diff options
Diffstat (limited to 'vnet/vnet/adj')
-rw-r--r-- | vnet/vnet/adj/adj.c | 343 | ||||
-rw-r--r-- | vnet/vnet/adj/adj.h | 100 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_alloc.c | 236 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_alloc.h | 53 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_glean.c | 246 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_glean.h | 56 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_internal.h | 97 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_midchain.c | 226 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_midchain.h | 71 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_nbr.c | 835 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_nbr.h | 116 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_rewrite.c | 52 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_rewrite.h | 49 | ||||
-rw-r--r-- | vnet/vnet/adj/adj_types.h | 38 |
14 files changed, 2518 insertions, 0 deletions
diff --git a/vnet/vnet/adj/adj.c b/vnet/vnet/adj/adj.c new file mode 100644 index 00000000000..b552fdb2bbc --- /dev/null +++ b/vnet/vnet/adj/adj.c @@ -0,0 +1,343 @@ +/* + * 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_alloc.h> +#include <vnet/adj/adj_internal.h> +#include <vnet/adj/adj_glean.h> +#include <vnet/adj/adj_midchain.h> +#include <vnet/fib/fib_node_list.h> + +/* + * Special Adj with index zero. we need to define this since the v4 mtrie + * assumes an index of 0 implies the ply is empty. therefore all 'real' + * adjs need a non-zero index. + */ +static ip_adjacency_t *special_v4_miss_adj_with_index_zero; + +/* Adjacency packet/byte counters indexed by adjacency index. */ +vlib_combined_counter_main_t adjacency_counters; + +always_inline void +adj_poison (ip_adjacency_t * adj) +{ + if (CLIB_DEBUG > 0) + { + u32 save_handle = adj->heap_handle;; + + memset (adj, 0xfe, sizeof (adj[0])); + + adj->heap_handle = save_handle; + } +} + +ip_adjacency_t * +adj_alloc (fib_protocol_t proto) +{ + ip_adjacency_t *adj; + + adj = aa_alloc(); + + adj_poison(adj); + + /* Make sure certain fields are always initialized. */ + /* Validate adjacency counters. */ + vlib_validate_combined_counter(&adjacency_counters, + adj->heap_handle); + + adj->rewrite_header.sw_if_index = ~0; + adj->mcast_group_index = ~0; + adj->saved_lookup_next_index = 0; + adj->n_adj = 1; + + fib_node_init(&adj->ia_node, + FIB_NODE_TYPE_ADJ); + adj->ia_nh_proto = proto; + + 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) +{ + vnet_main_t * vnm = va_arg (*args, vnet_main_t *); + u32 adj_index = va_arg (*args, u32); + format_ip_adjacency_flags_t fiaf = va_arg (*args, format_ip_adjacency_flags_t); + ip_adjacency_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_vnet_sw_interface_name, + vnm, + vnet_get_sw_interface(vnm, + adj->rewrite_header.sw_if_index)); + break; + + case IP_LOOKUP_NEXT_MIDCHAIN: + s = format (s, "%U", format_adj_midchain, adj_index, 2); + break; + default: + break; + } + s = format (s, " index:%d", adj_index); + + if (fiaf & FORMAT_IP_ADJACENCY_DETAIL) + { + s = format (s, " locks:%d", adj->ia_node.fn_locks); + s = format(s, "\nchildren:\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) +{ + ASSERT(0 == fib_node_list_get_size(adj->ia_node.fn_children)); + ADJ_DBG(adj, "last-lock-gone"); + + 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->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; + default: + /* + * type not stored in any DB from which we need to remove it + */ + break; + } + + fib_node_deinit(&adj->ia_node); + aa_free(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); + ASSERT(adj->heap_handle!=0); + + 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); + ASSERT(adj->heap_handle!=0); + + ADJ_DBG(adj, "unlock"); + ASSERT(adj); + ASSERT(adj->heap_handle!=0); + + 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); +} + +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(); + + /* + * 4 special adjs for v4 and v6 resp. + */ + aa_bootstrap(8); + special_v4_miss_adj_with_index_zero = adj_alloc(FIB_PROTOCOL_IP4); + + return (NULL); +} + +VLIB_INIT_FUNCTION (adj_module_init); + +/* + * DEPRECATED: DO NOT USE + * + * Create new block of given number of contiguous adjacencies. + */ +ip_adjacency_t * +ip_add_adjacency (ip_lookup_main_t * lm, + ip_adjacency_t * copy_adj, + u32 n_adj, + u32 * adj_index_return) +{ + ip_adjacency_t * adj; + u32 ai, i, handle; + + ASSERT(1==n_adj); + + adj = aa_alloc (); + handle = ai = adj->heap_handle; + + /* Validate adjacency counters. */ + vlib_validate_combined_counter (&adjacency_counters, ai + n_adj - 1); + + for (i = 0; i < n_adj; i++) + { + /* Make sure certain fields are always initialized. */ + adj[i].rewrite_header.sw_if_index = ~0; + adj[i].mcast_group_index = ~0; + adj[i].saved_lookup_next_index = 0; + + if (copy_adj) + adj[i] = copy_adj[i]; + + adj[i].heap_handle = handle; + adj[i].n_adj = n_adj; + + /* Zero possibly stale counters for re-used adjacencies. */ + vlib_zero_combined_counter (&adjacency_counters, ai + i); + } + + *adj_index_return = ai; + return adj; +} diff --git a/vnet/vnet/adj/adj.h b/vnet/vnet/adj/adj.h new file mode 100644 index 00000000000..3a1236497e1 --- /dev/null +++ b/vnet/vnet/adj/adj.h @@ -0,0 +1,100 @@ +/* + * 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. + * - rewrite: an adj with no key, but with a rewrite string. + * + * 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/ip/lookup.h> +#include <vnet/adj/adj_types.h> +#include <vnet/adj/adj_nbr.h> +#include <vnet/adj/adj_rewrite.h> +#include <vnet/adj/adj_glean.h> + +/** + * @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 + * The global adjacnecy heap. Exposed for fast/inline data-plane access + */ +extern ip_adjacency_t *adj_heap; + +/** + * @brief + * Adjacency packet counters + */ +extern vlib_combined_counter_main_t adjacency_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_heap, adj_index)); +} + +#endif diff --git a/vnet/vnet/adj/adj_alloc.c b/vnet/vnet/adj/adj_alloc.c new file mode 100644 index 00000000000..5cc8cf6ef04 --- /dev/null +++ b/vnet/vnet/adj/adj_alloc.c @@ -0,0 +1,236 @@ +/* + * 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_alloc.h> +#include <vnet/ip/ip.h> + +/* + * the single adj heap + */ +ip_adjacency_t *adj_heap; + +/* + * any operation which could cause the adj vector to be reallocated + * must have a worker thread barrier + */ +static inline int will_reallocate (ip_adjacency_t * adjs, u32 n) +{ + uword aligned_header_bytes, new_data_bytes; + uword data_bytes; + aa_header_t * ah = aa_header (adjs); + + if (adjs == 0) + return 1; + + data_bytes = (vec_len (adjs) + n) * sizeof (*adjs); + + aligned_header_bytes = vec_header_bytes (aa_aligned_header_bytes); + + new_data_bytes = data_bytes + aligned_header_bytes; + + ASSERT (clib_mem_is_heap_object (_vec_find(ah))); + + if (PREDICT_TRUE(new_data_bytes <= clib_mem_size (_vec_find(ah)))) + return 0; + + return 1; +} + +ip_adjacency_t * +aa_alloc (void) +{ + vlib_main_t * vm = &vlib_global_main; + aa_header_t * ah = aa_header (adj_heap); + ip_adjacency_t * adj_block; + u32 freelist_length; + int need_barrier_sync = 0; + u32 n = 1; + + ASSERT(os_get_cpu_number() == 0); + ASSERT (clib_mem_is_heap_object (_vec_find(ah))); + + /* If we don't have a freelist of size N, fresh allocation is required */ + if (vec_len (ah->free_indices_by_size) <= n) + { + if (will_reallocate (adj_heap, n)) + { + need_barrier_sync = 1; + vlib_worker_thread_barrier_sync (vm); + } + /* Workers wont look at the freelists... */ + vec_validate (ah->free_indices_by_size, n); + vec_add2_ha (adj_heap, adj_block, n, aa_aligned_header_bytes, + CLIB_CACHE_LINE_BYTES); + if (need_barrier_sync) + vlib_worker_thread_barrier_release (vm); + goto out; + } + /* See if we have a free adj block to dole out */ + if ((freelist_length = vec_len(ah->free_indices_by_size[n]))) + { + u32 index = ah->free_indices_by_size[n][freelist_length-1]; + + adj_block = &adj_heap[index]; + _vec_len(ah->free_indices_by_size[n]) -= 1; + goto out; + } + /* Allocate a new block of size N */ + if (will_reallocate (adj_heap, n)) + { + need_barrier_sync = 1; + vlib_worker_thread_barrier_sync (vm); + } + vec_add2_ha (adj_heap, adj_block, n, aa_aligned_header_bytes, + CLIB_CACHE_LINE_BYTES); + + if (need_barrier_sync) + vlib_worker_thread_barrier_release (vm); + + out: + memset (adj_block, 0, n * (sizeof(*adj_block))); + adj_block->heap_handle = adj_block - adj_heap; + adj_block->n_adj = n; + + /* + * the adj heap may have realloc'd. recache. + */ + ip4_main.lookup_main.adjacency_heap = adj_heap; + ip6_main.lookup_main.adjacency_heap = adj_heap; + + return (adj_block); +} + +void aa_free (ip_adjacency_t * adj) +{ + aa_header_t * ah = aa_header (adj_heap); + + ASSERT (adj_heap && adj && (adj->heap_handle < vec_len (adj_heap))); + ASSERT (adj->heap_handle != 0); + + vec_add1 (ah->free_indices_by_size[adj->n_adj], adj->heap_handle); + adj->heap_handle = 0; +} + +void aa_bootstrap (u32 n) +{ + ip_adjacency_t * adj_block; + aa_header_t * ah; + int i; + + vec_add2_ha (adj_heap, adj_block, n, aa_aligned_header_bytes, + CLIB_CACHE_LINE_BYTES); + + memset (adj_block, 0, n * sizeof(*adj_block)); + ah = aa_header (adj_heap); + memset (ah, 0, sizeof (*ah)); + + vec_validate (ah->free_indices_by_size, 1); + + for (i = 0 ; i < vec_len (adj_heap); i++) + { + adj_block->n_adj = 1; + adj_block->heap_handle = ~0; + /* Euchre the allocator into returning 0, 1, 2, etc. */ + vec_add1 (ah->free_indices_by_size[1], n - (i+1)); + } + + ip4_main.lookup_main.adjacency_heap = adj_heap; + ip6_main.lookup_main.adjacency_heap = adj_heap; +} + +u8 * format_adjacency_alloc (u8 * s, va_list * args) +{ + vnet_main_t * vnm = va_arg (*args, vnet_main_t *); + int verbose = va_arg (*args, int); + ip_adjacency_t * adj; + u32 inuse = 0, freed = 0; + u32 on_freelist = 0; + int i, j; + aa_header_t * ah = aa_header (adj_heap); + + for (i = 0; i < vec_len (adj_heap); i += adj->n_adj) + { + adj = adj_heap + i; + if ((i == 0) || adj->heap_handle) + inuse += adj->n_adj; + else + freed += adj->n_adj; + } + + for (i = 1; i < vec_len(ah->free_indices_by_size); i++) + { + for (j = 0; j < vec_len(ah->free_indices_by_size[i]); j++) + { + adj = adj_heap + ah->free_indices_by_size[i][j]; + ASSERT(adj->heap_handle == 0); + on_freelist += adj->n_adj; + } + } + + s = format (s, "adj_heap: %d total, %d in use, %d free, %d on freelists\n", + vec_len(adj_heap), inuse, freed, on_freelist); + if (verbose) + { + for (i = 0; i < vec_len (adj_heap); i += adj->n_adj) + { + adj = adj_heap + i; + if ((i == 0) || adj->heap_handle) + { + if (adj->n_adj > 1) + s = format (s, "[%d-%d] ", i, i+adj->n_adj-1); + else + s = format (s, "[%d] ", i); + + for (j = 0; j < adj->n_adj; j++) + { + if (j > 0) + s = format (s, " "); + + s = format(s, "%U\n", format_ip_adjacency, + vnm, i+j, FORMAT_IP_ADJACENCY_NONE); + } + } + } + } + return s; +} + +static clib_error_t * +show_adjacency_alloc_command_fn (vlib_main_t * vm, + unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + int verbose = 0; + vnet_main_t *vnm = vnet_get_main(); + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "verbose")) + verbose = 1; + else + return clib_error_return (0, "unknown input `%U'", + format_unformat_error, input); + } + + vlib_cli_output (vm, "%U", format_adjacency_alloc, vnm, verbose); + + return 0; +} + +VLIB_CLI_COMMAND (show_adjacency_alloc_command, static) = { + .path = "show adjacency alloc", + .short_help = "show adjacency alloc", + .function = show_adjacency_alloc_command_fn, +}; diff --git a/vnet/vnet/adj/adj_alloc.h b/vnet/vnet/adj/adj_alloc.h new file mode 100644 index 00000000000..7d1a3fb3133 --- /dev/null +++ b/vnet/vnet/adj/adj_alloc.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_alloc_h__ +#define __adj_alloc_h__ + +/** + * @brief + * Adjacency allocator: heap-like in that the code + * will dole out contiguous chunks of n items. In the interests of + * thread safety, we don't bother about coalescing free blocks of size r + * into free blocks of size s, where r < s. + * + * We include explicit references to worker thread barrier synchronization + * where necessary. + */ + +#include <vppinfra/vec.h> +#include <vlib/vlib.h> +#include <vnet/ip/lookup.h> + +typedef struct { + u32 ** free_indices_by_size; +} aa_header_t; + +#define aa_aligned_header_bytes \ + vec_aligned_header_bytes (sizeof (aa_header_t), sizeof (void *)) + +/* Pool header from user pointer */ +static inline aa_header_t * aa_header (void * v) +{ + return vec_aligned_header (v, sizeof (aa_header_t), sizeof (void *)); +} + +extern ip_adjacency_t *aa_alloc(void); +extern void aa_free (ip_adjacency_t * adj); +extern void aa_bootstrap (u32 n); + +format_function_t format_adj_allocation; + +#endif /* __adj_alloc_h__ */ diff --git a/vnet/vnet/adj/adj_glean.c b/vnet/vnet/adj/adj_glean.c new file mode 100644 index 00000000000..6eb6718e216 --- /dev/null +++ b/vnet/vnet/adj/adj_glean.c @@ -0,0 +1,246 @@ +/* + * 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_alloc.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->heap_handle; + + 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->heap_handle); + + return (adj->heap_handle); +} + +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->heap_handle, &bw_ctx); + } + + return (NULL); +} + +VNET_SW_INTERFACE_ADMIN_UP_DOWN_FUNCTION(adj_glean_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->heap_handle, &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, " glean: %U", + 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/vnet/vnet/adj/adj_glean.h b/vnet/vnet/adj/adj_glean.h new file mode 100644 index 00000000000..ce3534ecee6 --- /dev/null +++ b/vnet/vnet/adj/adj_glean.h @@ -0,0 +1,56 @@ +/* + * 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 + * Module initialisation + */ +extern void adj_glean_module_init(void); + +#endif diff --git a/vnet/vnet/adj/adj_internal.h b/vnet/vnet/adj/adj_internal.h new file mode 100644 index 00000000000..79042d1fd2a --- /dev/null +++ b/vnet/vnet/adj/adj_internal.h @@ -0,0 +1,97 @@ +/* + * 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/ip/ip.h> +#include <vnet/mpls/mpls.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->heap_handle, _adj, \ + ##_args); \ +} +#else +#define ADJ_DBG(_e, _fmt, _args...) +#endif + +static inline vlib_node_registration_t* +adj_get_rewrite_node (fib_link_t linkt) +{ + switch (linkt) { + case FIB_LINK_IP4: + return (&ip4_rewrite_node); + case FIB_LINK_IP6: + return (&ip6_rewrite_node); + case FIB_LINK_MPLS: + return (&mpls_output_node); + } + ASSERT(0); + return (NULL); +} + +static inline vnet_l3_packet_type_t +adj_fib_link_2_vnet (fib_link_t linkt) +{ + switch (linkt) + { + case FIB_LINK_IP4: + return (VNET_L3_PACKET_TYPE_IP4); + case FIB_LINK_IP6: + return (VNET_L3_PACKET_TYPE_IP6); + case FIB_LINK_MPLS: + return (VNET_L3_PACKET_TYPE_MPLS_UNICAST); + } + return (0); +} + +static inline vnet_l3_packet_type_t +adj_fib_proto_2_nd (fib_protocol_t fp) +{ + switch (fp) + { + case FIB_PROTOCOL_IP4: + return (VNET_L3_PACKET_TYPE_ARP); + case FIB_PROTOCOL_IP6: + return (VNET_L3_PACKET_TYPE_IP6); + case FIB_PROTOCOL_MPLS: + return (VNET_L3_PACKET_TYPE_MPLS_UNICAST); + } + return (0); +} + +extern ip_adjacency_t * adj_alloc(fib_protocol_t proto); + +extern void adj_nbr_remove(fib_protocol_t nh_proto, + fib_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); + +#endif diff --git a/vnet/vnet/adj/adj_midchain.c b/vnet/vnet/adj/adj_midchain.c new file mode 100644 index 00000000000..4b9b6a414d2 --- /dev/null +++ b/vnet/vnet/adj/adj_midchain.c @@ -0,0 +1,226 @@ +/* + * 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/dpo/drop_dpo.h> +#include <vnet/fib/fib_walk.h> + +static inline u32 +adj_get_midchain_node (fib_link_t link) +{ + switch (link) { + case FIB_LINK_IP4: + return (ip4_midchain_node.index); + case FIB_LINK_IP6: + return (ip6_midchain_node.index); + case FIB_LINK_MPLS: + return (mpls_midchain_node.index); + } + ASSERT(0); + return (0); +} + +/** + * 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, + u32 post_rewrite_node, + u8 *rewrite) +{ + ip_adjacency_t *adj; + + ASSERT(ADJ_INDEX_INVALID != adj_index); + + adj = adj_get(adj_index); + adj->lookup_next_index = IP_LOOKUP_NEXT_MIDCHAIN; + adj->sub_type.midchain.tx_function_node = post_rewrite_node; + + if (NULL != rewrite) + { + /* + * new rewrite provided. + * use a dummy rewrite header to get the interface to print into. + */ + ip_adjacency_t dummy; + dpo_id_t tmp = DPO_NULL; + + vnet_rewrite_for_tunnel(vnet_get_main(), + adj->rewrite_header.sw_if_index, + adj_get_midchain_node(adj->ia_link), + adj->sub_type.midchain.tx_function_node, + &dummy.rewrite_header, + rewrite, + vec_len(rewrite)); + + /* + * this is an update of an existing rewrite. + * packets are in flight. we'll need to briefly stack on the drop DPO + * whilst the rewrite is written, so any packets that see the partial update + * are binned. + */ + if (!dpo_id_is_valid(&adj->sub_type.midchain.next_dpo)) + { + /* + * not stacked yet. stack on the drop + */ + dpo_stack(DPO_ADJACENCY_MIDCHAIN, + fib_proto_to_dpo(adj->ia_nh_proto), + &adj->sub_type.midchain.next_dpo, + drop_dpo_get(fib_proto_to_dpo(adj->ia_nh_proto))); + } + + dpo_copy(&tmp, &adj->sub_type.midchain.next_dpo); + dpo_stack(DPO_ADJACENCY_MIDCHAIN, + fib_proto_to_dpo(adj->ia_nh_proto), + &adj->sub_type.midchain.next_dpo, + drop_dpo_get(fib_proto_to_dpo(adj->ia_nh_proto))); + + CLIB_MEMORY_BARRIER(); + + clib_memcpy(&adj->rewrite_header, + &dummy.rewrite_header, + VLIB_BUFFER_PRE_DATA_SIZE); + + CLIB_MEMORY_BARRIER(); + + /* + * The graph arc used/created here is from the post-rewirte node to the + * child's registered node. This is because post adj processing the next + * node is the interface's specific node, then the post-write-node (aka + * the interface's tx-function) - from there we need to get to the stacked + * child's node. + */ + dpo_stack_from_node(adj->sub_type.midchain.tx_function_node, + &adj->sub_type.midchain.next_dpo, + &tmp); + dpo_reset(&tmp); + } + else + { + ASSERT(0); + } + + /* + * time for walkies fido. + */ + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = FIB_NODE_BW_REASON_ADJ_UPDATE, + }; + + fib_walk_sync(FIB_NODE_TYPE_ADJ, adj->heap_handle, &bw_ctx); +} + +/** + * 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); + + dpo_stack_from_node(adj->sub_type.midchain.tx_function_node, + &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); + vnet_main_t * vnm = vnet_get_main(); + ip_adjacency_t * adj = adj_get(index); + + s = format (s, "%U", format_fib_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, + vnm->vlib_main, &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 * const midchain_nodes[DPO_PROTO_NUM] = +{ + [DPO_PROTO_IP4] = midchain_ip4_nodes, + [DPO_PROTO_IP6] = midchain_ip6_nodes, + [DPO_PROTO_MPLS] = midchain_mpls_nodes, +}; + +void +adj_midchain_module_init (void) +{ + dpo_register(DPO_ADJACENCY_MIDCHAIN, &adj_midchain_dpo_vft, midchain_nodes); +} diff --git a/vnet/vnet/adj/adj_midchain.h b/vnet/vnet/adj/adj_midchain.h new file mode 100644 index 00000000000..adf86f1d007 --- /dev/null +++ b/vnet/vnet/adj/adj_midchain.h @@ -0,0 +1,71 @@ +/* + * 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, + u32 post_rewrite_node, + 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 + * 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/vnet/vnet/adj/adj_nbr.c b/vnet/vnet/adj/adj_nbr.c new file mode 100644 index 00000000000..7da1becd4c1 --- /dev/null +++ b/vnet/vnet/adj/adj_nbr.c @@ -0,0 +1,835 @@ +/* + * 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, + fib_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 (fib_protocol_t nh_proto, + fib_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); + + 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, + fib_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 vlib_node_registration_t* +adj_get_nd_node (fib_protocol_t proto) +{ + switch (proto) { + case FIB_PROTOCOL_IP4: + return (&ip4_arp_node); + case FIB_PROTOCOL_IP6: + return (&ip6_discover_neighbor_node); + case FIB_PROTOCOL_MPLS: + break; + } + ASSERT(0); + return (NULL); +} + +static void +adj_ip4_nbr_probe (ip_adjacency_t *adj) +{ + vnet_main_t * vnm = vnet_get_main(); + ip4_main_t * im = &ip4_main; + ip_interface_address_t * ia; + ethernet_arp_header_t * h; + vnet_hw_interface_t * hi; + vnet_sw_interface_t * si; + ip4_address_t * src; + vlib_buffer_t * b; + vlib_main_t * vm; + u32 bi = 0; + + vm = vlib_get_main(); + + si = vnet_get_sw_interface (vnm, + adj->rewrite_header.sw_if_index); + + if (!(si->flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP)) + { + return; + } + + src = + ip4_interface_address_matching_destination(im, + &adj->sub_type.nbr.next_hop.ip4, + adj->rewrite_header.sw_if_index, + &ia); + if (! src) + { + return; + } + + h = vlib_packet_template_get_packet (vm, &im->ip4_arp_request_packet_template, &bi); + + hi = vnet_get_sup_hw_interface (vnm, adj->rewrite_header.sw_if_index); + + clib_memcpy (h->ip4_over_ethernet[0].ethernet, + hi->hw_address, + sizeof (h->ip4_over_ethernet[0].ethernet)); + + h->ip4_over_ethernet[0].ip4 = src[0]; + h->ip4_over_ethernet[1].ip4 = adj->sub_type.nbr.next_hop.ip4; + + b = vlib_get_buffer (vm, bi); + vnet_buffer (b)->sw_if_index[VLIB_RX] = + vnet_buffer (b)->sw_if_index[VLIB_TX] = + adj->rewrite_header.sw_if_index; + + /* Add encapsulation string for software interface (e.g. ethernet header). */ + vnet_rewrite_one_header (adj[0], h, sizeof (ethernet_header_t)); + vlib_buffer_advance (b, -adj->rewrite_header.data_bytes); + + { + vlib_frame_t * f = vlib_get_frame_to_node (vm, hi->output_node_index); + u32 * to_next = vlib_frame_vector_args (f); + to_next[0] = bi; + f->n_vectors = 1; + vlib_put_frame_to_node (vm, hi->output_node_index, f); + } +} + +static void +adj_ip6_nbr_probe (ip_adjacency_t *adj) +{ + icmp6_neighbor_solicitation_header_t * h; + vnet_main_t * vnm = vnet_get_main(); + ip6_main_t * im = &ip6_main; + ip_interface_address_t * ia; + ip6_address_t * dst, *src; + vnet_hw_interface_t * hi; + vnet_sw_interface_t * si; + vlib_buffer_t * b; + int bogus_length; + vlib_main_t * vm; + u32 bi = 0; + + vm = vlib_get_main(); + + si = vnet_get_sw_interface(vnm, adj->rewrite_header.sw_if_index); + dst = &adj->sub_type.nbr.next_hop.ip6; + + if (!(si->flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP)) + { + return; + } + src = ip6_interface_address_matching_destination(im, dst, + adj->rewrite_header.sw_if_index, + &ia); + if (! src) + { + return; + } + + h = vlib_packet_template_get_packet(vm, + &im->discover_neighbor_packet_template, + &bi); + + hi = vnet_get_sup_hw_interface(vnm, adj->rewrite_header.sw_if_index); + + h->ip.dst_address.as_u8[13] = dst->as_u8[13]; + h->ip.dst_address.as_u8[14] = dst->as_u8[14]; + h->ip.dst_address.as_u8[15] = dst->as_u8[15]; + h->ip.src_address = src[0]; + h->neighbor.target_address = dst[0]; + + clib_memcpy (h->link_layer_option.ethernet_address, + hi->hw_address, + vec_len(hi->hw_address)); + + h->neighbor.icmp.checksum = + ip6_tcp_udp_icmp_compute_checksum(vm, 0, &h->ip, &bogus_length); + ASSERT(bogus_length == 0); + + b = vlib_get_buffer (vm, bi); + vnet_buffer (b)->sw_if_index[VLIB_RX] = + vnet_buffer (b)->sw_if_index[VLIB_TX] = + adj->rewrite_header.sw_if_index; + + /* Add encapsulation string for software interface (e.g. ethernet header). */ + vnet_rewrite_one_header(adj[0], h, sizeof (ethernet_header_t)); + vlib_buffer_advance(b, -adj->rewrite_header.data_bytes); + + { + vlib_frame_t * f = vlib_get_frame_to_node(vm, hi->output_node_index); + u32 * to_next = vlib_frame_vector_args(f); + to_next[0] = bi; + f->n_vectors = 1; + vlib_put_frame_to_node(vm, hi->output_node_index, f); + } +} + +static ip_adjacency_t* +adj_nbr_alloc (fib_protocol_t nh_proto, + fib_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->heap_handle); + + /* + * 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; + memset(&adj->sub_type.midchain.next_dpo, 0, + sizeof(adj->sub_type.midchain.next_dpo)); + + return (adj); +} + +/* + * adj_add_for_nbr + * + * 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 + * - fib_index; this is broken. i will fix it. + * the adj lookup currently occurs in the FIB. + */ +adj_index_t +adj_nbr_add_or_lock (fib_protocol_t nh_proto, + fib_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) + { + adj = adj_nbr_alloc(nh_proto, link_type, nh_addr, sw_if_index); + + /* + * If there is no next-hop, this is the 'auto-adj' used on p2p + * links instead of a glean. + */ + if (ip46_address_is_zero(nh_addr)) + { + adj->lookup_next_index = IP_LOOKUP_NEXT_REWRITE; + + vnet_rewrite_for_sw_interface(vnet_get_main(), + adj_fib_link_2_vnet(link_type), + sw_if_index, + adj_get_rewrite_node(link_type)->index, + VNET_REWRITE_FOR_SW_INTERFACE_ADDRESS_BROADCAST, + &adj->rewrite_header, + sizeof (adj->rewrite_data)); + } + else + { + vnet_rewrite_for_sw_interface(vnet_get_main(), + adj_fib_proto_2_nd(nh_proto), + sw_if_index, + adj_get_nd_node(nh_proto)->index, + VNET_REWRITE_FOR_SW_INTERFACE_ADDRESS_BROADCAST, + &adj->rewrite_header, + sizeof (adj->rewrite_data)); + + switch (nh_proto) + { + case FIB_PROTOCOL_IP4: + adj_ip4_nbr_probe(adj); + break; + case FIB_PROTOCOL_IP6: + adj_ip6_nbr_probe(adj); + break; + case FIB_PROTOCOL_MPLS: + break; + } + } + } + else + { + adj = adj_get(adj_index); + } + + adj_lock(adj->heap_handle); + + return (adj->heap_handle); +} + +adj_index_t +adj_nbr_add_or_lock_w_rewrite (fib_protocol_t nh_proto, + fib_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->heap_handle); + adj_nbr_update_rewrite(adj->heap_handle, rewrite); + + return (adj->heap_handle); +} + +/** + * 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, + u8 *rewrite) +{ + ip_adjacency_t *adj; + + ASSERT(ADJ_INDEX_INVALID != adj_index); + + adj = adj_get(adj_index); + + if (NULL != rewrite) + { + /* + * new rewrite provided. + * use a dummy rewrite header to get the interface to print into. + */ + ip_adjacency_t dummy; + + vnet_rewrite_for_sw_interface(vnet_get_main(), + adj_fib_link_2_vnet(adj->ia_link), + adj->rewrite_header.sw_if_index, + adj_get_rewrite_node(adj->ia_link)->index, + rewrite, + &dummy.rewrite_header, + sizeof (dummy.rewrite_data)); + + if (IP_LOOKUP_NEXT_REWRITE == adj->lookup_next_index) + { + /* + * this is an update of an existing rewrite. + * we can't just paste in the new rewrite as that is not atomic. + * So we briefly swap the ADJ to ARP type, paste, then swap back. + */ + adj->lookup_next_index = IP_LOOKUP_NEXT_ARP; + CLIB_MEMORY_BARRIER(); + } + /* + * else + * this is the first time the rewrite is added. + * paste it on then swap the next type. + */ + clib_memcpy(&adj->rewrite_header, + &dummy.rewrite_header, + VLIB_BUFFER_PRE_DATA_SIZE); + + adj->lookup_next_index = IP_LOOKUP_NEXT_REWRITE; + } + else + { + /* + * clear the rewrite. + */ + adj->lookup_next_index = IP_LOOKUP_NEXT_ARP; + CLIB_MEMORY_BARRIER(); + + adj->rewrite_header.data_bytes = 0; + } + + /* + * time for walkies fido. + * The link type MPLS Adj never has children. So if it is this adj + * that is updated, we need to walk from its IP sibling. + */ + if (FIB_LINK_MPLS == adj->ia_link) + { + adj_index = 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); + + ASSERT(ADJ_INDEX_INVALID != adj_index); + } + + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = FIB_NODE_BW_REASON_FLAG_ADJ_UPDATE, + /* + * This walk only needs to go back one level, but there is no control here. + * the first receiving fib_entry_t will quash the walk + */ + }; + + fib_walk_sync(FIB_NODE_TYPE_ADJ, adj_index, &bw_ctx); +} + +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); +} + +/** + * Context for the state change walk of the DB + */ +typedef struct adj_nbr_interface_state_change_ctx_t_ +{ + /** + * Flags passed from the vnet notifiy function + */ + int flags; +} adj_nbr_interface_state_change_ctx_t; + +static void +adj_nbr_interface_state_change_one (BVT(clib_bihash_kv) * kvp, + void *arg) +{ + /* + * Back walk the graph to inform the forwarding entries + * that this interface state has changed. + */ + adj_nbr_interface_state_change_ctx_t *ctx = arg; + + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = (ctx->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, kvp->value, &bw_ctx); +} + +static clib_error_t * +adj_nbr_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++) + { + if (!ADJ_NBR_ITF_OK(proto, sw_if_index)) + continue; + + adj_nbr_interface_state_change_ctx_t ctx = { + .flags = flags, + }; + + BV(clib_bihash_foreach_key_value_pair) ( + adj_nbr_tables[proto][sw_if_index], + adj_nbr_interface_state_change_one, + &ctx); + } + + return (NULL); +} + +VNET_SW_INTERFACE_ADMIN_UP_DOWN_FUNCTION(adj_nbr_interface_state_change); + +static void +adj_nbr_interface_delete_one (BVT(clib_bihash_kv) * kvp, + 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, kvp->value, &bw_ctx); +} + +/** + * 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++) + { + if (!ADJ_NBR_ITF_OK(proto, sw_if_index)) + continue; + + BV(clib_bihash_foreach_key_value_pair) ( + adj_nbr_tables[proto][sw_if_index], + adj_nbr_interface_delete_one, + NULL); + } + + return (NULL); + +} + +VNET_SW_INTERFACE_ADD_DEL_FUNCTION(adj_nbr_interface_add_del); + + +static void +adj_nbr_show_one (BVT(clib_bihash_kv) * kvp, + void *arg) +{ + vlib_cli_output (arg, "[@%d] %U", + kvp->value, + format_ip_adjacency, + vnet_get_main(), kvp->value, + FORMAT_IP_ADJACENCY_NONE); +} + +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; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "%d", &ai)) + ; + else + break; + } + + if (ADJ_INDEX_INVALID != ai) + { + vlib_cli_output (vm, "[@%d] %U", + ai, + + format_ip_adjacency, + vnet_get_main(), ai, + FORMAT_IP_ADJACENCY_DETAIL); + } + else + { + fib_protocol_t proto; + + for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++) + { + u32 sw_if_index; + + vec_foreach_index(sw_if_index, adj_nbr_tables[proto]) + { + if (!ADJ_NBR_ITF_OK(proto, sw_if_index)) + continue; + + BV(clib_bihash_foreach_key_value_pair) ( + adj_nbr_tables[proto][sw_if_index], + adj_nbr_show_one, + vm); + } + } + } + + return 0; +} + +VLIB_CLI_COMMAND (ip4_show_fib_command, static) = { + .path = "show adj nbr", + .short_help = "show adj nbr [<adj_index>] [sw_if_index <index>]", + .function = adj_nbr_show, +}; + +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_fib_link, adj->ia_link); + s = format (s, ": via %U", + format_ip46_address, &adj->sub_type.nbr.next_hop); + 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); + vnet_main_t * vnm = vnet_get_main(); + ip_adjacency_t * adj = adj_get(index); + + s = format (s, "%U", format_fib_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, + vnm->vlib_main, &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); +} + +const static dpo_vft_t adj_nbr_dpo_vft = { + .dv_lock = adj_dpo_lock, + .dv_unlock = adj_dpo_unlock, + .dv_format = format_adj_nbr, +}; +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-transit", + 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 * const nbr_nodes[DPO_PROTO_NUM] = +{ + [DPO_PROTO_IP4] = nbr_ip4_nodes, + [DPO_PROTO_IP6] = nbr_ip6_nodes, + [DPO_PROTO_MPLS] = nbr_mpls_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/vnet/vnet/adj/adj_nbr.h b/vnet/vnet/adj/adj_nbr.h new file mode 100644 index 00000000000..331423bd036 --- /dev/null +++ b/vnet/vnet/adj/adj_nbr.h @@ -0,0 +1,116 @@ +/* + * 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, + fib_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, + fib_link_t link_type, + const ip46_address_t *nh_addr, + u32 sw_if_index, + u8 *rewrite); + +/** + * @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, + 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 + * 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/vnet/vnet/adj/adj_rewrite.c b/vnet/vnet/adj/adj_rewrite.c new file mode 100644 index 00000000000..db802e33665 --- /dev/null +++ b/vnet/vnet/adj/adj_rewrite.c @@ -0,0 +1,52 @@ +/* + * 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_alloc.h> +#include <vnet/adj/adj_internal.h> + +/** + * adj_rewrite_add_and_lock + * + * A rewrite sub-type has the rewrite string provided, but no key + */ +adj_index_t +adj_rewrite_add_and_lock (fib_protocol_t nh_proto, + fib_link_t link_type, + u32 sw_if_index, + u8 *rewrite) +{ + ip_adjacency_t *adj; + + adj = adj_alloc(nh_proto); + + adj->lookup_next_index = IP_LOOKUP_NEXT_REWRITE; + adj->ia_link = link_type; + adj->rewrite_header.sw_if_index = sw_if_index; + + ASSERT(NULL != rewrite); + + vnet_rewrite_for_sw_interface(vnet_get_main(), + adj_fib_link_2_vnet(link_type), + adj->rewrite_header.sw_if_index, + adj_get_rewrite_node(link_type)->index, + rewrite, + &adj->rewrite_header, + sizeof (adj->rewrite_data)); + + adj_lock(adj->heap_handle); + + return (adj->heap_handle); +} diff --git a/vnet/vnet/adj/adj_rewrite.h b/vnet/vnet/adj/adj_rewrite.h new file mode 100644 index 00000000000..f8df255150d --- /dev/null +++ b/vnet/vnet/adj/adj_rewrite.h @@ -0,0 +1,49 @@ +/* + * 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 + * A rewrite adjacency has no key, and thus cannot be 'found' from the + * FIB resolution code. the client therefore needs to maange these adjacencies + */ + +#ifndef __ADJ_REWRITE_H__ +#define __ADJ_REWRITE_H__ + +#include <vnet/adj/adj_types.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 sw_if_index + * The interface on which the peer resides + * + * @param rewrite + * The rewrite to prepend to packets + */ +extern adj_index_t adj_rewrite_add_and_lock(fib_protocol_t nh_proto, + fib_link_t link_type, + u32 sw_if_index, + u8 *rewrite); + +#endif diff --git a/vnet/vnet/adj/adj_types.h b/vnet/vnet/adj/adj_types.h new file mode 100644 index 00000000000..a7234663d29 --- /dev/null +++ b/vnet/vnet/adj/adj_types.h @@ -0,0 +1,38 @@ +/* + * 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) + +#endif |