l3_forward.rst revision 8b25d1ad
1..  BSD LICENSE
2    Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
3    All rights reserved.
4
5    Redistribution and use in source and binary forms, with or without
6    modification, are permitted provided that the following conditions
7    are met:
8
9    * Redistributions of source code must retain the above copyright
10    notice, this list of conditions and the following disclaimer.
11    * Redistributions in binary form must reproduce the above copyright
12    notice, this list of conditions and the following disclaimer in
13    the documentation and/or other materials provided with the
14    distribution.
15    * Neither the name of Intel Corporation nor the names of its
16    contributors may be used to endorse or promote products derived
17    from this software without specific prior written permission.
18
19    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
22    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
23    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
25    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
29    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30
31L3 Forwarding Sample Application
32================================
33
34The L3 Forwarding application is a simple example of packet processing using the DPDK.
35The application performs L3 forwarding.
36
37Overview
38--------
39
40The application demonstrates the use of the hash and LPM libraries in the DPDK to implement packet forwarding.
41The initialization and run-time paths are very similar to those of the :doc:`l2_forward_real_virtual`.
42The main difference from the L2 Forwarding sample application is that the forwarding decision
43is made based on information read from the input packet.
44
45The lookup method is either hash-based or LPM-based and is selected at compile time. When the selected lookup method is hash-based,
46a hash object is used to emulate the flow classification stage.
47The hash object is used in correlation with a flow table to map each input packet to its flow at runtime.
48
49The hash lookup key is represented by a DiffServ 5-tuple composed of the following fields read from the input packet:
50Source IP Address, Destination IP Address, Protocol, Source Port and Destination Port.
51The ID of the output interface for the input packet is read from the identified flow table entry.
52The set of flows used by the application is statically configured and loaded into the hash at initialization time.
53When the selected lookup method is LPM based, an LPM object is used to emulate the forwarding stage for IPv4 packets.
54The LPM object is used as the routing table to identify the next hop for each input packet at runtime.
55
56The LPM lookup key is represented by the Destination IP Address field read from the input packet.
57The ID of the output interface for the input packet is the next hop returned by the LPM lookup.
58The set of LPM rules used by the application is statically configured and loaded into the LPM object at initialization time.
59
60In the sample application, hash-based forwarding supports IPv4 and IPv6. LPM-based forwarding supports IPv4 only.
61
62Compiling the Application
63-------------------------
64
65To compile the application:
66
67#.  Go to the sample application directory:
68
69    .. code-block:: console
70
71        export RTE_SDK=/path/to/rte_sdk
72        cd ${RTE_SDK}/examples/l3fwd
73
74#.  Set the target (a default target is used if not specified). For example:
75
76    .. code-block:: console
77
78        export RTE_TARGET=x86_64-native-linuxapp-gcc
79
80    See the *DPDK Getting Started Guide* for possible RTE_TARGET values.
81
82#.  Build the application:
83
84    .. code-block:: console
85
86        make
87
88Running the Application
89-----------------------
90
91The application has a number of command line options:
92
93.. code-block:: console
94
95    ./build/l3fwd [EAL options] -- -p PORTMASK [-P]  --config(port,queue,lcore)[,(port,queue,lcore)] [--enable-jumbo [--max-pkt-len PKTLEN]]  [--no-numa][--hash-entry-num][--ipv6] [--parse-ptype]
96
97where,
98
99*   -p PORTMASK: Hexadecimal bitmask of ports to configure
100
101*   -P: optional, sets all ports to promiscuous mode so that packets are accepted regardless of the packet's Ethernet MAC destination address.
102    Without this option, only packets with the Ethernet MAC destination address set to the Ethernet address of the port are accepted.
103
104*   --config (port,queue,lcore)[,(port,queue,lcore)]: determines which queues from which ports are mapped to which cores
105
106*   --enable-jumbo: optional, enables jumbo frames
107
108*   --max-pkt-len: optional, maximum packet length in decimal (64-9600)
109
110*   --no-numa: optional, disables numa awareness
111
112*   --hash-entry-num: optional, specifies the hash entry number in hexadecimal to be setup
113
114*   --ipv6: optional, set it if running ipv6 packets
115
116*   --parse-ptype: optional, set it if use software way to analyze packet type
117
118For example, consider a dual processor socket platform where cores 0-7 and 16-23 appear on socket 0, while cores 8-15 and 24-31 appear on socket 1.
119Let's say that the programmer wants to use memory from both NUMA nodes, the platform has only two ports, one connected to each NUMA node,
120and the programmer wants to use two cores from each processor socket to do the packet processing.
121
122To enable L3 forwarding between two ports, using two cores, cores 1 and 2, from each processor,
123while also taking advantage of local memory access by optimizing around NUMA, the programmer must enable two queues from each port,
124pin to the appropriate cores and allocate memory from the appropriate NUMA node. This is achieved using the following command:
125
126.. code-block:: console
127
128    ./build/l3fwd -c 606 -n 4 -- -p 0x3 --config="(0,0,1),(0,1,2),(1,0,9),(1,1,10)"
129
130In this command:
131
132*   The -c option enables cores 0, 1, 2, 3
133
134*   The -p option enables ports 0 and 1
135
136*   The --config option enables two queues on each port and maps each (port,queue) pair to a specific core.
137    Logic to enable multiple RX queues using RSS and to allocate memory from the correct NUMA nodes
138    is included in the application and is done transparently.
139    The following table shows the mapping in this example:
140
141+----------+-----------+-----------+-------------------------------------+
142| **Port** | **Queue** | **lcore** | **Description**                     |
143|          |           |           |                                     |
144+----------+-----------+-----------+-------------------------------------+
145| 0        | 0         | 0         | Map queue 0 from port 0 to lcore 0. |
146|          |           |           |                                     |
147+----------+-----------+-----------+-------------------------------------+
148| 0        | 1         | 2         | Map queue 1 from port 0 to lcore 2. |
149|          |           |           |                                     |
150+----------+-----------+-----------+-------------------------------------+
151| 1        | 0         | 1         | Map queue 0 from port 1 to lcore 1. |
152|          |           |           |                                     |
153+----------+-----------+-----------+-------------------------------------+
154| 1        | 1         | 3         | Map queue 1 from port 1 to lcore 3. |
155|          |           |           |                                     |
156+----------+-----------+-----------+-------------------------------------+
157
158Refer to the *DPDK Getting Started Guide* for general information on running applications and
159the Environment Abstraction Layer (EAL) options.
160
161.. _l3_fwd_explanation:
162
163Explanation
164-----------
165
166The following sections provide some explanation of the sample application code. As mentioned in the overview section,
167the initialization and run-time paths are very similar to those of the :doc:`l2_forward_real_virtual`.
168The following sections describe aspects that are specific to the L3 Forwarding sample application.
169
170Hash Initialization
171~~~~~~~~~~~~~~~~~~~
172
173The hash object is created and loaded with the pre-configured entries read from a global array,
174and then generate the expected 5-tuple as key to keep consistence with those of real flow
175for the convenience to execute hash performance test on 4M/8M/16M flows.
176
177.. note::
178
179    The Hash initialization will setup both ipv4 and ipv6 hash table,
180    and populate the either table depending on the value of variable ipv6.
181    To support the hash performance test with up to 8M single direction flows/16M bi-direction flows,
182    populate_ipv4_many_flow_into_table() function will populate the hash table with specified hash table entry number(default 4M).
183
184.. note::
185
186    Value of global variable ipv6 can be specified with --ipv6 in the command line.
187    Value of global variable hash_entry_number,
188    which is used to specify the total hash entry number for all used ports in hash performance test,
189    can be specified with --hash-entry-num VALUE in command line, being its default value 4.
190
191.. code-block:: c
192
193    #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
194
195        static void
196        setup_hash(int socketid)
197        {
198            // ...
199
200            if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
201                if (ipv6 == 0) {
202                    /* populate the ipv4 hash */
203                    populate_ipv4_many_flow_into_table(ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
204                } else {
205                    /* populate the ipv6 hash */
206                    populate_ipv6_many_flow_into_table( ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
207                }
208            } else
209                if (ipv6 == 0) {
210                    /* populate the ipv4 hash */
211                    populate_ipv4_few_flow_into_table(ipv4_l3fwd_lookup_struct[socketid]);
212                } else {
213                    /* populate the ipv6 hash */
214                    populate_ipv6_few_flow_into_table(ipv6_l3fwd_lookup_struct[socketid]);
215                }
216            }
217        }
218    #endif
219
220LPM Initialization
221~~~~~~~~~~~~~~~~~~
222
223The LPM object is created and loaded with the pre-configured entries read from a global array.
224
225.. code-block:: c
226
227    #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
228
229    static void
230    setup_lpm(int socketid)
231    {
232        unsigned i;
233        int ret;
234        char s[64];
235
236        /* create the LPM table */
237
238        snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
239
240        ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid, IPV4_L3FWD_LPM_MAX_RULES, 0);
241
242        if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
243            rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
244                " on socket %d\n", socketid);
245
246        /* populate the LPM table */
247
248        for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
249            /* skip unused ports */
250
251            if ((1 << ipv4_l3fwd_route_array[i].if_out & enabled_port_mask) == 0)
252                continue;
253
254            ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid], ipv4_l3fwd_route_array[i].ip,
255           	                    ipv4_l3fwd_route_array[i].depth, ipv4_l3fwd_route_array[i].if_out);
256
257            if (ret < 0) {
258                rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
259                        "l3fwd LPM table on socket %d\n", i, socketid);
260            }
261
262            printf("LPM: Adding route 0x%08x / %d (%d)\n",
263                (unsigned)ipv4_l3fwd_route_array[i].ip, ipv4_l3fwd_route_array[i].depth, ipv4_l3fwd_route_array[i].if_out);
264        }
265    }
266    #endif
267
268Packet Forwarding for Hash-based Lookups
269~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
270
271For each input packet, the packet forwarding operation is done by the l3fwd_simple_forward()
272or simple_ipv4_fwd_4pkts() function for IPv4 packets or the simple_ipv6_fwd_4pkts() function for IPv6 packets.
273The l3fwd_simple_forward() function provides the basic functionality for both IPv4 and IPv6 packet forwarding
274for any number of burst packets received,
275and the packet forwarding decision (that is, the identification of the output interface for the packet)
276for hash-based lookups is done by the  get_ipv4_dst_port() or get_ipv6_dst_port() function.
277The get_ipv4_dst_port() function is shown below:
278
279.. code-block:: c
280
281    static inline uint8_t
282    get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t *ipv4_l3fwd_lookup_struct)
283    {
284        int ret = 0;
285        union ipv4_5tuple_host key;
286
287        ipv4_hdr = (uint8_t \*)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
288
289        m128i data = _mm_loadu_si128(( m128i*)(ipv4_hdr));
290
291        /* Get 5 tuple: dst port, src port, dst IP address, src IP address and protocol */
292
293        key.xmm = _mm_and_si128(data, mask0);
294
295        /* Find destination port */
296
297        ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
298
299        return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
300    }
301
302The get_ipv6_dst_port() function is similar to the get_ipv4_dst_port() function.
303
304The simple_ipv4_fwd_4pkts() and simple_ipv6_fwd_4pkts() function are optimized for continuous 4 valid ipv4 and ipv6 packets,
305they leverage the multiple buffer optimization to boost the performance of forwarding packets with the exact match on hash table.
306The key code snippet of simple_ipv4_fwd_4pkts() is shown below:
307
308.. code-block:: c
309
310    static inline void
311    simple_ipv4_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
312    {
313        // ...
314
315        data[0] = _mm_loadu_si128(( m128i*)(rte_pktmbuf_mtod(m[0], unsigned char *) + sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
316        data[1] = _mm_loadu_si128(( m128i*)(rte_pktmbuf_mtod(m[1], unsigned char *) + sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
317        data[2] = _mm_loadu_si128(( m128i*)(rte_pktmbuf_mtod(m[2], unsigned char *) + sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
318        data[3] = _mm_loadu_si128(( m128i*)(rte_pktmbuf_mtod(m[3], unsigned char *) + sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
319
320        key[0].xmm = _mm_and_si128(data[0], mask0);
321        key[1].xmm = _mm_and_si128(data[1], mask0);
322        key[2].xmm = _mm_and_si128(data[2], mask0);
323        key[3].xmm = _mm_and_si128(data[3], mask0);
324
325        const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
326
327        rte_hash_lookup_bulk(qconf->ipv4_lookup_struct, &key_array[0], 4, ret);
328
329        dst_port[0] = (ret[0] < 0)? portid:ipv4_l3fwd_out_if[ret[0]];
330        dst_port[1] = (ret[1] < 0)? portid:ipv4_l3fwd_out_if[ret[1]];
331        dst_port[2] = (ret[2] < 0)? portid:ipv4_l3fwd_out_if[ret[2]];
332        dst_port[3] = (ret[3] < 0)? portid:ipv4_l3fwd_out_if[ret[3]];
333
334        // ...
335    }
336
337The simple_ipv6_fwd_4pkts() function is similar to the simple_ipv4_fwd_4pkts() function.
338
339Known issue: IP packets with extensions or IP packets which are not TCP/UDP cannot work well at this mode.
340
341Packet Forwarding for LPM-based Lookups
342~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
343
344For each input packet, the packet forwarding operation is done by the l3fwd_simple_forward() function,
345but the packet forwarding decision (that is, the identification of the output interface for the packet)
346for LPM-based lookups is done by the get_ipv4_dst_port() function below:
347
348.. code-block:: c
349
350    static inline uint8_t
351    get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid, lookup_struct_t *ipv4_l3fwd_lookup_struct)
352    {
353        uint8_t next_hop;
354
355        return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct, rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)? next_hop : portid);
356    }
357