 seconds, which were previously accepted, now fail
with -EINVAL and leave the current value unchanged.

Suggested-by: Eric Dumazet <edumazet@kernel.org>
Signed-off-by: Yingjie Wang <x-professor@outlook.com>
---
 Documentation/networking/mptcp-sysctl.rst | 1 +
 net/mptcp/ctrl.c                          | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/Documentation/networking/mptcp-sysctl.rst b/Documentation/networking/mptcp-sysctl.rst
index b9b5f58e0625..ca00253c8ae9 100644
--- a/Documentation/networking/mptcp-sysctl.rst
+++ b/Documentation/networking/mptcp-sysctl.rst
@@ -63,6 +63,7 @@ blackhole_timeout - INTEGER (seconds)
 	blackhole issue goes away.
 
 	0 to disable the blackhole detection. This is a per-namespace sysctl.
+	The maximum value is one day (86400 seconds).
 
 	Default: 3600
 
diff --git a/net/mptcp/ctrl.c b/net/mptcp/ctrl.c
index 0b10e94e7f43..1f167f72c1f5 100644
--- a/net/mptcp/ctrl.c
+++ b/net/mptcp/ctrl.c
@@ -20,6 +20,7 @@ static int mptcp_pernet_id;
 
 #ifdef CONFIG_SYSCTL
 static int mptcp_pm_type_max = __MPTCP_PM_TYPE_MAX;
+static int mptcp_blackhole_timeout_max = 24 * 3600;
 #endif
 
 struct mptcp_pernet {
@@ -350,6 +351,7 @@ static struct ctl_table mptcp_sysctl_table[] = {
 		.mode = 0644,
 		.proc_handler = proc_blackhole_detect_timeout,
 		.extra1 = SYSCTL_ZERO,
+		.extra2 = &mptcp_blackhole_timeout_max,
 	},
 	{
 		.procname = "syn_retrans_before_tcp_fallback",
-- 
2.43.0


^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2026-11-16 16:46 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-16 16:46 [PATCH net-next v3 0/3] tcp/mptcp: prevent blackhole timeout arithmetic wraparound Yingjie Wang
2026-11-16 16:46 ` [PATCH net-next v3 1/3] tcp: avoid wrapping Fast Open blackhole timeout Yingjie Wang
2026-11-16 16:46 ` [PATCH net-next v3 2/3] mptcp: avoid wrapping " Yingjie Wang
2026-11-16 16:46 ` [PATCH net-next v3 3/3] mptcp: limit blackhole_timeout to one day Yingjie Wang

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®