FreeBSD/Linux Kernel Cross Reference
sys/net/netmap_user.h
1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 * $FreeBSD: releng/12.0/sys/net/netmap_user.h 339906 2018-10-30 08:36:36Z vmaffione $
32 *
33 * Functions and macros to manipulate netmap structures and packets
34 * in userspace. See netmap(4) for more information.
35 *
36 * The address of the struct netmap_if, say nifp, is computed from the
37 * value returned from ioctl(.., NIOCREG, ...) and the mmap region:
38 * ioctl(fd, NIOCREG, &req);
39 * mem = mmap(0, ... );
40 * nifp = NETMAP_IF(mem, req.nr_nifp);
41 * (so simple, we could just do it manually)
42 *
43 * From there:
44 * struct netmap_ring *NETMAP_TXRING(nifp, index)
45 * struct netmap_ring *NETMAP_RXRING(nifp, index)
46 * we can access ring->cur, ring->head, ring->tail, etc.
47 *
48 * ring->slot[i] gives us the i-th slot (we can access
49 * directly len, flags, buf_idx)
50 *
51 * char *buf = NETMAP_BUF(ring, x) returns a pointer to
52 * the buffer numbered x
53 *
54 * All ring indexes (head, cur, tail) should always move forward.
55 * To compute the next index in a circular ring you can use
56 * i = nm_ring_next(ring, i);
57 *
58 * To ease porting apps from pcap to netmap we supply a few fuctions
59 * that can be called to open, close, read and write on netmap in a way
60 * similar to libpcap. Note that the read/write function depend on
61 * an ioctl()/select()/poll() being issued to refill rings or push
62 * packets out.
63 *
64 * In order to use these, include #define NETMAP_WITH_LIBS
65 * in the source file that invokes these functions.
66 */
67
68 #ifndef _NET_NETMAP_USER_H_
69 #define _NET_NETMAP_USER_H_
70
71 #define NETMAP_DEVICE_NAME "/dev/netmap"
72
73 #ifdef __CYGWIN__
74 /*
75 * we can compile userspace apps with either cygwin or msvc,
76 * and we use _WIN32 to identify windows specific code
77 */
78 #ifndef _WIN32
79 #define _WIN32
80 #endif /* _WIN32 */
81
82 #endif /* __CYGWIN__ */
83
84 #ifdef _WIN32
85 #undef NETMAP_DEVICE_NAME
86 #define NETMAP_DEVICE_NAME "/proc/sys/DosDevices/Global/netmap"
87 #include <windows.h>
88 #include <WinDef.h>
89 #include <sys/cygwin.h>
90 #endif /* _WIN32 */
91
92 #include <stdint.h>
93 #include <sys/socket.h> /* apple needs sockaddr */
94 #include <net/if.h> /* IFNAMSIZ */
95 #include <ctype.h>
96
97 #ifndef likely
98 #define likely(x) __builtin_expect(!!(x), 1)
99 #define unlikely(x) __builtin_expect(!!(x), 0)
100 #endif /* likely and unlikely */
101
102 #include <net/netmap.h>
103
104 /* helper macro */
105 #define _NETMAP_OFFSET(type, ptr, offset) \
106 ((type)(void *)((char *)(ptr) + (offset)))
107
108 #define NETMAP_IF(_base, _ofs) _NETMAP_OFFSET(struct netmap_if *, _base, _ofs)
109
110 #define NETMAP_TXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
111 nifp, (nifp)->ring_ofs[index] )
112
113 #define NETMAP_RXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
114 nifp, (nifp)->ring_ofs[index + (nifp)->ni_tx_rings + 1] )
115
116 #define NETMAP_BUF(ring, index) \
117 ((char *)(ring) + (ring)->buf_ofs + ((index)*(ring)->nr_buf_size))
118
119 #define NETMAP_BUF_IDX(ring, buf) \
120 ( ((char *)(buf) - ((char *)(ring) + (ring)->buf_ofs) ) / \
121 (ring)->nr_buf_size )
122
123
124 static inline uint32_t
125 nm_ring_next(struct netmap_ring *r, uint32_t i)
126 {
127 return ( unlikely(i + 1 == r->num_slots) ? 0 : i + 1);
128 }
129
130
131 /*
132 * Return 1 if we have pending transmissions in the tx ring.
133 * When everything is complete ring->head = ring->tail + 1 (modulo ring size)
134 */
135 static inline int
136 nm_tx_pending(struct netmap_ring *r)
137 {
138 return nm_ring_next(r, r->tail) != r->head;
139 }
140
141
142 static inline uint32_t
143 nm_ring_space(struct netmap_ring *ring)
144 {
145 int ret = ring->tail - ring->cur;
146 if (ret < 0)
147 ret += ring->num_slots;
148 return ret;
149 }
150
151
152 #ifdef NETMAP_WITH_LIBS
153 /*
154 * Support for simple I/O libraries.
155 * Include other system headers required for compiling this.
156 */
157
158 #ifndef HAVE_NETMAP_WITH_LIBS
159 #define HAVE_NETMAP_WITH_LIBS
160
161 #include <stdio.h>
162 #include <sys/time.h>
163 #include <sys/mman.h>
164 #include <string.h> /* memset */
165 #include <sys/ioctl.h>
166 #include <sys/errno.h> /* EINVAL */
167 #include <fcntl.h> /* O_RDWR */
168 #include <unistd.h> /* close() */
169 #include <signal.h>
170 #include <stdlib.h>
171
172 #ifndef ND /* debug macros */
173 /* debug support */
174 #define ND(_fmt, ...) do {} while(0)
175 #define D(_fmt, ...) \
176 do { \
177 struct timeval _t0; \
178 gettimeofday(&_t0, NULL); \
179 fprintf(stderr, "%03d.%06d %s [%d] " _fmt "\n", \
180 (int)(_t0.tv_sec % 1000), (int)_t0.tv_usec, \
181 __FUNCTION__, __LINE__, ##__VA_ARGS__); \
182 } while (0)
183
184 /* Rate limited version of "D", lps indicates how many per second */
185 #define RD(lps, format, ...) \
186 do { \
187 static int __t0, __cnt; \
188 struct timeval __xxts; \
189 gettimeofday(&__xxts, NULL); \
190 if (__t0 != __xxts.tv_sec) { \
191 __t0 = __xxts.tv_sec; \
192 __cnt = 0; \
193 } \
194 if (__cnt++ < lps) { \
195 D(format, ##__VA_ARGS__); \
196 } \
197 } while (0)
198 #endif
199
200 struct nm_pkthdr { /* first part is the same as pcap_pkthdr */
201 struct timeval ts;
202 uint32_t caplen;
203 uint32_t len;
204
205 uint64_t flags; /* NM_MORE_PKTS etc */
206 #define NM_MORE_PKTS 1
207 struct nm_desc *d;
208 struct netmap_slot *slot;
209 uint8_t *buf;
210 };
211
212 struct nm_stat { /* same as pcap_stat */
213 u_int ps_recv;
214 u_int ps_drop;
215 u_int ps_ifdrop;
216 #ifdef WIN32 /* XXX or _WIN32 ? */
217 u_int bs_capt;
218 #endif /* WIN32 */
219 };
220
221 #define NM_ERRBUF_SIZE 512
222
223 struct nm_desc {
224 struct nm_desc *self; /* point to self if netmap. */
225 int fd;
226 void *mem;
227 uint32_t memsize;
228 int done_mmap; /* set if mem is the result of mmap */
229 struct netmap_if * const nifp;
230 uint16_t first_tx_ring, last_tx_ring, cur_tx_ring;
231 uint16_t first_rx_ring, last_rx_ring, cur_rx_ring;
232 struct nmreq req; /* also contains the nr_name = ifname */
233 struct nm_pkthdr hdr;
234
235 /*
236 * The memory contains netmap_if, rings and then buffers.
237 * Given a pointer (e.g. to nm_inject) we can compare with
238 * mem/buf_start/buf_end to tell if it is a buffer or
239 * some other descriptor in our region.
240 * We also store a pointer to some ring as it helps in the
241 * translation from buffer indexes to addresses.
242 */
243 struct netmap_ring * const some_ring;
244 void * const buf_start;
245 void * const buf_end;
246 /* parameters from pcap_open_live */
247 int snaplen;
248 int promisc;
249 int to_ms;
250 char *errbuf;
251
252 /* save flags so we can restore them on close */
253 uint32_t if_flags;
254 uint32_t if_reqcap;
255 uint32_t if_curcap;
256
257 struct nm_stat st;
258 char msg[NM_ERRBUF_SIZE];
259 };
260
261 /*
262 * when the descriptor is open correctly, d->self == d
263 * Eventually we should also use some magic number.
264 */
265 #define P2NMD(p) ((struct nm_desc *)(p))
266 #define IS_NETMAP_DESC(d) ((d) && P2NMD(d)->self == P2NMD(d))
267 #define NETMAP_FD(d) (P2NMD(d)->fd)
268
269
270 /*
271 * this is a slightly optimized copy routine which rounds
272 * to multiple of 64 bytes and is often faster than dealing
273 * with other odd sizes. We assume there is enough room
274 * in the source and destination buffers.
275 */
276 static inline void
277 nm_pkt_copy(const void *_src, void *_dst, int l)
278 {
279 const uint64_t *src = (const uint64_t *)_src;
280 uint64_t *dst = (uint64_t *)_dst;
281
282 if (unlikely(l >= 1024 || l % 64)) {
283 memcpy(dst, src, l);
284 return;
285 }
286 for (; likely(l > 0); l-=64) {
287 *dst++ = *src++;
288 *dst++ = *src++;
289 *dst++ = *src++;
290 *dst++ = *src++;
291 *dst++ = *src++;
292 *dst++ = *src++;
293 *dst++ = *src++;
294 *dst++ = *src++;
295 }
296 }
297
298
299 /*
300 * The callback, invoked on each received packet. Same as libpcap
301 */
302 typedef void (*nm_cb_t)(u_char *, const struct nm_pkthdr *, const u_char *d);
303
304 /*
305 *--- the pcap-like API ---
306 *
307 * nm_open() opens a file descriptor, binds to a port and maps memory.
308 *
309 * ifname (netmap:foo or vale:foo) is the port name
310 * a suffix can indicate the follwing:
311 * ^ bind the host (sw) ring pair
312 * * bind host and NIC ring pairs
313 * -NN bind individual NIC ring pair
314 * {NN bind master side of pipe NN
315 * }NN bind slave side of pipe NN
316 * a suffix starting with / and the following flags,
317 * in any order:
318 * x exclusive access
319 * z zero copy monitor (both tx and rx)
320 * t monitor tx side (copy monitor)
321 * r monitor rx side (copy monitor)
322 * R bind only RX ring(s)
323 * T bind only TX ring(s)
324 *
325 * req provides the initial values of nmreq before parsing ifname.
326 * Remember that the ifname parsing will override the ring
327 * number in nm_ringid, and part of nm_flags;
328 * flags special functions, normally 0
329 * indicates which fields of *arg are significant
330 * arg special functions, normally NULL
331 * if passed a netmap_desc with mem != NULL,
332 * use that memory instead of mmap.
333 */
334
335 static struct nm_desc *nm_open(const char *ifname, const struct nmreq *req,
336 uint64_t flags, const struct nm_desc *arg);
337
338 /*
339 * nm_open can import some fields from the parent descriptor.
340 * These flags control which ones.
341 * Also in flags you can specify NETMAP_NO_TX_POLL and NETMAP_DO_RX_POLL,
342 * which set the initial value for these flags.
343 * Note that the 16 low bits of the flags are reserved for data
344 * that may go into the nmreq.
345 */
346 enum {
347 NM_OPEN_NO_MMAP = 0x040000, /* reuse mmap from parent */
348 NM_OPEN_IFNAME = 0x080000, /* nr_name, nr_ringid, nr_flags */
349 NM_OPEN_ARG1 = 0x100000,
350 NM_OPEN_ARG2 = 0x200000,
351 NM_OPEN_ARG3 = 0x400000,
352 NM_OPEN_RING_CFG = 0x800000, /* tx|rx rings|slots */
353 };
354
355
356 /*
357 * nm_close() closes and restores the port to its previous state
358 */
359
360 static int nm_close(struct nm_desc *);
361
362 /*
363 * nm_mmap() do mmap or inherit from parent if the nr_arg2
364 * (memory block) matches.
365 */
366
367 static int nm_mmap(struct nm_desc *, const struct nm_desc *);
368
369 /*
370 * nm_inject() is the same as pcap_inject()
371 * nm_dispatch() is the same as pcap_dispatch()
372 * nm_nextpkt() is the same as pcap_next()
373 */
374
375 static int nm_inject(struct nm_desc *, const void *, size_t);
376 static int nm_dispatch(struct nm_desc *, int, nm_cb_t, u_char *);
377 static u_char *nm_nextpkt(struct nm_desc *, struct nm_pkthdr *);
378
379 #ifdef _WIN32
380
381 intptr_t _get_osfhandle(int); /* defined in io.h in windows */
382
383 /*
384 * In windows we do not have yet native poll support, so we keep track
385 * of file descriptors associated to netmap ports to emulate poll on
386 * them and fall back on regular poll on other file descriptors.
387 */
388 struct win_netmap_fd_list {
389 struct win_netmap_fd_list *next;
390 int win_netmap_fd;
391 HANDLE win_netmap_handle;
392 };
393
394 /*
395 * list head containing all the netmap opened fd and their
396 * windows HANDLE counterparts
397 */
398 static struct win_netmap_fd_list *win_netmap_fd_list_head;
399
400 static void
401 win_insert_fd_record(int fd)
402 {
403 struct win_netmap_fd_list *curr;
404
405 for (curr = win_netmap_fd_list_head; curr; curr = curr->next) {
406 if (fd == curr->win_netmap_fd) {
407 return;
408 }
409 }
410 curr = calloc(1, sizeof(*curr));
411 curr->next = win_netmap_fd_list_head;
412 curr->win_netmap_fd = fd;
413 curr->win_netmap_handle = IntToPtr(_get_osfhandle(fd));
414 win_netmap_fd_list_head = curr;
415 }
416
417 void
418 win_remove_fd_record(int fd)
419 {
420 struct win_netmap_fd_list *curr = win_netmap_fd_list_head;
421 struct win_netmap_fd_list *prev = NULL;
422 for (; curr ; prev = curr, curr = curr->next) {
423 if (fd != curr->win_netmap_fd)
424 continue;
425 /* found the entry */
426 if (prev == NULL) { /* we are freeing the first entry */
427 win_netmap_fd_list_head = curr->next;
428 } else {
429 prev->next = curr->next;
430 }
431 free(curr);
432 break;
433 }
434 }
435
436
437 HANDLE
438 win_get_netmap_handle(int fd)
439 {
440 struct win_netmap_fd_list *curr;
441
442 for (curr = win_netmap_fd_list_head; curr; curr = curr->next) {
443 if (fd == curr->win_netmap_fd) {
444 return curr->win_netmap_handle;
445 }
446 }
447 return NULL;
448 }
449
450 /*
451 * we need to wrap ioctl and mmap, at least for the netmap file descriptors
452 */
453
454 /*
455 * use this function only from netmap_user.h internal functions
456 * same as ioctl, returns 0 on success and -1 on error
457 */
458 static int
459 win_nm_ioctl_internal(HANDLE h, int32_t ctlCode, void *arg)
460 {
461 DWORD bReturn = 0, szIn, szOut;
462 BOOL ioctlReturnStatus;
463 void *inParam = arg, *outParam = arg;
464
465 switch (ctlCode) {
466 case NETMAP_POLL:
467 szIn = sizeof(POLL_REQUEST_DATA);
468 szOut = sizeof(POLL_REQUEST_DATA);
469 break;
470 case NETMAP_MMAP:
471 szIn = 0;
472 szOut = sizeof(void*);
473 inParam = NULL; /* nothing on input */
474 break;
475 case NIOCTXSYNC:
476 case NIOCRXSYNC:
477 szIn = 0;
478 szOut = 0;
479 break;
480 case NIOCREGIF:
481 szIn = sizeof(struct nmreq);
482 szOut = sizeof(struct nmreq);
483 break;
484 case NIOCCONFIG:
485 D("unsupported NIOCCONFIG!");
486 return -1;
487
488 default: /* a regular ioctl */
489 D("invalid ioctl %x on netmap fd", ctlCode);
490 return -1;
491 }
492
493 ioctlReturnStatus = DeviceIoControl(h,
494 ctlCode, inParam, szIn,
495 outParam, szOut,
496 &bReturn, NULL);
497 // XXX note windows returns 0 on error or async call, 1 on success
498 // we could call GetLastError() to figure out what happened
499 return ioctlReturnStatus ? 0 : -1;
500 }
501
502 /*
503 * this function is what must be called from user-space programs
504 * same as ioctl, returns 0 on success and -1 on error
505 */
506 static int
507 win_nm_ioctl(int fd, int32_t ctlCode, void *arg)
508 {
509 HANDLE h = win_get_netmap_handle(fd);
510
511 if (h == NULL) {
512 return ioctl(fd, ctlCode, arg);
513 } else {
514 return win_nm_ioctl_internal(h, ctlCode, arg);
515 }
516 }
517
518 #define ioctl win_nm_ioctl /* from now on, within this file ... */
519
520 /*
521 * We cannot use the native mmap on windows
522 * The only parameter used is "fd", the other ones are just declared to
523 * make this signature comparable to the FreeBSD/Linux one
524 */
525 static void *
526 win32_mmap_emulated(void *addr, size_t length, int prot, int flags, int fd, int32_t offset)
527 {
528 HANDLE h = win_get_netmap_handle(fd);
529
530 if (h == NULL) {
531 return mmap(addr, length, prot, flags, fd, offset);
532 } else {
533 MEMORY_ENTRY ret;
534
535 return win_nm_ioctl_internal(h, NETMAP_MMAP, &ret) ?
536 NULL : ret.pUsermodeVirtualAddress;
537 }
538 }
539
540 #define mmap win32_mmap_emulated
541
542 #include <sys/poll.h> /* XXX needed to use the structure pollfd */
543
544 static int
545 win_nm_poll(struct pollfd *fds, int nfds, int timeout)
546 {
547 HANDLE h;
548
549 if (nfds != 1 || fds == NULL || (h = win_get_netmap_handle(fds->fd)) == NULL) {;
550 return poll(fds, nfds, timeout);
551 } else {
552 POLL_REQUEST_DATA prd;
553
554 prd.timeout = timeout;
555 prd.events = fds->events;
556
557 win_nm_ioctl_internal(h, NETMAP_POLL, &prd);
558 if ((prd.revents == POLLERR) || (prd.revents == STATUS_TIMEOUT)) {
559 return -1;
560 }
561 return 1;
562 }
563 }
564
565 #define poll win_nm_poll
566
567 static int
568 win_nm_open(char* pathname, int flags)
569 {
570
571 if (strcmp(pathname, NETMAP_DEVICE_NAME) == 0) {
572 int fd = open(NETMAP_DEVICE_NAME, O_RDWR);
573 if (fd < 0) {
574 return -1;
575 }
576
577 win_insert_fd_record(fd);
578 return fd;
579 } else {
580 return open(pathname, flags);
581 }
582 }
583
584 #define open win_nm_open
585
586 static int
587 win_nm_close(int fd)
588 {
589 if (fd != -1) {
590 close(fd);
591 if (win_get_netmap_handle(fd) != NULL) {
592 win_remove_fd_record(fd);
593 }
594 }
595 return 0;
596 }
597
598 #define close win_nm_close
599
600 #endif /* _WIN32 */
601
602 static int
603 nm_is_identifier(const char *s, const char *e)
604 {
605 for (; s != e; s++) {
606 if (!isalnum(*s) && *s != '_') {
607 return 0;
608 }
609 }
610
611 return 1;
612 }
613
614 #define MAXERRMSG 80
615 static int
616 nm_parse(const char *ifname, struct nm_desc *d, char *err)
617 {
618 int is_vale;
619 const char *port = NULL;
620 const char *vpname = NULL;
621 u_int namelen;
622 uint32_t nr_ringid = 0, nr_flags;
623 char errmsg[MAXERRMSG] = "";
624 long num;
625 uint16_t nr_arg2 = 0;
626 enum { P_START, P_RNGSFXOK, P_GETNUM, P_FLAGS, P_FLAGSOK, P_MEMID } p_state;
627
628 errno = 0;
629
630 is_vale = (ifname[0] == 'v');
631 if (is_vale) {
632 port = index(ifname, ':');
633 if (port == NULL) {
634 snprintf(errmsg, MAXERRMSG,
635 "missing ':' in vale name");
636 goto fail;
637 }
638
639 if (!nm_is_identifier(ifname + 4, port)) {
640 snprintf(errmsg, MAXERRMSG, "invalid bridge name");
641 goto fail;
642 }
643
644 vpname = ++port;
645 } else {
646 ifname += 7;
647 port = ifname;
648 }
649
650 /* scan for a separator */
651 for (; *port && !index("-*^{}/@", *port); port++)
652 ;
653
654 if (is_vale && !nm_is_identifier(vpname, port)) {
655 snprintf(errmsg, MAXERRMSG, "invalid bridge port name");
656 goto fail;
657 }
658
659 namelen = port - ifname;
660 if (namelen >= sizeof(d->req.nr_name)) {
661 snprintf(errmsg, MAXERRMSG, "name too long");
662 goto fail;
663 }
664 memcpy(d->req.nr_name, ifname, namelen);
665 d->req.nr_name[namelen] = '\0';
666
667 p_state = P_START;
668 nr_flags = NR_REG_ALL_NIC; /* default for no suffix */
669 while (*port) {
670 switch (p_state) {
671 case P_START:
672 switch (*port) {
673 case '^': /* only SW ring */
674 nr_flags = NR_REG_SW;
675 p_state = P_RNGSFXOK;
676 break;
677 case '*': /* NIC and SW */
678 nr_flags = NR_REG_NIC_SW;
679 p_state = P_RNGSFXOK;
680 break;
681 case '-': /* one NIC ring pair */
682 nr_flags = NR_REG_ONE_NIC;
683 p_state = P_GETNUM;
684 break;
685 case '{': /* pipe (master endpoint) */
686 nr_flags = NR_REG_PIPE_MASTER;
687 p_state = P_GETNUM;
688 break;
689 case '}': /* pipe (slave endoint) */
690 nr_flags = NR_REG_PIPE_SLAVE;
691 p_state = P_GETNUM;
692 break;
693 case '/': /* start of flags */
694 p_state = P_FLAGS;
695 break;
696 case '@': /* start of memid */
697 p_state = P_MEMID;
698 break;
699 default:
700 snprintf(errmsg, MAXERRMSG, "unknown modifier: '%c'", *port);
701 goto fail;
702 }
703 port++;
704 break;
705 case P_RNGSFXOK:
706 switch (*port) {
707 case '/':
708 p_state = P_FLAGS;
709 break;
710 case '@':
711 p_state = P_MEMID;
712 break;
713 default:
714 snprintf(errmsg, MAXERRMSG, "unexpected character: '%c'", *port);
715 goto fail;
716 }
717 port++;
718 break;
719 case P_GETNUM:
720 num = strtol(port, (char **)&port, 10);
721 if (num < 0 || num >= NETMAP_RING_MASK) {
722 snprintf(errmsg, MAXERRMSG, "'%ld' out of range [0, %d)",
723 num, NETMAP_RING_MASK);
724 goto fail;
725 }
726 nr_ringid = num & NETMAP_RING_MASK;
727 p_state = P_RNGSFXOK;
728 break;
729 case P_FLAGS:
730 case P_FLAGSOK:
731 if (*port == '@') {
732 port++;
733 p_state = P_MEMID;
734 break;
735 }
736 switch (*port) {
737 case 'x':
738 nr_flags |= NR_EXCLUSIVE;
739 break;
740 case 'z':
741 nr_flags |= NR_ZCOPY_MON;
742 break;
743 case 't':
744 nr_flags |= NR_MONITOR_TX;
745 break;
746 case 'r':
747 nr_flags |= NR_MONITOR_RX;
748 break;
749 case 'R':
750 nr_flags |= NR_RX_RINGS_ONLY;
751 break;
752 case 'T':
753 nr_flags |= NR_TX_RINGS_ONLY;
754 break;
755 default:
756 snprintf(errmsg, MAXERRMSG, "unrecognized flag: '%c'", *port);
757 goto fail;
758 }
759 port++;
760 p_state = P_FLAGSOK;
761 break;
762 case P_MEMID:
763 if (nr_arg2 != 0) {
764 snprintf(errmsg, MAXERRMSG, "double setting of memid");
765 goto fail;
766 }
767 num = strtol(port, (char **)&port, 10);
768 if (num <= 0) {
769 snprintf(errmsg, MAXERRMSG, "invalid memid %ld, must be >0", num);
770 goto fail;
771 }
772 nr_arg2 = num;
773 p_state = P_RNGSFXOK;
774 break;
775 }
776 }
777 if (p_state != P_START && p_state != P_RNGSFXOK && p_state != P_FLAGSOK) {
778 snprintf(errmsg, MAXERRMSG, "unexpected end of port name");
779 goto fail;
780 }
781 ND("flags: %s %s %s %s",
782 (nr_flags & NR_EXCLUSIVE) ? "EXCLUSIVE" : "",
783 (nr_flags & NR_ZCOPY_MON) ? "ZCOPY_MON" : "",
784 (nr_flags & NR_MONITOR_TX) ? "MONITOR_TX" : "",
785 (nr_flags & NR_MONITOR_RX) ? "MONITOR_RX" : "");
786
787 d->req.nr_flags |= nr_flags;
788 d->req.nr_ringid |= nr_ringid;
789 d->req.nr_arg2 = nr_arg2;
790
791 d->self = d;
792
793 return 0;
794 fail:
795 if (!errno)
796 errno = EINVAL;
797 if (err)
798 strncpy(err, errmsg, MAXERRMSG);
799 return -1;
800 }
801
802 /*
803 * Try to open, return descriptor if successful, NULL otherwise.
804 * An invalid netmap name will return errno = 0;
805 * You can pass a pointer to a pre-filled nm_desc to add special
806 * parameters. Flags is used as follows
807 * NM_OPEN_NO_MMAP use the memory from arg, only XXX avoid mmap
808 * if the nr_arg2 (memory block) matches.
809 * NM_OPEN_ARG1 use req.nr_arg1 from arg
810 * NM_OPEN_ARG2 use req.nr_arg2 from arg
811 * NM_OPEN_RING_CFG user ring config from arg
812 */
813 static struct nm_desc *
814 nm_open(const char *ifname, const struct nmreq *req,
815 uint64_t new_flags, const struct nm_desc *arg)
816 {
817 struct nm_desc *d = NULL;
818 const struct nm_desc *parent = arg;
819 char errmsg[MAXERRMSG] = "";
820 uint32_t nr_reg;
821
822 if (strncmp(ifname, "netmap:", 7) &&
823 strncmp(ifname, NM_BDG_NAME, strlen(NM_BDG_NAME))) {
824 errno = 0; /* name not recognised, not an error */
825 return NULL;
826 }
827
828 d = (struct nm_desc *)calloc(1, sizeof(*d));
829 if (d == NULL) {
830 snprintf(errmsg, MAXERRMSG, "nm_desc alloc failure");
831 errno = ENOMEM;
832 return NULL;
833 }
834 d->self = d; /* set this early so nm_close() works */
835 d->fd = open(NETMAP_DEVICE_NAME, O_RDWR);
836 if (d->fd < 0) {
837 snprintf(errmsg, MAXERRMSG, "cannot open /dev/netmap: %s", strerror(errno));
838 goto fail;
839 }
840
841 if (req)
842 d->req = *req;
843
844 if (!(new_flags & NM_OPEN_IFNAME)) {
845 if (nm_parse(ifname, d, errmsg) < 0)
846 goto fail;
847 }
848
849 d->req.nr_version = NETMAP_API;
850 d->req.nr_ringid &= NETMAP_RING_MASK;
851
852 /* optionally import info from parent */
853 if (IS_NETMAP_DESC(parent) && new_flags) {
854 if (new_flags & NM_OPEN_ARG1)
855 D("overriding ARG1 %d", parent->req.nr_arg1);
856 d->req.nr_arg1 = new_flags & NM_OPEN_ARG1 ?
857 parent->req.nr_arg1 : 4;
858 if (new_flags & NM_OPEN_ARG2) {
859 D("overriding ARG2 %d", parent->req.nr_arg2);
860 d->req.nr_arg2 = parent->req.nr_arg2;
861 }
862 if (new_flags & NM_OPEN_ARG3)
863 D("overriding ARG3 %d", parent->req.nr_arg3);
864 d->req.nr_arg3 = new_flags & NM_OPEN_ARG3 ?
865 parent->req.nr_arg3 : 0;
866 if (new_flags & NM_OPEN_RING_CFG) {
867 D("overriding RING_CFG");
868 d->req.nr_tx_slots = parent->req.nr_tx_slots;
869 d->req.nr_rx_slots = parent->req.nr_rx_slots;
870 d->req.nr_tx_rings = parent->req.nr_tx_rings;
871 d->req.nr_rx_rings = parent->req.nr_rx_rings;
872 }
873 if (new_flags & NM_OPEN_IFNAME) {
874 D("overriding ifname %s ringid 0x%x flags 0x%x",
875 parent->req.nr_name, parent->req.nr_ringid,
876 parent->req.nr_flags);
877 memcpy(d->req.nr_name, parent->req.nr_name,
878 sizeof(d->req.nr_name));
879 d->req.nr_ringid = parent->req.nr_ringid;
880 d->req.nr_flags = parent->req.nr_flags;
881 }
882 }
883 /* add the *XPOLL flags */
884 d->req.nr_ringid |= new_flags & (NETMAP_NO_TX_POLL | NETMAP_DO_RX_POLL);
885
886 if (ioctl(d->fd, NIOCREGIF, &d->req)) {
887 snprintf(errmsg, MAXERRMSG, "NIOCREGIF failed: %s", strerror(errno));
888 goto fail;
889 }
890
891 nr_reg = d->req.nr_flags & NR_REG_MASK;
892
893 if (nr_reg == NR_REG_SW) { /* host stack */
894 d->first_tx_ring = d->last_tx_ring = d->req.nr_tx_rings;
895 d->first_rx_ring = d->last_rx_ring = d->req.nr_rx_rings;
896 } else if (nr_reg == NR_REG_ALL_NIC) { /* only nic */
897 d->first_tx_ring = 0;
898 d->first_rx_ring = 0;
899 d->last_tx_ring = d->req.nr_tx_rings - 1;
900 d->last_rx_ring = d->req.nr_rx_rings - 1;
901 } else if (nr_reg == NR_REG_NIC_SW) {
902 d->first_tx_ring = 0;
903 d->first_rx_ring = 0;
904 d->last_tx_ring = d->req.nr_tx_rings;
905 d->last_rx_ring = d->req.nr_rx_rings;
906 } else if (nr_reg == NR_REG_ONE_NIC) {
907 /* XXX check validity */
908 d->first_tx_ring = d->last_tx_ring =
909 d->first_rx_ring = d->last_rx_ring = d->req.nr_ringid & NETMAP_RING_MASK;
910 } else { /* pipes */
911 d->first_tx_ring = d->last_tx_ring = 0;
912 d->first_rx_ring = d->last_rx_ring = 0;
913 }
914
915 /* if parent is defined, do nm_mmap() even if NM_OPEN_NO_MMAP is set */
916 if ((!(new_flags & NM_OPEN_NO_MMAP) || parent) && nm_mmap(d, parent)) {
917 snprintf(errmsg, MAXERRMSG, "mmap failed: %s", strerror(errno));
918 goto fail;
919 }
920
921
922 #ifdef DEBUG_NETMAP_USER
923 { /* debugging code */
924 int i;
925
926 D("%s tx %d .. %d %d rx %d .. %d %d", ifname,
927 d->first_tx_ring, d->last_tx_ring, d->req.nr_tx_rings,
928 d->first_rx_ring, d->last_rx_ring, d->req.nr_rx_rings);
929 for (i = 0; i <= d->req.nr_tx_rings; i++) {
930 struct netmap_ring *r = NETMAP_TXRING(d->nifp, i);
931 D("TX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
932 }
933 for (i = 0; i <= d->req.nr_rx_rings; i++) {
934 struct netmap_ring *r = NETMAP_RXRING(d->nifp, i);
935 D("RX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
936 }
937 }
938 #endif /* debugging */
939
940 d->cur_tx_ring = d->first_tx_ring;
941 d->cur_rx_ring = d->first_rx_ring;
942 return d;
943
944 fail:
945 nm_close(d);
946 if (errmsg[0])
947 D("%s %s", errmsg, ifname);
948 if (errno == 0)
949 errno = EINVAL;
950 return NULL;
951 }
952
953
954 static int
955 nm_close(struct nm_desc *d)
956 {
957 /*
958 * ugly trick to avoid unused warnings
959 */
960 static void *__xxzt[] __attribute__ ((unused)) =
961 { (void *)nm_open, (void *)nm_inject,
962 (void *)nm_dispatch, (void *)nm_nextpkt } ;
963
964 if (d == NULL || d->self != d)
965 return EINVAL;
966 if (d->done_mmap && d->mem)
967 munmap(d->mem, d->memsize);
968 if (d->fd != -1) {
969 close(d->fd);
970 }
971
972 bzero(d, sizeof(*d));
973 free(d);
974 return 0;
975 }
976
977
978 static int
979 nm_mmap(struct nm_desc *d, const struct nm_desc *parent)
980 {
981 //XXX TODO: check if mmap is already done
982
983 if (IS_NETMAP_DESC(parent) && parent->mem &&
984 parent->req.nr_arg2 == d->req.nr_arg2) {
985 /* do not mmap, inherit from parent */
986 D("do not mmap, inherit from parent");
987 d->memsize = parent->memsize;
988 d->mem = parent->mem;
989 } else {
990 /* XXX TODO: check if memsize is too large (or there is overflow) */
991 d->memsize = d->req.nr_memsize;
992 d->mem = mmap(0, d->memsize, PROT_WRITE | PROT_READ, MAP_SHARED,
993 d->fd, 0);
994 if (d->mem == MAP_FAILED) {
995 goto fail;
996 }
997 d->done_mmap = 1;
998 }
999 {
1000 struct netmap_if *nifp = NETMAP_IF(d->mem, d->req.nr_offset);
1001 struct netmap_ring *r = NETMAP_RXRING(nifp, d->first_rx_ring);
1002 if ((void *)r == (void *)nifp) {
1003 /* the descriptor is open for TX only */
1004 r = NETMAP_TXRING(nifp, d->first_tx_ring);
1005 }
1006
1007 *(struct netmap_if **)(uintptr_t)&(d->nifp) = nifp;
1008 *(struct netmap_ring **)(uintptr_t)&d->some_ring = r;
1009 *(void **)(uintptr_t)&d->buf_start = NETMAP_BUF(r, 0);
1010 *(void **)(uintptr_t)&d->buf_end =
1011 (char *)d->mem + d->memsize;
1012 }
1013
1014 return 0;
1015
1016 fail:
1017 return EINVAL;
1018 }
1019
1020 /*
1021 * Same prototype as pcap_inject(), only need to cast.
1022 */
1023 static int
1024 nm_inject(struct nm_desc *d, const void *buf, size_t size)
1025 {
1026 u_int c, n = d->last_tx_ring - d->first_tx_ring + 1,
1027 ri = d->cur_tx_ring;
1028
1029 for (c = 0; c < n ; c++, ri++) {
1030 /* compute current ring to use */
1031 struct netmap_ring *ring;
1032 uint32_t i, j, idx;
1033 size_t rem;
1034
1035 if (ri > d->last_tx_ring)
1036 ri = d->first_tx_ring;
1037 ring = NETMAP_TXRING(d->nifp, ri);
1038 rem = size;
1039 j = ring->cur;
1040 while (rem > ring->nr_buf_size && j != ring->tail) {
1041 rem -= ring->nr_buf_size;
1042 j = nm_ring_next(ring, j);
1043 }
1044 if (j == ring->tail && rem > 0)
1045 continue;
1046 i = ring->cur;
1047 while (i != j) {
1048 idx = ring->slot[i].buf_idx;
1049 ring->slot[i].len = ring->nr_buf_size;
1050 ring->slot[i].flags = NS_MOREFRAG;
1051 nm_pkt_copy(buf, NETMAP_BUF(ring, idx), ring->nr_buf_size);
1052 i = nm_ring_next(ring, i);
1053 buf = (char *)buf + ring->nr_buf_size;
1054 }
1055 idx = ring->slot[i].buf_idx;
1056 ring->slot[i].len = rem;
1057 ring->slot[i].flags = 0;
1058 nm_pkt_copy(buf, NETMAP_BUF(ring, idx), rem);
1059 ring->head = ring->cur = nm_ring_next(ring, i);
1060 d->cur_tx_ring = ri;
1061 return size;
1062 }
1063 return 0; /* fail */
1064 }
1065
1066
1067 /*
1068 * Same prototype as pcap_dispatch(), only need to cast.
1069 */
1070 static int
1071 nm_dispatch(struct nm_desc *d, int cnt, nm_cb_t cb, u_char *arg)
1072 {
1073 int n = d->last_rx_ring - d->first_rx_ring + 1;
1074 int c, got = 0, ri = d->cur_rx_ring;
1075 d->hdr.buf = NULL;
1076 d->hdr.flags = NM_MORE_PKTS;
1077 d->hdr.d = d;
1078
1079 if (cnt == 0)
1080 cnt = -1;
1081 /* cnt == -1 means infinite, but rings have a finite amount
1082 * of buffers and the int is large enough that we never wrap,
1083 * so we can omit checking for -1
1084 */
1085 for (c=0; c < n && cnt != got; c++, ri++) {
1086 /* compute current ring to use */
1087 struct netmap_ring *ring;
1088
1089 if (ri > d->last_rx_ring)
1090 ri = d->first_rx_ring;
1091 ring = NETMAP_RXRING(d->nifp, ri);
1092 for ( ; !nm_ring_empty(ring) && cnt != got; got++) {
1093 u_int idx, i;
1094 if (d->hdr.buf) { /* from previous round */
1095 cb(arg, &d->hdr, d->hdr.buf);
1096 }
1097 i = ring->cur;
1098 idx = ring->slot[i].buf_idx;
1099 /* d->cur_rx_ring doesn't change inside this loop, but
1100 * set it here, so it reflects d->hdr.buf's ring */
1101 d->cur_rx_ring = ri;
1102 d->hdr.slot = &ring->slot[i];
1103 d->hdr.buf = (u_char *)NETMAP_BUF(ring, idx);
1104 // __builtin_prefetch(buf);
1105 d->hdr.len = d->hdr.caplen = ring->slot[i].len;
1106 d->hdr.ts = ring->ts;
1107 ring->head = ring->cur = nm_ring_next(ring, i);
1108 }
1109 }
1110 if (d->hdr.buf) { /* from previous round */
1111 d->hdr.flags = 0;
1112 cb(arg, &d->hdr, d->hdr.buf);
1113 }
1114 return got;
1115 }
1116
1117 static u_char *
1118 nm_nextpkt(struct nm_desc *d, struct nm_pkthdr *hdr)
1119 {
1120 int ri = d->cur_rx_ring;
1121
1122 do {
1123 /* compute current ring to use */
1124 struct netmap_ring *ring = NETMAP_RXRING(d->nifp, ri);
1125 if (!nm_ring_empty(ring)) {
1126 u_int i = ring->cur;
1127 u_int idx = ring->slot[i].buf_idx;
1128 u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
1129
1130 // __builtin_prefetch(buf);
1131 hdr->ts = ring->ts;
1132 hdr->len = hdr->caplen = ring->slot[i].len;
1133 ring->cur = nm_ring_next(ring, i);
1134 /* we could postpone advancing head if we want
1135 * to hold the buffer. This can be supported in
1136 * the future.
1137 */
1138 ring->head = ring->cur;
1139 d->cur_rx_ring = ri;
1140 return buf;
1141 }
1142 ri++;
1143 if (ri > d->last_rx_ring)
1144 ri = d->first_rx_ring;
1145 } while (ri != d->cur_rx_ring);
1146 return NULL; /* nothing found */
1147 }
1148
1149 #endif /* !HAVE_NETMAP_WITH_LIBS */
1150
1151 #endif /* NETMAP_WITH_LIBS */
1152
1153 #endif /* _NET_NETMAP_USER_H_ */
Cache object: b24a3a5be80d51aeb6642c37729d14d5
|