diff options
Diffstat (limited to 'src/vnet')
-rw-r--r-- | src/vnet/api_errno.h | 5 | ||||
-rw-r--r-- | src/vnet/l2/l2.api | 65 | ||||
-rw-r--r-- | src/vnet/l2/l2_api.c | 83 | ||||
-rw-r--r-- | src/vnet/l2/l2_bd.c | 37 | ||||
-rw-r--r-- | src/vnet/l2/l2_fib.c | 394 | ||||
-rw-r--r-- | src/vnet/l2/l2_fib.h | 37 | ||||
-rw-r--r-- | src/vnet/l2/l2_fwd.c | 3 | ||||
-rw-r--r-- | src/vnet/l2/l2_learn.c | 30 | ||||
-rw-r--r-- | src/vnet/l2/l2_learn.h | 5 |
9 files changed, 497 insertions, 162 deletions
diff --git a/src/vnet/api_errno.h b/src/vnet/api_errno.h index 747c65e7fb5..22522f34be9 100644 --- a/src/vnet/api_errno.h +++ b/src/vnet/api_errno.h @@ -112,8 +112,9 @@ _(BD_ALREADY_EXISTS, -119, "Bridge domain already exists") \ _(BD_IN_USE, -120, "Bridge domain has member interfaces") \ _(BD_NOT_MODIFIABLE, -121, "Bridge domain 0 can't be deleted/modified") \ _(BD_ID_EXCEED_MAX, -122, "Bridge domain ID exceed 16M limit") \ -_(UNSUPPORTED, -123, "Unsupported") \ -_(SUBIF_DOESNT_EXIST, -124, "Subinterface doesn't exist") +_(SUBIF_DOESNT_EXIST, -123, "Subinterface doesn't exist") \ +_(L2_MACS_EVENT_CLINET_PRESENT, -124, "Client already exist for L2 MACs events") \ +_(UNSUPPORTED, -125, "Unsupported") typedef enum { diff --git a/src/vnet/l2/l2.api b/src/vnet/l2/l2.api index bb3990c6fa4..e508bfb529c 100644 --- a/src/vnet/l2/l2.api +++ b/src/vnet/l2/l2.api @@ -133,12 +133,64 @@ autoreply define l2fib_add_del u8 bvi_mac; }; -/** \brief Set L2 flags request !!! TODO - need more info, feature bits in l2_input.h +/** \brief Register to recive L2 MAC events for leanred and aged MAC + Will also change MAC learn limit to L2LEARN_INFORM_LIMIT + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param learn_limit - MAC learn limit, 0 => default to 1000 + @param scan_delay - event scan delay in 10 msec unit, 0 => default to 100 msec + @param max_macs_in_event - in units of 10 mac entries, 0 => default to 100 entries + @param enable_disable - 1 => register for MAC events, 0 => cancel registration + @param pid - sender's pid +*/ +autoreply define want_l2_macs_events +{ + u32 client_index; + u32 context; + u32 learn_limit; + u8 scan_delay; + u8 max_macs_in_event; + u8 enable_disable; + u32 pid; +}; + +/** \brief Entry for learned or aged MAC in L2 MAC Events + @param sw_if_index - sw_if_index in the domain + @param mac_addr - mac_address + @is_del - 0 => newly learned MAC, 1 => aged out MAC +*/ +typeonly define mac_entry +{ + u32 sw_if_index; + u8 mac_addr[6]; + u8 is_del; + u8 spare; +}; + +/** \brief L2 MAC event for a list of learned or aged MACs + @param client_index - opaque cookie to identify the sender + @param pid - client pid registered to receive notification + @param n_macs - number of learned/aged MAC enntries + @param mac - array of learned/aged MAC entries +*/ +define l2_macs_event +{ + u32 client_index; + u32 pid; + u32 n_macs; + vl_api_mac_entry_t mac[n_macs]; +}; + +/** \brief Set interface L2 flags (such as L2_LEARN, L2_FWD, + L2_FLOOD, L2_UU_FLOOD, or L2_ARP_TERM bits). This can be used + to disable one or more of the features represented by the + flag bits on an interface to override what is set as default + for all interfaces in the bridge domain @param client_index - opaque cookie to identify the sender @param context - sender context, to match reply w/ request @param sw_if_index - interface @param is_set - if non-zero, set the bits, else clear them - @param feature_bitmap - non-zero bits to set or clear + @param feature_bitmap - non-zero bits (as above) to set or clear */ define l2_flags { @@ -149,9 +201,10 @@ define l2_flags u32 feature_bitmap; }; -/** \brief Set L2 bits response +/** \brief Set interface L2 flags response @param context - sender context, to match reply w/ request @param retval - return code for the set l2 bits request + @param resulting_feature_bitmap - the internal l2 feature bitmap after the request is implemented */ define l2_flags_reply { @@ -250,12 +303,12 @@ manual_print manual_endian define bridge_domain_details }; /** \brief Set bridge flags (such as L2_LEARN, L2_FWD, L2_FLOOD, - L2_UU_FLOOD, or L2_ARP_TERM) request + L2_UU_FLOOD, or L2_ARP_TERM bits) request @param client_index - opaque cookie to identify the sender @param context - sender context, to match reply w/ request @param bd_id - the bridge domain to set the flags for @param is_set - if non-zero, set the flags, else clear them - @param feature_bitmap - bits that are non-zero to set or clear + @param feature_bitmap - bits (as above) that are non-zero to set or clear */ define bridge_flags { @@ -269,7 +322,7 @@ define bridge_flags /** \brief Set bridge flags response @param context - sender context, to match reply w/ request @param retval - return code for the set bridge flags request - @param resulting_feature_bitmap - the feature bitmap value after the request is implemented + @param resulting_feature_bitmap - the internal L2 feature bitmap after the request is implemented */ define bridge_flags_reply { diff --git a/src/vnet/l2/l2_api.c b/src/vnet/l2/l2_api.c index a0b40d6d840..c81cbad77d7 100644 --- a/src/vnet/l2/l2_api.c +++ b/src/vnet/l2/l2_api.c @@ -25,6 +25,7 @@ #include <vnet/l2/l2_input.h> #include <vnet/l2/l2_fib.h> #include <vnet/l2/l2_vtr.h> +#include <vnet/l2/l2_learn.h> #include <vnet/vnet_msg_enum.h> @@ -55,6 +56,7 @@ _(L2FIB_FLUSH_ALL, l2fib_flush_all) \ _(L2FIB_FLUSH_INT, l2fib_flush_int) \ _(L2FIB_FLUSH_BD, l2fib_flush_bd) \ _(L2FIB_ADD_DEL, l2fib_add_del) \ +_(WANT_L2_MACS_EVENTS, want_l2_macs_events) \ _(L2_FLAGS, l2_flags) \ _(BRIDGE_DOMAIN_ADD_DEL, bridge_domain_add_del) \ _(BRIDGE_DOMAIN_DUMP, bridge_domain_dump) \ @@ -221,8 +223,8 @@ vl_api_l2fib_add_del_t_handler (vl_api_l2fib_add_del_t * mp) goto bad_sw_if_index; } } - u32 static_mac = mp->static_mac ? 1 : 0; - u32 bvi_mac = mp->bvi_mac ? 1 : 0; + u8 static_mac = mp->static_mac ? 1 : 0; + u8 bvi_mac = mp->bvi_mac ? 1 : 0; l2fib_add_fwd_entry (mac, bd_index, sw_if_index, static_mac, bvi_mac); } @@ -238,6 +240,58 @@ vl_api_l2fib_add_del_t_handler (vl_api_l2fib_add_del_t * mp) } static void +vl_api_want_l2_macs_events_t_handler (vl_api_want_l2_macs_events_t * mp) +{ + int rv = 0; + vl_api_want_l2_macs_events_reply_t *rmp; + l2learn_main_t *lm = &l2learn_main; + l2fib_main_t *fm = &l2fib_main; + u32 pid = ntohl (mp->pid); + u32 learn_limit = ntohl (mp->learn_limit); + + if (mp->enable_disable) + { + if (lm->client_pid == 0) + { + lm->client_pid = pid; + lm->client_index = mp->client_index; + + if (mp->max_macs_in_event) + fm->max_macs_in_event = mp->max_macs_in_event * 10; + else + fm->max_macs_in_event = L2FIB_EVENT_MAX_MACS_DEFAULT; + + if (mp->scan_delay) + fm->event_scan_delay = (f64) (mp->scan_delay) * 10e-3; + else + fm->event_scan_delay = L2FIB_EVENT_SCAN_DELAY_DEFAULT; + + /* change learn limit and flush all learned MACs */ + if (learn_limit && (learn_limit < L2LEARN_DEFAULT_LIMIT)) + lm->global_learn_limit = learn_limit; + else + lm->global_learn_limit = L2FIB_EVENT_LEARN_LIMIT_DEFAULT; + + l2fib_flush_all_mac (vlib_get_main ()); + } + else if (lm->client_pid != pid) + { + rv = VNET_API_ERROR_L2_MACS_EVENT_CLINET_PRESENT; + goto exit; + } + } + else if (lm->client_pid) + { + lm->client_pid = 0; + lm->client_index = 0; + lm->global_learn_limit = L2LEARN_DEFAULT_LIMIT; + } + +exit: + REPLY_MACRO (VL_API_WANT_L2_MACS_EVENTS_REPLY); +} + +static void vl_api_l2fib_flush_int_t_handler (vl_api_l2fib_flush_int_t * mp) { int rv = 0; @@ -293,8 +347,25 @@ vl_api_l2_flags_t_handler (vl_api_l2_flags_t * mp) VALIDATE_SW_IF_INDEX (mp); u32 sw_if_index = ntohl (mp->sw_if_index); - u32 flags = ntohl (mp->feature_bitmap) & L2INPUT_VALID_MASK; - rbm = l2input_intf_bitmap_enable (sw_if_index, flags, mp->is_set); + u32 flags = ntohl (mp->feature_bitmap); + u32 bitmap = 0; + + if (flags & L2_LEARN) + bitmap |= L2INPUT_FEAT_LEARN; + + if (flags & L2_FWD) + bitmap |= L2INPUT_FEAT_FWD; + + if (flags & L2_FLOOD) + bitmap |= L2INPUT_FEAT_FLOOD; + + if (flags & L2_UU_FLOOD) + bitmap |= L2INPUT_FEAT_UU_FLOOD; + + if (flags & L2_ARP_TERM) + bitmap |= L2INPUT_FEAT_ARP_TERM; + + rbm = l2input_intf_bitmap_enable (sw_if_index, bitmap, mp->is_set); BAD_SW_IF_INDEX_LABEL; @@ -455,13 +526,13 @@ vl_api_bridge_flags_t_handler (vl_api_bridge_flags_t * mp) goto out; } - bd_set_flags (vm, bd_index, flags, mp->is_set); + u32 bitmap = bd_set_flags (vm, bd_index, flags, mp->is_set); out: /* *INDENT-OFF* */ REPLY_MACRO2(VL_API_BRIDGE_FLAGS_REPLY, ({ - rmp->resulting_feature_bitmap = ntohl(flags); + rmp->resulting_feature_bitmap = ntohl(bitmap); })); /* *INDENT-ON* */ } diff --git a/src/vnet/l2/l2_bd.c b/src/vnet/l2/l2_bd.c index a87d02f2ffd..6e0db058e31 100644 --- a/src/vnet/l2/l2_bd.c +++ b/src/vnet/l2/l2_bd.c @@ -263,7 +263,7 @@ bd_set_flags (vlib_main_t * vm, u32 bd_index, u32 flags, u32 enable) bd_config->feature_bitmap &= ~feature_bitmap; } - return 0; + return bd_config->feature_bitmap; } /** @@ -328,12 +328,7 @@ bd_learn (vlib_main_t * vm, } /* set the bridge domain flag */ - if (bd_set_flags (vm, bd_index, L2_LEARN, enable)) - { - error = - clib_error_return (0, "bridge-domain id %d out of range", bd_index); - goto done; - } + bd_set_flags (vm, bd_index, L2_LEARN, enable); done: return error; @@ -397,12 +392,7 @@ bd_fwd (vlib_main_t * vm, unformat_input_t * input, vlib_cli_command_t * cmd) } /* set the bridge domain flag */ - if (bd_set_flags (vm, bd_index, L2_FWD, enable)) - { - error = - clib_error_return (0, "bridge-domain id %d out of range", bd_index); - goto done; - } + bd_set_flags (vm, bd_index, L2_FWD, enable); done: return error; @@ -468,12 +458,7 @@ bd_flood (vlib_main_t * vm, } /* set the bridge domain flag */ - if (bd_set_flags (vm, bd_index, L2_FLOOD, enable)) - { - error = - clib_error_return (0, "bridge-domain id %d out of range", bd_index); - goto done; - } + bd_set_flags (vm, bd_index, L2_FLOOD, enable); done: return error; @@ -538,12 +523,7 @@ bd_uu_flood (vlib_main_t * vm, } /* set the bridge domain flag */ - if (bd_set_flags (vm, bd_index, L2_UU_FLOOD, enable)) - { - error = - clib_error_return (0, "bridge-domain id %d out of range", bd_index); - goto done; - } + bd_set_flags (vm, bd_index, L2_UU_FLOOD, enable); done: return error; @@ -605,12 +585,7 @@ bd_arp_term (vlib_main_t * vm, enable = 0; /* set the bridge domain flag */ - if (bd_set_flags (vm, bd_index, L2_ARP_TERM, enable)) - { - error = - clib_error_return (0, "bridge-domain id %d out of range", bd_index); - goto done; - } + bd_set_flags (vm, bd_index, L2_ARP_TERM, enable); done: return error; diff --git a/src/vnet/l2/l2_fib.c b/src/vnet/l2/l2_fib.c index 7e59b09848a..8aa0ac295dd 100644 --- a/src/vnet/l2/l2_fib.c +++ b/src/vnet/l2/l2_fib.c @@ -31,6 +31,17 @@ #include <vppinfra/bihash_template.c> +#include <vlibmemory/api.h> +#include <vnet/vnet_msg_enum.h> + +#define vl_typedefs /* define message structures */ +#include <vnet/vnet_all_api_h.h> +#undef vl_typedefs + +#define vl_endianfun /* define message structures */ +#include <vnet/vnet_all_api_h.h> +#undef vl_endianfun + /** * @file * @brief Ethernet MAC Address FIB Table Management. @@ -117,6 +128,7 @@ show_l2fib (vlib_main_t * vm, int i, j, k; u8 verbose = 0; u8 raw = 0; + u8 learn = 0; u32 bd_id, bd_index = ~0; u8 now = (u8) (vlib_time_now (vm) / 60); u8 *s = 0; @@ -127,12 +139,18 @@ show_l2fib (vlib_main_t * vm, verbose = 1; else if (unformat (input, "bd_index %d", &bd_index)) verbose = 1; + else if (unformat (input, "learn")) + { + learn = 1; + verbose = 0; + } else if (unformat (input, "bd_id %d", &bd_id)) { uword *p = hash_get (bdm->bd_index_by_bd_id, bd_id); if (p) { - verbose = 1; + if (learn == 0) + verbose = 1; bd_index = p[0]; } else @@ -155,7 +173,7 @@ show_l2fib (vlib_main_t * vm, if (v->kvp[k].key == ~0ULL && v->kvp[k].value == ~0ULL) continue; - if (verbose && first_entry) + if ((verbose || learn) && first_entry) { first_entry = 0; vlib_cli_output (vm, @@ -168,13 +186,19 @@ show_l2fib (vlib_main_t * vm, key.raw = v->kvp[k].key; result.raw = v->kvp[k].value; - if (verbose + if ((verbose || learn) & ((bd_index >> 31) || (bd_index == key.fields.bd_index))) { + if (learn && result.fields.age_not) + { + total_entries++; + continue; /* skip provisioned macs */ + } + bd_config = vec_elt_at_index (l2input_main.bd_configs, key.fields.bd_index); - if (bd_config->mac_age && !result.fields.static_mac) + if (bd_config->mac_age && !result.fields.age_not) { i16 delta = now - result.fields.timestamp; delta += delta < 0 ? 256 : 0; @@ -206,9 +230,19 @@ show_l2fib (vlib_main_t * vm, if (total_entries == 0) vlib_cli_output (vm, "no l2fib entries"); else - vlib_cli_output (vm, - "%lld l2fib entries with %d learned (or non-static) entries", - total_entries, l2learn_main.global_learn_count); + { + l2learn_main_t *lm = &l2learn_main; + vlib_cli_output (vm, "L2FIB total/learned entries: %d/%d " + "Last scan time: %.4esec Learn limit: %d ", + total_entries, lm->global_learn_count, + msm->age_scan_duration, lm->global_learn_limit); + if (lm->client_pid) + vlib_cli_output (vm, "L2MAC events client PID: %d " + "Last e-scan time: %.4esec Delay: %.2esec " + "Max macs in event: %d", + lm->client_pid, msm->evt_scan_duration, + msm->event_scan_delay, msm->max_macs_in_event); + } if (raw) vlib_cli_output (vm, "Raw Hash Table:\n%U\n", @@ -242,7 +276,7 @@ show_l2fib (vlib_main_t * vm, /* *INDENT-OFF* */ VLIB_CLI_COMMAND (show_l2fib_cli, static) = { .path = "show l2fib", - .short_help = "show l2fib [verbose | bd_id <nn> | bd_index <nn> | raw]", + .short_help = "show l2fib [verbose | learn | bd_id <nn> | bd_index <nn> | raw", .function = show_l2fib, }; /* *INDENT-ON* */ @@ -309,36 +343,39 @@ l2fib_cur_seq_num (u32 bd_index, u32 sw_if_index) */ void l2fib_add_entry (u64 mac, u32 bd_index, - u32 sw_if_index, u32 static_mac, u32 filter_mac, u32 bvi_mac) + u32 sw_if_index, u8 static_mac, u8 filter_mac, u8 bvi_mac) { l2fib_entry_key_t key; l2fib_entry_result_t result; __attribute__ ((unused)) u32 bucket_contents; - l2fib_main_t *mp = &l2fib_main; + l2fib_main_t *fm = &l2fib_main; + l2learn_main_t *lm = &l2learn_main; BVT (clib_bihash_kv) kv; /* set up key */ key.raw = l2fib_make_key ((u8 *) & mac, bd_index); + /* check if entry alread exist */ + if (BV (clib_bihash_search) (&fm->mac_table, &kv, &kv)) + { + /* decrement counter if overwriting a learned mac */ + result.raw = kv.value; + if ((result.fields.age_not == 0) && (lm->global_learn_count)) + lm->global_learn_count--; + } + /* set up result */ result.raw = 0; /* clear all fields */ result.fields.sw_if_index = sw_if_index; result.fields.static_mac = static_mac; result.fields.filter = filter_mac; result.fields.bvi = bvi_mac; - if (!static_mac) - result.fields.sn = l2fib_cur_seq_num (bd_index, sw_if_index); + result.fields.age_not = 1; /* no aging for provisioned entry */ kv.key = key.raw; kv.value = result.raw; - BV (clib_bihash_add_del) (&mp->mac_table, &kv, 1 /* is_add */ ); - - /* increment counter if dynamically learned mac */ - if (result.fields.static_mac == 0) - { - l2learn_main.global_learn_count++; - } + BV (clib_bihash_add_del) (&fm->mac_table, &kv, 1 /* is_add */ ); } /** @@ -630,13 +667,8 @@ l2fib_del_entry_by_key (u64 raw_key) result.raw = kv.value; /* decrement counter if dynamically learned mac */ - if (result.fields.static_mac == 0) - { - if (l2learn_main.global_learn_count > 0) - { - l2learn_main.global_learn_count--; - } - } + if ((result.fields.age_not == 0) && (l2learn_main.global_learn_count)) + l2learn_main.global_learn_count--; /* Remove entry from hash table */ BV (clib_bihash_add_del) (&mp->mac_table, &kv, 0 /* is_add */ ); @@ -910,111 +942,273 @@ BVT (clib_bihash) * get_mac_table (void) return &mp->mac_table; } +static_always_inline void * +allocate_mac_evt_buf (u32 client, u32 client_index) +{ + l2fib_main_t *fm = &l2fib_main; + vl_api_l2_macs_event_t *mp = vl_msg_api_alloc + (sizeof (*mp) + (fm->max_macs_in_event * sizeof (vl_api_mac_entry_t))); + mp->_vl_msg_id = htons (VL_API_L2_MACS_EVENT); + mp->pid = htonl (client); + mp->client_index = client_index; + return mp; +} + +static_always_inline f64 +l2fib_scan (vlib_main_t * vm, f64 start_time, u8 event_only) +{ + l2fib_main_t *fm = &l2fib_main; + l2learn_main_t *lm = &l2learn_main; + + BVT (clib_bihash) * h = &fm->mac_table; + int i, j, k; + f64 last_start = start_time; + f64 accum_t = 0; + f64 delta_t = 0; + u32 evt_idx = 0; + u32 learn_count = 0; + u32 client = lm->client_pid; + u32 cl_idx = lm->client_index; + vl_api_l2_macs_event_t *mp = 0; + unix_shared_memory_queue_t *q = 0; + + if (client) + { + mp = allocate_mac_evt_buf (client, cl_idx); + q = vl_api_client_index_to_input_queue (lm->client_index); + } + + for (i = 0; i < h->nbuckets; i++) + { + /* allow no more than 20us without a pause */ + delta_t = vlib_time_now (vm) - last_start; + if (delta_t > 20e-6) + { + vlib_process_suspend (vm, 100e-6); /* suspend for 100 us */ + last_start = vlib_time_now (vm); + accum_t += delta_t; + } + + if (i < (h->nbuckets - 3)) + { + BVT (clib_bihash_bucket) * b = &h->buckets[i + 3]; + CLIB_PREFETCH (b, CLIB_CACHE_LINE_BYTES, LOAD); + b = &h->buckets[i + 1]; + if (b->offset) + { + BVT (clib_bihash_value) * v = + BV (clib_bihash_get_value) (h, b->offset); + CLIB_PREFETCH (v, CLIB_CACHE_LINE_BYTES, LOAD); + } + } + + BVT (clib_bihash_bucket) * b = &h->buckets[i]; + if (b->offset == 0) + continue; + BVT (clib_bihash_value) * v = BV (clib_bihash_get_value) (h, b->offset); + for (j = 0; j < (1 << b->log2_pages); j++) + { + for (k = 0; k < BIHASH_KVP_PER_PAGE; k++) + { + if (v->kvp[k].key == ~0ULL && v->kvp[k].value == ~0ULL) + continue; + + l2fib_entry_key_t key = {.raw = v->kvp[k].key }; + l2fib_entry_result_t result = {.raw = v->kvp[k].value }; + + if (result.fields.age_not == 0) + learn_count++; + + if (PREDICT_FALSE (evt_idx >= fm->max_macs_in_event)) + { + /* evet message full, sent it and start a new one */ + if (q && (q->cursize < q->maxsize)) + { + mp->n_macs = htonl (evt_idx); + vl_msg_api_send_shmem (q, (u8 *) & mp); + mp = allocate_mac_evt_buf (client, cl_idx); + } + else + { + clib_warning ("MAC event to pid %d queue stuffed!" + " %d MAC entries lost", client, evt_idx); + } + evt_idx = 0; + } + + if (client) + { + if (result.fields.lrn_evt) + { + /* copy mac entry to event msg */ + clib_memcpy (mp->mac[evt_idx].mac_addr, key.fields.mac, + 6); + mp->mac[evt_idx].is_del = 0; + mp->mac[evt_idx].sw_if_index = + htonl (result.fields.sw_if_index); + /* clear event bit and update mac entry */ + result.fields.lrn_evt = 0; + BVT (clib_bihash_kv) kv; + kv.key = key.raw; + kv.value = result.raw; + BV (clib_bihash_add_del) (&fm->mac_table, &kv, 1); + evt_idx++; + continue; /* skip aging */ + } + } + + if (event_only || result.fields.age_not) + continue; /* skip aging - static_mac alsways age_not */ + + /* start aging processing */ + u32 bd_index = key.fields.bd_index; + u32 sw_if_index = result.fields.sw_if_index; + u16 sn = l2fib_cur_seq_num (bd_index, sw_if_index).as_u16; + if (result.fields.sn.as_u16 != sn) + goto age_out; /* stale mac */ + + l2_bridge_domain_t *bd_config = + vec_elt_at_index (l2input_main.bd_configs, bd_index); + + if (bd_config->mac_age == 0) + continue; /* skip aging */ + + i16 delta = (u8) (start_time / 60) - result.fields.timestamp; + delta += delta < 0 ? 256 : 0; + + if (delta < bd_config->mac_age) + continue; /* still valid */ + + age_out: + if (client) + { + /* copy mac entry to event msg */ + clib_memcpy (mp->mac[evt_idx].mac_addr, key.fields.mac, 6); + mp->mac[evt_idx].is_del = 1; + mp->mac[evt_idx].sw_if_index = + htonl (result.fields.sw_if_index); + evt_idx++; + } + /* delete mac entry */ + BVT (clib_bihash_kv) kv; + kv.key = key.raw; + BV (clib_bihash_add_del) (&fm->mac_table, &kv, 0); + learn_count--; + } + v++; + } + } + + /* keep learn count consistent */ + l2learn_main.global_learn_count = learn_count; + + if (mp) + { + /* send any outstanding mac event message else free message buffer */ + if (evt_idx) + { + if (q && (q->cursize < q->maxsize)) + { + mp->n_macs = htonl (evt_idx); + vl_msg_api_send_shmem (q, (u8 *) & mp); + } + else + { + clib_warning ("MAC event to pid %d queue stuffed!" + " %d MAC entries lost", client, evt_idx); + vl_msg_api_free (mp); + } + } + else + vl_msg_api_free (mp); + } + return delta_t + accum_t; +} + +/* Type of scan */ +#define SCAN_MAC_AGE 0 +#define SCAN_MAC_EVENT 1 + +/* Maximum f64 value */ +#define TIME_MAX (1.7976931348623157e+308) + static uword l2fib_mac_age_scanner_process (vlib_main_t * vm, vlib_node_runtime_t * rt, vlib_frame_t * f) { uword event_type, *event_data = 0; - l2fib_main_t *msm = &l2fib_main; + l2fib_main_t *fm = &l2fib_main; + l2learn_main_t *lm = &l2learn_main; bool enabled = 0; - f64 start_time, last_run_duration = 0, t; + bool scan = SCAN_MAC_AGE; /* SCAN_FOR_AGE or SCAN_FOR_EVENT */ + f64 start_time, next_age_scan_time = TIME_MAX; while (1) { if (enabled) - vlib_process_wait_for_event_or_clock (vm, 60 - last_run_duration); + { + if (lm->client_pid) /* mac event client waiting */ + vlib_process_wait_for_event_or_clock (vm, fm->event_scan_delay); + else /* agin only */ + { + f64 t = next_age_scan_time - vlib_time_now (vm); + if (t < fm->event_scan_delay) + t = fm->event_scan_delay; + vlib_process_wait_for_event_or_clock (vm, t); + } + } else vlib_process_wait_for_event (vm); event_type = vlib_process_get_events (vm, &event_data); vec_reset_length (event_data); + start_time = vlib_time_now (vm); + switch (event_type) { - case ~0: + case ~0: /* timer expired */ + if ((lm->client_pid == 0) || (start_time >= next_age_scan_time)) + { + scan = SCAN_MAC_AGE; + if (enabled) + next_age_scan_time = start_time + L2FIB_AGE_SCAN_INTERVAL; + else + next_age_scan_time = TIME_MAX; + } + else + scan = SCAN_MAC_EVENT; break; + case L2_MAC_AGE_PROCESS_EVENT_START: + scan = SCAN_MAC_AGE; + next_age_scan_time = start_time + L2FIB_AGE_SCAN_INTERVAL; enabled = 1; break; + case L2_MAC_AGE_PROCESS_EVENT_STOP: enabled = 0; + next_age_scan_time = TIME_MAX; + l2fib_main.age_scan_duration = 0; + l2fib_main.evt_scan_duration = 0; continue; + case L2_MAC_AGE_PROCESS_EVENT_ONE_PASS: - enabled = 0; + scan = SCAN_MAC_AGE; + if (enabled) + next_age_scan_time = start_time + L2FIB_AGE_SCAN_INTERVAL; + else + next_age_scan_time = TIME_MAX; break; + default: ASSERT (0); } - last_run_duration = start_time = vlib_time_now (vm); - BVT (clib_bihash) * h = &msm->mac_table; - int i, j, k; - for (i = 0; i < h->nbuckets; i++) - { - /* Allow no more than 10us without a pause */ - t = vlib_time_now (vm); - if (t > start_time + 10e-6) - { - vlib_process_suspend (vm, 100e-6); /* suspend for 100 us */ - start_time = vlib_time_now (vm); - } - - if (i < (h->nbuckets - 3)) - { - BVT (clib_bihash_bucket) * b = &h->buckets[i + 3]; - CLIB_PREFETCH (b, CLIB_CACHE_LINE_BYTES, LOAD); - b = &h->buckets[i + 1]; - if (b->offset) - { - BVT (clib_bihash_value) * v = - BV (clib_bihash_get_value) (h, b->offset); - CLIB_PREFETCH (v, CLIB_CACHE_LINE_BYTES, LOAD); - } - } - - BVT (clib_bihash_bucket) * b = &h->buckets[i]; - if (b->offset == 0) - continue; - BVT (clib_bihash_value) * v = - BV (clib_bihash_get_value) (h, b->offset); - for (j = 0; j < (1 << b->log2_pages); j++) - { - for (k = 0; k < BIHASH_KVP_PER_PAGE; k++) - { - if (v->kvp[k].key == ~0ULL && v->kvp[k].value == ~0ULL) - continue; - - l2fib_entry_key_t key = {.raw = v->kvp[k].key }; - l2fib_entry_result_t result = {.raw = v->kvp[k].value }; - - if (result.fields.static_mac) - continue; - - u32 bd_index = key.fields.bd_index; - u32 sw_if_index = result.fields.sw_if_index; - u16 sn = l2fib_cur_seq_num (bd_index, sw_if_index).as_u16; - if (result.fields.sn.as_u16 != sn) - { - l2fib_del_entry_by_key (key.raw); - continue; - } - l2_bridge_domain_t *bd_config = - vec_elt_at_index (l2input_main.bd_configs, bd_index); - - if (bd_config->mac_age == 0) - continue; - - i16 delta = - (u8) (start_time / 60) - result.fields.timestamp; - delta += delta < 0 ? 256 : 0; - - if (delta > bd_config->mac_age) - l2fib_del_entry_by_key (key.raw); - } - v++; - } - } - last_run_duration = vlib_time_now (vm) - last_run_duration; + if (scan == SCAN_MAC_EVENT) + l2fib_main.evt_scan_duration = l2fib_scan (vm, start_time, 1); + else + l2fib_main.age_scan_duration = l2fib_scan (vm, start_time, 0); } return 0; } diff --git a/src/vnet/l2/l2_fib.h b/src/vnet/l2/l2_fib.h index ee6f0dc56a4..49a8b5b628c 100644 --- a/src/vnet/l2/l2_fib.h +++ b/src/vnet/l2/l2_fib.h @@ -27,6 +27,18 @@ #define L2FIB_NUM_BUCKETS (64 * 1024) #define L2FIB_MEMORY_SIZE (256<<20) +/* Ager scan interval is 1 minute for aging */ +#define L2FIB_AGE_SCAN_INTERVAL (60.0) + +/* MAC event scan delay is 100 msec unless specified by MAC event client */ +#define L2FIB_EVENT_SCAN_DELAY_DEFAULT (0.1) + +/* Max MACs in a event message is 100 unless specified by MAC event client */ +#define L2FIB_EVENT_MAX_MACS_DEFAULT (100) + +/* MAC event learn limit is 1000 unless specified by MAC event client */ +#define L2FIB_EVENT_LEARN_LIMIT_DEFAULT (1000) + typedef struct { @@ -36,6 +48,16 @@ typedef struct /* per swif vector of sequence number for interface based flush of MACs */ u8 *swif_seq_num; + /* last event or ager scan duration */ + f64 evt_scan_duration; + f64 age_scan_duration; + + /* delay between event scans, default to 100 msec */ + f64 event_scan_delay; + + /* max macs in evet message, default to 100 entries */ + u32 max_macs_in_event; + /* convenience variables */ vlib_main_t *vlib_main; vnet_main_t *vnet_main; @@ -89,12 +111,15 @@ typedef struct { struct { - u32 sw_if_index; /* output sw_if_index (L3 interface if bvi==1) */ + u32 sw_if_index; /* output sw_if_index (L3 intf if bvi==1) */ - u8 static_mac:1; /* static mac, no dataplane learning */ + u8 static_mac:1; /* static mac, no MAC move */ + u8 age_not:1; /* not subject to age */ u8 bvi:1; /* mac is for a bridged virtual interface */ u8 filter:1; /* drop packets to/from this mac */ - u8 unused1:5; + u8 lrn_evt:1; /* MAC learned to be sent in L2 MAC event */ + u8 unused:3; + u8 timestamp; /* timestamp for aging */ l2fib_seq_num_t sn; /* bd/int seq num */ } fields; @@ -348,11 +373,11 @@ void l2fib_clear_table (void); void l2fib_add_entry (u64 mac, u32 bd_index, - u32 sw_if_index, u32 static_mac, u32 drop_mac, u32 bvi_mac); + u32 sw_if_index, u8 static_mac, u8 drop_mac, u8 bvi_mac); static inline void -l2fib_add_fwd_entry (u64 mac, u32 bd_index, u32 sw_if_index, u32 static_mac, - u32 bvi_mac) +l2fib_add_fwd_entry (u64 mac, u32 bd_index, u32 sw_if_index, u8 static_mac, + u8 bvi_mac) { l2fib_add_entry (mac, bd_index, sw_if_index, static_mac, 0, bvi_mac); } diff --git a/src/vnet/l2/l2_fwd.c b/src/vnet/l2/l2_fwd.c index 8140728b5e1..2bb7307c68c 100644 --- a/src/vnet/l2/l2_fwd.c +++ b/src/vnet/l2/l2_fwd.c @@ -141,8 +141,9 @@ l2fwd_process (vlib_main_t * vm, vnet_buffer (b0)->sw_if_index[VLIB_TX] = result0->fields.sw_if_index; *next0 = L2FWD_NEXT_L2_OUTPUT; int l2fib_seq_num_valid = 1; + /* check l2fib seq num for stale entries */ - if (!result0->fields.static_mac) + if (!result0->fields.age_not) { l2fib_seq_num_t in_sn = {.as_u16 = vnet_buffer (b0)->l2.l2fib_sn }; l2fib_seq_num_t expected_sn = { diff --git a/src/vnet/l2/l2_learn.c b/src/vnet/l2/l2_learn.c index 65406292cc7..47c036b0bf1 100644 --- a/src/vnet/l2/l2_learn.c +++ b/src/vnet/l2/l2_learn.c @@ -123,11 +123,9 @@ l2learn_process (vlib_node_runtime_t * node, /* Check mac table lookup result */ if (PREDICT_TRUE (result0->fields.sw_if_index == sw_if_index0)) { - /* - * The entry was in the table, and the sw_if_index matched, the normal case - */ + /* Entry in L2FIB with matching sw_if_index matched - normal fast path */ counter_base[L2LEARN_ERROR_HIT] += 1; - int update = !result0->fields.static_mac && + int update = !result0->fields.age_not && /* static_mac always age_not */ (result0->fields.timestamp != timestamp || result0->fields.sn.as_u16 != vnet_buffer (b0)->l2.l2fib_sn); @@ -136,10 +134,10 @@ l2learn_process (vlib_node_runtime_t * node, } else if (result0->raw == ~0) { - /* The entry was not in table, so add it */ + /* Entry not in L2FIB - add it */ counter_base[L2LEARN_ERROR_MISS] += 1; - if (msm->global_learn_count == msm->global_learn_limit) + if (msm->global_learn_count >= msm->global_learn_limit) { /* * Global limit reached. Do not learn the mac but forward the packet. @@ -149,15 +147,22 @@ l2learn_process (vlib_node_runtime_t * node, return; } + /* Do not learn if mac is 0 */ + l2fib_entry_key_t key = *key0; + key.fields.bd_index = 0; + if (key.raw == 0) + return; + /* It is ok to learn */ msm->global_learn_count++; result0->raw = 0; /* clear all fields */ result0->fields.sw_if_index = sw_if_index0; + result0->fields.lrn_evt = (msm->client_pid != 0); cached_key->raw = ~0; /* invalidate the cache */ } else { - /* The entry was in the table, but with the wrong sw_if_index mapping (mac move) */ + /* Entry in L2FIB with different sw_if_index - mac move or filter */ if (result0->fields.filter) { ASSERT (result0->fields.sw_if_index == ~0); @@ -167,8 +172,6 @@ l2learn_process (vlib_node_runtime_t * node, return; } - counter_base[L2LEARN_ERROR_MAC_MOVE] += 1; - if (result0->fields.static_mac) { /* @@ -185,6 +188,13 @@ l2learn_process (vlib_node_runtime_t * node, * TODO: check global/bridge domain/interface learn limits */ result0->fields.sw_if_index = sw_if_index0; + if (result0->fields.age_not) /* The mac was provisioned */ + { + msm->global_learn_count++; + result0->fields.age_not = 0; + } + result0->fields.lrn_evt = (msm->client_pid != 0); + counter_base[L2LEARN_ERROR_MAC_MOVE] += 1; } /* Update the entry */ @@ -479,7 +489,7 @@ VLIB_NODE_FUNCTION_MULTIARCH (l2learn_node, l2learn_node_fn) * Set the default number of dynamically learned macs to the number * of buckets. */ - mp->global_learn_limit = L2FIB_NUM_BUCKETS * 16; + mp->global_learn_limit = L2LEARN_DEFAULT_LIMIT; return 0; } diff --git a/src/vnet/l2/l2_learn.h b/src/vnet/l2/l2_learn.h index 0d95de04b66..000ab59ee7f 100644 --- a/src/vnet/l2/l2_learn.h +++ b/src/vnet/l2/l2_learn.h @@ -34,6 +34,10 @@ typedef struct /* maximum number of dynamically learned mac entries */ u32 global_learn_limit; + /* client waiting for L2 MAC events for learned and aged MACs */ + u32 client_pid; + u32 client_index; + /* Next nodes for each feature */ u32 feat_next_node_index[32]; @@ -42,6 +46,7 @@ typedef struct vnet_main_t *vnet_main; } l2learn_main_t; +#define L2LEARN_DEFAULT_LIMIT (L2FIB_NUM_BUCKETS * 16) l2learn_main_t l2learn_main; |