FD.io VPP  v17.10-9-gd594711
Vector Packet Processing
device.c
Go to the documentation of this file.
1 /*
2  *------------------------------------------------------------------
3  * af_packet.c - linux kernel packet interface
4  *
5  * Copyright (c) 2016 Cisco and/or its affiliates.
6  * Licensed under the Apache License, Version 2.0 (the "License");
7  * you may not use this file except in compliance with the License.
8  * You may obtain a copy of the License at:
9  *
10  * http://www.apache.org/licenses/LICENSE-2.0
11  *
12  * Unless required by applicable law or agreed to in writing, software
13  * distributed under the License is distributed on an "AS IS" BASIS,
14  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15  * See the License for the specific language governing permissions and
16  * limitations under the License.
17  *------------------------------------------------------------------
18  */
19 
20 #include <linux/if_packet.h>
21 #include <sys/socket.h>
22 #include <sys/ioctl.h>
23 #include <net/if.h>
24 #include <net/if_arp.h>
25 
26 #include <vlib/vlib.h>
27 #include <vlib/unix/unix.h>
28 #include <vnet/ip/ip.h>
29 #include <vnet/ethernet/ethernet.h>
30 
32 
33 #define foreach_af_packet_tx_func_error \
34 _(FRAME_NOT_READY, "tx frame not ready") \
35 _(TXRING_EAGAIN, "tx sendto temporary failure") \
36 _(TXRING_FATAL, "tx sendto fatal failure") \
37 _(TXRING_OVERRUN, "tx ring overrun")
38 
39 typedef enum
40 {
41 #define _(f,s) AF_PACKET_TX_ERROR_##f,
43 #undef _
46 
48 #define _(n,s) s,
50 #undef _
51 };
52 
53 
54 static u8 *
55 format_af_packet_device_name (u8 * s, va_list * args)
56 {
57  u32 i = va_arg (*args, u32);
60 
61  s = format (s, "host-%s", apif->host_if_name);
62  return s;
63 }
64 
65 static u8 *
66 format_af_packet_device (u8 * s, va_list * args)
67 {
68  s = format (s, "Linux PACKET socket interface");
69  return s;
70 }
71 
72 static u8 *
73 format_af_packet_tx_trace (u8 * s, va_list * args)
74 {
75  s = format (s, "Unimplemented...");
76  return s;
77 }
78 
79 static uword
81  vlib_node_runtime_t * node, vlib_frame_t * frame)
82 {
84  u32 *buffers = vlib_frame_args (frame);
85  u32 n_left = frame->n_vectors;
86  u32 n_sent = 0;
87  vnet_interface_output_runtime_t *rd = (void *) node->runtime_data;
88  af_packet_if_t *apif =
90  int block = 0;
91  u32 block_size = apif->tx_req->tp_block_size;
92  u32 frame_size = apif->tx_req->tp_frame_size;
93  u32 frame_num = apif->tx_req->tp_frame_nr;
94  u8 *block_start = apif->tx_ring + block * block_size;
95  u32 tx_frame = apif->next_tx_frame;
96  struct tpacket2_hdr *tph;
97  u32 frame_not_ready = 0;
98 
99  clib_spinlock_lock_if_init (&apif->lockp);
100 
101  while (n_left > 0)
102  {
103  u32 len;
104  u32 offset = 0;
105  vlib_buffer_t *b0;
106  n_left--;
107  u32 bi = buffers[0];
108  buffers++;
109 
110  tph = (struct tpacket2_hdr *) (block_start + tx_frame * frame_size);
111 
112  if (PREDICT_FALSE
113  (tph->tp_status & (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING)))
114  {
115  frame_not_ready++;
116  goto next;
117  }
118 
119  do
120  {
121  b0 = vlib_get_buffer (vm, bi);
122  len = b0->current_length;
123  clib_memcpy ((u8 *) tph +
124  TPACKET_ALIGN (sizeof (struct tpacket2_hdr)) + offset,
125  vlib_buffer_get_current (b0), len);
126  offset += len;
127  }
128  while ((bi =
129  (b0->flags & VLIB_BUFFER_NEXT_PRESENT) ? b0->next_buffer : 0));
130 
131  tph->tp_len = tph->tp_snaplen = offset;
132  tph->tp_status = TP_STATUS_SEND_REQUEST;
133  n_sent++;
134  next:
135  /* check if we've exhausted the ring */
136  if (PREDICT_FALSE (frame_not_ready + n_sent == frame_num))
137  break;
138 
139  tx_frame = (tx_frame + 1) % frame_num;
140  }
141 
143 
144  if (PREDICT_TRUE (n_sent))
145  {
146  apif->next_tx_frame = tx_frame;
147 
148  if (PREDICT_FALSE (sendto (apif->fd, NULL, 0,
149  MSG_DONTWAIT, NULL, 0) == -1))
150  {
151  /* Uh-oh, drop & move on, but count whether it was fatal or not.
152  * Note that we have no reliable way to properly determine the
153  * disposition of the packets we just enqueued for delivery.
154  */
155  vlib_error_count (vm, node->node_index,
156  unix_error_is_fatal (errno) ?
157  AF_PACKET_TX_ERROR_TXRING_FATAL :
158  AF_PACKET_TX_ERROR_TXRING_EAGAIN, n_sent);
159  }
160  }
161 
162  clib_spinlock_unlock_if_init (&apif->lockp);
163 
164  if (PREDICT_FALSE (frame_not_ready))
165  vlib_error_count (vm, node->node_index,
166  AF_PACKET_TX_ERROR_FRAME_NOT_READY, frame_not_ready);
167 
168  if (PREDICT_FALSE (frame_not_ready + n_sent == frame_num))
169  vlib_error_count (vm, node->node_index, AF_PACKET_TX_ERROR_TXRING_OVERRUN,
170  n_left);
171 
172  vlib_buffer_free (vm, vlib_frame_args (frame), frame->n_vectors);
173  return frame->n_vectors;
174 }
175 
176 static void
178  u32 node_index)
179 {
181  vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index);
182  af_packet_if_t *apif =
184 
185  /* Shut off redirection */
186  if (node_index == ~0)
187  {
188  apif->per_interface_next_index = node_index;
189  return;
190  }
191 
194  node_index);
195 }
196 
197 static void
199 {
200  /* Nothing for now */
201 }
202 
203 static clib_error_t *
205  u32 flags)
206 {
208  vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index);
209  af_packet_if_t *apif =
211  u32 hw_flags;
212  int rv, fd = socket (AF_UNIX, SOCK_DGRAM, 0);
213  struct ifreq ifr;
214 
215  if (0 > fd)
216  {
217  clib_unix_warning ("af_packet_%s could not open socket",
218  apif->host_if_name);
219  return 0;
220  }
221 
222  /* if interface is a bridge ignore */
223  if (apif->host_if_index < 0)
224  goto error; /* no error */
225 
226  /* use host_if_index in case host name has changed */
227  ifr.ifr_ifindex = apif->host_if_index;
228  if ((rv = ioctl (fd, SIOCGIFNAME, &ifr)) < 0)
229  {
230  clib_unix_warning ("af_packet_%s ioctl could not retrieve eth name",
231  apif->host_if_name);
232  goto error;
233  }
234 
235  apif->is_admin_up = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) != 0;
236 
237  if ((rv = ioctl (fd, SIOCGIFFLAGS, &ifr)) < 0)
238  {
239  clib_unix_warning ("af_packet_%s error: %d",
240  apif->is_admin_up ? "up" : "down", rv);
241  goto error;
242  }
243 
244  if (apif->is_admin_up)
245  {
247  ifr.ifr_flags |= IFF_UP;
248  }
249  else
250  {
251  hw_flags = 0;
252  ifr.ifr_flags &= ~IFF_UP;
253  }
254 
255  if ((rv = ioctl (fd, SIOCSIFFLAGS, &ifr)) < 0)
256  {
257  clib_unix_warning ("af_packet_%s error: %d",
258  apif->is_admin_up ? "up" : "down", rv);
259  goto error;
260  }
261 
262  vnet_hw_interface_set_flags (vnm, hw_if_index, hw_flags);
263 
264 error:
265  if (0 <= fd)
266  close (fd);
267 
268  return 0; /* no error */
269 }
270 
271 static clib_error_t *
273  u32 hw_if_index,
274  struct vnet_sw_interface_t *st, int is_add)
275 {
276  /* Nothing for now */
277  return 0;
278 }
279 
281  (struct vnet_hw_interface_t *hi, char *address)
282 {
284  af_packet_if_t *apif =
286  int rv, fd = socket (AF_UNIX, SOCK_DGRAM, 0);
287  struct ifreq ifr;
288 
289  if (0 > fd)
290  {
291  clib_unix_warning ("af_packet_%s could not open socket",
292  apif->host_if_name);
293  return 0;
294  }
295 
296  /* if interface is a bridge ignore */
297  if (apif->host_if_index < 0)
298  goto error; /* no error */
299 
300  /* use host_if_index in case host name has changed */
301  ifr.ifr_ifindex = apif->host_if_index;
302  if ((rv = ioctl (fd, SIOCGIFNAME, &ifr)) < 0)
303  {
305  ("af_packet_%s ioctl could not retrieve eth name, error: %d",
306  apif->host_if_name, rv);
307  goto error;
308  }
309 
310  clib_memcpy (ifr.ifr_hwaddr.sa_data, address, 6);
311  ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
312 
313  if ((rv = ioctl (fd, SIOCSIFHWADDR, &ifr)) < 0)
314  {
315  clib_unix_warning ("af_packet_%s ioctl could not set mac, error: %d",
316  apif->host_if_name, rv);
317  goto error;
318  }
319 
320 error:
321 
322  if (0 <= fd)
323  close (fd);
324 
325  return 0; /* no error */
326 }
327 
328 /* *INDENT-OFF* */
330  .name = "af-packet",
331  .tx_function = af_packet_interface_tx,
332  .format_device_name = format_af_packet_device_name,
333  .format_device = format_af_packet_device,
334  .format_tx_trace = format_af_packet_tx_trace,
335  .tx_function_n_errors = AF_PACKET_TX_N_ERROR,
336  .tx_function_error_strings = af_packet_tx_func_error_strings,
337  .rx_redirect_to_node = af_packet_set_interface_next_node,
338  .clear_counters = af_packet_clear_hw_interface_counters,
339  .admin_up_down_function = af_packet_interface_admin_up_down,
340  .subif_add_del_function = af_packet_subif_add_del_function,
341  .mac_addr_change_function = af_packet_set_mac_address_function,
342 };
343 
346 /* *INDENT-ON* */
347 
348 /*
349  * fd.io coding-style-patch-verification: ON
350  *
351  * Local Variables:
352  * eval: (c-set-style "gnu")
353  * End:
354  */
sll srl srl sll sra u16x4 i
Definition: vector_sse2.h:337
clib_error_t * vnet_hw_interface_set_flags(vnet_main_t *vnm, u32 hw_if_index, u32 flags)
Definition: interface.c:538
static void vlib_buffer_free(vlib_main_t *vm, u32 *buffers, u32 n_buffers)
Free buffers Frees the entire buffer chain for each buffer.
Definition: buffer_funcs.h:317
u8 runtime_data[0]
Function dependent node-runtime data.
Definition: node.h:464
#define PREDICT_TRUE(x)
Definition: clib.h:98
static void vlib_error_count(vlib_main_t *vm, uword node_index, uword counter, uword increment)
Definition: error_funcs.h:57
#define NULL
Definition: clib.h:55
static u8 * format_af_packet_tx_trace(u8 *s, va_list *args)
Definition: device.c:73
static_always_inline void clib_spinlock_unlock_if_init(clib_spinlock_t *p)
Definition: lock.h:85
static vnet_hw_interface_t * vnet_get_hw_interface(vnet_main_t *vnm, u32 hw_if_index)
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:419
static void af_packet_set_interface_next_node(vnet_main_t *vnm, u32 hw_if_index, u32 node_index)
Definition: device.c:177
#define VNET_HW_INTERFACE_FLAG_LINK_UP
Definition: interface.h:394
static clib_error_t * af_packet_set_mac_address_function(struct vnet_hw_interface_t *hi, char *address)
Definition: device.c:281
static uword vlib_node_add_next(vlib_main_t *vm, uword node, uword next_node)
Definition: node_funcs.h:1108
VNET_DEVICE_CLASS(af_packet_device_class)
u8 * host_if_name
Definition: af_packet.h:26
af_packet_if_t * interfaces
Definition: af_packet.h:47
#define VLIB_BUFFER_NEXT_PRESENT
Definition: buffer.h:95
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:458
u16 current_length
Nbytes between current data and the end of this buffer.
Definition: buffer.h:72
static void af_packet_clear_hw_interface_counters(u32 instance)
Definition: device.c:198
static void * vlib_buffer_get_current(vlib_buffer_t *b)
Get pointer to current data to process.
Definition: buffer.h:193
#define PREDICT_FALSE(x)
Definition: clib.h:97
static char * af_packet_tx_func_error_strings[]
Definition: device.c:47
static u8 * format_af_packet_device_name(u8 *s, va_list *args)
Definition: device.c:55
u32 node_index
Node index.
Definition: node.h:437
u16 n_vectors
Definition: node.h:344
vlib_main_t * vm
Definition: buffer.c:283
static clib_error_t * af_packet_subif_add_del_function(vnet_main_t *vnm, u32 hw_if_index, struct vnet_sw_interface_t *st, int is_add)
Definition: device.c:272
#define clib_memcpy(a, b, c)
Definition: string.h:69
u32 per_interface_next_index
Definition: af_packet.h:40
vlib_node_registration_t af_packet_input_node
(constructor) VLIB_REGISTER_NODE (af_packet_input_node)
Definition: node.c:289
#define foreach_af_packet_tx_func_error
Definition: device.c:33
int host_if_index
Definition: af_packet.h:27
#define VNET_SW_INTERFACE_FLAG_ADMIN_UP
Definition: interface.h:572
unsigned int u32
Definition: types.h:88
static word unix_error_is_fatal(word error)
Definition: error.h:118
u32 next_buffer
Next buffer for this linked-list of buffers.
Definition: buffer.h:109
af_packet_main_t af_packet_main
Definition: af_packet.h:59
static void * vlib_frame_args(vlib_frame_t *f)
Get pointer to frame scalar data.
Definition: node_funcs.h:284
af_packet_tx_func_error_t
Definition: device.c:39
static vlib_main_t * vlib_get_main(void)
Definition: global_funcs.h:23
vnet_device_class_t af_packet_device_class
u64 uword
Definition: types.h:112
template key/value backing page structure
Definition: bihash_doc.h:44
unsigned char u8
Definition: types.h:56
static clib_error_t * af_packet_interface_admin_up_down(vnet_main_t *vnm, u32 hw_if_index, u32 flags)
Definition: device.c:204
#define clib_unix_warning(format, args...)
Definition: error.h:68
struct clib_bihash_value offset
template key/value backing page structure
#define CLIB_MEMORY_BARRIER()
Definition: clib.h:101
static uword af_packet_interface_tx(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame)
Definition: device.c:80
u32 flags
Definition: vhost-user.h:77
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:75
static_always_inline void clib_spinlock_lock_if_init(clib_spinlock_t *p)
Definition: lock.h:65
static vlib_buffer_t * vlib_get_buffer(vlib_main_t *vm, u32 buffer_index)
Translate buffer index into buffer pointer.
Definition: buffer_funcs.h:57
static u8 * format_af_packet_device(u8 *s, va_list *args)
Definition: device.c:66
#define VLIB_DEVICE_TX_FUNCTION_MULTIARCH(dev, fn)
Definition: interface.h:232