FD.io VPP  v17.04.2-2-ga8f93f8
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 {
71  vnet_main_t *vnm = vnet_get_main ();
72  u32 idx = uf->private_data;
73  af_packet_if_t *apif = pool_elt_at_index (apm->interfaces, idx);
74 
77 
78  /* Schedule the rx node */
80 
81  return 0;
82 }
83 
84 static int
86  tpacket_req_t * tx_req, int *fd, u8 ** ring)
87 {
88  int ret, err;
89  struct sockaddr_ll sll;
90  uint host_if_index;
91  int ver = TPACKET_V2;
92  socklen_t req_sz = sizeof (struct tpacket_req);
93  u32 ring_sz = rx_req->tp_block_size * rx_req->tp_block_nr +
94  tx_req->tp_block_size * tx_req->tp_block_nr;
95 
96  host_if_index = if_nametoindex ((const char *) name);
97 
98  if (!host_if_index)
99  {
100  DBG_SOCK ("Wrong host interface name");
101  ret = VNET_API_ERROR_INVALID_INTERFACE;
102  goto error;
103  }
104 
105  if ((*fd = socket (AF_PACKET, SOCK_RAW, htons (ETH_P_ALL))) < 0)
106  {
107  DBG_SOCK ("Failed to create socket");
108  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
109  goto error;
110  }
111 
112  if ((err =
113  setsockopt (*fd, SOL_PACKET, PACKET_VERSION, &ver, sizeof (ver))) < 0)
114  {
115  DBG_SOCK ("Failed to set rx packet interface version");
116  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
117  goto error;
118  }
119 
120  int opt = 1;
121  if ((err =
122  setsockopt (*fd, SOL_PACKET, PACKET_LOSS, &opt, sizeof (opt))) < 0)
123  {
124  DBG_SOCK ("Failed to set packet tx ring error handling option");
125  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
126  goto error;
127  }
128 
129  if ((err =
130  setsockopt (*fd, SOL_PACKET, PACKET_RX_RING, rx_req, req_sz)) < 0)
131  {
132  DBG_SOCK ("Failed to set packet rx ring options");
133  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
134  goto error;
135  }
136 
137  if ((err =
138  setsockopt (*fd, SOL_PACKET, PACKET_TX_RING, tx_req, req_sz)) < 0)
139  {
140  DBG_SOCK ("Failed to set packet rx ring options");
141  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
142  goto error;
143  }
144 
145  *ring =
146  mmap (NULL, ring_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED, *fd,
147  0);
148  if (*ring == MAP_FAILED)
149  {
150  DBG_SOCK ("mmap failure");
151  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
152  goto error;
153  }
154 
155  memset (&sll, 0, sizeof (sll));
156  sll.sll_family = PF_PACKET;
157  sll.sll_protocol = htons (ETH_P_ALL);
158  sll.sll_ifindex = host_if_index;
159 
160  if ((err = bind (*fd, (struct sockaddr *) &sll, sizeof (sll))) < 0)
161  {
162  DBG_SOCK ("Failed to bind rx packet socket (error %d)", err);
163  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
164  goto error;
165  }
166 
167  return 0;
168 error:
169  if (*fd >= 0)
170  close (*fd);
171  *fd = -1;
172  return ret;
173 }
174 
175 int
176 af_packet_create_if (vlib_main_t * vm, u8 * host_if_name, u8 * hw_addr_set,
177  u32 * sw_if_index)
178 {
180  int ret, fd = -1;
181  struct tpacket_req *rx_req = 0;
182  struct tpacket_req *tx_req = 0;
183  u8 *ring = 0;
184  af_packet_if_t *apif = 0;
185  u8 hw_addr[6];
186  clib_error_t *error;
189  vnet_main_t *vnm = vnet_get_main ();
190  uword *p;
191  uword if_index;
192  u8 *host_if_name_dup = vec_dup (host_if_name);
193 
194  p = mhash_get (&apm->if_index_by_host_if_name, host_if_name);
195  if (p)
196  {
197  return VNET_API_ERROR_SUBIF_ALREADY_EXISTS;
198  }
199 
200  vec_validate (rx_req, 0);
201  rx_req->tp_block_size = AF_PACKET_RX_BLOCK_SIZE;
202  rx_req->tp_frame_size = AF_PACKET_RX_FRAME_SIZE;
203  rx_req->tp_block_nr = AF_PACKET_RX_BLOCK_NR;
204  rx_req->tp_frame_nr = AF_PACKET_RX_FRAME_NR;
205 
206  vec_validate (tx_req, 0);
207  tx_req->tp_block_size = AF_PACKET_TX_BLOCK_SIZE;
208  tx_req->tp_frame_size = AF_PACKET_TX_FRAME_SIZE;
209  tx_req->tp_block_nr = AF_PACKET_TX_BLOCK_NR;
210  tx_req->tp_frame_nr = AF_PACKET_TX_FRAME_NR;
211 
212  ret = create_packet_v2_sock (host_if_name, rx_req, tx_req, &fd, &ring);
213 
214  if (ret != 0)
215  goto error;
216 
217  /* So far everything looks good, let's create interface */
218  pool_get (apm->interfaces, apif);
219  if_index = apif - apm->interfaces;
220 
221  apif->fd = fd;
222  apif->rx_ring = ring;
223  apif->tx_ring = ring + rx_req->tp_block_size * rx_req->tp_block_nr;
224  apif->rx_req = rx_req;
225  apif->tx_req = tx_req;
226  apif->host_if_name = host_if_name_dup;
227  apif->per_interface_next_index = ~0;
228  apif->next_tx_frame = 0;
229  apif->next_rx_frame = 0;
230 
231  if (tm->n_vlib_mains > 1)
232  {
235  memset ((void *) apif->lockp, 0, CLIB_CACHE_LINE_BYTES);
236  }
237 
238  {
239  unix_file_t template = { 0 };
241  template.file_descriptor = fd;
242  template.private_data = if_index;
244  apif->unix_file_index = unix_file_add (&unix_main, &template);
245  }
246 
247  /*use configured or generate random MAC address */
248  if (hw_addr_set)
249  clib_memcpy (hw_addr, hw_addr_set, 6);
250  else
251  {
252  f64 now = vlib_time_now (vm);
253  u32 rnd;
254  rnd = (u32) (now * 1e6);
255  rnd = random_u32 (&rnd);
256 
257  clib_memcpy (hw_addr + 2, &rnd, sizeof (rnd));
258  hw_addr[0] = 2;
259  hw_addr[1] = 0xfe;
260  }
261 
263  if_index, hw_addr, &apif->hw_if_index,
265 
266  if (error)
267  {
268  memset (apif, 0, sizeof (*apif));
269  pool_put (apm->interfaces, apif);
270  clib_error_report (error);
271  ret = VNET_API_ERROR_SYSCALL_ERROR_1;
272  goto error;
273  }
274 
275  sw = vnet_get_hw_sw_interface (vnm, apif->hw_if_index);
276  apif->sw_if_index = sw->sw_if_index;
278  vnet_device_input_assign_thread (apif->hw_if_index, 0, /* queue */
279  ~0 /* any cpu */ );
280 
283 
284  mhash_set_mem (&apm->if_index_by_host_if_name, host_if_name_dup, &if_index,
285  0);
286  if (sw_if_index)
287  *sw_if_index = apif->sw_if_index;
288 
289  return 0;
290 
291 error:
292  vec_free (host_if_name_dup);
293  vec_free (rx_req);
294  vec_free (tx_req);
295  return ret;
296 }
297 
298 int
300 {
301  vnet_main_t *vnm = vnet_get_main ();
303  af_packet_if_t *apif;
304  uword *p;
305  uword if_index;
306  u32 ring_sz;
307 
308  p = mhash_get (&apm->if_index_by_host_if_name, host_if_name);
309  if (p == NULL)
310  {
311  clib_warning ("Host interface %s does not exist", host_if_name);
312  return VNET_API_ERROR_SYSCALL_ERROR_1;
313  }
314  apif = pool_elt_at_index (apm->interfaces, p[0]);
315  if_index = apif - apm->interfaces;
316 
317  /* bring down the interface */
319 
320  /* clean up */
321  if (apif->unix_file_index != ~0)
322  {
324  apif->unix_file_index = ~0;
325  }
326  else
327  close (apif->fd);
328 
329  ring_sz = apif->rx_req->tp_block_size * apif->rx_req->tp_block_nr +
330  apif->tx_req->tp_block_size * apif->tx_req->tp_block_nr;
331  if (munmap (apif->rx_ring, ring_sz))
332  clib_warning ("Host interface %s could not free rx/tx ring",
333  host_if_name);
334  apif->rx_ring = NULL;
335  apif->tx_ring = NULL;
336  apif->fd = -1;
337 
338  vec_free (apif->rx_req);
339  apif->rx_req = NULL;
340  vec_free (apif->tx_req);
341  apif->tx_req = NULL;
342 
343  vec_free (apif->host_if_name);
344  apif->host_if_name = NULL;
345 
346  mhash_unset (&apm->if_index_by_host_if_name, host_if_name, &if_index);
347 
349 
350  pool_put (apm->interfaces, apif);
351 
352  return 0;
353 }
354 
355 static clib_error_t *
357 {
360 
361  memset (apm, 0, sizeof (af_packet_main_t));
362 
364 
367 
368  return 0;
369 }
370 
372 
373 /*
374  * fd.io coding-style-patch-verification: ON
375  *
376  * Local Variables:
377  * eval: (c-set-style "gnu")
378  * End:
379  */
u32 ** rx_buffers
Definition: af_packet.h:50
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:436
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:530
unix_file_function_t * read_function
Definition: unix.h:62
uword * pending_input_bitmap
Definition: af_packet.h:47
void ethernet_delete_interface(vnet_main_t *vnm, u32 hw_if_index)
Definition: interface.c:295
vnet_main_t * vnet_get_main(void)
Definition: misc.c:46
#define AF_PACKET_TX_BLOCK_NR
Definition: af_packet.c:34
unix_main_t unix_main
Definition: main.c:59
#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:185
struct tpacket_req * tx_req
Definition: af_packet.h:27
#define AF_PACKET_RX_FRAME_NR
Definition: af_packet.c:43
u32 unix_file_index
Definition: af_packet.h:32
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:26
#define VNET_HW_INTERFACE_FLAG_LINK_UP
Definition: interface.h:379
#define vec_validate_aligned(V, I, A)
Make sure vector is long enough for given index (no header, specified alignment)
Definition: vec.h:447
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:200
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:24
af_packet_if_t * interfaces
Definition: af_packet.h:44
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:34
static_always_inline void vnet_device_input_set_interrupt_pending(vnet_main_t *vnm, u32 hw_if_index, u16 queue_id)
Definition: devices.h:115
static uword unix_file_add(unix_main_t *um, unix_file_t *template)
Definition: unix.h:136
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:397
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 pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:241
#define vec_dup(V)
Return copy of vector (no header, no alignment)
Definition: vec.h:374
#define AF_PACKET_RX_BLOCK_NR
Definition: af_packet.c:42
static void vnet_set_device_input_node(u32 hw_if_index, u32 node_index)
Definition: devices.h:75
#define UNIX_FILE_EVENT_EDGE_TRIGGERED
Definition: unix.h:56
struct tpacket_req tpacket_req_t
Definition: af_packet.c:57
u32 next_tx_frame
Definition: af_packet.h:35
vlib_main_t * vm
Definition: buffer.c:276
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:340
#define clib_warning(format, args...)
Definition: error.h:59
#define clib_memcpy(a, b, c)
Definition: string.h:69
#define AF_PACKET_RX_BLOCK_SIZE
Definition: af_packet.c:45
u32 per_interface_next_index
Definition: af_packet.h:37
vlib_node_registration_t af_packet_input_node
(constructor) VLIB_REGISTER_NODE (af_packet_input_node)
Definition: node.c:266
unsigned int u32
Definition: types.h:88
#define DBG_SOCK(args...)
Definition: af_packet.c:51
static uword * mhash_get(mhash_t *h, const void *key)
Definition: mhash.h:110
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:246
af_packet_main_t af_packet_main
Definition: af_packet.h:56
#define clib_error_report(e)
Definition: error.h:125
mhash_t if_index_by_host_if_name
Definition: af_packet.h:53
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:85
unsigned char u8
Definition: types.h:56
Definition: unix.h:49
#define AF_PACKET_RX_FRAME_SIZE
Definition: af_packet.c:41
static void * clib_mem_alloc_aligned(uword size, uword align)
Definition: mem.h:117
static u32 random_u32(u32 *seed)
32-bit random number generator
Definition: random.h:69
volatile u32 * lockp
Definition: af_packet.h:23
static vlib_thread_main_t * vlib_get_thread_main()
Definition: global_funcs.h:32
static clib_error_t * af_packet_init(vlib_main_t *vm)
Definition: af_packet.c:356
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:176
u32 flags
Definition: vhost-user.h:78
#define CLIB_CACHE_LINE_BYTES
Definition: cache.h:67
void vnet_device_input_assign_thread(u32 hw_if_index, u16 queue_id, uword cpu_index)
Definition: devices.c:106
#define AF_PACKET_TX_FRAME_NR
Definition: af_packet.c:35
static void unix_file_del(unix_main_t *um, unix_file_t *f)
Definition: unix.h:146
int af_packet_delete_if(vlib_main_t *vm, u8 *host_if_name)
Definition: af_packet.c:299