diff options
Diffstat (limited to 'src/vnet/adj')
-rw-r--r-- | src/vnet/adj/adj.c | 607 | ||||
-rw-r--r-- | src/vnet/adj/adj.h | 379 | ||||
-rw-r--r-- | src/vnet/adj/adj_bfd.c | 184 | ||||
-rw-r--r-- | src/vnet/adj/adj_delegate.c | 144 | ||||
-rw-r--r-- | src/vnet/adj/adj_delegate.h | 104 | ||||
-rw-r--r-- | src/vnet/adj/adj_glean.c | 285 | ||||
-rw-r--r-- | src/vnet/adj/adj_glean.h | 61 | ||||
-rw-r--r-- | src/vnet/adj/adj_internal.h | 112 | ||||
-rw-r--r-- | src/vnet/adj/adj_l2.c | 192 | ||||
-rw-r--r-- | src/vnet/adj/adj_l2.h | 24 | ||||
-rw-r--r-- | src/vnet/adj/adj_mcast.c | 483 | ||||
-rw-r--r-- | src/vnet/adj/adj_mcast.h | 114 | ||||
-rw-r--r-- | src/vnet/adj/adj_midchain.c | 666 | ||||
-rw-r--r-- | src/vnet/adj/adj_midchain.h | 82 | ||||
-rw-r--r-- | src/vnet/adj/adj_nbr.c | 1124 | ||||
-rw-r--r-- | src/vnet/adj/adj_nbr.h | 176 | ||||
-rw-r--r-- | src/vnet/adj/adj_nsh.c | 211 | ||||
-rw-r--r-- | src/vnet/adj/adj_nsh.h | 31 | ||||
-rw-r--r-- | src/vnet/adj/adj_types.h | 53 | ||||
-rw-r--r-- | src/vnet/adj/rewrite.c | 234 | ||||
-rw-r--r-- | src/vnet/adj/rewrite.h | 350 |
21 files changed, 5616 insertions, 0 deletions
diff --git a/src/vnet/adj/adj.c b/src/vnet/adj/adj.c new file mode 100644 index 00000000..f8496913 --- /dev/null +++ b/src/vnet/adj/adj.c @@ -0,0 +1,607 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/adj/adj.h> +#include <vnet/adj/adj_internal.h> +#include <vnet/adj/adj_glean.h> +#include <vnet/adj/adj_midchain.h> +#include <vnet/adj/adj_mcast.h> +#include <vnet/adj/adj_delegate.h> +#include <vnet/fib/fib_node_list.h> + +/* Adjacency packet/byte counters indexed by adjacency index. */ +vlib_combined_counter_main_t adjacency_counters; + +/* + * the single adj pool + */ +ip_adjacency_t *adj_pool; + +/** + * @brief Global Config for enabling per-adjacency counters. + * By default these are disabled. + */ +int adj_per_adj_counters; + +always_inline void +adj_poison (ip_adjacency_t * adj) +{ + if (CLIB_DEBUG > 0) + { + memset (adj, 0xfe, sizeof (adj[0])); + } +} + +ip_adjacency_t * +adj_alloc (fib_protocol_t proto) +{ + ip_adjacency_t *adj; + + pool_get_aligned(adj_pool, adj, CLIB_CACHE_LINE_BYTES); + + adj_poison(adj); + + /* Make sure certain fields are always initialized. */ + /* Validate adjacency counters. */ + vlib_validate_combined_counter(&adjacency_counters, + adj_get_index(adj)); + + fib_node_init(&adj->ia_node, + FIB_NODE_TYPE_ADJ); + + adj->ia_nh_proto = proto; + adj->ia_flags = 0; + adj->rewrite_header.sw_if_index = ~0; + adj->rewrite_header.flags = 0; + adj->lookup_next_index = 0; + adj->ia_delegates = NULL; + + /* lest it become a midchain in the future */ + memset(&adj->sub_type.midchain.next_dpo, 0, + sizeof(adj->sub_type.midchain.next_dpo)); + + return (adj); +} + +static int +adj_index_is_special (adj_index_t adj_index) +{ + if (ADJ_INDEX_INVALID == adj_index) + return (!0); + + return (0); +} + +/** + * @brief Pretty print helper function for formatting specific adjacencies. + * @param s - input string to format + * @param args - other args passed to format function such as: + * - vnet_main_t + * - ip_lookup_main_t + * - adj_index + */ +u8 * +format_ip_adjacency (u8 * s, va_list * args) +{ + format_ip_adjacency_flags_t fiaf; + ip_adjacency_t * adj; + u32 adj_index; + + adj_index = va_arg (*args, u32); + fiaf = va_arg (*args, format_ip_adjacency_flags_t); + adj = adj_get(adj_index); + + switch (adj->lookup_next_index) + { + case IP_LOOKUP_NEXT_REWRITE: + s = format (s, "%U", format_adj_nbr, adj_index, 0); + break; + case IP_LOOKUP_NEXT_ARP: + s = format (s, "%U", format_adj_nbr_incomplete, adj_index, 0); + break; + case IP_LOOKUP_NEXT_GLEAN: + s = format (s, "%U", format_adj_glean, adj_index, 0); + break; + case IP_LOOKUP_NEXT_MIDCHAIN: + s = format (s, "%U", format_adj_midchain, adj_index, 2); + break; + case IP_LOOKUP_NEXT_MCAST: + s = format (s, "%U", format_adj_mcast, adj_index, 0); + break; + case IP_LOOKUP_NEXT_MCAST_MIDCHAIN: + s = format (s, "%U", format_adj_mcast_midchain, adj_index, 0); + break; + default: + break; + } + + if (fiaf & FORMAT_IP_ADJACENCY_DETAIL) + { + adj_delegate_type_t adt; + adj_delegate_t *aed; + vlib_counter_t counts; + + vlib_get_combined_counter(&adjacency_counters, adj_index, &counts); + s = format (s, "\n counts:[%Ld:%Ld]", counts.packets, counts.bytes); + s = format (s, "\n locks:%d", adj->ia_node.fn_locks); + s = format(s, "\n delegates:\n "); + FOR_EACH_ADJ_DELEGATE(adj, adt, aed, + { + s = format(s, " %U\n", format_adj_deletegate, aed); + }); + + s = format(s, "\n children:\n "); + s = fib_node_children_format(adj->ia_node.fn_children, s); + } + + return s; +} + +/* + * adj_last_lock_gone + * + * last lock/reference to the adj has gone, we no longer need it. + */ +static void +adj_last_lock_gone (ip_adjacency_t *adj) +{ + vlib_main_t * vm = vlib_get_main(); + + ASSERT(0 == fib_node_list_get_size(adj->ia_node.fn_children)); + ADJ_DBG(adj, "last-lock-gone"); + + vlib_worker_thread_barrier_sync (vm); + + switch (adj->lookup_next_index) + { + case IP_LOOKUP_NEXT_MIDCHAIN: + dpo_reset(&adj->sub_type.midchain.next_dpo); + /* FALL THROUGH */ + case IP_LOOKUP_NEXT_ARP: + case IP_LOOKUP_NEXT_REWRITE: + /* + * complete and incomplete nbr adjs + */ + adj_nbr_remove(adj_get_index(adj), + adj->ia_nh_proto, + adj->ia_link, + &adj->sub_type.nbr.next_hop, + adj->rewrite_header.sw_if_index); + break; + case IP_LOOKUP_NEXT_GLEAN: + adj_glean_remove(adj->ia_nh_proto, + adj->rewrite_header.sw_if_index); + break; + case IP_LOOKUP_NEXT_MCAST: + case IP_LOOKUP_NEXT_MCAST_MIDCHAIN: + adj_mcast_remove(adj->ia_nh_proto, + adj->rewrite_header.sw_if_index); + break; + case IP_LOOKUP_NEXT_DROP: + case IP_LOOKUP_NEXT_PUNT: + case IP_LOOKUP_NEXT_LOCAL: + case IP_LOOKUP_NEXT_ICMP_ERROR: + case IP_LOOKUP_N_NEXT: + /* + * type not stored in any DB from which we need to remove it + */ + break; + } + + vlib_worker_thread_barrier_release(vm); + + fib_node_deinit(&adj->ia_node); + ASSERT(0 == vec_len(adj->ia_delegates)); + vec_free(adj->ia_delegates); + pool_put(adj_pool, adj); +} + +void +adj_lock (adj_index_t adj_index) +{ + ip_adjacency_t *adj; + + if (adj_index_is_special(adj_index)) + { + return; + } + + adj = adj_get(adj_index); + ASSERT(adj); + + ADJ_DBG(adj, "lock"); + fib_node_lock(&adj->ia_node); +} + +void +adj_unlock (adj_index_t adj_index) +{ + ip_adjacency_t *adj; + + if (adj_index_is_special(adj_index)) + { + return; + } + + adj = adj_get(adj_index); + ASSERT(adj); + + ADJ_DBG(adj, "unlock"); + ASSERT(adj); + + fib_node_unlock(&adj->ia_node); +} + +u32 +adj_child_add (adj_index_t adj_index, + fib_node_type_t child_type, + fib_node_index_t child_index) +{ + ASSERT(ADJ_INDEX_INVALID != adj_index); + if (adj_index_is_special(adj_index)) + { + return (~0); + } + + return (fib_node_child_add(FIB_NODE_TYPE_ADJ, + adj_index, + child_type, + child_index)); +} + +void +adj_child_remove (adj_index_t adj_index, + u32 sibling_index) +{ + if (adj_index_is_special(adj_index)) + { + return; + } + + fib_node_child_remove(FIB_NODE_TYPE_ADJ, + adj_index, + sibling_index); +} + +/* + * Context for the walk to update the cached feture flags. + */ +typedef struct adj_feature_update_t_ +{ + u8 arc; + u8 enable; +} adj_feature_update_ctx_t; + +static adj_walk_rc_t +adj_feature_update_walk_cb (adj_index_t ai, + void *arg) +{ + adj_feature_update_ctx_t *ctx = arg; + ip_adjacency_t *adj; + + adj = adj_get(ai); + + /* + * this ugly mess matches the feature arc that is changing with affected + * adjacencies + */ + if (((ctx->arc == ip6_main.lookup_main.output_feature_arc_index) && + (VNET_LINK_IP6 == adj->ia_link)) || + ((ctx->arc == ip4_main.lookup_main.output_feature_arc_index) && + (VNET_LINK_IP4 == adj->ia_link)) || + ((ctx->arc == mpls_main.output_feature_arc_index) && + (VNET_LINK_MPLS == adj->ia_link))) + { + if (ctx->enable) + adj->rewrite_header.flags |= VNET_REWRITE_HAS_FEATURES; + else + adj->rewrite_header.flags &= ~VNET_REWRITE_HAS_FEATURES; + } + return (ADJ_WALK_RC_CONTINUE); +} + +void +adj_feature_update (u32 sw_if_index, + u8 arc_index, + u8 is_enable) +{ + /* + * Walk all the adjacencies on the interface to update the cached + * 'has-features' flag + */ + adj_feature_update_ctx_t ctx = { + .arc = arc_index, + .enable = is_enable, + }; + adj_walk (sw_if_index, adj_feature_update_walk_cb, &ctx); +} + +/** + * @brief Walk the Adjacencies on a given interface + */ +void +adj_walk (u32 sw_if_index, + adj_walk_cb_t cb, + void *ctx) +{ + /* + * walk all the neighbor adjacencies + */ + fib_protocol_t proto; + + FOR_EACH_FIB_IP_PROTOCOL(proto) + { + adj_nbr_walk(sw_if_index, proto, cb, ctx); + adj_mcast_walk(sw_if_index, proto, cb, ctx); + } +} + +/** + * @brief Return the link type of the adjacency + */ +vnet_link_t +adj_get_link_type (adj_index_t ai) +{ + const ip_adjacency_t *adj; + + adj = adj_get(ai); + + return (adj->ia_link); +} + +/** + * @brief Return the sw interface index of the adjacency. + */ +u32 +adj_get_sw_if_index (adj_index_t ai) +{ + const ip_adjacency_t *adj; + + adj = adj_get(ai); + + return (adj->rewrite_header.sw_if_index); +} + +/** + * @brief Return true if the adjacency is 'UP', i.e. can be used for forwarding + * 0 is down, !0 is up. + */ +int +adj_is_up (adj_index_t ai) +{ + const adj_delegate_t *aed; + + aed = adj_delegate_get(adj_get(ai), ADJ_DELEGATE_BFD); + + if (NULL == aed) + { + /* + * no BFD tracking - resolved + */ + return (!0); + } + else + { + /* + * defer to the state of the BFD tracking + */ + return (ADJ_BFD_STATE_UP == aed->ad_bfd_state); + } +} + +/** + * @brief Return the rewrite string of the adjacency + */ +const u8* +adj_get_rewrite (adj_index_t ai) +{ + vnet_rewrite_header_t *rw; + ip_adjacency_t *adj; + + adj = adj_get(ai); + rw = &adj->rewrite_header; + + ASSERT (rw->data_bytes != 0xfefe); + + return (rw->data - rw->data_bytes); +} + +static fib_node_t * +adj_get_node (fib_node_index_t index) +{ + ip_adjacency_t *adj; + + adj = adj_get(index); + + return (&adj->ia_node); +} + +#define ADJ_FROM_NODE(_node) \ + ((ip_adjacency_t*)((char*)_node - STRUCT_OFFSET_OF(ip_adjacency_t, ia_node))) + +static void +adj_node_last_lock_gone (fib_node_t *node) +{ + adj_last_lock_gone(ADJ_FROM_NODE(node)); +} + +static fib_node_back_walk_rc_t +adj_back_walk_notify (fib_node_t *node, + fib_node_back_walk_ctx_t *ctx) +{ + /* + * Que pasa. yo soj en el final! + */ + ASSERT(0); + + return (FIB_NODE_BACK_WALK_CONTINUE); +} + +/* + * Adjacency's graph node virtual function table + */ +static const fib_node_vft_t adj_vft = { + .fnv_get = adj_get_node, + .fnv_last_lock = adj_node_last_lock_gone, + .fnv_back_walk = adj_back_walk_notify, +}; + +static clib_error_t * +adj_module_init (vlib_main_t * vm) +{ + fib_node_register_type(FIB_NODE_TYPE_ADJ, &adj_vft); + + adj_nbr_module_init(); + adj_glean_module_init(); + adj_midchain_module_init(); + adj_mcast_module_init(); + + return (NULL); +} + +VLIB_INIT_FUNCTION (adj_module_init); + +static clib_error_t * +adj_show (vlib_main_t * vm, + unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + adj_index_t ai = ADJ_INDEX_INVALID; + u32 sw_if_index = ~0; + int summary = 0; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "%d", &ai)) + ; + else if (unformat (input, "sum")) + summary = 1; + else if (unformat (input, "summary")) + summary = 1; + else if (unformat (input, "%U", + unformat_vnet_sw_interface, vnet_get_main(), + &sw_if_index)) + ; + else + break; + } + + if (summary) + { + vlib_cli_output (vm, "Number of adjacenies: %d", pool_elts(adj_pool)); + vlib_cli_output (vm, "Per-adjacency counters: %s", + (adj_are_counters_enabled() ? + "enabled": + "disabled")); + } + else + { + if (ADJ_INDEX_INVALID != ai) + { + if (pool_is_free_index(adj_pool, ai)) + { + vlib_cli_output (vm, "adjacency %d invalid", ai); + return 0; + } + + vlib_cli_output (vm, "[@%d] %U", + ai, + format_ip_adjacency, ai, + FORMAT_IP_ADJACENCY_DETAIL); + } + else + { + /* *INDENT-OFF* */ + pool_foreach_index(ai, adj_pool, + ({ + if (~0 != sw_if_index && + sw_if_index != adj_get_sw_if_index(ai)) + { + } + else + { + vlib_cli_output (vm, "[@%d] %U", + ai, + format_ip_adjacency, ai, + FORMAT_IP_ADJACENCY_NONE); + } + })); + /* *INDENT-ON* */ + } + } + return 0; +} + +/*? + * Show all adjacencies. + * @cliexpar + * @cliexstart{sh adj} + * [@0] + * [@1] glean: loop0 + * [@2] ipv4 via 1.0.0.2 loop0: IP4: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc + * [@3] mpls via 1.0.0.2 loop0: MPLS: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc + * [@4] ipv4 via 1.0.0.3 loop0: IP4: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc + * [@5] mpls via 1.0.0.3 loop0: MPLS: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc + * @cliexend + ?*/ +VLIB_CLI_COMMAND (adj_show_command, static) = { + .path = "show adj", + .short_help = "show adj [<adj_index>] [interface] [summary]", + .function = adj_show, +}; + +/** + * @brief CLI invoked function to enable/disable per-adj counters + */ +static clib_error_t * +adj_cli_counters_set (vlib_main_t * vm, + unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + clib_error_t *error = NULL; + int enable = ~0; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "enable")) + enable = 1; + else if (unformat (input, "disable")) + enable = 0; + else + break; + } + + if (enable != ~0) + { + /* user requested something sensible */ + adj_per_adj_counters = enable; + } + else + { + error = clib_error_return (0, "specify 'enable' or 'disable'"); + } + + return (error); +} + +/*? + * Enabe/disble per-adjacency counters. This is optional because it comes with + * a non-negligible performance cost. + ?*/ +VLIB_CLI_COMMAND (adj_cli_counters_set_command, static) = { + .path = "adjacency counters", + .short_help = "adjacency counters [enable|disable]", + .function = adj_cli_counters_set, +}; diff --git a/src/vnet/adj/adj.h b/src/vnet/adj/adj.h new file mode 100644 index 00000000..ed5eb1f1 --- /dev/null +++ b/src/vnet/adj/adj.h @@ -0,0 +1,379 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +/** + * An adjacency is a representation of an attached L3 peer. + * + * Adjacency Sub-types: + * - neighbour: a representation of an attached L3 peer. + * Key:{addr,interface,link/ether-type} + * SHARED + * - glean: used to drive ARP/ND for packets destined to a local sub-net. + * 'glean' mean use the packet's destination address as the target + * address in the ARP packet. + * UNSHARED. Only one per-interface. + * - midchain: a nighbour adj on a virtual/tunnel interface. + * + * The API to create and update the adjacency is very sub-type specific. This + * is intentional as it encourages the user to carefully consider which adjacency + * sub-type they are really using, and hence assign it data in the appropriate + * sub-type space in the union of sub-types. This prevents the adj becoming a + * disorganised dumping group for 'my features needs a u16 somewhere' data. It + * is important to enforce this approach as space in the adjacency is a premium, + * as we need it to fit in 1 cache line. + * + * the API is also based around an index to an ajdacency not a raw pointer. This + * is so the user doesn't suffer the same limp inducing firearm injuries that + * the author suffered as the adjacenices can realloc. + */ + +#ifndef __ADJ_H__ +#define __ADJ_H__ + +#include <vnet/adj/adj_types.h> +#include <vnet/adj/adj_nbr.h> +#include <vnet/adj/adj_glean.h> +#include <vnet/adj/rewrite.h> + +/** @brief Common (IP4/IP6) next index stored in adjacency. */ +typedef enum +{ + /** Adjacency to drop this packet. */ + IP_LOOKUP_NEXT_DROP, + /** Adjacency to punt this packet. */ + IP_LOOKUP_NEXT_PUNT, + + /** This packet is for one of our own IP addresses. */ + IP_LOOKUP_NEXT_LOCAL, + + /** This packet matches an "incomplete adjacency" and packets + need to be passed to ARP to find rewrite string for + this destination. */ + IP_LOOKUP_NEXT_ARP, + + /** This packet matches an "interface route" and packets + need to be passed to ARP to find rewrite string for + this destination. */ + IP_LOOKUP_NEXT_GLEAN, + + /** This packet is to be rewritten and forwarded to the next + processing node. This is typically the output interface but + might be another node for further output processing. */ + IP_LOOKUP_NEXT_REWRITE, + + /** This packets follow a mid-chain adjacency */ + IP_LOOKUP_NEXT_MIDCHAIN, + + /** This packets needs to go to ICMP error */ + IP_LOOKUP_NEXT_ICMP_ERROR, + + /** Multicast Adjacency. */ + IP_LOOKUP_NEXT_MCAST, + + /** Multicast Midchain Adjacency. An Adjacency for sending macst packets + * on a tunnel/virtual interface */ + IP_LOOKUP_NEXT_MCAST_MIDCHAIN, + + IP_LOOKUP_N_NEXT, +} __attribute__ ((packed)) ip_lookup_next_t; + +typedef enum +{ + IP4_LOOKUP_N_NEXT = IP_LOOKUP_N_NEXT, +} ip4_lookup_next_t; + +typedef enum +{ + /* Hop-by-hop header handling */ + IP6_LOOKUP_NEXT_HOP_BY_HOP = IP_LOOKUP_N_NEXT, + IP6_LOOKUP_NEXT_ADD_HOP_BY_HOP, + IP6_LOOKUP_NEXT_POP_HOP_BY_HOP, + IP6_LOOKUP_N_NEXT, +} ip6_lookup_next_t; + +#define IP4_LOOKUP_NEXT_NODES { \ + [IP_LOOKUP_NEXT_DROP] = "ip4-drop", \ + [IP_LOOKUP_NEXT_PUNT] = "ip4-punt", \ + [IP_LOOKUP_NEXT_LOCAL] = "ip4-local", \ + [IP_LOOKUP_NEXT_ARP] = "ip4-arp", \ + [IP_LOOKUP_NEXT_GLEAN] = "ip4-glean", \ + [IP_LOOKUP_NEXT_REWRITE] = "ip4-rewrite", \ + [IP_LOOKUP_NEXT_MCAST] = "ip4-rewrite-mcast", \ + [IP_LOOKUP_NEXT_MIDCHAIN] = "ip4-midchain", \ + [IP_LOOKUP_NEXT_MCAST_MIDCHAIN] = "ip4-mcast-midchain", \ + [IP_LOOKUP_NEXT_ICMP_ERROR] = "ip4-icmp-error", \ +} + +#define IP6_LOOKUP_NEXT_NODES { \ + [IP_LOOKUP_NEXT_DROP] = "ip6-drop", \ + [IP_LOOKUP_NEXT_PUNT] = "ip6-punt", \ + [IP_LOOKUP_NEXT_LOCAL] = "ip6-local", \ + [IP_LOOKUP_NEXT_ARP] = "ip6-discover-neighbor", \ + [IP_LOOKUP_NEXT_GLEAN] = "ip6-glean", \ + [IP_LOOKUP_NEXT_REWRITE] = "ip6-rewrite", \ + [IP_LOOKUP_NEXT_MCAST] = "ip6-rewrite-mcast", \ + [IP_LOOKUP_NEXT_MIDCHAIN] = "ip6-midchain", \ + [IP_LOOKUP_NEXT_MCAST_MIDCHAIN] = "ip6-mcast-midchain", \ + [IP_LOOKUP_NEXT_ICMP_ERROR] = "ip6-icmp-error", \ + [IP6_LOOKUP_NEXT_HOP_BY_HOP] = "ip6-hop-by-hop", \ + [IP6_LOOKUP_NEXT_ADD_HOP_BY_HOP] = "ip6-add-hop-by-hop", \ + [IP6_LOOKUP_NEXT_POP_HOP_BY_HOP] = "ip6-pop-hop-by-hop", \ +} + +/** + * Forward delcartion + */ +struct ip_adjacency_t_; + +/** + * @brief A function type for post-rewrite fixups on midchain adjacency + */ +typedef void (*adj_midchain_fixup_t) (vlib_main_t * vm, + struct ip_adjacency_t_ * adj, + vlib_buffer_t * b0); + +/** + * @brief Flags on an IP adjacency + */ +typedef enum ip_adjacency_flags_t_ +{ + ADJ_FLAG_NONE = 0, + + /** + * Currently a sync walk is active. Used to prevent re-entrant walking + */ + ADJ_FLAG_SYNC_WALK_ACTIVE = (1 << 0), + + /** + * Packets TX through the midchain do not increment the interface + * counters. This should be used when the adj is associated with an L2 + * interface and that L2 interface is in a bridege domain. In that case + * the packet will have traversed the interface's TX node, and hence have + * been counted, before it traverses ths midchain + */ + ADJ_FLAG_MIDCHAIN_NO_COUNT = (1 << 1), +} __attribute__ ((packed)) adj_flags_t; + +/** + * @brief IP unicast adjacency. + * @note cache aligned. + * + * An adjacency is a represenation of a peer on a particular link. + */ +typedef struct ip_adjacency_t_ +{ + CLIB_CACHE_LINE_ALIGN_MARK (cacheline0); + + /** + * Linkage into the FIB node grpah. First member since this type + * has 8 byte alignment requirements. + */ + fib_node_t ia_node; + + /** + * Next hop after ip4-lookup. + * This is not accessed in the rewrite nodes. + * 1-bytes + */ + ip_lookup_next_t lookup_next_index; + + /** + * link/ether-type + * 1 bytes + */ + vnet_link_t ia_link; + + /** + * The protocol of the neighbor/peer. i.e. the protocol with + * which to interpret the 'next-hop' attirbutes of the sub-types. + * 1-btyes + */ + fib_protocol_t ia_nh_proto; + + /** + * Flags on the adjacency + * 1-bytes + */ + adj_flags_t ia_flags; + + union + { + /** + * IP_LOOKUP_NEXT_ARP/IP_LOOKUP_NEXT_REWRITE + * + * neighbour adjacency sub-type; + */ + struct + { + ip46_address_t next_hop; + } nbr; + /** + * IP_LOOKUP_NEXT_MIDCHAIN + * + * A nbr adj that is also recursive. Think tunnels. + * A nbr adj can transition to be of type MDICHAIN + * so be sure to leave the two structs with the next_hop + * fields aligned. + */ + struct + { + /** + * The recursive next-hop. + * This field MUST be at the same memory location as + * sub_type.nbr.next_hop + */ + ip46_address_t next_hop; + /** + * The next DPO to use + */ + dpo_id_t next_dpo; + /** + * A function to perform the post-rewrite fixup + */ + adj_midchain_fixup_t fixup_func; + } midchain; + /** + * IP_LOOKUP_NEXT_GLEAN + * + * Glean the address to ARP for from the packet's destination. + * Technically these aren't adjacencies, i.e. they are not a + * representation of a peer. One day we might untangle this coupling + * and use a new Glean DPO. + */ + struct + { + ip46_address_t receive_addr; + } glean; + } sub_type; + + CLIB_CACHE_LINE_ALIGN_MARK (cacheline1); + + /* Rewrite in second/third cache lines */ + vnet_declare_rewrite (VLIB_BUFFER_PRE_DATA_SIZE); + + /** + * more control plane members that do not fit on the first cacheline + */ + /** + * A sorted vector of delegates + */ + struct adj_delegate_t_ *ia_delegates; + +} ip_adjacency_t; + +STATIC_ASSERT ((STRUCT_OFFSET_OF (ip_adjacency_t, cacheline0) == 0), + "IP adjacency cachline 0 is not offset"); +STATIC_ASSERT ((STRUCT_OFFSET_OF (ip_adjacency_t, cacheline1) == + CLIB_CACHE_LINE_BYTES), + "IP adjacency cachline 1 is more than one cachline size offset"); + +/** + * @brief + * Take a reference counting lock on the adjacency + */ +extern void adj_lock(adj_index_t adj_index); +/** + * @brief + * Release a reference counting lock on the adjacency + */ +extern void adj_unlock(adj_index_t adj_index); + +/** + * @brief + * Add a child dependent to an adjacency. The child will + * thus be informed via its registerd back-walk function + * when the adjacency state changes. + */ +extern u32 adj_child_add(adj_index_t adj_index, + fib_node_type_t type, + fib_node_index_t child_index); +/** + * @brief + * Remove a child dependent + */ +extern void adj_child_remove(adj_index_t adj_index, + u32 sibling_index); + +/** + * @brief Walk the Adjacencies on a given interface + */ +extern void adj_walk (u32 sw_if_index, + adj_walk_cb_t cb, + void *ctx); + +/** + * @brief Return the link type of the adjacency + */ +extern vnet_link_t adj_get_link_type (adj_index_t ai); + +/** + * @brief Return the sw interface index of the adjacency. + */ +extern u32 adj_get_sw_if_index (adj_index_t ai); + +/** + * @brief Return true if the adjacency is 'UP', i.e. can be used for forwarding. + * 0 is down, !0 is up. + */ +extern int adj_is_up (adj_index_t ai); + +/** + * @brief Return the link type of the adjacency + */ +extern const u8* adj_get_rewrite (adj_index_t ai); + +/** + * @brief Notify the adjacency subsystem that the features settings for + * an interface have changed + */ +extern void adj_feature_update (u32 sw_if_index, u8 arc_index, u8 is_enable); + +/** + * @brief + * The global adjacnecy pool. Exposed for fast/inline data-plane access + */ +extern ip_adjacency_t *adj_pool; + +/** + * @brief + * Adjacency packet counters + */ +extern vlib_combined_counter_main_t adjacency_counters; + +/** + * @brief Global Config for enabling per-adjacency counters + * This is configurable because it comes with a non-negligible + * performance cost. */ +extern int adj_per_adj_counters; + +/** + * @brief + * Get a pointer to an adjacency object from its index + */ +static inline ip_adjacency_t * +adj_get (adj_index_t adj_index) +{ + return (vec_elt_at_index(adj_pool, adj_index)); +} + +/** + * @brief Get the global configuration option for enabling per-adj counters + */ +static inline int +adj_are_counters_enabled (void) +{ + return (adj_per_adj_counters); +} + +#endif diff --git a/src/vnet/adj/adj_bfd.c b/src/vnet/adj/adj_bfd.c new file mode 100644 index 00000000..3d294c46 --- /dev/null +++ b/src/vnet/adj/adj_bfd.c @@ -0,0 +1,184 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/bfd/bfd_main.h> + +#include <vnet/adj/adj_delegate.h> +#include <vnet/adj/adj_nbr.h> +#include <vnet/fib/fib_walk.h> + +static adj_bfd_state_t +adj_bfd_bfd_state_to_fib (bfd_state_e bstate) +{ + switch (bstate) + { + case BFD_STATE_up: + return (ADJ_BFD_STATE_UP); + case BFD_STATE_down: + case BFD_STATE_admin_down: + case BFD_STATE_init: + return (ADJ_BFD_STATE_DOWN); + } + return (ADJ_BFD_STATE_DOWN); +} + +static void +adj_bfd_update_walk (adj_index_t ai) +{ + /* + * initiate a backwalk of dependent children + * to notify of the state change of this adj. + */ + fib_node_back_walk_ctx_t ctx = { + .fnbw_reason = FIB_NODE_BW_REASON_FLAG_ADJ_UPDATE, + }; + fib_walk_sync(FIB_NODE_TYPE_ADJ, ai, &ctx); +} + +/** + * @brief Callback function registered with BFD module to receive notifications + * of the CRUD of BFD sessions + * would be static but for the fact it's called from the unit-tests + */ +void +adj_bfd_notify (bfd_listen_event_e event, + const bfd_session_t *session) +{ + const bfd_udp_key_t *key; + fib_protocol_t fproto; + adj_delegate_t *aed; + adj_index_t ai; + + if (BFD_HOP_TYPE_SINGLE != session->hop_type) + { + /* + * multi-hop BFD sessions attach directly to the FIB entry + * single-hop adj to the associate adjacency. + */ + return; + } + + key = &session->udp.key; + + fproto = (ip46_address_is_ip4 (&key->peer_addr) ? + FIB_PROTOCOL_IP4: + FIB_PROTOCOL_IP6); + + /* + * find the adj that corresponds to the BFD session. + */ + ai = adj_nbr_add_or_lock(fproto, + fib_proto_to_link(fproto), + &key->peer_addr, + key->sw_if_index); + + switch (event) + { + case BFD_LISTEN_EVENT_CREATE: + /* + * The creation of a new session + */ + if ((ADJ_INDEX_INVALID != ai) && + (aed = adj_delegate_get(adj_get(ai), + ADJ_DELEGATE_BFD))) + { + /* + * already got state for this adj + */ + } + else + { + /* + * lock the adj. add the delegate. + * Lockinging the adj prevents it being removed and thus maintains + * the BFD derived states + */ + adj_lock(ai); + + aed = adj_delegate_find_or_add(adj_get(ai), ADJ_DELEGATE_BFD); + + /* + * pretend the session is up and skip the walk. + * If we set it down then we get traffic loss on new children. + * if we walk then we lose traffic for existing children. Wait + * for the first BFD UP/DOWN before we let the session's state + * influence forwarding. + */ + aed->ad_bfd_state = ADJ_BFD_STATE_UP; + aed->ad_bfd_index = session->bs_idx; + } + break; + + case BFD_LISTEN_EVENT_UPDATE: + /* + * state change up/dowm and + */ + aed = adj_delegate_get(adj_get(ai), ADJ_DELEGATE_BFD); + + if (NULL != aed) + { + aed->ad_bfd_state = adj_bfd_bfd_state_to_fib(session->local_state); + adj_bfd_update_walk(ai); + } + /* + * else + * not an adj with BFD state + */ + break; + + case BFD_LISTEN_EVENT_DELETE: + /* + * session has been removed. + */ + + if (adj_delegate_get(adj_get(ai), ADJ_DELEGATE_BFD)) + { + /* + * has an associated BFD tracking delegate + * remove the BFD tracking deletgate, update children, then + * unlock the adj + */ + adj_delegate_remove(adj_get(ai), ADJ_DELEGATE_BFD); + + adj_bfd_update_walk(ai); + adj_unlock(ai); + } + /* + * else + * no BFD associated state + */ + break; + } + + /* + * unlock match of the add-or-lock at the start + */ + adj_unlock(ai); +} + +static clib_error_t * +adj_bfd_main_init (vlib_main_t * vm) +{ + clib_error_t * error = NULL; + + if ((error = vlib_call_init_function (vm, bfd_main_init))) + return (error); + + bfd_register_listener(adj_bfd_notify); + + return (error); +} + +VLIB_INIT_FUNCTION (adj_bfd_main_init); diff --git a/src/vnet/adj/adj_delegate.c b/src/vnet/adj/adj_delegate.c new file mode 100644 index 00000000..701b36e2 --- /dev/null +++ b/src/vnet/adj/adj_delegate.c @@ -0,0 +1,144 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/adj/adj_delegate.h> +#include <vnet/adj/adj.h> +#include <vnet/adj/adj_internal.h> + +static adj_delegate_t * +adj_delegate_find_i (const ip_adjacency_t *adj, + adj_delegate_type_t type, + u32 *index) +{ + adj_delegate_t *delegate; + int ii; + + ii = 0; + vec_foreach(delegate, adj->ia_delegates) + { + if (delegate->ad_type == type) + { + if (NULL != index) + *index = ii; + + return (delegate); + } + else + { + ii++; + } + } + + return (NULL); +} + +adj_delegate_t * +adj_delegate_get (const ip_adjacency_t *adj, + adj_delegate_type_t type) +{ + return (adj_delegate_find_i(adj, type, NULL)); +} + +void +adj_delegate_remove (ip_adjacency_t *adj, + adj_delegate_type_t type) +{ + adj_delegate_t *aed; + u32 index = ~0; + + aed = adj_delegate_find_i(adj, type, &index); + + ASSERT(NULL != aed); + + vec_del1(adj->ia_delegates, index); +} + +static int +adj_delegate_cmp_for_sort (void * v1, + void * v2) +{ + adj_delegate_t *delegate1 = v1, *delegate2 = v2; + + return (delegate1->ad_type - delegate2->ad_type); +} + +static void +adj_delegate_init (ip_adjacency_t *adj, + adj_delegate_type_t type) + +{ + adj_delegate_t delegate = { + .ad_adj_index = adj_get_index(adj), + .ad_type = type, + }; + + vec_add1(adj->ia_delegates, delegate); + vec_sort_with_function(adj->ia_delegates, + adj_delegate_cmp_for_sort); +} + +adj_delegate_t * +adj_delegate_find_or_add (ip_adjacency_t *adj, + adj_delegate_type_t adt) +{ + adj_delegate_t *delegate; + + delegate = adj_delegate_get(adj, adt); + + if (NULL == delegate) + { + adj_delegate_init(adj, adt); + } + + return (adj_delegate_get(adj, adt)); +} + +/** + * typedef for printing a delegate + */ +typedef u8 * (*adj_delegate_format_t)(const adj_delegate_t *aed, + u8 *s); + +/** + * Print a delegate that represents BFD tracking + */ +static u8 * +adj_delegate_fmt_bfd (const adj_delegate_t *aed, + u8 *s) +{ + s = format(s, "BFD:[state:%d index:%d]", + aed->ad_bfd_state, + aed->ad_bfd_index); + + return (s); +} + +/** + * A delegate type to formatter map + */ +static adj_delegate_format_t aed_formatters[] = +{ + [ADJ_DELEGATE_BFD] = adj_delegate_fmt_bfd, +}; + +u8 * +format_adj_deletegate (u8 * s, va_list * args) +{ + adj_delegate_t *aed; + + aed = va_arg (*args, adj_delegate_t *); + + return (aed_formatters[aed->ad_type](aed, s)); +} diff --git a/src/vnet/adj/adj_delegate.h b/src/vnet/adj/adj_delegate.h new file mode 100644 index 00000000..17651203 --- /dev/null +++ b/src/vnet/adj/adj_delegate.h @@ -0,0 +1,104 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __ADJ_DELEGATE_T__ +#define __ADJ_DELEGATE_T__ + +#include <vnet/adj/adj.h> + +/** + * Delegate types + */ +typedef enum adj_delegate_type_t_ { + /** + * BFD session state + */ + ADJ_DELEGATE_BFD, +} adj_delegate_type_t; + +#define FOR_EACH_ADJ_DELEGATE(_adj, _adt, _aed, _body) \ +{ \ + for (_adt = ADJ_DELEGATE_BFD; \ + _adt <= ADJ_DELEGATE_BFD; \ + _adt++) \ + { \ + _aed = adj_delegate_get(_adj, _adt); \ + if (NULL != _aed) { \ + _body; \ + } \ + } \ +} + +/** + * Distillation of the BFD session states into a go/no-go for using + * the associated tracked adjacency + */ +typedef enum adj_bfd_state_t_ +{ + ADJ_BFD_STATE_DOWN, + ADJ_BFD_STATE_UP, +} adj_bfd_state_t; + +/** + * A Delagate is a means to implement the Delagation design pattern; + * the extension of an object's functionality through the composition of, + * and delgation to, other objects. + * These 'other' objects are delegates. Delagates are thus attached to + * ADJ objects to extend their functionality. + */ +typedef struct adj_delegate_t_ +{ + /** + * The ADJ entry object to which the delagate is attached + */ + adj_index_t ad_adj_index; + + /** + * The delagate type + */ + adj_delegate_type_t ad_type; + + /** + * A union of data for the different delegate types + */ + union + { + /** + * BFD delegate daa + */ + struct { + /** + * BFD session state + */ + adj_bfd_state_t ad_bfd_state; + /** + * BFD session index + */ + u32 ad_bfd_index; + }; + }; +} adj_delegate_t; + +extern void adj_delegate_remove(ip_adjacency_t *adj, + adj_delegate_type_t type); + +extern adj_delegate_t *adj_delegate_find_or_add(ip_adjacency_t *adj, + adj_delegate_type_t fdt); +extern adj_delegate_t *adj_delegate_get(const ip_adjacency_t *adj, + adj_delegate_type_t type); + +extern u8 *format_adj_deletegate(u8 * s, va_list * args); + +#endif diff --git a/src/vnet/adj/adj_glean.c b/src/vnet/adj/adj_glean.c new file mode 100644 index 00000000..8d86e2a9 --- /dev/null +++ b/src/vnet/adj/adj_glean.c @@ -0,0 +1,285 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/adj/adj.h> +#include <vnet/adj/adj_internal.h> +#include <vnet/fib/fib_walk.h> + +/* + * The 'DB' of all glean adjs. + * There is only one glean per-interface per-protocol, so this is a per-interface + * vector + */ +static adj_index_t *adj_gleans[FIB_PROTOCOL_MAX]; + +static inline vlib_node_registration_t* +adj_get_glean_node (fib_protocol_t proto) +{ + switch (proto) { + case FIB_PROTOCOL_IP4: + return (&ip4_glean_node); + case FIB_PROTOCOL_IP6: + return (&ip6_glean_node); + case FIB_PROTOCOL_MPLS: + break; + } + ASSERT(0); + return (NULL); +} + +/* + * adj_glean_add_or_lock + * + * The next_hop address here is used for source address selection in the DP. + * The glean adj is added to an interface's connected prefix, the next-hop + * passed here is the local prefix on the same interface. + */ +adj_index_t +adj_glean_add_or_lock (fib_protocol_t proto, + u32 sw_if_index, + const ip46_address_t *nh_addr) +{ + ip_adjacency_t * adj; + + vec_validate_init_empty(adj_gleans[proto], sw_if_index, ADJ_INDEX_INVALID); + + if (ADJ_INDEX_INVALID == adj_gleans[proto][sw_if_index]) + { + adj = adj_alloc(proto); + + adj->lookup_next_index = IP_LOOKUP_NEXT_GLEAN; + adj->ia_nh_proto = proto; + adj_gleans[proto][sw_if_index] = adj_get_index(adj); + + if (NULL != nh_addr) + { + adj->sub_type.glean.receive_addr = *nh_addr; + } + + adj->rewrite_header.data_bytes = 0; + + vnet_rewrite_for_sw_interface(vnet_get_main(), + adj_fib_proto_2_nd(proto), + sw_if_index, + adj_get_glean_node(proto)->index, + VNET_REWRITE_FOR_SW_INTERFACE_ADDRESS_BROADCAST, + &adj->rewrite_header, + sizeof (adj->rewrite_data)); + } + else + { + adj = adj_get(adj_gleans[proto][sw_if_index]); + } + + adj_lock(adj_get_index(adj)); + + return (adj_get_index(adj)); +} + +void +adj_glean_remove (fib_protocol_t proto, + u32 sw_if_index) +{ + ASSERT(sw_if_index < vec_len(adj_gleans[proto])); + + adj_gleans[proto][sw_if_index] = ADJ_INDEX_INVALID; +} + +static clib_error_t * +adj_glean_interface_state_change (vnet_main_t * vnm, + u32 sw_if_index, + u32 flags) +{ + /* + * for each glean on the interface trigger a walk back to the children + */ + fib_protocol_t proto; + ip_adjacency_t *adj; + + + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + if (sw_if_index >= vec_len(adj_gleans[proto]) || + ADJ_INDEX_INVALID == adj_gleans[proto][sw_if_index]) + continue; + + adj = adj_get(adj_gleans[proto][sw_if_index]); + + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP ? + FIB_NODE_BW_REASON_FLAG_INTERFACE_UP : + FIB_NODE_BW_REASON_FLAG_INTERFACE_DOWN), + }; + + fib_walk_sync(FIB_NODE_TYPE_ADJ, adj_get_index(adj), &bw_ctx); + } + + return (NULL); +} + +VNET_SW_INTERFACE_ADMIN_UP_DOWN_FUNCTION(adj_glean_interface_state_change); + +/** + * @brief Invoked on each SW interface of a HW interface when the + * HW interface state changes + */ +static void +adj_nbr_hw_sw_interface_state_change (vnet_main_t * vnm, + u32 sw_if_index, + void *arg) +{ + adj_glean_interface_state_change(vnm, sw_if_index, (uword) arg); +} + +/** + * @brief Registered callback for HW interface state changes + */ +static clib_error_t * +adj_glean_hw_interface_state_change (vnet_main_t * vnm, + u32 hw_if_index, + u32 flags) +{ + /* + * walk SW interfaces on the HW + */ + uword sw_flags; + + sw_flags = ((flags & VNET_HW_INTERFACE_FLAG_LINK_UP) ? + VNET_SW_INTERFACE_FLAG_ADMIN_UP : + 0); + + vnet_hw_interface_walk_sw(vnm, hw_if_index, + adj_nbr_hw_sw_interface_state_change, + (void*) sw_flags); + + return (NULL); +} + +VNET_HW_INTERFACE_LINK_UP_DOWN_FUNCTION( + adj_glean_hw_interface_state_change); + +static clib_error_t * +adj_glean_interface_delete (vnet_main_t * vnm, + u32 sw_if_index, + u32 is_add) +{ + /* + * for each glean on the interface trigger a walk back to the children + */ + fib_protocol_t proto; + ip_adjacency_t *adj; + + if (is_add) + { + /* + * not interested in interface additions. we will not back walk + * to resolve paths through newly added interfaces. Why? The control + * plane should have the brains to add interfaces first, then routes. + * So the case where there are paths with a interface that matches + * one just created is the case where the path resolved through an + * interface that was deleted, and still has not been removed. The + * new interface added, is NO GUARANTEE that the interface being + * added now, even though it may have the same sw_if_index, is the + * same interface that the path needs. So tough! + * If the control plane wants these routes to resolve it needs to + * remove and add them again. + */ + return (NULL); + } + + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + if (sw_if_index >= vec_len(adj_gleans[proto]) || + ADJ_INDEX_INVALID == adj_gleans[proto][sw_if_index]) + continue; + + adj = adj_get(adj_gleans[proto][sw_if_index]); + + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = FIB_NODE_BW_REASON_FLAG_INTERFACE_DELETE, + }; + + fib_walk_sync(FIB_NODE_TYPE_ADJ, adj_get_index(adj), &bw_ctx); + } + + return (NULL); +} + +VNET_SW_INTERFACE_ADD_DEL_FUNCTION(adj_glean_interface_delete); + +u8* +format_adj_glean (u8* s, va_list *ap) +{ + index_t index = va_arg(*ap, index_t); + CLIB_UNUSED(u32 indent) = va_arg(*ap, u32); + vnet_main_t * vnm = vnet_get_main(); + ip_adjacency_t * adj = adj_get(index); + + return (format(s, "%U-glean: %U", + format_fib_protocol, adj->ia_nh_proto, + format_vnet_sw_interface_name, + vnm, + vnet_get_sw_interface(vnm, + adj->rewrite_header.sw_if_index))); +} + + +static void +adj_dpo_lock (dpo_id_t *dpo) +{ + adj_lock(dpo->dpoi_index); +} +static void +adj_dpo_unlock (dpo_id_t *dpo) +{ + adj_unlock(dpo->dpoi_index); +} + +const static dpo_vft_t adj_glean_dpo_vft = { + .dv_lock = adj_dpo_lock, + .dv_unlock = adj_dpo_unlock, + .dv_format = format_adj_glean, +}; + +/** + * @brief The per-protocol VLIB graph nodes that are assigned to a glean + * object. + * + * this means that these graph nodes are ones from which a glean is the + * parent object in the DPO-graph. + */ +const static char* const glean_ip4_nodes[] = +{ + "ip4-glean", + NULL, +}; +const static char* const glean_ip6_nodes[] = +{ + "ip6-glean", + NULL, +}; + +const static char* const * const glean_nodes[DPO_PROTO_NUM] = +{ + [DPO_PROTO_IP4] = glean_ip4_nodes, + [DPO_PROTO_IP6] = glean_ip6_nodes, + [DPO_PROTO_MPLS] = NULL, +}; + +void +adj_glean_module_init (void) +{ + dpo_register(DPO_ADJACENCY_GLEAN, &adj_glean_dpo_vft, glean_nodes); +} diff --git a/src/vnet/adj/adj_glean.h b/src/vnet/adj/adj_glean.h new file mode 100644 index 00000000..640bd2f9 --- /dev/null +++ b/src/vnet/adj/adj_glean.h @@ -0,0 +1,61 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/** + * @brief Glean Adjacency + * + * A gleean adjacency represent the need to discover new peers on an + * attached link. Packets that hit a glean adjacency will generate an + * ARP/ND packet addessesed to the packet's destination address. + * Note this is different to an incomplete neighbour adjacency, which + * does not send ARP/ND requests to the packet's destination address, + * but instead to the next-hop address of the adjacency itself. + */ + +#ifndef __ADJ_GLEAN_H__ +#define __ADJ_GLEAN_H__ + +#include <vnet/adj/adj_types.h> + +/** + * @brief + * Add (and lock) a new or lock an existing glean adjacency + * + * @param proto + * The protocol for the neighbours that we wish to glean + * + * @param sw_if_index + * The interface on which to glean + * + * @param nh_addr + * the address applied to the interface on which to glean. This + * as the source address in packets when the ARP/ND packet is sent + */ +extern adj_index_t adj_glean_add_or_lock(fib_protocol_t proto, + u32 sw_if_index, + const ip46_address_t *nh_addr); + +/** + * @brief Format/display a glean adjacency. + */ +extern u8* format_adj_glean(u8* s, va_list *ap); + +/** + * @brief + * Module initialisation + */ +extern void adj_glean_module_init(void); + +#endif diff --git a/src/vnet/adj/adj_internal.h b/src/vnet/adj/adj_internal.h new file mode 100644 index 00000000..2c123c54 --- /dev/null +++ b/src/vnet/adj/adj_internal.h @@ -0,0 +1,112 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __ADJ_INTERNAL_H__ +#define __ADJ_INTERNAL_H__ + +#include <vnet/adj/adj.h> +#include <vnet/adj/adj_mcast.h> +#include <vnet/ip/ip.h> +#include <vnet/mpls/mpls.h> +#include <vnet/adj/adj_l2.h> +#include <vnet/adj/adj_nsh.h> + +/** + * big switch to turn on Adjacency debugging + */ +#undef ADJ_DEBUG + +/* + * Debug macro + */ +#ifdef ADJ_DEBUG +#define ADJ_DBG(_adj, _fmt, _args...) \ +{ \ + clib_warning("adj:[%d:%p]:" _fmt, \ + _adj - adj_pool, _adj, \ + ##_args); \ +} +#else +#define ADJ_DBG(_e, _fmt, _args...) +#endif + +static inline u32 +adj_get_rewrite_node (vnet_link_t linkt) +{ + switch (linkt) { + case VNET_LINK_IP4: + return (ip4_rewrite_node.index); + case VNET_LINK_IP6: + return (ip6_rewrite_node.index); + case VNET_LINK_MPLS: + return (mpls_output_node.index); + case VNET_LINK_ETHERNET: + return (adj_l2_rewrite_node.index); + case VNET_LINK_NSH: + return (adj_nsh_rewrite_node.index); + case VNET_LINK_ARP: + break; + } + ASSERT(0); + return (0); +} + +static inline vnet_link_t +adj_fib_proto_2_nd (fib_protocol_t fp) +{ + switch (fp) + { + case FIB_PROTOCOL_IP4: + return (VNET_LINK_ARP); + case FIB_PROTOCOL_IP6: + return (VNET_LINK_IP6); + case FIB_PROTOCOL_MPLS: + return (VNET_LINK_MPLS); + } + return (0); +} + +/** + * @brief + * Get a pointer to an adjacency object from its index + */ +static inline adj_index_t +adj_get_index (ip_adjacency_t *adj) +{ + return (adj - adj_pool); +} + +extern void adj_nbr_update_rewrite_internal(ip_adjacency_t *adj, + ip_lookup_next_t adj_next_index, + u32 complete_next_index, + u32 next_index, + u8 *rewrite); +extern void adj_midchain_setup(adj_index_t adj_index, + adj_midchain_fixup_t fixup, + adj_flags_t flags); + +extern ip_adjacency_t * adj_alloc(fib_protocol_t proto); + +extern void adj_nbr_remove(adj_index_t ai, + fib_protocol_t nh_proto, + vnet_link_t link_type, + const ip46_address_t *nh_addr, + u32 sw_if_index); +extern void adj_glean_remove(fib_protocol_t proto, + u32 sw_if_index); +extern void adj_mcast_remove(fib_protocol_t proto, + u32 sw_if_index); + +#endif diff --git a/src/vnet/adj/adj_l2.c b/src/vnet/adj/adj_l2.c new file mode 100644 index 00000000..20d70dd4 --- /dev/null +++ b/src/vnet/adj/adj_l2.c @@ -0,0 +1,192 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/vnet.h> +#include <vnet/adj/adj_l2.h> +#include <vnet/ethernet/ethernet.h> +#include <vnet/ip/ip.h> + +/** + * @brief Trace data for a L2 Midchain + */ +typedef struct adj_l2_trace_t_ { + /** Adjacency index taken. */ + u32 adj_index; +} adj_l2_trace_t; + +static u8 * +format_adj_l2_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + adj_l2_trace_t * t = va_arg (*args, adj_l2_trace_t *); + + s = format (s, "adj-idx %d : %U", + t->adj_index, + format_ip_adjacency, t->adj_index, FORMAT_IP_ADJACENCY_NONE); + return s; +} + +typedef enum adj_l2_rewrite_next_t_ +{ + ADJ_L2_REWRITE_NEXT_DROP, +} adj_l2_rewrite_next_t; + +always_inline uword +adj_l2_rewrite_inline (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame, + int is_midchain) +{ + u32 * from = vlib_frame_vector_args (frame); + u32 n_left_from, n_left_to_next, * to_next, next_index; + u32 thread_index = vlib_get_thread_index(); + ethernet_main_t * em = ðernet_main; + + n_left_from = frame->n_vectors; + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + ip_adjacency_t * adj0; + vlib_buffer_t * p0; + char *h0; + u32 pi0, rw_len0, adj_index0, next0 = 0; + u32 tx_sw_if_index0; + + pi0 = to_next[0] = from[0]; + from += 1; + n_left_from -= 1; + to_next += 1; + n_left_to_next -= 1; + + p0 = vlib_get_buffer (vm, pi0); + h0 = vlib_buffer_get_current (p0); + + adj_index0 = vnet_buffer (p0)->ip.adj_index[VLIB_TX]; + + adj0 = adj_get (adj_index0); + + /* Guess we are only writing on simple Ethernet header. */ + vnet_rewrite_one_header (adj0[0], h0, + sizeof (ethernet_header_t)); + + /* Update packet buffer attributes/set output interface. */ + rw_len0 = adj0[0].rewrite_header.data_bytes; + vnet_buffer(p0)->ip.save_rewrite_length = rw_len0; + vnet_buffer(p0)->sw_if_index[VLIB_TX] = adj0->rewrite_header.sw_if_index; + + vlib_increment_combined_counter(&adjacency_counters, + thread_index, + adj_index0, + /* packet increment */ 0, + /* byte increment */ rw_len0); + + /* Check MTU of outgoing interface. */ + if (PREDICT_TRUE((vlib_buffer_length_in_chain (vm, p0) <= + adj0[0].rewrite_header.max_l3_packet_bytes))) + { + /* Don't adjust the buffer for ttl issue; icmp-error node wants + * to see the IP headerr */ + p0->current_data -= rw_len0; + p0->current_length += rw_len0; + tx_sw_if_index0 = adj0[0].rewrite_header.sw_if_index; + + if (is_midchain) + { + adj0->sub_type.midchain.fixup_func(vm, adj0, p0); + } + + vnet_buffer (p0)->sw_if_index[VLIB_TX] = tx_sw_if_index0; + + /* + * Follow the feature ARC. this will result eventually in + * the midchain-tx node + */ + vnet_feature_arc_start(em->output_feature_arc_index, tx_sw_if_index0, &next0, p0); + } + else + { + /* can't fragment L2 */ + next0 = ADJ_L2_REWRITE_NEXT_DROP; + } + + if (PREDICT_FALSE(p0->flags & VLIB_BUFFER_IS_TRACED)) + { + adj_l2_trace_t *tr = vlib_add_trace (vm, node, + p0, sizeof (*tr)); + tr->adj_index = vnet_buffer(p0)->ip.adj_index[VLIB_TX]; + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + pi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + return frame->n_vectors; +} + +static uword +adj_l2_rewrite (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + return adj_l2_rewrite_inline (vm, node, frame, 0); +} + +static uword +adj_l2_midchain (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + return adj_l2_rewrite_inline (vm, node, frame, 1); +} + +VLIB_REGISTER_NODE (adj_l2_rewrite_node) = { + .function = adj_l2_rewrite, + .name = "adj-l2-rewrite", + .vector_size = sizeof (u32), + + .format_trace = format_adj_l2_trace, + + .n_next_nodes = 1, + .next_nodes = { + [ADJ_L2_REWRITE_NEXT_DROP] = "error-drop", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (adj_l2_rewrite_node, adj_l2_rewrite) + +VLIB_REGISTER_NODE (adj_l2_midchain_node) = { + .function = adj_l2_midchain, + .name = "adj-l2-midchain", + .vector_size = sizeof (u32), + + .format_trace = format_adj_l2_trace, + + .n_next_nodes = 1, + .next_nodes = { + [ADJ_L2_REWRITE_NEXT_DROP] = "error-drop", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (adj_l2_midchain_node, adj_l2_midchain) diff --git a/src/vnet/adj/adj_l2.h b/src/vnet/adj/adj_l2.h new file mode 100644 index 00000000..3aa1c74b --- /dev/null +++ b/src/vnet/adj/adj_l2.h @@ -0,0 +1,24 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __ADJ_L2_H__ +#define __ADJ_L2_H__ + +#include <vnet/adj/adj.h> + +extern vlib_node_registration_t adj_l2_midchain_node; +extern vlib_node_registration_t adj_l2_rewrite_node; + +#endif diff --git a/src/vnet/adj/adj_mcast.c b/src/vnet/adj/adj_mcast.c new file mode 100644 index 00000000..da06cd00 --- /dev/null +++ b/src/vnet/adj/adj_mcast.c @@ -0,0 +1,483 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/adj/adj_mcast.h> +#include <vnet/adj/adj_internal.h> +#include <vnet/fib/fib_walk.h> +#include <vnet/ip/ip.h> + +/* + * The 'DB' of all mcast adjs. + * There is only one mcast per-interface per-protocol, so this is a per-interface + * vector + */ +static adj_index_t *adj_mcasts[FIB_PROTOCOL_MAX]; + +static u32 +adj_get_mcast_node (fib_protocol_t proto) +{ + switch (proto) { + case FIB_PROTOCOL_IP4: + return (ip4_rewrite_mcast_node.index); + case FIB_PROTOCOL_IP6: + return (ip6_rewrite_mcast_node.index); + case FIB_PROTOCOL_MPLS: + break; + } + ASSERT(0); + return (0); +} + +/* + * adj_mcast_add_or_lock + * + * The next_hop address here is used for source address selection in the DP. + * The mcast adj is added to an interface's connected prefix, the next-hop + * passed here is the local prefix on the same interface. + */ +adj_index_t +adj_mcast_add_or_lock (fib_protocol_t proto, + vnet_link_t link_type, + u32 sw_if_index) +{ + ip_adjacency_t * adj; + + vec_validate_init_empty(adj_mcasts[proto], sw_if_index, ADJ_INDEX_INVALID); + + if (ADJ_INDEX_INVALID == adj_mcasts[proto][sw_if_index]) + { + vnet_main_t *vnm; + + vnm = vnet_get_main(); + adj = adj_alloc(proto); + + adj->lookup_next_index = IP_LOOKUP_NEXT_MCAST; + adj->ia_nh_proto = proto; + adj->ia_link = link_type; + adj_mcasts[proto][sw_if_index] = adj_get_index(adj); + adj_lock(adj_get_index(adj)); + + vnet_rewrite_init(vnm, sw_if_index, + adj_get_mcast_node(proto), + vnet_tx_node_index_for_sw_interface(vnm, sw_if_index), + &adj->rewrite_header); + + /* + * we need a rewrite where the destination IP address is converted + * to the appropriate link-layer address. This is interface specific. + * So ask the interface to do it. + */ + vnet_update_adjacency_for_sw_interface(vnm, sw_if_index, + adj_get_index(adj)); + } + else + { + adj = adj_get(adj_mcasts[proto][sw_if_index]); + adj_lock(adj_get_index(adj)); + } + + return (adj_get_index(adj)); +} + +/** + * adj_mcast_update_rewrite + * + * Update the adjacency's rewrite string. A NULL string implies the + * rewirte is reset (i.e. when ARP/ND etnry is gone). + * NB: the adj being updated may be handling traffic in the DP. + */ +void +adj_mcast_update_rewrite (adj_index_t adj_index, + u8 *rewrite, + u8 offset, + u32 mask) +{ + ip_adjacency_t *adj; + + ASSERT(ADJ_INDEX_INVALID != adj_index); + + adj = adj_get(adj_index); + + /* + * update the adj's rewrite string and build the arc + * from the rewrite node to the interface's TX node + */ + adj_nbr_update_rewrite_internal(adj, IP_LOOKUP_NEXT_MCAST, + adj_get_mcast_node(adj->ia_nh_proto), + vnet_tx_node_index_for_sw_interface( + vnet_get_main(), + adj->rewrite_header.sw_if_index), + rewrite); + /* + * set the fields corresponding to the mcast IP address rewrite + * The mask must be stored in network byte order, since the packet's + * IP address will also be in network order. + */ + adj->rewrite_header.dst_mcast_offset = offset; + adj->rewrite_header.dst_mcast_mask = clib_host_to_net_u32(mask); +} + +/** + * adj_mcast_midchain_update_rewrite + * + * Update the adjacency's rewrite string. A NULL string implies the + * rewirte is reset (i.e. when ARP/ND etnry is gone). + * NB: the adj being updated may be handling traffic in the DP. + */ +void +adj_mcast_midchain_update_rewrite (adj_index_t adj_index, + adj_midchain_fixup_t fixup, + adj_flags_t flags, + u8 *rewrite, + u8 offset, + u32 mask) +{ + ip_adjacency_t *adj; + + ASSERT(ADJ_INDEX_INVALID != adj_index); + + adj = adj_get(adj_index); + + /* + * one time only update. since we don't support chainging the tunnel + * src,dst, this is all we need. + */ + ASSERT(adj->lookup_next_index == IP_LOOKUP_NEXT_MCAST); + /* + * tunnels can always provide a rewrite. + */ + ASSERT(NULL != rewrite); + + adj_midchain_setup(adj_index, fixup, flags); + + /* + * update the adj's rewrite string and build the arc + * from the rewrite node to the interface's TX node + */ + adj_nbr_update_rewrite_internal(adj, IP_LOOKUP_NEXT_MCAST_MIDCHAIN, + adj_get_mcast_node(adj->ia_nh_proto), + vnet_tx_node_index_for_sw_interface( + vnet_get_main(), + adj->rewrite_header.sw_if_index), + rewrite); + + /* + * set the fields corresponding to the mcast IP address rewrite + * The mask must be stored in network byte order, since the packet's + * IP address will also be in network order. + */ + adj->rewrite_header.dst_mcast_offset = offset; + adj->rewrite_header.dst_mcast_mask = clib_host_to_net_u32(mask); +} + +void +adj_mcast_remove (fib_protocol_t proto, + u32 sw_if_index) +{ + ASSERT(sw_if_index < vec_len(adj_mcasts[proto])); + + adj_mcasts[proto][sw_if_index] = ADJ_INDEX_INVALID; +} + +static clib_error_t * +adj_mcast_interface_state_change (vnet_main_t * vnm, + u32 sw_if_index, + u32 flags) +{ + /* + * for each mcast on the interface trigger a walk back to the children + */ + fib_protocol_t proto; + ip_adjacency_t *adj; + + + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + if (sw_if_index >= vec_len(adj_mcasts[proto]) || + ADJ_INDEX_INVALID == adj_mcasts[proto][sw_if_index]) + continue; + + adj = adj_get(adj_mcasts[proto][sw_if_index]); + + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP ? + FIB_NODE_BW_REASON_FLAG_INTERFACE_UP : + FIB_NODE_BW_REASON_FLAG_INTERFACE_DOWN), + }; + + fib_walk_sync(FIB_NODE_TYPE_ADJ, adj_get_index(adj), &bw_ctx); + } + + return (NULL); +} + +VNET_SW_INTERFACE_ADMIN_UP_DOWN_FUNCTION(adj_mcast_interface_state_change); + +/** + * @brief Invoked on each SW interface of a HW interface when the + * HW interface state changes + */ +static void +adj_mcast_hw_sw_interface_state_change (vnet_main_t * vnm, + u32 sw_if_index, + void *arg) +{ + adj_mcast_interface_state_change(vnm, sw_if_index, (uword) arg); +} + +/** + * @brief Registered callback for HW interface state changes + */ +static clib_error_t * +adj_mcast_hw_interface_state_change (vnet_main_t * vnm, + u32 hw_if_index, + u32 flags) +{ + /* + * walk SW interfaces on the HW + */ + uword sw_flags; + + sw_flags = ((flags & VNET_HW_INTERFACE_FLAG_LINK_UP) ? + VNET_SW_INTERFACE_FLAG_ADMIN_UP : + 0); + + vnet_hw_interface_walk_sw(vnm, hw_if_index, + adj_mcast_hw_sw_interface_state_change, + (void*) sw_flags); + + return (NULL); +} + +VNET_HW_INTERFACE_LINK_UP_DOWN_FUNCTION( + adj_mcast_hw_interface_state_change); + +static clib_error_t * +adj_mcast_interface_delete (vnet_main_t * vnm, + u32 sw_if_index, + u32 is_add) +{ + /* + * for each mcast on the interface trigger a walk back to the children + */ + fib_protocol_t proto; + ip_adjacency_t *adj; + + if (is_add) + { + /* + * not interested in interface additions. we will not back walk + * to resolve paths through newly added interfaces. Why? The control + * plane should have the brains to add interfaces first, then routes. + * So the case where there are paths with a interface that matches + * one just created is the case where the path resolved through an + * interface that was deleted, and still has not been removed. The + * new interface added, is NO GUARANTEE that the interface being + * added now, even though it may have the same sw_if_index, is the + * same interface that the path needs. So tough! + * If the control plane wants these routes to resolve it needs to + * remove and add them again. + */ + return (NULL); + } + + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + if (sw_if_index >= vec_len(adj_mcasts[proto]) || + ADJ_INDEX_INVALID == adj_mcasts[proto][sw_if_index]) + continue; + + adj = adj_get(adj_mcasts[proto][sw_if_index]); + + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = FIB_NODE_BW_REASON_FLAG_INTERFACE_DELETE, + }; + + fib_walk_sync(FIB_NODE_TYPE_ADJ, adj_get_index(adj), &bw_ctx); + } + + return (NULL); +} + +VNET_SW_INTERFACE_ADD_DEL_FUNCTION(adj_mcast_interface_delete); + +/** + * @brief Walk the multicast Adjacencies on a given interface + */ +void +adj_mcast_walk (u32 sw_if_index, + fib_protocol_t proto, + adj_walk_cb_t cb, + void *ctx) +{ + if (vec_len(adj_mcasts[proto]) > sw_if_index) + { + if (ADJ_INDEX_INVALID != adj_mcasts[proto][sw_if_index]) + { + cb(adj_mcasts[proto][sw_if_index], ctx); + } + } +} + +u8* +format_adj_mcast (u8* s, va_list *ap) +{ + index_t index = va_arg(*ap, index_t); + CLIB_UNUSED(u32 indent) = va_arg(*ap, u32); + ip_adjacency_t * adj = adj_get(index); + + s = format(s, "%U-mcast: ", + format_fib_protocol, adj->ia_nh_proto); + if (adj->rewrite_header.flags & VNET_REWRITE_HAS_FEATURES) + s = format(s, "[features] "); + s = format (s, "%U", + format_vnet_rewrite, + &adj->rewrite_header, sizeof (adj->rewrite_data), 0); + + return (s); +} + +u8* +format_adj_mcast_midchain (u8* s, va_list *ap) +{ + index_t index = va_arg(*ap, index_t); + CLIB_UNUSED(u32 indent) = va_arg(*ap, u32); + vnet_main_t * vnm = vnet_get_main(); + ip_adjacency_t * adj = adj_get(index); + + s = format(s, "%U-mcast-midchain: ", + format_fib_protocol, adj->ia_nh_proto); + s = format (s, "%U", + format_vnet_rewrite, + vnm->vlib_main, &adj->rewrite_header, + sizeof (adj->rewrite_data), 0); + s = format (s, "\n%Ustacked-on:\n%U%U", + format_white_space, indent, + format_white_space, indent+2, + format_dpo_id, &adj->sub_type.midchain.next_dpo, indent+2); + + return (s); +} + + +static void +adj_dpo_lock (dpo_id_t *dpo) +{ + adj_lock(dpo->dpoi_index); +} +static void +adj_dpo_unlock (dpo_id_t *dpo) +{ + adj_unlock(dpo->dpoi_index); +} + +const static dpo_vft_t adj_mcast_dpo_vft = { + .dv_lock = adj_dpo_lock, + .dv_unlock = adj_dpo_unlock, + .dv_format = format_adj_mcast, +}; +const static dpo_vft_t adj_mcast_midchain_dpo_vft = { + .dv_lock = adj_dpo_lock, + .dv_unlock = adj_dpo_unlock, + .dv_format = format_adj_mcast_midchain, +}; + +/** + * @brief The per-protocol VLIB graph nodes that are assigned to a mcast + * object. + * + * this means that these graph nodes are ones from which a mcast is the + * parent object in the DPO-graph. + */ +const static char* const adj_mcast_ip4_nodes[] = +{ + "ip4-rewrite-mcast", + NULL, +}; +const static char* const adj_mcast_ip6_nodes[] = +{ + "ip6-rewrite-mcast", + NULL, +}; + +const static char* const * const adj_mcast_nodes[DPO_PROTO_NUM] = +{ + [DPO_PROTO_IP4] = adj_mcast_ip4_nodes, + [DPO_PROTO_IP6] = adj_mcast_ip6_nodes, + [DPO_PROTO_MPLS] = NULL, +}; + +/** + * @brief The per-protocol VLIB graph nodes that are assigned to a mcast + * object. + * + * this means that these graph nodes are ones from which a mcast is the + * parent object in the DPO-graph. + */ +const static char* const adj_mcast_midchain_ip4_nodes[] = +{ + "ip4-mcast-midchain", + NULL, +}; +const static char* const adj_mcast_midchain_ip6_nodes[] = +{ + "ip6-mcast-midchain", + NULL, +}; + +const static char* const * const adj_mcast_midchain_nodes[DPO_PROTO_NUM] = +{ + [DPO_PROTO_IP4] = adj_mcast_midchain_ip4_nodes, + [DPO_PROTO_IP6] = adj_mcast_midchain_ip6_nodes, + [DPO_PROTO_MPLS] = NULL, +}; + +/** + * @brief Return the size of the adj DB. + * This is only for testing purposes so an efficient implementation is not needed + */ +u32 +adj_mcast_db_size (void) +{ + u32 n_adjs, sw_if_index; + fib_protocol_t proto; + + n_adjs = 0; + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + for (sw_if_index = 0; + sw_if_index < vec_len(adj_mcasts[proto]); + sw_if_index++) + { + if (ADJ_INDEX_INVALID != adj_mcasts[proto][sw_if_index]) + { + n_adjs++; + } + } + } + + return (n_adjs); +} + +void +adj_mcast_module_init (void) +{ + dpo_register(DPO_ADJACENCY_MCAST, + &adj_mcast_dpo_vft, + adj_mcast_nodes); + dpo_register(DPO_ADJACENCY_MCAST_MIDCHAIN, + &adj_mcast_midchain_dpo_vft, + adj_mcast_midchain_nodes); +} diff --git a/src/vnet/adj/adj_mcast.h b/src/vnet/adj/adj_mcast.h new file mode 100644 index 00000000..bfb0d6f6 --- /dev/null +++ b/src/vnet/adj/adj_mcast.h @@ -0,0 +1,114 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/** + * @brief Mcast Adjacency + * + * The multicast adjacency forwards IP traffic on an interface toward a multicast + * group address. This is a different type of adjacency to a unicast adjacency + * since the application of the MAC header is different, and so the VLIB node + * visited is also different. DPO types have different VLIB nodes. + */ + +#ifndef __ADJ_MCAST_H__ +#define __ADJ_MCAST_H__ + +#include <vnet/adj/adj_types.h> +#include <vnet/adj/adj_midchain.h> + +/** + * @brief + * Add (and lock) a new or lock an existing mcast adjacency + * + * @param proto + * The protocol for the neighbours that we wish to mcast + * + * @param link_type + * A description of the protocol of the packets that will forward + * through this adj. On an ethernet interface this is the MAC header's + * ether-type + * + * @param sw_if_index + * The interface on which to mcast + */ +extern adj_index_t adj_mcast_add_or_lock(fib_protocol_t proto, + vnet_link_t link_type, + u32 sw_if_index); + +/** + * @brief + * Update the rewrite string for an existing adjacecny. + * + * @param + * The index of the adj to update + * + * @param + * The new rewrite + * + * @param + * The offset in the rewrite a which to write in packet's + * IP Address + * + * @param + * The mask to apply to the packet berfore the rewrite. + */ +extern void adj_mcast_update_rewrite(adj_index_t adj_index, + u8 *rewrite, + u8 offset, + u32 mask); + +/** + * @brief + * Update the rewrite string for an existing adjacecny and + * Convert the adjacency into a midchain + * + * @param + * The index of the adj to update + * + * @param + * The new rewrite + */ +extern void adj_mcast_midchain_update_rewrite(adj_index_t adj_index, + adj_midchain_fixup_t fixup, + adj_flags_t flags, + u8 *rewrite, + u8 offset, + u32 mask); +/** + * @brief Walk the multicast Adjacencies on a given interface + */ +extern void adj_mcast_walk (u32 sw_if_index, + fib_protocol_t adj_nh_proto, + adj_walk_cb_t cb, + void *ctx); + +/** + * @brief Format/display a mcast adjacency. + */ +extern u8* format_adj_mcast(u8* s, va_list *ap); +extern u8* format_adj_mcast_midchain(u8* s, va_list *ap); + +/** + * @brief Get the sze of the mcast adj DB. Test purposes only. + */ +extern u32 adj_mcast_db_size(void); + +/** + * @brief + * Module initialisation + */ +extern void adj_mcast_module_init(void); + +#endif diff --git a/src/vnet/adj/adj_midchain.c b/src/vnet/adj/adj_midchain.c new file mode 100644 index 00000000..e9a510b0 --- /dev/null +++ b/src/vnet/adj/adj_midchain.c @@ -0,0 +1,666 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/adj/adj_nbr.h> +#include <vnet/adj/adj_internal.h> +#include <vnet/adj/adj_l2.h> +#include <vnet/adj/adj_nsh.h> +#include <vnet/adj/adj_midchain.h> +#include <vnet/ethernet/arp_packet.h> +#include <vnet/dpo/drop_dpo.h> +#include <vnet/fib/fib_walk.h> + +/** + * The two midchain tx feature node indices + */ +static u32 adj_midchain_tx_feature_node[VNET_LINK_NUM]; +static u32 adj_midchain_tx_no_count_feature_node[VNET_LINK_NUM]; + +/** + * @brief Trace data for packets traversing the midchain tx node + */ +typedef struct adj_midchain_tx_trace_t_ +{ + /** + * @brief the midchain adj we are traversing + */ + adj_index_t ai; +} adj_midchain_tx_trace_t; + +always_inline uword +adj_midchain_tx_inline (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame, + int interface_count) +{ + u32 * from, * to_next, n_left_from, n_left_to_next; + u32 next_index; + vnet_main_t *vnm = vnet_get_main (); + vnet_interface_main_t *im = &vnm->interface_main; + u32 thread_index = vm->thread_index; + + /* Vector of buffer / pkt indices we're supposed to process */ + from = vlib_frame_vector_args (frame); + + /* Number of buffers / pkts */ + n_left_from = frame->n_vectors; + + /* Speculatively send the first buffer to the last disposition we used */ + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + /* set up to enqueue to our disposition with index = next_index */ + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from >= 8 && n_left_to_next > 4) + { + u32 bi0, adj_index0, next0; + const ip_adjacency_t * adj0; + const dpo_id_t *dpo0; + vlib_buffer_t * b0; + u32 bi1, adj_index1, next1; + const ip_adjacency_t * adj1; + const dpo_id_t *dpo1; + vlib_buffer_t * b1; + u32 bi2, adj_index2, next2; + const ip_adjacency_t * adj2; + const dpo_id_t *dpo2; + vlib_buffer_t * b2; + u32 bi3, adj_index3, next3; + const ip_adjacency_t * adj3; + const dpo_id_t *dpo3; + vlib_buffer_t * b3; + + /* Prefetch next iteration. */ + { + vlib_buffer_t * p4, * p5; + vlib_buffer_t * p6, * p7; + + p4 = vlib_get_buffer (vm, from[4]); + p5 = vlib_get_buffer (vm, from[5]); + p6 = vlib_get_buffer (vm, from[6]); + p7 = vlib_get_buffer (vm, from[7]); + + vlib_prefetch_buffer_header (p4, LOAD); + vlib_prefetch_buffer_header (p5, LOAD); + vlib_prefetch_buffer_header (p6, LOAD); + vlib_prefetch_buffer_header (p7, LOAD); + } + + bi0 = from[0]; + to_next[0] = bi0; + bi1 = from[1]; + to_next[1] = bi1; + bi2 = from[2]; + to_next[2] = bi2; + bi3 = from[3]; + to_next[3] = bi3; + + from += 4; + to_next += 4; + n_left_from -= 4; + n_left_to_next -= 4; + + b0 = vlib_get_buffer(vm, bi0); + b1 = vlib_get_buffer(vm, bi1); + b2 = vlib_get_buffer(vm, bi2); + b3 = vlib_get_buffer(vm, bi3); + + /* Follow the DPO on which the midchain is stacked */ + adj_index0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX]; + adj_index1 = vnet_buffer(b1)->ip.adj_index[VLIB_TX]; + adj_index2 = vnet_buffer(b2)->ip.adj_index[VLIB_TX]; + adj_index3 = vnet_buffer(b3)->ip.adj_index[VLIB_TX]; + + adj0 = adj_get(adj_index0); + adj1 = adj_get(adj_index1); + adj2 = adj_get(adj_index2); + adj3 = adj_get(adj_index3); + + dpo0 = &adj0->sub_type.midchain.next_dpo; + dpo1 = &adj1->sub_type.midchain.next_dpo; + dpo2 = &adj2->sub_type.midchain.next_dpo; + dpo3 = &adj3->sub_type.midchain.next_dpo; + + next0 = dpo0->dpoi_next_node; + next1 = dpo1->dpoi_next_node; + next2 = dpo2->dpoi_next_node; + next3 = dpo3->dpoi_next_node; + + vnet_buffer(b1)->ip.adj_index[VLIB_TX] = dpo1->dpoi_index; + vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index; + vnet_buffer(b2)->ip.adj_index[VLIB_TX] = dpo2->dpoi_index; + vnet_buffer(b3)->ip.adj_index[VLIB_TX] = dpo3->dpoi_index; + + if (interface_count) + { + vlib_increment_combined_counter (im->combined_sw_if_counters + + VNET_INTERFACE_COUNTER_TX, + thread_index, + adj0->rewrite_header.sw_if_index, + 1, + vlib_buffer_length_in_chain (vm, b0)); + vlib_increment_combined_counter (im->combined_sw_if_counters + + VNET_INTERFACE_COUNTER_TX, + thread_index, + adj1->rewrite_header.sw_if_index, + 1, + vlib_buffer_length_in_chain (vm, b1)); + vlib_increment_combined_counter (im->combined_sw_if_counters + + VNET_INTERFACE_COUNTER_TX, + thread_index, + adj2->rewrite_header.sw_if_index, + 1, + vlib_buffer_length_in_chain (vm, b2)); + vlib_increment_combined_counter (im->combined_sw_if_counters + + VNET_INTERFACE_COUNTER_TX, + thread_index, + adj3->rewrite_header.sw_if_index, + 1, + vlib_buffer_length_in_chain (vm, b3)); + } + + if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED)) + { + adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node, + b0, sizeof (*tr)); + tr->ai = adj_index0; + } + if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED)) + { + adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node, + b1, sizeof (*tr)); + tr->ai = adj_index1; + } + if (PREDICT_FALSE(b2->flags & VLIB_BUFFER_IS_TRACED)) + { + adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node, + b2, sizeof (*tr)); + tr->ai = adj_index2; + } + if (PREDICT_FALSE(b3->flags & VLIB_BUFFER_IS_TRACED)) + { + adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node, + b3, sizeof (*tr)); + tr->ai = adj_index3; + } + + vlib_validate_buffer_enqueue_x4 (vm, node, next_index, + to_next, n_left_to_next, + bi0, bi1, bi2, bi3, + next0, next1, next2, next3); + } + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0, adj_index0, next0; + const ip_adjacency_t * adj0; + const dpo_id_t *dpo0; + vlib_buffer_t * b0; + + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer(vm, bi0); + + /* Follow the DPO on which the midchain is stacked */ + adj_index0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX]; + adj0 = adj_get(adj_index0); + dpo0 = &adj0->sub_type.midchain.next_dpo; + next0 = dpo0->dpoi_next_node; + vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index; + + if (interface_count) + { + vlib_increment_combined_counter (im->combined_sw_if_counters + + VNET_INTERFACE_COUNTER_TX, + thread_index, + adj0->rewrite_header.sw_if_index, + 1, + vlib_buffer_length_in_chain (vm, b0)); + } + + if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED)) + { + adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node, + b0, sizeof (*tr)); + tr->ai = adj_index0; + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + return frame->n_vectors; +} + +static u8 * +format_adj_midchain_tx_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + adj_midchain_tx_trace_t *tr = va_arg (*args, adj_midchain_tx_trace_t*); + + s = format(s, "adj-midchain:[%d]:%U", tr->ai, + format_ip_adjacency, tr->ai, + FORMAT_IP_ADJACENCY_NONE); + + return (s); +} + +static uword +adj_midchain_tx (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + return (adj_midchain_tx_inline(vm, node, frame, 1)); +} + +VLIB_REGISTER_NODE (adj_midchain_tx_node, static) = { + .function = adj_midchain_tx, + .name = "adj-midchain-tx", + .vector_size = sizeof (u32), + + .format_trace = format_adj_midchain_tx_trace, + + .n_next_nodes = 1, + .next_nodes = { + [0] = "error-drop", + }, +}; + +static uword +adj_midchain_tx_no_count (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + return (adj_midchain_tx_inline(vm, node, frame, 0)); +} + +VLIB_REGISTER_NODE (adj_midchain_tx_no_count_node, static) = { + .function = adj_midchain_tx_no_count, + .name = "adj-midchain-tx-no-count", + .vector_size = sizeof (u32), + + .format_trace = format_adj_midchain_tx_trace, + + .n_next_nodes = 1, + .next_nodes = { + [0] = "error-drop", + }, +}; + +VNET_FEATURE_INIT (adj_midchain_tx_ip4, static) = { + .arc_name = "ip4-output", + .node_name = "adj-midchain-tx", + .runs_before = VNET_FEATURES ("interface-output"), + .feature_index_ptr = &adj_midchain_tx_feature_node[VNET_LINK_IP4], +}; +VNET_FEATURE_INIT (adj_midchain_tx_no_count_ip4, static) = { + .arc_name = "ip4-output", + .node_name = "adj-midchain-tx-no-count", + .runs_before = VNET_FEATURES ("interface-output"), + .feature_index_ptr = &adj_midchain_tx_no_count_feature_node[VNET_LINK_IP4], +}; +VNET_FEATURE_INIT (adj_midchain_tx_ip6, static) = { + .arc_name = "ip6-output", + .node_name = "adj-midchain-tx", + .runs_before = VNET_FEATURES ("interface-output"), + .feature_index_ptr = &adj_midchain_tx_feature_node[VNET_LINK_IP6], +}; +VNET_FEATURE_INIT (adj_midchain_tx_no_count_ip6, static) = { + .arc_name = "ip6-output", + .node_name = "adj-midchain-tx-no-count", + .runs_before = VNET_FEATURES ("interface-output"), + .feature_index_ptr = &adj_midchain_tx_no_count_feature_node[VNET_LINK_IP6], +}; +VNET_FEATURE_INIT (adj_midchain_tx_mpls, static) = { + .arc_name = "mpls-output", + .node_name = "adj-midchain-tx", + .runs_before = VNET_FEATURES ("interface-output"), + .feature_index_ptr = &adj_midchain_tx_feature_node[VNET_LINK_MPLS], +}; +VNET_FEATURE_INIT (adj_midchain_tx_no_count_mpls, static) = { + .arc_name = "mpls-output", + .node_name = "adj-midchain-tx-no-count", + .runs_before = VNET_FEATURES ("interface-output"), + .feature_index_ptr = &adj_midchain_tx_no_count_feature_node[VNET_LINK_MPLS], +}; +VNET_FEATURE_INIT (adj_midchain_tx_ethernet, static) = { + .arc_name = "ethernet-output", + .node_name = "adj-midchain-tx", + .runs_before = VNET_FEATURES ("error-drop"), + .feature_index_ptr = &adj_midchain_tx_feature_node[VNET_LINK_ETHERNET], +}; +VNET_FEATURE_INIT (adj_midchain_tx_no_count_ethernet, static) = { + .arc_name = "ethernet-output", + .node_name = "adj-midchain-tx-no-count", + .runs_before = VNET_FEATURES ("error-drop"), + .feature_index_ptr = &adj_midchain_tx_no_count_feature_node[VNET_LINK_ETHERNET], +}; +VNET_FEATURE_INIT (adj_midchain_tx_nsh, static) = { + .arc_name = "nsh-output", + .node_name = "adj-midchain-tx", + .runs_before = VNET_FEATURES ("error-drop"), + .feature_index_ptr = &adj_midchain_tx_feature_node[VNET_LINK_NSH], +}; +VNET_FEATURE_INIT (adj_midchain_tx_no_count_nsh, static) = { + .arc_name = "nsh-output", + .node_name = "adj-midchain-tx-no-count", + .runs_before = VNET_FEATURES ("error-drop"), + .feature_index_ptr = &adj_midchain_tx_no_count_feature_node[VNET_LINK_NSH], +}; + +static inline u32 +adj_get_midchain_node (vnet_link_t link) +{ + switch (link) { + case VNET_LINK_IP4: + return (ip4_midchain_node.index); + case VNET_LINK_IP6: + return (ip6_midchain_node.index); + case VNET_LINK_MPLS: + return (mpls_midchain_node.index); + case VNET_LINK_ETHERNET: + return (adj_l2_midchain_node.index); + case VNET_LINK_NSH: + return (adj_nsh_midchain_node.index); + case VNET_LINK_ARP: + break; + } + ASSERT(0); + return (0); +} + +static u8 +adj_midchain_get_feature_arc_index_for_link_type (const ip_adjacency_t *adj) +{ + u8 arc = (u8) ~0; + switch (adj->ia_link) + { + case VNET_LINK_IP4: + { + arc = ip4_main.lookup_main.output_feature_arc_index; + break; + } + case VNET_LINK_IP6: + { + arc = ip6_main.lookup_main.output_feature_arc_index; + break; + } + case VNET_LINK_MPLS: + { + arc = mpls_main.output_feature_arc_index; + break; + } + case VNET_LINK_ETHERNET: + { + arc = ethernet_main.output_feature_arc_index; + break; + } + case VNET_LINK_NSH: + { + arc = nsh_main_dummy.output_feature_arc_index; + break; + } + case VNET_LINK_ARP: + ASSERT(0); + break; + } + + ASSERT (arc != (u8) ~0); + + return (arc); +} + +static u32 +adj_nbr_midchain_get_tx_node (ip_adjacency_t *adj) +{ + return ((adj->ia_flags & ADJ_FLAG_MIDCHAIN_NO_COUNT) ? + adj_midchain_tx_no_count_node.index : + adj_midchain_tx_node.index); +} + +static u32 +adj_nbr_midchain_get_feature_node (ip_adjacency_t *adj) +{ + if (adj->ia_flags & ADJ_FLAG_MIDCHAIN_NO_COUNT) + { + return (adj_midchain_tx_no_count_feature_node[adj->ia_link]); + } + + return (adj_midchain_tx_feature_node[adj->ia_link]); +} + +/** + * adj_midchain_setup + * + * Setup the adj as a mid-chain + */ +void +adj_midchain_setup (adj_index_t adj_index, + adj_midchain_fixup_t fixup, + adj_flags_t flags) +{ + u32 feature_index, tx_node; + ip_adjacency_t *adj; + u8 arc_index; + + ASSERT(ADJ_INDEX_INVALID != adj_index); + + adj = adj_get(adj_index); + + adj->sub_type.midchain.fixup_func = fixup; + adj->ia_flags |= flags; + + arc_index = adj_midchain_get_feature_arc_index_for_link_type (adj); + feature_index = adj_nbr_midchain_get_feature_node(adj); + tx_node = adj_nbr_midchain_get_tx_node(adj); + + vnet_feature_enable_disable_with_index (arc_index, feature_index, + adj->rewrite_header.sw_if_index, + 1 /* enable */, 0, 0); + + /* + * stack the midchain on the drop so it's ready to forward in the adj-midchain-tx. + * The graph arc used/created here is from the midchain-tx node to the + * child's registered node. This is because post adj processing the next + * node are any output features, then the midchain-tx. from there we + * need to get to the stacked child's node. + */ + dpo_stack_from_node(tx_node, + &adj->sub_type.midchain.next_dpo, + drop_dpo_get(vnet_link_to_dpo_proto(adj->ia_link))); +} + +/** + * adj_nbr_midchain_update_rewrite + * + * Update the adjacency's rewrite string. A NULL string implies the + * rewrite is reset (i.e. when ARP/ND etnry is gone). + * NB: the adj being updated may be handling traffic in the DP. + */ +void +adj_nbr_midchain_update_rewrite (adj_index_t adj_index, + adj_midchain_fixup_t fixup, + adj_flags_t flags, + u8 *rewrite) +{ + ip_adjacency_t *adj; + + ASSERT(ADJ_INDEX_INVALID != adj_index); + + adj = adj_get(adj_index); + + /* + * one time only update. since we don't support chainging the tunnel + * src,dst, this is all we need. + */ + ASSERT(adj->lookup_next_index == IP_LOOKUP_NEXT_ARP); + /* + * tunnels can always provide a rewrite. + */ + ASSERT(NULL != rewrite); + + adj_midchain_setup(adj_index, fixup, flags); + + /* + * update the rewirte with the workers paused. + */ + adj_nbr_update_rewrite_internal(adj, + IP_LOOKUP_NEXT_MIDCHAIN, + adj_get_midchain_node(adj->ia_link), + adj_nbr_midchain_get_tx_node(adj), + rewrite); +} + +/** + * adj_nbr_midchain_unstack + * + * Unstack the adj. stack it on drop + */ +void +adj_nbr_midchain_unstack (adj_index_t adj_index) +{ + ip_adjacency_t *adj; + + ASSERT(ADJ_INDEX_INVALID != adj_index); + + adj = adj_get(adj_index); + + /* + * stack on the drop + */ + dpo_stack(DPO_ADJACENCY_MIDCHAIN, + vnet_link_to_dpo_proto(adj->ia_link), + &adj->sub_type.midchain.next_dpo, + drop_dpo_get(vnet_link_to_dpo_proto(adj->ia_link))); + CLIB_MEMORY_BARRIER(); +} + +/** + * adj_nbr_midchain_stack + */ +void +adj_nbr_midchain_stack (adj_index_t adj_index, + const dpo_id_t *next) +{ + ip_adjacency_t *adj; + + ASSERT(ADJ_INDEX_INVALID != adj_index); + + adj = adj_get(adj_index); + + ASSERT((IP_LOOKUP_NEXT_MIDCHAIN == adj->lookup_next_index) || + (IP_LOOKUP_NEXT_MCAST_MIDCHAIN == adj->lookup_next_index)); + + dpo_stack_from_node(adj_nbr_midchain_get_tx_node(adj), + &adj->sub_type.midchain.next_dpo, + next); +} + +u8* +format_adj_midchain (u8* s, va_list *ap) +{ + index_t index = va_arg(*ap, index_t); + u32 indent = va_arg(*ap, u32); + ip_adjacency_t * adj = adj_get(index); + + s = format (s, "%U", format_vnet_link, adj->ia_link); + s = format (s, " via %U ", + format_ip46_address, &adj->sub_type.nbr.next_hop); + s = format (s, " %U", + format_vnet_rewrite, + &adj->rewrite_header, sizeof (adj->rewrite_data), indent); + s = format (s, "\n%Ustacked-on:\n%U%U", + format_white_space, indent, + format_white_space, indent+2, + format_dpo_id, &adj->sub_type.midchain.next_dpo, indent+2); + + return (s); +} + +static void +adj_dpo_lock (dpo_id_t *dpo) +{ + adj_lock(dpo->dpoi_index); +} +static void +adj_dpo_unlock (dpo_id_t *dpo) +{ + adj_unlock(dpo->dpoi_index); +} + +const static dpo_vft_t adj_midchain_dpo_vft = { + .dv_lock = adj_dpo_lock, + .dv_unlock = adj_dpo_unlock, + .dv_format = format_adj_midchain, +}; + +/** + * @brief The per-protocol VLIB graph nodes that are assigned to a midchain + * object. + * + * this means that these graph nodes are ones from which a midchain is the + * parent object in the DPO-graph. + */ +const static char* const midchain_ip4_nodes[] = +{ + "ip4-midchain", + NULL, +}; +const static char* const midchain_ip6_nodes[] = +{ + "ip6-midchain", + NULL, +}; +const static char* const midchain_mpls_nodes[] = +{ + "mpls-midchain", + NULL, +}; +const static char* const midchain_ethernet_nodes[] = +{ + "adj-l2-midchain", + NULL, +}; +const static char* const midchain_nsh_nodes[] = +{ + "adj-nsh-midchain", + NULL, +}; + +const static char* const * const midchain_nodes[DPO_PROTO_NUM] = +{ + [DPO_PROTO_IP4] = midchain_ip4_nodes, + [DPO_PROTO_IP6] = midchain_ip6_nodes, + [DPO_PROTO_MPLS] = midchain_mpls_nodes, + [DPO_PROTO_ETHERNET] = midchain_ethernet_nodes, + [DPO_PROTO_NSH] = midchain_nsh_nodes, +}; + +void +adj_midchain_module_init (void) +{ + dpo_register(DPO_ADJACENCY_MIDCHAIN, &adj_midchain_dpo_vft, midchain_nodes); +} diff --git a/src/vnet/adj/adj_midchain.h b/src/vnet/adj/adj_midchain.h new file mode 100644 index 00000000..27ca1d33 --- /dev/null +++ b/src/vnet/adj/adj_midchain.h @@ -0,0 +1,82 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +/** + * Midchain Adjacency sub-type. These adjs represent an L3 peer on a + * tunnel interface. The tunnel's adjacency is thus not the end of the chain, + * and needs to stack on/link to another chain (or portion of the graph) to + * reach the tunnel's destination. + */ + +#ifndef __ADJ_MIDCHAIN_H__ +#define __ADJ_MIDCHAIN_H__ + +#include <vnet/adj/adj.h> + +/** + * @brief + * Convert an existing neighbour adjacency into a midchain + * + * @param adj_index + * The index of the neighbour adjacency. + * + * @param post_rewrite_node + * The VLIB graph node that provides the post-encap fixup. + * where 'fixup' is e.g., correcting chksum, length, etc. + * + * @param rewrite + * The rewrite. + */ +extern void adj_nbr_midchain_update_rewrite(adj_index_t adj_index, + adj_midchain_fixup_t fixup, + adj_flags_t flags, + u8 *rewrite); + +/** + * @brief + * [re]stack a midchain. 'Stacking' is the act of forming parent-child + * relationships in the data-plane graph. + * + * @param adj_index + * The index of the midchain to stack + * + * @param dpo + * The parent DPO to stack onto (i.e. become a child of). + */ +extern void adj_nbr_midchain_stack(adj_index_t adj_index, + const dpo_id_t *dpo); + +/** + * @brief + * unstack a midchain. This will break the chain between the midchain and + * the next graph section. This is a implemented as stack-on-drop + * + * @param adj_index + * The index of the midchain to stack + */ +extern void adj_nbr_midchain_unstack(adj_index_t adj_index); + +/** + * @brief + * Module initialisation + */ +extern void adj_midchain_module_init(void); + +/** + * @brief + * Format a midchain adjacency + */ +extern u8* format_adj_midchain(u8* s, va_list *ap); + +#endif diff --git a/src/vnet/adj/adj_nbr.c b/src/vnet/adj/adj_nbr.c new file mode 100644 index 00000000..3d450d1f --- /dev/null +++ b/src/vnet/adj/adj_nbr.c @@ -0,0 +1,1124 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/adj/adj_nbr.h> +#include <vnet/adj/adj_internal.h> +#include <vnet/ethernet/arp_packet.h> +#include <vnet/fib/fib_walk.h> + +/* + * Vector Hash tables of neighbour (traditional) adjacencies + * Key: interface(for the vector index), address (and its proto), + * link-type/ether-type. + */ +static BVT(clib_bihash) **adj_nbr_tables[FIB_PROTOCOL_MAX]; + +// FIXME SIZE APPROPRIATELY. ASK DAVEB. +#define ADJ_NBR_DEFAULT_HASH_NUM_BUCKETS (64 * 64) +#define ADJ_NBR_DEFAULT_HASH_MEMORY_SIZE (32<<20) + + +#define ADJ_NBR_SET_KEY(_key, _lt, _nh) \ +{ \ + _key.key[0] = (_nh)->as_u64[0]; \ + _key.key[1] = (_nh)->as_u64[1]; \ + _key.key[2] = (_lt); \ +} + +#define ADJ_NBR_ITF_OK(_proto, _itf) \ + (((_itf) < vec_len(adj_nbr_tables[_proto])) && \ + (NULL != adj_nbr_tables[_proto][sw_if_index])) + +static void +adj_nbr_insert (fib_protocol_t nh_proto, + vnet_link_t link_type, + const ip46_address_t *nh_addr, + u32 sw_if_index, + adj_index_t adj_index) +{ + BVT(clib_bihash_kv) kv; + + if (sw_if_index >= vec_len(adj_nbr_tables[nh_proto])) + { + vec_validate(adj_nbr_tables[nh_proto], sw_if_index); + } + if (NULL == adj_nbr_tables[nh_proto][sw_if_index]) + { + adj_nbr_tables[nh_proto][sw_if_index] = + clib_mem_alloc_aligned(sizeof(BVT(clib_bihash)), + CLIB_CACHE_LINE_BYTES); + memset(adj_nbr_tables[nh_proto][sw_if_index], + 0, + sizeof(BVT(clib_bihash))); + + BV(clib_bihash_init) (adj_nbr_tables[nh_proto][sw_if_index], + "Adjacency Neighbour table", + ADJ_NBR_DEFAULT_HASH_NUM_BUCKETS, + ADJ_NBR_DEFAULT_HASH_MEMORY_SIZE); + } + + ADJ_NBR_SET_KEY(kv, link_type, nh_addr); + kv.value = adj_index; + + BV(clib_bihash_add_del) (adj_nbr_tables[nh_proto][sw_if_index], &kv, 1); +} + +void +adj_nbr_remove (adj_index_t ai, + fib_protocol_t nh_proto, + vnet_link_t link_type, + const ip46_address_t *nh_addr, + u32 sw_if_index) +{ + BVT(clib_bihash_kv) kv; + + if (!ADJ_NBR_ITF_OK(nh_proto, sw_if_index)) + return; + + ADJ_NBR_SET_KEY(kv, link_type, nh_addr); + kv.value = ai; + + BV(clib_bihash_add_del) (adj_nbr_tables[nh_proto][sw_if_index], &kv, 0); +} + +static adj_index_t +adj_nbr_find (fib_protocol_t nh_proto, + vnet_link_t link_type, + const ip46_address_t *nh_addr, + u32 sw_if_index) +{ + BVT(clib_bihash_kv) kv; + + ADJ_NBR_SET_KEY(kv, link_type, nh_addr); + + if (!ADJ_NBR_ITF_OK(nh_proto, sw_if_index)) + return (ADJ_INDEX_INVALID); + + if (BV(clib_bihash_search)(adj_nbr_tables[nh_proto][sw_if_index], + &kv, &kv) < 0) + { + return (ADJ_INDEX_INVALID); + } + else + { + return (kv.value); + } +} + +static inline u32 +adj_get_nd_node (fib_protocol_t proto) +{ + switch (proto) { + case FIB_PROTOCOL_IP4: + return (ip4_arp_node.index); + case FIB_PROTOCOL_IP6: + return (ip6_discover_neighbor_node.index); + case FIB_PROTOCOL_MPLS: + break; + } + ASSERT(0); + return (ip4_arp_node.index); +} + +/** + * @brief Check and set feature flags if o/p interface has any o/p features. + */ +static void +adj_nbr_evaluate_feature (adj_index_t ai) +{ + ip_adjacency_t *adj; + vnet_feature_main_t *fm = &feature_main; + i16 feature_count; + u8 arc_index; + u32 sw_if_index; + + adj = adj_get(ai); + + switch (adj->ia_link) + { + case VNET_LINK_IP4: + arc_index = ip4_main.lookup_main.output_feature_arc_index; + break; + case VNET_LINK_IP6: + arc_index = ip6_main.lookup_main.output_feature_arc_index; + break; + case VNET_LINK_MPLS: + arc_index = mpls_main.output_feature_arc_index; + break; + default: + return; + } + + sw_if_index = adj->rewrite_header.sw_if_index; + if (vec_len(fm->feature_count_by_sw_if_index[arc_index]) > sw_if_index) + { + feature_count = fm->feature_count_by_sw_if_index[arc_index][sw_if_index]; + if (feature_count > 0) + adj->rewrite_header.flags |= VNET_REWRITE_HAS_FEATURES; + } + + return; +} + +static ip_adjacency_t* +adj_nbr_alloc (fib_protocol_t nh_proto, + vnet_link_t link_type, + const ip46_address_t *nh_addr, + u32 sw_if_index) +{ + ip_adjacency_t *adj; + + adj = adj_alloc(nh_proto); + + adj_nbr_insert(nh_proto, link_type, nh_addr, + sw_if_index, + adj_get_index(adj)); + + /* + * since we just added the ADJ we have no rewrite string for it, + * so its for ARP + */ + adj->lookup_next_index = IP_LOOKUP_NEXT_ARP; + adj->sub_type.nbr.next_hop = *nh_addr; + adj->ia_link = link_type; + adj->ia_nh_proto = nh_proto; + adj->rewrite_header.sw_if_index = sw_if_index; + + adj_nbr_evaluate_feature (adj_get_index(adj)); + return (adj); +} + +/* + * adj_nbr_add_or_lock + * + * Add an adjacency for the neighbour requested. + * + * The key for an adj is: + * - the Next-hops protocol (i.e. v4 or v6) + * - the address of the next-hop + * - the interface the next-hop is reachable through + */ +adj_index_t +adj_nbr_add_or_lock (fib_protocol_t nh_proto, + vnet_link_t link_type, + const ip46_address_t *nh_addr, + u32 sw_if_index) +{ + adj_index_t adj_index; + ip_adjacency_t *adj; + + adj_index = adj_nbr_find(nh_proto, link_type, nh_addr, sw_if_index); + + if (ADJ_INDEX_INVALID == adj_index) + { + vnet_main_t *vnm; + + vnm = vnet_get_main(); + adj = adj_nbr_alloc(nh_proto, link_type, nh_addr, sw_if_index); + adj_index = adj_get_index(adj); + adj_lock(adj_index); + + vnet_rewrite_init(vnm, sw_if_index, + adj_get_nd_node(nh_proto), + vnet_tx_node_index_for_sw_interface(vnm, sw_if_index), + &adj->rewrite_header); + + /* + * we need a rewrite where the destination IP address is converted + * to the appropriate link-layer address. This is interface specific. + * So ask the interface to do it. + */ + vnet_update_adjacency_for_sw_interface(vnm, sw_if_index, adj_index); + } + else + { + adj_lock(adj_index); + } + + return (adj_index); +} + +adj_index_t +adj_nbr_add_or_lock_w_rewrite (fib_protocol_t nh_proto, + vnet_link_t link_type, + const ip46_address_t *nh_addr, + u32 sw_if_index, + u8 *rewrite) +{ + adj_index_t adj_index; + ip_adjacency_t *adj; + + adj_index = adj_nbr_find(nh_proto, link_type, nh_addr, sw_if_index); + + if (ADJ_INDEX_INVALID == adj_index) + { + adj = adj_nbr_alloc(nh_proto, link_type, nh_addr, sw_if_index); + adj->rewrite_header.sw_if_index = sw_if_index; + } + else + { + adj = adj_get(adj_index); + } + + adj_lock(adj_get_index(adj)); + adj_nbr_update_rewrite(adj_get_index(adj), + ADJ_NBR_REWRITE_FLAG_COMPLETE, + rewrite); + + return (adj_get_index(adj)); +} + +/** + * adj_nbr_update_rewrite + * + * Update the adjacency's rewrite string. A NULL string implies the + * rewirte is reset (i.e. when ARP/ND etnry is gone). + * NB: the adj being updated may be handling traffic in the DP. + */ +void +adj_nbr_update_rewrite (adj_index_t adj_index, + adj_nbr_rewrite_flag_t flags, + u8 *rewrite) +{ + ip_adjacency_t *adj; + + ASSERT(ADJ_INDEX_INVALID != adj_index); + + adj = adj_get(adj_index); + + if (flags & ADJ_NBR_REWRITE_FLAG_COMPLETE) + { + /* + * update the adj's rewrite string and build the arc + * from the rewrite node to the interface's TX node + */ + adj_nbr_update_rewrite_internal(adj, IP_LOOKUP_NEXT_REWRITE, + adj_get_rewrite_node(adj->ia_link), + vnet_tx_node_index_for_sw_interface( + vnet_get_main(), + adj->rewrite_header.sw_if_index), + rewrite); + } + else + { + adj_nbr_update_rewrite_internal(adj, IP_LOOKUP_NEXT_ARP, + adj_get_nd_node(adj->ia_nh_proto), + vnet_tx_node_index_for_sw_interface( + vnet_get_main(), + adj->rewrite_header.sw_if_index), + rewrite); + } +} + +/** + * adj_nbr_update_rewrite_internal + * + * Update the adjacency's rewrite string. A NULL string implies the + * rewirte is reset (i.e. when ARP/ND etnry is gone). + * NB: the adj being updated may be handling traffic in the DP. + */ +void +adj_nbr_update_rewrite_internal (ip_adjacency_t *adj, + ip_lookup_next_t adj_next_index, + u32 this_node, + u32 next_node, + u8 *rewrite) +{ + ip_adjacency_t *walk_adj; + adj_index_t walk_ai; + vlib_main_t * vm; + u32 old_next; + int do_walk; + + vm = vlib_get_main(); + old_next = adj->lookup_next_index; + + walk_ai = adj_get_index(adj); + if (VNET_LINK_MPLS == adj->ia_link) + { + /* + * The link type MPLS has no children in the control plane graph, it only + * has children in the data-palne graph. The backwalk is up the former. + * So we need to walk from its IP cousin. + */ + walk_ai = adj_nbr_find(adj->ia_nh_proto, + fib_proto_to_link(adj->ia_nh_proto), + &adj->sub_type.nbr.next_hop, + adj->rewrite_header.sw_if_index); + } + + /* + * Don't call the walk re-entrantly + */ + if (ADJ_INDEX_INVALID != walk_ai) + { + walk_adj = adj_get(walk_ai); + if (ADJ_FLAG_SYNC_WALK_ACTIVE & walk_adj->ia_flags) + { + do_walk = 0; + } + else + { + /* + * Prevent re-entrant walk of the same adj + */ + walk_adj->ia_flags |= ADJ_FLAG_SYNC_WALK_ACTIVE; + do_walk = 1; + } + } + else + { + do_walk = 0; + } + + /* + * lock the adjacencies that are affected by updates this walk will provoke. + * Since the aim of the walk is to update children to link to a different + * DPO, this adj will no longer be in use and its lock count will drop to 0. + * We don't want it to be deleted as part of this endevour. + */ + adj_lock(adj_get_index(adj)); + adj_lock(walk_ai); + + /* + * Updating a rewrite string is not atomic; + * - the rewrite string is too long to write in one instruction + * - when swapping from incomplete to complete, we also need to update + * the VLIB graph next-index of the adj. + * ideally we would only want to suspend forwarding via this adj whilst we + * do this, but we do not have that level of granularity - it's suspend all + * worker threads or nothing. + * The other chioces are: + * - to mark the adj down and back walk so child load-balances drop this adj + * from the set. + * - update the next_node index of this adj to point to error-drop + * both of which will mean for MAC change we will drop for this adj + * which is not acceptable. However, when the adj changes type (from + * complete to incomplete and vice-versa) the child DPOs, which have the + * VLIB graph next node index, will be sending packets to the wrong graph + * node. So from the options above, updating the next_node of the adj to + * be drop will work, but it relies on each graph node v4/v6/mpls, rewrite/ + * arp/midchain always be valid w.r.t. a mis-match of adj type and node type + * (i.e. a rewrite adj in the arp node). This is not enforcable. Getting it + * wrong will lead to hard to find bugs since its a race condition. So we + * choose the more reliable method of updating the children to use the drop, + * then switching adj's type, then updating the children again. Did I mention + * that this doesn't happen often... + * So we need to distinguish between the two cases: + * 1 - mac change + * 2 - adj type change + */ + if (do_walk && + old_next != adj_next_index && + ADJ_INDEX_INVALID != walk_ai) + { + /* + * the adj is changing type. we need to fix all children so that they + * stack momentarily on a drop, while the adj changes. If we don't do + * this the children will send packets to a VLIB graph node that does + * not correspond to the adj's type - and it goes downhill from there. + */ + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = FIB_NODE_BW_REASON_FLAG_ADJ_DOWN, + /* + * force this walk to be synchrous. if we don't and a node in the graph + * (a heavily shared path-list) chooses to back-ground the walk (make it + * async) then it will pause and we will do the adj update below, before + * all the children are updated. not good. + */ + .fnbw_flags = FIB_NODE_BW_FLAG_FORCE_SYNC, + }; + + fib_walk_sync(FIB_NODE_TYPE_ADJ, walk_ai, &bw_ctx); + } + + /* + * If we are just updating the MAC string of the adj (which we also can't + * do atomically), then we need to stop packets switching through the adj. + * We can't do that on a per-adj basis, so it's all the packets. + * If we are updating the type, and we walked back to the children above, + * then this barrier serves to flush the queues/frames. + */ + vlib_worker_thread_barrier_sync(vm); + + adj->lookup_next_index = adj_next_index; + + if (NULL != rewrite) + { + /* + * new rewrite provided. + * fill in the adj's rewrite string, and build the VLIB graph arc. + */ + vnet_rewrite_set_data_internal(&adj->rewrite_header, + sizeof(adj->rewrite_data), + rewrite, + vec_len(rewrite)); + vec_free(rewrite); + } + else + { + vnet_rewrite_clear_data_internal(&adj->rewrite_header, + sizeof(adj->rewrite_data)); + } + adj->rewrite_header.next_index = vlib_node_add_next(vlib_get_main(), + this_node, + next_node); + + /* + * done with the rewirte update - let the workers loose. + */ + vlib_worker_thread_barrier_release(vm); + + if (do_walk && + (old_next != adj->lookup_next_index) && + (ADJ_INDEX_INVALID != walk_ai)) + { + /* + * backwalk to the children so they can stack on the now updated + * adjacency + */ + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = FIB_NODE_BW_REASON_FLAG_ADJ_UPDATE, + }; + + fib_walk_sync(FIB_NODE_TYPE_ADJ, walk_ai, &bw_ctx); + } + /* + * Prevent re-entrant walk of the same adj + */ + if (do_walk) + { + walk_adj->ia_flags &= ~ADJ_FLAG_SYNC_WALK_ACTIVE; + } + + adj_unlock(adj_get_index(adj)); + adj_unlock(walk_ai); +} + +typedef struct adj_db_count_ctx_t_ { + u64 count; +} adj_db_count_ctx_t; + +static void +adj_db_count (BVT(clib_bihash_kv) * kvp, + void *arg) +{ + adj_db_count_ctx_t * ctx = arg; + ctx->count++; +} + +u32 +adj_nbr_db_size (void) +{ + adj_db_count_ctx_t ctx = { + .count = 0, + }; + fib_protocol_t proto; + u32 sw_if_index = 0; + + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + vec_foreach_index(sw_if_index, adj_nbr_tables[proto]) + { + if (NULL != adj_nbr_tables[proto][sw_if_index]) + { + BV(clib_bihash_foreach_key_value_pair) ( + adj_nbr_tables[proto][sw_if_index], + adj_db_count, + &ctx); + } + } + } + return (ctx.count); +} + +/** + * @brief Context for a walk of the adjacency neighbour DB + */ +typedef struct adj_walk_ctx_t_ +{ + adj_walk_cb_t awc_cb; + void *awc_ctx; +} adj_walk_ctx_t; + +static void +adj_nbr_walk_cb (BVT(clib_bihash_kv) * kvp, + void *arg) +{ + adj_walk_ctx_t *ctx = arg; + + // FIXME: can't stop early... + ctx->awc_cb(kvp->value, ctx->awc_ctx); +} + +void +adj_nbr_walk (u32 sw_if_index, + fib_protocol_t adj_nh_proto, + adj_walk_cb_t cb, + void *ctx) +{ + if (!ADJ_NBR_ITF_OK(adj_nh_proto, sw_if_index)) + return; + + adj_walk_ctx_t awc = { + .awc_ctx = ctx, + .awc_cb = cb, + }; + + BV(clib_bihash_foreach_key_value_pair) ( + adj_nbr_tables[adj_nh_proto][sw_if_index], + adj_nbr_walk_cb, + &awc); +} + +/** + * @brief Context for a walk of the adjacency neighbour DB + */ +typedef struct adj_walk_nh_ctx_t_ +{ + adj_walk_cb_t awc_cb; + void *awc_ctx; + const ip46_address_t *awc_nh; +} adj_walk_nh_ctx_t; + +static void +adj_nbr_walk_nh_cb (BVT(clib_bihash_kv) * kvp, + void *arg) +{ + ip_adjacency_t *adj; + adj_walk_nh_ctx_t *ctx = arg; + + adj = adj_get(kvp->value); + + if (!ip46_address_cmp(&adj->sub_type.nbr.next_hop, ctx->awc_nh)) + ctx->awc_cb(kvp->value, ctx->awc_ctx); +} + +/** + * @brief Walk adjacencies on a link with a given v4 next-hop. + * that is visit the adjacencies with different link types. + */ +void +adj_nbr_walk_nh4 (u32 sw_if_index, + const ip4_address_t *addr, + adj_walk_cb_t cb, + void *ctx) +{ + if (!ADJ_NBR_ITF_OK(FIB_PROTOCOL_IP4, sw_if_index)) + return; + + ip46_address_t nh = { + .ip4 = *addr, + }; + + adj_walk_nh_ctx_t awc = { + .awc_ctx = ctx, + .awc_cb = cb, + .awc_nh = &nh, + }; + + BV(clib_bihash_foreach_key_value_pair) ( + adj_nbr_tables[FIB_PROTOCOL_IP4][sw_if_index], + adj_nbr_walk_nh_cb, + &awc); +} + +/** + * @brief Walk adjacencies on a link with a given v6 next-hop. + * that is visit the adjacencies with different link types. + */ +void +adj_nbr_walk_nh6 (u32 sw_if_index, + const ip6_address_t *addr, + adj_walk_cb_t cb, + void *ctx) +{ + if (!ADJ_NBR_ITF_OK(FIB_PROTOCOL_IP6, sw_if_index)) + return; + + ip46_address_t nh = { + .ip6 = *addr, + }; + + adj_walk_nh_ctx_t awc = { + .awc_ctx = ctx, + .awc_cb = cb, + .awc_nh = &nh, + }; + + BV(clib_bihash_foreach_key_value_pair) ( + adj_nbr_tables[FIB_PROTOCOL_IP6][sw_if_index], + adj_nbr_walk_nh_cb, + &awc); +} + +/** + * @brief Walk adjacencies on a link with a given next-hop. + * that is visit the adjacencies with different link types. + */ +void +adj_nbr_walk_nh (u32 sw_if_index, + fib_protocol_t adj_nh_proto, + const ip46_address_t *nh, + adj_walk_cb_t cb, + void *ctx) +{ + if (!ADJ_NBR_ITF_OK(adj_nh_proto, sw_if_index)) + return; + + adj_walk_nh_ctx_t awc = { + .awc_ctx = ctx, + .awc_cb = cb, + .awc_nh = nh, + }; + + BV(clib_bihash_foreach_key_value_pair) ( + adj_nbr_tables[adj_nh_proto][sw_if_index], + adj_nbr_walk_nh_cb, + &awc); +} + +/** + * Flags associated with the interface state walks + */ +typedef enum adj_nbr_interface_flags_t_ +{ + ADJ_NBR_INTERFACE_UP = (1 << 0), +} adj_nbr_interface_flags_t; + +/** + * Context for the state change walk of the DB + */ +typedef struct adj_nbr_interface_state_change_ctx_t_ +{ + /** + * Flags on the interface + */ + adj_nbr_interface_flags_t flags; +} adj_nbr_interface_state_change_ctx_t; + +static adj_walk_rc_t +adj_nbr_interface_state_change_one (adj_index_t ai, + void *arg) +{ + /* + * Back walk the graph to inform the forwarding entries + * that this interface state has changed. Do this synchronously + * since this is the walk that provides convergence + */ + adj_nbr_interface_state_change_ctx_t *ctx = arg; + + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = ((ctx->flags & ADJ_NBR_INTERFACE_UP) ? + FIB_NODE_BW_REASON_FLAG_INTERFACE_UP : + FIB_NODE_BW_REASON_FLAG_INTERFACE_DOWN), + /* + * the force sync applies only as far as the first fib_entry. + * And it's the fib_entry's we need to converge away from + * the adjacencies on the now down link + */ + .fnbw_flags = (!(ctx->flags & ADJ_NBR_INTERFACE_UP) ? + FIB_NODE_BW_FLAG_FORCE_SYNC : + 0), + }; + + fib_walk_sync(FIB_NODE_TYPE_ADJ, ai, &bw_ctx); + + return (ADJ_WALK_RC_CONTINUE); +} + +/** + * @brief Registered function for SW interface state changes + */ +static clib_error_t * +adj_nbr_sw_interface_state_change (vnet_main_t * vnm, + u32 sw_if_index, + u32 flags) +{ + fib_protocol_t proto; + + /* + * walk each adj on the interface and trigger a walk from that adj + */ + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + adj_nbr_interface_state_change_ctx_t ctx = { + .flags = ((flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) ? + ADJ_NBR_INTERFACE_UP : + 0), + }; + + adj_nbr_walk(sw_if_index, proto, + adj_nbr_interface_state_change_one, + &ctx); + } + + return (NULL); +} + +VNET_SW_INTERFACE_ADMIN_UP_DOWN_FUNCTION_PRIO( + adj_nbr_sw_interface_state_change, + VNET_ITF_FUNC_PRIORITY_HIGH); + +/** + * @brief Invoked on each SW interface of a HW interface when the + * HW interface state changes + */ +static void +adj_nbr_hw_sw_interface_state_change (vnet_main_t * vnm, + u32 sw_if_index, + void *arg) +{ + adj_nbr_interface_state_change_ctx_t *ctx = arg; + fib_protocol_t proto; + + /* + * walk each adj on the interface and trigger a walk from that adj + */ + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + adj_nbr_walk(sw_if_index, proto, + adj_nbr_interface_state_change_one, + ctx); + } +} + +/** + * @brief Registered callback for HW interface state changes + */ +static clib_error_t * +adj_nbr_hw_interface_state_change (vnet_main_t * vnm, + u32 hw_if_index, + u32 flags) +{ + /* + * walk SW interface on the HW + */ + adj_nbr_interface_state_change_ctx_t ctx = { + .flags = ((flags & VNET_HW_INTERFACE_FLAG_LINK_UP) ? + ADJ_NBR_INTERFACE_UP : + 0), + }; + + vnet_hw_interface_walk_sw(vnm, hw_if_index, + adj_nbr_hw_sw_interface_state_change, + &ctx); + + return (NULL); +} + +VNET_HW_INTERFACE_LINK_UP_DOWN_FUNCTION_PRIO( + adj_nbr_hw_interface_state_change, + VNET_ITF_FUNC_PRIORITY_HIGH); + +static adj_walk_rc_t +adj_nbr_interface_delete_one (adj_index_t ai, + void *arg) +{ + /* + * Back walk the graph to inform the forwarding entries + * that this interface has been deleted. + */ + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = FIB_NODE_BW_REASON_FLAG_INTERFACE_DELETE, + }; + + fib_walk_sync(FIB_NODE_TYPE_ADJ, ai, &bw_ctx); + + return (ADJ_WALK_RC_CONTINUE); +} + +/** + * adj_nbr_interface_add_del + * + * Registered to receive interface Add and delete notifications + */ +static clib_error_t * +adj_nbr_interface_add_del (vnet_main_t * vnm, + u32 sw_if_index, + u32 is_add) +{ + fib_protocol_t proto; + + if (is_add) + { + /* + * not interested in interface additions. we will not back walk + * to resolve paths through newly added interfaces. Why? The control + * plane should have the brains to add interfaces first, then routes. + * So the case where there are paths with a interface that matches + * one just created is the case where the path resolved through an + * interface that was deleted, and still has not been removed. The + * new interface added, is NO GUARANTEE that the interface being + * added now, even though it may have the same sw_if_index, is the + * same interface that the path needs. So tough! + * If the control plane wants these routes to resolve it needs to + * remove and add them again. + */ + return (NULL); + } + + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + adj_nbr_walk(sw_if_index, proto, + adj_nbr_interface_delete_one, + NULL); + } + + return (NULL); + +} + +VNET_SW_INTERFACE_ADD_DEL_FUNCTION(adj_nbr_interface_add_del); + + +static adj_walk_rc_t +adj_nbr_show_one (adj_index_t ai, + void *arg) +{ + vlib_cli_output (arg, "[@%d] %U", + ai, + format_ip_adjacency, ai, + FORMAT_IP_ADJACENCY_NONE); + + return (ADJ_WALK_RC_CONTINUE); +} + +static clib_error_t * +adj_nbr_show (vlib_main_t * vm, + unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + adj_index_t ai = ADJ_INDEX_INVALID; + u32 sw_if_index = ~0; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "%d", &ai)) + ; + else if (unformat (input, "%U", + unformat_vnet_sw_interface, vnet_get_main(), + &sw_if_index)) + ; + else + break; + } + + if (ADJ_INDEX_INVALID != ai) + { + vlib_cli_output (vm, "[@%d] %U", + ai, + format_ip_adjacency, ai, + FORMAT_IP_ADJACENCY_DETAIL); + } + else if (~0 != sw_if_index) + { + fib_protocol_t proto; + + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + adj_nbr_walk(sw_if_index, proto, + adj_nbr_show_one, + vm); + } + } + else + { + fib_protocol_t proto; + + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + vec_foreach_index(sw_if_index, adj_nbr_tables[proto]) + { + adj_nbr_walk(sw_if_index, proto, + adj_nbr_show_one, + vm); + } + } + } + + return 0; +} + +/*? + * Show all neighbour adjacencies. + * @cliexpar + * @cliexstart{sh adj nbr} + * [@2] ipv4 via 1.0.0.2 loop0: IP4: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc + * [@3] mpls via 1.0.0.2 loop0: MPLS_UNICAST: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc + * [@4] ipv4 via 1.0.0.3 loop0: IP4: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc + * [@5] mpls via 1.0.0.3 loop0: MPLS_UNICAST: 00:00:22:aa:bb:cc -> 00:00:11:aa:bb:cc + * @cliexend + ?*/ +VLIB_CLI_COMMAND (ip4_show_fib_command, static) = { + .path = "show adj nbr", + .short_help = "show adj nbr [<adj_index>] [interface]", + .function = adj_nbr_show, +}; + +static ip46_type_t +adj_proto_to_46 (fib_protocol_t proto) +{ + switch (proto) + { + case FIB_PROTOCOL_IP4: + return (IP46_TYPE_IP4); + case FIB_PROTOCOL_IP6: + return (IP46_TYPE_IP6); + default: + return (IP46_TYPE_IP4); + } + return (IP46_TYPE_IP4); +} + +u8* +format_adj_nbr_incomplete (u8* s, va_list *ap) +{ + index_t index = va_arg(*ap, index_t); + CLIB_UNUSED(u32 indent) = va_arg(*ap, u32); + vnet_main_t * vnm = vnet_get_main(); + ip_adjacency_t * adj = adj_get(index); + + s = format (s, "arp-%U", format_vnet_link, adj->ia_link); + s = format (s, ": via %U", + format_ip46_address, &adj->sub_type.nbr.next_hop, + adj_proto_to_46(adj->ia_nh_proto)); + s = format (s, " %U", + format_vnet_sw_interface_name, + vnm, + vnet_get_sw_interface(vnm, + adj->rewrite_header.sw_if_index)); + + return (s); +} + +u8* +format_adj_nbr (u8* s, va_list *ap) +{ + index_t index = va_arg(*ap, index_t); + CLIB_UNUSED(u32 indent) = va_arg(*ap, u32); + ip_adjacency_t * adj = adj_get(index); + + s = format (s, "%U", format_vnet_link, adj->ia_link); + s = format (s, " via %U ", + format_ip46_address, &adj->sub_type.nbr.next_hop, + adj_proto_to_46(adj->ia_nh_proto)); + s = format (s, "%U", + format_vnet_rewrite, + &adj->rewrite_header, sizeof (adj->rewrite_data), 0); + + return (s); +} + +static void +adj_dpo_lock (dpo_id_t *dpo) +{ + adj_lock(dpo->dpoi_index); +} +static void +adj_dpo_unlock (dpo_id_t *dpo) +{ + adj_unlock(dpo->dpoi_index); +} + +static void +adj_mem_show (void) +{ + fib_show_memory_usage("Adjacency", + pool_elts(adj_pool), + pool_len(adj_pool), + sizeof(ip_adjacency_t)); +} + +const static dpo_vft_t adj_nbr_dpo_vft = { + .dv_lock = adj_dpo_lock, + .dv_unlock = adj_dpo_unlock, + .dv_format = format_adj_nbr, + .dv_mem_show = adj_mem_show, +}; +const static dpo_vft_t adj_nbr_incompl_dpo_vft = { + .dv_lock = adj_dpo_lock, + .dv_unlock = adj_dpo_unlock, + .dv_format = format_adj_nbr_incomplete, +}; + +/** + * @brief The per-protocol VLIB graph nodes that are assigned to an adjacency + * object. + * + * this means that these graph nodes are ones from which a nbr is the + * parent object in the DPO-graph. + */ +const static char* const nbr_ip4_nodes[] = +{ + "ip4-rewrite", + NULL, +}; +const static char* const nbr_ip6_nodes[] = +{ + "ip6-rewrite", + NULL, +}; +const static char* const nbr_mpls_nodes[] = +{ + "mpls-output", + NULL, +}; +const static char* const nbr_ethernet_nodes[] = +{ + "adj-l2-rewrite", + NULL, +}; +const static char* const * const nbr_nodes[DPO_PROTO_NUM] = +{ + [DPO_PROTO_IP4] = nbr_ip4_nodes, + [DPO_PROTO_IP6] = nbr_ip6_nodes, + [DPO_PROTO_MPLS] = nbr_mpls_nodes, + [DPO_PROTO_ETHERNET] = nbr_ethernet_nodes, +}; + +const static char* const nbr_incomplete_ip4_nodes[] = +{ + "ip4-arp", + NULL, +}; +const static char* const nbr_incomplete_ip6_nodes[] = +{ + "ip6-discover-neighbor", + NULL, +}; +const static char* const nbr_incomplete_mpls_nodes[] = +{ + "mpls-adj-incomplete", + NULL, +}; + +const static char* const * const nbr_incomplete_nodes[DPO_PROTO_NUM] = +{ + [DPO_PROTO_IP4] = nbr_incomplete_ip4_nodes, + [DPO_PROTO_IP6] = nbr_incomplete_ip6_nodes, + [DPO_PROTO_MPLS] = nbr_incomplete_mpls_nodes, +}; + +void +adj_nbr_module_init (void) +{ + dpo_register(DPO_ADJACENCY, + &adj_nbr_dpo_vft, + nbr_nodes); + dpo_register(DPO_ADJACENCY_INCOMPLETE, + &adj_nbr_incompl_dpo_vft, + nbr_incomplete_nodes); +} diff --git a/src/vnet/adj/adj_nbr.h b/src/vnet/adj/adj_nbr.h new file mode 100644 index 00000000..293766b8 --- /dev/null +++ b/src/vnet/adj/adj_nbr.h @@ -0,0 +1,176 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +/** + * @brief + * Neighbour Adjacency sub-type. These adjs represent an L3 peer on a + * connected link. + */ + +#ifndef __ADJ_NBR_H__ +#define __ADJ_NBR_H__ + +#include <vnet/vnet.h> +#include <vnet/adj/adj_types.h> +#include <vnet/fib/fib_node.h> +#include <vnet/dpo/dpo.h> + +/** + * @brief + * Add (and lock) a new or lock an existing neighbour adjacency + * + * @param nh_proto + * The protocol for the next-hop address (v4 or v6) + * + * @param link_type + * A description of the protocol of the packets that will forward + * through this adj. On an ethernet interface this is the MAC header's + * ether-type + * + * @param nh_addr + * The address of the next-hop/peer to send the packet to + * + * @param sw_if_index + * The interface on which the peer resides + */ +extern adj_index_t adj_nbr_add_or_lock(fib_protocol_t nh_proto, + vnet_link_t link_type, + const ip46_address_t *nh_addr, + u32 sw_if_index); + +/** + * @brief + * Add (and lock) a new or lock an existing neighbour adjacency + * + * @param nh_proto + * The protocol for the next-hop address (v4 or v6) + * + * @param link_type + * A description of the protocol of the packets that will forward + * through this adj. On an ethernet interface this is the MAC header's + * ether-type + * + * @param nh_addr + * The address of the next-hop/peer to send the packet to + * + * @param sw_if_index + * The interface on which the peer resides + * + * @param rewrite + * The rewrite to prepend to packets + */ +extern adj_index_t adj_nbr_add_or_lock_w_rewrite(fib_protocol_t nh_proto, + vnet_link_t link_type, + const ip46_address_t *nh_addr, + u32 sw_if_index, + u8 *rewrite); +/** + * @brief When adding a rewrite to an adjacency these are flags that + * apply to that rewrite + */ +typedef enum adj_nbr_rewrite_flag_t_ +{ + ADJ_NBR_REWRITE_FLAG_NONE, + + /** + * An indication that the rewrite is incomplete, i.e. that it describes the + * ARP/ND rewrite when probing. + */ + ADJ_NBR_REWRITE_FLAG_INCOMPLETE = ADJ_NBR_REWRITE_FLAG_NONE, + + /** + * An indication that the rewrite is complete, i.e. that it fully describes + * the link-layer addressing for the desintation. + * The opposite of this is an incomplete rewrite that describes the ARP/ND + * rewrite when probing. + */ + ADJ_NBR_REWRITE_FLAG_COMPLETE = (1 << 0), +} adj_nbr_rewrite_flag_t; + +/** + * @brief + * Update the rewrite string for an existing adjacecny. + * + * @param + * The index of the adj to update + * + * @param + * The new rewrite + */ +extern void adj_nbr_update_rewrite(adj_index_t adj_index, + adj_nbr_rewrite_flag_t flags, + u8 *rewrite); + +/** + * @brief + * Format aa incomplete neigbour (ARP) adjacency + */ +extern u8* format_adj_nbr_incomplete(u8* s, va_list *ap); + +/** + * @brief + * Format a neigbour (REWRITE) adjacency + */ +extern u8* format_adj_nbr(u8* s, va_list *ap); + +/** + * @brief Walk the neighbour Adjacencies on a given interface + */ +extern void adj_nbr_walk (u32 sw_if_index, + fib_protocol_t adj_nh_proto, + adj_walk_cb_t cb, + void *ctx); +/** + * @brief Walk the neighbour Adjacencies on a given interface with a given next-hop + */ +void +adj_nbr_walk_nh (u32 sw_if_index, + fib_protocol_t adj_nh_proto, + const ip46_address_t *nh, + adj_walk_cb_t cb, + void *ctx); + +/** + * @brief Walk adjacencies on a link with a given v4 next-hop. + * that is visit the adjacencies with different link types. + */ +void +adj_nbr_walk_nh4 (u32 sw_if_index, + const ip4_address_t *addr, + adj_walk_cb_t cb, + void *ctx); + +/** + * @brief Walk adjacencies on a link with a given v6 next-hop. + * that is visit the adjacencies with different link types. + */ +void +adj_nbr_walk_nh6 (u32 sw_if_index, + const ip6_address_t *addr, + adj_walk_cb_t cb, + void *ctx); + +/** + * @brief + * Module initialisation + */ +extern void adj_nbr_module_init(void); + +/** + * @brief + * Return the size of the adjacency database. for testing purposes + */ +extern u32 adj_nbr_db_size(void); + +#endif diff --git a/src/vnet/adj/adj_nsh.c b/src/vnet/adj/adj_nsh.c new file mode 100644 index 00000000..128570b0 --- /dev/null +++ b/src/vnet/adj/adj_nsh.c @@ -0,0 +1,211 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/vnet.h> +#include <vnet/adj/adj_nsh.h> +#include <vnet/ip/ip.h> + +nsh_main_dummy_t nsh_main_dummy; + +/** + * @brief Trace data for a NSH Midchain + */ +typedef struct adj_nsh_trace_t_ { + /** Adjacency index taken. */ + u32 adj_index; +} adj_nsh_trace_t; + +static u8 * +format_adj_nsh_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + adj_nsh_trace_t * t = va_arg (*args, adj_nsh_trace_t *); + + s = format (s, "adj-idx %d : %U", + t->adj_index, + format_ip_adjacency, t->adj_index, FORMAT_IP_ADJACENCY_NONE); + return s; +} + +typedef enum adj_nsh_rewrite_next_t_ +{ + ADJ_NSH_REWRITE_NEXT_DROP, +} adj_gpe_rewrite_next_t; + +always_inline uword +adj_nsh_rewrite_inline (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame, + int is_midchain) +{ + u32 * from = vlib_frame_vector_args (frame); + u32 n_left_from, n_left_to_next, * to_next, next_index; + u32 thread_index = vlib_get_thread_index(); + + n_left_from = frame->n_vectors; + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_left_from > 0 && n_left_to_next > 0) + { + ip_adjacency_t * adj0; + vlib_buffer_t * p0; + char *h0; + u32 pi0, rw_len0, adj_index0, next0 = 0; + u32 tx_sw_if_index0; + + pi0 = to_next[0] = from[0]; + from += 1; + n_left_from -= 1; + to_next += 1; + n_left_to_next -= 1; + + p0 = vlib_get_buffer (vm, pi0); + h0 = vlib_buffer_get_current (p0); + + adj_index0 = vnet_buffer (p0)->ip.adj_index[VLIB_TX]; + + /* We should never rewrite a pkt using the MISS adjacency */ + ASSERT(adj_index0); + + adj0 = adj_get (adj_index0); + + /* Guess we are only writing on simple IP4 header. */ + vnet_rewrite_one_header(adj0[0], h0, sizeof(ip4_header_t)); + + /* Update packet buffer attributes/set output interface. */ + rw_len0 = adj0[0].rewrite_header.data_bytes; + vnet_buffer(p0)->ip.save_rewrite_length = rw_len0; + + vlib_increment_combined_counter(&adjacency_counters, + thread_index, + adj_index0, + /* packet increment */ 0, + /* byte increment */ rw_len0); + + /* Check MTU of outgoing interface. */ + if (PREDICT_TRUE((vlib_buffer_length_in_chain (vm, p0) <= + adj0[0].rewrite_header.max_l3_packet_bytes))) + { + /* Don't adjust the buffer for ttl issue; icmp-error node wants + * to see the IP headerr */ + p0->current_data -= rw_len0; + p0->current_length += rw_len0; + tx_sw_if_index0 = adj0[0].rewrite_header.sw_if_index; + + if (is_midchain) + { + adj0->sub_type.midchain.fixup_func(vm, adj0, p0); + } + + vnet_buffer (p0)->sw_if_index[VLIB_TX] = tx_sw_if_index0; + + /* + * Follow the feature ARC. this will result eventually in + * the midchain-tx node + */ + vnet_feature_arc_start (nsh_main_dummy.output_feature_arc_index, + tx_sw_if_index0, &next0, p0); + } + else + { + /* can't fragment NSH */ + next0 = ADJ_NSH_REWRITE_NEXT_DROP; + } + + if (PREDICT_FALSE(p0->flags & VLIB_BUFFER_IS_TRACED)) + { + adj_nsh_trace_t *tr = vlib_add_trace (vm, node, + p0, sizeof (*tr)); + tr->adj_index = vnet_buffer(p0)->ip.adj_index[VLIB_TX]; + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + pi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + return frame->n_vectors; +} + +static uword +adj_nsh_rewrite (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + return adj_nsh_rewrite_inline (vm, node, frame, 0); +} + +static uword +adj_nsh_midchain (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + return adj_nsh_rewrite_inline (vm, node, frame, 1); +} + +VLIB_REGISTER_NODE (adj_nsh_rewrite_node) = { + .function = adj_nsh_rewrite, + .name = "adj-nsh-rewrite", + .vector_size = sizeof (u32), + + .format_trace = format_adj_nsh_trace, + + .n_next_nodes = 1, + .next_nodes = { + [ADJ_NSH_REWRITE_NEXT_DROP] = "error-drop", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (adj_nsh_rewrite_node, adj_nsh_rewrite) + +VLIB_REGISTER_NODE (adj_nsh_midchain_node) = { + .function = adj_nsh_midchain, + .name = "adj-nsh-midchain", + .vector_size = sizeof (u32), + + .format_trace = format_adj_nsh_trace, + + .n_next_nodes = 1, + .next_nodes = { + [ADJ_NSH_REWRITE_NEXT_DROP] = "error-drop", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (adj_nsh_midchain_node, adj_nsh_midchain) + +/* Built-in ip4 tx feature path definition */ +/* *INDENT-OFF* */ +VNET_FEATURE_ARC_INIT (nsh_output, static) = +{ + .arc_name = "nsh-output", + .start_nodes = VNET_FEATURES ("adj-nsh-midchain"), + .arc_index_ptr = &nsh_main_dummy.output_feature_arc_index, +}; + +VNET_FEATURE_INIT (nsh_tx_drop, static) = +{ + .arc_name = "nsh-output", + .node_name = "error-drop", + .runs_before = 0, /* not before any other features */ +}; +/* *INDENT-ON* */ diff --git a/src/vnet/adj/adj_nsh.h b/src/vnet/adj/adj_nsh.h new file mode 100644 index 00000000..5501fbb9 --- /dev/null +++ b/src/vnet/adj/adj_nsh.h @@ -0,0 +1,31 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __ADJ_NSH_H__ +#define __ADJ_NSH_H__ + +#include <vnet/adj/adj.h> + +extern vlib_node_registration_t adj_nsh_midchain_node; +extern vlib_node_registration_t adj_nsh_rewrite_node; + +typedef struct _nsh_main_dummy +{ + u8 output_feature_arc_index; +} nsh_main_dummy_t; + +extern nsh_main_dummy_t nsh_main_dummy; + +#endif diff --git a/src/vnet/adj/adj_types.h b/src/vnet/adj/adj_types.h new file mode 100644 index 00000000..cf90c084 --- /dev/null +++ b/src/vnet/adj/adj_types.h @@ -0,0 +1,53 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __ADJ_TYPES_H__ +#define __ADJ_TYPES_H__ + +#include <vnet/vnet.h> + +/** + * @brief An index for adjacencies. + * Alas 'C' is not typesafe enough to b0rk when a u32 is used instead of + * an adi_index_t. However, for us humans, we can glean much more intent + * from the declaration + * foo bar(adj_index_t t); + * than we can from + * foo bar(u32 t); + */ +typedef u32 adj_index_t; + +/** + * @brief Invalid ADJ index - used when no adj is known + * likewise blazoned capitals INVALID speak volumes where ~0 does not. + */ +#define ADJ_INDEX_INVALID ((u32)~0) + +/** + * @brief return codes from a adjacency walker callback function + */ +typedef enum adj_walk_rc_t_ +{ + ADJ_WALK_RC_STOP, + ADJ_WALK_RC_CONTINUE, +} adj_walk_rc_t; + +/** + * @brief Call back function when walking adjacencies + */ +typedef adj_walk_rc_t (*adj_walk_cb_t)(adj_index_t ai, + void *ctx); + +#endif diff --git a/src/vnet/adj/rewrite.c b/src/vnet/adj/rewrite.c new file mode 100644 index 00000000..47fb74df --- /dev/null +++ b/src/vnet/adj/rewrite.c @@ -0,0 +1,234 @@ +/* + * Copyright (c) 2015 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +/* + * rewrite.c: packet rewrite + * + * Copyright (c) 2008 Eliot Dresselhaus + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice shall be + * included in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +#include <vnet/vnet.h> +#include <vnet/ip/lookup.h> + +void +vnet_rewrite_copy_slow_path (vnet_rewrite_data_t * p0, + vnet_rewrite_data_t * rw0, + word n_left, uword most_likely_size) +{ + uword n_done = + round_pow2 (most_likely_size, sizeof (rw0[0])) / sizeof (rw0[0]); + + p0 -= n_done; + rw0 -= n_done; + + /* As we enter the cleanup loop, p0 and rw0 point to the last chunk written + by the fast path. Hence, the constant 1, which the + vnet_rewrite_copy_one macro renders as p0[-1] = rw0[-1]. */ + + while (n_left > 0) + { + vnet_rewrite_copy_one (p0, rw0, 1); + p0--; + rw0--; + n_left--; + } +} + +u8 * +format_vnet_rewrite (u8 * s, va_list * args) +{ + vnet_rewrite_header_t *rw = va_arg (*args, vnet_rewrite_header_t *); + u32 max_data_bytes = va_arg (*args, u32); + CLIB_UNUSED (uword indent) = va_arg (*args, u32); + vnet_main_t *vnm = vnet_get_main (); + + if (rw->sw_if_index != ~0) + { + vnet_sw_interface_t *si; + si = vnet_get_sw_interface_safe (vnm, rw->sw_if_index); + if (NULL != si) + s = format (s, "%U: ", format_vnet_sw_interface_name, vnm, si); + else + s = format (s, "DELETED:%d", rw->sw_if_index); + } + + /* Format rewrite string. */ + if (rw->data_bytes > 0) + + s = format (s, "%U", + format_hex_bytes, + rw->data + max_data_bytes - rw->data_bytes, rw->data_bytes); + + return s; +} + +u32 +vnet_tx_node_index_for_sw_interface (vnet_main_t * vnm, u32 sw_if_index) +{ + vnet_hw_interface_t *hw = vnet_get_sup_hw_interface (vnm, sw_if_index); + return (hw->output_node_index); +} + +void +vnet_rewrite_init (vnet_main_t * vnm, + u32 sw_if_index, + u32 this_node, u32 next_node, vnet_rewrite_header_t * rw) +{ + rw->sw_if_index = sw_if_index; + rw->next_index = vlib_node_add_next (vnm->vlib_main, this_node, next_node); + rw->max_l3_packet_bytes = + vnet_sw_interface_get_mtu (vnm, sw_if_index, VLIB_TX); +} + +void +vnet_rewrite_for_sw_interface (vnet_main_t * vnm, + vnet_link_t link_type, + u32 sw_if_index, + u32 node_index, + void *dst_address, + vnet_rewrite_header_t * rw, + u32 max_rewrite_bytes) +{ + + vnet_hw_interface_t *hw = vnet_get_sup_hw_interface (vnm, sw_if_index); + vnet_hw_interface_class_t *hc = + vnet_get_hw_interface_class (vnm, hw->hw_class_index); + u8 *rewrite = NULL; + + vnet_rewrite_init (vnm, sw_if_index, node_index, + vnet_tx_node_index_for_sw_interface (vnm, sw_if_index), + rw); + + ASSERT (hc->build_rewrite); + rewrite = hc->build_rewrite (vnm, sw_if_index, link_type, dst_address); + + ASSERT (vec_len (rewrite) < max_rewrite_bytes); + vnet_rewrite_set_data_internal (rw, max_rewrite_bytes, rewrite, + vec_len (rewrite)); + vec_free (rewrite); +} + +void +vnet_rewrite_for_tunnel (vnet_main_t * vnm, + u32 tx_sw_if_index, + u32 rewrite_node_index, + u32 post_rewrite_node_index, + vnet_rewrite_header_t * rw, + u8 * rewrite_data, u32 rewrite_length) +{ + ip_adjacency_t *adj = 0; + /* + * Installed into vnet_buffer(b)->sw_if_index[VLIB_TX] e.g. + * by ip4_rewrite_inline. If the post-rewrite node injects into + * ipX-forward, this will be interpreted as a FIB number. + */ + rw->sw_if_index = tx_sw_if_index; + rw->next_index = vlib_node_add_next (vnm->vlib_main, rewrite_node_index, + post_rewrite_node_index); + rw->max_l3_packet_bytes = (u16) ~ 0; /* we can't know at this point */ + + ASSERT (rewrite_length < sizeof (adj->rewrite_data)); + /* Leave room for ethernet + VLAN tag */ + vnet_rewrite_set_data_internal (rw, sizeof (adj->rewrite_data), + rewrite_data, rewrite_length); +} + +void +serialize_vnet_rewrite (serialize_main_t * m, va_list * va) +{ + vnet_rewrite_header_t *rw = va_arg (*va, vnet_rewrite_header_t *); + u32 max_data_bytes = va_arg (*va, u32); + u8 *p; + + serialize_integer (m, rw->sw_if_index, sizeof (rw->sw_if_index)); + serialize_integer (m, rw->data_bytes, sizeof (rw->data_bytes)); + serialize_integer (m, rw->max_l3_packet_bytes, + sizeof (rw->max_l3_packet_bytes)); + p = serialize_get (m, rw->data_bytes); + clib_memcpy (p, vnet_rewrite_get_data_internal (rw, max_data_bytes), + rw->data_bytes); +} + +void +unserialize_vnet_rewrite (serialize_main_t * m, va_list * va) +{ + vnet_rewrite_header_t *rw = va_arg (*va, vnet_rewrite_header_t *); + u32 max_data_bytes = va_arg (*va, u32); + u8 *p; + + /* It is up to user to fill these in. */ + rw->next_index = ~0; + + unserialize_integer (m, &rw->sw_if_index, sizeof (rw->sw_if_index)); + unserialize_integer (m, &rw->data_bytes, sizeof (rw->data_bytes)); + unserialize_integer (m, &rw->max_l3_packet_bytes, + sizeof (rw->max_l3_packet_bytes)); + p = unserialize_get (m, rw->data_bytes); + clib_memcpy (vnet_rewrite_get_data_internal (rw, max_data_bytes), p, + rw->data_bytes); +} + +u8 * +vnet_build_rewrite_for_sw_interface (vnet_main_t * vnm, + u32 sw_if_index, + vnet_link_t link_type, + const void *dst_address) +{ + vnet_hw_interface_t *hw = vnet_get_sup_hw_interface (vnm, sw_if_index); + vnet_hw_interface_class_t *hc = + vnet_get_hw_interface_class (vnm, hw->hw_class_index); + + ASSERT (hc->build_rewrite); + return (hc->build_rewrite (vnm, sw_if_index, link_type, dst_address)); +} + + +void +vnet_update_adjacency_for_sw_interface (vnet_main_t * vnm, + u32 sw_if_index, u32 ai) +{ + vnet_hw_interface_t *hw = vnet_get_sup_hw_interface (vnm, sw_if_index); + vnet_hw_interface_class_t *hc = + vnet_get_hw_interface_class (vnm, hw->hw_class_index); + + ASSERT (hc->update_adjacency); + hc->update_adjacency (vnm, sw_if_index, ai); +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/adj/rewrite.h b/src/vnet/adj/rewrite.h new file mode 100644 index 00000000..1dea72f5 --- /dev/null +++ b/src/vnet/adj/rewrite.h @@ -0,0 +1,350 @@ +/* + * Copyright (c) 2015 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +/* + * rewrite.h: packet rewrite + * + * Copyright (c) 2008 Eliot Dresselhaus + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice shall be + * included in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +#ifndef included_vnet_rewrite_h +#define included_vnet_rewrite_h + +#include <vlib/vlib.h> +#include <vnet/l3_types.h> + +/* Consider using vector types for speed? */ +typedef uword vnet_rewrite_data_t; + +/** + * Flags associated with the rewrite/adjacency + */ +typedef enum vnet_rewrite_flags_t_ +{ + /** + * This adjacency/interface has output features configured + */ + VNET_REWRITE_HAS_FEATURES = (1 << 0), +} __attribute__ ((packed)) vnet_rewrite_flags_t; + +/* *INDENT-OFF* */ +typedef CLIB_PACKED (struct { + /* Interface to mark re-written packets with. */ + u32 sw_if_index; + + /* Next node to feed after packet rewrite is done. */ + u16 next_index; + + /* Number of bytes in rewrite data. */ + u16 data_bytes; + + /* Max packet size layer 3 (MTU) for output interface. + Used for MTU check after packet rewrite. */ + u16 max_l3_packet_bytes; + + /* Data-plane flags on the adjacency/rewrite */ + vnet_rewrite_flags_t flags; + + /* When dynamically writing a multicast destination L2 addresss + * this is the offset from the IP address at which to write in the + * IP->MAC address translation. + */ + u8 dst_mcast_offset; + + /* The mask to apply to the lower 4 bytes of the IP address before ORing + * into the destinaiton MAC address */ + u32 dst_mcast_mask; + + /* Rewrite string starting at end and going backwards. */ + u8 data[0]; +}) vnet_rewrite_header_t; +/* *INDENT-ON* */ + +/** + * At 16 bytes of rewrite herader we have enought space left for a IPv6 + * (40 bytes) + LISP-GPE (8 bytes) in the cache line + */ +STATIC_ASSERT (sizeof (vnet_rewrite_header_t) <= 16, + "Rewrite header too big"); + +/* + Helper macro for declaring rewrite string w/ given max-size. + + Typical usage: + typedef struct { + // + int a, b; + + // Total adjacency is 64 bytes. + vnet_rewrite_declare(64 - 2*sizeof(int)) rw; + } my_adjacency_t; +*/ +#define vnet_declare_rewrite(total_bytes) \ +struct { \ + vnet_rewrite_header_t rewrite_header; \ + \ + u8 rewrite_data[(total_bytes) - sizeof (vnet_rewrite_header_t)]; \ +} + +always_inline void +vnet_rewrite_clear_data_internal (vnet_rewrite_header_t * rw, int max_size) +{ + /* Sanity check values carefully for this memset operation */ + ASSERT ((max_size > 0) && (max_size < VLIB_BUFFER_PRE_DATA_SIZE)); + + rw->data_bytes = 0; + memset (rw->data, 0xfe, max_size); +} + +always_inline void +vnet_rewrite_set_data_internal (vnet_rewrite_header_t * rw, + int max_size, void *data, int data_bytes) +{ + /* Sanity check values carefully for this memset operation */ + ASSERT ((max_size > 0) && (max_size < VLIB_BUFFER_PRE_DATA_SIZE)); + ASSERT ((data_bytes >= 0) && (data_bytes < max_size)); + + rw->data_bytes = data_bytes; + clib_memcpy (rw->data + max_size - data_bytes, data, data_bytes); + memset (rw->data, 0xfe, max_size - data_bytes); +} + +#define vnet_rewrite_set_data(rw,data,data_bytes) \ + vnet_rewrite_set_data_internal (&((rw).rewrite_header), \ + sizeof ((rw).rewrite_data), \ + (data), \ + (data_bytes)) + +always_inline void * +vnet_rewrite_get_data_internal (vnet_rewrite_header_t * rw, int max_size) +{ + ASSERT (rw->data_bytes <= max_size); + return rw->data + max_size - rw->data_bytes; +} + +#define vnet_rewrite_get_data(rw) \ + vnet_rewrite_get_data_internal (&((rw).rewrite_header), sizeof ((rw).rewrite_data)) + +always_inline void +vnet_rewrite_copy_one (vnet_rewrite_data_t * p0, vnet_rewrite_data_t * rw0, + int i) +{ + p0[-i] = rw0[-i]; +} + +void vnet_rewrite_copy_slow_path (vnet_rewrite_data_t * p0, + vnet_rewrite_data_t * rw0, + word n_left, uword most_likely_size); + +/* *INDENT-OFF* */ +typedef CLIB_PACKED (struct { + u64 a; + u32 b; + u16 c; +}) eh_copy_t; +/* *INDENT-ON* */ + +always_inline void +_vnet_rewrite_one_header (vnet_rewrite_header_t * h0, + void *packet0, int max_size, int most_likely_size) +{ + vnet_rewrite_data_t *p0 = packet0; + vnet_rewrite_data_t *rw0 = (vnet_rewrite_data_t *) (h0->data + max_size); + word n_left0; + + /* 0xfefe => poisoned adjacency => crash */ + ASSERT (h0->data_bytes != 0xfefe); + + if (PREDICT_TRUE (h0->data_bytes == sizeof (eh_copy_t))) + { + eh_copy_t *s, *d; + s = (eh_copy_t *) (h0->data + max_size - sizeof (eh_copy_t)); + d = (eh_copy_t *) (((u8 *) packet0) - sizeof (eh_copy_t)); + clib_memcpy (d, s, sizeof (eh_copy_t)); + return; + } + + +#define _(i) \ + do { \ + if (most_likely_size > ((i)-1)*sizeof (vnet_rewrite_data_t)) \ + vnet_rewrite_copy_one (p0, rw0, (i)); \ + } while (0) + + _(4); + _(3); + _(2); + _(1); + +#undef _ + + n_left0 = (int) + (((int) h0->data_bytes - most_likely_size) + (sizeof (rw0[0]) - 1)) + / (int) sizeof (rw0[0]); + if (PREDICT_FALSE (n_left0 > 0)) + vnet_rewrite_copy_slow_path (p0, rw0, n_left0, most_likely_size); +} + +always_inline void +_vnet_rewrite_two_headers (vnet_rewrite_header_t * h0, + vnet_rewrite_header_t * h1, + void *packet0, + void *packet1, int max_size, int most_likely_size) +{ + vnet_rewrite_data_t *p0 = packet0; + vnet_rewrite_data_t *p1 = packet1; + vnet_rewrite_data_t *rw0 = (vnet_rewrite_data_t *) (h0->data + max_size); + vnet_rewrite_data_t *rw1 = (vnet_rewrite_data_t *) (h1->data + max_size); + word n_left0, n_left1; + int slow_path; + + /* 0xfefe => poisoned adjacency => crash */ + ASSERT (h0->data_bytes != 0xfefe); + ASSERT (h1->data_bytes != 0xfefe); + + /* Arithmetic calculation: bytes0 == bytes1 == 14 */ + slow_path = h0->data_bytes ^ h1->data_bytes; + slow_path += h0->data_bytes ^ sizeof (eh_copy_t); + + if (PREDICT_TRUE (slow_path == 0)) + { + eh_copy_t *s0, *d0, *s1, *d1; + s0 = (eh_copy_t *) (h0->data + max_size - sizeof (eh_copy_t)); + d0 = (eh_copy_t *) (((u8 *) packet0) - sizeof (eh_copy_t)); + clib_memcpy (d0, s0, sizeof (eh_copy_t)); + s1 = (eh_copy_t *) (h1->data + max_size - sizeof (eh_copy_t)); + d1 = (eh_copy_t *) (((u8 *) packet1) - sizeof (eh_copy_t)); + clib_memcpy (d1, s1, sizeof (eh_copy_t)); + return; + } + +#define _(i) \ + do { \ + if (most_likely_size > ((i)-1)*sizeof (vnet_rewrite_data_t)) \ + { \ + vnet_rewrite_copy_one (p0, rw0, (i)); \ + vnet_rewrite_copy_one (p1, rw1, (i)); \ + } \ + } while (0) + + _(4); + _(3); + _(2); + _(1); + +#undef _ + + n_left0 = (int) + (((int) h0->data_bytes - most_likely_size) + (sizeof (rw0[0]) - 1)) + / (int) sizeof (rw0[0]); + n_left1 = (int) + (((int) h1->data_bytes - most_likely_size) + (sizeof (rw1[0]) - 1)) + / (int) sizeof (rw1[0]); + + if (PREDICT_FALSE (n_left0 > 0 || n_left1 > 0)) + { + vnet_rewrite_copy_slow_path (p0, rw0, n_left0, most_likely_size); + vnet_rewrite_copy_slow_path (p1, rw1, n_left1, most_likely_size); + } +} + +#define vnet_rewrite_one_header(rw0,p0,most_likely_size) \ + _vnet_rewrite_one_header (&((rw0).rewrite_header), (p0), \ + sizeof ((rw0).rewrite_data), \ + (most_likely_size)) + +#define vnet_rewrite_two_headers(rw0,rw1,p0,p1,most_likely_size) \ + _vnet_rewrite_two_headers (&((rw0).rewrite_header), &((rw1).rewrite_header), \ + (p0), (p1), \ + sizeof ((rw0).rewrite_data), \ + (most_likely_size)) + +always_inline void +_vnet_fixup_one_header (vnet_rewrite_header_t * h0, + u8 * addr, u32 addr_len, u8 * packet0) +{ + if (PREDICT_TRUE (h0->dst_mcast_mask)) + { + /* location to write to in the packet */ + u8 *p0 = packet0 - h0->dst_mcast_offset; + u32 *p1 = (u32 *) p0; + /* location to copy from in the L3 dest address */ + u32 *a0 = (u32 *) (addr + addr_len - sizeof (h0->dst_mcast_mask)); + + *p1 |= (*a0 & h0->dst_mcast_mask); + } +} + +#define vnet_fixup_one_header(rw0,addr,p0) \ + _vnet_fixup_one_header (&((rw0).rewrite_header), \ + (u8*)(addr), sizeof((*addr)), \ + (u8*)(p0)) + +#define VNET_REWRITE_FOR_SW_INTERFACE_ADDRESS_BROADCAST ((void *) 0) +/** Deprecated */ +void vnet_rewrite_for_sw_interface (struct vnet_main_t *vnm, + vnet_link_t packet_type, + u32 sw_if_index, + u32 node_index, + void *dst_address, + vnet_rewrite_header_t * rw, + u32 max_rewrite_bytes); + +u32 vnet_tx_node_index_for_sw_interface (struct vnet_main_t *vnm, + u32 sw_if_index); + +void vnet_rewrite_init (struct vnet_main_t *vnm, + u32 sw_if_index, + u32 this_node, + u32 next_node, vnet_rewrite_header_t * rw); + +u8 *vnet_build_rewrite_for_sw_interface (struct vnet_main_t *vnm, + u32 sw_if_index, + vnet_link_t packet_type, + const void *dst_address); +void vnet_update_adjacency_for_sw_interface (struct vnet_main_t *vnm, + u32 sw_if_index, u32 ai); + +format_function_t format_vnet_rewrite; + +serialize_function_t serialize_vnet_rewrite, unserialize_vnet_rewrite; + +#endif /* included_vnet_rewrite_h */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ |