ligang/vpp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

udp/session: refactor to support dgram mode

Browse Source 
- adds session layer support for datagram based protocols
- updates udp to work in pure connectionless and datagram mode. The
  existing connected mode is now 'accessible' for apps as a dummy UDPC,
  as in, connected udp, protocol.
- updates udp_echo, echo client, echo server code to work in datagram
  mode.

Change-Id: I2960c0d2d246cb166005f545794ec31fe0d546dd
Signed-off-by: Florin Coras <fcoras@cisco.com>
This commit is contained in:
Florin Coras
2018-04-09 09:24:52 -07:00
committed by Damjan Marion
parent 684d08c7e5
commit 7fb0fe1f69
21 changed files with 1105 additions and 447 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/svm/svm_fifo.c
+18 -1
View File
@@ -169,6 +169,9 @@ format_svm_fifo (u8 * s, va_list * args)
svm_fifo_t *f = va_arg (*args, svm_fifo_t *);
int verbose = va_arg (*args, int);
if (!s)
return s;
s = format (s, "cursize %u nitems %u has_event %d\n",
f->cursize, f->nitems, f->has_event);
s = format (s, " head %d tail %d\n", f->head, f->tail);
@@ -459,7 +462,7 @@ svm_fifo_enqueue_internal (svm_fifo_t * f, u32 max_bytes,
f->ooos_newest = OOO_SEGMENT_INVALID_INDEX;
if (PREDICT_FALSE (cursize == f->nitems))
return -2; /* fifo stuffed */
return SVM_FIFO_FULL;
nitems = f->nitems;
@@ -615,6 +618,20 @@ svm_fifo_enqueue_with_offset (svm_fifo_t * f,
copy_from_here);
}
void
svm_fifo_overwrite_head (svm_fifo_t * f, u8 * data, u32 len)
{
u32 first_chunk;
ASSERT (len <= f->nitems);
if (len < f->nitems - f->head)
clib_memcpy (&f->data[f->head], data, len);
else
{
first_chunk = len - (f->nitems - f->head);
clib_memcpy (&f->data[f->head], data, first_chunk);
clib_memcpy (f->data, data + first_chunk, len - first_chunk);
}
}
static int
svm_fifo_dequeue_internal (svm_fifo_t * f, u32 max_bytes, u8 * copy_here)
src/svm/svm_fifo.h
+6
View File
@@ -80,6 +80,11 @@ typedef struct _svm_fifo
CLIB_CACHE_LINE_ALIGN_MARK (data);
} svm_fifo_t;
typedef enum
{
SVM_FIFO_FULL = -2,
} svm_fifo_err_t;
#if SVM_FIFO_TRACE
#define svm_fifo_trace_add(_f, _s, _l, _t) \
{ \
@@ -150,6 +155,7 @@ int svm_fifo_dequeue_drop (svm_fifo_t * f, u32 max_bytes);
u32 svm_fifo_number_ooo_segments (svm_fifo_t * f);
ooo_segment_t *svm_fifo_first_ooo_segment (svm_fifo_t * f);
void svm_fifo_init_pointers (svm_fifo_t * f, u32 pointer);
void svm_fifo_overwrite_head (svm_fifo_t * f, u8 * data, u32 len);
format_function_t format_svm_fifo;
src/tests/vnet/session/udp_echo.c
+303 -172
View File
File diff suppressed because it is too large Load Diff
src/vnet/session-apps/echo_client.c
+65 -47
View File
@@ -46,23 +46,21 @@ static void
send_data_chunk (echo_client_main_t * ecm, session_t * s)
{
u8 *test_data = ecm->connect_test_data;
int test_buf_offset;
int test_buf_len, test_buf_offset, rv;
u32 bytes_this_chunk;
session_fifo_event_t evt;
svm_fifo_t *txf;
int rv;
ASSERT (vec_len (test_data) > 0);
test_buf_len = vec_len (test_data);
test_buf_offset = s->bytes_sent % test_buf_len;
bytes_this_chunk = clib_min (test_buf_len - test_buf_offset,
s->bytes_to_send);
test_buf_offset = s->bytes_sent % vec_len (test_data);
bytes_this_chunk = vec_len (test_data) - test_buf_offset;
bytes_this_chunk = bytes_this_chunk < s->bytes_to_send
? bytes_this_chunk : s->bytes_to_send;
txf = s->server_tx_fifo;
rv = svm_fifo_enqueue_nowait (txf, bytes_this_chunk,
test_data + test_buf_offset);
if (!ecm->is_dgram)
rv = app_send_stream (&s->data, test_data + test_buf_offset,
bytes_this_chunk, 0);
else
rv = app_send_dgram (&s->data, test_data + test_buf_offset,
bytes_this_chunk, 0);
/* If we managed to enqueue data... */
if (rv > 0)
@@ -89,35 +87,24 @@ send_data_chunk (echo_client_main_t * ecm, session_t * s)
ed->data[1] = s->bytes_sent;
ed->data[2] = s->bytes_to_send;
}
/* Poke the session layer */
if (svm_fifo_set_event (txf))
{
/* Fabricate TX event, send to vpp */
evt.fifo = txf;
evt.event_type = FIFO_EVENT_APP_TX;
if (svm_queue_add
(ecm->vpp_event_queue[txf->master_thread_index], (u8 *) & evt,
0 /* do wait for mutex */ ))
clib_warning ("could not enqueue event");
}
}
}
static void
receive_data_chunk (echo_client_main_t * ecm, session_t * s)
{
svm_fifo_t *rx_fifo = s->server_rx_fifo;
u32 my_thread_index = vlib_get_thread_index ();
svm_fifo_t *rx_fifo = s->data.rx_fifo;
u32 thread_index = vlib_get_thread_index ();
int n_read, i;
if (ecm->test_bytes)
{
n_read = svm_fifo_dequeue_nowait (rx_fifo,
vec_len (ecm->rx_buf
[my_thread_index]),
ecm->rx_buf[my_thread_index]);
if (!ecm->is_dgram)
n_read = app_recv_stream (&s->data, ecm->rx_buf[thread_index],
vec_len (ecm->rx_buf[thread_index]));
else
n_read = app_recv_dgram (&s->data, ecm->rx_buf[thread_index],
vec_len (ecm->rx_buf[thread_index]));
}
else
{
@@ -148,17 +135,18 @@ receive_data_chunk (echo_client_main_t * ecm, session_t * s)
{
for (i = 0; i < n_read; i++)
{
if (ecm->rx_buf[my_thread_index][i]
if (ecm->rx_buf[thread_index][i]
!= ((s->bytes_received + i) & 0xff))
{
clib_warning ("read %d error at byte %lld, 0x%x not 0x%x",
n_read, s->bytes_received + i,
ecm->rx_buf[my_thread_index][i],
ecm->rx_buf[thread_index][i],
((s->bytes_received + i) & 0xff));
ecm->test_failed = 1;
}
}
}
ASSERT (n_read <= s->bytes_to_receive);
s->bytes_to_receive -= n_read;
s->bytes_received += n_read;
}
@@ -230,20 +218,15 @@ echo_client_node_fn (vlib_main_t * vm, vlib_node_runtime_t * node,
}
if (sp->bytes_to_receive > 0)
{
receive_data_chunk (ecm, sp);
delete_session = 0;
}
if (PREDICT_FALSE (delete_session == 1))
{
u32 index, thread_index;
stream_session_t *s;
__sync_fetch_and_add (&ecm->tx_total, sp->bytes_sent);
__sync_fetch_and_add (&ecm->rx_total, sp->bytes_received);
session_parse_handle (sp->vpp_session_handle,
&index, &thread_index);
s = session_get_if_valid (index, thread_index);
s = session_get_from_handle_if_valid (sp->vpp_session_handle);
if (s)
{
@@ -312,8 +295,8 @@ echo_clients_init (vlib_main_t * vm)
num_threads = 1 /* main thread */ + vtm->n_threads;
/* Init test data. Bigecmuffer */
vec_validate (ecm->connect_test_data, 1024 * 1024 - 1);
/* Init test data. Big buffer */
vec_validate (ecm->connect_test_data, 4 * 1024 * 1024 - 1);
for (i = 0; i < vec_len (ecm->connect_test_data); i++)
ecm->connect_test_data[i] = i & 0xff;
@@ -363,12 +346,22 @@ echo_clients_session_connected_callback (u32 app_index, u32 api_context,
session_index = session - ecm->sessions;
session->bytes_to_send = ecm->bytes_to_send;
session->bytes_to_receive = ecm->no_return ? 0ULL : ecm->bytes_to_send;
session->server_rx_fifo = s->server_rx_fifo;
session->server_rx_fifo->client_session_index = session_index;
session->server_tx_fifo = s->server_tx_fifo;
session->server_tx_fifo->client_session_index = session_index;
session->data.rx_fifo = s->server_rx_fifo;
session->data.rx_fifo->client_session_index = session_index;
session->data.tx_fifo = s->server_tx_fifo;
session->data.tx_fifo->client_session_index = session_index;
session->data.vpp_evt_q = ecm->vpp_event_queue[thread_index];
session->vpp_session_handle = session_handle (s);
if (ecm->is_dgram)
{
transport_connection_t *tc;
tc = session_get_transport (s);
clib_memcpy (&session->data.transport, tc,
sizeof (session->data.transport));
session->data.is_dgram = 1;
}
vec_add1 (ecm->connection_index_by_thread[thread_index], session_index);
__sync_fetch_and_add (&ecm->ready_connections, 1);
if (ecm->ready_connections == ecm->expected_connections)
@@ -410,6 +403,28 @@ echo_clients_session_disconnect_callback (stream_session_t * s)
static int
echo_clients_rx_callback (stream_session_t * s)
{
echo_client_main_t *ecm = &echo_client_main;
session_t *sp;
sp = pool_elt_at_index (ecm->sessions,
s->server_rx_fifo->client_session_index);
receive_data_chunk (ecm, sp);
if (svm_fifo_max_dequeue (s->server_rx_fifo))
{
session_fifo_event_t evt;
svm_queue_t *q;
if (svm_fifo_set_event (s->server_rx_fifo))
{
evt.fifo = s->server_rx_fifo;
evt.event_type = FIFO_EVENT_BUILTIN_RX;
q = session_manager_get_vpp_event_queue (s->thread_index);
if (PREDICT_FALSE (q->cursize == q->maxsize))
clib_warning ("out of event queue space");
else if (svm_queue_add (q, (u8 *) & evt, 0))
clib_warning ("failed to enqueue self-tap");
}
}
return 0;
}
@@ -544,7 +559,7 @@ echo_clients_connect (vlib_main_t * vm, u32 n_clients)
}
#define ec_cli_output(_fmt, _args...) \
if (!ecm->no_output) \
if (!ecm->no_output) \
vlib_cli_output(vm, _fmt, ##_args)
static clib_error_t *
@@ -663,6 +678,9 @@ echo_clients_command_fn (vlib_main_t * vm,
ecm->connect_uri = format (0, "%s%c", default_uri, 0);
}
if (ecm->connect_uri[0] == 'u' && ecm->connect_uri[3] != 'c')
ecm->is_dgram = 1;
#if ECHO_CLIENT_PTHREAD
echo_clients_start_tx_pthread ();
#endif
src/vnet/session-apps/echo_client.h
+4 -5
View File
@@ -30,15 +30,13 @@
typedef struct
{
app_session_t data;
u64 bytes_to_send;
u64 bytes_sent;
u64 bytes_to_receive;
u64 bytes_received;
svm_fifo_t *server_rx_fifo;
svm_fifo_t *server_tx_fifo;
u64 vpp_session_handle;
u8 thread_index;
} session_t;
typedef struct
@@ -46,7 +44,7 @@ typedef struct
/*
* Application setup parameters
*/
svm_queue_t *vl_input_queue; /**< vpe input queue */
svm_queue_t *vl_input_queue; /**< vpe input queue */
svm_queue_t **vpp_event_queue;
u32 cli_node_index; /**< cli process node index */
@@ -65,6 +63,7 @@ typedef struct
u32 private_segment_count; /**< Number of private fifo segs */
u32 private_segment_size; /**< size of private fifo segs */
u32 tls_engine; /**< TLS engine mbedtls/openssl */
u8 is_dgram;
/*
* Test state variables
*/
src/vnet/session-apps/echo_server.c
+62 -42
View File
@@ -34,13 +34,14 @@ typedef struct
* Config params
*/
u8 no_echo; /**< Don't echo traffic */
u32 fifo_size; /**< Fifo size */
u32 fifo_size; /**< Fifo size */
u32 rcv_buffer_size; /**< Rcv buffer size */
u32 prealloc_fifos; /**< Preallocate fifos */
u32 private_segment_count; /**< Number of private segments */
u32 private_segment_size; /**< Size of private segments */
char *server_uri; /**< Server URI */
u32 tls_engine; /**< TLS engine: mbedtls/openssl */
u8 is_dgram; /**< set if transport is dgram */
/*
* Test state
*/
@@ -126,25 +127,13 @@ test_bytes (echo_server_main_t * esm, int actual_transfer)
}
/*
* If no-echo, just read the data and be done with it
* If no-echo, just drop the data and be done with it.
*/
int
echo_server_builtin_server_rx_callback_no_echo (stream_session_t * s)
{
echo_server_main_t *esm = &echo_server_main;
u32 my_thread_id = vlib_get_thread_index ();
int actual_transfer;
svm_fifo_t *rx_fifo;
rx_fifo = s->server_rx_fifo;
do
{
actual_transfer =
svm_fifo_dequeue_nowait (rx_fifo, esm->rcv_buffer_size,
esm->rx_buf[my_thread_id]);
}
while (actual_transfer > 0);
svm_fifo_t *rx_fifo = s->server_rx_fifo;
svm_fifo_dequeue_drop (rx_fifo, svm_fifo_max_dequeue (rx_fifo));
return 0;
}
@@ -157,6 +146,8 @@ echo_server_rx_callback (stream_session_t * s)
echo_server_main_t *esm = &echo_server_main;
session_fifo_event_t evt;
u32 thread_index = vlib_get_thread_index ();
app_session_transport_t at;
svm_queue_t *q;
ASSERT (s->thread_index == thread_index);
@@ -166,14 +157,29 @@ echo_server_rx_callback (stream_session_t * s)
ASSERT (rx_fifo->master_thread_index == thread_index);
ASSERT (tx_fifo->master_thread_index == thread_index);
max_dequeue = svm_fifo_max_dequeue (s->server_rx_fifo);
max_enqueue = svm_fifo_max_enqueue (s->server_tx_fifo);
max_enqueue = svm_fifo_max_enqueue (tx_fifo);
if (!esm->is_dgram)
{
max_dequeue = svm_fifo_max_dequeue (rx_fifo);
}
else
{
session_dgram_pre_hdr_t ph;
svm_fifo_peek (rx_fifo, 0, sizeof (ph), (u8 *) & ph);
max_dequeue = ph.data_length - ph.data_offset;
if (!esm->vpp_queue[s->thread_index])
{
q = session_manager_get_vpp_event_queue (s->thread_index);
esm->vpp_queue[s->thread_index] = q;
}
max_enqueue -= sizeof (session_dgram_hdr_t);
}
if (PREDICT_FALSE (max_dequeue == 0))
return 0;
/* Number of bytes we're going to copy */
max_transfer = (max_dequeue < max_enqueue) ? max_dequeue : max_enqueue;
max_transfer = clib_min (max_dequeue, max_enqueue);
/* No space in tx fifo */
if (PREDICT_FALSE (max_transfer == 0))
@@ -184,16 +190,16 @@ echo_server_rx_callback (stream_session_t * s)
/* Program self-tap to retry */
if (svm_fifo_set_event (rx_fifo))
{
svm_queue_t *q;
evt.fifo = rx_fifo;
evt.event_type = FIFO_EVENT_BUILTIN_RX;
q = esm->vpp_queue[thread_index];
q = esm->vpp_queue[s->thread_index];
if (PREDICT_FALSE (q->cursize == q->maxsize))
clib_warning ("out of event queue space");
else if (svm_queue_add (q, (u8 *) & evt, 0))
clib_warning ("failed to enqueue self-tap");
vec_validate (esm->rx_retries[s->thread_index], s->session_index);
if (esm->rx_retries[thread_index][s->session_index] == 500000)
{
clib_warning ("session stuck: %U", format_stream_session, s, 2);
@@ -205,36 +211,47 @@ echo_server_rx_callback (stream_session_t * s)
return 0;
}
_vec_len (esm->rx_buf[thread_index]) = max_transfer;
actual_transfer = svm_fifo_dequeue_nowait (rx_fifo, max_transfer,
esm->rx_buf[thread_index]);
vec_validate (esm->rx_buf[thread_index], max_transfer);
if (!esm->is_dgram)
{
actual_transfer = app_recv_stream_raw (rx_fifo,
esm->rx_buf[thread_index],
max_transfer,
0 /* don't clear event */ );
}
else
{
actual_transfer = app_recv_dgram_raw (rx_fifo,
esm->rx_buf[thread_index],
max_transfer, &at,
0 /* don't clear event */ );
}
ASSERT (actual_transfer == max_transfer);
// test_bytes (esm, actual_transfer);
/* test_bytes (esm, actual_transfer); */
/*
* Echo back
*/
n_written = svm_fifo_enqueue_nowait (tx_fifo, actual_transfer,
esm->rx_buf[thread_index]);
if (n_written != max_transfer)
clib_warning ("short trout!");
if (svm_fifo_set_event (tx_fifo))
if (!esm->is_dgram)
{
/* Fabricate TX event, send to vpp */
evt.fifo = tx_fifo;
evt.event_type = FIFO_EVENT_APP_TX;
if (svm_queue_add (esm->vpp_queue[s->thread_index],
(u8 *) & evt, 0 /* do wait for mutex */ ))
clib_warning ("failed to enqueue tx evt");
n_written = app_send_stream_raw (tx_fifo,
esm->vpp_queue[thread_index],
esm->rx_buf[thread_index],
actual_transfer, 0);
}
else
{
n_written = app_send_dgram_raw (tx_fifo, &at,
esm->vpp_queue[s->thread_index],
esm->rx_buf[thread_index],
actual_transfer, 0);
}
if (PREDICT_FALSE (n_written < max_dequeue))
if (n_written != max_transfer)
clib_warning ("short trout! written %u read %u", n_written, max_transfer);
if (PREDICT_FALSE (svm_fifo_max_dequeue (rx_fifo)))
goto rx_event;
return 0;
@@ -411,6 +428,7 @@ echo_server_create_command_fn (vlib_main_t * vm, unformat_input_t * input,
esm->private_segment_count = 0;
esm->private_segment_size = 0;
esm->tls_engine = TLS_ENGINE_OPENSSL;
esm->is_dgram = 0;
vec_free (esm->server_uri);
while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
@@ -479,6 +497,8 @@ echo_server_create_command_fn (vlib_main_t * vm, unformat_input_t * input,
clib_warning ("No uri provided! Using default: %s", default_uri);
esm->server_uri = (char *) format (0, "%s%c", default_uri, 0);
}
if (esm->server_uri[0] == 'u' && esm->server_uri[3] != 'c')
esm->is_dgram = 1;
rv = echo_server_create (vm, appns_id, appns_flags, appns_secret);
vec_free (appns_id);
src/vnet/session/application.c
+8 -5
View File
@@ -445,13 +445,9 @@ application_start_listen (application_t * srv, session_endpoint_t * sep,
s = listen_session_new (0, sst);
s->app_index = srv->index;
if (stream_session_listen (s, sep))
goto err;
/* Allocate segment manager. All sessions derived out of a listen session
* have fifos allocated by the same segment manager. */
sm = application_alloc_segment_manager (srv);
if (sm == 0)
if (!(sm = application_alloc_segment_manager (srv)))
goto err;
/* Add to app's listener table. Useful to find all child listeners
@@ -459,6 +455,13 @@ application_start_listen (application_t * srv, session_endpoint_t * sep,
handle = listen_session_get_handle (s);
hash_set (srv->listeners_table, handle, segment_manager_index (sm));
if (stream_session_listen (s, sep))
{
segment_manager_del (sm);
hash_unset (srv->listeners_table, handle);
goto err;
}
*res = handle;
return 0;
src/vnet/session/application_interface.h
+157
View File
@@ -173,6 +173,163 @@ extern const u32 test_srv_crt_rsa_len;
extern const char test_srv_key_rsa[];
extern const u32 test_srv_key_rsa_len;
typedef struct app_session_transport_
{
ip46_address_t rmt_ip; /**< remote ip */
ip46_address_t lcl_ip; /**< local ip */
u16 rmt_port; /**< remote port */
u16 lcl_port; /**< local port */
u8 is_ip4; /**< set if uses ip4 networking */
} app_session_transport_t;
typedef struct app_session_
{
svm_fifo_t *rx_fifo; /**< rx fifo */
svm_fifo_t *tx_fifo; /**< tx fifo */
session_type_t session_type; /**< session type */
volatile u8 session_state; /**< session state */
u32 session_index; /**< index in owning pool */
app_session_transport_t transport; /**< transport info */
svm_queue_t *vpp_evt_q; /**< vpp event queue for session */
u8 is_dgram; /**< set if it works in dgram mode */
} app_session_t;
always_inline int
app_send_dgram_raw (svm_fifo_t * f, app_session_transport_t * at,
svm_queue_t * vpp_evt_q, u8 * data, u32 len, u8 noblock)
{
u32 max_enqueue, actual_write;
session_dgram_hdr_t hdr;
session_fifo_event_t evt;
int rv;
max_enqueue = svm_fifo_max_enqueue (f);
if (svm_fifo_max_enqueue (f) <= sizeof (session_dgram_hdr_t))
return 0;
max_enqueue -= sizeof (session_dgram_hdr_t);
actual_write = clib_min (len, max_enqueue);
hdr.data_length = actual_write;
hdr.data_offset = 0;
clib_memcpy (&hdr.rmt_ip, &at->rmt_ip, sizeof (ip46_address_t));
hdr.is_ip4 = at->is_ip4;
hdr.rmt_port = at->rmt_port;
clib_memcpy (&hdr.lcl_ip, &at->lcl_ip, sizeof (ip46_address_t));
hdr.lcl_port = at->lcl_port;
rv = svm_fifo_enqueue_nowait (f, sizeof (hdr), (u8 *) & hdr);
if (rv <= 0)
return 0;
ASSERT (rv == sizeof (hdr));
if ((rv = svm_fifo_enqueue_nowait (f, actual_write, data)) > 0)
{
if (svm_fifo_set_event (f))
{
evt.fifo = f;
evt.event_type = FIFO_EVENT_APP_TX;
svm_queue_add (vpp_evt_q, (u8 *) & evt, noblock);
}
}
return rv;
}
always_inline int
app_send_dgram (app_session_t * s, u8 * data, u32 len, u8 noblock)
{
return app_send_dgram_raw (s->tx_fifo, &s->transport, s->vpp_evt_q, data,
len, noblock);
}
always_inline int
app_send_stream_raw (svm_fifo_t * f, svm_queue_t * vpp_evt_q, u8 * data,
u32 len, u8 noblock)
{
session_fifo_event_t evt;
int rv;
if ((rv = svm_fifo_enqueue_nowait (f, len, data)) > 0)
{
if (svm_fifo_set_event (f))
{
evt.fifo = f;
evt.event_type = FIFO_EVENT_APP_TX;
svm_queue_add (vpp_evt_q, (u8 *) & evt, noblock);
}
}
return rv;
}
always_inline int
app_send_stream (app_session_t * s, u8 * data, u32 len, u8 noblock)
{
return app_send_stream_raw (s->tx_fifo, s->vpp_evt_q, data, len, noblock);
}
always_inline int
app_send (app_session_t * s, u8 * data, u32 len, u8 noblock)
{
if (s->is_dgram)
return app_send_dgram (s, data, len, noblock);
return app_send_stream (s, data, len, noblock);
}
always_inline int
app_recv_dgram_raw (svm_fifo_t * f, u8 * buf, u32 len,
app_session_transport_t * at, u8 clear_evt)
{
session_dgram_pre_hdr_t ph;
u32 max_deq;
int rv;
if (clear_evt)
svm_fifo_unset_event (f);
max_deq = svm_fifo_max_dequeue (f);
if (max_deq < sizeof (session_dgram_hdr_t))
return 0;
svm_fifo_peek (f, 0, sizeof (ph), (u8 *) & ph);
ASSERT (ph.data_length >= ph.data_offset);
if (!ph.data_offset)
svm_fifo_peek (f, sizeof (ph), sizeof (*at), (u8 *) at);
len = clib_min (len, ph.data_length - ph.data_offset);
rv = svm_fifo_peek (f, ph.data_offset + SESSION_CONN_HDR_LEN, len, buf);
ph.data_offset += rv;
if (ph.data_offset == ph.data_length)
svm_fifo_dequeue_drop (f, ph.data_length + SESSION_CONN_HDR_LEN);
else
svm_fifo_overwrite_head (f, (u8 *) & ph, sizeof (ph));
return rv;
}
always_inline int
app_recv_dgram (app_session_t * s, u8 * buf, u32 len)
{
return app_recv_dgram_raw (s->rx_fifo, buf, len, &s->transport, 1);
}
always_inline int
app_recv_stream_raw (svm_fifo_t * f, u8 * buf, u32 len, u8 clear_evt)
{
if (clear_evt)
svm_fifo_unset_event (f);
return svm_fifo_dequeue_nowait (f, len, buf);
}
always_inline int
app_recv_stream (app_session_t * s, u8 * buf, u32 len)
{
return app_recv_stream_raw (s->rx_fifo, buf, len, 1);
}
always_inline int
app_recv (app_session_t * s, u8 * data, u32 len)
{
if (s->is_dgram)
return app_recv_dgram (s, data, len);
return app_recv_stream (s, data, len);
}
#endif /* __included_uri_h__ */
/*
src/vnet/session/session.api
+23 -5
View File
@@ -13,7 +13,7 @@
* limitations under the License.
*/
option version = "1.0.2";
option version = "1.0.3";
/** \brief client->vpp, attach application to session layer
@param client_index - opaque cookie to identify the sender
@@ -119,13 +119,25 @@ autoreply define unmap_segment {
"tcp://::/0/80" [ipv6] etc.
@param options - socket options, fifo sizes, etc.
*/
autoreply define bind_uri {
define bind_uri {
u32 client_index;
u32 context;
u32 accept_cookie;
u8 uri[128];
};
define bind_uri_reply {
u32 context;
u64 handle;
i32 retval;
u64 rx_fifo;
u64 tx_fifo;
u8 lcl_is_ip4;
u8 lcl_ip[16];
u16 lcl_port;
u64 vpp_evt_q;
};
/** \brief Unbind a given URI
@param client_index - opaque cookie to identify the sender
@param context - sender context, to match reply w/ request
@@ -314,8 +326,12 @@ autoreply define connect_sock {
@param context - sender context, to match reply w/ request
@param handle - bind handle
@param retval - return code for the request
@param event_queue_address - vpp event queue address or 0 if this
connection shouldn't send events
@param lcl_is_ip4 - local ip address type
@param lcl_ip6 - local ip address
@param lcl_port - local port
@param rx_fifo - rx fifo address if allocated (connectionless)
@param tx_fifo - tx fifo address if allocated (connectionless)
@param vpp_evt_q - vpp event queue address (connectionless)
@param segment_name_length - length of segment name
@param segment_name - name of segment client needs to attach to
*/
@@ -323,10 +339,12 @@ define bind_sock_reply {
u32 context;
u64 handle;
i32 retval;
u64 server_event_queue_address;
u8 lcl_is_ip4;
u8 lcl_ip[16];
u16 lcl_port;
u64 rx_fifo;
u64 tx_fifo;
u64 vpp_evt_q;
u32 segment_size;
u8 segment_name_length;
u8 segment_name[128];
src/vnet/session/session.c
+71 -24
View File
@@ -355,14 +355,19 @@ session_enqueue_stream_connection (transport_connection_t * tc,
return enqueued;
}
int
session_enqueue_dgram_connection (stream_session_t * s, vlib_buffer_t * b,
u8 proto, u8 queue_event)
session_enqueue_dgram_connection (stream_session_t * s,
session_dgram_hdr_t * hdr,
vlib_buffer_t * b, u8 proto, u8 queue_event)
{
int enqueued = 0, rv, in_order_off;
if (svm_fifo_max_enqueue (s->server_rx_fifo) < b->current_length)
return -1;
ASSERT (svm_fifo_max_enqueue (s->server_rx_fifo)
>= b->current_length + sizeof (*hdr));
svm_fifo_enqueue_nowait (s->server_rx_fifo, sizeof (session_dgram_hdr_t),
(u8 *) hdr);
enqueued = svm_fifo_enqueue_nowait (s->server_rx_fifo, b->current_length,
vlib_buffer_get_current (b));
if (PREDICT_FALSE ((b->flags & VLIB_BUFFER_NEXT_PRESENT) && enqueued >= 0))
@@ -530,6 +535,16 @@ session_manager_flush_enqueue_events (u8 transport_proto, u32 thread_index)
return errors;
}
int
session_manager_flush_all_enqueue_events (u8 transport_proto)
{
vlib_thread_main_t *vtm = vlib_get_thread_main ();
int i, errors = 0;
for (i = 0; i < 1 + vtm->n_threads; i++)
errors += session_manager_flush_enqueue_events (transport_proto, i);
return errors;
}
/**
* Init fifo tail and head pointers
*
@@ -825,7 +840,7 @@ session_open_cl (u32 app_index, session_endpoint_t * rmt, u32 opaque)
if (session_alloc_and_init (sm, tc, 1, &s))
return -1;
s->app_index = app->index;
s->session_state = SESSION_STATE_CONNECTING_READY;
s->session_state = SESSION_STATE_OPENED;
/* Tell the app about the new event fifo for this session */
app->cb_fns.session_connected_callback (app->index, opaque, s, 0);
@@ -841,10 +856,6 @@ session_open_vc (u32 app_index, session_endpoint_t * rmt, u32 opaque)
u64 handle;
int rv;
/* TODO until udp is fixed */
if (rmt->transport_proto == TRANSPORT_PROTO_UDP)
return session_open_cl (app_index, rmt, opaque);
tep = session_endpoint_to_transport (rmt);
rv = tp_vfts[rmt->transport_proto].open (tep);
if (rv < 0)
@@ -912,14 +923,6 @@ session_open (u32 app_index, session_endpoint_t * rmt, u32 opaque)
return session_open_srv_fns[tst] (app_index, rmt, opaque);
}
/**
* Ask transport to listen on local transport endpoint.
*
* @param s Session for which listen will be called. Note that unlike
* established sessions, listen sessions are not associated to a
* thread.
* @param tep Local endpoint to be listened on.
*/
int
session_listen_vc (stream_session_t * s, session_endpoint_t * sep)
{
@@ -947,6 +950,40 @@ session_listen_vc (stream_session_t * s, session_endpoint_t * sep)
return 0;
}
int
session_listen_cl (stream_session_t * s, session_endpoint_t * sep)
{
transport_connection_t *tc;
application_t *server;
segment_manager_t *sm;
u32 tci;
/* Transport bind/listen */
tci = tp_vfts[sep->transport_proto].bind (s->session_index,
session_endpoint_to_transport
(sep));
if (tci == (u32) ~ 0)
return -1;
/* Attach transport to session */
s->connection_index = tci;
tc = tp_vfts[sep->transport_proto].get_listener (tci);
/* Weird but handle it ... */
if (tc == 0)
return -1;
server = application_get (s->app_index);
sm = application_get_listen_segment_manager (server, s);
if (session_alloc_fifos (sm, s))
return -1;
/* Add to the main lookup table */
session_lookup_add_connection (tc, s->session_index);
return 0;
}
int
session_listen_app (stream_session_t * s, session_endpoint_t * sep)
{
@@ -965,11 +1002,19 @@ typedef int (*session_listen_service_fn) (stream_session_t *,
static session_listen_service_fn
session_listen_srv_fns[TRANSPORT_N_SERVICES] = {
session_listen_vc,
session_listen_vc,
session_listen_cl,
session_listen_app,
};
/* *INDENT-ON* */
/**
* Ask transport to listen on local transport endpoint.
*
* @param s Session for which listen will be called. Note that unlike
* established sessions, listen sessions are not associated to a
* thread.
* @param tep Local endpoint to be listened on.
*/
int
stream_session_listen (stream_session_t * s, session_endpoint_t * sep)
{
@@ -1125,7 +1170,8 @@ session_manager_get_evt_q_segment (void)
static session_fifo_rx_fn *session_tx_fns[TRANSPORT_TX_N_FNS] = {
session_tx_fifo_peek_and_snd,
session_tx_fifo_dequeue_and_snd,
session_tx_fifo_dequeue_internal
session_tx_fifo_dequeue_internal,
session_tx_fifo_dequeue_and_snd
};
/* *INDENT-ON* */
@@ -1228,11 +1274,12 @@ session_manager_main_enable (vlib_main_t * vm)
vec_validate (smm->peekers_rw_locks, num_threads - 1);
for (i = 0; i < TRANSPORT_N_PROTO; i++)
for (j = 0; j < num_threads; j++)
{
vec_validate (smm->session_to_enqueue[i], num_threads - 1);
vec_validate (smm->current_enqueue_epoch[i], num_threads - 1);
}
{
vec_validate (smm->current_enqueue_epoch[i], num_threads - 1);
vec_validate (smm->session_to_enqueue[i], num_threads - 1);
for (j = 0; j < num_threads; j++)
smm->current_enqueue_epoch[i][j] = 1;
}
for (i = 0; i < num_threads; i++)
{
src/vnet/session/session.h
+38 -2
View File
@@ -204,6 +204,31 @@ struct _session_manager_main
};
typedef struct session_dgram_pre_hdr_
{
u32 data_length;
u32 data_offset;
} session_dgram_pre_hdr_t;
/* *INDENT-OFF* */
typedef CLIB_PACKED (struct session_dgram_header_
{
u32 data_length;
u32 data_offset;
ip46_address_t rmt_ip;
ip46_address_t lcl_ip;
u16 rmt_port;
u16 lcl_port;
u8 is_ip4;
}) session_dgram_hdr_t;
/* *INDENT-ON* */
#define SESSION_CONN_ID_LEN 37
#define SESSION_CONN_HDR_LEN 45
STATIC_ASSERT (sizeof (session_dgram_hdr_t) == (SESSION_CONN_ID_LEN + 8),
"session conn id wrong length");
extern session_manager_main_t session_manager_main;
extern vlib_node_registration_t session_queue_node;
@@ -342,6 +367,14 @@ session_has_transport (stream_session_t * s)
return (session_get_transport_proto (s) != TRANSPORT_PROTO_NONE);
}
always_inline transport_service_type_t
session_transport_service_type (stream_session_t * s)
{
transport_proto_t tp;
tp = session_get_transport_proto (s);
return transport_protocol_service_type (tp);
}
/**
* Acquires a lock that blocks a session pool from expanding.
*
@@ -442,8 +475,10 @@ int
session_enqueue_stream_connection (transport_connection_t * tc,
vlib_buffer_t * b, u32 offset,
u8 queue_event, u8 is_in_order);
int session_enqueue_dgram_connection (stream_session_t * s, vlib_buffer_t * b,
u8 proto, u8 queue_event);
int session_enqueue_dgram_connection (stream_session_t * s,
session_dgram_hdr_t * hdr,
vlib_buffer_t * b, u8 proto,
u8 queue_event);
int stream_session_peek_bytes (transport_connection_t * tc, u8 * buffer,
u32 offset, u32 max_bytes);
u32 stream_session_dequeue_drop (transport_connection_t * tc, u32 max_bytes);
@@ -490,6 +525,7 @@ session_manager_get_vpp_event_queue (u32 thread_index)
}
int session_manager_flush_enqueue_events (u8 proto, u32 thread_index);
int session_manager_flush_all_enqueue_events (u8 transport_proto);
always_inline u64
listen_session_get_handle (stream_session_t * s)
src/vnet/session/session_api.c
+37 -4
View File
@@ -480,9 +480,12 @@ done:
static void
vl_api_bind_uri_t_handler (vl_api_bind_uri_t * mp)
{
vl_api_bind_uri_reply_t *rmp;
transport_connection_t *tc = 0;
vnet_bind_args_t _a, *a = &_a;
application_t *app;
vl_api_bind_uri_reply_t *rmp;
stream_session_t *s;
application_t *app = 0;
svm_queue_t *vpp_evt_q;
int rv;
if (session_manager_is_enabled () == 0)
@@ -505,7 +508,30 @@ vl_api_bind_uri_t_handler (vl_api_bind_uri_t * mp)
}
done:
REPLY_MACRO (VL_API_BIND_URI_REPLY);
/* *INDENT-OFF* */
REPLY_MACRO2 (VL_API_BIND_URI_REPLY, ({
if (!rv)
{
rmp->handle = a->handle;
rmp->lcl_is_ip4 = tc->is_ip4;
rmp->lcl_port = tc->lcl_port;
if (app && application_has_global_scope (app))
{
s = listen_session_get_from_handle (a->handle);
tc = listen_session_get_transport (s);
clib_memcpy (rmp->lcl_ip, &tc->lcl_ip, sizeof(tc->lcl_ip));
if (session_transport_service_type (s) == TRANSPORT_SERVICE_CL)
{
rmp->rx_fifo = pointer_to_uword (s->server_rx_fifo);
rmp->tx_fifo = pointer_to_uword (s->server_tx_fifo);
vpp_evt_q = session_manager_get_vpp_event_queue (0);
rmp->vpp_evt_q = pointer_to_uword (vpp_evt_q);
}
}
}
}));
/* *INDENT-ON* */
}
static void
@@ -733,6 +759,7 @@ vl_api_bind_sock_t_handler (vl_api_bind_sock_t * mp)
stream_session_t *s;
transport_connection_t *tc = 0;
ip46_address_t *ip46;
svm_queue_t *vpp_evt_q;
if (session_manager_is_enabled () == 0)
{
@@ -775,8 +802,14 @@ done:
{
s = listen_session_get_from_handle (a->handle);
tc = listen_session_get_transport (s);
rmp->lcl_is_ip4 = tc->is_ip4;
clib_memcpy (rmp->lcl_ip, &tc->lcl_ip, sizeof (tc->lcl_ip));
if (session_transport_service_type (s) == TRANSPORT_SERVICE_CL)
{
rmp->rx_fifo = pointer_to_uword (s->server_rx_fifo);
rmp->tx_fifo = pointer_to_uword (s->server_tx_fifo);
vpp_evt_q = session_manager_get_vpp_event_queue (0);
rmp->vpp_evt_q = pointer_to_uword (vpp_evt_q);
}
}
}
}));
src/vnet/session/session_cli.c
+5
View File
@@ -23,6 +23,9 @@ format_stream_session_fifos (u8 * s, va_list * args)
session_fifo_event_t _e, *e = &_e;
u8 found;
if (!ss->server_rx_fifo || !ss->server_tx_fifo)
return s;
s = format (s, " Rx fifo: %U", format_svm_fifo, ss->server_rx_fifo, 1);
if (verbose > 2 && ss->server_rx_fifo->has_event)
{
@@ -76,6 +79,8 @@ format_stream_session (u8 * s, va_list * args)
{
s = format (s, "%-40U%v", tp_vft->format_listener, ss->connection_index,
str);
if (verbose > 1)
s = format (s, "\n%U", format_stream_session_fifos, ss, verbose);
}
else if (ss->session_state == SESSION_STATE_CONNECTING)
{
src/vnet/session/session_node.c
+103 -31
View File
@@ -70,7 +70,7 @@ session_tx_fifo_chain_tail (session_manager_main_t * smm, vlib_main_t * vm,
vlib_buffer_t * b0, u32 bi0, u8 n_bufs_per_seg,
u32 left_from_seg, u32 * left_to_snd0,
u16 * n_bufs, u32 * tx_offset, u16 deq_per_buf,
u8 peek_data)
u8 peek_data, transport_tx_fn_type_t tx_type)
{
vlib_buffer_t *chain_b0, *prev_b0;
u32 chain_bi0, to_deq;
@@ -102,7 +102,23 @@ session_tx_fifo_chain_tail (session_manager_main_t * smm, vlib_main_t * vm,
}
else
{
n_bytes_read = svm_fifo_dequeue_nowait (fifo, len_to_deq0, data0);
if (tx_type == TRANSPORT_TX_DGRAM)
{
session_dgram_hdr_t *hdr;
u16 deq_now;
hdr = (session_dgram_hdr_t *) svm_fifo_head (fifo);
deq_now = clib_min (hdr->data_length - hdr->data_offset,
len_to_deq0);
n_bytes_read = svm_fifo_peek (fifo, hdr->data_offset, deq_now,
data0);
ASSERT (n_bytes_read > 0);
hdr->data_offset += n_bytes_read;
if (hdr->data_offset == hdr->data_length)
svm_fifo_dequeue_drop (fifo, hdr->data_length);
}
else
n_bytes_read = svm_fifo_dequeue_nowait (fifo, len_to_deq0, data0);
}
ASSERT (n_bytes_read == len_to_deq0);
chain_b0->current_length = n_bytes_read;
@@ -145,12 +161,35 @@ session_tx_fifo_read_and_snd_i (vlib_main_t * vm, vlib_node_runtime_t * node,
int i, n_bytes_read;
u32 n_bytes_per_buf, deq_per_buf, deq_per_first_buf;
u32 bufs_alloc, bufs_now;
session_dgram_hdr_t hdr;
next_index = next0 = smm->session_type_to_next[s0->session_type];
tp = session_get_transport_proto (s0);
transport_vft = transport_protocol_get_vft (tp);
tc0 = transport_vft->get_connection (s0->connection_index, thread_index);
if (peek_data)
{
if (PREDICT_FALSE (s0->session_state < SESSION_STATE_READY))
{
/* Can retransmit for closed sessions but can't send new data if
* session is not ready or closed */
vec_add1 (smm->pending_event_vector[thread_index], *e0);
return 0;
}
tc0 =
transport_vft->get_connection (s0->connection_index, thread_index);
}
else
{
if (s0->session_state == SESSION_STATE_LISTENING)
tc0 = transport_vft->get_listener (s0->connection_index);
else
{
if (PREDICT_FALSE (s0->session_state == SESSION_STATE_CLOSED))
return 0;
tc0 = transport_vft->get_connection (s0->connection_index,
thread_index);
}
}
/* Make sure we have space to send and there's something to dequeue */
snd_mss0 = transport_vft->send_mss (tc0);
@@ -168,20 +207,26 @@ session_tx_fifo_read_and_snd_i (vlib_main_t * vm, vlib_node_runtime_t * node,
/* Check how much we can pull. */
max_dequeue0 = svm_fifo_max_dequeue (s0->server_tx_fifo);
if (peek_data)
{
/* Offset in rx fifo from where to peek data */
tx_offset = transport_vft->tx_fifo_offset (tc0);
if (PREDICT_FALSE (tx_offset >= max_dequeue0))
max_dequeue0 = 0;
else
max_dequeue0 -= tx_offset;
return 0;
max_dequeue0 -= tx_offset;
}
/* Nothing to read return */
if (max_dequeue0 == 0)
return 0;
else
{
if (transport_vft->tx_type == TRANSPORT_TX_DGRAM)
{
if (max_dequeue0 < sizeof (hdr))
return 0;
svm_fifo_peek (s0->server_tx_fifo, 0, sizeof (hdr), (u8 *) & hdr);
ASSERT (hdr.data_length > hdr.data_offset);
max_dequeue0 = hdr.data_length - hdr.data_offset;
}
}
ASSERT (max_dequeue0 > 0);
/* Ensure we're not writing more than transport window allows */
if (max_dequeue0 < snd_space0)
@@ -286,14 +331,42 @@ session_tx_fifo_read_and_snd_i (vlib_main_t * vm, vlib_node_runtime_t * node,
}
else
{
n_bytes_read = svm_fifo_dequeue_nowait (s0->server_tx_fifo,
len_to_deq0, data0);
if (n_bytes_read <= 0)
goto dequeue_fail;
if (transport_vft->tx_type == TRANSPORT_TX_DGRAM)
{
svm_fifo_t *f = s0->server_tx_fifo;
u16 deq_now;
u32 offset;
ASSERT (hdr.data_length > hdr.data_offset);
deq_now = clib_min (hdr.data_length - hdr.data_offset,
len_to_deq0);
offset = hdr.data_offset + SESSION_CONN_HDR_LEN;
n_bytes_read = svm_fifo_peek (f, offset, deq_now, data0);
if (PREDICT_FALSE (n_bytes_read <= 0))
goto dequeue_fail;
if (s0->session_state == SESSION_STATE_LISTENING)
{
ip_copy (&tc0->rmt_ip, &hdr.rmt_ip, tc0->is_ip4);
tc0->rmt_port = hdr.rmt_port;
}
hdr.data_offset += n_bytes_read;
if (hdr.data_offset == hdr.data_length)
{
offset = hdr.data_length + SESSION_CONN_HDR_LEN;
svm_fifo_dequeue_drop (f, offset);
}
}
else
{
n_bytes_read = svm_fifo_dequeue_nowait (s0->server_tx_fifo,
len_to_deq0, data0);
if (n_bytes_read <= 0)
goto dequeue_fail;
}
}
b0->current_length = n_bytes_read;
left_to_snd0 -= n_bytes_read;
*n_tx_packets = *n_tx_packets + 1;
@@ -307,7 +380,8 @@ session_tx_fifo_read_and_snd_i (vlib_main_t * vm, vlib_node_runtime_t * node,
s0->server_tx_fifo, b0, bi0,
n_bufs_per_seg, left_for_seg,
&left_to_snd0, &n_bufs, &tx_offset,
deq_per_buf, peek_data);
deq_per_buf, peek_data,
transport_vft->tx_type);
}
/* Ask transport to push header after current_length and
@@ -345,12 +419,18 @@ session_tx_fifo_read_and_snd_i (vlib_main_t * vm, vlib_node_runtime_t * node,
/* If we couldn't dequeue all bytes mark as partially read */
if (max_len_to_snd0 < max_dequeue0)
if (svm_fifo_set_event (s0->server_tx_fifo))
vec_add1 (smm->pending_event_vector[thread_index], *e0);
if (!peek_data && transport_vft->tx_type == TRANSPORT_TX_DGRAM)
{
/* If we don't already have new event */
if (svm_fifo_set_event (s0->server_tx_fifo))
{
vec_add1 (smm->pending_event_vector[thread_index], *e0);
}
/* Fix dgram pre header */
if (max_len_to_snd0 < max_dequeue0)
svm_fifo_overwrite_head (s0->server_tx_fifo, (u8 *) & hdr,
sizeof (session_dgram_pre_hdr_t));
/* More data needs to be read */
else if (svm_fifo_max_dequeue (s0->server_tx_fifo) > 0)
vec_add1 (smm->pending_event_vector[thread_index], *e0);
}
return 0;
@@ -360,7 +440,6 @@ dequeue_fail:
* read, return buff to free list and return
*/
clib_warning ("dequeue fail");
if (svm_fifo_set_event (s0->server_tx_fifo))
{
vec_add1 (smm->pending_event_vector[thread_index], *e0);
@@ -638,13 +717,6 @@ skip_dequeue:
clib_warning ("It's dead, Jim!");
continue;
}
/* Can retransmit for closed sessions but can't do anything if
* session is not ready or closed */
if (PREDICT_FALSE (s0->session_state < SESSION_STATE_READY))
{
vec_add1 (smm->pending_event_vector[my_thread_index], *e0);
continue;
}
/* Spray packets in per session type frames, since they go to
* different nodes */
rv = (smm->session_tx_fns[s0->session_type]) (vm, node, smm, e0, s0,
src/vnet/session/stream_session.h
+8 -19
View File
@@ -31,30 +31,19 @@ typedef enum
SESSION_STATE_CONNECTING,
SESSION_STATE_ACCEPTING,
SESSION_STATE_READY,
SESSION_STATE_CONNECTING_READY,
SESSION_STATE_OPENED,
SESSION_STATE_CLOSED,
SESSION_STATE_N_STATES,
} stream_session_state_t;
/* TODO convert to macro once cleanup completed */
typedef struct app_session_
typedef struct generic_session_
{
/** fifo pointers. Once allocated, these do not move */
svm_fifo_t *server_rx_fifo;
svm_fifo_t *server_tx_fifo;
/** Type */
session_type_t session_type;
/** State */
volatile u8 session_state;
/** Session index in owning pool */
u32 session_index;
/** Application index */
u32 app_index;
} app_session_t;
svm_fifo_t *rx_fifo; /**< rx fifo */
svm_fifo_t *tx_fifo; /**< tx fifo */
session_type_t session_type; /**< session type */
volatile u8 session_state; /**< session state */
u32 session_index; /**< index in owning pool */
} generic_session_t;
typedef struct _stream_session_t
{
src/vnet/session/transport.c
+16
View File
@@ -57,6 +57,9 @@ format_transport_proto (u8 * s, va_list * args)
case TRANSPORT_PROTO_SCTP:
s = format (s, "SCTP");
break;
case TRANSPORT_PROTO_UDPC:
s = format (s, "UDPC");
break;
}
return s;
}
@@ -76,6 +79,9 @@ format_transport_proto_short (u8 * s, va_list * args)
case TRANSPORT_PROTO_SCTP:
s = format (s, "S");
break;
case TRANSPORT_PROTO_UDPC:
s = format (s, "U");
break;
}
return s;
}
@@ -100,6 +106,10 @@ unformat_transport_proto (unformat_input_t * input, va_list * args)
*proto = TRANSPORT_PROTO_TLS;
else if (unformat (input, "TLS"))
*proto = TRANSPORT_PROTO_TLS;
else if (unformat (input, "udpc"))
*proto = TRANSPORT_PROTO_UDPC;
else if (unformat (input, "UDPC"))
*proto = TRANSPORT_PROTO_UDPC;
else
return 0;
return 1;
@@ -185,6 +195,12 @@ transport_protocol_get_vft (transport_proto_t transport_proto)
return &tp_vfts[transport_proto];
}
transport_service_type_t
transport_protocol_service_type (transport_proto_t tp)
{
return tp_vfts[tp].service_type;
}
#define PORT_MASK ((1 << 16)- 1)
void
src/vnet/session/transport.h
+4 -15
View File
@@ -35,10 +35,10 @@ typedef struct _transport_connection
{
ip46_address_t rmt_ip; /**< Remote IP */
ip46_address_t lcl_ip; /**< Local IP */
u16 lcl_port; /**< Local port */
u16 rmt_port; /**< Remote port */
u8 proto; /**< Protocol id */
u16 lcl_port; /**< Local port */
u8 is_ip4; /**< Flag if IP4 connection */
u8 proto; /**< Protocol id */
u32 fib_index; /**< Network namespace */
};
/*
@@ -88,6 +88,7 @@ typedef enum _transport_proto
TRANSPORT_PROTO_SCTP,
TRANSPORT_PROTO_NONE,
TRANSPORT_PROTO_TLS,
TRANSPORT_PROTO_UDPC,
TRANSPORT_N_PROTO
} transport_proto_t;
@@ -99,7 +100,7 @@ uword unformat_transport_proto (unformat_input_t * input, va_list * args);
_(u32, sw_if_index) /**< interface endpoint is associated with */ \
_(ip46_address_t, ip) /**< ip address */ \
_(u32, fib_index) /**< fib table endpoint is associated with */ \
_(u8, is_ip4) /**< 1 if ip4 */ \
_(u8, is_ip4) /**< set if ip4 */ \
_(u16, port) /**< port in net order */ \
typedef struct _transport_endpoint
@@ -125,18 +126,6 @@ transport_endpoint_fib_proto (transport_endpoint_t * tep)
return tep->is_ip4 ? FIB_PROTOCOL_IP4 : FIB_PROTOCOL_IP6;
}
always_inline u8
transport_is_stream (u8 proto)
{
return ((proto == TRANSPORT_PROTO_TCP) || (proto == TRANSPORT_PROTO_SCTP));
}
always_inline u8
transport_is_dgram (u8 proto)
{
return (proto == TRANSPORT_PROTO_UDP);
}
int transport_alloc_local_port (u8 proto, ip46_address_t * ip);
int transport_alloc_local_endpoint (u8 proto, transport_endpoint_t * rmt,
ip46_address_t * lcl_addr,
src/vnet/session/transport_interface.h
+4 -2
View File
@@ -23,7 +23,8 @@ typedef enum transport_dequeue_type_
{
TRANSPORT_TX_PEEK, /**< reliable transport protos */
TRANSPORT_TX_DEQUEUE, /**< unreliable transport protos */
TRANSPORT_TX_INTERNAL, /**< apps acting as transports */
TRANSPORT_TX_INTERNAL, /**< apps acting as transports */
TRANSPORT_TX_DGRAM, /**< datagram mode */
TRANSPORT_TX_N_FNS
} transport_tx_fn_type_t;
@@ -31,7 +32,7 @@ typedef enum transport_service_type_
{
TRANSPORT_SERVICE_VC, /**< virtual circuit service */
TRANSPORT_SERVICE_CL, /**< connectionless service */
TRANSPORT_SERVICE_APP, /**< app transport service */
TRANSPORT_SERVICE_APP, /**< app transport service */
TRANSPORT_N_SERVICES
} transport_service_type_t;
@@ -96,6 +97,7 @@ void transport_register_protocol (transport_proto_t transport_proto,
const transport_proto_vft_t * vft,
fib_protocol_t fib_proto, u32 output_node);
transport_proto_vft_t *transport_protocol_get_vft (transport_proto_t tp);
transport_service_type_t transport_protocol_service_type (transport_proto_t);
void transport_update_time (f64 time_now, u8 thread_index);
void transport_enable_disable (vlib_main_t * vm, u8 is_en);
src/vnet/udp/udp.c
+53 -3
View File
@@ -49,6 +49,7 @@ udp_connection_alloc (u32 thread_index)
uc->c_c_index = uc - um->connections[thread_index];
uc->c_thread_index = thread_index;
uc->c_proto = TRANSPORT_PROTO_UDP;
clib_spinlock_init (&uc->rx_lock);
return uc;
}
@@ -92,6 +93,7 @@ udp_session_bind (u32 session_index, transport_endpoint_t * lcl)
listener->c_proto = TRANSPORT_PROTO_UDP;
listener->c_s_index = session_index;
listener->c_fib_index = lcl->fib_index;
clib_spinlock_init (&listener->rx_lock);
node_index = lcl->is_ip4 ? udp4_input_node.index : udp6_input_node.index;
udp_register_dst_port (vm, clib_net_to_host_u16 (lcl->port), node_index,
@@ -140,7 +142,7 @@ udp_push_header (transport_connection_t * tc, vlib_buffer_t * b)
vnet_buffer (b)->l3_hdr_offset = (u8 *) ih - b->data;
}
vnet_buffer (b)->sw_if_index[VLIB_RX] = 0;
vnet_buffer (b)->sw_if_index[VLIB_TX] = ~0;
vnet_buffer (b)->sw_if_index[VLIB_TX] = uc->c_fib_index;
b->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED;
return 0;
@@ -210,7 +212,7 @@ format_udp_connection (u8 * s, va_list * args)
if (verbose)
{
if (verbose == 1)
s = format (s, "%-15s", "-");
s = format (s, "%-15s\n", "-");
else
s = format (s, "\n");
}
@@ -322,8 +324,52 @@ const static transport_proto_vft_t udp_proto = {
.format_connection = format_udp_session,
.format_half_open = format_udp_half_open_session,
.format_listener = format_udp_listener_session,
.tx_type = TRANSPORT_TX_DGRAM,
.service_type = TRANSPORT_SERVICE_CL,
};
/* *INDENT-ON* */
int
udpc_connection_open (transport_endpoint_t * rmt)
{
udp_connection_t *uc;
u32 uc_index;
uc_index = udp_open_connection (rmt);
uc = udp_connection_get (uc_index, vlib_get_thread_index ());
uc->is_connected = 1;
return uc_index;
}
u32
udpc_connection_listen (u32 session_index, transport_endpoint_t * lcl)
{
udp_connection_t *listener;
u32 li;
li = udp_session_bind (session_index, lcl);
listener = udp_listener_get (li);
listener->is_connected = 1;
return li;
}
/* *INDENT-OFF* */
const static transport_proto_vft_t udpc_proto = {
.bind = udpc_connection_listen,
.open = udpc_connection_open,
.unbind = udp_session_unbind,
.push_header = udp_push_header,
.get_connection = udp_session_get,
.get_listener = udp_session_get_listener,
.get_half_open = udp_half_open_session_get_transport,
.close = udp_session_close,
.cleanup = udp_session_cleanup,
.send_mss = udp_send_mss,
.send_space = udp_send_space,
.format_connection = format_udp_session,
.format_half_open = format_udp_half_open_session,
.format_listener = format_udp_listener_session,
.tx_type = TRANSPORT_TX_DEQUEUE,
.service_type = TRANSPORT_SERVICE_VC,
.service_type = TRANSPORT_SERVICE_CL,
};
/* *INDENT-ON* */
@@ -361,6 +407,10 @@ udp_init (vlib_main_t * vm)
FIB_PROTOCOL_IP4, ip4_lookup_node.index);
transport_register_protocol (TRANSPORT_PROTO_UDP, &udp_proto,
FIB_PROTOCOL_IP6, ip6_lookup_node.index);
transport_register_protocol (TRANSPORT_PROTO_UDPC, &udpc_proto,
FIB_PROTOCOL_IP4, ip4_lookup_node.index);
transport_register_protocol (TRANSPORT_PROTO_UDPC, &udpc_proto,
FIB_PROTOCOL_IP6, ip6_lookup_node.index);
/*
* Initialize data structures
src/vnet/udp/udp.h
+4 -4
View File
@@ -36,9 +36,9 @@ typedef enum
typedef struct
{
transport_connection_t connection; /** must be first */
/** ersatz MTU to limit fifo pushes to test data size */
u32 mtu;
transport_connection_t connection; /**< must be first */
clib_spinlock_t rx_lock; /**< rx fifo lock */
u8 is_connected; /**< connected mode */
} udp_connection_t;
#define foreach_udp4_dst_port \
@@ -207,7 +207,7 @@ udp_pool_remove_peeker (u32 thread_index)
}
always_inline udp_connection_t *
udp_conenction_clone_safe (u32 connection_index, u32 thread_index)
udp_connection_clone_safe (u32 connection_index, u32 thread_index)
{
udp_connection_t *old_c, *new_c;
u32 current_thread_index = vlib_get_thread_index ();
src/vnet/udp/udp_input.c
+116 -66
View File
@@ -13,20 +13,19 @@
* limitations under the License.
*/
#include <vlibmemory/api.h>
#include <vlib/vlib.h>
#include <vnet/vnet.h>
#include <vnet/pg/pg.h>
#include <vnet/ip/ip.h>
#include <vnet/udp/udp.h>
#include <vppinfra/hash.h>
#include <vppinfra/error.h>
#include <vppinfra/elog.h>
#include <vnet/vnet.h>
#include <vnet/pg/pg.h>
#include <vnet/ip/ip.h>
#include <vnet/udp/udp.h>
#include <vnet/udp/udp_packet.h>
#include <vlibmemory/api.h>
#include "../session/application_interface.h"
#include <vnet/session/session.h>
static char *udp_error_strings[] = {
#define udp_error(n,s) s,
@@ -106,9 +105,11 @@ udp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
ip6_header_t *ip60;
u8 *data0;
stream_session_t *s0;
transport_connection_t *tc0 = 0;
udp_connection_t *child0, *new_uc0;
int written0;
udp_connection_t *uc0, *child0, *new_uc0;
transport_connection_t *tc0;
int wrote0;
void *rmt_addr, *lcl_addr;
session_dgram_hdr_t hdr0;
/* speculatively enqueue b0 to the current next frame */
bi0 = from[0];
@@ -127,11 +128,14 @@ udp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
if (is_ip4)
{
/* $$$$ fixme: udp_local doesn't do ip options correctly anyhow */
/* TODO: must fix once udp_local does ip options correctly */
ip40 = (ip4_header_t *) (((u8 *) udp0) - sizeof (*ip40));
s0 = session_lookup_safe4 (fib_index0, &ip40->dst_address,
&ip40->src_address, udp0->dst_port,
udp0->src_port, TRANSPORT_PROTO_UDP);
lcl_addr = &ip40->dst_address;
rmt_addr = &ip40->src_address;
}
else
{
@@ -139,67 +143,83 @@ udp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
s0 = session_lookup_safe6 (fib_index0, &ip60->dst_address,
&ip60->src_address, udp0->dst_port,
udp0->src_port, TRANSPORT_PROTO_UDP);
lcl_addr = &ip60->dst_address;
rmt_addr = &ip60->src_address;
}
if (PREDICT_FALSE (s0 == 0))
if (PREDICT_FALSE (!s0))
{
error0 = UDP_ERROR_NO_LISTENER;
goto trace0;
}
if (PREDICT_TRUE (s0->session_state == SESSION_STATE_READY))
if (s0->session_state == SESSION_STATE_OPENED)
{
/* TODO optimization: move cl session to right thread
* However, since such a move would affect the session handle,
* which we pass 'raw' to the app, we'd also have notify the
* app of the change or change the way we pass handles to apps.
*/
tc0 = session_get_transport (s0);
uc0 = udp_get_connection_from_transport (tc0);
if (uc0->is_connected)
{
/*
* Clone the transport. It will be cleaned up with the
* session once we notify the session layer.
*/
new_uc0 = udp_connection_clone_safe (s0->connection_index,
s0->thread_index);
ASSERT (s0->session_index == new_uc0->c_s_index);
/*
* Drop the 'lock' on pool resize
*/
session_pool_remove_peeker (s0->thread_index);
session_dgram_connect_notify (&new_uc0->connection,
s0->thread_index, &s0);
tc0 = &new_uc0->connection;
}
}
else if (s0->session_state == SESSION_STATE_READY)
{
tc0 = session_get_transport (s0);
}
else if (s0->session_state == SESSION_STATE_CONNECTING_READY)
{
/*
* Clone the transport. It will be cleaned up with the
* session once we notify the session layer.
*/
new_uc0 = udp_conenction_clone_safe (s0->connection_index,
s0->thread_index);
ASSERT (s0->session_index == new_uc0->c_s_index);
/*
* Drop the 'lock' on pool resize
*/
session_pool_remove_peeker (s0->thread_index);
session_dgram_connect_notify (&new_uc0->connection,
s0->thread_index, &s0);
tc0 = &new_uc0->connection;
uc0 = udp_get_connection_from_transport (tc0);
}
else if (s0->session_state == SESSION_STATE_LISTENING)
{
tc0 = listen_session_get_transport (s0);
child0 = udp_connection_alloc (my_thread_index);
if (is_ip4)
uc0 = udp_get_connection_from_transport (tc0);
if (uc0->is_connected)
{
ip_set (&child0->c_lcl_ip, &ip40->dst_address, 1);
ip_set (&child0->c_rmt_ip, &ip40->src_address, 1);
}
else
{
ip_set (&child0->c_lcl_ip, &ip60->dst_address, 0);
ip_set (&child0->c_rmt_ip, &ip60->src_address, 0);
}
child0->c_lcl_port = udp0->dst_port;
child0->c_rmt_port = udp0->src_port;
child0->c_is_ip4 = is_ip4;
child0->mtu = 1460; /* $$$$ policy */
child0 = udp_connection_alloc (my_thread_index);
if (is_ip4)
{
ip_set (&child0->c_lcl_ip, &ip40->dst_address, 1);
ip_set (&child0->c_rmt_ip, &ip40->src_address, 1);
}
else
{
ip_set (&child0->c_lcl_ip, &ip60->dst_address, 0);
ip_set (&child0->c_rmt_ip, &ip60->src_address, 0);
}
child0->c_lcl_port = udp0->dst_port;
child0->c_rmt_port = udp0->src_port;
child0->c_is_ip4 = is_ip4;
if (stream_session_accept
(&child0->connection, tc0->s_index, 1))
{
error0 = UDP_ERROR_CREATE_SESSION;
goto trace0;
if (stream_session_accept (&child0->connection,
tc0->s_index, 1))
{
error0 = UDP_ERROR_CREATE_SESSION;
goto trace0;
}
s0 =
session_get (child0->c_s_index, child0->c_thread_index);
s0->session_state = SESSION_STATE_READY;
tc0 = &child0->connection;
uc0 = udp_get_connection_from_transport (tc0);
error0 = UDP_ERROR_LISTENER;
}
s0 = session_get (child0->c_s_index, child0->c_thread_index);
s0->session_state = SESSION_STATE_READY;
tc0 = &child0->connection;
error0 = UDP_ERROR_LISTENER;
}
else
{
@@ -207,15 +227,48 @@ udp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
goto trace0;
}
written0 = session_enqueue_dgram_connection (s0, b0, tc0->proto,
1 /* queue evt */ );
if (PREDICT_FALSE (written0 < 0))
if (!uc0->is_connected)
{
error0 = UDP_ERROR_FIFO_FULL;
goto trace0;
if (svm_fifo_max_enqueue (s0->server_rx_fifo)
< b0->current_length + sizeof (session_dgram_hdr_t))
{
error0 = UDP_ERROR_FIFO_FULL;
goto trace0;
}
hdr0.data_length = b0->current_length;
hdr0.data_offset = 0;
ip_set (&hdr0.lcl_ip, lcl_addr, is_ip4);
ip_set (&hdr0.rmt_ip, rmt_addr, is_ip4);
hdr0.lcl_port = udp0->dst_port;
hdr0.rmt_port = udp0->src_port;
hdr0.is_ip4 = is_ip4;
clib_spinlock_lock (&uc0->rx_lock);
wrote0 = session_enqueue_dgram_connection (s0, &hdr0, b0,
TRANSPORT_PROTO_UDP,
1 /* queue evt */ );
clib_spinlock_unlock (&uc0->rx_lock);
ASSERT (wrote0 > 0);
if (s0->session_state != SESSION_STATE_LISTENING)
session_pool_remove_peeker (s0->thread_index);
}
else
{
if (svm_fifo_max_enqueue (s0->server_rx_fifo)
< b0->current_length)
{
error0 = UDP_ERROR_FIFO_FULL;
goto trace0;
}
wrote0 = session_enqueue_stream_connection (tc0, b0, 0,
1 /* queue evt */ ,
1 /* in order */ );
ASSERT (wrote0 > 0);
}
trace0:
b0->error = node->errors[error0];
if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE)
@@ -224,7 +277,7 @@ udp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
udp_input_trace_t *t = vlib_add_trace (vm, node, b0,
sizeof (*t));
t->connection = tc0 ? tc0->c_index : ~0;
t->connection = s0 ? s0->connection_index : ~0;
t->disposition = error0;
t->thread_index = my_thread_index;
}
@@ -237,14 +290,11 @@ udp46_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
vlib_put_next_frame (vm, node, next_index, n_left_to_next);
}
errors = session_manager_flush_enqueue_events (TRANSPORT_PROTO_UDP,
my_thread_index);
errors = session_manager_flush_all_enqueue_events (TRANSPORT_PROTO_UDP);
udp_input_inc_counter (vm, is_ip4, UDP_ERROR_EVENT_FIFO_FULL, errors);
return frame->n_vectors;
}
vlib_node_registration_t udp4_input_node;
vlib_node_registration_t udp6_input_node;
static uword
udp4_input (vlib_main_t * vm, vlib_node_runtime_t * node,