On Wed, Aug 2, 2023 at 10:14 PM Jesper Dangaard Brouer [off-list ref] wrote:

On 02/08/2023 00.07, Andrew Kanner wrote:

quoted

Using the syzkaller repro with reduced packet size it was discovered
that XDP_PACKET_HEADROOM is not checked in tun_can_build_skb(),
although pad may be incremented in tun_build_skb(). This may end up
with exceeding the PAGE_SIZE limit in tun_build_skb().

Fixes: 7df13219d757 ("tun: reserve extra headroom only when XDP is set")
Link: https://syzkaller.appspot.com/bug?extid=f817490f5bd20541b90a
Signed-off-by: Andrew Kanner <redacted>
---

Notes:
     v3 -> v4:
     * fall back to v1, fixing only missing XDP_PACKET_HEADROOM in pad and
       removing bpf_xdp_adjust_tail() check for frame_sz.
     * added rcu read lock, noted by Jason Wang [off-list ref] in v1
     * I decided to leave the packet length check in tun_can_build_skb()
       instead of moving to tun_build_skb() suggested by Jason Wang
       [off-list ref]. Otherwise extra packets will be dropped
       without falling back to tun_alloc_skb(). And in the discussion of v3
       Jesper Dangaard Brouer [off-list ref] noticed that XDP is ok
       with a higher order pages if it's a contiguous physical memory
       allocation, so falling to tun_alloc_skb() -> do_xdp_generic() should
       be ok.

     v2 -> v3:
     * attach the forgotten changelog

     v1 -> v2:
     * merged 2 patches in 1, fixing both issues: WARN_ON_ONCE with
       syzkaller repro and missing XDP_PACKET_HEADROOM in pad
     * changed the title and description of the execution path, suggested
       by Jason Wang [off-list ref]
     * move the limit check from tun_can_build_skb() to tun_build_skb() to
       remove duplication and locking issue, and also drop the packet in
       case of a failed check - noted by Jason Wang [off-list ref]

  drivers/net/tun.c | 11 ++++++++++-
  1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index d75456adc62a..a1d04bc9485f 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c

@@ -1582,6 +1582,9 @@ static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
  static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
                            int len, int noblock, bool zerocopy)
  {
+     struct bpf_prog *xdp_prog;
+     int pad = TUN_RX_PAD;
+
      if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
              return false;

@@ -1594,7 +1597,13 @@ static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
      if (zerocopy)
              return false;

-     if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
+     rcu_read_lock();
+     xdp_prog = rcu_dereference(tun->xdp_prog);
+     if (xdp_prog)
+             pad += XDP_PACKET_HEADROOM;
+     rcu_read_unlock();
+

Isolated seen, I guess, this is a correct fix to 7df13219d757.

I think so.

Actually, I think we can probably always count XDP_PACKET_HEADROOM
here. Since there's a window that XDP program might be attached in the
middle of tun_can_build_skb() and tun_build_skb().

quoted

+     if (SKB_DATA_ALIGN(len + pad) +
          SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
              return false;

Question to Jason Wang:
Why fall back (to e.g. tun_alloc_skb()) when size is above PAGE_SIZE?

AFAIK tun_build_skb() *can* create get larger packets than PAGE_SIZE
from it's page_frag.  Is there a reason for this limitation?

I couldn't recall but I think we can relax.

Thanks

(To Andrew, I assume a change in this area is another patch).

--Jesper