xref: /freebsd/sys/net/netmap_user.h (revision 2ff63af9b88c7413b7d71715b5532625752a248e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (C) 2011-2016 Universita` di Pisa
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  *   1. Redistributions of source code must retain the above copyright
12  *      notice, this list of conditions and the following disclaimer.
13  *   2. Redistributions in binary form must reproduce the above copyright
14  *      notice, this list of conditions and the following disclaimer in the
15  *      documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 /*
31  *
32  * Functions and macros to manipulate netmap structures and packets
33  * in userspace. See netmap(4) for more information.
34  *
35  * The address of the struct netmap_if, say nifp, is computed from the
36  * value returned from ioctl(.., NIOCREG, ...) and the mmap region:
37  *	ioctl(fd, NIOCREG, &req);
38  *	mem = mmap(0, ... );
39  *	nifp = NETMAP_IF(mem, req.nr_nifp);
40  *		(so simple, we could just do it manually)
41  *
42  * From there:
43  *	struct netmap_ring *NETMAP_TXRING(nifp, index)
44  *	struct netmap_ring *NETMAP_RXRING(nifp, index)
45  *		we can access ring->cur, ring->head, ring->tail, etc.
46  *
47  *	ring->slot[i] gives us the i-th slot (we can access
48  *		directly len, flags, buf_idx)
49  *
50  *	char *buf = NETMAP_BUF(ring, x) returns a pointer to
51  *		the buffer numbered x
52  *
53  * All ring indexes (head, cur, tail) should always move forward.
54  * To compute the next index in a circular ring you can use
55  *	i = nm_ring_next(ring, i);
56  *
57  * To ease porting apps from pcap to netmap we supply a few functions
58  * that can be called to open, close, read and write on netmap in a way
59  * similar to libpcap. Note that the read/write function depend on
60  * an ioctl()/select()/poll() being issued to refill rings or push
61  * packets out.
62  *
63  * In order to use these, include #define NETMAP_WITH_LIBS
64  * in the source file that invokes these functions.
65  */
66 
67 #ifndef _NET_NETMAP_USER_H_
68 #define _NET_NETMAP_USER_H_
69 
70 #define NETMAP_DEVICE_NAME "/dev/netmap"
71 
72 #ifdef __CYGWIN__
73 /*
74  * we can compile userspace apps with either cygwin or msvc,
75  * and we use _WIN32 to identify windows specific code
76  */
77 #ifndef _WIN32
78 #define _WIN32
79 #endif	/* _WIN32 */
80 
81 #endif	/* __CYGWIN__ */
82 
83 #ifdef _WIN32
84 #undef NETMAP_DEVICE_NAME
85 #define NETMAP_DEVICE_NAME "/proc/sys/DosDevices/Global/netmap"
86 #include <windows.h>
87 #include <WinDef.h>
88 #include <sys/cygwin.h>
89 #endif /* _WIN32 */
90 
91 #include <stdint.h>
92 #include <sys/socket.h>		/* apple needs sockaddr */
93 #include <net/if.h>		/* IFNAMSIZ */
94 #include <ctype.h>
95 #include <string.h>	/* memset */
96 #include <sys/time.h>   /* gettimeofday */
97 
98 #ifndef likely
99 #define likely(x)	__builtin_expect(!!(x), 1)
100 #define unlikely(x)	__builtin_expect(!!(x), 0)
101 #endif /* likely and unlikely */
102 
103 #include <net/netmap.h>
104 
105 /* helper macro */
106 #define _NETMAP_OFFSET(type, ptr, offset) \
107 	((type)(void *)((char *)(ptr) + (offset)))
108 
109 #define NETMAP_IF(_base, _ofs)	_NETMAP_OFFSET(struct netmap_if *, _base, _ofs)
110 
111 #define NETMAP_TXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
112 	nifp, (nifp)->ring_ofs[index] )
113 
114 #define NETMAP_RXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *,	\
115 	nifp, (nifp)->ring_ofs[index + (nifp)->ni_tx_rings + 		\
116 		(nifp)->ni_host_tx_rings] )
117 
118 #define NETMAP_BUF(ring, index)				\
119 	((char *)(ring) + (ring)->buf_ofs + ((size_t)(index)*(ring)->nr_buf_size))
120 
121 #define NETMAP_BUF_IDX(ring, buf)			\
122 	( ((char *)(buf) - ((char *)(ring) + (ring)->buf_ofs) ) / \
123 		(ring)->nr_buf_size )
124 
125 /* read the offset field in a ring's slot */
126 #define NETMAP_ROFFSET(ring, slot)			\
127 	((slot)->ptr & (ring)->offset_mask)
128 
129 /* update the offset field in a ring's slot */
130 #define NETMAP_WOFFSET(ring, slot, offset)		\
131 	do { (slot)->ptr = ((slot)->ptr & ~(ring)->offset_mask) | \
132 		((offset) & (ring)->offset_mask); } while (0)
133 
134 /* obtain the start of the buffer pointed to by  a ring's slot, taking the
135  * offset field into account
136  */
137 #define NETMAP_BUF_OFFSET(ring, slot)			\
138 	(NETMAP_BUF(ring, (slot)->buf_idx) + NETMAP_ROFFSET(ring, slot))
139 
140 
141 static inline uint32_t
nm_ring_next(struct netmap_ring * r,uint32_t i)142 nm_ring_next(struct netmap_ring *r, uint32_t i)
143 {
144 	return ( unlikely(i + 1 == r->num_slots) ? 0 : i + 1);
145 }
146 
147 
148 /*
149  * Return 1 if we have pending transmissions in the tx ring.
150  * When everything is complete ring->head = ring->tail + 1 (modulo ring size)
151  */
152 static inline int
nm_tx_pending(struct netmap_ring * r)153 nm_tx_pending(struct netmap_ring *r)
154 {
155 	return nm_ring_next(r, r->tail) != r->head;
156 }
157 
158 /* Compute the number of slots available in the netmap ring. We use
159  * ring->head as explained in the comment above nm_ring_empty(). */
160 static inline uint32_t
nm_ring_space(struct netmap_ring * ring)161 nm_ring_space(struct netmap_ring *ring)
162 {
163         int ret = ring->tail - ring->head;
164         if (ret < 0)
165                 ret += ring->num_slots;
166         return ret;
167 }
168 
169 #ifndef ND /* debug macros */
170 /* debug support */
171 #define ND(_fmt, ...) do {} while(0)
172 #define D(_fmt, ...)						\
173 	do {							\
174 		struct timeval _t0;				\
175 		gettimeofday(&_t0, NULL);			\
176 		fprintf(stderr, "%03d.%06d %s [%d] " _fmt "\n",	\
177 		    (int)(_t0.tv_sec % 1000), (int)_t0.tv_usec,	\
178 		    __FUNCTION__, __LINE__, ##__VA_ARGS__);	\
179         } while (0)
180 
181 /* Rate limited version of "D", lps indicates how many per second */
182 #define RD(lps, format, ...)                                    \
183     do {                                                        \
184         static int __t0, __cnt;                                 \
185         struct timeval __xxts;                                  \
186         gettimeofday(&__xxts, NULL);                            \
187         if (__t0 != __xxts.tv_sec) {                            \
188             __t0 = __xxts.tv_sec;                               \
189             __cnt = 0;                                          \
190         }                                                       \
191         if (__cnt++ < lps) {                                    \
192             D(format, ##__VA_ARGS__);                           \
193         }                                                       \
194     } while (0)
195 #endif
196 
197 /*
198  * this is a slightly optimized copy routine which rounds
199  * to multiple of 64 bytes and is often faster than dealing
200  * with other odd sizes. We assume there is enough room
201  * in the source and destination buffers.
202  */
203 static inline void
nm_pkt_copy(const void * _src,void * _dst,int l)204 nm_pkt_copy(const void *_src, void *_dst, int l)
205 {
206 	const uint64_t *src = (const uint64_t *)_src;
207 	uint64_t *dst = (uint64_t *)_dst;
208 
209 	if (unlikely(l >= 1024 || l % 64)) {
210 		memcpy(dst, src, l);
211 		return;
212 	}
213 	for (; likely(l > 0); l-=64) {
214 		*dst++ = *src++;
215 		*dst++ = *src++;
216 		*dst++ = *src++;
217 		*dst++ = *src++;
218 		*dst++ = *src++;
219 		*dst++ = *src++;
220 		*dst++ = *src++;
221 		*dst++ = *src++;
222 	}
223 }
224 
225 #ifdef NETMAP_WITH_LIBS
226 /*
227  * Support for simple I/O libraries.
228  * Include other system headers required for compiling this.
229  */
230 
231 #ifndef HAVE_NETMAP_WITH_LIBS
232 #define HAVE_NETMAP_WITH_LIBS
233 
234 #include <stdio.h>
235 #include <sys/time.h>
236 #include <sys/mman.h>
237 #include <sys/ioctl.h>
238 #include <sys/errno.h>	/* EINVAL */
239 #include <fcntl.h>	/* O_RDWR */
240 #include <unistd.h>	/* close() */
241 #include <signal.h>
242 #include <stdlib.h>
243 
244 struct nm_pkthdr {	/* first part is the same as pcap_pkthdr */
245 	struct timeval	ts;
246 	uint32_t	caplen;
247 	uint32_t	len;
248 
249 	uint64_t flags;	/* NM_MORE_PKTS etc */
250 #define NM_MORE_PKTS	1
251 	struct nm_desc *d;
252 	struct netmap_slot *slot;
253 	uint8_t *buf;
254 };
255 
256 struct nm_stat {	/* same as pcap_stat	*/
257 	u_int	ps_recv;
258 	u_int	ps_drop;
259 	u_int	ps_ifdrop;
260 #ifdef WIN32 /* XXX or _WIN32 ? */
261 	u_int	bs_capt;
262 #endif /* WIN32 */
263 };
264 
265 #define NM_ERRBUF_SIZE	512
266 
267 struct nm_desc {
268 	struct nm_desc *self; /* point to self if netmap. */
269 	int fd;
270 	void *mem;
271 	size_t memsize;
272 	int done_mmap;	/* set if mem is the result of mmap */
273 	struct netmap_if * const nifp;
274 	uint16_t first_tx_ring, last_tx_ring, cur_tx_ring;
275 	uint16_t first_rx_ring, last_rx_ring, cur_rx_ring;
276 	struct nmreq req;	/* also contains the nr_name = ifname */
277 	struct nm_pkthdr hdr;
278 
279 	/*
280 	 * The memory contains netmap_if, rings and then buffers.
281 	 * Given a pointer (e.g. to nm_inject) we can compare with
282 	 * mem/buf_start/buf_end to tell if it is a buffer or
283 	 * some other descriptor in our region.
284 	 * We also store a pointer to some ring as it helps in the
285 	 * translation from buffer indexes to addresses.
286 	 */
287 	struct netmap_ring * const some_ring;
288 	void * const buf_start;
289 	void * const buf_end;
290 	/* parameters from pcap_open_live */
291 	int snaplen;
292 	int promisc;
293 	int to_ms;
294 	char *errbuf;
295 
296 	/* save flags so we can restore them on close */
297 	uint32_t if_flags;
298         uint32_t if_reqcap;
299         uint32_t if_curcap;
300 
301 	struct nm_stat st;
302 	char msg[NM_ERRBUF_SIZE];
303 };
304 
305 /*
306  * when the descriptor is open correctly, d->self == d
307  * Eventually we should also use some magic number.
308  */
309 #define P2NMD(p)		((const struct nm_desc *)(p))
310 #define IS_NETMAP_DESC(d)	((d) && P2NMD(d)->self == P2NMD(d))
311 #define NETMAP_FD(d)		(P2NMD(d)->fd)
312 
313 /*
314  * The callback, invoked on each received packet. Same as libpcap
315  */
316 typedef void (*nm_cb_t)(u_char *, const struct nm_pkthdr *, const u_char *d);
317 
318 /*
319  *--- the pcap-like API ---
320  *
321  * nm_open() opens a file descriptor, binds to a port and maps memory.
322  *
323  * ifname	(netmap:foo or vale:foo) is the port name
324  *		a suffix can indicate the following:
325  *		^		bind the host (sw) ring pair
326  *		*		bind host and NIC ring pairs
327  *		-NN		bind individual NIC ring pair
328  *		{NN		bind master side of pipe NN
329  *		}NN		bind slave side of pipe NN
330  *		a suffix starting with / and the following flags,
331  *		in any order:
332  *		x		exclusive access
333  *		z		zero copy monitor (both tx and rx)
334  *		t		monitor tx side (copy monitor)
335  *		r		monitor rx side (copy monitor)
336  *		R		bind only RX ring(s)
337  *		T		bind only TX ring(s)
338  *
339  * req		provides the initial values of nmreq before parsing ifname.
340  *		Remember that the ifname parsing will override the ring
341  *		number in nm_ringid, and part of nm_flags;
342  * flags	special functions, normally 0
343  *		indicates which fields of *arg are significant
344  * arg		special functions, normally NULL
345  *		if passed a netmap_desc with mem != NULL,
346  *		use that memory instead of mmap.
347  */
348 
349 static struct nm_desc *nm_open(const char *ifname, const struct nmreq *req,
350 	uint64_t flags, const struct nm_desc *arg);
351 
352 /*
353  * nm_open can import some fields from the parent descriptor.
354  * These flags control which ones.
355  * Also in flags you can specify NETMAP_NO_TX_POLL and NETMAP_DO_RX_POLL,
356  * which set the initial value for these flags.
357  * Note that the 16 low bits of the flags are reserved for data
358  * that may go into the nmreq.
359  */
360 enum {
361 	NM_OPEN_NO_MMAP =	0x040000, /* reuse mmap from parent */
362 	NM_OPEN_IFNAME =	0x080000, /* nr_name, nr_ringid, nr_flags */
363 	NM_OPEN_ARG1 =		0x100000,
364 	NM_OPEN_ARG2 =		0x200000,
365 	NM_OPEN_ARG3 =		0x400000,
366 	NM_OPEN_RING_CFG =	0x800000, /* tx|rx rings|slots */
367 };
368 
369 
370 /*
371  * nm_close()	closes and restores the port to its previous state
372  */
373 
374 static int nm_close(struct nm_desc *);
375 
376 /*
377  * nm_mmap()    do mmap or inherit from parent if the nr_arg2
378  *              (memory block) matches.
379  */
380 
381 static int nm_mmap(struct nm_desc *, const struct nm_desc *);
382 
383 /*
384  * nm_inject() is the same as pcap_inject()
385  * nm_dispatch() is the same as pcap_dispatch()
386  * nm_nextpkt() is the same as pcap_next()
387  */
388 
389 static int nm_inject(struct nm_desc *, const void *, size_t);
390 static int nm_dispatch(struct nm_desc *, int, nm_cb_t, u_char *);
391 static u_char *nm_nextpkt(struct nm_desc *, struct nm_pkthdr *);
392 
393 #ifdef _WIN32
394 
395 intptr_t _get_osfhandle(int); /* defined in io.h in windows */
396 
397 /*
398  * In windows we do not have yet native poll support, so we keep track
399  * of file descriptors associated to netmap ports to emulate poll on
400  * them and fall back on regular poll on other file descriptors.
401  */
402 struct win_netmap_fd_list {
403 	struct win_netmap_fd_list *next;
404 	int win_netmap_fd;
405 	HANDLE win_netmap_handle;
406 };
407 
408 /*
409  * list head containing all the netmap opened fd and their
410  * windows HANDLE counterparts
411  */
412 static struct win_netmap_fd_list *win_netmap_fd_list_head;
413 
414 static void
win_insert_fd_record(int fd)415 win_insert_fd_record(int fd)
416 {
417 	struct win_netmap_fd_list *curr;
418 
419 	for (curr = win_netmap_fd_list_head; curr; curr = curr->next) {
420 		if (fd == curr->win_netmap_fd) {
421 			return;
422 		}
423 	}
424 	curr = calloc(1, sizeof(*curr));
425 	curr->next = win_netmap_fd_list_head;
426 	curr->win_netmap_fd = fd;
427 	curr->win_netmap_handle = IntToPtr(_get_osfhandle(fd));
428 	win_netmap_fd_list_head = curr;
429 }
430 
431 void
win_remove_fd_record(int fd)432 win_remove_fd_record(int fd)
433 {
434 	struct win_netmap_fd_list *curr = win_netmap_fd_list_head;
435 	struct win_netmap_fd_list *prev = NULL;
436 	for (; curr ; prev = curr, curr = curr->next) {
437 		if (fd != curr->win_netmap_fd)
438 			continue;
439 		/* found the entry */
440 		if (prev == NULL) { /* we are freeing the first entry */
441 			win_netmap_fd_list_head = curr->next;
442 		} else {
443 			prev->next = curr->next;
444 		}
445 		free(curr);
446 		break;
447 	}
448 }
449 
450 
451 HANDLE
win_get_netmap_handle(int fd)452 win_get_netmap_handle(int fd)
453 {
454 	struct win_netmap_fd_list *curr;
455 
456 	for (curr = win_netmap_fd_list_head; curr; curr = curr->next) {
457 		if (fd == curr->win_netmap_fd) {
458 			return curr->win_netmap_handle;
459 		}
460 	}
461 	return NULL;
462 }
463 
464 /*
465  * we need to wrap ioctl and mmap, at least for the netmap file descriptors
466  */
467 
468 /*
469  * use this function only from netmap_user.h internal functions
470  * same as ioctl, returns 0 on success and -1 on error
471  */
472 static int
win_nm_ioctl_internal(HANDLE h,int32_t ctlCode,void * arg)473 win_nm_ioctl_internal(HANDLE h, int32_t ctlCode, void *arg)
474 {
475 	DWORD bReturn = 0, szIn, szOut;
476 	BOOL ioctlReturnStatus;
477 	void *inParam = arg, *outParam = arg;
478 
479 	switch (ctlCode) {
480 	case NETMAP_POLL:
481 		szIn = sizeof(POLL_REQUEST_DATA);
482 		szOut = sizeof(POLL_REQUEST_DATA);
483 		break;
484 	case NETMAP_MMAP:
485 		szIn = 0;
486 		szOut = sizeof(void*);
487 		inParam = NULL; /* nothing on input */
488 		break;
489 	case NIOCTXSYNC:
490 	case NIOCRXSYNC:
491 		szIn = 0;
492 		szOut = 0;
493 		break;
494 	case NIOCREGIF:
495 		szIn = sizeof(struct nmreq);
496 		szOut = sizeof(struct nmreq);
497 		break;
498 	case NIOCCONFIG:
499 		D("unsupported NIOCCONFIG!");
500 		return -1;
501 
502 	default: /* a regular ioctl */
503 		D("invalid ioctl %x on netmap fd", ctlCode);
504 		return -1;
505 	}
506 
507 	ioctlReturnStatus = DeviceIoControl(h,
508 				ctlCode, inParam, szIn,
509 				outParam, szOut,
510 				&bReturn, NULL);
511 	// XXX note windows returns 0 on error or async call, 1 on success
512 	// we could call GetLastError() to figure out what happened
513 	return ioctlReturnStatus ? 0 : -1;
514 }
515 
516 /*
517  * this function is what must be called from user-space programs
518  * same as ioctl, returns 0 on success and -1 on error
519  */
520 static int
win_nm_ioctl(int fd,int32_t ctlCode,void * arg)521 win_nm_ioctl(int fd, int32_t ctlCode, void *arg)
522 {
523 	HANDLE h = win_get_netmap_handle(fd);
524 
525 	if (h == NULL) {
526 		return ioctl(fd, ctlCode, arg);
527 	} else {
528 		return win_nm_ioctl_internal(h, ctlCode, arg);
529 	}
530 }
531 
532 #define ioctl win_nm_ioctl /* from now on, within this file ... */
533 
534 /*
535  * We cannot use the native mmap on windows
536  * The only parameter used is "fd", the other ones are just declared to
537  * make this signature comparable to the FreeBSD/Linux one
538  */
539 static void *
win32_mmap_emulated(void * addr,size_t length,int prot,int flags,int fd,int32_t offset)540 win32_mmap_emulated(void *addr, size_t length, int prot, int flags, int fd, int32_t offset)
541 {
542 	HANDLE h = win_get_netmap_handle(fd);
543 
544 	if (h == NULL) {
545 		return mmap(addr, length, prot, flags, fd, offset);
546 	} else {
547 		MEMORY_ENTRY ret;
548 
549 		return win_nm_ioctl_internal(h, NETMAP_MMAP, &ret) ?
550 			NULL : ret.pUsermodeVirtualAddress;
551 	}
552 }
553 
554 #define mmap win32_mmap_emulated
555 
556 #include <sys/poll.h> /* XXX needed to use the structure pollfd */
557 
558 static int
win_nm_poll(struct pollfd * fds,int nfds,int timeout)559 win_nm_poll(struct pollfd *fds, int nfds, int timeout)
560 {
561 	HANDLE h;
562 
563 	if (nfds != 1 || fds == NULL || (h = win_get_netmap_handle(fds->fd)) == NULL) {;
564 		return poll(fds, nfds, timeout);
565 	} else {
566 		POLL_REQUEST_DATA prd;
567 
568 		prd.timeout = timeout;
569 		prd.events = fds->events;
570 
571 		win_nm_ioctl_internal(h, NETMAP_POLL, &prd);
572 		if ((prd.revents == POLLERR) || (prd.revents == STATUS_TIMEOUT)) {
573 			return -1;
574 		}
575 		return 1;
576 	}
577 }
578 
579 #define poll win_nm_poll
580 
581 static int
win_nm_open(char * pathname,int flags)582 win_nm_open(char* pathname, int flags)
583 {
584 
585 	if (strcmp(pathname, NETMAP_DEVICE_NAME) == 0) {
586 		int fd = open(NETMAP_DEVICE_NAME, O_RDWR);
587 		if (fd < 0) {
588 			return -1;
589 		}
590 
591 		win_insert_fd_record(fd);
592 		return fd;
593 	} else {
594 		return open(pathname, flags);
595 	}
596 }
597 
598 #define open win_nm_open
599 
600 static int
win_nm_close(int fd)601 win_nm_close(int fd)
602 {
603 	if (fd != -1) {
604 		close(fd);
605 		if (win_get_netmap_handle(fd) != NULL) {
606 			win_remove_fd_record(fd);
607 		}
608 	}
609 	return 0;
610 }
611 
612 #define close win_nm_close
613 
614 #endif /* _WIN32 */
615 
616 static int
nm_is_identifier(const char * s,const char * e)617 nm_is_identifier(const char *s, const char *e)
618 {
619 	for (; s != e; s++) {
620 		if (!isalnum(*s) && *s != '_') {
621 			return 0;
622 		}
623 	}
624 
625 	return 1;
626 }
627 
628 #define MAXERRMSG 80
629 static int
nm_parse(const char * ifname,struct nm_desc * d,char * err)630 nm_parse(const char *ifname, struct nm_desc *d, char *err)
631 {
632 	int is_vale;
633 	const char *port = NULL;
634 	const char *vpname = NULL;
635 	u_int namelen;
636 	uint32_t nr_ringid = 0, nr_flags;
637 	char errmsg[MAXERRMSG] = "", *tmp;
638 	long num;
639 	uint16_t nr_arg2 = 0;
640 	enum { P_START, P_RNGSFXOK, P_GETNUM, P_FLAGS, P_FLAGSOK, P_MEMID } p_state;
641 
642 	errno = 0;
643 
644 	is_vale = (ifname[0] == 'v');
645 	if (is_vale) {
646 		port = index(ifname, ':');
647 		if (port == NULL) {
648 			snprintf(errmsg, MAXERRMSG,
649 				 "missing ':' in vale name");
650 			goto fail;
651 		}
652 
653 		if (!nm_is_identifier(ifname + 4, port)) {
654 			snprintf(errmsg, MAXERRMSG, "invalid bridge name");
655 			goto fail;
656 		}
657 
658 		vpname = ++port;
659 	} else {
660 		ifname += 7;
661 		port = ifname;
662 	}
663 
664 	/* scan for a separator */
665 	for (; *port && !index("-*^{}/@", *port); port++)
666 		;
667 
668 	if (is_vale && !nm_is_identifier(vpname, port)) {
669 		snprintf(errmsg, MAXERRMSG, "invalid bridge port name");
670 		goto fail;
671 	}
672 
673 	namelen = port - ifname;
674 	if (namelen >= sizeof(d->req.nr_name)) {
675 		snprintf(errmsg, MAXERRMSG, "name too long");
676 		goto fail;
677 	}
678 	memcpy(d->req.nr_name, ifname, namelen);
679 	d->req.nr_name[namelen] = '\0';
680 
681 	p_state = P_START;
682 	nr_flags = NR_REG_ALL_NIC; /* default for no suffix */
683 	while (*port) {
684 		switch (p_state) {
685 		case P_START:
686 			switch (*port) {
687 			case '^': /* only SW ring */
688 				nr_flags = NR_REG_SW;
689 				p_state = P_RNGSFXOK;
690 				break;
691 			case '*': /* NIC and SW */
692 				nr_flags = NR_REG_NIC_SW;
693 				p_state = P_RNGSFXOK;
694 				break;
695 			case '-': /* one NIC ring pair */
696 				nr_flags = NR_REG_ONE_NIC;
697 				p_state = P_GETNUM;
698 				break;
699 			case '{': /* pipe (master endpoint) */
700 				nr_flags = NR_REG_PIPE_MASTER;
701 				p_state = P_GETNUM;
702 				break;
703 			case '}': /* pipe (slave endpoint) */
704 				nr_flags = NR_REG_PIPE_SLAVE;
705 				p_state = P_GETNUM;
706 				break;
707 			case '/': /* start of flags */
708 				p_state = P_FLAGS;
709 				break;
710 			case '@': /* start of memid */
711 				p_state = P_MEMID;
712 				break;
713 			default:
714 				snprintf(errmsg, MAXERRMSG, "unknown modifier: '%c'", *port);
715 				goto fail;
716 			}
717 			port++;
718 			break;
719 		case P_RNGSFXOK:
720 			switch (*port) {
721 			case '/':
722 				p_state = P_FLAGS;
723 				break;
724 			case '@':
725 				p_state = P_MEMID;
726 				break;
727 			default:
728 				snprintf(errmsg, MAXERRMSG, "unexpected character: '%c'", *port);
729 				goto fail;
730 			}
731 			port++;
732 			break;
733 		case P_GETNUM:
734 			num = strtol(port, &tmp, 10);
735 			if (num < 0 || num >= NETMAP_RING_MASK) {
736 				snprintf(errmsg, MAXERRMSG, "'%ld' out of range [0, %d)",
737 						num, NETMAP_RING_MASK);
738 				goto fail;
739 			}
740 			port = tmp;
741 			nr_ringid = num & NETMAP_RING_MASK;
742 			p_state = P_RNGSFXOK;
743 			break;
744 		case P_FLAGS:
745 		case P_FLAGSOK:
746 			if (*port == '@') {
747 				port++;
748 				p_state = P_MEMID;
749 				break;
750 			}
751 			switch (*port) {
752 			case 'x':
753 				nr_flags |= NR_EXCLUSIVE;
754 				break;
755 			case 'z':
756 				nr_flags |= NR_ZCOPY_MON;
757 				break;
758 			case 't':
759 				nr_flags |= NR_MONITOR_TX;
760 				break;
761 			case 'r':
762 				nr_flags |= NR_MONITOR_RX;
763 				break;
764 			case 'R':
765 				nr_flags |= NR_RX_RINGS_ONLY;
766 				break;
767 			case 'T':
768 				nr_flags |= NR_TX_RINGS_ONLY;
769 				break;
770 			default:
771 				snprintf(errmsg, MAXERRMSG, "unrecognized flag: '%c'", *port);
772 				goto fail;
773 			}
774 			port++;
775 			p_state = P_FLAGSOK;
776 			break;
777 		case P_MEMID:
778 			if (nr_arg2 != 0) {
779 				snprintf(errmsg, MAXERRMSG, "double setting of memid");
780 				goto fail;
781 			}
782 			num = strtol(port, &tmp, 10);
783 			if (num <= 0) {
784 				snprintf(errmsg, MAXERRMSG, "invalid memid %ld, must be >0", num);
785 				goto fail;
786 			}
787 			port = tmp;
788 			nr_arg2 = num;
789 			p_state = P_RNGSFXOK;
790 			break;
791 		}
792 	}
793 	if (p_state != P_START && p_state != P_RNGSFXOK && p_state != P_FLAGSOK) {
794 		snprintf(errmsg, MAXERRMSG, "unexpected end of port name");
795 		goto fail;
796 	}
797 	ND("flags: %s %s %s %s",
798 			(nr_flags & NR_EXCLUSIVE) ? "EXCLUSIVE" : "",
799 			(nr_flags & NR_ZCOPY_MON) ? "ZCOPY_MON" : "",
800 			(nr_flags & NR_MONITOR_TX) ? "MONITOR_TX" : "",
801 			(nr_flags & NR_MONITOR_RX) ? "MONITOR_RX" : "");
802 
803 	d->req.nr_flags |= nr_flags;
804 	d->req.nr_ringid |= nr_ringid;
805 	d->req.nr_arg2 = nr_arg2;
806 
807 	d->self = d;
808 
809 	return 0;
810 fail:
811 	if (!errno)
812 		errno = EINVAL;
813 	if (err)
814 		strncpy(err, errmsg, MAXERRMSG);
815 	return -1;
816 }
817 
818 /*
819  * Try to open, return descriptor if successful, NULL otherwise.
820  * An invalid netmap name will return errno = 0;
821  * You can pass a pointer to a pre-filled nm_desc to add special
822  * parameters. Flags is used as follows
823  * NM_OPEN_NO_MMAP	use the memory from arg, only XXX avoid mmap
824  *			if the nr_arg2 (memory block) matches.
825  * NM_OPEN_ARG1		use req.nr_arg1 from arg
826  * NM_OPEN_ARG2		use req.nr_arg2 from arg
827  * NM_OPEN_RING_CFG	user ring config from arg
828  */
829 static struct nm_desc *
nm_open(const char * ifname,const struct nmreq * req,uint64_t new_flags,const struct nm_desc * arg)830 nm_open(const char *ifname, const struct nmreq *req,
831 	uint64_t new_flags, const struct nm_desc *arg)
832 {
833 	struct nm_desc *d = NULL;
834 	const struct nm_desc *parent = arg;
835 	char errmsg[MAXERRMSG] = "";
836 	uint32_t nr_reg;
837 
838 	if (strncmp(ifname, "netmap:", 7) &&
839 			strncmp(ifname, NM_BDG_NAME, strlen(NM_BDG_NAME))) {
840 		errno = 0; /* name not recognised, not an error */
841 		return NULL;
842 	}
843 
844 	d = (struct nm_desc *)calloc(1, sizeof(*d));
845 	if (d == NULL) {
846 		snprintf(errmsg, MAXERRMSG, "nm_desc alloc failure");
847 		errno = ENOMEM;
848 		return NULL;
849 	}
850 	d->self = d;	/* set this early so nm_close() works */
851 	d->fd = open(NETMAP_DEVICE_NAME, O_RDWR);
852 	if (d->fd < 0) {
853 		snprintf(errmsg, MAXERRMSG, "cannot open /dev/netmap: %s", strerror(errno));
854 		goto fail;
855 	}
856 
857 	if (req)
858 		d->req = *req;
859 
860 	if (!(new_flags & NM_OPEN_IFNAME)) {
861 		if (nm_parse(ifname, d, errmsg) < 0)
862 			goto fail;
863 	}
864 
865 	d->req.nr_version = NETMAP_API;
866 	d->req.nr_ringid &= NETMAP_RING_MASK;
867 
868 	/* optionally import info from parent */
869 	if (IS_NETMAP_DESC(parent) && new_flags) {
870 		if (new_flags & NM_OPEN_ARG1)
871 			D("overriding ARG1 %d", parent->req.nr_arg1);
872 		d->req.nr_arg1 = new_flags & NM_OPEN_ARG1 ?
873 			parent->req.nr_arg1 : 4;
874 		if (new_flags & NM_OPEN_ARG2) {
875 			D("overriding ARG2 %d", parent->req.nr_arg2);
876 			d->req.nr_arg2 =  parent->req.nr_arg2;
877 		}
878 		if (new_flags & NM_OPEN_ARG3)
879 			D("overriding ARG3 %d", parent->req.nr_arg3);
880 		d->req.nr_arg3 = new_flags & NM_OPEN_ARG3 ?
881 			parent->req.nr_arg3 : 0;
882 		if (new_flags & NM_OPEN_RING_CFG) {
883 			D("overriding RING_CFG");
884 			d->req.nr_tx_slots = parent->req.nr_tx_slots;
885 			d->req.nr_rx_slots = parent->req.nr_rx_slots;
886 			d->req.nr_tx_rings = parent->req.nr_tx_rings;
887 			d->req.nr_rx_rings = parent->req.nr_rx_rings;
888 		}
889 		if (new_flags & NM_OPEN_IFNAME) {
890 			D("overriding ifname %s ringid 0x%x flags 0x%x",
891 				parent->req.nr_name, parent->req.nr_ringid,
892 				parent->req.nr_flags);
893 			memcpy(d->req.nr_name, parent->req.nr_name,
894 				sizeof(d->req.nr_name));
895 			d->req.nr_ringid = parent->req.nr_ringid;
896 			d->req.nr_flags = parent->req.nr_flags;
897 		}
898 	}
899 	/* add the *XPOLL flags */
900 	d->req.nr_ringid |= new_flags & (NETMAP_NO_TX_POLL | NETMAP_DO_RX_POLL);
901 
902 	if (ioctl(d->fd, NIOCREGIF, &d->req)) {
903 		snprintf(errmsg, MAXERRMSG, "NIOCREGIF failed: %s", strerror(errno));
904 		goto fail;
905 	}
906 
907 	nr_reg = d->req.nr_flags & NR_REG_MASK;
908 
909 	if (nr_reg == NR_REG_SW) { /* host stack */
910 		d->first_tx_ring = d->last_tx_ring = d->req.nr_tx_rings;
911 		d->first_rx_ring = d->last_rx_ring = d->req.nr_rx_rings;
912 	} else if (nr_reg ==  NR_REG_ALL_NIC) { /* only nic */
913 		d->first_tx_ring = 0;
914 		d->first_rx_ring = 0;
915 		d->last_tx_ring = d->req.nr_tx_rings - 1;
916 		d->last_rx_ring = d->req.nr_rx_rings - 1;
917 	} else if (nr_reg ==  NR_REG_NIC_SW) {
918 		d->first_tx_ring = 0;
919 		d->first_rx_ring = 0;
920 		d->last_tx_ring = d->req.nr_tx_rings;
921 		d->last_rx_ring = d->req.nr_rx_rings;
922 	} else if (nr_reg == NR_REG_ONE_NIC) {
923 		/* XXX check validity */
924 		d->first_tx_ring = d->last_tx_ring =
925 		d->first_rx_ring = d->last_rx_ring = d->req.nr_ringid & NETMAP_RING_MASK;
926 	} else { /* pipes */
927 		d->first_tx_ring = d->last_tx_ring = 0;
928 		d->first_rx_ring = d->last_rx_ring = 0;
929 	}
930 
931         /* if parent is defined, do nm_mmap() even if NM_OPEN_NO_MMAP is set */
932 	if ((!(new_flags & NM_OPEN_NO_MMAP) || parent) && nm_mmap(d, parent)) {
933 	        snprintf(errmsg, MAXERRMSG, "mmap failed: %s", strerror(errno));
934 		goto fail;
935 	}
936 
937 
938 #ifdef DEBUG_NETMAP_USER
939     { /* debugging code */
940 	int i;
941 
942 	D("%s tx %d .. %d %d rx %d .. %d %d", ifname,
943 		d->first_tx_ring, d->last_tx_ring, d->req.nr_tx_rings,
944                 d->first_rx_ring, d->last_rx_ring, d->req.nr_rx_rings);
945 	for (i = 0; i <= d->req.nr_tx_rings; i++) {
946 		struct netmap_ring *r = NETMAP_TXRING(d->nifp, i);
947 		D("TX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
948 	}
949 	for (i = 0; i <= d->req.nr_rx_rings; i++) {
950 		struct netmap_ring *r = NETMAP_RXRING(d->nifp, i);
951 		D("RX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
952 	}
953     }
954 #endif /* debugging */
955 
956 	d->cur_tx_ring = d->first_tx_ring;
957 	d->cur_rx_ring = d->first_rx_ring;
958 	return d;
959 
960 fail:
961 	nm_close(d);
962 	if (errmsg[0])
963 		D("%s %s", errmsg, ifname);
964 	if (errno == 0)
965 		errno = EINVAL;
966 	return NULL;
967 }
968 
969 
970 static int
nm_close(struct nm_desc * d)971 nm_close(struct nm_desc *d)
972 {
973 	/*
974 	 * ugly trick to avoid unused warnings
975 	 */
976 	static void *__xxzt[] __attribute__ ((unused))  =
977 		{ (void *)nm_open, (void *)nm_inject,
978 		  (void *)nm_dispatch, (void *)nm_nextpkt } ;
979 
980 	if (d == NULL || d->self != d)
981 		return EINVAL;
982 	if (d->done_mmap && d->mem)
983 		munmap(d->mem, d->memsize);
984 	if (d->fd != -1) {
985 		close(d->fd);
986 	}
987 
988 	bzero((char *)d, sizeof(*d));
989 	free(d);
990 	return 0;
991 }
992 
993 
994 static int
nm_mmap(struct nm_desc * d,const struct nm_desc * parent)995 nm_mmap(struct nm_desc *d, const struct nm_desc *parent)
996 {
997 	if (d->done_mmap)
998 		return 0;
999 
1000 	if (IS_NETMAP_DESC(parent) && parent->mem &&
1001 	    parent->req.nr_arg2 == d->req.nr_arg2) {
1002 		/* do not mmap, inherit from parent */
1003 		D("do not mmap, inherit from parent");
1004 		d->memsize = parent->memsize;
1005 		d->mem = parent->mem;
1006 	} else {
1007 		/* XXX TODO: check if memsize is too large (or there is overflow) */
1008 		d->memsize = d->req.nr_memsize;
1009 		d->mem = mmap(0, d->memsize, PROT_WRITE | PROT_READ, MAP_SHARED,
1010 				d->fd, 0);
1011 		if (d->mem == MAP_FAILED) {
1012 			goto fail;
1013 		}
1014 		d->done_mmap = 1;
1015 	}
1016 	{
1017 		struct netmap_if *nifp = NETMAP_IF(d->mem, d->req.nr_offset);
1018 		struct netmap_ring *r = NETMAP_RXRING(nifp, d->first_rx_ring);
1019 		if ((void *)r == (void *)nifp) {
1020 			/* the descriptor is open for TX only */
1021 			r = NETMAP_TXRING(nifp, d->first_tx_ring);
1022 		}
1023 
1024 		*(struct netmap_if **)(uintptr_t)&(d->nifp) = nifp;
1025 		*(struct netmap_ring **)(uintptr_t)&d->some_ring = r;
1026 		*(void **)(uintptr_t)&d->buf_start = NETMAP_BUF(r, 0);
1027 		*(void **)(uintptr_t)&d->buf_end =
1028 			(char *)d->mem + d->memsize;
1029 	}
1030 
1031 	return 0;
1032 
1033 fail:
1034 	return EINVAL;
1035 }
1036 
1037 /*
1038  * Same prototype as pcap_inject(), only need to cast.
1039  */
1040 static int
nm_inject(struct nm_desc * d,const void * buf,size_t size)1041 nm_inject(struct nm_desc *d, const void *buf, size_t size)
1042 {
1043 	u_int c, n = d->last_tx_ring - d->first_tx_ring + 1,
1044 		ri = d->cur_tx_ring;
1045 
1046 	for (c = 0; c < n ; c++, ri++) {
1047 		/* compute current ring to use */
1048 		struct netmap_ring *ring;
1049 		uint32_t i, j, idx;
1050 		size_t rem;
1051 
1052 		if (ri > d->last_tx_ring)
1053 			ri = d->first_tx_ring;
1054 		ring = NETMAP_TXRING(d->nifp, ri);
1055 		rem = size;
1056 		j = ring->cur;
1057 		while (rem > ring->nr_buf_size && j != ring->tail) {
1058 			rem -= ring->nr_buf_size;
1059 			j = nm_ring_next(ring, j);
1060 		}
1061 		if (j == ring->tail && rem > 0)
1062 			continue;
1063 		i = ring->cur;
1064 		while (i != j) {
1065 			idx = ring->slot[i].buf_idx;
1066 			ring->slot[i].len = ring->nr_buf_size;
1067 			ring->slot[i].flags = NS_MOREFRAG;
1068 			nm_pkt_copy(buf, NETMAP_BUF(ring, idx), ring->nr_buf_size);
1069 			i = nm_ring_next(ring, i);
1070 			buf = (const char *)buf + ring->nr_buf_size;
1071 		}
1072 		idx = ring->slot[i].buf_idx;
1073 		ring->slot[i].len = rem;
1074 		ring->slot[i].flags = 0;
1075 		nm_pkt_copy(buf, NETMAP_BUF(ring, idx), rem);
1076 		ring->head = ring->cur = nm_ring_next(ring, i);
1077 		d->cur_tx_ring = ri;
1078 		return size;
1079 	}
1080 	return 0; /* fail */
1081 }
1082 
1083 
1084 /*
1085  * Same prototype as pcap_dispatch(), only need to cast.
1086  */
1087 static int
nm_dispatch(struct nm_desc * d,int cnt,nm_cb_t cb,u_char * arg)1088 nm_dispatch(struct nm_desc *d, int cnt, nm_cb_t cb, u_char *arg)
1089 {
1090 	int n = d->last_rx_ring - d->first_rx_ring + 1;
1091 	int c, got = 0, ri = d->cur_rx_ring;
1092 	d->hdr.buf = NULL;
1093 	d->hdr.flags = NM_MORE_PKTS;
1094 	d->hdr.d = d;
1095 
1096 	if (cnt == 0)
1097 		cnt = -1;
1098 	/* cnt == -1 means infinite, but rings have a finite amount
1099 	 * of buffers and the int is large enough that we never wrap,
1100 	 * so we can omit checking for -1
1101 	 */
1102 	for (c=0; c < n && cnt != got; c++, ri++) {
1103 		/* compute current ring to use */
1104 		struct netmap_ring *ring;
1105 
1106 		if (ri > d->last_rx_ring)
1107 			ri = d->first_rx_ring;
1108 		ring = NETMAP_RXRING(d->nifp, ri);
1109 		for ( ; !nm_ring_empty(ring) && cnt != got; got++) {
1110 			u_int idx, i;
1111 			u_char *oldbuf;
1112 			struct netmap_slot *slot;
1113 			if (d->hdr.buf) { /* from previous round */
1114 				cb(arg, &d->hdr, d->hdr.buf);
1115 			}
1116 			i = ring->cur;
1117 			slot = &ring->slot[i];
1118 			idx = slot->buf_idx;
1119 			/* d->cur_rx_ring doesn't change inside this loop, but
1120 			 * set it here, so it reflects d->hdr.buf's ring */
1121 			d->cur_rx_ring = ri;
1122 			d->hdr.slot = slot;
1123 			oldbuf = d->hdr.buf = (u_char *)NETMAP_BUF(ring, idx);
1124 			// __builtin_prefetch(buf);
1125 			d->hdr.len = d->hdr.caplen = slot->len;
1126 			while (slot->flags & NS_MOREFRAG) {
1127 				u_char *nbuf;
1128 				u_int oldlen = slot->len;
1129 				i = nm_ring_next(ring, i);
1130 				slot = &ring->slot[i];
1131 				d->hdr.len += slot->len;
1132 				nbuf = (u_char *)NETMAP_BUF(ring, slot->buf_idx);
1133 				if (oldbuf != NULL && (uint32_t)(nbuf - oldbuf) == ring->nr_buf_size &&
1134 						oldlen == ring->nr_buf_size) {
1135 					d->hdr.caplen += slot->len;
1136 					oldbuf = nbuf;
1137 				} else {
1138 					oldbuf = NULL;
1139 				}
1140 			}
1141 			d->hdr.ts = ring->ts;
1142 			ring->head = ring->cur = nm_ring_next(ring, i);
1143 		}
1144 	}
1145 	if (d->hdr.buf) { /* from previous round */
1146 		d->hdr.flags = 0;
1147 		cb(arg, &d->hdr, d->hdr.buf);
1148 	}
1149 	return got;
1150 }
1151 
1152 static u_char *
nm_nextpkt(struct nm_desc * d,struct nm_pkthdr * hdr)1153 nm_nextpkt(struct nm_desc *d, struct nm_pkthdr *hdr)
1154 {
1155 	int ri = d->cur_rx_ring;
1156 
1157 	do {
1158 		/* compute current ring to use */
1159 		struct netmap_ring *ring = NETMAP_RXRING(d->nifp, ri);
1160 		if (!nm_ring_empty(ring)) {
1161 			u_int i = ring->cur;
1162 			u_int idx = ring->slot[i].buf_idx;
1163 			u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
1164 
1165 			// __builtin_prefetch(buf);
1166 			hdr->ts = ring->ts;
1167 			hdr->len = hdr->caplen = ring->slot[i].len;
1168 			ring->cur = nm_ring_next(ring, i);
1169 			/* we could postpone advancing head if we want
1170 			 * to hold the buffer. This can be supported in
1171 			 * the future.
1172 			 */
1173 			ring->head = ring->cur;
1174 			d->cur_rx_ring = ri;
1175 			return buf;
1176 		}
1177 		ri++;
1178 		if (ri > d->last_rx_ring)
1179 			ri = d->first_rx_ring;
1180 	} while (ri != d->cur_rx_ring);
1181 	return NULL; /* nothing found */
1182 }
1183 
1184 #endif /* !HAVE_NETMAP_WITH_LIBS */
1185 
1186 #endif /* NETMAP_WITH_LIBS */
1187 
1188 #endif /* _NET_NETMAP_USER_H_ */
1189