xref: /dpdk/drivers/net/tap/tap_flow.c (revision c7e9729d)
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 			.tpid = -1,
274 			/* DEI matching is not supported */
275 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
276 			.tci = 0xffef,
277 #else
278 			.tci = 0xefff,
279 #endif
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(&spec->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 	/* TC does not support tpid masking. Only accept if exact match. */
582 	if (mask->tpid && mask->tpid != 0xffff)
583 		return -1;
584 	/* Double-tagging not supported. */
585 	if (spec && mask->tpid && spec->tpid != htons(ETH_P_8021Q))
586 		return -1;
587 	info->vlan = 1;
588 	if (!flow)
589 		return 0;
590 	msg = &flow->msg;
591 	msg->t.tcm_info = TC_H_MAKE(msg->t.tcm_info, htons(ETH_P_8021Q));
592 #define VLAN_PRIO(tci) ((tci) >> 13)
593 #define VLAN_ID(tci) ((tci) & 0xfff)
594 	if (!spec)
595 		return 0;
596 	if (spec->tci) {
597 		uint16_t tci = ntohs(spec->tci) & mask->tci;
598 		uint16_t prio = VLAN_PRIO(tci);
599 		uint8_t vid = VLAN_ID(tci);
600 
601 		if (prio)
602 			tap_nlattr_add8(&msg->nh,
603 					TCA_FLOWER_KEY_VLAN_PRIO, prio);
604 		if (vid)
605 			tap_nlattr_add16(&msg->nh,
606 					 TCA_FLOWER_KEY_VLAN_ID, vid);
607 	}
608 	return 0;
609 }
610 
611 /**
612  * Make as much checks as possible on an IPv4 item, and if a flow is provided,
613  * fill it appropriately with IPv4 info.
614  *
615  * @param[in] item
616  *   Item specification.
617  * @param[in, out] data
618  *   Additional data structure to tell next layers we've been here.
619  *
620  * @return
621  *   0 if checks are alright, -1 otherwise.
622  */
623 static int
624 tap_flow_create_ipv4(const struct rte_flow_item *item, void *data)
625 {
626 	struct convert_data *info = (struct convert_data *)data;
627 	const struct rte_flow_item_ipv4 *spec = item->spec;
628 	const struct rte_flow_item_ipv4 *mask = item->mask;
629 	struct rte_flow *flow = info->flow;
630 	struct nlmsg *msg;
631 
632 	/* use default mask if none provided */
633 	if (!mask)
634 		mask = tap_flow_items[RTE_FLOW_ITEM_TYPE_IPV4].default_mask;
635 	/* check that previous eth type is compatible with ipv4 */
636 	if (info->eth_type && info->eth_type != htons(ETH_P_IP))
637 		return -1;
638 	/* store ip_proto for consistency if udp/tcp pattern item comes next */
639 	if (spec)
640 		info->ip_proto = spec->hdr.next_proto_id;
641 	if (!flow)
642 		return 0;
643 	msg = &flow->msg;
644 	if (!info->eth_type)
645 		info->eth_type = htons(ETH_P_IP);
646 	if (!spec)
647 		return 0;
648 	if (spec->hdr.dst_addr) {
649 		tap_nlattr_add32(&msg->nh, TCA_FLOWER_KEY_IPV4_DST,
650 			     spec->hdr.dst_addr);
651 		tap_nlattr_add32(&msg->nh, TCA_FLOWER_KEY_IPV4_DST_MASK,
652 			     mask->hdr.dst_addr);
653 	}
654 	if (spec->hdr.src_addr) {
655 		tap_nlattr_add32(&msg->nh, TCA_FLOWER_KEY_IPV4_SRC,
656 			     spec->hdr.src_addr);
657 		tap_nlattr_add32(&msg->nh, TCA_FLOWER_KEY_IPV4_SRC_MASK,
658 			     mask->hdr.src_addr);
659 	}
660 	if (spec->hdr.next_proto_id)
661 		tap_nlattr_add8(&msg->nh, TCA_FLOWER_KEY_IP_PROTO,
662 			    spec->hdr.next_proto_id);
663 	return 0;
664 }
665 
666 /**
667  * Make as much checks as possible on an IPv6 item, and if a flow is provided,
668  * fill it appropriately with IPv6 info.
669  *
670  * @param[in] item
671  *   Item specification.
672  * @param[in, out] data
673  *   Additional data structure to tell next layers we've been here.
674  *
675  * @return
676  *   0 if checks are alright, -1 otherwise.
677  */
678 static int
679 tap_flow_create_ipv6(const struct rte_flow_item *item, void *data)
680 {
681 	struct convert_data *info = (struct convert_data *)data;
682 	const struct rte_flow_item_ipv6 *spec = item->spec;
683 	const struct rte_flow_item_ipv6 *mask = item->mask;
684 	struct rte_flow *flow = info->flow;
685 	uint8_t empty_addr[16] = { 0 };
686 	struct nlmsg *msg;
687 
688 	/* use default mask if none provided */
689 	if (!mask)
690 		mask = tap_flow_items[RTE_FLOW_ITEM_TYPE_IPV6].default_mask;
691 	/* check that previous eth type is compatible with ipv6 */
692 	if (info->eth_type && info->eth_type != htons(ETH_P_IPV6))
693 		return -1;
694 	/* store ip_proto for consistency if udp/tcp pattern item comes next */
695 	if (spec)
696 		info->ip_proto = spec->hdr.proto;
697 	if (!flow)
698 		return 0;
699 	msg = &flow->msg;
700 	if (!info->eth_type)
701 		info->eth_type = htons(ETH_P_IPV6);
702 	if (!spec)
703 		return 0;
704 	if (memcmp(spec->hdr.dst_addr, empty_addr, 16)) {
705 		tap_nlattr_add(&msg->nh, TCA_FLOWER_KEY_IPV6_DST,
706 			   sizeof(spec->hdr.dst_addr), &spec->hdr.dst_addr);
707 		tap_nlattr_add(&msg->nh, TCA_FLOWER_KEY_IPV6_DST_MASK,
708 			   sizeof(mask->hdr.dst_addr), &mask->hdr.dst_addr);
709 	}
710 	if (memcmp(spec->hdr.src_addr, empty_addr, 16)) {
711 		tap_nlattr_add(&msg->nh, TCA_FLOWER_KEY_IPV6_SRC,
712 			   sizeof(spec->hdr.src_addr), &spec->hdr.src_addr);
713 		tap_nlattr_add(&msg->nh, TCA_FLOWER_KEY_IPV6_SRC_MASK,
714 			   sizeof(mask->hdr.src_addr), &mask->hdr.src_addr);
715 	}
716 	if (spec->hdr.proto)
717 		tap_nlattr_add8(&msg->nh,
718 				TCA_FLOWER_KEY_IP_PROTO, spec->hdr.proto);
719 	return 0;
720 }
721 
722 /**
723  * Make as much checks as possible on a UDP item, and if a flow is provided,
724  * fill it appropriately with UDP info.
725  *
726  * @param[in] item
727  *   Item specification.
728  * @param[in, out] data
729  *   Additional data structure to tell next layers we've been here.
730  *
731  * @return
732  *   0 if checks are alright, -1 otherwise.
733  */
734 static int
735 tap_flow_create_udp(const struct rte_flow_item *item, void *data)
736 {
737 	struct convert_data *info = (struct convert_data *)data;
738 	const struct rte_flow_item_udp *spec = item->spec;
739 	const struct rte_flow_item_udp *mask = item->mask;
740 	struct rte_flow *flow = info->flow;
741 	struct nlmsg *msg;
742 
743 	/* use default mask if none provided */
744 	if (!mask)
745 		mask = tap_flow_items[RTE_FLOW_ITEM_TYPE_UDP].default_mask;
746 	/* check that previous ip_proto is compatible with udp */
747 	if (info->ip_proto && info->ip_proto != IPPROTO_UDP)
748 		return -1;
749 	/* TC does not support UDP port masking. Only accept if exact match. */
750 	if ((mask->hdr.src_port && mask->hdr.src_port != 0xffff) ||
751 	    (mask->hdr.dst_port && mask->hdr.dst_port != 0xffff))
752 		return -1;
753 	if (!flow)
754 		return 0;
755 	msg = &flow->msg;
756 	tap_nlattr_add8(&msg->nh, TCA_FLOWER_KEY_IP_PROTO, IPPROTO_UDP);
757 	if (!spec)
758 		return 0;
759 	if (spec->hdr.dst_port & mask->hdr.dst_port)
760 		tap_nlattr_add16(&msg->nh, TCA_FLOWER_KEY_UDP_DST,
761 			     spec->hdr.dst_port);
762 	if (spec->hdr.src_port & mask->hdr.src_port)
763 		tap_nlattr_add16(&msg->nh, TCA_FLOWER_KEY_UDP_SRC,
764 			     spec->hdr.src_port);
765 	return 0;
766 }
767 
768 /**
769  * Make as much checks as possible on a TCP item, and if a flow is provided,
770  * fill it appropriately with TCP info.
771  *
772  * @param[in] item
773  *   Item specification.
774  * @param[in, out] data
775  *   Additional data structure to tell next layers we've been here.
776  *
777  * @return
778  *   0 if checks are alright, -1 otherwise.
779  */
780 static int
781 tap_flow_create_tcp(const struct rte_flow_item *item, void *data)
782 {
783 	struct convert_data *info = (struct convert_data *)data;
784 	const struct rte_flow_item_tcp *spec = item->spec;
785 	const struct rte_flow_item_tcp *mask = item->mask;
786 	struct rte_flow *flow = info->flow;
787 	struct nlmsg *msg;
788 
789 	/* use default mask if none provided */
790 	if (!mask)
791 		mask = tap_flow_items[RTE_FLOW_ITEM_TYPE_TCP].default_mask;
792 	/* check that previous ip_proto is compatible with tcp */
793 	if (info->ip_proto && info->ip_proto != IPPROTO_TCP)
794 		return -1;
795 	/* TC does not support TCP port masking. Only accept if exact match. */
796 	if ((mask->hdr.src_port && mask->hdr.src_port != 0xffff) ||
797 	    (mask->hdr.dst_port && mask->hdr.dst_port != 0xffff))
798 		return -1;
799 	if (!flow)
800 		return 0;
801 	msg = &flow->msg;
802 	tap_nlattr_add8(&msg->nh, TCA_FLOWER_KEY_IP_PROTO, IPPROTO_TCP);
803 	if (!spec)
804 		return 0;
805 	if (spec->hdr.dst_port & mask->hdr.dst_port)
806 		tap_nlattr_add16(&msg->nh, TCA_FLOWER_KEY_TCP_DST,
807 			     spec->hdr.dst_port);
808 	if (spec->hdr.src_port & mask->hdr.src_port)
809 		tap_nlattr_add16(&msg->nh, TCA_FLOWER_KEY_TCP_SRC,
810 			     spec->hdr.src_port);
811 	return 0;
812 }
813 
814 /**
815  * Check support for a given item.
816  *
817  * @param[in] item
818  *   Item specification.
819  * @param size
820  *   Bit-Mask size in bytes.
821  * @param[in] supported_mask
822  *   Bit-mask covering supported fields to compare with spec, last and mask in
823  *   \item.
824  * @param[in] default_mask
825  *   Bit-mask default mask if none is provided in \item.
826  *
827  * @return
828  *   0 on success.
829  */
830 static int
831 tap_flow_item_validate(const struct rte_flow_item *item,
832 		       unsigned int size,
833 		       const uint8_t *supported_mask,
834 		       const uint8_t *default_mask)
835 {
836 	int ret = 0;
837 
838 	/* An empty layer is allowed, as long as all fields are NULL */
839 	if (!item->spec && (item->mask || item->last))
840 		return -1;
841 	/* Is the item spec compatible with what the NIC supports? */
842 	if (item->spec && !item->mask) {
843 		unsigned int i;
844 		const uint8_t *spec = item->spec;
845 
846 		for (i = 0; i < size; ++i)
847 			if ((spec[i] | supported_mask[i]) != supported_mask[i])
848 				return -1;
849 		/* Is the default mask compatible with what the NIC supports? */
850 		for (i = 0; i < size; i++)
851 			if ((default_mask[i] | supported_mask[i]) !=
852 			    supported_mask[i])
853 				return -1;
854 	}
855 	/* Is the item last compatible with what the NIC supports? */
856 	if (item->last && !item->mask) {
857 		unsigned int i;
858 		const uint8_t *spec = item->last;
859 
860 		for (i = 0; i < size; ++i)
861 			if ((spec[i] | supported_mask[i]) != supported_mask[i])
862 				return -1;
863 	}
864 	/* Is the item mask compatible with what the NIC supports? */
865 	if (item->mask) {
866 		unsigned int i;
867 		const uint8_t *spec = item->mask;
868 
869 		for (i = 0; i < size; ++i)
870 			if ((spec[i] | supported_mask[i]) != supported_mask[i])
871 				return -1;
872 	}
873 	/**
874 	 * Once masked, Are item spec and item last equal?
875 	 * TC does not support range so anything else is invalid.
876 	 */
877 	if (item->spec && item->last) {
878 		uint8_t spec[size];
879 		uint8_t last[size];
880 		const uint8_t *apply = default_mask;
881 		unsigned int i;
882 
883 		if (item->mask)
884 			apply = item->mask;
885 		for (i = 0; i < size; ++i) {
886 			spec[i] = ((const uint8_t *)item->spec)[i] & apply[i];
887 			last[i] = ((const uint8_t *)item->last)[i] & apply[i];
888 		}
889 		ret = memcmp(spec, last, size);
890 	}
891 	return ret;
892 }
893 
894 /**
895  * Configure the kernel with a TC action and its configured parameters
896  * Handled actions: "gact", "mirred", "skbedit", "bpf"
897  *
898  * @param[in] flow
899  *   Pointer to rte flow containing the netlink message
900  *
901  * @param[in, out] act_index
902  *   Pointer to action sequence number in the TC command
903  *
904  * @param[in] adata
905  *  Pointer to struct holding the action parameters
906  *
907  * @return
908  *   -1 on failure, 0 on success
909  */
910 static int
911 add_action(struct rte_flow *flow, size_t *act_index, struct action_data *adata)
912 {
913 	struct nlmsg *msg = &flow->msg;
914 
915 	if (tap_nlattr_nested_start(msg, (*act_index)++) < 0)
916 		return -1;
917 
918 	tap_nlattr_add(&msg->nh, TCA_ACT_KIND,
919 				strlen(adata->id) + 1, adata->id);
920 	if (tap_nlattr_nested_start(msg, TCA_ACT_OPTIONS) < 0)
921 		return -1;
922 	if (strcmp("gact", adata->id) == 0) {
923 		tap_nlattr_add(&msg->nh, TCA_GACT_PARMS, sizeof(adata->gact),
924 			   &adata->gact);
925 	} else if (strcmp("mirred", adata->id) == 0) {
926 		if (adata->mirred.eaction == TCA_EGRESS_MIRROR)
927 			adata->mirred.action = TC_ACT_PIPE;
928 		else /* REDIRECT */
929 			adata->mirred.action = TC_ACT_STOLEN;
930 		tap_nlattr_add(&msg->nh, TCA_MIRRED_PARMS,
931 			   sizeof(adata->mirred),
932 			   &adata->mirred);
933 	} else if (strcmp("skbedit", adata->id) == 0) {
934 		tap_nlattr_add(&msg->nh, TCA_SKBEDIT_PARMS,
935 			   sizeof(adata->skbedit.skbedit),
936 			   &adata->skbedit.skbedit);
937 		tap_nlattr_add16(&msg->nh, TCA_SKBEDIT_QUEUE_MAPPING,
938 			     adata->skbedit.queue);
939 	} else if (strcmp("bpf", adata->id) == 0) {
940 		tap_nlattr_add32(&msg->nh, TCA_ACT_BPF_FD, adata->bpf.bpf_fd);
941 		tap_nlattr_add(&msg->nh, TCA_ACT_BPF_NAME,
942 			   strlen(adata->bpf.annotation) + 1,
943 			   adata->bpf.annotation);
944 		tap_nlattr_add(&msg->nh, TCA_ACT_BPF_PARMS,
945 			   sizeof(adata->bpf.bpf),
946 			   &adata->bpf.bpf);
947 	} else {
948 		return -1;
949 	}
950 	tap_nlattr_nested_finish(msg); /* nested TCA_ACT_OPTIONS */
951 	tap_nlattr_nested_finish(msg); /* nested act_index */
952 	return 0;
953 }
954 
955 /**
956  * Helper function to send a serie of TC actions to the kernel
957  *
958  * @param[in] flow
959  *   Pointer to rte flow containing the netlink message
960  *
961  * @param[in] nb_actions
962  *   Number of actions in an array of action structs
963  *
964  * @param[in] data
965  *   Pointer to an array of action structs
966  *
967  * @param[in] classifier_actions
968  *   The classifier on behave of which the actions are configured
969  *
970  * @return
971  *   -1 on failure, 0 on success
972  */
973 static int
974 add_actions(struct rte_flow *flow, int nb_actions, struct action_data *data,
975 	    int classifier_action)
976 {
977 	struct nlmsg *msg = &flow->msg;
978 	size_t act_index = 1;
979 	int i;
980 
981 	if (tap_nlattr_nested_start(msg, classifier_action) < 0)
982 		return -1;
983 	for (i = 0; i < nb_actions; i++)
984 		if (add_action(flow, &act_index, data + i) < 0)
985 			return -1;
986 	tap_nlattr_nested_finish(msg); /* nested TCA_FLOWER_ACT */
987 	return 0;
988 }
989 
990 /**
991  * Validate a flow supported by TC.
992  * If flow param is not NULL, then also fill the netlink message inside.
993  *
994  * @param pmd
995  *   Pointer to private structure.
996  * @param[in] attr
997  *   Flow rule attributes.
998  * @param[in] pattern
999  *   Pattern specification (list terminated by the END pattern item).
1000  * @param[in] actions
1001  *   Associated actions (list terminated by the END action).
1002  * @param[out] error
1003  *   Perform verbose error reporting if not NULL.
1004  * @param[in, out] flow
1005  *   Flow structure to update.
1006  * @param[in] mirred
1007  *   If set to TCA_EGRESS_REDIR, provided actions will be replaced with a
1008  *   redirection to the tap netdevice, and the TC rule will be configured
1009  *   on the remote netdevice in pmd.
1010  *   If set to TCA_EGRESS_MIRROR, provided actions will be replaced with a
1011  *   mirroring to the tap netdevice, and the TC rule will be configured
1012  *   on the remote netdevice in pmd. Matching packets will thus be duplicated.
1013  *   If set to 0, the standard behavior is to be used: set correct actions for
1014  *   the TC rule, and apply it on the tap netdevice.
1015  *
1016  * @return
1017  *   0 on success, a negative errno value otherwise and rte_errno is set.
1018  */
1019 static int
1020 priv_flow_process(struct pmd_internals *pmd,
1021 		  const struct rte_flow_attr *attr,
1022 		  const struct rte_flow_item items[],
1023 		  const struct rte_flow_action actions[],
1024 		  struct rte_flow_error *error,
1025 		  struct rte_flow *flow,
1026 		  int mirred)
1027 {
1028 	const struct tap_flow_items *cur_item = tap_flow_items;
1029 	struct convert_data data = {
1030 		.eth_type = 0,
1031 		.ip_proto = 0,
1032 		.flow = flow,
1033 	};
1034 	int action = 0; /* Only one action authorized for now */
1035 
1036 	if (attr->group > MAX_GROUP) {
1037 		rte_flow_error_set(
1038 			error, EINVAL, RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1039 			NULL, "group value too big: cannot exceed 15");
1040 		return -rte_errno;
1041 	}
1042 	if (attr->priority > MAX_PRIORITY) {
1043 		rte_flow_error_set(
1044 			error, EINVAL, RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1045 			NULL, "priority value too big");
1046 		return -rte_errno;
1047 	} else if (flow) {
1048 		uint16_t group = attr->group << GROUP_SHIFT;
1049 		uint16_t prio = group | (attr->priority +
1050 				RSS_PRIORITY_OFFSET + PRIORITY_OFFSET);
1051 		flow->msg.t.tcm_info = TC_H_MAKE(prio << 16,
1052 						 flow->msg.t.tcm_info);
1053 	}
1054 	if (flow) {
1055 		if (mirred) {
1056 			/*
1057 			 * If attr->ingress, the rule applies on remote ingress
1058 			 * to match incoming packets
1059 			 * If attr->egress, the rule applies on tap ingress (as
1060 			 * seen from the kernel) to deal with packets going out
1061 			 * from the DPDK app.
1062 			 */
1063 			flow->msg.t.tcm_parent = TC_H_MAKE(TC_H_INGRESS, 0);
1064 		} else {
1065 			/* Standard rule on tap egress (kernel standpoint). */
1066 			flow->msg.t.tcm_parent =
1067 				TC_H_MAKE(MULTIQ_MAJOR_HANDLE, 0);
1068 		}
1069 		/* use flower filter type */
1070 		tap_nlattr_add(&flow->msg.nh, TCA_KIND, sizeof("flower"), "flower");
1071 		if (tap_nlattr_nested_start(&flow->msg, TCA_OPTIONS) < 0)
1072 			goto exit_item_not_supported;
1073 	}
1074 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; ++items) {
1075 		const struct tap_flow_items *token = NULL;
1076 		unsigned int i;
1077 		int err = 0;
1078 
1079 		if (items->type == RTE_FLOW_ITEM_TYPE_VOID)
1080 			continue;
1081 		for (i = 0;
1082 		     cur_item->items &&
1083 		     cur_item->items[i] != RTE_FLOW_ITEM_TYPE_END;
1084 		     ++i) {
1085 			if (cur_item->items[i] == items->type) {
1086 				token = &tap_flow_items[items->type];
1087 				break;
1088 			}
1089 		}
1090 		if (!token)
1091 			goto exit_item_not_supported;
1092 		cur_item = token;
1093 		err = tap_flow_item_validate(
1094 			items, cur_item->mask_sz,
1095 			(const uint8_t *)cur_item->mask,
1096 			(const uint8_t *)cur_item->default_mask);
1097 		if (err)
1098 			goto exit_item_not_supported;
1099 		if (flow && cur_item->convert) {
1100 			err = cur_item->convert(items, &data);
1101 			if (err)
1102 				goto exit_item_not_supported;
1103 		}
1104 	}
1105 	if (flow) {
1106 		if (data.vlan) {
1107 			tap_nlattr_add16(&flow->msg.nh, TCA_FLOWER_KEY_ETH_TYPE,
1108 				     htons(ETH_P_8021Q));
1109 			tap_nlattr_add16(&flow->msg.nh,
1110 				     TCA_FLOWER_KEY_VLAN_ETH_TYPE,
1111 				     data.eth_type ?
1112 				     data.eth_type : htons(ETH_P_ALL));
1113 		} else if (data.eth_type) {
1114 			tap_nlattr_add16(&flow->msg.nh, TCA_FLOWER_KEY_ETH_TYPE,
1115 				     data.eth_type);
1116 		}
1117 	}
1118 	if (mirred && flow) {
1119 		struct action_data adata = {
1120 			.id = "mirred",
1121 			.mirred = {
1122 				.eaction = mirred,
1123 			},
1124 		};
1125 
1126 		/*
1127 		 * If attr->egress && mirred, then this is a special
1128 		 * case where the rule must be applied on the tap, to
1129 		 * redirect packets coming from the DPDK App, out
1130 		 * through the remote netdevice.
1131 		 */
1132 		adata.mirred.ifindex = attr->ingress ? pmd->if_index :
1133 			pmd->remote_if_index;
1134 		if (mirred == TCA_EGRESS_MIRROR)
1135 			adata.mirred.action = TC_ACT_PIPE;
1136 		else
1137 			adata.mirred.action = TC_ACT_STOLEN;
1138 		if (add_actions(flow, 1, &adata, TCA_FLOWER_ACT) < 0)
1139 			goto exit_action_not_supported;
1140 		else
1141 			goto end;
1142 	}
1143 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; ++actions) {
1144 		int err = 0;
1145 
1146 		if (actions->type == RTE_FLOW_ACTION_TYPE_VOID) {
1147 			continue;
1148 		} else if (actions->type == RTE_FLOW_ACTION_TYPE_DROP) {
1149 			if (action)
1150 				goto exit_action_not_supported;
1151 			action = 1;
1152 			if (flow) {
1153 				struct action_data adata = {
1154 					.id = "gact",
1155 					.gact = {
1156 						.action = TC_ACT_SHOT,
1157 					},
1158 				};
1159 
1160 				err = add_actions(flow, 1, &adata,
1161 						  TCA_FLOWER_ACT);
1162 			}
1163 		} else if (actions->type == RTE_FLOW_ACTION_TYPE_PASSTHRU) {
1164 			if (action)
1165 				goto exit_action_not_supported;
1166 			action = 1;
1167 			if (flow) {
1168 				struct action_data adata = {
1169 					.id = "gact",
1170 					.gact = {
1171 						/* continue */
1172 						.action = TC_ACT_UNSPEC,
1173 					},
1174 				};
1175 
1176 				err = add_actions(flow, 1, &adata,
1177 						  TCA_FLOWER_ACT);
1178 			}
1179 		} else if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
1180 			const struct rte_flow_action_queue *queue =
1181 				(const struct rte_flow_action_queue *)
1182 				actions->conf;
1183 
1184 			if (action)
1185 				goto exit_action_not_supported;
1186 			action = 1;
1187 			if (!queue ||
1188 			    (queue->index > pmd->dev->data->nb_rx_queues - 1))
1189 				goto exit_action_not_supported;
1190 			if (flow) {
1191 				struct action_data adata = {
1192 					.id = "skbedit",
1193 					.skbedit = {
1194 						.skbedit = {
1195 							.action = TC_ACT_PIPE,
1196 						},
1197 						.queue = queue->index,
1198 					},
1199 				};
1200 
1201 				err = add_actions(flow, 1, &adata,
1202 					TCA_FLOWER_ACT);
1203 			}
1204 		} else if (actions->type == RTE_FLOW_ACTION_TYPE_RSS) {
1205 			const struct rte_flow_action_rss *rss =
1206 				(const struct rte_flow_action_rss *)
1207 				actions->conf;
1208 
1209 			if (action++)
1210 				goto exit_action_not_supported;
1211 
1212 			if (!pmd->rss_enabled) {
1213 				err = rss_enable(pmd, attr, error);
1214 				if (err)
1215 					goto exit_action_not_supported;
1216 			}
1217 			if (flow && rss)
1218 				err = rss_add_actions(flow, pmd, rss, error);
1219 		} else {
1220 			goto exit_action_not_supported;
1221 		}
1222 		if (err)
1223 			goto exit_action_not_supported;
1224 	}
1225 end:
1226 	if (flow)
1227 		tap_nlattr_nested_finish(&flow->msg); /* nested TCA_OPTIONS */
1228 	return 0;
1229 exit_item_not_supported:
1230 	rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1231 			   items, "item not supported");
1232 	return -rte_errno;
1233 exit_action_not_supported:
1234 	rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
1235 			   actions, "action not supported");
1236 	return -rte_errno;
1237 }
1238 
1239 
1240 
1241 /**
1242  * Validate a flow.
1243  *
1244  * @see rte_flow_validate()
1245  * @see rte_flow_ops
1246  */
1247 static int
1248 tap_flow_validate(struct rte_eth_dev *dev,
1249 		  const struct rte_flow_attr *attr,
1250 		  const struct rte_flow_item items[],
1251 		  const struct rte_flow_action actions[],
1252 		  struct rte_flow_error *error)
1253 {
1254 	struct pmd_internals *pmd = dev->data->dev_private;
1255 
1256 	return priv_flow_process(pmd, attr, items, actions, error, NULL, 0);
1257 }
1258 
1259 /**
1260  * Set a unique handle in a flow.
1261  *
1262  * The kernel supports TC rules with equal priority, as long as they use the
1263  * same matching fields (e.g.: dst mac and ipv4) with different values (and
1264  * full mask to ensure no collision is possible).
1265  * In those rules, the handle (uint32_t) is the part that would identify
1266  * specifically each rule.
1267  *
1268  * On 32-bit architectures, the handle can simply be the flow's pointer address.
1269  * On 64-bit architectures, we rely on jhash(flow) to find a (sufficiently)
1270  * unique handle.
1271  *
1272  * @param[in, out] flow
1273  *   The flow that needs its handle set.
1274  */
1275 static void
1276 tap_flow_set_handle(struct rte_flow *flow)
1277 {
1278 	uint32_t handle = 0;
1279 
1280 	if (sizeof(flow) > 4)
1281 		handle = rte_jhash(&flow, sizeof(flow), 1);
1282 	else
1283 		handle = (uintptr_t)flow;
1284 	/* must be at least 1 to avoid letting the kernel choose one for us */
1285 	if (!handle)
1286 		handle = 1;
1287 	flow->msg.t.tcm_handle = handle;
1288 }
1289 
1290 /**
1291  * Free the flow opened file descriptors and allocated memory
1292  *
1293  * @param[in] flow
1294  *   Pointer to the flow to free
1295  *
1296  */
1297 static void
1298 tap_flow_free(struct pmd_internals *pmd, struct rte_flow *flow)
1299 {
1300 	int i;
1301 
1302 	if (!flow)
1303 		return;
1304 
1305 	if (pmd->rss_enabled) {
1306 		/* Close flow BPF file descriptors */
1307 		for (i = 0; i < SEC_MAX; i++)
1308 			if (flow->bpf_fd[i] != 0) {
1309 				close(flow->bpf_fd[i]);
1310 				flow->bpf_fd[i] = 0;
1311 			}
1312 
1313 		/* Release the map key for this RSS rule */
1314 		bpf_rss_key(KEY_CMD_RELEASE, &flow->key_idx);
1315 		flow->key_idx = 0;
1316 	}
1317 
1318 	/* Free flow allocated memory */
1319 	rte_free(flow);
1320 }
1321 
1322 /**
1323  * Create a flow.
1324  *
1325  * @see rte_flow_create()
1326  * @see rte_flow_ops
1327  */
1328 static struct rte_flow *
1329 tap_flow_create(struct rte_eth_dev *dev,
1330 		const struct rte_flow_attr *attr,
1331 		const struct rte_flow_item items[],
1332 		const struct rte_flow_action actions[],
1333 		struct rte_flow_error *error)
1334 {
1335 	struct pmd_internals *pmd = dev->data->dev_private;
1336 	struct rte_flow *remote_flow = NULL;
1337 	struct rte_flow *flow = NULL;
1338 	struct nlmsg *msg = NULL;
1339 	int err;
1340 
1341 	if (!pmd->if_index) {
1342 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE,
1343 				   NULL,
1344 				   "can't create rule, ifindex not found");
1345 		goto fail;
1346 	}
1347 	/*
1348 	 * No rules configured through standard rte_flow should be set on the
1349 	 * priorities used by implicit rules.
1350 	 */
1351 	if ((attr->group == MAX_GROUP) &&
1352 	    attr->priority > (MAX_PRIORITY - TAP_REMOTE_MAX_IDX)) {
1353 		rte_flow_error_set(
1354 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1355 			NULL, "priority value too big");
1356 		goto fail;
1357 	}
1358 	flow = rte_malloc(__func__, sizeof(struct rte_flow), 0);
1359 	if (!flow) {
1360 		rte_flow_error_set(error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1361 				   NULL, "cannot allocate memory for rte_flow");
1362 		goto fail;
1363 	}
1364 	msg = &flow->msg;
1365 	tc_init_msg(msg, pmd->if_index, RTM_NEWTFILTER,
1366 		    NLM_F_REQUEST | NLM_F_ACK | NLM_F_EXCL | NLM_F_CREATE);
1367 	msg->t.tcm_info = TC_H_MAKE(0, htons(ETH_P_ALL));
1368 	tap_flow_set_handle(flow);
1369 	if (priv_flow_process(pmd, attr, items, actions, error, flow, 0))
1370 		goto fail;
1371 	err = tap_nl_send(pmd->nlsk_fd, &msg->nh);
1372 	if (err < 0) {
1373 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE,
1374 				   NULL, "couldn't send request to kernel");
1375 		goto fail;
1376 	}
1377 	err = tap_nl_recv_ack(pmd->nlsk_fd);
1378 	if (err < 0) {
1379 		RTE_LOG(ERR, PMD,
1380 			"Kernel refused TC filter rule creation (%d): %s\n",
1381 			errno, strerror(errno));
1382 		rte_flow_error_set(error, EEXIST, RTE_FLOW_ERROR_TYPE_HANDLE,
1383 				   NULL,
1384 				   "overlapping rules or Kernel too old for flower support");
1385 		goto fail;
1386 	}
1387 	LIST_INSERT_HEAD(&pmd->flows, flow, next);
1388 	/**
1389 	 * If a remote device is configured, a TC rule with identical items for
1390 	 * matching must be set on that device, with a single action: redirect
1391 	 * to the local pmd->if_index.
1392 	 */
1393 	if (pmd->remote_if_index) {
1394 		remote_flow = rte_malloc(__func__, sizeof(struct rte_flow), 0);
1395 		if (!remote_flow) {
1396 			rte_flow_error_set(
1397 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1398 				"cannot allocate memory for rte_flow");
1399 			goto fail;
1400 		}
1401 		msg = &remote_flow->msg;
1402 		/* set the rule if_index for the remote netdevice */
1403 		tc_init_msg(
1404 			msg, pmd->remote_if_index, RTM_NEWTFILTER,
1405 			NLM_F_REQUEST | NLM_F_ACK | NLM_F_EXCL | NLM_F_CREATE);
1406 		msg->t.tcm_info = TC_H_MAKE(0, htons(ETH_P_ALL));
1407 		tap_flow_set_handle(remote_flow);
1408 		if (priv_flow_process(pmd, attr, items, NULL,
1409 				      error, remote_flow, TCA_EGRESS_REDIR)) {
1410 			rte_flow_error_set(
1411 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1412 				NULL, "rte flow rule validation failed");
1413 			goto fail;
1414 		}
1415 		err = tap_nl_send(pmd->nlsk_fd, &msg->nh);
1416 		if (err < 0) {
1417 			rte_flow_error_set(
1418 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1419 				NULL, "Failure sending nl request");
1420 			goto fail;
1421 		}
1422 		err = tap_nl_recv_ack(pmd->nlsk_fd);
1423 		if (err < 0) {
1424 			RTE_LOG(ERR, PMD,
1425 				"Kernel refused TC filter rule creation (%d): %s\n",
1426 				errno, strerror(errno));
1427 			rte_flow_error_set(
1428 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1429 				NULL,
1430 				"overlapping rules or Kernel too old for flower support");
1431 			goto fail;
1432 		}
1433 		flow->remote_flow = remote_flow;
1434 	}
1435 	return flow;
1436 fail:
1437 	if (remote_flow)
1438 		rte_free(remote_flow);
1439 	if (flow)
1440 		tap_flow_free(pmd, flow);
1441 	return NULL;
1442 }
1443 
1444 /**
1445  * Destroy a flow using pointer to pmd_internal.
1446  *
1447  * @param[in, out] pmd
1448  *   Pointer to private structure.
1449  * @param[in] flow
1450  *   Pointer to the flow to destroy.
1451  * @param[in, out] error
1452  *   Pointer to the flow error handler
1453  *
1454  * @return 0 if the flow could be destroyed, -1 otherwise.
1455  */
1456 static int
1457 tap_flow_destroy_pmd(struct pmd_internals *pmd,
1458 		     struct rte_flow *flow,
1459 		     struct rte_flow_error *error)
1460 {
1461 	struct rte_flow *remote_flow = flow->remote_flow;
1462 	int ret = 0;
1463 
1464 	LIST_REMOVE(flow, next);
1465 	flow->msg.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
1466 	flow->msg.nh.nlmsg_type = RTM_DELTFILTER;
1467 
1468 	ret = tap_nl_send(pmd->nlsk_fd, &flow->msg.nh);
1469 	if (ret < 0) {
1470 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE,
1471 				   NULL, "couldn't send request to kernel");
1472 		goto end;
1473 	}
1474 	ret = tap_nl_recv_ack(pmd->nlsk_fd);
1475 	/* If errno is ENOENT, the rule is already no longer in the kernel. */
1476 	if (ret < 0 && errno == ENOENT)
1477 		ret = 0;
1478 	if (ret < 0) {
1479 		RTE_LOG(ERR, PMD,
1480 			"Kernel refused TC filter rule deletion (%d): %s\n",
1481 			errno, strerror(errno));
1482 		rte_flow_error_set(
1483 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1484 			"couldn't receive kernel ack to our request");
1485 		goto end;
1486 	}
1487 
1488 	if (remote_flow) {
1489 		remote_flow->msg.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
1490 		remote_flow->msg.nh.nlmsg_type = RTM_DELTFILTER;
1491 
1492 		ret = tap_nl_send(pmd->nlsk_fd, &remote_flow->msg.nh);
1493 		if (ret < 0) {
1494 			rte_flow_error_set(
1495 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1496 				NULL, "Failure sending nl request");
1497 			goto end;
1498 		}
1499 		ret = tap_nl_recv_ack(pmd->nlsk_fd);
1500 		if (ret < 0 && errno == ENOENT)
1501 			ret = 0;
1502 		if (ret < 0) {
1503 			RTE_LOG(ERR, PMD,
1504 				"Kernel refused TC filter rule deletion (%d): %s\n",
1505 				errno, strerror(errno));
1506 			rte_flow_error_set(
1507 				error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1508 				NULL, "Failure trying to receive nl ack");
1509 			goto end;
1510 		}
1511 	}
1512 end:
1513 	if (remote_flow)
1514 		rte_free(remote_flow);
1515 	tap_flow_free(pmd, flow);
1516 	return ret;
1517 }
1518 
1519 /**
1520  * Destroy a flow.
1521  *
1522  * @see rte_flow_destroy()
1523  * @see rte_flow_ops
1524  */
1525 static int
1526 tap_flow_destroy(struct rte_eth_dev *dev,
1527 		 struct rte_flow *flow,
1528 		 struct rte_flow_error *error)
1529 {
1530 	struct pmd_internals *pmd = dev->data->dev_private;
1531 
1532 	return tap_flow_destroy_pmd(pmd, flow, error);
1533 }
1534 
1535 /**
1536  * Enable/disable flow isolation.
1537  *
1538  * @see rte_flow_isolate()
1539  * @see rte_flow_ops
1540  */
1541 static int
1542 tap_flow_isolate(struct rte_eth_dev *dev,
1543 		 int set,
1544 		 struct rte_flow_error *error __rte_unused)
1545 {
1546 	struct pmd_internals *pmd = dev->data->dev_private;
1547 
1548 	if (set)
1549 		pmd->flow_isolate = 1;
1550 	else
1551 		pmd->flow_isolate = 0;
1552 	/*
1553 	 * If netdevice is there, setup appropriate flow rules immediately.
1554 	 * Otherwise it will be set when bringing up the netdevice (tun_alloc).
1555 	 */
1556 	if (!pmd->rxq[0].fd)
1557 		return 0;
1558 	if (set) {
1559 		struct rte_flow *flow;
1560 
1561 		while (1) {
1562 			flow = LIST_FIRST(&pmd->implicit_flows);
1563 			if (!flow)
1564 				break;
1565 			/*
1566 			 * Remove all implicit rules on the remote.
1567 			 * Keep the local rule to redirect packets on TX.
1568 			 * Keep also the last implicit local rule: ISOLATE.
1569 			 */
1570 			if (flow->msg.t.tcm_ifindex == pmd->if_index)
1571 				break;
1572 			if (tap_flow_destroy_pmd(pmd, flow, NULL) < 0)
1573 				goto error;
1574 		}
1575 		/* Switch the TC rule according to pmd->flow_isolate */
1576 		if (tap_flow_implicit_create(pmd, TAP_ISOLATE) == -1)
1577 			goto error;
1578 	} else {
1579 		/* Switch the TC rule according to pmd->flow_isolate */
1580 		if (tap_flow_implicit_create(pmd, TAP_ISOLATE) == -1)
1581 			goto error;
1582 		if (!pmd->remote_if_index)
1583 			return 0;
1584 		if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0)
1585 			goto error;
1586 		if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
1587 			goto error;
1588 		if (tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0)
1589 			goto error;
1590 		if (tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0)
1591 			goto error;
1592 		if (dev->data->promiscuous &&
1593 		    tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC) < 0)
1594 			goto error;
1595 		if (dev->data->all_multicast &&
1596 		    tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI) < 0)
1597 			goto error;
1598 	}
1599 	return 0;
1600 error:
1601 	pmd->flow_isolate = 0;
1602 	return rte_flow_error_set(
1603 		error, ENOTSUP, RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1604 		"TC rule creation failed");
1605 }
1606 
1607 /**
1608  * Destroy all flows.
1609  *
1610  * @see rte_flow_flush()
1611  * @see rte_flow_ops
1612  */
1613 int
1614 tap_flow_flush(struct rte_eth_dev *dev, struct rte_flow_error *error)
1615 {
1616 	struct pmd_internals *pmd = dev->data->dev_private;
1617 	struct rte_flow *flow;
1618 
1619 	while (!LIST_EMPTY(&pmd->flows)) {
1620 		flow = LIST_FIRST(&pmd->flows);
1621 		if (tap_flow_destroy(dev, flow, error) < 0)
1622 			return -1;
1623 	}
1624 	return 0;
1625 }
1626 
1627 /**
1628  * Add an implicit flow rule on the remote device to make sure traffic gets to
1629  * the tap netdevice from there.
1630  *
1631  * @param pmd
1632  *   Pointer to private structure.
1633  * @param[in] idx
1634  *   The idx in the implicit_rte_flows array specifying which rule to apply.
1635  *
1636  * @return -1 if the rule couldn't be applied, 0 otherwise.
1637  */
1638 int tap_flow_implicit_create(struct pmd_internals *pmd,
1639 			     enum implicit_rule_index idx)
1640 {
1641 	uint16_t flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_EXCL | NLM_F_CREATE;
1642 	struct rte_flow_action *actions = implicit_rte_flows[idx].actions;
1643 	struct rte_flow_action isolate_actions[2] = {
1644 		[1] = {
1645 			.type = RTE_FLOW_ACTION_TYPE_END,
1646 		},
1647 	};
1648 	struct rte_flow_item *items = implicit_rte_flows[idx].items;
1649 	struct rte_flow_attr *attr = &implicit_rte_flows[idx].attr;
1650 	struct rte_flow_item_eth eth_local = { .type = 0 };
1651 	uint16_t if_index = pmd->remote_if_index;
1652 	struct rte_flow *remote_flow = NULL;
1653 	struct nlmsg *msg = NULL;
1654 	int err = 0;
1655 	struct rte_flow_item items_local[2] = {
1656 		[0] = {
1657 			.type = items[0].type,
1658 			.spec = &eth_local,
1659 			.mask = items[0].mask,
1660 		},
1661 		[1] = {
1662 			.type = items[1].type,
1663 		}
1664 	};
1665 
1666 	remote_flow = rte_malloc(__func__, sizeof(struct rte_flow), 0);
1667 	if (!remote_flow) {
1668 		RTE_LOG(ERR, PMD, "Cannot allocate memory for rte_flow\n");
1669 		goto fail;
1670 	}
1671 	msg = &remote_flow->msg;
1672 	if (idx == TAP_REMOTE_TX) {
1673 		if_index = pmd->if_index;
1674 	} else if (idx == TAP_ISOLATE) {
1675 		if_index = pmd->if_index;
1676 		/* Don't be exclusive for this rule, it can be changed later. */
1677 		flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE;
1678 		isolate_actions[0].type = pmd->flow_isolate ?
1679 			RTE_FLOW_ACTION_TYPE_DROP :
1680 			RTE_FLOW_ACTION_TYPE_PASSTHRU;
1681 		actions = isolate_actions;
1682 	} else if (idx == TAP_REMOTE_LOCAL_MAC) {
1683 		/*
1684 		 * eth addr couldn't be set in implicit_rte_flows[] as it is not
1685 		 * known at compile time.
1686 		 */
1687 		memcpy(&eth_local.dst, &pmd->eth_addr, sizeof(pmd->eth_addr));
1688 		items = items_local;
1689 	}
1690 	tc_init_msg(msg, if_index, RTM_NEWTFILTER, flags);
1691 	msg->t.tcm_info = TC_H_MAKE(0, htons(ETH_P_ALL));
1692 	/*
1693 	 * The ISOLATE rule is always present and must have a static handle, as
1694 	 * the action is changed whether the feature is enabled (DROP) or
1695 	 * disabled (PASSTHRU).
1696 	 * There is just one REMOTE_PROMISCUOUS rule in all cases. It should
1697 	 * have a static handle such that adding it twice will fail with EEXIST
1698 	 * with any kernel version. Remark: old kernels may falsely accept the
1699 	 * same REMOTE_PROMISCUOUS rules if they had different handles.
1700 	 */
1701 	if (idx == TAP_ISOLATE)
1702 		remote_flow->msg.t.tcm_handle = ISOLATE_HANDLE;
1703 	else if (idx == TAP_REMOTE_PROMISC)
1704 		remote_flow->msg.t.tcm_handle = REMOTE_PROMISCUOUS_HANDLE;
1705 	else
1706 		tap_flow_set_handle(remote_flow);
1707 	if (priv_flow_process(pmd, attr, items, actions, NULL,
1708 			      remote_flow, implicit_rte_flows[idx].mirred)) {
1709 		RTE_LOG(ERR, PMD, "rte flow rule validation failed\n");
1710 		goto fail;
1711 	}
1712 	err = tap_nl_send(pmd->nlsk_fd, &msg->nh);
1713 	if (err < 0) {
1714 		RTE_LOG(ERR, PMD, "Failure sending nl request\n");
1715 		goto fail;
1716 	}
1717 	err = tap_nl_recv_ack(pmd->nlsk_fd);
1718 	if (err < 0) {
1719 		/* Silently ignore re-entering remote promiscuous rule */
1720 		if (errno == EEXIST && idx == TAP_REMOTE_PROMISC)
1721 			goto success;
1722 		RTE_LOG(ERR, PMD,
1723 			"Kernel refused TC filter rule creation (%d): %s\n",
1724 			errno, strerror(errno));
1725 		goto fail;
1726 	}
1727 	LIST_INSERT_HEAD(&pmd->implicit_flows, remote_flow, next);
1728 success:
1729 	return 0;
1730 fail:
1731 	if (remote_flow)
1732 		rte_free(remote_flow);
1733 	return -1;
1734 }
1735 
1736 /**
1737  * Remove specific implicit flow rule on the remote device.
1738  *
1739  * @param[in, out] pmd
1740  *   Pointer to private structure.
1741  * @param[in] idx
1742  *   The idx in the implicit_rte_flows array specifying which rule to remove.
1743  *
1744  * @return -1 if one of the implicit rules couldn't be created, 0 otherwise.
1745  */
1746 int tap_flow_implicit_destroy(struct pmd_internals *pmd,
1747 			      enum implicit_rule_index idx)
1748 {
1749 	struct rte_flow *remote_flow;
1750 	int cur_prio = -1;
1751 	int idx_prio = implicit_rte_flows[idx].attr.priority + PRIORITY_OFFSET;
1752 
1753 	for (remote_flow = LIST_FIRST(&pmd->implicit_flows);
1754 	     remote_flow;
1755 	     remote_flow = LIST_NEXT(remote_flow, next)) {
1756 		cur_prio = (remote_flow->msg.t.tcm_info >> 16) & PRIORITY_MASK;
1757 		if (cur_prio != idx_prio)
1758 			continue;
1759 		return tap_flow_destroy_pmd(pmd, remote_flow, NULL);
1760 	}
1761 	return 0;
1762 }
1763 
1764 /**
1765  * Destroy all implicit flows.
1766  *
1767  * @see rte_flow_flush()
1768  */
1769 int
1770 tap_flow_implicit_flush(struct pmd_internals *pmd, struct rte_flow_error *error)
1771 {
1772 	struct rte_flow *remote_flow;
1773 
1774 	while (!LIST_EMPTY(&pmd->implicit_flows)) {
1775 		remote_flow = LIST_FIRST(&pmd->implicit_flows);
1776 		if (tap_flow_destroy_pmd(pmd, remote_flow, error) < 0)
1777 			return -1;
1778 	}
1779 	return 0;
1780 }
1781 
1782 #define MAX_RSS_KEYS 256
1783 #define KEY_IDX_OFFSET (3 * MAX_RSS_KEYS)
1784 #define SEC_NAME_CLS_Q "cls_q"
1785 
1786 const char *sec_name[SEC_MAX] = {
1787 	[SEC_L3_L4] = "l3_l4",
1788 };
1789 
1790 /**
1791  * Enable RSS on tap: create TC rules for queuing.
1792  *
1793  * @param[in, out] pmd
1794  *   Pointer to private structure.
1795  *
1796  * @param[in] attr
1797  *   Pointer to rte_flow to get flow group
1798  *
1799  * @param[out] error
1800  *   Pointer to error reporting if not NULL.
1801  *
1802  * @return 0 on success, negative value on failure.
1803  */
1804 static int rss_enable(struct pmd_internals *pmd,
1805 			const struct rte_flow_attr *attr,
1806 			struct rte_flow_error *error)
1807 {
1808 	struct rte_flow *rss_flow = NULL;
1809 	struct nlmsg *msg = NULL;
1810 	/* 4096 is the maximum number of instructions for a BPF program */
1811 	char annotation[64];
1812 	int i;
1813 	int err = 0;
1814 
1815 	/* unlimit locked memory */
1816 	struct rlimit memlock_limit = {
1817 		.rlim_cur = RLIM_INFINITY,
1818 		.rlim_max = RLIM_INFINITY,
1819 	};
1820 	setrlimit(RLIMIT_MEMLOCK, &memlock_limit);
1821 
1822 	 /* Get a new map key for a new RSS rule */
1823 	err = bpf_rss_key(KEY_CMD_INIT, NULL);
1824 	if (err < 0) {
1825 		rte_flow_error_set(
1826 			error, EINVAL, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1827 			"Failed to initialize BPF RSS keys");
1828 
1829 		return -1;
1830 	}
1831 
1832 	/*
1833 	 *  Create BPF RSS MAP
1834 	 */
1835 	pmd->map_fd = tap_flow_bpf_rss_map_create(sizeof(__u32), /* key size */
1836 				sizeof(struct rss_key),
1837 				MAX_RSS_KEYS);
1838 	if (pmd->map_fd < 0) {
1839 		RTE_LOG(ERR, PMD,
1840 			"Failed to create BPF map (%d): %s\n",
1841 				errno, strerror(errno));
1842 		rte_flow_error_set(
1843 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1844 			"Kernel too old or not configured "
1845 			"to support BPF maps");
1846 
1847 		return -ENOTSUP;
1848 	}
1849 
1850 	/*
1851 	 * Add a rule per queue to match reclassified packets and direct them to
1852 	 * the correct queue.
1853 	 */
1854 	for (i = 0; i < pmd->dev->data->nb_rx_queues; i++) {
1855 		pmd->bpf_fd[i] = tap_flow_bpf_cls_q(i);
1856 		if (pmd->bpf_fd[i] < 0) {
1857 			RTE_LOG(ERR, PMD,
1858 				"Failed to load BPF section %s for queue %d",
1859 				SEC_NAME_CLS_Q, i);
1860 			rte_flow_error_set(
1861 				error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE,
1862 				NULL,
1863 				"Kernel too old or not configured "
1864 				"to support BPF programs loading");
1865 
1866 			return -ENOTSUP;
1867 		}
1868 
1869 		rss_flow = rte_malloc(__func__, sizeof(struct rte_flow), 0);
1870 		if (!rss_flow) {
1871 			RTE_LOG(ERR, PMD,
1872 				"Cannot allocate memory for rte_flow");
1873 			return -1;
1874 		}
1875 		msg = &rss_flow->msg;
1876 		tc_init_msg(msg, pmd->if_index, RTM_NEWTFILTER, NLM_F_REQUEST |
1877 			    NLM_F_ACK | NLM_F_EXCL | NLM_F_CREATE);
1878 		msg->t.tcm_info = TC_H_MAKE(0, htons(ETH_P_ALL));
1879 		tap_flow_set_handle(rss_flow);
1880 		uint16_t group = attr->group << GROUP_SHIFT;
1881 		uint16_t prio = group | (i + PRIORITY_OFFSET);
1882 		msg->t.tcm_info = TC_H_MAKE(prio << 16, msg->t.tcm_info);
1883 		msg->t.tcm_parent = TC_H_MAKE(MULTIQ_MAJOR_HANDLE, 0);
1884 
1885 		tap_nlattr_add(&msg->nh, TCA_KIND, sizeof("bpf"), "bpf");
1886 		if (tap_nlattr_nested_start(msg, TCA_OPTIONS) < 0)
1887 			return -1;
1888 		tap_nlattr_add32(&msg->nh, TCA_BPF_FD, pmd->bpf_fd[i]);
1889 		snprintf(annotation, sizeof(annotation), "[%s%d]",
1890 			SEC_NAME_CLS_Q, i);
1891 		tap_nlattr_add(&msg->nh, TCA_BPF_NAME, strlen(annotation) + 1,
1892 			   annotation);
1893 		/* Actions */
1894 		{
1895 			struct action_data adata = {
1896 				.id = "skbedit",
1897 				.skbedit = {
1898 					.skbedit = {
1899 						.action = TC_ACT_PIPE,
1900 					},
1901 					.queue = i,
1902 				},
1903 			};
1904 			if (add_actions(rss_flow, 1, &adata, TCA_BPF_ACT) < 0)
1905 				return -1;
1906 		}
1907 		tap_nlattr_nested_finish(msg); /* nested TCA_OPTIONS */
1908 
1909 		/* Netlink message is now ready to be sent */
1910 		if (tap_nl_send(pmd->nlsk_fd, &msg->nh) < 0)
1911 			return -1;
1912 		err = tap_nl_recv_ack(pmd->nlsk_fd);
1913 		if (err < 0) {
1914 			RTE_LOG(ERR, PMD,
1915 				"Kernel refused TC filter rule creation (%d): %s\n",
1916 				errno, strerror(errno));
1917 			return err;
1918 		}
1919 		LIST_INSERT_HEAD(&pmd->rss_flows, rss_flow, next);
1920 	}
1921 
1922 	pmd->rss_enabled = 1;
1923 	return err;
1924 }
1925 
1926 /**
1927  * Manage bpf RSS keys repository with operations: init, get, release
1928  *
1929  * @param[in] cmd
1930  *   Command on RSS keys: init, get, release
1931  *
1932  * @param[in, out] key_idx
1933  *   Pointer to RSS Key index (out for get command, in for release command)
1934  *
1935  * @return -1 if couldn't get, release or init the RSS keys, 0 otherwise.
1936  */
1937 static int bpf_rss_key(enum bpf_rss_key_e cmd, __u32 *key_idx)
1938 {
1939 	__u32 i;
1940 	int err = 0;
1941 	static __u32 num_used_keys;
1942 	static __u32 rss_keys[MAX_RSS_KEYS] = {KEY_STAT_UNSPEC};
1943 	static __u32 rss_keys_initialized;
1944 	__u32 key;
1945 
1946 	switch (cmd) {
1947 	case KEY_CMD_GET:
1948 		if (!rss_keys_initialized) {
1949 			err = -1;
1950 			break;
1951 		}
1952 
1953 		if (num_used_keys == RTE_DIM(rss_keys)) {
1954 			err = -1;
1955 			break;
1956 		}
1957 
1958 		*key_idx = num_used_keys % RTE_DIM(rss_keys);
1959 		while (rss_keys[*key_idx] == KEY_STAT_USED)
1960 			*key_idx = (*key_idx + 1) % RTE_DIM(rss_keys);
1961 
1962 		rss_keys[*key_idx] = KEY_STAT_USED;
1963 
1964 		/*
1965 		 * Add an offset to key_idx in order to handle a case of
1966 		 * RSS and non RSS flows mixture.
1967 		 * If a non RSS flow is destroyed it has an eBPF map
1968 		 * index 0 (initialized on flow creation) and might
1969 		 * unintentionally remove RSS entry 0 from eBPF map.
1970 		 * To avoid this issue, add an offset to the real index
1971 		 * during a KEY_CMD_GET operation and subtract this offset
1972 		 * during a KEY_CMD_RELEASE operation in order to restore
1973 		 * the real index.
1974 		 */
1975 		*key_idx += KEY_IDX_OFFSET;
1976 		num_used_keys++;
1977 	break;
1978 
1979 	case KEY_CMD_RELEASE:
1980 		if (!rss_keys_initialized)
1981 			break;
1982 
1983 		/*
1984 		 * Subtract offest to restore real key index
1985 		 * If a non RSS flow is falsely trying to release map
1986 		 * entry 0 - the offset subtraction will calculate the real
1987 		 * map index as an out-of-range value and the release operation
1988 		 * will be silently ignored.
1989 		 */
1990 		key = *key_idx - KEY_IDX_OFFSET;
1991 		if (key >= RTE_DIM(rss_keys))
1992 			break;
1993 
1994 		if (rss_keys[key] == KEY_STAT_USED) {
1995 			rss_keys[key] = KEY_STAT_AVAILABLE;
1996 			num_used_keys--;
1997 		}
1998 	break;
1999 
2000 	case KEY_CMD_INIT:
2001 		for (i = 0; i < RTE_DIM(rss_keys); i++)
2002 			rss_keys[i] = KEY_STAT_AVAILABLE;
2003 
2004 		rss_keys_initialized = 1;
2005 		num_used_keys = 0;
2006 	break;
2007 
2008 	case KEY_CMD_DEINIT:
2009 		for (i = 0; i < RTE_DIM(rss_keys); i++)
2010 			rss_keys[i] = KEY_STAT_UNSPEC;
2011 
2012 		rss_keys_initialized = 0;
2013 		num_used_keys = 0;
2014 	break;
2015 
2016 	default:
2017 		break;
2018 	}
2019 
2020 	return err;
2021 }
2022 
2023 /**
2024  * Add RSS hash calculations and queue selection
2025  *
2026  * @param[in, out] pmd
2027  *   Pointer to internal structure. Used to set/get RSS map fd
2028  *
2029  * @param[in] rss
2030  *   Pointer to RSS flow actions
2031  *
2032  * @param[out] error
2033  *   Pointer to error reporting if not NULL.
2034  *
2035  * @return 0 on success, negative value on failure
2036  */
2037 static int rss_add_actions(struct rte_flow *flow, struct pmd_internals *pmd,
2038 			   const struct rte_flow_action_rss *rss,
2039 			   struct rte_flow_error *error)
2040 {
2041 	/* 4096 is the maximum number of instructions for a BPF program */
2042 	int i;
2043 	int err;
2044 	struct rss_key rss_entry = { .hash_fields = 0,
2045 				     .key_size = 0 };
2046 
2047 	/* Get a new map key for a new RSS rule */
2048 	err = bpf_rss_key(KEY_CMD_GET, &flow->key_idx);
2049 	if (err < 0) {
2050 		rte_flow_error_set(
2051 			error, EINVAL, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2052 			"Failed to get BPF RSS key");
2053 
2054 		return -1;
2055 	}
2056 
2057 	/* Update RSS map entry with queues */
2058 	rss_entry.nb_queues = rss->num;
2059 	for (i = 0; i < rss->num; i++)
2060 		rss_entry.queues[i] = rss->queue[i];
2061 	rss_entry.hash_fields =
2062 		(1 << HASH_FIELD_IPV4_L3_L4) | (1 << HASH_FIELD_IPV6_L3_L4);
2063 
2064 	/* Add this RSS entry to map */
2065 	err = tap_flow_bpf_update_rss_elem(pmd->map_fd,
2066 				&flow->key_idx, &rss_entry);
2067 
2068 	if (err) {
2069 		RTE_LOG(ERR, PMD,
2070 			"Failed to update BPF map entry #%u (%d): %s\n",
2071 			flow->key_idx, errno, strerror(errno));
2072 		rte_flow_error_set(
2073 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2074 			"Kernel too old or not configured "
2075 			"to support BPF maps updates");
2076 
2077 		return -ENOTSUP;
2078 	}
2079 
2080 
2081 	/*
2082 	 * Load bpf rules to calculate hash for this key_idx
2083 	 */
2084 
2085 	flow->bpf_fd[SEC_L3_L4] =
2086 		tap_flow_bpf_calc_l3_l4_hash(flow->key_idx, pmd->map_fd);
2087 	if (flow->bpf_fd[SEC_L3_L4] < 0) {
2088 		RTE_LOG(ERR, PMD,
2089 			"Failed to load BPF section %s (%d): %s\n",
2090 				sec_name[SEC_L3_L4], errno, strerror(errno));
2091 		rte_flow_error_set(
2092 			error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2093 			"Kernel too old or not configured "
2094 			"to support BPF program loading");
2095 
2096 		return -ENOTSUP;
2097 	}
2098 
2099 	/* Actions */
2100 	{
2101 		struct action_data adata[] = {
2102 			{
2103 				.id = "bpf",
2104 				.bpf = {
2105 					.bpf_fd = flow->bpf_fd[SEC_L3_L4],
2106 					.annotation = sec_name[SEC_L3_L4],
2107 					.bpf = {
2108 						.action = TC_ACT_PIPE,
2109 					},
2110 				},
2111 			},
2112 		};
2113 
2114 		if (add_actions(flow, RTE_DIM(adata), adata,
2115 			TCA_FLOWER_ACT) < 0)
2116 			return -1;
2117 	}
2118 
2119 	return 0;
2120 }
2121 
2122 /**
2123  * Manage filter operations.
2124  *
2125  * @param dev
2126  *   Pointer to Ethernet device structure.
2127  * @param filter_type
2128  *   Filter type.
2129  * @param filter_op
2130  *   Operation to perform.
2131  * @param arg
2132  *   Pointer to operation-specific structure.
2133  *
2134  * @return
2135  *   0 on success, negative errno value on failure.
2136  */
2137 int
2138 tap_dev_filter_ctrl(struct rte_eth_dev *dev,
2139 		    enum rte_filter_type filter_type,
2140 		    enum rte_filter_op filter_op,
2141 		    void *arg)
2142 {
2143 	switch (filter_type) {
2144 	case RTE_ETH_FILTER_GENERIC:
2145 		if (filter_op != RTE_ETH_FILTER_GET)
2146 			return -EINVAL;
2147 		*(const void **)arg = &tap_flow_ops;
2148 		return 0;
2149 	default:
2150 		RTE_LOG(ERR, PMD, "%p: filter type (%d) not supported\n",
2151 			(void *)dev, filter_type);
2152 	}
2153 	return -EINVAL;
2154 }
2155 
2156