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 #define NETMAP_DEVICE_NAME "/dev/netmap"
69 #ifdef __CYGWIN__
70 /*
71 * we can compile userspace apps with either cygwin or msvc,
72 * and we use _WIN32 to identify windows specific code
73 */
74 #ifndef _WIN32
75 #define _WIN32
76 #endif /* _WIN32 */
77
78 #endif /* __CYGWIN__ */
79
80 #ifdef _WIN32
81 #undef NETMAP_DEVICE_NAME
82 #define NETMAP_DEVICE_NAME "/proc/sys/DosDevices/Global/netmap"
83 #include <windows.h>
84 #include <WinDef.h>
85 #include <sys/cygwin.h>
86 //#include <netioapi.h>
87 //#include <winsock.h>
88 //#define IFNAMSIZ 256
89 #endif
90
91 #include <stdint.h>
92 #include <sys/socket.h> /* apple needs sockaddr */
93 #include <net/if.h> /* IFNAMSIZ */
94
95 #ifndef likely
96 #define likely(x) __builtin_expect(!!(x), 1)
97 #define unlikely(x) __builtin_expect(!!(x), 0)
98 #endif /* likely and unlikely */
99
100 #include "netmap.h"
101
102 /* helper macro */
103 #define _NETMAP_OFFSET(type, ptr, offset) \
104 ((type)(void *)((char *)(ptr) + (offset)))
105
106 #define NETMAP_IF(_base, _ofs) _NETMAP_OFFSET(struct netmap_if *, _base, _ofs)
107
108 #define NETMAP_TXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
109 nifp, (nifp)->ring_ofs[index] )
110
111 #define NETMAP_RXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
112 nifp, (nifp)->ring_ofs[index + (nifp)->ni_tx_rings + 1] )
113
114 #define NETMAP_BUF(ring, index) \
115 ((char *)(ring) + (ring)->buf_ofs + ((index)*(ring)->nr_buf_size))
116
117 #define NETMAP_BUF_IDX(ring, buf) \
118 ( ((char *)(buf) - ((char *)(ring) + (ring)->buf_ofs) ) / \
119 (ring)->nr_buf_size )
120
121
122 static inline uint32_t
nm_ring_next(struct netmap_ring * r,uint32_t i)123 nm_ring_next(struct netmap_ring *r, uint32_t i)
124 {
125 return ( unlikely(i + 1 == r->num_slots) ? 0 : i + 1);
126 }
127
128
129 /*
130 * Return 1 if we have pending transmissions in the tx ring.
131 * When everything is complete ring->head = ring->tail + 1 (modulo ring size)
132 */
133 static inline int
nm_tx_pending(struct netmap_ring * r)134 nm_tx_pending(struct netmap_ring *r)
135 {
136 return nm_ring_next(r, r->tail) != r->head;
137 }
138
139
140 static inline uint32_t
nm_ring_space(struct netmap_ring * ring)141 nm_ring_space(struct netmap_ring *ring)
142 {
143 int ret = ring->tail - ring->cur;
144 if (ret < 0)
145 ret += ring->num_slots;
146 return ret;
147 }
148
149
150 #ifdef NETMAP_WITH_LIBS
151 /*
152 * Support for simple I/O libraries.
153 * Include other system headers required for compiling this.
154 */
155
156 #ifndef HAVE_NETMAP_WITH_LIBS
157 #define HAVE_NETMAP_WITH_LIBS
158
159 #include <stdio.h>
160 #include <sys/time.h>
161 #include <sys/mman.h>
162 #include <string.h> /* memset */
163 #include <sys/ioctl.h>
164 #include <sys/errno.h> /* EINVAL */
165 #include <fcntl.h> /* O_RDWR */
166 #include <unistd.h> /* close() */
167 #include <signal.h>
168 #include <stdlib.h>
169
170 #ifndef ND /* debug macros */
171 /* debug support */
172 #define ND(_fmt, ...) do {} while(0)
173 #define D(_fmt, ...) \
174 do { \
175 struct timeval _t0; \
176 gettimeofday(&_t0, NULL); \
177 fprintf(stderr, "%03d.%06d %s [%d] " _fmt "\n", \
178 (int)(_t0.tv_sec % 1000), (int)_t0.tv_usec, \
179 __FUNCTION__, __LINE__, ##__VA_ARGS__); \
180 } while (0)
181
182 /* Rate limited version of "D", lps indicates how many per second */
183 #define RD(lps, format, ...) \
184 do { \
185 static int __t0, __cnt; \
186 struct timeval __xxts; \
187 gettimeofday(&__xxts, NULL); \
188 if (__t0 != __xxts.tv_sec) { \
189 __t0 = __xxts.tv_sec; \
190 __cnt = 0; \
191 } \
192 if (__cnt++ < lps) { \
193 D(format, ##__VA_ARGS__); \
194 } \
195 } while (0)
196 #endif
197
198 struct nm_pkthdr { /* same as pcap_pkthdr */
199 struct timeval ts;
200 uint32_t caplen;
201 uint32_t len;
202 };
203
204 struct nm_stat { /* same as pcap_stat */
205 u_int ps_recv;
206 u_int ps_drop;
207 u_int ps_ifdrop;
208 #ifdef WIN32
209 u_int bs_capt;
210 #endif /* WIN32 */
211 };
212
213 #define NM_ERRBUF_SIZE 512
214
215 struct nm_desc {
216 struct nm_desc *self; /* point to self if netmap. */
217 int fd;
218 void *mem;
219 uint32_t memsize;
220 int done_mmap; /* set if mem is the result of mmap */
221 struct netmap_if * const nifp;
222 uint16_t first_tx_ring, last_tx_ring, cur_tx_ring;
223 uint16_t first_rx_ring, last_rx_ring, cur_rx_ring;
224 struct nmreq req; /* also contains the nr_name = ifname */
225 struct nm_pkthdr hdr;
226
227 /*
228 * The memory contains netmap_if, rings and then buffers.
229 * Given a pointer (e.g. to nm_inject) we can compare with
230 * mem/buf_start/buf_end to tell if it is a buffer or
231 * some other descriptor in our region.
232 * We also store a pointer to some ring as it helps in the
233 * translation from buffer indexes to addresses.
234 */
235 struct netmap_ring * const some_ring;
236 void * const buf_start;
237 void * const buf_end;
238 /* parameters from pcap_open_live */
239 int snaplen;
240 int promisc;
241 int to_ms;
242 char *errbuf;
243
244 /* save flags so we can restore them on close */
245 uint32_t if_flags;
246 uint32_t if_reqcap;
247 uint32_t if_curcap;
248
249 struct nm_stat st;
250 char msg[NM_ERRBUF_SIZE];
251 };
252
253 /*
254 * when the descriptor is open correctly, d->self == d
255 * Eventually we should also use some magic number.
256 */
257 #define P2NMD(p) ((struct nm_desc *)(p))
258 #define IS_NETMAP_DESC(d) ((d) && P2NMD(d)->self == P2NMD(d))
259 #define NETMAP_FD(d) (P2NMD(d)->fd)
260
261
262 /*
263 * this is a slightly optimized copy routine which rounds
264 * to multiple of 64 bytes and is often faster than dealing
265 * with other odd sizes. We assume there is enough room
266 * in the source and destination buffers.
267 *
268 * XXX only for multiples of 64 bytes, non overlapped.
269 */
270 static inline void
nm_pkt_copy(const void * _src,void * _dst,int l)271 nm_pkt_copy(const void *_src, void *_dst, int l)
272 {
273 const uint64_t *src = (const uint64_t *)_src;
274 uint64_t *dst = (uint64_t *)_dst;
275
276 if (unlikely(l >= 1024)) {
277 memcpy(dst, src, l);
278 return;
279 }
280 for (; likely(l > 0); l-=64) {
281 *dst++ = *src++;
282 *dst++ = *src++;
283 *dst++ = *src++;
284 *dst++ = *src++;
285 *dst++ = *src++;
286 *dst++ = *src++;
287 *dst++ = *src++;
288 *dst++ = *src++;
289 }
290 }
291
292
293 /*
294 * The callback, invoked on each received packet. Same as libpcap
295 */
296 typedef void (*nm_cb_t)(u_char *, const struct nm_pkthdr *, const u_char *d);
297
298 /*
299 *--- the pcap-like API ---
300 *
301 * nm_open() opens a file descriptor, binds to a port and maps memory.
302 *
303 * ifname (netmap:foo or vale:foo) is the port name
304 * a suffix can indicate the follwing:
305 * ^ bind the host (sw) ring pair
306 * * bind host and NIC ring pairs (transparent)
307 * -NN bind individual NIC ring pair
308 * {NN bind master side of pipe NN
309 * }NN bind slave side of pipe NN
310 * a suffix starting with / and the following flags,
311 * in any order:
312 * x exclusive access
313 * z zero copy monitor
314 * t monitor tx side
315 * r monitor rx side
316 * R bind only RX ring(s)
317 * T bind only TX ring(s)
318 *
319 * req provides the initial values of nmreq before parsing ifname.
320 * Remember that the ifname parsing will override the ring
321 * number in nm_ringid, and part of nm_flags;
322 * flags special functions, normally 0
323 * indicates which fields of *arg are significant
324 * arg special functions, normally NULL
325 * if passed a netmap_desc with mem != NULL,
326 * use that memory instead of mmap.
327 */
328
329 static struct nm_desc *nm_open(const char *ifname, const struct nmreq *req,
330 uint64_t flags, const struct nm_desc *arg);
331
332 /*
333 * nm_open can import some fields from the parent descriptor.
334 * These flags control which ones.
335 * Also in flags you can specify NETMAP_NO_TX_POLL and NETMAP_DO_RX_POLL,
336 * which set the initial value for these flags.
337 * Note that the 16 low bits of the flags are reserved for data
338 * that may go into the nmreq.
339 */
340 enum {
341 NM_OPEN_NO_MMAP = 0x040000, /* reuse mmap from parent */
342 NM_OPEN_IFNAME = 0x080000, /* nr_name, nr_ringid, nr_flags */
343 NM_OPEN_ARG1 = 0x100000,
344 NM_OPEN_ARG2 = 0x200000,
345 NM_OPEN_ARG3 = 0x400000,
346 NM_OPEN_RING_CFG = 0x800000, /* tx|rx rings|slots */
347 };
348
349
350 /*
351 * nm_close() closes and restores the port to its previous state
352 */
353
354 static int nm_close(struct nm_desc *);
355
356 /*
357 * nm_mmap() do mmap or inherit from parent if the nr_arg2
358 * (memory block) matches.
359 */
360
361 static int nm_mmap(struct nm_desc *, const struct nm_desc *);
362
363 /*
364 * nm_inject() is the same as pcap_inject()
365 * nm_dispatch() is the same as pcap_dispatch()
366 * nm_nextpkt() is the same as pcap_next()
367 */
368
369 static int nm_inject(struct nm_desc *, const void *, size_t);
370 static int nm_dispatch(struct nm_desc *, int, nm_cb_t, u_char *);
371 static u_char *nm_nextpkt(struct nm_desc *, struct nm_pkthdr *);
372
373 #ifdef _WIN32
374
375 intptr_t _get_osfhandle(int); /* defined in io.h in windows */
376
377 /*
378 * In windows we do not have yet native poll support, so we keep track
379 * of file descriptors associated to netmap ports to emulate poll on
380 * them and fall back on regular poll on other file descriptors.
381 */
382 struct win_netmap_fd_list {
383 struct win_netmap_fd_list *next;
384 int win_netmap_fd;
385 HANDLE win_netmap_handle;
386 };
387
388 /*
389 * list head containing all the netmap opened fd and their
390 * windows HANDLE counterparts
391 */
392 static struct win_netmap_fd_list *win_netmap_fd_list_head;
393
394 static void
win_insert_fd_record(int fd)395 win_insert_fd_record(int fd)
396 {
397 struct win_netmap_fd_list *curr;
398
399 for (curr = win_netmap_fd_list_head; curr; curr = curr->next) {
400 if (fd == curr->win_netmap_fd) {
401 return;
402 }
403 }
404 curr = calloc(1, sizeof(*curr));
405 curr->next = win_netmap_fd_list_head;
406 curr->win_netmap_fd = fd;
407 curr->win_netmap_handle = IntToPtr(_get_osfhandle(fd));
408 win_netmap_fd_list_head = curr;
409 }
410
411 void
win_remove_fd_record(int fd)412 win_remove_fd_record(int fd)
413 {
414 struct win_netmap_fd_list *curr = win_netmap_fd_list_head;
415 struct win_netmap_fd_list *prev = NULL;
416 for (; curr ; prev = curr, curr = curr->next) {
417 if (fd != curr->win_netmap_fd)
418 continue;
419 /* found the entry */
420 if (prev == NULL) { /* we are freeing the first entry */
421 win_netmap_fd_list_head = curr->next;
422 } else {
423 prev->next = curr->next;
424 }
425 free(curr);
426 break;
427 }
428 }
429
430
431 HANDLE
win_get_netmap_handle(int fd)432 win_get_netmap_handle(int fd)
433 {
434 struct win_netmap_fd_list *curr;
435
436 for (curr = win_netmap_fd_list_head; curr; curr = curr->next) {
437 if (fd == curr->win_netmap_fd) {
438 return curr->win_netmap_handle;
439 }
440 }
441 return NULL;
442 }
443
444 /*
445 * we need to wrap ioctl and mmap, at least for the netmap file descriptors
446 */
447
448 /*
449 * use this function only from netmap_user.h internal functions
450 * same as ioctl, returns 0 on success and -1 on error
451 */
452 static int
win_nm_ioctl_internal(HANDLE h,int32_t ctlCode,void * arg)453 win_nm_ioctl_internal(HANDLE h, int32_t ctlCode, void *arg)
454 {
455 DWORD bReturn = 0, szIn, szOut;
456 BOOL ioctlReturnStatus;
457 void *inParam = arg, *outParam = arg;
458
459 switch (ctlCode) {
460 case NETMAP_POLL:
461 szIn = sizeof(POLL_REQUEST_DATA);
462 szOut = sizeof(POLL_REQUEST_DATA);
463 break;
464 case NETMAP_MMAP:
465 szIn = 0;
466 szOut = sizeof(void*);
467 inParam = NULL; /* nothing on input */
468 break;
469 case NIOCTXSYNC:
470 case NIOCRXSYNC:
471 szIn = 0;
472 szOut = 0;
473 break;
474 case NIOCREGIF:
475 szIn = sizeof(struct nmreq);
476 szOut = sizeof(struct nmreq);
477 break;
478 case NIOCCONFIG:
479 D("unsupported NIOCCONFIG!");
480 return -1;
481
482 default: /* a regular ioctl */
483 D("invalid ioctl %x on netmap fd", ctlCode);
484 return -1;
485 }
486
487 ioctlReturnStatus = DeviceIoControl(h,
488 ctlCode, inParam, szIn,
489 outParam, szOut,
490 &bReturn, NULL);
491 // XXX note windows returns 0 on error or async call, 1 on success
492 // we could call GetLastError() to figure out what happened
493 return ioctlReturnStatus ? 0 : -1;
494 }
495
496 /*
497 * this function is what must be called from user-space programs
498 * same as ioctl, returns 0 on success and -1 on error
499 */
500 static int
win_nm_ioctl(int fd,int32_t ctlCode,void * arg)501 win_nm_ioctl(int fd, int32_t ctlCode, void *arg)
502 {
503 HANDLE h = win_get_netmap_handle(fd);
504
505 if (h == NULL) {
506 return ioctl(fd, ctlCode, arg);
507 } else {
508 return win_nm_ioctl_internal(h, ctlCode, arg);
509 }
510 }
511
512 #define ioctl win_nm_ioctl /* from now on, within this file ... */
513
514 /*
515 * We cannot use the native mmap on windows
516 * The only parameter used is "fd", the other ones are just declared to
517 * make this signature comparable to the FreeBSD/Linux one
518 */
519 static void *
win32_mmap_emulated(void * addr,size_t length,int prot,int flags,int fd,int32_t offset)520 win32_mmap_emulated(void *addr, size_t length, int prot, int flags, int fd, int32_t offset)
521 {
522 HANDLE h = win_get_netmap_handle(fd);
523
524 if (h == NULL) {
525 return mmap(addr, length, prot, flags, fd, offset);
526 } else {
527 MEMORY_ENTRY ret;
528
529 return win_nm_ioctl_internal(h, NETMAP_MMAP, &ret) ?
530 NULL : ret.pUsermodeVirtualAddress;
531 }
532 }
533
534 #define mmap win32_mmap_emulated
535
536 #include <sys/poll.h> /* XXX needed to use the structure pollfd */
537
538 static int
win_nm_poll(struct pollfd * fds,int nfds,int timeout)539 win_nm_poll(struct pollfd *fds, int nfds, int timeout)
540 {
541 HANDLE h;
542
543 if (nfds != 1 || fds == NULL || (h = win_get_netmap_handle(fds->fd)) == NULL) {;
544 return poll(fds, nfds, timeout);
545 } else {
546 POLL_REQUEST_DATA prd;
547
548 prd.timeout = timeout;
549 prd.events = fds->events;
550
551 win_nm_ioctl_internal(h, NETMAP_POLL, &prd);
552 if ((prd.revents == POLLERR) || (prd.revents == STATUS_TIMEOUT)) {
553 return -1;
554 }
555 return 1;
556 }
557 }
558
559 #define poll win_nm_poll
560
561 static int
win_nm_open(char * pathname,int flags)562 win_nm_open(char* pathname, int flags){
563
564 if (strcmp(pathname, NETMAP_DEVICE_NAME) == 0){
565 int fd = open(NETMAP_DEVICE_NAME, O_RDWR);
566 if (fd < 0) {
567 return -1;
568 }
569
570 win_insert_fd_record(fd);
571 return fd;
572 }
573 else {
574
575 return open(pathname, flags);
576 }
577 }
578
579 #define open win_nm_open
580
581 static int
win_nm_close(int fd)582 win_nm_close(int fd){
583 if (fd != -1){
584 close(fd);
585 if (win_get_netmap_handle(fd) != NULL){
586 win_remove_fd_record(fd);
587 }
588 }
589 return 0;
590 }
591
592 #define close win_nm_close
593
594 #endif /* _WIN32 */
595
596 /*
597 * Try to open, return descriptor if successful, NULL otherwise.
598 * An invalid netmap name will return errno = 0;
599 * You can pass a pointer to a pre-filled nm_desc to add special
600 * parameters. Flags is used as follows
601 * NM_OPEN_NO_MMAP use the memory from arg, only XXX avoid mmap
602 * if the nr_arg2 (memory block) matches.
603 * NM_OPEN_ARG1 use req.nr_arg1 from arg
604 * NM_OPEN_ARG2 use req.nr_arg2 from arg
605 * NM_OPEN_RING_CFG user ring config from arg
606 */
607 static struct nm_desc *
nm_open(const char * ifname,const struct nmreq * req,uint64_t new_flags,const struct nm_desc * arg)608 nm_open(const char *ifname, const struct nmreq *req,
609 uint64_t new_flags, const struct nm_desc *arg)
610 {
611 struct nm_desc *d = NULL;
612 const struct nm_desc *parent = arg;
613 u_int namelen;
614 uint32_t nr_ringid = 0, nr_flags, nr_reg;
615 const char *port = NULL;
616 #define MAXERRMSG 80
617 char errmsg[MAXERRMSG] = "";
618 enum { P_START, P_RNGSFXOK, P_GETNUM, P_FLAGS, P_FLAGSOK } p_state;
619 long num;
620
621 if (strncmp(ifname, "netmap:", 7) && strncmp(ifname, "vale", 4)) {
622 errno = 0; /* name not recognised, not an error */
623 return NULL;
624 }
625 if (ifname[0] == 'n')
626 ifname += 7;
627 /* scan for a separator */
628 for (port = ifname; *port && !index("-*^{}/", *port); port++)
629 ;
630 namelen = port - ifname;
631 if (namelen >= sizeof(d->req.nr_name)) {
632 snprintf(errmsg, MAXERRMSG, "name too long");
633 goto fail;
634 }
635 p_state = P_START;
636 nr_flags = NR_REG_ALL_NIC; /* default for no suffix */
637 while (*port) {
638 switch (p_state) {
639 case P_START:
640 switch (*port) {
641 case '^': /* only SW ring */
642 nr_flags = NR_REG_SW;
643 p_state = P_RNGSFXOK;
644 break;
645 case '*': /* NIC and SW */
646 nr_flags = NR_REG_NIC_SW;
647 p_state = P_RNGSFXOK;
648 break;
649 case '-': /* one NIC ring pair */
650 nr_flags = NR_REG_ONE_NIC;
651 p_state = P_GETNUM;
652 break;
653 case '{': /* pipe (master endpoint) */
654 nr_flags = NR_REG_PIPE_MASTER;
655 p_state = P_GETNUM;
656 break;
657 case '}': /* pipe (slave endoint) */
658 nr_flags = NR_REG_PIPE_SLAVE;
659 p_state = P_GETNUM;
660 break;
661 case '/': /* start of flags */
662 p_state = P_FLAGS;
663 break;
664 default:
665 snprintf(errmsg, MAXERRMSG, "unknown modifier: '%c'", *port);
666 goto fail;
667 }
668 port++;
669 break;
670 case P_RNGSFXOK:
671 switch (*port) {
672 case '/':
673 p_state = P_FLAGS;
674 break;
675 default:
676 snprintf(errmsg, MAXERRMSG, "unexpected character: '%c'", *port);
677 goto fail;
678 }
679 port++;
680 break;
681 case P_GETNUM:
682 num = strtol(port, (char **)&port, 10);
683 if (num < 0 || num >= NETMAP_RING_MASK) {
684 snprintf(errmsg, MAXERRMSG, "'%ld' out of range [0, %d)",
685 num, NETMAP_RING_MASK);
686 goto fail;
687 }
688 nr_ringid = num & NETMAP_RING_MASK;
689 p_state = P_RNGSFXOK;
690 break;
691 case P_FLAGS:
692 case P_FLAGSOK:
693 switch (*port) {
694 case 'x':
695 nr_flags |= NR_EXCLUSIVE;
696 break;
697 case 'z':
698 nr_flags |= NR_ZCOPY_MON;
699 break;
700 case 't':
701 nr_flags |= NR_MONITOR_TX;
702 break;
703 case 'r':
704 nr_flags |= NR_MONITOR_RX;
705 break;
706 case 'R':
707 nr_flags |= NR_RX_RINGS_ONLY;
708 break;
709 case 'T':
710 nr_flags |= NR_TX_RINGS_ONLY;
711 break;
712 default:
713 snprintf(errmsg, MAXERRMSG, "unrecognized flag: '%c'", *port);
714 goto fail;
715 }
716 port++;
717 p_state = P_FLAGSOK;
718 break;
719 }
720 }
721 if (p_state != P_START && p_state != P_RNGSFXOK && p_state != P_FLAGSOK) {
722 snprintf(errmsg, MAXERRMSG, "unexpected end of port name");
723 goto fail;
724 }
725 ND("flags: %s %s %s %s",
726 (nr_flags & NR_EXCLUSIVE) ? "EXCLUSIVE" : "",
727 (nr_flags & NR_ZCOPY_MON) ? "ZCOPY_MON" : "",
728 (nr_flags & NR_MONITOR_TX) ? "MONITOR_TX" : "",
729 (nr_flags & NR_MONITOR_RX) ? "MONITOR_RX" : "");
730 d = (struct nm_desc *)calloc(1, sizeof(*d));
731 if (d == NULL) {
732 snprintf(errmsg, MAXERRMSG, "nm_desc alloc failure");
733 errno = ENOMEM;
734 return NULL;
735 }
736 d->self = d; /* set this early so nm_close() works */
737 d->fd = open(NETMAP_DEVICE_NAME, O_RDWR);
738 if (d->fd < 0) {
739 snprintf(errmsg, MAXERRMSG, "cannot open /dev/netmap: %s", strerror(errno));
740 goto fail;
741 }
742
743 if (req)
744 d->req = *req;
745 d->req.nr_version = NETMAP_API;
746 d->req.nr_ringid &= ~NETMAP_RING_MASK;
747
748 /* these fields are overridden by ifname and flags processing */
749 d->req.nr_ringid |= nr_ringid;
750 d->req.nr_flags |= nr_flags;
751 memcpy(d->req.nr_name, ifname, namelen);
752 d->req.nr_name[namelen] = '\0';
753 /* optionally import info from parent */
754 if (IS_NETMAP_DESC(parent) && new_flags) {
755 if (new_flags & NM_OPEN_ARG1)
756 D("overriding ARG1 %d", parent->req.nr_arg1);
757 d->req.nr_arg1 = new_flags & NM_OPEN_ARG1 ?
758 parent->req.nr_arg1 : 4;
759 if (new_flags & NM_OPEN_ARG2)
760 D("overriding ARG2 %d", parent->req.nr_arg2);
761 d->req.nr_arg2 = new_flags & NM_OPEN_ARG2 ?
762 parent->req.nr_arg2 : 0;
763 if (new_flags & NM_OPEN_ARG3)
764 D("overriding ARG3 %d", parent->req.nr_arg3);
765 d->req.nr_arg3 = new_flags & NM_OPEN_ARG3 ?
766 parent->req.nr_arg3 : 0;
767 if (new_flags & NM_OPEN_RING_CFG) {
768 D("overriding RING_CFG");
769 d->req.nr_tx_slots = parent->req.nr_tx_slots;
770 d->req.nr_rx_slots = parent->req.nr_rx_slots;
771 d->req.nr_tx_rings = parent->req.nr_tx_rings;
772 d->req.nr_rx_rings = parent->req.nr_rx_rings;
773 }
774 if (new_flags & NM_OPEN_IFNAME) {
775 D("overriding ifname %s ringid 0x%x flags 0x%x",
776 parent->req.nr_name, parent->req.nr_ringid,
777 parent->req.nr_flags);
778 memcpy(d->req.nr_name, parent->req.nr_name,
779 sizeof(d->req.nr_name));
780 d->req.nr_ringid = parent->req.nr_ringid;
781 d->req.nr_flags = parent->req.nr_flags;
782 }
783 }
784 /* add the *XPOLL flags */
785 d->req.nr_ringid |= new_flags & (NETMAP_NO_TX_POLL | NETMAP_DO_RX_POLL);
786
787 if (ioctl(d->fd, NIOCREGIF, &d->req)) {
788 snprintf(errmsg, MAXERRMSG, "NIOCREGIF failed: %s", strerror(errno));
789 goto fail;
790 }
791
792 /* if parent is defined, do nm_mmap() even if NM_OPEN_NO_MMAP is set */
793 if ((!(new_flags & NM_OPEN_NO_MMAP) || parent) && nm_mmap(d, parent)) {
794 snprintf(errmsg, MAXERRMSG, "mmap failed: %s", strerror(errno));
795 goto fail;
796 }
797
798 nr_reg = d->req.nr_flags & NR_REG_MASK;
799
800 if (nr_reg == NR_REG_SW) { /* host stack */
801 d->first_tx_ring = d->last_tx_ring = d->req.nr_tx_rings;
802 d->first_rx_ring = d->last_rx_ring = d->req.nr_rx_rings;
803 } else if (nr_reg == NR_REG_ALL_NIC) { /* only nic */
804 d->first_tx_ring = 0;
805 d->first_rx_ring = 0;
806 d->last_tx_ring = d->req.nr_tx_rings - 1;
807 d->last_rx_ring = d->req.nr_rx_rings - 1;
808 } else if (nr_reg == NR_REG_NIC_SW) {
809 d->first_tx_ring = 0;
810 d->first_rx_ring = 0;
811 d->last_tx_ring = d->req.nr_tx_rings;
812 d->last_rx_ring = d->req.nr_rx_rings;
813 } else if (nr_reg == NR_REG_ONE_NIC) {
814 /* XXX check validity */
815 d->first_tx_ring = d->last_tx_ring =
816 d->first_rx_ring = d->last_rx_ring = d->req.nr_ringid & NETMAP_RING_MASK;
817 } else { /* pipes */
818 d->first_tx_ring = d->last_tx_ring = 0;
819 d->first_rx_ring = d->last_rx_ring = 0;
820 }
821
822 #ifdef DEBUG_NETMAP_USER
823 { /* debugging code */
824 int i;
825
826 D("%s tx %d .. %d %d rx %d .. %d %d", ifname,
827 d->first_tx_ring, d->last_tx_ring, d->req.nr_tx_rings,
828 d->first_rx_ring, d->last_rx_ring, d->req.nr_rx_rings);
829 for (i = 0; i <= d->req.nr_tx_rings; i++) {
830 struct netmap_ring *r = NETMAP_TXRING(d->nifp, i);
831 D("TX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
832 }
833 for (i = 0; i <= d->req.nr_rx_rings; i++) {
834 struct netmap_ring *r = NETMAP_RXRING(d->nifp, i);
835 D("RX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
836 }
837 }
838 #endif /* debugging */
839
840 d->cur_tx_ring = d->first_tx_ring;
841 d->cur_rx_ring = d->first_rx_ring;
842 return d;
843
844 fail:
845 nm_close(d);
846 if (errmsg[0])
847 D("%s %s", errmsg, ifname);
848 if (errno == 0)
849 errno = EINVAL;
850 return NULL;
851 }
852
853
854 static int
nm_close(struct nm_desc * d)855 nm_close(struct nm_desc *d)
856 {
857 /*
858 * ugly trick to avoid unused warnings
859 */
860 static void *__xxzt[] __attribute__ ((unused)) =
861 { (void *)nm_open, (void *)nm_inject,
862 (void *)nm_dispatch, (void *)nm_nextpkt } ;
863
864 if (d == NULL || d->self != d)
865 return EINVAL;
866 if (d->done_mmap && d->mem)
867 munmap(d->mem, d->memsize);
868 if (d->fd != -1){
869 close(d->fd);
870 }
871
872 bzero(d, sizeof(*d));
873 free(d);
874 return 0;
875 }
876
877
878 static int
nm_mmap(struct nm_desc * d,const struct nm_desc * parent)879 nm_mmap(struct nm_desc *d, const struct nm_desc *parent)
880 {
881 //XXX TODO: check if mmap is already done
882
883 if (IS_NETMAP_DESC(parent) && parent->mem &&
884 parent->req.nr_arg2 == d->req.nr_arg2) {
885 /* do not mmap, inherit from parent */
886 D("do not mmap, inherit from parent");
887 d->memsize = parent->memsize;
888 d->mem = parent->mem;
889 } else {
890 /* XXX TODO: check if memsize is too large (or there is overflow) */
891 d->memsize = d->req.nr_memsize;
892 d->mem = mmap(0, d->memsize, PROT_WRITE | PROT_READ, MAP_SHARED,
893 d->fd, 0);
894 if (d->mem == MAP_FAILED) {
895 goto fail;
896 }
897 d->done_mmap = 1;
898 }
899 {
900 struct netmap_if *nifp = NETMAP_IF(d->mem, d->req.nr_offset);
901 struct netmap_ring *r = NETMAP_RXRING(nifp, );
902
903 *(struct netmap_if **)(uintptr_t)&(d->nifp) = nifp;
904 *(struct netmap_ring **)(uintptr_t)&d->some_ring = r;
905 *(void **)(uintptr_t)&d->buf_start = NETMAP_BUF(r, 0);
906 *(void **)(uintptr_t)&d->buf_end =
907 (char *)d->mem + d->memsize;
908 }
909
910 return 0;
911
912 fail:
913 return EINVAL;
914 }
915
916 /*
917 * Same prototype as pcap_inject(), only need to cast.
918 */
919 static int
nm_inject(struct nm_desc * d,const void * buf,size_t size)920 nm_inject(struct nm_desc *d, const void *buf, size_t size)
921 {
922 u_int c, n = d->last_tx_ring - d->first_tx_ring + 1;
923
924 for (c = 0; c < n ; c++) {
925 /* compute current ring to use */
926 struct netmap_ring *ring;
927 uint32_t i, idx;
928 uint32_t ri = d->cur_tx_ring + c;
929
930 if (ri > d->last_tx_ring)
931 ri = d->first_tx_ring;
932 ring = NETMAP_TXRING(d->nifp, ri);
933 if (nm_ring_empty(ring)) {
934 continue;
935 }
936 i = ring->cur;
937 idx = ring->slot[i].buf_idx;
938 ring->slot[i].len = size;
939 nm_pkt_copy(buf, NETMAP_BUF(ring, idx), size);
940 d->cur_tx_ring = ri;
941 ring->head = ring->cur = nm_ring_next(ring, i);
942 return size;
943 }
944 return 0; /* fail */
945 }
946
947
948 /*
949 * Same prototype as pcap_dispatch(), only need to cast.
950 */
951 static int
nm_dispatch(struct nm_desc * d,int cnt,nm_cb_t cb,u_char * arg)952 nm_dispatch(struct nm_desc *d, int cnt, nm_cb_t cb, u_char *arg)
953 {
954 int n = d->last_rx_ring - d->first_rx_ring + 1;
955 int c, got = 0, ri = d->cur_rx_ring;
956
957 if (cnt == 0)
958 cnt = -1;
959 /* cnt == -1 means infinite, but rings have a finite amount
960 * of buffers and the int is large enough that we never wrap,
961 * so we can omit checking for -1
962 */
963 for (c=0; c < n && cnt != got; c++) {
964 /* compute current ring to use */
965 struct netmap_ring *ring;
966
967 ri = d->cur_rx_ring + c;
968 if (ri > d->last_rx_ring)
969 ri = d->first_rx_ring;
970 ring = NETMAP_RXRING(d->nifp, ri);
971 for ( ; !nm_ring_empty(ring) && cnt != got; got++) {
972 u_int i = ring->cur;
973 u_int idx = ring->slot[i].buf_idx;
974 u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
975
976 // __builtin_prefetch(buf);
977 d->hdr.len = d->hdr.caplen = ring->slot[i].len;
978 d->hdr.ts = ring->ts;
979 cb(arg, &d->hdr, buf);
980 ring->head = ring->cur = nm_ring_next(ring, i);
981 }
982 }
983 d->cur_rx_ring = ri;
984 return got;
985 }
986
987 static u_char *
nm_nextpkt(struct nm_desc * d,struct nm_pkthdr * hdr)988 nm_nextpkt(struct nm_desc *d, struct nm_pkthdr *hdr)
989 {
990 int ri = d->cur_rx_ring;
991
992 do {
993 /* compute current ring to use */
994 struct netmap_ring *ring = NETMAP_RXRING(d->nifp, ri);
995 if (!nm_ring_empty(ring)) {
996 u_int i = ring->cur;
997 u_int idx = ring->slot[i].buf_idx;
998 u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
999
1000 // __builtin_prefetch(buf);
1001 hdr->ts = ring->ts;
1002 hdr->len = hdr->caplen = ring->slot[i].len;
1003 ring->cur = nm_ring_next(ring, i);
1004 /* we could postpone advancing head if we want
1005 * to hold the buffer. This can be supported in
1006 * the future.
1007 */
1008 ring->head = ring->cur;
1009 d->cur_rx_ring = ri;
1010 return buf;
1011 }
1012 ri++;
1013 if (ri > d->last_rx_ring)
1014 ri = d->first_rx_ring;
1015 } while (ri != d->cur_rx_ring);
1016 return NULL; /* nothing found */
1017 }
1018
1019 #endif /* !HAVE_NETMAP_WITH_LIBS */
1020
1021 #endif /* NETMAP_WITH_LIBS */
1022
1023 #endif /* _NET_NETMAP_USER_H_ */
1024