diff options
Diffstat (limited to 'src/vnet/session')
| -rw-r--r-- | src/vnet/session/application.c | 657 | ||||
| -rw-r--r-- | src/vnet/session/application.h | 136 | ||||
| -rw-r--r-- | src/vnet/session/application_interface.c | 406 | ||||
| -rw-r--r-- | src/vnet/session/application_interface.h | 183 | ||||
| -rw-r--r-- | src/vnet/session/segment_manager.c | 636 | ||||
| -rw-r--r-- | src/vnet/session/segment_manager.h | 131 | ||||
| -rw-r--r-- | src/vnet/session/session.api | 331 | ||||
| -rw-r--r-- | src/vnet/session/session.c | 1036 | ||||
| -rw-r--r-- | src/vnet/session/session.h | 436 | ||||
| -rwxr-xr-x | src/vnet/session/session_api.c | 763 | ||||
| -rwxr-xr-x | src/vnet/session/session_cli.c | 494 | ||||
| -rw-r--r-- | src/vnet/session/session_debug.h | 142 | ||||
| -rw-r--r-- | src/vnet/session/session_lookup.c | 619 | ||||
| -rw-r--r-- | src/vnet/session/session_lookup.h | 100 | ||||
| -rw-r--r-- | src/vnet/session/session_node.c | 707 | ||||
| -rw-r--r-- | src/vnet/session/stream_session.h | 92 | ||||
| -rw-r--r-- | src/vnet/session/transport.h | 94 | ||||
| -rw-r--r-- | src/vnet/session/transport_interface.c | 109 | ||||
| -rw-r--r-- | src/vnet/session/transport_interface.h | 82 |
19 files changed, 7154 insertions, 0 deletions
diff --git a/src/vnet/session/application.c b/src/vnet/session/application.c new file mode 100644 index 00000000..2b789c5f --- /dev/null +++ b/src/vnet/session/application.c @@ -0,0 +1,657 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/session/application.h> +#include <vnet/session/application_interface.h> +#include <vnet/session/session.h> + +/** + * Pool from which we allocate all applications + */ +static application_t *app_pool; + +/** + * Hash table of apps by api client index + */ +static uword *app_by_api_client_index; + +/** + * Default application event queue size + */ +static u32 default_app_evt_queue_size = 128; + +int +application_api_queue_is_full (application_t * app) +{ + unix_shared_memory_queue_t *q; + + /* builtin servers are always OK */ + if (app->api_client_index == ~0) + return 0; + + q = vl_api_client_index_to_input_queue (app->api_client_index); + if (!q) + return 1; + + if (q->cursize == q->maxsize) + return 1; + return 0; +} + +static void +application_table_add (application_t * app) +{ + hash_set (app_by_api_client_index, app->api_client_index, app->index); +} + +static void +application_table_del (application_t * app) +{ + hash_unset (app_by_api_client_index, app->api_client_index); +} + +application_t * +application_lookup (u32 api_client_index) +{ + uword *p; + p = hash_get (app_by_api_client_index, api_client_index); + if (p) + return application_get (p[0]); + + return 0; +} + +application_t * +application_new () +{ + application_t *app; + pool_get (app_pool, app); + memset (app, 0, sizeof (*app)); + app->index = application_get_index (app); + app->connects_seg_manager = APP_INVALID_SEGMENT_MANAGER_INDEX; + app->first_segment_manager = APP_INVALID_SEGMENT_MANAGER_INDEX; + if (CLIB_DEBUG > 1) + clib_warning ("[%d] New app (%d)", getpid (), app->index); + return app; +} + +void +application_del (application_t * app) +{ + segment_manager_t *sm; + u64 handle; + u32 index, *handles = 0; + int i; + vnet_unbind_args_t _a, *a = &_a; + + /* + * The app event queue allocated in first segment is cleared with + * the segment manager. No need to explicitly free it. + */ + if (CLIB_DEBUG > 1) + clib_warning ("[%d] Delete app (%d)", getpid (), app->index); + + /* + * Listener cleanup + */ + + /* *INDENT-OFF* */ + hash_foreach (handle, index, app->listeners_table, + ({ + vec_add1 (handles, handle); + sm = segment_manager_get (index); + sm->app_index = SEGMENT_MANAGER_INVALID_APP_INDEX; + })); + /* *INDENT-ON* */ + + for (i = 0; i < vec_len (handles); i++) + { + a->app_index = app->index; + a->handle = handles[i]; + /* seg manager is removed when unbind completes */ + vnet_unbind (a); + } + + /* + * Connects segment manager cleanup + */ + + if (app->connects_seg_manager != APP_INVALID_SEGMENT_MANAGER_INDEX) + { + sm = segment_manager_get (app->connects_seg_manager); + sm->app_index = SEGMENT_MANAGER_INVALID_APP_INDEX; + segment_manager_init_del (sm); + } + + + /* If first segment manager is used by a listener */ + if (app->first_segment_manager != APP_INVALID_SEGMENT_MANAGER_INDEX + && app->first_segment_manager != app->connects_seg_manager) + { + sm = segment_manager_get (app->first_segment_manager); + /* .. and has no fifos, e.g. it might be used for redirected sessions, + * remove it */ + if (!segment_manager_has_fifos (sm)) + { + sm->app_index = SEGMENT_MANAGER_INVALID_APP_INDEX; + segment_manager_del (sm); + } + } + + application_table_del (app); + pool_put (app_pool, app); +} + +static void +application_verify_cb_fns (session_cb_vft_t * cb_fns) +{ + if (cb_fns->session_accept_callback == 0) + clib_warning ("No accept callback function provided"); + if (cb_fns->session_connected_callback == 0) + clib_warning ("No session connected callback function provided"); + if (cb_fns->session_disconnect_callback == 0) + clib_warning ("No session disconnect callback function provided"); + if (cb_fns->session_reset_callback == 0) + clib_warning ("No session reset callback function provided"); +} + +int +application_init (application_t * app, u32 api_client_index, u64 * options, + session_cb_vft_t * cb_fns) +{ + segment_manager_t *sm; + segment_manager_properties_t *props; + u32 app_evt_queue_size, first_seg_size; + u32 default_rx_fifo_size = 16 << 10, default_tx_fifo_size = 16 << 10; + int rv; + + app_evt_queue_size = options[APP_EVT_QUEUE_SIZE] > 0 ? + options[APP_EVT_QUEUE_SIZE] : default_app_evt_queue_size; + + /* Setup segment manager */ + sm = segment_manager_new (); + sm->app_index = app->index; + props = &app->sm_properties; + props->add_segment_size = options[SESSION_OPTIONS_ADD_SEGMENT_SIZE]; + props->rx_fifo_size = options[SESSION_OPTIONS_RX_FIFO_SIZE]; + props->rx_fifo_size = + props->rx_fifo_size ? props->rx_fifo_size : default_rx_fifo_size; + props->tx_fifo_size = options[SESSION_OPTIONS_TX_FIFO_SIZE]; + props->tx_fifo_size = + props->tx_fifo_size ? props->tx_fifo_size : default_tx_fifo_size; + props->add_segment = props->add_segment_size != 0; + props->preallocated_fifo_pairs = options[APP_OPTIONS_PREALLOC_FIFO_PAIRS]; + props->use_private_segment = options[APP_OPTIONS_FLAGS] + & APP_OPTIONS_FLAGS_BUILTIN_APP; + props->private_segment_count = options[APP_OPTIONS_PRIVATE_SEGMENT_COUNT]; + props->private_segment_size = options[APP_OPTIONS_PRIVATE_SEGMENT_SIZE]; + + first_seg_size = options[SESSION_OPTIONS_SEGMENT_SIZE]; + if ((rv = segment_manager_init (sm, props, first_seg_size))) + return rv; + sm->first_is_protected = 1; + + app->first_segment_manager = segment_manager_index (sm); + app->api_client_index = api_client_index; + app->flags = options[APP_OPTIONS_FLAGS]; + app->cb_fns = *cb_fns; + + /* Allocate app event queue in the first shared-memory segment */ + app->event_queue = segment_manager_alloc_queue (sm, app_evt_queue_size); + + /* Check that the obvious things are properly set up */ + application_verify_cb_fns (cb_fns); + + /* Add app to lookup by api_client_index table */ + application_table_add (app); + + return 0; +} + +application_t * +application_get (u32 index) +{ + return pool_elt_at_index (app_pool, index); +} + +application_t * +application_get_if_valid (u32 index) +{ + if (pool_is_free_index (app_pool, index)) + return 0; + + return pool_elt_at_index (app_pool, index); +} + +u32 +application_get_index (application_t * app) +{ + return app - app_pool; +} + +static segment_manager_t * +application_alloc_segment_manager (application_t * app) +{ + segment_manager_t *sm = 0; + + /* If the first segment manager is not in use, don't allocate a new one */ + if (app->first_segment_manager != APP_INVALID_SEGMENT_MANAGER_INDEX + && app->first_segment_manager_in_use == 0) + { + sm = segment_manager_get (app->first_segment_manager); + app->first_segment_manager_in_use = 1; + return sm; + } + + sm = segment_manager_new (); + sm->properties = &app->sm_properties; + + return sm; +} + +/** + * Start listening local transport endpoint for requested transport. + * + * Creates a 'dummy' stream session with state LISTENING to be used in session + * lookups, prior to establishing connection. Requests transport to build + * it's own specific listening connection. + */ +int +application_start_listen (application_t * srv, session_type_t session_type, + transport_endpoint_t * tep, u64 * res) +{ + segment_manager_t *sm; + stream_session_t *s; + u64 handle; + + s = listen_session_new (session_type); + s->app_index = srv->index; + + if (stream_session_listen (s, tep)) + 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) + goto err; + + /* Add to app's listener table. Useful to find all child listeners + * when app goes down, although, just for unbinding this is not needed */ + handle = listen_session_get_handle (s); + hash_set (srv->listeners_table, handle, segment_manager_index (sm)); + + *res = handle; + return 0; + +err: + listen_session_del (s); + return -1; +} + +/** + * Stop listening on session associated to handle + */ +int +application_stop_listen (application_t * srv, u64 handle) +{ + stream_session_t *listener; + uword *indexp; + segment_manager_t *sm; + + if (srv && hash_get (srv->listeners_table, handle) == 0) + { + clib_warning ("app doesn't own handle %llu!", handle); + return -1; + } + + listener = listen_session_get_from_handle (handle); + stream_session_stop_listen (listener); + + indexp = hash_get (srv->listeners_table, handle); + ASSERT (indexp); + + sm = segment_manager_get (*indexp); + if (srv->first_segment_manager == *indexp) + { + /* Delete sessions but don't remove segment manager */ + srv->first_segment_manager_in_use = 0; + segment_manager_del_sessions (sm); + } + else + { + segment_manager_init_del (sm); + } + hash_unset (srv->listeners_table, handle); + listen_session_del (listener); + + return 0; +} + +int +application_open_session (application_t * app, session_type_t sst, + transport_endpoint_t * tep, u32 api_context) +{ + segment_manager_t *sm; + transport_connection_t *tc = 0; + int rv; + + /* Make sure we have a segment manager for connects */ + if (app->connects_seg_manager == (u32) ~ 0) + { + sm = application_alloc_segment_manager (app); + if (sm == 0) + return -1; + app->connects_seg_manager = segment_manager_index (sm); + } + + if ((rv = stream_session_open (app->index, sst, tep, &tc))) + return rv; + + /* Store api_context for when the reply comes. Not the nicest thing + * but better than allocating a separate half-open pool. */ + tc->s_index = api_context; + + return 0; +} + +segment_manager_t * +application_get_connect_segment_manager (application_t * app) +{ + ASSERT (app->connects_seg_manager != (u32) ~ 0); + return segment_manager_get (app->connects_seg_manager); +} + +segment_manager_t * +application_get_listen_segment_manager (application_t * app, + stream_session_t * s) +{ + uword *smp; + smp = hash_get (app->listeners_table, listen_session_get_handle (s)); + ASSERT (smp != 0); + return segment_manager_get (*smp); +} + +static u8 * +app_get_name_from_reg_index (application_t * app) +{ + u8 *app_name; + + vl_api_registration_t *regp; + regp = vl_api_client_index_to_registration (app->api_client_index); + if (!regp) + app_name = format (0, "builtin-%d%c", app->index, 0); + else + app_name = format (0, "%s%c", regp->name, 0); + + return app_name; +} + +int +application_is_proxy (application_t * app) +{ + return !(app->flags & APP_OPTIONS_FLAGS_IS_PROXY); +} + +int +application_add_segment_notify (u32 app_index, u32 fifo_segment_index) +{ + application_t *app = application_get (app_index); + u32 seg_size = 0; + u8 *seg_name; + + /* Send an API message to the external app, to map new segment */ + ASSERT (app->cb_fns.add_segment_callback); + + segment_manager_get_segment_info (fifo_segment_index, &seg_name, &seg_size); + return app->cb_fns.add_segment_callback (app->api_client_index, seg_name, + seg_size); +} + +u8 * +format_application_listener (u8 * s, va_list * args) +{ + application_t *app = va_arg (*args, application_t *); + u64 handle = va_arg (*args, u64); + u32 index = va_arg (*args, u32); + int verbose = va_arg (*args, int); + stream_session_t *listener; + u8 *app_name, *str; + + if (app == 0) + { + if (verbose) + s = format (s, "%-40s%-20s%-15s%-15s%-10s", "Connection", "App", + "API Client", "ListenerID", "SegManager"); + else + s = format (s, "%-40s%-20s", "Connection", "App"); + + return s; + } + + app_name = app_get_name_from_reg_index (app); + listener = listen_session_get_from_handle (handle); + str = format (0, "%U", format_stream_session, listener, verbose); + + if (verbose) + { + s = format (s, "%-40s%-20s%-15u%-15u%-10u", str, app_name, + app->api_client_index, handle, index); + } + else + s = format (s, "%-40s%-20s", str, app_name); + + vec_free (app_name); + return s; +} + +void +application_format_connects (application_t * app, int verbose) +{ + vlib_main_t *vm = vlib_get_main (); + segment_manager_t *sm; + u8 *app_name, *s = 0; + int j; + + /* Header */ + if (app == 0) + { + if (verbose) + vlib_cli_output (vm, "%-40s%-20s%-15s%-10s", "Connection", "App", + "API Client", "SegManager"); + else + vlib_cli_output (vm, "%-40s%-20s", "Connection", "App"); + return; + } + + /* make sure */ + if (app->connects_seg_manager == (u32) ~ 0) + return; + + app_name = app_get_name_from_reg_index (app); + + /* Across all fifo segments */ + sm = segment_manager_get (app->connects_seg_manager); + for (j = 0; j < vec_len (sm->segment_indices); j++) + { + svm_fifo_segment_private_t *fifo_segment; + svm_fifo_t *fifo; + u8 *str; + + fifo_segment = svm_fifo_segment_get_segment (sm->segment_indices[j]); + fifo = svm_fifo_segment_get_fifo_list (fifo_segment); + while (fifo) + { + u32 session_index, thread_index; + stream_session_t *session; + + session_index = fifo->master_session_index; + thread_index = fifo->master_thread_index; + + session = stream_session_get (session_index, thread_index); + str = format (0, "%U", format_stream_session, session, verbose); + + if (verbose) + s = format (s, "%-40s%-20s%-15u%-10u", str, app_name, + app->api_client_index, app->connects_seg_manager); + else + s = format (s, "%-40s%-20s", str, app_name); + + vlib_cli_output (vm, "%v", s); + vec_reset_length (s); + vec_free (str); + + fifo = fifo->next; + } + vec_free (s); + } + + vec_free (app_name); +} + +u8 * +format_application (u8 * s, va_list * args) +{ + application_t *app = va_arg (*args, application_t *); + CLIB_UNUSED (int verbose) = va_arg (*args, int); + u8 *app_name; + + if (app == 0) + { + if (verbose) + s = format (s, "%-10s%-20s%-15s%-15s%-15s%-15s", "Index", "Name", + "API Client", "Add seg size", "Rx fifo size", + "Tx fifo size"); + else + s = format (s, "%-10s%-20s%-20s", "Index", "Name", "API Client"); + return s; + } + + app_name = app_get_name_from_reg_index (app); + if (verbose) + s = format (s, "%-10d%-20s%-15d%-15d%-15d%-15d", app->index, app_name, + app->api_client_index, app->sm_properties.add_segment_size, + app->sm_properties.rx_fifo_size, + app->sm_properties.tx_fifo_size); + else + s = format (s, "%-10d%-20s%-20d", app->index, app_name, + app->api_client_index); + return s; +} + +static clib_error_t * +show_app_command_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + application_t *app; + int do_server = 0; + int do_client = 0; + int verbose = 0; + + if (!session_manager_is_enabled ()) + { + clib_error_return (0, "session layer is not enabled"); + } + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "server")) + do_server = 1; + else if (unformat (input, "client")) + do_client = 1; + else if (unformat (input, "verbose")) + verbose = 1; + else + break; + } + + if (do_server) + { + u64 handle; + u32 index; + if (pool_elts (app_pool)) + { + vlib_cli_output (vm, "%U", format_application_listener, + 0 /* header */ , 0, 0, + verbose); + /* *INDENT-OFF* */ + pool_foreach (app, app_pool, + ({ + /* App's listener sessions */ + if (hash_elts (app->listeners_table) == 0) + continue; + hash_foreach (handle, index, app->listeners_table, + ({ + vlib_cli_output (vm, "%U", format_application_listener, app, + handle, index, verbose); + })); + })); + /* *INDENT-ON* */ + } + else + vlib_cli_output (vm, "No active server bindings"); + } + + if (do_client) + { + if (pool_elts (app_pool)) + { + application_format_connects (0, verbose); + + /* *INDENT-OFF* */ + pool_foreach (app, app_pool, + ({ + if (app->connects_seg_manager == (u32)~0) + continue; + application_format_connects (app, verbose); + })); + /* *INDENT-ON* */ + } + else + vlib_cli_output (vm, "No active client bindings"); + } + + /* Print app related info */ + if (!do_server && !do_client) + { + vlib_cli_output (vm, "%U", format_application, 0, verbose); + pool_foreach (app, app_pool, ( + { + vlib_cli_output (vm, "%U", + format_application, app, + verbose); + } + )); + } + + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (show_app_command, static) = +{ + .path = "show app", + .short_help = "show app [server|client] [verbose]", + .function = show_app_command_fn, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/application.h b/src/vnet/session/application.h new file mode 100644 index 00000000..e030c376 --- /dev/null +++ b/src/vnet/session/application.h @@ -0,0 +1,136 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef SRC_VNET_SESSION_APPLICATION_H_ +#define SRC_VNET_SESSION_APPLICATION_H_ + +#include <vnet/vnet.h> +#include <vnet/session/session.h> +#include <vnet/session/segment_manager.h> + +typedef enum +{ + APP_SERVER, + APP_CLIENT, + APP_N_TYPES +} application_type_t; + +typedef struct _stream_session_cb_vft +{ + /** Notify server of new segment */ + int (*add_segment_callback) (u32 api_client_index, const u8 * seg_name, + u32 seg_size); + + /** Notify server of newly accepted session */ + int (*session_accept_callback) (stream_session_t * new_session); + + /* Connection request callback */ + int (*session_connected_callback) (u32 app_index, u32 opaque, + stream_session_t * s, u8 code); + + /** Notify app that session is closing */ + void (*session_disconnect_callback) (stream_session_t * s); + + /** Notify app that session was reset */ + void (*session_reset_callback) (stream_session_t * s); + + /* Direct RX callback, for built-in servers */ + int (*builtin_server_rx_callback) (stream_session_t * session); + + /* Redirect connection to local server */ + int (*redirect_connect_callback) (u32 api_client_index, void *mp); +} session_cb_vft_t; + +typedef struct _application +{ + /** Index in server pool */ + u32 index; + + /** Flags */ + u32 flags; + + /* + * Binary API interface to external app + */ + + /** Binary API connection index, ~0 if internal */ + u32 api_client_index; + + /** Application listens for events on this svm queue */ + unix_shared_memory_queue_t *event_queue; + + /* + * Callbacks: shoulder-taps for the server/client + */ + + session_cb_vft_t cb_fns; + + /* + * svm segment management + */ + u32 connects_seg_manager; + + /** Lookup tables for listeners. Value is segment manager index */ + uword *listeners_table; + + /** First segment manager has in the the first segment the application's + * event fifo. Depending on what the app does, it may be either used for + * a listener or for connects. */ + u32 first_segment_manager; + u8 first_segment_manager_in_use; + + /** Segment manager properties. Shared by all segment managers */ + segment_manager_properties_t sm_properties; +} application_t; + +#define APP_INVALID_SEGMENT_MANAGER_INDEX ((u32) ~0) + +application_t *application_new (); +int +application_init (application_t * app, u32 api_client_index, u64 * options, + session_cb_vft_t * cb_fns); +void application_del (application_t * app); +application_t *application_get (u32 index); +application_t *application_get_if_valid (u32 index); +application_t *application_lookup (u32 api_client_index); +u32 application_get_index (application_t * app); + +int +application_start_listen (application_t * app, session_type_t session_type, + transport_endpoint_t * tep, u64 * handle); +int application_stop_listen (application_t * srv, u64 handle); +int +application_open_session (application_t * app, session_type_t sst, + transport_endpoint_t * tep, u32 api_context); +int application_api_queue_is_full (application_t * app); + +segment_manager_t *application_get_listen_segment_manager (application_t * + app, + stream_session_t * + s); +segment_manager_t *application_get_connect_segment_manager (application_t * + app); +int application_is_proxy (application_t * app); +int application_add_segment_notify (u32 app_index, u32 fifo_segment_index); + +#endif /* SRC_VNET_SESSION_APPLICATION_H_ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/application_interface.c b/src/vnet/session/application_interface.c new file mode 100644 index 00000000..7e7449aa --- /dev/null +++ b/src/vnet/session/application_interface.c @@ -0,0 +1,406 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#include <vnet/session/application_interface.h> + +#include <vnet/session/session.h> +#include <vlibmemory/api.h> +#include <vnet/dpo/load_balance.h> +#include <vnet/fib/ip4_fib.h> + +/** @file + VPP's application/session API bind/unbind/connect/disconnect calls +*/ + +static u8 +ip_is_zero (ip46_address_t * ip46_address, u8 is_ip4) +{ + if (is_ip4) + return (ip46_address->ip4.as_u32 == 0); + else + return (ip46_address->as_u64[0] == 0 && ip46_address->as_u64[1] == 0); +} + +static u8 +ip_is_local (ip46_address_t * ip46_address, u8 is_ip4) +{ + fib_node_index_t fei; + fib_entry_flag_t flags; + fib_prefix_t prefix; + + /* Check if requester is local */ + if (is_ip4) + { + prefix.fp_len = 32; + prefix.fp_proto = FIB_PROTOCOL_IP4; + } + else + { + prefix.fp_len = 128; + prefix.fp_proto = FIB_PROTOCOL_IP6; + } + + clib_memcpy (&prefix.fp_addr, ip46_address, sizeof (ip46_address_t)); + fei = fib_table_lookup (0, &prefix); + flags = fib_entry_get_flags (fei); + + return (flags & FIB_ENTRY_FLAG_LOCAL); +} + +int +api_parse_session_handle (u64 handle, u32 * session_index, u32 * thread_index) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + stream_session_t *pool; + + *thread_index = handle & 0xFFFFFFFF; + *session_index = handle >> 32; + + if (*thread_index >= vec_len (smm->sessions)) + return VNET_API_ERROR_INVALID_VALUE; + + pool = smm->sessions[*thread_index]; + + if (pool_is_free_index (pool, *session_index)) + return VNET_API_ERROR_INVALID_VALUE_2; + + return 0; +} + +int +vnet_bind_i (u32 app_index, session_type_t sst, + transport_endpoint_t * tep, u64 * handle) +{ + application_t *app; + stream_session_t *listener; + + app = application_get_if_valid (app_index); + if (!app) + { + clib_warning ("app not attached"); + return VNET_API_ERROR_APPLICATION_NOT_ATTACHED; + } + + listener = stream_session_lookup_listener (&tep->ip, tep->port, sst); + if (listener) + return VNET_API_ERROR_ADDRESS_IN_USE; + + if (!ip_is_zero (&tep->ip, tep->is_ip4) + && !ip_is_local (&tep->ip, tep->is_ip4)) + return VNET_API_ERROR_INVALID_VALUE_2; + + /* Setup listen path down to transport */ + return application_start_listen (app, sst, tep, handle); +} + +int +vnet_unbind_i (u32 app_index, u64 handle) +{ + application_t *app = application_get_if_valid (app_index); + + if (!app) + { + clib_warning ("app (%d) not attached", app_index); + return VNET_API_ERROR_APPLICATION_NOT_ATTACHED; + } + + /* Clear the listener */ + return application_stop_listen (app, handle); +} + +int +vnet_connect_i (u32 app_index, u32 api_context, session_type_t sst, + transport_endpoint_t * tep, void *mp) +{ + stream_session_t *listener; + application_t *server, *app; + + /* + * Figure out if connecting to a local server + */ + listener = stream_session_lookup_listener (&tep->ip, tep->port, sst); + if (listener) + { + server = application_get (listener->app_index); + + /* + * Server is willing to have a direct fifo connection created + * instead of going through the state machine, etc. + */ + if (server->flags & APP_OPTIONS_FLAGS_USE_FIFO) + return server->cb_fns. + redirect_connect_callback (server->api_client_index, mp); + } + + /* + * Not connecting to a local server. Create regular session + */ + app = application_get (app_index); + return application_open_session (app, sst, tep, api_context); +} + +/** + * unformat a vnet URI + * + * fifo://name + * tcp://ip46-addr:port + * udp://ip46-addr:port + * + * u8 ip46_address[16]; + * u16 port_in_host_byte_order; + * stream_session_type_t sst; + * u8 *fifo_name; + * + * if (unformat (input, "%U", unformat_vnet_uri, &ip46_address, + * &sst, &port, &fifo_name)) + * etc... + * + */ +uword +unformat_vnet_uri (unformat_input_t * input, va_list * args) +{ + session_type_t *sst = va_arg (*args, session_type_t *); + transport_endpoint_t *tep = va_arg (*args, transport_endpoint_t *); + + if (unformat (input, "tcp://%U/%d", unformat_ip4_address, &tep->ip.ip4, + &tep->port)) + { + *sst = SESSION_TYPE_IP4_TCP; + tep->port = clib_host_to_net_u16 (tep->port); + tep->is_ip4 = 1; + return 1; + } + if (unformat (input, "udp://%U/%d", unformat_ip4_address, &tep->ip.ip4, + &tep->port)) + { + *sst = SESSION_TYPE_IP4_UDP; + tep->port = clib_host_to_net_u16 (tep->port); + tep->is_ip4 = 1; + return 1; + } + if (unformat (input, "udp://%U/%d", unformat_ip6_address, &tep->ip.ip6, + &tep->port)) + { + *sst = SESSION_TYPE_IP6_UDP; + tep->port = clib_host_to_net_u16 (tep->port); + return 1; + } + if (unformat (input, "tcp://%U/%d", unformat_ip6_address, &tep->ip.ip6, + &tep->port)) + { + *sst = SESSION_TYPE_IP6_TCP; + tep->port = clib_host_to_net_u16 (tep->port); + return 1; + } + + return 0; +} + +static u8 *cache_uri; +static session_type_t cache_sst; +static transport_endpoint_t *cache_tep; + +int +parse_uri (char *uri, session_type_t * sst, transport_endpoint_t * tep) +{ + unformat_input_t _input, *input = &_input; + + if (cache_uri && !strncmp (uri, (char *) cache_uri, vec_len (cache_uri))) + { + *sst = cache_sst; + *tep = *cache_tep; + return 0; + } + + /* Make sure */ + uri = (char *) format (0, "%s%c", uri, 0); + + /* Parse uri */ + unformat_init_string (input, uri, strlen (uri)); + if (!unformat (input, "%U", unformat_vnet_uri, sst, tep)) + { + unformat_free (input); + return VNET_API_ERROR_INVALID_VALUE; + } + unformat_free (input); + + vec_free (cache_uri); + cache_uri = (u8 *) uri; + cache_sst = *sst; + if (cache_tep) + clib_mem_free (cache_tep); + cache_tep = clib_mem_alloc (sizeof (*tep)); + *cache_tep = *tep; + + return 0; +} + +/** + * Attaches application. + * + * Allocates a vpp app, i.e., a structure that keeps back pointers + * to external app and a segment manager for shared memory fifo based + * communication with the external app. + */ +int +vnet_application_attach (vnet_app_attach_args_t * a) +{ + application_t *app = 0; + segment_manager_t *sm; + u8 *seg_name; + int rv; + + app = application_new (); + if ((rv = application_init (app, a->api_client_index, a->options, + a->session_cb_vft))) + return rv; + + a->app_event_queue_address = pointer_to_uword (app->event_queue); + sm = segment_manager_get (app->first_segment_manager); + segment_manager_get_segment_info (sm->segment_indices[0], + &seg_name, &a->segment_size); + + a->segment_name_length = vec_len (seg_name); + a->segment_name = seg_name; + ASSERT (vec_len (a->segment_name) <= 128); + a->app_index = app->index; + return 0; +} + +int +vnet_application_detach (vnet_app_detach_args_t * a) +{ + application_t *app; + app = application_get_if_valid (a->app_index); + + if (!app) + { + clib_warning ("app not attached"); + return VNET_API_ERROR_APPLICATION_NOT_ATTACHED; + } + + application_del (app); + return 0; +} + +int +vnet_bind_uri (vnet_bind_args_t * a) +{ + session_type_t sst = SESSION_N_TYPES; + transport_endpoint_t tep; + int rv; + + memset (&tep, 0, sizeof (tep)); + rv = parse_uri (a->uri, &sst, &tep); + if (rv) + return rv; + + if ((rv = vnet_bind_i (a->app_index, sst, &tep, &a->handle))) + return rv; + + return 0; +} + +int +vnet_unbind_uri (vnet_unbind_args_t * a) +{ + session_type_t sst = SESSION_N_TYPES; + stream_session_t *listener; + transport_endpoint_t tep; + int rv; + + rv = parse_uri (a->uri, &sst, &tep); + if (rv) + return rv; + + listener = stream_session_lookup_listener (&tep.ip, + clib_host_to_net_u16 (tep.port), + sst); + if (!listener) + return VNET_API_ERROR_ADDRESS_NOT_IN_USE; + + return vnet_unbind_i (a->app_index, listen_session_get_handle (listener)); +} + +int +vnet_connect_uri (vnet_connect_args_t * a) +{ + transport_endpoint_t tep; + session_type_t sst; + int rv; + + /* Parse uri */ + memset (&tep, 0, sizeof (tep)); + rv = parse_uri (a->uri, &sst, &tep); + if (rv) + return rv; + + return vnet_connect_i (a->app_index, a->api_context, sst, &tep, a->mp); +} + +int +vnet_disconnect_session (vnet_disconnect_args_t * a) +{ + u32 index, thread_index; + stream_session_t *s; + + stream_session_parse_handle (a->handle, &index, &thread_index); + s = stream_session_get_if_valid (index, thread_index); + + if (!s || s->app_index != a->app_index) + return VNET_API_ERROR_INVALID_VALUE; + + /* We're peeking into another's thread pool. Make sure */ + ASSERT (s->session_index == index); + + session_send_session_evt_to_thread (a->handle, FIFO_EVENT_DISCONNECT, + thread_index); + return 0; +} + +int +vnet_bind (vnet_bind_args_t * a) +{ + session_type_t sst = SESSION_N_TYPES; + int rv; + + sst = session_type_from_proto_and_ip (a->proto, a->tep.is_ip4); + if ((rv = vnet_bind_i (a->app_index, sst, &a->tep, &a->handle))) + return rv; + + return 0; +} + +int +vnet_unbind (vnet_unbind_args_t * a) +{ + return vnet_unbind_i (a->app_index, a->handle); +} + +int +vnet_connect (vnet_connect_args_t * a) +{ + session_type_t sst; + + sst = session_type_from_proto_and_ip (a->proto, a->tep.is_ip4); + return vnet_connect_i (a->app_index, a->api_context, sst, &a->tep, a->mp); +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/application_interface.h b/src/vnet/session/application_interface.h new file mode 100644 index 00000000..1d63f6cc --- /dev/null +++ b/src/vnet/session/application_interface.h @@ -0,0 +1,183 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef __included_uri_h__ +#define __included_uri_h__ + +#include <vlib/vlib.h> +#include <vnet/vnet.h> +#include <svm/svm_fifo_segment.h> +#include <vnet/session/session.h> +#include <vnet/session/application.h> +#include <vnet/session/transport.h> + +typedef struct _vnet_app_attach_args_t +{ + /** Binary API client index */ + u32 api_client_index; + + /** Application and segment manager options */ + u64 *options; + + /** Session to application callback functions */ + session_cb_vft_t *session_cb_vft; + + /** Flag that indicates if app is builtin */ + u8 builtin; + + /* + * Results + */ + u8 *segment_name; + u32 segment_name_length; + u32 segment_size; + u64 app_event_queue_address; + u32 app_index; +} vnet_app_attach_args_t; + +typedef struct _vnet_app_detach_args_t +{ + u32 app_index; +} vnet_app_detach_args_t; + +typedef struct _vnet_bind_args_t +{ + union + { + char *uri; + struct + { + transport_endpoint_t tep; + transport_proto_t proto; + }; + }; + + u32 app_index; + + /* + * Results + */ + char *segment_name; + u32 segment_name_length; + u64 server_event_queue_address; + u64 handle; +} vnet_bind_args_t; + +typedef struct _vnet_unbind_args_t +{ + union + { + char *uri; + u64 handle; + }; + u32 app_index; +} vnet_unbind_args_t; + +typedef struct _vnet_connect_args +{ + union + { + char *uri; + struct + { + transport_endpoint_t tep; + transport_proto_t proto; + }; + }; + u32 app_index; + u32 api_context; + + /* Used for redirects */ + void *mp; + + /* used for proxy connections */ + u64 server_handle; +} vnet_connect_args_t; + +typedef struct _vnet_disconnect_args_t +{ + u64 handle; + u32 app_index; +} vnet_disconnect_args_t; + +/* Application attach options */ +typedef enum +{ + APP_EVT_QUEUE_SIZE, + APP_OPTIONS_FLAGS, + APP_OPTIONS_PREALLOC_FIFO_PAIRS, + APP_OPTIONS_PRIVATE_SEGMENT_COUNT, + APP_OPTIONS_PRIVATE_SEGMENT_SIZE, + SESSION_OPTIONS_SEGMENT_SIZE, + SESSION_OPTIONS_ADD_SEGMENT_SIZE, + SESSION_OPTIONS_RX_FIFO_SIZE, + SESSION_OPTIONS_TX_FIFO_SIZE, + SESSION_OPTIONS_PREALLOCATED_FIFO_PAIRS, + SESSION_OPTIONS_ACCEPT_COOKIE, + SESSION_OPTIONS_N_OPTIONS +} app_attach_options_index_t; + +#define foreach_app_options_flags \ + _(USE_FIFO, "Use FIFO with redirects") \ + _(ADD_SEGMENT, "Add segment and signal app if needed") \ + _(BUILTIN_APP, "Application is builtin") \ + _(IS_PROXY, "Application is proxying") + +typedef enum _app_options +{ +#define _(sym, str) APP_OPTIONS_##sym, + foreach_app_options_flags +#undef _ +} app_options_t; + +typedef enum _app_options_flags +{ +#define _(sym, str) APP_OPTIONS_FLAGS_##sym = 1 << APP_OPTIONS_##sym, + foreach_app_options_flags +#undef _ +} app_options_flags_t; + +///** Server can handle delegated connect requests from local clients */ +//#define APP_OPTIONS_FLAGS_USE_FIFO (1<<0) +// +///** Server wants vpp to add segments when out of memory for fifos */ +//#define APP_OPTIONS_FLAGS_ADD_SEGMENT (1<<1) + +#define VNET_CONNECT_REDIRECTED 123 + +int vnet_application_attach (vnet_app_attach_args_t * a); +int vnet_application_detach (vnet_app_detach_args_t * a); + +int vnet_bind_uri (vnet_bind_args_t *); +int vnet_unbind_uri (vnet_unbind_args_t * a); +int vnet_connect_uri (vnet_connect_args_t * a); +int vnet_disconnect_session (vnet_disconnect_args_t * a); + +int vnet_bind (vnet_bind_args_t * a); +int vnet_connect (vnet_connect_args_t * a); +int vnet_unbind (vnet_unbind_args_t * a); + +int +api_parse_session_handle (u64 handle, u32 * session_index, + u32 * thread_index); + +#endif /* __included_uri_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/segment_manager.c b/src/vnet/session/segment_manager.c new file mode 100644 index 00000000..48d02755 --- /dev/null +++ b/src/vnet/session/segment_manager.c @@ -0,0 +1,636 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/session/segment_manager.h> +#include <vnet/session/session.h> +#include <vnet/session/application.h> + +/** + * Counter used to build segment names + */ +u32 segment_name_counter = 0; + +/** + * Pool of segment managers + */ +segment_manager_t *segment_managers = 0; + +/** + * Process private segment index + */ +u32 *private_segment_indices; + +/** + * Default fifo and segment size. TODO config. + */ +u32 default_fifo_size = 1 << 16; +u32 default_segment_size = 1 << 20; + +void +segment_manager_get_segment_info (u32 index, u8 ** name, u32 * size) +{ + svm_fifo_segment_private_t *s; + s = svm_fifo_segment_get_segment (index); + *name = s->h->segment_name; + *size = s->ssvm.ssvm_size; +} + +always_inline int +session_manager_add_segment_i (segment_manager_t * sm, u32 segment_size, + u8 * segment_name) +{ + svm_fifo_segment_create_args_t _ca, *ca = &_ca; + int rv; + + memset (ca, 0, sizeof (*ca)); + + if (!sm->properties->use_private_segment) + { + ca->segment_name = (char *) segment_name; + ca->segment_size = segment_size; + ca->rx_fifo_size = sm->properties->rx_fifo_size; + ca->tx_fifo_size = sm->properties->tx_fifo_size; + ca->preallocated_fifo_pairs = sm->properties->preallocated_fifo_pairs; + + rv = svm_fifo_segment_create (ca); + if (rv) + { + clib_warning ("svm_fifo_segment_create ('%s', %d) failed", + ca->segment_name, ca->segment_size); + return VNET_API_ERROR_SVM_SEGMENT_CREATE_FAIL; + } + } + else + { + u32 rx_fifo_size, tx_fifo_size, rx_rounded_data_size, + tx_rounded_data_size; + u32 approx_segment_count; + u64 approx_total_size; + + ca->segment_name = "process-private-segment"; + ca->segment_size = ~0; + ca->rx_fifo_size = sm->properties->rx_fifo_size; + ca->tx_fifo_size = sm->properties->tx_fifo_size; + ca->preallocated_fifo_pairs = sm->properties->preallocated_fifo_pairs; + ca->private_segment_count = sm->properties->private_segment_count; + ca->private_segment_size = sm->properties->private_segment_size; + + /* Default to a small private segment */ + if (ca->private_segment_size == 0) + ca->private_segment_size = 128 << 20; + + /* Calculate space requirements */ + rx_rounded_data_size = (1 << (max_log2 (ca->rx_fifo_size))); + tx_rounded_data_size = (1 << (max_log2 (ca->tx_fifo_size))); + + rx_fifo_size = sizeof (svm_fifo_t) + rx_rounded_data_size; + tx_fifo_size = sizeof (svm_fifo_t) + tx_rounded_data_size; + + approx_total_size = (u64) ca->preallocated_fifo_pairs + * (rx_fifo_size + tx_fifo_size); + approx_segment_count = + (approx_total_size + + (ca->private_segment_size - 1)) / (u64) ca->private_segment_size; + + /* The user asked us to figure it out... */ + if (ca->private_segment_count == 0) + { + ca->private_segment_count = approx_segment_count; + } + /* Follow directions, but issue a warning */ + else if (approx_segment_count != ca->private_segment_count) + { + clib_warning + ("Honoring segment count %u, but calculated count was %u", + ca->private_segment_count, approx_segment_count); + } + + if (svm_fifo_segment_create_process_private (ca)) + clib_warning ("Failed to create process private segment"); + + ASSERT (vec_len (ca->new_segment_indices)); + } + vec_append (sm->segment_indices, ca->new_segment_indices); + vec_free (ca->new_segment_indices); + return 0; +} + +int +session_manager_add_segment (segment_manager_t * sm) +{ + u8 *segment_name; + svm_fifo_segment_create_args_t _ca, *ca = &_ca; + u32 add_segment_size; + int rv; + + memset (ca, 0, sizeof (*ca)); + segment_name = format (0, "%d-%d%c", getpid (), segment_name_counter++, 0); + add_segment_size = sm->properties->add_segment_size ? + sm->properties->add_segment_size : default_segment_size; + + rv = session_manager_add_segment_i (sm, add_segment_size, segment_name); + vec_free (segment_name); + return rv; +} + +int +session_manager_add_first_segment (segment_manager_t * sm, u32 segment_size) +{ + u8 *segment_name; + int rv; + + segment_name = format (0, "%d-%d%c", getpid (), segment_name_counter++, 0); + rv = session_manager_add_segment_i (sm, segment_size, segment_name); + vec_free (segment_name); + return rv; +} + +segment_manager_t * +segment_manager_new () +{ + segment_manager_t *sm; + pool_get (segment_managers, sm); + memset (sm, 0, sizeof (*sm)); + return sm; +} + +/** + * Initializes segment manager based on options provided. + * Returns error if svm segment allocation fails. + */ +int +segment_manager_init (segment_manager_t * sm, + segment_manager_properties_t * properties, + u32 first_seg_size) +{ + int rv; + + /* app allocates these */ + sm->properties = properties; + + first_seg_size = first_seg_size > 0 ? first_seg_size : default_segment_size; + + rv = session_manager_add_first_segment (sm, first_seg_size); + if (rv) + { + clib_warning ("Failed to allocate segment"); + return rv; + } + + clib_spinlock_init (&sm->lockp); + return 0; +} + +u8 +segment_manager_has_fifos (segment_manager_t * sm) +{ + svm_fifo_segment_private_t *segment; + int i; + + for (i = 0; i < vec_len (sm->segment_indices); i++) + { + segment = svm_fifo_segment_get_segment (sm->segment_indices[i]); + if (CLIB_DEBUG && i && !svm_fifo_segment_has_fifos (segment) + && !(segment->h->flags & FIFO_SEGMENT_F_IS_PREALLOCATED)) + clib_warning ("segment %d has no fifos!", sm->segment_indices[i]); + if (svm_fifo_segment_has_fifos (segment)) + return 1; + } + return 0; +} + +static u8 +segment_manager_app_detached (segment_manager_t * sm) +{ + return (sm->app_index == SEGMENT_MANAGER_INVALID_APP_INDEX); +} + +static void +segment_manager_del_segment (segment_manager_t * sm, u32 segment_index) +{ + svm_fifo_segment_private_t *fifo_segment; + u32 svm_segment_index; + clib_spinlock_lock (&sm->lockp); + svm_segment_index = sm->segment_indices[segment_index]; + fifo_segment = svm_fifo_segment_get_segment (svm_segment_index); + if (!fifo_segment + || ((fifo_segment->h->flags & FIFO_SEGMENT_F_IS_PREALLOCATED) + && !segment_manager_app_detached (sm))) + { + clib_spinlock_unlock (&sm->lockp); + return; + } + svm_fifo_segment_delete (fifo_segment); + vec_del1 (sm->segment_indices, segment_index); + clib_spinlock_unlock (&sm->lockp); +} + +/** + * Initiate disconnects for all sessions 'owned' by a segment manager + */ +void +segment_manager_del_sessions (segment_manager_t * sm) +{ + int j; + svm_fifo_segment_private_t *fifo_segment; + svm_fifo_t *fifo; + + ASSERT (vec_len (sm->segment_indices)); + + /* Across all fifo segments used by the server */ + for (j = 0; j < vec_len (sm->segment_indices); j++) + { + fifo_segment = svm_fifo_segment_get_segment (sm->segment_indices[j]); + fifo = svm_fifo_segment_get_fifo_list (fifo_segment); + + /* + * Remove any residual sessions from the session lookup table + * Don't bother deleting the individual fifos, we're going to + * throw away the fifo segment in a minute. + */ + while (fifo) + { + u32 session_index, thread_index; + stream_session_t *session; + + session_index = fifo->master_session_index; + thread_index = fifo->master_thread_index; + session = stream_session_get (session_index, thread_index); + + /* Instead of directly removing the session call disconnect */ + if (session->session_state != SESSION_STATE_CLOSED) + { + session->session_state = SESSION_STATE_CLOSED; + session_send_session_evt_to_thread (stream_session_handle + (session), + FIFO_EVENT_DISCONNECT, + thread_index); + } + fifo = fifo->next; + } + + /* Instead of removing the segment, test when cleaning up disconnected + * sessions if the segment can be removed. + */ + } +} + +/** + * Removes segment manager. + * + * Since the fifos allocated in the segment keep backpointers to the sessions + * prior to removing the segment, we call session disconnect. This + * subsequently propagates into transport. + */ +void +segment_manager_del (segment_manager_t * sm) +{ + int i; + + ASSERT (!segment_manager_has_fifos (sm) + && segment_manager_app_detached (sm)); + + /* If we have empty preallocated segments that haven't been removed, remove + * them now. Apart from that, the first segment in the first segment manager + * is not removed when all fifos are removed. It can only be removed when + * the manager is explicitly deleted/detached by the app. */ + for (i = vec_len (sm->segment_indices) - 1; i >= 0; i--) + { + if (CLIB_DEBUG) + { + svm_fifo_segment_private_t *segment; + segment = svm_fifo_segment_get_segment (sm->segment_indices[i]); + ASSERT (!svm_fifo_segment_has_fifos (segment)); + } + segment_manager_del_segment (sm, i); + } + clib_spinlock_free (&sm->lockp); + if (CLIB_DEBUG) + memset (sm, 0xfe, sizeof (*sm)); + pool_put (segment_managers, sm); +} + +void +segment_manager_init_del (segment_manager_t * sm) +{ + if (segment_manager_has_fifos (sm)) + segment_manager_del_sessions (sm); + else + { + ASSERT (!sm->first_is_protected || segment_manager_app_detached (sm)); + segment_manager_del (sm); + } +} + +int +segment_manager_alloc_session_fifos (segment_manager_t * sm, + svm_fifo_t ** server_rx_fifo, + svm_fifo_t ** server_tx_fifo, + u32 * fifo_segment_index) +{ + svm_fifo_segment_private_t *fifo_segment; + u32 fifo_size, sm_index; + u8 added_a_segment = 0; + int i; + + ASSERT (vec_len (sm->segment_indices)); + + /* Make sure we don't have multiple threads trying to allocate segments + * at the same time. */ + clib_spinlock_lock (&sm->lockp); + + /* Allocate svm fifos */ +again: + for (i = 0; i < vec_len (sm->segment_indices); i++) + { + *fifo_segment_index = sm->segment_indices[i]; + fifo_segment = svm_fifo_segment_get_segment (*fifo_segment_index); + + fifo_size = sm->properties->rx_fifo_size; + fifo_size = (fifo_size == 0) ? default_fifo_size : fifo_size; + *server_rx_fifo = + svm_fifo_segment_alloc_fifo (fifo_segment, fifo_size, + FIFO_SEGMENT_RX_FREELIST); + + fifo_size = sm->properties->tx_fifo_size; + fifo_size = (fifo_size == 0) ? default_fifo_size : fifo_size; + *server_tx_fifo = + svm_fifo_segment_alloc_fifo (fifo_segment, fifo_size, + FIFO_SEGMENT_TX_FREELIST); + + if (*server_rx_fifo == 0) + { + /* This would be very odd, but handle it... */ + if (*server_tx_fifo != 0) + { + svm_fifo_segment_free_fifo (fifo_segment, *server_tx_fifo, + FIFO_SEGMENT_TX_FREELIST); + *server_tx_fifo = 0; + } + continue; + } + if (*server_tx_fifo == 0) + { + if (*server_rx_fifo != 0) + { + svm_fifo_segment_free_fifo (fifo_segment, *server_rx_fifo, + FIFO_SEGMENT_RX_FREELIST); + *server_rx_fifo = 0; + } + continue; + } + break; + } + + /* See if we're supposed to create another segment */ + if (*server_rx_fifo == 0) + { + if (sm->properties->add_segment && !sm->properties->use_private_segment) + { + if (added_a_segment) + { + clib_warning ("added a segment, still can't allocate a fifo"); + clib_spinlock_unlock (&sm->lockp); + return SESSION_ERROR_NEW_SEG_NO_SPACE; + } + + if (session_manager_add_segment (sm)) + { + clib_spinlock_unlock (&sm->lockp); + return VNET_API_ERROR_URI_FIFO_CREATE_FAILED; + } + + added_a_segment = 1; + goto again; + } + else + { + clib_warning ("No space to allocate fifos!"); + clib_spinlock_unlock (&sm->lockp); + return SESSION_ERROR_NO_SPACE; + } + } + + /* Backpointers to segment manager */ + sm_index = segment_manager_index (sm); + (*server_tx_fifo)->segment_manager = sm_index; + (*server_rx_fifo)->segment_manager = sm_index; + + clib_spinlock_unlock (&sm->lockp); + + if (added_a_segment) + return application_add_segment_notify (sm->app_index, + *fifo_segment_index); + + return 0; +} + +void +segment_manager_dealloc_fifos (u32 svm_segment_index, svm_fifo_t * rx_fifo, + svm_fifo_t * tx_fifo) +{ + segment_manager_t *sm; + svm_fifo_segment_private_t *fifo_segment; + u32 i, segment_index = ~0; + u8 is_first; + + sm = segment_manager_get_if_valid (rx_fifo->segment_manager); + + /* It's possible to have no segment manager if the session was removed + * as result of a detach. */ + if (!sm) + return; + + fifo_segment = svm_fifo_segment_get_segment (svm_segment_index); + svm_fifo_segment_free_fifo (fifo_segment, rx_fifo, + FIFO_SEGMENT_RX_FREELIST); + svm_fifo_segment_free_fifo (fifo_segment, tx_fifo, + FIFO_SEGMENT_TX_FREELIST); + + /* + * Try to remove svm segment if it has no fifos. This can be done only if + * the segment is not the first in the segment manager or if it is first + * and it is not protected. Moreover, if the segment is first and the app + * has detached from the segment manager, remove the segment manager. + */ + if (!svm_fifo_segment_has_fifos (fifo_segment)) + { + is_first = sm->segment_indices[0] == svm_segment_index; + + /* Remove segment if it holds no fifos or first but not protected */ + if (!is_first || !sm->first_is_protected) + { + /* Find the segment manager segment index */ + for (i = 0; i < vec_len (sm->segment_indices); i++) + if (sm->segment_indices[i] == svm_segment_index) + { + segment_index = i; + break; + } + ASSERT (segment_index != (u32) ~ 0); + segment_manager_del_segment (sm, segment_index); + } + + /* Remove segment manager if no sessions and detached from app */ + if (segment_manager_app_detached (sm) + && !segment_manager_has_fifos (sm)) + segment_manager_del (sm); + } +} + +/** + * Allocates shm queue in the first segment + */ +unix_shared_memory_queue_t * +segment_manager_alloc_queue (segment_manager_t * sm, u32 queue_size) +{ + ssvm_shared_header_t *sh; + svm_fifo_segment_private_t *segment; + unix_shared_memory_queue_t *q; + void *oldheap; + + ASSERT (sm->segment_indices != 0); + + segment = svm_fifo_segment_get_segment (sm->segment_indices[0]); + sh = segment->ssvm.sh; + + oldheap = ssvm_push_heap (sh); + q = unix_shared_memory_queue_init (queue_size, + sizeof (session_fifo_event_t), + 0 /* consumer pid */ , + 0 /* signal when queue non-empty */ ); + ssvm_pop_heap (oldheap); + return q; +} + +/** + * Frees shm queue allocated in the first segment + */ +void +segment_manager_dealloc_queue (segment_manager_t * sm, + unix_shared_memory_queue_t * q) +{ + ssvm_shared_header_t *sh; + svm_fifo_segment_private_t *segment; + void *oldheap; + + ASSERT (sm->segment_indices != 0); + + segment = svm_fifo_segment_get_segment (sm->segment_indices[0]); + sh = segment->ssvm.sh; + + oldheap = ssvm_push_heap (sh); + unix_shared_memory_queue_free (q); + ssvm_pop_heap (oldheap); +} + +static clib_error_t * +segment_manager_show_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + svm_fifo_segment_private_t *segments, *seg; + segment_manager_t *sm; + u8 show_segments = 0, verbose = 0, *name; + uword address; + u64 size; + u32 active_fifos; + u32 free_fifos; + + mheap_t *heap_header; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "segments")) + show_segments = 1; + else if (unformat (input, "verbose")) + verbose = 1; + else + return clib_error_return (0, "unknown input `%U'", + format_unformat_error, input); + } + vlib_cli_output (vm, "%d segment managers allocated", + pool_elts (segment_managers)); + if (verbose && pool_elts (segment_managers)) + { + vlib_cli_output (vm, "%-10s%=15s%=12s", "Index", "App Index", + "Segments"); + + /* *INDENT-OFF* */ + pool_foreach (sm, segment_managers, ({ + vlib_cli_output (vm, "%-10d%=15d%=12d", segment_manager_index(sm), + sm->app_index, vec_len (sm->segment_indices)); + })); + /* *INDENT-ON* */ + + } + if (show_segments) + { + segments = svm_fifo_segment_segments_pool (); + vlib_cli_output (vm, "%d svm fifo segments allocated", + pool_elts (segments)); + vlib_cli_output (vm, "%-20s%=12s%=16s%=16s%=16s", "Name", + "HeapSize (M)", "ActiveFifos", "FreeFifos", "Address"); + + /* *INDENT-OFF* */ + pool_foreach (seg, segments, ({ + if (seg->h->flags & FIFO_SEGMENT_F_IS_PRIVATE) + { + address = pointer_to_uword (seg->ssvm.sh->heap); + if (seg->h->flags & FIFO_SEGMENT_F_IS_MAIN_HEAP) + name = format (0, "main heap"); + else + name = format (0, "private heap"); + heap_header = mheap_header (seg->ssvm.sh->heap); + size = heap_header->max_size; + } + else + { + address = seg->ssvm.sh->ssvm_va; + size = seg->ssvm.ssvm_size; + name = seg->ssvm.sh->name; + } + active_fifos = svm_fifo_segment_num_fifos (seg); + free_fifos = svm_fifo_segment_num_free_fifos (seg, ~0 /* size */); + vlib_cli_output (vm, "%-20v%=16llu%=16u%=16u%16llx", + name, size >> 20ULL, active_fifos, free_fifos, + address); + if (verbose) + vlib_cli_output (vm, "%U", + format_svm_fifo_segment, seg, verbose); + if (seg->h->flags & FIFO_SEGMENT_F_IS_PRIVATE) + vec_free (name); + })); + /* *INDENT-ON* */ + + } + return 0; +} + + /* *INDENT-OFF* */ +VLIB_CLI_COMMAND (segment_manager_show_command, static) = +{ + .path = "show segment-manager", + .short_help = "show segment-manager [segments][verbose]", + .function = segment_manager_show_fn, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/segment_manager.h b/src/vnet/session/segment_manager.h new file mode 100644 index 00000000..6e5b8989 --- /dev/null +++ b/src/vnet/session/segment_manager.h @@ -0,0 +1,131 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef SRC_VNET_SESSION_SEGMENT_MANAGER_H_ +#define SRC_VNET_SESSION_SEGMENT_MANAGER_H_ + +#include <vnet/vnet.h> +#include <svm/svm_fifo_segment.h> + +#include <vlibmemory/unix_shared_memory_queue.h> +#include <vlibmemory/api.h> +#include <vppinfra/lock.h> + +typedef struct _segment_manager_properties +{ + /** Session fifo sizes. */ + u32 rx_fifo_size; + u32 tx_fifo_size; + + /** Preallocated pool sizes */ + u32 preallocated_fifo_pairs; + + /** Configured additional segment size */ + u32 add_segment_size; + + /** Flag that indicates if additional segments should be created */ + u8 add_segment; + + /** Use private memory segment instead of shared memory */ + u8 use_private_segment; + + /** Use one or more private mheaps, instead of the global heap */ + u32 private_segment_count; + u32 private_segment_size; +} segment_manager_properties_t; + +typedef struct _segment_manager +{ + clib_spinlock_t lockp; + + /** segments mapped by this manager */ + u32 *segment_indices; + + /** Owner app index */ + u32 app_index; + + /** + * Pointer to manager properties. Could be shared among all of + * an app's segment managers s + */ + segment_manager_properties_t *properties; + + /** + * First segment should not be deleted unless segment manger is deleted. + * This also indicates that the segment manager is the first to have been + * allocated for the app. + */ + u8 first_is_protected; +} segment_manager_t; + +#define SEGMENT_MANAGER_INVALID_APP_INDEX ((u32) ~0) + +/** Pool of segment managers */ +extern segment_manager_t *segment_managers; + +always_inline segment_manager_t * +segment_manager_get (u32 index) +{ + return pool_elt_at_index (segment_managers, index); +} + +always_inline segment_manager_t * +segment_manager_get_if_valid (u32 index) +{ + if (pool_is_free_index (segment_managers, index)) + return 0; + return pool_elt_at_index (segment_managers, index); +} + +always_inline u32 +segment_manager_index (segment_manager_t * sm) +{ + return sm - segment_managers; +} + +segment_manager_t *segment_manager_new (); +int +segment_manager_init (segment_manager_t * sm, + segment_manager_properties_t * properties, + u32 seg_size); + +void segment_manager_get_segment_info (u32 index, u8 ** name, u32 * size); +int +session_manager_add_first_segment (segment_manager_t * sm, u32 segment_size); +int session_manager_add_segment (segment_manager_t * sm); +void segment_manager_del_sessions (segment_manager_t * sm); +void segment_manager_del (segment_manager_t * sm); +void segment_manager_init_del (segment_manager_t * sm); +u8 segment_manager_has_fifos (segment_manager_t * sm); +int +segment_manager_alloc_session_fifos (segment_manager_t * sm, + svm_fifo_t ** server_rx_fifo, + svm_fifo_t ** server_tx_fifo, + u32 * fifo_segment_index); +void +segment_manager_dealloc_fifos (u32 svm_segment_index, svm_fifo_t * rx_fifo, + svm_fifo_t * tx_fifo); +unix_shared_memory_queue_t *segment_manager_alloc_queue (segment_manager_t * + sm, u32 queue_size); +void segment_manager_dealloc_queue (segment_manager_t * sm, + unix_shared_memory_queue_t * q); + +#endif /* SRC_VNET_SESSION_SEGMENT_MANAGER_H_ */ +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session.api b/src/vnet/session/session.api new file mode 100644 index 00000000..30d2ae96 --- /dev/null +++ b/src/vnet/session/session.api @@ -0,0 +1,331 @@ +/* + * Copyright (c) 2015-2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/** \brief client->vpp, attach application to session layer + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param initial_segment_size - size of the initial shm segment to be + allocated + @param options - segment size, fifo sizes, etc. +*/ + define application_attach { + u32 client_index; + u32 context; + u32 initial_segment_size; + u64 options[16]; + }; + + /** \brief Application attach reply + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param app_event_queue_address - vpp event queue address or 0 if this + connection shouldn't send events + @param segment_size - size of first shm segment + @param segment_name_length - length of segment name + @param segment_name - name of segment client needs to attach to +*/ +define application_attach_reply { + u32 context; + i32 retval; + u64 app_event_queue_address; + u32 segment_size; + u8 segment_name_length; + u8 segment_name[128]; +}; + + /** \brief client->vpp, attach application to session layer + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request +*/ +autoreply define application_detach { + u32 client_index; + u32 context; + }; + +/** \brief vpp->client, please map an additional shared memory segment + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param segment_name - +*/ +autoreply define map_another_segment { + u32 client_index; + u32 context; + u32 segment_size; + u8 segment_name[128]; +}; + + /** \brief Bind to a given URI + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param accept_cookie - sender accept cookie, to identify this bind flavor + @param uri - a URI, e.g. "tcp://0.0.0.0/0/80" [ipv4] + "tcp://::/0/80" [ipv6] etc. + @param options - socket options, fifo sizes, etc. +*/ +autoreply define bind_uri { + u32 client_index; + u32 context; + u32 accept_cookie; + u8 uri[128]; +}; + +/** \brief Unbind a given URI + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param uri - a URI, e.g. "tcp://0.0.0.0/0/80" [ipv4] + "tcp://::/0/80" [ipv6], etc. + @param options - socket options, fifo sizes, etc. +*/ +autoreply define unbind_uri { + u32 client_index; + u32 context; + u8 uri[128]; +}; + +/** \brief Connect to a given URI + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param accept_cookie - sender accept cookie, to identify this bind flavor + @param uri - a URI, e.g. "tcp4://0.0.0.0/0/80" + "tcp6://::/0/80" [ipv6], etc. + @param options - socket options, fifo sizes, etc. passed by vpp to the + server when redirecting connects + @param client_queue_address - binary API client queue address. Used by + local server when connect was redirected. +*/ +autoreply define connect_uri { + u32 client_index; + u32 context; + u8 uri[128]; + u64 client_queue_address; + u64 options[16]; +}; + +/** \brief vpp->client, accept this session + @param context - sender context, to match reply w/ request + @param listener_handle - tells client which listener this pertains to + @param handle - unique session identifier + @param session_thread_index - thread index of new session + @param rx_fifo_address - rx (vpp -> vpp-client) fifo address + @param tx_fifo_address - tx (vpp-client -> vpp) fifo address + @param vpp_event_queue_address - vpp's event queue address + @param port - remote port + @param is_ip4 - 1 if the ip is ip4 + @param ip - remote ip +*/ +define accept_session { + u32 client_index; + u32 context; + u64 listener_handle; + u64 handle; + u64 server_rx_fifo; + u64 server_tx_fifo; + u64 vpp_event_queue_address; + u16 port; + u8 is_ip4; + u8 ip[16]; +}; + +/** \brief client->vpp, reply to an accept message + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param session_index - session index from accept_session / connect_reply + @param session_thread_index - thread index from accept_session / + connect_reply +*/ +define accept_session_reply { + u32 context; + i32 retval; + u64 handle; +}; + +/** \brief bidirectional disconnect API + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param handle - session handle obtained from accept/connect +*/ +define disconnect_session { + u32 client_index; + u32 context; + u64 handle; +}; + +/** \brief bidirectional disconnect reply API + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param handle - session handle +*/ +define disconnect_session_reply { + u32 client_index; + u32 context; + i32 retval; + u64 handle; +}; + +/** \brief vpp->client reset session API + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param handle - session handle obtained via accept/connects +*/ +define reset_session { + u32 client_index; + u32 context; + u64 handle; +}; + +/** \brief client->vpp reset session reply + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param handle - session handle obtained via accept/connect +*/ +define reset_session_reply { + u32 client_index; + u32 context; + i32 retval; + u64 handle; +}; + +/** \brief Bind to an ip:port pair for a given transport protocol + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param vrf - bind namespace + @param is_ip4 - flag that is 1 if ip address family is IPv4 + @param ip - ip address + @param port - port + @param proto - protocol 0 - TCP 1 - UDP + @param options - socket options, fifo sizes, etc. +*/ +define bind_sock { + u32 client_index; + u32 context; + u32 vrf; + u8 is_ip4; + u8 ip[16]; + u16 port; + u8 proto; + u64 options[16]; +}; + +/** \brief Unbind + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param handle - bind handle obtained from bind reply +*/ +autoreply define unbind_sock { + u32 client_index; + u32 context; + u64 handle; +}; + +/** \brief Connect to a remote peer + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param app_connect - application connection id to be returned in reply + @param vrf - connection namespace + @param is_ip4 - flag that is 1 if ip address family is IPv4 + @param ip - ip address + @param port - port + @param proto - protocol 0 - TCP 1 - UDP + @param client_queue_address - client's API queue address. Non-zero when + used to perform redirects + @param options - socket options, fifo sizes, etc. when doing redirects +*/ +autoreply define connect_sock { + u32 client_index; + u32 context; + u32 vrf; + u8 is_ip4; + u8 ip[16]; + u16 port; + u8 proto; + u64 client_queue_address; + u64 options[16]; +}; + +/** \brief Bind reply + @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 segment_name_length - length of segment name + @param segment_name - name of segment client needs to attach to +*/ +define bind_sock_reply { + u32 context; + u64 handle; + i32 retval; + u64 server_event_queue_address; + u32 segment_size; + u8 segment_name_length; + u8 segment_name[128]; +}; + +/* Dummy connect message -- needed to satisfy api generators +* +* NEVER USED, doxygen tags elided on purpose. +*/ +define connect_session { + u32 client_index; + u32 context; +}; + +/** \brief vpp/server->client, connect reply -- used for all connect_* messages + @param context - sender context, to match reply w/ request + @param retval - return code for the request + @param handle - connection handle + @param server_rx_fifo - rx (vpp -> vpp-client) fifo address + @param server_tx_fifo - tx (vpp-client -> vpp) fifo address + @param vpp_event_queue_address - vpp's event queue address + @param segment_size - size of segment to be attached. Only for redirects. + @param segment_name_length - non-zero if the client needs to attach to + the fifo segment + @param segment_name - set if the client needs to attach to the segment +*/ +define connect_session_reply { + u32 context; + i32 retval; + u64 handle; + u64 server_rx_fifo; + u64 server_tx_fifo; + u64 vpp_event_queue_address; + u32 segment_size; + u8 segment_name_length; + u8 segment_name[128]; +}; + +/** \brief enable/disable session layer + @param client_index - opaque cookie to identify the sender + client to vpp direction only + @param context - sender context, to match reply w/ request + @param is_enable - disable session layer if 0, enable otherwise +*/ +autoreply define session_enable_disable { + u32 client_index; + u32 context; + u8 is_enable; +}; + +/* + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session.c b/src/vnet/session/session.c new file mode 100644 index 00000000..dc930ce8 --- /dev/null +++ b/src/vnet/session/session.c @@ -0,0 +1,1036 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +/** + * @file + * @brief Session and session manager + */ + +#include <vnet/session/session.h> +#include <vnet/session/session_debug.h> +#include <vnet/session/application.h> +#include <vlibmemory/api.h> +#include <vnet/dpo/load_balance.h> +#include <vnet/fib/ip4_fib.h> +#include <vnet/tcp/tcp.h> + +session_manager_main_t session_manager_main; +extern transport_proto_vft_t *tp_vfts; + +int +stream_session_create_i (segment_manager_t * sm, transport_connection_t * tc, + u8 alloc_fifos, stream_session_t ** ret_s) +{ + session_manager_main_t *smm = &session_manager_main; + svm_fifo_t *server_rx_fifo = 0, *server_tx_fifo = 0; + u32 fifo_segment_index; + u32 pool_index; + stream_session_t *s; + u64 value; + u32 thread_index = tc->thread_index; + int rv; + + ASSERT (thread_index == vlib_get_thread_index ()); + + /* Create the session */ + pool_get_aligned (smm->sessions[thread_index], s, CLIB_CACHE_LINE_BYTES); + memset (s, 0, sizeof (*s)); + pool_index = s - smm->sessions[thread_index]; + + /* Allocate fifos */ + if (alloc_fifos) + { + if ((rv = segment_manager_alloc_session_fifos (sm, &server_rx_fifo, + &server_tx_fifo, + &fifo_segment_index))) + { + pool_put (smm->sessions[thread_index], s); + return rv; + } + /* Initialize backpointers */ + server_rx_fifo->master_session_index = pool_index; + server_rx_fifo->master_thread_index = thread_index; + + server_tx_fifo->master_session_index = pool_index; + server_tx_fifo->master_thread_index = thread_index; + + s->server_rx_fifo = server_rx_fifo; + s->server_tx_fifo = server_tx_fifo; + s->svm_segment_index = fifo_segment_index; + } + + /* Initialize state machine, such as it is... */ + s->session_type = session_type_from_proto_and_ip (tc->transport_proto, + tc->is_ip4); + s->session_state = SESSION_STATE_CONNECTING; + s->thread_index = thread_index; + s->session_index = pool_index; + + /* Attach transport to session */ + s->connection_index = tc->c_index; + + /* Attach session to transport */ + tc->s_index = s->session_index; + + /* Add to the main lookup table */ + value = stream_session_handle (s); + stream_session_table_add_for_tc (tc, value); + + *ret_s = s; + + return 0; +} + +/** + * Discards bytes from buffer chain + * + * It discards n_bytes_to_drop starting at first buffer after chain_b + */ +always_inline void +session_enqueue_discard_chain_bytes (vlib_main_t * vm, vlib_buffer_t * b, + vlib_buffer_t ** chain_b, + u32 n_bytes_to_drop) +{ + vlib_buffer_t *next = *chain_b; + u32 to_drop = n_bytes_to_drop; + ASSERT (b->flags & VLIB_BUFFER_NEXT_PRESENT); + while (to_drop && (next->flags & VLIB_BUFFER_NEXT_PRESENT)) + { + next = vlib_get_buffer (vm, next->next_buffer); + if (next->current_length > to_drop) + { + vlib_buffer_advance (next, to_drop); + to_drop = 0; + } + else + { + to_drop -= next->current_length; + next->current_length = 0; + } + } + *chain_b = next; + + if (to_drop == 0) + b->total_length_not_including_first_buffer -= n_bytes_to_drop; +} + +/** + * Enqueue buffer chain tail + */ +always_inline int +session_enqueue_chain_tail (stream_session_t * s, vlib_buffer_t * b, + u32 offset, u8 is_in_order) +{ + vlib_buffer_t *chain_b; + u32 chain_bi, len, diff; + vlib_main_t *vm = vlib_get_main (); + u8 *data; + u32 written = 0; + int rv = 0; + + if (is_in_order && offset) + { + diff = offset - b->current_length; + if (diff > b->total_length_not_including_first_buffer) + return 0; + chain_b = b; + session_enqueue_discard_chain_bytes (vm, b, &chain_b, diff); + chain_bi = vlib_get_buffer_index (vm, chain_b); + } + else + chain_bi = b->next_buffer; + + do + { + chain_b = vlib_get_buffer (vm, chain_bi); + data = vlib_buffer_get_current (chain_b); + len = chain_b->current_length; + if (!len) + continue; + if (is_in_order) + { + rv = svm_fifo_enqueue_nowait (s->server_rx_fifo, len, data); + if (rv == len) + { + written += rv; + } + else if (rv < len) + { + return (rv > 0) ? (written + rv) : written; + } + else if (rv > len) + { + written += rv; + + /* written more than what was left in chain */ + if (written > b->total_length_not_including_first_buffer) + return written; + + /* drop the bytes that have already been delivered */ + session_enqueue_discard_chain_bytes (vm, b, &chain_b, rv - len); + } + } + else + { + rv = svm_fifo_enqueue_with_offset (s->server_rx_fifo, offset, len, + data); + if (rv) + { + clib_warning ("failed to enqueue multi-buffer seg"); + return -1; + } + offset += len; + } + } + while ((chain_bi = (chain_b->flags & VLIB_BUFFER_NEXT_PRESENT) + ? chain_b->next_buffer : 0)); + + if (is_in_order) + return written; + + return 0; +} + +/* + * Enqueue data for delivery to session peer. Does not notify peer of enqueue + * event but on request can queue notification events for later delivery by + * calling stream_server_flush_enqueue_events(). + * + * @param tc Transport connection which is to be enqueued data + * @param b Buffer to be enqueued + * @param offset Offset at which to start enqueueing if out-of-order + * @param queue_event Flag to indicate if peer is to be notified or if event + * is to be queued. The former is useful when more data is + * enqueued and only one event is to be generated. + * @param is_in_order Flag to indicate if data is in order + * @return Number of bytes enqueued or a negative value if enqueueing failed. + */ +int +stream_session_enqueue_data (transport_connection_t * tc, vlib_buffer_t * b, + u32 offset, u8 queue_event, u8 is_in_order) +{ + stream_session_t *s; + int enqueued = 0, rv, in_order_off; + + s = stream_session_get (tc->s_index, tc->thread_index); + + if (is_in_order) + { + 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)) + { + in_order_off = enqueued > b->current_length ? enqueued : 0; + rv = session_enqueue_chain_tail (s, b, in_order_off, 1); + if (rv > 0) + enqueued += rv; + } + } + else + { + rv = svm_fifo_enqueue_with_offset (s->server_rx_fifo, offset, + b->current_length, + vlib_buffer_get_current (b)); + if (PREDICT_FALSE ((b->flags & VLIB_BUFFER_NEXT_PRESENT) && !rv)) + session_enqueue_chain_tail (s, b, offset + b->current_length, 0); + /* if something was enqueued, report even this as success for ooo + * segment handling */ + return rv; + } + + if (queue_event) + { + /* Queue RX event on this fifo. Eventually these will need to be flushed + * by calling stream_server_flush_enqueue_events () */ + session_manager_main_t *smm = vnet_get_session_manager_main (); + u32 thread_index = s->thread_index; + u32 my_enqueue_epoch = smm->current_enqueue_epoch[thread_index]; + + if (s->enqueue_epoch != my_enqueue_epoch) + { + s->enqueue_epoch = my_enqueue_epoch; + vec_add1 (smm->session_indices_to_enqueue_by_thread[thread_index], + s - smm->sessions[thread_index]); + } + } + + return enqueued; +} + +/** Check if we have space in rx fifo to push more bytes */ +u8 +stream_session_no_space (transport_connection_t * tc, u32 thread_index, + u16 data_len) +{ + stream_session_t *s = stream_session_get (tc->s_index, thread_index); + + if (PREDICT_FALSE (s->session_state != SESSION_STATE_READY)) + return 1; + + if (data_len > svm_fifo_max_enqueue (s->server_rx_fifo)) + return 1; + + return 0; +} + +u32 +stream_session_tx_fifo_max_dequeue (transport_connection_t * tc) +{ + stream_session_t *s = stream_session_get (tc->s_index, tc->thread_index); + if (!s->server_tx_fifo) + return 0; + return svm_fifo_max_dequeue (s->server_tx_fifo); +} + +int +stream_session_peek_bytes (transport_connection_t * tc, u8 * buffer, + u32 offset, u32 max_bytes) +{ + stream_session_t *s = stream_session_get (tc->s_index, tc->thread_index); + return svm_fifo_peek (s->server_tx_fifo, offset, max_bytes, buffer); +} + +u32 +stream_session_dequeue_drop (transport_connection_t * tc, u32 max_bytes) +{ + stream_session_t *s = stream_session_get (tc->s_index, tc->thread_index); + return svm_fifo_dequeue_drop (s->server_tx_fifo, max_bytes); +} + +/** + * Notify session peer that new data has been enqueued. + * + * @param s Stream session for which the event is to be generated. + * @param block Flag to indicate if call should block if event queue is full. + * + * @return 0 on succes or negative number if failed to send notification. + */ +static int +stream_session_enqueue_notify (stream_session_t * s, u8 block) +{ + application_t *app; + session_fifo_event_t evt; + unix_shared_memory_queue_t *q; + static u32 serial_number; + + if (PREDICT_FALSE (s->session_state == SESSION_STATE_CLOSED)) + { + /* Session is closed so app will never clean up. Flush rx fifo */ + u32 to_dequeue = svm_fifo_max_dequeue (s->server_rx_fifo); + if (to_dequeue) + svm_fifo_dequeue_drop (s->server_rx_fifo, to_dequeue); + return 0; + } + + /* Get session's server */ + app = application_get_if_valid (s->app_index); + + if (PREDICT_FALSE (app == 0)) + { + clib_warning ("invalid s->app_index = %d", s->app_index); + return 0; + } + + /* Built-in server? Hand event to the callback... */ + if (app->cb_fns.builtin_server_rx_callback) + return app->cb_fns.builtin_server_rx_callback (s); + + /* If no event, send one */ + if (svm_fifo_set_event (s->server_rx_fifo)) + { + /* Fabricate event */ + evt.fifo = s->server_rx_fifo; + evt.event_type = FIFO_EVENT_APP_RX; + evt.event_id = serial_number++; + + /* Add event to server's event queue */ + q = app->event_queue; + + /* Based on request block (or not) for lack of space */ + if (block || PREDICT_TRUE (q->cursize < q->maxsize)) + unix_shared_memory_queue_add (app->event_queue, (u8 *) & evt, + 0 /* do wait for mutex */ ); + else + { + clib_warning ("fifo full"); + return -1; + } + } + + /* *INDENT-OFF* */ + SESSION_EVT_DBG(SESSION_EVT_ENQ, s, ({ + ed->data[0] = evt.event_id; + ed->data[1] = svm_fifo_max_dequeue (s->server_rx_fifo); + })); + /* *INDENT-ON* */ + + return 0; +} + +/** + * Flushes queue of sessions that are to be notified of new data + * enqueued events. + * + * @param thread_index Thread index for which the flush is to be performed. + * @return 0 on success or a positive number indicating the number of + * failures due to API queue being full. + */ +int +session_manager_flush_enqueue_events (u32 thread_index) +{ + session_manager_main_t *smm = &session_manager_main; + u32 *session_indices_to_enqueue; + int i, errors = 0; + + session_indices_to_enqueue = + smm->session_indices_to_enqueue_by_thread[thread_index]; + + for (i = 0; i < vec_len (session_indices_to_enqueue); i++) + { + stream_session_t *s0; + + /* Get session */ + s0 = stream_session_get_if_valid (session_indices_to_enqueue[i], + thread_index); + if (s0 == 0 || stream_session_enqueue_notify (s0, 0 /* don't block */ )) + { + errors++; + } + } + + vec_reset_length (session_indices_to_enqueue); + + smm->session_indices_to_enqueue_by_thread[thread_index] = + session_indices_to_enqueue; + + /* Increment enqueue epoch for next round */ + smm->current_enqueue_epoch[thread_index]++; + + return errors; +} + +/** + * Init fifo tail and head pointers + * + * Useful if transport uses absolute offsets for tracking ooo segments. + */ +void +stream_session_init_fifos_pointers (transport_connection_t * tc, + u32 rx_pointer, u32 tx_pointer) +{ + stream_session_t *s; + s = stream_session_get (tc->s_index, tc->thread_index); + svm_fifo_init_pointers (s->server_rx_fifo, rx_pointer); + svm_fifo_init_pointers (s->server_tx_fifo, tx_pointer); +} + +int +stream_session_connect_notify (transport_connection_t * tc, u8 is_fail) +{ + application_t *app; + stream_session_t *new_s = 0; + u64 handle; + u32 opaque = 0; + int error = 0; + u8 st; + + st = session_type_from_proto_and_ip (tc->transport_proto, tc->is_ip4); + handle = stream_session_half_open_lookup_handle (&tc->lcl_ip, &tc->rmt_ip, + tc->lcl_port, tc->rmt_port, + st); + if (handle == HALF_OPEN_LOOKUP_INVALID_VALUE) + { + TCP_DBG ("half-open was removed!"); + return -1; + } + + /* Cleanup half-open table */ + stream_session_half_open_table_del (tc); + + /* Get the app's index from the handle we stored when opening connection + * and the opaque (api_context for external apps) from transport session + * index */ + app = application_get_if_valid (handle >> 32); + if (!app) + return -1; + + opaque = tc->s_index; + + if (!is_fail) + { + segment_manager_t *sm; + u8 alloc_fifos; + sm = application_get_connect_segment_manager (app); + alloc_fifos = application_is_proxy (app); + /* Create new session (svm segments are allocated if needed) */ + if (stream_session_create_i (sm, tc, alloc_fifos, &new_s)) + { + is_fail = 1; + error = -1; + } + else + new_s->app_index = app->index; + } + + /* Notify client application */ + if (app->cb_fns.session_connected_callback (app->index, opaque, new_s, + is_fail)) + { + clib_warning ("failed to notify app"); + if (!is_fail) + stream_session_disconnect (new_s); + } + else + { + if (!is_fail) + new_s->session_state = SESSION_STATE_READY; + } + + return error; +} + +void +stream_session_accept_notify (transport_connection_t * tc) +{ + application_t *server; + stream_session_t *s; + + s = stream_session_get (tc->s_index, tc->thread_index); + server = application_get (s->app_index); + server->cb_fns.session_accept_callback (s); +} + +/** + * Notification from transport that connection is being closed. + * + * A disconnect is sent to application but state is not removed. Once + * disconnect is acknowledged by application, session disconnect is called. + * Ultimately this leads to close being called on transport (passive close). + */ +void +stream_session_disconnect_notify (transport_connection_t * tc) +{ + application_t *server; + stream_session_t *s; + + s = stream_session_get (tc->s_index, tc->thread_index); + server = application_get (s->app_index); + server->cb_fns.session_disconnect_callback (s); +} + +/** + * Cleans up session and lookup table. + */ +void +stream_session_delete (stream_session_t * s) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + int rv; + + /* Delete from the main lookup table. */ + if ((rv = stream_session_table_del (s))) + clib_warning ("hash delete error, rv %d", rv); + + /* Cleanup fifo segments */ + segment_manager_dealloc_fifos (s->svm_segment_index, s->server_rx_fifo, + s->server_tx_fifo); + + pool_put (smm->sessions[s->thread_index], s); + if (CLIB_DEBUG) + memset (s, 0xFA, sizeof (*s)); +} + +/** + * Notification from transport that connection is being deleted + * + * This removes the session if it is still valid. It should be called only on + * previously fully established sessions. For instance failed connects should + * call stream_session_connect_notify and indicate that the connect has + * failed. + */ +void +stream_session_delete_notify (transport_connection_t * tc) +{ + stream_session_t *s; + + /* App might've been removed already */ + s = stream_session_get_if_valid (tc->s_index, tc->thread_index); + if (!s) + return; + stream_session_delete (s); +} + +/** + * Notify application that connection has been reset. + */ +void +stream_session_reset_notify (transport_connection_t * tc) +{ + stream_session_t *s; + application_t *app; + s = stream_session_get (tc->s_index, tc->thread_index); + + app = application_get (s->app_index); + app->cb_fns.session_reset_callback (s); +} + +/** + * Accept a stream session. Optionally ping the server by callback. + */ +int +stream_session_accept (transport_connection_t * tc, u32 listener_index, + u8 sst, u8 notify) +{ + application_t *server; + stream_session_t *s, *listener; + segment_manager_t *sm; + + int rv; + + /* Find the server */ + listener = listen_session_get (sst, listener_index); + server = application_get (listener->app_index); + + sm = application_get_listen_segment_manager (server, listener); + if ((rv = stream_session_create_i (sm, tc, 1, &s))) + return rv; + + s->app_index = server->index; + s->listener_index = listener_index; + s->session_state = SESSION_STATE_ACCEPTING; + + /* Shoulder-tap the server */ + if (notify) + { + server->cb_fns.session_accept_callback (s); + } + + return 0; +} + +/** + * Ask transport to open connection to remote transport endpoint. + * + * Stores handle for matching request with reply since the call can be + * asynchronous. For instance, for TCP the 3-way handshake must complete + * before reply comes. Session is only created once connection is established. + * + * @param app_index Index of the application requesting the connect + * @param st Session type requested. + * @param tep Remote transport endpoint + * @param res Resulting transport connection . + */ +int +stream_session_open (u32 app_index, session_type_t st, + transport_endpoint_t * rmt, + transport_connection_t ** res) +{ + transport_connection_t *tc; + int rv; + u64 handle; + + rv = tp_vfts[st].open (rmt); + if (rv < 0) + { + clib_warning ("Transport failed to open connection."); + return VNET_API_ERROR_SESSION_CONNECT_FAIL; + } + + tc = tp_vfts[st].get_half_open ((u32) rv); + + /* Save app and tc index. The latter is needed to help establish the + * connection while the former is needed when the connect notify comes + * and we have to notify the external app */ + handle = (((u64) app_index) << 32) | (u64) tc->c_index; + + /* Add to the half-open lookup table */ + stream_session_half_open_table_add (tc, handle); + + *res = tc; + + return 0; +} + +/** + * 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, transport_endpoint_t * tep) +{ + transport_connection_t *tc; + u32 tci; + + /* Transport bind/listen */ + tci = tp_vfts[s->session_type].bind (s->session_index, tep); + + if (tci == (u32) ~ 0) + return -1; + + /* Attach transport to session */ + s->connection_index = tci; + tc = tp_vfts[s->session_type].get_listener (tci); + + /* Weird but handle it ... */ + if (tc == 0) + return -1; + + /* Add to the main lookup table */ + stream_session_table_add_for_tc (tc, s->session_index); + + return 0; +} + +/** + * Ask transport to stop listening on local transport endpoint. + * + * @param s Session to stop listening on. It must be in state LISTENING. + */ +int +stream_session_stop_listen (stream_session_t * s) +{ + transport_connection_t *tc; + + if (s->session_state != SESSION_STATE_LISTENING) + { + clib_warning ("not a listening session"); + return -1; + } + + tc = tp_vfts[s->session_type].get_listener (s->connection_index); + if (!tc) + { + clib_warning ("no transport"); + return VNET_API_ERROR_ADDRESS_NOT_IN_USE; + } + + stream_session_table_del_for_tc (tc); + tp_vfts[s->session_type].unbind (s->connection_index); + return 0; +} + +void +session_send_session_evt_to_thread (u64 session_handle, + fifo_event_type_t evt_type, + u32 thread_index) +{ + static u16 serial_number = 0; + u32 tries = 0; + session_fifo_event_t evt; + unix_shared_memory_queue_t *q; + + /* Fabricate event */ + evt.session_handle = session_handle; + evt.event_type = evt_type; + evt.event_id = serial_number++; + + q = session_manager_get_vpp_event_queue (thread_index); + while (unix_shared_memory_queue_add (q, (u8 *) & evt, 1)) + { + if (tries++ == 3) + { + TCP_DBG ("failed to enqueue evt"); + break; + } + } +} + +/** + * Disconnect session and propagate to transport. This should eventually + * result in a delete notification that allows us to cleanup session state. + * Called for both active/passive disconnects. + * + * Should be called from the session's thread. + */ +void +stream_session_disconnect (stream_session_t * s) +{ + s->session_state = SESSION_STATE_CLOSED; + tp_vfts[s->session_type].close (s->connection_index, s->thread_index); +} + +/** + * Cleanup transport and session state. + * + * Notify transport of the cleanup, wait for a delete notify to actually + * remove the session state. + */ +void +stream_session_cleanup (stream_session_t * s) +{ + int rv; + + s->session_state = SESSION_STATE_CLOSED; + + /* Delete from the main lookup table to avoid more enqueues */ + rv = stream_session_table_del (s); + if (rv) + clib_warning ("hash delete error, rv %d", rv); + + tp_vfts[s->session_type].cleanup (s->connection_index, s->thread_index); +} + +/** + * Allocate vpp event queue (once) per worker thread + */ +void +session_vpp_event_queue_allocate (session_manager_main_t * smm, + u32 thread_index) +{ + api_main_t *am = &api_main; + void *oldheap; + u32 event_queue_length = 2048; + + if (smm->vpp_event_queues[thread_index] == 0) + { + /* Allocate event fifo in the /vpe-api shared-memory segment */ + oldheap = svm_push_data_heap (am->vlib_rp); + + if (smm->configured_event_queue_length) + event_queue_length = smm->configured_event_queue_length; + + smm->vpp_event_queues[thread_index] = + unix_shared_memory_queue_init + (event_queue_length, + sizeof (session_fifo_event_t), 0 /* consumer pid */ , + 0 /* (do not) send signal when queue non-empty */ ); + + svm_pop_heap (oldheap); + } +} + +session_type_t +session_type_from_proto_and_ip (transport_proto_t proto, u8 is_ip4) +{ + if (proto == TRANSPORT_PROTO_TCP) + { + if (is_ip4) + return SESSION_TYPE_IP4_TCP; + else + return SESSION_TYPE_IP6_TCP; + } + else + { + if (is_ip4) + return SESSION_TYPE_IP4_UDP; + else + return SESSION_TYPE_IP6_UDP; + } + + return SESSION_N_TYPES; +} + +static clib_error_t * +session_manager_main_enable (vlib_main_t * vm) +{ + session_manager_main_t *smm = &session_manager_main; + vlib_thread_main_t *vtm = vlib_get_thread_main (); + u32 num_threads; + u32 preallocated_sessions_per_worker; + int i; + + num_threads = 1 /* main thread */ + vtm->n_threads; + + if (num_threads < 1) + return clib_error_return (0, "n_thread_stacks not set"); + + /* $$$ config parameters */ + svm_fifo_segment_init (0x200000000ULL /* first segment base VA */ , + 20 /* timeout in seconds */ ); + + /* configure per-thread ** vectors */ + vec_validate (smm->sessions, num_threads - 1); + vec_validate (smm->session_indices_to_enqueue_by_thread, num_threads - 1); + vec_validate (smm->tx_buffers, num_threads - 1); + vec_validate (smm->pending_event_vector, num_threads - 1); + vec_validate (smm->free_event_vector, num_threads - 1); + vec_validate (smm->current_enqueue_epoch, num_threads - 1); + vec_validate (smm->vpp_event_queues, num_threads - 1); + + for (i = 0; i < num_threads; i++) + { + vec_validate (smm->free_event_vector[i], 0); + _vec_len (smm->free_event_vector[i]) = 0; + vec_validate (smm->pending_event_vector[i], 0); + _vec_len (smm->pending_event_vector[i]) = 0; + } + +#if SESSION_DBG + vec_validate (smm->last_event_poll_by_thread, num_threads - 1); +#endif + + /* Allocate vpp event queues */ + for (i = 0; i < vec_len (smm->vpp_event_queues); i++) + session_vpp_event_queue_allocate (smm, i); + + /* Preallocate sessions */ + if (smm->preallocated_sessions) + { + if (num_threads == 1) + { + pool_init_fixed (smm->sessions[0], smm->preallocated_sessions); + } + else + { + int j; + preallocated_sessions_per_worker = + (1.1 * (f64) smm->preallocated_sessions / + (f64) (num_threads - 1)); + + for (j = 1; j < num_threads; j++) + { + pool_init_fixed (smm->sessions[j], + preallocated_sessions_per_worker); + } + } + } + + session_lookup_init (); + + smm->is_enabled = 1; + + /* Enable TCP transport */ + vnet_tcp_enable_disable (vm, 1); + + return 0; +} + +void +session_node_enable_disable (u8 is_en) +{ + u8 state = is_en ? VLIB_NODE_STATE_POLLING : VLIB_NODE_STATE_DISABLED; + /* *INDENT-OFF* */ + foreach_vlib_main (({ + vlib_node_set_state (this_vlib_main, session_queue_node.index, + state); + })); + /* *INDENT-ON* */ +} + +clib_error_t * +vnet_session_enable_disable (vlib_main_t * vm, u8 is_en) +{ + if (is_en) + { + if (session_manager_main.is_enabled) + return 0; + + session_node_enable_disable (is_en); + + return session_manager_main_enable (vm); + } + else + { + session_manager_main.is_enabled = 0; + session_node_enable_disable (is_en); + } + + return 0; +} + +clib_error_t * +session_manager_main_init (vlib_main_t * vm) +{ + session_manager_main_t *smm = &session_manager_main; + smm->is_enabled = 0; + return 0; +} + +VLIB_INIT_FUNCTION (session_manager_main_init); + +static clib_error_t * +session_config_fn (vlib_main_t * vm, unformat_input_t * input) +{ + session_manager_main_t *smm = &session_manager_main; + u32 nitems; + uword tmp; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "event-queue-length %d", &nitems)) + { + if (nitems >= 2048) + smm->configured_event_queue_length = nitems; + else + clib_warning ("event queue length %d too small, ignored", nitems); + } + else if (unformat (input, "preallocated-sessions %d", + &smm->preallocated_sessions)) + ; + else if (unformat (input, "v4-session-table-buckets %d", + &smm->configured_v4_session_table_buckets)) + ; + else if (unformat (input, "v4-halfopen-table-buckets %d", + &smm->configured_v4_halfopen_table_buckets)) + ; + else if (unformat (input, "v6-session-table-buckets %d", + &smm->configured_v6_session_table_buckets)) + ; + else if (unformat (input, "v6-halfopen-table-buckets %d", + &smm->configured_v6_halfopen_table_buckets)) + ; + else if (unformat (input, "v4-session-table-memory %U", + unformat_memory_size, &tmp)) + { + if (tmp >= 0x100000000) + return clib_error_return (0, "memory size %llx (%lld) too large", + tmp, tmp); + smm->configured_v4_session_table_memory = tmp; + } + else if (unformat (input, "v4-halfopen-table-memory %U", + unformat_memory_size, &tmp)) + { + if (tmp >= 0x100000000) + return clib_error_return (0, "memory size %llx (%lld) too large", + tmp, tmp); + smm->configured_v4_halfopen_table_memory = tmp; + } + else if (unformat (input, "v6-session-table-memory %U", + unformat_memory_size, &tmp)) + { + if (tmp >= 0x100000000) + return clib_error_return (0, "memory size %llx (%lld) too large", + tmp, tmp); + smm->configured_v6_session_table_memory = tmp; + } + else if (unformat (input, "v6-halfopen-table-memory %U", + unformat_memory_size, &tmp)) + { + if (tmp >= 0x100000000) + return clib_error_return (0, "memory size %llx (%lld) too large", + tmp, tmp); + smm->configured_v6_halfopen_table_memory = tmp; + } + else + return clib_error_return (0, "unknown input `%U'", + format_unformat_error, input); + } + return 0; +} + +VLIB_CONFIG_FUNCTION (session_config_fn, "session"); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session.h b/src/vnet/session/session.h new file mode 100644 index 00000000..83addec2 --- /dev/null +++ b/src/vnet/session/session.h @@ -0,0 +1,436 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef __included_session_h__ +#define __included_session_h__ + +#include <vnet/session/stream_session.h> +#include <vnet/session/session_lookup.h> +#include <vnet/session/transport_interface.h> +#include <vlibmemory/unix_shared_memory_queue.h> +#include <vnet/session/session_debug.h> +#include <vnet/session/segment_manager.h> + +#define HALF_OPEN_LOOKUP_INVALID_VALUE ((u64)~0) +#define INVALID_INDEX ((u32)~0) + +/* TODO decide how much since we have pre-data as well */ +#define MAX_HDRS_LEN 100 /* Max number of bytes for headers */ + +typedef enum +{ + FIFO_EVENT_APP_RX, + FIFO_EVENT_APP_TX, + FIFO_EVENT_TIMEOUT, + FIFO_EVENT_DISCONNECT, + FIFO_EVENT_BUILTIN_RX, + FIFO_EVENT_RPC, +} fifo_event_type_t; + +static inline const char * +fifo_event_type_str (fifo_event_type_t et) +{ + switch (et) + { + case FIFO_EVENT_APP_RX: + return "FIFO_EVENT_APP_RX"; + case FIFO_EVENT_APP_TX: + return "FIFO_EVENT_APP_TX"; + case FIFO_EVENT_TIMEOUT: + return "FIFO_EVENT_TIMEOUT"; + case FIFO_EVENT_DISCONNECT: + return "FIFO_EVENT_DISCONNECT"; + case FIFO_EVENT_BUILTIN_RX: + return "FIFO_EVENT_BUILTIN_RX"; + case FIFO_EVENT_RPC: + return "FIFO_EVENT_RPC"; + default: + return "UNKNOWN FIFO EVENT"; + } +} + +#define foreach_session_input_error \ +_(NO_SESSION, "No session drops") \ +_(NO_LISTENER, "No listener for dst port drops") \ +_(ENQUEUED, "Packets pushed into rx fifo") \ +_(NOT_READY, "Session not ready packets") \ +_(FIFO_FULL, "Packets dropped for lack of rx fifo space") \ +_(EVENT_FIFO_FULL, "Events not sent for lack of event fifo space") \ +_(API_QUEUE_FULL, "Sessions not created for lack of API queue space") \ +_(NEW_SEG_NO_SPACE, "Created segment, couldn't allocate a fifo pair") \ +_(NO_SPACE, "Couldn't allocate a fifo pair") + +typedef enum +{ +#define _(sym,str) SESSION_ERROR_##sym, + foreach_session_input_error +#undef _ + SESSION_N_ERROR, +} session_error_t; + +/* Event queue input node static next indices */ +typedef enum +{ + SESSION_QUEUE_NEXT_DROP, + SESSION_QUEUE_NEXT_TCP_IP4_OUTPUT, + SESSION_QUEUE_NEXT_IP4_LOOKUP, + SESSION_QUEUE_NEXT_TCP_IP6_OUTPUT, + SESSION_QUEUE_NEXT_IP6_LOOKUP, + SESSION_QUEUE_N_NEXT, +} session_queue_next_t; + +typedef struct +{ + void *fp; + void *arg; +} rpc_args_t; + +/* *INDENT-OFF* */ +typedef CLIB_PACKED (struct { + union + { + svm_fifo_t * fifo; + u64 session_handle; + rpc_args_t rpc_args; + }; + u8 event_type; + u16 event_id; +}) session_fifo_event_t; +/* *INDENT-ON* */ + +/* Forward definition */ +typedef struct _session_manager_main session_manager_main_t; + +typedef int + (session_fifo_rx_fn) (vlib_main_t * vm, vlib_node_runtime_t * node, + session_manager_main_t * smm, + session_fifo_event_t * e0, stream_session_t * s0, + u32 thread_index, int *n_tx_pkts); + +extern session_fifo_rx_fn session_tx_fifo_peek_and_snd; +extern session_fifo_rx_fn session_tx_fifo_dequeue_and_snd; + +u8 session_node_lookup_fifo_event (svm_fifo_t * f, session_fifo_event_t * e); + +struct _session_manager_main +{ + /** Per worker thread session pools */ + stream_session_t **sessions; + + /** Pool of listen sessions. Same type as stream sessions to ease lookups */ + stream_session_t *listen_sessions[SESSION_N_TYPES]; + + /** Sparse vector to map dst port to stream server */ + u16 *stream_server_by_dst_port[SESSION_N_TYPES]; + + /** per-worker enqueue epoch counters */ + u8 *current_enqueue_epoch; + + /** Per-worker thread vector of sessions to enqueue */ + u32 **session_indices_to_enqueue_by_thread; + + /** per-worker tx buffer free lists */ + u32 **tx_buffers; + + /** Per worker-thread vector of partially read events */ + session_fifo_event_t **free_event_vector; + + /** per-worker active event vectors */ + session_fifo_event_t **pending_event_vector; + + /** vpp fifo event queue */ + unix_shared_memory_queue_t **vpp_event_queues; + + /** vpp fifo event queue configured length */ + u32 configured_event_queue_length; + + /** session table size parameters */ + u32 configured_v4_session_table_buckets; + u32 configured_v4_session_table_memory; + u32 configured_v4_halfopen_table_buckets; + u32 configured_v4_halfopen_table_memory; + u32 configured_v6_session_table_buckets; + u32 configured_v6_session_table_memory; + u32 configured_v6_halfopen_table_buckets; + u32 configured_v6_halfopen_table_memory; + + /** Unique segment name counter */ + u32 unique_segment_name_counter; + + /** Per transport rx function that can either dequeue or peek */ + session_fifo_rx_fn *session_tx_fns[SESSION_N_TYPES]; + + /** Session manager is enabled */ + u8 is_enabled; + + /** Preallocate session config parameter */ + u32 preallocated_sessions; + +#if SESSION_DBG + /** + * last event poll time by thread + * Debug only. Will cause false cache-line sharing as-is + */ + f64 *last_event_poll_by_thread; +#endif + +}; + +extern session_manager_main_t session_manager_main; +extern vlib_node_registration_t session_queue_node; + +/* + * Session manager function + */ +always_inline session_manager_main_t * +vnet_get_session_manager_main () +{ + return &session_manager_main; +} + +always_inline u8 +stream_session_is_valid (u32 si, u8 thread_index) +{ + stream_session_t *s; + s = pool_elt_at_index (session_manager_main.sessions[thread_index], si); + if (s->thread_index != thread_index || s->session_index != si + /* || s->server_rx_fifo->master_session_index != si + || s->server_tx_fifo->master_session_index != si + || s->server_rx_fifo->master_thread_index != thread_index + || s->server_tx_fifo->master_thread_index != thread_index */ ) + return 0; + return 1; +} + +always_inline stream_session_t * +stream_session_get (u32 si, u32 thread_index) +{ + ASSERT (stream_session_is_valid (si, thread_index)); + return pool_elt_at_index (session_manager_main.sessions[thread_index], si); +} + +always_inline stream_session_t * +stream_session_get_if_valid (u64 si, u32 thread_index) +{ + if (thread_index >= vec_len (session_manager_main.sessions)) + return 0; + + if (pool_is_free_index (session_manager_main.sessions[thread_index], si)) + return 0; + + ASSERT (stream_session_is_valid (si, thread_index)); + return pool_elt_at_index (session_manager_main.sessions[thread_index], si); +} + +always_inline u64 +stream_session_handle (stream_session_t * s) +{ + return ((u64) s->thread_index << 32) | (u64) s->session_index; +} + +always_inline u32 +stream_session_index_from_handle (u64 handle) +{ + return handle & 0xFFFFFFFF; +} + +always_inline u32 +stream_session_thread_from_handle (u64 handle) +{ + return handle >> 32; +} + +always_inline void +stream_session_parse_handle (u64 handle, u32 * index, u32 * thread_index) +{ + *index = stream_session_index_from_handle (handle); + *thread_index = stream_session_thread_from_handle (handle); +} + +always_inline stream_session_t * +stream_session_get_from_handle (u64 handle) +{ + session_manager_main_t *smm = &session_manager_main; + return pool_elt_at_index (smm->sessions[stream_session_thread_from_handle + (handle)], + stream_session_index_from_handle (handle)); +} + +always_inline stream_session_t * +stream_session_listener_get (u8 sst, u64 si) +{ + return pool_elt_at_index (session_manager_main.listen_sessions[sst], si); +} + +always_inline u32 +stream_session_get_index (stream_session_t * s) +{ + if (s->session_state == SESSION_STATE_LISTENING) + return s - session_manager_main.listen_sessions[s->session_type]; + + return s - session_manager_main.sessions[s->thread_index]; +} + +always_inline u32 +stream_session_max_rx_enqueue (transport_connection_t * tc) +{ + stream_session_t *s = stream_session_get (tc->s_index, tc->thread_index); + return svm_fifo_max_enqueue (s->server_rx_fifo); +} + +always_inline u32 +stream_session_rx_fifo_size (transport_connection_t * tc) +{ + stream_session_t *s = stream_session_get (tc->s_index, tc->thread_index); + return s->server_rx_fifo->nitems; +} + +u32 stream_session_tx_fifo_max_dequeue (transport_connection_t * tc); + +int +stream_session_enqueue_data (transport_connection_t * tc, vlib_buffer_t * b, + u32 offset, u8 queue_event, u8 is_in_order); +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); + +int stream_session_connect_notify (transport_connection_t * tc, u8 is_fail); +void stream_session_init_fifos_pointers (transport_connection_t * tc, + u32 rx_pointer, u32 tx_pointer); + +void stream_session_accept_notify (transport_connection_t * tc); +void stream_session_disconnect_notify (transport_connection_t * tc); +void stream_session_delete_notify (transport_connection_t * tc); +void stream_session_reset_notify (transport_connection_t * tc); +int +stream_session_accept (transport_connection_t * tc, u32 listener_index, + u8 sst, u8 notify); +int +stream_session_open (u32 app_index, session_type_t st, + transport_endpoint_t * tep, + transport_connection_t ** tc); +int stream_session_listen (stream_session_t * s, transport_endpoint_t * tep); +int stream_session_stop_listen (stream_session_t * s); +void stream_session_disconnect (stream_session_t * s); +void stream_session_cleanup (stream_session_t * s); +void session_send_session_evt_to_thread (u64 session_handle, + fifo_event_type_t evt_type, + u32 thread_index); + +u8 *format_stream_session (u8 * s, va_list * args); +uword unformat_stream_session (unformat_input_t * input, va_list * args); +uword unformat_transport_connection (unformat_input_t * input, + va_list * args); + +int +send_session_connected_callback (u32 app_index, u32 api_context, + stream_session_t * s, u8 is_fail); + + +clib_error_t *vnet_session_enable_disable (vlib_main_t * vm, u8 is_en); + +always_inline unix_shared_memory_queue_t * +session_manager_get_vpp_event_queue (u32 thread_index) +{ + return session_manager_main.vpp_event_queues[thread_index]; +} + +int session_manager_flush_enqueue_events (u32 thread_index); + +always_inline u64 +listen_session_get_handle (stream_session_t * s) +{ + ASSERT (s->session_state == SESSION_STATE_LISTENING); + return ((u64) s->session_type << 32) | s->session_index; +} + +always_inline stream_session_t * +listen_session_get_from_handle (u64 handle) +{ + session_manager_main_t *smm = &session_manager_main; + stream_session_t *s; + u32 type, index; + type = handle >> 32; + index = handle & 0xFFFFFFFF; + + if (pool_is_free_index (smm->listen_sessions[type], index)) + return 0; + + s = pool_elt_at_index (smm->listen_sessions[type], index); + ASSERT (s->session_state == SESSION_STATE_LISTENING); + return s; +} + +always_inline stream_session_t * +listen_session_new (session_type_t type) +{ + stream_session_t *s; + pool_get_aligned (session_manager_main.listen_sessions[type], s, + CLIB_CACHE_LINE_BYTES); + memset (s, 0, sizeof (*s)); + + s->session_type = type; + s->session_state = SESSION_STATE_LISTENING; + s->session_index = s - session_manager_main.listen_sessions[type]; + + return s; +} + +always_inline stream_session_t * +listen_session_get (session_type_t type, u32 index) +{ + return pool_elt_at_index (session_manager_main.listen_sessions[type], + index); +} + +always_inline void +listen_session_del (stream_session_t * s) +{ + pool_put (session_manager_main.listen_sessions[s->session_type], s); +} + +always_inline stream_session_t * +session_manager_get_listener (u8 type, u32 index) +{ + return pool_elt_at_index (session_manager_main.listen_sessions[type], + index); +} + +always_inline void +session_manager_set_transport_rx_fn (u8 type, u8 is_peek) +{ + /* If an offset function is provided, then peek instead of dequeue */ + session_manager_main.session_tx_fns[type] = (is_peek) ? + session_tx_fifo_peek_and_snd : session_tx_fifo_dequeue_and_snd; +} + +session_type_t +session_type_from_proto_and_ip (transport_proto_t proto, u8 is_ip4); + +always_inline u8 +session_manager_is_enabled () +{ + return session_manager_main.is_enabled == 1; +} + +#endif /* __included_session_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session_api.c b/src/vnet/session/session_api.c new file mode 100755 index 00000000..250f9906 --- /dev/null +++ b/src/vnet/session/session_api.c @@ -0,0 +1,763 @@ +/* + * Copyright (c) 2015-2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/vnet.h> +#include <vlibmemory/api.h> +#include <vnet/session/application.h> + +#include <vnet/vnet_msg_enum.h> +#include "application_interface.h" + +#define vl_typedefs /* define message structures */ +#include <vnet/vnet_all_api_h.h> +#undef vl_typedefs + +#define vl_endianfun /* define message structures */ +#include <vnet/vnet_all_api_h.h> +#undef vl_endianfun + +/* instantiate all the print functions we know about */ +#define vl_print(handle, ...) vlib_cli_output (handle, __VA_ARGS__) +#define vl_printfun +#include <vnet/vnet_all_api_h.h> +#undef vl_printfun + +#include <vlibapi/api_helper_macros.h> + +#define foreach_session_api_msg \ +_(MAP_ANOTHER_SEGMENT_REPLY, map_another_segment_reply) \ +_(APPLICATION_ATTACH, application_attach) \ +_(APPLICATION_DETACH, application_detach) \ +_(BIND_URI, bind_uri) \ +_(UNBIND_URI, unbind_uri) \ +_(CONNECT_URI, connect_uri) \ +_(DISCONNECT_SESSION, disconnect_session) \ +_(DISCONNECT_SESSION_REPLY, disconnect_session_reply) \ +_(ACCEPT_SESSION_REPLY, accept_session_reply) \ +_(RESET_SESSION_REPLY, reset_session_reply) \ +_(BIND_SOCK, bind_sock) \ +_(UNBIND_SOCK, unbind_sock) \ +_(CONNECT_SOCK, connect_sock) \ +_(SESSION_ENABLE_DISABLE, session_enable_disable) \ + +static int +send_add_segment_callback (u32 api_client_index, const u8 * segment_name, + u32 segment_size) +{ + vl_api_map_another_segment_t *mp; + unix_shared_memory_queue_t *q; + + q = vl_api_client_index_to_input_queue (api_client_index); + + if (!q) + return -1; + + mp = vl_msg_api_alloc (sizeof (*mp)); + memset (mp, 0, sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_MAP_ANOTHER_SEGMENT); + mp->segment_size = segment_size; + strncpy ((char *) mp->segment_name, (char *) segment_name, + sizeof (mp->segment_name) - 1); + + vl_msg_api_send_shmem (q, (u8 *) & mp); + + return 0; +} + +static int +send_session_accept_callback (stream_session_t * s) +{ + vl_api_accept_session_t *mp; + unix_shared_memory_queue_t *q, *vpp_queue; + application_t *server = application_get (s->app_index); + transport_connection_t *tc; + transport_proto_vft_t *tp_vft; + stream_session_t *listener; + + q = vl_api_client_index_to_input_queue (server->api_client_index); + vpp_queue = session_manager_get_vpp_event_queue (s->thread_index); + + if (!q) + return -1; + + mp = vl_msg_api_alloc (sizeof (*mp)); + memset (mp, 0, sizeof (*mp)); + + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_ACCEPT_SESSION); + mp->context = server->index; + listener = listen_session_get (s->session_type, s->listener_index); + tp_vft = session_get_transport_vft (s->session_type); + tc = tp_vft->get_connection (s->connection_index, s->thread_index); + mp->listener_handle = listen_session_get_handle (listener); + mp->handle = stream_session_handle (s); + mp->server_rx_fifo = pointer_to_uword (s->server_rx_fifo); + mp->server_tx_fifo = pointer_to_uword (s->server_tx_fifo); + mp->vpp_event_queue_address = pointer_to_uword (vpp_queue); + mp->port = tc->rmt_port; + mp->is_ip4 = tc->is_ip4; + clib_memcpy (&mp->ip, &tc->rmt_ip, sizeof (tc->rmt_ip)); + vl_msg_api_send_shmem (q, (u8 *) & mp); + + return 0; +} + +static void +send_session_disconnect_callback (stream_session_t * s) +{ + vl_api_disconnect_session_t *mp; + unix_shared_memory_queue_t *q; + application_t *app = application_get (s->app_index); + + q = vl_api_client_index_to_input_queue (app->api_client_index); + + if (!q) + return; + + mp = vl_msg_api_alloc (sizeof (*mp)); + memset (mp, 0, sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_DISCONNECT_SESSION); + mp->handle = stream_session_handle (s); + vl_msg_api_send_shmem (q, (u8 *) & mp); +} + +static void +send_session_reset_callback (stream_session_t * s) +{ + vl_api_reset_session_t *mp; + unix_shared_memory_queue_t *q; + application_t *app = application_get (s->app_index); + + q = vl_api_client_index_to_input_queue (app->api_client_index); + + if (!q) + return; + + mp = vl_msg_api_alloc (sizeof (*mp)); + memset (mp, 0, sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_RESET_SESSION); + mp->handle = stream_session_handle (s); + vl_msg_api_send_shmem (q, (u8 *) & mp); +} + +int +send_session_connected_callback (u32 app_index, u32 api_context, + stream_session_t * s, u8 is_fail) +{ + vl_api_connect_session_reply_t *mp; + unix_shared_memory_queue_t *q; + application_t *app; + unix_shared_memory_queue_t *vpp_queue; + + app = application_get (app_index); + q = vl_api_client_index_to_input_queue (app->api_client_index); + + if (!q) + return -1; + + mp = vl_msg_api_alloc (sizeof (*mp)); + mp->_vl_msg_id = clib_host_to_net_u16 (VL_API_CONNECT_SESSION_REPLY); + mp->context = api_context; + if (!is_fail) + { + vpp_queue = session_manager_get_vpp_event_queue (s->thread_index); + mp->server_rx_fifo = pointer_to_uword (s->server_rx_fifo); + mp->server_tx_fifo = pointer_to_uword (s->server_tx_fifo); + mp->handle = stream_session_handle (s); + mp->vpp_event_queue_address = pointer_to_uword (vpp_queue); + mp->retval = 0; + } + else + { + mp->retval = clib_host_to_net_u32 (VNET_API_ERROR_SESSION_CONNECT_FAIL); + } + + vl_msg_api_send_shmem (q, (u8 *) & mp); + return 0; +} + +/** + * Redirect a connect_uri message to the indicated server. + * Only sent if the server has bound the related port with + * URI_OPTIONS_FLAGS_USE_FIFO + */ +static int +redirect_connect_callback (u32 server_api_client_index, void *mp_arg) +{ + vl_api_connect_uri_t *mp = mp_arg; + unix_shared_memory_queue_t *server_q, *client_q; + vlib_main_t *vm = vlib_get_main (); + f64 timeout = vlib_time_now (vm) + 0.5; + application_t *app; + int rv = 0; + + server_q = vl_api_client_index_to_input_queue (server_api_client_index); + + if (!server_q) + { + rv = VNET_API_ERROR_INVALID_VALUE; + goto out; + } + + client_q = vl_api_client_index_to_input_queue (mp->client_index); + if (!client_q) + { + rv = VNET_API_ERROR_INVALID_VALUE_2; + goto out; + } + + /* Tell the server the client's API queue address, so it can reply */ + mp->client_queue_address = pointer_to_uword (client_q); + app = application_lookup (mp->client_index); + if (!app) + { + clib_warning ("no client application"); + return -1; + } + + mp->options[SESSION_OPTIONS_RX_FIFO_SIZE] = app->sm_properties.rx_fifo_size; + mp->options[SESSION_OPTIONS_TX_FIFO_SIZE] = app->sm_properties.tx_fifo_size; + + /* + * Bounce message handlers MUST NOT block the data-plane. + * Spin waiting for the queue lock, but + */ + + while (vlib_time_now (vm) < timeout) + { + rv = + unix_shared_memory_queue_add (server_q, (u8 *) & mp, 1 /*nowait */ ); + switch (rv) + { + /* correctly enqueued */ + case 0: + return VNET_CONNECT_REDIRECTED; + + /* continue spinning, wait for pthread_mutex_trylock to work */ + case -1: + continue; + + /* queue stuffed, drop the msg */ + case -2: + rv = VNET_API_ERROR_QUEUE_FULL; + goto out; + } + } +out: + /* Dispose of the message */ + vl_msg_api_free (mp); + return rv; +} + +static session_cb_vft_t uri_session_cb_vft = { + .session_accept_callback = send_session_accept_callback, + .session_disconnect_callback = send_session_disconnect_callback, + .session_connected_callback = send_session_connected_callback, + .session_reset_callback = send_session_reset_callback, + .add_segment_callback = send_add_segment_callback, + .redirect_connect_callback = redirect_connect_callback +}; + +static void +vl_api_session_enable_disable_t_handler (vl_api_session_enable_disable_t * mp) +{ + vl_api_session_enable_disable_reply_t *rmp; + vlib_main_t *vm = vlib_get_main (); + int rv = 0; + + vnet_session_enable_disable (vm, mp->is_enable); + REPLY_MACRO (VL_API_SESSION_ENABLE_DISABLE_REPLY); +} + +static void +vl_api_application_attach_t_handler (vl_api_application_attach_t * mp) +{ + vl_api_application_attach_reply_t *rmp; + vnet_app_attach_args_t _a, *a = &_a; + int rv; + + if (session_manager_is_enabled () == 0) + { + rv = VNET_API_ERROR_FEATURE_DISABLED; + goto done; + } + + STATIC_ASSERT (sizeof (u64) * SESSION_OPTIONS_N_OPTIONS <= + sizeof (mp->options), + "Out of options, fix api message definition"); + + memset (a, 0, sizeof (*a)); + + a->api_client_index = mp->client_index; + a->options = mp->options; + a->session_cb_vft = &uri_session_cb_vft; + + rv = vnet_application_attach (a); + +done: + + /* *INDENT-OFF* */ + REPLY_MACRO2 (VL_API_APPLICATION_ATTACH_REPLY, ({ + if (!rv) + { + rmp->segment_name_length = 0; + /* $$$$ policy? */ + rmp->segment_size = a->segment_size; + if (a->segment_name_length) + { + memcpy (rmp->segment_name, a->segment_name, + a->segment_name_length); + rmp->segment_name_length = a->segment_name_length; + } + rmp->app_event_queue_address = a->app_event_queue_address; + } + })); + /* *INDENT-ON* */ +} + +static void +vl_api_application_detach_t_handler (vl_api_application_detach_t * mp) +{ + vl_api_application_detach_reply_t *rmp; + int rv = VNET_API_ERROR_INVALID_VALUE_2; + vnet_app_detach_args_t _a, *a = &_a; + application_t *app; + + if (session_manager_is_enabled () == 0) + { + rv = VNET_API_ERROR_FEATURE_DISABLED; + goto done; + } + + app = application_lookup (mp->client_index); + if (app) + { + a->app_index = app->index; + rv = vnet_application_detach (a); + } + +done: + REPLY_MACRO (VL_API_APPLICATION_DETACH_REPLY); +} + +static void +vl_api_bind_uri_t_handler (vl_api_bind_uri_t * mp) +{ + vl_api_bind_uri_reply_t *rmp; + vnet_bind_args_t _a, *a = &_a; + application_t *app; + int rv; + + if (session_manager_is_enabled () == 0) + { + rv = VNET_API_ERROR_FEATURE_DISABLED; + goto done; + } + + app = application_lookup (mp->client_index); + if (app) + { + memset (a, 0, sizeof (*a)); + a->uri = (char *) mp->uri; + a->app_index = app->index; + rv = vnet_bind_uri (a); + } + else + { + rv = VNET_API_ERROR_APPLICATION_NOT_ATTACHED; + } + +done: + REPLY_MACRO (VL_API_BIND_URI_REPLY); +} + +static void +vl_api_unbind_uri_t_handler (vl_api_unbind_uri_t * mp) +{ + vl_api_unbind_uri_reply_t *rmp; + application_t *app; + vnet_unbind_args_t _a, *a = &_a; + int rv; + + if (session_manager_is_enabled () == 0) + { + rv = VNET_API_ERROR_FEATURE_DISABLED; + goto done; + } + + app = application_lookup (mp->client_index); + if (app) + { + a->uri = (char *) mp->uri; + a->app_index = app->index; + rv = vnet_unbind_uri (a); + } + else + { + rv = VNET_API_ERROR_APPLICATION_NOT_ATTACHED; + } + +done: + REPLY_MACRO (VL_API_UNBIND_URI_REPLY); +} + +static void +vl_api_connect_uri_t_handler (vl_api_connect_uri_t * mp) +{ + vl_api_connect_session_reply_t *rmp; + vnet_connect_args_t _a, *a = &_a; + application_t *app; + int rv; + + if (session_manager_is_enabled () == 0) + { + rv = VNET_API_ERROR_FEATURE_DISABLED; + goto done; + } + + app = application_lookup (mp->client_index); + if (app) + { + a->uri = (char *) mp->uri; + a->api_context = mp->context; + a->app_index = app->index; + a->mp = mp; + rv = vnet_connect_uri (a); + } + else + { + rv = VNET_API_ERROR_APPLICATION_NOT_ATTACHED; + } + + if (rv == 0 || rv == VNET_CONNECT_REDIRECTED) + return; + + /* Got some error, relay it */ + +done: + /* *INDENT-OFF* */ + REPLY_MACRO (VL_API_CONNECT_SESSION_REPLY); + /* *INDENT-ON* */ +} + +static void +vl_api_disconnect_session_t_handler (vl_api_disconnect_session_t * mp) +{ + vl_api_disconnect_session_reply_t *rmp; + vnet_disconnect_args_t _a, *a = &_a; + application_t *app; + int rv = 0; + + if (session_manager_is_enabled () == 0) + { + rv = VNET_API_ERROR_FEATURE_DISABLED; + goto done; + } + + app = application_lookup (mp->client_index); + if (app) + { + a->handle = mp->handle; + a->app_index = app->index; + rv = vnet_disconnect_session (a); + } + else + { + rv = VNET_API_ERROR_APPLICATION_NOT_ATTACHED; + } + +done: + REPLY_MACRO (VL_API_DISCONNECT_SESSION_REPLY); +} + +static void +vl_api_disconnect_session_reply_t_handler (vl_api_disconnect_session_reply_t * + mp) +{ + vnet_disconnect_args_t _a, *a = &_a; + application_t *app; + + /* Client objected to disconnecting the session, log and continue */ + if (mp->retval) + { + clib_warning ("client retval %d", mp->retval); + return; + } + + /* Disconnect has been confirmed. Confirm close to transport */ + app = application_lookup (mp->client_index); + if (app) + { + a->handle = mp->handle; + a->app_index = app->index; + vnet_disconnect_session (a); + } +} + +static void +vl_api_reset_session_reply_t_handler (vl_api_reset_session_reply_t * mp) +{ + application_t *app; + stream_session_t *s; + u32 index, thread_index; + + app = application_lookup (mp->client_index); + if (!app) + return; + + stream_session_parse_handle (mp->handle, &index, &thread_index); + s = stream_session_get_if_valid (index, thread_index); + if (s == 0 || app->index != s->app_index) + { + clib_warning ("Invalid session!"); + return; + } + + /* Client objected to resetting the session, log and continue */ + if (mp->retval) + { + clib_warning ("client retval %d", mp->retval); + return; + } + + /* This comes as a response to a reset, transport only waiting for + * confirmation to remove connection state, no need to disconnect */ + stream_session_cleanup (s); +} + +static void +vl_api_accept_session_reply_t_handler (vl_api_accept_session_reply_t * mp) +{ + stream_session_t *s; + u32 session_index, thread_index; + vnet_disconnect_args_t _a, *a = &_a; + + /* Server isn't interested, kill the session */ + if (mp->retval) + { + a->app_index = mp->context; + a->handle = mp->handle; + vnet_disconnect_session (a); + } + else + { + stream_session_parse_handle (mp->handle, &session_index, &thread_index); + s = stream_session_get_if_valid (session_index, thread_index); + if (!s) + { + clib_warning ("session doesn't exist"); + return; + } + if (s->app_index != mp->context) + { + clib_warning ("app doesn't own session"); + return; + } + /* XXX volatile? */ + s->session_state = SESSION_STATE_READY; + } +} + +static void +vl_api_map_another_segment_reply_t_handler (vl_api_map_another_segment_reply_t + * mp) +{ + clib_warning ("not implemented"); +} + +static void +vl_api_bind_sock_t_handler (vl_api_bind_sock_t * mp) +{ + vl_api_bind_sock_reply_t *rmp; + vnet_bind_args_t _a, *a = &_a; + int rv = VNET_API_ERROR_APPLICATION_NOT_ATTACHED; + application_t *app; + + if (session_manager_is_enabled () == 0) + { + rv = VNET_API_ERROR_FEATURE_DISABLED; + goto done; + } + + app = application_lookup (mp->client_index); + if (app) + { + ip46_address_t *ip46 = (ip46_address_t *) mp->ip; + + memset (a, 0, sizeof (*a)); + a->tep.is_ip4 = mp->is_ip4; + a->tep.ip = *ip46; + a->tep.port = mp->port; + a->tep.vrf = mp->vrf; + a->app_index = app->index; + + rv = vnet_bind (a); + } +done: + /* *INDENT-OFF* */ + REPLY_MACRO2 (VL_API_BIND_SOCK_REPLY,({ + if (!rv) + rmp->handle = a->handle; + })); + /* *INDENT-ONF* */ +} + +static void +vl_api_unbind_sock_t_handler (vl_api_unbind_sock_t * mp) +{ + vl_api_unbind_sock_reply_t *rmp; + vnet_unbind_args_t _a, *a = &_a; + application_t *app; + int rv = VNET_API_ERROR_APPLICATION_NOT_ATTACHED; + + if (session_manager_is_enabled () == 0) + { + rv = VNET_API_ERROR_FEATURE_DISABLED; + goto done; + } + + app = application_lookup (mp->client_index); + if (app) + { + a->app_index = mp->client_index; + a->handle = mp->handle; + rv = vnet_unbind (a); + } + +done: + REPLY_MACRO (VL_API_UNBIND_SOCK_REPLY); +} + +static void +vl_api_connect_sock_t_handler (vl_api_connect_sock_t * mp) +{ + vl_api_connect_session_reply_t *rmp; + vnet_connect_args_t _a, *a = &_a; + application_t *app; + int rv; + + if (session_manager_is_enabled () == 0) + { + rv = VNET_API_ERROR_FEATURE_DISABLED; + goto done; + } + + app = application_lookup (mp->client_index); + if (app) + { + unix_shared_memory_queue_t *client_q; + ip46_address_t *ip46 = (ip46_address_t *) mp->ip; + + client_q = vl_api_client_index_to_input_queue (mp->client_index); + mp->client_queue_address = pointer_to_uword (client_q); + a->tep.is_ip4 = mp->is_ip4; + a->tep.ip = *ip46; + a->tep.port = mp->port; + a->tep.vrf = mp->vrf; + a->api_context = mp->context; + a->app_index = app->index; + a->proto = mp->proto; + a->mp = mp; + rv = vnet_connect (a); + } + else + { + rv = VNET_API_ERROR_APPLICATION_NOT_ATTACHED; + } + + if (rv == 0 || rv == VNET_CONNECT_REDIRECTED) + return; + + /* Got some error, relay it */ + +done: + REPLY_MACRO (VL_API_CONNECT_SESSION_REPLY); +} + +static clib_error_t * +application_reaper_cb (u32 client_index) +{ + application_t *app = application_lookup (client_index); + vnet_app_detach_args_t _a, *a = &_a; + if (app) + { + a->app_index = app->index; + vnet_application_detach (a); + } + return 0; +} + +VL_MSG_API_REAPER_FUNCTION (application_reaper_cb); + +#define vl_msg_name_crc_list +#include <vnet/vnet_all_api_h.h> +#undef vl_msg_name_crc_list + +static void +setup_message_id_table (api_main_t * am) +{ +#define _(id,n,crc) vl_msg_api_add_msg_name_crc (am, #n "_" #crc, id); + foreach_vl_msg_name_crc_session; +#undef _ +} + +/* + * session_api_hookup + * Add uri's API message handlers to the table. + * vlib has alread mapped shared memory and + * added the client registration handlers. + * See .../open-repo/vlib/memclnt_vlib.c:memclnt_process() + */ +static clib_error_t * +session_api_hookup (vlib_main_t * vm) +{ + api_main_t *am = &api_main; + +#define _(N,n) \ + vl_msg_api_set_handlers(VL_API_##N, #n, \ + vl_api_##n##_t_handler, \ + vl_noop_handler, \ + vl_api_##n##_t_endian, \ + vl_api_##n##_t_print, \ + sizeof(vl_api_##n##_t), 1); + foreach_session_api_msg; +#undef _ + + /* + * Messages which bounce off the data-plane to + * an API client. Simply tells the message handling infra not + * to free the message. + * + * Bounced message handlers MUST NOT block the data plane + */ + am->message_bounce[VL_API_CONNECT_URI] = 1; + am->message_bounce[VL_API_CONNECT_SOCK] = 1; + + /* + * Set up the (msg_name, crc, message-id) table + */ + setup_message_id_table (am); + + return 0; +} + +VLIB_API_INIT_FUNCTION (session_api_hookup); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session_cli.c b/src/vnet/session/session_cli.c new file mode 100755 index 00000000..8c30a1df --- /dev/null +++ b/src/vnet/session/session_cli.c @@ -0,0 +1,494 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#include <vnet/session/application.h> +#include <vnet/session/session.h> + +u8 * +format_stream_session_fifos (u8 * s, va_list * args) +{ + stream_session_t *ss = va_arg (*args, stream_session_t *); + int verbose = va_arg (*args, int); + session_fifo_event_t _e, *e = &_e; + u8 found; + + s = format (s, " Rx fifo: %U", format_svm_fifo, ss->server_rx_fifo, 1); + if (verbose > 2 && ss->server_rx_fifo->has_event) + { + found = session_node_lookup_fifo_event (ss->server_rx_fifo, e); + s = format (s, " session node event: %s\n", + found ? "found" : "not found"); + } + s = format (s, " Tx fifo: %U", format_svm_fifo, ss->server_tx_fifo, 1); + if (verbose > 2 && ss->server_tx_fifo->has_event) + { + found = session_node_lookup_fifo_event (ss->server_tx_fifo, e); + s = format (s, " session node event: %s\n", + found ? "found" : "not found"); + } + return s; +} + +/** + * Format stream session as per the following format + * + * verbose: + * "Connection", "Rx fifo", "Tx fifo", "Session Index" + * non-verbose: + * "Connection" + */ +u8 * +format_stream_session (u8 * s, va_list * args) +{ + stream_session_t *ss = va_arg (*args, stream_session_t *); + int verbose = va_arg (*args, int); + transport_proto_vft_t *tp_vft; + u8 *str = 0; + tp_vft = session_get_transport_vft (ss->session_type); + + if (verbose == 1 && ss->session_state >= SESSION_STATE_ACCEPTING) + str = format (0, "%-10u%-10u%-10lld", + svm_fifo_max_dequeue (ss->server_rx_fifo), + svm_fifo_max_enqueue (ss->server_tx_fifo), + stream_session_get_index (ss)); + + if (ss->session_state == SESSION_STATE_READY + || ss->session_state == SESSION_STATE_ACCEPTING + || ss->session_state == SESSION_STATE_CLOSED) + { + s = format (s, "%U", tp_vft->format_connection, ss->connection_index, + ss->thread_index, verbose); + if (verbose == 1) + s = format (s, "%v", str); + if (verbose > 1) + s = format (s, "%U", format_stream_session_fifos, ss, verbose); + } + else if (ss->session_state == SESSION_STATE_LISTENING) + { + s = format (s, "%-40U%v", tp_vft->format_listener, ss->connection_index, + str); + } + else if (ss->session_state == SESSION_STATE_CONNECTING) + { + s = format (s, "%-40U%v", tp_vft->format_half_open, + ss->connection_index, str); + } + else + { + clib_warning ("Session in state: %d!", ss->session_state); + } + vec_free (str); + + return s; +} + +uword +unformat_stream_session_id (unformat_input_t * input, va_list * args) +{ + u8 *proto = va_arg (*args, u8 *); + ip46_address_t *lcl = va_arg (*args, ip46_address_t *); + ip46_address_t *rmt = va_arg (*args, ip46_address_t *); + u16 *lcl_port = va_arg (*args, u16 *); + u16 *rmt_port = va_arg (*args, u16 *); + u8 *is_ip4 = va_arg (*args, u8 *); + u8 tuple_is_set = 0; + + memset (lcl, 0, sizeof (*lcl)); + memset (rmt, 0, sizeof (*rmt)); + + if (unformat (input, "tcp")) + { + *proto = TRANSPORT_PROTO_TCP; + } + if (unformat (input, "udp")) + { + *proto = TRANSPORT_PROTO_UDP; + } + if (unformat (input, "%U:%d->%U:%d", unformat_ip4_address, &lcl->ip4, + lcl_port, unformat_ip4_address, &rmt->ip4, rmt_port)) + { + *is_ip4 = 1; + tuple_is_set = 1; + } + else if (unformat (input, "%U:%d->%U:%d", unformat_ip6_address, &lcl->ip6, + lcl_port, unformat_ip6_address, &rmt->ip6, rmt_port)) + { + *is_ip4 = 0; + tuple_is_set = 1; + } + + return tuple_is_set; +} + +uword +unformat_stream_session (unformat_input_t * input, va_list * args) +{ + stream_session_t **result = va_arg (*args, stream_session_t **); + stream_session_t *s; + u8 proto = ~0; + ip46_address_t lcl, rmt; + u32 lcl_port = 0, rmt_port = 0; + u8 is_ip4 = 0, s_type = ~0; + + if (!unformat (input, "%U", unformat_stream_session_id, &proto, &lcl, &rmt, + &lcl_port, &rmt_port, &is_ip4)) + return 0; + + s_type = session_type_from_proto_and_ip (proto, is_ip4); + if (is_ip4) + s = stream_session_lookup4 (&lcl.ip4, &rmt.ip4, + clib_host_to_net_u16 (lcl_port), + clib_host_to_net_u16 (rmt_port), s_type); + else + s = stream_session_lookup6 (&lcl.ip6, &rmt.ip6, + clib_host_to_net_u16 (lcl_port), + clib_host_to_net_u16 (rmt_port), s_type); + if (s) + { + *result = s; + return 1; + } + return 0; +} + +uword +unformat_transport_connection (unformat_input_t * input, va_list * args) +{ + transport_connection_t **result = va_arg (*args, transport_connection_t **); + u32 suggested_proto = va_arg (*args, u32); + transport_connection_t *tc; + u8 proto = ~0; + ip46_address_t lcl, rmt; + u32 lcl_port = 0, rmt_port = 0; + u8 is_ip4 = 0, s_type = ~0; + + if (!unformat (input, "%U", unformat_stream_session_id, &proto, &lcl, &rmt, + &lcl_port, &rmt_port, &is_ip4)) + return 0; + + proto = (proto == (u8) ~ 0) ? suggested_proto : proto; + if (proto == (u8) ~ 0) + return 0; + s_type = session_type_from_proto_and_ip (proto, is_ip4); + if (is_ip4) + tc = stream_session_lookup_transport4 (&lcl.ip4, &rmt.ip4, + clib_host_to_net_u16 (lcl_port), + clib_host_to_net_u16 (rmt_port), + s_type); + else + tc = stream_session_lookup_transport6 (&lcl.ip6, &rmt.ip6, + clib_host_to_net_u16 (lcl_port), + clib_host_to_net_u16 (rmt_port), + s_type); + + if (tc) + { + *result = tc; + return 1; + } + return 0; +} + +static clib_error_t * +show_session_command_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + session_manager_main_t *smm = &session_manager_main; + int verbose = 0, i; + stream_session_t *pool; + stream_session_t *s; + u8 *str = 0, one_session = 0; + + if (!smm->is_enabled) + { + return clib_error_return (0, "session layer is not enabled"); + } + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "verbose %d", &verbose)) + ; + else if (unformat (input, "verbose")) + verbose = 1; + else if (unformat (input, "%U", unformat_stream_session, &s)) + { + one_session = 1; + } + else + return clib_error_return (0, "unknown input `%U'", + format_unformat_error, input); + } + + if (one_session) + { + vlib_cli_output (vm, "%U", format_stream_session, s, 3); + return 0; + } + + for (i = 0; i < vec_len (smm->sessions); i++) + { + u32 once_per_pool; + pool = smm->sessions[i]; + + once_per_pool = 1; + + if (pool_elts (pool)) + { + + vlib_cli_output (vm, "Thread %d: %d active sessions", + i, pool_elts (pool)); + if (verbose) + { + if (once_per_pool && verbose == 1) + { + str = format (str, "%-50s%-15s%-10s%-10s%-10s", + "Connection", "State", "Rx-f", "Tx-f", + "S-idx"); + vlib_cli_output (vm, "%v", str); + vec_reset_length (str); + once_per_pool = 0; + } + + /* *INDENT-OFF* */ + pool_foreach (s, pool, + ({ + vec_reset_length (str); + str = format (str, "%U", format_stream_session, s, verbose); + vlib_cli_output (vm, "%v", str); + })); + /* *INDENT-ON* */ + } + } + else + vlib_cli_output (vm, "Thread %d: no active sessions", i); + vec_reset_length (str); + } + vec_free (str); + + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (vlib_cli_show_session_command) = +{ + .path = "show session", + .short_help = "show session [verbose [nnn]]", + .function = show_session_command_fn, +}; +/* *INDENT-ON* */ + +static int +clear_session (stream_session_t * s) +{ + application_t *server = application_get (s->app_index); + server->cb_fns.session_disconnect_callback (s); + return 0; +} + +static clib_error_t * +clear_session_command_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + session_manager_main_t *smm = &session_manager_main; + u32 thread_index = 0, clear_all = 0; + u32 session_index = ~0; + stream_session_t **pool, *session; + + if (!smm->is_enabled) + { + return clib_error_return (0, "session layer is not enabled"); + } + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "thread %d", &thread_index)) + ; + else if (unformat (input, "session %d", &session_index)) + ; + else if (unformat (input, "all")) + clear_all = 1; + else + return clib_error_return (0, "unknown input `%U'", + format_unformat_error, input); + } + + if (!clear_all && session_index == ~0) + return clib_error_return (0, "session <nn> required, but not set."); + + if (session_index != ~0) + { + session = stream_session_get_if_valid (session_index, thread_index); + if (!session) + return clib_error_return (0, "no session %d on thread %d", + session_index, thread_index); + clear_session (session); + } + + if (clear_all) + { + /* *INDENT-OFF* */ + vec_foreach (pool, smm->sessions) + { + pool_foreach(session, *pool, ({ + clear_session (session); + })); + }; + /* *INDENT-ON* */ + } + + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (clear_session_command, static) = +{ + .path = "clear session", + .short_help = "clear session thread <thread> session <index>", + .function = clear_session_command_fn, +}; +/* *INDENT-ON* */ + +static clib_error_t * +show_session_fifo_trace_command_fn (vlib_main_t * vm, + unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + stream_session_t *s = 0; + u8 is_rx = 0, *str = 0; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "%U", unformat_stream_session, &s)) + ; + else if (unformat (input, "rx")) + is_rx = 1; + else if (unformat (input, "tx")) + is_rx = 0; + else + return clib_error_return (0, "unknown input `%U'", + format_unformat_error, input); + } + + if (!SVM_FIFO_TRACE) + { + vlib_cli_output (vm, "fifo tracing not enabled"); + return 0; + } + + if (!s) + { + vlib_cli_output (vm, "could not find session"); + return 0; + } + + str = is_rx ? + svm_fifo_dump_trace (str, s->server_rx_fifo) : + svm_fifo_dump_trace (str, s->server_tx_fifo); + + vlib_cli_output (vm, "%v", str); + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (show_session_fifo_trace_command, static) = +{ + .path = "show session fifo trace", + .short_help = "show session fifo trace <session>", + .function = show_session_fifo_trace_command_fn, +}; +/* *INDENT-ON* */ + +static clib_error_t * +session_replay_fifo_command_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + stream_session_t *s = 0; + u8 is_rx = 0, *str = 0; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "%U", unformat_stream_session, &s)) + ; + else if (unformat (input, "rx")) + is_rx = 1; + else + return clib_error_return (0, "unknown input `%U'", + format_unformat_error, input); + } + + if (!SVM_FIFO_TRACE) + { + vlib_cli_output (vm, "fifo tracing not enabled"); + return 0; + } + + if (!s) + { + vlib_cli_output (vm, "could not find session"); + return 0; + } + + str = is_rx ? + svm_fifo_replay (str, s->server_rx_fifo, 0, 1) : + svm_fifo_replay (str, s->server_tx_fifo, 0, 1); + + vlib_cli_output (vm, "%v", str); + return 0; +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (session_replay_fifo_trace_command, static) = +{ + .path = "session replay fifo", + .short_help = "session replay fifo <session>", + .function = session_replay_fifo_command_fn, +}; +/* *INDENT-ON* */ + +static clib_error_t * +session_enable_disable_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + u8 is_en = 1; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "enable")) + is_en = 1; + else if (unformat (input, "disable")) + is_en = 0; + else + return clib_error_return (0, "unknown input `%U'", + format_unformat_error, input); + } + + return vnet_session_enable_disable (vm, is_en); +} + +/* *INDENT-OFF* */ +VLIB_CLI_COMMAND (session_enable_disable_command, static) = +{ + .path = "session", + .short_help = "session [enable|disable]", + .function = session_enable_disable_fn, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session_debug.h b/src/vnet/session/session_debug.h new file mode 100644 index 00000000..eb11f1a0 --- /dev/null +++ b/src/vnet/session/session_debug.h @@ -0,0 +1,142 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef SRC_VNET_SESSION_SESSION_DEBUG_H_ +#define SRC_VNET_SESSION_SESSION_DEBUG_H_ + +#include <vnet/session/transport.h> +#include <vlib/vlib.h> + +#define foreach_session_dbg_evt \ + _(ENQ, "enqueue") \ + _(DEQ, "dequeue") \ + _(DEQ_NODE, "dequeue") \ + _(POLL_GAP_TRACK, "poll gap track") \ + +typedef enum _session_evt_dbg +{ +#define _(sym, str) SESSION_EVT_##sym, + foreach_session_dbg_evt +#undef _ +} session_evt_dbg_e; + +#define SESSION_DBG (0) +#define SESSION_DEQ_NODE_EVTS (0) +#define SESSION_EVT_POLL_DBG (1) + +#if TRANSPORT_DEBUG && SESSION_DBG + +#define DEC_SESSION_ETD(_s, _e, _size) \ + struct \ + { \ + u32 data[_size]; \ + } * ed; \ + transport_proto_vft_t *vft = \ + session_get_transport_vft (_s->session_type); \ + transport_connection_t *_tc = \ + vft->get_connection (_s->connection_index, _s->thread_index); \ + ed = ELOG_TRACK_DATA (&vlib_global_main.elog_main, \ + _e, _tc->elog_track) + +#define DEC_SESSION_ED(_e, _size) \ + struct \ + { \ + u32 data[_size]; \ + } * ed; \ + ed = ELOG_DATA (&vlib_global_main.elog_main, _e) + +#define SESSION_EVT_DEQ_HANDLER(_s, _body) \ +{ \ + ELOG_TYPE_DECLARE (_e) = \ + { \ + .format = "deq: id %d len %d rd %d wnd %d", \ + .format_args = "i4i4i4i4", \ + }; \ + DEC_SESSION_ETD(_s, _e, 4); \ + do { _body; } while (0); \ +} + +#define SESSION_EVT_ENQ_HANDLER(_s, _body) \ +{ \ + ELOG_TYPE_DECLARE (_e) = \ + { \ + .format = "enq: id %d length %d", \ + .format_args = "i4i4", \ + }; \ + DEC_SESSION_ETD(_s, _e, 2); \ + do { _body; } while (0); \ +} + +#if SESSION_DEQ_NODE_EVTS +#define SESSION_EVT_DEQ_NODE_HANDLER(_node_evt) \ +{ \ + ELOG_TYPE_DECLARE (_e) = \ + { \ + .format = "deq-node: %s", \ + .format_args = "t4", \ + .n_enum_strings = 2, \ + .enum_strings = { \ + "start", \ + "end", \ + }, \ + }; \ + DEC_SESSION_ED(_e, 1); \ + ed->data[0] = _node_evt; \ +} +#else +#define SESSION_EVT_DEQ_NODE_HANDLER(_node_evt) +#endif + +#if SESSION_DBG && SESSION_EVT_POLL_DBG +#define SESSION_EVT_POLL_GAP(_smm, _my_thread_index) \ +{ \ + ELOG_TYPE_DECLARE (_e) = \ + { \ + .format = "nixon-gap: %d MS", \ + .format_args = "i4", \ + }; \ + DEC_SESSION_ED(_e, 1); \ + ed->data[0] = (u32) ((now - \ + _smm->last_event_poll_by_thread[my_thread_index])*1000.0); \ +} +#define SESSION_EVT_POLL_GAP_TRACK_HANDLER(_smm, _my_thread_index) \ +{ \ + if (PREDICT_TRUE( \ + smm->last_event_poll_by_thread[my_thread_index] != 0.0)) \ + if (now > smm->last_event_poll_by_thread[_my_thread_index] + 500e-6)\ + SESSION_EVT_POLL_GAP(smm, my_thread_index); \ + _smm->last_event_poll_by_thread[my_thread_index] = now; \ +} + +#else +#define SESSION_EVT_POLL_GAP(_smm, _my_thread_index) +#define SESSION_EVT_POLL_GAP_TRACK_HANDLER(_smm, _my_thread_index) +#endif + +#define CONCAT_HELPER(_a, _b) _a##_b +#define CC(_a, _b) CONCAT_HELPER(_a, _b) +#define SESSION_EVT_DBG(_evt, _args...) CC(_evt, _HANDLER)(_args) + +#else +#define SESSION_EVT_DBG(_evt, _args...) +#endif + +#endif /* SRC_VNET_SESSION_SESSION_DEBUG_H_ */ +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session_lookup.c b/src/vnet/session/session_lookup.c new file mode 100644 index 00000000..4487b1c3 --- /dev/null +++ b/src/vnet/session/session_lookup.c @@ -0,0 +1,619 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/** Generate typed init functions for multiple hash table styles... */ +#include <vppinfra/bihash_16_8.h> +#include <vppinfra/bihash_template.h> + +#include <vppinfra/bihash_template.c> + +#undef __included_bihash_template_h__ + +#include <vppinfra/bihash_48_8.h> +#include <vppinfra/bihash_template.h> + +#include <vppinfra/bihash_template.c> +#include <vnet/session/session_lookup.h> +#include <vnet/session/session.h> + +static session_lookup_t session_lookup; +extern transport_proto_vft_t *tp_vfts; + +/* *INDENT-OFF* */ +/* 16 octets */ +typedef CLIB_PACKED (struct { + union + { + struct + { + ip4_address_t src; + ip4_address_t dst; + u16 src_port; + u16 dst_port; + /* align by making this 4 octets even though its a 1-bit field + * NOTE: avoid key overlap with other transports that use 5 tuples for + * session identification. + */ + u32 proto; + }; + u64 as_u64[2]; + }; +}) v4_connection_key_t; + +typedef CLIB_PACKED (struct { + union + { + struct + { + /* 48 octets */ + ip6_address_t src; + ip6_address_t dst; + u16 src_port; + u16 dst_port; + u32 proto; + u64 unused; + }; + u64 as_u64[6]; + }; +}) v6_connection_key_t; +/* *INDENT-ON* */ + +typedef clib_bihash_kv_16_8_t session_kv4_t; +typedef clib_bihash_kv_48_8_t session_kv6_t; + +always_inline void +make_v4_ss_kv (session_kv4_t * kv, ip4_address_t * lcl, ip4_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto) +{ + v4_connection_key_t *key = (v4_connection_key_t *) kv->key; + + key->src.as_u32 = lcl->as_u32; + key->dst.as_u32 = rmt->as_u32; + key->src_port = lcl_port; + key->dst_port = rmt_port; + key->proto = proto; + + kv->value = ~0ULL; +} + +always_inline void +make_v4_listener_kv (session_kv4_t * kv, ip4_address_t * lcl, u16 lcl_port, + u8 proto) +{ + v4_connection_key_t *key = (v4_connection_key_t *) kv->key; + + key->src.as_u32 = lcl->as_u32; + key->dst.as_u32 = 0; + key->src_port = lcl_port; + key->dst_port = 0; + key->proto = proto; + + kv->value = ~0ULL; +} + +always_inline void +make_v4_ss_kv_from_tc (session_kv4_t * kv, transport_connection_t * t) +{ + make_v4_ss_kv (kv, &t->lcl_ip.ip4, &t->rmt_ip.ip4, t->lcl_port, t->rmt_port, + session_type_from_proto_and_ip (t->transport_proto, 1)); +} + +always_inline void +make_v6_ss_kv (session_kv6_t * kv, ip6_address_t * lcl, ip6_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto) +{ + v6_connection_key_t *key = (v6_connection_key_t *) kv->key; + + key->src.as_u64[0] = lcl->as_u64[0]; + key->src.as_u64[1] = lcl->as_u64[1]; + key->dst.as_u64[0] = rmt->as_u64[0]; + key->dst.as_u64[1] = rmt->as_u64[1]; + key->src_port = lcl_port; + key->dst_port = rmt_port; + key->proto = proto; + key->unused = 0; + + kv->value = ~0ULL; +} + +always_inline void +make_v6_listener_kv (session_kv6_t * kv, ip6_address_t * lcl, u16 lcl_port, + u8 proto) +{ + v6_connection_key_t *key = (v6_connection_key_t *) kv->key; + + key->src.as_u64[0] = lcl->as_u64[0]; + key->src.as_u64[1] = lcl->as_u64[1]; + key->dst.as_u64[0] = 0; + key->dst.as_u64[1] = 0; + key->src_port = lcl_port; + key->dst_port = 0; + key->proto = proto; + key->unused = 0; + + kv->value = ~0ULL; +} + +always_inline void +make_v6_ss_kv_from_tc (session_kv6_t * kv, transport_connection_t * t) +{ + make_v6_ss_kv (kv, &t->lcl_ip.ip6, &t->rmt_ip.ip6, t->lcl_port, t->rmt_port, + session_type_from_proto_and_ip (t->transport_proto, 0)); +} + +/* + * Session lookup key; (src-ip, dst-ip, src-port, dst-port, session-type) + * Value: (owner thread index << 32 | session_index); + */ +void +stream_session_table_add_for_tc (transport_connection_t * tc, u64 value) +{ + session_lookup_t *sl = &session_lookup; + session_kv4_t kv4; + session_kv6_t kv6; + + if (tc->is_ip4) + { + make_v4_ss_kv_from_tc (&kv4, tc); + kv4.value = value; + clib_bihash_add_del_16_8 (&sl->v4_session_hash, &kv4, 1 /* is_add */ ); + } + else + { + make_v6_ss_kv_from_tc (&kv6, tc); + kv6.value = value; + clib_bihash_add_del_48_8 (&sl->v6_session_hash, &kv6, 1 /* is_add */ ); + } +} + +void +stream_session_table_add (session_manager_main_t * smm, stream_session_t * s, + u64 value) +{ + transport_connection_t *tc; + + tc = tp_vfts[s->session_type].get_connection (s->connection_index, + s->thread_index); + stream_session_table_add_for_tc (tc, value); +} + +int +stream_session_table_del_for_tc (transport_connection_t * tc) +{ + session_lookup_t *sl = &session_lookup; + session_kv4_t kv4; + session_kv6_t kv6; + + if (tc->is_ip4) + { + make_v4_ss_kv_from_tc (&kv4, tc); + return clib_bihash_add_del_16_8 (&sl->v4_session_hash, &kv4, + 0 /* is_add */ ); + } + else + { + make_v6_ss_kv_from_tc (&kv6, tc); + return clib_bihash_add_del_48_8 (&sl->v6_session_hash, &kv6, + 0 /* is_add */ ); + } + + return 0; +} + +int +stream_session_table_del (stream_session_t * s) +{ + transport_connection_t *ts; + ts = tp_vfts[s->session_type].get_connection (s->connection_index, + s->thread_index); + return stream_session_table_del_for_tc (ts); +} + + +void +stream_session_half_open_table_add (transport_connection_t * tc, u64 value) +{ + session_lookup_t *sl = &session_lookup; + session_kv4_t kv4; + session_kv6_t kv6; + + if (tc->is_ip4) + { + make_v4_ss_kv_from_tc (&kv4, tc); + kv4.value = value; + (void) clib_bihash_add_del_16_8 (&sl->v4_half_open_hash, &kv4, + 1 /* is_add */ ); + } + else + { + make_v6_ss_kv_from_tc (&kv6, tc); + kv6.value = value; + (void) clib_bihash_add_del_48_8 (&sl->v6_half_open_hash, &kv6, + 1 /* is_add */ ); + } +} + +void +stream_session_half_open_table_del (transport_connection_t * tc) +{ + session_lookup_t *sl = &session_lookup; + session_kv4_t kv4; + session_kv6_t kv6; + + if (tc->is_ip4) + { + make_v4_ss_kv_from_tc (&kv4, tc); + clib_bihash_add_del_16_8 (&sl->v4_half_open_hash, &kv4, + 0 /* is_add */ ); + } + else + { + make_v6_ss_kv_from_tc (&kv6, tc); + clib_bihash_add_del_48_8 (&sl->v6_half_open_hash, &kv6, + 0 /* is_add */ ); + } +} + +stream_session_t * +stream_session_lookup_listener4 (ip4_address_t * lcl, u16 lcl_port, u8 proto) +{ + session_lookup_t *sl = &session_lookup; + session_kv4_t kv4; + int rv; + + make_v4_listener_kv (&kv4, lcl, lcl_port, proto); + rv = clib_bihash_search_inline_16_8 (&sl->v4_session_hash, &kv4); + if (rv == 0) + return session_manager_get_listener (proto, (u32) kv4.value); + + /* Zero out the lcl ip */ + kv4.key[0] = 0; + rv = clib_bihash_search_inline_16_8 (&sl->v4_session_hash, &kv4); + if (rv == 0) + return session_manager_get_listener (proto, (u32) kv4.value); + + return 0; +} + +/** Looks up a session based on the 5-tuple passed as argument. + * + * First it tries to find an established session, if this fails, it tries + * finding a listener session if this fails, it tries a lookup with a + * wildcarded local source (listener bound to all interfaces) + */ +stream_session_t * +stream_session_lookup4 (ip4_address_t * lcl, ip4_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto) +{ + session_lookup_t *sl = &session_lookup; + session_kv4_t kv4; + stream_session_t *s; + int rv; + + /* Lookup session amongst established ones */ + make_v4_ss_kv (&kv4, lcl, rmt, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_16_8 (&sl->v4_session_hash, &kv4); + if (rv == 0) + return stream_session_get_from_handle (kv4.value); + + /* If nothing is found, check if any listener is available */ + if ((s = stream_session_lookup_listener4 (lcl, lcl_port, proto))) + return s; + + /* Finally, try half-open connections */ + rv = clib_bihash_search_inline_16_8 (&sl->v4_half_open_hash, &kv4); + if (rv == 0) + return stream_session_get_from_handle (kv4.value); + return 0; +} + +stream_session_t * +stream_session_lookup_listener6 (ip6_address_t * lcl, u16 lcl_port, u8 proto) +{ + session_lookup_t *sl = &session_lookup; + session_kv6_t kv6; + int rv; + + make_v6_listener_kv (&kv6, lcl, lcl_port, proto); + rv = clib_bihash_search_inline_48_8 (&sl->v6_session_hash, &kv6); + if (rv == 0) + return session_manager_get_listener (proto, (u32) kv6.value); + + /* Zero out the lcl ip */ + kv6.key[0] = kv6.key[1] = 0; + rv = clib_bihash_search_inline_48_8 (&sl->v6_session_hash, &kv6); + if (rv == 0) + return session_manager_get_listener (proto, (u32) kv6.value); + + return 0; +} + +/* Looks up a session based on the 5-tuple passed as argument. + * First it tries to find an established session, if this fails, it tries + * finding a listener session if this fails, it tries a lookup with a + * wildcarded local source (listener bound to all interfaces) */ +stream_session_t * +stream_session_lookup6 (ip6_address_t * lcl, ip6_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto) +{ + session_lookup_t *sl = &session_lookup; + session_kv6_t kv6; + stream_session_t *s; + int rv; + + make_v6_ss_kv (&kv6, lcl, rmt, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_48_8 (&sl->v6_session_hash, &kv6); + if (rv == 0) + return stream_session_get_from_handle (kv6.value); + + /* If nothing is found, check if any listener is available */ + if ((s = stream_session_lookup_listener6 (lcl, lcl_port, proto))) + return s; + + /* Finally, try half-open connections */ + rv = clib_bihash_search_inline_48_8 (&sl->v6_half_open_hash, &kv6); + if (rv == 0) + return stream_session_get_from_handle (kv6.value); + return 0; +} + +stream_session_t * +stream_session_lookup_listener (ip46_address_t * lcl, u16 lcl_port, u8 proto) +{ + switch (proto) + { + case SESSION_TYPE_IP4_UDP: + case SESSION_TYPE_IP4_TCP: + return stream_session_lookup_listener4 (&lcl->ip4, lcl_port, proto); + break; + case SESSION_TYPE_IP6_UDP: + case SESSION_TYPE_IP6_TCP: + return stream_session_lookup_listener6 (&lcl->ip6, lcl_port, proto); + break; + } + return 0; +} + +u64 +stream_session_half_open_lookup_handle (ip46_address_t * lcl, + ip46_address_t * rmt, u16 lcl_port, + u16 rmt_port, u8 proto) +{ + session_lookup_t *sl = &session_lookup; + session_kv4_t kv4; + session_kv6_t kv6; + int rv; + + switch (proto) + { + case SESSION_TYPE_IP4_UDP: + case SESSION_TYPE_IP4_TCP: + make_v4_ss_kv (&kv4, &lcl->ip4, &rmt->ip4, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_16_8 (&sl->v4_half_open_hash, &kv4); + + if (rv == 0) + return kv4.value; + + return HALF_OPEN_LOOKUP_INVALID_VALUE; + break; + case SESSION_TYPE_IP6_UDP: + case SESSION_TYPE_IP6_TCP: + make_v6_ss_kv (&kv6, &lcl->ip6, &rmt->ip6, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_48_8 (&sl->v6_half_open_hash, &kv6); + + if (rv == 0) + return kv6.value; + + return HALF_OPEN_LOOKUP_INVALID_VALUE; + break; + } + return HALF_OPEN_LOOKUP_INVALID_VALUE; +} + +transport_connection_t * +stream_session_half_open_lookup (ip46_address_t * lcl, ip46_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto) +{ + u64 handle; + handle = + stream_session_half_open_lookup_handle (lcl, rmt, lcl_port, rmt_port, + proto); + if (handle != HALF_OPEN_LOOKUP_INVALID_VALUE) + return tp_vfts[proto].get_half_open (handle & 0xFFFFFFFF); + return 0; +} + +always_inline stream_session_t * +stream_session_get_tsi (u64 ti_and_si, u32 thread_index) +{ + ASSERT ((u32) (ti_and_si >> 32) == thread_index); + return pool_elt_at_index (session_manager_main.sessions[thread_index], + ti_and_si & 0xFFFFFFFFULL); +} + +transport_connection_t * +stream_session_lookup_transport_wt4 (ip4_address_t * lcl, ip4_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto, + u32 my_thread_index) +{ + session_lookup_t *sl = &session_lookup; + session_kv4_t kv4; + stream_session_t *s; + int rv; + + /* Lookup session amongst established ones */ + make_v4_ss_kv (&kv4, lcl, rmt, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_16_8 (&sl->v4_session_hash, &kv4); + if (rv == 0) + { + s = stream_session_get_tsi (kv4.value, my_thread_index); + return tp_vfts[s->session_type].get_connection (s->connection_index, + my_thread_index); + } + + /* If nothing is found, check if any listener is available */ + s = stream_session_lookup_listener4 (lcl, lcl_port, proto); + if (s) + return tp_vfts[s->session_type].get_listener (s->connection_index); + + /* Finally, try half-open connections */ + rv = clib_bihash_search_inline_16_8 (&sl->v4_half_open_hash, &kv4); + if (rv == 0) + return tp_vfts[proto].get_half_open (kv4.value & 0xFFFFFFFF); + return 0; +} + +transport_connection_t * +stream_session_lookup_transport4 (ip4_address_t * lcl, ip4_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto) +{ + session_lookup_t *sl = &session_lookup; + session_kv4_t kv4; + stream_session_t *s; + int rv; + + /* Lookup session amongst established ones */ + make_v4_ss_kv (&kv4, lcl, rmt, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_16_8 (&sl->v4_session_hash, &kv4); + if (rv == 0) + { + s = stream_session_get_from_handle (kv4.value); + return tp_vfts[s->session_type].get_connection (s->connection_index, + s->thread_index); + } + + /* If nothing is found, check if any listener is available */ + s = stream_session_lookup_listener4 (lcl, lcl_port, proto); + if (s) + return tp_vfts[s->session_type].get_listener (s->connection_index); + + /* Finally, try half-open connections */ + rv = clib_bihash_search_inline_16_8 (&sl->v4_half_open_hash, &kv4); + if (rv == 0) + return tp_vfts[proto].get_half_open (kv4.value & 0xFFFFFFFF); + return 0; +} + +transport_connection_t * +stream_session_lookup_transport_wt6 (ip6_address_t * lcl, ip6_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto, + u32 my_thread_index) +{ + session_lookup_t *sl = &session_lookup; + stream_session_t *s; + session_kv6_t kv6; + int rv; + + make_v6_ss_kv (&kv6, lcl, rmt, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_48_8 (&sl->v6_session_hash, &kv6); + if (rv == 0) + { + s = stream_session_get_tsi (kv6.value, my_thread_index); + return tp_vfts[s->session_type].get_connection (s->connection_index, + my_thread_index); + } + + /* If nothing is found, check if any listener is available */ + s = stream_session_lookup_listener6 (lcl, lcl_port, proto); + if (s) + return tp_vfts[s->session_type].get_listener (s->connection_index); + + /* Finally, try half-open connections */ + rv = clib_bihash_search_inline_48_8 (&sl->v6_half_open_hash, &kv6); + if (rv == 0) + return tp_vfts[proto].get_half_open (kv6.value & 0xFFFFFFFF); + + return 0; +} + +transport_connection_t * +stream_session_lookup_transport6 (ip6_address_t * lcl, ip6_address_t * rmt, + u16 lcl_port, u16 rmt_port, u8 proto) +{ + session_lookup_t *sl = &session_lookup; + stream_session_t *s; + session_kv6_t kv6; + int rv; + + make_v6_ss_kv (&kv6, lcl, rmt, lcl_port, rmt_port, proto); + rv = clib_bihash_search_inline_48_8 (&sl->v6_session_hash, &kv6); + if (rv == 0) + { + s = stream_session_get_from_handle (kv6.value); + return tp_vfts[s->session_type].get_connection (s->connection_index, + s->thread_index); + } + + /* If nothing is found, check if any listener is available */ + s = stream_session_lookup_listener6 (lcl, lcl_port, proto); + if (s) + return tp_vfts[s->session_type].get_listener (s->connection_index); + + /* Finally, try half-open connections */ + rv = clib_bihash_search_inline_48_8 (&sl->v6_half_open_hash, &kv6); + if (rv == 0) + return tp_vfts[proto].get_half_open (kv6.value & 0xFFFFFFFF); + + return 0; +} + +#define foreach_hash_table_parameter \ + _(v4,session,buckets,20000) \ + _(v4,session,memory,(64<<20)) \ + _(v6,session,buckets,20000) \ + _(v6,session,memory,(64<<20)) \ + _(v4,halfopen,buckets,20000) \ + _(v4,halfopen,memory,(64<<20)) \ + _(v6,halfopen,buckets,20000) \ + _(v6,halfopen,memory,(64<<20)) + +void +session_lookup_init (void) +{ + session_lookup_t *sl = &session_lookup; + +#define _(af,table,parm,value) \ + u32 configured_##af##_##table##_table_##parm = value; + foreach_hash_table_parameter; +#undef _ + +#define _(af,table,parm,value) \ + if (session_manager_main.configured_##af##_##table##_table_##parm) \ + configured_##af##_##table##_table_##parm = \ + session_manager_main.configured_##af##_##table##_table_##parm; + foreach_hash_table_parameter; +#undef _ + + clib_bihash_init_16_8 (&sl->v4_session_hash, "v4 session table", + configured_v4_session_table_buckets, + configured_v4_session_table_memory); + clib_bihash_init_48_8 (&sl->v6_session_hash, "v6 session table", + configured_v6_session_table_buckets, + configured_v6_session_table_memory); + clib_bihash_init_16_8 (&sl->v4_half_open_hash, "v4 half-open table", + configured_v4_halfopen_table_buckets, + configured_v4_halfopen_table_memory); + clib_bihash_init_48_8 (&sl->v6_half_open_hash, "v6 half-open table", + configured_v6_halfopen_table_buckets, + configured_v6_halfopen_table_memory); +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session_lookup.h b/src/vnet/session/session_lookup.h new file mode 100644 index 00000000..cf1dc013 --- /dev/null +++ b/src/vnet/session/session_lookup.h @@ -0,0 +1,100 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef SRC_VNET_SESSION_SESSION_LOOKUP_H_ +#define SRC_VNET_SESSION_SESSION_LOOKUP_H_ + +#include <vnet/session/stream_session.h> +#include <vnet/session/transport.h> + +typedef struct _session_lookup +{ + /** Lookup tables for established sessions and listeners */ + clib_bihash_16_8_t v4_session_hash; + clib_bihash_48_8_t v6_session_hash; + + /** Lookup tables for half-open sessions */ + clib_bihash_16_8_t v4_half_open_hash; + clib_bihash_48_8_t v6_half_open_hash; +} session_lookup_t; + +stream_session_t *stream_session_lookup_listener4 (ip4_address_t * lcl, + u16 lcl_port, u8 proto); +stream_session_t *stream_session_lookup4 (ip4_address_t * lcl, + ip4_address_t * rmt, u16 lcl_port, + u16 rmt_port, u8 proto); +stream_session_t *stream_session_lookup_listener6 (ip6_address_t * lcl, + u16 lcl_port, u8 proto); +stream_session_t *stream_session_lookup6 (ip6_address_t * lcl, + ip6_address_t * rmt, u16 lcl_port, + u16 rmt_port, u8 proto); +transport_connection_t *stream_session_lookup_transport_wt4 (ip4_address_t * + lcl, + ip4_address_t * + rmt, + u16 lcl_port, + u16 rmt_port, + u8 proto, + u32 + thread_index); +transport_connection_t *stream_session_lookup_transport4 (ip4_address_t * lcl, + ip4_address_t * rmt, + u16 lcl_port, + u16 rmt_port, + u8 proto); +transport_connection_t *stream_session_lookup_transport_wt6 (ip6_address_t * + lcl, + ip6_address_t * + rmt, + u16 lcl_port, + u16 rmt_port, + u8 proto, + u32 + thread_index); +transport_connection_t *stream_session_lookup_transport6 (ip6_address_t * lcl, + ip6_address_t * rmt, + u16 lcl_port, + u16 rmt_port, + u8 proto); + +stream_session_t *stream_session_lookup_listener (ip46_address_t * lcl, + u16 lcl_port, u8 proto); +u64 stream_session_half_open_lookup_handle (ip46_address_t * lcl, + ip46_address_t * rmt, + u16 lcl_port, + u16 rmt_port, u8 proto); +transport_connection_t *stream_session_half_open_lookup (ip46_address_t * lcl, + ip46_address_t * rmt, + u16 lcl_port, + u16 rmt_port, + u8 proto); +void stream_session_table_add_for_tc (transport_connection_t * tc, u64 value); +int stream_session_table_del_for_tc (transport_connection_t * tc); +int stream_session_table_del (stream_session_t * s); +void stream_session_half_open_table_del (transport_connection_t * tc); +void stream_session_half_open_table_add (transport_connection_t * tc, + u64 value); + +void session_lookup_init (void); + +#endif /* SRC_VNET_SESSION_SESSION_LOOKUP_H_ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/session_node.c b/src/vnet/session/session_node.c new file mode 100644 index 00000000..d0155849 --- /dev/null +++ b/src/vnet/session/session_node.c @@ -0,0 +1,707 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <math.h> +#include <vlib/vlib.h> +#include <vnet/vnet.h> +#include <vnet/tcp/tcp.h> +#include <vppinfra/elog.h> +#include <vnet/session/application.h> +#include <vnet/session/session_debug.h> +#include <vlibmemory/unix_shared_memory_queue.h> + +vlib_node_registration_t session_queue_node; + +typedef struct +{ + u32 session_index; + u32 server_thread_index; +} session_queue_trace_t; + +/* packet trace format function */ +static u8 * +format_session_queue_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 *); + session_queue_trace_t *t = va_arg (*args, session_queue_trace_t *); + + s = format (s, "SESSION_QUEUE: session index %d, server thread index %d", + t->session_index, t->server_thread_index); + return s; +} + +vlib_node_registration_t session_queue_node; + +#define foreach_session_queue_error \ +_(TX, "Packets transmitted") \ +_(TIMER, "Timer events") \ +_(NO_BUFFER, "Out of buffers") + +typedef enum +{ +#define _(sym,str) SESSION_QUEUE_ERROR_##sym, + foreach_session_queue_error +#undef _ + SESSION_QUEUE_N_ERROR, +} session_queue_error_t; + +static char *session_queue_error_strings[] = { +#define _(sym,string) string, + foreach_session_queue_error +#undef _ +}; + +static u32 session_type_to_next[] = { + SESSION_QUEUE_NEXT_TCP_IP4_OUTPUT, + SESSION_QUEUE_NEXT_IP4_LOOKUP, + SESSION_QUEUE_NEXT_TCP_IP6_OUTPUT, + SESSION_QUEUE_NEXT_IP6_LOOKUP, +}; + +always_inline void +session_tx_fifo_chain_tail (session_manager_main_t * smm, vlib_main_t * vm, + u8 thread_index, svm_fifo_t * fifo, + 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) +{ + vlib_buffer_t *chain_b0, *prev_b0; + u32 chain_bi0, to_deq; + u16 len_to_deq0, n_bytes_read; + u8 *data0, j; + + b0->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID; + b0->total_length_not_including_first_buffer = 0; + + chain_bi0 = bi0; + chain_b0 = b0; + to_deq = left_from_seg; + for (j = 1; j < n_bufs_per_seg; j++) + { + prev_b0 = chain_b0; + len_to_deq0 = clib_min (to_deq, deq_per_buf); + + *n_bufs -= 1; + chain_bi0 = smm->tx_buffers[thread_index][*n_bufs]; + _vec_len (smm->tx_buffers[thread_index]) = *n_bufs; + + chain_b0 = vlib_get_buffer (vm, chain_bi0); + chain_b0->current_data = 0; + data0 = vlib_buffer_get_current (chain_b0); + if (peek_data) + { + n_bytes_read = svm_fifo_peek (fifo, *tx_offset, len_to_deq0, data0); + *tx_offset += n_bytes_read; + } + 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; + b0->total_length_not_including_first_buffer += chain_b0->current_length; + + /* update previous buffer */ + prev_b0->next_buffer = chain_bi0; + prev_b0->flags |= VLIB_BUFFER_NEXT_PRESENT; + + /* update current buffer */ + chain_b0->next_buffer = 0; + + to_deq -= n_bytes_read; + if (to_deq == 0) + break; + } + ASSERT (to_deq == 0 + && b0->total_length_not_including_first_buffer == left_from_seg); + *left_to_snd0 -= left_from_seg; +} + +always_inline int +session_tx_fifo_read_and_snd_i (vlib_main_t * vm, vlib_node_runtime_t * node, + session_manager_main_t * smm, + session_fifo_event_t * e0, + stream_session_t * s0, u32 thread_index, + int *n_tx_packets, u8 peek_data) +{ + u32 n_trace = vlib_get_trace_count (vm, node); + u32 left_to_snd0, max_len_to_snd0, len_to_deq0, snd_space0; + u32 n_bufs_per_evt, n_frames_per_evt, n_bufs_per_frame; + transport_connection_t *tc0; + transport_proto_vft_t *transport_vft; + u32 next_index, next0, *to_next, n_left_to_next, bi0; + vlib_buffer_t *b0; + u32 tx_offset = 0, max_dequeue0, n_bytes_per_seg, left_for_seg; + u16 snd_mss0, n_bufs_per_seg, n_bufs; + u8 *data0; + int i, n_bytes_read; + u32 n_bytes_per_buf, deq_per_buf, deq_per_first_buf; + u32 buffers_allocated, buffers_allocated_this_call; + + next_index = next0 = session_type_to_next[s0->session_type]; + + transport_vft = session_get_transport_vft (s0->session_type); + 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); + snd_space0 = transport_vft->send_space (tc0); + + /* Can't make any progress */ + if (snd_space0 == 0 || snd_mss0 == 0) + { + vec_add1 (smm->pending_event_vector[thread_index], *e0); + return 0; + } + + /* 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; + } + + /* Nothing to read return */ + if (max_dequeue0 == 0) + { + svm_fifo_unset_event (s0->server_tx_fifo); + return 0; + } + + /* Ensure we're not writing more than transport window allows */ + if (max_dequeue0 < snd_space0) + { + /* Constrained by tx queue. Try to send only fully formed segments */ + max_len_to_snd0 = (max_dequeue0 > snd_mss0) ? + max_dequeue0 - max_dequeue0 % snd_mss0 : max_dequeue0; + /* TODO Nagle ? */ + } + else + { + /* Expectation is that snd_space0 is already a multiple of snd_mss */ + max_len_to_snd0 = snd_space0; + } + + n_bytes_per_buf = vlib_buffer_free_list_buffer_size + (vm, VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX); + ASSERT (n_bytes_per_buf > MAX_HDRS_LEN); + n_bytes_per_seg = MAX_HDRS_LEN + snd_mss0; + n_bufs_per_seg = ceil ((double) n_bytes_per_seg / n_bytes_per_buf); + n_bufs_per_evt = ceil ((double) max_len_to_snd0 / n_bytes_per_seg); + n_frames_per_evt = ceil ((double) n_bufs_per_evt / VLIB_FRAME_SIZE); + n_bufs_per_frame = n_bufs_per_seg * VLIB_FRAME_SIZE; + + deq_per_buf = clib_min (snd_mss0, n_bytes_per_buf); + deq_per_first_buf = clib_min (snd_mss0, n_bytes_per_buf - MAX_HDRS_LEN); + + n_bufs = vec_len (smm->tx_buffers[thread_index]); + left_to_snd0 = max_len_to_snd0; + for (i = 0; i < n_frames_per_evt; i++) + { + /* Make sure we have at least one full frame of buffers ready */ + if (PREDICT_FALSE (n_bufs < n_bufs_per_frame)) + { + vec_validate (smm->tx_buffers[thread_index], + n_bufs + n_bufs_per_frame - 1); + buffers_allocated = 0; + do + { + buffers_allocated_this_call = vlib_buffer_alloc (vm, + &smm->tx_buffers + [thread_index] + [n_bufs + + buffers_allocated], + n_bufs_per_frame + - + buffers_allocated); + buffers_allocated += buffers_allocated_this_call; + } + while (buffers_allocated_this_call > 0 + && ((buffers_allocated + n_bufs < n_bufs_per_frame))); + + n_bufs += buffers_allocated; + _vec_len (smm->tx_buffers[thread_index]) = n_bufs; + + if (PREDICT_FALSE (n_bufs < n_bufs_per_frame)) + { + vec_add1 (smm->pending_event_vector[thread_index], *e0); + return -1; + } + ASSERT (n_bufs >= n_bufs_per_frame); + } + /* Allow enqueuing of a new event */ + svm_fifo_unset_event (s0->server_tx_fifo); + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + while (left_to_snd0 && n_left_to_next) + { + /* + * Handle first buffer in chain separately + */ + + /* Get free buffer */ + ASSERT (n_bufs >= 1); + bi0 = smm->tx_buffers[thread_index][--n_bufs]; + _vec_len (smm->tx_buffers[thread_index]) = n_bufs; + + /* usual speculation, or the enqueue_x1 macro will barf */ + to_next[0] = bi0; + to_next += 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + b0->error = 0; + b0->flags |= VNET_BUFFER_F_LOCALLY_ORIGINATED; + b0->current_data = 0; + b0->total_length_not_including_first_buffer = 0; + + len_to_deq0 = clib_min (left_to_snd0, deq_per_first_buf); + data0 = vlib_buffer_make_headroom (b0, MAX_HDRS_LEN); + if (peek_data) + { + n_bytes_read = svm_fifo_peek (s0->server_tx_fifo, tx_offset, + len_to_deq0, data0); + if (n_bytes_read <= 0) + goto dequeue_fail; + /* Keep track of progress locally, transport is also supposed to + * increment it independently when pushing the header */ + tx_offset += n_bytes_read; + } + 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; + + /* + * Fill in the remaining buffers in the chain, if any + */ + if (PREDICT_FALSE (n_bufs_per_seg > 1 && left_to_snd0)) + { + left_for_seg = clib_min (snd_mss0 - n_bytes_read, left_to_snd0); + session_tx_fifo_chain_tail (smm, vm, thread_index, + 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); + } + + /* Ask transport to push header after current_length and + * total_length_not_including_first_buffer are updated */ + transport_vft->push_header (tc0, b0); + + /* *INDENT-OFF* */ + SESSION_EVT_DBG(SESSION_EVT_DEQ, s0, ({ + ed->data[0] = e0->event_id; + ed->data[1] = max_dequeue0; + ed->data[2] = len_to_deq0; + ed->data[3] = left_to_snd0; + })); + /* *INDENT-ON* */ + + VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0); + if (VLIB_BUFFER_TRACE_TRAJECTORY) + b0->pre_data[1] = 3; + + if (PREDICT_FALSE (n_trace > 0)) + { + session_queue_trace_t *t0; + vlib_trace_buffer (vm, node, next_index, b0, + 1 /* follow_chain */ ); + vlib_set_trace_count (vm, node, --n_trace); + t0 = vlib_add_trace (vm, node, b0, sizeof (*t0)); + t0->session_index = s0->session_index; + t0->server_thread_index = s0->thread_index; + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + bi0, next0); + } + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + /* If we couldn't dequeue all bytes mark as partially read */ + if (max_len_to_snd0 < max_dequeue0) + { + /* 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); + } + } + return 0; + +dequeue_fail: + /* + * Can't read from fifo. If we don't already have an event, save as partially + * 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); + } + vlib_put_next_frame (vm, node, next_index, n_left_to_next + 1); + _vec_len (smm->tx_buffers[thread_index]) += 1; + + return 0; +} + +int +session_tx_fifo_peek_and_snd (vlib_main_t * vm, vlib_node_runtime_t * node, + session_manager_main_t * smm, + session_fifo_event_t * e0, + stream_session_t * s0, u32 thread_index, + int *n_tx_pkts) +{ + return session_tx_fifo_read_and_snd_i (vm, node, smm, e0, s0, thread_index, + n_tx_pkts, 1); +} + +int +session_tx_fifo_dequeue_and_snd (vlib_main_t * vm, vlib_node_runtime_t * node, + session_manager_main_t * smm, + session_fifo_event_t * e0, + stream_session_t * s0, u32 thread_index, + int *n_tx_pkts) +{ + return session_tx_fifo_read_and_snd_i (vm, node, smm, e0, s0, thread_index, + n_tx_pkts, 0); +} + +always_inline stream_session_t * +session_event_get_session (session_fifo_event_t * e, u8 thread_index) +{ + return stream_session_get_if_valid (e->fifo->master_session_index, + thread_index); +} + +void +dump_thread_0_event_queue (void) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + vlib_main_t *vm = &vlib_global_main; + u32 my_thread_index = vm->thread_index; + session_fifo_event_t _e, *e = &_e; + stream_session_t *s0; + int i, index; + i8 *headp; + + unix_shared_memory_queue_t *q; + q = smm->vpp_event_queues[my_thread_index]; + + index = q->head; + + for (i = 0; i < q->cursize; i++) + { + headp = (i8 *) (&q->data[0] + q->elsize * index); + clib_memcpy (e, headp, q->elsize); + + switch (e->event_type) + { + case FIFO_EVENT_APP_TX: + s0 = session_event_get_session (e, my_thread_index); + fformat (stdout, "[%04d] TX session %d\n", i, s0->session_index); + break; + + case FIFO_EVENT_DISCONNECT: + s0 = stream_session_get_from_handle (e->session_handle); + fformat (stdout, "[%04d] disconnect session %d\n", i, + s0->session_index); + break; + + case FIFO_EVENT_BUILTIN_RX: + s0 = session_event_get_session (e, my_thread_index); + fformat (stdout, "[%04d] builtin_rx %d\n", i, s0->session_index); + break; + + case FIFO_EVENT_RPC: + fformat (stdout, "[%04d] RPC call %llx with %llx\n", + i, (u64) (e->rpc_args.fp), (u64) (e->rpc_args.arg)); + break; + + default: + fformat (stdout, "[%04d] unhandled event type %d\n", + i, e->event_type); + break; + } + + index++; + + if (index == q->maxsize) + index = 0; + } +} + +static u8 +session_node_cmp_event (session_fifo_event_t * e, svm_fifo_t * f) +{ + stream_session_t *s; + switch (e->event_type) + { + case FIFO_EVENT_APP_RX: + case FIFO_EVENT_APP_TX: + case FIFO_EVENT_BUILTIN_RX: + if (e->fifo == f) + return 1; + break; + case FIFO_EVENT_DISCONNECT: + break; + case FIFO_EVENT_RPC: + s = stream_session_get_from_handle (e->session_handle); + if (!s) + { + clib_warning ("session has event but doesn't exist!"); + break; + } + if (s->server_rx_fifo == f || s->server_tx_fifo == f) + return 1; + break; + default: + break; + } + return 0; +} + +u8 +session_node_lookup_fifo_event (svm_fifo_t * f, session_fifo_event_t * e) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + unix_shared_memory_queue_t *q; + session_fifo_event_t *pending_event_vector, *evt; + int i, index, found = 0; + i8 *headp; + u8 thread_index; + + ASSERT (e); + thread_index = f->master_thread_index; + /* + * Search evt queue + */ + q = smm->vpp_event_queues[thread_index]; + index = q->head; + for (i = 0; i < q->cursize; i++) + { + headp = (i8 *) (&q->data[0] + q->elsize * index); + clib_memcpy (e, headp, q->elsize); + found = session_node_cmp_event (e, f); + if (found) + break; + if (++index == q->maxsize) + index = 0; + } + /* + * Search pending events vector + */ + pending_event_vector = smm->pending_event_vector[thread_index]; + vec_foreach (evt, pending_event_vector) + { + found = session_node_cmp_event (evt, f); + if (found) + { + clib_memcpy (e, evt, sizeof (*evt)); + break; + } + } + return found; +} + +static uword +session_queue_node_fn (vlib_main_t * vm, vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + session_manager_main_t *smm = vnet_get_session_manager_main (); + session_fifo_event_t *my_pending_event_vector, *e; + session_fifo_event_t *my_fifo_events; + u32 n_to_dequeue, n_events; + unix_shared_memory_queue_t *q; + application_t *app; + int n_tx_packets = 0; + u32 my_thread_index = vm->thread_index; + int i, rv; + f64 now = vlib_time_now (vm); + void (*fp) (void *); + + SESSION_EVT_DBG (SESSION_EVT_POLL_GAP_TRACK, smm, my_thread_index); + + /* + * Update TCP time + */ + tcp_update_time (now, my_thread_index); + + /* + * Get vpp queue events + */ + q = smm->vpp_event_queues[my_thread_index]; + if (PREDICT_FALSE (q == 0)) + return 0; + + my_fifo_events = smm->free_event_vector[my_thread_index]; + + /* min number of events we can dequeue without blocking */ + n_to_dequeue = q->cursize; + my_pending_event_vector = smm->pending_event_vector[my_thread_index]; + + if (n_to_dequeue == 0 && vec_len (my_pending_event_vector) == 0) + return 0; + + SESSION_EVT_DBG (SESSION_EVT_DEQ_NODE, 0); + + /* + * If we didn't manage to process previous events try going + * over them again without dequeuing new ones. + */ + /* XXX: Block senders to sessions that can't keep up */ + if (0 && vec_len (my_pending_event_vector) >= 100) + { + clib_warning ("too many fifo events unsolved"); + goto skip_dequeue; + } + + /* See you in the next life, don't be late */ + if (pthread_mutex_trylock (&q->mutex)) + return 0; + + for (i = 0; i < n_to_dequeue; i++) + { + vec_add2 (my_fifo_events, e, 1); + unix_shared_memory_queue_sub_raw (q, (u8 *) e); + } + + /* The other side of the connection is not polling */ + if (q->cursize < (q->maxsize / 8)) + (void) pthread_cond_broadcast (&q->condvar); + pthread_mutex_unlock (&q->mutex); + + vec_append (my_fifo_events, my_pending_event_vector); + + _vec_len (my_pending_event_vector) = 0; + smm->pending_event_vector[my_thread_index] = my_pending_event_vector; + +skip_dequeue: + n_events = vec_len (my_fifo_events); + for (i = 0; i < n_events; i++) + { + stream_session_t *s0; /* $$$ prefetch 1 ahead maybe */ + session_fifo_event_t *e0; + + e0 = &my_fifo_events[i]; + + switch (e0->event_type) + { + case FIFO_EVENT_APP_TX: + s0 = session_event_get_session (e0, my_thread_index); + + if (PREDICT_FALSE (!s0)) + { + 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)) + 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, + my_thread_index, + &n_tx_packets); + /* Out of buffers */ + if (PREDICT_FALSE (rv < 0)) + { + vlib_node_increment_counter (vm, node->node_index, + SESSION_QUEUE_ERROR_NO_BUFFER, 1); + continue; + } + break; + case FIFO_EVENT_DISCONNECT: + s0 = stream_session_get_from_handle (e0->session_handle); + stream_session_disconnect (s0); + break; + case FIFO_EVENT_BUILTIN_RX: + s0 = session_event_get_session (e0, my_thread_index); + if (PREDICT_FALSE (!s0)) + continue; + svm_fifo_unset_event (s0->server_rx_fifo); + app = application_get (s0->app_index); + app->cb_fns.builtin_server_rx_callback (s0); + break; + case FIFO_EVENT_RPC: + fp = e0->rpc_args.fp; + (*fp) (e0->rpc_args.arg); + break; + + default: + clib_warning ("unhandled event type %d", e0->event_type); + } + } + + _vec_len (my_fifo_events) = 0; + smm->free_event_vector[my_thread_index] = my_fifo_events; + + vlib_node_increment_counter (vm, session_queue_node.index, + SESSION_QUEUE_ERROR_TX, n_tx_packets); + + SESSION_EVT_DBG (SESSION_EVT_DEQ_NODE, 1); + + return n_tx_packets; +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (session_queue_node) = +{ + .function = session_queue_node_fn, + .name = "session-queue", + .format_trace = format_session_queue_trace, + .type = VLIB_NODE_TYPE_INPUT, + .n_errors = ARRAY_LEN (session_queue_error_strings), + .error_strings = session_queue_error_strings, + .n_next_nodes = SESSION_QUEUE_N_NEXT, + .state = VLIB_NODE_STATE_DISABLED, + .next_nodes = + { + [SESSION_QUEUE_NEXT_DROP] = "error-drop", + [SESSION_QUEUE_NEXT_IP4_LOOKUP] = "ip4-lookup", + [SESSION_QUEUE_NEXT_IP6_LOOKUP] = "ip6-lookup", + [SESSION_QUEUE_NEXT_TCP_IP4_OUTPUT] = "tcp4-output", + [SESSION_QUEUE_NEXT_TCP_IP6_OUTPUT] = "tcp6-output", + }, +}; +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/stream_session.h b/src/vnet/session/stream_session.h new file mode 100644 index 00000000..275052d3 --- /dev/null +++ b/src/vnet/session/stream_session.h @@ -0,0 +1,92 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef SRC_VNET_SESSION_STREAM_SESSION_H_ +#define SRC_VNET_SESSION_STREAM_SESSION_H_ + +#include <vnet/vnet.h> +#include <svm/svm_fifo.h> + +#define foreach_session_type \ + _(IP4_TCP, ip4_tcp) \ + _(IP4_UDP, ip4_udp) \ + _(IP6_TCP, ip6_tcp) \ + _(IP6_UDP, ip6_udp) + +typedef enum +{ +#define _(A, a) SESSION_TYPE_##A, + foreach_session_type +#undef _ + SESSION_N_TYPES, +} session_type_t; + +/* + * Application session state + */ +typedef enum +{ + SESSION_STATE_LISTENING, + SESSION_STATE_CONNECTING, + SESSION_STATE_ACCEPTING, + SESSION_STATE_READY, + SESSION_STATE_CLOSED, + SESSION_STATE_N_STATES, +} stream_session_state_t; + +typedef struct _stream_session_t +{ + /** fifo pointers. Once allocated, these do not move */ + svm_fifo_t *server_rx_fifo; + svm_fifo_t *server_tx_fifo; + + /** Type */ + u8 session_type; + + /** State */ + volatile u8 session_state; + + u8 thread_index; + + /** To avoid n**2 "one event per frame" check */ + u8 enqueue_epoch; + + /** svm segment index where fifos were allocated */ + u32 svm_segment_index; + + /** Session index in per_thread pool */ + u32 session_index; + + /** Transport specific */ + u32 connection_index; + + /** stream server pool index */ + u32 app_index; + + /** Parent listener session if the result of an accept */ + u32 listener_index; + + CLIB_CACHE_LINE_ALIGN_MARK (pad); +} stream_session_t; + +#endif /* SRC_VNET_SESSION_STREAM_SESSION_H_ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/transport.h b/src/vnet/session/transport.h new file mode 100644 index 00000000..e2c47949 --- /dev/null +++ b/src/vnet/session/transport.h @@ -0,0 +1,94 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef VNET_VNET_URI_TRANSPORT_H_ +#define VNET_VNET_URI_TRANSPORT_H_ + +#include <vnet/vnet.h> +#include <vnet/ip/ip.h> +#include <vppinfra/bihash_16_8.h> +#include <vppinfra/bihash_48_8.h> +#include <vnet/tcp/tcp_debug.h> + +/* + * Protocol independent transport properties associated to a session + */ +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 transport_proto; /**< Protocol id */ + u8 is_ip4; /**< Flag if IP4 connection */ + u32 vrf; /**< FIB table id */ + + u32 s_index; /**< Parent session index */ + u32 c_index; /**< Connection index in transport pool */ + u32 thread_index; /**< Worker-thread index */ + + fib_node_index_t rmt_fei; /**< FIB entry index for rmt */ + dpo_id_t rmt_dpo; /**< Forwarding DPO for rmt */ + +#if TRANSPORT_DEBUG + elog_track_t elog_track; /**< Event logging */ + u32 cc_stat_tstamp; /**< CC stats timestamp */ +#endif + + /** Macros for 'derived classes' where base is named "connection" */ +#define c_lcl_ip connection.lcl_ip +#define c_rmt_ip connection.rmt_ip +#define c_lcl_ip4 connection.lcl_ip.ip4 +#define c_rmt_ip4 connection.rmt_ip.ip4 +#define c_lcl_ip6 connection.lcl_ip.ip6 +#define c_rmt_ip6 connection.rmt_ip.ip6 +#define c_lcl_port connection.lcl_port +#define c_rmt_port connection.rmt_port +#define c_transport_proto connection.transport_proto +#define c_vrf connection.vrf +#define c_state connection.state +#define c_s_index connection.s_index +#define c_c_index connection.c_index +#define c_is_ip4 connection.is_ip4 +#define c_thread_index connection.thread_index +#define c_elog_track connection.elog_track +#define c_cc_stat_tstamp connection.cc_stat_tstamp +#define c_rmt_fei connection.rmt_fei +#define c_rmt_dpo connection.rmt_dpo +} transport_connection_t; + +typedef enum _transport_proto +{ + TRANSPORT_PROTO_TCP, + TRANSPORT_PROTO_UDP +} transport_proto_t; + +typedef struct _transport_endpoint +{ + ip46_address_t ip; /** ip address */ + u16 port; /** port in net order */ + u8 is_ip4; /** 1 if ip4 */ + u32 vrf; /** fib table the endpoint is associated with */ +} transport_endpoint_t; + +#endif /* VNET_VNET_URI_TRANSPORT_H_ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/transport_interface.c b/src/vnet/session/transport_interface.c new file mode 100644 index 00000000..ef8d1e49 --- /dev/null +++ b/src/vnet/session/transport_interface.c @@ -0,0 +1,109 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include <vnet/session/transport_interface.h> +#include <vnet/session/session.h> + +/** + * Per-type vector of transport protocol virtual function tables + */ +transport_proto_vft_t *tp_vfts; + +u32 +transport_endpoint_lookup (transport_endpoint_table_t * ht, + ip46_address_t * ip, u16 port) +{ + clib_bihash_kv_24_8_t kv; + int rv; + + kv.key[0] = ip->as_u64[0]; + kv.key[1] = ip->as_u64[1]; + kv.key[2] = port; + + rv = clib_bihash_search_inline_24_8 (ht, &kv); + if (rv == 0) + return kv.value; + + return TRANSPORT_ENDPOINT_INVALID_INDEX; +} + +void +transport_endpoint_table_add (transport_endpoint_table_t * ht, + transport_endpoint_t * te, u32 value) +{ + clib_bihash_kv_24_8_t kv; + + kv.key[0] = te->ip.as_u64[0]; + kv.key[1] = te->ip.as_u64[1]; + kv.key[2] = te->port; + kv.value = value; + + clib_bihash_add_del_24_8 (ht, &kv, 1); +} + +void +transport_endpoint_table_del (transport_endpoint_table_t * ht, + transport_endpoint_t * te) +{ + clib_bihash_kv_24_8_t kv; + + kv.key[0] = te->ip.as_u64[0]; + kv.key[1] = te->ip.as_u64[1]; + kv.key[2] = te->port; + + clib_bihash_add_del_24_8 (ht, &kv, 0); +} + +/** + * Register transport virtual function table. + * + * @param type - session type (not protocol type) + * @param vft - virtual function table + */ +void +session_register_transport (transport_proto_t transport_proto, u8 is_ip4, + const transport_proto_vft_t * vft) +{ + u8 session_type; + session_type = session_type_from_proto_and_ip (transport_proto, is_ip4); + + vec_validate (tp_vfts, session_type); + tp_vfts[session_type] = *vft; + + /* If an offset function is provided, then peek instead of dequeue */ + session_manager_set_transport_rx_fn (session_type, + vft->tx_fifo_offset != 0); +} + +/** + * Get transport virtual function table + * + * @param type - session type (not protocol type) + */ +transport_proto_vft_t * +session_get_transport_vft (u8 session_type) +{ + if (session_type >= vec_len (tp_vfts)) + return 0; + return &tp_vfts[session_type]; +} + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/session/transport_interface.h b/src/vnet/session/transport_interface.h new file mode 100644 index 00000000..661221c4 --- /dev/null +++ b/src/vnet/session/transport_interface.h @@ -0,0 +1,82 @@ +/* + * Copyright (c) 2017 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef SRC_VNET_SESSION_TRANSPORT_INTERFACE_H_ +#define SRC_VNET_SESSION_TRANSPORT_INTERFACE_H_ + +#include <vnet/vnet.h> +#include <vnet/session/transport.h> + +/* + * Transport protocol virtual function table + */ +typedef struct _transport_proto_vft +{ + /* + * Setup + */ + u32 (*bind) (u32 session_index, transport_endpoint_t * lcl); + u32 (*unbind) (u32); + int (*open) (transport_endpoint_t * rmt); + void (*close) (u32 conn_index, u32 thread_index); + void (*cleanup) (u32 conn_index, u32 thread_index); + + /* + * Transmission + */ + u32 (*push_header) (transport_connection_t * tconn, vlib_buffer_t * b); + u16 (*send_mss) (transport_connection_t * tc); + u32 (*send_space) (transport_connection_t * tc); + u32 (*tx_fifo_offset) (transport_connection_t * tc); + + /* + * Connection retrieval + */ + transport_connection_t *(*get_connection) (u32 conn_idx, u32 thread_idx); + transport_connection_t *(*get_listener) (u32 conn_index); + transport_connection_t *(*get_half_open) (u32 conn_index); + + /* + * Format + */ + u8 *(*format_connection) (u8 * s, va_list * args); + u8 *(*format_listener) (u8 * s, va_list * args); + u8 *(*format_half_open) (u8 * s, va_list * args); +} transport_proto_vft_t; + +typedef clib_bihash_24_8_t transport_endpoint_table_t; + +#define TRANSPORT_ENDPOINT_INVALID_INDEX ((u32)~0) + +u32 transport_endpoint_lookup (transport_endpoint_table_t * ht, + ip46_address_t * ip, u16 port); +void transport_endpoint_table_add (transport_endpoint_table_t * ht, + transport_endpoint_t * te, u32 value); +void transport_endpoint_table_del (transport_endpoint_table_t * ht, + transport_endpoint_t * te); + +void session_register_transport (transport_proto_t transport_proto, u8 is_ip4, + const transport_proto_vft_t * vft); +transport_proto_vft_t *session_get_transport_vft (u8 session_type); + +#endif /* SRC_VNET_SESSION_TRANSPORT_INTERFACE_H_ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ |