Line data Source code
1 : /*
2 : *------------------------------------------------------------------
3 : * Copyright (c) 2016 Cisco and/or its affiliates.
4 : * Licensed under the Apache License, Version 2.0 (the "License");
5 : * you may not use this file except in compliance with the License.
6 : * You may obtain a copy of the License at:
7 : *
8 : * http://www.apache.org/licenses/LICENSE-2.0
9 : *
10 : * Unless required by applicable law or agreed to in writing, software
11 : * distributed under the License is distributed on an "AS IS" BASIS,
12 : * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 : * See the License for the specific language governing permissions and
14 : * limitations under the License.
15 : *------------------------------------------------------------------
16 : */
17 :
18 : #include <sys/types.h>
19 : #include <sys/stat.h>
20 : #include <fcntl.h>
21 : #include <net/if.h>
22 : #include <linux/if_tun.h>
23 : #include <sys/ioctl.h>
24 : #include <sys/eventfd.h>
25 :
26 : #include <vlib/vlib.h>
27 : #include <vlib/unix/unix.h>
28 : #include <vnet/ethernet/ethernet.h>
29 : #include <vnet/feature/feature.h>
30 : #include <vnet/interface/rx_queue_funcs.h>
31 : #include <vnet/ip/ip4_packet.h>
32 : #include <vnet/ip/ip6_packet.h>
33 : #include <vnet/udp/udp_packet.h>
34 : #include <vnet/tcp/tcp_packet.h>
35 : #include <vnet/devices/virtio/virtio.h>
36 : #include <vnet/devices/virtio/virtio_inline.h>
37 :
38 : static char *virtio_input_error_strings[] = {
39 : #define _(n, s) s,
40 : foreach_virtio_input_error
41 : #undef _
42 : };
43 :
44 : typedef struct
45 : {
46 : u32 next_index;
47 : u32 hw_if_index;
48 : u16 ring;
49 : u16 len;
50 : vnet_virtio_net_hdr_v1_t hdr;
51 : } virtio_input_trace_t;
52 :
53 : static u8 *
54 0 : format_virtio_input_trace (u8 * s, va_list * args)
55 : {
56 0 : CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
57 0 : CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
58 0 : virtio_input_trace_t *t = va_arg (*args, virtio_input_trace_t *);
59 0 : u32 indent = format_get_indent (s);
60 :
61 0 : s = format (s, "virtio: hw_if_index %d next-index %d vring %u len %u",
62 0 : t->hw_if_index, t->next_index, t->ring, t->len);
63 0 : s = format (s, "\n%Uhdr: flags 0x%02x gso_type 0x%02x hdr_len %u "
64 : "gso_size %u csum_start %u csum_offset %u num_buffers %u",
65 : format_white_space, indent + 2,
66 0 : t->hdr.flags, t->hdr.gso_type, t->hdr.hdr_len, t->hdr.gso_size,
67 0 : t->hdr.csum_start, t->hdr.csum_offset, t->hdr.num_buffers);
68 0 : return s;
69 : }
70 :
71 : static_always_inline void
72 29224600 : virtio_needs_csum (vlib_buffer_t *b0, vnet_virtio_net_hdr_v1_t *hdr,
73 : u8 *l4_proto, u8 *l4_hdr_sz, virtio_if_type_t type)
74 : {
75 29224600 : if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
76 : {
77 0 : u16 ethertype = 0, l2hdr_sz = 0;
78 0 : vnet_buffer_oflags_t oflags = 0;
79 :
80 0 : if (type == VIRTIO_IF_TYPE_TUN)
81 : {
82 0 : switch (b0->data[0] & 0xf0)
83 : {
84 0 : case 0x40:
85 0 : ethertype = ETHERNET_TYPE_IP4;
86 0 : break;
87 0 : case 0x60:
88 0 : ethertype = ETHERNET_TYPE_IP6;
89 0 : break;
90 : }
91 0 : }
92 : else
93 : {
94 0 : ethernet_header_t *eh = (ethernet_header_t *) b0->data;
95 0 : ethertype = clib_net_to_host_u16 (eh->type);
96 0 : l2hdr_sz = sizeof (ethernet_header_t);
97 :
98 0 : if (ethernet_frame_is_tagged (ethertype))
99 : {
100 0 : ethernet_vlan_header_t *vlan =
101 : (ethernet_vlan_header_t *) (eh + 1);
102 :
103 0 : ethertype = clib_net_to_host_u16 (vlan->type);
104 0 : l2hdr_sz += sizeof (*vlan);
105 0 : if (ethertype == ETHERNET_TYPE_VLAN)
106 : {
107 0 : vlan++;
108 0 : ethertype = clib_net_to_host_u16 (vlan->type);
109 0 : l2hdr_sz += sizeof (*vlan);
110 : }
111 : }
112 : }
113 :
114 0 : vnet_buffer (b0)->l2_hdr_offset = 0;
115 0 : vnet_buffer (b0)->l3_hdr_offset = l2hdr_sz;
116 :
117 0 : if (PREDICT_TRUE (ethertype == ETHERNET_TYPE_IP4))
118 : {
119 0 : ip4_header_t *ip4 = (ip4_header_t *) (b0->data + l2hdr_sz);
120 0 : vnet_buffer (b0)->l4_hdr_offset = l2hdr_sz + ip4_header_bytes (ip4);
121 0 : *l4_proto = ip4->protocol;
122 0 : oflags |= VNET_BUFFER_OFFLOAD_F_IP_CKSUM;
123 0 : b0->flags |=
124 : (VNET_BUFFER_F_IS_IP4 | VNET_BUFFER_F_L2_HDR_OFFSET_VALID |
125 : VNET_BUFFER_F_L3_HDR_OFFSET_VALID |
126 : VNET_BUFFER_F_L4_HDR_OFFSET_VALID);
127 : }
128 0 : else if (PREDICT_TRUE (ethertype == ETHERNET_TYPE_IP6))
129 : {
130 0 : ip6_header_t *ip6 = (ip6_header_t *) (b0->data + l2hdr_sz);
131 0 : vnet_buffer (b0)->l4_hdr_offset = l2hdr_sz + sizeof (ip6_header_t);
132 : /* FIXME IPv6 EH traversal */
133 0 : *l4_proto = ip6->protocol;
134 0 : b0->flags |= (VNET_BUFFER_F_IS_IP6 |
135 : VNET_BUFFER_F_L2_HDR_OFFSET_VALID
136 : | VNET_BUFFER_F_L3_HDR_OFFSET_VALID |
137 : VNET_BUFFER_F_L4_HDR_OFFSET_VALID);
138 : }
139 0 : if (*l4_proto == IP_PROTOCOL_TCP)
140 : {
141 0 : oflags |= VNET_BUFFER_OFFLOAD_F_TCP_CKSUM;
142 0 : tcp_header_t *tcp =
143 0 : (tcp_header_t *) (b0->data + vnet_buffer (b0)->l4_hdr_offset);
144 0 : *l4_hdr_sz = tcp_header_bytes (tcp);
145 : }
146 0 : else if (*l4_proto == IP_PROTOCOL_UDP)
147 : {
148 0 : oflags |= VNET_BUFFER_OFFLOAD_F_UDP_CKSUM;
149 0 : *l4_hdr_sz = sizeof (udp_header_t);
150 : }
151 0 : if (oflags)
152 0 : vnet_buffer_offload_flags_set (b0, oflags);
153 : }
154 29224600 : }
155 :
156 : static_always_inline void
157 29224600 : fill_gso_buffer_flags (vlib_buffer_t *b0, vnet_virtio_net_hdr_v1_t *hdr,
158 : u8 l4_proto, u8 l4_hdr_sz)
159 : {
160 29224600 : if (hdr->gso_type == VIRTIO_NET_HDR_GSO_TCPV4)
161 : {
162 0 : ASSERT (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM);
163 0 : vnet_buffer2 (b0)->gso_size = hdr->gso_size;
164 0 : vnet_buffer2 (b0)->gso_l4_hdr_sz = l4_hdr_sz;
165 0 : b0->flags |= VNET_BUFFER_F_GSO | VNET_BUFFER_F_IS_IP4;
166 : }
167 29224600 : if (hdr->gso_type == VIRTIO_NET_HDR_GSO_TCPV6)
168 : {
169 0 : ASSERT (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM);
170 0 : vnet_buffer2 (b0)->gso_size = hdr->gso_size;
171 0 : vnet_buffer2 (b0)->gso_l4_hdr_sz = l4_hdr_sz;
172 0 : b0->flags |= VNET_BUFFER_F_GSO | VNET_BUFFER_F_IS_IP6;
173 : }
174 29224600 : }
175 :
176 : static_always_inline u16
177 533412000 : virtio_n_left_to_process (vnet_virtio_vring_t *vring, const int packed)
178 : {
179 533412000 : if (packed)
180 0 : return vring->desc_in_use;
181 : else
182 533412000 : return vring->used->idx - vring->last_used_idx;
183 : }
184 :
185 : static_always_inline u16
186 114009000 : virtio_get_slot_id (vnet_virtio_vring_t *vring, const int packed, u16 last,
187 : u16 mask)
188 : {
189 114009000 : if (packed)
190 0 : return vring->packed_desc[last].id;
191 : else
192 114009000 : return vring->used->ring[last & mask].id;
193 : }
194 :
195 : static_always_inline u16
196 114009000 : virtio_get_len (vnet_virtio_vring_t *vring, const int packed, const int hdr_sz,
197 : u16 last, u16 mask)
198 : {
199 114009000 : if (packed)
200 0 : return vring->packed_desc[last].len - hdr_sz;
201 : else
202 114009000 : return vring->used->ring[last & mask].len - hdr_sz;
203 : }
204 :
205 : #define increment_last(last, packed, vring) \
206 : do \
207 : { \
208 : last++; \
209 : if (packed && last >= vring->queue_size) \
210 : { \
211 : last = 0; \
212 : vring->used_wrap_counter ^= 1; \
213 : } \
214 : } \
215 : while (0)
216 :
217 : static_always_inline void
218 1085710 : virtio_device_input_ethernet (vlib_main_t *vm, vlib_node_runtime_t *node,
219 : const u32 next_index, const u32 sw_if_index,
220 : const u32 hw_if_index)
221 : {
222 : vlib_next_frame_t *nf;
223 : vlib_frame_t *f;
224 : ethernet_input_frame_t *ef;
225 :
226 1085710 : if (PREDICT_FALSE (VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT != next_index))
227 127136 : return;
228 :
229 958572 : nf = vlib_node_runtime_get_next_frame (
230 : vm, node, VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT);
231 958572 : f = vlib_get_frame (vm, nf->frame);
232 958572 : f->flags = ETH_INPUT_FRAME_F_SINGLE_SW_IF_IDX;
233 :
234 958572 : ef = vlib_frame_scalar_args (f);
235 958572 : ef->sw_if_index = sw_if_index;
236 958572 : ef->hw_if_index = hw_if_index;
237 958572 : vlib_frame_no_append (f);
238 : }
239 :
240 : static_always_inline uword
241 533412000 : virtio_device_input_gso_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
242 : vlib_frame_t *frame, virtio_if_t *vif,
243 : vnet_virtio_vring_t *vring,
244 : virtio_if_type_t type, int gso_enabled,
245 : int checksum_offload_enabled, int packed)
246 : {
247 533412000 : vnet_main_t *vnm = vnet_get_main ();
248 533412000 : u32 thread_index = vm->thread_index;
249 533412000 : uword n_trace = vlib_get_trace_count (vm, node);
250 : u32 next_index;
251 533412000 : const int hdr_sz = vif->virtio_net_hdr_sz;
252 533412000 : u32 *to_next = 0;
253 533412000 : u32 n_rx_packets = 0;
254 533412000 : u32 n_rx_bytes = 0;
255 533412000 : u16 mask = vring->queue_size - 1;
256 533412000 : u16 last = vring->last_used_idx;
257 533412000 : u16 n_left = virtio_n_left_to_process (vring, packed);
258 533412000 : vlib_buffer_t bt = {};
259 :
260 533412000 : if (n_left == 0)
261 532326000 : return 0;
262 :
263 1085710 : if (type == VIRTIO_IF_TYPE_TUN)
264 : {
265 127136 : next_index = VNET_DEVICE_INPUT_NEXT_IP4_INPUT;
266 : }
267 : else
268 : {
269 958572 : next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
270 958572 : if (PREDICT_FALSE (vif->per_interface_next_index != ~0))
271 0 : next_index = vif->per_interface_next_index;
272 :
273 : /* only for l2, redirect if feature path enabled */
274 958572 : vnet_feature_start_device_input_x1 (vif->sw_if_index, &next_index, &bt);
275 : }
276 :
277 2171420 : while (n_left)
278 : {
279 : u32 n_left_to_next;
280 1085710 : u32 next0 = next_index;
281 :
282 1085710 : vlib_get_new_next_frame (vm, node, next_index, to_next, n_left_to_next);
283 :
284 55433900 : while (n_left && n_left_to_next)
285 : {
286 54348200 : if (packed)
287 : {
288 0 : vnet_virtio_vring_packed_desc_t *d = &vring->packed_desc[last];
289 0 : u16 flags = d->flags;
290 0 : if ((flags & VRING_DESC_F_AVAIL) !=
291 0 : (vring->used_wrap_counter << 7)
292 0 : || (flags & VRING_DESC_F_USED) !=
293 0 : (vring->used_wrap_counter << 15))
294 : {
295 0 : n_left = 0;
296 0 : break;
297 : }
298 : }
299 54348200 : u8 l4_proto = 0, l4_hdr_sz = 0;
300 54348200 : u16 num_buffers = 1;
301 : vnet_virtio_net_hdr_v1_t *hdr;
302 54348200 : u16 slot = virtio_get_slot_id (vring, packed, last, mask);
303 54348200 : u16 len = virtio_get_len (vring, packed, hdr_sz, last, mask);
304 54348200 : u32 bi0 = vring->buffers[slot];
305 54348200 : vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
306 54348200 : hdr = vlib_buffer_get_current (b0);
307 54348200 : if (hdr_sz == sizeof (vnet_virtio_net_hdr_v1_t))
308 54348200 : num_buffers = hdr->num_buffers;
309 :
310 54348200 : b0->flags = VLIB_BUFFER_TOTAL_LENGTH_VALID;
311 54348200 : b0->current_data = 0;
312 54348200 : b0->current_length = len;
313 :
314 54348200 : if (checksum_offload_enabled)
315 29224600 : virtio_needs_csum (b0, hdr, &l4_proto, &l4_hdr_sz, type);
316 :
317 54348200 : if (gso_enabled)
318 29224600 : fill_gso_buffer_flags (b0, hdr, l4_proto, l4_hdr_sz);
319 :
320 54348200 : vnet_buffer (b0)->sw_if_index[VLIB_RX] = vif->sw_if_index;
321 54348200 : vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~ 0;
322 :
323 : /* if multisegment packet */
324 54348200 : if (PREDICT_FALSE (num_buffers > 1))
325 : {
326 : vlib_buffer_t *pb, *cb;
327 29072300 : pb = b0;
328 29072300 : b0->total_length_not_including_first_buffer = 0;
329 88733100 : while (num_buffers > 1)
330 : {
331 59660800 : increment_last (last, packed, vring);
332 59660800 : u16 cslot = virtio_get_slot_id (vring, packed, last, mask);
333 : /* hdr size is 0 after 1st packet in chain buffers */
334 59660800 : u16 clen = virtio_get_len (vring, packed, 0, last, mask);
335 59660800 : u32 cbi = vring->buffers[cslot];
336 59660800 : cb = vlib_get_buffer (vm, cbi);
337 :
338 : /* current buffer */
339 59660800 : cb->current_length = clen;
340 :
341 : /* previous buffer */
342 59660800 : pb->next_buffer = cbi;
343 59660800 : pb->flags |= VLIB_BUFFER_NEXT_PRESENT;
344 :
345 : /* first buffer */
346 59660800 : b0->total_length_not_including_first_buffer += clen;
347 :
348 59660800 : pb = cb;
349 59660800 : vring->desc_in_use--;
350 59660800 : num_buffers--;
351 59660800 : n_left--;
352 : }
353 29072300 : len += b0->total_length_not_including_first_buffer;
354 : }
355 :
356 54348200 : if (type == VIRTIO_IF_TYPE_TUN)
357 : {
358 5528250 : switch (b0->data[0] & 0xf0)
359 : {
360 3080500 : case 0x40:
361 3080500 : next0 = VNET_DEVICE_INPUT_NEXT_IP4_INPUT;
362 3080500 : break;
363 2447750 : case 0x60:
364 2447750 : next0 = VNET_DEVICE_INPUT_NEXT_IP6_INPUT;
365 2447750 : break;
366 0 : default:
367 0 : next0 = VNET_DEVICE_INPUT_NEXT_DROP;
368 0 : break;
369 : }
370 :
371 5528250 : if (PREDICT_FALSE (vif->per_interface_next_index != ~0))
372 0 : next0 = vif->per_interface_next_index;
373 : }
374 : else
375 : {
376 : /* copy feature arc data from template */
377 48819900 : b0->current_config_index = bt.current_config_index;
378 48819900 : vnet_buffer (b0)->feature_arc_index =
379 48819900 : vnet_buffer (&bt)->feature_arc_index;
380 : }
381 :
382 : /* trace */
383 54348200 : if (PREDICT_FALSE (n_trace > 0 && vlib_trace_buffer (vm, node, next0, b0, /* follow_chain */
384 : 1)))
385 : {
386 : virtio_input_trace_t *tr;
387 0 : vlib_set_trace_count (vm, node, --n_trace);
388 0 : tr = vlib_add_trace (vm, node, b0, sizeof (*tr));
389 0 : tr->next_index = next0;
390 0 : tr->hw_if_index = vif->hw_if_index;
391 0 : tr->len = len;
392 0 : clib_memcpy_fast (&tr->hdr, hdr, (hdr_sz == 12) ? 12 : 10);
393 : }
394 :
395 : /* enqueue buffer */
396 54348200 : to_next[0] = bi0;
397 54348200 : vring->desc_in_use--;
398 54348200 : to_next += 1;
399 54348200 : n_left_to_next--;
400 54348200 : n_left--;
401 54348200 : increment_last (last, packed, vring);
402 :
403 : /* only tun interfaces may have different next index */
404 54348200 : if (type == VIRTIO_IF_TYPE_TUN)
405 5528250 : vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
406 : n_left_to_next, bi0, next0);
407 :
408 : /* next packet */
409 54348200 : n_rx_packets++;
410 54348200 : n_rx_bytes += len;
411 : }
412 1085710 : virtio_device_input_ethernet (vm, node, next_index, vif->sw_if_index,
413 : vif->hw_if_index);
414 1085710 : vlib_put_next_frame (vm, node, next_index, n_left_to_next);
415 : }
416 1085710 : vring->last_used_idx = last;
417 :
418 1085710 : vring->total_packets += n_rx_packets;
419 1085710 : vlib_increment_combined_counter (vnm->interface_main.combined_sw_if_counters
420 : + VNET_INTERFACE_COUNTER_RX, thread_index,
421 : vif->sw_if_index, n_rx_packets,
422 : n_rx_bytes);
423 :
424 1085710 : return n_rx_packets;
425 : }
426 :
427 : static_always_inline uword
428 533412000 : virtio_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
429 : vlib_frame_t * frame, virtio_if_t * vif, u16 qid,
430 : virtio_if_type_t type)
431 : {
432 533412000 : vnet_virtio_vring_t *vring = vec_elt_at_index (vif->rxq_vrings, qid);
433 533412000 : const int hdr_sz = vif->virtio_net_hdr_sz;
434 : uword rv;
435 :
436 533412000 : if (vif->is_packed)
437 : {
438 0 : if (vif->gso_enabled)
439 : rv =
440 0 : virtio_device_input_gso_inline (vm, node, frame, vif, vring, type,
441 : 1, 1, 1);
442 0 : else if (vif->csum_offload_enabled)
443 : rv =
444 0 : virtio_device_input_gso_inline (vm, node, frame, vif, vring, type,
445 : 0, 1, 1);
446 : else
447 : rv =
448 0 : virtio_device_input_gso_inline (vm, node, frame, vif, vring, type,
449 : 0, 0, 1);
450 :
451 0 : virtio_refill_vring_packed (vm, vif, type, vring, hdr_sz,
452 : node->node_index);
453 : }
454 : else
455 : {
456 533412000 : if (vif->gso_enabled)
457 : rv =
458 296785000 : virtio_device_input_gso_inline (vm, node, frame, vif, vring, type,
459 : 1, 1, 0);
460 236627000 : else if (vif->csum_offload_enabled)
461 : rv =
462 0 : virtio_device_input_gso_inline (vm, node, frame, vif, vring, type,
463 : 0, 1, 0);
464 : else
465 : rv =
466 236627000 : virtio_device_input_gso_inline (vm, node, frame, vif, vring, type,
467 : 0, 0, 0);
468 :
469 533412000 : virtio_refill_vring_split (vm, vif, type, vring, hdr_sz,
470 : node->node_index);
471 : }
472 533412000 : return rv;
473 : }
474 :
475 320232236 : VLIB_NODE_FN (virtio_input_node) (vlib_main_t * vm,
476 : vlib_node_runtime_t * node,
477 : vlib_frame_t * frame)
478 : {
479 320230000 : u32 n_rx = 0;
480 320230000 : virtio_main_t *vim = &virtio_main;
481 : vnet_hw_if_rxq_poll_vector_t *p,
482 320230000 : *pv = vnet_hw_if_get_rxq_poll_vector (vm, node);
483 :
484 854268000 : vec_foreach (p, pv)
485 : {
486 : virtio_if_t *vif;
487 534038000 : vif = vec_elt_at_index (vim->interfaces, p->dev_instance);
488 534038000 : if (vif->flags & VIRTIO_IF_FLAG_ADMIN_UP)
489 : {
490 533412000 : if (vif->type == VIRTIO_IF_TYPE_TAP)
491 422720000 : n_rx += virtio_device_input_inline (
492 422720000 : vm, node, frame, vif, p->queue_id, VIRTIO_IF_TYPE_TAP);
493 110692000 : else if (vif->type == VIRTIO_IF_TYPE_PCI)
494 0 : n_rx += virtio_device_input_inline (
495 0 : vm, node, frame, vif, p->queue_id, VIRTIO_IF_TYPE_PCI);
496 110692000 : else if (vif->type == VIRTIO_IF_TYPE_TUN)
497 110692000 : n_rx += virtio_device_input_inline (
498 110692000 : vm, node, frame, vif, p->queue_id, VIRTIO_IF_TYPE_TUN);
499 : }
500 : }
501 :
502 320230000 : return n_rx;
503 : }
504 :
505 : /* *INDENT-OFF* */
506 178120 : VLIB_REGISTER_NODE (virtio_input_node) = {
507 : .name = "virtio-input",
508 : .sibling_of = "device-input",
509 : .format_trace = format_virtio_input_trace,
510 : .flags = VLIB_NODE_FLAG_TRACE_SUPPORTED,
511 : .type = VLIB_NODE_TYPE_INPUT,
512 : .state = VLIB_NODE_STATE_INTERRUPT,
513 : .n_errors = VIRTIO_INPUT_N_ERROR,
514 : .error_strings = virtio_input_error_strings,
515 : };
516 : /* *INDENT-ON* */
517 :
518 : /*
519 : * fd.io coding-style-patch-verification: ON
520 : *
521 : * Local Variables:
522 : * eval: (c-set-style "gnu")
523 : * End:
524 : */
|
