1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2020 Inspur Corporation
3 */
4
5 #include <rte_malloc.h>
6 #include <rte_mbuf.h>
7 #include <rte_ethdev.h>
8
9 #include "gro_udp4.h"
10
11 void *
gro_udp4_tbl_create(uint16_t socket_id,uint16_t max_flow_num,uint16_t max_item_per_flow)12 gro_udp4_tbl_create(uint16_t socket_id,
13 uint16_t max_flow_num,
14 uint16_t max_item_per_flow)
15 {
16 struct gro_udp4_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_UDP4_TBL_MAX_ITEM_NUM);
22
23 if (entries_num == 0)
24 return NULL;
25
26 tbl = rte_zmalloc_socket(__func__,
27 sizeof(struct gro_udp4_tbl),
28 RTE_CACHE_LINE_SIZE,
29 socket_id);
30 if (tbl == NULL)
31 return NULL;
32
33 size = sizeof(struct gro_udp4_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_udp4_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
gro_udp4_tbl_destroy(void * tbl)63 gro_udp4_tbl_destroy(void *tbl)
64 {
65 struct gro_udp4_tbl *udp_tbl = tbl;
66
67 if (udp_tbl) {
68 rte_free(udp_tbl->items);
69 rte_free(udp_tbl->flows);
70 }
71 rte_free(udp_tbl);
72 }
73
74 static inline uint32_t
find_an_empty_item(struct gro_udp4_tbl * tbl)75 find_an_empty_item(struct gro_udp4_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
find_an_empty_flow(struct gro_udp4_tbl * tbl)87 find_an_empty_flow(struct gro_udp4_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
insert_new_item(struct gro_udp4_tbl * tbl,struct rte_mbuf * pkt,uint64_t start_time,uint32_t prev_idx,uint16_t frag_offset,uint8_t is_last_frag)99 insert_new_item(struct gro_udp4_tbl *tbl,
100 struct rte_mbuf *pkt,
101 uint64_t start_time,
102 uint32_t prev_idx,
103 uint16_t frag_offset,
104 uint8_t is_last_frag)
105 {
106 uint32_t item_idx;
107
108 item_idx = find_an_empty_item(tbl);
109 if (unlikely(item_idx == INVALID_ARRAY_INDEX))
110 return INVALID_ARRAY_INDEX;
111
112 tbl->items[item_idx].firstseg = pkt;
113 tbl->items[item_idx].lastseg = rte_pktmbuf_lastseg(pkt);
114 tbl->items[item_idx].start_time = start_time;
115 tbl->items[item_idx].next_pkt_idx = INVALID_ARRAY_INDEX;
116 tbl->items[item_idx].frag_offset = frag_offset;
117 tbl->items[item_idx].is_last_frag = is_last_frag;
118 tbl->items[item_idx].nb_merged = 1;
119 tbl->item_num++;
120
121 /* if the previous packet exists, chain them together. */
122 if (prev_idx != INVALID_ARRAY_INDEX) {
123 tbl->items[item_idx].next_pkt_idx =
124 tbl->items[prev_idx].next_pkt_idx;
125 tbl->items[prev_idx].next_pkt_idx = item_idx;
126 }
127
128 return item_idx;
129 }
130
131 static inline uint32_t
delete_item(struct gro_udp4_tbl * tbl,uint32_t item_idx,uint32_t prev_item_idx)132 delete_item(struct gro_udp4_tbl *tbl, uint32_t item_idx,
133 uint32_t prev_item_idx)
134 {
135 uint32_t next_idx = tbl->items[item_idx].next_pkt_idx;
136
137 /* NULL indicates an empty item */
138 tbl->items[item_idx].firstseg = NULL;
139 tbl->item_num--;
140 if (prev_item_idx != INVALID_ARRAY_INDEX)
141 tbl->items[prev_item_idx].next_pkt_idx = next_idx;
142
143 return next_idx;
144 }
145
146 static inline uint32_t
insert_new_flow(struct gro_udp4_tbl * tbl,struct udp4_flow_key * src,uint32_t item_idx)147 insert_new_flow(struct gro_udp4_tbl *tbl,
148 struct udp4_flow_key *src,
149 uint32_t item_idx)
150 {
151 struct udp4_flow_key *dst;
152 uint32_t flow_idx;
153
154 flow_idx = find_an_empty_flow(tbl);
155 if (unlikely(flow_idx == INVALID_ARRAY_INDEX))
156 return INVALID_ARRAY_INDEX;
157
158 dst = &(tbl->flows[flow_idx].key);
159
160 rte_ether_addr_copy(&(src->eth_saddr), &(dst->eth_saddr));
161 rte_ether_addr_copy(&(src->eth_daddr), &(dst->eth_daddr));
162 dst->ip_src_addr = src->ip_src_addr;
163 dst->ip_dst_addr = src->ip_dst_addr;
164 dst->ip_id = src->ip_id;
165
166 tbl->flows[flow_idx].start_index = item_idx;
167 tbl->flow_num++;
168
169 return flow_idx;
170 }
171
172 /*
173 * update the packet length for the flushed packet.
174 */
175 static inline void
update_header(struct gro_udp4_item * item)176 update_header(struct gro_udp4_item *item)
177 {
178 struct rte_ipv4_hdr *ipv4_hdr;
179 struct rte_mbuf *pkt = item->firstseg;
180 uint16_t frag_offset;
181
182 ipv4_hdr = (struct rte_ipv4_hdr *)(rte_pktmbuf_mtod(pkt, char *) +
183 pkt->l2_len);
184 ipv4_hdr->total_length = rte_cpu_to_be_16(pkt->pkt_len -
185 pkt->l2_len);
186
187 /* Clear MF bit if it is last fragment */
188 if (item->is_last_frag) {
189 frag_offset = rte_be_to_cpu_16(ipv4_hdr->fragment_offset);
190 ipv4_hdr->fragment_offset =
191 rte_cpu_to_be_16(frag_offset & ~RTE_IPV4_HDR_MF_FLAG);
192 }
193 }
194
195 int32_t
gro_udp4_reassemble(struct rte_mbuf * pkt,struct gro_udp4_tbl * tbl,uint64_t start_time)196 gro_udp4_reassemble(struct rte_mbuf *pkt,
197 struct gro_udp4_tbl *tbl,
198 uint64_t start_time)
199 {
200 struct rte_ether_hdr *eth_hdr;
201 struct rte_ipv4_hdr *ipv4_hdr;
202 uint16_t ip_dl;
203 uint16_t ip_id, hdr_len;
204 uint16_t frag_offset = 0;
205 uint8_t is_last_frag;
206
207 struct udp4_flow_key key;
208 uint32_t cur_idx, prev_idx, item_idx;
209 uint32_t i, max_flow_num, remaining_flow_num;
210 int cmp;
211 uint8_t find;
212
213 eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
214 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)eth_hdr + pkt->l2_len);
215 hdr_len = pkt->l2_len + pkt->l3_len;
216
217 /*
218 * Don't process non-fragment packet.
219 */
220 if (!is_ipv4_fragment(ipv4_hdr))
221 return -1;
222
223 /*
224 * Don't process the packet whose payload length is less than or
225 * equal to 0.
226 */
227 if (pkt->pkt_len <= hdr_len)
228 return -1;
229
230 ip_dl = rte_be_to_cpu_16(ipv4_hdr->total_length);
231 if (ip_dl <= pkt->l3_len)
232 return -1;
233
234 ip_dl -= pkt->l3_len;
235 ip_id = rte_be_to_cpu_16(ipv4_hdr->packet_id);
236 frag_offset = rte_be_to_cpu_16(ipv4_hdr->fragment_offset);
237 is_last_frag = ((frag_offset & RTE_IPV4_HDR_MF_FLAG) == 0) ? 1 : 0;
238 frag_offset = (uint16_t)(frag_offset & RTE_IPV4_HDR_OFFSET_MASK) << 3;
239
240 rte_ether_addr_copy(&(eth_hdr->src_addr), &(key.eth_saddr));
241 rte_ether_addr_copy(&(eth_hdr->dst_addr), &(key.eth_daddr));
242 key.ip_src_addr = ipv4_hdr->src_addr;
243 key.ip_dst_addr = ipv4_hdr->dst_addr;
244 key.ip_id = ip_id;
245
246 /* Search for a matched flow. */
247 max_flow_num = tbl->max_flow_num;
248 remaining_flow_num = tbl->flow_num;
249 find = 0;
250 for (i = 0; i < max_flow_num && remaining_flow_num; i++) {
251 if (tbl->flows[i].start_index != INVALID_ARRAY_INDEX) {
252 if (is_same_udp4_flow(tbl->flows[i].key, key)) {
253 find = 1;
254 break;
255 }
256 remaining_flow_num--;
257 }
258 }
259
260 /*
261 * Fail to find a matched flow. Insert a new flow and store the
262 * packet into the flow.
263 */
264 if (find == 0) {
265 item_idx = insert_new_item(tbl, pkt, start_time,
266 INVALID_ARRAY_INDEX, frag_offset,
267 is_last_frag);
268 if (unlikely(item_idx == INVALID_ARRAY_INDEX))
269 return -1;
270 if (insert_new_flow(tbl, &key, item_idx) ==
271 INVALID_ARRAY_INDEX) {
272 /*
273 * Fail to insert a new flow, so delete the
274 * stored packet.
275 */
276 delete_item(tbl, item_idx, INVALID_ARRAY_INDEX);
277 return -1;
278 }
279 return 0;
280 }
281
282 /*
283 * Check all packets in the flow and try to find a neighbor for
284 * the input packet.
285 */
286 cur_idx = tbl->flows[i].start_index;
287 prev_idx = cur_idx;
288 do {
289 cmp = udp4_check_neighbor(&(tbl->items[cur_idx]),
290 frag_offset, ip_dl, 0);
291 if (cmp) {
292 if (merge_two_udp4_packets(&(tbl->items[cur_idx]),
293 pkt, cmp, frag_offset,
294 is_last_frag, 0))
295 return 1;
296 /*
297 * Fail to merge the two packets, as the packet
298 * length is greater than the max value. Store
299 * the packet into the flow.
300 */
301 if (insert_new_item(tbl, pkt, start_time, prev_idx,
302 frag_offset, is_last_frag) ==
303 INVALID_ARRAY_INDEX)
304 return -1;
305 return 0;
306 }
307
308 /* Ensure inserted items are ordered by frag_offset */
309 if (frag_offset
310 < tbl->items[cur_idx].frag_offset) {
311 break;
312 }
313
314 prev_idx = cur_idx;
315 cur_idx = tbl->items[cur_idx].next_pkt_idx;
316 } while (cur_idx != INVALID_ARRAY_INDEX);
317
318 /* Fail to find a neighbor, so store the packet into the flow. */
319 if (cur_idx == tbl->flows[i].start_index) {
320 /* Insert it before the first packet of the flow */
321 item_idx = insert_new_item(tbl, pkt, start_time,
322 INVALID_ARRAY_INDEX, frag_offset,
323 is_last_frag);
324 if (unlikely(item_idx == INVALID_ARRAY_INDEX))
325 return -1;
326 tbl->items[item_idx].next_pkt_idx = cur_idx;
327 tbl->flows[i].start_index = item_idx;
328 } else {
329 if (insert_new_item(tbl, pkt, start_time, prev_idx,
330 frag_offset, is_last_frag)
331 == INVALID_ARRAY_INDEX)
332 return -1;
333 }
334
335 return 0;
336 }
337
338 static int
gro_udp4_merge_items(struct gro_udp4_tbl * tbl,uint32_t start_idx)339 gro_udp4_merge_items(struct gro_udp4_tbl *tbl,
340 uint32_t start_idx)
341 {
342 uint16_t frag_offset;
343 uint8_t is_last_frag;
344 int16_t ip_dl;
345 struct rte_mbuf *pkt;
346 int cmp;
347 uint32_t item_idx;
348 uint16_t hdr_len;
349
350 item_idx = tbl->items[start_idx].next_pkt_idx;
351 while (item_idx != INVALID_ARRAY_INDEX) {
352 pkt = tbl->items[item_idx].firstseg;
353 hdr_len = pkt->l2_len + pkt->l3_len;
354 ip_dl = pkt->pkt_len - hdr_len;
355 frag_offset = tbl->items[item_idx].frag_offset;
356 is_last_frag = tbl->items[item_idx].is_last_frag;
357 cmp = udp4_check_neighbor(&(tbl->items[start_idx]),
358 frag_offset, ip_dl, 0);
359 if (cmp) {
360 if (merge_two_udp4_packets(
361 &(tbl->items[start_idx]),
362 pkt, cmp, frag_offset,
363 is_last_frag, 0)) {
364 item_idx = delete_item(tbl, item_idx,
365 INVALID_ARRAY_INDEX);
366 tbl->items[start_idx].next_pkt_idx
367 = item_idx;
368 } else
369 return 0;
370 } else
371 return 0;
372 }
373
374 return 0;
375 }
376
377 uint16_t
gro_udp4_tbl_timeout_flush(struct gro_udp4_tbl * tbl,uint64_t flush_timestamp,struct rte_mbuf ** out,uint16_t nb_out)378 gro_udp4_tbl_timeout_flush(struct gro_udp4_tbl *tbl,
379 uint64_t flush_timestamp,
380 struct rte_mbuf **out,
381 uint16_t nb_out)
382 {
383 uint16_t k = 0;
384 uint32_t i, j;
385 uint32_t max_flow_num = tbl->max_flow_num;
386
387 for (i = 0; i < max_flow_num; i++) {
388 if (unlikely(tbl->flow_num == 0))
389 return k;
390
391 j = tbl->flows[i].start_index;
392 while (j != INVALID_ARRAY_INDEX) {
393 if (tbl->items[j].start_time <= flush_timestamp) {
394 gro_udp4_merge_items(tbl, j);
395 out[k++] = tbl->items[j].firstseg;
396 if (tbl->items[j].nb_merged > 1)
397 update_header(&(tbl->items[j]));
398 /*
399 * Delete the packet and get the next
400 * packet in the flow.
401 */
402 j = delete_item(tbl, j, INVALID_ARRAY_INDEX);
403 tbl->flows[i].start_index = j;
404 if (j == INVALID_ARRAY_INDEX)
405 tbl->flow_num--;
406
407 if (unlikely(k == nb_out))
408 return k;
409 } else
410 /*
411 * Flushing packets does not strictly follow
412 * timestamp. It does not flush left packets of
413 * the flow this time once it finds one item
414 * whose start_time is greater than
415 * flush_timestamp. So go to check other flows.
416 */
417 break;
418 }
419 }
420 return k;
421 }
422
423 uint32_t
gro_udp4_tbl_pkt_count(void * tbl)424 gro_udp4_tbl_pkt_count(void *tbl)
425 {
426 struct gro_udp4_tbl *gro_tbl = tbl;
427
428 if (gro_tbl)
429 return gro_tbl->item_num;
430
431 return 0;
432 }
433