Netdev List
 help / color / mirror / Atom feed
* [PATCH net-next v2] net: avoid hashing the namespace pointer for device names
@ 2026-11-18  6:26 Zhengchuan Liang
  2026-11-18  8:35 ` Eric Dumazet
  2026-11-22  6:43 ` netdev-bot+sashiko
  0 siblings, 2 replies; 3+ messages in thread
From: Zhengchuan Liang @ 2026-11-18  6:26 UTC (permalink / raw)
  To: David S . Miller
  Cc: Eric Dumazet, Jakub Kicinski, Paolo Abeni, Simon Horman, netdev,
	linux-kernel, Zhengchuan Liang

dev_name_hash() salts device-name hashes with the network namespace
pointer. An unprivileged user can time SIOCGIFINDEX lookups for chosen
missing names and identify collisions with a resident name such as "lo".
Ranking this collision pattern across possible KASLR slides recovers the
init_net pointer and thus the kernel text base.

Use an unsalted name hash. Each network namespace has a separate hash
table, so this does not change lookup or namespace-isolation semantics.

Found by AI-assisted code review and reproduced unprivileged on an
unmodified kernel without fault injection. The reproducer and networking
selftests pass with this change.

Fixes: 8387ff2577eb ("vfs: make the string hashes salt the hash")
Cc: stable@vger.kernel.org
Assisted-by: LLM
Signed-off-by: Zhengchuan Liang <zcliangcn@gmail.com>
---
Changes in v2:
- Pass NULL instead of net_hash_mix(), which would expose a seed shared
  by other per-network namespace hash tables.
- Target net-next, as requested by Eric.

v1: https://lore.kernel.org/all/62774f8c6cbad549d1a95f88df99ab7745d7f6b8.1790917676.git.zcliangcn@gmail.com/

 net/core/dev.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index d88af07aea29..12100fbf9022 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -195,7 +195,7 @@ static inline void dev_base_seq_inc(struct net *net)
 
 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
 {
-	unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
+	unsigned int hash = full_name_hash(NULL, name, strnlen(name, IFNAMSIZ));
 
 	return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
 }
-- 
2.34.1


^ permalink raw reply related	[flat|nested] 3+ messages in thread

* Re: [PATCH net-next v2] net: avoid hashing the namespace pointer for device names
  2026-11-18  6:26 [PATCH net-next v2] net: avoid hashing the namespace pointer for device names Zhengchuan Liang
@ 2026-11-18  8:35 ` Eric Dumazet
  2026-11-22  6:43 ` netdev-bot+sashiko
  1 sibling, 0 replies; 3+ messages in thread
From: Eric Dumazet @ 2026-11-18  8:35 UTC (permalink / raw)
  To: Zhengchuan Liang
  Cc: David S . Miller, Jakub Kicinski, Paolo Abeni, Simon Horman,
	netdev, linux-kernel

Le mer. 18 nov. 2026 à 07:27, Zhengchuan Liang <zcliangcn@gmail.com> a écrit :
>
> dev_name_hash() salts device-name hashes with the network namespace
> pointer. An unprivileged user can time SIOCGIFINDEX lookups for chosen
> missing names and identify collisions with a resident name such as "lo".
> Ranking this collision pattern across possible KASLR slides recovers the
> init_net pointer and thus the kernel text base.
>
> Use an unsalted name hash. Each network namespace has a separate hash
> table, so this does not change lookup or namespace-isolation semantics.
>
> Found by AI-assisted code review and reproduced unprivileged on an
> unmodified kernel without fault injection. The reproducer and networking
> selftests pass with this change.
>
> Fixes: 8387ff2577eb ("vfs: make the string hashes salt the hash")
> Cc: stable@vger.kernel.org

As discussed in v1, local KASLR attacks are not a serious concern,
this is why I asked for net-next.

A "Cc: stable" tag does not belong in a net-next patch.

Maintainers can probably remove it when applying, no need for a V3.

Reviewed-by: Eric Dumazet <edumazet@kernel.org>

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH net-next v2] net: avoid hashing the namespace pointer for device names
  2026-11-18  6:26 [PATCH net-next v2] net: avoid hashing the namespace pointer for device names Zhengchuan Liang
  2026-11-18  8:35 ` Eric Dumazet
@ 2026-11-22  6:43 ` netdev-bot+sashiko
  1 sibling, 0 replies; 3+ messages in thread
From: netdev-bot+sashiko @ 2026-11-22  6:43 UTC (permalink / raw)
  To: zcliangcn; +Cc: davem, edumazet, kuba, pabeni, horms, netdev, linux-kernel

Thank you for your contribution! Sashiko AI review found 1 potential
issue(s) to consider.

Critical: 0 · High: 0 · Medium: 1 · Low: 0

If you believe none of the issues are legitimate, please reply to each
issue of Medium severity or higher and explain why the AI is wrong.
If any of them are legitimate and you plan to respin, please let patchwork
know by including "pw-bot: cr" as a separate line at the end of your reply
(one such reply per series is enough).

- [Medium] The patch fixes only dev_name_hash().

--

net: avoid hashing the namespace pointer for device names

This commit changes dev_name_hash() to call full_name_hash() with no salt,
where it used to salt with the struct net pointer. The goal is to stop
unprivileged users from recovering init_net and the kernel text base by
timing name lookups that collide in the hash table.

> diff --git a/net/core/dev.c b/net/core/dev.c
> index d88af07aea290..12100fbf9022a 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -195,7 +195,7 @@ static inline void dev_base_seq_inc(struct net *net)
>  
>  static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
>  {
> -	unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
> +	unsigned int hash = full_name_hash(NULL, name, strnlen(name, IFNAMSIZ));
>  
>  	return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
>  }

[Severity: Medium]
This fixes dev_name_hash(). Is there still a similar name hash in Open
vSwitch that uses the struct net pointer as its seed?

net/openvswitch/vport.c:hash_bucket() {
	unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
	return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
}

jhash() takes a u32 initval, so the seed is the low 32 bits of the net
pointer. On x86-64 those low bits contain the KASLR slide of init_net.

ovs_vport_locate() then walks the bucket with strcmp() and net_eq(). The
lookup time therefore depends on whether a probe name shares a bucket with
a resident vport:

net/openvswitch/vport.c:ovs_vport_locate() {
	struct hlist_head *bucket = hash_bucket(net, name);
	...
	hlist_for_each_entry_rcu(vport, bucket, hash_node,
				 lockdep_ovsl_is_held())
		if (!strcmp(name, ovs_vport_name(vport)) &&
		    net_eq(ovs_dp_get_net(vport->dp), net))
			return vport;
	...
}

Unprivileged users can reach this lookup with a name they choose, since
both commands use .flags = 0:

  OVS_VPORT_CMD_GET -> lookup_vport() -> ovs_vport_locate()
  OVS_DP_CMD_GET -> lookup_datapath() -> ovs_vport_locate()

Take a host with OVS vports in init_net, such as ovs-system or br-int.
Could the collision-timing method in the commit message still recover
init_net through this path? It has not been shown that one extra strcmp()
can be measured through genetlink, which adds more timing noise.

A grep of net/ suggests this is the only hash left that uses the net
pointer as its seed. The commit message doesn't mention it.

dev_table is one global table of VPORT_HASH_BUCKETS buckets, shared by
every namespace. Would it need a random seed instead of NULL? Options
include get_random_u32() in ovs_vport_init() or net_hash_mix(). This
could be done as a separate follow-up.

-- 
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/c9b3c1a0e3410ffa70db448e2299ed420528b5ff.1794981913.git.zcliangcn%40gmail.com

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2026-11-22  6:43 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-18  6:26 [PATCH net-next v2] net: avoid hashing the namespace pointer for device names Zhengchuan Liang
2026-11-18  8:35 ` Eric Dumazet
2026-11-22  6:43 ` netdev-bot+sashiko

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