Re: [PATCH] datapath: Fix for force/commit action failures

2017-07-13 Thread Joe Stringer 
On 13 July 2017 at 15:38, Greg Rose  wrote:
> On 07/13/2017 11:03 AM, Joe Stringer wrote:
>>
>> On 13 July 2017 at 11:01, Greg Rose  wrote:
>>>
>>> On 07/13/2017 10:46 AM, Joe Stringer wrote:


 On 13 July 2017 at 09:25, Greg Rose  wrote:
>
>
> When there is an established connection in direction A->B, it is
> possible to receive a packet on port B which then executes
> ct(commit,force) without first performing ct() - ie, a lookup.
> In this case, we would expect that this packet can delete the existing
> entry so that we can commit a connection with direction B->A. However,
> currently we only perform a check in skb_nfct_cached() for whether
> OVS_CS_F_TRACKED is set and OVS_CS_F_INVALID is not set, ie that a
> lookup previously occurred. In the above scenario, a lookup has not
> occurred but we should still be able to statelessly look up the
> existing entry and potentially delete the entry if it is in the
> opposite direction.
>
> This patch extends the check to also hint that if the action has the
> force flag set, then we will lookup the existing entry so that the
> force check at the end of skb_nfct_cached has the ability to delete
> the connection.
>
> CC: d...@openvswitch.org
> CC: Pravin Shalar 
> Signed-off-by: Joe Stringer 
> Signed-off-by: Greg Rose 
> ---



 A couple more administrative notes, on netdev the module name in the
 patch subject for openvswitch is "openvswitch" rather than datapath;
>>>
>>>
>>>
>>> Right you are.
>>>
 and patches rather than having just "PATCH" as the subject prefix
 should state the tree. In this case, it's a bugfix so it should be
 "PATCH net".
>>>
>>>
>>>
>>> I knew that... forgot the format patch option to add it.  Net-next
>>> is closed so that would be mandatory.
>>>
>>>   Furthermore, if you're able to figure out which commit


 introduced the issue (I believe it's introduced by the force commit
 patch), then you should place the "Fixes: " tag. I can give you some
 pointers off-list on how to do this (git blame and some basic
 formatting of the targeted patch should do the trick - this tag
 expects a 12-digit hash).

 For reference, I ended up looking it up during review, this is the
 line you'd add:
 Fixes: dd41d33f0b03 ("openvswitch: Add force commit.")
>>>
>>>
>>>
>>> Oh, thanks!
>>>
>>>

>net/openvswitch/conntrack.c | 12 
>1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
> index 08679eb..9041cf8 100644
> --- a/net/openvswitch/conntrack.c
> +++ b/net/openvswitch/conntrack.c
> @@ -641,17 +641,21 @@ static bool skb_nfct_cached(struct net *net,
>   ct = nf_ct_get(skb, );
>   /* If no ct, check if we have evidence that an existing
> conntrack entry
>* might be found for this skb.  This happens when we lose a
> skb->_nfct
> -* due to an upcall.  If the connection was not confirmed, it
> is
> not
> -* cached and needs to be run through conntrack again.
> +* due to an upcall, or if the direction is being forced.  If
> the
> +* connection was not confirmed, it is not cached and needs to
> be
> run
> +* through conntrack again.
>*/
> -   if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
> +   if ((!ct && (key->ct_state & OVS_CS_F_TRACKED) &&
>   !(key->ct_state & OVS_CS_F_INVALID) &&
> -   key->ct_zone == info->zone.id) {
> +key->ct_zone == info->zone.id) ||
> +(!key->ct_state && info->force)) {
>   ct = ovs_ct_find_existing(net, >zone,
> info->family, skb,
> !!(key->ct_state
>& OVS_CS_F_NAT_MASK));
>   if (ct)
>   nf_ct_get(skb, );
> +   else
> +   return false;
>   }
>   if (!ct)
>   return false;



 I was just wondering if this has the potential to prevent
 nf_conntrack_in() from being called at all in this case, which is also
 not quite right. In the original case of (!ct && (key->ct_state &
 OVS_CS_F_TRACKED) && !(key->ct_state & OVS_CS_F_TRACKED)), which I'll
 refer to as "ct_executed", we explicitly want to avoid running
 nf_conntrack_in() if we already ran it, because the connection tracker
 doesn't expect to see the same packet twice (there's also things like
 stats/accounting, and potentially L4 state machines that

Re: [PATCH] datapath: Fix for force/commit action failures

2017-07-13 Thread Greg Rose 

On 07/13/2017 11:03 AM, Joe Stringer wrote:

On 13 July 2017 at 11:01, Greg Rose  wrote:

On 07/13/2017 10:46 AM, Joe Stringer wrote:


On 13 July 2017 at 09:25, Greg Rose  wrote:


When there is an established connection in direction A->B, it is
possible to receive a packet on port B which then executes
ct(commit,force) without first performing ct() - ie, a lookup.
In this case, we would expect that this packet can delete the existing
entry so that we can commit a connection with direction B->A. However,
currently we only perform a check in skb_nfct_cached() for whether
OVS_CS_F_TRACKED is set and OVS_CS_F_INVALID is not set, ie that a
lookup previously occurred. In the above scenario, a lookup has not
occurred but we should still be able to statelessly look up the
existing entry and potentially delete the entry if it is in the
opposite direction.

This patch extends the check to also hint that if the action has the
force flag set, then we will lookup the existing entry so that the
force check at the end of skb_nfct_cached has the ability to delete
the connection.

CC: d...@openvswitch.org
CC: Pravin Shalar 
Signed-off-by: Joe Stringer 
Signed-off-by: Greg Rose 
---



A couple more administrative notes, on netdev the module name in the
patch subject for openvswitch is "openvswitch" rather than datapath;



Right you are.


and patches rather than having just "PATCH" as the subject prefix
should state the tree. In this case, it's a bugfix so it should be
"PATCH net".



I knew that... forgot the format patch option to add it.  Net-next
is closed so that would be mandatory.

  Furthermore, if you're able to figure out which commit


introduced the issue (I believe it's introduced by the force commit
patch), then you should place the "Fixes: " tag. I can give you some
pointers off-list on how to do this (git blame and some basic
formatting of the targeted patch should do the trick - this tag
expects a 12-digit hash).

For reference, I ended up looking it up during review, this is the
line you'd add:
Fixes: dd41d33f0b03 ("openvswitch: Add force commit.")



Oh, thanks!





   net/openvswitch/conntrack.c | 12 
   1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
index 08679eb..9041cf8 100644
--- a/net/openvswitch/conntrack.c
+++ b/net/openvswitch/conntrack.c
@@ -641,17 +641,21 @@ static bool skb_nfct_cached(struct net *net,
  ct = nf_ct_get(skb, );
  /* If no ct, check if we have evidence that an existing
conntrack entry
   * might be found for this skb.  This happens when we lose a
skb->_nfct
-* due to an upcall.  If the connection was not confirmed, it is
not
-* cached and needs to be run through conntrack again.
+* due to an upcall, or if the direction is being forced.  If the
+* connection was not confirmed, it is not cached and needs to be
run
+* through conntrack again.
   */
-   if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
+   if ((!ct && (key->ct_state & OVS_CS_F_TRACKED) &&
  !(key->ct_state & OVS_CS_F_INVALID) &&
-   key->ct_zone == info->zone.id) {
+key->ct_zone == info->zone.id) ||
+(!key->ct_state && info->force)) {
  ct = ovs_ct_find_existing(net, >zone,
info->family, skb,
!!(key->ct_state
   & OVS_CS_F_NAT_MASK));
  if (ct)
  nf_ct_get(skb, );
+   else
+   return false;
  }
  if (!ct)
  return false;



I was just wondering if this has the potential to prevent
nf_conntrack_in() from being called at all in this case, which is also
not quite right. In the original case of (!ct && (key->ct_state &
OVS_CS_F_TRACKED) && !(key->ct_state & OVS_CS_F_TRACKED)), which I'll
refer to as "ct_executed", we explicitly want to avoid running
nf_conntrack_in() if we already ran it, because the connection tracker
doesn't expect to see the same packet twice (there's also things like
stats/accounting, and potentially L4 state machines that could get
messed up by multiple calls). By the time the info->force and
direction check happens at the end of the function, "ct_executed" is
implied to be true. However, in this new case, ct_executed is actually
false - because there was no ct() before the ct(force,commit). In this
case, we only want to look up the existing entry to see if it should
be deleted; if it should not be deleted, then we still haven't yet
done the nf_conntrack_in() call so we should return false and the
caller, __ovs_ct_lookup() should make the call to nf_conntrack_in().

What I mean is something like the following incremental on your patch:

diff --git a/net/openvswitch/conntrack.c

Re: [PATCH] datapath: Fix for force/commit action failures

2017-07-13 Thread Joe Stringer 
On 13 July 2017 at 11:01, Greg Rose  wrote:
> On 07/13/2017 10:46 AM, Joe Stringer wrote:
>>
>> On 13 July 2017 at 09:25, Greg Rose  wrote:
>>>
>>> When there is an established connection in direction A->B, it is
>>> possible to receive a packet on port B which then executes
>>> ct(commit,force) without first performing ct() - ie, a lookup.
>>> In this case, we would expect that this packet can delete the existing
>>> entry so that we can commit a connection with direction B->A. However,
>>> currently we only perform a check in skb_nfct_cached() for whether
>>> OVS_CS_F_TRACKED is set and OVS_CS_F_INVALID is not set, ie that a
>>> lookup previously occurred. In the above scenario, a lookup has not
>>> occurred but we should still be able to statelessly look up the
>>> existing entry and potentially delete the entry if it is in the
>>> opposite direction.
>>>
>>> This patch extends the check to also hint that if the action has the
>>> force flag set, then we will lookup the existing entry so that the
>>> force check at the end of skb_nfct_cached has the ability to delete
>>> the connection.
>>>
>>> CC: d...@openvswitch.org
>>> CC: Pravin Shalar 
>>> Signed-off-by: Joe Stringer 
>>> Signed-off-by: Greg Rose 
>>> ---
>>
>>
>> A couple more administrative notes, on netdev the module name in the
>> patch subject for openvswitch is "openvswitch" rather than datapath;
>
>
> Right you are.
>
>> and patches rather than having just "PATCH" as the subject prefix
>> should state the tree. In this case, it's a bugfix so it should be
>> "PATCH net".
>
>
> I knew that... forgot the format patch option to add it.  Net-next
> is closed so that would be mandatory.
>
>  Furthermore, if you're able to figure out which commit
>>
>> introduced the issue (I believe it's introduced by the force commit
>> patch), then you should place the "Fixes: " tag. I can give you some
>> pointers off-list on how to do this (git blame and some basic
>> formatting of the targeted patch should do the trick - this tag
>> expects a 12-digit hash).
>>
>> For reference, I ended up looking it up during review, this is the
>> line you'd add:
>> Fixes: dd41d33f0b03 ("openvswitch: Add force commit.")
>
>
> Oh, thanks!
>
>
>>
>>>   net/openvswitch/conntrack.c | 12 
>>>   1 file changed, 8 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
>>> index 08679eb..9041cf8 100644
>>> --- a/net/openvswitch/conntrack.c
>>> +++ b/net/openvswitch/conntrack.c
>>> @@ -641,17 +641,21 @@ static bool skb_nfct_cached(struct net *net,
>>>  ct = nf_ct_get(skb, );
>>>  /* If no ct, check if we have evidence that an existing
>>> conntrack entry
>>>   * might be found for this skb.  This happens when we lose a
>>> skb->_nfct
>>> -* due to an upcall.  If the connection was not confirmed, it is
>>> not
>>> -* cached and needs to be run through conntrack again.
>>> +* due to an upcall, or if the direction is being forced.  If the
>>> +* connection was not confirmed, it is not cached and needs to be
>>> run
>>> +* through conntrack again.
>>>   */
>>> -   if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
>>> +   if ((!ct && (key->ct_state & OVS_CS_F_TRACKED) &&
>>>  !(key->ct_state & OVS_CS_F_INVALID) &&
>>> -   key->ct_zone == info->zone.id) {
>>> +key->ct_zone == info->zone.id) ||
>>> +(!key->ct_state && info->force)) {
>>>  ct = ovs_ct_find_existing(net, >zone,
>>> info->family, skb,
>>>!!(key->ct_state
>>>   & OVS_CS_F_NAT_MASK));
>>>  if (ct)
>>>  nf_ct_get(skb, );
>>> +   else
>>> +   return false;
>>>  }
>>>  if (!ct)
>>>  return false;
>>
>>
>> I was just wondering if this has the potential to prevent
>> nf_conntrack_in() from being called at all in this case, which is also
>> not quite right. In the original case of (!ct && (key->ct_state &
>> OVS_CS_F_TRACKED) && !(key->ct_state & OVS_CS_F_TRACKED)), which I'll
>> refer to as "ct_executed", we explicitly want to avoid running
>> nf_conntrack_in() if we already ran it, because the connection tracker
>> doesn't expect to see the same packet twice (there's also things like
>> stats/accounting, and potentially L4 state machines that could get
>> messed up by multiple calls). By the time the info->force and
>> direction check happens at the end of the function, "ct_executed" is
>> implied to be true. However, in this new case, ct_executed is actually
>> false - because there was no ct() before the ct(force,commit). In this
>> case, we only want to look up the existing entry to see if it should
>> be deleted; if it should not be

Re: [PATCH] datapath: Fix for force/commit action failures

2017-07-13 Thread Greg Rose 

On 07/13/2017 10:46 AM, Joe Stringer wrote:

On 13 July 2017 at 09:25, Greg Rose  wrote:

When there is an established connection in direction A->B, it is
possible to receive a packet on port B which then executes
ct(commit,force) without first performing ct() - ie, a lookup.
In this case, we would expect that this packet can delete the existing
entry so that we can commit a connection with direction B->A. However,
currently we only perform a check in skb_nfct_cached() for whether
OVS_CS_F_TRACKED is set and OVS_CS_F_INVALID is not set, ie that a
lookup previously occurred. In the above scenario, a lookup has not
occurred but we should still be able to statelessly look up the
existing entry and potentially delete the entry if it is in the
opposite direction.

This patch extends the check to also hint that if the action has the
force flag set, then we will lookup the existing entry so that the
force check at the end of skb_nfct_cached has the ability to delete
the connection.

CC: d...@openvswitch.org
CC: Pravin Shalar 
Signed-off-by: Joe Stringer 
Signed-off-by: Greg Rose 
---


A couple more administrative notes, on netdev the module name in the
patch subject for openvswitch is "openvswitch" rather than datapath;


Right you are.


and patches rather than having just "PATCH" as the subject prefix
should state the tree. In this case, it's a bugfix so it should be
"PATCH net".


I knew that... forgot the format patch option to add it.  Net-next
is closed so that would be mandatory.

 Furthermore, if you're able to figure out which commit

introduced the issue (I believe it's introduced by the force commit
patch), then you should place the "Fixes: " tag. I can give you some
pointers off-list on how to do this (git blame and some basic
formatting of the targeted patch should do the trick - this tag
expects a 12-digit hash).

For reference, I ended up looking it up during review, this is the
line you'd add:
Fixes: dd41d33f0b03 ("openvswitch: Add force commit.")


Oh, thanks!




  net/openvswitch/conntrack.c | 12 
  1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
index 08679eb..9041cf8 100644
--- a/net/openvswitch/conntrack.c
+++ b/net/openvswitch/conntrack.c
@@ -641,17 +641,21 @@ static bool skb_nfct_cached(struct net *net,
 ct = nf_ct_get(skb, );
 /* If no ct, check if we have evidence that an existing conntrack entry
  * might be found for this skb.  This happens when we lose a skb->_nfct
-* due to an upcall.  If the connection was not confirmed, it is not
-* cached and needs to be run through conntrack again.
+* due to an upcall, or if the direction is being forced.  If the
+* connection was not confirmed, it is not cached and needs to be run
+* through conntrack again.
  */
-   if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
+   if ((!ct && (key->ct_state & OVS_CS_F_TRACKED) &&
 !(key->ct_state & OVS_CS_F_INVALID) &&
-   key->ct_zone == info->zone.id) {
+key->ct_zone == info->zone.id) ||
+(!key->ct_state && info->force)) {
 ct = ovs_ct_find_existing(net, >zone, info->family, skb,
   !!(key->ct_state
  & OVS_CS_F_NAT_MASK));
 if (ct)
 nf_ct_get(skb, );
+   else
+   return false;
 }
 if (!ct)
 return false;


I was just wondering if this has the potential to prevent
nf_conntrack_in() from being called at all in this case, which is also
not quite right. In the original case of (!ct && (key->ct_state &
OVS_CS_F_TRACKED) && !(key->ct_state & OVS_CS_F_TRACKED)), which I'll
refer to as "ct_executed", we explicitly want to avoid running
nf_conntrack_in() if we already ran it, because the connection tracker
doesn't expect to see the same packet twice (there's also things like
stats/accounting, and potentially L4 state machines that could get
messed up by multiple calls). By the time the info->force and
direction check happens at the end of the function, "ct_executed" is
implied to be true. However, in this new case, ct_executed is actually
false - because there was no ct() before the ct(force,commit). In this
case, we only want to look up the existing entry to see if it should
be deleted; if it should not be deleted, then we still haven't yet
done the nf_conntrack_in() call so we should return false and the
caller, __ovs_ct_lookup() should make the call to nf_conntrack_in().

What I mean is something like the following incremental on your patch:

diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
index 9041cf8b822f..98783f114824 100644
--- a/net/openvswitch/conntrack.c
+++ b/net/openvswitch/conntrack.c
@@ -637,6

Re: [PATCH] datapath: Fix for force/commit action failures

2017-07-13 Thread Joe Stringer 
On 13 July 2017 at 09:25, Greg Rose  wrote:
> When there is an established connection in direction A->B, it is
> possible to receive a packet on port B which then executes
> ct(commit,force) without first performing ct() - ie, a lookup.
> In this case, we would expect that this packet can delete the existing
> entry so that we can commit a connection with direction B->A. However,
> currently we only perform a check in skb_nfct_cached() for whether
> OVS_CS_F_TRACKED is set and OVS_CS_F_INVALID is not set, ie that a
> lookup previously occurred. In the above scenario, a lookup has not
> occurred but we should still be able to statelessly look up the
> existing entry and potentially delete the entry if it is in the
> opposite direction.
>
> This patch extends the check to also hint that if the action has the
> force flag set, then we will lookup the existing entry so that the
> force check at the end of skb_nfct_cached has the ability to delete
> the connection.
>
> CC: d...@openvswitch.org
> CC: Pravin Shalar 
> Signed-off-by: Joe Stringer 
> Signed-off-by: Greg Rose 
> ---

A couple more administrative notes, on netdev the module name in the
patch subject for openvswitch is "openvswitch" rather than datapath;
and patches rather than having just "PATCH" as the subject prefix
should state the tree. In this case, it's a bugfix so it should be
"PATCH net". Furthermore, if you're able to figure out which commit
introduced the issue (I believe it's introduced by the force commit
patch), then you should place the "Fixes: " tag. I can give you some
pointers off-list on how to do this (git blame and some basic
formatting of the targeted patch should do the trick - this tag
expects a 12-digit hash).

For reference, I ended up looking it up during review, this is the
line you'd add:
Fixes: dd41d33f0b03 ("openvswitch: Add force commit.")

>  net/openvswitch/conntrack.c | 12 
>  1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
> index 08679eb..9041cf8 100644
> --- a/net/openvswitch/conntrack.c
> +++ b/net/openvswitch/conntrack.c
> @@ -641,17 +641,21 @@ static bool skb_nfct_cached(struct net *net,
> ct = nf_ct_get(skb, );
> /* If no ct, check if we have evidence that an existing conntrack 
> entry
>  * might be found for this skb.  This happens when we lose a 
> skb->_nfct
> -* due to an upcall.  If the connection was not confirmed, it is not
> -* cached and needs to be run through conntrack again.
> +* due to an upcall, or if the direction is being forced.  If the
> +* connection was not confirmed, it is not cached and needs to be run
> +* through conntrack again.
>  */
> -   if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
> +   if ((!ct && (key->ct_state & OVS_CS_F_TRACKED) &&
> !(key->ct_state & OVS_CS_F_INVALID) &&
> -   key->ct_zone == info->zone.id) {
> +key->ct_zone == info->zone.id) ||
> +(!key->ct_state && info->force)) {
> ct = ovs_ct_find_existing(net, >zone, info->family, skb,
>   !!(key->ct_state
>  & OVS_CS_F_NAT_MASK));
> if (ct)
> nf_ct_get(skb, );
> +   else
> +   return false;
> }
> if (!ct)
> return false;

I was just wondering if this has the potential to prevent
nf_conntrack_in() from being called at all in this case, which is also
not quite right. In the original case of (!ct && (key->ct_state &
OVS_CS_F_TRACKED) && !(key->ct_state & OVS_CS_F_TRACKED)), which I'll
refer to as "ct_executed", we explicitly want to avoid running
nf_conntrack_in() if we already ran it, because the connection tracker
doesn't expect to see the same packet twice (there's also things like
stats/accounting, and potentially L4 state machines that could get
messed up by multiple calls). By the time the info->force and
direction check happens at the end of the function, "ct_executed" is
implied to be true. However, in this new case, ct_executed is actually
false - because there was no ct() before the ct(force,commit). In this
case, we only want to look up the existing entry to see if it should
be deleted; if it should not be deleted, then we still haven't yet
done the nf_conntrack_in() call so we should return false and the
caller, __ovs_ct_lookup() should make the call to nf_conntrack_in().

What I mean is something like the following incremental on your patch:

diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
index 9041cf8b822f..98783f114824 100644
--- a/net/openvswitch/conntrack.c
+++ b/net/openvswitch/conntrack.c
@@ -637,6 +637,7 @@ static bool skb_nfct_cached(struct net *net,
{
   enum ip_conntrack_info ctinfo;

Re: [ovs-dev] [PATCH] datapath: Fix for force/commit action failures

2017-07-13 Thread Greg Rose 

On 07/13/2017 10:08 AM, Darrell Ball wrote:



On 7/13/17, 9:25 AM, "ovs-dev-boun...@openvswitch.org on behalf of Greg Rose" 
 wrote:

 When there is an established connection in direction A->B, it is
 possible to receive a packet on port B which then executes
 ct(commit,force) without first performing ct() - ie, a lookup.
 In this case, we would expect that this packet can delete the existing
 entry so that we can commit a connection with direction B->A. However,
 currently we only perform a check in skb_nfct_cached() for whether
 OVS_CS_F_TRACKED is set and OVS_CS_F_INVALID is not set, ie that a
 lookup previously occurred. In the above scenario, a lookup has not
 occurred but we should still be able to statelessly look up the
 existing entry and potentially delete the entry if it is in the
 opposite direction.
 
 This patch extends the check to also hint that if the action has the

 force flag set, then we will lookup the existing entry so that the
 force check at the end of skb_nfct_cached has the ability to delete
 the connection.
 
 CC: d...@openvswitch.org

 CC: Pravin Shalar 
 Signed-off-by: Joe Stringer 
 Signed-off-by: Greg Rose 
 ---
  net/openvswitch/conntrack.c | 12 
  1 file changed, 8 insertions(+), 4 deletions(-)
 
 diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c

 index 08679eb..9041cf8 100644
 --- a/net/openvswitch/conntrack.c
 +++ b/net/openvswitch/conntrack.c
 @@ -641,17 +641,21 @@ static bool skb_nfct_cached(struct net *net,
ct = nf_ct_get(skb, );
/* If no ct, check if we have evidence that an existing conntrack entry
 * might be found for this skb.  This happens when we lose a skb->_nfct
 -   * due to an upcall.  If the connection was not confirmed, it is not
 -   * cached and needs to be run through conntrack again.
 +   * due to an upcall, or if the direction is being forced.  If the
 +   * connection was not confirmed, it is not cached and needs to be run
 +   * through conntrack again.
 */
 -  if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
 +  if ((!ct && (key->ct_state & OVS_CS_F_TRACKED) &&
!(key->ct_state & OVS_CS_F_INVALID) &&
 -  key->ct_zone == info->zone.id) {
 +   key->ct_zone == info->zone.id) ||
 +   (!key->ct_state && info->force)) {
ct = ovs_ct_find_existing(net, >zone, info->family, skb,
  !!(key->ct_state
 & OVS_CS_F_NAT_MASK));
if (ct)
nf_ct_get(skb, );
 +  else
 +  return false;

the above else case looks redundant since it maps to the following check
if (!ct)
return false;
which services a superset of the code flow.


Sure, we can let it fall through... missed that.

After waiting for more review I'll send a V2 patch.

Thanks for the review Darrell

- Greg



}
if (!ct)
return false;
 --
 1.8.3.1
 
 ___

 dev mailing list
 d...@openvswitch.org
 
https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.openvswitch.org_mailman_listinfo_ovs-2Ddev=DwICAg=uilaK90D4TOVoH58JNXRgQ=BVhFA09CGX7JQ5Ih-uZnsw=5da6ykiSQJBoJXmpq6iVknmPAc5JDzlVngmp8j_xcXA=PsX-njQ_hFqy8P77KNEyX0i7u165p2Wrbg4uAYqfbGs=

Re: [ovs-dev] [PATCH] datapath: Fix for force/commit action failures

2017-07-13 Thread Darrell Ball 


On 7/13/17, 9:25 AM, "ovs-dev-boun...@openvswitch.org on behalf of Greg Rose" 
 wrote:

When there is an established connection in direction A->B, it is
possible to receive a packet on port B which then executes
ct(commit,force) without first performing ct() - ie, a lookup.
In this case, we would expect that this packet can delete the existing
entry so that we can commit a connection with direction B->A. However,
currently we only perform a check in skb_nfct_cached() for whether
OVS_CS_F_TRACKED is set and OVS_CS_F_INVALID is not set, ie that a
lookup previously occurred. In the above scenario, a lookup has not
occurred but we should still be able to statelessly look up the
existing entry and potentially delete the entry if it is in the
opposite direction.

This patch extends the check to also hint that if the action has the
force flag set, then we will lookup the existing entry so that the
force check at the end of skb_nfct_cached has the ability to delete
the connection.

CC: d...@openvswitch.org
CC: Pravin Shalar 
Signed-off-by: Joe Stringer 
Signed-off-by: Greg Rose 
---
 net/openvswitch/conntrack.c | 12 
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
index 08679eb..9041cf8 100644
--- a/net/openvswitch/conntrack.c
+++ b/net/openvswitch/conntrack.c
@@ -641,17 +641,21 @@ static bool skb_nfct_cached(struct net *net,
ct = nf_ct_get(skb, );
/* If no ct, check if we have evidence that an existing conntrack entry
 * might be found for this skb.  This happens when we lose a skb->_nfct
-* due to an upcall.  If the connection was not confirmed, it is not
-* cached and needs to be run through conntrack again.
+* due to an upcall, or if the direction is being forced.  If the
+* connection was not confirmed, it is not cached and needs to be run
+* through conntrack again.
 */
-   if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
+   if ((!ct && (key->ct_state & OVS_CS_F_TRACKED) &&
!(key->ct_state & OVS_CS_F_INVALID) &&
-   key->ct_zone == info->zone.id) {
+key->ct_zone == info->zone.id) ||
+(!key->ct_state && info->force)) {
ct = ovs_ct_find_existing(net, >zone, info->family, skb,
  !!(key->ct_state
 & OVS_CS_F_NAT_MASK));
if (ct)
nf_ct_get(skb, );
+   else
+   return false;

the above else case looks redundant since it maps to the following check
if (!ct)
return false;
which services a superset of the code flow.

}
if (!ct)
return false;
-- 
1.8.3.1

___
dev mailing list
d...@openvswitch.org

https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.openvswitch.org_mailman_listinfo_ovs-2Ddev=DwICAg=uilaK90D4TOVoH58JNXRgQ=BVhFA09CGX7JQ5Ih-uZnsw=5da6ykiSQJBoJXmpq6iVknmPAc5JDzlVngmp8j_xcXA=PsX-njQ_hFqy8P77KNEyX0i7u165p2Wrbg4uAYqfbGs=

[PATCH] datapath: Fix for force/commit action failures

2017-07-13 Thread Greg Rose 
When there is an established connection in direction A->B, it is
possible to receive a packet on port B which then executes
ct(commit,force) without first performing ct() - ie, a lookup.
In this case, we would expect that this packet can delete the existing
entry so that we can commit a connection with direction B->A. However,
currently we only perform a check in skb_nfct_cached() for whether
OVS_CS_F_TRACKED is set and OVS_CS_F_INVALID is not set, ie that a
lookup previously occurred. In the above scenario, a lookup has not
occurred but we should still be able to statelessly look up the
existing entry and potentially delete the entry if it is in the
opposite direction.

This patch extends the check to also hint that if the action has the
force flag set, then we will lookup the existing entry so that the
force check at the end of skb_nfct_cached has the ability to delete
the connection.

CC: d...@openvswitch.org
CC: Pravin Shalar 
Signed-off-by: Joe Stringer 
Signed-off-by: Greg Rose 
---
 net/openvswitch/conntrack.c | 12 
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
index 08679eb..9041cf8 100644
--- a/net/openvswitch/conntrack.c
+++ b/net/openvswitch/conntrack.c
@@ -641,17 +641,21 @@ static bool skb_nfct_cached(struct net *net,
ct = nf_ct_get(skb, );
/* If no ct, check if we have evidence that an existing conntrack entry
 * might be found for this skb.  This happens when we lose a skb->_nfct
-* due to an upcall.  If the connection was not confirmed, it is not
-* cached and needs to be run through conntrack again.
+* due to an upcall, or if the direction is being forced.  If the
+* connection was not confirmed, it is not cached and needs to be run
+* through conntrack again.
 */
-   if (!ct && key->ct_state & OVS_CS_F_TRACKED &&
+   if ((!ct && (key->ct_state & OVS_CS_F_TRACKED) &&
!(key->ct_state & OVS_CS_F_INVALID) &&
-   key->ct_zone == info->zone.id) {
+key->ct_zone == info->zone.id) ||
+(!key->ct_state && info->force)) {
ct = ovs_ct_find_existing(net, >zone, info->family, skb,
  !!(key->ct_state
 & OVS_CS_F_NAT_MASK));
if (ct)
nf_ct_get(skb, );
+   else
+   return false;
}
if (!ct)
return false;
-- 
1.8.3.1