xref: /dpdk/lib/gro/gro_tcp4.c (revision 09442498ef736d0a96632cf8b8c15d8ca78a6468)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2017 Intel Corporation
3  */
4 
5 #include <rte_malloc.h>
6 #include <rte_mbuf.h>
7 #include <rte_ethdev.h>
8 
9 #include "gro_tcp4.h"
10 
11 void *
12 gro_tcp4_tbl_create(uint16_t socket_id,
13 		uint16_t max_flow_num,
14 		uint16_t max_item_per_flow)
15 {
16 	struct gro_tcp4_tbl *tbl;
17 	size_t size;
18 	uint32_t entries_num, i;
19 
20 	entries_num = max_flow_num * max_item_per_flow;
21 	entries_num = RTE_MIN(entries_num, GRO_TCP4_TBL_MAX_ITEM_NUM);
22 
23 	if (entries_num == 0)
24 		return NULL;
25 
26 	tbl = rte_zmalloc_socket(__func__,
27 			sizeof(struct gro_tcp4_tbl),
28 			RTE_CACHE_LINE_SIZE,
29 			socket_id);
30 	if (tbl == NULL)
31 		return NULL;
32 
33 	size = sizeof(struct gro_tcp4_item) * entries_num;
34 	tbl->items = rte_zmalloc_socket(__func__,
35 			size,
36 			RTE_CACHE_LINE_SIZE,
37 			socket_id);
38 	if (tbl->items == NULL) {
39 		rte_free(tbl);
40 		return NULL;
41 	}
42 	tbl->max_item_num = entries_num;
43 
44 	size = sizeof(struct gro_tcp4_flow) * entries_num;
45 	tbl->flows = rte_zmalloc_socket(__func__,
46 			size,
47 			RTE_CACHE_LINE_SIZE,
48 			socket_id);
49 	if (tbl->flows == NULL) {
50 		rte_free(tbl->items);
51 		rte_free(tbl);
52 		return NULL;
53 	}
54 	/* INVALID_ARRAY_INDEX indicates an empty flow */
55 	for (i = 0; i < entries_num; i++)
56 		tbl->flows[i].start_index = INVALID_ARRAY_INDEX;
57 	tbl->max_flow_num = entries_num;
58 
59 	return tbl;
60 }
61 
62 void
63 gro_tcp4_tbl_destroy(void *tbl)
64 {
65 	struct gro_tcp4_tbl *tcp_tbl = tbl;
66 
67 	if (tcp_tbl) {
68 		rte_free(tcp_tbl->items);
69 		rte_free(tcp_tbl->flows);
70 	}
71 	rte_free(tcp_tbl);
72 }
73 
74 static inline uint32_t
75 find_an_empty_item(struct gro_tcp4_tbl *tbl)
76 {
77 	uint32_t i;
78 	uint32_t max_item_num = tbl->max_item_num;
79 
80 	for (i = 0; i < max_item_num; i++)
81 		if (tbl->items[i].firstseg == NULL)
82 			return i;
83 	return INVALID_ARRAY_INDEX;
84 }
85 
86 static inline uint32_t
87 find_an_empty_flow(struct gro_tcp4_tbl *tbl)
88 {
89 	uint32_t i;
90 	uint32_t max_flow_num = tbl->max_flow_num;
91 
92 	for (i = 0; i < max_flow_num; i++)
93 		if (tbl->flows[i].start_index == INVALID_ARRAY_INDEX)
94 			return i;
95 	return INVALID_ARRAY_INDEX;
96 }
97 
98 static inline uint32_t
99 insert_new_item(struct gro_tcp4_tbl *tbl,
100 		struct rte_mbuf *pkt,
101 		uint64_t start_time,
102 		uint32_t prev_idx,
103 		uint32_t sent_seq,
104 		uint16_t ip_id,
105 		uint8_t is_atomic)
106 {
107 	uint32_t item_idx;
108 
109 	item_idx = find_an_empty_item(tbl);
110 	if (item_idx == INVALID_ARRAY_INDEX)
111 		return INVALID_ARRAY_INDEX;
112 
113 	tbl->items[item_idx].firstseg = pkt;
114 	tbl->items[item_idx].lastseg = rte_pktmbuf_lastseg(pkt);
115 	tbl->items[item_idx].start_time = start_time;
116 	tbl->items[item_idx].next_pkt_idx = INVALID_ARRAY_INDEX;
117 	tbl->items[item_idx].sent_seq = sent_seq;
118 	tbl->items[item_idx].ip_id = ip_id;
119 	tbl->items[item_idx].nb_merged = 1;
120 	tbl->items[item_idx].is_atomic = is_atomic;
121 	tbl->item_num++;
122 
123 	/* if the previous packet exists, chain them together. */
124 	if (prev_idx != INVALID_ARRAY_INDEX) {
125 		tbl->items[item_idx].next_pkt_idx =
126 			tbl->items[prev_idx].next_pkt_idx;
127 		tbl->items[prev_idx].next_pkt_idx = item_idx;
128 	}
129 
130 	return item_idx;
131 }
132 
133 static inline uint32_t
134 delete_item(struct gro_tcp4_tbl *tbl, uint32_t item_idx,
135 		uint32_t prev_item_idx)
136 {
137 	uint32_t next_idx = tbl->items[item_idx].next_pkt_idx;
138 
139 	/* NULL indicates an empty item */
140 	tbl->items[item_idx].firstseg = NULL;
141 	tbl->item_num--;
142 	if (prev_item_idx != INVALID_ARRAY_INDEX)
143 		tbl->items[prev_item_idx].next_pkt_idx = next_idx;
144 
145 	return next_idx;
146 }
147 
148 static inline uint32_t
149 insert_new_flow(struct gro_tcp4_tbl *tbl,
150 		struct tcp4_flow_key *src,
151 		uint32_t item_idx)
152 {
153 	struct tcp4_flow_key *dst;
154 	uint32_t flow_idx;
155 
156 	flow_idx = find_an_empty_flow(tbl);
157 	if (unlikely(flow_idx == INVALID_ARRAY_INDEX))
158 		return INVALID_ARRAY_INDEX;
159 
160 	dst = &(tbl->flows[flow_idx].key);
161 
162 	rte_ether_addr_copy(&(src->eth_saddr), &(dst->eth_saddr));
163 	rte_ether_addr_copy(&(src->eth_daddr), &(dst->eth_daddr));
164 	dst->ip_src_addr = src->ip_src_addr;
165 	dst->ip_dst_addr = src->ip_dst_addr;
166 	dst->recv_ack = src->recv_ack;
167 	dst->src_port = src->src_port;
168 	dst->dst_port = src->dst_port;
169 
170 	tbl->flows[flow_idx].start_index = item_idx;
171 	tbl->flow_num++;
172 
173 	return flow_idx;
174 }
175 
176 /*
177  * update the packet length for the flushed packet.
178  */
179 static inline void
180 update_header(struct gro_tcp4_item *item)
181 {
182 	struct rte_ipv4_hdr *ipv4_hdr;
183 	struct rte_mbuf *pkt = item->firstseg;
184 
185 	ipv4_hdr = (struct rte_ipv4_hdr *)(rte_pktmbuf_mtod(pkt, char *) +
186 			pkt->l2_len);
187 	ipv4_hdr->total_length = rte_cpu_to_be_16(pkt->pkt_len -
188 			pkt->l2_len);
189 }
190 
191 int32_t
192 gro_tcp4_reassemble(struct rte_mbuf *pkt,
193 		struct gro_tcp4_tbl *tbl,
194 		uint64_t start_time)
195 {
196 	struct rte_ether_hdr *eth_hdr;
197 	struct rte_ipv4_hdr *ipv4_hdr;
198 	struct rte_tcp_hdr *tcp_hdr;
199 	uint32_t sent_seq;
200 	int32_t tcp_dl;
201 	uint16_t ip_id, hdr_len, frag_off, ip_tlen;
202 	uint8_t is_atomic;
203 
204 	struct tcp4_flow_key key;
205 	uint32_t cur_idx, prev_idx, item_idx;
206 	uint32_t i, max_flow_num, remaining_flow_num;
207 	int cmp;
208 	uint8_t find;
209 
210 	/*
211 	 * Don't process the packet whose TCP header length is greater
212 	 * than 60 bytes or less than 20 bytes.
213 	 */
214 	if (unlikely(INVALID_TCP_HDRLEN(pkt->l4_len)))
215 		return -1;
216 
217 	eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
218 	ipv4_hdr = (struct rte_ipv4_hdr *)((char *)eth_hdr + pkt->l2_len);
219 	tcp_hdr = (struct rte_tcp_hdr *)((char *)ipv4_hdr + pkt->l3_len);
220 	hdr_len = pkt->l2_len + pkt->l3_len + pkt->l4_len;
221 
222 	/*
223 	 * Don't process the packet which has FIN, SYN, RST, PSH, URG, ECE
224 	 * or CWR set.
225 	 */
226 	if (tcp_hdr->tcp_flags != RTE_TCP_ACK_FLAG)
227 		return -1;
228 
229 	/* trim the tail padding bytes */
230 	ip_tlen = rte_be_to_cpu_16(ipv4_hdr->total_length);
231 	if (pkt->pkt_len > (uint32_t)(ip_tlen + pkt->l2_len))
232 		rte_pktmbuf_trim(pkt, pkt->pkt_len - ip_tlen - pkt->l2_len);
233 
234 	/*
235 	 * Don't process the packet whose payload length is less than or
236 	 * equal to 0.
237 	 */
238 	tcp_dl = pkt->pkt_len - hdr_len;
239 	if (tcp_dl <= 0)
240 		return -1;
241 
242 	/*
243 	 * Save IPv4 ID for the packet whose DF bit is 0. For the packet
244 	 * whose DF bit is 1, IPv4 ID is ignored.
245 	 */
246 	frag_off = rte_be_to_cpu_16(ipv4_hdr->fragment_offset);
247 	is_atomic = (frag_off & RTE_IPV4_HDR_DF_FLAG) == RTE_IPV4_HDR_DF_FLAG;
248 	ip_id = is_atomic ? 0 : rte_be_to_cpu_16(ipv4_hdr->packet_id);
249 	sent_seq = rte_be_to_cpu_32(tcp_hdr->sent_seq);
250 
251 	rte_ether_addr_copy(&(eth_hdr->src_addr), &(key.eth_saddr));
252 	rte_ether_addr_copy(&(eth_hdr->dst_addr), &(key.eth_daddr));
253 	key.ip_src_addr = ipv4_hdr->src_addr;
254 	key.ip_dst_addr = ipv4_hdr->dst_addr;
255 	key.src_port = tcp_hdr->src_port;
256 	key.dst_port = tcp_hdr->dst_port;
257 	key.recv_ack = tcp_hdr->recv_ack;
258 
259 	/* Search for a matched flow. */
260 	max_flow_num = tbl->max_flow_num;
261 	remaining_flow_num = tbl->flow_num;
262 	find = 0;
263 	for (i = 0; i < max_flow_num && remaining_flow_num; i++) {
264 		if (tbl->flows[i].start_index != INVALID_ARRAY_INDEX) {
265 			if (is_same_tcp4_flow(tbl->flows[i].key, key)) {
266 				find = 1;
267 				break;
268 			}
269 			remaining_flow_num--;
270 		}
271 	}
272 
273 	/*
274 	 * Fail to find a matched flow. Insert a new flow and store the
275 	 * packet into the flow.
276 	 */
277 	if (find == 0) {
278 		item_idx = insert_new_item(tbl, pkt, start_time,
279 				INVALID_ARRAY_INDEX, sent_seq, ip_id,
280 				is_atomic);
281 		if (item_idx == INVALID_ARRAY_INDEX)
282 			return -1;
283 		if (insert_new_flow(tbl, &key, item_idx) ==
284 				INVALID_ARRAY_INDEX) {
285 			/*
286 			 * Fail to insert a new flow, so delete the
287 			 * stored packet.
288 			 */
289 			delete_item(tbl, item_idx, INVALID_ARRAY_INDEX);
290 			return -1;
291 		}
292 		return 0;
293 	}
294 
295 	/*
296 	 * Check all packets in the flow and try to find a neighbor for
297 	 * the input packet.
298 	 */
299 	cur_idx = tbl->flows[i].start_index;
300 	prev_idx = cur_idx;
301 	do {
302 		cmp = check_seq_option(&(tbl->items[cur_idx]), tcp_hdr,
303 				sent_seq, ip_id, pkt->l4_len, tcp_dl, 0,
304 				is_atomic);
305 		if (cmp) {
306 			if (merge_two_tcp4_packets(&(tbl->items[cur_idx]),
307 						pkt, cmp, sent_seq, ip_id, 0))
308 				return 1;
309 			/*
310 			 * Fail to merge the two packets, as the packet
311 			 * length is greater than the max value. Store
312 			 * the packet into the flow.
313 			 */
314 			if (insert_new_item(tbl, pkt, start_time, cur_idx,
315 						sent_seq, ip_id, is_atomic) ==
316 					INVALID_ARRAY_INDEX)
317 				return -1;
318 			return 0;
319 		}
320 		prev_idx = cur_idx;
321 		cur_idx = tbl->items[cur_idx].next_pkt_idx;
322 	} while (cur_idx != INVALID_ARRAY_INDEX);
323 
324 	/* Fail to find a neighbor, so store the packet into the flow. */
325 	if (insert_new_item(tbl, pkt, start_time, prev_idx, sent_seq,
326 				ip_id, is_atomic) == INVALID_ARRAY_INDEX)
327 		return -1;
328 
329 	return 0;
330 }
331 
332 uint16_t
333 gro_tcp4_tbl_timeout_flush(struct gro_tcp4_tbl *tbl,
334 		uint64_t flush_timestamp,
335 		struct rte_mbuf **out,
336 		uint16_t nb_out)
337 {
338 	uint16_t k = 0;
339 	uint32_t i, j;
340 	uint32_t max_flow_num = tbl->max_flow_num;
341 
342 	for (i = 0; i < max_flow_num; i++) {
343 		if (unlikely(tbl->flow_num == 0))
344 			return k;
345 
346 		j = tbl->flows[i].start_index;
347 		while (j != INVALID_ARRAY_INDEX) {
348 			if (tbl->items[j].start_time <= flush_timestamp) {
349 				out[k++] = tbl->items[j].firstseg;
350 				if (tbl->items[j].nb_merged > 1)
351 					update_header(&(tbl->items[j]));
352 				/*
353 				 * Delete the packet and get the next
354 				 * packet in the flow.
355 				 */
356 				j = delete_item(tbl, j, INVALID_ARRAY_INDEX);
357 				tbl->flows[i].start_index = j;
358 				if (j == INVALID_ARRAY_INDEX)
359 					tbl->flow_num--;
360 
361 				if (unlikely(k == nb_out))
362 					return k;
363 			} else
364 				/*
365 				 * The left packets in this flow won't be
366 				 * timeout. Go to check other flows.
367 				 */
368 				break;
369 		}
370 	}
371 	return k;
372 }
373 
374 uint32_t
375 gro_tcp4_tbl_pkt_count(void *tbl)
376 {
377 	struct gro_tcp4_tbl *gro_tbl = tbl;
378 
379 	if (gro_tbl)
380 		return gro_tbl->item_num;
381 
382 	return 0;
383 }
384