FD.io VPP  v19.04.4-rc0-5-ge88582fac
Vector Packet Processing
node.c
Go to the documentation of this file.
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>
35 
36 
37 #define foreach_virtio_input_error \
38  _(UNKNOWN, "unknown")
39 
40 typedef enum
41 {
42 #define _(f,s) VIRTIO_INPUT_ERROR_##f,
44 #undef _
47 
48 static char *virtio_input_error_strings[] = {
49 #define _(n,s) s,
51 #undef _
52 };
53 
54 typedef struct
55 {
60  struct virtio_net_hdr_v1 hdr;
62 
63 static u8 *
64 format_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 
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 
89 more:
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 =
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 =
122  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  }
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 
145 fill_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  {
153  u16 ethertype = clib_net_to_host_u16 (eh->type);
154  u16 l2hdr_sz = sizeof (ethernet_header_t);
155 
156  vnet_buffer (b0)->l2_hdr_offset = 0;
157  vnet_buffer (b0)->l3_hdr_offset = l2hdr_sz;
158  if (PREDICT_TRUE (ethertype == ETHERNET_TYPE_IP4))
159  {
160  ip4_header_t *ip4 = (ip4_header_t *) (b0->data + l2hdr_sz);
161  vnet_buffer (b0)->l4_hdr_offset = l2hdr_sz + ip4_header_bytes (ip4);
162  l4_proto = ip4->protocol;
163  b0->flags |=
164  (VNET_BUFFER_F_IS_IP4 | VNET_BUFFER_F_L2_HDR_OFFSET_VALID
165  | VNET_BUFFER_F_L3_HDR_OFFSET_VALID |
166  VNET_BUFFER_F_L4_HDR_OFFSET_VALID);
167  b0->flags |= VNET_BUFFER_F_OFFLOAD_IP_CKSUM;
168  }
169  else if (PREDICT_TRUE (ethertype == ETHERNET_TYPE_IP6))
170  {
171  ip6_header_t *ip6 = (ip6_header_t *) (b0->data + l2hdr_sz);
172  /* FIXME IPv6 EH traversal */
173  vnet_buffer (b0)->l4_hdr_offset = l2hdr_sz + sizeof (ip6_header_t);
174  l4_proto = ip6->protocol;
175  b0->flags |=
176  (VNET_BUFFER_F_IS_IP6 | VNET_BUFFER_F_L2_HDR_OFFSET_VALID
177  | VNET_BUFFER_F_L3_HDR_OFFSET_VALID |
178  VNET_BUFFER_F_L4_HDR_OFFSET_VALID);
179  b0->flags |= VNET_BUFFER_F_OFFLOAD_IP_CKSUM;
180  }
181  if (l4_proto == IP_PROTOCOL_TCP)
182  {
183  b0->flags |= VNET_BUFFER_F_OFFLOAD_TCP_CKSUM;
184  tcp_header_t *tcp = (tcp_header_t *) (b0->data +
186  (b0)->l4_hdr_offset);
187  l4_hdr_sz = tcp_header_bytes (tcp);
188  tcp->checksum = 0;
189  }
190  else if (l4_proto == IP_PROTOCOL_UDP)
191  {
192  b0->flags |= VNET_BUFFER_F_OFFLOAD_UDP_CKSUM;
193  udp_header_t *udp = (udp_header_t *) (b0->data +
195  (b0)->l4_hdr_offset);
196  l4_hdr_sz = sizeof (*udp);
197  udp->checksum = 0;
198  }
199  }
200 
201  if (hdr->gso_type == VIRTIO_NET_HDR_GSO_TCPV4)
202  {
203  ASSERT (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM);
204  vnet_buffer2 (b0)->gso_size = hdr->gso_size;
205  vnet_buffer2 (b0)->gso_l4_hdr_sz = l4_hdr_sz;
206  b0->flags |= VNET_BUFFER_F_GSO;
207  b0->flags |= VNET_BUFFER_F_IS_IP4;
208  }
209  if (hdr->gso_type == VIRTIO_NET_HDR_GSO_TCPV6)
210  {
211  ASSERT (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM);
212  vnet_buffer2 (b0)->gso_size = hdr->gso_size;
213  vnet_buffer2 (b0)->gso_l4_hdr_sz = l4_hdr_sz;
214  b0->flags |= VNET_BUFFER_F_GSO;
215  b0->flags |= VNET_BUFFER_F_IS_IP6;
216  }
217 }
218 
219 
222  vlib_frame_t * frame, virtio_if_t * vif, u16 qid,
223  int gso_enabled)
224 {
225  vnet_main_t *vnm = vnet_get_main ();
226  u32 thread_index = vm->thread_index;
227  uword n_trace = vlib_get_trace_count (vm, node);
228  virtio_vring_t *vring = vec_elt_at_index (vif->rxq_vrings, qid);
230  const int hdr_sz = vif->virtio_net_hdr_sz;
231  u32 *to_next = 0;
232  u32 n_rx_packets = 0;
233  u32 n_rx_bytes = 0;
234  u16 mask = vring->size - 1;
235  u16 last = vring->last_used_idx;
236  u16 n_left = vring->used->idx - last;
237 
238  if ((vring->used->flags & VIRTIO_RING_FLAG_MASK_INT) == 0 &&
239  vring->last_kick_avail_idx != vring->avail->idx)
240  virtio_kick (vm, vring, vif);
241 
242  if (n_left == 0)
243  goto refill;
244 
245  while (n_left)
246  {
247  u32 n_left_to_next;
248  u32 next0 = next_index;
249  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
250 
251  while (n_left && n_left_to_next)
252  {
253  u16 num_buffers = 1;
254  struct vring_used_elem *e = &vring->used->ring[last & mask];
255  struct virtio_net_hdr_v1 *hdr;
256  u16 slot = e->id;
257  u16 len = e->len - hdr_sz;
258  u32 bi0 = vring->buffers[slot];
259  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
260  hdr = vlib_buffer_get_current (b0);
261  if (hdr_sz == sizeof (struct virtio_net_hdr_v1))
262  num_buffers = hdr->num_buffers;
263 
264  b0->current_data += hdr_sz;
265  b0->current_length = len;
267  b0->flags = VLIB_BUFFER_TOTAL_LENGTH_VALID;
268 
269  if (gso_enabled)
270  fill_gso_buffer_flags (b0, hdr);
271 
272  vnet_buffer (b0)->sw_if_index[VLIB_RX] = vif->sw_if_index;
273  vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~ 0;
274 
275  /* if multisegment packet */
276  if (PREDICT_FALSE (num_buffers > 1))
277  {
278  vlib_buffer_t *pb, *cb;
279  pb = b0;
280  while (num_buffers > 1)
281  {
282  last++;
283  e = &vring->used->ring[last & mask];
284  u32 cbi = vring->buffers[e->id];
285  cb = vlib_get_buffer (vm, cbi);
286 
287  /* current buffer */
288  cb->current_length = e->len;
289 
290  /* previous buffer */
291  pb->next_buffer = cbi;
292  pb->flags |= VLIB_BUFFER_NEXT_PRESENT;
293 
294  /* first buffer */
296 
297  pb = cb;
298  vring->desc_in_use--;
299  num_buffers--;
300  n_left--;
301  }
302  }
303 
304  if (PREDICT_FALSE (vif->per_interface_next_index != ~0))
305  next0 = vif->per_interface_next_index;
306  else
307  /* redirect if feature path enabled */
309  /* trace */
311 
312  if (PREDICT_FALSE (n_trace > 0))
313  {
315  vlib_trace_buffer (vm, node, next0, b0,
316  /* follow_chain */ 0);
317  vlib_set_trace_count (vm, node, --n_trace);
318  tr = vlib_add_trace (vm, node, b0, sizeof (*tr));
319  tr->next_index = next0;
320  tr->hw_if_index = vif->hw_if_index;
321  tr->len = len;
322  clib_memcpy_fast (&tr->hdr, hdr, hdr_sz);
323  }
324 
325  /* enqueue buffer */
326  to_next[0] = bi0;
327  vring->desc_in_use--;
328  to_next += 1;
329  n_left_to_next--;
330  n_left--;
331  last++;
332 
333  /* enqueue */
334  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
335  n_left_to_next, bi0, next0);
336 
337  /* next packet */
338  n_rx_packets++;
339  n_rx_bytes += len;
340  }
341  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
342  }
343  vring->last_used_idx = last;
344 
346  + VNET_INTERFACE_COUNTER_RX, thread_index,
347  vif->sw_if_index, n_rx_packets,
348  n_rx_bytes);
349 
350 refill:
351  virtio_refill_vring (vm, vif, vring, hdr_sz);
352 
353  return n_rx_packets;
354 }
355 
357  vlib_node_runtime_t * node,
358  vlib_frame_t * frame)
359 {
360  u32 n_rx = 0;
361  virtio_main_t *nm = &virtio_main;
362  vnet_device_input_runtime_t *rt = (void *) node->runtime_data;
364 
366  {
367  virtio_if_t *vif;
368  vif = vec_elt_at_index (nm->interfaces, dq->dev_instance);
369  if (vif->flags & VIRTIO_IF_FLAG_ADMIN_UP)
370  {
371  if (vif->gso_enabled)
372  n_rx += virtio_device_input_inline (vm, node, frame, vif,
373  dq->queue_id, 1);
374  else
375  n_rx += virtio_device_input_inline (vm, node, frame, vif,
376  dq->queue_id, 0);
377  }
378  }
379 
380  return n_rx;
381 }
382 
383 /* *INDENT-OFF* */
385  .name = "virtio-input",
386  .sibling_of = "device-input",
387  .format_trace = format_virtio_input_trace,
388  .type = VLIB_NODE_TYPE_INPUT,
389  .state = VLIB_NODE_STATE_INTERRUPT,
390  .n_errors = VIRTIO_INPUT_N_ERROR,
391  .error_strings = virtio_input_error_strings,
392 };
393 /* *INDENT-ON* */
394 
395 /*
396  * fd.io coding-style-patch-verification: ON
397  *
398  * Local Variables:
399  * eval: (c-set-style "gnu")
400  * End:
401  */
u32 per_interface_next_index
Definition: virtio.h:150
u32 flags
buffer flags: VLIB_BUFFER_FREE_LIST_INDEX_MASK: bits used to store free list index, VLIB_BUFFER_IS_TRACED: trace this buffer.
Definition: buffer.h:124
struct vring_used * used
Definition: virtio.h:105
static uword vlib_buffer_get_current_pa(vlib_main_t *vm, vlib_buffer_t *b)
Definition: buffer_funcs.h:427
vlib_node_registration_t virtio_input_node
(constructor) VLIB_REGISTER_NODE (virtio_input_node)
Definition: node.c:384
virtio_if_t * interfaces
Definition: virtio.h:191
vnet_device_and_queue_t * devices_and_queues
Definition: devices.h:69
#define clib_min(x, y)
Definition: clib.h:295
#define CLIB_UNUSED(x)
Definition: clib.h:82
static u32 vlib_get_trace_count(vlib_main_t *vm, vlib_node_runtime_t *rt)
Definition: trace_funcs.h:156
static void vlib_increment_combined_counter(vlib_combined_counter_main_t *cm, u32 thread_index, u32 index, u64 n_packets, u64 n_bytes)
Increment a combined counter.
Definition: counter.h:220
vnet_main_t * vnet_get_main(void)
Definition: misc.c:47
#define vnet_buffer2(b)
Definition: buffer.h:428
vnet_interface_main_t interface_main
Definition: vnet.h:56
#define PREDICT_TRUE(x)
Definition: clib.h:112
i16 current_data
signed offset in data[], pre_data[] that we are currently processing.
Definition: buffer.h:110
int gso_enabled
Definition: virtio.h:181
#define CLIB_MEMORY_STORE_BARRIER()
Definition: clib.h:118
u32 dev_instance
Definition: virtio.h:139
#define clib_memcpy_fast(a, b, c)
Definition: string.h:81
u32 thread_index
Definition: main.h:197
u16 current_length
Nbytes between current data and the end of this buffer.
Definition: buffer.h:113
u8 data[0]
Packet data.
Definition: buffer.h:181
static heap_elt_t * last(heap_header_t *h)
Definition: heap.c:53
static u32 format_get_indent(u8 *s)
Definition: format.h:72
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:424
#define VLIB_NODE_FN(node)
Definition: node.h:201
static u32 vlib_buffer_alloc_to_ring_from_pool(vlib_main_t *vm, u32 *ring, u32 start, u32 ring_size, u32 n_buffers, u8 buffer_pool_index)
Allocate buffers into ring from specific buffer pool.
Definition: buffer_funcs.h:657
struct _tcp_header tcp_header_t
unsigned char u8
Definition: types.h:56
static void vlib_trace_buffer(vlib_main_t *vm, vlib_node_runtime_t *r, u32 next_index, vlib_buffer_t *b, int follow_chain)
Definition: trace_funcs.h:114
struct vring_avail * avail
Definition: virtio.h:106
#define static_always_inline
Definition: clib.h:99
u32 hw_if_index
Definition: virtio.h:140
static_always_inline void fill_gso_buffer_flags(vlib_buffer_t *b0, struct virtio_net_hdr_v1 *hdr)
Definition: node.c:145
vlib_combined_counter_main_t * combined_sw_if_counters
Definition: interface.h:824
u8 * format_white_space(u8 *s, va_list *va)
Definition: std-formats.c:113
#define vec_elt_at_index(v, i)
Get vector value at index i checking that i is in bounds.
unsigned int u32
Definition: types.h:88
static u8 * format_virtio_input_trace(u8 *s, va_list *args)
Definition: node.c:64
static char * virtio_input_error_strings[]
Definition: node.c:48
#define foreach_virtio_input_error
Definition: node.c:37
virtio_input_error_t
Definition: node.c:40
unsigned short u16
Definition: types.h:57
static void * vlib_buffer_get_current(vlib_buffer_t *b)
Get pointer to current data to process.
Definition: buffer.h:229
#define PREDICT_FALSE(x)
Definition: clib.h:111
u8 buffer_pool_index
Definition: virtio.h:112
#define vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next, n_left_to_next, bi0, next0)
Finish enqueueing one buffer forward in the graph.
Definition: buffer_node.h:218
#define vlib_get_next_frame(vm, node, next_index, vectors, n_vectors_left)
Get pointer to next frame vector data by (vlib_node_runtime_t, next_index).
Definition: node_funcs.h:338
u8 len
Definition: ip_types.api:49
u16 desc_next
Definition: virtio.h:109
u16 virtio_net_hdr_sz
Definition: virtio.h:143
static_always_inline u32 vlib_buffer_get_default_data_size(vlib_main_t *vm)
Definition: buffer_funcs.h:96
virtio_vring_t * rxq_vrings
Definition: virtio.h:161
u16 last_used_idx
Definition: virtio.h:118
#define VLIB_REGISTER_NODE(x,...)
Definition: node.h:169
vlib_main_t * vm
Definition: buffer.c:312
#define VIRTIO_RING_FLAG_MASK_INT
Definition: virtio.h:99
u32 flags
Definition: virtio.h:137
u16 last_kick_avail_idx
Definition: virtio.h:119
void vlib_put_next_frame(vlib_main_t *vm, vlib_node_runtime_t *r, u32 next_index, u32 n_vectors_left)
Release pointer to next frame vector data.
Definition: main.c:458
virtio_if_type_t type
Definition: virtio.h:144
#define ASSERT(truth)
struct virtio_net_hdr_v1 hdr
Definition: node.c:60
static uword pointer_to_uword(const void *p)
Definition: types.h:131
virtio_main_t virtio_main
Definition: virtio.c:37
static void * vlib_add_trace(vlib_main_t *vm, vlib_node_runtime_t *r, vlib_buffer_t *b, u32 n_data_bytes)
Definition: trace_funcs.h:57
#define foreach_device_and_queue(var, vec)
Definition: devices.h:161
Definition: defs.h:47
u32 next_buffer
Next buffer for this linked-list of buffers.
Definition: buffer.h:140
#define VLIB_BUFFER_TRACE_TRAJECTORY_INIT(b)
Definition: buffer.h:451
VLIB buffer representation.
Definition: buffer.h:102
u64 uword
Definition: types.h:112
static_always_inline uword virtio_device_input_inline(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame, virtio_if_t *vif, u16 qid, int gso_enabled)
Definition: node.c:221
u32 * buffers
Definition: virtio.h:117
#define vnet_buffer(b)
Definition: buffer.h:369
u32 sw_if_index
Definition: virtio.h:141
static_always_inline void vnet_feature_start_device_input_x1(u32 sw_if_index, u32 *next0, vlib_buffer_t *b0)
Definition: feature.h:308
static int tcp_header_bytes(tcp_header_t *t)
Definition: tcp_packet.h:93
static_always_inline void virtio_refill_vring(vlib_main_t *vm, virtio_if_t *vif, virtio_vring_t *vring, const int hdr_sz)
Definition: node.c:82
struct vring_desc * desc
Definition: virtio.h:104
static int ip4_header_bytes(const ip4_header_t *i)
Definition: ip4_packet.h:235
static void vlib_set_trace_count(vlib_main_t *vm, vlib_node_runtime_t *rt, u32 count)
Definition: trace_funcs.h:172
u32 total_length_not_including_first_buffer
Only valid for first buffer in chain.
Definition: buffer.h:167
static vlib_buffer_t * vlib_get_buffer(vlib_main_t *vm, u32 buffer_index)
Translate buffer index into buffer pointer.
Definition: buffer_funcs.h:85
u16 desc_in_use
Definition: virtio.h:108
Definition: defs.h:46
static_always_inline void virtio_kick(vlib_main_t *vm, virtio_vring_t *vring, virtio_if_t *vif)
Definition: virtio.h:216