mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH net] mac802154: remove redundant security_enabled assignment
@ 2026-11-16  6:52 Petr Vaganov
  2026-11-17 11:06 ` Miquel Raynal
  2026-11-18  2:04 ` netdev-bot+sashiko
  0 siblings, 2 replies; 3+ messages in thread
From: Petr Vaganov @ 2026-11-16  6:52 UTC (permalink / raw)
  To: Alexander Aring, Stefan Schmidt, Miquel Raynal, linux-wpan
  Cc: Petr Vaganov, David S. Miller, Eric Dumazet, Jakub Kicinski,
	Paolo Abeni, Simon Horman, Phoebe Buckheister, netdev,
	linux-kernel, lvc-project, stable

During fuzz testing, the following issue was discovered:

BUG: KMSAN: uninit-value in ieee802154_hdr_push_sechdr net/ieee802154/header_ops.c:54 [inline]
BUG: KMSAN: uninit-value in ieee802154_hdr_push+0xa62/0xb20 net/ieee802154/header_ops.c:108
 ieee802154_hdr_push_sechdr net/ieee802154/header_ops.c:54 [inline]
 ieee802154_hdr_push+0xa62/0xb20 net/ieee802154/header_ops.c:108
 ieee802154_header_create+0xa83/0xd20 net/mac802154/iface.c:396
 wpan_dev_hard_header include/net/cfg802154.h:525 [inline]
 dgram_sendmsg+0xc6d/0x15d0 net/ieee802154/socket.c:677
 ieee802154_sock_sendmsg+0x8f/0xd0 net/ieee802154/socket.c:96
 sock_sendmsg_nosec net/socket.c:711 [inline]
 __sock_sendmsg+0x2f3/0x360 net/socket.c:726
 ____sys_sendmsg+0x960/0xc30 net/socket.c:2581
 ___sys_sendmsg+0x285/0x3e0 net/socket.c:2635
 __sys_sendmmsg+0x33c/0x950 net/socket.c:2724
 __do_sys_sendmmsg net/socket.c:2751 [inline]
 __se_sys_sendmmsg net/socket.c:2748 [inline]
 __x64_sys_sendmmsg+0xc6/0x150 net/socket.c:2748
 x64_sys_call+0x12a8/0x3cf0 arch/x86/include/generated/asm/syscalls_64.h:308
 do_syscall_x64 arch/x86/entry/common.c:47 [inline]
 do_syscall_64+0xd9/0x1c0 arch/x86/entry/common.c:78
 entry_SYSCALL_64_after_hwframe+0x77/0x7f

Uninit was stored to memory at:
 ieee802154_hdr_push_sechdr net/ieee802154/header_ops.c:51 [inline]
 ieee802154_hdr_push+0xa5b/0xb20 net/ieee802154/header_ops.c:108
 ieee802154_header_create+0xa83/0xd20 net/mac802154/iface.c:396
 wpan_dev_hard_header include/net/cfg802154.h:525 [inline]
 dgram_sendmsg+0xc6d/0x15d0 net/ieee802154/socket.c:677
 ieee802154_sock_sendmsg+0x8f/0xd0 net/ieee802154/socket.c:96
 sock_sendmsg_nosec net/socket.c:711 [inline]
 __sock_sendmsg+0x2f3/0x360 net/socket.c:726
 ____sys_sendmsg+0x960/0xc30 net/socket.c:2581
 ___sys_sendmsg+0x285/0x3e0 net/socket.c:2635
 __sys_sendmmsg+0x33c/0x950 net/socket.c:2724
 __do_sys_sendmmsg net/socket.c:2751 [inline]
 __se_sys_sendmmsg net/socket.c:2748 [inline]
 __x64_sys_sendmmsg+0xc6/0x150 net/socket.c:2748
 x64_sys_call+0x12a8/0x3cf0 arch/x86/include/generated/asm/syscalls_64.h:308
 do_syscall_x64 arch/x86/entry/common.c:47 [inline]
 do_syscall_64+0xd9/0x1c0 arch/x86/entry/common.c:78
 entry_SYSCALL_64_after_hwframe+0x77/0x7f

Commit f30be4d53cad ("mac802154: integrate llsec with wpan devices")
introduced mac802154_set_header_security() as the sole function
responsible for enabling frame-level security: it sets both
hdr->fc.security_enabled = 1 and initializes hdr->sec together,
maintaining the invariant that security_enabled = 1 implies hdr->sec
is initialized.

However, ieee802154_header_create() was left with a pre-existing
assignment:

    hdr.fc.security_enabled = cb->secen;

This assignment can set security_enabled = 1 (from a stale cb->secen)
even when mac802154_set_header_security() returns 0 without initializing
hdr.sec, violating that invariant. ieee802154_hdr_push_sechdr() then
reads uninitialized hdr.sec fields.

The stale value in cb->secen arises from a specific setsockopt()
sequence: calling WPAN_SECURITY_ON (secen_override=1, secen=1) and
then WPAN_SECURITY_DEFAULT (secen_override=0, secen unchanged at 1).
This undocumented combination likely explains why the bug was rarely
triggered.

Remove the redundant assignment; mac802154_set_header_security() is the
only correct place to set security_enabled.

Found by Linux Verification Center (linuxtesting.org) with Syzkaller.

Closes: https://syzkaller.appspot.com/bug?extid=60a66d44892b66b56545
Closes: https://syzkaller.appspot.com/bug?id=23f21e76e9d2d1ed587cb5e71515260d6d025b63
Fixes: f30be4d53cad ("mac802154: integrate llsec with wpan devices")
Cc: stable@vger.kernel.org
Signed-off-by: Petr Vaganov <petrvaganoff@gmail.com>
---
 net/mac802154/iface.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/net/mac802154/iface.c b/net/mac802154/iface.c
index f437b8daa00d..7b7332805ecc 100644
--- a/net/mac802154/iface.c
+++ b/net/mac802154/iface.c
@@ -368,7 +368,6 @@ static int ieee802154_header_create(struct sk_buff *skb,
 
 	memset(&hdr.fc, 0, sizeof(hdr.fc));
 	hdr.fc.type = cb->type;
-	hdr.fc.security_enabled = cb->secen;
 	hdr.fc.ack_request = cb->ackreq;
 	hdr.seq = atomic_inc_return(&dev->ieee802154_ptr->dsn) & 0xFF;
 
-- 
2.55.0

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

* Re: [PATCH net] mac802154: remove redundant security_enabled assignment
  2026-11-16  6:52 [PATCH net] mac802154: remove redundant security_enabled assignment Petr Vaganov
@ 2026-11-17 11:06 ` Miquel Raynal
  2026-11-18  2:04 ` netdev-bot+sashiko
  1 sibling, 0 replies; 3+ messages in thread
From: Miquel Raynal @ 2026-11-17 11:06 UTC (permalink / raw)
  To: Petr Vaganov
  Cc: Alexander Aring, Stefan Schmidt, linux-wpan, David S. Miller,
	Eric Dumazet, Jakub Kicinski, Paolo Abeni, Simon Horman,
	Phoebe Buckheister, netdev, linux-kernel, lvc-project, stable

On 16/11/2026 at 13:52:34 +07, Petr Vaganov <petrvaganoff@gmail.com> wrote:

> During fuzz testing, the following issue was discovered:
>
> BUG: KMSAN: uninit-value in ieee802154_hdr_push_sechdr net/ieee802154/header_ops.c:54 [inline]
> BUG: KMSAN: uninit-value in ieee802154_hdr_push+0xa62/0xb20 net/ieee802154/header_ops.c:108
>  ieee802154_hdr_push_sechdr net/ieee802154/header_ops.c:54 [inline]
>  ieee802154_hdr_push+0xa62/0xb20 net/ieee802154/header_ops.c:108
>  ieee802154_header_create+0xa83/0xd20 net/mac802154/iface.c:396
>  wpan_dev_hard_header include/net/cfg802154.h:525 [inline]
>  dgram_sendmsg+0xc6d/0x15d0 net/ieee802154/socket.c:677
>  ieee802154_sock_sendmsg+0x8f/0xd0 net/ieee802154/socket.c:96
>  sock_sendmsg_nosec net/socket.c:711 [inline]
>  __sock_sendmsg+0x2f3/0x360 net/socket.c:726
>  ____sys_sendmsg+0x960/0xc30 net/socket.c:2581
>  ___sys_sendmsg+0x285/0x3e0 net/socket.c:2635
>  __sys_sendmmsg+0x33c/0x950 net/socket.c:2724
>  __do_sys_sendmmsg net/socket.c:2751 [inline]
>  __se_sys_sendmmsg net/socket.c:2748 [inline]
>  __x64_sys_sendmmsg+0xc6/0x150 net/socket.c:2748
>  x64_sys_call+0x12a8/0x3cf0 arch/x86/include/generated/asm/syscalls_64.h:308
>  do_syscall_x64 arch/x86/entry/common.c:47 [inline]
>  do_syscall_64+0xd9/0x1c0 arch/x86/entry/common.c:78
>  entry_SYSCALL_64_after_hwframe+0x77/0x7f
>
> Uninit was stored to memory at:
>  ieee802154_hdr_push_sechdr net/ieee802154/header_ops.c:51 [inline]
>  ieee802154_hdr_push+0xa5b/0xb20 net/ieee802154/header_ops.c:108
>  ieee802154_header_create+0xa83/0xd20 net/mac802154/iface.c:396
>  wpan_dev_hard_header include/net/cfg802154.h:525 [inline]
>  dgram_sendmsg+0xc6d/0x15d0 net/ieee802154/socket.c:677
>  ieee802154_sock_sendmsg+0x8f/0xd0 net/ieee802154/socket.c:96
>  sock_sendmsg_nosec net/socket.c:711 [inline]
>  __sock_sendmsg+0x2f3/0x360 net/socket.c:726
>  ____sys_sendmsg+0x960/0xc30 net/socket.c:2581
>  ___sys_sendmsg+0x285/0x3e0 net/socket.c:2635
>  __sys_sendmmsg+0x33c/0x950 net/socket.c:2724
>  __do_sys_sendmmsg net/socket.c:2751 [inline]
>  __se_sys_sendmmsg net/socket.c:2748 [inline]
>  __x64_sys_sendmmsg+0xc6/0x150 net/socket.c:2748
>  x64_sys_call+0x12a8/0x3cf0 arch/x86/include/generated/asm/syscalls_64.h:308
>  do_syscall_x64 arch/x86/entry/common.c:47 [inline]
>  do_syscall_64+0xd9/0x1c0 arch/x86/entry/common.c:78
>  entry_SYSCALL_64_after_hwframe+0x77/0x7f
>
> Commit f30be4d53cad ("mac802154: integrate llsec with wpan devices")
> introduced mac802154_set_header_security() as the sole function
> responsible for enabling frame-level security: it sets both
> hdr->fc.security_enabled = 1 and initializes hdr->sec together,
> maintaining the invariant that security_enabled = 1 implies hdr->sec
> is initialized.
>
> However, ieee802154_header_create() was left with a pre-existing
> assignment:
>
>     hdr.fc.security_enabled = cb->secen;
>
> This assignment can set security_enabled = 1 (from a stale cb->secen)
> even when mac802154_set_header_security() returns 0 without initializing
> hdr.sec, violating that invariant. ieee802154_hdr_push_sechdr() then
> reads uninitialized hdr.sec fields.
>
> The stale value in cb->secen arises from a specific setsockopt()
> sequence: calling WPAN_SECURITY_ON (secen_override=1, secen=1) and
> then WPAN_SECURITY_DEFAULT (secen_override=0, secen unchanged at 1).
> This undocumented combination likely explains why the bug was rarely
> triggered.
>
> Remove the redundant assignment; mac802154_set_header_security() is the
> only correct place to set security_enabled.
>
> Found by Linux Verification Center (linuxtesting.org) with Syzkaller.
>
> Closes: https://syzkaller.appspot.com/bug?extid=60a66d44892b66b56545
> Closes: https://syzkaller.appspot.com/bug?id=23f21e76e9d2d1ed587cb5e71515260d6d025b63
> Fixes: f30be4d53cad ("mac802154: integrate llsec with wpan devices")
> Cc: stable@vger.kernel.org
> Signed-off-by: Petr Vaganov <petrvaganoff@gmail.com>
> ---
>  net/mac802154/iface.c | 1 -
>  1 file changed, 1 deletion(-)
>
> diff --git a/net/mac802154/iface.c b/net/mac802154/iface.c
> index f437b8daa00d..7b7332805ecc 100644
> --- a/net/mac802154/iface.c
> +++ b/net/mac802154/iface.c
> @@ -368,7 +368,6 @@ static int ieee802154_header_create(struct sk_buff *skb,
>  
>  	memset(&hdr.fc, 0, sizeof(hdr.fc));
>  	hdr.fc.type = cb->type;
> -	hdr.fc.security_enabled = cb->secen;
>  	hdr.fc.ack_request = cb->ackreq;
>  	hdr.seq = atomic_inc_return(&dev->ieee802154_ptr->dsn) & 0xFF;

There has been a similar patch in 2023 but it was never updated by the
author. Fixes tag is better here, it should be taken.

Reviewed-by: Miquel Raynal <miquel.raynal@bootlin.com>

Thanks,
Miquèl

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

* Re: [PATCH net] mac802154: remove redundant security_enabled assignment
  2026-11-16  6:52 [PATCH net] mac802154: remove redundant security_enabled assignment Petr Vaganov
  2026-11-17 11:06 ` Miquel Raynal
@ 2026-11-18  2:04 ` netdev-bot+sashiko
  1 sibling, 0 replies; 3+ messages in thread
From: netdev-bot+sashiko @ 2026-11-18  2:04 UTC (permalink / raw)
  To: petrvaganoff
  Cc: alex.aring, stefan, miquel.raynal, linux-wpan, davem, edumazet,
	kuba, pabeni, horms, phoebe.buckheister, netdev, linux-kernel,
	lvc-project, stable

Thank you for your contribution! Sashiko AI review found 2 potential
issue(s) to consider.

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

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).

- [High] The fix relies on an invariant that does not hold.
- [Low] The subject 'mac802154: remove redundant security_enabled
  assignment' presents a bug fix as a cleanup.

--

mac802154: remove redundant security_enabled assignment

The commit removes the assignment hdr.fc.security_enabled = cb->secen from
ieee802154_header_create(). After that, mac802154_set_header_security() is
the only place that sets the security bit. This targets a KMSAN uninit-value
report in ieee802154_hdr_push_sechdr(), triggered by a stale cb->secen after
WPAN_SECURITY_ON followed by WPAN_SECURITY_DEFAULT.

> mac802154: remove redundant security_enabled assignment

[Severity: Low]
This isn't a bug, but is "redundant" the right word for this line?

Removing it changes the transmitted frame. When
mac802154_set_header_security() takes this early return, the removed line
was the only thing setting fc.security_enabled:

	if (!params.enabled ||
	    (cb->secen_override && !cb->secen) ||
	    !params.out_level)
		return 0;

This is a fix tagged for stable. Would a subject like "mac802154: fix
uninit-value in ieee802154_header_create()" be more accurate?

> diff --git a/net/mac802154/iface.c b/net/mac802154/iface.c
> index f437b8daa00d0..7b7332805ecce 100644
> --- a/net/mac802154/iface.c
> +++ b/net/mac802154/iface.c
> @@ -368,7 +368,6 @@ static int ieee802154_header_create(struct sk_buff *skb,
>  
>  	memset(&hdr.fc, 0, sizeof(hdr.fc));
>  	hdr.fc.type = cb->type;
> -	hdr.fc.security_enabled = cb->secen;

[Severity: High]
Does this fully close the uninitialized hdr.sec read in
ieee802154_hdr_push_sechdr()?

The commit message relies on mac802154_set_header_security() keeping the
invariant that security_enabled = 1 implies hdr->sec is initialized. That
function only writes some of the fields:

net/mac802154/iface.c:mac802154_set_header_security() {
    ...
	hdr->fc.security_enabled = 1;
	hdr->sec.level = level;
	hdr->sec.key_id_mode = params.out_key.mode;
	if (params.out_key.mode == IEEE802154_SCF_KEY_SHORT_INDEX)
		hdr->sec.short_src = params.out_key.short_source;
	else if (params.out_key.mode == IEEE802154_SCF_KEY_HW_INDEX)
		hdr->sec.extended_src = params.out_key.extended_source;
	hdr->sec.key_id = params.out_key.id;
    ...
}

It never writes sec.frame_counter or the reserved bits that share a byte
with level and key_id_mode. In ieee802154_header_create(), hdr is on the
stack and only hdr.fc is cleared by the memset above.
mac802154_header_create() has the same pattern.

So on the normal llsec TX path (llsec enabled, out_level != 0),
ieee802154_hdr_push_sechdr() still copies stack memory into the skb:

	memcpy(buf, hdr, 1);
	memcpy(buf + 1, &hdr->frame_counter, 4);

frame_counter is overwritten only later, in mac802154_llsec_encrypt(). By
then, packet taps in dev_queue_xmit_nit() have already seen the header, as
the TODO comment in net/mac802154/tx.c notes:

dgram_sendmsg()
  wpan_dev_hard_header()
    ieee802154_header_create()
      mac802154_set_header_security()
      ieee802154_hdr_push()
        ieee802154_hdr_push_sechdr()
  dev_queue_xmit()
    dev_queue_xmit_nit()
    ieee802154_subif_start_xmit()
      mac802154_llsec_encrypt()

If encryption fails with -ENOKEY or -EOVERFLOW, mac802154_llsec_encrypt()
returns before it pushes the header again. The 4 uninitialized
frame_counter bytes then stay in the data that AF_PACKET taps copy to
userspace.

The reserved bits survive the memcpy(hdr, buf, 1) in
ieee802154_hdr_get_sechdr(). mac802154_llsec_encrypt() then pushes them
again, so they also go out over the air and into the CCM AAD.

Should hdr (or at least hdr.sec) be zero-initialized in both
ieee802154_header_create() and mac802154_header_create()? Another option
is to memset hdr->sec in mac802154_set_header_security() before filling it.

>  	hdr.fc.ack_request = cb->ackreq;
>  	hdr.seq = atomic_inc_return(&dev->ieee802154_ptr->dsn) & 0xFF;
>

-- 
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261116065235.27348-1-petrvaganoff%40gmail.com

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

end of thread, other threads:[~2026-11-18  2:04 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-16  6:52 [PATCH net] mac802154: remove redundant security_enabled assignment Petr Vaganov
2026-11-17 11:06 ` Miquel Raynal
2026-11-18  2:04 ` 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

all inboxes | Powered by JetHome®