node.c revision efa119db
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/devices/devices.h>
30#include <vnet/feature/feature.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/devices/virtio/virtio.h>
35
36
37#define foreach_virtio_input_error \
38  _(UNKNOWN, "unknown")
39
40typedef enum
41{
42#define _(f,s) VIRTIO_INPUT_ERROR_##f,
43  foreach_virtio_input_error
44#undef _
45    VIRTIO_INPUT_N_ERROR,
46} virtio_input_error_t;
47
48static char *virtio_input_error_strings[] = {
49#define _(n,s) s,
50  foreach_virtio_input_error
51#undef _
52};
53
54typedef struct
55{
56  u32 next_index;
57  u32 hw_if_index;
58  u16 ring;
59  u16 len;
60  struct virtio_net_hdr_v1 hdr;
61} virtio_input_trace_t;
62
63static u8 *
64format_virtio_input_trace (u8 * s, va_list * args)
65{
66  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
67  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
68  virtio_input_trace_t *t = va_arg (*args, virtio_input_trace_t *);
69  u32 indent = format_get_indent (s);
70
71  s = format (s, "virtio: hw_if_index %d next-index %d vring %u len %u",
72	      t->hw_if_index, t->next_index, t->ring, t->len);
73  s = format (s, "\n%Uhdr: flags 0x%02x gso_type 0x%02x hdr_len %u "
74	      "gso_size %u csum_start %u csum_offset %u num_buffers %u",
75	      format_white_space, indent + 2,
76	      t->hdr.flags, t->hdr.gso_type, t->hdr.hdr_len, t->hdr.gso_size,
77	      t->hdr.csum_start, t->hdr.csum_offset, t->hdr.num_buffers);
78  return s;
79}
80
81static_always_inline void
82virtio_refill_vring (vlib_main_t * vm, virtio_if_t * vif,
83		     virtio_vring_t * vring, const int hdr_sz)
84{
85  u16 used, next, avail, n_slots;
86  u16 sz = vring->size;
87  u16 mask = sz - 1;
88
89more:
90  used = vring->desc_in_use;
91
92  if (sz - used < sz / 8)
93    return;
94
95  /* deliver free buffers in chunks of 64 */
96  n_slots = clib_min (sz - used, 64);
97
98  next = vring->desc_next;
99  avail = vring->avail->idx;
100  n_slots =
101    vlib_buffer_alloc_to_ring_from_pool (vm, vring->buffers, next,
102					 vring->size, n_slots,
103					 vring->buffer_pool_index);
104
105  if (n_slots == 0)
106    return;
107
108  while (n_slots)
109    {
110      struct vring_desc *d = &vring->desc[next];;
111      vlib_buffer_t *b = vlib_get_buffer (vm, vring->buffers[next]);
112      /*
113       * current_data may not be initialized with 0 and may contain
114       * previous offset. Here we want to make sure, it should be 0
115       * initialized.
116       */
117      b->current_data = 0;
118      b->current_data -= hdr_sz;
119      memset (vlib_buffer_get_current (b), 0, hdr_sz);
120      d->addr =
121	((vif->type == VIRTIO_IF_TYPE_PCI) ? vlib_buffer_get_current_pa (vm,
122									 b) :
123	 pointer_to_uword (vlib_buffer_get_current (b)));
124      d->len = vlib_buffer_get_default_data_size (vm) + hdr_sz;
125      d->flags = VRING_DESC_F_WRITE;
126      vring->avail->ring[avail & mask] = next;
127      avail++;
128      next = (next + 1) & mask;
129      n_slots--;
130      used++;
131    }
132  CLIB_MEMORY_STORE_BARRIER ();
133  vring->avail->idx = avail;
134  vring->desc_next = next;
135  vring->desc_in_use = used;
136
137  if ((vring->used->flags & VIRTIO_RING_FLAG_MASK_INT) == 0)
138    {
139      virtio_kick (vm, vring, vif);
140    }
141  goto more;
142}
143
144static_always_inline void
145fill_gso_buffer_flags (vlib_buffer_t * b0, struct virtio_net_hdr_v1 *hdr)
146{
147  u8 l4_proto = 0;
148  u8 l4_hdr_sz = 0;
149  if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
150
151    {
152      ethernet_header_t *eh = (ethernet_header_t *) b0->data;
153      u16 ethertype = clib_net_to_host_u16 (eh->type);
154      u16 l2hdr_sz = sizeof (ethernet_header_t);
155
156      if (ethernet_frame_is_tagged (ethertype))
157	{
158	  ethernet_vlan_header_t *vlan = (ethernet_vlan_header_t *) (eh + 1);
159
160	  ethertype = clib_net_to_host_u16 (vlan->type);
161	  l2hdr_sz += sizeof (*vlan);
162	  if (ethertype == ETHERNET_TYPE_VLAN)
163	    {
164	      vlan++;
165	      ethertype = clib_net_to_host_u16 (vlan->type);
166	      l2hdr_sz += sizeof (*vlan);
167	    }
168	}
169
170      vnet_buffer (b0)->l2_hdr_offset = 0;
171      vnet_buffer (b0)->l3_hdr_offset = l2hdr_sz;
172      if (PREDICT_TRUE (ethertype == ETHERNET_TYPE_IP4))
173	{
174	  ip4_header_t *ip4 = (ip4_header_t *) (b0->data + l2hdr_sz);
175	  vnet_buffer (b0)->l4_hdr_offset = l2hdr_sz + ip4_header_bytes (ip4);
176	  l4_proto = ip4->protocol;
177	  b0->flags |=
178	    (VNET_BUFFER_F_IS_IP4 | VNET_BUFFER_F_L2_HDR_OFFSET_VALID
179	     | VNET_BUFFER_F_L3_HDR_OFFSET_VALID |
180	     VNET_BUFFER_F_L4_HDR_OFFSET_VALID);
181	  b0->flags |= VNET_BUFFER_F_OFFLOAD_IP_CKSUM;
182	}
183      else if (PREDICT_TRUE (ethertype == ETHERNET_TYPE_IP6))
184	{
185	  ip6_header_t *ip6 = (ip6_header_t *) (b0->data + l2hdr_sz);
186	  /* FIXME IPv6 EH traversal */
187	  vnet_buffer (b0)->l4_hdr_offset = l2hdr_sz + sizeof (ip6_header_t);
188	  l4_proto = ip6->protocol;
189	  b0->flags |=
190	    (VNET_BUFFER_F_IS_IP6 | VNET_BUFFER_F_L2_HDR_OFFSET_VALID
191	     | VNET_BUFFER_F_L3_HDR_OFFSET_VALID |
192	     VNET_BUFFER_F_L4_HDR_OFFSET_VALID);
193	}
194      if (l4_proto == IP_PROTOCOL_TCP)
195	{
196	  b0->flags |= VNET_BUFFER_F_OFFLOAD_TCP_CKSUM;
197	  tcp_header_t *tcp = (tcp_header_t *) (b0->data +
198						vnet_buffer
199						(b0)->l4_hdr_offset);
200	  l4_hdr_sz = tcp_header_bytes (tcp);
201	  tcp->checksum = 0;
202	}
203      else if (l4_proto == IP_PROTOCOL_UDP)
204	{
205	  b0->flags |= VNET_BUFFER_F_OFFLOAD_UDP_CKSUM;
206	  udp_header_t *udp = (udp_header_t *) (b0->data +
207						vnet_buffer
208						(b0)->l4_hdr_offset);
209	  l4_hdr_sz = sizeof (*udp);
210	  udp->checksum = 0;
211	}
212    }
213
214  if (hdr->gso_type == VIRTIO_NET_HDR_GSO_TCPV4)
215    {
216      ASSERT (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM);
217      vnet_buffer2 (b0)->gso_size = hdr->gso_size;
218      vnet_buffer2 (b0)->gso_l4_hdr_sz = l4_hdr_sz;
219      b0->flags |= VNET_BUFFER_F_GSO;
220      b0->flags |= VNET_BUFFER_F_IS_IP4;
221    }
222  if (hdr->gso_type == VIRTIO_NET_HDR_GSO_TCPV6)
223    {
224      ASSERT (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM);
225      vnet_buffer2 (b0)->gso_size = hdr->gso_size;
226      vnet_buffer2 (b0)->gso_l4_hdr_sz = l4_hdr_sz;
227      b0->flags |= VNET_BUFFER_F_GSO;
228      b0->flags |= VNET_BUFFER_F_IS_IP6;
229    }
230}
231
232
233static_always_inline uword
234virtio_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
235			    vlib_frame_t * frame, virtio_if_t * vif, u16 qid,
236			    int gso_enabled)
237{
238  vnet_main_t *vnm = vnet_get_main ();
239  u32 thread_index = vm->thread_index;
240  uword n_trace = vlib_get_trace_count (vm, node);
241  virtio_vring_t *vring = vec_elt_at_index (vif->rxq_vrings, qid);
242  u32 next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
243  const int hdr_sz = vif->virtio_net_hdr_sz;
244  u32 *to_next = 0;
245  u32 n_rx_packets = 0;
246  u32 n_rx_bytes = 0;
247  u16 mask = vring->size - 1;
248  u16 last = vring->last_used_idx;
249  u16 n_left = vring->used->idx - last;
250
251  if ((vring->used->flags & VIRTIO_RING_FLAG_MASK_INT) == 0 &&
252      vring->last_kick_avail_idx != vring->avail->idx)
253    virtio_kick (vm, vring, vif);
254
255  if (n_left == 0)
256    goto refill;
257
258  while (n_left)
259    {
260      u32 n_left_to_next;
261      u32 next0 = next_index;
262      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
263
264      while (n_left && n_left_to_next)
265	{
266	  u16 num_buffers = 1;
267	  struct vring_used_elem *e = &vring->used->ring[last & mask];
268	  struct virtio_net_hdr_v1 *hdr;
269	  u16 slot = e->id;
270	  u16 len = e->len - hdr_sz;
271	  u32 bi0 = vring->buffers[slot];
272	  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
273	  hdr = vlib_buffer_get_current (b0);
274	  if (hdr_sz == sizeof (struct virtio_net_hdr_v1))
275	    num_buffers = hdr->num_buffers;
276
277	  b0->current_data += hdr_sz;
278	  b0->current_length = len;
279	  b0->total_length_not_including_first_buffer = 0;
280	  b0->flags = VLIB_BUFFER_TOTAL_LENGTH_VALID;
281
282	  if (gso_enabled)
283	    fill_gso_buffer_flags (b0, hdr);
284
285	  vnet_buffer (b0)->sw_if_index[VLIB_RX] = vif->sw_if_index;
286	  vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~ 0;
287
288	  /* if multisegment packet */
289	  if (PREDICT_FALSE (num_buffers > 1))
290	    {
291	      vlib_buffer_t *pb, *cb;
292	      pb = b0;
293	      while (num_buffers > 1)
294		{
295		  last++;
296		  e = &vring->used->ring[last & mask];
297		  u32 cbi = vring->buffers[e->id];
298		  cb = vlib_get_buffer (vm, cbi);
299
300		  /* current buffer */
301		  cb->current_length = e->len;
302
303		  /* previous buffer */
304		  pb->next_buffer = cbi;
305		  pb->flags |= VLIB_BUFFER_NEXT_PRESENT;
306
307		  /* first buffer */
308		  b0->total_length_not_including_first_buffer += e->len;
309
310		  pb = cb;
311		  vring->desc_in_use--;
312		  num_buffers--;
313		  n_left--;
314		}
315	    }
316
317	  if (PREDICT_FALSE (vif->per_interface_next_index != ~0))
318	    next0 = vif->per_interface_next_index;
319	  else
320	    /* redirect if feature path enabled */
321	    vnet_feature_start_device_input_x1 (vif->sw_if_index, &next0, b0);
322	  /* trace */
323	  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0);
324
325	  if (PREDICT_FALSE (n_trace > 0))
326	    {
327	      virtio_input_trace_t *tr;
328	      vlib_trace_buffer (vm, node, next0, b0,
329				 /* follow_chain */ 1);
330	      vlib_set_trace_count (vm, node, --n_trace);
331	      tr = vlib_add_trace (vm, node, b0, sizeof (*tr));
332	      tr->next_index = next0;
333	      tr->hw_if_index = vif->hw_if_index;
334	      tr->len = len + b0->total_length_not_including_first_buffer;
335	      clib_memcpy_fast (&tr->hdr, hdr, hdr_sz);
336	    }
337
338	  /* enqueue buffer */
339	  to_next[0] = bi0;
340	  vring->desc_in_use--;
341	  to_next += 1;
342	  n_left_to_next--;
343	  n_left--;
344	  last++;
345
346	  /* enqueue */
347	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
348					   n_left_to_next, bi0, next0);
349
350	  /* next packet */
351	  n_rx_packets++;
352	  n_rx_bytes += (len + b0->total_length_not_including_first_buffer);
353	}
354      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
355    }
356  vring->last_used_idx = last;
357
358  vlib_increment_combined_counter (vnm->interface_main.combined_sw_if_counters
359				   + VNET_INTERFACE_COUNTER_RX, thread_index,
360				   vif->sw_if_index, n_rx_packets,
361				   n_rx_bytes);
362
363refill:
364  virtio_refill_vring (vm, vif, vring, hdr_sz);
365
366  return n_rx_packets;
367}
368
369VLIB_NODE_FN (virtio_input_node) (vlib_main_t * vm,
370				  vlib_node_runtime_t * node,
371				  vlib_frame_t * frame)
372{
373  u32 n_rx = 0;
374  virtio_main_t *nm = &virtio_main;
375  vnet_device_input_runtime_t *rt = (void *) node->runtime_data;
376  vnet_device_and_queue_t *dq;
377
378  foreach_device_and_queue (dq, rt->devices_and_queues)
379  {
380    virtio_if_t *vif;
381    vif = vec_elt_at_index (nm->interfaces, dq->dev_instance);
382    if (vif->flags & VIRTIO_IF_FLAG_ADMIN_UP)
383      {
384	if (vif->gso_enabled)
385	  n_rx += virtio_device_input_inline (vm, node, frame, vif,
386					      dq->queue_id, 1);
387	else
388	  n_rx += virtio_device_input_inline (vm, node, frame, vif,
389					      dq->queue_id, 0);
390      }
391  }
392
393  return n_rx;
394}
395
396/* *INDENT-OFF* */
397VLIB_REGISTER_NODE (virtio_input_node) = {
398  .name = "virtio-input",
399  .sibling_of = "device-input",
400  .format_trace = format_virtio_input_trace,
401  .type = VLIB_NODE_TYPE_INPUT,
402  .state = VLIB_NODE_STATE_INTERRUPT,
403  .n_errors = VIRTIO_INPUT_N_ERROR,
404  .error_strings = virtio_input_error_strings,
405};
406/* *INDENT-ON* */
407
408/*
409 * fd.io coding-style-patch-verification: ON
410 *
411 * Local Variables:
412 * eval: (c-set-style "gnu")
413 * End:
414 */
415