FD.io VPP  v16.12-rc0-308-g931be3a
Vector Packet Processing
af_packet.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_ether.h>
21 #include <linux/if_packet.h>
22 
23 #include <vlib/vlib.h>
24 #include <vlib/unix/unix.h>
25 #include <vnet/ip/ip.h>
26 #include <vnet/ethernet/ethernet.h>
27 
29 
30 #define AF_PACKET_DEBUG_SOCKET 0
31 
32 #define AF_PACKET_TX_FRAMES_PER_BLOCK 1024
33 #define AF_PACKET_TX_FRAME_SIZE (2048 * 5)
34 #define AF_PACKET_TX_BLOCK_NR 1
35 #define AF_PACKET_TX_FRAME_NR (AF_PACKET_TX_BLOCK_NR * \
36  AF_PACKET_TX_FRAMES_PER_BLOCK)
37 #define AF_PACKET_TX_BLOCK_SIZE (AF_PACKET_TX_FRAME_SIZE * \
38  AF_PACKET_TX_FRAMES_PER_BLOCK)
39 
40 #define AF_PACKET_RX_FRAMES_PER_BLOCK 1024
41 #define AF_PACKET_RX_FRAME_SIZE (2048 * 5)
42 #define AF_PACKET_RX_BLOCK_NR 1
43 #define AF_PACKET_RX_FRAME_NR (AF_PACKET_RX_BLOCK_NR * \
44  AF_PACKET_RX_FRAMES_PER_BLOCK)
45 #define AF_PACKET_RX_BLOCK_SIZE (AF_PACKET_RX_FRAME_SIZE * \
46  AF_PACKET_RX_FRAMES_PER_BLOCK)
47 
48 #if AF_PACKET_DEBUG_SOCKET == 1
49 #define DBG_SOCK(args...) clib_warning(args);
50 #else
51 #define DBG_SOCK(args...)
52 #endif
53 
54 /*defined in net/if.h but clashes with dpdk headers */
55 unsigned int if_nametoindex (const char *ifname);
56 
57 typedef struct tpacket_req tpacket_req_t;
58 
59 static u32
61  u32 flags)
62 {
63  /* nothing for now */
64  return 0;
65 }
66 
67 static clib_error_t *
69 {
70  vlib_main_t *vm = vlib_get_main ();
72  u32 idx = uf->private_data;
73 
76 
77  /* Schedule the rx node */
79 
80  return 0;
81 }
82 
83 static int
85  tpacket_req_t * tx_req, int *fd, u8 ** ring)
86 {
87  int ret, err;
88  struct sockaddr_ll sll;
89  uint host_if_index;
90  int ver = TPACKET_V2;
91  socklen_t req_sz = sizeof (struct tpacket_req);
92  u32 ring_sz = rx_req->tp_block_size * rx_req->tp_block_nr +
93  tx_req->tp_block_size * tx_req->tp_block_nr;
94 
95  host_if_index = if_nametoindex ((const char *) name);
96 
97  if (!host_if_index)
98  {
99  DBG_SOCK ("Wrong host interface name");
100  ret = VNET_API_ERROR_INVALID_INTERFACE;
101  goto error;
102  }
103 
104  if ((*fd = socket (AF_PACKET, SOCK_RAW, htons (ETH_P_ALL))) < 0)
105  {
106  DBG_SOCK ("Failed to create socket");
107  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
108  goto error;
109  }
110 
111  if ((err =
112  setsockopt (*fd, SOL_PACKET, PACKET_VERSION, &ver, sizeof (ver))) < 0)
113  {
114  DBG_SOCK ("Failed to set rx packet interface version");
115  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
116  goto error;
117  }
118 
119  int opt = 1;
120  if ((err =
121  setsockopt (*fd, SOL_PACKET, PACKET_LOSS, &opt, sizeof (opt))) < 0)
122  {
123  DBG_SOCK ("Failed to set packet tx ring error handling option");
124  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
125  goto error;
126  }
127 
128  if ((err =
129  setsockopt (*fd, SOL_PACKET, PACKET_RX_RING, rx_req, req_sz)) < 0)
130  {
131  DBG_SOCK ("Failed to set packet rx ring options");
132  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
133  goto error;
134  }
135 
136  if ((err =
137  setsockopt (*fd, SOL_PACKET, PACKET_TX_RING, tx_req, req_sz)) < 0)
138  {
139  DBG_SOCK ("Failed to set packet rx ring options");
140  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
141  goto error;
142  }
143 
144  *ring =
145  mmap (NULL, ring_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED, *fd,
146  0);
147  if (*ring == MAP_FAILED)
148  {
149  DBG_SOCK ("mmap failure");
150  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
151  goto error;
152  }
153 
154  memset (&sll, 0, sizeof (sll));
155  sll.sll_family = PF_PACKET;
156  sll.sll_protocol = htons (ETH_P_ALL);
157  sll.sll_ifindex = host_if_index;
158 
159  if ((err = bind (*fd, (struct sockaddr *) &sll, sizeof (sll))) < 0)
160  {
161  DBG_SOCK ("Failed to bind rx packet socket (error %d)", err);
162  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
163  goto error;
164  }
165 
166  return 0;
167 error:
168  if (*fd >= 0)
169  close (*fd);
170  *fd = -1;
171  return ret;
172 }
173 
174 int
175 af_packet_create_if (vlib_main_t * vm, u8 * host_if_name, u8 * hw_addr_set,
176  u32 * sw_if_index)
177 {
179  int ret, fd = -1;
180  struct tpacket_req *rx_req = 0;
181  struct tpacket_req *tx_req = 0;
182  u8 *ring = 0;
183  af_packet_if_t *apif = 0;
184  u8 hw_addr[6];
185  clib_error_t *error;
187  vnet_main_t *vnm = vnet_get_main ();
188  uword *p;
189  uword if_index;
190  u8 *host_if_name_dup = vec_dup (host_if_name);
191 
192  p = mhash_get (&apm->if_index_by_host_if_name, host_if_name);
193  if (p)
194  {
195  return VNET_API_ERROR_SUBIF_ALREADY_EXISTS;
196  }
197 
198  vec_validate (rx_req, 0);
199  rx_req->tp_block_size = AF_PACKET_RX_BLOCK_SIZE;
200  rx_req->tp_frame_size = AF_PACKET_RX_FRAME_SIZE;
201  rx_req->tp_block_nr = AF_PACKET_RX_BLOCK_NR;
202  rx_req->tp_frame_nr = AF_PACKET_RX_FRAME_NR;
203 
204  vec_validate (tx_req, 0);
205  tx_req->tp_block_size = AF_PACKET_TX_BLOCK_SIZE;
206  tx_req->tp_frame_size = AF_PACKET_TX_FRAME_SIZE;
207  tx_req->tp_block_nr = AF_PACKET_TX_BLOCK_NR;
208  tx_req->tp_frame_nr = AF_PACKET_TX_FRAME_NR;
209 
210  ret = create_packet_v2_sock (host_if_name, rx_req, tx_req, &fd, &ring);
211 
212  if (ret != 0)
213  goto error;
214 
215  /* So far everything looks good, let's create interface */
216  pool_get (apm->interfaces, apif);
217  if_index = apif - apm->interfaces;
218 
219  apif->fd = fd;
220  apif->rx_ring = ring;
221  apif->tx_ring = ring + rx_req->tp_block_size * rx_req->tp_block_nr;
222  apif->rx_req = rx_req;
223  apif->tx_req = tx_req;
224  apif->host_if_name = host_if_name_dup;
225  apif->per_interface_next_index = ~0;
226  apif->next_tx_frame = 0;
227  apif->next_rx_frame = 0;
228 
229  {
230  unix_file_t template = { 0 };
232  template.file_descriptor = fd;
233  template.private_data = if_index;
235  apif->unix_file_index = unix_file_add (&unix_main, &template);
236  }
237 
238  /*use configured or generate random MAC address */
239  if (hw_addr_set)
240  clib_memcpy (hw_addr, hw_addr_set, 6);
241  else
242  {
243  f64 now = vlib_time_now (vm);
244  u32 rnd;
245  rnd = (u32) (now * 1e6);
246  rnd = random_u32 (&rnd);
247 
248  clib_memcpy (hw_addr + 2, &rnd, sizeof (rnd));
249  hw_addr[0] = 2;
250  hw_addr[1] = 0xfe;
251  }
252 
254  if_index, hw_addr, &apif->hw_if_index,
256 
257  if (error)
258  {
259  memset (apif, 0, sizeof (*apif));
260  pool_put (apm->interfaces, apif);
261  clib_error_report (error);
262  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
263  goto error;
264  }
265 
266  sw = vnet_get_hw_sw_interface (vnm, apif->hw_if_index);
267  apif->sw_if_index = sw->sw_if_index;
268 
271 
272  mhash_set_mem (&apm->if_index_by_host_if_name, host_if_name_dup, &if_index,
273  0);
274  if (sw_if_index)
275  *sw_if_index = apif->sw_if_index;
276  return 0;
277 
278 error:
279  vec_free (host_if_name_dup);
280  vec_free (rx_req);
281  vec_free (tx_req);
282  return ret;
283 }
284 
285 int
286 af_packet_delete_if (vlib_main_t * vm, u8 * host_if_name)
287 {
288  vnet_main_t *vnm = vnet_get_main ();
290  af_packet_if_t *apif;
291  uword *p;
292  uword if_index;
293  u32 ring_sz;
294 
295  p = mhash_get (&apm->if_index_by_host_if_name, host_if_name);
296  if (p == NULL)
297  {
298  clib_warning ("Host interface %s does not exist", host_if_name);
299  return VNET_API_ERROR_SYSCALL_ERROR_1;
300  }
301  apif = pool_elt_at_index (apm->interfaces, p[0]);
302  if_index = apif - apm->interfaces;
303 
304  /* bring down the interface */
306 
307  /* clean up */
308  if (apif->unix_file_index != ~0)
309  {
311  apif->unix_file_index = ~0;
312  }
313  else
314  close (apif->fd);
315 
316  ring_sz = apif->rx_req->tp_block_size * apif->rx_req->tp_block_nr +
317  apif->tx_req->tp_block_size * apif->tx_req->tp_block_nr;
318  if (munmap (apif->rx_ring, ring_sz))
319  clib_warning ("Host interface %s could not free rx/tx ring",
320  host_if_name);
321  apif->rx_ring = NULL;
322  apif->tx_ring = NULL;
323  apif->fd = -1;
324 
325  vec_free (apif->rx_req);
326  apif->rx_req = NULL;
327  vec_free (apif->tx_req);
328  apif->tx_req = NULL;
329 
330  vec_free (apif->host_if_name);
331  apif->host_if_name = NULL;
332 
333  mhash_unset (&apm->if_index_by_host_if_name, host_if_name, &if_index);
334 
336 
337  pool_put (apm->interfaces, apif);
338 
339  return 0;
340 }
341 
342 static clib_error_t *
344 {
347 
348  memset (apm, 0, sizeof (af_packet_main_t));
349 
351 
354 
355  return 0;
356 }
357 
359 
360 /*
361  * fd.io coding-style-patch-verification: ON
362  *
363  * Local Variables:
364  * eval: (c-set-style "gnu")
365  * End:
366  */
u32 ** rx_buffers
Definition: af_packet.h:49
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:396
unix_file_t * file_pool
Definition: unix.h:89
vmrglw vmrglh hi
clib_error_t * vnet_hw_interface_set_flags(vnet_main_t *vnm, u32 hw_if_index, u32 flags)
Definition: interface.c:522
unix_file_function_t * read_function
Definition: unix.h:62
uword * pending_input_bitmap
Definition: af_packet.h:46
void ethernet_delete_interface(vnet_main_t *vnm, u32 hw_if_index)
Definition: interface.c:254
static vlib_main_t * vlib_get_main(void)
Definition: global_funcs.h:23
#define AF_PACKET_TX_BLOCK_NR
Definition: af_packet.c:34
static void vlib_node_set_interrupt_pending(vlib_main_t *vm, u32 node_index)
Definition: node_funcs.h:181
#define NULL
Definition: clib.h:55
uword mhash_unset(mhash_t *h, void *key, uword *old_value)
Definition: mhash.c:353
static f64 vlib_time_now(vlib_main_t *vm)
Definition: main.h:182
struct tpacket_req * tx_req
Definition: af_packet.h:26
#define AF_PACKET_RX_FRAME_NR
Definition: af_packet.c:43
u32 unix_file_index
Definition: af_packet.h:31
static clib_error_t * af_packet_fd_read_ready(unix_file_t *uf)
Definition: af_packet.c:68
static void mhash_init_vec_string(mhash_t *h, uword n_value_bytes)
Definition: mhash.h:84
static uword * clib_bitmap_set(uword *ai, uword i, uword value)
Sets the ith bit of a bitmap to new_value Removes trailing zeros from the bitmap. ...
Definition: bitmap.h:167
struct tpacket_req * rx_req
Definition: af_packet.h:25
#define clib_error_report(e)
Definition: error.h:125
#define VNET_HW_INTERFACE_FLAG_LINK_UP
Definition: interface.h:348
#define vec_validate_aligned(V, I, A)
Make sure vector is long enough for given index (no header, specified alignment)
Definition: vec.h:407
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:200
static uword unix_file_add(unix_main_t *um, unix_file_t *template)
Definition: unix.h:136
static vnet_sw_interface_t * vnet_get_hw_sw_interface(vnet_main_t *vnm, u32 hw_if_index)
u8 * host_if_name
Definition: af_packet.h:23
af_packet_if_t * interfaces
Definition: af_packet.h:43
vnet_main_t * vnet_get_main(void)
Definition: misc.c:46
uword flags
Definition: error.h:83
#define AF_PACKET_TX_BLOCK_SIZE
Definition: af_packet.c:37
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:111
u32 next_rx_frame
Definition: af_packet.h:33
#define clib_warning(format, args...)
Definition: error.h:59
static void unix_file_del(unix_main_t *um, unix_file_t *f)
Definition: unix.h:146
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:369
uword private_data
Definition: unix.h:59
uword mhash_set_mem(mhash_t *h, void *key, uword *new_value, uword *old_value)
Definition: mhash.c:271
#define UNIX_FILE_EVENT_EDGE_TRIGGERED
Definition: unix.h:56
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:214
#define vec_dup(V)
Return copy of vector (no header, no alignment)
Definition: vec.h:334
#define AF_PACKET_RX_BLOCK_NR
Definition: af_packet.c:42
struct tpacket_req tpacket_req_t
Definition: af_packet.c:57
u32 next_tx_frame
Definition: af_packet.h:34
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:300
static vlib_thread_main_t * vlib_get_thread_main()
Definition: global_funcs.h:32
#define clib_memcpy(a, b, c)
Definition: string.h:64
#define AF_PACKET_RX_BLOCK_SIZE
Definition: af_packet.c:45
static uword * mhash_get(mhash_t *h, void *key)
Definition: mhash.h:110
u32 per_interface_next_index
Definition: af_packet.h:36
vlib_node_registration_t af_packet_input_node
(constructor) VLIB_REGISTER_NODE (af_packet_input_node)
Definition: node.c:283
unsigned int u32
Definition: types.h:88
#define DBG_SOCK(args...)
Definition: af_packet.c:51
clib_error_t * ethernet_register_interface(vnet_main_t *vnm, u32 dev_class_index, u32 dev_instance, u8 *address, u32 *hw_if_index_return, ethernet_flag_change_function_t flag_change)
Definition: interface.c:205
af_packet_main_t af_packet_main
Definition: af_packet.h:55
unix_main_t unix_main
Definition: main.c:57
mhash_t if_index_by_host_if_name
Definition: af_packet.h:52
unsigned int if_nametoindex(const char *ifname)
vnet_device_class_t af_packet_device_class
u64 uword
Definition: types.h:112
#define AF_PACKET_TX_FRAME_SIZE
Definition: af_packet.c:33
double f64
Definition: types.h:142
static int create_packet_v2_sock(u8 *name, tpacket_req_t *rx_req, tpacket_req_t *tx_req, int *fd, u8 **ring)
Definition: af_packet.c:84
unsigned char u8
Definition: types.h:56
Definition: unix.h:49
#define AF_PACKET_RX_FRAME_SIZE
Definition: af_packet.c:41
static u32 random_u32(u32 *seed)
32-bit random number generator
Definition: random.h:69
static clib_error_t * af_packet_init(vlib_main_t *vm)
Definition: af_packet.c:343
static u32 af_packet_eth_flag_change(vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 flags)
Definition: af_packet.c:60
int af_packet_create_if(vlib_main_t *vm, u8 *host_if_name, u8 *hw_addr_set, u32 *sw_if_index)
Definition: af_packet.c:175
u32 flags
Definition: vhost-user.h:75
#define CLIB_CACHE_LINE_BYTES
Definition: cache.h:67
#define AF_PACKET_TX_FRAME_NR
Definition: af_packet.c:35
int af_packet_delete_if(vlib_main_t *vm, u8 *host_if_name)
Definition: af_packet.c:286