xref: /dpdk/drivers/net/tap/tap_flow.c (revision cc9ecbb48ee3a8fb80df6c470141260df3eacec0)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2017 6WIND S.A.
3  * Copyright 2017 Mellanox Technologies, Ltd
4  */
5 
6 #include <errno.h>
7 #include <string.h>
8 #include <unistd.h>
9 #include <sys/queue.h>
10 #include <sys/resource.h>
11 
12 #include <rte_byteorder.h>
13 #include <rte_jhash.h>
14 #include <rte_malloc.h>
15 #include <rte_eth_tap.h>
16 #include <tap_flow.h>
17 #include <tap_autoconf.h>
18 #include <tap_tcmsgs.h>
19 #include <tap_rss.h>
20 
21 #ifndef HAVE_TC_FLOWER
22 /*
23  * For kernels < 4.2, this enum is not defined. Runtime checks will be made to
24  * avoid sending TC messages the kernel cannot understand.
25  */
26 enum {
27 	TCA_FLOWER_UNSPEC,
28 	TCA_FLOWER_CLASSID,
29 	TCA_FLOWER_INDEV,
30 	TCA_FLOWER_ACT,
31 	TCA_FLOWER_KEY_ETH_DST,         /* ETH_ALEN */
32 	TCA_FLOWER_KEY_ETH_DST_MASK,    /* ETH_ALEN */
33 	TCA_FLOWER_KEY_ETH_SRC,         /* ETH_ALEN */
34 	TCA_FLOWER_KEY_ETH_SRC_MASK,    /* ETH_ALEN */
35 	TCA_FLOWER_KEY_ETH_TYPE,        /* be16 */
36 	TCA_FLOWER_KEY_IP_PROTO,        /* u8 */
37 	TCA_FLOWER_KEY_IPV4_SRC,        /* be32 */
38 	TCA_FLOWER_KEY_IPV4_SRC_MASK,   /* be32 */
39 	TCA_FLOWER_KEY_IPV4_DST,        /* be32 */
40 	TCA_FLOWER_KEY_IPV4_DST_MASK,   /* be32 */
41 	TCA_FLOWER_KEY_IPV6_SRC,        /* struct in6_addr */
42 	TCA_FLOWER_KEY_IPV6_SRC_MASK,   /* struct in6_addr */
43 	TCA_FLOWER_KEY_IPV6_DST,        /* struct in6_addr */
44 	TCA_FLOWER_KEY_IPV6_DST_MASK,   /* struct in6_addr */
45 	TCA_FLOWER_KEY_TCP_SRC,         /* be16 */
46 	TCA_FLOWER_KEY_TCP_DST,         /* be16 */
47 	TCA_FLOWER_KEY_UDP_SRC,         /* be16 */
48 	TCA_FLOWER_KEY_UDP_DST,         /* be16 */
49 };
50 #endif
51 #ifndef HAVE_TC_VLAN_ID
52 enum {
53 	/* TCA_FLOWER_FLAGS, */
54 	TCA_FLOWER_KEY_VLAN_ID = TCA_FLOWER_KEY_UDP_DST + 2, /* be16 */
55 	TCA_FLOWER_KEY_VLAN_PRIO,       /* u8   */
56 	TCA_FLOWER_KEY_VLAN_ETH_TYPE,   /* be16 */
57 };
58 #endif
59 /*
60  * For kernels < 4.2 BPF related enums may not be defined.
61  * Runtime checks will be carried out to gracefully report on TC messages that
62  * are rejected by the kernel. Rejection reasons may be due to:
63  * 1. enum is not defined
64  * 2. enum is defined but kernel is not configured to support BPF system calls,
65  *    BPF classifications or BPF actions.
66  */
67 #ifndef HAVE_TC_BPF
68 enum {
69 	TCA_BPF_UNSPEC,
70 	TCA_BPF_ACT,
71 	TCA_BPF_POLICE,
72 	TCA_BPF_CLASSID,
73 	TCA_BPF_OPS_LEN,
74 	TCA_BPF_OPS,
75 };
76 #endif
77 #ifndef HAVE_TC_BPF_FD
78 enum {
79 	TCA_BPF_FD = TCA_BPF_OPS + 1,
80 	TCA_BPF_NAME,
81 };
82 #endif
83 #ifndef HAVE_TC_ACT_BPF
84 #define tc_gen \
85 	__u32                 index; \
86 	__u32                 capab; \
87 	int                   action; \
88 	int                   refcnt; \
89 	int                   bindcnt
90 
91 struct tc_act_bpf {
92 	tc_gen;
93 };
94 
95 enum {
96 	TCA_ACT_BPF_UNSPEC,
97 	TCA_ACT_BPF_TM,
98 	TCA_ACT_BPF_PARMS,
99 	TCA_ACT_BPF_OPS_LEN,
100 	TCA_ACT_BPF_OPS,
101 };
102 
103 #endif
104 #ifndef HAVE_TC_ACT_BPF_FD
105 enum {
106 	TCA_ACT_BPF_FD = TCA_ACT_BPF_OPS + 1,
107 	TCA_ACT_BPF_NAME,
108 };
109 #endif
110 
111 /* RSS key management */
112 enum bpf_rss_key_e {
113 	KEY_CMD_GET = 1,
114 	KEY_CMD_RELEASE,
115 	KEY_CMD_INIT,
116 	KEY_CMD_DEINIT,
117 };
118 
119 enum key_status_e {
120 	KEY_STAT_UNSPEC,
121 	KEY_STAT_USED,
122 	KEY_STAT_AVAILABLE,
123 };
124 
125 #define ISOLATE_HANDLE 1
126 #define REMOTE_PROMISCUOUS_HANDLE 2
127 
128 struct rte_flow {
129 	LIST_ENTRY(rte_flow) next; /* Pointer to the next rte_flow structure */
130 	struct rte_flow *remote_flow; /* associated remote flow */
131 	int bpf_fd[SEC_MAX]; /* list of bfs fds per ELF section */
132 	uint32_t key_idx; /* RSS rule key index into BPF map */
133 	struct nlmsg msg;
134 };
135 
136 struct convert_data {
137 	uint16_t eth_type;
138 	uint16_t ip_proto;
139 	uint8_t vlan;
140 	struct rte_flow *flow;
141 };
142 
143 struct remote_rule {
144 	struct rte_flow_attr attr;
145 	struct rte_flow_item items[2];
146 	struct rte_flow_action actions[2];
147 	int mirred;
148 };
149 
150 struct action_data {
151 	char id[16];
152 
153 	union {
154 		struct tc_gact gact;
155 		struct tc_mirred mirred;
156 		struct skbedit {
157 			struct tc_skbedit skbedit;
158 			uint16_t queue;
159 		} skbedit;
160 		struct bpf {
161 			struct tc_act_bpf bpf;
162 			int bpf_fd;
163 			const char *annotation;
164 		} bpf;
165 	};
166 };
167 
168 static int tap_flow_create_eth(const struct rte_flow_item *item, void *data);
169 static int tap_flow_create_vlan(const struct rte_flow_item *item, void *data);
170 static int tap_flow_create_ipv4(const struct rte_flow_item *item, void *data);
171 static int tap_flow_create_ipv6(const struct rte_flow_item *item, void *data);
172 static int tap_flow_create_udp(const struct rte_flow_item *item, void *data);
173 static int tap_flow_create_tcp(const struct rte_flow_item *item, void *data);
174 static int
175 tap_flow_validate(struct rte_eth_dev *dev,
176 		  const struct rte_flow_attr *attr,
177 		  const struct rte_flow_item items[],
178 		  const struct rte_flow_action actions[],
179 		  struct rte_flow_error *error);
180 
181 static struct rte_flow *
182 tap_flow_create(struct rte_eth_dev *dev,
183 		const struct rte_flow_attr *attr,
184 		const struct rte_flow_item items[],
185 		const struct rte_flow_action actions[],
186 		struct rte_flow_error *error);
187 
188 static void
189 tap_flow_free(struct pmd_internals *pmd,
190 	struct rte_flow *flow);
191 
192 static int
193 tap_flow_destroy(struct rte_eth_dev *dev,
194 		 struct rte_flow *flow,
195 		 struct rte_flow_error *error);
196 
197 static int
198 tap_flow_isolate(struct rte_eth_dev *dev,
199 		 int set,
200 		 struct rte_flow_error *error);
201 
202 static int bpf_rss_key(enum bpf_rss_key_e cmd, __u32 *key_idx);
203 static int rss_enable(struct pmd_internals *pmd,
204 			const struct rte_flow_attr *attr,
205 			struct rte_flow_error *error);
206 static int rss_add_actions(struct rte_flow *flow, struct pmd_internals *pmd,
207 			const struct rte_flow_action_rss *rss,
208 			struct rte_flow_error *error);
209 
210 static const struct rte_flow_ops tap_flow_ops = {
211 	.validate = tap_flow_validate,
212 	.create = tap_flow_create,
213 	.destroy = tap_flow_destroy,
214 	.flush = tap_flow_flush,
215 	.isolate = tap_flow_isolate,
216 };
217 
218 /* Static initializer for items. */
219 #define ITEMS(...) \
220 	(const enum rte_flow_item_type []){ \
221 		__VA_ARGS__, RTE_FLOW_ITEM_TYPE_END, \
222 	}
223 
224 /* Structure to generate a simple graph of layers supported by the NIC. */
225 struct tap_flow_items {
226 	/* Bit-mask corresponding to what is supported for this item. */
227 	const void *mask;
228 	const unsigned int mask_sz; /* Bit-mask size in bytes. */
229 	/*
230 	 * Bit-mask corresponding to the default mask, if none is provided
231 	 * along with the item.
232 	 */
233 	const void *default_mask;
234 	/**
235 	 * Conversion function from rte_flow to netlink attributes.
236 	 *
237 	 * @param item
238 	 *   rte_flow item to convert.
239 	 * @param data
240 	 *   Internal structure to store the conversion.
241 	 *
242 	 * @return
243 	 *   0 on success, negative value otherwise.
244 	 */
245 	int (*convert)(const struct rte_flow_item *item, void *data);
246 	/** List of possible following items.  */
247 	const enum rte_flow_item_type *const items;
248 };
249 
250 /* Graph of supported items and associated actions. */
251 static const struct tap_flow_items tap_flow_items[] = {
252 	[RTE_FLOW_ITEM_TYPE_END] = {
253 		.items = ITEMS(RTE_FLOW_ITEM_TYPE_ETH),
254 	},
255 	[RTE_FLOW_ITEM_TYPE_ETH] = {
256 		.items = ITEMS(
257 			RTE_FLOW_ITEM_TYPE_VLAN,
258 			RTE_FLOW_ITEM_TYPE_IPV4,
259 			RTE_FLOW_ITEM_TYPE_IPV6),
260 		.mask = &(const struct rte_flow_item_eth){
261 			.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
262 			.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
263 			.type = -1,
264 		},
265 		.mask_sz = sizeof(struct rte_flow_item_eth),
266 		.default_mask = &rte_flow_item_eth_mask,
267 		.convert = tap_flow_create_eth,
268 	},
269 	[RTE_FLOW_ITEM_TYPE_VLAN] = {
270 		.items = ITEMS(RTE_FLOW_ITEM_TYPE_IPV4,
271 			       RTE_FLOW_ITEM_TYPE_IPV6),
272 		.mask = &(const struct rte_flow_item_vlan){
273 			/* DEI matching is not supported */
274 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
275 			.tci = 0xffef,
276 #else
277 			.tci = 0xefff,
278 #endif
279 			.inner_type = -1,
280 		},
281 		.mask_sz = sizeof(struct rte_flow_item_vlan),
282 		.default_mask = &rte_flow_item_vlan_mask,
283 		.convert = tap_flow_create_vlan,
284 	},
285 	[RTE_FLOW_ITEM_TYPE_IPV4] = {
286 		.items = ITEMS(RTE_FLOW_ITEM_TYPE_UDP,
287 			       RTE_FLOW_ITEM_TYPE_TCP),
288 		.mask = &(const struct rte_flow_item_ipv4){
289 			.hdr = {
290 				.src_addr = -1,
291 				.dst_addr = -1,
292 				.next_proto_id = -1,
293 			},
294 		},
295 		.mask_sz = sizeof(struct rte_flow_item_ipv4),
296 		.default_mask = &rte_flow_item_ipv4_mask,
297 		.convert = tap_flow_create_ipv4,
298 	},
299 	[RTE_FLOW_ITEM_TYPE_IPV6] = {
300 		.items = ITEMS(RTE_FLOW_ITEM_TYPE_UDP,
301 			       RTE_FLOW_ITEM_TYPE_TCP),
302 		.mask = &(const struct rte_flow_item_ipv6){
303 			.hdr = {
304 				.src_addr = {
305 					"\xff\xff\xff\xff\xff\xff\xff\xff"
306 					"\xff\xff\xff\xff\xff\xff\xff\xff",
307 				},
308 				.dst_addr = {
309 					"\xff\xff\xff\xff\xff\xff\xff\xff"
310 					"\xff\xff\xff\xff\xff\xff\xff\xff",
311 				},
312 				.proto = -1,
313 			},
314 		},
315 		.mask_sz = sizeof(struct rte_flow_item_ipv6),
316 		.default_mask = &rte_flow_item_ipv6_mask,
317 		.convert = tap_flow_create_ipv6,
318 	},
319 	[RTE_FLOW_ITEM_TYPE_UDP] = {
320 		.mask = &(const struct rte_flow_item_udp){
321 			.hdr = {
322 				.src_port = -1,
323 				.dst_port = -1,
324 			},
325 		},
326 		.mask_sz = sizeof(struct rte_flow_item_udp),
327 		.default_mask = &rte_flow_item_udp_mask,
328 		.convert = tap_flow_create_udp,
329 	},
330 	[RTE_FLOW_ITEM_TYPE_TCP] = {
331 		.mask = &(const struct rte_flow_item_tcp){
332 			.hdr = {
333 				.src_port = -1,
334 				.dst_port = -1,
335 			},
336 		},
337 		.mask_sz = sizeof(struct rte_flow_item_tcp),
338 		.default_mask = &rte_flow_item_tcp_mask,
339 		.convert = tap_flow_create_tcp,
340 	},
341 };
342 
343 /*
344  *                TC rules, by growing priority
345  *
346  *        Remote netdevice                  Tap netdevice
347  * +-------------+-------------+  +-------------+-------------+
348  * |   Ingress   |   Egress    |  |   Ingress   |   Egress    |
349  * |-------------|-------------|  |-------------|-------------|
350  * |             |  \       /  |  |             |  REMOTE TX  | prio 1
351  * |             |   \     /   |  |             |   \     /   | prio 2
352  * |  EXPLICIT   |    \   /    |  |  EXPLICIT   |    \   /    |   .
353  * |             |     \ /     |  |             |     \ /     |   .
354  * |    RULES    |      X      |  |    RULES    |      X      |   .
355  * |      .      |     / \     |  |      .      |     / \     |   .
356  * |      .      |    /   \    |  |      .      |    /   \    |   .
357  * |      .      |   /     \   |  |      .      |   /     \   |   .
358  * |      .      |  /       \  |  |      .      |  /       \  |   .
359  *
360  *      ....           ....           ....           ....
361  *
362  * |      .      |  \       /  |  |      .      |  \       /  |   .
363  * |      .      |   \     /   |  |      .      |   \     /   |   .
364  * |             |    \   /    |  |             |    \   /    |
365  * |  LOCAL_MAC  |     \ /     |  |    \   /    |     \ /     | last prio - 5
366  * |   PROMISC   |      X      |  |     \ /     |      X      | last prio - 4
367  * |   ALLMULTI  |     / \     |  |      X      |     / \     | last prio - 3
368  * |  BROADCAST  |    /   \    |  |     / \     |    /   \    | last prio - 2
369  * | BROADCASTV6 |   /     \   |  |    /   \    |   /     \   | last prio - 1
370  * |     xx      |  /       \  |  |   ISOLATE   |  /       \  | last prio
371  * +-------------+-------------+  +-------------+-------------+
372  *
373  * The implicit flow rules are stored in a list in with mandatorily the last two
374  * being the ISOLATE and REMOTE_TX rules. e.g.:
375  *
376  * LOCAL_MAC -> BROADCAST -> BROADCASTV6 -> REMOTE_TX -> ISOLATE -> NULL
377  *
378  * That enables tap_flow_isolate() to remove implicit rules by popping the list
379  * head and remove it as long as it applies on the remote netdevice. The
380  * implicit rule for TX redirection is not removed, as isolate concerns only
381  * incoming traffic.
382  */
383 
384 static struct remote_rule implicit_rte_flows[TAP_REMOTE_MAX_IDX] = {
385 	[TAP_REMOTE_LOCAL_MAC] = {
386 		.attr = {
387 			.group = MAX_GROUP,
388 			.priority = PRIORITY_MASK - TAP_REMOTE_LOCAL_MAC,
389 			.ingress = 1,
390 		},
391 		.items[0] = {
392 			.type = RTE_FLOW_ITEM_TYPE_ETH,
393 			.mask =  &(const struct rte_flow_item_eth){
394 				.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
395 			},
396 		},
397 		.items[1] = {
398 			.type = RTE_FLOW_ITEM_TYPE_END,
399 		},
400 		.mirred = TCA_EGRESS_REDIR,
401 	},
402 	[TAP_REMOTE_BROADCAST] = {
403 		.attr = {
404 			.group = MAX_GROUP,
405 			.priority = PRIORITY_MASK - TAP_REMOTE_BROADCAST,
406 			.ingress = 1,
407 		},
408 		.items[0] = {
409 			.type = RTE_FLOW_ITEM_TYPE_ETH,
410 			.mask =  &(const struct rte_flow_item_eth){
411 				.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
412 			},
413 			.spec = &(const struct rte_flow_item_eth){
414 				.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
415 			},
416 		},
417 		.items[1] = {
418 			.type = RTE_FLOW_ITEM_TYPE_END,
419 		},
420 		.mirred = TCA_EGRESS_MIRROR,
421 	},
422 	[TAP_REMOTE_BROADCASTV6] = {
423 		.attr = {
424 			.group = MAX_GROUP,
425 			.priority = PRIORITY_MASK - TAP_REMOTE_BROADCASTV6,
426 			.ingress = 1,
427 		},
428 		.items[0] = {
429 			.type = RTE_FLOW_ITEM_TYPE_ETH,
430 			.mask =  &(const struct rte_flow_item_eth){
431 				.dst.addr_bytes = "\x33\x33\x00\x00\x00\x00",
432 			},
433 			.spec = &(const struct rte_flow_item_eth){
434 				.dst.addr_bytes = "\x33\x33\x00\x00\x00\x00",
435 			},
436 		},
437 		.items[1] = {
438 			.type = RTE_FLOW_ITEM_TYPE_END,
439 		},
440 		.mirred = TCA_EGRESS_MIRROR,
441 	},
442 	[TAP_REMOTE_PROMISC] = {
443 		.attr = {
444 			.group = MAX_GROUP,
445 			.priority = PRIORITY_MASK - TAP_REMOTE_PROMISC,
446 			.ingress = 1,
447 		},
448 		.items[0] = {
449 			.type = RTE_FLOW_ITEM_TYPE_VOID,
450 		},
451 		.items[1] = {
452 			.type = RTE_FLOW_ITEM_TYPE_END,
453 		},
454 		.mirred = TCA_EGRESS_MIRROR,
455 	},
456 	[TAP_REMOTE_ALLMULTI] = {
457 		.attr = {
458 			.group = MAX_GROUP,
459 			.priority = PRIORITY_MASK - TAP_REMOTE_ALLMULTI,
460 			.ingress = 1,
461 		},
462 		.items[0] = {
463 			.type = RTE_FLOW_ITEM_TYPE_ETH,
464 			.mask =  &(const struct rte_flow_item_eth){
465 				.dst.addr_bytes = "\x01\x00\x00\x00\x00\x00",
466 			},
467 			.spec = &(const struct rte_flow_item_eth){
468 				.dst.addr_bytes = "\x01\x00\x00\x00\x00\x00",
469 			},
470 		},
471 		.items[1] = {
472 			.type = RTE_FLOW_ITEM_TYPE_END,
473 		},
474 		.mirred = TCA_EGRESS_MIRROR,
475 	},
476 	[TAP_REMOTE_TX] = {
477 		.attr = {
478 			.group = 0,
479 			.priority = TAP_REMOTE_TX,
480 			.egress = 1,
481 		},
482 		.items[0] = {
483 			.type = RTE_FLOW_ITEM_TYPE_VOID,
484 		},
485 		.items[1] = {
486 			.type = RTE_FLOW_ITEM_TYPE_END,
487 		},
488 		.mirred = TCA_EGRESS_MIRROR,
489 	},
490 	[TAP_ISOLATE] = {
491 		.attr = {
492 			.group = MAX_GROUP,
493 			.priority = PRIORITY_MASK - TAP_ISOLATE,
494 			.ingress = 1,
495 		},
496 		.items[0] = {
497 			.type = RTE_FLOW_ITEM_TYPE_VOID,
498 		},
499 		.items[1] = {
500 			.type = RTE_FLOW_ITEM_TYPE_END,
501 		},
502 	},
503 };
504 
505 /**
506  * Make as much checks as possible on an Ethernet item, and if a flow is
507  * provided, fill it appropriately with Ethernet info.
508  *
509  * @param[in] item
510  *   Item specification.
511  * @param[in, out] data
512  *   Additional data structure to tell next layers we've been here.
513  *
514  * @return
515  *   0 if checks are alright, -1 otherwise.
516  */
517 static int
518 tap_flow_create_eth(const struct rte_flow_item *item, void *data)
519 {
520 	struct convert_data *info = (struct convert_data *)data;
521 	const struct rte_flow_item_eth *spec = item->spec;
522 	const struct rte_flow_item_eth *mask = item->mask;
523 	struct rte_flow *flow = info->flow;
524 	struct nlmsg *msg;
525 
526 	/* use default mask if none provided */
527 	if (!mask)
528 		mask = tap_flow_items[RTE_FLOW_ITEM_TYPE_ETH].default_mask;
529 	/* TC does not support eth_type masking. Only accept if exact match. */
530 	if (mask->type && mask->type != 0xffff)
531 		return -1;
532 	if (!spec)
533 		return 0;
534 	/* store eth_type for consistency if ipv4/6 pattern item comes next */
535 	if (spec->type & mask->type)
536 		info->eth_type = spec->type;
537 	if (!flow)
538 		return 0;
539 	msg = &flow->msg;
540 	if (!is_zero_ether_addr(&mask->dst)) {
541 		tap_nlattr_add(&msg->nh, TCA_FLOWER_KEY_ETH_DST, ETHER_ADDR_LEN,
542 			   &spec->dst.addr_bytes);
543 		tap_nlattr_add(&msg->nh,
544 			   TCA_FLOWER_KEY_ETH_DST_MASK, ETHER_ADDR_LEN,
545 			   &mask->dst.addr_bytes);
546 	}
547 	if (!is_zero_ether_addr(&mask->src)) {
548 		tap_nlattr_add(&msg->nh, TCA_FLOWER_KEY_ETH_SRC, ETHER_ADDR_LEN,
549 			   &spec->src.addr_bytes);
550 		tap_nlattr_add(&msg->nh,
551 			   TCA_FLOWER_KEY_ETH_SRC_MASK, ETHER_ADDR_LEN,
552 			   &mask->src.addr_bytes);
553 	}
554 	return 0;
555 }
556 
557 /**
558  * Make as much checks as possible on a VLAN item, and if a flow is provided,
559  * fill it appropriately with VLAN info.
560  *
561  * @param[in] item
562  *   Item specification.
563  * @param[in, out] data
564  *   Additional data structure to tell next layers we've been here.
565  *
566  * @return
567  *   0 if checks are alright, -1 otherwise.
568  */
569 static int
570 tap_flow_create_vlan(const struct rte_flow_item *item, void *data)
571 {
572 	struct convert_data *info = (struct convert_data *)data;
573 	const struct rte_flow_item_vlan *spec = item->spec;
574 	const struct rte_flow_item_vlan *mask = item->mask;
575 	struct rte_flow *flow = info->flow;
576 	struct nlmsg *msg;
577 
578 	/* use default mask if none provided */
579 	if (!mask)
580 		mask = tap_flow_items[RTE_FLOW_ITEM_TYPE_VLAN].default_mask;
581 	/* Outer TPID cannot be matched. */
582 	if (info->eth_type)
583 		return -1;
584 	/* Double-tagging not supported. */
585 	if (info->vlan)
586 		return -1;
587 	info->vlan = 1;
588 	if (mask->inner_type) {
589 		/* TC does not support partial eth_type masking */
590 		if (mask->inner_type != RTE_BE16(0xffff))
591 			return -1;
592 		info->eth_type = spec->inner_type;
593 	}
594 	if (!flow)
595 		return 0;
596 	msg = &flow->msg;
597 	msg->t.tcm_info = TC_H_MAKE(msg->t.tcm_info, htons(ETH_P_8021Q));
598 #define VLAN_PRIO(tci) ((tci) >> 13)
599 #define VLAN_ID(tci) ((tci) & 0xfff)
600 	if (!spec)
601 		return 0;
602 	if (spec->tci) {
603 		uint16_t tci = ntohs(spec->tci) & mask->tci;
604 		uint16_t prio = VLAN_PRIO(tci);
605 		uint8_t vid = VLAN_ID(tci);
606 
607 		if (prio)
608 			tap_nlattr_add8(&msg->nh,
609 					TCA_FLOWER_KEY_VLAN_PRIO, prio);
610 		if (vid)
611 			tap_nlattr_add16(&msg->nh,
612 					 TCA_FLOWER_KEY_VLAN_ID, vid);
613 	}
614 	return 0;
615 }
616 
617 /**
618  * Make as much checks as possible on an IPv4 item, and if a flow is provided,
619  * fill it appropriately with IPv4 info.
620  *
621  * @param[in] item
622  *   Item specification.
623  * @param[in, out] data
624  *   Additional data structure to tell next layers we've been here.
625  *
626  * @return
627  *   0 if checks are alright, -1 otherwise.
628  */
629 static int
630 tap_flow_create_ipv4(const struct rte_flow_item *item, void *data)
631 {
632 	struct convert_data *info = (struct convert_data *)data;
633 	const struct rte_flow_item_ipv4 *spec = item->spec;
634 	const struct rte_flow_item_ipv4 *mask = item->mask;
635 	struct rte_flow *flow = info->flow;
636 	struct nlmsg *msg;
637 
638 	/* use default mask if none provided */
639 	if (!mask)
640 		mask = tap_flow_items[RTE_FLOW_ITEM_TYPE_IPV4].default_mask;
641 	/* check that previous eth type is compatible with ipv4 */
642 	if (info->eth_type && info->eth_type != htons(ETH_P_IP))
643 		return -1;
644 	/* store ip_proto for consistency if udp/tcp pattern item comes next */
645 	if (spec)
646 		info->ip_proto = spec->hdr.next_proto_id;
647 	if (!flow)
648 		return 0;
649 	msg = &flow->msg;
650 	if (!info->eth_type)
651 		info->eth_type = htons(ETH_P_IP);
652 	if (!spec)
653 		return 0;
654 	if (mask->hdr.dst_addr) {
655 		tap_nlattr_add32(&msg->nh, TCA_FLOWER_KEY_IPV4_DST,
656 			     spec->hdr.dst_addr);
657 		tap_nlattr_add32(&msg->nh, TCA_FLOWER_KEY_IPV4_DST_MASK,
658 			     mask->hdr.dst_addr);
659 	}
660 	if (mask->hdr.src_addr) {
661 		tap_nlattr_add32(&msg->nh, TCA_FLOWER_KEY_IPV4_SRC,
662 			     spec->hdr.src_addr);
663 		tap_nlattr_add32(&msg->nh, TCA_FLOWER_KEY_IPV4_SRC_MASK,
664 			     mask->hdr.src_addr);
665 	}
666 	if (spec->hdr.next_proto_id)
667 		tap_nlattr_add8(&msg->nh, TCA_FLOWER_KEY_IP_PROTO,
668 			    spec->hdr.next_proto_id);
669 	return 0;
670 }
671 
672 /**
673  * Make as much checks as possible on an IPv6 item, and if a flow is provided,
674  * fill it appropriately with IPv6 info.
675  *
676  * @param[in] item
677  *   Item specification.
678  * @param[in, out] data
679  *   Additional data structure to tell next layers we've been here.
680  *
681  * @return
682  *   0 if checks are alright, -1 otherwise.
683  */
684 static int
685 tap_flow_create_ipv6(const struct rte_flow_item *item, void *data)
686 {
687 	struct convert_data *info = (struct convert_data *)data;
688 	const struct rte_flow_item_ipv6 *spec = item->spec;
689 	const struct rte_flow_item_ipv6 *mask = item->mask;
690 	struct rte_flow *flow = info->flow;
691 	uint8_t empty_addr[16] = { 0 };
692 	struct nlmsg *msg;
693 
694 	/* use default mask if none provided */
695 	if (!mask)
696 		mask = tap_flow_items[RTE_FLOW_ITEM_TYPE_IPV6].default_mask;
697 	/* check that previous eth type is compatible with ipv6 */
698 	if (info->eth_type && info->eth_type != htons(ETH_P_IPV6))
699 		return -1;
700 	/* store ip_proto for consistency if udp/tcp pattern item comes next */
701 	if (spec)
702 		info->ip_proto = spec->hdr.proto;
703 	if (!flow)
704 		return 0;
705 	msg = &flow->msg;
706 	if (!info->eth_type)
707 		info->eth_type = htons(ETH_P_IPV6);
708 	if (!spec)
709 		return 0;
710 	if (memcmp(mask->hdr.dst_addr, empty_addr, 16)) {
711 		tap_nlattr_add(&msg->nh, TCA_FLOWER_KEY_IPV6_DST,
712 			   sizeof(spec->hdr.dst_addr), &spec->hdr.dst_addr);
713 		tap_nlattr_add(&msg->nh, TCA_FLOWER_KEY_IPV6_DST_MASK,
714 			   sizeof(mask->hdr.dst_addr), &mask->hdr.dst_addr);
715 	}
716 	if (memcmp(mask->hdr.src_addr, empty_addr, 16)) {
717 		tap_nlattr_add(&msg->nh, TCA_FLOWER_KEY_IPV6_SRC,
718 			   sizeof(spec->hdr.src_addr), &spec->hdr.src_addr);
719 		tap_nlattr_add(&msg->nh, TCA_FLOWER_KEY_IPV6_SRC_MASK,
720 			   sizeof(mask->hdr.src_addr), &mask->hdr.src_addr);
721 	}
722 	if (spec->hdr.proto)
723 		tap_nlattr_add8(&msg->nh,
724 				TCA_FLOWER_KEY_IP_PROTO, spec->hdr.proto);
725 	return 0;
726 }
727 
728 /**
729  * Make as much checks as possible on a UDP item, and if a flow is provided,
730  * fill it appropriately with UDP info.
731  *
732  * @param[in] item
733  *   Item specification.
734  * @param[in, out] data
735  *   Additional data structure to tell next layers we've been here.
736  *
737  * @return
738  *   0 if checks are alright, -1 otherwise.
739  */
740 static int
741 tap_flow_create_udp(const struct rte_flow_item *item, void *data)
742 {
743 	struct convert_data *info = (struct convert_data *)data;
744 	const struct rte_flow_item_udp *spec = item->spec;
745 	const struct rte_flow_item_udp *mask = item->mask;
746 	struct rte_flow *flow = info->flow;
747 	struct nlmsg *msg;
748 
749 	/* use default mask if none provided */
750 	if (!mask)
751 		mask = tap_flow_items[RTE_FLOW_ITEM_TYPE_UDP].default_mask;
752 	/* check that previous ip_proto is compatible with udp */
753 	if (info->ip_proto && info->ip_proto != IPPROTO_UDP)
754 		return -1;
755 	/* TC does not support UDP port masking. Only accept if exact match. */
756 	if ((mask->hdr.src_port && mask->hdr.src_port != 0xffff) ||
757 	    (mask->hdr.dst_port && mask->hdr.dst_port != 0xffff))
758 		return -1;
759 	if (!flow)
760 		return 0;
761 	msg = &flow->msg;
762 	tap_nlattr_add8(&msg->nh, TCA_FLOWER_KEY_IP_PROTO, IPPROTO_UDP);
763 	if (!spec)
764 		return 0;
765 	if (mask->hdr.dst_port)
766 		tap_nlattr_add16(&msg->nh, TCA_FLOWER_KEY_UDP_DST,
767 			     spec->hdr.dst_port);
768 	if (mask->hdr.src_port)
769 		tap_nlattr_add16(&msg->nh, TCA_FLOWER_KEY_UDP_SRC,
770 			     spec->hdr.src_port);
771 	return 0;
772 }
773 
774 /**
775  * Make as much checks as possible on a TCP item, and if a flow is provided,
776  * fill it appropriately with TCP info.
777  *
778  * @param[in] item
779  *   Item specification.
780  * @param[in, out] data
781  *   Additional data structure to tell next layers we've been here.
782  *
783  * @return
784  *   0 if checks are alright, -1 otherwise.
785  */
786 static int
787 tap_flow_create_tcp(const struct rte_flow_item *item, void *data)
788 {
789 	struct convert_data *info = (struct convert_data *)data;
790 	const struct rte_flow_item_tcp *spec = item->spec;
791 	const struct rte_flow_item_tcp *mask = item->mask;
792 	struct rte_flow *flow = info->flow;
793 	struct nlmsg *msg;
794 
795 	/* use default mask if none provided */
796 	if (!mask)
797 		mask = tap_flow_items[RTE_FLOW_ITEM_TYPE_TCP].default_mask;
798 	/* check that previous ip_proto is compatible with tcp */
799 	if (info->ip_proto && info->ip_proto != IPPROTO_TCP)
800 		return -1;
801 	/* TC does not support TCP port masking. Only accept if exact match. */
802 	if ((mask->hdr.src_port && mask->hdr.src_port != 0xffff) ||
803 	    (mask->hdr.dst_port && mask->hdr.dst_port != 0xffff))
804 		return -1;
805 	if (!flow)
806 		return 0;
807 	msg = &flow->msg;
808 	tap_nlattr_add8(&msg->nh, TCA_FLOWER_KEY_IP_PROTO, IPPROTO_TCP);
809 	if (!spec)
810 		return 0;
811 	if (mask->hdr.dst_port)
812 		tap_nlattr_add16(&msg->nh, TCA_FLOWER_KEY_TCP_DST,
813 			     spec->hdr.dst_port);
814 	if (mask->hdr.src_port)
815 		tap_nlattr_add16(&msg->nh, TCA_FLOWER_KEY_TCP_SRC,
816 			     spec->hdr.src_port);
817 	return 0;
818 }
819 
820 /**
821  * Check support for a given item.
822  *
823  * @param[in] item
824  *   Item specification.
825  * @param size
826  *   Bit-Mask size in bytes.
827  * @param[in] supported_mask
828  *   Bit-mask covering supported fields to compare with spec, last and mask in
829  *   \item.
830  * @param[in] default_mask
831  *   Bit-mask default mask if none is provided in \item.
832  *
833  * @return
834  *   0 on success.
835  */
836 static int
837 tap_flow_item_validate(const struct rte_flow_item *item,
838 		       unsigned int size,
839 		       const uint8_t *supported_mask,
840 		       const uint8_t *default_mask)
841 {
842 	int ret = 0;
843 
844 	/* An empty layer is allowed, as long as all fields are NULL */
845 	if (!item->spec && (item->mask || item->last))
846 		return -1;
847 	/* Is the item spec compatible with what the NIC supports? */
848 	if (item->spec && !item->mask) {
849 		unsigned int i;
850 		const uint8_t *spec = item->spec;
851 
852 		for (i = 0; i < size; ++i)
853 			if ((spec[i] | supported_mask[i]) != supported_mask[i])
854 				return -1;
855 		/* Is the default mask compatible with what the NIC supports? */
856 		for (i = 0; i < size; i++)
857 			if ((default_mask[i] | supported_mask[i]) !=
858 			    supported_mask[i])
859 				return -1;
860 	}
861 	/* Is the item last compatible with what the NIC supports? */
862 	if (item->last && !item->mask) {
863 		unsigned int i;
864 		const uint8_t *spec = item->last;
865 
866 		for (i = 0; i < size; ++i)
867 			if ((spec[i] | supported_mask[i]) != supported_mask[i])
868 				return -1;
869 	}
870 	/* Is the item mask compatible with what the NIC supports? */
871 	if (item->mask) {
872 		unsigned int i;
873 		const uint8_t *spec = item->mask;
874 
875 		for (i = 0; i < size; ++i)
876 			if ((spec[i] | supported_mask[i]) != supported_mask[i])
877 				return -1;
878 	}
879 	/**
880 	 * Once masked, Are item spec and item last equal?
881 	 * TC does not support range so anything else is invalid.
882 	 */
883 	if (item->spec && item->last) {
884 		uint8_t spec[size];
885 		uint8_t last[size];
886 		const uint8_t *apply = default_mask;
887 		unsigned int i;
888 
889 		if (item->mask)
890 			apply = item->mask;
891 		for (i = 0; i < size; ++i) {
892 			spec[i] = ((const uint8_t *)item->spec)[i] & apply[i];
893 			last[i] = ((const uint8_t *)item->last)[i] & apply[i];
894 		}
895 		ret = memcmp(spec, last, size);
896 	}
897 	return ret;
898 }
899 
900 /**
901  * Configure the kernel with a TC action and its configured parameters
902  * Handled actions: "gact", "mirred", "skbedit", "bpf"
903  *
904  * @param[in] flow
905  *   Pointer to rte flow containing the netlink message
906  *
907  * @param[in, out] act_index
908  *   Pointer to action sequence number in the TC command
909  *
910  * @param[in] adata
911  *  Pointer to struct holding the action parameters
912  *
913  * @return
914  *   -1 on failure, 0 on success
915  */
916 static int
917 add_action(struct rte_flow *flow, size_t *act_index, struct action_data *adata)
918 {
919 	struct nlmsg *msg = &flow->msg;
920 
921 	if (tap_nlattr_nested_start(msg, (*act_index)++) < 0)
922 		return -1;
923 
924 	tap_nlattr_add(&msg->nh, TCA_ACT_KIND,
925 				strlen(adata->id) + 1, adata->id);
926 	if (tap_nlattr_nested_start(msg, TCA_ACT_OPTIONS) < 0)
927 		return -1;
928 	if (strcmp("gact", adata->id) == 0) {
929 		tap_nlattr_add(&msg->nh, TCA_GACT_PARMS, sizeof(adata->gact),
930 			   &adata->gact);
931 	} else if (strcmp("mirred", adata->id) == 0) {
932 		if (adata->mirred.eaction == TCA_EGRESS_MIRROR)
933 			adata->mirred.action = TC_ACT_PIPE;
934 		else /* REDIRECT */
935 			adata->mirred.action = TC_ACT_STOLEN;
936 		tap_nlattr_add(&msg->nh, TCA_MIRRED_PARMS,
937 			   sizeof(adata->mirred),
938 			   &adata->mirred);
939 	} else if (strcmp("skbedit", adata->id) == 0) {
940 		tap_nlattr_add(&msg->nh, TCA_SKBEDIT_PARMS,
941 			   sizeof(adata->skbedit.skbedit),
942 			   &adata->skbedit.skbedit);
943 		tap_nlattr_add16(&msg->nh, TCA_SKBEDIT_QUEUE_MAPPING,
944 			     adata->skbedit.queue);
945 	} else if (strcmp("bpf", adata->id) == 0) {
946 		tap_nlattr_add32(&msg->nh, TCA_ACT_BPF_FD, adata->bpf.bpf_fd);
947 		tap_nlattr_add(&msg->nh, TCA_ACT_BPF_NAME,
948 			   strlen(adata->bpf.annotation) + 1,
949 			   adata->bpf.annotation);
950 		tap_nlattr_add(&msg->nh, TCA_ACT_BPF_PARMS,
951 			   sizeof(adata->bpf.bpf),
952 			   &adata->bpf.bpf);
953 	} else {
954 		return -1;
955 	}
956 	tap_nlattr_nested_finish(msg); /* nested TCA_ACT_OPTIONS */
957 	tap_nlattr_nested_finish(msg); /* nested act_index */
958 	return 0;
959 }
960 
961 /**
962  * Helper function to send a serie of TC actions to the kernel
963  *
964  * @param[in] flow
965  *   Pointer to rte flow containing the netlink message
966  *
967  * @param[in] nb_actions
968  *   Number of actions in an array of action structs
969  *
970  * @param[in] data
971  *   Pointer to an array of action structs
972  *
973  * @param[in] classifier_actions
974  *   The classifier on behave of which the actions are configured
975  *
976  * @return
977  *   -1 on failure, 0 on success
978  */
979 static int
980 add_actions(struct rte_flow *flow, int nb_actions, struct action_data *data,
981 	    int classifier_action)
982 {
983 	struct nlmsg *msg = &flow->msg;
984 	size_t act_index = 1;
985 	int i;
986 
987 	if (tap_nlattr_nested_start(msg, classifier_action) < 0)
988 		return -1;
989 	for (i = 0; i < nb_actions; i++)
990 		if (add_action(flow, &act_index, data + i) < 0)
991 			return -1;
992 	tap_nlattr_nested_finish(msg); /* nested TCA_FLOWER_ACT */
993 	return 0;
994 }
995 
996 /**
997  * Validate a flow supported by TC.
998  * If flow param is not NULL, then also fill the netlink message inside.
999  *
1000  * @param pmd
1001  *   Pointer to private structure.
1002  * @param[in] attr
1003  *   Flow rule attributes.
1004  * @param[in] pattern
1005  *   Pattern specification (list terminated by the END pattern item).
1006  * @param[in] actions
1007  *   Associated actions (list terminated by the END action).
1008  * @param[out] error
1009  *   Perform verbose error reporting if not NULL.
1010  * @param[in, out] flow
1011  *   Flow structure to update.
1012  * @param[in] mirred
1013  *   If set to TCA_EGRESS_REDIR, provided actions will be replaced with a
1014  *   redirection to the tap netdevice, and the TC rule will be configured
1015  *   on the remote netdevice in pmd.
1016  *   If set to TCA_EGRESS_MIRROR, provided actions will be replaced with a
1017  *   mirroring to the tap netdevice, and the TC rule will be configured
1018  *   on the remote netdevice in pmd. Matching packets will thus be duplicated.
1019  *   If set to 0, the standard behavior is to be used: set correct actions for
1020  *   the TC rule, and apply it on the tap netdevice.
1021  *
1022  * @return
1023  *   0 on success, a negative errno value otherwise and rte_errno is set.
1024  */
1025 static int
1026 priv_flow_process(struct pmd_internals *pmd,
1027 		  const struct rte_flow_attr *attr,
1028 		  const struct rte_flow_item items[],
1029 		  const struct rte_flow_action actions[],
1030 		  struct rte_flow_error *error,
1031 		  struct rte_flow *flow,
1032 		  int mirred)
1033 {
1034 	const struct tap_flow_items *cur_item = tap_flow_items;
1035 	struct convert_data data = {
1036 		.eth_type = 0,
1037 		.ip_proto = 0,
1038 		.flow = flow,
1039 	};
1040 	int action = 0; /* Only one action authorized for now */
1041 
1042 	if (attr->transfer) {
1043 		rte_flow_error_set(
1044 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1045 			NULL, "transfer is not supported");
1046 		return -rte_errno;
1047 	}
1048 	if (attr->group > MAX_GROUP) {
1049 		rte_flow_error_set(
1050 			error, EINVAL, RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1051 			NULL, "group value too big: cannot exceed 15");
1052 		return -rte_errno;
1053 	}
1054 	if (attr->priority > MAX_PRIORITY) {
1055 		rte_flow_error_set(
1056 			error, EINVAL, RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1057 			NULL, "priority value too big");
1058 		return -rte_errno;
1059 	} else if (flow) {
1060 		uint16_t group = attr->group << GROUP_SHIFT;
1061 		uint16_t prio = group | (attr->priority +
1062 				RSS_PRIORITY_OFFSET + PRIORITY_OFFSET);
1063 		flow->msg.t.tcm_info = TC_H_MAKE(prio << 16,
1064 						 flow->msg.t.tcm_info);
1065 	}
1066 	if (flow) {
1067 		if (mirred) {
1068 			/*
1069 			 * If attr->ingress, the rule applies on remote ingress
1070 			 * to match incoming packets
1071 			 * If attr->egress, the rule applies on tap ingress (as
1072 			 * seen from the kernel) to deal with packets going out
1073 			 * from the DPDK app.
1074 			 */
1075 			flow->msg.t.tcm_parent = TC_H_MAKE(TC_H_INGRESS, 0);
1076 		} else {
1077 			/* Standard rule on tap egress (kernel standpoint). */
1078 			flow->msg.t.tcm_parent =
1079 				TC_H_MAKE(MULTIQ_MAJOR_HANDLE, 0);
1080 		}
1081 		/* use flower filter type */
1082 		tap_nlattr_add(&flow->msg.nh, TCA_KIND, sizeof("flower"), "flower");
1083 		if (tap_nlattr_nested_start(&flow->msg, TCA_OPTIONS) < 0)
1084 			goto exit_item_not_supported;
1085 	}
1086 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; ++items) {
1087 		const struct tap_flow_items *token = NULL;
1088 		unsigned int i;
1089 		int err = 0;
1090 
1091 		if (items->type == RTE_FLOW_ITEM_TYPE_VOID)
1092 			continue;
1093 		for (i = 0;
1094 		     cur_item->items &&
1095 		     cur_item->items[i] != RTE_FLOW_ITEM_TYPE_END;
1096 		     ++i) {
1097 			if (cur_item->items[i] == items->type) {
1098 				token = &tap_flow_items[items->type];
1099 				break;
1100 			}
1101 		}
1102 		if (!token)
1103 			goto exit_item_not_supported;
1104 		cur_item = token;
1105 		err = tap_flow_item_validate(
1106 			items, cur_item->mask_sz,
1107 			(const uint8_t *)cur_item->mask,
1108 			(const uint8_t *)cur_item->default_mask);
1109 		if (err)
1110 			goto exit_item_not_supported;
1111 		if (flow && cur_item->convert) {
1112 			err = cur_item->convert(items, &data);
1113 			if (err)
1114 				goto exit_item_not_supported;
1115 		}
1116 	}
1117 	if (flow) {
1118 		if (data.vlan) {
1119 			tap_nlattr_add16(&flow->msg.nh, TCA_FLOWER_KEY_ETH_TYPE,
1120 				     htons(ETH_P_8021Q));
1121 			tap_nlattr_add16(&flow->msg.nh,
1122 				     TCA_FLOWER_KEY_VLAN_ETH_TYPE,
1123 				     data.eth_type ?
1124 				     data.eth_type : htons(ETH_P_ALL));
1125 		} else if (data.eth_type) {
1126 			tap_nlattr_add16(&flow->msg.nh, TCA_FLOWER_KEY_ETH_TYPE,
1127 				     data.eth_type);
1128 		}
1129 	}
1130 	if (mirred && flow) {
1131 		struct action_data adata = {
1132 			.id = "mirred",
1133 			.mirred = {
1134 				.eaction = mirred,
1135 			},
1136 		};
1137 
1138 		/*
1139 		 * If attr->egress && mirred, then this is a special
1140 		 * case where the rule must be applied on the tap, to
1141 		 * redirect packets coming from the DPDK App, out
1142 		 * through the remote netdevice.
1143 		 */
1144 		adata.mirred.ifindex = attr->ingress ? pmd->if_index :
1145 			pmd->remote_if_index;
1146 		if (mirred == TCA_EGRESS_MIRROR)
1147 			adata.mirred.action = TC_ACT_PIPE;
1148 		else
1149 			adata.mirred.action = TC_ACT_STOLEN;
1150 		if (add_actions(flow, 1, &adata, TCA_FLOWER_ACT) < 0)
1151 			goto exit_action_not_supported;
1152 		else
1153 			goto end;
1154 	}
1155 actions:
1156 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; ++actions) {
1157 		int err = 0;
1158 
1159 		if (actions->type == RTE_FLOW_ACTION_TYPE_VOID) {
1160 			continue;
1161 		} else if (actions->type == RTE_FLOW_ACTION_TYPE_DROP) {
1162 			if (action)
1163 				goto exit_action_not_supported;
1164 			action = 1;
1165 			if (flow) {
1166 				struct action_data adata = {
1167 					.id = "gact",
1168 					.gact = {
1169 						.action = TC_ACT_SHOT,
1170 					},
1171 				};
1172 
1173 				err = add_actions(flow, 1, &adata,
1174 						  TCA_FLOWER_ACT);
1175 			}
1176 		} else if (actions->type == RTE_FLOW_ACTION_TYPE_PASSTHRU) {
1177 			if (action)
1178 				goto exit_action_not_supported;
1179 			action = 1;
1180 			if (flow) {
1181 				struct action_data adata = {
1182 					.id = "gact",
1183 					.gact = {
1184 						/* continue */
1185 						.action = TC_ACT_UNSPEC,
1186 					},
1187 				};
1188 
1189 				err = add_actions(flow, 1, &adata,
1190 						  TCA_FLOWER_ACT);
1191 			}
1192 		} else if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
1193 			const struct rte_flow_action_queue *queue =
1194 				(const struct rte_flow_action_queue *)
1195 				actions->conf;
1196 
1197 			if (action)
1198 				goto exit_action_not_supported;
1199 			action = 1;
1200 			if (!queue ||
1201 			    (queue->index > pmd->dev->data->nb_rx_queues - 1))
1202 				goto exit_action_not_supported;
1203 			if (flow) {
1204 				struct action_data adata = {
1205 					.id = "skbedit",
1206 					.skbedit = {
1207 						.skbedit = {
1208 							.action = TC_ACT_PIPE,
1209 						},
1210 						.queue = queue->index,
1211 					},
1212 				};
1213 
1214 				err = add_actions(flow, 1, &adata,
1215 					TCA_FLOWER_ACT);
1216 			}
1217 		} else if (actions->type == RTE_FLOW_ACTION_TYPE_RSS) {
1218 			const struct rte_flow_action_rss *rss =
1219 				(const struct rte_flow_action_rss *)
1220 				actions->conf;
1221 
1222 			if (action++)
1223 				goto exit_action_not_supported;
1224 
1225 			if (!pmd->rss_enabled) {
1226 				err = rss_enable(pmd, attr, error);
1227 				if (err)
1228 					goto exit_action_not_supported;
1229 			}
1230 			if (flow)
1231 				err = rss_add_actions(flow, pmd, rss, error);
1232 		} else {
1233 			goto exit_action_not_supported;
1234 		}
1235 		if (err)
1236 			goto exit_action_not_supported;
1237 	}
1238 	/* When fate is unknown, drop traffic. */
1239 	if (!action) {
1240 		static const struct rte_flow_action drop[] = {
1241 			{ .type = RTE_FLOW_ACTION_TYPE_DROP, },
1242 			{ .type = RTE_FLOW_ACTION_TYPE_END, },
1243 		};
1244 
1245 		actions = drop;
1246 		goto actions;
1247 	}
1248 end:
1249 	if (flow)
1250 		tap_nlattr_nested_finish(&flow->msg); /* nested TCA_OPTIONS */
1251 	return 0;
1252 exit_item_not_supported:
1253 	rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1254 			   items, "item not supported");
1255 	return -rte_errno;
1256 exit_action_not_supported:
1257 	rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
1258 			   actions, "action not supported");
1259 	return -rte_errno;
1260 }
1261 
1262 
1263 
1264 /**
1265  * Validate a flow.
1266  *
1267  * @see rte_flow_validate()
1268  * @see rte_flow_ops
1269  */
1270 static int
1271 tap_flow_validate(struct rte_eth_dev *dev,
1272 		  const struct rte_flow_attr *attr,
1273 		  const struct rte_flow_item items[],
1274 		  const struct rte_flow_action actions[],
1275 		  struct rte_flow_error *error)
1276 {
1277 	struct pmd_internals *pmd = dev->data->dev_private;
1278 
1279 	return priv_flow_process(pmd, attr, items, actions, error, NULL, 0);
1280 }
1281 
1282 /**
1283  * Set a unique handle in a flow.
1284  *
1285  * The kernel supports TC rules with equal priority, as long as they use the
1286  * same matching fields (e.g.: dst mac and ipv4) with different values (and
1287  * full mask to ensure no collision is possible).
1288  * In those rules, the handle (uint32_t) is the part that would identify
1289  * specifically each rule.
1290  *
1291  * On 32-bit architectures, the handle can simply be the flow's pointer address.
1292  * On 64-bit architectures, we rely on jhash(flow) to find a (sufficiently)
1293  * unique handle.
1294  *
1295  * @param[in, out] flow
1296  *   The flow that needs its handle set.
1297  */
1298 static void
1299 tap_flow_set_handle(struct rte_flow *flow)
1300 {
1301 	uint32_t handle = 0;
1302 
1303 	if (sizeof(flow) > 4)
1304 		handle = rte_jhash(&flow, sizeof(flow), 1);
1305 	else
1306 		handle = (uintptr_t)flow;
1307 	/* must be at least 1 to avoid letting the kernel choose one for us */
1308 	if (!handle)
1309 		handle = 1;
1310 	flow->msg.t.tcm_handle = handle;
1311 }
1312 
1313 /**
1314  * Free the flow opened file descriptors and allocated memory
1315  *
1316  * @param[in] flow
1317  *   Pointer to the flow to free
1318  *
1319  */
1320 static void
1321 tap_flow_free(struct pmd_internals *pmd, struct rte_flow *flow)
1322 {
1323 	int i;
1324 
1325 	if (!flow)
1326 		return;
1327 
1328 	if (pmd->rss_enabled) {
1329 		/* Close flow BPF file descriptors */
1330 		for (i = 0; i < SEC_MAX; i++)
1331 			if (flow->bpf_fd[i] != 0) {
1332 				close(flow->bpf_fd[i]);
1333 				flow->bpf_fd[i] = 0;
1334 			}
1335 
1336 		/* Release the map key for this RSS rule */
1337 		bpf_rss_key(KEY_CMD_RELEASE, &flow->key_idx);
1338 		flow->key_idx = 0;
1339 	}
1340 
1341 	/* Free flow allocated memory */
1342 	rte_free(flow);
1343 }
1344 
1345 /**
1346  * Create a flow.
1347  *
1348  * @see rte_flow_create()
1349  * @see rte_flow_ops
1350  */
1351 static struct rte_flow *
1352 tap_flow_create(struct rte_eth_dev *dev,
1353 		const struct rte_flow_attr *attr,
1354 		const struct rte_flow_item items[],
1355 		const struct rte_flow_action actions[],
1356 		struct rte_flow_error *error)
1357 {
1358 	struct pmd_internals *pmd = dev->data->dev_private;
1359 	struct rte_flow *remote_flow = NULL;
1360 	struct rte_flow *flow = NULL;
1361 	struct nlmsg *msg = NULL;
1362 	int err;
1363 
1364 	if (!pmd->if_index) {
1365 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE,
1366 				   NULL,
1367 				   "can't create rule, ifindex not found");
1368 		goto fail;
1369 	}
1370 	/*
1371 	 * No rules configured through standard rte_flow should be set on the
1372 	 * priorities used by implicit rules.
1373 	 */
1374 	if ((attr->group == MAX_GROUP) &&
1375 	    attr->priority > (MAX_PRIORITY - TAP_REMOTE_MAX_IDX)) {
1376 		rte_flow_error_set(
1377 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1378 			NULL, "priority value too big");
1379 		goto fail;
1380 	}
1381 	flow = rte_malloc(__func__, sizeof(struct rte_flow), 0);
1382 	if (!flow) {
1383 		rte_flow_error_set(error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1384 				   NULL, "cannot allocate memory for rte_flow");
1385 		goto fail;
1386 	}
1387 	msg = &flow->msg;
1388 	tc_init_msg(msg, pmd->if_index, RTM_NEWTFILTER,
1389 		    NLM_F_REQUEST | NLM_F_ACK | NLM_F_EXCL | NLM_F_CREATE);
1390 	msg->t.tcm_info = TC_H_MAKE(0, htons(ETH_P_ALL));
1391 	tap_flow_set_handle(flow);
1392 	if (priv_flow_process(pmd, attr, items, actions, error, flow, 0))
1393 		goto fail;
1394 	err = tap_nl_send(pmd->nlsk_fd, &msg->nh);
1395 	if (err < 0) {
1396 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE,
1397 				   NULL, "couldn't send request to kernel");
1398 		goto fail;
1399 	}
1400 	err = tap_nl_recv_ack(pmd->nlsk_fd);
1401 	if (err < 0) {
1402 		TAP_LOG(ERR,
1403 			"Kernel refused TC filter rule creation (%d): %s",
1404 			errno, strerror(errno));
1405 		rte_flow_error_set(error, EEXIST, RTE_FLOW_ERROR_TYPE_HANDLE,
1406 				   NULL,
1407 				   "overlapping rules or Kernel too old for flower support");
1408 		goto fail;
1409 	}
1410 	LIST_INSERT_HEAD(&pmd->flows, flow, next);
1411 	/**
1412 	 * If a remote device is configured, a TC rule with identical items for
1413 	 * matching must be set on that device, with a single action: redirect
1414 	 * to the local pmd->if_index.
1415 	 */
1416 	if (pmd->remote_if_index) {
1417 		remote_flow = rte_malloc(__func__, sizeof(struct rte_flow), 0);
1418 		if (!remote_flow) {
1419 			rte_flow_error_set(
1420 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1421 				"cannot allocate memory for rte_flow");
1422 			goto fail;
1423 		}
1424 		msg = &remote_flow->msg;
1425 		/* set the rule if_index for the remote netdevice */
1426 		tc_init_msg(
1427 			msg, pmd->remote_if_index, RTM_NEWTFILTER,
1428 			NLM_F_REQUEST | NLM_F_ACK | NLM_F_EXCL | NLM_F_CREATE);
1429 		msg->t.tcm_info = TC_H_MAKE(0, htons(ETH_P_ALL));
1430 		tap_flow_set_handle(remote_flow);
1431 		if (priv_flow_process(pmd, attr, items, NULL,
1432 				      error, remote_flow, TCA_EGRESS_REDIR)) {
1433 			rte_flow_error_set(
1434 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1435 				NULL, "rte flow rule validation failed");
1436 			goto fail;
1437 		}
1438 		err = tap_nl_send(pmd->nlsk_fd, &msg->nh);
1439 		if (err < 0) {
1440 			rte_flow_error_set(
1441 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1442 				NULL, "Failure sending nl request");
1443 			goto fail;
1444 		}
1445 		err = tap_nl_recv_ack(pmd->nlsk_fd);
1446 		if (err < 0) {
1447 			TAP_LOG(ERR,
1448 				"Kernel refused TC filter rule creation (%d): %s",
1449 				errno, strerror(errno));
1450 			rte_flow_error_set(
1451 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1452 				NULL,
1453 				"overlapping rules or Kernel too old for flower support");
1454 			goto fail;
1455 		}
1456 		flow->remote_flow = remote_flow;
1457 	}
1458 	return flow;
1459 fail:
1460 	if (remote_flow)
1461 		rte_free(remote_flow);
1462 	if (flow)
1463 		tap_flow_free(pmd, flow);
1464 	return NULL;
1465 }
1466 
1467 /**
1468  * Destroy a flow using pointer to pmd_internal.
1469  *
1470  * @param[in, out] pmd
1471  *   Pointer to private structure.
1472  * @param[in] flow
1473  *   Pointer to the flow to destroy.
1474  * @param[in, out] error
1475  *   Pointer to the flow error handler
1476  *
1477  * @return 0 if the flow could be destroyed, -1 otherwise.
1478  */
1479 static int
1480 tap_flow_destroy_pmd(struct pmd_internals *pmd,
1481 		     struct rte_flow *flow,
1482 		     struct rte_flow_error *error)
1483 {
1484 	struct rte_flow *remote_flow = flow->remote_flow;
1485 	int ret = 0;
1486 
1487 	LIST_REMOVE(flow, next);
1488 	flow->msg.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
1489 	flow->msg.nh.nlmsg_type = RTM_DELTFILTER;
1490 
1491 	ret = tap_nl_send(pmd->nlsk_fd, &flow->msg.nh);
1492 	if (ret < 0) {
1493 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE,
1494 				   NULL, "couldn't send request to kernel");
1495 		goto end;
1496 	}
1497 	ret = tap_nl_recv_ack(pmd->nlsk_fd);
1498 	/* If errno is ENOENT, the rule is already no longer in the kernel. */
1499 	if (ret < 0 && errno == ENOENT)
1500 		ret = 0;
1501 	if (ret < 0) {
1502 		TAP_LOG(ERR,
1503 			"Kernel refused TC filter rule deletion (%d): %s",
1504 			errno, strerror(errno));
1505 		rte_flow_error_set(
1506 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1507 			"couldn't receive kernel ack to our request");
1508 		goto end;
1509 	}
1510 
1511 	if (remote_flow) {
1512 		remote_flow->msg.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
1513 		remote_flow->msg.nh.nlmsg_type = RTM_DELTFILTER;
1514 
1515 		ret = tap_nl_send(pmd->nlsk_fd, &remote_flow->msg.nh);
1516 		if (ret < 0) {
1517 			rte_flow_error_set(
1518 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1519 				NULL, "Failure sending nl request");
1520 			goto end;
1521 		}
1522 		ret = tap_nl_recv_ack(pmd->nlsk_fd);
1523 		if (ret < 0 && errno == ENOENT)
1524 			ret = 0;
1525 		if (ret < 0) {
1526 			TAP_LOG(ERR,
1527 				"Kernel refused TC filter rule deletion (%d): %s",
1528 				errno, strerror(errno));
1529 			rte_flow_error_set(
1530 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1531 				NULL, "Failure trying to receive nl ack");
1532 			goto end;
1533 		}
1534 	}
1535 end:
1536 	if (remote_flow)
1537 		rte_free(remote_flow);
1538 	tap_flow_free(pmd, flow);
1539 	return ret;
1540 }
1541 
1542 /**
1543  * Destroy a flow.
1544  *
1545  * @see rte_flow_destroy()
1546  * @see rte_flow_ops
1547  */
1548 static int
1549 tap_flow_destroy(struct rte_eth_dev *dev,
1550 		 struct rte_flow *flow,
1551 		 struct rte_flow_error *error)
1552 {
1553 	struct pmd_internals *pmd = dev->data->dev_private;
1554 
1555 	return tap_flow_destroy_pmd(pmd, flow, error);
1556 }
1557 
1558 /**
1559  * Enable/disable flow isolation.
1560  *
1561  * @see rte_flow_isolate()
1562  * @see rte_flow_ops
1563  */
1564 static int
1565 tap_flow_isolate(struct rte_eth_dev *dev,
1566 		 int set,
1567 		 struct rte_flow_error *error __rte_unused)
1568 {
1569 	struct pmd_internals *pmd = dev->data->dev_private;
1570 
1571 	/* normalize 'set' variable to contain 0 or 1 values */
1572 	if (set)
1573 		set = 1;
1574 	/* if already in the right isolation mode - nothing to do */
1575 	if ((set ^ pmd->flow_isolate) == 0)
1576 		return 0;
1577 	/* mark the isolation mode for tap_flow_implicit_create() */
1578 	pmd->flow_isolate = set;
1579 	/*
1580 	 * If netdevice is there, setup appropriate flow rules immediately.
1581 	 * Otherwise it will be set when bringing up the netdevice (tun_alloc).
1582 	 */
1583 	if (!pmd->rxq[0].fd)
1584 		return 0;
1585 	if (set) {
1586 		struct rte_flow *remote_flow;
1587 
1588 		while (1) {
1589 			remote_flow = LIST_FIRST(&pmd->implicit_flows);
1590 			if (!remote_flow)
1591 				break;
1592 			/*
1593 			 * Remove all implicit rules on the remote.
1594 			 * Keep the local rule to redirect packets on TX.
1595 			 * Keep also the last implicit local rule: ISOLATE.
1596 			 */
1597 			if (remote_flow->msg.t.tcm_ifindex == pmd->if_index)
1598 				break;
1599 			if (tap_flow_destroy_pmd(pmd, remote_flow, NULL) < 0)
1600 				goto error;
1601 		}
1602 		/* Switch the TC rule according to pmd->flow_isolate */
1603 		if (tap_flow_implicit_create(pmd, TAP_ISOLATE) == -1)
1604 			goto error;
1605 	} else {
1606 		/* Switch the TC rule according to pmd->flow_isolate */
1607 		if (tap_flow_implicit_create(pmd, TAP_ISOLATE) == -1)
1608 			goto error;
1609 		if (!pmd->remote_if_index)
1610 			return 0;
1611 		if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0)
1612 			goto error;
1613 		if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
1614 			goto error;
1615 		if (tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0)
1616 			goto error;
1617 		if (tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0)
1618 			goto error;
1619 		if (dev->data->promiscuous &&
1620 		    tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC) < 0)
1621 			goto error;
1622 		if (dev->data->all_multicast &&
1623 		    tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI) < 0)
1624 			goto error;
1625 	}
1626 	return 0;
1627 error:
1628 	pmd->flow_isolate = 0;
1629 	return rte_flow_error_set(
1630 		error, ENOTSUP, RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1631 		"TC rule creation failed");
1632 }
1633 
1634 /**
1635  * Destroy all flows.
1636  *
1637  * @see rte_flow_flush()
1638  * @see rte_flow_ops
1639  */
1640 int
1641 tap_flow_flush(struct rte_eth_dev *dev, struct rte_flow_error *error)
1642 {
1643 	struct pmd_internals *pmd = dev->data->dev_private;
1644 	struct rte_flow *flow;
1645 
1646 	while (!LIST_EMPTY(&pmd->flows)) {
1647 		flow = LIST_FIRST(&pmd->flows);
1648 		if (tap_flow_destroy(dev, flow, error) < 0)
1649 			return -1;
1650 	}
1651 	return 0;
1652 }
1653 
1654 /**
1655  * Add an implicit flow rule on the remote device to make sure traffic gets to
1656  * the tap netdevice from there.
1657  *
1658  * @param pmd
1659  *   Pointer to private structure.
1660  * @param[in] idx
1661  *   The idx in the implicit_rte_flows array specifying which rule to apply.
1662  *
1663  * @return -1 if the rule couldn't be applied, 0 otherwise.
1664  */
1665 int tap_flow_implicit_create(struct pmd_internals *pmd,
1666 			     enum implicit_rule_index idx)
1667 {
1668 	uint16_t flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_EXCL | NLM_F_CREATE;
1669 	struct rte_flow_action *actions = implicit_rte_flows[idx].actions;
1670 	struct rte_flow_action isolate_actions[2] = {
1671 		[1] = {
1672 			.type = RTE_FLOW_ACTION_TYPE_END,
1673 		},
1674 	};
1675 	struct rte_flow_item *items = implicit_rte_flows[idx].items;
1676 	struct rte_flow_attr *attr = &implicit_rte_flows[idx].attr;
1677 	struct rte_flow_item_eth eth_local = { .type = 0 };
1678 	uint16_t if_index = pmd->remote_if_index;
1679 	struct rte_flow *remote_flow = NULL;
1680 	struct nlmsg *msg = NULL;
1681 	int err = 0;
1682 	struct rte_flow_item items_local[2] = {
1683 		[0] = {
1684 			.type = items[0].type,
1685 			.spec = &eth_local,
1686 			.mask = items[0].mask,
1687 		},
1688 		[1] = {
1689 			.type = items[1].type,
1690 		}
1691 	};
1692 
1693 	remote_flow = rte_malloc(__func__, sizeof(struct rte_flow), 0);
1694 	if (!remote_flow) {
1695 		TAP_LOG(ERR, "Cannot allocate memory for rte_flow");
1696 		goto fail;
1697 	}
1698 	msg = &remote_flow->msg;
1699 	if (idx == TAP_REMOTE_TX) {
1700 		if_index = pmd->if_index;
1701 	} else if (idx == TAP_ISOLATE) {
1702 		if_index = pmd->if_index;
1703 		/* Don't be exclusive for this rule, it can be changed later. */
1704 		flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE;
1705 		isolate_actions[0].type = pmd->flow_isolate ?
1706 			RTE_FLOW_ACTION_TYPE_DROP :
1707 			RTE_FLOW_ACTION_TYPE_PASSTHRU;
1708 		actions = isolate_actions;
1709 	} else if (idx == TAP_REMOTE_LOCAL_MAC) {
1710 		/*
1711 		 * eth addr couldn't be set in implicit_rte_flows[] as it is not
1712 		 * known at compile time.
1713 		 */
1714 		memcpy(&eth_local.dst, &pmd->eth_addr, sizeof(pmd->eth_addr));
1715 		items = items_local;
1716 	}
1717 	tc_init_msg(msg, if_index, RTM_NEWTFILTER, flags);
1718 	msg->t.tcm_info = TC_H_MAKE(0, htons(ETH_P_ALL));
1719 	/*
1720 	 * The ISOLATE rule is always present and must have a static handle, as
1721 	 * the action is changed whether the feature is enabled (DROP) or
1722 	 * disabled (PASSTHRU).
1723 	 * There is just one REMOTE_PROMISCUOUS rule in all cases. It should
1724 	 * have a static handle such that adding it twice will fail with EEXIST
1725 	 * with any kernel version. Remark: old kernels may falsely accept the
1726 	 * same REMOTE_PROMISCUOUS rules if they had different handles.
1727 	 */
1728 	if (idx == TAP_ISOLATE)
1729 		remote_flow->msg.t.tcm_handle = ISOLATE_HANDLE;
1730 	else if (idx == TAP_REMOTE_PROMISC)
1731 		remote_flow->msg.t.tcm_handle = REMOTE_PROMISCUOUS_HANDLE;
1732 	else
1733 		tap_flow_set_handle(remote_flow);
1734 	if (priv_flow_process(pmd, attr, items, actions, NULL,
1735 			      remote_flow, implicit_rte_flows[idx].mirred)) {
1736 		TAP_LOG(ERR, "rte flow rule validation failed");
1737 		goto fail;
1738 	}
1739 	err = tap_nl_send(pmd->nlsk_fd, &msg->nh);
1740 	if (err < 0) {
1741 		TAP_LOG(ERR, "Failure sending nl request");
1742 		goto fail;
1743 	}
1744 	err = tap_nl_recv_ack(pmd->nlsk_fd);
1745 	if (err < 0) {
1746 		/* Silently ignore re-entering existing rule */
1747 		if (errno == EEXIST)
1748 			goto success;
1749 		TAP_LOG(ERR,
1750 			"Kernel refused TC filter rule creation (%d): %s",
1751 			errno, strerror(errno));
1752 		goto fail;
1753 	}
1754 	LIST_INSERT_HEAD(&pmd->implicit_flows, remote_flow, next);
1755 success:
1756 	return 0;
1757 fail:
1758 	if (remote_flow)
1759 		rte_free(remote_flow);
1760 	return -1;
1761 }
1762 
1763 /**
1764  * Remove specific implicit flow rule on the remote device.
1765  *
1766  * @param[in, out] pmd
1767  *   Pointer to private structure.
1768  * @param[in] idx
1769  *   The idx in the implicit_rte_flows array specifying which rule to remove.
1770  *
1771  * @return -1 if one of the implicit rules couldn't be created, 0 otherwise.
1772  */
1773 int tap_flow_implicit_destroy(struct pmd_internals *pmd,
1774 			      enum implicit_rule_index idx)
1775 {
1776 	struct rte_flow *remote_flow;
1777 	int cur_prio = -1;
1778 	int idx_prio = implicit_rte_flows[idx].attr.priority + PRIORITY_OFFSET;
1779 
1780 	for (remote_flow = LIST_FIRST(&pmd->implicit_flows);
1781 	     remote_flow;
1782 	     remote_flow = LIST_NEXT(remote_flow, next)) {
1783 		cur_prio = (remote_flow->msg.t.tcm_info >> 16) & PRIORITY_MASK;
1784 		if (cur_prio != idx_prio)
1785 			continue;
1786 		return tap_flow_destroy_pmd(pmd, remote_flow, NULL);
1787 	}
1788 	return 0;
1789 }
1790 
1791 /**
1792  * Destroy all implicit flows.
1793  *
1794  * @see rte_flow_flush()
1795  */
1796 int
1797 tap_flow_implicit_flush(struct pmd_internals *pmd, struct rte_flow_error *error)
1798 {
1799 	struct rte_flow *remote_flow;
1800 
1801 	while (!LIST_EMPTY(&pmd->implicit_flows)) {
1802 		remote_flow = LIST_FIRST(&pmd->implicit_flows);
1803 		if (tap_flow_destroy_pmd(pmd, remote_flow, error) < 0)
1804 			return -1;
1805 	}
1806 	return 0;
1807 }
1808 
1809 #define MAX_RSS_KEYS 256
1810 #define KEY_IDX_OFFSET (3 * MAX_RSS_KEYS)
1811 #define SEC_NAME_CLS_Q "cls_q"
1812 
1813 const char *sec_name[SEC_MAX] = {
1814 	[SEC_L3_L4] = "l3_l4",
1815 };
1816 
1817 /**
1818  * Enable RSS on tap: create TC rules for queuing.
1819  *
1820  * @param[in, out] pmd
1821  *   Pointer to private structure.
1822  *
1823  * @param[in] attr
1824  *   Pointer to rte_flow to get flow group
1825  *
1826  * @param[out] error
1827  *   Pointer to error reporting if not NULL.
1828  *
1829  * @return 0 on success, negative value on failure.
1830  */
1831 static int rss_enable(struct pmd_internals *pmd,
1832 			const struct rte_flow_attr *attr,
1833 			struct rte_flow_error *error)
1834 {
1835 	struct rte_flow *rss_flow = NULL;
1836 	struct nlmsg *msg = NULL;
1837 	/* 4096 is the maximum number of instructions for a BPF program */
1838 	char annotation[64];
1839 	int i;
1840 	int err = 0;
1841 
1842 	/* unlimit locked memory */
1843 	struct rlimit memlock_limit = {
1844 		.rlim_cur = RLIM_INFINITY,
1845 		.rlim_max = RLIM_INFINITY,
1846 	};
1847 	setrlimit(RLIMIT_MEMLOCK, &memlock_limit);
1848 
1849 	 /* Get a new map key for a new RSS rule */
1850 	err = bpf_rss_key(KEY_CMD_INIT, NULL);
1851 	if (err < 0) {
1852 		rte_flow_error_set(
1853 			error, EINVAL, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1854 			"Failed to initialize BPF RSS keys");
1855 
1856 		return -1;
1857 	}
1858 
1859 	/*
1860 	 *  Create BPF RSS MAP
1861 	 */
1862 	pmd->map_fd = tap_flow_bpf_rss_map_create(sizeof(__u32), /* key size */
1863 				sizeof(struct rss_key),
1864 				MAX_RSS_KEYS);
1865 	if (pmd->map_fd < 0) {
1866 		TAP_LOG(ERR,
1867 			"Failed to create BPF map (%d): %s",
1868 				errno, strerror(errno));
1869 		rte_flow_error_set(
1870 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1871 			"Kernel too old or not configured "
1872 			"to support BPF maps");
1873 
1874 		return -ENOTSUP;
1875 	}
1876 
1877 	/*
1878 	 * Add a rule per queue to match reclassified packets and direct them to
1879 	 * the correct queue.
1880 	 */
1881 	for (i = 0; i < pmd->dev->data->nb_rx_queues; i++) {
1882 		pmd->bpf_fd[i] = tap_flow_bpf_cls_q(i);
1883 		if (pmd->bpf_fd[i] < 0) {
1884 			TAP_LOG(ERR,
1885 				"Failed to load BPF section %s for queue %d",
1886 				SEC_NAME_CLS_Q, i);
1887 			rte_flow_error_set(
1888 				error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE,
1889 				NULL,
1890 				"Kernel too old or not configured "
1891 				"to support BPF programs loading");
1892 
1893 			return -ENOTSUP;
1894 		}
1895 
1896 		rss_flow = rte_malloc(__func__, sizeof(struct rte_flow), 0);
1897 		if (!rss_flow) {
1898 			TAP_LOG(ERR,
1899 				"Cannot allocate memory for rte_flow");
1900 			return -1;
1901 		}
1902 		msg = &rss_flow->msg;
1903 		tc_init_msg(msg, pmd->if_index, RTM_NEWTFILTER, NLM_F_REQUEST |
1904 			    NLM_F_ACK | NLM_F_EXCL | NLM_F_CREATE);
1905 		msg->t.tcm_info = TC_H_MAKE(0, htons(ETH_P_ALL));
1906 		tap_flow_set_handle(rss_flow);
1907 		uint16_t group = attr->group << GROUP_SHIFT;
1908 		uint16_t prio = group | (i + PRIORITY_OFFSET);
1909 		msg->t.tcm_info = TC_H_MAKE(prio << 16, msg->t.tcm_info);
1910 		msg->t.tcm_parent = TC_H_MAKE(MULTIQ_MAJOR_HANDLE, 0);
1911 
1912 		tap_nlattr_add(&msg->nh, TCA_KIND, sizeof("bpf"), "bpf");
1913 		if (tap_nlattr_nested_start(msg, TCA_OPTIONS) < 0)
1914 			return -1;
1915 		tap_nlattr_add32(&msg->nh, TCA_BPF_FD, pmd->bpf_fd[i]);
1916 		snprintf(annotation, sizeof(annotation), "[%s%d]",
1917 			SEC_NAME_CLS_Q, i);
1918 		tap_nlattr_add(&msg->nh, TCA_BPF_NAME, strlen(annotation) + 1,
1919 			   annotation);
1920 		/* Actions */
1921 		{
1922 			struct action_data adata = {
1923 				.id = "skbedit",
1924 				.skbedit = {
1925 					.skbedit = {
1926 						.action = TC_ACT_PIPE,
1927 					},
1928 					.queue = i,
1929 				},
1930 			};
1931 			if (add_actions(rss_flow, 1, &adata, TCA_BPF_ACT) < 0)
1932 				return -1;
1933 		}
1934 		tap_nlattr_nested_finish(msg); /* nested TCA_OPTIONS */
1935 
1936 		/* Netlink message is now ready to be sent */
1937 		if (tap_nl_send(pmd->nlsk_fd, &msg->nh) < 0)
1938 			return -1;
1939 		err = tap_nl_recv_ack(pmd->nlsk_fd);
1940 		if (err < 0) {
1941 			TAP_LOG(ERR,
1942 				"Kernel refused TC filter rule creation (%d): %s",
1943 				errno, strerror(errno));
1944 			return err;
1945 		}
1946 		LIST_INSERT_HEAD(&pmd->rss_flows, rss_flow, next);
1947 	}
1948 
1949 	pmd->rss_enabled = 1;
1950 	return err;
1951 }
1952 
1953 /**
1954  * Manage bpf RSS keys repository with operations: init, get, release
1955  *
1956  * @param[in] cmd
1957  *   Command on RSS keys: init, get, release
1958  *
1959  * @param[in, out] key_idx
1960  *   Pointer to RSS Key index (out for get command, in for release command)
1961  *
1962  * @return -1 if couldn't get, release or init the RSS keys, 0 otherwise.
1963  */
1964 static int bpf_rss_key(enum bpf_rss_key_e cmd, __u32 *key_idx)
1965 {
1966 	__u32 i;
1967 	int err = 0;
1968 	static __u32 num_used_keys;
1969 	static __u32 rss_keys[MAX_RSS_KEYS] = {KEY_STAT_UNSPEC};
1970 	static __u32 rss_keys_initialized;
1971 	__u32 key;
1972 
1973 	switch (cmd) {
1974 	case KEY_CMD_GET:
1975 		if (!rss_keys_initialized) {
1976 			err = -1;
1977 			break;
1978 		}
1979 
1980 		if (num_used_keys == RTE_DIM(rss_keys)) {
1981 			err = -1;
1982 			break;
1983 		}
1984 
1985 		*key_idx = num_used_keys % RTE_DIM(rss_keys);
1986 		while (rss_keys[*key_idx] == KEY_STAT_USED)
1987 			*key_idx = (*key_idx + 1) % RTE_DIM(rss_keys);
1988 
1989 		rss_keys[*key_idx] = KEY_STAT_USED;
1990 
1991 		/*
1992 		 * Add an offset to key_idx in order to handle a case of
1993 		 * RSS and non RSS flows mixture.
1994 		 * If a non RSS flow is destroyed it has an eBPF map
1995 		 * index 0 (initialized on flow creation) and might
1996 		 * unintentionally remove RSS entry 0 from eBPF map.
1997 		 * To avoid this issue, add an offset to the real index
1998 		 * during a KEY_CMD_GET operation and subtract this offset
1999 		 * during a KEY_CMD_RELEASE operation in order to restore
2000 		 * the real index.
2001 		 */
2002 		*key_idx += KEY_IDX_OFFSET;
2003 		num_used_keys++;
2004 	break;
2005 
2006 	case KEY_CMD_RELEASE:
2007 		if (!rss_keys_initialized)
2008 			break;
2009 
2010 		/*
2011 		 * Subtract offest to restore real key index
2012 		 * If a non RSS flow is falsely trying to release map
2013 		 * entry 0 - the offset subtraction will calculate the real
2014 		 * map index as an out-of-range value and the release operation
2015 		 * will be silently ignored.
2016 		 */
2017 		key = *key_idx - KEY_IDX_OFFSET;
2018 		if (key >= RTE_DIM(rss_keys))
2019 			break;
2020 
2021 		if (rss_keys[key] == KEY_STAT_USED) {
2022 			rss_keys[key] = KEY_STAT_AVAILABLE;
2023 			num_used_keys--;
2024 		}
2025 	break;
2026 
2027 	case KEY_CMD_INIT:
2028 		for (i = 0; i < RTE_DIM(rss_keys); i++)
2029 			rss_keys[i] = KEY_STAT_AVAILABLE;
2030 
2031 		rss_keys_initialized = 1;
2032 		num_used_keys = 0;
2033 	break;
2034 
2035 	case KEY_CMD_DEINIT:
2036 		for (i = 0; i < RTE_DIM(rss_keys); i++)
2037 			rss_keys[i] = KEY_STAT_UNSPEC;
2038 
2039 		rss_keys_initialized = 0;
2040 		num_used_keys = 0;
2041 	break;
2042 
2043 	default:
2044 		break;
2045 	}
2046 
2047 	return err;
2048 }
2049 
2050 /**
2051  * Add RSS hash calculations and queue selection
2052  *
2053  * @param[in, out] pmd
2054  *   Pointer to internal structure. Used to set/get RSS map fd
2055  *
2056  * @param[in] rss
2057  *   Pointer to RSS flow actions
2058  *
2059  * @param[out] error
2060  *   Pointer to error reporting if not NULL.
2061  *
2062  * @return 0 on success, negative value on failure
2063  */
2064 static int rss_add_actions(struct rte_flow *flow, struct pmd_internals *pmd,
2065 			   const struct rte_flow_action_rss *rss,
2066 			   struct rte_flow_error *error)
2067 {
2068 	/* 4096 is the maximum number of instructions for a BPF program */
2069 	unsigned int i;
2070 	int err;
2071 	struct rss_key rss_entry = { .hash_fields = 0,
2072 				     .key_size = 0 };
2073 
2074 	/* Check supported RSS features */
2075 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT)
2076 		return rte_flow_error_set
2077 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2078 			 "non-default RSS hash functions are not supported");
2079 	if (rss->level)
2080 		return rte_flow_error_set
2081 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2082 			 "a nonzero RSS encapsulation level is not supported");
2083 
2084 	/* Get a new map key for a new RSS rule */
2085 	err = bpf_rss_key(KEY_CMD_GET, &flow->key_idx);
2086 	if (err < 0) {
2087 		rte_flow_error_set(
2088 			error, EINVAL, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2089 			"Failed to get BPF RSS key");
2090 
2091 		return -1;
2092 	}
2093 
2094 	/* Update RSS map entry with queues */
2095 	rss_entry.nb_queues = rss->queue_num;
2096 	for (i = 0; i < rss->queue_num; i++)
2097 		rss_entry.queues[i] = rss->queue[i];
2098 	rss_entry.hash_fields =
2099 		(1 << HASH_FIELD_IPV4_L3_L4) | (1 << HASH_FIELD_IPV6_L3_L4);
2100 
2101 	/* Add this RSS entry to map */
2102 	err = tap_flow_bpf_update_rss_elem(pmd->map_fd,
2103 				&flow->key_idx, &rss_entry);
2104 
2105 	if (err) {
2106 		TAP_LOG(ERR,
2107 			"Failed to update BPF map entry #%u (%d): %s",
2108 			flow->key_idx, errno, strerror(errno));
2109 		rte_flow_error_set(
2110 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2111 			"Kernel too old or not configured "
2112 			"to support BPF maps updates");
2113 
2114 		return -ENOTSUP;
2115 	}
2116 
2117 
2118 	/*
2119 	 * Load bpf rules to calculate hash for this key_idx
2120 	 */
2121 
2122 	flow->bpf_fd[SEC_L3_L4] =
2123 		tap_flow_bpf_calc_l3_l4_hash(flow->key_idx, pmd->map_fd);
2124 	if (flow->bpf_fd[SEC_L3_L4] < 0) {
2125 		TAP_LOG(ERR,
2126 			"Failed to load BPF section %s (%d): %s",
2127 				sec_name[SEC_L3_L4], errno, strerror(errno));
2128 		rte_flow_error_set(
2129 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2130 			"Kernel too old or not configured "
2131 			"to support BPF program loading");
2132 
2133 		return -ENOTSUP;
2134 	}
2135 
2136 	/* Actions */
2137 	{
2138 		struct action_data adata[] = {
2139 			{
2140 				.id = "bpf",
2141 				.bpf = {
2142 					.bpf_fd = flow->bpf_fd[SEC_L3_L4],
2143 					.annotation = sec_name[SEC_L3_L4],
2144 					.bpf = {
2145 						.action = TC_ACT_PIPE,
2146 					},
2147 				},
2148 			},
2149 		};
2150 
2151 		if (add_actions(flow, RTE_DIM(adata), adata,
2152 			TCA_FLOWER_ACT) < 0)
2153 			return -1;
2154 	}
2155 
2156 	return 0;
2157 }
2158 
2159 /**
2160  * Manage filter operations.
2161  *
2162  * @param dev
2163  *   Pointer to Ethernet device structure.
2164  * @param filter_type
2165  *   Filter type.
2166  * @param filter_op
2167  *   Operation to perform.
2168  * @param arg
2169  *   Pointer to operation-specific structure.
2170  *
2171  * @return
2172  *   0 on success, negative errno value on failure.
2173  */
2174 int
2175 tap_dev_filter_ctrl(struct rte_eth_dev *dev,
2176 		    enum rte_filter_type filter_type,
2177 		    enum rte_filter_op filter_op,
2178 		    void *arg)
2179 {
2180 	switch (filter_type) {
2181 	case RTE_ETH_FILTER_GENERIC:
2182 		if (filter_op != RTE_ETH_FILTER_GET)
2183 			return -EINVAL;
2184 		*(const void **)arg = &tap_flow_ops;
2185 		return 0;
2186 	default:
2187 		TAP_LOG(ERR, "%p: filter type (%d) not supported",
2188 			dev, filter_type);
2189 	}
2190 	return -EINVAL;
2191 }
2192