diff options
Diffstat (limited to 'src/vnet/ipsec')
| -rw-r--r-- | src/vnet/ipsec/esp.h | 86 | ||||
| -rw-r--r-- | src/vnet/ipsec/esp_decrypt.c | 1573 | ||||
| -rw-r--r-- | src/vnet/ipsec/esp_encrypt.c | 671 | ||||
| -rw-r--r-- | src/vnet/ipsec/ipsec.c | 83 | ||||
| -rw-r--r-- | src/vnet/ipsec/ipsec.h | 9 | ||||
| -rw-r--r-- | src/vnet/ipsec/ipsec_cli.c | 38 | ||||
| -rw-r--r-- | src/vnet/ipsec/ipsec_sa.c | 64 | ||||
| -rw-r--r-- | src/vnet/ipsec/ipsec_sa.h | 46 |
8 files changed, 1862 insertions, 708 deletions
diff --git a/src/vnet/ipsec/esp.h b/src/vnet/ipsec/esp.h index b1736453b93..6c47bb05428 100644 --- a/src/vnet/ipsec/esp.h +++ b/src/vnet/ipsec/esp.h @@ -105,29 +105,101 @@ esp_seq_advance (ipsec_sa_t * sa) return 0; } -always_inline void -esp_aad_fill (vnet_crypto_op_t * op, - const esp_header_t * esp, const ipsec_sa_t * sa) +always_inline u16 +esp_aad_fill (u8 * data, const esp_header_t * esp, const ipsec_sa_t * sa) { esp_aead_t *aad; - aad = (esp_aead_t *) op->aad; + aad = (esp_aead_t *) data; aad->data[0] = esp->spi; if (ipsec_sa_is_set_USE_ESN (sa)) { /* SPI, seq-hi, seq-low */ - aad->data[1] = clib_host_to_net_u32 (sa->seq_hi); + aad->data[1] = (u32) clib_host_to_net_u32 (sa->seq_hi); aad->data[2] = esp->seq; - op->aad_len = 12; + return 12; } else { /* SPI, seq-low */ aad->data[1] = esp->seq; - op->aad_len = 8; + return 8; } } + +/** + * The post data structure to for esp_encrypt/decrypt_inline to write to + * vib_buffer_t opaque unused field, and for post nodes to pick up after + * dequeue. + **/ +typedef struct +{ + union + { + struct + { + u8 icv_sz; + u8 iv_sz; + ipsec_sa_flags_t flags; + u32 sa_index; + }; + u64 sa_data; + }; + + u32 seq; + i16 current_data; + i16 current_length; + u16 hdr_sz; + u16 is_chain; + u32 protect_index; +} esp_decrypt_packet_data_t; + +STATIC_ASSERT_SIZEOF (esp_decrypt_packet_data_t, 3 * sizeof (u64)); + +/* we are forced to store the decrypt post data into 2 separate places - + vlib_opaque and opaque2. */ +typedef struct +{ + vlib_buffer_t *lb; + u32 free_buffer_index; + u8 icv_removed; +} esp_decrypt_packet_data2_t; + +typedef union +{ + u16 next_index; + esp_decrypt_packet_data_t decrypt_data; +} esp_post_data_t; + +STATIC_ASSERT (sizeof (esp_post_data_t) <= + STRUCT_SIZE_OF (vnet_buffer_opaque_t, unused), + "Custom meta-data too large for vnet_buffer_opaque_t"); + +#define esp_post_data(b) \ + ((esp_post_data_t *)((u8 *)((b)->opaque) \ + + STRUCT_OFFSET_OF (vnet_buffer_opaque_t, unused))) + +STATIC_ASSERT (sizeof (esp_decrypt_packet_data2_t) <= + STRUCT_SIZE_OF (vnet_buffer_opaque2_t, unused), + "Custom meta-data too large for vnet_buffer_opaque2_t"); + +#define esp_post_data2(b) \ + ((esp_decrypt_packet_data2_t *)((u8 *)((b)->opaque2) \ + + STRUCT_OFFSET_OF (vnet_buffer_opaque2_t, unused))) + +typedef struct +{ + /* esp post node index for async crypto */ + u32 esp4_post_next; + u32 esp6_post_next; + u32 esp4_tun_post_next; + u32 esp6_tun_post_next; +} esp_async_post_next_t; + +extern esp_async_post_next_t esp_encrypt_async_next; +extern esp_async_post_next_t esp_decrypt_async_next; + #endif /* __ESP_H__ */ /* diff --git a/src/vnet/ipsec/esp_decrypt.c b/src/vnet/ipsec/esp_decrypt.c index 0d14fe56740..b3c031e54af 100644 --- a/src/vnet/ipsec/esp_decrypt.c +++ b/src/vnet/ipsec/esp_decrypt.c @@ -32,7 +32,8 @@ _(DROP, "error-drop") \ _(IP4_INPUT, "ip4-input-no-checksum") \ _(IP6_INPUT, "ip6-input") \ _(L2_INPUT, "l2-input") \ -_(HANDOFF, "handoff") +_(HANDOFF, "handoff") \ +_(PENDING, "pending") #define _(v, s) ESP_DECRYPT_NEXT_##v, typedef enum @@ -42,9 +43,23 @@ typedef enum ESP_DECRYPT_N_NEXT, } esp_decrypt_next_t; +#define foreach_esp_decrypt_post_next \ +_(DROP, "error-drop") \ +_(IP4_INPUT, "ip4-input-no-checksum") \ +_(IP6_INPUT, "ip6-input") \ +_(L2_INPUT, "l2-input") + +#define _(v, s) ESP_DECRYPT_POST_NEXT_##v, +typedef enum +{ + foreach_esp_decrypt_post_next +#undef _ + ESP_DECRYPT_POST_N_NEXT, +} esp_decrypt_post_next_t; #define foreach_esp_decrypt_error \ _(RX_PKTS, "ESP pkts received") \ + _(RX_POST_PKTS, "ESP-POST pkts received") \ _(DECRYPTION_FAILED, "ESP decryption failed") \ _(INTEG_ERROR, "Integrity check failed") \ _(CRYPTO_ENGINE_ERROR, "crypto engine error (packet dropped)") \ @@ -96,32 +111,6 @@ format_esp_decrypt_trace (u8 * s, va_list * args) return s; } -typedef struct -{ - vlib_buffer_t *lb; - union - { - struct - { - u8 icv_sz; - u8 iv_sz; - ipsec_sa_flags_t flags; - u32 sa_index; - }; - u64 sa_data; - }; - - u32 seq; - u32 free_buffer_index; - i16 current_data; - i16 current_length; - u16 hdr_sz; - u8 icv_removed; - u8 __unused; -} esp_decrypt_packet_data_t; - -STATIC_ASSERT_SIZEOF (esp_decrypt_packet_data_t, 4 * sizeof (u64)); - #define ESP_ENCRYPT_PD_F_FD_TRANSPORT (1 << 2) static_always_inline void @@ -214,10 +203,10 @@ esp_remove_tail (vlib_main_t * vm, vlib_buffer_t * b, vlib_buffer_t * last, return pointer to it */ static_always_inline u8 * esp_move_icv (vlib_main_t * vm, vlib_buffer_t * first, - esp_decrypt_packet_data_t * pd, u16 icv_sz, u16 * dif) + esp_decrypt_packet_data2_t * pd2, u16 icv_sz, u16 * dif) { vlib_buffer_t *before_last, *bp; - u16 last_sz = pd->lb->current_length; + u16 last_sz = pd2->lb->current_length; u16 first_sz = icv_sz - last_sz; bp = before_last = first; @@ -227,42 +216,43 @@ esp_move_icv (vlib_main_t * vm, vlib_buffer_t * first, bp = vlib_get_buffer (vm, bp->next_buffer); } - u8 *lb_curr = vlib_buffer_get_current (pd->lb); + u8 *lb_curr = vlib_buffer_get_current (pd2->lb); memmove (lb_curr + first_sz, lb_curr, last_sz); clib_memcpy_fast (lb_curr, vlib_buffer_get_tail (before_last) - first_sz, first_sz); before_last->current_length -= first_sz; + clib_memset (vlib_buffer_get_tail (before_last), 0, first_sz); if (dif) dif[0] = first_sz; - pd->lb = before_last; - pd->icv_removed = 1; - pd->free_buffer_index = before_last->next_buffer; + pd2->lb = before_last; + pd2->icv_removed = 1; + pd2->free_buffer_index = before_last->next_buffer; before_last->flags &= ~VLIB_BUFFER_NEXT_PRESENT; return lb_curr; } -static_always_inline int +static_always_inline i16 esp_insert_esn (vlib_main_t * vm, ipsec_sa_t * sa, - esp_decrypt_packet_data_t * pd, vnet_crypto_op_t * op, - u16 * len, vlib_buffer_t * b, u8 * payload) + esp_decrypt_packet_data2_t * pd2, u32 * data_len, + u8 ** digest, u16 * len, vlib_buffer_t * b, u8 * payload) { if (!ipsec_sa_is_set_USE_ESN (sa)) - return 1; + return 0; /* shift ICV by 4 bytes to insert ESN */ u32 seq_hi = clib_host_to_net_u32 (sa->seq_hi); u8 tmp[ESP_MAX_ICV_SIZE], sz = sizeof (sa->seq_hi); - if (pd->icv_removed) + if (pd2->icv_removed) { - u16 space_left = vlib_buffer_space_left_at_end (vm, pd->lb); + u16 space_left = vlib_buffer_space_left_at_end (vm, pd2->lb); if (space_left >= sz) { - clib_memcpy_fast (vlib_buffer_get_tail (pd->lb), &seq_hi, sz); - op->len += sz; + clib_memcpy_fast (vlib_buffer_get_tail (pd2->lb), &seq_hi, sz); + *data_len += sz; } else - return 0; + return sz; len[0] = b->current_length; } @@ -271,39 +261,39 @@ esp_insert_esn (vlib_main_t * vm, ipsec_sa_t * sa, clib_memcpy_fast (tmp, payload + len[0], ESP_MAX_ICV_SIZE); clib_memcpy_fast (payload + len[0], &seq_hi, sz); clib_memcpy_fast (payload + len[0] + sz, tmp, ESP_MAX_ICV_SIZE); - op->len += sz; - op->digest += sz; + *data_len += sz; + *digest += sz; } - return 1; + return sz; } static_always_inline u8 * esp_move_icv_esn (vlib_main_t * vm, vlib_buffer_t * first, - esp_decrypt_packet_data_t * pd, u16 icv_sz, ipsec_sa_t * sa, - u8 * extra_esn, vnet_crypto_op_t * op) + esp_decrypt_packet_data2_t * pd2, u16 icv_sz, + ipsec_sa_t * sa, u8 * extra_esn, u32 * len) { u16 dif = 0; - u8 *digest = esp_move_icv (vm, first, pd, icv_sz, &dif); + u8 *digest = esp_move_icv (vm, first, pd2, icv_sz, &dif); if (dif) - op->len -= dif; + *len -= dif; if (ipsec_sa_is_set_USE_ESN (sa)) { u8 sz = sizeof (sa->seq_hi); u32 seq_hi = clib_host_to_net_u32 (sa->seq_hi); - u16 space_left = vlib_buffer_space_left_at_end (vm, pd->lb); + u16 space_left = vlib_buffer_space_left_at_end (vm, pd2->lb); if (space_left >= sz) { - clib_memcpy_fast (vlib_buffer_get_tail (pd->lb), &seq_hi, sz); - op->len += sz; + clib_memcpy_fast (vlib_buffer_get_tail (pd2->lb), &seq_hi, sz); + *len += sz; } else { /* no space for ESN at the tail, use the next buffer * (with ICV data) */ - ASSERT (pd->icv_removed); - vlib_buffer_t *tmp = vlib_get_buffer (vm, pd->free_buffer_index); + ASSERT (pd2->icv_removed); + vlib_buffer_t *tmp = vlib_get_buffer (vm, pd2->free_buffer_index); clib_memcpy_fast (vlib_buffer_get_current (tmp) - sz, &seq_hi, sz); extra_esn[0] = 1; } @@ -311,10 +301,735 @@ esp_move_icv_esn (vlib_main_t * vm, vlib_buffer_t * first, return digest; } +static_always_inline int +esp_decrypt_chain_integ (vlib_main_t * vm, ipsec_per_thread_data_t * ptd, + esp_decrypt_packet_data2_t * pd2, + ipsec_sa_t * sa0, vlib_buffer_t * b, u8 icv_sz, + u8 * start_src, u32 start_len, + u8 ** digest, u16 * n_ch, u32 * integ_total_len) +{ + vnet_crypto_op_chunk_t *ch; + vlib_buffer_t *cb = vlib_get_buffer (vm, b->next_buffer); + u16 n_chunks = 1; + u32 total_len; + vec_add2 (ptd->chunks, ch, 1); + total_len = ch->len = start_len; + ch->src = start_src; + + while (1) + { + vec_add2 (ptd->chunks, ch, 1); + n_chunks += 1; + ch->src = vlib_buffer_get_current (cb); + if (pd2->lb == cb) + { + if (pd2->icv_removed) + ch->len = cb->current_length; + else + ch->len = cb->current_length - icv_sz; + if (ipsec_sa_is_set_USE_ESN (sa0)) + { + u32 seq_hi = clib_host_to_net_u32 (sa0->seq_hi); + u8 tmp[ESP_MAX_ICV_SIZE], sz = sizeof (sa0->seq_hi); + u8 *esn; + vlib_buffer_t *tmp_b; + u16 space_left = vlib_buffer_space_left_at_end (vm, pd2->lb); + if (space_left < sz) + { + if (pd2->icv_removed) + { + /* use pre-data area from the last bufer + that was removed from the chain */ + tmp_b = vlib_get_buffer (vm, pd2->free_buffer_index); + esn = tmp_b->data - sz; + } + else + { + /* no space, need to allocate new buffer */ + u32 tmp_bi = 0; + if (vlib_buffer_alloc (vm, &tmp_bi, 1) != 1) + return -1; + tmp_b = vlib_get_buffer (vm, tmp_bi); + esn = tmp_b->data; + pd2->free_buffer_index = tmp_bi; + } + clib_memcpy_fast (esn, &seq_hi, sz); + + vec_add2 (ptd->chunks, ch, 1); + n_chunks += 1; + ch->src = esn; + ch->len = sz; + } + else + { + if (pd2->icv_removed) + { + clib_memcpy_fast (vlib_buffer_get_tail + (pd2->lb), &seq_hi, sz); + } + else + { + clib_memcpy_fast (tmp, *digest, ESP_MAX_ICV_SIZE); + clib_memcpy_fast (*digest, &seq_hi, sz); + clib_memcpy_fast (*digest + sz, tmp, ESP_MAX_ICV_SIZE); + *digest += sz; + } + ch->len += sz; + } + } + total_len += ch->len; + break; + } + else + total_len += ch->len = cb->current_length; + + if (!(cb->flags & VLIB_BUFFER_NEXT_PRESENT)) + break; + + cb = vlib_get_buffer (vm, cb->next_buffer); + } + + if (n_ch) + *n_ch = n_chunks; + if (integ_total_len) + *integ_total_len = total_len; + + return 0; +} + +static_always_inline u32 +esp_decrypt_chain_crypto (vlib_main_t * vm, ipsec_per_thread_data_t * ptd, + esp_decrypt_packet_data2_t * pd2, + ipsec_sa_t * sa0, vlib_buffer_t * b, u8 icv_sz, + u8 * start, u32 start_len, u8 ** tag, u16 * n_ch) +{ + vnet_crypto_op_chunk_t *ch; + vlib_buffer_t *cb = b; + u16 n_chunks = 1; + u32 total_len; + vec_add2 (ptd->chunks, ch, 1); + total_len = ch->len = start_len; + ch->src = ch->dst = start; + cb = vlib_get_buffer (vm, cb->next_buffer); + n_chunks = 1; + + while (1) + { + vec_add2 (ptd->chunks, ch, 1); + n_chunks += 1; + ch->src = ch->dst = vlib_buffer_get_current (cb); + if (pd2->lb == cb) + { + if (ipsec_sa_is_set_IS_AEAD (sa0)) + { + if (pd2->lb->current_length < icv_sz) + { + u16 dif = 0; + *tag = esp_move_icv (vm, b, pd2, icv_sz, &dif); + + /* this chunk does not contain crypto data */ + n_chunks -= 1; + /* and fix previous chunk's length as it might have + been changed */ + ASSERT (n_chunks > 0); + if (pd2->lb == b) + { + total_len -= dif; + ch[-1].len -= dif; + } + else + { + total_len = total_len + pd2->lb->current_length - + ch[-1].len; + ch[-1].len = pd2->lb->current_length; + } + break; + } + else + *tag = vlib_buffer_get_tail (pd2->lb) - icv_sz; + } + + if (pd2->icv_removed) + total_len += ch->len = cb->current_length; + else + total_len += ch->len = cb->current_length - icv_sz; + } + else + total_len += ch->len = cb->current_length; + + if (!(cb->flags & VLIB_BUFFER_NEXT_PRESENT)) + break; + + cb = vlib_get_buffer (vm, cb->next_buffer); + } + + if (n_ch) + *n_ch = n_chunks; + + return total_len; +} + +static_always_inline void +esp_decrypt_prepare_sync_op (vlib_main_t * vm, vlib_node_runtime_t * node, + ipsec_per_thread_data_t * ptd, + vnet_crypto_op_t *** crypto_ops, + vnet_crypto_op_t *** integ_ops, + vnet_crypto_op_t * op, + ipsec_sa_t * sa0, u8 * payload, + u16 len, u8 icv_sz, u8 iv_sz, + esp_decrypt_packet_data_t * pd, + esp_decrypt_packet_data2_t * pd2, + vlib_buffer_t * b, u16 * next, u32 index) +{ + const u8 esp_sz = sizeof (esp_header_t); + + if (PREDICT_TRUE (sa0->integ_op_id != VNET_CRYPTO_OP_NONE)) + { + vnet_crypto_op_init (op, sa0->integ_op_id); + op->key_index = sa0->integ_key_index; + op->src = payload; + op->flags = VNET_CRYPTO_OP_FLAG_HMAC_CHECK; + op->user_data = index; + op->digest = payload + len; + op->digest_len = icv_sz; + op->len = len; + + if (pd->is_chain) + { + /* buffer is chained */ + op->len = pd->current_length; + + /* special case when ICV is splitted and needs to be reassembled + * first -> move it to the last buffer. Also take into account + * that ESN needs to be added after encrypted data and may or + * may not fit in the tail.*/ + if (pd2->lb->current_length < icv_sz) + { + u8 extra_esn = 0; + op->digest = + esp_move_icv_esn (vm, b, pd2, icv_sz, sa0, + &extra_esn, &op->len); + + if (extra_esn) + { + /* esn is in the last buffer, that was unlinked from + * the chain */ + op->len = b->current_length; + } + else + { + if (pd2->lb == b) + { + /* we now have a single buffer of crypto data, adjust + * the length (second buffer contains only ICV) */ + *integ_ops = &ptd->integ_ops; + *crypto_ops = &ptd->crypto_ops; + len = b->current_length; + goto out; + } + } + } + else + op->digest = vlib_buffer_get_tail (pd2->lb) - icv_sz; + + op->flags |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; + op->chunk_index = vec_len (ptd->chunks); + if (esp_decrypt_chain_integ (vm, ptd, pd2, sa0, b, icv_sz, + payload, pd->current_length, + &op->digest, &op->n_chunks, 0) < 0) + { + b->error = node->errors[ESP_DECRYPT_ERROR_NO_BUFFERS]; + next[0] = ESP_DECRYPT_NEXT_DROP; + return; + } + } + else + esp_insert_esn (vm, sa0, pd2, &op->len, &op->digest, &len, b, + payload); + out: + vec_add_aligned (*(integ_ops[0]), op, 1, CLIB_CACHE_LINE_BYTES); + } + + payload += esp_sz; + len -= esp_sz; + + if (sa0->crypto_dec_op_id != VNET_CRYPTO_OP_NONE) + { + vnet_crypto_op_init (op, sa0->crypto_dec_op_id); + op->key_index = sa0->crypto_key_index; + op->iv = payload; + + if (ipsec_sa_is_set_IS_AEAD (sa0)) + { + esp_header_t *esp0; + esp_aead_t *aad; + u8 *scratch; + + /* + * construct the AAD and the nonce (Salt || IV) in a scratch + * space in front of the IP header. + */ + scratch = payload - esp_sz; + esp0 = (esp_header_t *) (scratch); + + scratch -= (sizeof (*aad) + pd->hdr_sz); + op->aad = scratch; + + op->aad_len = esp_aad_fill (op->aad, esp0, sa0); + + /* + * we don't need to refer to the ESP header anymore so we + * can overwrite it with the salt and use the IV where it is + * to form the nonce = (Salt + IV) + */ + op->iv -= sizeof (sa0->salt); + clib_memcpy_fast (op->iv, &sa0->salt, sizeof (sa0->salt)); + + op->tag = payload + len; + op->tag_len = 16; + } + op->src = op->dst = payload += iv_sz; + op->len = len - iv_sz; + op->user_data = index; + + if (pd->is_chain && (pd2->lb != b)) + { + /* buffer is chained */ + op->flags |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; + op->chunk_index = vec_len (ptd->chunks); + esp_decrypt_chain_crypto (vm, ptd, pd2, sa0, b, icv_sz, + payload, len - pd->iv_sz + pd->icv_sz, + &op->tag, &op->n_chunks); + } + + vec_add_aligned (*(crypto_ops[0]), op, 1, CLIB_CACHE_LINE_BYTES); + } +} + +static_always_inline int +esp_decrypt_prepare_async_frame (vlib_main_t * vm, + vlib_node_runtime_t * node, + ipsec_per_thread_data_t * ptd, + vnet_crypto_async_frame_t ** f, + ipsec_sa_t * sa0, u8 * payload, u16 len, + u8 icv_sz, u8 iv_sz, + esp_decrypt_packet_data_t * pd, + esp_decrypt_packet_data2_t * pd2, u32 bi, + vlib_buffer_t * b, u16 * next, + u16 async_next) +{ + const u8 esp_sz = sizeof (esp_header_t); + u32 current_protect_index = vnet_buffer (b)->ipsec.protect_index; + esp_decrypt_packet_data_t *async_pd = &(esp_post_data (b))->decrypt_data; + esp_decrypt_packet_data2_t *async_pd2 = esp_post_data2 (b); + u8 *tag = payload + len, *iv = payload + esp_sz, *aad = 0; + u32 key_index; + u32 crypto_len, integ_len = 0; + i16 crypto_start_offset, integ_start_offset = 0; + u8 flags = 0; + + if (!ipsec_sa_is_set_IS_AEAD (sa0)) + { + /* linked algs */ + key_index = sa0->linked_key_index; + integ_start_offset = payload - b->data; + integ_len = len; + + if (pd->is_chain) + { + /* buffer is chained */ + integ_len = pd->current_length; + + /* special case when ICV is splitted and needs to be reassembled + * first -> move it to the last buffer. Also take into account + * that ESN needs to be added after encrypted data and may or + * may not fit in the tail.*/ + if (pd2->lb->current_length < icv_sz) + { + u8 extra_esn = 0; + tag = esp_move_icv_esn (vm, b, pd2, icv_sz, sa0, + &extra_esn, &integ_len); + + if (extra_esn) + { + /* esn is in the last buffer, that was unlinked from + * the chain */ + integ_len = b->current_length; + } + else + { + if (pd2->lb == b) + { + /* we now have a single buffer of crypto data, adjust + * the length (second buffer contains only ICV) */ + len = b->current_length; + goto out; + } + } + } + else + tag = vlib_buffer_get_tail (pd2->lb) - icv_sz; + + flags |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; + if (esp_decrypt_chain_integ (vm, ptd, pd2, sa0, b, icv_sz, payload, + pd->current_length, &tag, + 0, &integ_len) < 0) + { + /* allocate buffer failed, will not add to frame and drop */ + b->error = node->errors[ESP_DECRYPT_ERROR_NO_BUFFERS]; + next[0] = ESP_DECRYPT_NEXT_DROP; + return 0; + } + } + else + esp_insert_esn (vm, sa0, pd2, &integ_len, &tag, &len, b, payload); + } + else + key_index = sa0->crypto_key_index; + +out: + /* crypto */ + payload += esp_sz; + len -= esp_sz; + iv = payload; + + if (ipsec_sa_is_set_IS_AEAD (sa0)) + { + esp_header_t *esp0; + u8 *scratch; + + /* + * construct the AAD and the nonce (Salt || IV) in a scratch + * space in front of the IP header. + */ + scratch = payload - esp_sz; + esp0 = (esp_header_t *) (scratch); + + scratch -= (sizeof (esp_aead_t) + pd->hdr_sz); + aad = scratch; + + esp_aad_fill (aad, esp0, sa0); + + /* + * we don't need to refer to the ESP header anymore so we + * can overwrite it with the salt and use the IV where it is + * to form the nonce = (Salt + IV) + */ + iv -= sizeof (sa0->salt); + clib_memcpy_fast (iv, &sa0->salt, sizeof (sa0->salt)); + + tag = payload + len; + } + + crypto_start_offset = (payload += iv_sz) - b->data; + crypto_len = len - iv_sz; + + if (pd->is_chain && (pd2->lb != b)) + { + /* buffer is chained */ + flags |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; + + crypto_len = esp_decrypt_chain_crypto (vm, ptd, pd2, sa0, b, icv_sz, + payload, + len - pd->iv_sz + pd->icv_sz, + &tag, 0); + } + + *async_pd = *pd; + *async_pd2 = *pd2; + pd->protect_index = current_protect_index; + next[0] = ESP_DECRYPT_NEXT_PENDING; + + /* for AEAD integ_len - crypto_len will be negative, it is ok since it + * is ignored by the engine. */ + return vnet_crypto_async_add_to_frame (vm, f, key_index, crypto_len, + integ_len - crypto_len, + crypto_start_offset, + integ_start_offset, + bi, async_next, iv, tag, aad, flags); +} + +static_always_inline void +esp_decrypt_post_crypto (vlib_main_t * vm, vlib_node_runtime_t * node, + esp_decrypt_packet_data_t * pd, + esp_decrypt_packet_data2_t * pd2, vlib_buffer_t * b, + u16 * next, int is_ip6, int is_tun, int is_async) +{ + ipsec_main_t *im = &ipsec_main; + ipsec_sa_t *sa0 = vec_elt_at_index (im->sad, pd->sa_index); + vlib_buffer_t *lb = b; + const u8 esp_sz = sizeof (esp_header_t); + const u8 tun_flags = IPSEC_SA_FLAG_IS_TUNNEL | IPSEC_SA_FLAG_IS_TUNNEL_V6; + u8 pad_length = 0, next_header = 0; + u16 icv_sz; + + /* + * redo the anti-reply check + * in this frame say we have sequence numbers, s, s+1, s+1, s+1 + * and s and s+1 are in the window. When we did the anti-replay + * check above we did so against the state of the window (W), + * after packet s-1. So each of the packets in the sequence will be + * accepted. + * This time s will be cheked against Ws-1, s+1 chceked against Ws + * (i.e. the window state is updated/advnaced) + * so this time the successive s+! packet will be dropped. + * This is a consequence of batching the decrypts. If the + * check-dcrypt-advance process was done for each packet it would + * be fine. But we batch the decrypts because it's much more efficient + * to do so in SW and if we offload to HW and the process is async. + * + * You're probably thinking, but this means an attacker can send the + * above sequence and cause VPP to perform decrpyts that will fail, + * and that's true. But if the attacker can determine s (a valid + * sequence number in the window) which is non-trivial, it can generate + * a sequence s, s+1, s+2, s+3, ... s+n and nothing will prevent any + * implementation, sequential or batching, from decrypting these. + */ + if (ipsec_sa_anti_replay_check (sa0, pd->seq)) + { + b->error = node->errors[ESP_DECRYPT_ERROR_REPLAY]; + next[0] = ESP_DECRYPT_NEXT_DROP; + return; + } + + ipsec_sa_anti_replay_advance (sa0, pd->seq); + + if (pd->is_chain) + { + lb = pd2->lb; + icv_sz = pd2->icv_removed ? 0 : pd->icv_sz; + if (pd2->free_buffer_index) + { + vlib_buffer_free_one (vm, pd2->free_buffer_index); + lb->next_buffer = 0; + } + if (lb->current_length < sizeof (esp_footer_t) + icv_sz) + { + /* esp footer is either splitted in two buffers or in the before + * last buffer */ + + vlib_buffer_t *before_last = b, *bp = b; + while (bp->flags & VLIB_BUFFER_NEXT_PRESENT) + { + before_last = bp; + bp = vlib_get_buffer (vm, bp->next_buffer); + } + u8 *bt = vlib_buffer_get_tail (before_last); + + if (lb->current_length == icv_sz) + { + esp_footer_t *f = (esp_footer_t *) (bt - sizeof (*f)); + pad_length = f->pad_length; + next_header = f->next_header; + } + else + { + pad_length = (bt - 1)[0]; + next_header = ((u8 *) vlib_buffer_get_current (lb))[0]; + } + } + else + { + esp_footer_t *f = + (esp_footer_t *) (lb->data + lb->current_data + + lb->current_length - sizeof (esp_footer_t) - + icv_sz); + pad_length = f->pad_length; + next_header = f->next_header; + } + } + else + { + icv_sz = pd->icv_sz; + esp_footer_t *f = + (esp_footer_t *) (lb->data + lb->current_data + lb->current_length - + sizeof (esp_footer_t) - icv_sz); + pad_length = f->pad_length; + next_header = f->next_header; + } + + u16 adv = pd->iv_sz + esp_sz; + u16 tail = sizeof (esp_footer_t) + pad_length + icv_sz; + u16 tail_orig = sizeof (esp_footer_t) + pad_length + pd->icv_sz; + b->flags &= ~VLIB_BUFFER_TOTAL_LENGTH_VALID; + + if ((pd->flags & tun_flags) == 0 && !is_tun) /* transport mode */ + { + u8 udp_sz = (is_ip6 == 0 && pd->flags & IPSEC_SA_FLAG_UDP_ENCAP) ? + sizeof (udp_header_t) : 0; + u16 ip_hdr_sz = pd->hdr_sz - udp_sz; + u8 *old_ip = b->data + pd->current_data - ip_hdr_sz - udp_sz; + u8 *ip = old_ip + adv + udp_sz; + + if (is_ip6 && ip_hdr_sz > 64) + memmove (ip, old_ip, ip_hdr_sz); + else + clib_memcpy_le64 (ip, old_ip, ip_hdr_sz); + + b->current_data = pd->current_data + adv - ip_hdr_sz; + b->current_length += ip_hdr_sz - adv; + esp_remove_tail (vm, b, lb, tail); + + if (is_ip6) + { + ip6_header_t *ip6 = (ip6_header_t *) ip; + u16 len = clib_net_to_host_u16 (ip6->payload_length); + len -= adv + tail_orig; + ip6->payload_length = clib_host_to_net_u16 (len); + ip6->protocol = next_header; + next[0] = ESP_DECRYPT_NEXT_IP6_INPUT; + } + else + { + ip4_header_t *ip4 = (ip4_header_t *) ip; + ip_csum_t sum = ip4->checksum; + u16 len = clib_net_to_host_u16 (ip4->length); + len = clib_host_to_net_u16 (len - adv - tail_orig - udp_sz); + sum = ip_csum_update (sum, ip4->protocol, next_header, + ip4_header_t, protocol); + sum = ip_csum_update (sum, ip4->length, len, ip4_header_t, length); + ip4->checksum = ip_csum_fold (sum); + ip4->protocol = next_header; + ip4->length = len; + next[0] = ESP_DECRYPT_NEXT_IP4_INPUT; + } + } + else + { + if (PREDICT_TRUE (next_header == IP_PROTOCOL_IP_IN_IP)) + { + next[0] = ESP_DECRYPT_NEXT_IP4_INPUT; + b->current_data = pd->current_data + adv; + b->current_length = pd->current_length - adv; + esp_remove_tail (vm, b, lb, tail); + } + else if (next_header == IP_PROTOCOL_IPV6) + { + next[0] = ESP_DECRYPT_NEXT_IP6_INPUT; + b->current_data = pd->current_data + adv; + b->current_length = pd->current_length - adv; + esp_remove_tail (vm, b, lb, tail); + } + else + { + if (is_tun && next_header == IP_PROTOCOL_GRE) + { + gre_header_t *gre; + + b->current_data = pd->current_data + adv; + b->current_length = pd->current_length - adv - tail; + + gre = vlib_buffer_get_current (b); + + vlib_buffer_advance (b, sizeof (*gre)); + + switch (clib_net_to_host_u16 (gre->protocol)) + { + case GRE_PROTOCOL_teb: + vnet_update_l2_len (b); + next[0] = ESP_DECRYPT_NEXT_L2_INPUT; + break; + case GRE_PROTOCOL_ip4: + next[0] = ESP_DECRYPT_NEXT_IP4_INPUT; + break; + case GRE_PROTOCOL_ip6: + next[0] = ESP_DECRYPT_NEXT_IP6_INPUT; + break; + default: + b->error = node->errors[ESP_DECRYPT_ERROR_UNSUP_PAYLOAD]; + next[0] = ESP_DECRYPT_NEXT_DROP; + break; + } + } + else + { + next[0] = ESP_DECRYPT_NEXT_DROP; + b->error = node->errors[ESP_DECRYPT_ERROR_UNSUP_PAYLOAD]; + return; + } + } + if (is_tun) + { + if (ipsec_sa_is_set_IS_PROTECT (sa0)) + { + /* + * There are two encap possibilities + * 1) the tunnel and ths SA are prodiving encap, i.e. it's + * MAC | SA-IP | TUN-IP | ESP | PAYLOAD + * implying the SA is in tunnel mode (on a tunnel interface) + * 2) only the tunnel provides encap + * MAC | TUN-IP | ESP | PAYLOAD + * implying the SA is in transport mode. + * + * For 2) we need only strip the tunnel encap and we're good. + * since the tunnel and crypto ecnap (int the tun=protect + * object) are the same and we verified above that these match + * for 1) we need to strip the SA-IP outer headers, to + * reveal the tunnel IP and then check that this matches + * the configured tunnel. + */ + const ipsec_tun_protect_t *itp; + + if (is_async) + itp = ipsec_tun_protect_get (pd->protect_index); + else + itp = + ipsec_tun_protect_get (vnet_buffer (b)-> + ipsec.protect_index); + + if (PREDICT_TRUE (next_header == IP_PROTOCOL_IP_IN_IP)) + { + const ip4_header_t *ip4; + + ip4 = vlib_buffer_get_current (b); + + if (!ip46_address_is_equal_v4 (&itp->itp_tun.src, + &ip4->dst_address) || + !ip46_address_is_equal_v4 (&itp->itp_tun.dst, + &ip4->src_address)) + { + next[0] = ESP_DECRYPT_NEXT_DROP; + b->error = node->errors[ESP_DECRYPT_ERROR_TUN_NO_PROTO]; + } + } + else if (next_header == IP_PROTOCOL_IPV6) + { + const ip6_header_t *ip6; + + ip6 = vlib_buffer_get_current (b); + + if (!ip46_address_is_equal_v6 (&itp->itp_tun.src, + &ip6->dst_address) || + !ip46_address_is_equal_v6 (&itp->itp_tun.dst, + &ip6->src_address)) + { + next[0] = ESP_DECRYPT_NEXT_DROP; + b->error = node->errors[ESP_DECRYPT_ERROR_TUN_NO_PROTO]; + } + } + } + } + } +} + +/* when submitting a frame is failed, drop all buffers in the frame */ +static_always_inline void +esp_async_recycle_failed_submit (vnet_crypto_async_frame_t * f, + vlib_buffer_t ** b, u16 * next) +{ + u32 n_drop = f->n_elts; + while (--n_drop) + { + (b - n_drop)[0]->error = ESP_DECRYPT_ERROR_CRYPTO_ENGINE_ERROR; + (next - n_drop)[0] = ESP_DECRYPT_NEXT_DROP; + } + vnet_crypto_async_reset_frame (f); +} + always_inline uword esp_decrypt_inline (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * from_frame, - int is_ip6, int is_tun) + int is_ip6, int is_tun, u16 async_next) { ipsec_main_t *im = &ipsec_main; u32 thread_index = vm->thread_index; @@ -325,20 +1040,26 @@ esp_decrypt_inline (vlib_main_t * vm, vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b = bufs; u16 nexts[VLIB_FRAME_SIZE], *next = nexts; esp_decrypt_packet_data_t pkt_data[VLIB_FRAME_SIZE], *pd = pkt_data; + esp_decrypt_packet_data2_t pkt_data2[VLIB_FRAME_SIZE], *pd2 = pkt_data2; esp_decrypt_packet_data_t cpd = { }; u32 current_sa_index = ~0, current_sa_bytes = 0, current_sa_pkts = 0; const u8 esp_sz = sizeof (esp_header_t); ipsec_sa_t *sa0 = 0; vnet_crypto_op_t _op, *op = &_op; - vnet_crypto_op_chunk_t *ch; vnet_crypto_op_t **crypto_ops = &ptd->crypto_ops; vnet_crypto_op_t **integ_ops = &ptd->integ_ops; + vnet_crypto_async_frame_t *async_frame = 0; + int is_async = im->async_mode; + vnet_crypto_async_op_id_t last_async_op = ~0; vlib_get_buffers (vm, from, b, n_left); - vec_reset_length (ptd->crypto_ops); - vec_reset_length (ptd->integ_ops); - vec_reset_length (ptd->chained_crypto_ops); - vec_reset_length (ptd->chained_integ_ops); + if (!is_async) + { + vec_reset_length (ptd->crypto_ops); + vec_reset_length (ptd->integ_ops); + vec_reset_length (ptd->chained_crypto_ops); + vec_reset_length (ptd->chained_integ_ops); + } vec_reset_length (ptd->chunks); clib_memset_u16 (nexts, -1, n_left); @@ -379,6 +1100,19 @@ esp_decrypt_inline (vlib_main_t * vm, cpd.iv_sz = sa0->crypto_iv_size; cpd.flags = sa0->flags; cpd.sa_index = current_sa_index; + + /* submit frame when op_id is different then the old one */ + if (is_async && last_async_op != sa0->crypto_async_dec_op_id) + { + if (async_frame && async_frame->n_elts) + { + if (vnet_crypto_async_submit_open_frame (vm, async_frame)) + esp_async_recycle_failed_submit (async_frame, b, next); + } + async_frame = + vnet_crypto_async_get_frame (vm, sa0->crypto_async_dec_op_id); + last_async_op = sa0->crypto_async_dec_op_id; + } } if (PREDICT_FALSE (~0 == sa0->decrypt_thread_index)) @@ -402,19 +1136,22 @@ esp_decrypt_inline (vlib_main_t * vm, pd->hdr_sz = pd->current_data - vnet_buffer (b[0])->l3_hdr_offset; payload = b[0]->data + pd->current_data; pd->seq = clib_host_to_net_u32 (((esp_header_t *) payload)->seq); - pd->free_buffer_index = 0; - pd->icv_removed = 0; + pd->is_chain = 0; + pd2->lb = b[0]; + pd2->free_buffer_index = 0; + pd2->icv_removed = 0; - pd->lb = b[0]; if (n_bufs > 1) { + pd->is_chain = 1; /* find last buffer in the chain */ - while (pd->lb->flags & VLIB_BUFFER_NEXT_PRESENT) - pd->lb = vlib_get_buffer (vm, pd->lb->next_buffer); + while (pd2->lb->flags & VLIB_BUFFER_NEXT_PRESENT) + pd2->lb = vlib_get_buffer (vm, pd2->lb->next_buffer); crypto_ops = &ptd->chained_crypto_ops; integ_ops = &ptd->chained_integ_ops; } + pd->current_length = b[0]->current_length; /* anti-reply check */ @@ -436,260 +1173,33 @@ esp_decrypt_inline (vlib_main_t * vm, current_sa_pkts += 1; current_sa_bytes += vlib_buffer_length_in_chain (vm, b[0]); - if (PREDICT_TRUE (sa0->integ_op_id != VNET_CRYPTO_OP_NONE)) + if (is_async) { - vnet_crypto_op_init (op, sa0->integ_op_id); - op->key_index = sa0->integ_key_index; - op->src = payload; - op->flags = VNET_CRYPTO_OP_FLAG_HMAC_CHECK; - op->user_data = b - bufs; - op->digest = payload + len; - op->digest_len = cpd.icv_sz; - op->len = len; - - if (pd->lb != b[0]) + int ret = esp_decrypt_prepare_async_frame (vm, node, ptd, + &async_frame, + sa0, payload, len, + cpd.icv_sz, + cpd.iv_sz, + pd, pd2, + from[b - bufs], + b[0], next, async_next); + if (PREDICT_FALSE (ret < 0)) { - /* buffer is chained */ - vlib_buffer_t *cb = b[0]; - op->len = pd->current_length; - - /* special case when ICV is splitted and needs to be reassembled - * first -> move it to the last buffer. Also take into account - * that ESN needs to be added after encrypted data and may or - * may not fit in the tail.*/ - if (pd->lb->current_length < cpd.icv_sz) - { - u8 extra_esn = 0; - op->digest = - esp_move_icv_esn (vm, b[0], pd, cpd.icv_sz, sa0, - &extra_esn, op); - - if (extra_esn) - { - /* esn is in the last buffer, that was unlinked from - * the chain */ - op->len = b[0]->current_length; - } - else - { - if (pd->lb == b[0]) - { - /* we now have a single buffer of crypto data, adjust - * the length (second buffer contains only ICV) */ - integ_ops = &ptd->integ_ops; - crypto_ops = &ptd->crypto_ops; - len = b[0]->current_length; - goto out; - } - } - } - else - op->digest = vlib_buffer_get_tail (pd->lb) - cpd.icv_sz; - - op->flags |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; - op->chunk_index = vec_len (ptd->chunks); - vec_add2 (ptd->chunks, ch, 1); - ch->len = pd->current_length; - ch->src = payload; - cb = vlib_get_buffer (vm, cb->next_buffer); - op->n_chunks = 1; - while (1) - { - vec_add2 (ptd->chunks, ch, 1); - op->n_chunks += 1; - ch->src = vlib_buffer_get_current (cb); - if (pd->lb == cb) - { - if (pd->icv_removed) - ch->len = cb->current_length; - else - ch->len = cb->current_length - cpd.icv_sz; - if (ipsec_sa_is_set_USE_ESN (sa0)) - { - u32 seq_hi = clib_host_to_net_u32 (sa0->seq_hi); - u8 tmp[ESP_MAX_ICV_SIZE], sz = sizeof (sa0->seq_hi); - u8 *esn; - vlib_buffer_t *tmp_b; - u16 space_left = vlib_buffer_space_left_at_end - (vm, pd->lb); - if (space_left < sz) - { - if (pd->icv_removed) - { - /* use pre-data area from the last bufer - that was removed from the chain */ - tmp_b = - vlib_get_buffer (vm, - pd->free_buffer_index); - esn = tmp_b->data - sz; - } - else - { - /* no space, need to allocate new buffer */ - u32 tmp_bi = 0; - if (1 != vlib_buffer_alloc (vm, &tmp_bi, 1)) - { - b[0]->error = node->errors - [ESP_DECRYPT_ERROR_NO_BUFFERS]; - next[0] = ESP_DECRYPT_NEXT_DROP; - goto next; - } - tmp_b = vlib_get_buffer (vm, tmp_bi); - esn = tmp_b->data; - pd->free_buffer_index = tmp_bi; - } - clib_memcpy_fast (esn, &seq_hi, sz); - - vec_add2 (ptd->chunks, ch, 1); - op->n_chunks += 1; - ch->src = esn; - ch->len = sz; - } - else - { - if (pd->icv_removed) - { - clib_memcpy_fast (vlib_buffer_get_tail - (pd->lb), &seq_hi, sz); - } - else - { - clib_memcpy_fast (tmp, op->digest, - ESP_MAX_ICV_SIZE); - clib_memcpy_fast (op->digest, &seq_hi, sz); - clib_memcpy_fast (op->digest + sz, tmp, - ESP_MAX_ICV_SIZE); - op->digest += sz; - } - ch->len += sz; - } - } - break; - } - else - ch->len = cb->current_length; - - if (!(cb->flags & VLIB_BUFFER_NEXT_PRESENT)) - break; - - cb = vlib_get_buffer (vm, cb->next_buffer); - } + esp_async_recycle_failed_submit (async_frame, b, next); + goto next; } - else - esp_insert_esn (vm, sa0, pd, op, &len, b[0], payload); - out: - vec_add_aligned (integ_ops[0], op, 1, CLIB_CACHE_LINE_BYTES); } - - payload += esp_sz; - len -= esp_sz; - - if (sa0->crypto_dec_op_id != VNET_CRYPTO_OP_NONE) - { - vnet_crypto_op_init (op, sa0->crypto_dec_op_id); - op->key_index = sa0->crypto_key_index; - op->iv = payload; - - if (ipsec_sa_is_set_IS_AEAD (sa0)) - { - esp_header_t *esp0; - esp_aead_t *aad; - u8 *scratch; - - /* - * construct the AAD and the nonce (Salt || IV) in a scratch - * space in front of the IP header. - */ - scratch = payload - esp_sz; - esp0 = (esp_header_t *) (scratch); - - scratch -= (sizeof (*aad) + pd->hdr_sz); - op->aad = scratch; - - esp_aad_fill (op, esp0, sa0); - - /* - * we don't need to refer to the ESP header anymore so we - * can overwrite it with the salt and use the IV where it is - * to form the nonce = (Salt + IV) - */ - op->iv -= sizeof (sa0->salt); - clib_memcpy_fast (op->iv, &sa0->salt, sizeof (sa0->salt)); - - op->tag = payload + len; - op->tag_len = 16; - } - op->src = op->dst = payload += cpd.iv_sz; - op->len = len - cpd.iv_sz; - op->user_data = b - bufs; - - if (pd->lb != b[0]) - { - /* buffer is chained */ - vlib_buffer_t *cb = b[0]; - op->flags |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; - op->chunk_index = vec_len (ptd->chunks); - vec_add2 (ptd->chunks, ch, 1); - ch->len = len - cpd.iv_sz + cpd.icv_sz; - ch->src = ch->dst = payload; - cb = vlib_get_buffer (vm, cb->next_buffer); - op->n_chunks = 1; - - while (1) - { - vec_add2 (ptd->chunks, ch, 1); - op->n_chunks += 1; - ch->src = ch->dst = vlib_buffer_get_current (cb); - if (pd->lb == cb) - { - if (ipsec_sa_is_set_IS_AEAD (sa0)) - { - if (pd->lb->current_length < cpd.icv_sz) - { - u16 dif = 0; - op->tag = - esp_move_icv (vm, b[0], pd, cpd.icv_sz, &dif); - - /* this chunk does not contain crypto data */ - op->n_chunks -= 1; - - /* and fix previous chunk's length as it might have - been changed */ - ASSERT (op->n_chunks > 0); - if (pd->lb == b[0]) - ch[-1].len -= dif; - else - ch[-1].len = pd->lb->current_length; - break; - } - else - op->tag = - vlib_buffer_get_tail (pd->lb) - cpd.icv_sz; - } - - if (pd->icv_removed) - ch->len = cb->current_length; - else - ch->len = cb->current_length - cpd.icv_sz; - } - else - ch->len = cb->current_length; - - if (!(cb->flags & VLIB_BUFFER_NEXT_PRESENT)) - break; - - cb = vlib_get_buffer (vm, cb->next_buffer); - } - } - - vec_add_aligned (crypto_ops[0], op, 1, CLIB_CACHE_LINE_BYTES); - } - + else + esp_decrypt_prepare_sync_op (vm, node, ptd, &crypto_ops, &integ_ops, + op, sa0, payload, len, cpd.icv_sz, + cpd.iv_sz, pd, pd2, b[0], next, + b - bufs); /* next */ next: n_left -= 1; next += 1; pd += 1; + pd2 += 1; b += 1; } @@ -698,15 +1208,35 @@ esp_decrypt_inline (vlib_main_t * vm, current_sa_index, current_sa_pkts, current_sa_bytes); - esp_process_ops (vm, node, ptd->integ_ops, bufs, nexts, - ESP_DECRYPT_ERROR_INTEG_ERROR); - esp_process_chained_ops (vm, node, ptd->chained_integ_ops, bufs, nexts, - ptd->chunks, ESP_DECRYPT_ERROR_INTEG_ERROR); + if (is_async) + { + if (async_frame && async_frame->n_elts) + { + if (vnet_crypto_async_submit_open_frame (vm, async_frame) < 0) + esp_async_recycle_failed_submit (async_frame, b, next); + } + + /* no post process in async */ + n_left = from_frame->n_vectors; + vlib_node_increment_counter (vm, node->node_index, + ESP_DECRYPT_ERROR_RX_PKTS, n_left); + vlib_buffer_enqueue_to_next (vm, node, from, nexts, n_left); - esp_process_ops (vm, node, ptd->crypto_ops, bufs, nexts, - ESP_DECRYPT_ERROR_DECRYPTION_FAILED); - esp_process_chained_ops (vm, node, ptd->chained_crypto_ops, bufs, nexts, - ptd->chunks, ESP_DECRYPT_ERROR_DECRYPTION_FAILED); + return n_left; + } + else + { + esp_process_ops (vm, node, ptd->integ_ops, bufs, nexts, + ESP_DECRYPT_ERROR_INTEG_ERROR); + esp_process_chained_ops (vm, node, ptd->chained_integ_ops, bufs, nexts, + ptd->chunks, ESP_DECRYPT_ERROR_INTEG_ERROR); + + esp_process_ops (vm, node, ptd->crypto_ops, bufs, nexts, + ESP_DECRYPT_ERROR_DECRYPTION_FAILED); + esp_process_chained_ops (vm, node, ptd->chained_crypto_ops, bufs, nexts, + ptd->chunks, + ESP_DECRYPT_ERROR_DECRYPTION_FAILED); + } /* Post decryption ronud - adjust packet data start and length and next node */ @@ -714,13 +1244,11 @@ esp_decrypt_inline (vlib_main_t * vm, n_left = from_frame->n_vectors; next = nexts; pd = pkt_data; + pd2 = pkt_data2; b = bufs; while (n_left) { - const u8 tun_flags = IPSEC_SA_FLAG_IS_TUNNEL | - IPSEC_SA_FLAG_IS_TUNNEL_V6; - if (n_left >= 2) { void *data = b[1]->data + pd[1].current_data; @@ -737,253 +1265,83 @@ esp_decrypt_inline (vlib_main_t * vm, CLIB_CACHE_LINE_BYTES * 2, LOAD); } - if (next[0] < ESP_DECRYPT_N_NEXT) - goto trace; - - sa0 = vec_elt_at_index (im->sad, pd->sa_index); + if (next[0] >= ESP_DECRYPT_N_NEXT) + esp_decrypt_post_crypto (vm, node, pd, pd2, b[0], next, is_ip6, + is_tun, 0); - /* - * redo the anti-reply check - * in this frame say we have sequence numbers, s, s+1, s+1, s+1 - * and s and s+1 are in the window. When we did the anti-replay - * check above we did so against the state of the window (W), - * after packet s-1. So each of the packets in the sequence will be - * accepted. - * This time s will be cheked against Ws-1, s+1 chceked against Ws - * (i.e. the window state is updated/advnaced) - * so this time the successive s+! packet will be dropped. - * This is a consequence of batching the decrypts. If the - * check-dcrypt-advance process was done for each packet it would - * be fine. But we batch the decrypts because it's much more efficient - * to do so in SW and if we offload to HW and the process is async. - * - * You're probably thinking, but this means an attacker can send the - * above sequence and cause VPP to perform decrpyts that will fail, - * and that's true. But if the attacker can determine s (a valid - * sequence number in the window) which is non-trivial, it can generate - * a sequence s, s+1, s+2, s+3, ... s+n and nothing will prevent any - * implementation, sequential or batching, from decrypting these. - */ - if (ipsec_sa_anti_replay_check (sa0, pd->seq)) + /* trace: */ + if (PREDICT_FALSE (b[0]->flags & VLIB_BUFFER_IS_TRACED)) { - b[0]->error = node->errors[ESP_DECRYPT_ERROR_REPLAY]; - next[0] = ESP_DECRYPT_NEXT_DROP; - goto trace; + esp_decrypt_trace_t *tr; + tr = vlib_add_trace (vm, node, b[0], sizeof (*tr)); + sa0 = pool_elt_at_index (im->sad, + vnet_buffer (b[0])->ipsec.sad_index); + tr->crypto_alg = sa0->crypto_alg; + tr->integ_alg = sa0->integ_alg; + tr->seq = pd->seq; + tr->sa_seq = sa0->last_seq; + tr->sa_seq_hi = sa0->seq_hi; } - ipsec_sa_anti_replay_advance (sa0, pd->seq); - - u8 pad_length = 0, next_header = 0; - u16 icv_sz = pd->icv_removed ? 0 : pd->icv_sz; + /* next */ + n_left -= 1; + next += 1; + pd += 1; + pd2 += 1; + b += 1; + } - if (pd->free_buffer_index) - vlib_buffer_free_one (vm, pd->free_buffer_index); + n_left = from_frame->n_vectors; + vlib_node_increment_counter (vm, node->node_index, + ESP_DECRYPT_ERROR_RX_PKTS, n_left); - if (pd->lb->current_length < sizeof (esp_footer_t) + icv_sz) - { - /* esp footer is either splitted in two buffers or in the before - * last buffer */ + vlib_buffer_enqueue_to_next (vm, node, from, nexts, n_left); - vlib_buffer_t *before_last = b[0], *bp = b[0]; - while (bp->flags & VLIB_BUFFER_NEXT_PRESENT) - { - before_last = bp; - bp = vlib_get_buffer (vm, bp->next_buffer); - } - u8 *bt = vlib_buffer_get_tail (before_last); + return n_left; +} - if (pd->lb->current_length == icv_sz) - { - esp_footer_t *f = (esp_footer_t *) (bt - sizeof (*f)); - pad_length = f->pad_length; - next_header = f->next_header; - } - else - { - pad_length = (bt - 1)[0]; - next_header = ((u8 *) vlib_buffer_get_current (pd->lb))[0]; - } - } - else - { - esp_footer_t *f = - (esp_footer_t *) (pd->lb->data + pd->lb->current_data + - pd->lb->current_length - sizeof (esp_footer_t) - - icv_sz); - pad_length = f->pad_length; - next_header = f->next_header; - } +always_inline uword +esp_decrypt_post_inline (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame, int is_ip6, int is_tun) +{ + ipsec_main_t *im = &ipsec_main; + u32 *from = vlib_frame_vector_args (from_frame); + u32 n_left = from_frame->n_vectors; + vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b = bufs; + u16 nexts[VLIB_FRAME_SIZE], *next = nexts; + vlib_get_buffers (vm, from, b, n_left); - u16 adv = pd->iv_sz + esp_sz; - u16 tail = sizeof (esp_footer_t) + pad_length + icv_sz; - u16 tail_orig = sizeof (esp_footer_t) + pad_length + pd->icv_sz; - b[0]->flags &= ~VLIB_BUFFER_TOTAL_LENGTH_VALID; + while (n_left > 0) + { + esp_decrypt_packet_data_t *pd = &(esp_post_data (b[0]))->decrypt_data; - if ((pd->flags & tun_flags) == 0 && !is_tun) /* transport mode */ + if (n_left > 2) { - u8 udp_sz = (is_ip6 == 0 && pd->flags & IPSEC_SA_FLAG_UDP_ENCAP) ? - sizeof (udp_header_t) : 0; - u16 ip_hdr_sz = pd->hdr_sz - udp_sz; - u8 *old_ip = b[0]->data + pd->current_data - ip_hdr_sz - udp_sz; - u8 *ip = old_ip + adv + udp_sz; - - if (is_ip6 && ip_hdr_sz > 64) - memmove (ip, old_ip, ip_hdr_sz); - else - clib_memcpy_le64 (ip, old_ip, ip_hdr_sz); - - b[0]->current_data = pd->current_data + adv - ip_hdr_sz; - b[0]->current_length += ip_hdr_sz - adv; - esp_remove_tail (vm, b[0], pd->lb, tail); - - if (is_ip6) - { - ip6_header_t *ip6 = (ip6_header_t *) ip; - u16 len = clib_net_to_host_u16 (ip6->payload_length); - len -= adv + tail_orig; - ip6->payload_length = clib_host_to_net_u16 (len); - ip6->protocol = next_header; - next[0] = ESP_DECRYPT_NEXT_IP6_INPUT; - } - else - { - ip4_header_t *ip4 = (ip4_header_t *) ip; - ip_csum_t sum = ip4->checksum; - u16 len = clib_net_to_host_u16 (ip4->length); - len = clib_host_to_net_u16 (len - adv - tail_orig - udp_sz); - sum = ip_csum_update (sum, ip4->protocol, next_header, - ip4_header_t, protocol); - sum = ip_csum_update (sum, ip4->length, len, - ip4_header_t, length); - ip4->checksum = ip_csum_fold (sum); - ip4->protocol = next_header; - ip4->length = len; - next[0] = ESP_DECRYPT_NEXT_IP4_INPUT; - } + vlib_prefetch_buffer_header (b[2], LOAD); + vlib_prefetch_buffer_header (b[1], LOAD); } + + if (!pd->is_chain) + esp_decrypt_post_crypto (vm, node, pd, 0, b[0], next, is_ip6, is_tun, + 1); else { - if (PREDICT_TRUE (next_header == IP_PROTOCOL_IP_IN_IP)) - { - next[0] = ESP_DECRYPT_NEXT_IP4_INPUT; - b[0]->current_data = pd->current_data + adv; - b[0]->current_length = pd->current_length - adv; - esp_remove_tail (vm, b[0], pd->lb, tail); - } - else if (next_header == IP_PROTOCOL_IPV6) - { - next[0] = ESP_DECRYPT_NEXT_IP6_INPUT; - b[0]->current_data = pd->current_data + adv; - b[0]->current_length = pd->current_length - adv; - esp_remove_tail (vm, b[0], pd->lb, tail); - } - else - { - if (is_tun && next_header == IP_PROTOCOL_GRE) - { - gre_header_t *gre; - - b[0]->current_data = pd->current_data + adv; - b[0]->current_length = pd->current_length - adv - tail; - - gre = vlib_buffer_get_current (b[0]); - - vlib_buffer_advance (b[0], sizeof (*gre)); - - switch (clib_net_to_host_u16 (gre->protocol)) - { - case GRE_PROTOCOL_teb: - vnet_update_l2_len (b[0]); - next[0] = ESP_DECRYPT_NEXT_L2_INPUT; - break; - case GRE_PROTOCOL_ip4: - next[0] = ESP_DECRYPT_NEXT_IP4_INPUT; - break; - case GRE_PROTOCOL_ip6: - next[0] = ESP_DECRYPT_NEXT_IP6_INPUT; - break; - default: - b[0]->error = - node->errors[ESP_DECRYPT_ERROR_UNSUP_PAYLOAD]; - next[0] = ESP_DECRYPT_NEXT_DROP; - break; - } - } - else - { - next[0] = ESP_DECRYPT_NEXT_DROP; - b[0]->error = node->errors[ESP_DECRYPT_ERROR_UNSUP_PAYLOAD]; - goto trace; - } - } - if (is_tun) - { - if (ipsec_sa_is_set_IS_PROTECT (sa0)) - { - /* - * There are two encap possibilities - * 1) the tunnel and ths SA are prodiving encap, i.e. it's - * MAC | SA-IP | TUN-IP | ESP | PAYLOAD - * implying the SA is in tunnel mode (on a tunnel interface) - * 2) only the tunnel provides encap - * MAC | TUN-IP | ESP | PAYLOAD - * implying the SA is in transport mode. - * - * For 2) we need only strip the tunnel encap and we're good. - * since the tunnel and crypto ecnap (int the tun=protect - * object) are the same and we verified above that these match - * for 1) we need to strip the SA-IP outer headers, to - * reveal the tunnel IP and then check that this matches - * the configured tunnel. - */ - const ipsec_tun_protect_t *itp; - - itp = ipsec_tun_protect_get - (vnet_buffer (b[0])->ipsec.protect_index); - - if (PREDICT_TRUE (next_header == IP_PROTOCOL_IP_IN_IP)) - { - const ip4_header_t *ip4; - - ip4 = vlib_buffer_get_current (b[0]); - - if (!ip46_address_is_equal_v4 (&itp->itp_tun.src, - &ip4->dst_address) || - !ip46_address_is_equal_v4 (&itp->itp_tun.dst, - &ip4->src_address)) - { - next[0] = ESP_DECRYPT_NEXT_DROP; - b[0]->error = - node->errors[ESP_DECRYPT_ERROR_TUN_NO_PROTO]; - } - } - else if (next_header == IP_PROTOCOL_IPV6) - { - const ip6_header_t *ip6; - - ip6 = vlib_buffer_get_current (b[0]); - - if (!ip46_address_is_equal_v6 (&itp->itp_tun.src, - &ip6->dst_address) || - !ip46_address_is_equal_v6 (&itp->itp_tun.dst, - &ip6->src_address)) - { - next[0] = ESP_DECRYPT_NEXT_DROP; - b[0]->error = - node->errors[ESP_DECRYPT_ERROR_TUN_NO_PROTO]; - } - } - } - } + esp_decrypt_packet_data2_t *pd2 = esp_post_data2 (b[0]); + esp_decrypt_post_crypto (vm, node, pd, pd2, b[0], next, is_ip6, + is_tun, 1); } - trace: + /*trace: */ if (PREDICT_FALSE (b[0]->flags & VLIB_BUFFER_IS_TRACED)) { + ipsec_sa_t *sa0 = pool_elt_at_index (im->sad, pd->sa_index); esp_decrypt_trace_t *tr; + esp_decrypt_packet_data_t *async_pd = + &(esp_post_data (b[0]))->decrypt_data; tr = vlib_add_trace (vm, node, b[0], sizeof (*tr)); - sa0 = pool_elt_at_index (im->sad, - vnet_buffer (b[0])->ipsec.sad_index); + sa0 = pool_elt_at_index (im->sad, async_pd->sa_index); + tr->crypto_alg = sa0->crypto_alg; tr->integ_alg = sa0->integ_alg; tr->seq = pd->seq; @@ -991,16 +1349,14 @@ esp_decrypt_inline (vlib_main_t * vm, tr->sa_seq_hi = sa0->seq_hi; } - /* next */ - n_left -= 1; - next += 1; - pd += 1; - b += 1; + n_left--; + next++; + b++; } n_left = from_frame->n_vectors; vlib_node_increment_counter (vm, node->node_index, - ESP_DECRYPT_ERROR_RX_PKTS, n_left); + ESP_DECRYPT_ERROR_RX_POST_PKTS, n_left); vlib_buffer_enqueue_to_next (vm, node, from, nexts, n_left); @@ -1011,31 +1367,80 @@ VLIB_NODE_FN (esp4_decrypt_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * from_frame) { - return esp_decrypt_inline (vm, node, from_frame, 0, 0); + return esp_decrypt_inline (vm, node, from_frame, 0, 0, + esp_decrypt_async_next.esp4_post_next); +} + +VLIB_NODE_FN (esp4_decrypt_post_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return esp_decrypt_post_inline (vm, node, from_frame, 0, 0); } VLIB_NODE_FN (esp4_decrypt_tun_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * from_frame) { - return esp_decrypt_inline (vm, node, from_frame, 0, 1); + return esp_decrypt_inline (vm, node, from_frame, 0, 1, + esp_decrypt_async_next.esp4_tun_post_next); +} + +VLIB_NODE_FN (esp4_decrypt_tun_post_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return esp_decrypt_post_inline (vm, node, from_frame, 0, 1); } VLIB_NODE_FN (esp6_decrypt_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * from_frame) { - return esp_decrypt_inline (vm, node, from_frame, 1, 0); + return esp_decrypt_inline (vm, node, from_frame, 1, 0, + esp_decrypt_async_next.esp6_post_next); +} + +VLIB_NODE_FN (esp6_decrypt_post_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return esp_decrypt_post_inline (vm, node, from_frame, 1, 0); } VLIB_NODE_FN (esp6_decrypt_tun_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * from_frame) { - return esp_decrypt_inline (vm, node, from_frame, 1, 1); + return esp_decrypt_inline (vm, node, from_frame, 1, 1, + esp_decrypt_async_next.esp6_tun_post_next); +} + +VLIB_NODE_FN (esp6_decrypt_tun_post_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return esp_decrypt_post_inline (vm, node, from_frame, 1, 1); +} + +VLIB_NODE_FN (esp_decrypt_pending_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return from_frame->n_vectors; } /* *INDENT-OFF* */ +VLIB_REGISTER_NODE (esp_decrypt_pending_node) = { + .name = "esp-decrypt-pending", + .vector_size = sizeof (u32), + .type = VLIB_NODE_TYPE_INTERNAL, + + .n_next_nodes = 0 +}; +/* *INDENT-ON* */ + +/* *INDENT-OFF* */ VLIB_REGISTER_NODE (esp4_decrypt_node) = { .name = "esp4-decrypt", .vector_size = sizeof (u32), @@ -1052,9 +1457,22 @@ VLIB_REGISTER_NODE (esp4_decrypt_node) = { [ESP_DECRYPT_NEXT_IP6_INPUT] = "ip6-input", [ESP_DECRYPT_NEXT_L2_INPUT] = "l2-input", [ESP_DECRYPT_NEXT_HANDOFF] = "esp4-decrypt-handoff", + [ESP_DECRYPT_NEXT_PENDING] = "esp-decrypt-pending" }, }; +VLIB_REGISTER_NODE (esp4_decrypt_post_node) = { + .name = "esp4-decrypt-post", + .vector_size = sizeof (u32), + .format_trace = format_esp_decrypt_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + + .n_errors = ARRAY_LEN(esp_decrypt_error_strings), + .error_strings = esp_decrypt_error_strings, + + .sibling_of = "esp4-decrypt", +}; + VLIB_REGISTER_NODE (esp6_decrypt_node) = { .name = "esp6-decrypt", .vector_size = sizeof (u32), @@ -1071,9 +1489,22 @@ VLIB_REGISTER_NODE (esp6_decrypt_node) = { [ESP_DECRYPT_NEXT_IP6_INPUT] = "ip6-input", [ESP_DECRYPT_NEXT_L2_INPUT] = "l2-input", [ESP_DECRYPT_NEXT_HANDOFF]= "esp6-decrypt-handoff", + [ESP_DECRYPT_NEXT_PENDING] = "esp-decrypt-pending" }, }; +VLIB_REGISTER_NODE (esp6_decrypt_post_node) = { + .name = "esp6-decrypt-post", + .vector_size = sizeof (u32), + .format_trace = format_esp_decrypt_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + + .n_errors = ARRAY_LEN(esp_decrypt_error_strings), + .error_strings = esp_decrypt_error_strings, + + .sibling_of = "esp6-decrypt", +}; + VLIB_REGISTER_NODE (esp4_decrypt_tun_node) = { .name = "esp4-decrypt-tun", .vector_size = sizeof (u32), @@ -1088,9 +1519,22 @@ VLIB_REGISTER_NODE (esp4_decrypt_tun_node) = { [ESP_DECRYPT_NEXT_IP6_INPUT] = "ip6-input", [ESP_DECRYPT_NEXT_L2_INPUT] = "l2-input", [ESP_DECRYPT_NEXT_HANDOFF] = "esp4-decrypt-tun-handoff", + [ESP_DECRYPT_NEXT_PENDING] = "esp-decrypt-pending" }, }; +VLIB_REGISTER_NODE (esp4_decrypt_tun_post_node) = { + .name = "esp4-decrypt-tun-post", + .vector_size = sizeof (u32), + .format_trace = format_esp_decrypt_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + + .n_errors = ARRAY_LEN(esp_decrypt_error_strings), + .error_strings = esp_decrypt_error_strings, + + .sibling_of = "esp4-decrypt-tun", +}; + VLIB_REGISTER_NODE (esp6_decrypt_tun_node) = { .name = "esp6-decrypt-tun", .vector_size = sizeof (u32), @@ -1105,8 +1549,21 @@ VLIB_REGISTER_NODE (esp6_decrypt_tun_node) = { [ESP_DECRYPT_NEXT_IP6_INPUT] = "ip6-input", [ESP_DECRYPT_NEXT_L2_INPUT] = "l2-input", [ESP_DECRYPT_NEXT_HANDOFF]= "esp6-decrypt-tun-handoff", + [ESP_DECRYPT_NEXT_PENDING] = "esp-decrypt-pending" }, }; + +VLIB_REGISTER_NODE (esp6_decrypt_tun_post_node) = { + .name = "esp6-decrypt-tun-post", + .vector_size = sizeof (u32), + .format_trace = format_esp_decrypt_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + + .n_errors = ARRAY_LEN(esp_decrypt_error_strings), + .error_strings = esp_decrypt_error_strings, + + .sibling_of = "esp6-decrypt-tun", +}; /* *INDENT-ON* */ /* diff --git a/src/vnet/ipsec/esp_encrypt.c b/src/vnet/ipsec/esp_encrypt.c index e40d30d7b3e..dce887f6fa8 100644 --- a/src/vnet/ipsec/esp_encrypt.c +++ b/src/vnet/ipsec/esp_encrypt.c @@ -28,6 +28,7 @@ #define foreach_esp_encrypt_next \ _(DROP, "error-drop") \ +_(PENDING, "pending") \ _(HANDOFF, "handoff") \ _(INTERFACE_OUTPUT, "interface-output") @@ -41,8 +42,10 @@ typedef enum #define foreach_esp_encrypt_error \ _(RX_PKTS, "ESP pkts received") \ + _(POST_RX_PKTS, "ESP-post pkts received") \ _(SEQ_CYCLED, "sequence number cycled (packet dropped)") \ _(CRYPTO_ENGINE_ERROR, "crypto engine error (packet dropped)") \ + _(CRYPTO_QUEUE_FULL, "crypto queue full (packet dropped)") \ _(NO_BUFFERS, "no buffers (packet dropped)") \ typedef enum @@ -70,6 +73,11 @@ typedef struct ipsec_integ_alg_t integ_alg; } esp_encrypt_trace_t; +typedef struct +{ + u32 next_index; +} esp_encrypt_post_trace_t; + /* packet trace format function */ static u8 * format_esp_encrypt_trace (u8 * s, va_list * args) @@ -88,6 +96,17 @@ format_esp_encrypt_trace (u8 * s, va_list * args) return s; } +static u8 * +format_esp_post_encrypt_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + esp_encrypt_post_trace_t *t = va_arg (*args, esp_encrypt_post_trace_t *); + + s = format (s, "esp-post: next node index %u", t->next_index); + return s; +} + /* pad packet in input buffer */ static_always_inline u8 * esp_add_footer_and_icv (vlib_main_t * vm, vlib_buffer_t ** last, @@ -276,9 +295,281 @@ typedef struct STATIC_ASSERT_SIZEOF (esp_gcm_nonce_t, 12); +static_always_inline u32 +esp_encrypt_chain_crypto (vlib_main_t * vm, ipsec_per_thread_data_t * ptd, + ipsec_sa_t * sa0, vlib_buffer_t * b, + vlib_buffer_t * lb, u8 icv_sz, u8 * start, + u32 start_len, u16 * n_ch) +{ + vnet_crypto_op_chunk_t *ch; + vlib_buffer_t *cb = b; + u32 n_chunks = 1; + u32 total_len; + vec_add2 (ptd->chunks, ch, 1); + total_len = ch->len = start_len; + ch->src = ch->dst = start; + cb = vlib_get_buffer (vm, cb->next_buffer); + + while (1) + { + vec_add2 (ptd->chunks, ch, 1); + n_chunks += 1; + if (lb == cb) + total_len += ch->len = cb->current_length - icv_sz; + else + total_len += ch->len = cb->current_length; + ch->src = ch->dst = vlib_buffer_get_current (cb); + + if (!(cb->flags & VLIB_BUFFER_NEXT_PRESENT)) + break; + + cb = vlib_get_buffer (vm, cb->next_buffer); + } + + if (n_ch) + *n_ch = n_chunks; + + return total_len; +} + +static_always_inline u32 +esp_encrypt_chain_integ (vlib_main_t * vm, ipsec_per_thread_data_t * ptd, + ipsec_sa_t * sa0, vlib_buffer_t * b, + vlib_buffer_t * lb, u8 icv_sz, u8 * start, + u32 start_len, u8 * digest, u16 * n_ch) +{ + vnet_crypto_op_chunk_t *ch; + vlib_buffer_t *cb = b; + u32 n_chunks = 1; + u32 total_len; + vec_add2 (ptd->chunks, ch, 1); + total_len = ch->len = start_len; + ch->src = start; + cb = vlib_get_buffer (vm, cb->next_buffer); + + while (1) + { + vec_add2 (ptd->chunks, ch, 1); + n_chunks += 1; + if (lb == cb) + { + total_len += ch->len = cb->current_length - icv_sz; + if (ipsec_sa_is_set_USE_ESN (sa0)) + { + u32 seq_hi = clib_net_to_host_u32 (sa0->seq_hi); + clib_memcpy_fast (digest, &seq_hi, sizeof (seq_hi)); + ch->len += sizeof (seq_hi); + total_len += sizeof (seq_hi); + } + } + else + total_len += ch->len = cb->current_length; + ch->src = vlib_buffer_get_current (cb); + + if (!(cb->flags & VLIB_BUFFER_NEXT_PRESENT)) + break; + + cb = vlib_get_buffer (vm, cb->next_buffer); + } + + if (n_ch) + *n_ch = n_chunks; + + return total_len; +} + +always_inline void +esp_prepare_sync_op (vlib_main_t * vm, ipsec_per_thread_data_t * ptd, + vnet_crypto_op_t ** crypto_ops, + vnet_crypto_op_t ** integ_ops, ipsec_sa_t * sa0, + u8 * payload, u16 payload_len, u8 iv_sz, u8 icv_sz, + vlib_buffer_t ** bufs, vlib_buffer_t ** b, + vlib_buffer_t * lb, u32 hdr_len, esp_header_t * esp, + esp_gcm_nonce_t * nonce) +{ + if (sa0->crypto_enc_op_id) + { + vnet_crypto_op_t *op; + vec_add2_aligned (crypto_ops[0], op, 1, CLIB_CACHE_LINE_BYTES); + vnet_crypto_op_init (op, sa0->crypto_enc_op_id); + + op->src = op->dst = payload; + op->key_index = sa0->crypto_key_index; + op->len = payload_len - icv_sz; + op->user_data = b - bufs; + + if (ipsec_sa_is_set_IS_AEAD (sa0)) + { + /* + * construct the AAD in a scratch space in front + * of the IP header. + */ + op->aad = payload - hdr_len - sizeof (esp_aead_t); + op->aad_len = esp_aad_fill (op->aad, esp, sa0); + + op->tag = payload + op->len; + op->tag_len = 16; + + u64 *iv = (u64 *) (payload - iv_sz); + nonce->salt = sa0->salt; + nonce->iv = *iv = clib_host_to_net_u64 (sa0->gcm_iv_counter++); + op->iv = (u8 *) nonce; + } + else + { + op->iv = payload - iv_sz; + op->flags = VNET_CRYPTO_OP_FLAG_INIT_IV; + } + + if (lb != b[0]) + { + /* is chained */ + op->flags |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; + op->chunk_index = vec_len (ptd->chunks); + op->tag = vlib_buffer_get_tail (lb) - icv_sz; + esp_encrypt_chain_crypto (vm, ptd, sa0, b[0], lb, icv_sz, payload, + payload_len, &op->n_chunks); + } + } + + if (sa0->integ_op_id) + { + vnet_crypto_op_t *op; + vec_add2_aligned (integ_ops[0], op, 1, CLIB_CACHE_LINE_BYTES); + vnet_crypto_op_init (op, sa0->integ_op_id); + op->src = payload - iv_sz - sizeof (esp_header_t); + op->digest = payload + payload_len - icv_sz; + op->key_index = sa0->integ_key_index; + op->digest_len = icv_sz; + op->len = payload_len - icv_sz + iv_sz + sizeof (esp_header_t); + op->user_data = b - bufs; + + if (lb != b[0]) + { + /* is chained */ + op->flags |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; + op->chunk_index = vec_len (ptd->chunks); + op->digest = vlib_buffer_get_tail (lb) - icv_sz; + + esp_encrypt_chain_integ (vm, ptd, sa0, b[0], lb, icv_sz, + payload - iv_sz - sizeof (esp_header_t), + payload_len + iv_sz + + sizeof (esp_header_t), op->digest, + &op->n_chunks); + } + else if (ipsec_sa_is_set_USE_ESN (sa0)) + { + u32 seq_hi = clib_net_to_host_u32 (sa0->seq_hi); + clib_memcpy_fast (op->digest, &seq_hi, sizeof (seq_hi)); + op->len += sizeof (seq_hi); + } + } +} + +static_always_inline int +esp_prepare_async_frame (vlib_main_t * vm, ipsec_per_thread_data_t * ptd, + vnet_crypto_async_frame_t ** async_frame, + ipsec_sa_t * sa, vlib_buffer_t * b, + esp_header_t * esp, u8 * payload, u32 payload_len, + u8 iv_sz, u8 icv_sz, u32 bi, u16 * next, u32 hdr_len, + u16 async_next, vlib_buffer_t * lb) +{ + esp_post_data_t *post = esp_post_data (b); + u8 *tag, *iv, *aad = 0; + u8 flag = 0; + u32 key_index; + i16 crypto_start_offset, integ_start_offset = 0; + u16 crypto_total_len, integ_total_len; + + post->next_index = next[0]; + next[0] = ESP_ENCRYPT_NEXT_PENDING; + + /* crypto */ + crypto_start_offset = payload - b->data; + crypto_total_len = integ_total_len = payload_len - icv_sz; + tag = payload + crypto_total_len; + + /* aead */ + if (ipsec_sa_is_set_IS_AEAD (sa)) + { + esp_gcm_nonce_t *nonce; + u64 *pkt_iv = (u64 *) (payload - iv_sz); + + aad = payload - hdr_len - sizeof (esp_aead_t); + esp_aad_fill (aad, esp, sa); + nonce = (esp_gcm_nonce_t *) (aad - sizeof (*nonce)); + nonce->salt = sa->salt; + nonce->iv = *pkt_iv = clib_host_to_net_u64 (sa->gcm_iv_counter++); + iv = (u8 *) nonce; + key_index = sa->crypto_key_index; + + if (lb != b) + { + /* chain */ + flag |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; + tag = vlib_buffer_get_tail (lb) - icv_sz; + crypto_total_len = esp_encrypt_chain_crypto (vm, ptd, sa, b, lb, + icv_sz, payload, + payload_len, 0); + } + goto out; + } + + /* cipher then hash */ + iv = payload - iv_sz; + integ_start_offset = crypto_start_offset - iv_sz - sizeof (esp_header_t); + integ_total_len += iv_sz + sizeof (esp_header_t); + flag |= VNET_CRYPTO_OP_FLAG_INIT_IV; + key_index = sa->linked_key_index; + + if (b != lb) + { + flag |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; + crypto_total_len = esp_encrypt_chain_crypto (vm, ptd, sa, b, lb, + icv_sz, payload, + payload_len, 0); + tag = vlib_buffer_get_tail (lb) - icv_sz; + integ_total_len = esp_encrypt_chain_integ (vm, ptd, sa, b, lb, icv_sz, + payload - iv_sz - + sizeof (esp_header_t), + payload_len + iv_sz + + sizeof (esp_header_t), + tag, 0); + } + else if (ipsec_sa_is_set_USE_ESN (sa) && !ipsec_sa_is_set_IS_AEAD (sa)) + { + u32 seq_hi = clib_net_to_host_u32 (sa->seq_hi); + clib_memcpy_fast (tag, &seq_hi, sizeof (seq_hi)); + integ_total_len += sizeof (seq_hi); + } + +out: + return vnet_crypto_async_add_to_frame (vm, async_frame, key_index, + crypto_total_len, + integ_total_len - crypto_total_len, + crypto_start_offset, + integ_start_offset, bi, async_next, + iv, tag, aad, flag); +} + +/* when submitting a frame is failed, drop all buffers in the frame */ +static_always_inline void +esp_async_recycle_failed_submit (vnet_crypto_async_frame_t * f, + vlib_buffer_t ** b, u16 * next) +{ + u32 n_drop = f->n_elts; + while (--n_drop) + { + (b - n_drop)[0]->error = ESP_ENCRYPT_ERROR_CRYPTO_ENGINE_ERROR; + (next - n_drop)[0] = ESP_ENCRYPT_NEXT_DROP; + } + vnet_crypto_async_reset_frame (f); +} + always_inline uword esp_encrypt_inline (vlib_main_t * vm, vlib_node_runtime_t * node, - vlib_frame_t * frame, int is_ip6, int is_tun) + vlib_frame_t * frame, int is_ip6, int is_tun, + u16 async_next) { ipsec_main_t *im = &ipsec_main; ipsec_per_thread_data_t *ptd = vec_elt_at_index (im->ptd, vm->thread_index); @@ -293,16 +584,21 @@ esp_encrypt_inline (vlib_main_t * vm, vlib_node_runtime_t * node, u32 current_sa_bytes = 0, spi = 0; u8 block_sz = 0, iv_sz = 0, icv_sz = 0; ipsec_sa_t *sa0 = 0; - vnet_crypto_op_chunk_t *ch; vlib_buffer_t *lb; vnet_crypto_op_t **crypto_ops = &ptd->crypto_ops; vnet_crypto_op_t **integ_ops = &ptd->integ_ops; + vnet_crypto_async_frame_t *async_frame = 0; + int is_async = im->async_mode; + vnet_crypto_async_op_id_t last_async_op = ~0; vlib_get_buffers (vm, from, b, n_left); - vec_reset_length (ptd->crypto_ops); - vec_reset_length (ptd->integ_ops); - vec_reset_length (ptd->chained_crypto_ops); - vec_reset_length (ptd->chained_integ_ops); + if (!is_async) + { + vec_reset_length (ptd->crypto_ops); + vec_reset_length (ptd->integ_ops); + vec_reset_length (ptd->chained_crypto_ops); + vec_reset_length (ptd->chained_integ_ops); + } vec_reset_length (ptd->chunks); while (n_left > 0) @@ -351,6 +647,20 @@ esp_encrypt_inline (vlib_main_t * vm, vlib_node_runtime_t * node, block_sz = sa0->crypto_block_size; icv_sz = sa0->integ_icv_size; iv_sz = sa0->crypto_iv_size; + + /* submit frame when op_id is different then the old one */ + if (is_async && sa0->crypto_async_enc_op_id != last_async_op) + { + if (async_frame && async_frame->n_elts) + { + if (vnet_crypto_async_submit_open_frame (vm, async_frame) + < 0) + esp_async_recycle_failed_submit (async_frame, b, next); + } + async_frame = + vnet_crypto_async_get_frame (vm, sa0->crypto_async_enc_op_id); + last_async_op = sa0->crypto_async_enc_op_id; + } } if (PREDICT_FALSE (~0 == sa0->encrypt_thread_index)) @@ -569,129 +879,25 @@ esp_encrypt_inline (vlib_main_t * vm, vlib_node_runtime_t * node, esp->spi = spi; esp->seq = clib_net_to_host_u32 (sa0->seq); - if (sa0->crypto_enc_op_id) + if (is_async) { - vnet_crypto_op_t *op; - vec_add2_aligned (crypto_ops[0], op, 1, CLIB_CACHE_LINE_BYTES); - vnet_crypto_op_init (op, sa0->crypto_enc_op_id); - - op->src = op->dst = payload; - op->key_index = sa0->crypto_key_index; - op->len = payload_len - icv_sz; - op->user_data = b - bufs; - - if (ipsec_sa_is_set_IS_AEAD (sa0)) - { - /* - * construct the AAD in a scratch space in front - * of the IP header. - */ - op->aad = payload - hdr_len - sizeof (esp_aead_t); - - esp_aad_fill (op, esp, sa0); - - op->tag = payload + op->len; - op->tag_len = 16; - - u64 *iv = (u64 *) (payload - iv_sz); - nonce->salt = sa0->salt; - nonce->iv = *iv = clib_host_to_net_u64 (sa0->gcm_iv_counter++); - op->iv = (u8 *) nonce; - nonce++; - } - else - { - op->iv = payload - iv_sz; - op->flags = VNET_CRYPTO_OP_FLAG_INIT_IV; - } + if (PREDICT_FALSE (sa0->crypto_async_enc_op_id == 0)) + goto trace; - if (lb != b[0]) + if (esp_prepare_async_frame (vm, ptd, &async_frame, sa0, b[0], esp, + payload, payload_len, iv_sz, + icv_sz, from[b - bufs], next, hdr_len, + async_next, lb)) { - /* is chained */ - vlib_buffer_t *cb = b[0]; - op->flags |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; - op->chunk_index = vec_len (ptd->chunks); - op->tag = vlib_buffer_get_tail (lb) - icv_sz; - vec_add2 (ptd->chunks, ch, 1); - ch->len = payload_len; - ch->src = ch->dst = payload; - cb = vlib_get_buffer (vm, cb->next_buffer); - op->n_chunks = 1; - - while (1) - { - vec_add2 (ptd->chunks, ch, 1); - op->n_chunks += 1; - if (lb == cb) - ch->len = cb->current_length - icv_sz; - else - ch->len = cb->current_length; - ch->src = ch->dst = vlib_buffer_get_current (cb); - - if (!(cb->flags & VLIB_BUFFER_NEXT_PRESENT)) - break; - - cb = vlib_get_buffer (vm, cb->next_buffer); - } + esp_async_recycle_failed_submit (async_frame, b, next); + goto trace; } } - - if (sa0->integ_op_id) + else { - vnet_crypto_op_t *op; - vec_add2_aligned (integ_ops[0], op, 1, CLIB_CACHE_LINE_BYTES); - vnet_crypto_op_init (op, sa0->integ_op_id); - op->src = payload - iv_sz - sizeof (esp_header_t); - op->digest = payload + payload_len - icv_sz; - op->key_index = sa0->integ_key_index; - op->digest_len = icv_sz; - op->len = payload_len - icv_sz + iv_sz + sizeof (esp_header_t); - op->user_data = b - bufs; - - if (lb != b[0]) - { - /* is chained */ - op->flags |= VNET_CRYPTO_OP_FLAG_CHAINED_BUFFERS; - vlib_buffer_t *cb = b[0]; - op->chunk_index = vec_len (ptd->chunks); - op->digest = vlib_buffer_get_tail (lb) - icv_sz; - vec_add2 (ptd->chunks, ch, 1); - ch->len = payload_len + iv_sz + sizeof (esp_header_t); - ch->src = payload - iv_sz - sizeof (esp_header_t); - cb = vlib_get_buffer (vm, cb->next_buffer); - op->n_chunks = 1; - - while (1) - { - vec_add2 (ptd->chunks, ch, 1); - op->n_chunks += 1; - if (lb == cb) - { - ch->len = cb->current_length - icv_sz; - if (ipsec_sa_is_set_USE_ESN (sa0)) - { - u32 seq_hi = clib_net_to_host_u32 (sa0->seq_hi); - clib_memcpy_fast (op->digest, &seq_hi, - sizeof (seq_hi)); - ch->len += sizeof (seq_hi); - } - } - else - ch->len = cb->current_length; - ch->src = vlib_buffer_get_current (cb); - - if (!(cb->flags & VLIB_BUFFER_NEXT_PRESENT)) - break; - - cb = vlib_get_buffer (vm, cb->next_buffer); - } - } - else if (ipsec_sa_is_set_USE_ESN (sa0)) - { - u32 seq_hi = clib_net_to_host_u32 (sa0->seq_hi); - clib_memcpy_fast (op->digest, &seq_hi, sizeof (seq_hi)); - op->len += sizeof (seq_hi); - } + esp_prepare_sync_op (vm, ptd, crypto_ops, integ_ops, sa0, payload, + payload_len, iv_sz, icv_sz, bufs, b, lb, + hdr_len, esp, nonce++); } vlib_buffer_advance (b[0], 0LL - hdr_len); @@ -721,14 +927,21 @@ esp_encrypt_inline (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_increment_combined_counter (&ipsec_sa_counters, thread_index, current_sa_index, current_sa_packets, current_sa_bytes); + if (!is_async) + { + esp_process_ops (vm, node, ptd->crypto_ops, bufs, nexts); + esp_process_chained_ops (vm, node, ptd->chained_crypto_ops, bufs, nexts, + ptd->chunks); - esp_process_ops (vm, node, ptd->crypto_ops, bufs, nexts); - esp_process_chained_ops (vm, node, ptd->chained_crypto_ops, bufs, nexts, - ptd->chunks); - - esp_process_ops (vm, node, ptd->integ_ops, bufs, nexts); - esp_process_chained_ops (vm, node, ptd->chained_integ_ops, bufs, nexts, - ptd->chunks); + esp_process_ops (vm, node, ptd->integ_ops, bufs, nexts); + esp_process_chained_ops (vm, node, ptd->chained_integ_ops, bufs, nexts, + ptd->chunks); + } + else if (async_frame && async_frame->n_elts) + { + if (vnet_crypto_async_submit_open_frame (vm, async_frame) < 0) + esp_async_recycle_failed_submit (async_frame, b, next); + } vlib_node_increment_counter (vm, node->node_index, ESP_ENCRYPT_ERROR_RX_PKTS, frame->n_vectors); @@ -737,11 +950,98 @@ esp_encrypt_inline (vlib_main_t * vm, vlib_node_runtime_t * node, return frame->n_vectors; } +always_inline uword +esp_encrypt_post_inline (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b = bufs; + u16 nexts[VLIB_FRAME_SIZE], *next = nexts; + u32 *from = vlib_frame_vector_args (frame); + u32 n_left = frame->n_vectors; + + vlib_get_buffers (vm, from, b, n_left); + + if (n_left >= 4) + { + vlib_prefetch_buffer_header (b[0], LOAD); + vlib_prefetch_buffer_header (b[1], LOAD); + vlib_prefetch_buffer_header (b[2], LOAD); + vlib_prefetch_buffer_header (b[3], LOAD); + } + + while (n_left > 8) + { + vlib_prefetch_buffer_header (b[4], LOAD); + vlib_prefetch_buffer_header (b[5], LOAD); + vlib_prefetch_buffer_header (b[6], LOAD); + vlib_prefetch_buffer_header (b[7], LOAD); + + next[0] = (esp_post_data (b[0]))->next_index; + next[1] = (esp_post_data (b[1]))->next_index; + next[2] = (esp_post_data (b[2]))->next_index; + next[3] = (esp_post_data (b[3]))->next_index; + + if (PREDICT_FALSE (node->flags & VLIB_NODE_FLAG_TRACE)) + { + if (b[0]->flags & VLIB_BUFFER_IS_TRACED) + { + esp_encrypt_post_trace_t *tr = vlib_add_trace (vm, node, b[0], + sizeof (*tr)); + tr->next_index = next[0]; + } + if (b[1]->flags & VLIB_BUFFER_IS_TRACED) + { + esp_encrypt_post_trace_t *tr = vlib_add_trace (vm, node, b[1], + sizeof (*tr)); + tr->next_index = next[1]; + } + if (b[2]->flags & VLIB_BUFFER_IS_TRACED) + { + esp_encrypt_post_trace_t *tr = vlib_add_trace (vm, node, b[2], + sizeof (*tr)); + tr->next_index = next[2]; + } + if (b[3]->flags & VLIB_BUFFER_IS_TRACED) + { + esp_encrypt_post_trace_t *tr = vlib_add_trace (vm, node, b[3], + sizeof (*tr)); + tr->next_index = next[3]; + } + } + + b += 4; + next += 4; + n_left -= 4; + } + + while (n_left > 0) + { + next[0] = (esp_post_data (b[0]))->next_index; + if (PREDICT_FALSE (b[0]->flags & VLIB_BUFFER_IS_TRACED)) + { + esp_encrypt_post_trace_t *tr = vlib_add_trace (vm, node, b[0], + sizeof (*tr)); + tr->next_index = next[0]; + } + + b += 1; + next += 1; + n_left -= 1; + } + + vlib_node_increment_counter (vm, node->node_index, + ESP_ENCRYPT_ERROR_POST_RX_PKTS, + frame->n_vectors); + vlib_buffer_enqueue_to_next (vm, node, from, nexts, frame->n_vectors); + return frame->n_vectors; +} + VLIB_NODE_FN (esp4_encrypt_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * from_frame) { - return esp_encrypt_inline (vm, node, from_frame, 0 /* is_ip6 */ , 0); + return esp_encrypt_inline (vm, node, from_frame, 0 /* is_ip6 */ , 0, + esp_encrypt_async_next.esp4_post_next); } /* *INDENT-OFF* */ @@ -759,15 +1059,37 @@ VLIB_REGISTER_NODE (esp4_encrypt_node) = { [ESP_ENCRYPT_NEXT_DROP] = "ip4-drop", [ESP_ENCRYPT_NEXT_HANDOFF] = "esp4-encrypt-handoff", [ESP_ENCRYPT_NEXT_INTERFACE_OUTPUT] = "interface-output", + [ESP_ENCRYPT_NEXT_PENDING] = "esp-encrypt-pending", }, }; /* *INDENT-ON* */ +VLIB_NODE_FN (esp4_encrypt_post_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return esp_encrypt_post_inline (vm, node, from_frame); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (esp4_encrypt_post_node) = { + .name = "esp4-encrypt-post", + .vector_size = sizeof (u32), + .format_trace = format_esp_post_encrypt_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + .sibling_of = "esp4-encrypt", + + .n_errors = ARRAY_LEN(esp_encrypt_error_strings), + .error_strings = esp_encrypt_error_strings, +}; +/* *INDENT-ON* */ + VLIB_NODE_FN (esp6_encrypt_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * from_frame) { - return esp_encrypt_inline (vm, node, from_frame, 1 /* is_ip6 */ , 0); + return esp_encrypt_inline (vm, node, from_frame, 1 /* is_ip6 */ , 0, + esp_encrypt_async_next.esp6_post_next); } /* *INDENT-OFF* */ @@ -785,15 +1107,37 @@ VLIB_REGISTER_NODE (esp6_encrypt_node) = { [ESP_ENCRYPT_NEXT_DROP] = "ip6-drop", [ESP_ENCRYPT_NEXT_HANDOFF] = "esp6-encrypt-handoff", [ESP_ENCRYPT_NEXT_INTERFACE_OUTPUT] = "interface-output", + [ESP_ENCRYPT_NEXT_PENDING] = "esp-encrypt-pending", }, }; /* *INDENT-ON* */ +VLIB_NODE_FN (esp6_encrypt_post_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return esp_encrypt_post_inline (vm, node, from_frame); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (esp6_encrypt_post_node) = { + .name = "esp6-encrypt-post", + .vector_size = sizeof (u32), + .format_trace = format_esp_post_encrypt_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + .sibling_of = "esp6-encrypt", + + .n_errors = ARRAY_LEN(esp_encrypt_error_strings), + .error_strings = esp_encrypt_error_strings, +}; +/* *INDENT-ON* */ + VLIB_NODE_FN (esp4_encrypt_tun_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * from_frame) { - return esp_encrypt_inline (vm, node, from_frame, 0 /* is_ip6 */ , 1); + return esp_encrypt_inline (vm, node, from_frame, 0 /* is_ip6 */ , 1, + esp_encrypt_async_next.esp4_tun_post_next); } /* *INDENT-OFF* */ @@ -811,9 +1155,29 @@ VLIB_REGISTER_NODE (esp4_encrypt_tun_node) = { [ESP_ENCRYPT_NEXT_DROP] = "ip4-drop", [ESP_ENCRYPT_NEXT_HANDOFF] = "esp4-encrypt-tun-handoff", [ESP_ENCRYPT_NEXT_INTERFACE_OUTPUT] = "error-drop", + [ESP_ENCRYPT_NEXT_PENDING] = "esp-encrypt-pending", }, }; +VLIB_NODE_FN (esp4_encrypt_tun_post_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return esp_encrypt_post_inline (vm, node, from_frame); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (esp4_encrypt_tun_post_node) = { + .name = "esp4-encrypt-tun-post", + .vector_size = sizeof (u32), + .format_trace = format_esp_post_encrypt_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + .sibling_of = "esp4-encrypt-tun", + + .n_errors = ARRAY_LEN(esp_encrypt_error_strings), + .error_strings = esp_encrypt_error_strings, +}; + VNET_FEATURE_INIT (esp4_encrypt_tun_feat_node, static) = { .arc_name = "ip4-output", @@ -840,7 +1204,8 @@ VLIB_NODE_FN (esp6_encrypt_tun_node) (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * from_frame) { - return esp_encrypt_inline (vm, node, from_frame, 1 /* is_ip6 */ , 1); + return esp_encrypt_inline (vm, node, from_frame, 1 /* is_ip6 */ , 1, + esp_encrypt_async_next.esp6_tun_post_next); } /* *INDENT-OFF* */ @@ -858,6 +1223,7 @@ VLIB_REGISTER_NODE (esp6_encrypt_tun_node) = { [ESP_ENCRYPT_NEXT_DROP] = "ip6-drop", [ESP_ENCRYPT_NEXT_HANDOFF] = "esp6-encrypt-tun-handoff", [ESP_ENCRYPT_NEXT_INTERFACE_OUTPUT] = "error-drop", + [ESP_ENCRYPT_NEXT_PENDING] = "esp-encrypt-pending", }, }; @@ -877,6 +1243,26 @@ VNET_FEATURE_INIT (esp4o6_encrypt_tun_feat_node, static) = /* *INDENT-ON* */ +VLIB_NODE_FN (esp6_encrypt_tun_post_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return esp_encrypt_post_inline (vm, node, from_frame); +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (esp6_encrypt_tun_post_node) = { + .name = "esp6-encrypt-tun-post", + .vector_size = sizeof (u32), + .format_trace = format_esp_post_encrypt_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + .sibling_of = "esp6-encrypt-tun", + + .n_errors = ARRAY_LEN(esp_encrypt_error_strings), + .error_strings = esp_encrypt_error_strings, +}; +/* *INDENT-ON* */ + typedef struct { u32 sa_index; @@ -1008,6 +1394,23 @@ VNET_FEATURE_INIT (esp6_no_crypto_tun_feat_node, static) = }; /* *INDENT-ON* */ +VLIB_NODE_FN (esp_encrypt_pending_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * from_frame) +{ + return from_frame->n_vectors; +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (esp_encrypt_pending_node) = { + .name = "esp-encrypt-pending", + .vector_size = sizeof (u32), + .type = VLIB_NODE_TYPE_INTERNAL, + + .n_next_nodes = 0 +}; +/* *INDENT-ON* */ + /* * fd.io coding-style-patch-verification: ON * diff --git a/src/vnet/ipsec/ipsec.c b/src/vnet/ipsec/ipsec.c index e7cb615f957..6b31926be4f 100644 --- a/src/vnet/ipsec/ipsec.c +++ b/src/vnet/ipsec/ipsec.c @@ -26,6 +26,8 @@ #include <vnet/ipsec/ah.h> ipsec_main_t ipsec_main; +esp_async_post_next_t esp_encrypt_async_next; +esp_async_post_next_t esp_decrypt_async_next; static clib_error_t * ipsec_check_ah_support (ipsec_sa_t * sa) @@ -173,7 +175,8 @@ ipsec_register_esp_backend (vlib_main_t * vm, ipsec_main_t * im, const char *esp6_decrypt_node_name, const char *esp6_decrypt_tun_node_name, check_support_cb_t esp_check_support_cb, - add_del_sa_sess_cb_t esp_add_del_sa_sess_cb) + add_del_sa_sess_cb_t esp_add_del_sa_sess_cb, + enable_disable_cb_t enable_disable_cb) { ipsec_esp_backend_t *b; @@ -206,6 +209,8 @@ ipsec_register_esp_backend (vlib_main_t * vm, ipsec_main_t * im, b->check_support_cb = esp_check_support_cb; b->add_del_sa_sess_cb = esp_add_del_sa_sess_cb; + b->enable_disable_cb = enable_disable_cb; + return b - im->esp_backends; } @@ -253,6 +258,18 @@ ipsec_select_esp_backend (ipsec_main_t * im, u32 backend_idx) if (pool_is_free_index (im->esp_backends, backend_idx)) return VNET_API_ERROR_INVALID_VALUE; + /* disable current backend */ + if (im->esp_current_backend != ~0) + { + ipsec_esp_backend_t *cb = pool_elt_at_index (im->esp_backends, + im->esp_current_backend); + if (cb->enable_disable_cb) + { + if ((cb->enable_disable_cb) (0) != 0) + return -1; + } + } + ipsec_esp_backend_t *b = pool_elt_at_index (im->esp_backends, backend_idx); im->esp_current_backend = backend_idx; im->esp4_encrypt_node_index = b->esp4_encrypt_node_index; @@ -273,9 +290,64 @@ ipsec_select_esp_backend (ipsec_main_t * im, u32 backend_idx) im->esp46_encrypt_tun_feature_index = b->esp46_encrypt_tun_feature_index; im->esp66_encrypt_tun_feature_index = b->esp66_encrypt_tun_feature_index; + if (b->enable_disable_cb) + { + if ((b->enable_disable_cb) (1) != 0) + return -1; + } return 0; } +void +ipsec_set_async_mode (u32 is_enabled) +{ + ipsec_main_t *im = &ipsec_main; + ipsec_sa_t *sa; + + /* lock all SAs before change im->async_mode */ + pool_foreach (sa, im->sad, ( + { + fib_node_lock (&sa->node); + })); + + im->async_mode = is_enabled; + + /* change SA crypto op data before unlock them */ + pool_foreach (sa, im->sad, ( + { + sa->crypto_op_data = is_enabled ? + sa->async_op_data.data : sa->sync_op_data.data; + fib_node_unlock (&sa->node); + })); +} + +static void +crypto_engine_backend_register_post_node (vlib_main_t * vm) +{ + esp_async_post_next_t *eit; + esp_async_post_next_t *dit; + + eit = &esp_encrypt_async_next; + eit->esp4_post_next = + vnet_crypto_register_post_node (vm, "esp4-encrypt-post"); + eit->esp6_post_next = + vnet_crypto_register_post_node (vm, "esp6-encrypt-post"); + eit->esp4_tun_post_next = + vnet_crypto_register_post_node (vm, "esp4-encrypt-tun-post"); + eit->esp6_tun_post_next = + vnet_crypto_register_post_node (vm, "esp6-encrypt-tun-post"); + + dit = &esp_decrypt_async_next; + dit->esp4_post_next = + vnet_crypto_register_post_node (vm, "esp4-decrypt-post"); + dit->esp6_post_next = + vnet_crypto_register_post_node (vm, "esp6-decrypt-post"); + dit->esp4_tun_post_next = + vnet_crypto_register_post_node (vm, "esp4-decrypt-tun-post"); + dit->esp6_tun_post_next = + vnet_crypto_register_post_node (vm, "esp6-decrypt-tun-post"); +} + static clib_error_t * ipsec_init (vlib_main_t * vm) { @@ -298,6 +370,9 @@ ipsec_init (vlib_main_t * vm) ASSERT (node); im->error_drop_node_index = node->index; + im->ah_current_backend = ~0; + im->esp_current_backend = ~0; + u32 idx = ipsec_register_ah_backend (vm, im, "crypto engine backend", "ah4-encrypt", "ah4-decrypt", @@ -320,7 +395,8 @@ ipsec_init (vlib_main_t * vm) "esp6-encrypt-tun", "esp6-decrypt", "esp6-decrypt-tun", - ipsec_check_esp_support, NULL); + ipsec_check_esp_support, + NULL, crypto_dispatch_enable_disable); im->esp_default_backend = idx; rv = ipsec_select_esp_backend (im, idx); @@ -454,6 +530,9 @@ ipsec_init (vlib_main_t * vm) im->esp6_dec_tun_fq_index = vlib_frame_queue_main_init (esp6_decrypt_tun_node.index, 0); + im->async_mode = 0; + crypto_engine_backend_register_post_node (vm); + return 0; } diff --git a/src/vnet/ipsec/ipsec.h b/src/vnet/ipsec/ipsec.h index f1b7dafa9be..712e16dab4c 100644 --- a/src/vnet/ipsec/ipsec.h +++ b/src/vnet/ipsec/ipsec.h @@ -28,6 +28,7 @@ typedef clib_error_t *(*add_del_sa_sess_cb_t) (u32 sa_index, u8 is_add); typedef clib_error_t *(*check_support_cb_t) (ipsec_sa_t * sa); +typedef clib_error_t *(*enable_disable_cb_t) (int is_enable); typedef struct { @@ -53,6 +54,8 @@ typedef struct add_del_sa_sess_cb_t add_del_sa_sess_cb; /* check support function */ check_support_cb_t check_support_cb; + /* enable or disable function */ + enable_disable_cb_t enable_disable_cb; u32 esp4_encrypt_node_index; u32 esp4_decrypt_node_index; u32 esp4_encrypt_next_index; @@ -194,6 +197,8 @@ typedef struct u32 esp6_enc_tun_fq_index; u32 esp4_dec_tun_fq_index; u32 esp6_dec_tun_fq_index; + + u8 async_mode; } ipsec_main_t; typedef enum ipsec_format_flags_t_ @@ -266,7 +271,8 @@ u32 ipsec_register_esp_backend (vlib_main_t * vm, ipsec_main_t * im, const char *esp6_decrypt_node_name, const char *esp6_decrypt_tun_node_name, check_support_cb_t esp_check_support_cb, - add_del_sa_sess_cb_t esp_add_del_sa_sess_cb); + add_del_sa_sess_cb_t esp_add_del_sa_sess_cb, + enable_disable_cb_t enable_disable_cb); int ipsec_select_ah_backend (ipsec_main_t * im, u32 ah_backend_idx); int ipsec_select_esp_backend (ipsec_main_t * im, u32 esp_backend_idx); @@ -282,6 +288,7 @@ ipsec_sa_get (u32 sa_index) void ipsec_add_feature (const char *arc_name, const char *node_name, u32 * out_feature_index); +void ipsec_set_async_mode (u32 is_enabled); #endif /* __IPSEC_H__ */ diff --git a/src/vnet/ipsec/ipsec_cli.c b/src/vnet/ipsec/ipsec_cli.c index 7779e79f067..13f9efdf566 100644 --- a/src/vnet/ipsec/ipsec_cli.c +++ b/src/vnet/ipsec/ipsec_cli.c @@ -414,6 +414,9 @@ show_ipsec_command_fn (vlib_main_t * vm, ipsec_spd_bindings_show_all (vm, im); ipsec_tun_protect_walk (ipsec_tun_protect_show_one, vm); + vlib_cli_output (vm, "IPSec async mode: %s", + (im->async_mode ? "on" : "off")); + return 0; } @@ -1038,6 +1041,41 @@ ipsec_cli_init (vlib_main_t * vm) VLIB_INIT_FUNCTION (ipsec_cli_init); +static clib_error_t * +set_async_mode_command_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + unformat_input_t _line_input, *line_input = &_line_input; + int async_enable = 0; + + if (!unformat_user (input, unformat_line_input, line_input)) + return 0; + + while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (line_input, "on")) + async_enable = 1; + else if (unformat (line_input, "off")) + async_enable = 0; + else + return (clib_error_return (0, "unknown input '%U'", + format_unformat_error, line_input)); + } + + vnet_crypto_request_async_mode (async_enable); + ipsec_set_async_mode (async_enable); + + unformat_free (line_input); + return (NULL); +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (set_async_mode_command, static) = { + .path = "set ipsec async mode", + .short_help = "set ipsec async mode on|off", + .function = set_async_mode_command_fn, +}; +/* *INDENT-ON* */ /* * fd.io coding-style-patch-verification: ON diff --git a/src/vnet/ipsec/ipsec_sa.c b/src/vnet/ipsec/ipsec_sa.c index 0e1e63d0db4..f60566cba79 100644 --- a/src/vnet/ipsec/ipsec_sa.c +++ b/src/vnet/ipsec/ipsec_sa.c @@ -100,8 +100,8 @@ ipsec_sa_set_crypto_alg (ipsec_sa_t * sa, ipsec_crypto_alg_t crypto_alg) sa->crypto_alg = crypto_alg; sa->crypto_iv_size = im->crypto_algs[crypto_alg].iv_size; sa->crypto_block_size = im->crypto_algs[crypto_alg].block_size; - sa->crypto_enc_op_id = im->crypto_algs[crypto_alg].enc_op_id; - sa->crypto_dec_op_id = im->crypto_algs[crypto_alg].dec_op_id; + sa->sync_op_data.crypto_enc_op_id = im->crypto_algs[crypto_alg].enc_op_id; + sa->sync_op_data.crypto_dec_op_id = im->crypto_algs[crypto_alg].dec_op_id; sa->crypto_calg = im->crypto_algs[crypto_alg].alg; ASSERT (sa->crypto_iv_size <= ESP_MAX_IV_SIZE); ASSERT (sa->crypto_block_size <= ESP_MAX_BLOCK_SIZE); @@ -118,11 +118,54 @@ ipsec_sa_set_integ_alg (ipsec_sa_t * sa, ipsec_integ_alg_t integ_alg) ipsec_main_t *im = &ipsec_main; sa->integ_alg = integ_alg; sa->integ_icv_size = im->integ_algs[integ_alg].icv_size; - sa->integ_op_id = im->integ_algs[integ_alg].op_id; + sa->sync_op_data.integ_op_id = im->integ_algs[integ_alg].op_id; sa->integ_calg = im->integ_algs[integ_alg].alg; ASSERT (sa->integ_icv_size <= ESP_MAX_ICV_SIZE); } +void +ipsec_sa_set_async_op_ids (ipsec_sa_t * sa) +{ + /* *INDENT-OFF* */ + if (ipsec_sa_is_set_USE_ESN (sa)) + { +#define _(n, s, k) \ + if( sa->sync_op_data.crypto_enc_op_id == VNET_CRYPTO_OP_##n##_ENC ) \ + sa->async_op_data.crypto_async_enc_op_id = \ + VNET_CRYPTO_OP_##n##_TAG16_AAD12_ENC; \ + if( sa->sync_op_data.crypto_dec_op_id == VNET_CRYPTO_OP_##n##_DEC ) \ + sa->async_op_data.crypto_async_dec_op_id = \ + VNET_CRYPTO_OP_##n##_TAG16_AAD12_DEC; + foreach_crypto_aead_alg +#undef _ + } + else + { +#define _(n, s, k) \ + if( sa->sync_op_data.crypto_enc_op_id == VNET_CRYPTO_OP_##n##_ENC ) \ + sa->async_op_data.crypto_async_enc_op_id = \ + VNET_CRYPTO_OP_##n##_TAG16_AAD8_ENC; \ + if( sa->sync_op_data.crypto_dec_op_id == VNET_CRYPTO_OP_##n##_DEC ) \ + sa->async_op_data.crypto_async_dec_op_id = \ + VNET_CRYPTO_OP_##n##_TAG16_AAD8_DEC; + foreach_crypto_aead_alg +#undef _ + } + +#define _(c, h, s, k ,d) \ + if( sa->sync_op_data.crypto_enc_op_id == VNET_CRYPTO_OP_##c##_ENC && \ + sa->sync_op_data.integ_op_id == VNET_CRYPTO_OP_##h##_HMAC) \ + sa->async_op_data.crypto_async_enc_op_id = \ + VNET_CRYPTO_OP_##c##_##h##_TAG##d##_ENC; \ + if( sa->sync_op_data.crypto_dec_op_id == VNET_CRYPTO_OP_##c##_DEC && \ + sa->sync_op_data.integ_op_id == VNET_CRYPTO_OP_##h##_HMAC) \ + sa->async_op_data.crypto_async_dec_op_id = \ + VNET_CRYPTO_OP_##c##_##h##_TAG##d##_DEC; + foreach_crypto_link_async_alg +#undef _ + /* *INDENT-ON* */ +} + int ipsec_sa_add_and_lock (u32 id, u32 spi, @@ -172,6 +215,8 @@ ipsec_sa_add_and_lock (u32 id, clib_memcpy (&sa->integ_key, ik, sizeof (sa->integ_key)); } ipsec_sa_set_crypto_alg (sa, crypto_alg); + ipsec_sa_set_async_op_ids (sa); + clib_memcpy (&sa->crypto_key, ck, sizeof (sa->crypto_key)); ip46_address_copy (&sa->tunnel_src_addr, tun_src); ip46_address_copy (&sa->tunnel_dst_addr, tun_dst); @@ -198,6 +243,19 @@ ipsec_sa_add_and_lock (u32 id, } } + if (sa->async_op_data.crypto_async_enc_op_id && + !ipsec_sa_is_set_IS_AEAD (sa)) + { //AES-CBC & HMAC + sa->async_op_data.linked_key_index = + vnet_crypto_key_add_linked (vm, sa->crypto_key_index, + sa->integ_key_index); + } + + if (im->async_mode) + sa->crypto_op_data = sa->async_op_data.data; + else + sa->crypto_op_data = sa->sync_op_data.data; + err = ipsec_check_support_cb (im, sa); if (err) { diff --git a/src/vnet/ipsec/ipsec_sa.h b/src/vnet/ipsec/ipsec_sa.h index 0997eb7d8bc..de9b51ea66a 100644 --- a/src/vnet/ipsec/ipsec_sa.h +++ b/src/vnet/ipsec/ipsec_sa.h @@ -127,9 +127,27 @@ typedef struct vnet_crypto_key_index_t crypto_key_index; vnet_crypto_key_index_t integ_key_index; - vnet_crypto_op_id_t crypto_enc_op_id:16; - vnet_crypto_op_id_t crypto_dec_op_id:16; - vnet_crypto_op_id_t integ_op_id:16; + + /* Union data shared by sync and async ops, updated when mode is + * changed. */ + union + { + struct + { + vnet_crypto_op_id_t crypto_enc_op_id:16; + vnet_crypto_op_id_t crypto_dec_op_id:16; + vnet_crypto_op_id_t integ_op_id:16; + }; + + struct + { + vnet_crypto_async_op_id_t crypto_async_enc_op_id:16; + vnet_crypto_async_op_id_t crypto_async_dec_op_id:16; + vnet_crypto_key_index_t linked_key_index; + }; + + u64 crypto_op_data; + }; /* data accessed by dataplane code should be above this comment */ CLIB_CACHE_LINE_ALIGN_MARK (cacheline1); @@ -165,6 +183,28 @@ typedef struct /* Salt used in GCM modes - stored in network byte order */ u32 salt; u64 gcm_iv_counter; + + union + { + struct + { + vnet_crypto_op_id_t crypto_enc_op_id:16; + vnet_crypto_op_id_t crypto_dec_op_id:16; + vnet_crypto_op_id_t integ_op_id:16; + }; + u64 data; + } sync_op_data; + + union + { + struct + { + vnet_crypto_async_op_id_t crypto_async_enc_op_id:16; + vnet_crypto_async_op_id_t crypto_async_dec_op_id:16; + vnet_crypto_key_index_t linked_key_index; + }; + u64 data; + } async_op_data; } ipsec_sa_t; STATIC_ASSERT_OFFSET_OF (ipsec_sa_t, cacheline1, CLIB_CACHE_LINE_BYTES); |