xref: /freebsd/sys/net/netmap_user.h (revision 3bdf775801b218aa5a89564839405b122f4b233e)
1 /*
2  * Copyright (C) 2011-2014 Universita` di Pisa. All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  *
8  *   1. Redistributions of source code must retain the above copyright
9  *      notice, this list of conditions and the following disclaimer.
10  *   2. Redistributions in binary form must reproduce the above copyright
11  *      notice, this list of conditions and the following disclaimer in the
12  *      documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 /*
28  * $FreeBSD$
29  *
30  * Functions and macros to manipulate netmap structures and packets
31  * in userspace. See netmap(4) for more information.
32  *
33  * The address of the struct netmap_if, say nifp, is computed from the
34  * value returned from ioctl(.., NIOCREG, ...) and the mmap region:
35  *	ioctl(fd, NIOCREG, &req);
36  *	mem = mmap(0, ... );
37  *	nifp = NETMAP_IF(mem, req.nr_nifp);
38  *		(so simple, we could just do it manually)
39  *
40  * From there:
41  *	struct netmap_ring *NETMAP_TXRING(nifp, index)
42  *	struct netmap_ring *NETMAP_RXRING(nifp, index)
43  *		we can access ring->nr_cur, ring->nr_avail, ring->nr_flags
44  *
45  *	ring->slot[i] gives us the i-th slot (we can access
46  *		directly len, flags, buf_idx)
47  *
48  *	char *buf = NETMAP_BUF(ring, x) returns a pointer to
49  *		the buffer numbered x
50  *
51  * All ring indexes (head, cur, tail) should always move forward.
52  * To compute the next index in a circular ring you can use
53  *	i = nm_ring_next(ring, i);
54  *
55  * To ease porting apps from pcap to netmap we supply a few fuctions
56  * that can be called to open, close, read and write on netmap in a way
57  * similar to libpcap. Note that the read/write function depend on
58  * an ioctl()/select()/poll() being issued to refill rings or push
59  * packets out.
60  *
61  * In order to use these, include #define NETMAP_WITH_LIBS
62  * in the source file that invokes these functions.
63  */
64 
65 #ifndef _NET_NETMAP_USER_H_
66 #define _NET_NETMAP_USER_H_
67 
68 #include <stdint.h>
69 #include <sys/socket.h>		/* apple needs sockaddr */
70 #include <net/if.h>		/* IFNAMSIZ */
71 
72 #ifndef likely
73 #define likely(x)	__builtin_expect(!!(x), 1)
74 #define unlikely(x)	__builtin_expect(!!(x), 0)
75 #endif /* likely and unlikely */
76 
77 #include <net/netmap.h>
78 
79 /* helper macro */
80 #define _NETMAP_OFFSET(type, ptr, offset) \
81 	((type)(void *)((char *)(ptr) + (offset)))
82 
83 #define NETMAP_IF(_base, _ofs)	_NETMAP_OFFSET(struct netmap_if *, _base, _ofs)
84 
85 #define NETMAP_TXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
86 	nifp, (nifp)->ring_ofs[index] )
87 
88 #define NETMAP_RXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *,	\
89 	nifp, (nifp)->ring_ofs[index + (nifp)->ni_tx_rings + 1] )
90 
91 #define NETMAP_BUF(ring, index)				\
92 	((char *)(ring) + (ring)->buf_ofs + ((index)*(ring)->nr_buf_size))
93 
94 #define NETMAP_BUF_IDX(ring, buf)			\
95 	( ((char *)(buf) - ((char *)(ring) + (ring)->buf_ofs) ) / \
96 		(ring)->nr_buf_size )
97 
98 
99 static inline uint32_t
100 nm_ring_next(struct netmap_ring *r, uint32_t i)
101 {
102 	return ( unlikely(i + 1 == r->num_slots) ? 0 : i + 1);
103 }
104 
105 
106 /*
107  * Return 1 if we have pending transmissions in the tx ring.
108  * When everything is complete ring->head = ring->tail + 1 (modulo ring size)
109  */
110 static inline int
111 nm_tx_pending(struct netmap_ring *r)
112 {
113 	return nm_ring_next(r, r->tail) != r->head;
114 }
115 
116 
117 static inline uint32_t
118 nm_ring_space(struct netmap_ring *ring)
119 {
120         int ret = ring->tail - ring->cur;
121         if (ret < 0)
122                 ret += ring->num_slots;
123         return ret;
124 }
125 
126 
127 #ifdef NETMAP_WITH_LIBS
128 /*
129  * Support for simple I/O libraries.
130  * Include other system headers required for compiling this.
131  */
132 
133 #ifndef HAVE_NETMAP_WITH_LIBS
134 #define HAVE_NETMAP_WITH_LIBS
135 
136 #include <sys/time.h>
137 #include <sys/mman.h>
138 #include <string.h>	/* memset */
139 #include <sys/ioctl.h>
140 #include <sys/errno.h>	/* EINVAL */
141 #include <fcntl.h>	/* O_RDWR */
142 #include <unistd.h>	/* close() */
143 #include <signal.h>
144 #include <stdlib.h>
145 
146 #ifndef ND /* debug macros */
147 /* debug support */
148 #define ND(_fmt, ...) do {} while(0)
149 #define D(_fmt, ...)						\
150 	do {							\
151 		struct timeval t0;				\
152 		gettimeofday(&t0, NULL);			\
153 		fprintf(stderr, "%03d.%06d %s [%d] " _fmt "\n",	\
154 		    (int)(t0.tv_sec % 1000), (int)t0.tv_usec,	\
155 		    __FUNCTION__, __LINE__, ##__VA_ARGS__);	\
156         } while (0)
157 
158 /* Rate limited version of "D", lps indicates how many per second */
159 #define RD(lps, format, ...)                                    \
160     do {                                                        \
161         static int t0, __cnt;                                   \
162         struct timeval __xxts;                                  \
163         gettimeofday(&__xxts, NULL);                            \
164         if (t0 != __xxts.tv_sec) {                              \
165             t0 = __xxts.tv_sec;                                 \
166             __cnt = 0;                                          \
167         }                                                       \
168         if (__cnt++ < lps) {                                    \
169             D(format, ##__VA_ARGS__);                           \
170         }                                                       \
171     } while (0)
172 #endif
173 
174 struct nm_pkthdr {	/* same as pcap_pkthdr */
175 	struct timeval	ts;
176 	uint32_t	caplen;
177 	uint32_t	len;
178 };
179 
180 struct nm_stat {	/* same as pcap_stat	*/
181 	u_int	ps_recv;
182 	u_int	ps_drop;
183 	u_int	ps_ifdrop;
184 #ifdef WIN32
185 	u_int	bs_capt;
186 #endif /* WIN32 */
187 };
188 
189 #define NM_ERRBUF_SIZE	512
190 
191 struct nm_desc {
192 	struct nm_desc *self; /* point to self if netmap. */
193 	int fd;
194 	void *mem;
195 	int memsize;
196 	int done_mmap;	/* set if mem is the result of mmap */
197 	struct netmap_if * const nifp;
198 	uint16_t first_tx_ring, last_tx_ring, cur_tx_ring;
199 	uint16_t first_rx_ring, last_rx_ring, cur_rx_ring;
200 	struct nmreq req;	/* also contains the nr_name = ifname */
201 	struct nm_pkthdr hdr;
202 
203 	/*
204 	 * The memory contains netmap_if, rings and then buffers.
205 	 * Given a pointer (e.g. to nm_inject) we can compare with
206 	 * mem/buf_start/buf_end to tell if it is a buffer or
207 	 * some other descriptor in our region.
208 	 * We also store a pointer to some ring as it helps in the
209 	 * translation from buffer indexes to addresses.
210 	 */
211 	struct netmap_ring * const some_ring;
212 	void * const buf_start;
213 	void * const buf_end;
214 	/* parameters from pcap_open_live */
215 	int snaplen;
216 	int promisc;
217 	int to_ms;
218 	char *errbuf;
219 
220 	/* save flags so we can restore them on close */
221 	uint32_t if_flags;
222         uint32_t if_reqcap;
223         uint32_t if_curcap;
224 
225 	struct nm_stat st;
226 	char msg[NM_ERRBUF_SIZE];
227 };
228 
229 /*
230  * when the descriptor is open correctly, d->self == d
231  * Eventually we should also use some magic number.
232  */
233 #define P2NMD(p)		((struct nm_desc *)(p))
234 #define IS_NETMAP_DESC(d)	((d) && P2NMD(d)->self == P2NMD(d))
235 #define NETMAP_FD(d)		(P2NMD(d)->fd)
236 
237 
238 /*
239  * this is a slightly optimized copy routine which rounds
240  * to multiple of 64 bytes and is often faster than dealing
241  * with other odd sizes. We assume there is enough room
242  * in the source and destination buffers.
243  *
244  * XXX only for multiples of 64 bytes, non overlapped.
245  */
246 static inline void
247 nm_pkt_copy(const void *_src, void *_dst, int l)
248 {
249 	const uint64_t *src = (const uint64_t *)_src;
250 	uint64_t *dst = (uint64_t *)_dst;
251 
252 	if (unlikely(l >= 1024)) {
253 		memcpy(dst, src, l);
254 		return;
255 	}
256 	for (; likely(l > 0); l-=64) {
257 		*dst++ = *src++;
258 		*dst++ = *src++;
259 		*dst++ = *src++;
260 		*dst++ = *src++;
261 		*dst++ = *src++;
262 		*dst++ = *src++;
263 		*dst++ = *src++;
264 		*dst++ = *src++;
265 	}
266 }
267 
268 
269 /*
270  * The callback, invoked on each received packet. Same as libpcap
271  */
272 typedef void (*nm_cb_t)(u_char *, const struct nm_pkthdr *, const u_char *d);
273 
274 /*
275  *--- the pcap-like API ---
276  *
277  * nm_open() opens a file descriptor, binds to a port and maps memory.
278  *
279  * ifname	(netmap:foo or vale:foo) is the port name
280  *		a suffix can indicate the follwing:
281  *		^		bind the host (sw) ring pair
282  *		*		bind host and NIC ring pairs (transparent)
283  *		-NN		bind individual NIC ring pair
284  *		{NN		bind master side of pipe NN
285  *		}NN		bind slave side of pipe NN
286  *
287  * req		provides the initial values of nmreq before parsing ifname.
288  *		Remember that the ifname parsing will override the ring
289  *		number in nm_ringid, and part of nm_flags;
290  * flags	special functions, normally 0
291  *		indicates which fields of *arg are significant
292  * arg		special functions, normally NULL
293  *		if passed a netmap_desc with mem != NULL,
294  *		use that memory instead of mmap.
295  */
296 
297 static struct nm_desc *nm_open(const char *ifname, const struct nmreq *req,
298 	uint64_t flags, const struct nm_desc *arg);
299 
300 /*
301  * nm_open can import some fields from the parent descriptor.
302  * These flags control which ones.
303  * Also in flags you can specify NETMAP_NO_TX_POLL and NETMAP_DO_RX_POLL,
304  * which set the initial value for these flags.
305  * Note that the 16 low bits of the flags are reserved for data
306  * that may go into the nmreq.
307  */
308 enum {
309 	NM_OPEN_NO_MMAP =	0x040000, /* reuse mmap from parent */
310 	NM_OPEN_IFNAME =	0x080000, /* nr_name, nr_ringid, nr_flags */
311 	NM_OPEN_ARG1 =		0x100000,
312 	NM_OPEN_ARG2 =		0x200000,
313 	NM_OPEN_ARG3 =		0x400000,
314 	NM_OPEN_RING_CFG =	0x800000, /* tx|rx rings|slots */
315 };
316 
317 
318 /*
319  * nm_close()	closes and restores the port to its previous state
320  */
321 
322 static int nm_close(struct nm_desc *);
323 
324 /*
325  * nm_inject() is the same as pcap_inject()
326  * nm_dispatch() is the same as pcap_dispatch()
327  * nm_nextpkt() is the same as pcap_next()
328  */
329 
330 static int nm_inject(struct nm_desc *, const void *, size_t);
331 static int nm_dispatch(struct nm_desc *, int, nm_cb_t, u_char *);
332 static u_char *nm_nextpkt(struct nm_desc *, struct nm_pkthdr *);
333 
334 
335 /*
336  * Try to open, return descriptor if successful, NULL otherwise.
337  * An invalid netmap name will return errno = 0;
338  * You can pass a pointer to a pre-filled nm_desc to add special
339  * parameters. Flags is used as follows
340  * NM_OPEN_NO_MMAP	use the memory from arg, only
341  *			if the nr_arg2 (memory block) matches.
342  * NM_OPEN_ARG1		use req.nr_arg1 from arg
343  * NM_OPEN_ARG2		use req.nr_arg2 from arg
344  * NM_OPEN_RING_CFG	user ring config from arg
345  */
346 static struct nm_desc *
347 nm_open(const char *ifname, const struct nmreq *req,
348 	uint64_t new_flags, const struct nm_desc *arg)
349 {
350 	struct nm_desc *d = NULL;
351 	const struct nm_desc *parent = arg;
352 	u_int namelen;
353 	uint32_t nr_ringid = 0, nr_flags;
354 	const char *port = NULL;
355 	const char *errmsg = NULL;
356 
357 	if (strncmp(ifname, "netmap:", 7) && strncmp(ifname, "vale", 4)) {
358 		errno = 0; /* name not recognised, not an error */
359 		return NULL;
360 	}
361 	if (ifname[0] == 'n')
362 		ifname += 7;
363 	/* scan for a separator */
364 	for (port = ifname; *port && !index("-*^{}", *port); port++)
365 		;
366 	namelen = port - ifname;
367 	if (namelen >= sizeof(d->req.nr_name)) {
368 		errmsg = "name too long";
369 		goto fail;
370 	}
371 	switch (*port) {
372 	default:  /* '\0', no suffix */
373 		nr_flags = NR_REG_ALL_NIC;
374 		break;
375 	case '-': /* one NIC */
376 		nr_flags = NR_REG_ONE_NIC;
377 		nr_ringid = atoi(port + 1);
378 		break;
379 	case '*': /* NIC and SW, ignore port */
380 		nr_flags = NR_REG_NIC_SW;
381 		if (port[1]) {
382 			errmsg = "invalid port for nic+sw";
383 			goto fail;
384 		}
385 		break;
386 	case '^': /* only sw ring */
387 		nr_flags = NR_REG_SW;
388 		if (port[1]) {
389 			errmsg = "invalid port for sw ring";
390 			goto fail;
391 		}
392 		break;
393 	case '{':
394 		nr_flags = NR_REG_PIPE_MASTER;
395 		nr_ringid = atoi(port + 1);
396 		break;
397 	case '}':
398 		nr_flags = NR_REG_PIPE_SLAVE;
399 		nr_ringid = atoi(port + 1);
400 		break;
401 	}
402 
403 	if (nr_ringid >= NETMAP_RING_MASK) {
404 		errmsg = "invalid ringid";
405 		goto fail;
406 	}
407 	/* add the *XPOLL flags */
408 	nr_ringid |= new_flags & (NETMAP_NO_TX_POLL | NETMAP_DO_RX_POLL);
409 
410 	d = (struct nm_desc *)calloc(1, sizeof(*d));
411 	if (d == NULL) {
412 		errmsg = "nm_desc alloc failure";
413 		errno = ENOMEM;
414 		return NULL;
415 	}
416 	d->self = d;	/* set this early so nm_close() works */
417 	d->fd = open("/dev/netmap", O_RDWR);
418 	if (d->fd < 0) {
419 		errmsg = "cannot open /dev/netmap";
420 		goto fail;
421 	}
422 
423 	if (req)
424 		d->req = *req;
425 	d->req.nr_version = NETMAP_API;
426 	d->req.nr_ringid &= ~NETMAP_RING_MASK;
427 
428 	/* these fields are overridden by ifname and flags processing */
429 	d->req.nr_ringid |= nr_ringid;
430 	d->req.nr_flags = nr_flags;
431 	memcpy(d->req.nr_name, ifname, namelen);
432 	d->req.nr_name[namelen] = '\0';
433 	/* optionally import info from parent */
434 	if (IS_NETMAP_DESC(parent) && new_flags) {
435 		if (new_flags & NM_OPEN_ARG1)
436 			D("overriding ARG1 %d", parent->req.nr_arg1);
437 		d->req.nr_arg1 = new_flags & NM_OPEN_ARG1 ?
438 			parent->req.nr_arg1 : 4;
439 		if (new_flags & NM_OPEN_ARG2)
440 			D("overriding ARG2 %d", parent->req.nr_arg2);
441 		d->req.nr_arg2 = new_flags & NM_OPEN_ARG2 ?
442 			parent->req.nr_arg2 : 0;
443 		if (new_flags & NM_OPEN_ARG3)
444 			D("overriding ARG3 %d", parent->req.nr_arg3);
445 		d->req.nr_arg3 = new_flags & NM_OPEN_ARG3 ?
446 			parent->req.nr_arg3 : 0;
447 		if (new_flags & NM_OPEN_RING_CFG) {
448 			D("overriding RING_CFG");
449 			d->req.nr_tx_slots = parent->req.nr_tx_slots;
450 			d->req.nr_rx_slots = parent->req.nr_rx_slots;
451 			d->req.nr_tx_rings = parent->req.nr_tx_rings;
452 			d->req.nr_rx_rings = parent->req.nr_rx_rings;
453 		}
454 		if (new_flags & NM_OPEN_IFNAME) {
455 			D("overriding ifname %s ringid 0x%x flags 0x%x",
456 				parent->req.nr_name, parent->req.nr_ringid,
457 				parent->req.nr_flags);
458 			memcpy(d->req.nr_name, parent->req.nr_name,
459 				sizeof(d->req.nr_name));
460 			d->req.nr_ringid = parent->req.nr_ringid;
461 			d->req.nr_flags = parent->req.nr_flags;
462 		}
463 	}
464 	if (ioctl(d->fd, NIOCREGIF, &d->req)) {
465 		errmsg = "NIOCREGIF failed";
466 		goto fail;
467 	}
468 
469 	if (IS_NETMAP_DESC(parent) && parent->mem &&
470 	    parent->req.nr_arg2 == d->req.nr_arg2) {
471 		/* do not mmap, inherit from parent */
472 		d->memsize = parent->memsize;
473 		d->mem = parent->mem;
474 	} else {
475 		d->memsize = d->req.nr_memsize;
476 		d->mem = mmap(0, d->memsize, PROT_WRITE | PROT_READ, MAP_SHARED,
477 				d->fd, 0);
478 		if (d->mem == NULL) {
479 			errmsg = "mmap failed";
480 			goto fail;
481 		}
482 		d->done_mmap = 1;
483 	}
484 	{
485 		struct netmap_if *nifp = NETMAP_IF(d->mem, d->req.nr_offset);
486 		struct netmap_ring *r = NETMAP_RXRING(nifp, );
487 
488 		*(struct netmap_if **)(uintptr_t)&(d->nifp) = nifp;
489 		*(struct netmap_ring **)(uintptr_t)&d->some_ring = r;
490 		*(void **)(uintptr_t)&d->buf_start = NETMAP_BUF(r, 0);
491 		*(void **)(uintptr_t)&d->buf_end =
492 			(char *)d->mem + d->memsize;
493 	}
494 
495 	if (nr_flags ==  NR_REG_SW) { /* host stack */
496 		d->first_tx_ring = d->last_tx_ring = d->req.nr_tx_rings;
497 		d->first_rx_ring = d->last_rx_ring = d->req.nr_rx_rings;
498 	} else if (nr_flags ==  NR_REG_ALL_NIC) { /* only nic */
499 		d->first_tx_ring = 0;
500 		d->first_rx_ring = 0;
501 		d->last_tx_ring = d->req.nr_tx_rings - 1;
502 		d->last_rx_ring = d->req.nr_rx_rings - 1;
503 	} else if (nr_flags ==  NR_REG_NIC_SW) {
504 		d->first_tx_ring = 0;
505 		d->first_rx_ring = 0;
506 		d->last_tx_ring = d->req.nr_tx_rings;
507 		d->last_rx_ring = d->req.nr_rx_rings;
508 	} else if (nr_flags == NR_REG_ONE_NIC) {
509 		/* XXX check validity */
510 		d->first_tx_ring = d->last_tx_ring =
511 		d->first_rx_ring = d->last_rx_ring = nr_ringid;
512 	} else { /* pipes */
513 		d->first_tx_ring = d->last_tx_ring = 0;
514 		d->first_rx_ring = d->last_rx_ring = 0;
515 	}
516 
517 #ifdef DEBUG_NETMAP_USER
518     { /* debugging code */
519 	int i;
520 
521 	D("%s tx %d .. %d %d rx %d .. %d %d", ifname,
522 		d->first_tx_ring, d->last_tx_ring, d->req.nr_tx_rings,
523                 d->first_rx_ring, d->last_rx_ring, d->req.nr_rx_rings);
524 	for (i = 0; i <= d->req.nr_tx_rings; i++) {
525 		struct netmap_ring *r = NETMAP_TXRING(d->nifp, i);
526 		D("TX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
527 	}
528 	for (i = 0; i <= d->req.nr_rx_rings; i++) {
529 		struct netmap_ring *r = NETMAP_RXRING(d->nifp, i);
530 		D("RX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
531 	}
532     }
533 #endif /* debugging */
534 
535 	d->cur_tx_ring = d->first_tx_ring;
536 	d->cur_rx_ring = d->first_rx_ring;
537 	return d;
538 
539 fail:
540 	nm_close(d);
541 	if (errmsg)
542 		D("%s %s", errmsg, ifname);
543 	errno = EINVAL;
544 	return NULL;
545 }
546 
547 
548 static int
549 nm_close(struct nm_desc *d)
550 {
551 	/*
552 	 * ugly trick to avoid unused warnings
553 	 */
554 	static void *__xxzt[] __attribute__ ((unused))  =
555 		{ (void *)nm_open, (void *)nm_inject,
556 		  (void *)nm_dispatch, (void *)nm_nextpkt } ;
557 
558 	if (d == NULL || d->self != d)
559 		return EINVAL;
560 	if (d->done_mmap && d->mem)
561 		munmap(d->mem, d->memsize);
562 	if (d->fd != -1)
563 		close(d->fd);
564 	bzero(d, sizeof(*d));
565 	free(d);
566 	return 0;
567 }
568 
569 
570 /*
571  * Same prototype as pcap_inject(), only need to cast.
572  */
573 static int
574 nm_inject(struct nm_desc *d, const void *buf, size_t size)
575 {
576 	u_int c, n = d->last_tx_ring - d->first_tx_ring + 1;
577 
578 	for (c = 0; c < n ; c++) {
579 		/* compute current ring to use */
580 		struct netmap_ring *ring;
581 		uint32_t i, idx;
582 		uint32_t ri = d->cur_tx_ring + c;
583 
584 		if (ri > d->last_tx_ring)
585 			ri = d->first_tx_ring;
586 		ring = NETMAP_TXRING(d->nifp, ri);
587 		if (nm_ring_empty(ring)) {
588 			continue;
589 		}
590 		i = ring->cur;
591 		idx = ring->slot[i].buf_idx;
592 		ring->slot[i].len = size;
593 		nm_pkt_copy(buf, NETMAP_BUF(ring, idx), size);
594 		d->cur_tx_ring = ri;
595 		ring->head = ring->cur = nm_ring_next(ring, i);
596 		return size;
597 	}
598 	return 0; /* fail */
599 }
600 
601 
602 /*
603  * Same prototype as pcap_dispatch(), only need to cast.
604  */
605 static int
606 nm_dispatch(struct nm_desc *d, int cnt, nm_cb_t cb, u_char *arg)
607 {
608 	int n = d->last_rx_ring - d->first_rx_ring + 1;
609 	int c, got = 0, ri = d->cur_rx_ring;
610 
611 	if (cnt == 0)
612 		cnt = -1;
613 	/* cnt == -1 means infinite, but rings have a finite amount
614 	 * of buffers and the int is large enough that we never wrap,
615 	 * so we can omit checking for -1
616 	 */
617 	for (c=0; c < n && cnt != got; c++) {
618 		/* compute current ring to use */
619 		struct netmap_ring *ring;
620 
621 		ri = d->cur_rx_ring + c;
622 		if (ri > d->last_rx_ring)
623 			ri = d->first_rx_ring;
624 		ring = NETMAP_RXRING(d->nifp, ri);
625 		for ( ; !nm_ring_empty(ring) && cnt != got; got++) {
626 			u_int i = ring->cur;
627 			u_int idx = ring->slot[i].buf_idx;
628 			u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
629 
630 			// __builtin_prefetch(buf);
631 			d->hdr.len = d->hdr.caplen = ring->slot[i].len;
632 			d->hdr.ts = ring->ts;
633 			cb(arg, &d->hdr, buf);
634 			ring->head = ring->cur = nm_ring_next(ring, i);
635 		}
636 	}
637 	d->cur_rx_ring = ri;
638 	return got;
639 }
640 
641 static u_char *
642 nm_nextpkt(struct nm_desc *d, struct nm_pkthdr *hdr)
643 {
644 	int ri = d->cur_rx_ring;
645 
646 	do {
647 		/* compute current ring to use */
648 		struct netmap_ring *ring = NETMAP_RXRING(d->nifp, ri);
649 		if (!nm_ring_empty(ring)) {
650 			u_int i = ring->cur;
651 			u_int idx = ring->slot[i].buf_idx;
652 			u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
653 
654 			// __builtin_prefetch(buf);
655 			hdr->ts = ring->ts;
656 			hdr->len = hdr->caplen = ring->slot[i].len;
657 			ring->cur = nm_ring_next(ring, i);
658 			/* we could postpone advancing head if we want
659 			 * to hold the buffer. This can be supported in
660 			 * the future.
661 			 */
662 			ring->head = ring->cur;
663 			d->cur_rx_ring = ri;
664 			return buf;
665 		}
666 		ri++;
667 		if (ri > d->last_rx_ring)
668 			ri = d->first_rx_ring;
669 	} while (ri != d->cur_rx_ring);
670 	return NULL; /* nothing found */
671 }
672 
673 #endif /* !HAVE_NETMAP_WITH_LIBS */
674 
675 #endif /* NETMAP_WITH_LIBS */
676 
677 #endif /* _NET_NETMAP_USER_H_ */
678