1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2018 Intel Corporation
3 */
4
5 #include <rte_malloc.h>
6 #include <rte_mbuf.h>
7 #include <rte_cycles.h>
8 #include <rte_ethdev.h>
9 #include <rte_udp.h>
10
11 #include "gro_vxlan_tcp4.h"
12
13 void *
gro_vxlan_tcp4_tbl_create(uint16_t socket_id,uint16_t max_flow_num,uint16_t max_item_per_flow)14 gro_vxlan_tcp4_tbl_create(uint16_t socket_id,
15 uint16_t max_flow_num,
16 uint16_t max_item_per_flow)
17 {
18 struct gro_vxlan_tcp4_tbl *tbl;
19 size_t size;
20 uint32_t entries_num, i;
21
22 entries_num = max_flow_num * max_item_per_flow;
23 entries_num = RTE_MIN(entries_num, GRO_VXLAN_TCP4_TBL_MAX_ITEM_NUM);
24
25 if (entries_num == 0)
26 return NULL;
27
28 tbl = rte_zmalloc_socket(__func__,
29 sizeof(struct gro_vxlan_tcp4_tbl),
30 RTE_CACHE_LINE_SIZE,
31 socket_id);
32 if (tbl == NULL)
33 return NULL;
34
35 size = sizeof(struct gro_vxlan_tcp4_item) * entries_num;
36 tbl->items = rte_zmalloc_socket(__func__,
37 size,
38 RTE_CACHE_LINE_SIZE,
39 socket_id);
40 if (tbl->items == NULL) {
41 rte_free(tbl);
42 return NULL;
43 }
44 tbl->max_item_num = entries_num;
45
46 size = sizeof(struct gro_vxlan_tcp4_flow) * entries_num;
47 tbl->flows = rte_zmalloc_socket(__func__,
48 size,
49 RTE_CACHE_LINE_SIZE,
50 socket_id);
51 if (tbl->flows == NULL) {
52 rte_free(tbl->items);
53 rte_free(tbl);
54 return NULL;
55 }
56
57 for (i = 0; i < entries_num; i++)
58 tbl->flows[i].start_index = INVALID_ARRAY_INDEX;
59 tbl->max_flow_num = entries_num;
60
61 return tbl;
62 }
63
64 void
gro_vxlan_tcp4_tbl_destroy(void * tbl)65 gro_vxlan_tcp4_tbl_destroy(void *tbl)
66 {
67 struct gro_vxlan_tcp4_tbl *vxlan_tbl = tbl;
68
69 if (vxlan_tbl) {
70 rte_free(vxlan_tbl->items);
71 rte_free(vxlan_tbl->flows);
72 }
73 rte_free(vxlan_tbl);
74 }
75
76 static inline uint32_t
find_an_empty_item(struct gro_vxlan_tcp4_tbl * tbl)77 find_an_empty_item(struct gro_vxlan_tcp4_tbl *tbl)
78 {
79 uint32_t max_item_num = tbl->max_item_num, i;
80
81 for (i = 0; i < max_item_num; i++)
82 if (tbl->items[i].inner_item.firstseg == NULL)
83 return i;
84 return INVALID_ARRAY_INDEX;
85 }
86
87 static inline uint32_t
find_an_empty_flow(struct gro_vxlan_tcp4_tbl * tbl)88 find_an_empty_flow(struct gro_vxlan_tcp4_tbl *tbl)
89 {
90 uint32_t max_flow_num = tbl->max_flow_num, i;
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
insert_new_item(struct gro_vxlan_tcp4_tbl * tbl,struct rte_mbuf * pkt,uint64_t start_time,uint32_t prev_idx,uint32_t sent_seq,uint16_t outer_ip_id,uint16_t ip_id,uint8_t outer_is_atomic,uint8_t is_atomic)99 insert_new_item(struct gro_vxlan_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 outer_ip_id,
105 uint16_t ip_id,
106 uint8_t outer_is_atomic,
107 uint8_t is_atomic)
108 {
109 uint32_t item_idx;
110
111 item_idx = find_an_empty_item(tbl);
112 if (unlikely(item_idx == INVALID_ARRAY_INDEX))
113 return INVALID_ARRAY_INDEX;
114
115 tbl->items[item_idx].inner_item.firstseg = pkt;
116 tbl->items[item_idx].inner_item.lastseg = rte_pktmbuf_lastseg(pkt);
117 tbl->items[item_idx].inner_item.start_time = start_time;
118 tbl->items[item_idx].inner_item.next_pkt_idx = INVALID_ARRAY_INDEX;
119 tbl->items[item_idx].inner_item.sent_seq = sent_seq;
120 tbl->items[item_idx].inner_item.ip_id = ip_id;
121 tbl->items[item_idx].inner_item.nb_merged = 1;
122 tbl->items[item_idx].inner_item.is_atomic = is_atomic;
123 tbl->items[item_idx].outer_ip_id = outer_ip_id;
124 tbl->items[item_idx].outer_is_atomic = outer_is_atomic;
125 tbl->item_num++;
126
127 /* If the previous packet exists, chain the new one with it. */
128 if (prev_idx != INVALID_ARRAY_INDEX) {
129 tbl->items[item_idx].inner_item.next_pkt_idx =
130 tbl->items[prev_idx].inner_item.next_pkt_idx;
131 tbl->items[prev_idx].inner_item.next_pkt_idx = item_idx;
132 }
133
134 return item_idx;
135 }
136
137 static inline uint32_t
delete_item(struct gro_vxlan_tcp4_tbl * tbl,uint32_t item_idx,uint32_t prev_item_idx)138 delete_item(struct gro_vxlan_tcp4_tbl *tbl,
139 uint32_t item_idx,
140 uint32_t prev_item_idx)
141 {
142 uint32_t next_idx = tbl->items[item_idx].inner_item.next_pkt_idx;
143
144 /* NULL indicates an empty item. */
145 tbl->items[item_idx].inner_item.firstseg = NULL;
146 tbl->item_num--;
147 if (prev_item_idx != INVALID_ARRAY_INDEX)
148 tbl->items[prev_item_idx].inner_item.next_pkt_idx = next_idx;
149
150 return next_idx;
151 }
152
153 static inline uint32_t
insert_new_flow(struct gro_vxlan_tcp4_tbl * tbl,struct vxlan_tcp4_flow_key * src,uint32_t item_idx)154 insert_new_flow(struct gro_vxlan_tcp4_tbl *tbl,
155 struct vxlan_tcp4_flow_key *src,
156 uint32_t item_idx)
157 {
158 struct vxlan_tcp4_flow_key *dst;
159 uint32_t flow_idx;
160
161 flow_idx = find_an_empty_flow(tbl);
162 if (unlikely(flow_idx == INVALID_ARRAY_INDEX))
163 return INVALID_ARRAY_INDEX;
164
165 dst = &(tbl->flows[flow_idx].key);
166
167 rte_ether_addr_copy(&(src->inner_key.eth_saddr),
168 &(dst->inner_key.eth_saddr));
169 rte_ether_addr_copy(&(src->inner_key.eth_daddr),
170 &(dst->inner_key.eth_daddr));
171 dst->inner_key.ip_src_addr = src->inner_key.ip_src_addr;
172 dst->inner_key.ip_dst_addr = src->inner_key.ip_dst_addr;
173 dst->inner_key.recv_ack = src->inner_key.recv_ack;
174 dst->inner_key.src_port = src->inner_key.src_port;
175 dst->inner_key.dst_port = src->inner_key.dst_port;
176
177 dst->vxlan_hdr.vx_flags = src->vxlan_hdr.vx_flags;
178 dst->vxlan_hdr.vx_vni = src->vxlan_hdr.vx_vni;
179 rte_ether_addr_copy(&(src->outer_eth_saddr), &(dst->outer_eth_saddr));
180 rte_ether_addr_copy(&(src->outer_eth_daddr), &(dst->outer_eth_daddr));
181 dst->outer_ip_src_addr = src->outer_ip_src_addr;
182 dst->outer_ip_dst_addr = src->outer_ip_dst_addr;
183 dst->outer_src_port = src->outer_src_port;
184 dst->outer_dst_port = src->outer_dst_port;
185
186 tbl->flows[flow_idx].start_index = item_idx;
187 tbl->flow_num++;
188
189 return flow_idx;
190 }
191
192 static inline int
is_same_vxlan_tcp4_flow(struct vxlan_tcp4_flow_key k1,struct vxlan_tcp4_flow_key k2)193 is_same_vxlan_tcp4_flow(struct vxlan_tcp4_flow_key k1,
194 struct vxlan_tcp4_flow_key k2)
195 {
196 return (rte_is_same_ether_addr(&k1.outer_eth_saddr,
197 &k2.outer_eth_saddr) &&
198 rte_is_same_ether_addr(&k1.outer_eth_daddr,
199 &k2.outer_eth_daddr) &&
200 (k1.outer_ip_src_addr == k2.outer_ip_src_addr) &&
201 (k1.outer_ip_dst_addr == k2.outer_ip_dst_addr) &&
202 (k1.outer_src_port == k2.outer_src_port) &&
203 (k1.outer_dst_port == k2.outer_dst_port) &&
204 (k1.vxlan_hdr.vx_flags == k2.vxlan_hdr.vx_flags) &&
205 (k1.vxlan_hdr.vx_vni == k2.vxlan_hdr.vx_vni) &&
206 is_same_tcp4_flow(k1.inner_key, k2.inner_key));
207 }
208
209 static inline int
check_vxlan_seq_option(struct gro_vxlan_tcp4_item * item,struct rte_tcp_hdr * tcp_hdr,uint32_t sent_seq,uint16_t outer_ip_id,uint16_t ip_id,uint16_t tcp_hl,uint16_t tcp_dl,uint8_t outer_is_atomic,uint8_t is_atomic)210 check_vxlan_seq_option(struct gro_vxlan_tcp4_item *item,
211 struct rte_tcp_hdr *tcp_hdr,
212 uint32_t sent_seq,
213 uint16_t outer_ip_id,
214 uint16_t ip_id,
215 uint16_t tcp_hl,
216 uint16_t tcp_dl,
217 uint8_t outer_is_atomic,
218 uint8_t is_atomic)
219 {
220 struct rte_mbuf *pkt = item->inner_item.firstseg;
221 int cmp;
222 uint16_t l2_offset;
223
224 /* Don't merge packets whose outer DF bits are different. */
225 if (unlikely(item->outer_is_atomic ^ outer_is_atomic))
226 return 0;
227
228 l2_offset = pkt->outer_l2_len + pkt->outer_l3_len;
229 cmp = check_seq_option(&item->inner_item, tcp_hdr, sent_seq, ip_id,
230 tcp_hl, tcp_dl, l2_offset, is_atomic);
231 if ((cmp > 0) && (outer_is_atomic ||
232 (outer_ip_id == item->outer_ip_id + 1)))
233 /* Append the new packet. */
234 return 1;
235 else if ((cmp < 0) && (outer_is_atomic ||
236 (outer_ip_id + item->inner_item.nb_merged ==
237 item->outer_ip_id)))
238 /* Prepend the new packet. */
239 return -1;
240
241 return 0;
242 }
243
244 static inline int
merge_two_vxlan_tcp4_packets(struct gro_vxlan_tcp4_item * item,struct rte_mbuf * pkt,int cmp,uint32_t sent_seq,uint16_t outer_ip_id,uint16_t ip_id)245 merge_two_vxlan_tcp4_packets(struct gro_vxlan_tcp4_item *item,
246 struct rte_mbuf *pkt,
247 int cmp,
248 uint32_t sent_seq,
249 uint16_t outer_ip_id,
250 uint16_t ip_id)
251 {
252 if (merge_two_tcp4_packets(&item->inner_item, pkt, cmp, sent_seq,
253 ip_id, pkt->outer_l2_len +
254 pkt->outer_l3_len)) {
255 /* Update the outer IPv4 ID to the large value. */
256 item->outer_ip_id = cmp > 0 ? outer_ip_id : item->outer_ip_id;
257 return 1;
258 }
259
260 return 0;
261 }
262
263 static inline void
update_vxlan_header(struct gro_vxlan_tcp4_item * item)264 update_vxlan_header(struct gro_vxlan_tcp4_item *item)
265 {
266 struct rte_ipv4_hdr *ipv4_hdr;
267 struct rte_udp_hdr *udp_hdr;
268 struct rte_mbuf *pkt = item->inner_item.firstseg;
269 uint16_t len;
270
271 /* Update the outer IPv4 header. */
272 len = pkt->pkt_len - pkt->outer_l2_len;
273 ipv4_hdr = (struct rte_ipv4_hdr *)(rte_pktmbuf_mtod(pkt, char *) +
274 pkt->outer_l2_len);
275 ipv4_hdr->total_length = rte_cpu_to_be_16(len);
276
277 /* Update the outer UDP header. */
278 len -= pkt->outer_l3_len;
279 udp_hdr = (struct rte_udp_hdr *)((char *)ipv4_hdr + pkt->outer_l3_len);
280 udp_hdr->dgram_len = rte_cpu_to_be_16(len);
281
282 /* Update the inner IPv4 header. */
283 len -= pkt->l2_len;
284 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)udp_hdr + pkt->l2_len);
285 ipv4_hdr->total_length = rte_cpu_to_be_16(len);
286 }
287
288 int32_t
gro_vxlan_tcp4_reassemble(struct rte_mbuf * pkt,struct gro_vxlan_tcp4_tbl * tbl,uint64_t start_time)289 gro_vxlan_tcp4_reassemble(struct rte_mbuf *pkt,
290 struct gro_vxlan_tcp4_tbl *tbl,
291 uint64_t start_time)
292 {
293 struct rte_ether_hdr *outer_eth_hdr, *eth_hdr;
294 struct rte_ipv4_hdr *outer_ipv4_hdr, *ipv4_hdr;
295 struct rte_tcp_hdr *tcp_hdr;
296 struct rte_udp_hdr *udp_hdr;
297 struct rte_vxlan_hdr *vxlan_hdr;
298 uint32_t sent_seq;
299 int32_t tcp_dl;
300 uint16_t frag_off, outer_ip_id, ip_id;
301 uint8_t outer_is_atomic, is_atomic;
302
303 struct vxlan_tcp4_flow_key key;
304 uint32_t cur_idx, prev_idx, item_idx;
305 uint32_t i, max_flow_num, remaining_flow_num;
306 int cmp;
307 uint16_t hdr_len;
308 uint8_t find;
309
310 /*
311 * Don't process the packet whose TCP header length is greater
312 * than 60 bytes or less than 20 bytes.
313 */
314 if (unlikely(INVALID_TCP_HDRLEN(pkt->l4_len)))
315 return -1;
316
317 outer_eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
318 outer_ipv4_hdr = (struct rte_ipv4_hdr *)((char *)outer_eth_hdr +
319 pkt->outer_l2_len);
320 udp_hdr = (struct rte_udp_hdr *)((char *)outer_ipv4_hdr +
321 pkt->outer_l3_len);
322 vxlan_hdr = (struct rte_vxlan_hdr *)((char *)udp_hdr +
323 sizeof(struct rte_udp_hdr));
324 eth_hdr = (struct rte_ether_hdr *)((char *)vxlan_hdr +
325 sizeof(struct rte_vxlan_hdr));
326 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)udp_hdr + pkt->l2_len);
327 tcp_hdr = (struct rte_tcp_hdr *)((char *)ipv4_hdr + pkt->l3_len);
328
329 /*
330 * Don't process the packet which has FIN, SYN, RST, PSH, URG,
331 * ECE or CWR set.
332 */
333 if (tcp_hdr->tcp_flags != RTE_TCP_ACK_FLAG)
334 return -1;
335
336 hdr_len = pkt->outer_l2_len + pkt->outer_l3_len + pkt->l2_len +
337 pkt->l3_len + pkt->l4_len;
338 /*
339 * Don't process the packet whose payload length is less than or
340 * equal to 0.
341 */
342 tcp_dl = pkt->pkt_len - hdr_len;
343 if (tcp_dl <= 0)
344 return -1;
345
346 /*
347 * Save IPv4 ID for the packet whose DF bit is 0. For the packet
348 * whose DF bit is 1, IPv4 ID is ignored.
349 */
350 frag_off = rte_be_to_cpu_16(outer_ipv4_hdr->fragment_offset);
351 outer_is_atomic =
352 (frag_off & RTE_IPV4_HDR_DF_FLAG) == RTE_IPV4_HDR_DF_FLAG;
353 outer_ip_id = outer_is_atomic ? 0 :
354 rte_be_to_cpu_16(outer_ipv4_hdr->packet_id);
355 frag_off = rte_be_to_cpu_16(ipv4_hdr->fragment_offset);
356 is_atomic = (frag_off & RTE_IPV4_HDR_DF_FLAG) == RTE_IPV4_HDR_DF_FLAG;
357 ip_id = is_atomic ? 0 : rte_be_to_cpu_16(ipv4_hdr->packet_id);
358
359 sent_seq = rte_be_to_cpu_32(tcp_hdr->sent_seq);
360
361 rte_ether_addr_copy(&(eth_hdr->s_addr), &(key.inner_key.eth_saddr));
362 rte_ether_addr_copy(&(eth_hdr->d_addr), &(key.inner_key.eth_daddr));
363 key.inner_key.ip_src_addr = ipv4_hdr->src_addr;
364 key.inner_key.ip_dst_addr = ipv4_hdr->dst_addr;
365 key.inner_key.recv_ack = tcp_hdr->recv_ack;
366 key.inner_key.src_port = tcp_hdr->src_port;
367 key.inner_key.dst_port = tcp_hdr->dst_port;
368
369 key.vxlan_hdr.vx_flags = vxlan_hdr->vx_flags;
370 key.vxlan_hdr.vx_vni = vxlan_hdr->vx_vni;
371 rte_ether_addr_copy(&(outer_eth_hdr->s_addr), &(key.outer_eth_saddr));
372 rte_ether_addr_copy(&(outer_eth_hdr->d_addr), &(key.outer_eth_daddr));
373 key.outer_ip_src_addr = outer_ipv4_hdr->src_addr;
374 key.outer_ip_dst_addr = outer_ipv4_hdr->dst_addr;
375 key.outer_src_port = udp_hdr->src_port;
376 key.outer_dst_port = udp_hdr->dst_port;
377
378 /* Search for a matched flow. */
379 max_flow_num = tbl->max_flow_num;
380 remaining_flow_num = tbl->flow_num;
381 find = 0;
382 for (i = 0; i < max_flow_num && remaining_flow_num; i++) {
383 if (tbl->flows[i].start_index != INVALID_ARRAY_INDEX) {
384 if (is_same_vxlan_tcp4_flow(tbl->flows[i].key, key)) {
385 find = 1;
386 break;
387 }
388 remaining_flow_num--;
389 }
390 }
391
392 /*
393 * Can't find a matched flow. Insert a new flow and store the
394 * packet into the flow.
395 */
396 if (find == 0) {
397 item_idx = insert_new_item(tbl, pkt, start_time,
398 INVALID_ARRAY_INDEX, sent_seq, outer_ip_id,
399 ip_id, outer_is_atomic, is_atomic);
400 if (item_idx == INVALID_ARRAY_INDEX)
401 return -1;
402 if (insert_new_flow(tbl, &key, item_idx) ==
403 INVALID_ARRAY_INDEX) {
404 /*
405 * Fail to insert a new flow, so
406 * delete the inserted packet.
407 */
408 delete_item(tbl, item_idx, INVALID_ARRAY_INDEX);
409 return -1;
410 }
411 return 0;
412 }
413
414 /* Check all packets in the flow and try to find a neighbor. */
415 cur_idx = tbl->flows[i].start_index;
416 prev_idx = cur_idx;
417 do {
418 cmp = check_vxlan_seq_option(&(tbl->items[cur_idx]), tcp_hdr,
419 sent_seq, outer_ip_id, ip_id, pkt->l4_len,
420 tcp_dl, outer_is_atomic, is_atomic);
421 if (cmp) {
422 if (merge_two_vxlan_tcp4_packets(&(tbl->items[cur_idx]),
423 pkt, cmp, sent_seq,
424 outer_ip_id, ip_id))
425 return 1;
426 /*
427 * Can't merge two packets, as the packet
428 * length will be greater than the max value.
429 * Insert the packet into the flow.
430 */
431 if (insert_new_item(tbl, pkt, start_time, prev_idx,
432 sent_seq, outer_ip_id,
433 ip_id, outer_is_atomic,
434 is_atomic) ==
435 INVALID_ARRAY_INDEX)
436 return -1;
437 return 0;
438 }
439 prev_idx = cur_idx;
440 cur_idx = tbl->items[cur_idx].inner_item.next_pkt_idx;
441 } while (cur_idx != INVALID_ARRAY_INDEX);
442
443 /* Can't find neighbor. Insert the packet into the flow. */
444 if (insert_new_item(tbl, pkt, start_time, prev_idx, sent_seq,
445 outer_ip_id, ip_id, outer_is_atomic,
446 is_atomic) == INVALID_ARRAY_INDEX)
447 return -1;
448
449 return 0;
450 }
451
452 uint16_t
gro_vxlan_tcp4_tbl_timeout_flush(struct gro_vxlan_tcp4_tbl * tbl,uint64_t flush_timestamp,struct rte_mbuf ** out,uint16_t nb_out)453 gro_vxlan_tcp4_tbl_timeout_flush(struct gro_vxlan_tcp4_tbl *tbl,
454 uint64_t flush_timestamp,
455 struct rte_mbuf **out,
456 uint16_t nb_out)
457 {
458 uint16_t k = 0;
459 uint32_t i, j;
460 uint32_t max_flow_num = tbl->max_flow_num;
461
462 for (i = 0; i < max_flow_num; i++) {
463 if (unlikely(tbl->flow_num == 0))
464 return k;
465
466 j = tbl->flows[i].start_index;
467 while (j != INVALID_ARRAY_INDEX) {
468 if (tbl->items[j].inner_item.start_time <=
469 flush_timestamp) {
470 out[k++] = tbl->items[j].inner_item.firstseg;
471 if (tbl->items[j].inner_item.nb_merged > 1)
472 update_vxlan_header(&(tbl->items[j]));
473 /*
474 * Delete the item and get the next packet
475 * index.
476 */
477 j = delete_item(tbl, j, INVALID_ARRAY_INDEX);
478 tbl->flows[i].start_index = j;
479 if (j == INVALID_ARRAY_INDEX)
480 tbl->flow_num--;
481
482 if (unlikely(k == nb_out))
483 return k;
484 } else
485 /*
486 * The left packets in the flow won't be
487 * timeout. Go to check other flows.
488 */
489 break;
490 }
491 }
492 return k;
493 }
494
495 uint32_t
gro_vxlan_tcp4_tbl_pkt_count(void * tbl)496 gro_vxlan_tcp4_tbl_pkt_count(void *tbl)
497 {
498 struct gro_vxlan_tcp4_tbl *gro_tbl = tbl;
499
500 if (gro_tbl)
501 return gro_tbl->item_num;
502
503 return 0;
504 }
505