net/sfc: support flow item VLAN in transfer rules
authorIvan Malov <ivan.malov@oktetlabs.ru>
Tue, 20 Oct 2020 09:13:27 +0000 (10:13 +0100)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 3 Nov 2020 22:24:25 +0000 (23:24 +0100)
Add support for this flow item to MAE-specific RTE flow implementation.

In a pattern, a L2 item preceding an item VLAN must have
correct "type" ("inner_type") set depending on the total
number of VLAN tags (double-tagging is supported):

"pattern eth type is X / vlan / end",
X = 0x8100, or 0x88a8, or 0x9100, or 0x9200, or 0x9300

"pattern eth type is X / vlan inner_type is 0x8100 / vlan / end"
X = 0x88a8, or 0x9100, or 0x9200, or 0x9300

Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andy Moreton <amoreton@xilinx.com>
doc/guides/nics/sfc_efx.rst
drivers/net/sfc/sfc_mae.c
drivers/net/sfc/sfc_mae.h

index e372627..cd61b4d 100644 (file)
@@ -200,6 +200,8 @@ Supported pattern items (***transfer*** rules):
 
 - ETH
 
+- VLAN (double-tagging is supported)
+
 Supported actions (***transfer*** rules):
 
 - OF_POP_VLAN
index 14e6d33..cc22fee 100644 (file)
@@ -260,6 +260,122 @@ sfc_mae_flow_cleanup(struct sfc_adapter *sa,
                efx_mae_match_spec_fini(sa->nic, spec_mae->match_spec);
 }
 
+static int
+sfc_mae_set_ethertypes(struct sfc_mae_parse_ctx *ctx)
+{
+       efx_mae_match_spec_t *efx_spec = ctx->match_spec_action;
+       struct sfc_mae_pattern_data *pdata = &ctx->pattern_data;
+       const efx_mae_field_id_t field_ids[] = {
+               EFX_MAE_FIELD_VLAN0_PROTO_BE,
+               EFX_MAE_FIELD_VLAN1_PROTO_BE,
+       };
+       const struct sfc_mae_ethertype *et;
+       unsigned int i;
+       int rc;
+
+       /*
+        * In accordance with RTE flow API convention, the innermost L2
+        * item's "type" ("inner_type") is a L3 EtherType. If there is
+        * no L3 item, it's 0x0000/0x0000.
+        */
+       et = &pdata->ethertypes[pdata->nb_vlan_tags];
+       rc = efx_mae_match_spec_field_set(efx_spec, EFX_MAE_FIELD_ETHER_TYPE_BE,
+                                         sizeof(et->value),
+                                         (const uint8_t *)&et->value,
+                                         sizeof(et->mask),
+                                         (const uint8_t *)&et->mask);
+       if (rc != 0)
+               return rc;
+
+       /*
+        * sfc_mae_rule_parse_item_vlan() has already made sure
+        * that pdata->nb_vlan_tags does not exceed this figure.
+        */
+       RTE_BUILD_BUG_ON(SFC_MAE_MATCH_VLAN_MAX_NTAGS != 2);
+
+       for (i = 0; i < pdata->nb_vlan_tags; ++i) {
+               et = &pdata->ethertypes[i];
+
+               rc = efx_mae_match_spec_field_set(efx_spec, field_ids[i],
+                                                 sizeof(et->value),
+                                                 (const uint8_t *)&et->value,
+                                                 sizeof(et->mask),
+                                                 (const uint8_t *)&et->mask);
+               if (rc != 0)
+                       return rc;
+       }
+
+       return 0;
+}
+
+static int
+sfc_mae_rule_process_pattern_data(struct sfc_mae_parse_ctx *ctx,
+                                 struct rte_flow_error *error)
+{
+       struct sfc_mae_pattern_data *pdata = &ctx->pattern_data;
+       struct sfc_mae_ethertype *ethertypes = pdata->ethertypes;
+       const rte_be16_t supported_tpids[] = {
+               /* VLAN standard TPID (always the first element) */
+               RTE_BE16(RTE_ETHER_TYPE_VLAN),
+
+               /* Double-tagging TPIDs */
+               RTE_BE16(RTE_ETHER_TYPE_QINQ),
+               RTE_BE16(RTE_ETHER_TYPE_QINQ1),
+               RTE_BE16(RTE_ETHER_TYPE_QINQ2),
+               RTE_BE16(RTE_ETHER_TYPE_QINQ3),
+       };
+       unsigned int nb_supported_tpids = RTE_DIM(supported_tpids);
+       unsigned int ethertype_idx;
+       int rc;
+
+       /*
+        * sfc_mae_rule_parse_item_vlan() has already made sure
+        * that pdata->nb_vlan_tags does not exceed this figure.
+        */
+       RTE_BUILD_BUG_ON(SFC_MAE_MATCH_VLAN_MAX_NTAGS != 2);
+
+       for (ethertype_idx = 0;
+            ethertype_idx < pdata->nb_vlan_tags; ++ethertype_idx) {
+               unsigned int tpid_idx;
+
+               /* Exact match is supported only. */
+               if (ethertypes[ethertype_idx].mask != RTE_BE16(0xffff)) {
+                       rc = EINVAL;
+                       goto fail;
+               }
+
+               for (tpid_idx = pdata->nb_vlan_tags - ethertype_idx - 1;
+                    tpid_idx < nb_supported_tpids; ++tpid_idx) {
+                       if (ethertypes[ethertype_idx].value ==
+                           supported_tpids[tpid_idx])
+                               break;
+               }
+
+               if (tpid_idx == nb_supported_tpids) {
+                       rc = EINVAL;
+                       goto fail;
+               }
+
+               nb_supported_tpids = 1;
+       }
+
+       /*
+        * Now, when the number of VLAN tags is known, set fields
+        * ETHER_TYPE, VLAN0_PROTO and VLAN1_PROTO so that the first
+        * one is either a valid L3 EtherType (or 0x0000/0x0000),
+        * and the last two are valid TPIDs (or 0x0000/0x0000).
+        */
+       rc = sfc_mae_set_ethertypes(ctx);
+       if (rc != 0)
+               goto fail;
+
+       return 0;
+
+fail:
+       return rte_flow_error_set(error, rc, RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+                                 "Failed to process pattern data");
+}
+
 static int
 sfc_mae_rule_parse_item_port_id(const struct rte_flow_item *item,
                                struct sfc_flow_parse_ctx *ctx,
@@ -486,6 +602,16 @@ sfc_mae_rule_parse_item_vf(const struct rte_flow_item *item,
        return 0;
 }
 
+/*
+ * Having this field ID in a field locator means that this
+ * locator cannot be used to actually set the field at the
+ * time when the corresponding item gets encountered. Such
+ * fields get stashed in the parsing context instead. This
+ * is required to resolve dependencies between the stashed
+ * fields. See sfc_mae_rule_process_pattern_data().
+ */
+#define SFC_MAE_FIELD_HANDLING_DEFERRED        EFX_MAE_FIELD_NIDS
+
 struct sfc_mae_field_locator {
        efx_mae_field_id_t              field_id;
        size_t                          size;
@@ -522,6 +648,9 @@ sfc_mae_parse_item(const struct sfc_mae_field_locator *field_locators,
        for (i = 0; i < nb_field_locators; ++i) {
                const struct sfc_mae_field_locator *fl = &field_locators[i];
 
+               if (fl->field_id == SFC_MAE_FIELD_HANDLING_DEFERRED)
+                       continue;
+
                rc = efx_mae_match_spec_field_set(efx_spec, fl->field_id,
                                                  fl->size, spec + fl->ofst,
                                                  fl->size, mask + fl->ofst);
@@ -539,7 +668,11 @@ sfc_mae_parse_item(const struct sfc_mae_field_locator *field_locators,
 
 static const struct sfc_mae_field_locator flocs_eth[] = {
        {
-               EFX_MAE_FIELD_ETHER_TYPE_BE,
+               /*
+                * This locator is used only for building supported fields mask.
+                * The field is handled by sfc_mae_rule_process_pattern_data().
+                */
+               SFC_MAE_FIELD_HANDLING_DEFERRED,
                RTE_SIZEOF_FIELD(struct rte_flow_item_eth, type),
                offsetof(struct rte_flow_item_eth, type),
        },
@@ -577,14 +710,128 @@ sfc_mae_rule_parse_item_eth(const struct rte_flow_item *item,
        if (rc != 0)
                return rc;
 
-       /* If "spec" is not set, could be any Ethernet */
-       if (spec == NULL)
+       if (spec != NULL) {
+               struct sfc_mae_pattern_data *pdata = &ctx_mae->pattern_data;
+               struct sfc_mae_ethertype *ethertypes = pdata->ethertypes;
+               const struct rte_flow_item_eth *item_spec;
+               const struct rte_flow_item_eth *item_mask;
+
+               item_spec = (const struct rte_flow_item_eth *)spec;
+               item_mask = (const struct rte_flow_item_eth *)mask;
+
+               ethertypes[0].value = item_spec->type;
+               ethertypes[0].mask = item_mask->type;
+       } else {
+               /*
+                * The specification is empty. This is wrong in the case
+                * when there are more network patterns in line. Other
+                * than that, any Ethernet can match. All of that is
+                * checked at the end of parsing.
+                */
                return 0;
+       }
 
        return sfc_mae_parse_item(flocs_eth, RTE_DIM(flocs_eth), spec, mask,
                                  ctx_mae->match_spec_action, error);
 }
 
+static const struct sfc_mae_field_locator flocs_vlan[] = {
+       /* Outermost tag */
+       {
+               EFX_MAE_FIELD_VLAN0_TCI_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_vlan, tci),
+               offsetof(struct rte_flow_item_vlan, tci),
+       },
+       {
+               /*
+                * This locator is used only for building supported fields mask.
+                * The field is handled by sfc_mae_rule_process_pattern_data().
+                */
+               SFC_MAE_FIELD_HANDLING_DEFERRED,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_vlan, inner_type),
+               offsetof(struct rte_flow_item_vlan, inner_type),
+       },
+
+       /* Innermost tag */
+       {
+               EFX_MAE_FIELD_VLAN1_TCI_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_vlan, tci),
+               offsetof(struct rte_flow_item_vlan, tci),
+       },
+       {
+               /*
+                * This locator is used only for building supported fields mask.
+                * The field is handled by sfc_mae_rule_process_pattern_data().
+                */
+               SFC_MAE_FIELD_HANDLING_DEFERRED,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_vlan, inner_type),
+               offsetof(struct rte_flow_item_vlan, inner_type),
+       },
+};
+
+static int
+sfc_mae_rule_parse_item_vlan(const struct rte_flow_item *item,
+                            struct sfc_flow_parse_ctx *ctx,
+                            struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       struct sfc_mae_pattern_data *pdata = &ctx_mae->pattern_data;
+       const struct sfc_mae_field_locator *flocs;
+       struct rte_flow_item_vlan supp_mask;
+       const uint8_t *spec = NULL;
+       const uint8_t *mask = NULL;
+       unsigned int nb_flocs;
+       int rc;
+
+       RTE_BUILD_BUG_ON(SFC_MAE_MATCH_VLAN_MAX_NTAGS != 2);
+
+       if (pdata->nb_vlan_tags == SFC_MAE_MATCH_VLAN_MAX_NTAGS) {
+               return rte_flow_error_set(error, ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Can't match that many VLAN tags");
+       }
+
+       nb_flocs = RTE_DIM(flocs_vlan) / SFC_MAE_MATCH_VLAN_MAX_NTAGS;
+       flocs = flocs_vlan + pdata->nb_vlan_tags * nb_flocs;
+
+       /* If parsing fails, this can remain incremented. */
+       ++pdata->nb_vlan_tags;
+
+       sfc_mae_item_build_supp_mask(flocs, nb_flocs,
+                                    &supp_mask, sizeof(supp_mask));
+
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask,
+                                &rte_flow_item_vlan_mask,
+                                sizeof(struct rte_flow_item_vlan), error);
+       if (rc != 0)
+               return rc;
+
+       if (spec != NULL) {
+               struct sfc_mae_ethertype *ethertypes = pdata->ethertypes;
+               const struct rte_flow_item_vlan *item_spec;
+               const struct rte_flow_item_vlan *item_mask;
+
+               item_spec = (const struct rte_flow_item_vlan *)spec;
+               item_mask = (const struct rte_flow_item_vlan *)mask;
+
+               ethertypes[pdata->nb_vlan_tags].value = item_spec->inner_type;
+               ethertypes[pdata->nb_vlan_tags].mask = item_mask->inner_type;
+       } else {
+               /*
+                * The specification is empty. This is wrong in the case
+                * when there are more network patterns in line. Other
+                * than that, any Ethernet can match. All of that is
+                * checked at the end of parsing.
+                */
+               return 0;
+       }
+
+       return sfc_mae_parse_item(flocs, nb_flocs, spec, mask,
+                                 ctx_mae->match_spec_action, error);
+}
+
 static const struct sfc_flow_item sfc_flow_items[] = {
        {
                .type = RTE_FLOW_ITEM_TYPE_PORT_ID,
@@ -637,6 +884,13 @@ static const struct sfc_flow_item sfc_flow_items[] = {
                .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
                .parse = sfc_mae_rule_parse_item_eth,
        },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_VLAN,
+               .prev_layer = SFC_FLOW_ITEM_L2,
+               .layer = SFC_FLOW_ITEM_L2,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_vlan,
+       },
 };
 
 int
@@ -670,6 +924,10 @@ sfc_mae_rule_parse_pattern(struct sfc_adapter *sa,
        if (rc != 0)
                goto fail_parse_pattern;
 
+       rc = sfc_mae_rule_process_pattern_data(&ctx_mae, error);
+       if (rc != 0)
+               goto fail_process_pattern_data;
+
        if (!efx_mae_match_spec_is_valid(sa->nic, ctx_mae.match_spec_action)) {
                rc = rte_flow_error_set(error, ENOTSUP,
                                        RTE_FLOW_ERROR_TYPE_ITEM, NULL,
@@ -682,6 +940,7 @@ sfc_mae_rule_parse_pattern(struct sfc_adapter *sa,
        return 0;
 
 fail_validate_match_spec_action:
+fail_process_pattern_data:
 fail_parse_pattern:
        efx_mae_match_spec_fini(sa->nic, ctx_mae.match_spec_action);
 
index f92e62d..e4e8ab6 100644 (file)
@@ -62,10 +62,57 @@ struct sfc_mae {
 struct sfc_adapter;
 struct sfc_flow_spec;
 
+/** This implementation supports double-tagging */
+#define SFC_MAE_MATCH_VLAN_MAX_NTAGS   (2)
+
+/** It is possible to keep track of one item ETH and two items VLAN */
+#define SFC_MAE_L2_MAX_NITEMS          (SFC_MAE_MATCH_VLAN_MAX_NTAGS + 1)
+
+/** Auxiliary entry format to keep track of L2 "type" ("inner_type") */
+struct sfc_mae_ethertype {
+       rte_be16_t      value;
+       rte_be16_t      mask;
+};
+
+struct sfc_mae_pattern_data {
+       /**
+        * Keeps track of "type" ("inner_type") mask and value for each
+        * parsed L2 item in a pattern. These values/masks get filled
+        * in MAE match specification at the end of parsing. Also, this
+        * information is used to conduct consistency checks:
+        *
+        * - If an item ETH is followed by a single item VLAN,
+        *   the former must have "type" set to one of supported
+        *   TPID values (0x8100, 0x88a8, 0x9100, 0x9200, 0x9300).
+        *
+        * - If an item ETH is followed by two items VLAN, the
+        *   item ETH must have "type" set to one of supported TPID
+        *   values (0x88a8, 0x9100, 0x9200, 0x9300), and the outermost
+        *   VLAN item must have "inner_type" set to TPID value 0x8100.
+        *
+        * In turn, mapping between RTE convention (above requirements) and
+        * MAE fields is non-trivial. The following scheme indicates
+        * which item EtherTypes go to which MAE fields in the case
+        * of single tag:
+        *
+        * ETH  (0x8100)        --> VLAN0_PROTO_BE
+        * VLAN (L3 EtherType)  --> ETHER_TYPE_BE
+        *
+        * Similarly, in the case of double tagging:
+        *
+        * ETH  (0x88a8)        --> VLAN0_PROTO_BE
+        * VLAN (0x8100)        --> VLAN1_PROTO_BE
+        * VLAN (L3 EtherType)  --> ETHER_TYPE_BE
+        */
+       struct sfc_mae_ethertype        ethertypes[SFC_MAE_L2_MAX_NITEMS];
+       unsigned int                    nb_vlan_tags;
+};
+
 struct sfc_mae_parse_ctx {
        struct sfc_adapter              *sa;
        efx_mae_match_spec_t            *match_spec_action;
        bool                            match_mport_set;
+       struct sfc_mae_pattern_data     pattern_data;
 };
 
 int sfc_mae_attach(struct sfc_adapter *sa);