DPDK provides a library for IPsec data-path processing. The library utilizes the existing DPDK crypto-dev and security API to provide the application with a transparent and high performant IPsec packet processing API. The library is concentrated on data-path protocols processing (ESP and AH), IKE protocol(s) implementation is out of scope for this library.
This API operates on the IPsec Security Association (SA) level. It provides functionality that allows user for given SA to process inbound and outbound IPsec packets.
To be more specific:
The SA level API is based on top of crypto-dev/security API and relies on them to perform actual cipher and integrity checking.
Due to the nature of the crypto-dev API (enqueue/dequeue model) the library introduces an asynchronous API for IPsec packets destined to be processed by the crypto-device.
The expected API call sequence for data-path processing would be:
/* enqueue for processing by crypto-device */
rte_ipsec_pkt_crypto_prepare(...);
rte_cryptodev_enqueue_burst(...);
/* dequeue from crypto-device and do final processing (if any) */
rte_cryptodev_dequeue_burst(...);
rte_ipsec_pkt_crypto_group(...); /* optional */
rte_ipsec_pkt_process(...);
For packets destined for inline processing no extra overhead is required and the synchronous API call: rte_ipsec_pkt_process() is sufficient for that case.
Note
For more details about the IPsec API, please refer to the DPDK API Reference.
The current implementation supports all four currently defined rte_security types:
In that mode the library functions perform
In that mode the library functions perform
In that mode the library functions perform
In that mode the library functions perform
To accommodate future custom implementations function pointers model is used for both crypto_prepare and process implementations.
SA database(SAD) is a table with <key, value> pairs.
Value is an opaque user provided pointer to the user defined SA data structure.
According to RFC4301 each SA can be uniquely identified by a key which is either:
- security parameter index(SPI)
- or SPI and destination IP(DIP)
- or SPI, DIP and source IP(SIP)
In case of multiple matches, longest matching key will be returned.
librte_ipsec SAD implementation provides ability to create/destroy SAD tables.
To create SAD table user has to specify how many entries of each key type is required and IP protocol type (IPv4/IPv6). As an example:
struct rte_ipsec_sad *sad;
struct rte_ipsec_sad_conf conf;
conf.socket_id = -1;
conf.max_sa[RTE_IPSEC_SAD_SPI_ONLY] = some_nb_rules_spi_only;
conf.max_sa[RTE_IPSEC_SAD_SPI_DIP] = some_nb_rules_spi_dip;
conf.max_sa[RTE_IPSEC_SAD_SPI_DIP_SIP] = some_nb_rules_spi_dip_sip;
conf.flags = RTE_IPSEC_SAD_FLAG_RW_CONCURRENCY;
sad = rte_ipsec_sad_create("test", &conf);
Note
for more information please refer to ipsec library API reference
Library also provides methods to add or delete key/value pairs from the SAD. To add user has to specify key, key type and a value which is an opaque pointer to SA. The key type reflects a set of tuple fields that will be used for lookup of the SA. As mentioned above there are 3 types of a key and the representation of a key type is:
RTE_IPSEC_SAD_SPI_ONLY,
RTE_IPSEC_SAD_SPI_DIP,
RTE_IPSEC_SAD_SPI_DIP_SIP,
As an example, to add new entry into the SAD for IPv4 addresses:
struct rte_ipsec_sa *sa;
union rte_ipsec_sad_key key;
key.v4.spi = rte_cpu_to_be_32(spi_val);
if (key_type >= RTE_IPSEC_SAD_SPI_DIP) /* DIP is optional*/
key.v4.dip = rte_cpu_to_be_32(dip_val);
if (key_type == RTE_IPSEC_SAD_SPI_DIP_SIP) /* SIP is optional*/
key.v4.sip = rte_cpu_to_be_32(sip_val);
rte_ipsec_sad_add(sad, &key, key_type, sa);
Note
By performance reason it is better to keep spi/dip/sip in net byte order to eliminate byteswap on lookup
To delete user has to specify key and key type.
Delete code would look like:
union rte_ipsec_sad_key key;
key.v4.spi = rte_cpu_to_be_32(necessary_spi);
if (key_type >= RTE_IPSEC_SAD_SPI_DIP) /* DIP is optional*/
key.v4.dip = rte_cpu_to_be_32(necessary_dip);
if (key_type == RTE_IPSEC_SAD_SPI_DIP_SIP) /* SIP is optional*/
key.v4.sip = rte_cpu_to_be_32(necessary_sip);
rte_ipsec_sad_del(sad, &key, key_type);
Library provides lookup by the given {SPI,DIP,SIP} tuple of inbound ipsec packet as a key.
The search key is represented by:
union rte_ipsec_sad_key {
struct rte_ipsec_sadv4_key v4;
struct rte_ipsec_sadv6_key v6;
};
where v4 is a tuple for IPv4:
struct rte_ipsec_sadv4_key {
uint32_t spi;
uint32_t dip;
uint32_t sip;
};
and v6 is a tuple for IPv6:
struct rte_ipsec_sadv6_key {
uint32_t spi;
uint8_t dip[16];
uint8_t sip[16];
};
As an example, lookup related code could look like that:
int i;
union rte_ipsec_sad_key keys[BURST_SZ];
const union rte_ipsec_sad_key *keys_p[BURST_SZ];
void *vals[BURST_SZ];
for (i = 0; i < BURST_SZ_MAX; i++) {
keys[i].v4.spi = esp_hdr[i]->spi;
keys[i].v4.dip = ipv4_hdr[i]->dst_addr;
keys[i].v4.sip = ipv4_hdr[i]->src_addr;
keys_p[i] = &keys[i];
}
rte_ipsec_sad_lookup(sad, keys_p, vals, BURST_SZ);
for (i = 0; i < BURST_SZ_MAX; i++) {
if (vals[i] == NULL)
printf("SA not found for key index %d\n", i);
else
printf("SA pointer is %p\n", vals[i]);
}
The following features are not properly supported in the current version: