[dpdk-dev] [RFC] ethdev: introduce conntrack flow action and item

[Bing Zhao]

Hello,

> -----Original Message-----
> From: Ajit Khaparde <ajit.khaparde at broadcom.com>
> Sent: Wednesday, March 24, 2021 7:27 AM
> To: Bing Zhao <bingz at nvidia.com>
> Cc: Ori Kam <orika at nvidia.com>; NBU-Contact-Thomas Monjalon
> <thomas at monjalon.net>; Ferruh Yigit <ferruh.yigit at intel.com>; Andrew
> Rybchenko <andrew.rybchenko at oktetlabs.ru>; dpdk-dev <dev at dpdk.org>
> Subject: Re: [dpdk-dev] [RFC] ethdev: introduce conntrack flow
> action and item
> 
> On Thu, Mar 18, 2021 at 12:30 AM Bing Zhao <bingz at nvidia.com> wrote:
> >
> > This commit introduced the conntrack action and item.
> >
> > Usually the HW offloading is stateless. For some stateful
> offloading
> > like a TCP connection, HW module will help provide the ability of
> a
> > full offloading w/o SW participation after the connection was
> > established.
> >
> > The basic usage is that in the first flow the application should
> add
> > the conntrack action and in the following flow(s) the application
> > should use the conntrack item to match on the result.
> >
> > A TCP connection has two directions traffic. To set a conntrack
> > action context correctly, information from packets of both
> directions
> > are required.
> >
> > The conntrack action should be created on one port and supply the
> > peer port as a parameter to the action. After context creating, it
> > could only be used between the ports (dual-port mode) or a single
> > port. The application should modify the action via
> action_ctx_update
> > interface before each use in dual-port mode, in order to set the
> > correct direction for the following rte flow.
> >
> > Query will be supported via action_ctx_query interface, about the
> > current packets information and connection status.
> >
> > For the packets received during the conntrack setup, it is
> suggested
> > to re-inject the packets in order to take full advantage of the
> > conntrack. Only the valid packets should pass the conntrack,
> packets
> > with invalid TCP information, like out of window, or with invalid
> > header, like malformed, should not pass.
> >
> > Testpmd command line example:
> >
> > set conntrack [index] enable is 1 last_seq is xxx last ack is xxx
> /
> > ... / orig_dir win_scale is xxx sent_end is xxx max_win is xxx ...
> /
> > rply_dir ... / end
> > flow action_ctx [CTX] create ingress ... / conntrack is [index] /
> end
> > flow create 0 group X ingress patterns ... / tcp / end actions
> action_ctx [CTX]
> > / jump group Y / end
> > flow create 0 group Y ingress patterns ... / ct is [Valid] / end
> actions
> > queue index [hairpin queue] / end
> >
> > Signed-off-by: Bing Zhao <bingz at nvidia.com>
> > ---
> >  lib/librte_ethdev/rte_flow.h | 191
> +++++++++++++++++++++++++++++++++++
> >  1 file changed, 191 insertions(+)
> >
> > diff --git a/lib/librte_ethdev/rte_flow.h
> b/lib/librte_ethdev/rte_flow.h
> > index 669e677e91..b2e4f0751a 100644
> > --- a/lib/librte_ethdev/rte_flow.h
> > +++ b/lib/librte_ethdev/rte_flow.h
> > @@ -550,6 +550,15 @@ enum rte_flow_item_type {
> >          * See struct rte_flow_item_geneve_opt
> >          */
> >         RTE_FLOW_ITEM_TYPE_GENEVE_OPT,
> > +
> > +       /**
> > +        * [META]
> > +        *
> > +        * Matches conntrack state.
> > +        *
> > +        * See struct rte_flow_item_conntrack.
> > +        */
> > +       RTE_FLOW_ITEM_TYPE_CONNTRACK,
> >  };
> >
> >  /**
> > @@ -1654,6 +1663,49 @@ rte_flow_item_geneve_opt_mask = {
> >  };
> >  #endif
> >
> > +/**
> > + * The packet is with valid.
> > + */
> > +#define RTE_FLOW_CONNTRACK_FLAG_STATE_VALID (1 << 0)
> > +/**
> > + * The state of the connection was changed.
> > + */
> > +#define RTE_FLOW_CONNTRACK_FLAG_STATE_CHANGED (1 << 1)
> > +/**
> > + * Error state was detected on this packet for this connection.
> > + */
> > +#define RTE_FLOW_CONNTRACK_FLAG_ERROR (1 << 2)
> > +/**
> > + * The HW connection tracking module is disabled.
> > + * It can be due to application command or an invalid state.
> > + */
> > +#define RTE_FLOW_CONNTRACK_FLAG_DISABLED (1 << 3)
> > +/**
> > + * The packet contains some bad field(s).
> > + */
> > +#define RTE_FLOW_CONNTRACK_FLAG_BAD_PKT (1 << 4)
> Why not an enum? We could use the bits, but group them under an enum?
> 

It could be. BTW, is there any convention to describe when to use #define macros and when to use enum types?

> > +
> > +/**
> > + * @warning
> > + * @b EXPERIMENTAL: this structure may change without prior
> notice
> > + *
> > + * RTE_FLOW_ITEM_TYPE_CONNTRACK
> > + *
> > + * Matches the state of a packet after it passed the connection
> tracking
> > + * examination. The state is a bit mask of one
> RTE_FLOW_CONNTRACK_FLAG*
> > + * or a reasonable combination of these bits.
> > + */
> > +struct rte_flow_item_conntrack {
> > +       uint32_t flags;
> > +};
> > +
> > +/** Default mask for RTE_FLOW_ITEM_TYPE_CONNTRACK. */
> > +#ifndef __cplusplus
> > +static const struct rte_flow_item_conntrack
> rte_flow_item_conntrack_mask = {
> > +       .flags = 0xffffffff,
> > +};
> > +#endif
> > +
> >  /**
> >   * Matching pattern item definition.
> >   *
> > @@ -2236,6 +2288,17 @@ enum rte_flow_action_type {
> >          * See struct rte_flow_action_modify_field.
> >          */
> >         RTE_FLOW_ACTION_TYPE_MODIFY_FIELD,
> > +
> > +       /**
> > +        * [META]
> > +        *
> > +        * Enable tracking a TCP connection state.
> > +        *
> > +        * Send packet to HW connection tracking module for
> examination.
> > +        *
> > +        * See struct rte_flow_action_conntrack.
> > +        */
> > +       RTE_FLOW_ACTION_TYPE_CONNTRACK,
> >  };
> >
> >  /**
> > @@ -2828,6 +2891,134 @@ struct rte_flow_action_set_dscp {
> >   */
> >  struct rte_flow_shared_action;
> >
> > +/**
> > + * The state of a TCP connection.
> > + */
> > +enum rte_flow_conntrack_state {
> > +       RTE_FLOW_CONNTRACK_STATE_SYN_RECV,
> > +       /**< SYN-ACK packet was seen. */
> > +       RTE_FLOW_CONNTRACK_STATE_ESTABLISHED,
> > +       /**< 3-way handshark was done. */
> > +       RTE_FLOW_CONNTRACK_STATE_FIN_WAIT,
> > +       /**< First FIN packet was received to close the connection.
> */
> > +       RTE_FLOW_CONNTRACK_STATE_CLOSE_WAIT,
> > +       /**< First FIN was ACKed. */
> > +       RTE_FLOW_CONNTRACK_STATE_LAST_ACK,
> > +       /**< After second FIN, waiting for the last ACK. */
> > +       RTE_FLOW_CONNTRACK_STATE_TIME_WAIT,
> > +       /**< Second FIN was ACKed, connection was closed. */
> > +};
> > +
> > +/**
> > + * The last passed TCP packet flags of a connection.
> > + */
> > +enum rte_flow_conntrack_index {
> > +       RTE_FLOW_CONNTRACK_INDEX_NONE = 0, /**< No Flag. */
> > +       RTE_FLOW_CONNTRACK_INDEX_SYN = (1 << 0), /**< With SYN
> flag. */
> > +       RTE_FLOW_CONNTRACK_INDEX_SYN_ACK = (1 << 1), /**< With
> SYN+ACK flag. */
> > +       RTE_FLOW_CONNTRACK_INDEX_FIN = (1 << 2), /**< With FIN
> flag. */
> > +       RTE_FLOW_CONNTRACK_INDEX_ACK = (1 << 3), /**< With ACK
> flag. */
> > +       RTE_FLOW_CONNTRACK_INDEX_RST = (1 << 4), /**< With RST
> flag. */
> > +};
> > +
> > +/**
> > + * @warning
> > + * @b EXPERIMENTAL: this structure may change without prior
> notice
> > + *
> > + * Configuration parameters for each direction of a TCP
> connection.
> > + */
> > +struct rte_flow_tcp_dir_param {
> > +       uint32_t scale:4; /**< TCP window scaling factor, 0xF to
> disable. */
> > +       uint32_t close_initiated:1; /**< The FIN was sent by this
> direction. */
> > +       uint32_t last_ack_seen:1;
> > +       /**< An ACK packet has been received by this side. */
> > +       uint32_t data_unacked:1;
> > +       /**< If set, indicates that there is unacked data of the
> connection. */
> > +       uint32_t sent_end;
> > +       /**< Maximal value of sequence + payload length over sent
> > +        * packets (next ACK from the opposite direction).
> > +        */
> > +       uint32_t reply_end;
> > +       /**< Maximal value of (ACK + window size) over received
> packet + length
> > +        * over sent packet (maximal sequence could be sent).
> > +        */
> > +       uint32_t max_win;
> > +       /**< Maximal value of actual window size over sent packets.
> */
> > +       uint32_t max_ack;
> > +       /**< Maximal value of ACK over sent packets. */
> > +};
> > +
> > +/**
> > + * @warning
> > + * @b EXPERIMENTAL: this structure may change without prior
> notice
> > + *
> > + * RTE_FLOW_ACTION_TYPE_CONNTRACK
> > + *
> > + * Configuration and initial state for the connection tracking
> module.
> > + * This structure could be used for both setting and query.
> Can we split the structure into set and query.
> Some of the fields seem to be relevant for a query.
> Also the names will be simpler and easier to understand that way.
> 

To my understanding, it may be better to have all of them in a single structure. Different HW may have different ability for querying, and most of the fields are used both for query and create/update.
If some field is not supported for querying, the PMD could return the default value, e.g., 0. And if we split query and create/update:
1. Query struct may also needs to be the UNION of different HW's query capacity
2. the 2 structures will have a lot of common fields (most of them)

> > + */
> > +struct rte_flow_action_conntrack {
> > +       uint16_t peer_port; /**< The peer port number, can be the
> same port. */
> > +       uint32_t is_original_dir:1;
> > +       /**< Direction of this connection when creating a flow,
> the value only
> > +        * affects the subsequent flows creation.
> > +        */
> > +       uint32_t enable:1;
> > +       /**< Enable / disable the conntrack HW module. When
> disabled, the
> > +        * result will always be RTE_FLOW_CONNTRACK_FLAG_DISABLED.
> > +        * In this state the HW will act as passthrough.
> > +        */
> We should be able to enable the block in HW implicitly based on the
> rte_flow_create.
> I don't think this is needed.
> 
> > +       uint32_t live_connection:1;
> > +       /**< At least one ack was seen, after the connection was
> established. */
> > +       uint32_t selective_ack:1;
> > +       /**< Enable selective ACK on this connection. */
> > +       uint32_t challenge_ack_passed:1;
> > +       /**< A challenge ack has passed. */
> > +       uint32_t last_direction:1;
> > +       /**< 1: The last packet is seen that comes from the
> original direction.
> > +        * 0: From the reply direction.
> > +        */
> > +       uint32_t liberal_mode:1;
> > +       /**< No TCP check will be done except the state change. */
> > +       enum rte_flow_conntrack_state state;
> initial_state or cur_state?
> 
> > +       /**< The current state of the connection. */
> > +       uint8_t max_ack_window;
> > +       /**< Scaling factor for maximal allowed ACK window. */
> > +       uint8_t retransmission_limit;
> > +       /**< Maximal allowed number of retransmission times. */
> > +       struct rte_flow_tcp_dir_param original_dir;
> > +       /**< TCP parameters of the original direction. */
> > +       struct rte_flow_tcp_dir_param reply_dir;
> > +       /**< TCP parameters of the reply direction. */
> > +       uint16_t last_window;
> > +       /**< The window value of the last packet passed this
> conntrack. */
> > +       enum rte_flow_conntrack_index last_index;
> Do you mean rte_flow_conntrack_last_state - as in last state as seen
> by HW block?
> Or maybe it is the TCP flag and not state?

They are a little different. It should be the 2nd, the TCP flag of the last packets passed the connection tracking module, not the connection state.

> 
> > +       uint32_t last_seq;
> > +       /**< The sequence of the last packet passed this conntrack.
> */
> > +       uint32_t last_ack;
> > +       /**< The acknowledgement of the last packet passed this
> conntrack. */
> > +       uint32_t last_end;
> > +       /**< The total value ACK + payload length of the last
> packet passed
> > +        * this conntrack.
> > +        */
> > +};
> > +
> > +/**
> > + * RTE_FLOW_ACTION_TYPE_CONNTRACK
> > + *
> > + * Wrapper structure for the context update interface.
> > + * Ports cannot support updating, and the only valid solution is
> to
> > + * destroy the old context and create a new one instead.
> > + */
> In that case why not destroy the flow and create a new one?

I may not quite understand your question but will try to answer a bit more detailed and please comment.
The connection tracking action context will be created before any flow creation, then it will be used by the flows:
1. Conntrack action will be used for the flows of bi-directional traffic, and when creating it, the information of TCP packets from both directions are needed.
2. This conntrack action could be used for multiple flows over single port or dual ports.
3. The flow could be destroyed w/o destroy the action and then it could be reused by a new flow(if needed).
4. One direction flow could be destroyed w/o destroying the opposite direction flow.

So if the user want to destroy the action context, it should call the destroy interface directly. The action context will be still "alive" after one flow that using it is destroyed. It couldn't be destroyed together with the flow.

> 
> > +struct rte_flow_modify_conntrack {
> > +       struct rte_flow_action_conntrack new_ct;
> > +       /**< New connection tracking parameters to be updated. */
> > +       uint32_t direction:1; /**< The direction field will be
> updated. */
> > +       uint32_t state:1;
> > +       /**< All the other fields except direction will be updated.
> */
> > +       uint32_t reserved:30; /**< Reserved bits for the future
> usage. */
> > +};
> > +
> >  /**
> >   * Field IDs for MODIFY_FIELD action.
> >   */
> > --
> > 2.19.0.windows.1
> >