xref: /freebsd/sys/net/netmap_user.h (revision 4436b51dff5736e74da464946049ea6899a88938)
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->cur, ring->head, ring->tail, etc.
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 <stdio.h>
137 #include <sys/time.h>
138 #include <sys/mman.h>
139 #include <string.h>	/* memset */
140 #include <sys/ioctl.h>
141 #include <sys/errno.h>	/* EINVAL */
142 #include <fcntl.h>	/* O_RDWR */
143 #include <unistd.h>	/* close() */
144 #include <signal.h>
145 #include <stdlib.h>
146 
147 #ifndef ND /* debug macros */
148 /* debug support */
149 #define ND(_fmt, ...) do {} while(0)
150 #define D(_fmt, ...)						\
151 	do {							\
152 		struct timeval _t0;				\
153 		gettimeofday(&_t0, NULL);			\
154 		fprintf(stderr, "%03d.%06d %s [%d] " _fmt "\n",	\
155 		    (int)(_t0.tv_sec % 1000), (int)_t0.tv_usec,	\
156 		    __FUNCTION__, __LINE__, ##__VA_ARGS__);	\
157         } while (0)
158 
159 /* Rate limited version of "D", lps indicates how many per second */
160 #define RD(lps, format, ...)                                    \
161     do {                                                        \
162         static int __t0, __cnt;                                 \
163         struct timeval __xxts;                                  \
164         gettimeofday(&__xxts, NULL);                            \
165         if (__t0 != __xxts.tv_sec) {                            \
166             __t0 = __xxts.tv_sec;                               \
167             __cnt = 0;                                          \
168         }                                                       \
169         if (__cnt++ < lps) {                                    \
170             D(format, ##__VA_ARGS__);                           \
171         }                                                       \
172     } while (0)
173 #endif
174 
175 struct nm_pkthdr {	/* same as pcap_pkthdr */
176 	struct timeval	ts;
177 	uint32_t	caplen;
178 	uint32_t	len;
179 };
180 
181 struct nm_stat {	/* same as pcap_stat	*/
182 	u_int	ps_recv;
183 	u_int	ps_drop;
184 	u_int	ps_ifdrop;
185 #ifdef WIN32
186 	u_int	bs_capt;
187 #endif /* WIN32 */
188 };
189 
190 #define NM_ERRBUF_SIZE	512
191 
192 struct nm_desc {
193 	struct nm_desc *self; /* point to self if netmap. */
194 	int fd;
195 	void *mem;
196 	uint32_t memsize;
197 	int done_mmap;	/* set if mem is the result of mmap */
198 	struct netmap_if * const nifp;
199 	uint16_t first_tx_ring, last_tx_ring, cur_tx_ring;
200 	uint16_t first_rx_ring, last_rx_ring, cur_rx_ring;
201 	struct nmreq req;	/* also contains the nr_name = ifname */
202 	struct nm_pkthdr hdr;
203 
204 	/*
205 	 * The memory contains netmap_if, rings and then buffers.
206 	 * Given a pointer (e.g. to nm_inject) we can compare with
207 	 * mem/buf_start/buf_end to tell if it is a buffer or
208 	 * some other descriptor in our region.
209 	 * We also store a pointer to some ring as it helps in the
210 	 * translation from buffer indexes to addresses.
211 	 */
212 	struct netmap_ring * const some_ring;
213 	void * const buf_start;
214 	void * const buf_end;
215 	/* parameters from pcap_open_live */
216 	int snaplen;
217 	int promisc;
218 	int to_ms;
219 	char *errbuf;
220 
221 	/* save flags so we can restore them on close */
222 	uint32_t if_flags;
223         uint32_t if_reqcap;
224         uint32_t if_curcap;
225 
226 	struct nm_stat st;
227 	char msg[NM_ERRBUF_SIZE];
228 };
229 
230 /*
231  * when the descriptor is open correctly, d->self == d
232  * Eventually we should also use some magic number.
233  */
234 #define P2NMD(p)		((struct nm_desc *)(p))
235 #define IS_NETMAP_DESC(d)	((d) && P2NMD(d)->self == P2NMD(d))
236 #define NETMAP_FD(d)		(P2NMD(d)->fd)
237 
238 
239 /*
240  * this is a slightly optimized copy routine which rounds
241  * to multiple of 64 bytes and is often faster than dealing
242  * with other odd sizes. We assume there is enough room
243  * in the source and destination buffers.
244  *
245  * XXX only for multiples of 64 bytes, non overlapped.
246  */
247 static inline void
248 nm_pkt_copy(const void *_src, void *_dst, int l)
249 {
250 	const uint64_t *src = (const uint64_t *)_src;
251 	uint64_t *dst = (uint64_t *)_dst;
252 
253 	if (unlikely(l >= 1024)) {
254 		memcpy(dst, src, l);
255 		return;
256 	}
257 	for (; likely(l > 0); l-=64) {
258 		*dst++ = *src++;
259 		*dst++ = *src++;
260 		*dst++ = *src++;
261 		*dst++ = *src++;
262 		*dst++ = *src++;
263 		*dst++ = *src++;
264 		*dst++ = *src++;
265 		*dst++ = *src++;
266 	}
267 }
268 
269 
270 /*
271  * The callback, invoked on each received packet. Same as libpcap
272  */
273 typedef void (*nm_cb_t)(u_char *, const struct nm_pkthdr *, const u_char *d);
274 
275 /*
276  *--- the pcap-like API ---
277  *
278  * nm_open() opens a file descriptor, binds to a port and maps memory.
279  *
280  * ifname	(netmap:foo or vale:foo) is the port name
281  *		a suffix can indicate the follwing:
282  *		^		bind the host (sw) ring pair
283  *		*		bind host and NIC ring pairs (transparent)
284  *		-NN		bind individual NIC ring pair
285  *		{NN		bind master side of pipe NN
286  *		}NN		bind slave side of pipe NN
287  *		a suffix starting with + and the following flags,
288  *		in any order:
289  *		x		exclusive access
290  *		z		zero copy monitor
291  *		t		monitor tx side
292  *		r		monitor rx side
293  *
294  * req		provides the initial values of nmreq before parsing ifname.
295  *		Remember that the ifname parsing will override the ring
296  *		number in nm_ringid, and part of nm_flags;
297  * flags	special functions, normally 0
298  *		indicates which fields of *arg are significant
299  * arg		special functions, normally NULL
300  *		if passed a netmap_desc with mem != NULL,
301  *		use that memory instead of mmap.
302  */
303 
304 static struct nm_desc *nm_open(const char *ifname, const struct nmreq *req,
305 	uint64_t flags, const struct nm_desc *arg);
306 
307 /*
308  * nm_open can import some fields from the parent descriptor.
309  * These flags control which ones.
310  * Also in flags you can specify NETMAP_NO_TX_POLL and NETMAP_DO_RX_POLL,
311  * which set the initial value for these flags.
312  * Note that the 16 low bits of the flags are reserved for data
313  * that may go into the nmreq.
314  */
315 enum {
316 	NM_OPEN_NO_MMAP =	0x040000, /* reuse mmap from parent */
317 	NM_OPEN_IFNAME =	0x080000, /* nr_name, nr_ringid, nr_flags */
318 	NM_OPEN_ARG1 =		0x100000,
319 	NM_OPEN_ARG2 =		0x200000,
320 	NM_OPEN_ARG3 =		0x400000,
321 	NM_OPEN_RING_CFG =	0x800000, /* tx|rx rings|slots */
322 };
323 
324 
325 /*
326  * nm_close()	closes and restores the port to its previous state
327  */
328 
329 static int nm_close(struct nm_desc *);
330 
331 /*
332  * nm_inject() is the same as pcap_inject()
333  * nm_dispatch() is the same as pcap_dispatch()
334  * nm_nextpkt() is the same as pcap_next()
335  */
336 
337 static int nm_inject(struct nm_desc *, const void *, size_t);
338 static int nm_dispatch(struct nm_desc *, int, nm_cb_t, u_char *);
339 static u_char *nm_nextpkt(struct nm_desc *, struct nm_pkthdr *);
340 
341 
342 /*
343  * Try to open, return descriptor if successful, NULL otherwise.
344  * An invalid netmap name will return errno = 0;
345  * You can pass a pointer to a pre-filled nm_desc to add special
346  * parameters. Flags is used as follows
347  * NM_OPEN_NO_MMAP	use the memory from arg, only
348  *			if the nr_arg2 (memory block) matches.
349  * NM_OPEN_ARG1		use req.nr_arg1 from arg
350  * NM_OPEN_ARG2		use req.nr_arg2 from arg
351  * NM_OPEN_RING_CFG	user ring config from arg
352  */
353 static struct nm_desc *
354 nm_open(const char *ifname, const struct nmreq *req,
355 	uint64_t new_flags, const struct nm_desc *arg)
356 {
357 	struct nm_desc *d = NULL;
358 	const struct nm_desc *parent = arg;
359 	u_int namelen;
360 	uint32_t nr_ringid = 0, nr_flags, nr_reg;
361 	const char *port = NULL;
362 #define MAXERRMSG 80
363 	char errmsg[MAXERRMSG] = "";
364 	enum { P_START, P_RNGSFXOK, P_GETNUM, P_FLAGS, P_FLAGSOK } p_state;
365 	long num;
366 
367 	if (strncmp(ifname, "netmap:", 7) && strncmp(ifname, "vale", 4)) {
368 		errno = 0; /* name not recognised, not an error */
369 		return NULL;
370 	}
371 	if (ifname[0] == 'n')
372 		ifname += 7;
373 	/* scan for a separator */
374 	for (port = ifname; *port && !index("-*^{}/", *port); port++)
375 		;
376 	namelen = port - ifname;
377 	if (namelen >= sizeof(d->req.nr_name)) {
378 		snprintf(errmsg, MAXERRMSG, "name too long");
379 		goto fail;
380 	}
381 	p_state = P_START;
382 	nr_flags = NR_REG_ALL_NIC; /* default for no suffix */
383 	while (*port) {
384 		switch (p_state) {
385 		case P_START:
386 			switch (*port) {
387 			case '^': /* only SW ring */
388 				nr_flags = NR_REG_SW;
389 				p_state = P_RNGSFXOK;
390 				break;
391 			case '*': /* NIC and SW */
392 				nr_flags = NR_REG_NIC_SW;
393 				p_state = P_RNGSFXOK;
394 				break;
395 			case '-': /* one NIC ring pair */
396 				nr_flags = NR_REG_ONE_NIC;
397 				p_state = P_GETNUM;
398 				break;
399 			case '{': /* pipe (master endpoint) */
400 				nr_flags = NR_REG_PIPE_MASTER;
401 				p_state = P_GETNUM;
402 				break;
403 			case '}': /* pipe (slave endoint) */
404 				nr_flags = NR_REG_PIPE_SLAVE;
405 				p_state = P_GETNUM;
406 				break;
407 			case '/': /* start of flags */
408 				p_state = P_FLAGS;
409 				break;
410 			default:
411 				snprintf(errmsg, MAXERRMSG, "unknown modifier: '%c'", *port);
412 				goto fail;
413 			}
414 			port++;
415 			break;
416 		case P_RNGSFXOK:
417 			switch (*port) {
418 			case '/':
419 				p_state = P_FLAGS;
420 				break;
421 			default:
422 				snprintf(errmsg, MAXERRMSG, "unexpected character: '%c'", *port);
423 				goto fail;
424 			}
425 			port++;
426 			break;
427 		case P_GETNUM:
428 			num = strtol(port, (char **)&port, 10);
429 			if (num < 0 || num >= NETMAP_RING_MASK) {
430 				snprintf(errmsg, MAXERRMSG, "'%ld' out of range [0, %d)",
431 						num, NETMAP_RING_MASK);
432 				goto fail;
433 			}
434 			nr_ringid = num & NETMAP_RING_MASK;
435 			p_state = P_RNGSFXOK;
436 			break;
437 		case P_FLAGS:
438 		case P_FLAGSOK:
439 			switch (*port) {
440 			case 'x':
441 				nr_flags |= NR_EXCLUSIVE;
442 				break;
443 			case 'z':
444 				nr_flags |= NR_ZCOPY_MON;
445 				break;
446 			case 't':
447 				nr_flags |= NR_MONITOR_TX;
448 				break;
449 			case 'r':
450 				nr_flags |= NR_MONITOR_RX;
451 				break;
452 			default:
453 				snprintf(errmsg, MAXERRMSG, "unrecognized flag: '%c'", *port);
454 				goto fail;
455 			}
456 			port++;
457 			p_state = P_FLAGSOK;
458 			break;
459 		}
460 	}
461 	if (p_state != P_START && p_state != P_RNGSFXOK && p_state != P_FLAGSOK) {
462 		snprintf(errmsg, MAXERRMSG, "unexpected end of port name");
463 		goto fail;
464 	}
465 	ND("flags: %s %s %s %s",
466 			(nr_flags & NR_EXCLUSIVE) ? "EXCLUSIVE" : "",
467 			(nr_flags & NR_ZCOPY_MON) ? "ZCOPY_MON" : "",
468 			(nr_flags & NR_MONITOR_TX) ? "MONITOR_TX" : "",
469 			(nr_flags & NR_MONITOR_RX) ? "MONITOR_RX" : "");
470 	d = (struct nm_desc *)calloc(1, sizeof(*d));
471 	if (d == NULL) {
472 		snprintf(errmsg, MAXERRMSG, "nm_desc alloc failure");
473 		errno = ENOMEM;
474 		return NULL;
475 	}
476 	d->self = d;	/* set this early so nm_close() works */
477 	d->fd = open("/dev/netmap", O_RDWR);
478 	if (d->fd < 0) {
479 		snprintf(errmsg, MAXERRMSG, "cannot open /dev/netmap: %s", strerror(errno));
480 		goto fail;
481 	}
482 
483 	if (req)
484 		d->req = *req;
485 	d->req.nr_version = NETMAP_API;
486 	d->req.nr_ringid &= ~NETMAP_RING_MASK;
487 
488 	/* these fields are overridden by ifname and flags processing */
489 	d->req.nr_ringid |= nr_ringid;
490 	d->req.nr_flags = nr_flags;
491 	memcpy(d->req.nr_name, ifname, namelen);
492 	d->req.nr_name[namelen] = '\0';
493 	/* optionally import info from parent */
494 	if (IS_NETMAP_DESC(parent) && new_flags) {
495 		if (new_flags & NM_OPEN_ARG1)
496 			D("overriding ARG1 %d", parent->req.nr_arg1);
497 		d->req.nr_arg1 = new_flags & NM_OPEN_ARG1 ?
498 			parent->req.nr_arg1 : 4;
499 		if (new_flags & NM_OPEN_ARG2)
500 			D("overriding ARG2 %d", parent->req.nr_arg2);
501 		d->req.nr_arg2 = new_flags & NM_OPEN_ARG2 ?
502 			parent->req.nr_arg2 : 0;
503 		if (new_flags & NM_OPEN_ARG3)
504 			D("overriding ARG3 %d", parent->req.nr_arg3);
505 		d->req.nr_arg3 = new_flags & NM_OPEN_ARG3 ?
506 			parent->req.nr_arg3 : 0;
507 		if (new_flags & NM_OPEN_RING_CFG) {
508 			D("overriding RING_CFG");
509 			d->req.nr_tx_slots = parent->req.nr_tx_slots;
510 			d->req.nr_rx_slots = parent->req.nr_rx_slots;
511 			d->req.nr_tx_rings = parent->req.nr_tx_rings;
512 			d->req.nr_rx_rings = parent->req.nr_rx_rings;
513 		}
514 		if (new_flags & NM_OPEN_IFNAME) {
515 			D("overriding ifname %s ringid 0x%x flags 0x%x",
516 				parent->req.nr_name, parent->req.nr_ringid,
517 				parent->req.nr_flags);
518 			memcpy(d->req.nr_name, parent->req.nr_name,
519 				sizeof(d->req.nr_name));
520 			d->req.nr_ringid = parent->req.nr_ringid;
521 			d->req.nr_flags = parent->req.nr_flags;
522 		}
523 	}
524 	/* add the *XPOLL flags */
525 	d->req.nr_ringid |= new_flags & (NETMAP_NO_TX_POLL | NETMAP_DO_RX_POLL);
526 
527 	if (ioctl(d->fd, NIOCREGIF, &d->req)) {
528 		snprintf(errmsg, MAXERRMSG, "NIOCREGIF failed: %s", strerror(errno));
529 		goto fail;
530 	}
531 
532 	if (IS_NETMAP_DESC(parent) && parent->mem &&
533 	    parent->req.nr_arg2 == d->req.nr_arg2) {
534 		/* do not mmap, inherit from parent */
535 		d->memsize = parent->memsize;
536 		d->mem = parent->mem;
537 	} else {
538 		/* XXX TODO: check if memsize is too large (or there is overflow) */
539 		d->memsize = d->req.nr_memsize;
540 		d->mem = mmap(0, d->memsize, PROT_WRITE | PROT_READ, MAP_SHARED,
541 				d->fd, 0);
542 		if (d->mem == MAP_FAILED) {
543 			snprintf(errmsg, MAXERRMSG, "mmap failed: %s", strerror(errno));
544 			goto fail;
545 		}
546 		d->done_mmap = 1;
547 	}
548 	{
549 		struct netmap_if *nifp = NETMAP_IF(d->mem, d->req.nr_offset);
550 		struct netmap_ring *r = NETMAP_RXRING(nifp, );
551 
552 		*(struct netmap_if **)(uintptr_t)&(d->nifp) = nifp;
553 		*(struct netmap_ring **)(uintptr_t)&d->some_ring = r;
554 		*(void **)(uintptr_t)&d->buf_start = NETMAP_BUF(r, 0);
555 		*(void **)(uintptr_t)&d->buf_end =
556 			(char *)d->mem + d->memsize;
557 	}
558 
559 	nr_reg = d->req.nr_flags & NR_REG_MASK;
560 
561 	if (nr_reg ==  NR_REG_SW) { /* host stack */
562 		d->first_tx_ring = d->last_tx_ring = d->req.nr_tx_rings;
563 		d->first_rx_ring = d->last_rx_ring = d->req.nr_rx_rings;
564 	} else if (nr_reg ==  NR_REG_ALL_NIC) { /* only nic */
565 		d->first_tx_ring = 0;
566 		d->first_rx_ring = 0;
567 		d->last_tx_ring = d->req.nr_tx_rings - 1;
568 		d->last_rx_ring = d->req.nr_rx_rings - 1;
569 	} else if (nr_reg ==  NR_REG_NIC_SW) {
570 		d->first_tx_ring = 0;
571 		d->first_rx_ring = 0;
572 		d->last_tx_ring = d->req.nr_tx_rings;
573 		d->last_rx_ring = d->req.nr_rx_rings;
574 	} else if (nr_reg == NR_REG_ONE_NIC) {
575 		/* XXX check validity */
576 		d->first_tx_ring = d->last_tx_ring =
577 		d->first_rx_ring = d->last_rx_ring = d->req.nr_ringid & NETMAP_RING_MASK;
578 	} else { /* pipes */
579 		d->first_tx_ring = d->last_tx_ring = 0;
580 		d->first_rx_ring = d->last_rx_ring = 0;
581 	}
582 
583 #ifdef DEBUG_NETMAP_USER
584     { /* debugging code */
585 	int i;
586 
587 	D("%s tx %d .. %d %d rx %d .. %d %d", ifname,
588 		d->first_tx_ring, d->last_tx_ring, d->req.nr_tx_rings,
589                 d->first_rx_ring, d->last_rx_ring, d->req.nr_rx_rings);
590 	for (i = 0; i <= d->req.nr_tx_rings; i++) {
591 		struct netmap_ring *r = NETMAP_TXRING(d->nifp, i);
592 		D("TX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
593 	}
594 	for (i = 0; i <= d->req.nr_rx_rings; i++) {
595 		struct netmap_ring *r = NETMAP_RXRING(d->nifp, i);
596 		D("RX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
597 	}
598     }
599 #endif /* debugging */
600 
601 	d->cur_tx_ring = d->first_tx_ring;
602 	d->cur_rx_ring = d->first_rx_ring;
603 	return d;
604 
605 fail:
606 	nm_close(d);
607 	if (errmsg[0])
608 		D("%s %s", errmsg, ifname);
609 	if (errno == 0)
610 		errno = EINVAL;
611 	return NULL;
612 }
613 
614 
615 static int
616 nm_close(struct nm_desc *d)
617 {
618 	/*
619 	 * ugly trick to avoid unused warnings
620 	 */
621 	static void *__xxzt[] __attribute__ ((unused))  =
622 		{ (void *)nm_open, (void *)nm_inject,
623 		  (void *)nm_dispatch, (void *)nm_nextpkt } ;
624 
625 	if (d == NULL || d->self != d)
626 		return EINVAL;
627 	if (d->done_mmap && d->mem)
628 		munmap(d->mem, d->memsize);
629 	if (d->fd != -1)
630 		close(d->fd);
631 	bzero(d, sizeof(*d));
632 	free(d);
633 	return 0;
634 }
635 
636 
637 /*
638  * Same prototype as pcap_inject(), only need to cast.
639  */
640 static int
641 nm_inject(struct nm_desc *d, const void *buf, size_t size)
642 {
643 	u_int c, n = d->last_tx_ring - d->first_tx_ring + 1;
644 
645 	for (c = 0; c < n ; c++) {
646 		/* compute current ring to use */
647 		struct netmap_ring *ring;
648 		uint32_t i, idx;
649 		uint32_t ri = d->cur_tx_ring + c;
650 
651 		if (ri > d->last_tx_ring)
652 			ri = d->first_tx_ring;
653 		ring = NETMAP_TXRING(d->nifp, ri);
654 		if (nm_ring_empty(ring)) {
655 			continue;
656 		}
657 		i = ring->cur;
658 		idx = ring->slot[i].buf_idx;
659 		ring->slot[i].len = size;
660 		nm_pkt_copy(buf, NETMAP_BUF(ring, idx), size);
661 		d->cur_tx_ring = ri;
662 		ring->head = ring->cur = nm_ring_next(ring, i);
663 		return size;
664 	}
665 	return 0; /* fail */
666 }
667 
668 
669 /*
670  * Same prototype as pcap_dispatch(), only need to cast.
671  */
672 static int
673 nm_dispatch(struct nm_desc *d, int cnt, nm_cb_t cb, u_char *arg)
674 {
675 	int n = d->last_rx_ring - d->first_rx_ring + 1;
676 	int c, got = 0, ri = d->cur_rx_ring;
677 
678 	if (cnt == 0)
679 		cnt = -1;
680 	/* cnt == -1 means infinite, but rings have a finite amount
681 	 * of buffers and the int is large enough that we never wrap,
682 	 * so we can omit checking for -1
683 	 */
684 	for (c=0; c < n && cnt != got; c++) {
685 		/* compute current ring to use */
686 		struct netmap_ring *ring;
687 
688 		ri = d->cur_rx_ring + c;
689 		if (ri > d->last_rx_ring)
690 			ri = d->first_rx_ring;
691 		ring = NETMAP_RXRING(d->nifp, ri);
692 		for ( ; !nm_ring_empty(ring) && cnt != got; got++) {
693 			u_int i = ring->cur;
694 			u_int idx = ring->slot[i].buf_idx;
695 			u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
696 
697 			// __builtin_prefetch(buf);
698 			d->hdr.len = d->hdr.caplen = ring->slot[i].len;
699 			d->hdr.ts = ring->ts;
700 			cb(arg, &d->hdr, buf);
701 			ring->head = ring->cur = nm_ring_next(ring, i);
702 		}
703 	}
704 	d->cur_rx_ring = ri;
705 	return got;
706 }
707 
708 static u_char *
709 nm_nextpkt(struct nm_desc *d, struct nm_pkthdr *hdr)
710 {
711 	int ri = d->cur_rx_ring;
712 
713 	do {
714 		/* compute current ring to use */
715 		struct netmap_ring *ring = NETMAP_RXRING(d->nifp, ri);
716 		if (!nm_ring_empty(ring)) {
717 			u_int i = ring->cur;
718 			u_int idx = ring->slot[i].buf_idx;
719 			u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
720 
721 			// __builtin_prefetch(buf);
722 			hdr->ts = ring->ts;
723 			hdr->len = hdr->caplen = ring->slot[i].len;
724 			ring->cur = nm_ring_next(ring, i);
725 			/* we could postpone advancing head if we want
726 			 * to hold the buffer. This can be supported in
727 			 * the future.
728 			 */
729 			ring->head = ring->cur;
730 			d->cur_rx_ring = ri;
731 			return buf;
732 		}
733 		ri++;
734 		if (ri > d->last_rx_ring)
735 			ri = d->first_rx_ring;
736 	} while (ri != d->cur_rx_ring);
737 	return NULL; /* nothing found */
738 }
739 
740 #endif /* !HAVE_NETMAP_WITH_LIBS */
741 
742 #endif /* NETMAP_WITH_LIBS */
743 
744 #endif /* _NET_NETMAP_USER_H_ */
745