On 4/13/2021 9:18 AM, Ori Kam wrote:
Hi Ferruh,
-----Original Message-----
From: Ferruh Yigit <ferruh.yi...@intel.com>
Subject: Re: [PATCH v2 1/2] ethdev: add packet integrity checks
On 4/13/2021 8:12 AM, Ori Kam wrote:
Hi Ferruh,
-----Original Message-----
From: Ferruh Yigit <ferruh.yi...@intel.com>
Subject: Re: [PATCH v2 1/2] ethdev: add packet integrity checks
On 4/12/2021 8:26 PM, Ori Kam wrote:
Hi Ferruh,
-----Original Message-----
From: Ferruh Yigit <ferruh.yi...@intel.com>
Subject: Re: [PATCH v2 1/2] ethdev: add packet integrity checks
On 4/11/2021 6:34 PM, Gregory Etelson wrote:
From: Ori Kam <or...@nvidia.com>
Currently, DPDK application can offload the checksum check,
and report it in the mbuf.
However, as more and more applications are offloading some or all
logic and action to the HW, there is a need to check the packet
integrity so the right decision can be taken.
The application logic can be positive meaning if the packet is
valid jump / do actions, or negative if packet is not valid
jump to SW / do actions (like drop) a, and add default flow
(match all in low priority) that will direct the miss packet
to the miss path.
Since currenlty rte_flow works in positive way the assumtion is
that the postive way will be the common way in this case also.
When thinking what is the best API to implement such feature,
we need to considure the following (in no specific order):
1. API breakage.
2. Simplicity.
3. Performance.
4. HW capabilities.
5. rte_flow limitation.
6. Flexability.
First option: Add integrity flags to each of the items.
For example add checksum_ok to ipv4 item.
Pros:
1. No new rte_flow item.
2. Simple in the way that on each item the app can see
what checks are available.
Cons:
1. API breakage.
2. increase number of flows, since app can't add global rule and
must have dedicated flow for each of the flow combinations, for
example
matching on icmp traffic or UDP/TCP traffic with IPv4 / IPv6 will
result in 5 flows.
Second option: dedicated item
Pros:
1. No API breakage, and there will be no for some time due to having
extra space. (by using bits)
2. Just one flow to support the icmp or UDP/TCP traffic with IPv4 /
IPv6.
3. Simplicity application can just look at one place to see all possible
checks.
4. Allow future support for more tests.
Cons:
1. New item, that holds number of fields from different items.
For starter the following bits are suggested:
1. packet_ok - means that all HW checks depending on packet layer have
passed. This may mean that in some HW such flow should be splited
to
number of flows or fail.
2. l2_ok - all check flor layer 2 have passed.
3. l3_ok - all check flor layer 2 have passed. If packet doens't have
l3 layer this check shoudl fail.
4. l4_ok - all check flor layer 2 have passed. If packet doesn't
have l4 layer this check should fail.
5. l2_crc_ok - the layer 2 crc is O.K. it is possible that the crc will
be O.K. but the l3_ok will be 0. it is not possible that l2_crc_ok will
be 0 and the l3_ok will be 0.
6. ipv4_csum_ok - IPv4 checksum is O.K.
7. l4_csum_ok - layer 4 checksum is O.K.
8. l3_len_OK - check that the reported layer 3 len is smaller than the
packet len.
Example of usage:
1. check packets from all possible layers for integrity.
flow create integrity spec packet_ok = 1 mask packet_ok = 1 .....
2. Check only packet with layer 4 (UDP / TCP)
flow create integrity spec l3_ok = 1, l4_ok = 1 mask l3_ok = 1 l4_ok =
1
Hi Ori,
Is the intention of the API just filtering, like apply some action to the
packets based on their integration status. Like drop packets their l2_crc
checksum failed? Here configuration is done by existing offload APIs.
Or is the intention to configure the integration check on NIC, like to say
enable layer 2 checks, and do the action based on integration check
status.
If I understand your question the first case is the one that this patch is
targeting.
meaning based on those bits route/apply actions to the packet while still in
the
HW.
This is not design to enable the queue status bits.
In the use case suggestion by this patch, just like you said the app
can decide to drop the packet before arriving to the queue, application
may
also
use the mark + queue action to mark to the SW what is the issue with this
packet.
I'm not sure I understand your comment about "here configuration is done
by
existing
offload API" do you mean like the drop / jump to table / any other rte_flow
action?
I am asking because difference between device configuration and packet
filtering
seems getting more blurred in the flow API.
Currently L4 checksum offload is requested by application via setting
'DEV_RX_OFFLOAD_TCP_CKSUM' (UDP/SCTP/...) offload flags. This is the
way
to
configure HW.
Is the intention of this patch doing packet filtering after device configured
with above offload API?
Or is the intention HW to be configured via flow API, like if "l4_ok = 1" is set
in the rule, will it enable L4 checks first and do the filtering later?
If not what is the expected behavior when integration checks are not
enabled
when the rule is created?
Let me try to explain it in a different way:
When application enables 'DEV_RX_OFFLOAD_TCP_CKSUM' (UDP/SCTP/...)
offload flags
It only means that the HW will report the checksum in the mbuf.
It is not only for reporting in mbuf, it also configures the HW to enable the
relevant checks. This is for device configuration.
Are these checks always enabled in mlx devices?
Yes, they are always on.
The RX offload just enables the RX burst to update the mbuf.
In any case if HW needs to be enabled, then
the PMD can enable the HW when first flow is inserted and just not
copy the value the mbuf.
This was my initial question, if the new rule is just for filtering or
"configuring device + filtering".
So, two questions:
1) Do we want to duplicate the way to configure the HW checks
2) Do we want want to use flow API as a device configuration interface, (as said
before I am feeling we are going that path more and more)
Since the HW is always on in your case that is not a big problem for you but (1)
can cause trouble for other vendors, for the cases like if HW check enabled via
flow API, later it is disabled by clearing the offload flag, what will be the
status of the flow rule?
Lets call this mode RX queue offload.
Now I'm introducing rte_flow offload,
This means that the application can create rte_flow that matches those
bits and based on that take different actions that are defined by the rte_flow.
Example for such a flow
Flow create 0 ingress pattern integrity spec packet_ok = 0 mask packet_ok =
1 / end actions count / drop / end
Offloading such a flow will result that all invalid packets will be counted and
dropped, in the HW
so even if the RX queue offload was enabled, no invalid packets will arrive to
the application.
In other case lets assume that the application wants all valid packets to jump
to the next group,
and all the reset of the packets will go to SW. (we can assume that later in
the pipeline also valid packets
will arrive to the application)
Flow create 0 ingress pattern integrity spec packet_ok = 1 mask packet_ok =
1 / end actions jump group 1 / end
Flow create 0 priority 1 pattern eth / end actions rss / end
In this case if the application enabled the RX offload then if the application
will receive invalid packet
also the flag in the mbuf will be set.
If application not enabled the Rx offload that means HW checks are not
enabled,
so HW can't know if a packet is OK or not, right.
In that case, for above rule, I expect none of the packets to match, so none
will jump to next group. Are we in the same page here?
Or do you expect above rule configure the HW to enable the relevant HW
checks first?
If this is required by HW then yes.
please see my answer above.
As you can see those two offloads mode are complementary to each other
and one doesn't force the other one in any
way.
I hope this is clearer.
Signed-off-by: Ori Kam <or...@nvidia.com>
---
v2: fix compilation error
---
doc/guides/prog_guide/rte_flow.rst | 19 ++++++++++++
lib/librte_ethdev/rte_flow.h | 47
++++++++++++++++++++++++++++++
2 files changed, 66 insertions(+)
diff --git a/doc/guides/prog_guide/rte_flow.rst
b/doc/guides/prog_guide/rte_flow.rst
index e1b93ecedf..87ef591405 100644
--- a/doc/guides/prog_guide/rte_flow.rst
+++ b/doc/guides/prog_guide/rte_flow.rst
@@ -1398,6 +1398,25 @@ Matches a eCPRI header.
- ``hdr``: eCPRI header definition (``rte_ecpri.h``).
- Default ``mask`` matches nothing, for all eCPRI messages.
+Item: ``PACKET_INTEGRITY_CHECKS``
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Matches packet integrity.
+
+- ``level``: the encapsulation level that should be checked. level 0
means
the
+ default PMD mode (Can be inner most / outermost). value of 1 means
outermost
+ and higher value means inner header. See also RSS level.
+- ``packet_ok``: All HW packet integrity checks have passed based on
the
max
+ layer of the packet.
+ layer of the packet.
+- ``l2_ok``: all layer 2 HW integrity checks passed.
+- ``l3_ok``: all layer 3 HW integrity checks passed.
+- ``l4_ok``: all layer 3 HW integrity checks passed.
s/layer 3/ layer 4/
Will fix.
+- ``l2_crc_ok``: layer 2 crc check passed.
+- ``ipv4_csum_ok``: ipv4 checksum check passed.
+- ``l4_csum_ok``: layer 4 checksum check passed.
+- ``l3_len_ok``: the layer 3 len is smaller than the packet len.
+
Actions
~~~~~~~
diff --git a/lib/librte_ethdev/rte_flow.h b/lib/librte_ethdev/rte_flow.h
index 6cc57136ac..77471af2c4 100644
--- a/lib/librte_ethdev/rte_flow.h
+++ b/lib/librte_ethdev/rte_flow.h
@@ -551,6 +551,15 @@ enum rte_flow_item_type {
* See struct rte_flow_item_geneve_opt
*/
RTE_FLOW_ITEM_TYPE_GENEVE_OPT,
+
+ /**
+ * [META]
+ *
+ * Matches on packet integrity.
+ *
+ * See struct rte_flow_item_integrity.
+ */
+ RTE_FLOW_ITEM_TYPE_INTEGRITY,
};
/**
@@ -1685,6 +1694,44 @@ rte_flow_item_geneve_opt_mask = {
};
#endif
+__extension__
+struct rte_flow_item_integrity {
+ uint32_t level;
+ /**< Packet encapsulation level the item should apply to.
+ * @see rte_flow_action_rss
+ */
+ union {
+ struct {
+ uint64_t packet_ok:1;
+ /** The packet is valid after passing all HW
checks. */
+ uint64_t l2_ok:1;
+ /**< L2 layer is valid after passing all HW
checks. */
+ uint64_t l3_ok:1;
+ /**< L3 layer is valid after passing all HW
checks. */
+ uint64_t l4_ok:1;
+ /**< L4 layer is valid after passing all HW
checks. */
+ uint64_t l2_crc_ok:1;
+ /**< L2 layer checksum is valid. */
+ uint64_t ipv4_csum_ok:1;
+ /**< L3 layer checksum is valid. */
+ uint64_t l4_csum_ok:1;
+ /**< L4 layer checksum is valid. */
+ uint64_t l3_len_ok:1;
+ /**< The l3 len is smaller than the packet len.
*/
packet len?
Do you mean replace the l3_len_ok with packet len?
no, I was trying to ask what is "packet len" here? frame length, or mbuf
buffer
length, or something else?
Frame length.
My only issue is that the check is comparing the l3 len to the packet len.
If you still think it is better to call it packet len, I'm also O.K with it.
+ uint64_t reserved:56;
+ };
+ uint64_t value;
+ };
+};
+
+#ifndef __cplusplus
+static const struct rte_flow_item_integrity
+rte_flow_item_integrity_mask = {
+ .level = 0,
+ .value = 0,
+};
+#endif
+
/**
* Matching pattern item definition.
*
