net/sfc: ignore spec bits not covered by mask
[dpdk.git] / drivers / net / sfc / sfc_flow.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2017-2018 Solarflare Communications Inc.
4  * All rights reserved.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9
10 #include <rte_byteorder.h>
11 #include <rte_tailq.h>
12 #include <rte_common.h>
13 #include <rte_ethdev_driver.h>
14 #include <rte_eth_ctrl.h>
15 #include <rte_ether.h>
16 #include <rte_flow.h>
17 #include <rte_flow_driver.h>
18
19 #include "efx.h"
20
21 #include "sfc.h"
22 #include "sfc_rx.h"
23 #include "sfc_filter.h"
24 #include "sfc_flow.h"
25 #include "sfc_log.h"
26
27 /*
28  * At now flow API is implemented in such a manner that each
29  * flow rule is converted to one or more hardware filters.
30  * All elements of flow rule (attributes, pattern items, actions)
31  * correspond to one or more fields in the efx_filter_spec_s structure
32  * that is responsible for the hardware filter.
33  * If some required field is unset in the flow rule, then a handful
34  * of filter copies will be created to cover all possible values
35  * of such a field.
36  */
37
38 enum sfc_flow_item_layers {
39         SFC_FLOW_ITEM_ANY_LAYER,
40         SFC_FLOW_ITEM_START_LAYER,
41         SFC_FLOW_ITEM_L2,
42         SFC_FLOW_ITEM_L3,
43         SFC_FLOW_ITEM_L4,
44 };
45
46 typedef int (sfc_flow_item_parse)(const struct rte_flow_item *item,
47                                   efx_filter_spec_t *spec,
48                                   struct rte_flow_error *error);
49
50 struct sfc_flow_item {
51         enum rte_flow_item_type type;           /* Type of item */
52         enum sfc_flow_item_layers layer;        /* Layer of item */
53         enum sfc_flow_item_layers prev_layer;   /* Previous layer of item */
54         sfc_flow_item_parse *parse;             /* Parsing function */
55 };
56
57 static sfc_flow_item_parse sfc_flow_parse_void;
58 static sfc_flow_item_parse sfc_flow_parse_eth;
59 static sfc_flow_item_parse sfc_flow_parse_vlan;
60 static sfc_flow_item_parse sfc_flow_parse_ipv4;
61 static sfc_flow_item_parse sfc_flow_parse_ipv6;
62 static sfc_flow_item_parse sfc_flow_parse_tcp;
63 static sfc_flow_item_parse sfc_flow_parse_udp;
64 static sfc_flow_item_parse sfc_flow_parse_vxlan;
65 static sfc_flow_item_parse sfc_flow_parse_geneve;
66 static sfc_flow_item_parse sfc_flow_parse_nvgre;
67
68 typedef int (sfc_flow_spec_set_vals)(struct sfc_flow_spec *spec,
69                                      unsigned int filters_count_for_one_val,
70                                      struct rte_flow_error *error);
71
72 typedef boolean_t (sfc_flow_spec_check)(efx_filter_match_flags_t match,
73                                         efx_filter_spec_t *spec,
74                                         struct sfc_filter *filter);
75
76 struct sfc_flow_copy_flag {
77         /* EFX filter specification match flag */
78         efx_filter_match_flags_t flag;
79         /* Number of values of corresponding field */
80         unsigned int vals_count;
81         /* Function to set values in specifications */
82         sfc_flow_spec_set_vals *set_vals;
83         /*
84          * Function to check that the specification is suitable
85          * for adding this match flag
86          */
87         sfc_flow_spec_check *spec_check;
88 };
89
90 static sfc_flow_spec_set_vals sfc_flow_set_unknown_dst_flags;
91 static sfc_flow_spec_check sfc_flow_check_unknown_dst_flags;
92 static sfc_flow_spec_set_vals sfc_flow_set_ethertypes;
93 static sfc_flow_spec_set_vals sfc_flow_set_ifrm_unknown_dst_flags;
94 static sfc_flow_spec_check sfc_flow_check_ifrm_unknown_dst_flags;
95
96 static boolean_t
97 sfc_flow_is_zero(const uint8_t *buf, unsigned int size)
98 {
99         uint8_t sum = 0;
100         unsigned int i;
101
102         for (i = 0; i < size; i++)
103                 sum |= buf[i];
104
105         return (sum == 0) ? B_TRUE : B_FALSE;
106 }
107
108 /*
109  * Validate item and prepare structures spec and mask for parsing
110  */
111 static int
112 sfc_flow_parse_init(const struct rte_flow_item *item,
113                     const void **spec_ptr,
114                     const void **mask_ptr,
115                     const void *supp_mask,
116                     const void *def_mask,
117                     unsigned int size,
118                     struct rte_flow_error *error)
119 {
120         const uint8_t *spec;
121         const uint8_t *mask;
122         const uint8_t *last;
123         uint8_t supp;
124         unsigned int i;
125
126         if (item == NULL) {
127                 rte_flow_error_set(error, EINVAL,
128                                    RTE_FLOW_ERROR_TYPE_ITEM, NULL,
129                                    "NULL item");
130                 return -rte_errno;
131         }
132
133         if ((item->last != NULL || item->mask != NULL) && item->spec == NULL) {
134                 rte_flow_error_set(error, EINVAL,
135                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
136                                    "Mask or last is set without spec");
137                 return -rte_errno;
138         }
139
140         /*
141          * If "mask" is not set, default mask is used,
142          * but if default mask is NULL, "mask" should be set
143          */
144         if (item->mask == NULL) {
145                 if (def_mask == NULL) {
146                         rte_flow_error_set(error, EINVAL,
147                                 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
148                                 "Mask should be specified");
149                         return -rte_errno;
150                 }
151
152                 mask = def_mask;
153         } else {
154                 mask = item->mask;
155         }
156
157         spec = item->spec;
158         last = item->last;
159
160         if (spec == NULL)
161                 goto exit;
162
163         /*
164          * If field values in "last" are either 0 or equal to the corresponding
165          * values in "spec" then they are ignored
166          */
167         if (last != NULL &&
168             !sfc_flow_is_zero(last, size) &&
169             memcmp(last, spec, size) != 0) {
170                 rte_flow_error_set(error, ENOTSUP,
171                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
172                                    "Ranging is not supported");
173                 return -rte_errno;
174         }
175
176         if (supp_mask == NULL) {
177                 rte_flow_error_set(error, EINVAL,
178                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
179                         "Supported mask for item should be specified");
180                 return -rte_errno;
181         }
182
183         /* Check that mask does not ask for more match than supp_mask */
184         for (i = 0; i < size; i++) {
185                 supp = ((const uint8_t *)supp_mask)[i];
186
187                 if (~supp & mask[i]) {
188                         rte_flow_error_set(error, ENOTSUP,
189                                            RTE_FLOW_ERROR_TYPE_ITEM, item,
190                                            "Item's field is not supported");
191                         return -rte_errno;
192                 }
193         }
194
195 exit:
196         *spec_ptr = spec;
197         *mask_ptr = mask;
198         return 0;
199 }
200
201 /*
202  * Protocol parsers.
203  * Masking is not supported, so masks in items should be either
204  * full or empty (zeroed) and set only for supported fields which
205  * are specified in the supp_mask.
206  */
207
208 static int
209 sfc_flow_parse_void(__rte_unused const struct rte_flow_item *item,
210                     __rte_unused efx_filter_spec_t *efx_spec,
211                     __rte_unused struct rte_flow_error *error)
212 {
213         return 0;
214 }
215
216 /**
217  * Convert Ethernet item to EFX filter specification.
218  *
219  * @param item[in]
220  *   Item specification. Outer frame specification may only comprise
221  *   source/destination addresses and Ethertype field.
222  *   Inner frame specification may contain destination address only.
223  *   There is support for individual/group mask as well as for empty and full.
224  *   If the mask is NULL, default mask will be used. Ranging is not supported.
225  * @param efx_spec[in, out]
226  *   EFX filter specification to update.
227  * @param[out] error
228  *   Perform verbose error reporting if not NULL.
229  */
230 static int
231 sfc_flow_parse_eth(const struct rte_flow_item *item,
232                    efx_filter_spec_t *efx_spec,
233                    struct rte_flow_error *error)
234 {
235         int rc;
236         const struct rte_flow_item_eth *spec = NULL;
237         const struct rte_flow_item_eth *mask = NULL;
238         const struct rte_flow_item_eth supp_mask = {
239                 .dst.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
240                 .src.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
241                 .type = 0xffff,
242         };
243         const struct rte_flow_item_eth ifrm_supp_mask = {
244                 .dst.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
245         };
246         const uint8_t ig_mask[EFX_MAC_ADDR_LEN] = {
247                 0x01, 0x00, 0x00, 0x00, 0x00, 0x00
248         };
249         const struct rte_flow_item_eth *supp_mask_p;
250         const struct rte_flow_item_eth *def_mask_p;
251         uint8_t *loc_mac = NULL;
252         boolean_t is_ifrm = (efx_spec->efs_encap_type !=
253                 EFX_TUNNEL_PROTOCOL_NONE);
254
255         if (is_ifrm) {
256                 supp_mask_p = &ifrm_supp_mask;
257                 def_mask_p = &ifrm_supp_mask;
258                 loc_mac = efx_spec->efs_ifrm_loc_mac;
259         } else {
260                 supp_mask_p = &supp_mask;
261                 def_mask_p = &rte_flow_item_eth_mask;
262                 loc_mac = efx_spec->efs_loc_mac;
263         }
264
265         rc = sfc_flow_parse_init(item,
266                                  (const void **)&spec,
267                                  (const void **)&mask,
268                                  supp_mask_p, def_mask_p,
269                                  sizeof(struct rte_flow_item_eth),
270                                  error);
271         if (rc != 0)
272                 return rc;
273
274         /* If "spec" is not set, could be any Ethernet */
275         if (spec == NULL)
276                 return 0;
277
278         if (is_same_ether_addr(&mask->dst, &supp_mask.dst)) {
279                 efx_spec->efs_match_flags |= is_ifrm ?
280                         EFX_FILTER_MATCH_IFRM_LOC_MAC :
281                         EFX_FILTER_MATCH_LOC_MAC;
282                 rte_memcpy(loc_mac, spec->dst.addr_bytes,
283                            EFX_MAC_ADDR_LEN);
284         } else if (memcmp(mask->dst.addr_bytes, ig_mask,
285                           EFX_MAC_ADDR_LEN) == 0) {
286                 if (is_unicast_ether_addr(&spec->dst))
287                         efx_spec->efs_match_flags |= is_ifrm ?
288                                 EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST :
289                                 EFX_FILTER_MATCH_UNKNOWN_UCAST_DST;
290                 else
291                         efx_spec->efs_match_flags |= is_ifrm ?
292                                 EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST :
293                                 EFX_FILTER_MATCH_UNKNOWN_MCAST_DST;
294         } else if (!is_zero_ether_addr(&mask->dst)) {
295                 goto fail_bad_mask;
296         }
297
298         /*
299          * ifrm_supp_mask ensures that the source address and
300          * ethertype masks are equal to zero in inner frame,
301          * so these fields are filled in only for the outer frame
302          */
303         if (is_same_ether_addr(&mask->src, &supp_mask.src)) {
304                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_MAC;
305                 rte_memcpy(efx_spec->efs_rem_mac, spec->src.addr_bytes,
306                            EFX_MAC_ADDR_LEN);
307         } else if (!is_zero_ether_addr(&mask->src)) {
308                 goto fail_bad_mask;
309         }
310
311         /*
312          * Ether type is in big-endian byte order in item and
313          * in little-endian in efx_spec, so byte swap is used
314          */
315         if (mask->type == supp_mask.type) {
316                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
317                 efx_spec->efs_ether_type = rte_bswap16(spec->type);
318         } else if (mask->type != 0) {
319                 goto fail_bad_mask;
320         }
321
322         return 0;
323
324 fail_bad_mask:
325         rte_flow_error_set(error, EINVAL,
326                            RTE_FLOW_ERROR_TYPE_ITEM, item,
327                            "Bad mask in the ETH pattern item");
328         return -rte_errno;
329 }
330
331 /**
332  * Convert VLAN item to EFX filter specification.
333  *
334  * @param item[in]
335  *   Item specification. Only VID field is supported.
336  *   The mask can not be NULL. Ranging is not supported.
337  * @param efx_spec[in, out]
338  *   EFX filter specification to update.
339  * @param[out] error
340  *   Perform verbose error reporting if not NULL.
341  */
342 static int
343 sfc_flow_parse_vlan(const struct rte_flow_item *item,
344                     efx_filter_spec_t *efx_spec,
345                     struct rte_flow_error *error)
346 {
347         int rc;
348         uint16_t vid;
349         const struct rte_flow_item_vlan *spec = NULL;
350         const struct rte_flow_item_vlan *mask = NULL;
351         const struct rte_flow_item_vlan supp_mask = {
352                 .tci = rte_cpu_to_be_16(ETH_VLAN_ID_MAX),
353                 .inner_type = RTE_BE16(0xffff),
354         };
355
356         rc = sfc_flow_parse_init(item,
357                                  (const void **)&spec,
358                                  (const void **)&mask,
359                                  &supp_mask,
360                                  NULL,
361                                  sizeof(struct rte_flow_item_vlan),
362                                  error);
363         if (rc != 0)
364                 return rc;
365
366         /*
367          * VID is in big-endian byte order in item and
368          * in little-endian in efx_spec, so byte swap is used.
369          * If two VLAN items are included, the first matches
370          * the outer tag and the next matches the inner tag.
371          */
372         if (mask->tci == supp_mask.tci) {
373                 vid = rte_bswap16(spec->tci);
374
375                 if (!(efx_spec->efs_match_flags &
376                       EFX_FILTER_MATCH_OUTER_VID)) {
377                         efx_spec->efs_match_flags |= EFX_FILTER_MATCH_OUTER_VID;
378                         efx_spec->efs_outer_vid = vid;
379                 } else if (!(efx_spec->efs_match_flags &
380                              EFX_FILTER_MATCH_INNER_VID)) {
381                         efx_spec->efs_match_flags |= EFX_FILTER_MATCH_INNER_VID;
382                         efx_spec->efs_inner_vid = vid;
383                 } else {
384                         rte_flow_error_set(error, EINVAL,
385                                            RTE_FLOW_ERROR_TYPE_ITEM, item,
386                                            "More than two VLAN items");
387                         return -rte_errno;
388                 }
389         } else {
390                 rte_flow_error_set(error, EINVAL,
391                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
392                                    "VLAN ID in TCI match is required");
393                 return -rte_errno;
394         }
395
396         if (efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE) {
397                 rte_flow_error_set(error, EINVAL,
398                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
399                                    "VLAN TPID matching is not supported");
400                 return -rte_errno;
401         }
402         if (mask->inner_type == supp_mask.inner_type) {
403                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
404                 efx_spec->efs_ether_type = rte_bswap16(spec->inner_type);
405         } else if (mask->inner_type) {
406                 rte_flow_error_set(error, EINVAL,
407                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
408                                    "Bad mask for VLAN inner_type");
409                 return -rte_errno;
410         }
411
412         return 0;
413 }
414
415 /**
416  * Convert IPv4 item to EFX filter specification.
417  *
418  * @param item[in]
419  *   Item specification. Only source and destination addresses and
420  *   protocol fields are supported. If the mask is NULL, default
421  *   mask will be used. Ranging is not supported.
422  * @param efx_spec[in, out]
423  *   EFX filter specification to update.
424  * @param[out] error
425  *   Perform verbose error reporting if not NULL.
426  */
427 static int
428 sfc_flow_parse_ipv4(const struct rte_flow_item *item,
429                     efx_filter_spec_t *efx_spec,
430                     struct rte_flow_error *error)
431 {
432         int rc;
433         const struct rte_flow_item_ipv4 *spec = NULL;
434         const struct rte_flow_item_ipv4 *mask = NULL;
435         const uint16_t ether_type_ipv4 = rte_cpu_to_le_16(EFX_ETHER_TYPE_IPV4);
436         const struct rte_flow_item_ipv4 supp_mask = {
437                 .hdr = {
438                         .src_addr = 0xffffffff,
439                         .dst_addr = 0xffffffff,
440                         .next_proto_id = 0xff,
441                 }
442         };
443
444         rc = sfc_flow_parse_init(item,
445                                  (const void **)&spec,
446                                  (const void **)&mask,
447                                  &supp_mask,
448                                  &rte_flow_item_ipv4_mask,
449                                  sizeof(struct rte_flow_item_ipv4),
450                                  error);
451         if (rc != 0)
452                 return rc;
453
454         /*
455          * Filtering by IPv4 source and destination addresses requires
456          * the appropriate ETHER_TYPE in hardware filters
457          */
458         if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE)) {
459                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
460                 efx_spec->efs_ether_type = ether_type_ipv4;
461         } else if (efx_spec->efs_ether_type != ether_type_ipv4) {
462                 rte_flow_error_set(error, EINVAL,
463                         RTE_FLOW_ERROR_TYPE_ITEM, item,
464                         "Ethertype in pattern with IPV4 item should be appropriate");
465                 return -rte_errno;
466         }
467
468         if (spec == NULL)
469                 return 0;
470
471         /*
472          * IPv4 addresses are in big-endian byte order in item and in
473          * efx_spec
474          */
475         if (mask->hdr.src_addr == supp_mask.hdr.src_addr) {
476                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_HOST;
477                 efx_spec->efs_rem_host.eo_u32[0] = spec->hdr.src_addr;
478         } else if (mask->hdr.src_addr != 0) {
479                 goto fail_bad_mask;
480         }
481
482         if (mask->hdr.dst_addr == supp_mask.hdr.dst_addr) {
483                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_HOST;
484                 efx_spec->efs_loc_host.eo_u32[0] = spec->hdr.dst_addr;
485         } else if (mask->hdr.dst_addr != 0) {
486                 goto fail_bad_mask;
487         }
488
489         if (mask->hdr.next_proto_id == supp_mask.hdr.next_proto_id) {
490                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
491                 efx_spec->efs_ip_proto = spec->hdr.next_proto_id;
492         } else if (mask->hdr.next_proto_id != 0) {
493                 goto fail_bad_mask;
494         }
495
496         return 0;
497
498 fail_bad_mask:
499         rte_flow_error_set(error, EINVAL,
500                            RTE_FLOW_ERROR_TYPE_ITEM, item,
501                            "Bad mask in the IPV4 pattern item");
502         return -rte_errno;
503 }
504
505 /**
506  * Convert IPv6 item to EFX filter specification.
507  *
508  * @param item[in]
509  *   Item specification. Only source and destination addresses and
510  *   next header fields are supported. If the mask is NULL, default
511  *   mask will be used. Ranging is not supported.
512  * @param efx_spec[in, out]
513  *   EFX filter specification to update.
514  * @param[out] error
515  *   Perform verbose error reporting if not NULL.
516  */
517 static int
518 sfc_flow_parse_ipv6(const struct rte_flow_item *item,
519                     efx_filter_spec_t *efx_spec,
520                     struct rte_flow_error *error)
521 {
522         int rc;
523         const struct rte_flow_item_ipv6 *spec = NULL;
524         const struct rte_flow_item_ipv6 *mask = NULL;
525         const uint16_t ether_type_ipv6 = rte_cpu_to_le_16(EFX_ETHER_TYPE_IPV6);
526         const struct rte_flow_item_ipv6 supp_mask = {
527                 .hdr = {
528                         .src_addr = { 0xff, 0xff, 0xff, 0xff,
529                                       0xff, 0xff, 0xff, 0xff,
530                                       0xff, 0xff, 0xff, 0xff,
531                                       0xff, 0xff, 0xff, 0xff },
532                         .dst_addr = { 0xff, 0xff, 0xff, 0xff,
533                                       0xff, 0xff, 0xff, 0xff,
534                                       0xff, 0xff, 0xff, 0xff,
535                                       0xff, 0xff, 0xff, 0xff },
536                         .proto = 0xff,
537                 }
538         };
539
540         rc = sfc_flow_parse_init(item,
541                                  (const void **)&spec,
542                                  (const void **)&mask,
543                                  &supp_mask,
544                                  &rte_flow_item_ipv6_mask,
545                                  sizeof(struct rte_flow_item_ipv6),
546                                  error);
547         if (rc != 0)
548                 return rc;
549
550         /*
551          * Filtering by IPv6 source and destination addresses requires
552          * the appropriate ETHER_TYPE in hardware filters
553          */
554         if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE)) {
555                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
556                 efx_spec->efs_ether_type = ether_type_ipv6;
557         } else if (efx_spec->efs_ether_type != ether_type_ipv6) {
558                 rte_flow_error_set(error, EINVAL,
559                         RTE_FLOW_ERROR_TYPE_ITEM, item,
560                         "Ethertype in pattern with IPV6 item should be appropriate");
561                 return -rte_errno;
562         }
563
564         if (spec == NULL)
565                 return 0;
566
567         /*
568          * IPv6 addresses are in big-endian byte order in item and in
569          * efx_spec
570          */
571         if (memcmp(mask->hdr.src_addr, supp_mask.hdr.src_addr,
572                    sizeof(mask->hdr.src_addr)) == 0) {
573                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_HOST;
574
575                 RTE_BUILD_BUG_ON(sizeof(efx_spec->efs_rem_host) !=
576                                  sizeof(spec->hdr.src_addr));
577                 rte_memcpy(&efx_spec->efs_rem_host, spec->hdr.src_addr,
578                            sizeof(efx_spec->efs_rem_host));
579         } else if (!sfc_flow_is_zero(mask->hdr.src_addr,
580                                      sizeof(mask->hdr.src_addr))) {
581                 goto fail_bad_mask;
582         }
583
584         if (memcmp(mask->hdr.dst_addr, supp_mask.hdr.dst_addr,
585                    sizeof(mask->hdr.dst_addr)) == 0) {
586                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_HOST;
587
588                 RTE_BUILD_BUG_ON(sizeof(efx_spec->efs_loc_host) !=
589                                  sizeof(spec->hdr.dst_addr));
590                 rte_memcpy(&efx_spec->efs_loc_host, spec->hdr.dst_addr,
591                            sizeof(efx_spec->efs_loc_host));
592         } else if (!sfc_flow_is_zero(mask->hdr.dst_addr,
593                                      sizeof(mask->hdr.dst_addr))) {
594                 goto fail_bad_mask;
595         }
596
597         if (mask->hdr.proto == supp_mask.hdr.proto) {
598                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
599                 efx_spec->efs_ip_proto = spec->hdr.proto;
600         } else if (mask->hdr.proto != 0) {
601                 goto fail_bad_mask;
602         }
603
604         return 0;
605
606 fail_bad_mask:
607         rte_flow_error_set(error, EINVAL,
608                            RTE_FLOW_ERROR_TYPE_ITEM, item,
609                            "Bad mask in the IPV6 pattern item");
610         return -rte_errno;
611 }
612
613 /**
614  * Convert TCP item to EFX filter specification.
615  *
616  * @param item[in]
617  *   Item specification. Only source and destination ports fields
618  *   are supported. If the mask is NULL, default mask will be used.
619  *   Ranging is not supported.
620  * @param efx_spec[in, out]
621  *   EFX filter specification to update.
622  * @param[out] error
623  *   Perform verbose error reporting if not NULL.
624  */
625 static int
626 sfc_flow_parse_tcp(const struct rte_flow_item *item,
627                    efx_filter_spec_t *efx_spec,
628                    struct rte_flow_error *error)
629 {
630         int rc;
631         const struct rte_flow_item_tcp *spec = NULL;
632         const struct rte_flow_item_tcp *mask = NULL;
633         const struct rte_flow_item_tcp supp_mask = {
634                 .hdr = {
635                         .src_port = 0xffff,
636                         .dst_port = 0xffff,
637                 }
638         };
639
640         rc = sfc_flow_parse_init(item,
641                                  (const void **)&spec,
642                                  (const void **)&mask,
643                                  &supp_mask,
644                                  &rte_flow_item_tcp_mask,
645                                  sizeof(struct rte_flow_item_tcp),
646                                  error);
647         if (rc != 0)
648                 return rc;
649
650         /*
651          * Filtering by TCP source and destination ports requires
652          * the appropriate IP_PROTO in hardware filters
653          */
654         if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
655                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
656                 efx_spec->efs_ip_proto = EFX_IPPROTO_TCP;
657         } else if (efx_spec->efs_ip_proto != EFX_IPPROTO_TCP) {
658                 rte_flow_error_set(error, EINVAL,
659                         RTE_FLOW_ERROR_TYPE_ITEM, item,
660                         "IP proto in pattern with TCP item should be appropriate");
661                 return -rte_errno;
662         }
663
664         if (spec == NULL)
665                 return 0;
666
667         /*
668          * Source and destination ports are in big-endian byte order in item and
669          * in little-endian in efx_spec, so byte swap is used
670          */
671         if (mask->hdr.src_port == supp_mask.hdr.src_port) {
672                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_PORT;
673                 efx_spec->efs_rem_port = rte_bswap16(spec->hdr.src_port);
674         } else if (mask->hdr.src_port != 0) {
675                 goto fail_bad_mask;
676         }
677
678         if (mask->hdr.dst_port == supp_mask.hdr.dst_port) {
679                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_PORT;
680                 efx_spec->efs_loc_port = rte_bswap16(spec->hdr.dst_port);
681         } else if (mask->hdr.dst_port != 0) {
682                 goto fail_bad_mask;
683         }
684
685         return 0;
686
687 fail_bad_mask:
688         rte_flow_error_set(error, EINVAL,
689                            RTE_FLOW_ERROR_TYPE_ITEM, item,
690                            "Bad mask in the TCP pattern item");
691         return -rte_errno;
692 }
693
694 /**
695  * Convert UDP item to EFX filter specification.
696  *
697  * @param item[in]
698  *   Item specification. Only source and destination ports fields
699  *   are supported. If the mask is NULL, default mask will be used.
700  *   Ranging is not supported.
701  * @param efx_spec[in, out]
702  *   EFX filter specification to update.
703  * @param[out] error
704  *   Perform verbose error reporting if not NULL.
705  */
706 static int
707 sfc_flow_parse_udp(const struct rte_flow_item *item,
708                    efx_filter_spec_t *efx_spec,
709                    struct rte_flow_error *error)
710 {
711         int rc;
712         const struct rte_flow_item_udp *spec = NULL;
713         const struct rte_flow_item_udp *mask = NULL;
714         const struct rte_flow_item_udp supp_mask = {
715                 .hdr = {
716                         .src_port = 0xffff,
717                         .dst_port = 0xffff,
718                 }
719         };
720
721         rc = sfc_flow_parse_init(item,
722                                  (const void **)&spec,
723                                  (const void **)&mask,
724                                  &supp_mask,
725                                  &rte_flow_item_udp_mask,
726                                  sizeof(struct rte_flow_item_udp),
727                                  error);
728         if (rc != 0)
729                 return rc;
730
731         /*
732          * Filtering by UDP source and destination ports requires
733          * the appropriate IP_PROTO in hardware filters
734          */
735         if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
736                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
737                 efx_spec->efs_ip_proto = EFX_IPPROTO_UDP;
738         } else if (efx_spec->efs_ip_proto != EFX_IPPROTO_UDP) {
739                 rte_flow_error_set(error, EINVAL,
740                         RTE_FLOW_ERROR_TYPE_ITEM, item,
741                         "IP proto in pattern with UDP item should be appropriate");
742                 return -rte_errno;
743         }
744
745         if (spec == NULL)
746                 return 0;
747
748         /*
749          * Source and destination ports are in big-endian byte order in item and
750          * in little-endian in efx_spec, so byte swap is used
751          */
752         if (mask->hdr.src_port == supp_mask.hdr.src_port) {
753                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_PORT;
754                 efx_spec->efs_rem_port = rte_bswap16(spec->hdr.src_port);
755         } else if (mask->hdr.src_port != 0) {
756                 goto fail_bad_mask;
757         }
758
759         if (mask->hdr.dst_port == supp_mask.hdr.dst_port) {
760                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_PORT;
761                 efx_spec->efs_loc_port = rte_bswap16(spec->hdr.dst_port);
762         } else if (mask->hdr.dst_port != 0) {
763                 goto fail_bad_mask;
764         }
765
766         return 0;
767
768 fail_bad_mask:
769         rte_flow_error_set(error, EINVAL,
770                            RTE_FLOW_ERROR_TYPE_ITEM, item,
771                            "Bad mask in the UDP pattern item");
772         return -rte_errno;
773 }
774
775 /*
776  * Filters for encapsulated packets match based on the EtherType and IP
777  * protocol in the outer frame.
778  */
779 static int
780 sfc_flow_set_match_flags_for_encap_pkts(const struct rte_flow_item *item,
781                                         efx_filter_spec_t *efx_spec,
782                                         uint8_t ip_proto,
783                                         struct rte_flow_error *error)
784 {
785         if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
786                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
787                 efx_spec->efs_ip_proto = ip_proto;
788         } else if (efx_spec->efs_ip_proto != ip_proto) {
789                 switch (ip_proto) {
790                 case EFX_IPPROTO_UDP:
791                         rte_flow_error_set(error, EINVAL,
792                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
793                                 "Outer IP header protocol must be UDP "
794                                 "in VxLAN/GENEVE pattern");
795                         return -rte_errno;
796
797                 case EFX_IPPROTO_GRE:
798                         rte_flow_error_set(error, EINVAL,
799                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
800                                 "Outer IP header protocol must be GRE "
801                                 "in NVGRE pattern");
802                         return -rte_errno;
803
804                 default:
805                         rte_flow_error_set(error, EINVAL,
806                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
807                                 "Only VxLAN/GENEVE/NVGRE tunneling patterns "
808                                 "are supported");
809                         return -rte_errno;
810                 }
811         }
812
813         if (efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE &&
814             efx_spec->efs_ether_type != EFX_ETHER_TYPE_IPV4 &&
815             efx_spec->efs_ether_type != EFX_ETHER_TYPE_IPV6) {
816                 rte_flow_error_set(error, EINVAL,
817                         RTE_FLOW_ERROR_TYPE_ITEM, item,
818                         "Outer frame EtherType in pattern with tunneling "
819                         "must be IPv4 or IPv6");
820                 return -rte_errno;
821         }
822
823         return 0;
824 }
825
826 static int
827 sfc_flow_set_efx_spec_vni_or_vsid(efx_filter_spec_t *efx_spec,
828                                   const uint8_t *vni_or_vsid_val,
829                                   const uint8_t *vni_or_vsid_mask,
830                                   const struct rte_flow_item *item,
831                                   struct rte_flow_error *error)
832 {
833         const uint8_t vni_or_vsid_full_mask[EFX_VNI_OR_VSID_LEN] = {
834                 0xff, 0xff, 0xff
835         };
836
837         if (memcmp(vni_or_vsid_mask, vni_or_vsid_full_mask,
838                    EFX_VNI_OR_VSID_LEN) == 0) {
839                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_VNI_OR_VSID;
840                 rte_memcpy(efx_spec->efs_vni_or_vsid, vni_or_vsid_val,
841                            EFX_VNI_OR_VSID_LEN);
842         } else if (!sfc_flow_is_zero(vni_or_vsid_mask, EFX_VNI_OR_VSID_LEN)) {
843                 rte_flow_error_set(error, EINVAL,
844                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
845                                    "Unsupported VNI/VSID mask");
846                 return -rte_errno;
847         }
848
849         return 0;
850 }
851
852 /**
853  * Convert VXLAN item to EFX filter specification.
854  *
855  * @param item[in]
856  *   Item specification. Only VXLAN network identifier field is supported.
857  *   If the mask is NULL, default mask will be used.
858  *   Ranging is not supported.
859  * @param efx_spec[in, out]
860  *   EFX filter specification to update.
861  * @param[out] error
862  *   Perform verbose error reporting if not NULL.
863  */
864 static int
865 sfc_flow_parse_vxlan(const struct rte_flow_item *item,
866                      efx_filter_spec_t *efx_spec,
867                      struct rte_flow_error *error)
868 {
869         int rc;
870         const struct rte_flow_item_vxlan *spec = NULL;
871         const struct rte_flow_item_vxlan *mask = NULL;
872         const struct rte_flow_item_vxlan supp_mask = {
873                 .vni = { 0xff, 0xff, 0xff }
874         };
875
876         rc = sfc_flow_parse_init(item,
877                                  (const void **)&spec,
878                                  (const void **)&mask,
879                                  &supp_mask,
880                                  &rte_flow_item_vxlan_mask,
881                                  sizeof(struct rte_flow_item_vxlan),
882                                  error);
883         if (rc != 0)
884                 return rc;
885
886         rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
887                                                      EFX_IPPROTO_UDP, error);
888         if (rc != 0)
889                 return rc;
890
891         efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_VXLAN;
892         efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
893
894         if (spec == NULL)
895                 return 0;
896
897         rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->vni,
898                                                mask->vni, item, error);
899
900         return rc;
901 }
902
903 /**
904  * Convert GENEVE item to EFX filter specification.
905  *
906  * @param item[in]
907  *   Item specification. Only Virtual Network Identifier and protocol type
908  *   fields are supported. But protocol type can be only Ethernet (0x6558).
909  *   If the mask is NULL, default mask will be used.
910  *   Ranging is not supported.
911  * @param efx_spec[in, out]
912  *   EFX filter specification to update.
913  * @param[out] error
914  *   Perform verbose error reporting if not NULL.
915  */
916 static int
917 sfc_flow_parse_geneve(const struct rte_flow_item *item,
918                       efx_filter_spec_t *efx_spec,
919                       struct rte_flow_error *error)
920 {
921         int rc;
922         const struct rte_flow_item_geneve *spec = NULL;
923         const struct rte_flow_item_geneve *mask = NULL;
924         const struct rte_flow_item_geneve supp_mask = {
925                 .protocol = RTE_BE16(0xffff),
926                 .vni = { 0xff, 0xff, 0xff }
927         };
928
929         rc = sfc_flow_parse_init(item,
930                                  (const void **)&spec,
931                                  (const void **)&mask,
932                                  &supp_mask,
933                                  &rte_flow_item_geneve_mask,
934                                  sizeof(struct rte_flow_item_geneve),
935                                  error);
936         if (rc != 0)
937                 return rc;
938
939         rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
940                                                      EFX_IPPROTO_UDP, error);
941         if (rc != 0)
942                 return rc;
943
944         efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_GENEVE;
945         efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
946
947         if (spec == NULL)
948                 return 0;
949
950         if (mask->protocol == supp_mask.protocol) {
951                 if (spec->protocol != rte_cpu_to_be_16(ETHER_TYPE_TEB)) {
952                         rte_flow_error_set(error, EINVAL,
953                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
954                                 "GENEVE encap. protocol must be Ethernet "
955                                 "(0x6558) in the GENEVE pattern item");
956                         return -rte_errno;
957                 }
958         } else if (mask->protocol != 0) {
959                 rte_flow_error_set(error, EINVAL,
960                         RTE_FLOW_ERROR_TYPE_ITEM, item,
961                         "Unsupported mask for GENEVE encap. protocol");
962                 return -rte_errno;
963         }
964
965         rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->vni,
966                                                mask->vni, item, error);
967
968         return rc;
969 }
970
971 /**
972  * Convert NVGRE item to EFX filter specification.
973  *
974  * @param item[in]
975  *   Item specification. Only virtual subnet ID field is supported.
976  *   If the mask is NULL, default mask will be used.
977  *   Ranging is not supported.
978  * @param efx_spec[in, out]
979  *   EFX filter specification to update.
980  * @param[out] error
981  *   Perform verbose error reporting if not NULL.
982  */
983 static int
984 sfc_flow_parse_nvgre(const struct rte_flow_item *item,
985                      efx_filter_spec_t *efx_spec,
986                      struct rte_flow_error *error)
987 {
988         int rc;
989         const struct rte_flow_item_nvgre *spec = NULL;
990         const struct rte_flow_item_nvgre *mask = NULL;
991         const struct rte_flow_item_nvgre supp_mask = {
992                 .tni = { 0xff, 0xff, 0xff }
993         };
994
995         rc = sfc_flow_parse_init(item,
996                                  (const void **)&spec,
997                                  (const void **)&mask,
998                                  &supp_mask,
999                                  &rte_flow_item_nvgre_mask,
1000                                  sizeof(struct rte_flow_item_nvgre),
1001                                  error);
1002         if (rc != 0)
1003                 return rc;
1004
1005         rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
1006                                                      EFX_IPPROTO_GRE, error);
1007         if (rc != 0)
1008                 return rc;
1009
1010         efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_NVGRE;
1011         efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
1012
1013         if (spec == NULL)
1014                 return 0;
1015
1016         rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->tni,
1017                                                mask->tni, item, error);
1018
1019         return rc;
1020 }
1021
1022 static const struct sfc_flow_item sfc_flow_items[] = {
1023         {
1024                 .type = RTE_FLOW_ITEM_TYPE_VOID,
1025                 .prev_layer = SFC_FLOW_ITEM_ANY_LAYER,
1026                 .layer = SFC_FLOW_ITEM_ANY_LAYER,
1027                 .parse = sfc_flow_parse_void,
1028         },
1029         {
1030                 .type = RTE_FLOW_ITEM_TYPE_ETH,
1031                 .prev_layer = SFC_FLOW_ITEM_START_LAYER,
1032                 .layer = SFC_FLOW_ITEM_L2,
1033                 .parse = sfc_flow_parse_eth,
1034         },
1035         {
1036                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
1037                 .prev_layer = SFC_FLOW_ITEM_L2,
1038                 .layer = SFC_FLOW_ITEM_L2,
1039                 .parse = sfc_flow_parse_vlan,
1040         },
1041         {
1042                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
1043                 .prev_layer = SFC_FLOW_ITEM_L2,
1044                 .layer = SFC_FLOW_ITEM_L3,
1045                 .parse = sfc_flow_parse_ipv4,
1046         },
1047         {
1048                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
1049                 .prev_layer = SFC_FLOW_ITEM_L2,
1050                 .layer = SFC_FLOW_ITEM_L3,
1051                 .parse = sfc_flow_parse_ipv6,
1052         },
1053         {
1054                 .type = RTE_FLOW_ITEM_TYPE_TCP,
1055                 .prev_layer = SFC_FLOW_ITEM_L3,
1056                 .layer = SFC_FLOW_ITEM_L4,
1057                 .parse = sfc_flow_parse_tcp,
1058         },
1059         {
1060                 .type = RTE_FLOW_ITEM_TYPE_UDP,
1061                 .prev_layer = SFC_FLOW_ITEM_L3,
1062                 .layer = SFC_FLOW_ITEM_L4,
1063                 .parse = sfc_flow_parse_udp,
1064         },
1065         {
1066                 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
1067                 .prev_layer = SFC_FLOW_ITEM_L4,
1068                 .layer = SFC_FLOW_ITEM_START_LAYER,
1069                 .parse = sfc_flow_parse_vxlan,
1070         },
1071         {
1072                 .type = RTE_FLOW_ITEM_TYPE_GENEVE,
1073                 .prev_layer = SFC_FLOW_ITEM_L4,
1074                 .layer = SFC_FLOW_ITEM_START_LAYER,
1075                 .parse = sfc_flow_parse_geneve,
1076         },
1077         {
1078                 .type = RTE_FLOW_ITEM_TYPE_NVGRE,
1079                 .prev_layer = SFC_FLOW_ITEM_L3,
1080                 .layer = SFC_FLOW_ITEM_START_LAYER,
1081                 .parse = sfc_flow_parse_nvgre,
1082         },
1083 };
1084
1085 /*
1086  * Protocol-independent flow API support
1087  */
1088 static int
1089 sfc_flow_parse_attr(const struct rte_flow_attr *attr,
1090                     struct rte_flow *flow,
1091                     struct rte_flow_error *error)
1092 {
1093         if (attr == NULL) {
1094                 rte_flow_error_set(error, EINVAL,
1095                                    RTE_FLOW_ERROR_TYPE_ATTR, NULL,
1096                                    "NULL attribute");
1097                 return -rte_errno;
1098         }
1099         if (attr->group != 0) {
1100                 rte_flow_error_set(error, ENOTSUP,
1101                                    RTE_FLOW_ERROR_TYPE_ATTR_GROUP, attr,
1102                                    "Groups are not supported");
1103                 return -rte_errno;
1104         }
1105         if (attr->priority != 0) {
1106                 rte_flow_error_set(error, ENOTSUP,
1107                                    RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY, attr,
1108                                    "Priorities are not supported");
1109                 return -rte_errno;
1110         }
1111         if (attr->egress != 0) {
1112                 rte_flow_error_set(error, ENOTSUP,
1113                                    RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attr,
1114                                    "Egress is not supported");
1115                 return -rte_errno;
1116         }
1117         if (attr->transfer != 0) {
1118                 rte_flow_error_set(error, ENOTSUP,
1119                                    RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER, attr,
1120                                    "Transfer is not supported");
1121                 return -rte_errno;
1122         }
1123         if (attr->ingress == 0) {
1124                 rte_flow_error_set(error, ENOTSUP,
1125                                    RTE_FLOW_ERROR_TYPE_ATTR_INGRESS, attr,
1126                                    "Only ingress is supported");
1127                 return -rte_errno;
1128         }
1129
1130         flow->spec.template.efs_flags |= EFX_FILTER_FLAG_RX;
1131         flow->spec.template.efs_rss_context = EFX_RSS_CONTEXT_DEFAULT;
1132
1133         return 0;
1134 }
1135
1136 /* Get item from array sfc_flow_items */
1137 static const struct sfc_flow_item *
1138 sfc_flow_get_item(enum rte_flow_item_type type)
1139 {
1140         unsigned int i;
1141
1142         for (i = 0; i < RTE_DIM(sfc_flow_items); i++)
1143                 if (sfc_flow_items[i].type == type)
1144                         return &sfc_flow_items[i];
1145
1146         return NULL;
1147 }
1148
1149 static int
1150 sfc_flow_parse_pattern(const struct rte_flow_item pattern[],
1151                        struct rte_flow *flow,
1152                        struct rte_flow_error *error)
1153 {
1154         int rc;
1155         unsigned int prev_layer = SFC_FLOW_ITEM_ANY_LAYER;
1156         boolean_t is_ifrm = B_FALSE;
1157         const struct sfc_flow_item *item;
1158
1159         if (pattern == NULL) {
1160                 rte_flow_error_set(error, EINVAL,
1161                                    RTE_FLOW_ERROR_TYPE_ITEM_NUM, NULL,
1162                                    "NULL pattern");
1163                 return -rte_errno;
1164         }
1165
1166         for (; pattern->type != RTE_FLOW_ITEM_TYPE_END; pattern++) {
1167                 item = sfc_flow_get_item(pattern->type);
1168                 if (item == NULL) {
1169                         rte_flow_error_set(error, ENOTSUP,
1170                                            RTE_FLOW_ERROR_TYPE_ITEM, pattern,
1171                                            "Unsupported pattern item");
1172                         return -rte_errno;
1173                 }
1174
1175                 /*
1176                  * Omitting one or several protocol layers at the beginning
1177                  * of pattern is supported
1178                  */
1179                 if (item->prev_layer != SFC_FLOW_ITEM_ANY_LAYER &&
1180                     prev_layer != SFC_FLOW_ITEM_ANY_LAYER &&
1181                     item->prev_layer != prev_layer) {
1182                         rte_flow_error_set(error, ENOTSUP,
1183                                            RTE_FLOW_ERROR_TYPE_ITEM, pattern,
1184                                            "Unexpected sequence of pattern items");
1185                         return -rte_errno;
1186                 }
1187
1188                 /*
1189                  * Allow only VOID and ETH pattern items in the inner frame.
1190                  * Also check that there is only one tunneling protocol.
1191                  */
1192                 switch (item->type) {
1193                 case RTE_FLOW_ITEM_TYPE_VOID:
1194                 case RTE_FLOW_ITEM_TYPE_ETH:
1195                         break;
1196
1197                 case RTE_FLOW_ITEM_TYPE_VXLAN:
1198                 case RTE_FLOW_ITEM_TYPE_GENEVE:
1199                 case RTE_FLOW_ITEM_TYPE_NVGRE:
1200                         if (is_ifrm) {
1201                                 rte_flow_error_set(error, EINVAL,
1202                                         RTE_FLOW_ERROR_TYPE_ITEM,
1203                                         pattern,
1204                                         "More than one tunneling protocol");
1205                                 return -rte_errno;
1206                         }
1207                         is_ifrm = B_TRUE;
1208                         break;
1209
1210                 default:
1211                         if (is_ifrm) {
1212                                 rte_flow_error_set(error, EINVAL,
1213                                         RTE_FLOW_ERROR_TYPE_ITEM,
1214                                         pattern,
1215                                         "There is an unsupported pattern item "
1216                                         "in the inner frame");
1217                                 return -rte_errno;
1218                         }
1219                         break;
1220                 }
1221
1222                 rc = item->parse(pattern, &flow->spec.template, error);
1223                 if (rc != 0)
1224                         return rc;
1225
1226                 if (item->layer != SFC_FLOW_ITEM_ANY_LAYER)
1227                         prev_layer = item->layer;
1228         }
1229
1230         return 0;
1231 }
1232
1233 static int
1234 sfc_flow_parse_queue(struct sfc_adapter *sa,
1235                      const struct rte_flow_action_queue *queue,
1236                      struct rte_flow *flow)
1237 {
1238         struct sfc_rxq *rxq;
1239
1240         if (queue->index >= sa->rxq_count)
1241                 return -EINVAL;
1242
1243         rxq = sa->rxq_info[queue->index].rxq;
1244         flow->spec.template.efs_dmaq_id = (uint16_t)rxq->hw_index;
1245
1246         return 0;
1247 }
1248
1249 static int
1250 sfc_flow_parse_rss(struct sfc_adapter *sa,
1251                    const struct rte_flow_action_rss *action_rss,
1252                    struct rte_flow *flow)
1253 {
1254         struct sfc_rss *rss = &sa->rss;
1255         unsigned int rxq_sw_index;
1256         struct sfc_rxq *rxq;
1257         unsigned int rxq_hw_index_min;
1258         unsigned int rxq_hw_index_max;
1259         efx_rx_hash_type_t efx_hash_types;
1260         const uint8_t *rss_key;
1261         struct sfc_flow_rss *sfc_rss_conf = &flow->rss_conf;
1262         unsigned int i;
1263
1264         if (action_rss->queue_num == 0)
1265                 return -EINVAL;
1266
1267         rxq_sw_index = sa->rxq_count - 1;
1268         rxq = sa->rxq_info[rxq_sw_index].rxq;
1269         rxq_hw_index_min = rxq->hw_index;
1270         rxq_hw_index_max = 0;
1271
1272         for (i = 0; i < action_rss->queue_num; ++i) {
1273                 rxq_sw_index = action_rss->queue[i];
1274
1275                 if (rxq_sw_index >= sa->rxq_count)
1276                         return -EINVAL;
1277
1278                 rxq = sa->rxq_info[rxq_sw_index].rxq;
1279
1280                 if (rxq->hw_index < rxq_hw_index_min)
1281                         rxq_hw_index_min = rxq->hw_index;
1282
1283                 if (rxq->hw_index > rxq_hw_index_max)
1284                         rxq_hw_index_max = rxq->hw_index;
1285         }
1286
1287         switch (action_rss->func) {
1288         case RTE_ETH_HASH_FUNCTION_DEFAULT:
1289         case RTE_ETH_HASH_FUNCTION_TOEPLITZ:
1290                 break;
1291         default:
1292                 return -EINVAL;
1293         }
1294
1295         if (action_rss->level)
1296                 return -EINVAL;
1297
1298         if (action_rss->types) {
1299                 int rc;
1300
1301                 rc = sfc_rx_hf_rte_to_efx(sa, action_rss->types,
1302                                           &efx_hash_types);
1303                 if (rc != 0)
1304                         return -rc;
1305         } else {
1306                 unsigned int i;
1307
1308                 efx_hash_types = 0;
1309                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
1310                         efx_hash_types |= rss->hf_map[i].efx;
1311         }
1312
1313         if (action_rss->key_len) {
1314                 if (action_rss->key_len != sizeof(rss->key))
1315                         return -EINVAL;
1316
1317                 rss_key = action_rss->key;
1318         } else {
1319                 rss_key = rss->key;
1320         }
1321
1322         flow->rss = B_TRUE;
1323
1324         sfc_rss_conf->rxq_hw_index_min = rxq_hw_index_min;
1325         sfc_rss_conf->rxq_hw_index_max = rxq_hw_index_max;
1326         sfc_rss_conf->rss_hash_types = efx_hash_types;
1327         rte_memcpy(sfc_rss_conf->rss_key, rss_key, sizeof(rss->key));
1328
1329         for (i = 0; i < RTE_DIM(sfc_rss_conf->rss_tbl); ++i) {
1330                 unsigned int nb_queues = action_rss->queue_num;
1331                 unsigned int rxq_sw_index = action_rss->queue[i % nb_queues];
1332                 struct sfc_rxq *rxq = sa->rxq_info[rxq_sw_index].rxq;
1333
1334                 sfc_rss_conf->rss_tbl[i] = rxq->hw_index - rxq_hw_index_min;
1335         }
1336
1337         return 0;
1338 }
1339
1340 static int
1341 sfc_flow_spec_flush(struct sfc_adapter *sa, struct sfc_flow_spec *spec,
1342                     unsigned int filters_count)
1343 {
1344         unsigned int i;
1345         int ret = 0;
1346
1347         for (i = 0; i < filters_count; i++) {
1348                 int rc;
1349
1350                 rc = efx_filter_remove(sa->nic, &spec->filters[i]);
1351                 if (ret == 0 && rc != 0) {
1352                         sfc_err(sa, "failed to remove filter specification "
1353                                 "(rc = %d)", rc);
1354                         ret = rc;
1355                 }
1356         }
1357
1358         return ret;
1359 }
1360
1361 static int
1362 sfc_flow_spec_insert(struct sfc_adapter *sa, struct sfc_flow_spec *spec)
1363 {
1364         unsigned int i;
1365         int rc = 0;
1366
1367         for (i = 0; i < spec->count; i++) {
1368                 rc = efx_filter_insert(sa->nic, &spec->filters[i]);
1369                 if (rc != 0) {
1370                         sfc_flow_spec_flush(sa, spec, i);
1371                         break;
1372                 }
1373         }
1374
1375         return rc;
1376 }
1377
1378 static int
1379 sfc_flow_spec_remove(struct sfc_adapter *sa, struct sfc_flow_spec *spec)
1380 {
1381         return sfc_flow_spec_flush(sa, spec, spec->count);
1382 }
1383
1384 static int
1385 sfc_flow_filter_insert(struct sfc_adapter *sa,
1386                        struct rte_flow *flow)
1387 {
1388         struct sfc_rss *rss = &sa->rss;
1389         struct sfc_flow_rss *flow_rss = &flow->rss_conf;
1390         uint32_t efs_rss_context = EFX_RSS_CONTEXT_DEFAULT;
1391         unsigned int i;
1392         int rc = 0;
1393
1394         if (flow->rss) {
1395                 unsigned int rss_spread = MIN(flow_rss->rxq_hw_index_max -
1396                                               flow_rss->rxq_hw_index_min + 1,
1397                                               EFX_MAXRSS);
1398
1399                 rc = efx_rx_scale_context_alloc(sa->nic,
1400                                                 EFX_RX_SCALE_EXCLUSIVE,
1401                                                 rss_spread,
1402                                                 &efs_rss_context);
1403                 if (rc != 0)
1404                         goto fail_scale_context_alloc;
1405
1406                 rc = efx_rx_scale_mode_set(sa->nic, efs_rss_context,
1407                                            rss->hash_alg,
1408                                            flow_rss->rss_hash_types, B_TRUE);
1409                 if (rc != 0)
1410                         goto fail_scale_mode_set;
1411
1412                 rc = efx_rx_scale_key_set(sa->nic, efs_rss_context,
1413                                           flow_rss->rss_key,
1414                                           sizeof(rss->key));
1415                 if (rc != 0)
1416                         goto fail_scale_key_set;
1417
1418                 /*
1419                  * At this point, fully elaborated filter specifications
1420                  * have been produced from the template. To make sure that
1421                  * RSS behaviour is consistent between them, set the same
1422                  * RSS context value everywhere.
1423                  */
1424                 for (i = 0; i < flow->spec.count; i++) {
1425                         efx_filter_spec_t *spec = &flow->spec.filters[i];
1426
1427                         spec->efs_rss_context = efs_rss_context;
1428                         spec->efs_dmaq_id = flow_rss->rxq_hw_index_min;
1429                         spec->efs_flags |= EFX_FILTER_FLAG_RX_RSS;
1430                 }
1431         }
1432
1433         rc = sfc_flow_spec_insert(sa, &flow->spec);
1434         if (rc != 0)
1435                 goto fail_filter_insert;
1436
1437         if (flow->rss) {
1438                 /*
1439                  * Scale table is set after filter insertion because
1440                  * the table entries are relative to the base RxQ ID
1441                  * and the latter is submitted to the HW by means of
1442                  * inserting a filter, so by the time of the request
1443                  * the HW knows all the information needed to verify
1444                  * the table entries, and the operation will succeed
1445                  */
1446                 rc = efx_rx_scale_tbl_set(sa->nic, efs_rss_context,
1447                                           flow_rss->rss_tbl,
1448                                           RTE_DIM(flow_rss->rss_tbl));
1449                 if (rc != 0)
1450                         goto fail_scale_tbl_set;
1451         }
1452
1453         return 0;
1454
1455 fail_scale_tbl_set:
1456         sfc_flow_spec_remove(sa, &flow->spec);
1457
1458 fail_filter_insert:
1459 fail_scale_key_set:
1460 fail_scale_mode_set:
1461         if (efs_rss_context != EFX_RSS_CONTEXT_DEFAULT)
1462                 efx_rx_scale_context_free(sa->nic, efs_rss_context);
1463
1464 fail_scale_context_alloc:
1465         return rc;
1466 }
1467
1468 static int
1469 sfc_flow_filter_remove(struct sfc_adapter *sa,
1470                        struct rte_flow *flow)
1471 {
1472         int rc = 0;
1473
1474         rc = sfc_flow_spec_remove(sa, &flow->spec);
1475         if (rc != 0)
1476                 return rc;
1477
1478         if (flow->rss) {
1479                 /*
1480                  * All specifications for a given flow rule have the same RSS
1481                  * context, so that RSS context value is taken from the first
1482                  * filter specification
1483                  */
1484                 efx_filter_spec_t *spec = &flow->spec.filters[0];
1485
1486                 rc = efx_rx_scale_context_free(sa->nic, spec->efs_rss_context);
1487         }
1488
1489         return rc;
1490 }
1491
1492 static int
1493 sfc_flow_parse_actions(struct sfc_adapter *sa,
1494                        const struct rte_flow_action actions[],
1495                        struct rte_flow *flow,
1496                        struct rte_flow_error *error)
1497 {
1498         int rc;
1499         boolean_t is_specified = B_FALSE;
1500
1501         if (actions == NULL) {
1502                 rte_flow_error_set(error, EINVAL,
1503                                    RTE_FLOW_ERROR_TYPE_ACTION_NUM, NULL,
1504                                    "NULL actions");
1505                 return -rte_errno;
1506         }
1507
1508         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1509                 /* This one may appear anywhere multiple times. */
1510                 if (actions->type == RTE_FLOW_ACTION_TYPE_VOID)
1511                         continue;
1512                 /* Fate-deciding actions may appear exactly once. */
1513                 if (is_specified) {
1514                         rte_flow_error_set
1515                                 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
1516                                  actions,
1517                                  "Cannot combine several fate-deciding actions,"
1518                                  "choose between QUEUE, RSS or DROP");
1519                         return -rte_errno;
1520                 }
1521                 switch (actions->type) {
1522                 case RTE_FLOW_ACTION_TYPE_QUEUE:
1523                         rc = sfc_flow_parse_queue(sa, actions->conf, flow);
1524                         if (rc != 0) {
1525                                 rte_flow_error_set(error, EINVAL,
1526                                         RTE_FLOW_ERROR_TYPE_ACTION, actions,
1527                                         "Bad QUEUE action");
1528                                 return -rte_errno;
1529                         }
1530
1531                         is_specified = B_TRUE;
1532                         break;
1533
1534                 case RTE_FLOW_ACTION_TYPE_RSS:
1535                         rc = sfc_flow_parse_rss(sa, actions->conf, flow);
1536                         if (rc != 0) {
1537                                 rte_flow_error_set(error, rc,
1538                                         RTE_FLOW_ERROR_TYPE_ACTION, actions,
1539                                         "Bad RSS action");
1540                                 return -rte_errno;
1541                         }
1542
1543                         is_specified = B_TRUE;
1544                         break;
1545
1546                 case RTE_FLOW_ACTION_TYPE_DROP:
1547                         flow->spec.template.efs_dmaq_id =
1548                                 EFX_FILTER_SPEC_RX_DMAQ_ID_DROP;
1549
1550                         is_specified = B_TRUE;
1551                         break;
1552
1553                 default:
1554                         rte_flow_error_set(error, ENOTSUP,
1555                                            RTE_FLOW_ERROR_TYPE_ACTION, actions,
1556                                            "Action is not supported");
1557                         return -rte_errno;
1558                 }
1559         }
1560
1561         /* When fate is unknown, drop traffic. */
1562         if (!is_specified) {
1563                 flow->spec.template.efs_dmaq_id =
1564                         EFX_FILTER_SPEC_RX_DMAQ_ID_DROP;
1565         }
1566
1567         return 0;
1568 }
1569
1570 /**
1571  * Set the EFX_FILTER_MATCH_UNKNOWN_UCAST_DST
1572  * and EFX_FILTER_MATCH_UNKNOWN_MCAST_DST match flags in the same
1573  * specifications after copying.
1574  *
1575  * @param spec[in, out]
1576  *   SFC flow specification to update.
1577  * @param filters_count_for_one_val[in]
1578  *   How many specifications should have the same match flag, what is the
1579  *   number of specifications before copying.
1580  * @param error[out]
1581  *   Perform verbose error reporting if not NULL.
1582  */
1583 static int
1584 sfc_flow_set_unknown_dst_flags(struct sfc_flow_spec *spec,
1585                                unsigned int filters_count_for_one_val,
1586                                struct rte_flow_error *error)
1587 {
1588         unsigned int i;
1589         static const efx_filter_match_flags_t vals[] = {
1590                 EFX_FILTER_MATCH_UNKNOWN_UCAST_DST,
1591                 EFX_FILTER_MATCH_UNKNOWN_MCAST_DST
1592         };
1593
1594         if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1595                 rte_flow_error_set(error, EINVAL,
1596                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1597                         "Number of specifications is incorrect while copying "
1598                         "by unknown destination flags");
1599                 return -rte_errno;
1600         }
1601
1602         for (i = 0; i < spec->count; i++) {
1603                 /* The check above ensures that divisor can't be zero here */
1604                 spec->filters[i].efs_match_flags |=
1605                         vals[i / filters_count_for_one_val];
1606         }
1607
1608         return 0;
1609 }
1610
1611 /**
1612  * Check that the following conditions are met:
1613  * - the list of supported filters has a filter
1614  *   with EFX_FILTER_MATCH_UNKNOWN_MCAST_DST flag instead of
1615  *   EFX_FILTER_MATCH_UNKNOWN_UCAST_DST, since this filter will also
1616  *   be inserted.
1617  *
1618  * @param match[in]
1619  *   The match flags of filter.
1620  * @param spec[in]
1621  *   Specification to be supplemented.
1622  * @param filter[in]
1623  *   SFC filter with list of supported filters.
1624  */
1625 static boolean_t
1626 sfc_flow_check_unknown_dst_flags(efx_filter_match_flags_t match,
1627                                  __rte_unused efx_filter_spec_t *spec,
1628                                  struct sfc_filter *filter)
1629 {
1630         unsigned int i;
1631         efx_filter_match_flags_t match_mcast_dst;
1632
1633         match_mcast_dst =
1634                 (match & ~EFX_FILTER_MATCH_UNKNOWN_UCAST_DST) |
1635                 EFX_FILTER_MATCH_UNKNOWN_MCAST_DST;
1636         for (i = 0; i < filter->supported_match_num; i++) {
1637                 if (match_mcast_dst == filter->supported_match[i])
1638                         return B_TRUE;
1639         }
1640
1641         return B_FALSE;
1642 }
1643
1644 /**
1645  * Set the EFX_FILTER_MATCH_ETHER_TYPE match flag and EFX_ETHER_TYPE_IPV4 and
1646  * EFX_ETHER_TYPE_IPV6 values of the corresponding field in the same
1647  * specifications after copying.
1648  *
1649  * @param spec[in, out]
1650  *   SFC flow specification to update.
1651  * @param filters_count_for_one_val[in]
1652  *   How many specifications should have the same EtherType value, what is the
1653  *   number of specifications before copying.
1654  * @param error[out]
1655  *   Perform verbose error reporting if not NULL.
1656  */
1657 static int
1658 sfc_flow_set_ethertypes(struct sfc_flow_spec *spec,
1659                         unsigned int filters_count_for_one_val,
1660                         struct rte_flow_error *error)
1661 {
1662         unsigned int i;
1663         static const uint16_t vals[] = {
1664                 EFX_ETHER_TYPE_IPV4, EFX_ETHER_TYPE_IPV6
1665         };
1666
1667         if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1668                 rte_flow_error_set(error, EINVAL,
1669                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1670                         "Number of specifications is incorrect "
1671                         "while copying by Ethertype");
1672                 return -rte_errno;
1673         }
1674
1675         for (i = 0; i < spec->count; i++) {
1676                 spec->filters[i].efs_match_flags |=
1677                         EFX_FILTER_MATCH_ETHER_TYPE;
1678
1679                 /*
1680                  * The check above ensures that
1681                  * filters_count_for_one_val is not 0
1682                  */
1683                 spec->filters[i].efs_ether_type =
1684                         vals[i / filters_count_for_one_val];
1685         }
1686
1687         return 0;
1688 }
1689
1690 /**
1691  * Set the EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST and
1692  * EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST match flags in the same
1693  * specifications after copying.
1694  *
1695  * @param spec[in, out]
1696  *   SFC flow specification to update.
1697  * @param filters_count_for_one_val[in]
1698  *   How many specifications should have the same match flag, what is the
1699  *   number of specifications before copying.
1700  * @param error[out]
1701  *   Perform verbose error reporting if not NULL.
1702  */
1703 static int
1704 sfc_flow_set_ifrm_unknown_dst_flags(struct sfc_flow_spec *spec,
1705                                     unsigned int filters_count_for_one_val,
1706                                     struct rte_flow_error *error)
1707 {
1708         unsigned int i;
1709         static const efx_filter_match_flags_t vals[] = {
1710                 EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST,
1711                 EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST
1712         };
1713
1714         if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1715                 rte_flow_error_set(error, EINVAL,
1716                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1717                         "Number of specifications is incorrect while copying "
1718                         "by inner frame unknown destination flags");
1719                 return -rte_errno;
1720         }
1721
1722         for (i = 0; i < spec->count; i++) {
1723                 /* The check above ensures that divisor can't be zero here */
1724                 spec->filters[i].efs_match_flags |=
1725                         vals[i / filters_count_for_one_val];
1726         }
1727
1728         return 0;
1729 }
1730
1731 /**
1732  * Check that the following conditions are met:
1733  * - the specification corresponds to a filter for encapsulated traffic
1734  * - the list of supported filters has a filter
1735  *   with EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST flag instead of
1736  *   EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST, since this filter will also
1737  *   be inserted.
1738  *
1739  * @param match[in]
1740  *   The match flags of filter.
1741  * @param spec[in]
1742  *   Specification to be supplemented.
1743  * @param filter[in]
1744  *   SFC filter with list of supported filters.
1745  */
1746 static boolean_t
1747 sfc_flow_check_ifrm_unknown_dst_flags(efx_filter_match_flags_t match,
1748                                       efx_filter_spec_t *spec,
1749                                       struct sfc_filter *filter)
1750 {
1751         unsigned int i;
1752         efx_tunnel_protocol_t encap_type = spec->efs_encap_type;
1753         efx_filter_match_flags_t match_mcast_dst;
1754
1755         if (encap_type == EFX_TUNNEL_PROTOCOL_NONE)
1756                 return B_FALSE;
1757
1758         match_mcast_dst =
1759                 (match & ~EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST) |
1760                 EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST;
1761         for (i = 0; i < filter->supported_match_num; i++) {
1762                 if (match_mcast_dst == filter->supported_match[i])
1763                         return B_TRUE;
1764         }
1765
1766         return B_FALSE;
1767 }
1768
1769 /*
1770  * Match flags that can be automatically added to filters.
1771  * Selecting the last minimum when searching for the copy flag ensures that the
1772  * EFX_FILTER_MATCH_UNKNOWN_UCAST_DST flag has a higher priority than
1773  * EFX_FILTER_MATCH_ETHER_TYPE. This is because the filter
1774  * EFX_FILTER_MATCH_UNKNOWN_UCAST_DST is at the end of the list of supported
1775  * filters.
1776  */
1777 static const struct sfc_flow_copy_flag sfc_flow_copy_flags[] = {
1778         {
1779                 .flag = EFX_FILTER_MATCH_UNKNOWN_UCAST_DST,
1780                 .vals_count = 2,
1781                 .set_vals = sfc_flow_set_unknown_dst_flags,
1782                 .spec_check = sfc_flow_check_unknown_dst_flags,
1783         },
1784         {
1785                 .flag = EFX_FILTER_MATCH_ETHER_TYPE,
1786                 .vals_count = 2,
1787                 .set_vals = sfc_flow_set_ethertypes,
1788                 .spec_check = NULL,
1789         },
1790         {
1791                 .flag = EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST,
1792                 .vals_count = 2,
1793                 .set_vals = sfc_flow_set_ifrm_unknown_dst_flags,
1794                 .spec_check = sfc_flow_check_ifrm_unknown_dst_flags,
1795         },
1796 };
1797
1798 /* Get item from array sfc_flow_copy_flags */
1799 static const struct sfc_flow_copy_flag *
1800 sfc_flow_get_copy_flag(efx_filter_match_flags_t flag)
1801 {
1802         unsigned int i;
1803
1804         for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1805                 if (sfc_flow_copy_flags[i].flag == flag)
1806                         return &sfc_flow_copy_flags[i];
1807         }
1808
1809         return NULL;
1810 }
1811
1812 /**
1813  * Make copies of the specifications, set match flag and values
1814  * of the field that corresponds to it.
1815  *
1816  * @param spec[in, out]
1817  *   SFC flow specification to update.
1818  * @param flag[in]
1819  *   The match flag to add.
1820  * @param error[out]
1821  *   Perform verbose error reporting if not NULL.
1822  */
1823 static int
1824 sfc_flow_spec_add_match_flag(struct sfc_flow_spec *spec,
1825                              efx_filter_match_flags_t flag,
1826                              struct rte_flow_error *error)
1827 {
1828         unsigned int i;
1829         unsigned int new_filters_count;
1830         unsigned int filters_count_for_one_val;
1831         const struct sfc_flow_copy_flag *copy_flag;
1832         int rc;
1833
1834         copy_flag = sfc_flow_get_copy_flag(flag);
1835         if (copy_flag == NULL) {
1836                 rte_flow_error_set(error, ENOTSUP,
1837                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1838                                    "Unsupported spec field for copying");
1839                 return -rte_errno;
1840         }
1841
1842         new_filters_count = spec->count * copy_flag->vals_count;
1843         if (new_filters_count > SF_FLOW_SPEC_NB_FILTERS_MAX) {
1844                 rte_flow_error_set(error, EINVAL,
1845                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1846                         "Too much EFX specifications in the flow rule");
1847                 return -rte_errno;
1848         }
1849
1850         /* Copy filters specifications */
1851         for (i = spec->count; i < new_filters_count; i++)
1852                 spec->filters[i] = spec->filters[i - spec->count];
1853
1854         filters_count_for_one_val = spec->count;
1855         spec->count = new_filters_count;
1856
1857         rc = copy_flag->set_vals(spec, filters_count_for_one_val, error);
1858         if (rc != 0)
1859                 return rc;
1860
1861         return 0;
1862 }
1863
1864 /**
1865  * Check that the given set of match flags missing in the original filter spec
1866  * could be covered by adding spec copies which specify the corresponding
1867  * flags and packet field values to match.
1868  *
1869  * @param miss_flags[in]
1870  *   Flags that are missing until the supported filter.
1871  * @param spec[in]
1872  *   Specification to be supplemented.
1873  * @param filter[in]
1874  *   SFC filter.
1875  *
1876  * @return
1877  *   Number of specifications after copy or 0, if the flags can not be added.
1878  */
1879 static unsigned int
1880 sfc_flow_check_missing_flags(efx_filter_match_flags_t miss_flags,
1881                              efx_filter_spec_t *spec,
1882                              struct sfc_filter *filter)
1883 {
1884         unsigned int i;
1885         efx_filter_match_flags_t copy_flags = 0;
1886         efx_filter_match_flags_t flag;
1887         efx_filter_match_flags_t match = spec->efs_match_flags | miss_flags;
1888         sfc_flow_spec_check *check;
1889         unsigned int multiplier = 1;
1890
1891         for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1892                 flag = sfc_flow_copy_flags[i].flag;
1893                 check = sfc_flow_copy_flags[i].spec_check;
1894                 if ((flag & miss_flags) == flag) {
1895                         if (check != NULL && (!check(match, spec, filter)))
1896                                 continue;
1897
1898                         copy_flags |= flag;
1899                         multiplier *= sfc_flow_copy_flags[i].vals_count;
1900                 }
1901         }
1902
1903         if (copy_flags == miss_flags)
1904                 return multiplier;
1905
1906         return 0;
1907 }
1908
1909 /**
1910  * Attempt to supplement the specification template to the minimally
1911  * supported set of match flags. To do this, it is necessary to copy
1912  * the specifications, filling them with the values of fields that
1913  * correspond to the missing flags.
1914  * The necessary and sufficient filter is built from the fewest number
1915  * of copies which could be made to cover the minimally required set
1916  * of flags.
1917  *
1918  * @param sa[in]
1919  *   SFC adapter.
1920  * @param spec[in, out]
1921  *   SFC flow specification to update.
1922  * @param error[out]
1923  *   Perform verbose error reporting if not NULL.
1924  */
1925 static int
1926 sfc_flow_spec_filters_complete(struct sfc_adapter *sa,
1927                                struct sfc_flow_spec *spec,
1928                                struct rte_flow_error *error)
1929 {
1930         struct sfc_filter *filter = &sa->filter;
1931         efx_filter_match_flags_t miss_flags;
1932         efx_filter_match_flags_t min_miss_flags = 0;
1933         efx_filter_match_flags_t match;
1934         unsigned int min_multiplier = UINT_MAX;
1935         unsigned int multiplier;
1936         unsigned int i;
1937         int rc;
1938
1939         match = spec->template.efs_match_flags;
1940         for (i = 0; i < filter->supported_match_num; i++) {
1941                 if ((match & filter->supported_match[i]) == match) {
1942                         miss_flags = filter->supported_match[i] & (~match);
1943                         multiplier = sfc_flow_check_missing_flags(miss_flags,
1944                                 &spec->template, filter);
1945                         if (multiplier > 0) {
1946                                 if (multiplier <= min_multiplier) {
1947                                         min_multiplier = multiplier;
1948                                         min_miss_flags = miss_flags;
1949                                 }
1950                         }
1951                 }
1952         }
1953
1954         if (min_multiplier == UINT_MAX) {
1955                 rte_flow_error_set(error, ENOTSUP,
1956                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1957                                    "Flow rule pattern is not supported");
1958                 return -rte_errno;
1959         }
1960
1961         for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1962                 efx_filter_match_flags_t flag = sfc_flow_copy_flags[i].flag;
1963
1964                 if ((flag & min_miss_flags) == flag) {
1965                         rc = sfc_flow_spec_add_match_flag(spec, flag, error);
1966                         if (rc != 0)
1967                                 return rc;
1968                 }
1969         }
1970
1971         return 0;
1972 }
1973
1974 /**
1975  * Check that set of match flags is referred to by a filter. Filter is
1976  * described by match flags with the ability to add OUTER_VID and INNER_VID
1977  * flags.
1978  *
1979  * @param match_flags[in]
1980  *   Set of match flags.
1981  * @param flags_pattern[in]
1982  *   Pattern of filter match flags.
1983  */
1984 static boolean_t
1985 sfc_flow_is_match_with_vids(efx_filter_match_flags_t match_flags,
1986                             efx_filter_match_flags_t flags_pattern)
1987 {
1988         if ((match_flags & flags_pattern) != flags_pattern)
1989                 return B_FALSE;
1990
1991         switch (match_flags & ~flags_pattern) {
1992         case 0:
1993         case EFX_FILTER_MATCH_OUTER_VID:
1994         case EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_INNER_VID:
1995                 return B_TRUE;
1996         default:
1997                 return B_FALSE;
1998         }
1999 }
2000
2001 /**
2002  * Check whether the spec maps to a hardware filter which is known to be
2003  * ineffective despite being valid.
2004  *
2005  * @param spec[in]
2006  *   SFC flow specification.
2007  */
2008 static boolean_t
2009 sfc_flow_is_match_flags_exception(struct sfc_flow_spec *spec)
2010 {
2011         unsigned int i;
2012         uint16_t ether_type;
2013         uint8_t ip_proto;
2014         efx_filter_match_flags_t match_flags;
2015
2016         for (i = 0; i < spec->count; i++) {
2017                 match_flags = spec->filters[i].efs_match_flags;
2018
2019                 if (sfc_flow_is_match_with_vids(match_flags,
2020                                                 EFX_FILTER_MATCH_ETHER_TYPE) ||
2021                     sfc_flow_is_match_with_vids(match_flags,
2022                                                 EFX_FILTER_MATCH_ETHER_TYPE |
2023                                                 EFX_FILTER_MATCH_LOC_MAC)) {
2024                         ether_type = spec->filters[i].efs_ether_type;
2025                         if (ether_type == EFX_ETHER_TYPE_IPV4 ||
2026                             ether_type == EFX_ETHER_TYPE_IPV6)
2027                                 return B_TRUE;
2028                 } else if (sfc_flow_is_match_with_vids(match_flags,
2029                                 EFX_FILTER_MATCH_ETHER_TYPE |
2030                                 EFX_FILTER_MATCH_IP_PROTO) ||
2031                            sfc_flow_is_match_with_vids(match_flags,
2032                                 EFX_FILTER_MATCH_ETHER_TYPE |
2033                                 EFX_FILTER_MATCH_IP_PROTO |
2034                                 EFX_FILTER_MATCH_LOC_MAC)) {
2035                         ip_proto = spec->filters[i].efs_ip_proto;
2036                         if (ip_proto == EFX_IPPROTO_TCP ||
2037                             ip_proto == EFX_IPPROTO_UDP)
2038                                 return B_TRUE;
2039                 }
2040         }
2041
2042         return B_FALSE;
2043 }
2044
2045 static int
2046 sfc_flow_validate_match_flags(struct sfc_adapter *sa,
2047                               struct rte_flow *flow,
2048                               struct rte_flow_error *error)
2049 {
2050         efx_filter_spec_t *spec_tmpl = &flow->spec.template;
2051         efx_filter_match_flags_t match_flags = spec_tmpl->efs_match_flags;
2052         int rc;
2053
2054         /* Initialize the first filter spec with template */
2055         flow->spec.filters[0] = *spec_tmpl;
2056         flow->spec.count = 1;
2057
2058         if (!sfc_filter_is_match_supported(sa, match_flags)) {
2059                 rc = sfc_flow_spec_filters_complete(sa, &flow->spec, error);
2060                 if (rc != 0)
2061                         return rc;
2062         }
2063
2064         if (sfc_flow_is_match_flags_exception(&flow->spec)) {
2065                 rte_flow_error_set(error, ENOTSUP,
2066                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2067                         "The flow rule pattern is unsupported");
2068                 return -rte_errno;
2069         }
2070
2071         return 0;
2072 }
2073
2074 static int
2075 sfc_flow_parse(struct rte_eth_dev *dev,
2076                const struct rte_flow_attr *attr,
2077                const struct rte_flow_item pattern[],
2078                const struct rte_flow_action actions[],
2079                struct rte_flow *flow,
2080                struct rte_flow_error *error)
2081 {
2082         struct sfc_adapter *sa = dev->data->dev_private;
2083         int rc;
2084
2085         rc = sfc_flow_parse_attr(attr, flow, error);
2086         if (rc != 0)
2087                 goto fail_bad_value;
2088
2089         rc = sfc_flow_parse_pattern(pattern, flow, error);
2090         if (rc != 0)
2091                 goto fail_bad_value;
2092
2093         rc = sfc_flow_parse_actions(sa, actions, flow, error);
2094         if (rc != 0)
2095                 goto fail_bad_value;
2096
2097         rc = sfc_flow_validate_match_flags(sa, flow, error);
2098         if (rc != 0)
2099                 goto fail_bad_value;
2100
2101         return 0;
2102
2103 fail_bad_value:
2104         return rc;
2105 }
2106
2107 static int
2108 sfc_flow_validate(struct rte_eth_dev *dev,
2109                   const struct rte_flow_attr *attr,
2110                   const struct rte_flow_item pattern[],
2111                   const struct rte_flow_action actions[],
2112                   struct rte_flow_error *error)
2113 {
2114         struct rte_flow flow;
2115
2116         memset(&flow, 0, sizeof(flow));
2117
2118         return sfc_flow_parse(dev, attr, pattern, actions, &flow, error);
2119 }
2120
2121 static struct rte_flow *
2122 sfc_flow_create(struct rte_eth_dev *dev,
2123                 const struct rte_flow_attr *attr,
2124                 const struct rte_flow_item pattern[],
2125                 const struct rte_flow_action actions[],
2126                 struct rte_flow_error *error)
2127 {
2128         struct sfc_adapter *sa = dev->data->dev_private;
2129         struct rte_flow *flow = NULL;
2130         int rc;
2131
2132         flow = rte_zmalloc("sfc_rte_flow", sizeof(*flow), 0);
2133         if (flow == NULL) {
2134                 rte_flow_error_set(error, ENOMEM,
2135                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2136                                    "Failed to allocate memory");
2137                 goto fail_no_mem;
2138         }
2139
2140         rc = sfc_flow_parse(dev, attr, pattern, actions, flow, error);
2141         if (rc != 0)
2142                 goto fail_bad_value;
2143
2144         TAILQ_INSERT_TAIL(&sa->filter.flow_list, flow, entries);
2145
2146         sfc_adapter_lock(sa);
2147
2148         if (sa->state == SFC_ADAPTER_STARTED) {
2149                 rc = sfc_flow_filter_insert(sa, flow);
2150                 if (rc != 0) {
2151                         rte_flow_error_set(error, rc,
2152                                 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2153                                 "Failed to insert filter");
2154                         goto fail_filter_insert;
2155                 }
2156         }
2157
2158         sfc_adapter_unlock(sa);
2159
2160         return flow;
2161
2162 fail_filter_insert:
2163         TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2164
2165 fail_bad_value:
2166         rte_free(flow);
2167         sfc_adapter_unlock(sa);
2168
2169 fail_no_mem:
2170         return NULL;
2171 }
2172
2173 static int
2174 sfc_flow_remove(struct sfc_adapter *sa,
2175                 struct rte_flow *flow,
2176                 struct rte_flow_error *error)
2177 {
2178         int rc = 0;
2179
2180         SFC_ASSERT(sfc_adapter_is_locked(sa));
2181
2182         if (sa->state == SFC_ADAPTER_STARTED) {
2183                 rc = sfc_flow_filter_remove(sa, flow);
2184                 if (rc != 0)
2185                         rte_flow_error_set(error, rc,
2186                                 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2187                                 "Failed to destroy flow rule");
2188         }
2189
2190         TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2191         rte_free(flow);
2192
2193         return rc;
2194 }
2195
2196 static int
2197 sfc_flow_destroy(struct rte_eth_dev *dev,
2198                  struct rte_flow *flow,
2199                  struct rte_flow_error *error)
2200 {
2201         struct sfc_adapter *sa = dev->data->dev_private;
2202         struct rte_flow *flow_ptr;
2203         int rc = EINVAL;
2204
2205         sfc_adapter_lock(sa);
2206
2207         TAILQ_FOREACH(flow_ptr, &sa->filter.flow_list, entries) {
2208                 if (flow_ptr == flow)
2209                         rc = 0;
2210         }
2211         if (rc != 0) {
2212                 rte_flow_error_set(error, rc,
2213                                    RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2214                                    "Failed to find flow rule to destroy");
2215                 goto fail_bad_value;
2216         }
2217
2218         rc = sfc_flow_remove(sa, flow, error);
2219
2220 fail_bad_value:
2221         sfc_adapter_unlock(sa);
2222
2223         return -rc;
2224 }
2225
2226 static int
2227 sfc_flow_flush(struct rte_eth_dev *dev,
2228                struct rte_flow_error *error)
2229 {
2230         struct sfc_adapter *sa = dev->data->dev_private;
2231         struct rte_flow *flow;
2232         int rc = 0;
2233         int ret = 0;
2234
2235         sfc_adapter_lock(sa);
2236
2237         while ((flow = TAILQ_FIRST(&sa->filter.flow_list)) != NULL) {
2238                 rc = sfc_flow_remove(sa, flow, error);
2239                 if (rc != 0)
2240                         ret = rc;
2241         }
2242
2243         sfc_adapter_unlock(sa);
2244
2245         return -ret;
2246 }
2247
2248 static int
2249 sfc_flow_isolate(struct rte_eth_dev *dev, int enable,
2250                  struct rte_flow_error *error)
2251 {
2252         struct sfc_adapter *sa = dev->data->dev_private;
2253         struct sfc_port *port = &sa->port;
2254         int ret = 0;
2255
2256         sfc_adapter_lock(sa);
2257         if (sa->state != SFC_ADAPTER_INITIALIZED) {
2258                 rte_flow_error_set(error, EBUSY,
2259                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2260                                    NULL, "please close the port first");
2261                 ret = -rte_errno;
2262         } else {
2263                 port->isolated = (enable) ? B_TRUE : B_FALSE;
2264         }
2265         sfc_adapter_unlock(sa);
2266
2267         return ret;
2268 }
2269
2270 const struct rte_flow_ops sfc_flow_ops = {
2271         .validate = sfc_flow_validate,
2272         .create = sfc_flow_create,
2273         .destroy = sfc_flow_destroy,
2274         .flush = sfc_flow_flush,
2275         .query = NULL,
2276         .isolate = sfc_flow_isolate,
2277 };
2278
2279 void
2280 sfc_flow_init(struct sfc_adapter *sa)
2281 {
2282         SFC_ASSERT(sfc_adapter_is_locked(sa));
2283
2284         TAILQ_INIT(&sa->filter.flow_list);
2285 }
2286
2287 void
2288 sfc_flow_fini(struct sfc_adapter *sa)
2289 {
2290         struct rte_flow *flow;
2291
2292         SFC_ASSERT(sfc_adapter_is_locked(sa));
2293
2294         while ((flow = TAILQ_FIRST(&sa->filter.flow_list)) != NULL) {
2295                 TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2296                 rte_free(flow);
2297         }
2298 }
2299
2300 void
2301 sfc_flow_stop(struct sfc_adapter *sa)
2302 {
2303         struct rte_flow *flow;
2304
2305         SFC_ASSERT(sfc_adapter_is_locked(sa));
2306
2307         TAILQ_FOREACH(flow, &sa->filter.flow_list, entries)
2308                 sfc_flow_filter_remove(sa, flow);
2309 }
2310
2311 int
2312 sfc_flow_start(struct sfc_adapter *sa)
2313 {
2314         struct rte_flow *flow;
2315         int rc = 0;
2316
2317         sfc_log_init(sa, "entry");
2318
2319         SFC_ASSERT(sfc_adapter_is_locked(sa));
2320
2321         TAILQ_FOREACH(flow, &sa->filter.flow_list, entries) {
2322                 rc = sfc_flow_filter_insert(sa, flow);
2323                 if (rc != 0)
2324                         goto fail_bad_flow;
2325         }
2326
2327         sfc_log_init(sa, "done");
2328
2329 fail_bad_flow:
2330         return rc;
2331 }