[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH] xen-netback: count number required slots for an skb more carefully




On 2013-9-4 5:53, Wei Liu wrote:
On Tue, Sep 03, 2013 at 06:29:50PM +0100, David Vrabel wrote:
From: David Vrabel <david.vrabel@xxxxxxxxxx>

When a VM is providing an iSCSI target and the LUN is used by the
backend domain, the generated skbs for direct I/O writes to the disk
have large, multi-page skb->data but no frags.

With some lengths and starting offsets, xen_netbk_count_skb_slots()
would be one short because the simple calculation of
DIV_ROUND_UP(skb_headlen(), PAGE_SIZE) was not accounting for the
decisions made by start_new_rx_buffer() which does not guarantee
responses are fully packed.

For example, a skb with length < 2 pages but which spans 3 pages would
be counted as requiring 2 slots but would actually use 3 slots.

skb->data:

     |        1111|222222222222|3333        |

Fully packed, this would need 2 slots:

     |111122222222|22223333    |

But because the 2nd page wholy fits into a slot it is not split across
slots and goes into a slot of its own:

     |1111        |222222222222|3333        |

Miscounting the number of slots means netback may push more responses
than the number of available requests.  This will cause the frontend
to get very confused and report "Too many frags/slots".  The frontend
never recovers and will eventually BUG.

Fix this by counting the number of required slots more carefully.  In
xen_netbk_count_skb_slots(), more closely follow the algorithm used by
xen_netbk_gop_skb() by introducing xen_netbk_count_frag_slots() which
is the dry-run equivalent of netbk_gop_frag_copy().

Phew! So this is backend miscounting bug. I thought it was a frontend
bug so it didn't ring a bell when we had our face-to-face discussion,
sorry. :-(

This bug was discussed back in July among Annie, Matt, Ian and I. We
finally agreed to take Matt's solution. Matt agreed to post final
version within a week but obviously he's too busy to do so. I was away
so I didn't follow closely. Eventually it fell through the crack. :-(

The fixes can be implemented in two ways, one is fix in xen_netbk_count_skb_slots to return correct slot count, my patch(http://lists.xen.org/archives/html/xen-devel/2013-07/msg00785.html) and David's fall in this way. The other way is fixed in netbk_gop_frag_copy which is Matt's(http://lists.xen.org/archives/html/xen-devel/2013-07/msg00760.html).

Matt, do you fancy sending the final version? IIRC the commit message
needs to be re-written. I personally still prefer Matt's solution as
it a) make efficient use of the ring, b) uses ring pointers to
calculate slots which is most accurate, c) removes the dependence on
MAX_SKB_FRAGS in guest RX path.

Anyway, we should get this fixed ASAP.

Totally agree. This issue is easy to be reproduced with large MTU. It is better to upstream the fix soon in case others hit it and waste time to fix it.

Thanks
Annie

Thanks
Wei.

REF:
<1373409659-22383-1-git-send-email-msw@xxxxxxxxxx>
<1373350520-19985-1-git-send-email-annie.li@xxxxxxxxxx>


Signed-off-by: David Vrabel <david.vrabel@xxxxxxxxxx>
---
[Resend to actually Cc netdev, sorry.]
---
  drivers/net/xen-netback/netback.c |   94 +++++++++++++++++++++++++------------
  1 files changed, 64 insertions(+), 30 deletions(-)

diff --git a/drivers/net/xen-netback/netback.c 
b/drivers/net/xen-netback/netback.c
index 64828de..f149ce5 100644
--- a/drivers/net/xen-netback/netback.c
+++ b/drivers/net/xen-netback/netback.c
@@ -361,6 +361,49 @@ static bool start_new_rx_buffer(int offset, unsigned long 
size, int head)
        return false;
  }
+struct xen_netbk_count_slot_state {
+       unsigned long copy_off;
+       bool head;
+};
+
+unsigned int xen_netbk_count_frag_slots(struct xenvif *vif,
+                                       unsigned long offset, unsigned long 
size,
+                                       struct xen_netbk_count_slot_state 
*state)
+{
+       unsigned count = 0;
+
+       offset &= ~PAGE_MASK;
+
+       while (size > 0) {
+               unsigned long bytes;
+
+               bytes = PAGE_SIZE - offset;
+
+               if (bytes > size)
+                       bytes = size;
+
+               if (start_new_rx_buffer(state->copy_off, bytes, state->head)) {
+                       count++;
+                       state->copy_off = 0;
+               }
+
+               if (state->copy_off + bytes > MAX_BUFFER_OFFSET)
+                       bytes = MAX_BUFFER_OFFSET - state->copy_off;
+
+               state->copy_off += bytes;
+
+               offset += bytes;
+               size -= bytes;
+
+               if (offset == PAGE_SIZE)
+                       offset = 0;
+
+               state->head = false;
+       }
+
+       return count;
+}
+
  /*
   * Figure out how many ring slots we're going to need to send @skb to
   * the guest. This function is essentially a dry run of
@@ -368,48 +411,39 @@ static bool start_new_rx_buffer(int offset, unsigned long 
size, int head)
   */
  unsigned int xen_netbk_count_skb_slots(struct xenvif *vif, struct sk_buff 
*skb)
  {
+       struct xen_netbk_count_slot_state state;
        unsigned int count;
-       int i, copy_off;
+       unsigned char *data;
+       unsigned i;
- count = DIV_ROUND_UP(skb_headlen(skb), PAGE_SIZE);
+       state.head = true;
+       state.copy_off = 0;
- copy_off = skb_headlen(skb) % PAGE_SIZE;
+       /* Slot for the first (partial) page of data. */
+       count = 1;
+ /* Need a slot for the GSO prefix for GSO extra data? */
        if (skb_shinfo(skb)->gso_size)
                count++;
- for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
-               unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
-               unsigned long offset = skb_shinfo(skb)->frags[i].page_offset;
-               unsigned long bytes;
-
-               offset &= ~PAGE_MASK;
-
-               while (size > 0) {
-                       BUG_ON(offset >= PAGE_SIZE);
-                       BUG_ON(copy_off > MAX_BUFFER_OFFSET);
-
-                       bytes = PAGE_SIZE - offset;
-
-                       if (bytes > size)
-                               bytes = size;
+       data = skb->data;
+       while (data < skb_tail_pointer(skb)) {
+               unsigned long offset = offset_in_page(data);
+               unsigned long size = PAGE_SIZE - offset;
- if (start_new_rx_buffer(copy_off, bytes, 0)) {
-                               count++;
-                               copy_off = 0;
-                       }
+               if (data + size > skb_tail_pointer(skb))
+                       size = skb_tail_pointer(skb) - data;
- if (copy_off + bytes > MAX_BUFFER_OFFSET)
-                               bytes = MAX_BUFFER_OFFSET - copy_off;
+               count += xen_netbk_count_frag_slots(vif, offset, size, &state);
- copy_off += bytes;
+               data += size;
+       }
- offset += bytes;
-                       size -= bytes;
+       for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+               unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
+               unsigned long offset = skb_shinfo(skb)->frags[i].page_offset;
- if (offset == PAGE_SIZE)
-                               offset = 0;
-               }
+               count += xen_netbk_count_frag_slots(vif, offset, size, &state);
        }
        return count;
  }
--
1.7.2.5

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.