[PATCH net] net/smc: Initialize smc hashtables before registering users

[PATCH net] net/smc: Initialize smc hashtables before registering users

[Alexandra Winter]

During initialisation of the SMC module initialize smc_v4/6_hashinfo before
calling smc_nl_init(), proto_register() or sock_register(), to avoid a race
that can cause use of an uninitialised pointer in case an smc protocol is
called before the module is done initialising.

syzbot report:
KASAN: null-ptr-deref in range [0x0000000000000008-0x000000000000000f]
Call Trace:
 <TASK>
 smc_diag_dump+0x59/0xa0 net/smc/smc_diag.c:236
 netlink_dump+0x647/0xd80 net/netlink/af_netlink.c:2325
 __netlink_dump_start+0x59f/0x780 net/netlink/af_netlink.c:2440
 netlink_dump_start include/linux/netlink.h:339 [inline]
 smc_diag_handler_dump+0x1ab/0x250 net/smc/smc_diag.c:251
 sock_diag_rcv_msg+0x3dc/0x5f0
 netlink_rcv_skb+0x1e3/0x430 net/netlink/af_netlink.c:2550
 netlink_unicast_kernel net/netlink/af_netlink.c:1331 [inline]
 netlink_unicast+0x7f0/0x990 net/netlink/af_netlink.c:1357
 netlink_sendmsg+0x8e4/0xcb0 net/netlink/af_netlink.c:1901

Fixes: f16a7dd5cf27 ("smc: netlink interface for SMC sockets")
Reported-by: syzbot+f69bfae0a4eb29976e44@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=f69bfae0a4eb29976e44
Signed-off-by: Alexandra Winter <wintera@linux.ibm.com>
---
 net/smc/af_smc.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/net/smc/af_smc.c b/net/smc/af_smc.c
index f97f77b041d9..b0f4405fb714 100644
--- a/net/smc/af_smc.c
+++ b/net/smc/af_smc.c
@@ -3524,6 +3524,9 @@ static int __init smc_init(void)
 		goto out_pernet_subsys_stat;
 	smc_clc_init();
 
+	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
+	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
+
 	rc = smc_nl_init();
 	if (rc)
 		goto out_ism;
@@ -3581,8 +3584,6 @@ static int __init smc_init(void)
 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
 		goto out_proto6;
 	}
-	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
-	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
 
 	rc = smc_ib_register_client();
 	if (rc) {
-- 
2.51.0

Re: [PATCH net] net/smc: Initialize smc hashtables before registering users

[Dust Li]

On 2025-12-17 12:48:19, Alexandra Winter wrote:
>During initialisation of the SMC module initialize smc_v4/6_hashinfo before
>calling smc_nl_init(), proto_register() or sock_register(), to avoid a race
>that can cause use of an uninitialised pointer in case an smc protocol is
>called before the module is done initialising.
>
>syzbot report:
>KASAN: null-ptr-deref in range [0x0000000000000008-0x000000000000000f]
>Call Trace:
> <TASK>
> smc_diag_dump+0x59/0xa0 net/smc/smc_diag.c:236
> netlink_dump+0x647/0xd80 net/netlink/af_netlink.c:2325
> __netlink_dump_start+0x59f/0x780 net/netlink/af_netlink.c:2440
> netlink_dump_start include/linux/netlink.h:339 [inline]
> smc_diag_handler_dump+0x1ab/0x250 net/smc/smc_diag.c:251
> sock_diag_rcv_msg+0x3dc/0x5f0
> netlink_rcv_skb+0x1e3/0x430 net/netlink/af_netlink.c:2550
> netlink_unicast_kernel net/netlink/af_netlink.c:1331 [inline]
> netlink_unicast+0x7f0/0x990 net/netlink/af_netlink.c:1357
> netlink_sendmsg+0x8e4/0xcb0 net/netlink/af_netlink.c:1901


I don't think this is related to smc_nl_init().

Here the calltrace is smc_diag_dump(), which was registered in
sock_diag_register(&smc_diag_handler).

But smc_nl_init() is registering the general netlink in SMC,
which is unrelated to smc_diag_dump().

I think the root cause should be related to the initializing between
smc_diag.ko and smc_v4/6_hashinfo.ht.
The change in your previous patch 'dibs: Register smc as dibs_client'
may change the possiblity to this bug.

Best regards,
Dust

>
>Fixes: f16a7dd5cf27 ("smc: netlink interface for SMC sockets")
>Reported-by: syzbot+f69bfae0a4eb29976e44@syzkaller.appspotmail.com
>Closes: https://syzkaller.appspot.com/bug?extid=f69bfae0a4eb29976e44
>Signed-off-by: Alexandra Winter <wintera@linux.ibm.com>
>---
> net/smc/af_smc.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
>diff --git a/net/smc/af_smc.c b/net/smc/af_smc.c
>index f97f77b041d9..b0f4405fb714 100644
>--- a/net/smc/af_smc.c
>+++ b/net/smc/af_smc.c
>@@ -3524,6 +3524,9 @@ static int __init smc_init(void)
> 		goto out_pernet_subsys_stat;
> 	smc_clc_init();
> 
>+	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
>+	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
>+
> 	rc = smc_nl_init();
> 	if (rc)
> 		goto out_ism;
>@@ -3581,8 +3584,6 @@ static int __init smc_init(void)
> 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
> 		goto out_proto6;
> 	}
>-	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
>-	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
> 
> 	rc = smc_ib_register_client();
> 	if (rc) {
>-- 
>2.51.0

Re: [PATCH net] net/smc: Initialize smc hashtables before registering users

[Alexandra Winter]

On 17.12.25 16:25, Dust Li wrote:
> On 2025-12-17 12:48:19, Alexandra Winter wrote:
>> During initialisation of the SMC module initialize smc_v4/6_hashinfo before
>> calling smc_nl_init(), proto_register() or sock_register(), to avoid a race
>> that can cause use of an uninitialised pointer in case an smc protocol is
>> called before the module is done initialising.
>>
>> syzbot report:
>> KASAN: null-ptr-deref in range [0x0000000000000008-0x000000000000000f]
>> Call Trace:
>> <TASK>
>> smc_diag_dump+0x59/0xa0 net/smc/smc_diag.c:236
>> netlink_dump+0x647/0xd80 net/netlink/af_netlink.c:2325
>> __netlink_dump_start+0x59f/0x780 net/netlink/af_netlink.c:2440
>> netlink_dump_start include/linux/netlink.h:339 [inline]
>> smc_diag_handler_dump+0x1ab/0x250 net/smc/smc_diag.c:251
>> sock_diag_rcv_msg+0x3dc/0x5f0
>> netlink_rcv_skb+0x1e3/0x430 net/netlink/af_netlink.c:2550
>> netlink_unicast_kernel net/netlink/af_netlink.c:1331 [inline]
>> netlink_unicast+0x7f0/0x990 net/netlink/af_netlink.c:1357
>> netlink_sendmsg+0x8e4/0xcb0 net/netlink/af_netlink.c:1901
> 
> I don't think this is related to smc_nl_init().
> 
> Here the calltrace is smc_diag_dump(), which was registered in
> sock_diag_register(&smc_diag_handler).
> 
> But smc_nl_init() is registering the general netlink in SMC,
> which is unrelated to smc_diag_dump().


I had assumed some dependency between the smc netlink diag socket and smc_nl_init()
and wrongly assumed that the smc_diag_init() and smc_init() could race.
I now understand that modprobe will ensure smc_diag_init() is called before smc_init(),
so you are right: this patch is indeed NOT a fix for this sysbot report [1]


> I think the root cause should be related to the initializing between
> smc_diag.ko and smc_v4/6_hashinfo.ht.

Given modprobe initializes the modules sequentially, I do not see how these could race.

I guess this syszbot report was fixed by
f584239a9ed2 ("net/smc: fix general protection fault in __smc_diag_dump")
as reported in [2] .

I'm not sure about the correct procedure, if nobody recommends a better action, I'll send a

#syz dup: general protection fault in __smc_diag_dump
to
syzbot+f69bfae0a4eb29976e44@syzkaller.appspotmail.com
(this one: general protection fault in smc_diag_dump_proto [1])


I still think initializing the hashtables before smc_nl_init()
makes sense. I'll resend this patch without mentioning syzbot.

-----
[1] https://syzkaller.appspot.com/bug?extid=f69bfae0a4eb29976e44
[2] https://syzkaller.appspot.com/bug?extid=f775be4458668f7d220e

Re: [PATCH net] net/smc: Initialize smc hashtables before registering users

[Dust Li]

On 2025-12-22 10:50:37, Alexandra Winter wrote:
>
>
>On 17.12.25 16:25, Dust Li wrote:
>> On 2025-12-17 12:48:19, Alexandra Winter wrote:
>>> During initialisation of the SMC module initialize smc_v4/6_hashinfo before
>>> calling smc_nl_init(), proto_register() or sock_register(), to avoid a race
>>> that can cause use of an uninitialised pointer in case an smc protocol is
>>> called before the module is done initialising.
>>>
>>> syzbot report:
>>> KASAN: null-ptr-deref in range [0x0000000000000008-0x000000000000000f]
>>> Call Trace:
>>> <TASK>
>>> smc_diag_dump+0x59/0xa0 net/smc/smc_diag.c:236
>>> netlink_dump+0x647/0xd80 net/netlink/af_netlink.c:2325
>>> __netlink_dump_start+0x59f/0x780 net/netlink/af_netlink.c:2440
>>> netlink_dump_start include/linux/netlink.h:339 [inline]
>>> smc_diag_handler_dump+0x1ab/0x250 net/smc/smc_diag.c:251
>>> sock_diag_rcv_msg+0x3dc/0x5f0
>>> netlink_rcv_skb+0x1e3/0x430 net/netlink/af_netlink.c:2550
>>> netlink_unicast_kernel net/netlink/af_netlink.c:1331 [inline]
>>> netlink_unicast+0x7f0/0x990 net/netlink/af_netlink.c:1357
>>> netlink_sendmsg+0x8e4/0xcb0 net/netlink/af_netlink.c:1901
>> 
>> I don't think this is related to smc_nl_init().
>> 
>> Here the calltrace is smc_diag_dump(), which was registered in
>> sock_diag_register(&smc_diag_handler).
>> 
>> But smc_nl_init() is registering the general netlink in SMC,
>> which is unrelated to smc_diag_dump().
>
>
>I had assumed some dependency between the smc netlink diag socket and smc_nl_init()
>and wrongly assumed that the smc_diag_init() and smc_init() could race.
>I now understand that modprobe will ensure smc_diag_init() is called before smc_init(),
>so you are right: this patch is indeed NOT a fix for this sysbot report [1]
>
>
>> I think the root cause should be related to the initializing between
>> smc_diag.ko and smc_v4/6_hashinfo.ht.
>
>Given modprobe initializes the modules sequentially, I do not see how these could race.
>
>I guess this syszbot report was fixed by
>f584239a9ed2 ("net/smc: fix general protection fault in __smc_diag_dump")
>as reported in [2] .
>
>I'm not sure about the correct procedure, if nobody recommends a better action, I'll send a
>
>#syz dup: general protection fault in __smc_diag_dump
>to
>syzbot+f69bfae0a4eb29976e44@syzkaller.appspotmail.com
>(this one: general protection fault in smc_diag_dump_proto [1])
>
>
>I still think initializing the hashtables before smc_nl_init()
>makes sense. I'll resend this patch without mentioning syzbot.

Agree.

Best regards,
Dust