1 .. SPDX-License-Identifier: GPL-2.0
2
3 ===========
4 Packet MMAP
5 ===========
6
7 Abstract
8 ========
9
10 This file documents the mmap() facility available with the PACKET
11 socket interface. This type of sockets is used for
12
13 i) capture network traffic with utilities like tcpdump,
14 ii) transmit network traffic, or any other that needs raw
15 access to network interface.
16
17 Howto can be found at:
18
19 https://web.archive.org/web/20220404160947/https://sites.google.com/site/packetmmap/
20
21 Please send your comments to
22 - Ulisses Alonso Camaró <uaca@i.hate.spam.alumni.uv.es>
23 - Johann Baudy
24
25 Why use PACKET_MMAP
26 ===================
27
28 Non PACKET_MMAP capture process (plain AF_PACKET) is very
29 inefficient. It uses very limited buffers and requires one system call to
30 capture each packet, it requires two if you want to get packet's timestamp
31 (like libpcap always does).
32
33 On the other hand PACKET_MMAP is very efficient. PACKET_MMAP provides a size
34 configurable circular buffer mapped in user space that can be used to either
35 send or receive packets. This way reading packets just needs to wait for them,
36 most of the time there is no need to issue a single system call. Concerning
37 transmission, multiple packets can be sent through one system call to get the
38 highest bandwidth. By using a shared buffer between the kernel and the user
39 also has the benefit of minimizing packet copies.
40
41 It's fine to use PACKET_MMAP to improve the performance of the capture and
42 transmission process, but it isn't everything. At least, if you are capturing
43 at high speeds (this is relative to the cpu speed), you should check if the
44 device driver of your network interface card supports some sort of interrupt
45 load mitigation or (even better) if it supports NAPI, also make sure it is
46 enabled. For transmission, check the MTU (Maximum Transmission Unit) used and
47 supported by devices of your network. CPU IRQ pinning of your network interface
48 card can also be an advantage.
49
50 How to use mmap() to improve capture process
51 ============================================
52
53 From the user standpoint, you should use the higher level libpcap library, which
54 is a de facto standard, portable across nearly all operating systems
55 including Win32.
56
57 Packet MMAP support was integrated into libpcap around the time of version 1.3.0;
58 TPACKET_V3 support was added in version 1.5.0
59
60 How to use mmap() directly to improve capture process
61 =====================================================
62
63 From the system calls stand point, the use of PACKET_MMAP involves
64 the following process::
65
66
67 [setup] socket() -------> creation of the capture socket
68 setsockopt() ---> allocation of the circular buffer (ring)
69 option: PACKET_RX_RING
70 mmap() ---------> mapping of the allocated buffer to the
71 user process
72
73 [capture] poll() ---------> to wait for incoming packets
74
75 [shutdown] close() --------> destruction of the capture socket and
76 deallocation of all associated
77 resources.
78
79
80 socket creation and destruction is straight forward, and is done
81 the same way with or without PACKET_MMAP::
82
83 int fd = socket(PF_PACKET, mode, htons(ETH_P_ALL));
84
85 where mode is SOCK_RAW for the raw interface were link level
86 information can be captured or SOCK_DGRAM for the cooked
87 interface where link level information capture is not
88 supported and a link level pseudo-header is provided
89 by the kernel.
90
91 The destruction of the socket and all associated resources
92 is done by a simple call to close(fd).
93
94 Similarly as without PACKET_MMAP, it is possible to use one socket
95 for capture and transmission. This can be done by mapping the
96 allocated RX and TX buffer ring with a single mmap() call.
97 See "Mapping and use of the circular buffer (ring)".
98
99 Next I will describe PACKET_MMAP settings and its constraints,
100 also the mapping of the circular buffer in the user process and
101 the use of this buffer.
102
103 How to use mmap() directly to improve transmission process
104 ==========================================================
105 Transmission process is similar to capture as shown below::
106
107 [setup] socket() -------> creation of the transmission socket
108 setsockopt() ---> allocation of the circular buffer (ring)
109 option: PACKET_TX_RING
110 bind() ---------> bind transmission socket with a network interface
111 mmap() ---------> mapping of the allocated buffer to the
112 user process
113
114 [transmission] poll() ---------> wait for free packets (optional)
115 send() ---------> send all packets that are set as ready in
116 the ring
117 The flag MSG_DONTWAIT can be used to return
118 before end of transfer.
119
120 [shutdown] close() --------> destruction of the transmission socket and
121 deallocation of all associated resources.
122
123 Socket creation and destruction is also straight forward, and is done
124 the same way as in capturing described in the previous paragraph::
125
126 int fd = socket(PF_PACKET, mode, 0);
127
128 The protocol can optionally be 0 in case we only want to transmit
129 via this socket, which avoids an expensive call to packet_rcv().
130 In this case, you also need to bind(2) the TX_RING with sll_protocol = 0
131 set. Otherwise, htons(ETH_P_ALL) or any other protocol, for example.
132
133 Binding the socket to your network interface is mandatory (with zero copy) to
134 know the header size of frames used in the circular buffer.
135
136 As capture, each frame contains two parts::
137
138 --------------------
139 | struct tpacket_hdr | Header. It contains the status of
140 | | of this frame
141 |--------------------|
142 | data buffer |
143 . . Data that will be sent over the network interface.
144 . .
145 --------------------
146
147 bind() associates the socket to your network interface thanks to
148 sll_ifindex parameter of struct sockaddr_ll.
149
150 Initialization example::
151
152 struct sockaddr_ll my_addr;
153 struct ifreq s_ifr;
154 ...
155
156 strscpy_pad (s_ifr.ifr_name, "eth0", sizeof(s_ifr.ifr_name));
157
158 /* get interface index of eth0 */
159 ioctl(this->socket, SIOCGIFINDEX, &s_ifr);
160
161 /* fill sockaddr_ll struct to prepare binding */
162 my_addr.sll_family = AF_PACKET;
163 my_addr.sll_protocol = htons(ETH_P_ALL);
164 my_addr.sll_ifindex = s_ifr.ifr_ifindex;
165
166 /* bind socket to eth0 */
167 bind(this->socket, (struct sockaddr *)&my_addr, sizeof(struct sockaddr_ll));
168
169 A complete tutorial is available at:
170 https://web.archive.org/web/20220404160947/https://sites.google.com/site/packetmmap/
171
172 By default, the user should put data at::
173
174 frame base + TPACKET_HDRLEN - sizeof(struct sockaddr_ll)
175
176 So, whatever you choose for the socket mode (SOCK_DGRAM or SOCK_RAW),
177 the beginning of the user data will be at::
178
179 frame base + TPACKET_ALIGN(sizeof(struct tpacket_hdr))
180
181 If you wish to put user data at a custom offset from the beginning of
182 the frame (for payload alignment with SOCK_RAW mode for instance) you
183 can set tp_net (with SOCK_DGRAM) or tp_mac (with SOCK_RAW). In order
184 to make this work it must be enabled previously with setsockopt()
185 and the PACKET_TX_HAS_OFF option.
186
187 PACKET_MMAP settings
188 ====================
189
190 To setup PACKET_MMAP from user level code is done with a call like
191
192 - Capture process::
193
194 setsockopt(fd, SOL_PACKET, PACKET_RX_RING, (void *) &req, sizeof(req))
195
196 - Transmission process::
197
198 setsockopt(fd, SOL_PACKET, PACKET_TX_RING, (void *) &req, sizeof(req))
199
200 The most significant argument in the previous call is the req parameter,
201 this parameter must to have the following structure::
202
203 struct tpacket_req
204 {
205 unsigned int tp_block_size; /* Minimal size of contiguous block */
206 unsigned int tp_block_nr; /* Number of blocks */
207 unsigned int tp_frame_size; /* Size of frame */
208 unsigned int tp_frame_nr; /* Total number of frames */
209 };
210
211 This structure is defined in /usr/include/linux/if_packet.h and establishes a
212 circular buffer (ring) of unswappable memory.
213 Being mapped in the capture process allows reading the captured frames and
214 related meta-information like timestamps without requiring a system call.
215
216 Frames are grouped in blocks. Each block is a physically contiguous
217 region of memory and holds tp_block_size/tp_frame_size frames. The total number
218 of blocks is tp_block_nr. Note that tp_frame_nr is a redundant parameter because::
219
220 frames_per_block = tp_block_size/tp_frame_size
221
222 indeed, packet_set_ring checks that the following condition is true::
223
224 frames_per_block * tp_block_nr == tp_frame_nr
225
226 Lets see an example, with the following values::
227
228 tp_block_size= 4096
229 tp_frame_size= 2048
230 tp_block_nr = 4
231 tp_frame_nr = 8
232
233 we will get the following buffer structure::
234
235 block #1 block #2
236 +---------+---------+ +---------+---------+
237 | frame 1 | frame 2 | | frame 3 | frame 4 |
238 +---------+---------+ +---------+---------+
239
240 block #3 block #4
241 +---------+---------+ +---------+---------+
242 | frame 5 | frame 6 | | frame 7 | frame 8 |
243 +---------+---------+ +---------+---------+
244
245 A frame can be of any size with the only condition it can fit in a block. A block
246 can only hold an integer number of frames, or in other words, a frame cannot
247 be spawned across two blocks, so there are some details you have to take into
248 account when choosing the frame_size. See "Mapping and use of the circular
249 buffer (ring)".
250
251 PACKET_MMAP setting constraints
252 ===============================
253
254 In kernel versions prior to 2.4.26 (for the 2.4 branch) and 2.6.5 (2.6 branch),
255 the PACKET_MMAP buffer could hold only 32768 frames in a 32 bit architecture or
256 16384 in a 64 bit architecture.
257
258 Block size limit
259 ----------------
260
261 As stated earlier, each block is a contiguous physical region of memory. These
262 memory regions are allocated with calls to the __get_free_pages() function. As
263 the name indicates, this function allocates pages of memory, and the second
264 argument is "order" or a power of two number of pages, that is
265 (for PAGE_SIZE == 4096) order=0 ==> 4096 bytes, order=1 ==> 8192 bytes,
266 order=2 ==> 16384 bytes, etc. The maximum size of a
267 region allocated by __get_free_pages is determined by the MAX_PAGE_ORDER macro.
268 More precisely the limit can be calculated as::
269
270 PAGE_SIZE << MAX_PAGE_ORDER
271
272 In a i386 architecture PAGE_SIZE is 4096 bytes
273 In a 2.4/i386 kernel MAX_PAGE_ORDER is 10
274 In a 2.6/i386 kernel MAX_PAGE_ORDER is 11
275
276 So get_free_pages can allocate as much as 4MB or 8MB in a 2.4/2.6 kernel
277 respectively, with an i386 architecture.
278
279 User space programs can include /usr/include/sys/user.h and
280 /usr/include/linux/mmzone.h to get PAGE_SIZE MAX_PAGE_ORDER declarations.
281
282 The pagesize can also be determined dynamically with the getpagesize (2)
283 system call.
284
285 Block number limit
286 ------------------
287
288 To understand the constraints of PACKET_MMAP, we have to see the structure
289 used to hold the pointers to each block.
290
291 Currently, this structure is a dynamically allocated vector with kmalloc
292 called pg_vec, its size limits the number of blocks that can be allocated::
293
294 +---+---+---+---+
295 | x | x | x | x |
296 +---+---+---+---+
297 | | | |
298 | | | v
299 | | v block #4
300 | v block #3
301 v block #2
302 block #1
303
304 kmalloc allocates any number of bytes of physically contiguous memory from
305 a pool of pre-determined sizes. This pool of memory is maintained by the slab
306 allocator which is at the end the responsible for doing the allocation and
307 hence which imposes the maximum memory that kmalloc can allocate.
308
309 In a 2.4/2.6 kernel and the i386 architecture, the limit is 131072 bytes. The
310 predetermined sizes that kmalloc uses can be checked in the "size-<bytes>"
311 entries of /proc/slabinfo
312
313 In a 32 bit architecture, pointers are 4 bytes long, so the total number of
314 pointers to blocks is::
315
316 131072/4 = 32768 blocks
317
318 PACKET_MMAP buffer size calculator
319 ==================================
320
321 Definitions:
322
323 ============== ================================================================
324 <size-max> is the maximum size of allocable with kmalloc
325 (see /proc/slabinfo)
326 <pointer size> depends on the architecture -- ``sizeof(void *)``
327 <page size> depends on the architecture -- PAGE_SIZE or getpagesize (2)
328 <max-order> is the value defined with MAX_PAGE_ORDER
329 <frame size> it's an upper bound of frame's capture size (more on this later)
330 ============== ================================================================
331
332 from these definitions we will derive::
333
334 <block number> = <size-max>/<pointer size>
335 <block size> = <pagesize> << <max-order>
336
337 so, the max buffer size is::
338
339 <block number> * <block size>
340
341 and, the number of frames be::
342
343 <block number> * <block size> / <frame size>
344
345 Suppose the following parameters, which apply for 2.6 kernel and an
346 i386 architecture::
347
348 <size-max> = 131072 bytes
349 <pointer size> = 4 bytes
350 <pagesize> = 4096 bytes
351 <max-order> = 11
352
353 and a value for <frame size> of 2048 bytes. These parameters will yield::
354
355 <block number> = 131072/4 = 32768 blocks
356 <block size> = 4096 << 11 = 8 MiB.
357
358 and hence the buffer will have a 262144 MiB size. So it can hold
359 262144 MiB / 2048 bytes = 134217728 frames
360
361 Actually, this buffer size is not possible with an i386 architecture.
362 Remember that the memory is allocated in kernel space, in the case of
363 an i386 kernel's memory size is limited to 1GiB.
364
365 All memory allocations are not freed until the socket is closed. The memory
366 allocations are done with GFP_KERNEL priority, this basically means that
367 the allocation can wait and swap other process' memory in order to allocate
368 the necessary memory, so normally limits can be reached.
369
370 Other constraints
371 -----------------
372
373 If you check the source code you will see that what I draw here as a frame
374 is not only the link level frame. At the beginning of each frame there is a
375 header called struct tpacket_hdr used in PACKET_MMAP to hold link level's frame
376 meta information like timestamp. So what we draw here a frame it's really
377 the following (from include/linux/if_packet.h)::
378
379 /*
380 Frame structure:
381
382 - Start. Frame must be aligned to TPACKET_ALIGNMENT=16
383 - struct tpacket_hdr
384 - pad to TPACKET_ALIGNMENT=16
385 - struct sockaddr_ll
386 - Gap, chosen so that packet data (Start+tp_net) aligns to
387 TPACKET_ALIGNMENT=16
388 - Start+tp_mac: [ Optional MAC header ]
389 - Start+tp_net: Packet data, aligned to TPACKET_ALIGNMENT=16.
390 - Pad to align to TPACKET_ALIGNMENT=16
391 */
392
393 The following are conditions that are checked in packet_set_ring
394
395 - tp_block_size must be a multiple of PAGE_SIZE (1)
396 - tp_frame_size must be greater than TPACKET_HDRLEN (obvious)
397 - tp_frame_size must be a multiple of TPACKET_ALIGNMENT
398 - tp_frame_nr must be exactly frames_per_block*tp_block_nr
399
400 Note that tp_block_size should be chosen to be a power of two or there will
401 be a waste of memory.
402
403 Mapping and use of the circular buffer (ring)
404 ---------------------------------------------
405
406 The mapping of the buffer in the user process is done with the conventional
407 mmap function. Even the circular buffer is compound of several physically
408 discontiguous blocks of memory, they are contiguous to the user space, hence
409 just one call to mmap is needed::
410
411 mmap(0, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
412
413 If tp_frame_size is a divisor of tp_block_size frames will be
414 contiguously spaced by tp_frame_size bytes. If not, each
415 tp_block_size/tp_frame_size frames there will be a gap between
416 the frames. This is because a frame cannot be spawn across two
417 blocks.
418
419 To use one socket for capture and transmission, the mapping of both the
420 RX and TX buffer ring has to be done with one call to mmap::
421
422 ...
423 setsockopt(fd, SOL_PACKET, PACKET_RX_RING, &foo, sizeof(foo));
424 setsockopt(fd, SOL_PACKET, PACKET_TX_RING, &bar, sizeof(bar));
425 ...
426 rx_ring = mmap(0, size * 2, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
427 tx_ring = rx_ring + size;
428
429 RX must be the first as the kernel maps the TX ring memory right
430 after the RX one.
431
432 At the beginning of each frame there is an status field (see
433 struct tpacket_hdr). If this field is 0 means that the frame is ready
434 to be used for the kernel, If not, there is a frame the user can read
435 and the following flags apply:
436
437 Capture process
438 ^^^^^^^^^^^^^^^
439
440 From include/linux/if_packet.h::
441
442 #define TP_STATUS_COPY (1 << 1)
443 #define TP_STATUS_LOSING (1 << 2)
444 #define TP_STATUS_CSUMNOTREADY (1 << 3)
445 #define TP_STATUS_CSUM_VALID (1 << 7)
446
447 ====================== =======================================================
448 TP_STATUS_COPY This flag indicates that the frame (and associated
449 meta information) has been truncated because it's
450 larger than tp_frame_size. This packet can be
451 read entirely with recvfrom().
452
453 In order to make this work it must to be
454 enabled previously with setsockopt() and
455 the PACKET_COPY_THRESH option.
456
457 The number of frames that can be buffered to
458 be read with recvfrom is limited like a normal socket.
459 See the SO_RCVBUF option in the socket (7) man page.
460
461 TP_STATUS_LOSING indicates there were packet drops from last time
462 statistics where checked with getsockopt() and
463 the PACKET_STATISTICS option.
464
465 TP_STATUS_CSUMNOTREADY currently it's used for outgoing IP packets which
466 its checksum will be done in hardware. So while
467 reading the packet we should not try to check the
468 checksum.
469
470 TP_STATUS_CSUM_VALID This flag indicates that at least the transport
471 header checksum of the packet has been already
472 validated on the kernel side. If the flag is not set
473 then we are free to check the checksum by ourselves
474 provided that TP_STATUS_CSUMNOTREADY is also not set.
475 ====================== =======================================================
476
477 for convenience there are also the following defines::
478
479 #define TP_STATUS_KERNEL 0
480 #define TP_STATUS_USER 1
481
482 The kernel initializes all frames to TP_STATUS_KERNEL, when the kernel
483 receives a packet it puts in the buffer and updates the status with
484 at least the TP_STATUS_USER flag. Then the user can read the packet,
485 once the packet is read the user must zero the status field, so the kernel
486 can use again that frame buffer.
487
488 The user can use poll (any other variant should apply too) to check if new
489 packets are in the ring::
490
491 struct pollfd pfd;
492
493 pfd.fd = fd;
494 pfd.revents = 0;
495 pfd.events = POLLIN|POLLRDNORM|POLLERR;
496
497 if (status == TP_STATUS_KERNEL)
498 retval = poll(&pfd, 1, timeout);
499
500 It doesn't incur in a race condition to first check the status value and
501 then poll for frames.
502
503 Transmission process
504 ^^^^^^^^^^^^^^^^^^^^
505
506 Those defines are also used for transmission::
507
508 #define TP_STATUS_AVAILABLE 0 // Frame is available
509 #define TP_STATUS_SEND_REQUEST 1 // Frame will be sent on next send()
510 #define TP_STATUS_SENDING 2 // Frame is currently in transmission
511 #define TP_STATUS_WRONG_FORMAT 4 // Frame format is not correct
512
513 First, the kernel initializes all frames to TP_STATUS_AVAILABLE. To send a
514 packet, the user fills a data buffer of an available frame, sets tp_len to
515 current data buffer size and sets its status field to TP_STATUS_SEND_REQUEST.
516 This can be done on multiple frames. Once the user is ready to transmit, it
517 calls send(). Then all buffers with status equal to TP_STATUS_SEND_REQUEST are
518 forwarded to the network device. The kernel updates each status of sent
519 frames with TP_STATUS_SENDING until the end of transfer.
520
521 At the end of each transfer, buffer status returns to TP_STATUS_AVAILABLE.
522
523 ::
524
525 header->tp_len = in_i_size;
526 header->tp_status = TP_STATUS_SEND_REQUEST;
527 retval = send(this->socket, NULL, 0, 0);
528
529 The user can also use poll() to check if a buffer is available:
530
531 (status == TP_STATUS_SENDING)
532
533 ::
534
535 struct pollfd pfd;
536 pfd.fd = fd;
537 pfd.revents = 0;
538 pfd.events = POLLOUT;
539 retval = poll(&pfd, 1, timeout);
540
541 What TPACKET versions are available and when to use them?
542 =========================================================
543
544 ::
545
546 int val = tpacket_version;
547 setsockopt(fd, SOL_PACKET, PACKET_VERSION, &val, sizeof(val));
548 getsockopt(fd, SOL_PACKET, PACKET_VERSION, &val, sizeof(val));
549
550 where 'tpacket_version' can be TPACKET_V1 (default), TPACKET_V2, TPACKET_V3.
551
552 TPACKET_V1:
553 - Default if not otherwise specified by setsockopt(2)
554 - RX_RING, TX_RING available
555
556 TPACKET_V1 --> TPACKET_V2:
557 - Made 64 bit clean due to unsigned long usage in TPACKET_V1
558 structures, thus this also works on 64 bit kernel with 32 bit
559 userspace and the like
560 - Timestamp resolution in nanoseconds instead of microseconds
561 - RX_RING, TX_RING available
562 - VLAN metadata information available for packets
563 (TP_STATUS_VLAN_VALID, TP_STATUS_VLAN_TPID_VALID),
564 in the tpacket2_hdr structure:
565
566 - TP_STATUS_VLAN_VALID bit being set into the tp_status field indicates
567 that the tp_vlan_tci field has valid VLAN TCI value
568 - TP_STATUS_VLAN_TPID_VALID bit being set into the tp_status field
569 indicates that the tp_vlan_tpid field has valid VLAN TPID value
570
571 - How to switch to TPACKET_V2:
572
573 1. Replace struct tpacket_hdr by struct tpacket2_hdr
574 2. Query header len and save
575 3. Set protocol version to 2, set up ring as usual
576 4. For getting the sockaddr_ll,
577 use ``(void *)hdr + TPACKET_ALIGN(hdrlen)`` instead of
578 ``(void *)hdr + TPACKET_ALIGN(sizeof(struct tpacket_hdr))``
579
580 TPACKET_V2 --> TPACKET_V3:
581 - Flexible buffer implementation for RX_RING:
582 1. Blocks can be configured with non-static frame-size
583 2. Read/poll is at a block-level (as opposed to packet-level)
584 3. Added poll timeout to avoid indefinite user-space wait
585 on idle links
586 4. Added user-configurable knobs:
587
588 4.1 block::timeout
589 4.2 tpkt_hdr::sk_rxhash
590
591 - RX Hash data available in user space
592 - TX_RING semantics are conceptually similar to TPACKET_V2;
593 use tpacket3_hdr instead of tpacket2_hdr, and TPACKET3_HDRLEN
594 instead of TPACKET2_HDRLEN. In the current implementation,
595 the tp_next_offset field in the tpacket3_hdr MUST be set to
596 zero, indicating that the ring does not hold variable sized frames.
597 Packets with non-zero values of tp_next_offset will be dropped.
598
599 AF_PACKET fanout mode
600 =====================
601
602 In the AF_PACKET fanout mode, packet reception can be load balanced among
603 processes. This also works in combination with mmap(2) on packet sockets.
604
605 Currently implemented fanout policies are:
606
607 - PACKET_FANOUT_HASH: schedule to socket by skb's packet hash
608 - PACKET_FANOUT_LB: schedule to socket by round-robin
609 - PACKET_FANOUT_CPU: schedule to socket by CPU packet arrives on
610 - PACKET_FANOUT_RND: schedule to socket by random selection
611 - PACKET_FANOUT_ROLLOVER: if one socket is full, rollover to another
612 - PACKET_FANOUT_QM: schedule to socket by skbs recorded queue_mapping
613
614 Minimal example code by David S. Miller (try things like "./test eth0 hash",
615 "./test eth0 lb", etc.)::
616
617 #include <stddef.h>
618 #include <stdlib.h>
619 #include <stdio.h>
620 #include <string.h>
621
622 #include <sys/types.h>
623 #include <sys/wait.h>
624 #include <sys/socket.h>
625 #include <sys/ioctl.h>
626
627 #include <unistd.h>
628
629 #include <linux/if_ether.h>
630 #include <linux/if_packet.h>
631
632 #include <net/if.h>
633
634 static const char *device_name;
635 static int fanout_type;
636 static int fanout_id;
637
638 #ifndef PACKET_FANOUT
639 # define PACKET_FANOUT 18
640 # define PACKET_FANOUT_HASH 0
641 # define PACKET_FANOUT_LB 1
642 #endif
643
644 static int setup_socket(void)
645 {
646 int err, fd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_IP));
647 struct sockaddr_ll ll;
648 struct ifreq ifr;
649 int fanout_arg;
650
651 if (fd < 0) {
652 perror("socket");
653 return EXIT_FAILURE;
654 }
655
656 memset(&ifr, 0, sizeof(ifr));
657 strcpy(ifr.ifr_name, device_name);
658 err = ioctl(fd, SIOCGIFINDEX, &ifr);
659 if (err < 0) {
660 perror("SIOCGIFINDEX");
661 return EXIT_FAILURE;
662 }
663
664 memset(&ll, 0, sizeof(ll));
665 ll.sll_family = AF_PACKET;
666 ll.sll_ifindex = ifr.ifr_ifindex;
667 err = bind(fd, (struct sockaddr *) &ll, sizeof(ll));
668 if (err < 0) {
669 perror("bind");
670 return EXIT_FAILURE;
671 }
672
673 fanout_arg = (fanout_id | (fanout_type << 16));
674 err = setsockopt(fd, SOL_PACKET, PACKET_FANOUT,
675 &fanout_arg, sizeof(fanout_arg));
676 if (err) {
677 perror("setsockopt");
678 return EXIT_FAILURE;
679 }
680
681 return fd;
682 }
683
684 static void fanout_thread(void)
685 {
686 int fd = setup_socket();
687 int limit = 10000;
688
689 if (fd < 0)
690 exit(fd);
691
692 while (limit-- > 0) {
693 char buf[1600];
694 int err;
695
696 err = read(fd, buf, sizeof(buf));
697 if (err < 0) {
698 perror("read");
699 exit(EXIT_FAILURE);
700 }
701 if ((limit % 10) == 0)
702 fprintf(stdout, "(%d) \n", getpid());
703 }
704
705 fprintf(stdout, "%d: Received 10000 packets\n", getpid());
706
707 close(fd);
708 exit(0);
709 }
710
711 int main(int argc, char **argp)
712 {
713 int fd, err;
714 int i;
715
716 if (argc != 3) {
717 fprintf(stderr, "Usage: %s INTERFACE {hash|lb}\n", argp[0]);
718 return EXIT_FAILURE;
719 }
720
721 if (!strcmp(argp[2], "hash"))
722 fanout_type = PACKET_FANOUT_HASH;
723 else if (!strcmp(argp[2], "lb"))
724 fanout_type = PACKET_FANOUT_LB;
725 else {
726 fprintf(stderr, "Unknown fanout type [%s]\n", argp[2]);
727 exit(EXIT_FAILURE);
728 }
729
730 device_name = argp[1];
731 fanout_id = getpid() & 0xffff;
732
733 for (i = 0; i < 4; i++) {
734 pid_t pid = fork();
735
736 switch (pid) {
737 case 0:
738 fanout_thread();
739
740 case -1:
741 perror("fork");
742 exit(EXIT_FAILURE);
743 }
744 }
745
746 for (i = 0; i < 4; i++) {
747 int status;
748
749 wait(&status);
750 }
751
752 return 0;
753 }
754
755 AF_PACKET TPACKET_V3 example
756 ============================
757
758 AF_PACKET's TPACKET_V3 ring buffer can be configured to use non-static frame
759 sizes by doing its own memory management. It is based on blocks where polling
760 works on a per block basis instead of per ring as in TPACKET_V2 and predecessor.
761
762 It is said that TPACKET_V3 brings the following benefits:
763
764 * ~15% - 20% reduction in CPU-usage
765 * ~20% increase in packet capture rate
766 * ~2x increase in packet density
767 * Port aggregation analysis
768 * Non static frame size to capture entire packet payload
769
770 So it seems to be a good candidate to be used with packet fanout.
771
772 Minimal example code by Daniel Borkmann based on Chetan Loke's lolpcap (compile
773 it with gcc -Wall -O2 blob.c, and try things like "./a.out eth0", etc.)::
774
775 /* Written from scratch, but kernel-to-user space API usage
776 * dissected from lolpcap:
777 * Copyright 2011, Chetan Loke <loke.chetan@gmail.com>
778 * License: GPL, version 2.0
779 */
780
781 #include <stdio.h>
782 #include <stdlib.h>
783 #include <stdint.h>
784 #include <string.h>
785 #include <assert.h>
786 #include <net/if.h>
787 #include <arpa/inet.h>
788 #include <netdb.h>
789 #include <poll.h>
790 #include <unistd.h>
791 #include <signal.h>
792 #include <inttypes.h>
793 #include <sys/socket.h>
794 #include <sys/mman.h>
795 #include <linux/if_packet.h>
796 #include <linux/if_ether.h>
797 #include <linux/ip.h>
798
799 #ifndef likely
800 # define likely(x) __builtin_expect(!!(x), 1)
801 #endif
802 #ifndef unlikely
803 # define unlikely(x) __builtin_expect(!!(x), 0)
804 #endif
805
806 struct block_desc {
807 uint32_t version;
808 uint32_t offset_to_priv;
809 struct tpacket_hdr_v1 h1;
810 };
811
812 struct ring {
813 struct iovec *rd;
814 uint8_t *map;
815 struct tpacket_req3 req;
816 };
817
818 static unsigned long packets_total = 0, bytes_total = 0;
819 static sig_atomic_t sigint = 0;
820
821 static void sighandler(int num)
822 {
823 sigint = 1;
824 }
825
826 static int setup_socket(struct ring *ring, char *netdev)
827 {
828 int err, i, fd, v = TPACKET_V3;
829 struct sockaddr_ll ll;
830 unsigned int blocksiz = 1 << 22, framesiz = 1 << 11;
831 unsigned int blocknum = 64;
832
833 fd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
834 if (fd < 0) {
835 perror("socket");
836 exit(1);
837 }
838
839 err = setsockopt(fd, SOL_PACKET, PACKET_VERSION, &v, sizeof(v));
840 if (err < 0) {
841 perror("setsockopt");
842 exit(1);
843 }
844
845 memset(&ring->req, 0, sizeof(ring->req));
846 ring->req.tp_block_size = blocksiz;
847 ring->req.tp_frame_size = framesiz;
848 ring->req.tp_block_nr = blocknum;
849 ring->req.tp_frame_nr = (blocksiz * blocknum) / framesiz;
850 ring->req.tp_retire_blk_tov = 60;
851 ring->req.tp_feature_req_word = TP_FT_REQ_FILL_RXHASH;
852
853 err = setsockopt(fd, SOL_PACKET, PACKET_RX_RING, &ring->req,
854 sizeof(ring->req));
855 if (err < 0) {
856 perror("setsockopt");
857 exit(1);
858 }
859
860 ring->map = mmap(NULL, ring->req.tp_block_size * ring->req.tp_block_nr,
861 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED, fd, 0);
862 if (ring->map == MAP_FAILED) {
863 perror("mmap");
864 exit(1);
865 }
866
867 ring->rd = malloc(ring->req.tp_block_nr * sizeof(*ring->rd));
868 assert(ring->rd);
869 for (i = 0; i < ring->req.tp_block_nr; ++i) {
870 ring->rd[i].iov_base = ring->map + (i * ring->req.tp_block_size);
871 ring->rd[i].iov_len = ring->req.tp_block_size;
872 }
873
874 memset(&ll, 0, sizeof(ll));
875 ll.sll_family = PF_PACKET;
876 ll.sll_protocol = htons(ETH_P_ALL);
877 ll.sll_ifindex = if_nametoindex(netdev);
878 ll.sll_hatype = 0;
879 ll.sll_pkttype = 0;
880 ll.sll_halen = 0;
881
882 err = bind(fd, (struct sockaddr *) &ll, sizeof(ll));
883 if (err < 0) {
884 perror("bind");
885 exit(1);
886 }
887
888 return fd;
889 }
890
891 static void display(struct tpacket3_hdr *ppd)
892 {
893 struct ethhdr *eth = (struct ethhdr *) ((uint8_t *) ppd + ppd->tp_mac);
894 struct iphdr *ip = (struct iphdr *) ((uint8_t *) eth + ETH_HLEN);
895
896 if (eth->h_proto == htons(ETH_P_IP)) {
897 struct sockaddr_in ss, sd;
898 char sbuff[NI_MAXHOST], dbuff[NI_MAXHOST];
899
900 memset(&ss, 0, sizeof(ss));
901 ss.sin_family = PF_INET;
902 ss.sin_addr.s_addr = ip->saddr;
903 getnameinfo((struct sockaddr *) &ss, sizeof(ss),
904 sbuff, sizeof(sbuff), NULL, 0, NI_NUMERICHOST);
905
906 memset(&sd, 0, sizeof(sd));
907 sd.sin_family = PF_INET;
908 sd.sin_addr.s_addr = ip->daddr;
909 getnameinfo((struct sockaddr *) &sd, sizeof(sd),
910 dbuff, sizeof(dbuff), NULL, 0, NI_NUMERICHOST);
911
912 printf("%s -> %s, ", sbuff, dbuff);
913 }
914
915 printf("rxhash: 0x%x\n", ppd->hv1.tp_rxhash);
916 }
917
918 static void walk_block(struct block_desc *pbd, const int block_num)
919 {
920 int num_pkts = pbd->h1.num_pkts, i;
921 unsigned long bytes = 0;
922 struct tpacket3_hdr *ppd;
923
924 ppd = (struct tpacket3_hdr *) ((uint8_t *) pbd +
925 pbd->h1.offset_to_first_pkt);
926 for (i = 0; i < num_pkts; ++i) {
927 bytes += ppd->tp_snaplen;
928 display(ppd);
929
930 ppd = (struct tpacket3_hdr *) ((uint8_t *) ppd +
931 ppd->tp_next_offset);
932 }
933
934 packets_total += num_pkts;
935 bytes_total += bytes;
936 }
937
938 static void flush_block(struct block_desc *pbd)
939 {
940 pbd->h1.block_status = TP_STATUS_KERNEL;
941 }
942
943 static void teardown_socket(struct ring *ring, int fd)
944 {
945 munmap(ring->map, ring->req.tp_block_size * ring->req.tp_block_nr);
946 free(ring->rd);
947 close(fd);
948 }
949
950 int main(int argc, char **argp)
951 {
952 int fd, err;
953 socklen_t len;
954 struct ring ring;
955 struct pollfd pfd;
956 unsigned int block_num = 0, blocks = 64;
957 struct block_desc *pbd;
958 struct tpacket_stats_v3 stats;
959
960 if (argc != 2) {
961 fprintf(stderr, "Usage: %s INTERFACE\n", argp[0]);
962 return EXIT_FAILURE;
963 }
964
965 signal(SIGINT, sighandler);
966
967 memset(&ring, 0, sizeof(ring));
968 fd = setup_socket(&ring, argp[argc - 1]);
969 assert(fd > 0);
970
971 memset(&pfd, 0, sizeof(pfd));
972 pfd.fd = fd;
973 pfd.events = POLLIN | POLLERR;
974 pfd.revents = 0;
975
976 while (likely(!sigint)) {
977 pbd = (struct block_desc *) ring.rd[block_num].iov_base;
978
979 if ((pbd->h1.block_status & TP_STATUS_USER) == 0) {
980 poll(&pfd, 1, -1);
981 continue;
982 }
983
984 walk_block(pbd, block_num);
985 flush_block(pbd);
986 block_num = (block_num + 1) % blocks;
987 }
988
989 len = sizeof(stats);
990 err = getsockopt(fd, SOL_PACKET, PACKET_STATISTICS, &stats, &len);
991 if (err < 0) {
992 perror("getsockopt");
993 exit(1);
994 }
995
996 fflush(stdout);
997 printf("\nReceived %u packets, %lu bytes, %u dropped, freeze_q_cnt: %u\n",
998 stats.tp_packets, bytes_total, stats.tp_drops,
999 stats.tp_freeze_q_cnt);
1000
1001 teardown_socket(&ring, fd);
1002 return 0;
1003 }
1004
1005 PACKET_QDISC_BYPASS
1006 ===================
1007
1008 If there is a requirement to load the network with many packets in a similar
1009 fashion as pktgen does, you might set the following option after socket
1010 creation::
1011
1012 int one = 1;
1013 setsockopt(fd, SOL_PACKET, PACKET_QDISC_BYPASS, &one, sizeof(one));
1014
1015 This has the side-effect, that packets sent through PF_PACKET will bypass the
1016 kernel's qdisc layer and are forcedly pushed to the driver directly. Meaning,
1017 packet are not buffered, tc disciplines are ignored, increased loss can occur
1018 and such packets are also not visible to other PF_PACKET sockets anymore. So,
1019 you have been warned; generally, this can be useful for stress testing various
1020 components of a system.
1021
1022 On default, PACKET_QDISC_BYPASS is disabled and needs to be explicitly enabled
1023 on PF_PACKET sockets.
1024
1025 PACKET_TIMESTAMP
1026 ================
1027
1028 The PACKET_TIMESTAMP setting determines the source of the timestamp in
1029 the packet meta information for mmap(2)ed RX_RING and TX_RINGs. If your
1030 NIC is capable of timestamping packets in hardware, you can request those
1031 hardware timestamps to be used. Note: you may need to enable the generation
1032 of hardware timestamps with SIOCSHWTSTAMP (see related information from
1033 Documentation/networking/timestamping.rst).
1034
1035 PACKET_TIMESTAMP accepts the same integer bit field as SO_TIMESTAMPING::
1036
1037 int req = SOF_TIMESTAMPING_RAW_HARDWARE;
1038 setsockopt(fd, SOL_PACKET, PACKET_TIMESTAMP, (void *) &req, sizeof(req))
1039
1040 For the mmap(2)ed ring buffers, such timestamps are stored in the
1041 ``tpacket{,2,3}_hdr`` structure's tp_sec and ``tp_{n,u}sec`` members.
1042 To determine what kind of timestamp has been reported, the tp_status field
1043 is binary or'ed with the following possible bits ...
1044
1045 ::
1046
1047 TP_STATUS_TS_RAW_HARDWARE
1048 TP_STATUS_TS_SOFTWARE
1049
1050 ... that are equivalent to its ``SOF_TIMESTAMPING_*`` counterparts. For the
1051 RX_RING, if neither is set (i.e. PACKET_TIMESTAMP is not set), then a
1052 software fallback was invoked *within* PF_PACKET's processing code (less
1053 precise).
1054
1055 Getting timestamps for the TX_RING works as follows: i) fill the ring frames,
1056 ii) call sendto() e.g. in blocking mode, iii) wait for status of relevant
1057 frames to be updated resp. the frame handed over to the application, iv) walk
1058 through the frames to pick up the individual hw/sw timestamps.
1059
1060 Only (!) if transmit timestamping is enabled, then these bits are combined
1061 with binary | with TP_STATUS_AVAILABLE, so you must check for that in your
1062 application (e.g. !(tp_status & (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING))
1063 in a first step to see if the frame belongs to the application, and then
1064 one can extract the type of timestamp in a second step from tp_status)!
1065
1066 If you don't care about them, thus having it disabled, checking for
1067 TP_STATUS_AVAILABLE resp. TP_STATUS_WRONG_FORMAT is sufficient. If in the
1068 TX_RING part only TP_STATUS_AVAILABLE is set, then the tp_sec and tp_{n,u}sec
1069 members do not contain a valid value. For TX_RINGs, by default no timestamp
1070 is generated!
1071
1072 See include/linux/net_tstamp.h and Documentation/networking/timestamping.rst
1073 for more information on hardware timestamps.
1074
1075 Miscellaneous bits
1076 ==================
1077
1078 - Packet sockets work well together with Linux socket filters, thus you also
1079 might want to have a look at Documentation/networking/filter.rst
1080
1081 THANKS
1082 ======
1083
1084 Jesse Brandeburg, for fixing my grammathical/spelling errors
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.