[v3] node: add ipv4 reassembly node

  From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add IPv4 reassembly node.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
Depends-on: series-28278

 v3 Changes:
 - Actually include the changes that fix compilation.
 v2 Changes:
 - Fix compilation.

 doc/guides/prog_guide/graph_lib.rst |   8 ++
 lib/node/ethdev_rx.c                |   1 +
 lib/node/ethdev_rx_priv.h           |   1 +
 lib/node/ip4_reassembly.c           | 175 ++++++++++++++++++++++++++++
 lib/node/ip4_reassembly_priv.h      |  28 +++++
 lib/node/meson.build                |   3 +-
 lib/node/rte_node_ip4_api.h         |  37 ++++++
 lib/node/version.map                |   1 +
 8 files changed, 253 insertions(+), 1 deletion(-)
 create mode 100644 lib/node/ip4_reassembly.c
 create mode 100644 lib/node/ip4_reassembly_priv.h

--
2.39.1
  

On Mon, Jun 5, 2023 at 4:57 PM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Add IPv4 reassembly node.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
> Depends-on: series-28278
>
>  v3 Changes:
>  - Actually include the changes that fix compilation.
>  v2 Changes:
>  - Fix compilation.
>
>  doc/guides/prog_guide/graph_lib.rst |   8 ++
>  lib/node/ethdev_rx.c                |   1 +
>  lib/node/ethdev_rx_priv.h           |   1 +
>  lib/node/ip4_reassembly.c           | 175 ++++++++++++++++++++++++++++
>  lib/node/ip4_reassembly_priv.h      |  28 +++++
>  lib/node/meson.build                |   3 +-
>  lib/node/rte_node_ip4_api.h         |  37 ++++++
>  lib/node/version.map                |   1 +
>  8 files changed, 253 insertions(+), 1 deletion(-)
>  create mode 100644 lib/node/ip4_reassembly.c
>  create mode 100644 lib/node/ip4_reassembly_priv.h
>
> diff --git a/doc/guides/prog_guide/graph_lib.rst b/doc/guides/prog_guide/graph_lib.rst
> index 841cabe259..49c470e287 100644
> --- a/doc/guides/prog_guide/graph_lib.rst
> +++ b/doc/guides/prog_guide/graph_lib.rst
> @@ -388,6 +388,14 @@ to determine the L2 header to be written to the packet before sending
>  the packet out to a particular ethdev_tx node.
>  ``rte_node_ip4_rewrite_add()`` is control path API to add next-hop info.
>
> +ip4_reassembly
> +~~~~~~~~~~~~~~
> +This node is an intermediate node that reassembles ipv4 fragmented packets,
> +non-fragmented packets pass through the node un-effected. The node rewrites
> +it's stream and moves it to the next node.
> +The fragment table and death row table should be setup via the
> +``rte_node_ip4_reassembly_configure`` API.
> +
>  ip6_lookup
>  ~~~~~~~~~~
>  This node is an intermediate node that does LPM lookup for the received
> diff --git a/lib/node/ethdev_rx.c b/lib/node/ethdev_rx.c
> index a19237b42f..6ed2e1834f 100644
> --- a/lib/node/ethdev_rx.c
> +++ b/lib/node/ethdev_rx.c
> @@ -218,6 +218,7 @@ static struct rte_node_register ethdev_rx_node_base = {
>                 /* Default pkt classification node */
>                 [ETHDEV_RX_NEXT_PKT_CLS] = "pkt_cls",
>                 [ETHDEV_RX_NEXT_IP4_LOOKUP] = "ip4_lookup",
> +               [ETHDEV_RX_NEXT_IP4_REASSEMBLY] = "ip4_reassembly",
>         },
>  };
>
> diff --git a/lib/node/ethdev_rx_priv.h b/lib/node/ethdev_rx_priv.h
> index 7f24cf962e..574a76c2a6 100644
> --- a/lib/node/ethdev_rx_priv.h
> +++ b/lib/node/ethdev_rx_priv.h
> @@ -39,6 +39,7 @@ struct ethdev_rx_node_elem {
>  enum ethdev_rx_next_nodes {
>         ETHDEV_RX_NEXT_IP4_LOOKUP,
>         ETHDEV_RX_NEXT_PKT_CLS,
> +       ETHDEV_RX_NEXT_IP4_REASSEMBLY,
>         ETHDEV_RX_NEXT_MAX,
>  };
>
> diff --git a/lib/node/ip4_reassembly.c b/lib/node/ip4_reassembly.c
> new file mode 100644
> index 0000000000..907db111c2
> --- /dev/null
> +++ b/lib/node/ip4_reassembly.c
> @@ -0,0 +1,175 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(C) 2023 Marvell.
> + */
> +
> +#include <arpa/inet.h>
> +#include <stdlib.h>
> +#include <sys/socket.h>
> +
> +#include <rte_cycles.h>
> +#include <rte_debug.h>
> +#include <rte_ethdev.h>
> +#include <rte_ether.h>
> +#include <rte_graph.h>
> +#include <rte_graph_worker.h>
> +#include <rte_ip.h>
> +#include <rte_ip_frag.h>
> +#include <rte_mbuf.h>
> +#include <rte_tcp.h>
> +#include <rte_udp.h>
> +
> +#include "rte_node_ip4_api.h"
> +
> +#include "ip4_reassembly_priv.h"
> +#include "node_private.h"
> +
> +struct ip4_reassembly_elem {
> +       struct ip4_reassembly_elem *next;
> +       struct ip4_reassembly_ctx ctx;
> +       rte_node_t node_id;
> +};
> +
> +/* IP4 reassembly global data struct */
> +struct ip4_reassembly_node_main {
> +       struct ip4_reassembly_elem *head;
> +};
> +
> +typedef struct ip4_reassembly_ctx ip4_reassembly_ctx_t;
> +typedef struct ip4_reassembly_elem ip4_reassembly_elem_t;
> +
> +static struct ip4_reassembly_node_main ip4_reassembly_main;
> +
> +static uint16_t
> +ip4_reassembly_node_process(struct rte_graph *graph, struct rte_node *node, void **objs,
> +                           uint16_t nb_objs)
> +{
> +#define PREFETCH_OFFSET 4
> +       struct rte_mbuf *mbuf, *mbuf_out;
> +       struct rte_ip_frag_death_row *dr;
> +       struct ip4_reassembly_ctx *ctx;
> +       struct rte_ipv4_hdr *ipv4_hdr;
> +       struct rte_ip_frag_tbl *tbl;
> +       uint16_t idx = 0;
> +       void **to_next;
> +       int i;
> +
> +       ctx = (struct ip4_reassembly_ctx *)node->ctx;
> +
> +       /* Get core specific reassembly tbl */
> +       tbl = ctx->tbl;
> +       dr = ctx->dr;
> +
> +       for (i = 0; i < PREFETCH_OFFSET && i < nb_objs; i++) {
> +               rte_prefetch0(rte_pktmbuf_mtod_offset((struct rte_mbuf *)objs[i], void *,
> +                                                     sizeof(struct rte_ether_hdr)));
> +       }
> +
> +       to_next = node->objs;
> +       for (i = 0; i < nb_objs - PREFETCH_OFFSET; i++) {
> +#if RTE_GRAPH_BURST_SIZE > 64
> +               /* Prefetch next-next mbufs */
> +               if (likely(i + 8 < nb_objs))
> +                       rte_prefetch0(objs[i + 8]);
> +#endif
> +               rte_prefetch0(rte_pktmbuf_mtod_offset((struct rte_mbuf *)objs[i + PREFETCH_OFFSET],
> +                                                     void *, sizeof(struct rte_ether_hdr)));
> +               mbuf = (struct rte_mbuf *)objs[i];
> +
> +               ipv4_hdr = rte_pktmbuf_mtod_offset(mbuf, struct rte_ipv4_hdr *,
> +                                                  sizeof(struct rte_ether_hdr));
> +               if (rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr)) {
> +                       /* prepare mbuf: setup l2_len/l3_len. */
> +                       mbuf->l2_len = sizeof(struct rte_ether_hdr);
> +                       mbuf->l3_len = sizeof(struct rte_ipv4_hdr);
> +
> +                       mbuf_out = rte_ipv4_frag_reassemble_packet(tbl, dr, mbuf, rte_rdtsc(),
> +                                                                  ipv4_hdr);
> +               } else {
> +                       mbuf_out = mbuf;
> +               }
> +
> +               if (mbuf_out)
> +                       to_next[idx++] = (void *)mbuf_out;
> +       }
> +
> +       for (; i < nb_objs; i++) {
> +               mbuf = (struct rte_mbuf *)objs[i];
> +
> +               ipv4_hdr = rte_pktmbuf_mtod_offset(mbuf, struct rte_ipv4_hdr *,
> +                                                  sizeof(struct rte_ether_hdr));
> +               if (rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr)) {
> +                       /* prepare mbuf: setup l2_len/l3_len. */
> +                       mbuf->l2_len = sizeof(struct rte_ether_hdr);
> +                       mbuf->l3_len = sizeof(struct rte_ipv4_hdr);
> +
> +                       mbuf_out = rte_ipv4_frag_reassemble_packet(tbl, dr, mbuf, rte_rdtsc(),
> +                                                                  ipv4_hdr);
> +               } else {
> +                       mbuf_out = mbuf;
> +               }
> +
> +               if (mbuf_out)
> +                       to_next[idx++] = (void *)mbuf_out;
> +       }
> +       node->idx = idx;
> +       rte_node_next_stream_move(graph, node, 0);
> +       rte_ip_frag_free_death_row(dr, 4);

Can these mbufs that are being freed sent to drop node so that it is
tracked and accounted as dropped ?

> +
> +       return idx;
> +}
> +
> +int
> +rte_node_ip4_reassembly_configure(struct rte_node_ip4_reassembly_cfg *cfg, uint16_t cnt)
> +{
> +       ip4_reassembly_elem_t *elem;
> +       int i;
> +
> +       for (i = 0; i < cnt; i++) {
> +               elem = malloc(sizeof(ip4_reassembly_elem_t));
> +               if (elem == NULL)
> +                       return -ENOMEM;
> +               elem->ctx.dr = cfg[i].dr;
> +               elem->ctx.tbl = cfg[i].tbl;
> +               elem->node_id = cfg[i].node_id;
> +               elem->next = ip4_reassembly_main.head;
> +               ip4_reassembly_main.head = elem;
> +       }
> +
> +       return 0;
> +}
> +
> +static int
> +ip4_reassembly_node_init(const struct rte_graph *graph, struct rte_node *node)
> +{
> +       ip4_reassembly_ctx_t *ctx = (ip4_reassembly_ctx_t *)node->ctx;
> +       ip4_reassembly_elem_t *elem = ip4_reassembly_main.head;
> +
> +       RTE_SET_USED(graph);
> +       while (elem) {
> +               if (elem->node_id == node->id) {
> +                       /* Update node specific context */
> +                       memcpy(ctx, &elem->ctx, sizeof(ip4_reassembly_ctx_t));
> +                       break;
> +               }
> +               elem = elem->next;
> +       }
> +
> +       return 0;
> +}
> +
> +static struct rte_node_register ip4_reassembly_node = {
> +       .process = ip4_reassembly_node_process,
> +       .name = "ip4_reassembly",
> +
> +       .init = ip4_reassembly_node_init,
> +
> +       .nb_edges = RTE_NODE_IP4_REASSEMBLY_NEXT_MAX,
> +};
> +
> +struct rte_node_register *
> +ip4_reassembly_node_get(void)
> +{
> +       return &ip4_reassembly_node;
> +}
> +
> +RTE_NODE_REGISTER(ip4_reassembly_node);
> diff --git a/lib/node/ip4_reassembly_priv.h b/lib/node/ip4_reassembly_priv.h
> new file mode 100644
> index 0000000000..1fa70274ba
> --- /dev/null
> +++ b/lib/node/ip4_reassembly_priv.h
> @@ -0,0 +1,28 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(C) 2023 Marvell International Ltd.
> + */
> +
> +#ifndef __INCLUDE_IP4_REASSEMBLY_PRIV_H__
> +#define __INCLUDE_IP4_REASSEMBLY_PRIV_H__
> +
> +/**
> + * @internal
> + *
> + * Ip4_reassembly context structure.
> + */
> +struct ip4_reassembly_ctx {
> +       struct rte_ip_frag_tbl *tbl;
> +       struct rte_ip_frag_death_row *dr;
> +};
> +
> +/**
> + * @internal
> + *
> + * Get the IP4 reassembly node
> + *
> + * @return
> + *   Pointer to the IP4 reassembly node.
> + */
> +struct rte_node_register *ip4_reassembly_node_get(void);
> +
> +#endif /* __INCLUDE_IP4_REASSEMBLY_PRIV_H__ */
> diff --git a/lib/node/meson.build b/lib/node/meson.build
> index b2f04269c5..94d3f36163 100644
> --- a/lib/node/meson.build
> +++ b/lib/node/meson.build
> @@ -19,8 +19,9 @@ sources = files(
>          'null.c',
>          'pkt_cls.c',
>          'pkt_drop.c',
> +        'ip4_reassembly.c'
>  )
>  headers = files('rte_node_ip4_api.h', 'rte_node_ip6_api.h', 'rte_node_eth_api.h')
>  # Strict-aliasing rules are violated by uint8_t[] to context size casts.
>  cflags += '-fno-strict-aliasing'
> -deps += ['graph', 'mbuf', 'lpm', 'ethdev', 'mempool', 'cryptodev']
> +deps += ['graph', 'mbuf', 'lpm', 'ethdev', 'mempool', 'cryptodev', 'ip_frag']
> diff --git a/lib/node/rte_node_ip4_api.h b/lib/node/rte_node_ip4_api.h
> index 46d0d8976b..cec5c4e309 100644
> --- a/lib/node/rte_node_ip4_api.h
> +++ b/lib/node/rte_node_ip4_api.h
> @@ -23,6 +23,8 @@ extern "C" {
>  #include <rte_common.h>
>  #include <rte_compat.h>
>
> +#include <rte_graph.h>
> +
>  /**
>   * IP4 lookup next nodes.
>   */
> @@ -35,6 +37,27 @@ enum rte_node_ip4_lookup_next {
>         /**< Number of next nodes of lookup node. */
>  };
>
> +/**
> + * IP4 reassembly next nodes.
> + */
> +enum rte_node_ip4_reassembly_next {
> +       RTE_NODE_IP4_REASSEMBLY_NEXT_MAX,
> +       /**< Number of next nodes of reassembly node. */
> +};
> +
> +/**
> + * Reassembly configure structure.
> + * @see rte_node_ip4_reassembly_configure
> + */
> +struct rte_node_ip4_reassembly_cfg {
> +       struct rte_ip_frag_tbl *tbl;
> +       /**< Reassembly fragmentation table. */
> +       struct rte_ip_frag_death_row *dr;
> +       /**< Reassembly deathrow table. */
> +       rte_node_t node_id;
> +       /**< Node identifier to configure. */
> +};
> +
>  /**
>   * Add ipv4 route to lookup table.
>   *
> @@ -73,6 +96,20 @@ __rte_experimental
>  int rte_node_ip4_rewrite_add(uint16_t next_hop, uint8_t *rewrite_data,
>                              uint8_t rewrite_len, uint16_t dst_port);
>
> +/**
> + * Add reassembly node configuration data.
> + *
> + * @param cfg
> + *   Pointer to the configuration structure.
> + * @param cnt
> + *   Number of configuration structures passed.
> + *
> + * @return
> + *   0 on success, negative otherwise.
> + */
> +__rte_experimental
> +int rte_node_ip4_reassembly_configure(struct rte_node_ip4_reassembly_cfg *cfg, uint16_t cnt);
> +
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/node/version.map b/lib/node/version.map
> index 40df308bfe..c10e2e78ea 100644
> --- a/lib/node/version.map
> +++ b/lib/node/version.map
> @@ -4,6 +4,7 @@ EXPERIMENTAL {
>         rte_node_eth_config;
>         rte_node_ip4_route_add;
>         rte_node_ip4_rewrite_add;
> +       rte_node_ip4_reassembly_configure;
>         rte_node_ip6_rewrite_add;
>         rte_node_ip6_route_add;
>         rte_node_logtype;
> --
> 2.39.1
>