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: stable/12/sys/net/netmap_user.h 366496 2020-10-06 19:11:55Z 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 #include <string.h> /* memset */
97 #include <sys/time.h> /* gettimeofday */
98
99 #ifndef likely
100 #define likely(x) __builtin_expect(!!(x), 1)
101 #define unlikely(x) __builtin_expect(!!(x), 0)
102 #endif /* likely and unlikely */
103
104 #include <net/netmap.h>
105
106 /* helper macro */
107 #define _NETMAP_OFFSET(type, ptr, offset) \
108 ((type)(void *)((char *)(ptr) + (offset)))
109
110 #define NETMAP_IF(_base, _ofs) _NETMAP_OFFSET(struct netmap_if *, _base, _ofs)
111
112 #define NETMAP_TXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
113 nifp, (nifp)->ring_ofs[index] )
114
115 #define NETMAP_RXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
116 nifp, (nifp)->ring_ofs[index + (nifp)->ni_tx_rings + \
117 (nifp)->ni_host_tx_rings] )
118
119 #define NETMAP_BUF(ring, index) \
120 ((char *)(ring) + (ring)->buf_ofs + ((size_t)(index)*(ring)->nr_buf_size))
121
122 #define NETMAP_BUF_IDX(ring, buf) \
123 ( ((char *)(buf) - ((char *)(ring) + (ring)->buf_ofs) ) / \
124 (ring)->nr_buf_size )
125
126
127 static inline uint32_t
nm_ring_next(struct netmap_ring * r,uint32_t i)128 nm_ring_next(struct netmap_ring *r, uint32_t i)
129 {
130 return ( unlikely(i + 1 == r->num_slots) ? 0 : i + 1);
131 }
132
133
134 /*
135 * Return 1 if we have pending transmissions in the tx ring.
136 * When everything is complete ring->head = ring->tail + 1 (modulo ring size)
137 */
138 static inline int
nm_tx_pending(struct netmap_ring * r)139 nm_tx_pending(struct netmap_ring *r)
140 {
141 return nm_ring_next(r, r->tail) != r->head;
142 }
143
144 /* Compute the number of slots available in the netmap ring. We use
145 * ring->head as explained in the comment above nm_ring_empty(). */
146 static inline uint32_t
nm_ring_space(struct netmap_ring * ring)147 nm_ring_space(struct netmap_ring *ring)
148 {
149 int ret = ring->tail - ring->head;
150 if (ret < 0)
151 ret += ring->num_slots;
152 return ret;
153 }
154
155 #ifndef ND /* debug macros */
156 /* debug support */
157 #define ND(_fmt, ...) do {} while(0)
158 #define D(_fmt, ...) \
159 do { \
160 struct timeval _t0; \
161 gettimeofday(&_t0, NULL); \
162 fprintf(stderr, "%03d.%06d %s [%d] " _fmt "\n", \
163 (int)(_t0.tv_sec % 1000), (int)_t0.tv_usec, \
164 __FUNCTION__, __LINE__, ##__VA_ARGS__); \
165 } while (0)
166
167 /* Rate limited version of "D", lps indicates how many per second */
168 #define RD(lps, format, ...) \
169 do { \
170 static int __t0, __cnt; \
171 struct timeval __xxts; \
172 gettimeofday(&__xxts, NULL); \
173 if (__t0 != __xxts.tv_sec) { \
174 __t0 = __xxts.tv_sec; \
175 __cnt = 0; \
176 } \
177 if (__cnt++ < lps) { \
178 D(format, ##__VA_ARGS__); \
179 } \
180 } while (0)
181 #endif
182
183 /*
184 * this is a slightly optimized copy routine which rounds
185 * to multiple of 64 bytes and is often faster than dealing
186 * with other odd sizes. We assume there is enough room
187 * in the source and destination buffers.
188 */
189 static inline void
nm_pkt_copy(const void * _src,void * _dst,int l)190 nm_pkt_copy(const void *_src, void *_dst, int l)
191 {
192 const uint64_t *src = (const uint64_t *)_src;
193 uint64_t *dst = (uint64_t *)_dst;
194
195 if (unlikely(l >= 1024 || l % 64)) {
196 memcpy(dst, src, l);
197 return;
198 }
199 for (; likely(l > 0); l-=64) {
200 *dst++ = *src++;
201 *dst++ = *src++;
202 *dst++ = *src++;
203 *dst++ = *src++;
204 *dst++ = *src++;
205 *dst++ = *src++;
206 *dst++ = *src++;
207 *dst++ = *src++;
208 }
209 }
210
211 #ifdef NETMAP_WITH_LIBS
212 /*
213 * Support for simple I/O libraries.
214 * Include other system headers required for compiling this.
215 */
216
217 #ifndef HAVE_NETMAP_WITH_LIBS
218 #define HAVE_NETMAP_WITH_LIBS
219
220 #include <stdio.h>
221 #include <sys/time.h>
222 #include <sys/mman.h>
223 #include <sys/ioctl.h>
224 #include <sys/errno.h> /* EINVAL */
225 #include <fcntl.h> /* O_RDWR */
226 #include <unistd.h> /* close() */
227 #include <signal.h>
228 #include <stdlib.h>
229
230 struct nm_pkthdr { /* first part is the same as pcap_pkthdr */
231 struct timeval ts;
232 uint32_t caplen;
233 uint32_t len;
234
235 uint64_t flags; /* NM_MORE_PKTS etc */
236 #define NM_MORE_PKTS 1
237 struct nm_desc *d;
238 struct netmap_slot *slot;
239 uint8_t *buf;
240 };
241
242 struct nm_stat { /* same as pcap_stat */
243 u_int ps_recv;
244 u_int ps_drop;
245 u_int ps_ifdrop;
246 #ifdef WIN32 /* XXX or _WIN32 ? */
247 u_int bs_capt;
248 #endif /* WIN32 */
249 };
250
251 #define NM_ERRBUF_SIZE 512
252
253 struct nm_desc {
254 struct nm_desc *self; /* point to self if netmap. */
255 int fd;
256 void *mem;
257 size_t memsize;
258 int done_mmap; /* set if mem is the result of mmap */
259 struct netmap_if * const nifp;
260 uint16_t first_tx_ring, last_tx_ring, cur_tx_ring;
261 uint16_t first_rx_ring, last_rx_ring, cur_rx_ring;
262 struct nmreq req; /* also contains the nr_name = ifname */
263 struct nm_pkthdr hdr;
264
265 /*
266 * The memory contains netmap_if, rings and then buffers.
267 * Given a pointer (e.g. to nm_inject) we can compare with
268 * mem/buf_start/buf_end to tell if it is a buffer or
269 * some other descriptor in our region.
270 * We also store a pointer to some ring as it helps in the
271 * translation from buffer indexes to addresses.
272 */
273 struct netmap_ring * const some_ring;
274 void * const buf_start;
275 void * const buf_end;
276 /* parameters from pcap_open_live */
277 int snaplen;
278 int promisc;
279 int to_ms;
280 char *errbuf;
281
282 /* save flags so we can restore them on close */
283 uint32_t if_flags;
284 uint32_t if_reqcap;
285 uint32_t if_curcap;
286
287 struct nm_stat st;
288 char msg[NM_ERRBUF_SIZE];
289 };
290
291 /*
292 * when the descriptor is open correctly, d->self == d
293 * Eventually we should also use some magic number.
294 */
295 #define P2NMD(p) ((const struct nm_desc *)(p))
296 #define IS_NETMAP_DESC(d) ((d) && P2NMD(d)->self == P2NMD(d))
297 #define NETMAP_FD(d) (P2NMD(d)->fd)
298
299
300
301
302 /*
303 * The callback, invoked on each received packet. Same as libpcap
304 */
305 typedef void (*nm_cb_t)(u_char *, const struct nm_pkthdr *, const u_char *d);
306
307 /*
308 *--- the pcap-like API ---
309 *
310 * nm_open() opens a file descriptor, binds to a port and maps memory.
311 *
312 * ifname (netmap:foo or vale:foo) is the port name
313 * a suffix can indicate the follwing:
314 * ^ bind the host (sw) ring pair
315 * * bind host and NIC ring pairs
316 * -NN bind individual NIC ring pair
317 * {NN bind master side of pipe NN
318 * }NN bind slave side of pipe NN
319 * a suffix starting with / and the following flags,
320 * in any order:
321 * x exclusive access
322 * z zero copy monitor (both tx and rx)
323 * t monitor tx side (copy monitor)
324 * r monitor rx side (copy monitor)
325 * R bind only RX ring(s)
326 * T bind only TX ring(s)
327 *
328 * req provides the initial values of nmreq before parsing ifname.
329 * Remember that the ifname parsing will override the ring
330 * number in nm_ringid, and part of nm_flags;
331 * flags special functions, normally 0
332 * indicates which fields of *arg are significant
333 * arg special functions, normally NULL
334 * if passed a netmap_desc with mem != NULL,
335 * use that memory instead of mmap.
336 */
337
338 static struct nm_desc *nm_open(const char *ifname, const struct nmreq *req,
339 uint64_t flags, const struct nm_desc *arg);
340
341 /*
342 * nm_open can import some fields from the parent descriptor.
343 * These flags control which ones.
344 * Also in flags you can specify NETMAP_NO_TX_POLL and NETMAP_DO_RX_POLL,
345 * which set the initial value for these flags.
346 * Note that the 16 low bits of the flags are reserved for data
347 * that may go into the nmreq.
348 */
349 enum {
350 NM_OPEN_NO_MMAP = 0x040000, /* reuse mmap from parent */
351 NM_OPEN_IFNAME = 0x080000, /* nr_name, nr_ringid, nr_flags */
352 NM_OPEN_ARG1 = 0x100000,
353 NM_OPEN_ARG2 = 0x200000,
354 NM_OPEN_ARG3 = 0x400000,
355 NM_OPEN_RING_CFG = 0x800000, /* tx|rx rings|slots */
356 };
357
358
359 /*
360 * nm_close() closes and restores the port to its previous state
361 */
362
363 static int nm_close(struct nm_desc *);
364
365 /*
366 * nm_mmap() do mmap or inherit from parent if the nr_arg2
367 * (memory block) matches.
368 */
369
370 static int nm_mmap(struct nm_desc *, const struct nm_desc *);
371
372 /*
373 * nm_inject() is the same as pcap_inject()
374 * nm_dispatch() is the same as pcap_dispatch()
375 * nm_nextpkt() is the same as pcap_next()
376 */
377
378 static int nm_inject(struct nm_desc *, const void *, size_t);
379 static int nm_dispatch(struct nm_desc *, int, nm_cb_t, u_char *);
380 static u_char *nm_nextpkt(struct nm_desc *, struct nm_pkthdr *);
381
382 #ifdef _WIN32
383
384 intptr_t _get_osfhandle(int); /* defined in io.h in windows */
385
386 /*
387 * In windows we do not have yet native poll support, so we keep track
388 * of file descriptors associated to netmap ports to emulate poll on
389 * them and fall back on regular poll on other file descriptors.
390 */
391 struct win_netmap_fd_list {
392 struct win_netmap_fd_list *next;
393 int win_netmap_fd;
394 HANDLE win_netmap_handle;
395 };
396
397 /*
398 * list head containing all the netmap opened fd and their
399 * windows HANDLE counterparts
400 */
401 static struct win_netmap_fd_list *win_netmap_fd_list_head;
402
403 static void
win_insert_fd_record(int fd)404 win_insert_fd_record(int fd)
405 {
406 struct win_netmap_fd_list *curr;
407
408 for (curr = win_netmap_fd_list_head; curr; curr = curr->next) {
409 if (fd == curr->win_netmap_fd) {
410 return;
411 }
412 }
413 curr = calloc(1, sizeof(*curr));
414 curr->next = win_netmap_fd_list_head;
415 curr->win_netmap_fd = fd;
416 curr->win_netmap_handle = IntToPtr(_get_osfhandle(fd));
417 win_netmap_fd_list_head = curr;
418 }
419
420 void
win_remove_fd_record(int fd)421 win_remove_fd_record(int fd)
422 {
423 struct win_netmap_fd_list *curr = win_netmap_fd_list_head;
424 struct win_netmap_fd_list *prev = NULL;
425 for (; curr ; prev = curr, curr = curr->next) {
426 if (fd != curr->win_netmap_fd)
427 continue;
428 /* found the entry */
429 if (prev == NULL) { /* we are freeing the first entry */
430 win_netmap_fd_list_head = curr->next;
431 } else {
432 prev->next = curr->next;
433 }
434 free(curr);
435 break;
436 }
437 }
438
439
440 HANDLE
win_get_netmap_handle(int fd)441 win_get_netmap_handle(int fd)
442 {
443 struct win_netmap_fd_list *curr;
444
445 for (curr = win_netmap_fd_list_head; curr; curr = curr->next) {
446 if (fd == curr->win_netmap_fd) {
447 return curr->win_netmap_handle;
448 }
449 }
450 return NULL;
451 }
452
453 /*
454 * we need to wrap ioctl and mmap, at least for the netmap file descriptors
455 */
456
457 /*
458 * use this function only from netmap_user.h internal functions
459 * same as ioctl, returns 0 on success and -1 on error
460 */
461 static int
win_nm_ioctl_internal(HANDLE h,int32_t ctlCode,void * arg)462 win_nm_ioctl_internal(HANDLE h, int32_t ctlCode, void *arg)
463 {
464 DWORD bReturn = 0, szIn, szOut;
465 BOOL ioctlReturnStatus;
466 void *inParam = arg, *outParam = arg;
467
468 switch (ctlCode) {
469 case NETMAP_POLL:
470 szIn = sizeof(POLL_REQUEST_DATA);
471 szOut = sizeof(POLL_REQUEST_DATA);
472 break;
473 case NETMAP_MMAP:
474 szIn = 0;
475 szOut = sizeof(void*);
476 inParam = NULL; /* nothing on input */
477 break;
478 case NIOCTXSYNC:
479 case NIOCRXSYNC:
480 szIn = 0;
481 szOut = 0;
482 break;
483 case NIOCREGIF:
484 szIn = sizeof(struct nmreq);
485 szOut = sizeof(struct nmreq);
486 break;
487 case NIOCCONFIG:
488 D("unsupported NIOCCONFIG!");
489 return -1;
490
491 default: /* a regular ioctl */
492 D("invalid ioctl %x on netmap fd", ctlCode);
493 return -1;
494 }
495
496 ioctlReturnStatus = DeviceIoControl(h,
497 ctlCode, inParam, szIn,
498 outParam, szOut,
499 &bReturn, NULL);
500 // XXX note windows returns 0 on error or async call, 1 on success
501 // we could call GetLastError() to figure out what happened
502 return ioctlReturnStatus ? 0 : -1;
503 }
504
505 /*
506 * this function is what must be called from user-space programs
507 * same as ioctl, returns 0 on success and -1 on error
508 */
509 static int
win_nm_ioctl(int fd,int32_t ctlCode,void * arg)510 win_nm_ioctl(int fd, int32_t ctlCode, void *arg)
511 {
512 HANDLE h = win_get_netmap_handle(fd);
513
514 if (h == NULL) {
515 return ioctl(fd, ctlCode, arg);
516 } else {
517 return win_nm_ioctl_internal(h, ctlCode, arg);
518 }
519 }
520
521 #define ioctl win_nm_ioctl /* from now on, within this file ... */
522
523 /*
524 * We cannot use the native mmap on windows
525 * The only parameter used is "fd", the other ones are just declared to
526 * make this signature comparable to the FreeBSD/Linux one
527 */
528 static void *
win32_mmap_emulated(void * addr,size_t length,int prot,int flags,int fd,int32_t offset)529 win32_mmap_emulated(void *addr, size_t length, int prot, int flags, int fd, int32_t offset)
530 {
531 HANDLE h = win_get_netmap_handle(fd);
532
533 if (h == NULL) {
534 return mmap(addr, length, prot, flags, fd, offset);
535 } else {
536 MEMORY_ENTRY ret;
537
538 return win_nm_ioctl_internal(h, NETMAP_MMAP, &ret) ?
539 NULL : ret.pUsermodeVirtualAddress;
540 }
541 }
542
543 #define mmap win32_mmap_emulated
544
545 #include <sys/poll.h> /* XXX needed to use the structure pollfd */
546
547 static int
win_nm_poll(struct pollfd * fds,int nfds,int timeout)548 win_nm_poll(struct pollfd *fds, int nfds, int timeout)
549 {
550 HANDLE h;
551
552 if (nfds != 1 || fds == NULL || (h = win_get_netmap_handle(fds->fd)) == NULL) {;
553 return poll(fds, nfds, timeout);
554 } else {
555 POLL_REQUEST_DATA prd;
556
557 prd.timeout = timeout;
558 prd.events = fds->events;
559
560 win_nm_ioctl_internal(h, NETMAP_POLL, &prd);
561 if ((prd.revents == POLLERR) || (prd.revents == STATUS_TIMEOUT)) {
562 return -1;
563 }
564 return 1;
565 }
566 }
567
568 #define poll win_nm_poll
569
570 static int
win_nm_open(char * pathname,int flags)571 win_nm_open(char* pathname, int flags)
572 {
573
574 if (strcmp(pathname, NETMAP_DEVICE_NAME) == 0) {
575 int fd = open(NETMAP_DEVICE_NAME, O_RDWR);
576 if (fd < 0) {
577 return -1;
578 }
579
580 win_insert_fd_record(fd);
581 return fd;
582 } else {
583 return open(pathname, flags);
584 }
585 }
586
587 #define open win_nm_open
588
589 static int
win_nm_close(int fd)590 win_nm_close(int fd)
591 {
592 if (fd != -1) {
593 close(fd);
594 if (win_get_netmap_handle(fd) != NULL) {
595 win_remove_fd_record(fd);
596 }
597 }
598 return 0;
599 }
600
601 #define close win_nm_close
602
603 #endif /* _WIN32 */
604
605 static int
nm_is_identifier(const char * s,const char * e)606 nm_is_identifier(const char *s, const char *e)
607 {
608 for (; s != e; s++) {
609 if (!isalnum(*s) && *s != '_') {
610 return 0;
611 }
612 }
613
614 return 1;
615 }
616
617 #define MAXERRMSG 80
618 static int
nm_parse(const char * ifname,struct nm_desc * d,char * err)619 nm_parse(const char *ifname, struct nm_desc *d, char *err)
620 {
621 int is_vale;
622 const char *port = NULL;
623 const char *vpname = NULL;
624 u_int namelen;
625 uint32_t nr_ringid = 0, nr_flags;
626 char errmsg[MAXERRMSG] = "", *tmp;
627 long num;
628 uint16_t nr_arg2 = 0;
629 enum { P_START, P_RNGSFXOK, P_GETNUM, P_FLAGS, P_FLAGSOK, P_MEMID } p_state;
630
631 errno = 0;
632
633 is_vale = (ifname[0] == 'v');
634 if (is_vale) {
635 port = index(ifname, ':');
636 if (port == NULL) {
637 snprintf(errmsg, MAXERRMSG,
638 "missing ':' in vale name");
639 goto fail;
640 }
641
642 if (!nm_is_identifier(ifname + 4, port)) {
643 snprintf(errmsg, MAXERRMSG, "invalid bridge name");
644 goto fail;
645 }
646
647 vpname = ++port;
648 } else {
649 ifname += 7;
650 port = ifname;
651 }
652
653 /* scan for a separator */
654 for (; *port && !index("-*^{}/@", *port); port++)
655 ;
656
657 if (is_vale && !nm_is_identifier(vpname, port)) {
658 snprintf(errmsg, MAXERRMSG, "invalid bridge port name");
659 goto fail;
660 }
661
662 namelen = port - ifname;
663 if (namelen >= sizeof(d->req.nr_name)) {
664 snprintf(errmsg, MAXERRMSG, "name too long");
665 goto fail;
666 }
667 memcpy(d->req.nr_name, ifname, namelen);
668 d->req.nr_name[namelen] = '\0';
669
670 p_state = P_START;
671 nr_flags = NR_REG_ALL_NIC; /* default for no suffix */
672 while (*port) {
673 switch (p_state) {
674 case P_START:
675 switch (*port) {
676 case '^': /* only SW ring */
677 nr_flags = NR_REG_SW;
678 p_state = P_RNGSFXOK;
679 break;
680 case '*': /* NIC and SW */
681 nr_flags = NR_REG_NIC_SW;
682 p_state = P_RNGSFXOK;
683 break;
684 case '-': /* one NIC ring pair */
685 nr_flags = NR_REG_ONE_NIC;
686 p_state = P_GETNUM;
687 break;
688 case '{': /* pipe (master endpoint) */
689 nr_flags = NR_REG_PIPE_MASTER;
690 p_state = P_GETNUM;
691 break;
692 case '}': /* pipe (slave endoint) */
693 nr_flags = NR_REG_PIPE_SLAVE;
694 p_state = P_GETNUM;
695 break;
696 case '/': /* start of flags */
697 p_state = P_FLAGS;
698 break;
699 case '@': /* start of memid */
700 p_state = P_MEMID;
701 break;
702 default:
703 snprintf(errmsg, MAXERRMSG, "unknown modifier: '%c'", *port);
704 goto fail;
705 }
706 port++;
707 break;
708 case P_RNGSFXOK:
709 switch (*port) {
710 case '/':
711 p_state = P_FLAGS;
712 break;
713 case '@':
714 p_state = P_MEMID;
715 break;
716 default:
717 snprintf(errmsg, MAXERRMSG, "unexpected character: '%c'", *port);
718 goto fail;
719 }
720 port++;
721 break;
722 case P_GETNUM:
723 num = strtol(port, &tmp, 10);
724 if (num < 0 || num >= NETMAP_RING_MASK) {
725 snprintf(errmsg, MAXERRMSG, "'%ld' out of range [0, %d)",
726 num, NETMAP_RING_MASK);
727 goto fail;
728 }
729 port = tmp;
730 nr_ringid = num & NETMAP_RING_MASK;
731 p_state = P_RNGSFXOK;
732 break;
733 case P_FLAGS:
734 case P_FLAGSOK:
735 if (*port == '@') {
736 port++;
737 p_state = P_MEMID;
738 break;
739 }
740 switch (*port) {
741 case 'x':
742 nr_flags |= NR_EXCLUSIVE;
743 break;
744 case 'z':
745 nr_flags |= NR_ZCOPY_MON;
746 break;
747 case 't':
748 nr_flags |= NR_MONITOR_TX;
749 break;
750 case 'r':
751 nr_flags |= NR_MONITOR_RX;
752 break;
753 case 'R':
754 nr_flags |= NR_RX_RINGS_ONLY;
755 break;
756 case 'T':
757 nr_flags |= NR_TX_RINGS_ONLY;
758 break;
759 default:
760 snprintf(errmsg, MAXERRMSG, "unrecognized flag: '%c'", *port);
761 goto fail;
762 }
763 port++;
764 p_state = P_FLAGSOK;
765 break;
766 case P_MEMID:
767 if (nr_arg2 != 0) {
768 snprintf(errmsg, MAXERRMSG, "double setting of memid");
769 goto fail;
770 }
771 num = strtol(port, &tmp, 10);
772 if (num <= 0) {
773 snprintf(errmsg, MAXERRMSG, "invalid memid %ld, must be >0", num);
774 goto fail;
775 }
776 port = tmp;
777 nr_arg2 = num;
778 p_state = P_RNGSFXOK;
779 break;
780 }
781 }
782 if (p_state != P_START && p_state != P_RNGSFXOK && p_state != P_FLAGSOK) {
783 snprintf(errmsg, MAXERRMSG, "unexpected end of port name");
784 goto fail;
785 }
786 ND("flags: %s %s %s %s",
787 (nr_flags & NR_EXCLUSIVE) ? "EXCLUSIVE" : "",
788 (nr_flags & NR_ZCOPY_MON) ? "ZCOPY_MON" : "",
789 (nr_flags & NR_MONITOR_TX) ? "MONITOR_TX" : "",
790 (nr_flags & NR_MONITOR_RX) ? "MONITOR_RX" : "");
791
792 d->req.nr_flags |= nr_flags;
793 d->req.nr_ringid |= nr_ringid;
794 d->req.nr_arg2 = nr_arg2;
795
796 d->self = d;
797
798 return 0;
799 fail:
800 if (!errno)
801 errno = EINVAL;
802 if (err)
803 strncpy(err, errmsg, MAXERRMSG);
804 return -1;
805 }
806
807 /*
808 * Try to open, return descriptor if successful, NULL otherwise.
809 * An invalid netmap name will return errno = 0;
810 * You can pass a pointer to a pre-filled nm_desc to add special
811 * parameters. Flags is used as follows
812 * NM_OPEN_NO_MMAP use the memory from arg, only XXX avoid mmap
813 * if the nr_arg2 (memory block) matches.
814 * NM_OPEN_ARG1 use req.nr_arg1 from arg
815 * NM_OPEN_ARG2 use req.nr_arg2 from arg
816 * NM_OPEN_RING_CFG user ring config from arg
817 */
818 static struct nm_desc *
nm_open(const char * ifname,const struct nmreq * req,uint64_t new_flags,const struct nm_desc * arg)819 nm_open(const char *ifname, const struct nmreq *req,
820 uint64_t new_flags, const struct nm_desc *arg)
821 {
822 struct nm_desc *d = NULL;
823 const struct nm_desc *parent = arg;
824 char errmsg[MAXERRMSG] = "";
825 uint32_t nr_reg;
826
827 if (strncmp(ifname, "netmap:", 7) &&
828 strncmp(ifname, NM_BDG_NAME, strlen(NM_BDG_NAME))) {
829 errno = 0; /* name not recognised, not an error */
830 return NULL;
831 }
832
833 d = (struct nm_desc *)calloc(1, sizeof(*d));
834 if (d == NULL) {
835 snprintf(errmsg, MAXERRMSG, "nm_desc alloc failure");
836 errno = ENOMEM;
837 return NULL;
838 }
839 d->self = d; /* set this early so nm_close() works */
840 d->fd = open(NETMAP_DEVICE_NAME, O_RDWR);
841 if (d->fd < 0) {
842 snprintf(errmsg, MAXERRMSG, "cannot open /dev/netmap: %s", strerror(errno));
843 goto fail;
844 }
845
846 if (req)
847 d->req = *req;
848
849 if (!(new_flags & NM_OPEN_IFNAME)) {
850 if (nm_parse(ifname, d, errmsg) < 0)
851 goto fail;
852 }
853
854 d->req.nr_version = NETMAP_API;
855 d->req.nr_ringid &= NETMAP_RING_MASK;
856
857 /* optionally import info from parent */
858 if (IS_NETMAP_DESC(parent) && new_flags) {
859 if (new_flags & NM_OPEN_ARG1)
860 D("overriding ARG1 %d", parent->req.nr_arg1);
861 d->req.nr_arg1 = new_flags & NM_OPEN_ARG1 ?
862 parent->req.nr_arg1 : 4;
863 if (new_flags & NM_OPEN_ARG2) {
864 D("overriding ARG2 %d", parent->req.nr_arg2);
865 d->req.nr_arg2 = parent->req.nr_arg2;
866 }
867 if (new_flags & NM_OPEN_ARG3)
868 D("overriding ARG3 %d", parent->req.nr_arg3);
869 d->req.nr_arg3 = new_flags & NM_OPEN_ARG3 ?
870 parent->req.nr_arg3 : 0;
871 if (new_flags & NM_OPEN_RING_CFG) {
872 D("overriding RING_CFG");
873 d->req.nr_tx_slots = parent->req.nr_tx_slots;
874 d->req.nr_rx_slots = parent->req.nr_rx_slots;
875 d->req.nr_tx_rings = parent->req.nr_tx_rings;
876 d->req.nr_rx_rings = parent->req.nr_rx_rings;
877 }
878 if (new_flags & NM_OPEN_IFNAME) {
879 D("overriding ifname %s ringid 0x%x flags 0x%x",
880 parent->req.nr_name, parent->req.nr_ringid,
881 parent->req.nr_flags);
882 memcpy(d->req.nr_name, parent->req.nr_name,
883 sizeof(d->req.nr_name));
884 d->req.nr_ringid = parent->req.nr_ringid;
885 d->req.nr_flags = parent->req.nr_flags;
886 }
887 }
888 /* add the *XPOLL flags */
889 d->req.nr_ringid |= new_flags & (NETMAP_NO_TX_POLL | NETMAP_DO_RX_POLL);
890
891 if (ioctl(d->fd, NIOCREGIF, &d->req)) {
892 snprintf(errmsg, MAXERRMSG, "NIOCREGIF failed: %s", strerror(errno));
893 goto fail;
894 }
895
896 nr_reg = d->req.nr_flags & NR_REG_MASK;
897
898 if (nr_reg == NR_REG_SW) { /* host stack */
899 d->first_tx_ring = d->last_tx_ring = d->req.nr_tx_rings;
900 d->first_rx_ring = d->last_rx_ring = d->req.nr_rx_rings;
901 } else if (nr_reg == NR_REG_ALL_NIC) { /* only nic */
902 d->first_tx_ring = 0;
903 d->first_rx_ring = 0;
904 d->last_tx_ring = d->req.nr_tx_rings - 1;
905 d->last_rx_ring = d->req.nr_rx_rings - 1;
906 } else if (nr_reg == NR_REG_NIC_SW) {
907 d->first_tx_ring = 0;
908 d->first_rx_ring = 0;
909 d->last_tx_ring = d->req.nr_tx_rings;
910 d->last_rx_ring = d->req.nr_rx_rings;
911 } else if (nr_reg == NR_REG_ONE_NIC) {
912 /* XXX check validity */
913 d->first_tx_ring = d->last_tx_ring =
914 d->first_rx_ring = d->last_rx_ring = d->req.nr_ringid & NETMAP_RING_MASK;
915 } else { /* pipes */
916 d->first_tx_ring = d->last_tx_ring = 0;
917 d->first_rx_ring = d->last_rx_ring = 0;
918 }
919
920 /* if parent is defined, do nm_mmap() even if NM_OPEN_NO_MMAP is set */
921 if ((!(new_flags & NM_OPEN_NO_MMAP) || parent) && nm_mmap(d, parent)) {
922 snprintf(errmsg, MAXERRMSG, "mmap failed: %s", strerror(errno));
923 goto fail;
924 }
925
926
927 #ifdef DEBUG_NETMAP_USER
928 { /* debugging code */
929 int i;
930
931 D("%s tx %d .. %d %d rx %d .. %d %d", ifname,
932 d->first_tx_ring, d->last_tx_ring, d->req.nr_tx_rings,
933 d->first_rx_ring, d->last_rx_ring, d->req.nr_rx_rings);
934 for (i = 0; i <= d->req.nr_tx_rings; i++) {
935 struct netmap_ring *r = NETMAP_TXRING(d->nifp, i);
936 D("TX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
937 }
938 for (i = 0; i <= d->req.nr_rx_rings; i++) {
939 struct netmap_ring *r = NETMAP_RXRING(d->nifp, i);
940 D("RX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
941 }
942 }
943 #endif /* debugging */
944
945 d->cur_tx_ring = d->first_tx_ring;
946 d->cur_rx_ring = d->first_rx_ring;
947 return d;
948
949 fail:
950 nm_close(d);
951 if (errmsg[0])
952 D("%s %s", errmsg, ifname);
953 if (errno == 0)
954 errno = EINVAL;
955 return NULL;
956 }
957
958
959 static int
nm_close(struct nm_desc * d)960 nm_close(struct nm_desc *d)
961 {
962 /*
963 * ugly trick to avoid unused warnings
964 */
965 static void *__xxzt[] __attribute__ ((unused)) =
966 { (void *)nm_open, (void *)nm_inject,
967 (void *)nm_dispatch, (void *)nm_nextpkt } ;
968
969 if (d == NULL || d->self != d)
970 return EINVAL;
971 if (d->done_mmap && d->mem)
972 munmap(d->mem, d->memsize);
973 if (d->fd != -1) {
974 close(d->fd);
975 }
976
977 bzero(d, sizeof(*d));
978 free(d);
979 return 0;
980 }
981
982
983 static int
nm_mmap(struct nm_desc * d,const struct nm_desc * parent)984 nm_mmap(struct nm_desc *d, const struct nm_desc *parent)
985 {
986 if (d->done_mmap)
987 return 0;
988
989 if (IS_NETMAP_DESC(parent) && parent->mem &&
990 parent->req.nr_arg2 == d->req.nr_arg2) {
991 /* do not mmap, inherit from parent */
992 D("do not mmap, inherit from parent");
993 d->memsize = parent->memsize;
994 d->mem = parent->mem;
995 } else {
996 /* XXX TODO: check if memsize is too large (or there is overflow) */
997 d->memsize = d->req.nr_memsize;
998 d->mem = mmap(0, d->memsize, PROT_WRITE | PROT_READ, MAP_SHARED,
999 d->fd, 0);
1000 if (d->mem == MAP_FAILED) {
1001 goto fail;
1002 }
1003 d->done_mmap = 1;
1004 }
1005 {
1006 struct netmap_if *nifp = NETMAP_IF(d->mem, d->req.nr_offset);
1007 struct netmap_ring *r = NETMAP_RXRING(nifp, d->first_rx_ring);
1008 if ((void *)r == (void *)nifp) {
1009 /* the descriptor is open for TX only */
1010 r = NETMAP_TXRING(nifp, d->first_tx_ring);
1011 }
1012
1013 *(struct netmap_if **)(uintptr_t)&(d->nifp) = nifp;
1014 *(struct netmap_ring **)(uintptr_t)&d->some_ring = r;
1015 *(void **)(uintptr_t)&d->buf_start = NETMAP_BUF(r, 0);
1016 *(void **)(uintptr_t)&d->buf_end =
1017 (char *)d->mem + d->memsize;
1018 }
1019
1020 return 0;
1021
1022 fail:
1023 return EINVAL;
1024 }
1025
1026 /*
1027 * Same prototype as pcap_inject(), only need to cast.
1028 */
1029 static int
nm_inject(struct nm_desc * d,const void * buf,size_t size)1030 nm_inject(struct nm_desc *d, const void *buf, size_t size)
1031 {
1032 u_int c, n = d->last_tx_ring - d->first_tx_ring + 1,
1033 ri = d->cur_tx_ring;
1034
1035 for (c = 0; c < n ; c++, ri++) {
1036 /* compute current ring to use */
1037 struct netmap_ring *ring;
1038 uint32_t i, j, idx;
1039 size_t rem;
1040
1041 if (ri > d->last_tx_ring)
1042 ri = d->first_tx_ring;
1043 ring = NETMAP_TXRING(d->nifp, ri);
1044 rem = size;
1045 j = ring->cur;
1046 while (rem > ring->nr_buf_size && j != ring->tail) {
1047 rem -= ring->nr_buf_size;
1048 j = nm_ring_next(ring, j);
1049 }
1050 if (j == ring->tail && rem > 0)
1051 continue;
1052 i = ring->cur;
1053 while (i != j) {
1054 idx = ring->slot[i].buf_idx;
1055 ring->slot[i].len = ring->nr_buf_size;
1056 ring->slot[i].flags = NS_MOREFRAG;
1057 nm_pkt_copy(buf, NETMAP_BUF(ring, idx), ring->nr_buf_size);
1058 i = nm_ring_next(ring, i);
1059 buf = (const char *)buf + ring->nr_buf_size;
1060 }
1061 idx = ring->slot[i].buf_idx;
1062 ring->slot[i].len = rem;
1063 ring->slot[i].flags = 0;
1064 nm_pkt_copy(buf, NETMAP_BUF(ring, idx), rem);
1065 ring->head = ring->cur = nm_ring_next(ring, i);
1066 d->cur_tx_ring = ri;
1067 return size;
1068 }
1069 return 0; /* fail */
1070 }
1071
1072
1073 /*
1074 * Same prototype as pcap_dispatch(), only need to cast.
1075 */
1076 static int
nm_dispatch(struct nm_desc * d,int cnt,nm_cb_t cb,u_char * arg)1077 nm_dispatch(struct nm_desc *d, int cnt, nm_cb_t cb, u_char *arg)
1078 {
1079 int n = d->last_rx_ring - d->first_rx_ring + 1;
1080 int c, got = 0, ri = d->cur_rx_ring;
1081 d->hdr.buf = NULL;
1082 d->hdr.flags = NM_MORE_PKTS;
1083 d->hdr.d = d;
1084
1085 if (cnt == 0)
1086 cnt = -1;
1087 /* cnt == -1 means infinite, but rings have a finite amount
1088 * of buffers and the int is large enough that we never wrap,
1089 * so we can omit checking for -1
1090 */
1091 for (c=0; c < n && cnt != got; c++, ri++) {
1092 /* compute current ring to use */
1093 struct netmap_ring *ring;
1094
1095 if (ri > d->last_rx_ring)
1096 ri = d->first_rx_ring;
1097 ring = NETMAP_RXRING(d->nifp, ri);
1098 for ( ; !nm_ring_empty(ring) && cnt != got; got++) {
1099 u_int idx, i;
1100 u_char *oldbuf;
1101 struct netmap_slot *slot;
1102 if (d->hdr.buf) { /* from previous round */
1103 cb(arg, &d->hdr, d->hdr.buf);
1104 }
1105 i = ring->cur;
1106 slot = &ring->slot[i];
1107 idx = slot->buf_idx;
1108 /* d->cur_rx_ring doesn't change inside this loop, but
1109 * set it here, so it reflects d->hdr.buf's ring */
1110 d->cur_rx_ring = ri;
1111 d->hdr.slot = slot;
1112 oldbuf = d->hdr.buf = (u_char *)NETMAP_BUF(ring, idx);
1113 // __builtin_prefetch(buf);
1114 d->hdr.len = d->hdr.caplen = slot->len;
1115 while (slot->flags & NS_MOREFRAG) {
1116 u_char *nbuf;
1117 u_int oldlen = slot->len;
1118 i = nm_ring_next(ring, i);
1119 slot = &ring->slot[i];
1120 d->hdr.len += slot->len;
1121 nbuf = (u_char *)NETMAP_BUF(ring, slot->buf_idx);
1122 if (oldbuf != NULL && nbuf - oldbuf == (int)ring->nr_buf_size &&
1123 oldlen == ring->nr_buf_size) {
1124 d->hdr.caplen += slot->len;
1125 oldbuf = nbuf;
1126 } else {
1127 oldbuf = NULL;
1128 }
1129 }
1130 d->hdr.ts = ring->ts;
1131 ring->head = ring->cur = nm_ring_next(ring, i);
1132 }
1133 }
1134 if (d->hdr.buf) { /* from previous round */
1135 d->hdr.flags = 0;
1136 cb(arg, &d->hdr, d->hdr.buf);
1137 }
1138 return got;
1139 }
1140
1141 static u_char *
nm_nextpkt(struct nm_desc * d,struct nm_pkthdr * hdr)1142 nm_nextpkt(struct nm_desc *d, struct nm_pkthdr *hdr)
1143 {
1144 int ri = d->cur_rx_ring;
1145
1146 do {
1147 /* compute current ring to use */
1148 struct netmap_ring *ring = NETMAP_RXRING(d->nifp, ri);
1149 if (!nm_ring_empty(ring)) {
1150 u_int i = ring->cur;
1151 u_int idx = ring->slot[i].buf_idx;
1152 u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
1153
1154 // __builtin_prefetch(buf);
1155 hdr->ts = ring->ts;
1156 hdr->len = hdr->caplen = ring->slot[i].len;
1157 ring->cur = nm_ring_next(ring, i);
1158 /* we could postpone advancing head if we want
1159 * to hold the buffer. This can be supported in
1160 * the future.
1161 */
1162 ring->head = ring->cur;
1163 d->cur_rx_ring = ri;
1164 return buf;
1165 }
1166 ri++;
1167 if (ri > d->last_rx_ring)
1168 ri = d->first_rx_ring;
1169 } while (ri != d->cur_rx_ring);
1170 return NULL; /* nothing found */
1171 }
1172
1173 #endif /* !HAVE_NETMAP_WITH_LIBS */
1174
1175 #endif /* NETMAP_WITH_LIBS */
1176
1177 #endif /* _NET_NETMAP_USER_H_ */
1178