mctp: prevent double key removal and unref
Jeremy Kerr
maintainerPATCH 1/1 mctp: prevent double key removal and unref
Currently, we have a bug where a simultaneous DROPTAG ioctl and socket close may race, as we attempt to remove a key from lists twice, and perform an unref for each removal operation. This may result in a uaf when we attempt the second unref.
This change fixes the race by making __mctp_key_remove tolerant to being called on a key that has already been removed from the socket/net lists, and only performs the unref when we do the actual remove. We also need to hold the list lock on the ioctl cleanup path.
This fix is based on a bug report and comprehensive analysis from butt3rflyh4ck <[email protected]>, found via syzkaller.
--- net/mctp/af_mctp.c | 23 ++++++++++++++++------- net/mctp/route.c | 10 +++++----- 2 files changed, 21 insertions(+), 12 deletions(-)
net/mctp/af_mctp.c +16 -7
@@ -295,11 +295,12 @@ __must_hold(&net->mctp.keys_lock)mctp_dev_release_key (key->dev, key);spin_unlock_irqrestore (&key->lock, flags); -hlist_del (&key->hlist); -hlist_del (&key->sklist); - -/* unref for the lists */ -mctp_key_unref (key); +if (!hlist_unhashed (&key->hlist)) { +hlist_del_init (&key->hlist); +hlist_del_init (&key->sklist); +/* unref for the lists */ +mctp_key_unref (key); + }kfree_skb (skb); }
@@ -373,9 +374,17 @@ static int mctp_ioctl_alloctag(struct mctp_sock *msk, unsigned long arg) ctl.tag = tag |MCTP_TAG_OWNER |MCTP_TAG_PREALLOC ;if (copy_to_user ((void __user *)arg, &ctl,sizeof (ctl))) { -spin_lock_irqsave (&key->lock, flags); -__mctp_key_remove (key, net, flags,MCTP_TRACE_KEY_DROPPED ); +unsigned long fl2; +/* Unwind our key allocation: the keys list lock needs to be +* taken before the individual key locks, and we need a valid +* flags value (fl2) to pass to __mctp_key_remove, hence the +* second spin_lock_irqsave() rather than a plain spin_lock(). +*/ +spin_lock_irqsave (&net->mctp.keys_lock, flags); +spin_lock_irqsave (&key->lock, fl2); +__mctp_key_remove (key, net, fl2,MCTP_TRACE_KEY_DROPPED );mctp_key_unref (key); +spin_unlock_irqrestore (&net->mctp.keys_lock, flags);return -EFAULT ; }
net/mctp/route.c +5 -5
@@ -228,12 +228,12 @@ __releases(&key->lock)if (!key->manual_alloc) {spin_lock_irqsave (&net->mctp.keys_lock, flags); -hlist_del (&key->hlist); -hlist_del (&key->sklist); +if (!hlist_unhashed (&key->hlist)) { +hlist_del_init (&key->hlist); +hlist_del_init (&key->sklist); +mctp_key_unref (key); + }spin_unlock_irqrestore (&net->mctp.keys_lock, flags); - -/* unref for the lists */ -mctp_key_unref (key); }/* and one for the local reference */
Signature · 1 line
2.35.1
This patch was applied to netdev/net.git (master) by David S. Miller <[email protected]>:
Hello:
This patch was applied to netdev/net.git (master)
by David S. Miller <[email protected]>:
Quoted 11 lines · On Wed, 12 Oct 2022 10:08:51 +0800 you wrote:
Currently, we have a bug where a simultaneous DROPTAG ioctl and socket close may race, as we attempt to remove a key from lists twice, and perform an unref for each removal operation. This may result in a uaf when we attempt the second unref.
This change fixes the race by making
__mctp_key_removetolerant to being called on a key that has already been removed from the socket/net lists, and only performs the unref when we do the actual remove. We also need to hold the list lock on the ioctl cleanup path.[...]
Here is the summary with links:
- [net] mctp: prevent double key removal and unref
https://git.kernel.org/netdev/net/c/3a732b46736c
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html