[PATCH net v2] net: atm: Fix incorrect net_device lec check

[PATCH net v2] net: atm: Fix incorrect net_device lec check

[Lin Ma]

There are two sites in atm mpoa code that believe the fetched object
net_device is of lec type. However, both of them do just name checking
to ensure that the device name starts with "lec" pattern string.

That is, malicious user can hijack this by creating another device
starting with that pattern, thereby causing type confusion. For example,
create a *team* interface with lecX name, bind that interface and send
messages will get a crash like below:

[   18.450000] kernel tried to execute NX-protected page - exploit attempt? (uid: 0)
[   18.452366] BUG: unable to handle page fault for address: ffff888005702a70
[   18.454253] #PF: supervisor instruction fetch in kernel mode
[   18.455058] #PF: error_code(0x0011) - permissions violation
[   18.455366] PGD 3801067 P4D 3801067 PUD 3802067 PMD 80000000056000e3
[   18.455725] Oops: 0011 [#1] PREEMPT SMP PTI
[   18.455966] CPU: 0 PID: 130 Comm: trigger Not tainted 6.1.90 #7
[   18.456921] RIP: 0010:0xffff888005702a70
[   18.457151] Code: .....
[   18.458168] RSP: 0018:ffffc90000677bf8 EFLAGS: 00010286
[   18.458461] RAX: ffff888005702a70 RBX: ffff888005702000 RCX: 000000000000001b
[   18.458850] RDX: ffffc90000677c10 RSI: ffff88800565e0a8 RDI: ffff888005702000
[   18.459248] RBP: ffffc90000677c68 R08: 0000000000000000 R09: 0000000000000000
[   18.459644] R10: 0000000000000000 R11: ffff888005702a70 R12: ffff88800556c000
[   18.460033] R13: ffff888005964900 R14: ffff8880054b4000 R15: ffff8880054b5000
[   18.460425] FS:  0000785e61b5a740(0000) GS:ffff88807dc00000(0000) knlGS:0000000000000000
[   18.460872] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   18.461183] CR2: ffff888005702a70 CR3: 00000000054c2000 CR4: 00000000000006f0
[   18.461580] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   18.461974] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[   18.462368] Call Trace:
[   18.462518]  <TASK>
[   18.462645]  ? __die_body+0x64/0xb0
[   18.462856]  ? page_fault_oops+0x353/0x3e0
[   18.463092]  ? exc_page_fault+0xaf/0xd0
[   18.463322]  ? asm_exc_page_fault+0x22/0x30
[   18.463589]  ? msg_from_mpoad+0x431/0x9d0
[   18.463820]  ? vcc_sendmsg+0x165/0x3b0
[   18.464031]  vcc_sendmsg+0x20a/0x3b0
[   18.464238]  ? wake_bit_function+0x80/0x80
[   18.464511]  __sys_sendto+0x38c/0x3a0
[   18.464729]  ? percpu_counter_add_batch+0x87/0xb0
[   18.465002]  __x64_sys_sendto+0x22/0x30
[   18.465219]  do_syscall_64+0x6c/0xa0
[   18.465465]  ? preempt_count_add+0x54/0xb0
[   18.465697]  ? up_read+0x37/0x80
[   18.465883]  ? do_user_addr_fault+0x25e/0x5b0
[   18.466126]  ? exit_to_user_mode_prepare+0x12/0xb0
[   18.466435]  entry_SYSCALL_64_after_hwframe+0x6e/0xd8
[   18.466727] RIP: 0033:0x785e61be4407
[   18.467948] RSP: 002b:00007ffe61ae2150 EFLAGS: 00000202 ORIG_RAX: 000000000000002c
[   18.468368] RAX: ffffffffffffffda RBX: 0000785e61b5a740 RCX: 0000785e61be4407
[   18.468758] RDX: 000000000000019c RSI: 00007ffe61ae21c0 RDI: 0000000000000003
[   18.469149] RBP: 00007ffe61ae2370 R08: 0000000000000000 R09: 0000000000000000
[   18.469542] R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000000000
[   18.469936] R13: 00007ffe61ae2498 R14: 0000785e61d74000 R15: 000057bddcbabd98

Refer to other net_device related subsystem, checking netlink_ops seems
like the correct way out, e.g., see how xgbe_netdev_event() validates
the netdev object. Hence, add correct comparison with lec_netdev_ops to
safeguard the casting.

By the way, this bug dates back to pre-git history (2.3.15), hence use
the first reference for tracking.

Signed-off-by: Lin Ma <linma@zju.edu.cn>
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
---
V1 -> V2: add null check suggested by Eric Dumazet <edumazet@google.com>,
          otherwise will crash

 net/atm/mpc.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/net/atm/mpc.c b/net/atm/mpc.c
index f6b447bba329..af409d6fa2dd 100644
--- a/net/atm/mpc.c
+++ b/net/atm/mpc.c
@@ -275,6 +275,9 @@ static struct net_device *find_lec_by_itfnum(int itf)
 	sprintf(name, "lec%d", itf);
 	dev = dev_get_by_name(&init_net, name);
 
+	if (!dev || dev->netdev_ops != lec_netdev_ops)
+		return NULL;
+
 	return dev;
 }
 
@@ -1006,7 +1009,7 @@ static int mpoa_event_listener(struct notifier_block *mpoa_notifier,
 	if (!net_eq(dev_net(dev), &init_net))
 		return NOTIFY_DONE;
 
-	if (strncmp(dev->name, "lec", 3))
+	if (dev->netdev_ops != lec_netdev_ops)
 		return NOTIFY_DONE; /* we are only interested in lec:s */
 
 	switch (event) {
-- 
2.17.1

Re: [PATCH net v2] net: atm: Fix incorrect net_device lec check

[Lin Ma]

Hi there,

Sorry again for the careless mistakes... the `lec_netdev_ops` is a static ops variable defined
elsewhere, hence the patch cannot compile correctly.

Preparing the v3 patch.

Regards
Lin

> V1 -> V2: add null check suggested by Eric Dumazet <edumazet@google.com>,
>           otherwise will crash
> 
>  net/atm/mpc.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/net/atm/mpc.c b/net/atm/mpc.c
> index f6b447bba329..af409d6fa2dd 100644
> --- a/net/atm/mpc.c
> +++ b/net/atm/mpc.c
> @@ -275,6 +275,9 @@ static struct net_device *find_lec_by_itfnum(int itf)
>  	sprintf(name, "lec%d", itf);
>  	dev = dev_get_by_name(&init_net, name);
>  
> +	if (!dev || dev->netdev_ops != lec_netdev_ops)
> +		return NULL;
> +
>  	return dev;
>  }
>  
> @@ -1006,7 +1009,7 @@ static int mpoa_event_listener(struct notifier_block *mpoa_notifier,
>  	if (!net_eq(dev_net(dev), &init_net))
>  		return NOTIFY_DONE;
>  
> -	if (strncmp(dev->name, "lec", 3))
> +	if (dev->netdev_ops != lec_netdev_ops)
>  		return NOTIFY_DONE; /* we are only interested in lec:s */
>  
>  	switch (event) {
> -- 
> 2.17.1