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