> -----邮件原件-----
> 发件人: Li,Rongqing
> 发送时间: 2020年7月21日 9:43
> 收件人: 'Magnus Karlsson' <magnus.karls...@gmail.com>
> 抄送: Network Development <netdev@vger.kernel.org>; intel-wired-lan
> <intel-wired-...@lists.osuosl.org>; Karlsson, Magnus
> <magnus.karls...@intel.com>; Björn Töpel <bjorn.to...@intel.com>
> 主题: 答复: [Intel-wired-lan] [PATCH 1/2] xdp: i40e: ixgbe: ixgbevf: not flip rx
> buffer for copy mode xdp
>
>
>
> > -----邮件原件-----
> > 发件人: Magnus Karlsson [mailto:magnus.karls...@gmail.com]
> > 发送时间: 2020年7月20日 15:21
> > 收件人: Li,Rongqing <lirongq...@baidu.com>
> > 抄送: Network Development <netdev@vger.kernel.org>; intel-wired-lan
> > <intel-wired-...@lists.osuosl.org>; Karlsson, Magnus
> > <magnus.karls...@intel.com>; Björn Töpel <bjorn.to...@intel.com>
> > 主题: Re: [Intel-wired-lan] [PATCH 1/2] xdp: i40e: ixgbe: ixgbevf: not
> > flip rx buffer for copy mode xdp
> >
> > On Fri, Jul 17, 2020 at 8:24 AM Li RongQing <lirongq...@baidu.com> wrote:
> > >
> > > i40e/ixgbe/ixgbevf_rx_buffer_flip in copy mode xdp can lead to data
> > > corruption, like the following flow:
> > >
> > > 1. first skb is not for xsk, and forwarded to another device
> > > or socket queue
> > > 2. seconds skb is for xsk, copy data to xsk memory, and page
> > > of skb->data is released
> > > 3. rx_buff is reusable since only first skb is in it, but
> > > *_rx_buffer_flip will make that page_offset is set to
> > > first skb data
> > > 4. then reuse rx buffer, first skb which still is living
> > > will be corrupted.
> e, but known size type */
> > > u32 id;
> > > @@ -73,6 +75,7 @@ struct xdp_buff {
> > > struct xdp_rxq_info *rxq;
> > > struct xdp_txq_info *txq;
> > > u32 frame_sz; /* frame size to deduce data_hard_end/reserved
> > > tailroom*/
> > > + u32 flags;
> >
> > RongQing,
> >
> > Sorry that I was not clear enough. Could you please submit the simple
> > patch you had, the one that only tests for the memory type.
> >
> > if (xdp->rxq->mem.type != MEM_TYPE_XSK_BUFF_POOL)
> > i40e_rx_buffer_flip(rx_ring, rx_buffer, size);
> >
> > I do not think that adding a flags field in the xdp_mem_info to fix an
> > Intel driver problem will be hugely popular. The struct is also meant
> > to contain long lived information, not things that will frequently change.
> >
>
>
> Thank you Magnus
>
> My original suggestion is wrong , it should be following
>
> if (xdp->rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL)
> i40e_rx_buffer_flip(rx_ring, rx_buffer, size);
>
>
> but I feel it is not enough to only check mem.type, it must ensure that
> map_type is BPF_MAP_TYPE_XSKMAP ? but it is not expose.
>
> other maptype, like BPF_MAP_TYPE_DEVMAP, and if mem.type is
> MEM_TYPE_PAGE_SHARED, not flip the rx buffer, will cause data corruption.
>
>
> -Li
>
>
How about this?
--- a/drivers/net/ethernet/intel/i40e/i40e_txrx.c
+++ b/drivers/net/ethernet/intel/i40e/i40e_txrx.c
@@ -2394,7 +2394,10 @@ static int i40e_clean_rx_irq(struct i40e_ring *rx_ring,
int budget)
if (xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR)) {
xdp_xmit |= xdp_res;
- i40e_rx_buffer_flip(rx_ring, rx_buffer, size);
+
+ if (xdp.rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL
||
+ xdp_get_map_type() != BPF_MAP_TYPE_XSKMAP)
+ i40e_rx_buffer_flip(rx_ring, rx_buffer,
size);
} else {
rx_buffer->pagecnt_bias++;
}
diff --git a/include/linux/filter.h b/include/linux/filter.h
index 259377723603..94f4435a77f3 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -919,6 +919,17 @@ static inline void xdp_clear_return_frame_no_direct(void)
ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
}
+static enum bpf_map_type xdp_get_map_type(void)
+{
+ struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+ struct bpf_map *map = READ_ONCE(ri->map);
+
+ if (map)
+ return map->map_type;
+ else
+ return BPF_MAP_TYPE_UNSPEC;
+}
+
static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
unsigned int pktlen)
