Re: [PATCH net v3 1/2] skmsg: prevent empty ingress skb from enqueuing

[Geliang Tang <geliang@kernel.org>]

On Tue, 2024-07-02 at 18:03 -0700, John Fastabend wrote:
> Geliang Tang wrote:
> > From: Geliang Tang <tanggeliang@kylinos.cn>
> > 
> > Run this BPF selftests (./test_progs -t sockmap_basic) on a
> > Loongarch
> > platform, a Kernel panic occurs:
> > 
> > '''
> > Oops[#1]:
> > CPU: 22 PID: 2824 Comm: test_progs Tainted: G           OE  6.10.0-
> > rc2+ #18
> > Hardware name: LOONGSON Dabieshan/Loongson-TC542F0, BIOS Loongson-
> > UDK2018
> >    ... ...
> >    ra: 90000000048bf6c0 sk_msg_recvmsg+0x120/0x560
> >   ERA: 9000000004162774 copy_page_to_iter+0x74/0x1c0
> >  CRMD: 000000b0 (PLV0 -IE -DA +PG DACF=CC DACM=CC -WE)
> >  PRMD: 0000000c (PPLV0 +PIE +PWE)
> >  EUEN: 00000007 (+FPE +SXE +ASXE -BTE)
> >  ECFG: 00071c1d (LIE=0,2-4,10-12 VS=7)
> > ESTAT: 00010000 [PIL] (IS= ECode=1 EsubCode=0)
> >  BADV: 0000000000000040
> >  PRID: 0014c011 (Loongson-64bit, Loongson-3C5000)
> > Modules linked in: bpf_testmod(OE) xt_CHECKSUM xt_MASQUERADE
> > xt_conntrack
> > Process test_progs (pid: 2824, threadinfo=0000000000863a31,
> > task=...)
> > Stack : ...
> >         ...
> > Call Trace:
> > [<9000000004162774>] copy_page_to_iter+0x74/0x1c0
> > [<90000000048bf6c0>] sk_msg_recvmsg+0x120/0x560
> > [<90000000049f2b90>] tcp_bpf_recvmsg_parser+0x170/0x4e0
> > [<90000000049aae34>] inet_recvmsg+0x54/0x100
> > [<900000000481ad5c>] sock_recvmsg+0x7c/0xe0
> > [<900000000481e1a8>] __sys_recvfrom+0x108/0x1c0
> > [<900000000481e27c>] sys_recvfrom+0x1c/0x40
> > [<9000000004c076ec>] do_syscall+0x8c/0xc0
> > [<9000000003731da4>] handle_syscall+0xc4/0x160
> > 
> > Code: ...
> > 
> > ---[ end trace 0000000000000000 ]---
> > Kernel panic - not syncing: Fatal exception
> > Kernel relocated by 0x3510000
> >  .text @ 0x9000000003710000
> >  .data @ 0x9000000004d70000
> >  .bss  @ 0x9000000006469400
> > ---[ end Kernel panic - not syncing: Fatal exception ]---
> > '''
> > 
> > This crash happens every time when running
> > sockmap_skb_verdict_shutdown
> > subtest in sockmap_basic.
> > 
> > This crash is because a NULL pointer is passed to page_address() in
> > sk_msg_recvmsg(). Due to the difference in architecture,
> > page_address(0)
> > will not trigger a panic on the X86 platform but will panic on the
> > Loogarch platform. So this bug was hidden on the x86 platform, but
> > now
> > it is exposed on the Loogarch platform.
> > 
> > The root cause is an empty skb (skb->len == 0) is put on the queue.
> > 
> > In this case, in sk_psock_skb_ingress_enqueue(), num_sge is zero,
> > and no
> > page is put to this sge (see sg_set_page in sg_set_page), but this
> > empty
> > sge is queued into ingress_msg list.
> > 
> > And in sk_msg_recvmsg(), this empty sge is used, and a NULL page is
> > got by sg_page(sge). Pass this NULL-page to copy_page_to_iter(), it
> > passed to kmap_local_page() and page_address(), then kernel panics.
> > 
> > To solve this, we should prevent empty skb from putting on the
> > queue. So
> > in sk_psock_verdict_recv(), if the skb->len is zero, drop this skb.
> > 
> > Fixes: ef5659280eb1 ("bpf, sockmap: Allow skipping sk_skb parser
> > program")
> > Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
> > ---
> >  net/core/skmsg.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/net/core/skmsg.c b/net/core/skmsg.c
> > index fd20aae30be2..44952cdd1425 100644
> > --- a/net/core/skmsg.c
> > +++ b/net/core/skmsg.c
> > @@ -1184,7 +1184,7 @@ static int sk_psock_verdict_recv(struct sock
> > *sk, struct sk_buff *skb)
> >  
> >  	rcu_read_lock();
> >  	psock = sk_psock(sk);
> > -	if (unlikely(!psock)) {
> > +	if (unlikely(!psock || !len)) {
> >  		len = 0;
> >  		tcp_eat_skb(sk, skb);
> >  		sock_drop(sk, skb);
> 
> The skb->len == 0 here is the FIN pkt right? We are using the EFAULT
> return
> triggered by copy_page_to_iter to check for is_fin in tcp_bpf.c.
> 
> The concern I have here is if we don't have the skb fin pkt on the
> recv
> queue we might go into wait_data and block instead of return to user
> when
> rcvmsg() is called from user. I wonder if we can write a test for
> this if
> we don't already have one we probably should create one.
> 
> Maybe a better fix assuming my assumption about fin being skb->len=0
> is
> correct?

Thanks John. Your fix is much better than mine. I'll use this as v5 and
update the commit log. I'll add your "Suggested-by" tag in it.

-Geliang

> 
> diff --git a/net/core/skmsg.c b/net/core/skmsg.c
> index fd20aae30be2..bbf40b999713 100644
> --- a/net/core/skmsg.c
> +++ b/net/core/skmsg.c
> @@ -434,7 +434,8 @@ int sk_msg_recvmsg(struct sock *sk, struct
> sk_psock *psock, struct msghdr *msg,
>                         page = sg_page(sge);
>                         if (copied + copy > len)
>                                 copy = len - copied;
> -                       copy = copy_page_to_iter(page, sge->offset,
> copy, iter);
> +                       if (copy)
> +                               copy = copy_page_to_iter(page, sge-
> >offset, copy, iter);
>                         if (!copy) {
>                                 copied = copied ? copied : -EFAULT;
>                                 goto out;
> 
> Thanks,
> John