summaryrefslogtreecommitdiffstats
path: root/src/plugins/wireguard/wireguard_output_tun.c
blob: c792d4b713ec281f9fab05d9b53447508626c55f (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
/*
 * Copyright (c) 2020 Doc.ai and/or its affiliates.
 * Copyright (c) 2020 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 <vlib/vlib.h>
#include <vnet/vnet.h>
#include <vppinfra/error.h>

#include <wireguard/wireguard.h>
#include <wireguard/wireguard_send.h>

#define foreach_wg_output_error                                         \
 _(NONE, "No error")							\
 _(PEER, "Peer error")                                                  \
 _(KEYPAIR, "Keypair error")                                            \
 _(TOO_BIG, "packet too big")                                           \

typedef enum
{
#define _(sym,str) WG_OUTPUT_ERROR_##sym,
  foreach_wg_output_error
#undef _
    WG_OUTPUT_N_ERROR,
} wg_output_error_t;

static char *wg_output_error_strings[] = {
#define _(sym,string) string,
  foreach_wg_output_error
#undef _
};

typedef enum
{
  WG_OUTPUT_NEXT_ERROR,
  WG_OUTPUT_NEXT_HANDOFF,
  WG_OUTPUT_NEXT_INTERFACE_OUTPUT,
  WG_OUTPUT_N_NEXT,
} wg_output_next_t;

typedef struct
{
  index_t peer;
  u8 header[sizeof (ip6_udp_header_t)];
  u8 is_ip4;
} wg_output_tun_trace_t;

u8 *
format_ip4_udp_header (u8 * s, va_list * args)
{
  ip4_udp_header_t *hdr4 = va_arg (*args, ip4_udp_header_t *);

  s = format (s, "%U:$U", format_ip4_header, &hdr4->ip4, format_udp_header,
	      &hdr4->udp);
  return (s);
}

u8 *
format_ip6_udp_header (u8 *s, va_list *args)
{
  ip6_udp_header_t *hdr6 = va_arg (*args, ip6_udp_header_t *);

  s = format (s, "%U:$U", format_ip6_header, &hdr6->ip6, format_udp_header,
	      &hdr6->udp);
  return (s);
}

/* packet trace format function */
static u8 *
format_wg_output_tun_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 *);

  wg_output_tun_trace_t *t = va_arg (*args, wg_output_tun_trace_t *);

  s = format (s, "peer: %d\n", t->peer);
  s = format (s, "  Encrypted packet: ");

  s = t->is_ip4 ? format (s, "%U", format_ip4_udp_header, t->header) :
		  format (s, "%U", format_ip6_udp_header, t->header);
  return s;
}

/* is_ip4 - inner header flag */
always_inline uword
wg_output_tun_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
		      vlib_frame_t *frame, u8 is_ip4)
{
  u32 n_left_from;
  u32 *from;
  ip4_udp_wg_header_t *hdr4_out = NULL;
  ip6_udp_wg_header_t *hdr6_out = NULL;
  message_data_t *message_data_wg = NULL;
  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b;
  u16 nexts[VLIB_FRAME_SIZE], *next;
  u32 thread_index = vm->thread_index;

  from = vlib_frame_vector_args (frame);
  n_left_from = frame->n_vectors;
  b = bufs;
  next = nexts;

  vlib_get_buffers (vm, from, bufs, n_left_from);

  wg_peer_t *peer = NULL;

  while (n_left_from > 0)
    {
      index_t peeri;
      u8 iph_offset = 0;
      u8 is_ip4_out = 1;
      u8 *plain_data;
      u16 plain_data_len;

      if (n_left_from > 2)
	{
	  u8 *p;
	  vlib_prefetch_buffer_header (b[2], LOAD);
	  p = vlib_buffer_get_current (b[1]);
	  CLIB_PREFETCH (p, CLIB_CACHE_LINE_BYTES, LOAD);
	}

      next[0] = WG_OUTPUT_NEXT_ERROR;
      peeri =
	wg_peer_get_by_adj_index (vnet_buffer (b[0])->ip.adj_index[VLIB_TX]);
      peer = wg_peer_get (peeri);

      if (wg_peer_is_dead (peer))
	{
	  b[0]->error = node->errors[WG_OUTPUT_ERROR_PEER];
	  goto out;
	}
      if (PREDICT_FALSE (~0 == peer->output_thread_index))
	{
	  /* this is the first packet to use this peer, claim the peer
	   * for this thread.
	   */
	  clib_atomic_cmp_and_swap (&peer->output_thread_index, ~0,
				    wg_peer_assign_thread (thread_index));
	}

      if (PREDICT_TRUE (thread_index != peer->output_thread_index))
	{
	  next[0] = WG_OUTPUT_NEXT_HANDOFF;
	  goto next;
	}

      if (PREDICT_FALSE (!peer->remote.r_current))
	{
	  wg_send_handshake_from_mt (peeri, false);
	  b[0]->error = node->errors[WG_OUTPUT_ERROR_KEYPAIR];
	  goto out;
	}

      is_ip4_out = ip46_address_is_ip4 (&peer->src.addr);
      if (is_ip4_out)
	{
	  hdr4_out = vlib_buffer_get_current (b[0]);
	  message_data_wg = &hdr4_out->wg;
	}
      else
	{
	  hdr6_out = vlib_buffer_get_current (b[0]);
	  message_data_wg = &hdr6_out->wg;
	}

      iph_offset = vnet_buffer (b[0])->ip.save_rewrite_length;
      plain_data = vlib_buffer_get_current (b[0]) + iph_offset;
      plain_data_len = vlib_buffer_length_in_chain (vm, b[0]) - iph_offset;

      size_t encrypted_packet_len = message_data_len (plain_data_len);

      /*
       * Ensure there is enough space to write the encrypted data
       * into the packet
       */
      if (PREDICT_FALSE (encrypted_packet_len >= WG_DEFAULT_DATA_SIZE) ||
	  PREDICT_FALSE ((b[0]->current_data + encrypted_packet_len) >=
			 vlib_buffer_get_default_data_size (vm)))
	{
	  b[0]->error = node->errors[WG_OUTPUT_ERROR_TOO_BIG];
	  goto out;
	}

      enum noise_state_crypt state;

      state = noise_remote_encrypt (
	vm, &peer->remote, &message_data_wg->receiver_index,
	&message_data_wg->counter, plain_data, plain_data_len, plain_data);

      if (PREDICT_FALSE (state == SC_KEEP_KEY_FRESH))
	{
	  wg_send_handshake_from_mt (peeri, false);
	}
      else if (PREDICT_FALSE (state == SC_FAILED))
	{
	  //TODO: Maybe wrong
	  wg_send_handshake_from_mt (peeri, false);
	  wg_peer_update_flags (peeri, WG_PEER_ESTABLISHED, false);
	  goto out;
	}

      /* Here we are sure that can send packet to next node */
      next[0] = WG_OUTPUT_NEXT_INTERFACE_OUTPUT;

      if (is_ip4_out)
	{
	  hdr4_out->wg.header.type = MESSAGE_DATA;
	  hdr4_out->udp.length = clib_host_to_net_u16 (encrypted_packet_len +
						       sizeof (udp_header_t));
	  b[0]->current_length =
	    (encrypted_packet_len + sizeof (ip4_udp_header_t));
	  ip4_header_set_len_w_chksum (
	    &hdr4_out->ip4, clib_host_to_net_u16 (b[0]->current_length));
	}
      else
	{
	  hdr6_out->wg.header.type = MESSAGE_DATA;
	  hdr6_out->udp.length = clib_host_to_net_u16 (encrypted_packet_len +
						       sizeof (udp_header_t));
	  b[0]->current_length =
	    (encrypted_packet_len + sizeof (ip6_udp_header_t));
	  hdr6_out->ip6.payload_length =
	    clib_host_to_net_u16 (b[0]->current_length);
	}

      wg_timers_any_authenticated_packet_sent (peer);
      wg_timers_data_sent (peer);
      wg_timers_any_authenticated_packet_traversal (peer);

    out:
      if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE)
			 && (b[0]->flags & VLIB_BUFFER_IS_TRACED)))
	{
	  wg_output_tun_trace_t *t =
	    vlib_add_trace (vm, node, b[0], sizeof (*t));

	  t->peer = peeri;
	  t->is_ip4 = is_ip4_out;
	  if (hdr4_out)
	    clib_memcpy (t->header, hdr4_out, sizeof (ip4_udp_header_t));
	  else if (hdr6_out)
	    clib_memcpy (t->header, hdr6_out, sizeof (ip6_udp_header_t));
	}

    next:
      n_left_from -= 1;
      next += 1;
      b += 1;
    }

  vlib_buffer_enqueue_to_next (vm, node, from, nexts, frame->n_vectors);
  return frame->n_vectors;
}

VLIB_NODE_FN (wg4_output_tun_node)
(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame)
{
  return wg_output_tun_inline (vm, node, frame, /* is_ip4 */ 1);
}

VLIB_NODE_FN (wg6_output_tun_node)
(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame)
{
  return wg_output_tun_inline (vm, node, frame, /* is_ip4 */ 0);
}

/* *INDENT-OFF* */
VLIB_REGISTER_NODE (wg4_output_tun_node) =
{
  .name = "wg4-output-tun",
  .vector_size = sizeof (u32),
  .format_trace = format_wg_output_tun_trace,
  .type = VLIB_NODE_TYPE_INTERNAL,
  .n_errors = ARRAY_LEN (wg_output_error_strings),
  .error_strings = wg_output_error_strings,
  .n_next_nodes = WG_OUTPUT_N_NEXT,
  .next_nodes = {
        [WG_OUTPUT_NEXT_HANDOFF] = "wg4-output-tun-handoff",
        [WG_OUTPUT_NEXT_INTERFACE_OUTPUT] = "adj-midchain-tx",
        [WG_OUTPUT_NEXT_ERROR] = "error-drop",
  },
};

VLIB_REGISTER_NODE (wg6_output_tun_node) =
{
  .name = "wg6-output-tun",
  .vector_size = sizeof (u32),
  .format_trace = format_wg_output_tun_trace,
  .type = VLIB_NODE_TYPE_INTERNAL,
  .n_errors = ARRAY_LEN (wg_output_error_strings),
  .error_strings = wg_output_error_strings,
  .n_next_nodes = WG_OUTPUT_N_NEXT,
  .next_nodes = {
        [WG_OUTPUT_NEXT_HANDOFF] = "wg6-output-tun-handoff",
        [WG_OUTPUT_NEXT_INTERFACE_OUTPUT] = "adj-midchain-tx",
        [WG_OUTPUT_NEXT_ERROR] = "error-drop",
  },
};
/* *INDENT-ON* */

/*
 * fd.io coding-style-patch-verification: ON
 *
 * Local Variables:
 * eval: (c-set-style "gnu")
 * End:
 */
class="n">head_index, t - tw->timers); #if TW_START_STOP_TRACE_SIZE > 0 TW (tw_timer_trace) (tw, timer_id, user_id, t - tw->timers); #endif return; } #endif glacier_ring_offset = interval >> (2 * TW_RING_SHIFT); ASSERT ((u64) glacier_ring_offset < TW_SLOTS_PER_RING); interval -= (((u64) glacier_ring_offset) << (2 * TW_RING_SHIFT)); #endif #if TW_TIMER_WHEELS > 1 slow_ring_offset = interval >> TW_RING_SHIFT; ASSERT ((u64) slow_ring_offset < TW_SLOTS_PER_RING); interval -= (((u64) slow_ring_offset) << TW_RING_SHIFT); #endif fast_ring_offset = interval & TW_RING_MASK; /* * Account for the current wheel positions(s) * This is made slightly complicated by the fact that the current * index vector will contain (TW_SLOTS_PER_RING, ...) when * the actual position is (0, ...) */ fast_ring_offset += tw->current_index[TW_TIMER_RING_FAST] & TW_RING_MASK; #if TW_TIMER_WHEELS > 1 carry = fast_ring_offset >= TW_SLOTS_PER_RING ? 1 : 0; fast_ring_offset %= TW_SLOTS_PER_RING; slow_ring_offset += (tw->current_index[TW_TIMER_RING_SLOW] & TW_RING_MASK) + carry; carry = slow_ring_offset >= TW_SLOTS_PER_RING ? 1 : 0; slow_ring_offset %= TW_SLOTS_PER_RING; #endif #if TW_TIMER_WHEELS > 2 glacier_ring_offset += (tw->current_index[TW_TIMER_RING_GLACIER] & TW_RING_MASK) + carry; glacier_ring_offset %= TW_SLOTS_PER_RING; #endif #if TW_TIMER_WHEELS > 2 if (glacier_ring_offset != (tw->current_index[TW_TIMER_RING_GLACIER] & TW_RING_MASK)) { /* We'll need slow and fast ring offsets later */ t->slow_ring_offset = slow_ring_offset; t->fast_ring_offset = fast_ring_offset; ts = &tw->w[TW_TIMER_RING_GLACIER][glacier_ring_offset]; timer_addhead (tw->timers, ts->head_index, t - tw->timers); #if TW_START_STOP_TRACE_SIZE > 0 TW (tw_timer_trace) (tw, timer_id, user_id, t - tw->timers); #endif return; } #endif #if TW_TIMER_WHEELS > 1 /* Timer expires more than 51.2 seconds from now? */ if (slow_ring_offset != (tw->current_index[TW_TIMER_RING_SLOW] & TW_RING_MASK)) { /* We'll need the fast ring offset later... */ t->fast_ring_offset = fast_ring_offset; ts = &tw->w[TW_TIMER_RING_SLOW][slow_ring_offset]; timer_addhead (tw->timers, ts->head_index, t - tw->timers); #if TW_START_STOP_TRACE_SIZE > 0 TW (tw_timer_trace) (tw, timer_id, user_id, t - tw->timers); #endif return; } #else fast_ring_offset %= TW_SLOTS_PER_RING; #endif /* Timer expires less than one fast-ring revolution from now */ ts = &tw->w[TW_TIMER_RING_FAST][fast_ring_offset]; timer_addhead (tw->timers, ts->head_index, t - tw->timers); #if TW_FAST_WHEEL_BITMAP tw->fast_slot_bitmap = clib_bitmap_set (tw->fast_slot_bitmap, fast_ring_offset, 1); #endif #if TW_START_STOP_TRACE_SIZE > 0 TW (tw_timer_trace) (tw, timer_id, user_id, t - tw->timers); #endif } /** * @brief Start a Tw Timer * @param tw_timer_wheel_t * tw timer wheel object pointer * @param u32 user_id user defined timer id, presumably for a tw session * @param u32 timer_id app-specific timer ID. 4 bits. * @param u64 interval timer interval in ticks * @returns handle needed to cancel the timer */ u32 TW (tw_timer_start) (TWT (tw_timer_wheel) * tw, u32 user_id, u32 timer_id, u64 interval) { TWT (tw_timer) * t; ASSERT (interval); pool_get (tw->timers, t); memset (t, 0xff, sizeof (*t)); t->user_handle = TW (make_internal_timer_handle) (user_id, timer_id); timer_add (tw, t, interval); return t - tw->timers; } #if TW_TIMER_SCAN_FOR_HANDLE > 0 int TW (scan_for_handle) (TWT (tw_timer_wheel) * tw, u32 handle) { int i, j; tw_timer_wheel_slot_t *ts; TWT (tw_timer) * t, *head; u32 next_index; int rv = 0; for (i = 0; i < TW_TIMER_WHEELS; i++) { for (j = 0; j < TW_SLOTS_PER_RING; j++) { ts = &tw->w[i][j]; head = pool_elt_at_index (tw->timers, ts->head_index); next_index = head->next; while (next_index != ts->head_index) { t = pool_elt_at_index (tw->timers, next_index); if (next_index == handle) { clib_warning ("handle %d found in ring %d slot %d", handle, i, j); clib_warning ("user handle 0x%x", t->user_handle); rv = 1; } next_index = t->next; } } } return rv; } #endif /* TW_TIMER_SCAN_FOR_HANDLE */ /** * @brief Stop a tw timer * @param tw_timer_wheel_t * tw timer wheel object pointer * @param u32 handle timer cancellation returned by tw_timer_start */ void TW (tw_timer_stop) (TWT (tw_timer_wheel) * tw, u32 handle) { TWT (tw_timer) * t; #if TW_TIMER_ALLOW_DUPLICATE_STOP /* * A vlib process may have its timer expire, and receive * an event before the expiration is processed. * That results in a duplicate tw_timer_stop. */ if (pool_is_free_index (tw->timers, handle)) return; #endif #if TW_START_STOP_TRACE_SIZE > 0 TW (tw_timer_trace) (tw, ~0, ~0, handle); #endif t = pool_elt_at_index (tw->timers, handle); /* in case of idiotic handle (e.g. passing a listhead index) */ ASSERT (t->user_handle != ~0); timer_remove (tw->timers, t); pool_put_index (tw->timers, handle); } int TW (tw_timer_handle_is_free) (TWT (tw_timer_wheel) * tw, u32 handle) { return pool_is_free_index (tw->timers, handle); } /** * @brief Update a tw timer * @param tw_timer_wheel_t * tw timer wheel object pointer * @param u32 handle timer returned by tw_timer_start * @param u32 interval timer interval in ticks */ void TW (tw_timer_update) (TWT (tw_timer_wheel) * tw, u32 handle, u64 interval) { TWT (tw_timer) * t; t = pool_elt_at_index (tw->timers, handle); timer_remove (tw->timers, t); timer_add (tw, t, interval); } /** * @brief Initialize a tw timer wheel template instance * @param tw_timer_wheel_t * tw timer wheel object pointer * @param void * expired_timer_callback. Passed a u32 * vector of * expired timer handles. The callback is optional. * @param f64 timer_interval_in_seconds */ void TW (tw_timer_wheel_init) (TWT (tw_timer_wheel) * tw, void *expired_timer_callback, f64 timer_interval_in_seconds, u32 max_expirations) { int ring, slot; tw_timer_wheel_slot_t *ts; TWT (tw_timer) * t; memset (tw, 0, sizeof (*tw)); tw->expired_timer_callback = expired_timer_callback; tw->max_expirations = max_expirations; if (timer_interval_in_seconds == 0.0) { clib_warning ("timer interval is zero"); abort (); } tw->timer_interval = timer_interval_in_seconds; tw->ticks_per_second = 1.0 / timer_interval_in_seconds; tw->first_expires_tick = ~0ULL; vec_validate (tw->expired_timer_handles, 0); _vec_len (tw->expired_timer_handles) = 0; for (ring = 0; ring < TW_TIMER_WHEELS; ring++) { for (slot = 0; slot < TW_SLOTS_PER_RING; slot++) { ts = &tw->w[ring][slot]; pool_get (tw->timers, t); memset (t, 0xff, sizeof (*t)); t->next = t->prev = t - tw->timers; ts->head_index = t - tw->timers; } } #if TW_OVERFLOW_VECTOR > 0 ts = &tw->overflow; pool_get (tw->timers, t); memset (t, 0xff, sizeof (*t)); t->next = t->prev = t - tw->timers; ts->head_index = t - tw->timers; #endif } /** * @brief Free a tw timer wheel template instance * @param tw_timer_wheel_t * tw timer wheel object pointer */ void TW (tw_timer_wheel_free) (TWT (tw_timer_wheel) * tw) { int i, j; tw_timer_wheel_slot_t *ts; TWT (tw_timer) * head, *t; u32 next_index; for (i = 0; i < TW_TIMER_WHEELS; i++) { for (j = 0; j < TW_SLOTS_PER_RING; j++) { ts = &tw->w[i][j]; head = pool_elt_at_index (tw->timers, ts->head_index); next_index = head->next; while (next_index != ts->head_index) { t = pool_elt_at_index (tw->timers, next_index); next_index = t->next; pool_put (tw->timers, t); } pool_put (tw->timers, head); } } #if TW_OVERFLOW_VECVOR > 0 ts = &tw->overflow; head = pool_elt_at_index (tw->timers, ts->head_index); next_index = head->next; while (next_index != ts->head_index) { t = pool_elt_at_index (tw->timers, next_index); next_index = t->next; pool_put (tw->timers, t); } pool_put (tw->timers, head); #endif memset (tw, 0, sizeof (*tw)); } /** * @brief Advance a tw timer wheel. Calls the expired timer callback * as needed. This routine should be called once every timer_interval seconds * @param tw_timer_wheel_t * tw timer wheel template instance pointer * @param f64 now the current time, e.g. from vlib_time_now(vm) * @returns u32 * vector of expired user handles */ static inline u32 * TW (tw_timer_expire_timers_internal) (TWT (tw_timer_wheel) * tw, f64 now, u32 * callback_vector_arg) { u32 nticks, i; tw_timer_wheel_slot_t *ts; TWT (tw_timer) * t, *head; u32 *callback_vector; u32 fast_wheel_index; u32 next_index; u32 slow_wheel_index __attribute__ ((unused)); u32 glacier_wheel_index __attribute__ ((unused)); /* Shouldn't happen */ if (PREDICT_FALSE (now < tw->next_run_time)) return callback_vector_arg; /* Number of ticks which have occurred */ nticks = tw->ticks_per_second * (now - tw->last_run_time); if (nticks == 0) return callback_vector_arg; /* Remember when we ran, compute next runtime */ tw->next_run_time = (now + tw->timer_interval); if (callback_vector_arg == 0) { _vec_len (tw->expired_timer_handles) = 0; callback_vector = tw->expired_timer_handles; } else callback_vector = callback_vector_arg; for (i = 0; i < nticks; i++) { fast_wheel_index = tw->current_index[TW_TIMER_RING_FAST]; if (TW_TIMER_WHEELS > 1) slow_wheel_index = tw->current_index[TW_TIMER_RING_SLOW]; if (TW_TIMER_WHEELS > 2) glacier_wheel_index = tw->current_index[TW_TIMER_RING_GLACIER]; #if TW_OVERFLOW_VECTOR > 0 /* Triple odometer-click? Process the overflow vector... */ if (PREDICT_FALSE (fast_wheel_index == TW_SLOTS_PER_RING && slow_wheel_index == TW_SLOTS_PER_RING && glacier_wheel_index == TW_SLOTS_PER_RING)) { u64 interval; u32 new_glacier_ring_offset, new_slow_ring_offset; u32 new_fast_ring_offset; ts = &tw->overflow; head = pool_elt_at_index (tw->timers, ts->head_index); next_index = head->next; /* Make slot empty */ head->next = head->prev = ts->head_index; /* traverse slot, place timers wherever they go */ while (next_index != head - tw->timers) { t = pool_elt_at_index (tw->timers, next_index); next_index = t->next; /* Remove from the overflow vector (hammer) */ t->next = t->prev = ~0; ASSERT (t->expiration_time >= tw->current_tick); interval = t->expiration_time - tw->current_tick; /* Right back onto the overflow vector? */ if (interval >= (1 << (3 * TW_RING_SHIFT))) { ts = &tw->overflow; timer_addhead (tw->timers, ts->head_index, t - tw->timers); continue; } /* Compute ring offsets */ new_glacier_ring_offset = interval >> (2 * TW_RING_SHIFT); interval -= (new_glacier_ring_offset << (2 * TW_RING_SHIFT)); /* Note: the wheels are at (0,0,0), no add-with-carry needed */ new_slow_ring_offset = interval >> TW_RING_SHIFT; interval -= (new_slow_ring_offset << TW_RING_SHIFT); new_fast_ring_offset = interval & TW_RING_MASK; t->slow_ring_offset = new_slow_ring_offset; t->fast_ring_offset = new_fast_ring_offset; /* Timer expires Right Now */ if (PREDICT_FALSE (t->slow_ring_offset == 0 && t->fast_ring_offset == 0 && new_glacier_ring_offset == 0)) { vec_add1 (callback_vector, t->user_handle); #if TW_START_STOP_TRACE_SIZE > 0 TW (tw_timer_trace) (tw, 0xfe, t->user_handle, t - tw->timers); #endif pool_put (tw->timers, t); } /* Timer moves to the glacier ring */ else if (new_glacier_ring_offset) { ts = &tw->w[TW_TIMER_RING_GLACIER][new_glacier_ring_offset]; timer_addhead (tw->timers, ts->head_index, t - tw->timers); } /* Timer moves to the slow ring */ else if (t->slow_ring_offset) { /* Add to slow ring */ ts = &tw->w[TW_TIMER_RING_SLOW][t->slow_ring_offset]; timer_addhead (tw->timers, ts->head_index, t - tw->timers); } /* Timer timer moves to the fast ring */ else { ts = &tw->w[TW_TIMER_RING_FAST][t->fast_ring_offset]; timer_addhead (tw->timers, ts->head_index, t - tw->timers); #if TW_FAST_WHEEL_BITMAP tw->fast_slot_bitmap = clib_bitmap_set (tw->fast_slot_bitmap, t->fast_ring_offset, 1); #endif } } } #endif #if TW_TIMER_WHEELS > 2 /* * Double odometer-click? Process one slot in the glacier ring... */ if (PREDICT_FALSE (fast_wheel_index == TW_SLOTS_PER_RING && slow_wheel_index == TW_SLOTS_PER_RING)) { glacier_wheel_index %= TW_SLOTS_PER_RING; ts = &tw->w[TW_TIMER_RING_GLACIER][glacier_wheel_index]; head = pool_elt_at_index (tw->timers, ts->head_index); next_index = head->next; /* Make slot empty */ head->next = head->prev = ts->head_index; /* traverse slot, deal timers into slow ring */ while (next_index != head - tw->timers) { t = pool_elt_at_index (tw->timers, next_index); next_index = t->next; /* Remove from glacier ring slot (hammer) */ t->next = t->prev = ~0; /* Timer expires Right Now */ if (PREDICT_FALSE (t->slow_ring_offset == 0 && t->fast_ring_offset == 0)) { vec_add1 (callback_vector, t->user_handle); #if TW_START_STOP_TRACE_SIZE > 0 TW (tw_timer_trace) (tw, 0xfe, t->user_handle, t - tw->timers); #endif pool_put (tw->timers, t); } /* Timer expires during slow-wheel tick 0 */ else if (PREDICT_FALSE (t->slow_ring_offset == 0)) { ts = &tw->w[TW_TIMER_RING_FAST][t->fast_ring_offset]; timer_addhead (tw->timers, ts->head_index, t - tw->timers); #if TW_FAST_WHEEL_BITMAP tw->fast_slot_bitmap = clib_bitmap_set (tw->fast_slot_bitmap, t->fast_ring_offset, 1); #endif } else /* typical case */ { /* Add to slow ring */ ts = &tw->w[TW_TIMER_RING_SLOW][t->slow_ring_offset]; timer_addhead (tw->timers, ts->head_index, t - tw->timers); } } } #endif #if TW_TIMER_WHEELS > 1 /* * Single odometer-click? Process a slot in the slow ring, */ if (PREDICT_FALSE (fast_wheel_index == TW_SLOTS_PER_RING)) { slow_wheel_index %= TW_SLOTS_PER_RING; ts = &tw->w[TW_TIMER_RING_SLOW][slow_wheel_index]; head = pool_elt_at_index (tw->timers, ts->head_index); next_index = head->next; /* Make slot empty */ head->next = head->prev = ts->head_index; /* traverse slot, deal timers into fast ring */ while (next_index != head - tw->timers) { t = pool_elt_at_index (tw->timers, next_index); next_index = t->next; /* Remove from sloe ring slot (hammer) */ t->next = t->prev = ~0; /* Timer expires Right Now */ if (PREDICT_FALSE (t->fast_ring_offset == 0)) { vec_add1 (callback_vector, t->user_handle); #if TW_START_STOP_TRACE_SIZE > 0 TW (tw_timer_trace) (tw, 0xfe, t->user_handle, t - tw->timers); #endif pool_put (tw->timers, t); } else /* typical case */ { /* Add to fast ring */ ts = &tw->w[TW_TIMER_RING_FAST][t->fast_ring_offset]; timer_addhead (tw->timers, ts->head_index, t - tw->timers); #if TW_FAST_WHEEL_BITMAP tw->fast_slot_bitmap = clib_bitmap_set (tw->fast_slot_bitmap, t->fast_ring_offset, 1); #endif } } } #endif /* Handle the fast ring */ fast_wheel_index %= TW_SLOTS_PER_RING; ts = &tw->w[TW_TIMER_RING_FAST][fast_wheel_index]; head = pool_elt_at_index (tw->timers, ts->head_index); next_index = head->next; /* Make slot empty */ head->next = head->prev = ts->head_index; /* Construct vector of expired timer handles to give the user */ while (next_index != ts->head_index) { t = pool_elt_at_index (tw->timers, next_index); next_index = t->next; vec_add1 (callback_vector, t->user_handle); #if TW_START_STOP_TRACE_SIZE > 0 TW (tw_timer_trace) (tw, 0xfe, t->user_handle, t - tw->timers); #endif pool_put (tw->timers, t); } /* If any timers expired, tell the user */ if (callback_vector_arg == 0 && vec_len (callback_vector)) { /* The callback is optional. We return the u32 * handle vector */ if (tw->expired_timer_callback) { tw->expired_timer_callback (callback_vector); vec_reset_length (callback_vector); } tw->expired_timer_handles = callback_vector; } #if TW_FAST_WHEEL_BITMAP tw->fast_slot_bitmap = clib_bitmap_set (tw->fast_slot_bitmap, fast_wheel_index, 0); #endif tw->current_tick++; fast_wheel_index++; tw->current_index[TW_TIMER_RING_FAST] = fast_wheel_index; #if TW_TIMER_WHEELS > 1 if (PREDICT_FALSE (fast_wheel_index == TW_SLOTS_PER_RING)) slow_wheel_index++; tw->current_index[TW_TIMER_RING_SLOW] = slow_wheel_index; #endif #if TW_TIMER_WHEELS > 2 if (PREDICT_FALSE (slow_wheel_index == TW_SLOTS_PER_RING)) glacier_wheel_index++; tw->current_index[TW_TIMER_RING_GLACIER] = glacier_wheel_index; #endif if (vec_len (callback_vector) >= tw->max_expirations) break; } if (callback_vector_arg == 0) tw->expired_timer_handles = callback_vector; tw->last_run_time += i * tw->timer_interval; return callback_vector; } u32 *TW (tw_timer_expire_timers) (TWT (tw_timer_wheel) * tw, f64 now) { return TW (tw_timer_expire_timers_internal) (tw, now, 0 /* no vector */ ); } u32 *TW (tw_timer_expire_timers_vec) (TWT (tw_timer_wheel) * tw, f64 now, u32 * vec) { return TW (tw_timer_expire_timers_internal) (tw, now, vec); } #if TW_FAST_WHEEL_BITMAP /** Returns an approximation to the first timer expiration in * timer-ticks from "now". To avoid wasting an unjustifiable * amount of time on the problem, we maintain an approximate fast-wheel slot * occupancy bitmap. We don't worry about clearing fast wheel bits * when timers are removed from fast wheel slots. */ u32 TW (tw_timer_first_expires_in_ticks) (TWT (tw_timer_wheel) * tw) { u32 first_expiring_index, fast_ring_index; i32 delta; if (clib_bitmap_is_zero (tw->fast_slot_bitmap)) return TW_SLOTS_PER_RING; fast_ring_index = tw->current_index[TW_TIMER_RING_FAST]; if (fast_ring_index == TW_SLOTS_PER_RING) fast_ring_index = 0; first_expiring_index = clib_bitmap_next_set (tw->fast_slot_bitmap, fast_ring_index); if (first_expiring_index == ~0 && fast_ring_index != 0) first_expiring_index = clib_bitmap_first_set (tw->fast_slot_bitmap); ASSERT (first_expiring_index != ~0); delta = (i32) first_expiring_index - (i32) fast_ring_index; if (delta < 0) delta += TW_SLOTS_PER_RING; ASSERT (delta >= 0); return (u32) delta; } #endif /* * fd.io coding-style-patch-verification: ON * * Local Variables: * eval: (c-set-style "gnu") * End: */