| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Tue, 18 Apr 2023 21:04:32 +0200
From: Eric Dumazet <edumazet@...gle.com>
To: Kuniyuki Iwashima <kuniyu@...zon.com>
Cc: davem@...emloft.net, kuba@...nel.org, kuni1840@...il.com,
netdev@...r.kernel.org, pabeni@...hat.com,
syzkaller@...glegroups.com, willemb@...gle.com
Subject: Re: [PATCH v2 net] tcp/udp: Fix memleaks of sk and zerocopy skbs with
TX timestamp.
On Tue, Apr 18, 2023 at 8:44 PM Kuniyuki Iwashima <kuniyu@...zon.com> wrote:
>
> From: Eric Dumazet <edumazet@...gle.com>
> Date: Tue, 18 Apr 2023 20:33:44 +0200
> > On Tue, Apr 18, 2023 at 8:09 PM Kuniyuki Iwashima <kuniyu@...zon.com> wrote:
> > >
> > > syzkaller reported [0] memory leaks of an UDP socket and ZEROCOPY
> > > skbs. We can reproduce the problem with these sequences:
> > >
> > > sk = socket(AF_INET, SOCK_DGRAM, 0)
> > > sk.setsockopt(SOL_SOCKET, SO_TIMESTAMPING, SOF_TIMESTAMPING_TX_SOFTWARE)
> > > sk.setsockopt(SOL_SOCKET, SO_ZEROCOPY, 1)
> > > sk.sendto(b'', MSG_ZEROCOPY, ('127.0.0.1', 53))
> > > sk.close()
> > >
> > > sendmsg() calls msg_zerocopy_alloc(), which allocates a skb, sets
> > > skb->cb->ubuf.refcnt to 1, and calls sock_hold(). Here, struct
> > > ubuf_info_msgzc indirectly holds a refcnt of the socket. When the
> > > skb is sent, __skb_tstamp_tx() clones it and puts the clone into
> > > the socket's error queue with the TX timestamp.
> > >
> > > When the original skb is received locally, skb_copy_ubufs() calls
> > > skb_unclone(), and pskb_expand_head() increments skb->cb->ubuf.refcnt.
> > > This additional count is decremented while freeing the skb, but struct
> > > ubuf_info_msgzc still has a refcnt, so __msg_zerocopy_callback() is
> > > not called.
> > >
> > > The last refcnt is not released unless we retrieve the TX timestamped
> > > skb by recvmsg(). When we close() the socket holding such skb, we
> > > never call sock_put() and leak the count, skb, and sk.
> > >
> > > To avoid this problem, we must (i) call skb_queue_purge() after
> > > flagging SOCK_DEAD during close() and (ii) make sure that TX tstamp
> > > skb is not queued when SOCK_DEAD is flagged. UDP lacks (i) and (ii),
> > > and TCP lacks (ii).
> > >
> > > Without (ii), a skb queued in a qdisc or device could be put into
> > > the error queue after skb_queue_purge().
> > >
> > > sendmsg() /* return immediately, but packets
> > > * are queued in a qdisc or device
> > > */
> > > close()
> > > skb_queue_purge()
> > > __skb_tstamp_tx()
> > > __skb_complete_tx_timestamp()
> > > sock_queue_err_skb()
> > > skb_queue_tail()
> > >
> > > Also, we need to check SOCK_DEAD under sk->sk_error_queue.lock
> > > in sock_queue_err_skb() to avoid this race.
> > >
> > > if (!sock_flag(sk, SOCK_DEAD))
> > > sock_set_flag(sk, SOCK_DEAD)
> > > skb_queue_purge()
> > >
> > > skb_queue_tail()
> > >
> > > [0]:
> >
> > > Fixes: f214f915e7db ("tcp: enable MSG_ZEROCOPY")
> > > Fixes: b5947e5d1e71 ("udp: msg_zerocopy")
> > > Reported-by: syzbot <syzkaller@...glegroups.com>
> > > Signed-off-by: Kuniyuki Iwashima <kuniyu@...zon.com>
> > > ---
> > > v2:
> > > * Move skb_queue_purge() after setting SOCK_DEAD in udp_destroy_sock()
> > > * Check SOCK_DEAD in sock_queue_err_skb() with sk_error_queue.lock
> > > * Add Fixes tag for TCP
> > >
> > > v1: https://lore.kernel.org/netdev/20230417171155.22916-1-kuniyu@amazon.com/
> > > ---
> > > net/core/skbuff.c | 15 ++++++++++++---
> > > net/ipv4/udp.c | 5 +++++
> > > 2 files changed, 17 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> > > index 4c0879798eb8..287b834df9c8 100644
> > > --- a/net/core/skbuff.c
> > > +++ b/net/core/skbuff.c
> > > @@ -4979,6 +4979,8 @@ static void skb_set_err_queue(struct sk_buff *skb)
> > > */
> > > int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
> > > {
> > > + unsigned long flags;
> > > +
> > > if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
> > > (unsigned int)READ_ONCE(sk->sk_rcvbuf))
> > > return -ENOMEM;
> > > @@ -4992,9 +4994,16 @@ int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
> > > /* before exiting rcu section, make sure dst is refcounted */
> > > skb_dst_force(skb);
> > >
> > > - skb_queue_tail(&sk->sk_error_queue, skb);
> > > - if (!sock_flag(sk, SOCK_DEAD))
> > > - sk_error_report(sk);
> > > + spin_lock_irqsave(&sk->sk_error_queue.lock, flags);
> > > + if (sock_flag(sk, SOCK_DEAD)) {
> >
> > SOCK_DEAD is set without holding sk_error_queue.lock, so I wonder why you
> > want to add a confusing construct.
> >
> > Just bail early ?
> >
> > diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> > index ef81452759be3fd251faaf76d89cfd002ee79256..fda05cb44f95821e98f8c5c05fba840a9d276abb
> > 100644
> > --- a/net/core/skbuff.c
> > +++ b/net/core/skbuff.c
> > @@ -4983,6 +4983,9 @@ int sock_queue_err_skb(struct sock *sk, struct
> > sk_buff *skb)
> > (unsigned int)READ_ONCE(sk->sk_rcvbuf))
> > return -ENOMEM;
> >
> > + if (sock_flag(sk, SOCK_DEAD))
> > + return -EINVAL;
> > +
>
> Isn't it possible that these sequences happen
>
> close()
> sock_set_flag(sk, SOCK_DEAD);
> skb_queue_purge(&sk->sk_error_queue)
>
> between the skb_queue_tail() below ? (2nd race mentioned in changelog)
>
> I thought we can guarantee the ordering by taking the same lock.
>
This is fragile.
We could very well rewrite skb_queue_purge() to not acquire the lock
in the common case.
I had the following in my tree for a while, to avoid many atomic and
irq masking operations...
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index ef81452759be3fd251faaf76d89cfd002ee79256..f570e7c89248e2d9d9ab50c2e86ac7019b837484
100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -3606,14 +3606,24 @@ EXPORT_SYMBOL(skb_dequeue_tail);
* skb_queue_purge - empty a list
* @list: list to empty
*
- * Delete all buffers on an &sk_buff list. Each buffer is removed from
- * the list and one reference dropped. This function takes the list
- * lock and is atomic with respect to other list locking functions.
+ * Delete all buffers on an &sk_buff list.
*/
void skb_queue_purge(struct sk_buff_head *list)
{
+ struct sk_buff_head priv;
+ unsigned long flags;
struct sk_buff *skb;
- while ((skb = skb_dequeue(list)) != NULL)
+
+ if (skb_queue_empty_lockless(list))
+ return;
+
+ __skb_queue_head_init(&priv);
+
+ spin_lock_irqsave(&list->lock, flags);
+ skb_queue_splice_init(list, &priv);
+ spin_unlock_irqrestore(&list->lock, flags);
+
+ while ((skb = __skb_dequeue(&priv)) != NULL)
kfree_skb(skb);
}
EXPORT_SYMBOL(skb_queue_purge);
Powered by blists - more mailing lists