1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright(c) 2010-2014 Intel Corporation.
4 IP Reassembly Sample Application
5 ================================
7 The L3 Forwarding application is a simple example of packet processing using the DPDK.
8 The application performs L3 forwarding with reassembly for fragmented IPv4 and IPv6 packets.
13 The application demonstrates the use of the DPDK libraries to implement packet forwarding
14 with reassembly for IPv4 and IPv6 fragmented packets.
15 The initialization and run- time paths are very similar to those of the :doc:`l2_forward_real_virtual`.
16 The main difference from the L2 Forwarding sample application is that
17 it reassembles fragmented IPv4 and IPv6 packets before forwarding.
18 The maximum allowed size of reassembled packet is 9.5 KB.
20 There are two key differences from the L2 Forwarding sample application:
22 * The first difference is that the forwarding decision is taken based on information read from the input packet's IP header.
24 * The second difference is that the application differentiates between IP and non-IP traffic by means of offload flags.
26 The Longest Prefix Match (LPM for IPv4, LPM6 for IPv6) table is used to store/lookup an outgoing port number, associated with that IPv4 address. Any unmatched packets are forwarded to the originating port.Compiling the Application
27 --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
30 Compiling the Application
31 -------------------------
33 To compile the sample application see :doc:`compiling`.
35 The application is located in the ``ip_reassembly`` sub-directory.
38 Running the Application
39 -----------------------
41 The application has a number of command line options:
43 .. code-block:: console
45 ./build/ip_reassembly [EAL options] -- -p PORTMASK [-q NQ] [--maxflows=FLOWS>] [--flowttl=TTL[(s|ms)]]
49 * -p PORTMASK: Hexadecimal bitmask of ports to configure
51 * -q NQ: Number of RX queues per lcore
53 * --maxflows=FLOWS: determines maximum number of active fragmented flows (1-65535). Default value: 4096.
55 * --flowttl=TTL[(s|ms)]: determines maximum Time To Live for fragmented packet.
56 If all fragments of the packet wouldn't appear within given time-out,
57 then they are considered as invalid and will be dropped.
58 Valid range is 1ms - 3600s. Default value: 1s.
60 To run the example in linuxapp environment with 2 lcores (2,4) over 2 ports(0,2) with 1 RX queue per lcore:
62 .. code-block:: console
64 ./build/ip_reassembly -l 2,4 -n 3 -- -p 5
65 EAL: coremask set to 14
66 EAL: Detected lcore 0 on socket 0
67 EAL: Detected lcore 1 on socket 1
68 EAL: Detected lcore 2 on socket 0
69 EAL: Detected lcore 3 on socket 1
70 EAL: Detected lcore 4 on socket 0
73 Initializing port 0 on lcore 2... Address:00:1B:21:76:FA:2C, rxq=0 txq=2,0 txq=4,1
74 done: Link Up - speed 10000 Mbps - full-duplex
75 Skipping disabled port 1
76 Initializing port 2 on lcore 4... Address:00:1B:21:5C:FF:54, rxq=0 txq=2,0 txq=4,1
77 done: Link Up - speed 10000 Mbps - full-duplex
78 Skipping disabled port 3IP_FRAG: Socket 0: adding route 100.10.0.0/16 (port 0)
79 IP_RSMBL: Socket 0: adding route 100.20.0.0/16 (port 1)
82 IP_RSMBL: Socket 0: adding route 0101:0101:0101:0101:0101:0101:0101:0101/48 (port 0)
83 IP_RSMBL: Socket 0: adding route 0201:0101:0101:0101:0101:0101:0101:0101/48 (port 1)
86 IP_RSMBL: entering main loop on lcore 4
87 IP_RSMBL: -- lcoreid=4 portid=2
88 IP_RSMBL: entering main loop on lcore 2
89 IP_RSMBL: -- lcoreid=2 portid=0
91 To run the example in linuxapp environment with 1 lcore (4) over 2 ports(0,2) with 2 RX queues per lcore:
93 .. code-block:: console
95 ./build/ip_reassembly -l 4 -n 3 -- -p 5 -q 2
97 To test the application, flows should be set up in the flow generator that match the values in the
98 l3fwd_ipv4_route_array and/or l3fwd_ipv6_route_array table.
100 Please note that in order to test this application,
101 the traffic generator should be generating valid fragmented IP packets.
102 For IPv6, the only supported case is when no other extension headers other than
103 fragment extension header are present in the packet.
105 The default l3fwd_ipv4_route_array table is:
109 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
110 {IPv4(100, 10, 0, 0), 16, 0},
111 {IPv4(100, 20, 0, 0), 16, 1},
112 {IPv4(100, 30, 0, 0), 16, 2},
113 {IPv4(100, 40, 0, 0), 16, 3},
114 {IPv4(100, 50, 0, 0), 16, 4},
115 {IPv4(100, 60, 0, 0), 16, 5},
116 {IPv4(100, 70, 0, 0), 16, 6},
117 {IPv4(100, 80, 0, 0), 16, 7},
120 The default l3fwd_ipv6_route_array table is:
124 struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
125 {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
126 {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
127 {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
128 {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
129 {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
130 {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
131 {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
132 {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
135 For example, for the fragmented input IPv4 packet with destination address: 100.10.1.1,
136 a reassembled IPv4 packet be sent out from port #0 to the destination address 100.10.1.1
137 once all the fragments are collected.
142 The following sections provide some explanation of the sample application code.
143 As mentioned in the overview section, the initialization and run-time paths are very similar to those of the :doc:`l2_forward_real_virtual`.
144 The following sections describe aspects that are specific to the IP reassemble sample application.
146 IPv4 Fragment Table Initialization
147 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
149 This application uses the rte_ip_frag library. Please refer to Programmer's Guide for more detailed explanation of how to use this library.
150 Fragment table maintains information about already received fragments of the packet.
151 Each IP packet is uniquely identified by triple <Source IP address>, <Destination IP address>, <ID>.
152 To avoid lock contention, each RX queue has its own Fragment Table,
153 e.g. the application can't handle the situation when different fragments of the same packet arrive through different RX queues.
154 Each table entry can hold information about packet consisting of up to RTE_LIBRTE_IP_FRAG_MAX_FRAGS fragments.
158 frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S * max_flow_ttl;
160 if ((qconf->frag_tbl[queue] = rte_ip_frag_tbl_create(max_flow_num, IPV4_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles, socket)) == NULL)
162 RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on " "lcore: %u for queue: %u failed\n", max_flow_num, lcore, queue);
166 Mempools Initialization
167 ~~~~~~~~~~~~~~~~~~~~~~~
169 The reassembly application demands a lot of mbuf's to be allocated.
170 At any given time up to (2 \* max_flow_num \* RTE_LIBRTE_IP_FRAG_MAX_FRAGS \* <maximum number of mbufs per packet>)
171 can be stored inside Fragment Table waiting for remaining fragments.
172 To keep mempool size under reasonable limits and to avoid situation when one RX queue can starve other queues,
173 each RX queue uses its own mempool.
177 nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * RTE_LIBRTE_IP_FRAG_MAX_FRAGS;
178 nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
179 nb_mbuf *= 2; /* ipv4 and ipv6 */
180 nb_mbuf += RTE_TEST_RX_DESC_DEFAULT + RTE_TEST_TX_DESC_DEFAULT;
181 nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
183 snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
185 if ((rxq->pool = rte_mempool_create(buf, nb_mbuf, MBUF_SIZE, 0, sizeof(struct rte_pktmbuf_pool_private), rte_pktmbuf_pool_init, NULL,
186 rte_pktmbuf_init, NULL, socket, MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET)) == NULL) {
188 RTE_LOG(ERR, IP_RSMBL, "mempool_create(%s) failed", buf);
192 Packet Reassembly and Forwarding
193 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
195 For each input packet, the packet forwarding operation is done by the l3fwd_simple_forward() function.
196 If the packet is an IPv4 or IPv6 fragment, then it calls rte_ipv4_reassemble_packet() for IPv4 packets,
197 or rte_ipv6_reassemble_packet() for IPv6 packets.
198 These functions either return a pointer to valid mbuf that contains reassembled packet,
199 or NULL (if the packet can't be reassembled for some reason).
200 Then l3fwd_simple_forward() continues with the code for the packet forwarding decision
201 (that is, the identification of the output interface for the packet) and
202 actual transmit of the packet.
204 The rte_ipv4_reassemble_packet() or rte_ipv6_reassemble_packet() are responsible for:
206 #. Searching the Fragment Table for entry with packet's <IP Source Address, IP Destination Address, Packet ID>
208 #. If the entry is found, then check if that entry already timed-out.
209 If yes, then free all previously received fragments,
210 and remove information about them from the entry.
212 #. If no entry with such key is found, then try to create a new one by one of two ways:
214 #. Use as empty entry
216 #. Delete a timed-out entry, free mbufs associated with it mbufs and store a new entry with specified key in it.
218 #. Update the entry with new fragment information and check
219 if a packet can be reassembled (the packet's entry contains all fragments).
221 #. If yes, then, reassemble the packet, mark table's entry as empty and return the reassembled mbuf to the caller.
223 #. If no, then just return a NULL to the caller.
225 If at any stage of packet processing a reassembly function encounters an error
226 (can't insert new entry into the Fragment table, or invalid/timed-out fragment),
227 then it will free all associated with the packet fragments,
228 mark the table entry as invalid and return NULL to the caller.
230 Debug logging and Statistics Collection
231 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
233 The RTE_LIBRTE_IP_FRAG_TBL_STAT controls statistics collection for the IP Fragment Table.
234 This macro is disabled by default.
235 To make ip_reassembly print the statistics to the standard output,
236 the user must send either an USR1, INT or TERM signal to the process.
237 For all of these signals, the ip_reassembly process prints Fragment table statistics for each RX queue,
238 plus the INT and TERM will cause process termination as usual.