/* Execute a group entry of type INDIRECT. */
static void
execute_indirect(struct group_entry *entry, struct packet *pkt) {
if (entry->desc->buckets_num > 0) {
struct ofl_bucket *bucket = entry->desc->buckets[0];
struct packet *p = packet_clone(pkt);
if (VLOG_IS_DBG_ENABLED(LOG_MODULE)) {
char *b = ofl_structs_bucket_to_string(bucket, entry->dp->exp);
VLOG_DBG_RL(LOG_MODULE, &rl, "Writing bucket: %s.", b);
free(b);
}
action_set_write_actions(p->action_set, bucket->actions_num, bucket->actions);
entry->stats->byte_count += p->buffer->size;
entry->stats->packet_count++;
entry->stats->counters[0]->byte_count += p->buffer->size;
entry->stats->counters[0]->packet_count++;
action_set_execute(p->action_set, p);
packet_destroy(p);
} else {
VLOG_DBG_RL(LOG_MODULE, &rl, "No bucket in group.");
}
}
/* Execute a group entry of type FAILFAST. */
static void
execute_ff(struct group_entry *entry, struct packet *pkt) {
size_t b = select_from_ff_group(entry);
if (b != -1) {
struct ofl_bucket *bucket = entry->desc->buckets[b];
struct packet *p = packet_clone(pkt);
if (VLOG_IS_DBG_ENABLED(LOG_MODULE)) {
char *b = ofl_structs_bucket_to_string(bucket, entry->dp->exp);
VLOG_DBG_RL(LOG_MODULE, &rl, "Writing bucket: %s.", b);
free(b);
}
action_set_write_actions(p->action_set, bucket->actions_num, bucket->actions);
entry->stats->byte_count += p->buffer->size;
entry->stats->packet_count++;
entry->stats->counters[b]->byte_count += p->buffer->size;
entry->stats->counters[b]->packet_count++;
/* Cookie field is set 0xffffffffffffffff
because we cannot associate to any
particular flow */
action_set_execute(p->action_set, p, 0xffffffffffffffff);
packet_destroy(p);
} else {
VLOG_DBG_RL(LOG_MODULE, &rl, "No bucket in group.");
}
}
static uint32_t
nx_entry_ok(const void *p, unsigned int match_len)
{
unsigned int payload_len;
ovs_be32 header_be;
uint32_t header;
if (match_len < 4) {
if (match_len) {
VLOG_DBG_RL(&rl, "nx_match ends with partial (%u-byte) nxm_header",
match_len);
}
return 0;
}
memcpy(&header_be, p, 4);
header = ntohl(header_be);
payload_len = NXM_LENGTH(header);
if (!payload_len) {
VLOG_DBG_RL(&rl, "nxm_entry %08"PRIx32" has invalid payload "
"length 0", header);
return 0;
}
if (match_len < payload_len + 4) {
VLOG_DBG_RL(&rl, "%"PRIu32"-byte nxm_entry but only "
"%u bytes left in nx_match", payload_len + 4, match_len);
return 0;
}
return header;
}
void
action_set_write_actions(struct action_set *set,
size_t actions_num,
struct ofl_action_header **actions) {
size_t i;
VLOG_DBG_RL(LOG_MODULE, &rl, "Writing to action set.");
for (i=0; i<actions_num; i++) {
action_set_write_action(set, actions[i]);
}
VLOG_DBG_RL(LOG_MODULE, &rl, action_set_to_string(set));
}
/* Free the specified entry which keeps state for one bundle ID. */
static void
bundle_table_entry_destroy(struct bundle_table_entry *entry) {
struct bundle_message *bundle_msg, *bundle_msg_next;
LIST_FOR_EACH_SAFE (bundle_msg, bundle_msg_next, struct bundle_message, node, &entry->bundle_message_list) {
VLOG_DBG_RL(LOG_MODULE, &rl, "Free message with type %u and length %u\n",
bundle_msg->message->type,
ntohs(bundle_msg->message->length));
list_remove(&bundle_msg->node);
bundle_message_free(bundle_msg);
}
list_remove(&entry->node);
VLOG_DBG_RL(LOG_MODULE, &rl, "Destroyed bundle table entry for bundle ID %u.\n", entry->bundle_id);
free(entry);
}
/* Helper function for nl_dump_next(). */
static int
nl_dump_recv(struct nl_dump *dump)
{
struct nlmsghdr *nlmsghdr;
int retval;
retval = nl_sock_recv__(dump->sock, &dump->buffer, true);
if (retval) {
return retval == EINTR ? EAGAIN : retval;
}
nlmsghdr = nl_msg_nlmsghdr(&dump->buffer);
if (dump->seq != nlmsghdr->nlmsg_seq) {
VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
nlmsghdr->nlmsg_seq, dump->seq);
return EAGAIN;
}
if (nl_msg_nlmsgerr(&dump->buffer, &retval)) {
VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
ovs_strerror(retval));
return retval && retval != EAGAIN ? retval : EPROTO;
}
return 0;
}
static void
resize(struct hmap *hmap, size_t new_mask, const char *where)
{
struct hmap tmp;
size_t i;
ovs_assert(is_pow2(new_mask + 1));
hmap_init(&tmp);
if (new_mask) {
tmp.buckets = xmalloc(sizeof *tmp.buckets * (new_mask + 1));
tmp.mask = new_mask;
for (i = 0; i <= tmp.mask; i++) {
tmp.buckets[i] = NULL;
}
}
for (i = 0; i <= hmap->mask; i++) {
struct hmap_node *node, *next;
int count = 0;
for (node = hmap->buckets[i]; node; node = next) {
next = node->next;
hmap_insert_fast(&tmp, node, node->hash);
count++;
}
if (count > 5) {
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
COVERAGE_INC(hmap_pathological);
VLOG_DBG_RL(&rl, "%s: %d nodes in bucket (%"PRIuSIZE" nodes, %"PRIuSIZE" buckets)",
where, count, hmap->n, hmap->mask + 1);
}
}
hmap_swap(hmap, &tmp);
hmap_destroy(&tmp);
}
/* Sends the 'request' member of the 'n' transactions in 'transactions' on
* 'sock', in order, and receives responses to all of them. Fills in the
* 'error' member of each transaction with 0 if it was successful, otherwise
* with a positive errno value. If 'reply' is nonnull, then it will be filled
* with the reply if the message receives a detailed reply. In other cases,
* i.e. where the request failed or had no reply beyond an indication of
* success, 'reply' will be cleared if it is nonnull.
*
* The caller is responsible for destroying each request and reply, and the
* transactions array itself.
*
* Before sending each message, this function will finalize nlmsg_len in each
* 'request' to match the ofpbuf's size, set nlmsg_pid to 'sock''s pid, and
* initialize nlmsg_seq.
*
* Bare Netlink is an unreliable transport protocol. This function layers
* reliable delivery and reply semantics on top of bare Netlink. See
* nl_sock_transact() for some caveats.
*/
void
nl_sock_transact_multiple(struct nl_sock *sock,
struct nl_transaction **transactions, size_t n)
{
int max_batch_count;
int error;
if (!n) {
return;
}
error = nl_sock_cow__(sock);
if (error) {
nl_sock_record_errors__(transactions, n, error);
return;
}
/* In theory, every request could have a 64 kB reply. But the default and
* maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
* be a bit below 128 kB, so that would only allow a single message in a
* "batch". So we assume that replies average (at most) 4 kB, which allows
* a good deal of batching.
*
* In practice, most of the requests that we batch either have no reply at
* all or a brief reply. */
max_batch_count = MAX(sock->rcvbuf / 4096, 1);
max_batch_count = MIN(max_batch_count, max_iovs);
while (n > 0) {
size_t count, bytes;
size_t done;
/* Batch up to 'max_batch_count' transactions. But cap it at about a
* page of requests total because big skbuffs are expensive to
* allocate in the kernel. */
#if defined(PAGESIZE)
enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
#else
enum { MAX_BATCH_BYTES = 4096 - 512 };
#endif
bytes = transactions[0]->request->size;
for (count = 1; count < n && count < max_batch_count; count++) {
if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
break;
}
bytes += transactions[count]->request->size;
}
error = nl_sock_transact_multiple__(sock, transactions, count, &done);
transactions += done;
n -= done;
if (error == ENOBUFS) {
VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
} else if (error) {
VLOG_ERR_RL(&rl, "transaction error (%s)", strerror(error));
nl_sock_record_errors__(transactions, n, error);
}
}
}
static enum ofperr
oxm_pull_match__(struct ofpbuf *b, bool strict, struct match *match)
{
struct ofp11_match_header *omh = ofpbuf_data(b);
uint8_t *p;
uint16_t match_len;
if (ofpbuf_size(b) < sizeof *omh) {
return OFPERR_OFPBMC_BAD_LEN;
}
match_len = ntohs(omh->length);
if (match_len < sizeof *omh) {
return OFPERR_OFPBMC_BAD_LEN;
}
if (omh->type != htons(OFPMT_OXM)) {
return OFPERR_OFPBMC_BAD_TYPE;
}
p = ofpbuf_try_pull(b, ROUND_UP(match_len, 8));
if (!p) {
VLOG_DBG_RL(&rl, "oxm length %u, rounded up to a "
"multiple of 8, is longer than space in message (max "
"length %"PRIu32")", match_len, ofpbuf_size(b));
return OFPERR_OFPBMC_BAD_LEN;
}
return nx_pull_raw(p + sizeof *omh, match_len - sizeof *omh,
strict, match, NULL, NULL);
}
void
group_entry_execute(struct group_entry *entry,
struct packet *packet) {
VLOG_DBG_RL(LOG_MODULE, &rl, "Executing group %u.", entry->stats->group_id);
/* NOTE: Packet is copied for all buckets now (even if there is only one).
* This allows execution of the original packet onward. It is not clear
* whether that is allowed or not according to the spec. though. */
switch (entry->desc->type) {
case (OFPGT_ALL): {
execute_all(entry, packet);
break;
}
case (OFPGT_SELECT): {
execute_select(entry, packet);
break;
}
case (OFPGT_INDIRECT): {
execute_indirect(entry, packet);
break;
}
case (OFPGT_FF): {
execute_ff(entry, packet);
break;
}
default: {
VLOG_WARN_RL(LOG_MODULE, &rl, "Trying to execute unknown group type (%u) in group (%u).", entry->desc->type, entry->stats->group_id);
}
}
}
/* Executes a group entry of type ALL. */
static void
execute_all(struct group_entry *entry, struct packet *pkt) {
size_t i;
/* TODO Zoltan: Currently packets are always cloned. However it should
* be possible to see if cloning is necessary, or not, based on bucket actions. */
for (i=0; i<entry->desc->buckets_num; i++) {
struct ofl_bucket *bucket = entry->desc->buckets[i];
struct packet *p = packet_clone(pkt);
if (VLOG_IS_DBG_ENABLED(LOG_MODULE)) {
char *b = ofl_structs_bucket_to_string(bucket, entry->dp->exp);
VLOG_DBG_RL(LOG_MODULE, &rl, "Writing bucket: %s.", b);
free(b);
}
action_set_write_actions(p->action_set, bucket->actions_num, bucket->actions);
entry->stats->byte_count += p->buffer->size;
entry->stats->packet_count++;
entry->stats->counters[i]->byte_count += p->buffer->size;
entry->stats->counters[i]->packet_count++;
/* Cookie field is set 0xffffffffffffffff
because we cannot associate to any
particular flow */
action_set_execute(p->action_set, p, 0xffffffffffffffff);
packet_destroy(p);
}
}
/* Attempts to make 'ml' learn from the fact that a frame from 'src_mac' was
* just observed arriving from 'src_port' on the given 'vlan'.
*
* Returns nonzero if we actually learned something from this, zero if it just
* confirms what we already knew. The nonzero return value is the tag of flows
* that now need revalidation.
*
* The 'vlan' parameter is used to maintain separate per-VLAN learning tables.
* Specify 0 if this behavior is undesirable.
*
* 'lock_type' specifies whether the entry should be locked or existing locks
* are check. */
tag_type
mac_learning_learn(struct mac_learning *ml,
const uint8_t src_mac[ETH_ADDR_LEN], uint16_t vlan,
uint16_t src_port, enum grat_arp_lock_type lock_type)
{
struct mac_entry *e;
struct list *bucket;
if (!is_learning_vlan(ml, vlan)) {
return 0;
}
if (eth_addr_is_multicast(src_mac)) {
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 30);
VLOG_DBG_RL(&rl, "multicast packet source "ETH_ADDR_FMT,
ETH_ADDR_ARGS(src_mac));
return 0;
}
bucket = mac_table_bucket(ml, src_mac, vlan);
e = search_bucket(bucket, src_mac, vlan);
if (!e) {
if (!list_is_empty(&ml->free)) {
e = mac_entry_from_lru_node(ml->free.next);
} else {
e = mac_entry_from_lru_node(ml->lrus.next);
list_remove(&e->hash_node);
}
memcpy(e->mac, src_mac, ETH_ADDR_LEN);
list_push_front(bucket, &e->hash_node);
e->port = -1;
e->vlan = vlan;
e->tag = make_unknown_mac_tag(ml, src_mac, vlan);
e->grat_arp_lock = TIME_MIN;
}
if (lock_type != GRAT_ARP_LOCK_CHECK || time_now() >= e->grat_arp_lock) {
/* Make the entry most-recently-used. */
list_remove(&e->lru_node);
list_push_back(&ml->lrus, &e->lru_node);
e->expires = time_now() + MAC_ENTRY_IDLE_TIME;
if (lock_type == GRAT_ARP_LOCK_SET) {
e->grat_arp_lock = time_now() + MAC_GRAT_ARP_LOCK_TIME;
}
/* Did we learn something? */
if (e->port != src_port) {
tag_type old_tag = e->tag;
e->port = src_port;
e->tag = tag_create_random();
COVERAGE_INC(mac_learning_learned);
return old_tag;
}
}
return 0;
}
/* Create the table entry for the specified bundle ID. */
static struct bundle_table_entry *
bundle_table_entry_create(uint32_t bundle_id, uint16_t flags) {
struct bundle_table_entry *entry;
entry = xmalloc(sizeof(struct bundle_table_entry));
list_init(&entry->node);
list_init(&entry->bundle_message_list);
entry->bundle_id = bundle_id;
entry->flags = flags;
entry->closed = false;
VLOG_DBG_RL(LOG_MODULE, &rl, "Created bundle table entry for bundle ID %u.\n", bundle_id);
return entry;
}
static enum ofperr
nx_pull_match__(struct ofpbuf *b, unsigned int match_len, bool strict,
struct match *match,
ovs_be64 *cookie, ovs_be64 *cookie_mask)
{
uint8_t *p = NULL;
if (match_len) {
p = ofpbuf_try_pull(b, ROUND_UP(match_len, 8));
if (!p) {
VLOG_DBG_RL(&rl, "nx_match length %u, rounded up to a "
"multiple of 8, is longer than space in message (max "
"length %"PRIu32")", match_len, ofpbuf_size(b));
return OFPERR_OFPBMC_BAD_LEN;
}
}
return nx_pull_raw(p, match_len, strict, match, cookie, cookie_mask);
}
/* Create a structure to retain an appended message. */
static struct bundle_message *
bundle_message_create(struct ofl_msg_bundle_add_msg *add_msg) {
struct bundle_message *msg;
size_t message_length;
msg = xmalloc(sizeof(struct bundle_message));
list_init(&msg->node);
message_length = ntohs(add_msg->message->length);
msg->message = xmalloc(message_length);
memcpy(msg->message, add_msg->message, message_length);
VLOG_DBG_RL(LOG_MODULE, &rl,
"Created %u byte bundle message entry for bundle ID %u.\n",
message_length,
add_msg->bundle_id);
return msg;
}
/* Processes 'msg', which should be an OpenFlow received on 'rconn', according
* to the learning switch state in 'sw'. The most likely result of processing
* is that flow-setup and packet-out OpenFlow messages will be sent out on
* 'rconn'. */
static void
lswitch_process_packet(struct lswitch *sw, const struct ofpbuf *msg)
{
enum ofptype type;
struct ofpbuf b;
b = *msg;
if (ofptype_pull(&type, &b)) {
return;
}
if (sw->state == S_FEATURES_REPLY
&& type != OFPTYPE_ECHO_REQUEST
&& type != OFPTYPE_FEATURES_REPLY) {
return;
}
if (type == OFPTYPE_ECHO_REQUEST) {
process_echo_request(sw, msg->data);
} else if (type == OFPTYPE_FEATURES_REPLY) {
if (sw->state == S_FEATURES_REPLY) {
if (!process_switch_features(sw, msg->data)) {
sw->state = S_SWITCHING;
} else {
rconn_disconnect(sw->rconn);
}
}
} else if (type == OFPTYPE_PACKET_IN) {
process_packet_in(sw, msg->data);
} else if (type == OFPTYPE_FLOW_REMOVED) {
/* Nothing to do. */
} else if (VLOG_IS_DBG_ENABLED()) {
char *s = ofp_to_string(msg->data, msg->size, 2);
VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
sw->datapath_id, s);
free(s);
}
}
static enum ofperr
nx_pull_raw(const uint8_t *p, unsigned int match_len, bool strict,
struct match *match, ovs_be64 *cookie, ovs_be64 *cookie_mask)
{
uint32_t header;
ovs_assert((cookie != NULL) == (cookie_mask != NULL));
match_init_catchall(match);
if (cookie) {
*cookie = *cookie_mask = htonll(0);
}
if (!match_len) {
return 0;
}
for (;
(header = nx_entry_ok(p, match_len)) != 0;
p += 4 + NXM_LENGTH(header), match_len -= 4 + NXM_LENGTH(header)) {
const struct mf_field *mf;
enum ofperr error;
mf = mf_from_nxm_header(header);
if (!mf) {
if (strict) {
error = OFPERR_OFPBMC_BAD_FIELD;
} else {
continue;
}
} else if (!mf_are_prereqs_ok(mf, &match->flow)) {
error = OFPERR_OFPBMC_BAD_PREREQ;
} else if (!mf_is_all_wild(mf, &match->wc)) {
error = OFPERR_OFPBMC_DUP_FIELD;
} else {
unsigned int width = mf->n_bytes;
union mf_value value;
memcpy(&value, p + 4, width);
if (!mf_is_value_valid(mf, &value)) {
error = OFPERR_OFPBMC_BAD_VALUE;
} else if (!NXM_HASMASK(header)) {
error = 0;
mf_set_value(mf, &value, match);
} else {
union mf_value mask;
memcpy(&mask, p + 4 + width, width);
if (!mf_is_mask_valid(mf, &mask)) {
error = OFPERR_OFPBMC_BAD_MASK;
} else {
error = check_mask_consistency(p, mf);
if (!error) {
mf_set(mf, &value, &mask, match);
}
}
}
}
/* Check if the match is for a cookie rather than a classifier rule. */
if ((header == NXM_NX_COOKIE || header == NXM_NX_COOKIE_W) && cookie) {
if (*cookie_mask) {
error = OFPERR_OFPBMC_DUP_FIELD;
} else {
unsigned int width = sizeof *cookie;
memcpy(cookie, p + 4, width);
if (NXM_HASMASK(header)) {
memcpy(cookie_mask, p + 4 + width, width);
} else {
*cookie_mask = OVS_BE64_MAX;
}
error = 0;
}
}
if (error) {
VLOG_DBG_RL(&rl, "bad nxm_entry %#08"PRIx32" (vendor=%"PRIu32", "
"field=%"PRIu32", hasmask=%"PRIu32", len=%"PRIu32"), "
"(%s)", header,
NXM_VENDOR(header), NXM_FIELD(header),
NXM_HASMASK(header), NXM_LENGTH(header),
ofperr_to_string(error));
return error;
}
}
return match_len ? OFPERR_OFPBMC_BAD_LEN : 0;
}
/* Tries to receive an openflow message from datapath 'dp_idx' on 'sock'. If
* successful, stores the received message into '*msgp' and returns 0. The
* caller is responsible for destroying the message with ofpbuf_delete(). On
* failure, returns a positive errno value and stores a null pointer into
* '*msgp'.
*
* Only Netlink messages with embedded OpenFlow messages are accepted. Other
* Netlink messages provoke errors.
*
* If 'wait' is true, dpif_recv_openflow waits for a message to be ready;
* otherwise, returns EAGAIN if the 'sock' receive buffer is empty. */
int
dpif_recv_openflow(struct dpif *dp, int dp_idx, struct ofpbuf **bufferp,
bool wait)
{
struct nlattr *attrs[ARRAY_SIZE(openflow_policy)];
struct ofpbuf *buffer;
struct ofp_header *oh;
uint16_t ofp_len;
buffer = *bufferp = NULL;
do {
int retval;
do {
ofpbuf_delete(buffer);
retval = nl_sock_recv(dp->sock, &buffer, wait);
} while (retval == ENOBUFS
|| (!retval
&& (nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE
|| nl_msg_nlmsgerr(buffer, NULL))));
if (retval) {
if (retval != EAGAIN) {
VLOG_WARN_RL(&rl, "dpif_recv_openflow: %s", strerror(retval));
}
return retval;
}
if (nl_msg_genlmsghdr(buffer) == NULL) {
VLOG_DBG_RL(&rl, "received packet too short for Generic Netlink");
goto error;
}
if (nl_msg_nlmsghdr(buffer)->nlmsg_type != openflow_family) {
VLOG_DBG_RL(&rl,
"received type (%"PRIu16") != openflow family (%d)",
nl_msg_nlmsghdr(buffer)->nlmsg_type, openflow_family);
goto error;
}
if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN,
openflow_policy, attrs,
ARRAY_SIZE(openflow_policy))) {
goto error;
}
} while (nl_attr_get_u32(attrs[DP_GENL_A_DP_IDX]) != dp_idx);
oh = buffer->data = (void *) nl_attr_get(attrs[DP_GENL_A_OPENFLOW]);
buffer->size = nl_attr_get_size(attrs[DP_GENL_A_OPENFLOW]);
ofp_len = ntohs(oh->length);
if (ofp_len != buffer->size) {
VLOG_WARN_RL(&rl,
"ofp_header.length %"PRIu16" != attribute length %zu\n",
ofp_len, buffer->size);
buffer->size = MIN(ofp_len, buffer->size);
}
*bufferp = buffer;
return 0;
error:
ofpbuf_delete(buffer);
return EPROTO;
}
/* Commit operation. */
static ofl_err
bundle_commit(struct datapath *dp,
struct bundle_table *table,
uint32_t bundle_id,
uint16_t flags,
const struct sender *sender) {
struct bundle_table_entry *entry;
struct bundle_message *bundle_msg;
struct ofl_msg_header *msg;
uint32_t xid;
ofl_err error;
ofl_err last_error = ofl_error(OFPET_BUNDLE_FAILED, OFPBFC_BAD_ID);
/* Find and process commit operation for bundle ID */
entry = bundle_table_entry_find(table, bundle_id);
if (entry != NULL) {
/* Ensure flags are consistent with flags specified previously */
if (entry->flags != flags) {
last_error = ofl_error(OFPET_BUNDLE_FAILED, OFPBFC_BAD_FLAGS);
} else {
/* Save state in case failure occurs */
last_error = 0;
dp_save_state(dp);
/* Commit all messages in bundle, stopping at first error */
LIST_FOR_EACH (bundle_msg, struct bundle_message, node, &entry->bundle_message_list) {
VLOG_DBG_RL(LOG_MODULE, &rl, "Commit of message with type %u and length %u\n",
bundle_msg->message->type,
ntohs(bundle_msg->message->length));
error = ofl_msg_unpack((uint8_t *)bundle_msg->message,
ntohs(bundle_msg->message->length),
&msg, &xid, dp->exp);
if (!error) {
/* This prototype only properly supports bundling of
* messages that do not generate replies (other than
* error replies). TODO: keep replies in a holding
* area and only release them to the controller when
* the commit succeeds. */
error = handle_control_msg(dp, msg, sender);
if (error) {
ofl_msg_free(msg, dp->exp);
}
}
if (error) {
last_error = error;
break;
}
}
/* Restore state if failures occurred */
if (last_error) {
dp_restore_state(dp);
} else {
/* TODO free memory used to save state without restoring
* (not required currently as variables used to save/restore
* state are re-used) */
}
/* We need to generate the error ourselves. The spec say that
* the error need to refer to the offending message in the budle.
* If we just return the error code, the error message would refer
* to the commit message. */
if (last_error) {
struct sender orig_sender = {.remote = sender->remote,
.conn_id = sender->conn_id,
.xid = xid};
struct ofl_msg_error orig_err =
{{.type = OFPT_ERROR},
.type = ofl_error_type(last_error),
.code = ofl_error_code(last_error),
.data_length = ntohs(bundle_msg->message->length),
.data = (uint8_t *)bundle_msg->message};
dp_send_message(dp, (struct ofl_msg_header *)&orig_err, &orig_sender);
/* Trigger second error message. */
last_error = ofl_error(OFPET_BUNDLE_FAILED, OFPBFC_MSG_FAILED);
}
}
/* Whether or not commit succeeded: free entry for bundle ID */
bundle_table_entry_destroy(entry);
}
return last_error;
}
void
pipeline_process_packet(struct pipeline *pl, struct packet *pkt) {
struct flow_table *table, *next_table;
if (VLOG_IS_DBG_ENABLED(LOG_MODULE)) {
char *pkt_str = packet_to_string(pkt);
VLOG_DBG_RL(LOG_MODULE, &rl, "processing packet: %s", pkt_str);
free(pkt_str);
}
if (!packet_handle_std_is_ttl_valid(pkt->handle_std)) {
if ((pl->dp->config.flags & OFPC_INVALID_TTL_TO_CONTROLLER) != 0) {
VLOG_DBG_RL(LOG_MODULE, &rl, "Packet has invalid TTL, sending to controller.");
send_packet_to_controller(pl, pkt, 0/*table_id*/, OFPR_INVALID_TTL);
} else {
VLOG_DBG_RL(LOG_MODULE, &rl, "Packet has invalid TTL, dropping.");
}
packet_destroy(pkt);
return;
}
next_table = pl->tables[0];
while (next_table != NULL) {
struct flow_entry *entry;
VLOG_DBG_RL(LOG_MODULE, &rl, "trying table %u.", next_table->stats->table_id);
pkt->table_id = next_table->stats->table_id;
table = next_table;
next_table = NULL;
entry = flow_table_lookup(table, pkt);
if (entry != NULL) {
if (VLOG_IS_DBG_ENABLED(LOG_MODULE)) {
char *m = ofl_structs_flow_stats_to_string(entry->stats, pkt->dp->exp);
VLOG_DBG_RL(LOG_MODULE, &rl, "found matching entry: %s.", m);
free(m);
}
execute_entry(pl, entry, &next_table, &pkt);
/* Packet could be destroyed by a meter instruction */
if (!pkt)
return;
if (next_table == NULL) {
/* Cookie field is set 0xffffffffffffffff
because we cannot associate it to any
particular flow */
action_set_execute(pkt->action_set, pkt, 0xffffffffffffffff);
packet_destroy(pkt);
return;
}
} else {
/* OpenFlow 1.3 default behavior on a table miss */
VLOG_DBG_RL(LOG_MODULE, &rl, "No matching entry found. Dropping packet.");
packet_destroy(pkt);
return;
}
}
VLOG_WARN_RL(LOG_MODULE, &rl, "Reached outside of pipeline processing cycle.");
}
static int
nl_sock_transact_multiple__(struct nl_sock *sock,
struct nl_transaction **transactions, size_t n,
size_t *done)
{
uint64_t tmp_reply_stub[1024 / 8];
struct nl_transaction tmp_txn;
struct ofpbuf tmp_reply;
uint32_t base_seq;
struct iovec iovs[MAX_IOVS];
struct msghdr msg;
int error;
int i;
base_seq = nl_sock_allocate_seq(sock, n);
*done = 0;
for (i = 0; i < n; i++) {
struct nl_transaction *txn = transactions[i];
struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
nlmsg->nlmsg_len = txn->request->size;
nlmsg->nlmsg_seq = base_seq + i;
nlmsg->nlmsg_pid = sock->pid;
iovs[i].iov_base = txn->request->data;
iovs[i].iov_len = txn->request->size;
}
memset(&msg, 0, sizeof msg);
msg.msg_iov = iovs;
msg.msg_iovlen = n;
do {
error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
} while (error == EINTR);
for (i = 0; i < n; i++) {
struct nl_transaction *txn = transactions[i];
log_nlmsg(__func__, error, txn->request->data, txn->request->size,
sock->protocol);
}
if (!error) {
COVERAGE_ADD(netlink_sent, n);
}
if (error) {
return error;
}
ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
tmp_txn.request = NULL;
tmp_txn.reply = &tmp_reply;
tmp_txn.error = 0;
while (n > 0) {
struct nl_transaction *buf_txn, *txn;
uint32_t seq;
/* Find a transaction whose buffer we can use for receiving a reply.
* If no such transaction is left, use tmp_txn. */
buf_txn = &tmp_txn;
for (i = 0; i < n; i++) {
if (transactions[i]->reply) {
buf_txn = transactions[i];
break;
}
}
/* Receive a reply. */
error = nl_sock_recv__(sock, buf_txn->reply, false);
if (error) {
if (error == EAGAIN) {
nl_sock_record_errors__(transactions, n, 0);
*done += n;
error = 0;
}
break;
}
/* Match the reply up with a transaction. */
seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
if (seq < base_seq || seq >= base_seq + n) {
VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
continue;
}
i = seq - base_seq;
txn = transactions[i];
/* Fill in the results for 'txn'. */
if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
if (txn->reply) {
ofpbuf_clear(txn->reply);
}
if (txn->error) {
VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
error, ovs_strerror(txn->error));
}
} else {
txn->error = 0;
if (txn->reply && txn != buf_txn) {
/* Swap buffers. */
struct ofpbuf *reply = buf_txn->reply;
buf_txn->reply = txn->reply;
txn->reply = reply;
}
}
/* Fill in the results for transactions before 'txn'. (We have to do
* this after the results for 'txn' itself because of the buffer swap
* above.) */
nl_sock_record_errors__(transactions, i, 0);
/* Advance. */
*done += i + 1;
transactions += i + 1;
n -= i + 1;
base_seq += i + 1;
}
ofpbuf_uninit(&tmp_reply);
return error;
}
void
meter_entry_apply(struct meter_entry *entry, struct packet **pkt){
size_t b;
bool drop = false;
entry->stats->packet_in_count++;
entry->stats->byte_in_count += (*pkt)->buffer->size;
b = choose_band(entry, *pkt);
if(b != -1){
struct ofl_meter_band_header *band_header = (struct ofl_meter_band_header*) entry->config->bands[b];
switch(band_header->type){
case OFPMBT_DROP:{
drop = true;
break;
}
case OFPMBT_DSCP_REMARK:{
packet_handle_std_validate((*pkt)->handle_std);
if ((*pkt)->handle_std->valid)
{
struct ofl_meter_band_dscp_remark *band_header = (struct ofl_meter_band_dscp_remark *) entry->config->bands[b];
/* Nothing prevent this band to be used for non-IP packets, so filter them out. Jean II */
if ((*pkt)->handle_std->proto->ipv4 != NULL) {
// Fetch dscp in ipv4 header
struct ip_header *ipv4 = (*pkt)->handle_std->proto->ipv4;
uint8_t old_drop = ipv4->ip_tos & 0x1C;
/* The spec says that we need to increase
the drop precedence of the packet.
We need a valid DSCP out of the process,
so we can only modify dscp if the
drop precedence is low (tos 0x***010**)
or medium (tos 0x***100**). Jean II */
if (((old_drop == 0x8) && (band_header->prec_level <= 2)) || ((old_drop == 0x10) && (band_header->prec_level <= 1))) {
uint8_t new_drop = old_drop + (band_header->prec_level << 3);
uint8_t new_tos = new_drop | (ipv4->ip_tos & 0xE3);
uint16_t old_val = htons((ipv4->ip_ihl_ver << 8) + ipv4->ip_tos);
uint16_t new_val = htons((ipv4->ip_ihl_ver << 8) + new_tos);
ipv4->ip_csum = recalc_csum16(ipv4->ip_csum, old_val, new_val);
ipv4->ip_tos = new_tos;
}
}
else if ((*pkt)->handle_std->proto->ipv6 != NULL){
struct ipv6_header *ipv6 = (*pkt)->handle_std->proto->ipv6;
uint32_t ipv6_ver_tc_fl = ntohl(ipv6->ipv6_ver_tc_fl);
uint32_t old_drop = ipv6_ver_tc_fl & 0x1C00000;
if (((old_drop == 0x800000) && (band_header->prec_level <= 2)) || ((old_drop == 0x1000000) && (band_header->prec_level <= 1))){
uint32_t prec_level = band_header->prec_level << 23;
uint32_t new_drop = old_drop + prec_level;
ipv6->ipv6_ver_tc_fl = htonl(new_drop | (ipv6_ver_tc_fl & 0xFE3FFFFF));
}
}
(*pkt)->handle_std->valid = false;
}
break;
}
case OFPMBT_EXPERIMENTER:{
break;
}
}
entry->stats->band_stats[b]->byte_band_count += (*pkt)->buffer->size;
entry->stats->band_stats[b]->packet_band_count++;
if (drop){
VLOG_DBG_RL(LOG_MODULE, &rl, "Dropping packet: rate %d", band_header->rate);
packet_destroy(*pkt);
*pkt = NULL;
}
}
}