2 * Copyright (C) 2011-2016 Universita` di Pisa
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * Functions and macros to manipulate netmap structures and packets
32 * in userspace. See netmap(4) for more information.
34 * The address of the struct netmap_if, say nifp, is computed from the
35 * value returned from ioctl(.., NIOCREG, ...) and the mmap region:
36 * ioctl(fd, NIOCREG, &req);
37 * mem = mmap(0, ... );
38 * nifp = NETMAP_IF(mem, req.nr_nifp);
39 * (so simple, we could just do it manually)
42 * struct netmap_ring *NETMAP_TXRING(nifp, index)
43 * struct netmap_ring *NETMAP_RXRING(nifp, index)
44 * we can access ring->cur, ring->head, ring->tail, etc.
46 * ring->slot[i] gives us the i-th slot (we can access
47 * directly len, flags, buf_idx)
49 * char *buf = NETMAP_BUF(ring, x) returns a pointer to
50 * the buffer numbered x
52 * All ring indexes (head, cur, tail) should always move forward.
53 * To compute the next index in a circular ring you can use
54 * i = nm_ring_next(ring, i);
56 * To ease porting apps from pcap to netmap we supply a few fuctions
57 * that can be called to open, close, read and write on netmap in a way
58 * similar to libpcap. Note that the read/write function depend on
59 * an ioctl()/select()/poll() being issued to refill rings or push
62 * In order to use these, include #define NETMAP_WITH_LIBS
63 * in the source file that invokes these functions.
66 #ifndef _NET_NETMAP_USER_H_
67 #define _NET_NETMAP_USER_H_
69 #define NETMAP_DEVICE_NAME "/dev/netmap"
73 * we can compile userspace apps with either cygwin or msvc,
74 * and we use _WIN32 to identify windows specific code
80 #endif /* __CYGWIN__ */
83 #undef NETMAP_DEVICE_NAME
84 #define NETMAP_DEVICE_NAME "/proc/sys/DosDevices/Global/netmap"
87 #include <sys/cygwin.h>
91 #include <sys/socket.h> /* apple needs sockaddr */
92 #include <net/if.h> /* IFNAMSIZ */
96 #define likely(x) __builtin_expect(!!(x), 1)
97 #define unlikely(x) __builtin_expect(!!(x), 0)
98 #endif /* likely and unlikely */
100 #include <net/netmap.h>
103 #define _NETMAP_OFFSET(type, ptr, offset) \
104 ((type)(void *)((char *)(ptr) + (offset)))
106 #define NETMAP_IF(_base, _ofs) _NETMAP_OFFSET(struct netmap_if *, _base, _ofs)
108 #define NETMAP_TXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
109 nifp, (nifp)->ring_ofs[index] )
111 #define NETMAP_RXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
112 nifp, (nifp)->ring_ofs[index + (nifp)->ni_tx_rings + 1] )
114 #define NETMAP_BUF(ring, index) \
115 ((char *)(ring) + (ring)->buf_ofs + ((index)*(ring)->nr_buf_size))
117 #define NETMAP_BUF_IDX(ring, buf) \
118 ( ((char *)(buf) - ((char *)(ring) + (ring)->buf_ofs) ) / \
119 (ring)->nr_buf_size )
122 static inline uint32_t
123 nm_ring_next(struct netmap_ring *r, uint32_t i)
125 return ( unlikely(i + 1 == r->num_slots) ? 0 : i + 1);
130 * Return 1 if we have pending transmissions in the tx ring.
131 * When everything is complete ring->head = ring->tail + 1 (modulo ring size)
134 nm_tx_pending(struct netmap_ring *r)
136 return nm_ring_next(r, r->tail) != r->head;
140 static inline uint32_t
141 nm_ring_space(struct netmap_ring *ring)
143 int ret = ring->tail - ring->cur;
145 ret += ring->num_slots;
150 #ifdef NETMAP_WITH_LIBS
152 * Support for simple I/O libraries.
153 * Include other system headers required for compiling this.
156 #ifndef HAVE_NETMAP_WITH_LIBS
157 #define HAVE_NETMAP_WITH_LIBS
160 #include <sys/time.h>
161 #include <sys/mman.h>
162 #include <string.h> /* memset */
163 #include <sys/ioctl.h>
164 #include <sys/errno.h> /* EINVAL */
165 #include <fcntl.h> /* O_RDWR */
166 #include <unistd.h> /* close() */
170 #ifndef ND /* debug macros */
172 #define ND(_fmt, ...) do {} while(0)
173 #define D(_fmt, ...) \
175 struct timeval _t0; \
176 gettimeofday(&_t0, NULL); \
177 fprintf(stderr, "%03d.%06d %s [%d] " _fmt "\n", \
178 (int)(_t0.tv_sec % 1000), (int)_t0.tv_usec, \
179 __FUNCTION__, __LINE__, ##__VA_ARGS__); \
182 /* Rate limited version of "D", lps indicates how many per second */
183 #define RD(lps, format, ...) \
185 static int __t0, __cnt; \
186 struct timeval __xxts; \
187 gettimeofday(&__xxts, NULL); \
188 if (__t0 != __xxts.tv_sec) { \
189 __t0 = __xxts.tv_sec; \
192 if (__cnt++ < lps) { \
193 D(format, ##__VA_ARGS__); \
198 struct nm_pkthdr { /* first part is the same as pcap_pkthdr */
203 uint64_t flags; /* NM_MORE_PKTS etc */
204 #define NM_MORE_PKTS 1
206 struct netmap_slot *slot;
210 struct nm_stat { /* same as pcap_stat */
214 #ifdef WIN32 /* XXX or _WIN32 ? */
219 #define NM_ERRBUF_SIZE 512
222 struct nm_desc *self; /* point to self if netmap. */
226 int done_mmap; /* set if mem is the result of mmap */
227 struct netmap_if * const nifp;
228 uint16_t first_tx_ring, last_tx_ring, cur_tx_ring;
229 uint16_t first_rx_ring, last_rx_ring, cur_rx_ring;
230 struct nmreq req; /* also contains the nr_name = ifname */
231 struct nm_pkthdr hdr;
234 * The memory contains netmap_if, rings and then buffers.
235 * Given a pointer (e.g. to nm_inject) we can compare with
236 * mem/buf_start/buf_end to tell if it is a buffer or
237 * some other descriptor in our region.
238 * We also store a pointer to some ring as it helps in the
239 * translation from buffer indexes to addresses.
241 struct netmap_ring * const some_ring;
242 void * const buf_start;
243 void * const buf_end;
244 /* parameters from pcap_open_live */
250 /* save flags so we can restore them on close */
256 char msg[NM_ERRBUF_SIZE];
260 * when the descriptor is open correctly, d->self == d
261 * Eventually we should also use some magic number.
263 #define P2NMD(p) ((struct nm_desc *)(p))
264 #define IS_NETMAP_DESC(d) ((d) && P2NMD(d)->self == P2NMD(d))
265 #define NETMAP_FD(d) (P2NMD(d)->fd)
269 * this is a slightly optimized copy routine which rounds
270 * to multiple of 64 bytes and is often faster than dealing
271 * with other odd sizes. We assume there is enough room
272 * in the source and destination buffers.
274 * XXX only for multiples of 64 bytes, non overlapped.
277 nm_pkt_copy(const void *_src, void *_dst, int l)
279 const uint64_t *src = (const uint64_t *)_src;
280 uint64_t *dst = (uint64_t *)_dst;
282 if (unlikely(l >= 1024)) {
286 for (; likely(l > 0); l-=64) {
300 * The callback, invoked on each received packet. Same as libpcap
302 typedef void (*nm_cb_t)(u_char *, const struct nm_pkthdr *, const u_char *d);
305 *--- the pcap-like API ---
307 * nm_open() opens a file descriptor, binds to a port and maps memory.
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,
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)
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.
335 static struct nm_desc *nm_open(const char *ifname, const struct nmreq *req,
336 uint64_t flags, const struct nm_desc *arg);
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.
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 */
357 * nm_close() closes and restores the port to its previous state
360 static int nm_close(struct nm_desc *);
363 * nm_mmap() do mmap or inherit from parent if the nr_arg2
364 * (memory block) matches.
367 static int nm_mmap(struct nm_desc *, const struct nm_desc *);
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()
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 *);
381 intptr_t _get_osfhandle(int); /* defined in io.h in windows */
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.
388 struct win_netmap_fd_list {
389 struct win_netmap_fd_list *next;
391 HANDLE win_netmap_handle;
395 * list head containing all the netmap opened fd and their
396 * windows HANDLE counterparts
398 static struct win_netmap_fd_list *win_netmap_fd_list_head;
401 win_insert_fd_record(int fd)
403 struct win_netmap_fd_list *curr;
405 for (curr = win_netmap_fd_list_head; curr; curr = curr->next) {
406 if (fd == curr->win_netmap_fd) {
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;
418 win_remove_fd_record(int fd)
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)
425 /* found the entry */
426 if (prev == NULL) { /* we are freeing the first entry */
427 win_netmap_fd_list_head = curr->next;
429 prev->next = curr->next;
438 win_get_netmap_handle(int fd)
440 struct win_netmap_fd_list *curr;
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;
451 * we need to wrap ioctl and mmap, at least for the netmap file descriptors
455 * use this function only from netmap_user.h internal functions
456 * same as ioctl, returns 0 on success and -1 on error
459 win_nm_ioctl_internal(HANDLE h, int32_t ctlCode, void *arg)
461 DWORD bReturn = 0, szIn, szOut;
462 BOOL ioctlReturnStatus;
463 void *inParam = arg, *outParam = arg;
467 szIn = sizeof(POLL_REQUEST_DATA);
468 szOut = sizeof(POLL_REQUEST_DATA);
472 szOut = sizeof(void*);
473 inParam = NULL; /* nothing on input */
481 szIn = sizeof(struct nmreq);
482 szOut = sizeof(struct nmreq);
485 D("unsupported NIOCCONFIG!");
488 default: /* a regular ioctl */
489 D("invalid ioctl %x on netmap fd", ctlCode);
493 ioctlReturnStatus = DeviceIoControl(h,
494 ctlCode, inParam, szIn,
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;
503 * this function is what must be called from user-space programs
504 * same as ioctl, returns 0 on success and -1 on error
507 win_nm_ioctl(int fd, int32_t ctlCode, void *arg)
509 HANDLE h = win_get_netmap_handle(fd);
512 return ioctl(fd, ctlCode, arg);
514 return win_nm_ioctl_internal(h, ctlCode, arg);
518 #define ioctl win_nm_ioctl /* from now on, within this file ... */
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
526 win32_mmap_emulated(void *addr, size_t length, int prot, int flags, int fd, int32_t offset)
528 HANDLE h = win_get_netmap_handle(fd);
531 return mmap(addr, length, prot, flags, fd, offset);
535 return win_nm_ioctl_internal(h, NETMAP_MMAP, &ret) ?
536 NULL : ret.pUsermodeVirtualAddress;
540 #define mmap win32_mmap_emulated
542 #include <sys/poll.h> /* XXX needed to use the structure pollfd */
545 win_nm_poll(struct pollfd *fds, int nfds, int timeout)
549 if (nfds != 1 || fds == NULL || (h = win_get_netmap_handle(fds->fd)) == NULL) {;
550 return poll(fds, nfds, timeout);
552 POLL_REQUEST_DATA prd;
554 prd.timeout = timeout;
555 prd.events = fds->events;
557 win_nm_ioctl_internal(h, NETMAP_POLL, &prd);
558 if ((prd.revents == POLLERR) || (prd.revents == STATUS_TIMEOUT)) {
565 #define poll win_nm_poll
568 win_nm_open(char* pathname, int flags)
571 if (strcmp(pathname, NETMAP_DEVICE_NAME) == 0) {
572 int fd = open(NETMAP_DEVICE_NAME, O_RDWR);
577 win_insert_fd_record(fd);
580 return open(pathname, flags);
584 #define open win_nm_open
591 if (win_get_netmap_handle(fd) != NULL) {
592 win_remove_fd_record(fd);
598 #define close win_nm_close
603 nm_is_identifier(const char *s, const char *e)
605 for (; s != e; s++) {
606 if (!isalnum(*s) && *s != '_') {
615 * Try to open, return descriptor if successful, NULL otherwise.
616 * An invalid netmap name will return errno = 0;
617 * You can pass a pointer to a pre-filled nm_desc to add special
618 * parameters. Flags is used as follows
619 * NM_OPEN_NO_MMAP use the memory from arg, only XXX avoid mmap
620 * if the nr_arg2 (memory block) matches.
621 * NM_OPEN_ARG1 use req.nr_arg1 from arg
622 * NM_OPEN_ARG2 use req.nr_arg2 from arg
623 * NM_OPEN_RING_CFG user ring config from arg
625 static struct nm_desc *
626 nm_open(const char *ifname, const struct nmreq *req,
627 uint64_t new_flags, const struct nm_desc *arg)
629 struct nm_desc *d = NULL;
630 const struct nm_desc *parent = arg;
632 uint32_t nr_ringid = 0, nr_flags, nr_reg;
633 const char *port = NULL;
634 const char *vpname = NULL;
636 char errmsg[MAXERRMSG] = "";
637 enum { P_START, P_RNGSFXOK, P_GETNUM, P_FLAGS, P_FLAGSOK, P_MEMID } p_state;
640 uint16_t nr_arg2 = 0;
642 if (strncmp(ifname, "netmap:", 7) &&
643 strncmp(ifname, NM_BDG_NAME, strlen(NM_BDG_NAME))) {
644 errno = 0; /* name not recognised, not an error */
648 is_vale = (ifname[0] == 'v');
650 port = index(ifname, ':');
652 snprintf(errmsg, MAXERRMSG,
653 "missing ':' in vale name");
657 if (!nm_is_identifier(ifname + 4, port)) {
658 snprintf(errmsg, MAXERRMSG, "invalid bridge name");
668 /* scan for a separator */
669 for (; *port && !index("-*^{}/@", *port); port++)
672 if (is_vale && !nm_is_identifier(vpname, port)) {
673 snprintf(errmsg, MAXERRMSG, "invalid bridge port name");
677 namelen = port - ifname;
678 if (namelen >= sizeof(d->req.nr_name)) {
679 snprintf(errmsg, MAXERRMSG, "name too long");
683 nr_flags = NR_REG_ALL_NIC; /* default for no suffix */
688 case '^': /* only SW ring */
689 nr_flags = NR_REG_SW;
690 p_state = P_RNGSFXOK;
692 case '*': /* NIC and SW */
693 nr_flags = NR_REG_NIC_SW;
694 p_state = P_RNGSFXOK;
696 case '-': /* one NIC ring pair */
697 nr_flags = NR_REG_ONE_NIC;
700 case '{': /* pipe (master endpoint) */
701 nr_flags = NR_REG_PIPE_MASTER;
704 case '}': /* pipe (slave endoint) */
705 nr_flags = NR_REG_PIPE_SLAVE;
708 case '/': /* start of flags */
711 case '@': /* start of memid */
715 snprintf(errmsg, MAXERRMSG, "unknown modifier: '%c'", *port);
729 snprintf(errmsg, MAXERRMSG, "unexpected character: '%c'", *port);
735 num = strtol(port, (char **)&port, 10);
736 if (num < 0 || num >= NETMAP_RING_MASK) {
737 snprintf(errmsg, MAXERRMSG, "'%ld' out of range [0, %d)",
738 num, NETMAP_RING_MASK);
741 nr_ringid = num & NETMAP_RING_MASK;
742 p_state = P_RNGSFXOK;
753 nr_flags |= NR_EXCLUSIVE;
756 nr_flags |= NR_ZCOPY_MON;
759 nr_flags |= NR_MONITOR_TX;
762 nr_flags |= NR_MONITOR_RX;
765 nr_flags |= NR_RX_RINGS_ONLY;
768 nr_flags |= NR_TX_RINGS_ONLY;
771 snprintf(errmsg, MAXERRMSG, "unrecognized flag: '%c'", *port);
779 snprintf(errmsg, MAXERRMSG, "double setting of memid");
782 num = strtol(port, (char **)&port, 10);
784 snprintf(errmsg, MAXERRMSG, "invalid memid %ld, must be >0", num);
788 p_state = P_RNGSFXOK;
792 if (p_state != P_START && p_state != P_RNGSFXOK && p_state != P_FLAGSOK) {
793 snprintf(errmsg, MAXERRMSG, "unexpected end of port name");
796 ND("flags: %s %s %s %s",
797 (nr_flags & NR_EXCLUSIVE) ? "EXCLUSIVE" : "",
798 (nr_flags & NR_ZCOPY_MON) ? "ZCOPY_MON" : "",
799 (nr_flags & NR_MONITOR_TX) ? "MONITOR_TX" : "",
800 (nr_flags & NR_MONITOR_RX) ? "MONITOR_RX" : "");
801 d = (struct nm_desc *)calloc(1, sizeof(*d));
803 snprintf(errmsg, MAXERRMSG, "nm_desc alloc failure");
807 d->self = d; /* set this early so nm_close() works */
808 d->fd = open(NETMAP_DEVICE_NAME, O_RDWR);
810 snprintf(errmsg, MAXERRMSG, "cannot open /dev/netmap: %s", strerror(errno));
816 d->req.nr_version = NETMAP_API;
817 d->req.nr_ringid &= ~NETMAP_RING_MASK;
819 /* these fields are overridden by ifname and flags processing */
820 d->req.nr_ringid |= nr_ringid;
821 d->req.nr_flags |= nr_flags;
823 d->req.nr_arg2 = nr_arg2;
824 memcpy(d->req.nr_name, ifname, namelen);
825 d->req.nr_name[namelen] = '\0';
826 /* optionally import info from parent */
827 if (IS_NETMAP_DESC(parent) && new_flags) {
828 if (new_flags & NM_OPEN_ARG1)
829 D("overriding ARG1 %d", parent->req.nr_arg1);
830 d->req.nr_arg1 = new_flags & NM_OPEN_ARG1 ?
831 parent->req.nr_arg1 : 4;
832 if (new_flags & NM_OPEN_ARG2)
833 D("overriding ARG2 %d", parent->req.nr_arg2);
834 d->req.nr_arg2 = new_flags & NM_OPEN_ARG2 ?
835 parent->req.nr_arg2 : 0;
836 if (new_flags & NM_OPEN_ARG3)
837 D("overriding ARG3 %d", parent->req.nr_arg3);
838 d->req.nr_arg3 = new_flags & NM_OPEN_ARG3 ?
839 parent->req.nr_arg3 : 0;
840 if (new_flags & NM_OPEN_RING_CFG) {
841 D("overriding RING_CFG");
842 d->req.nr_tx_slots = parent->req.nr_tx_slots;
843 d->req.nr_rx_slots = parent->req.nr_rx_slots;
844 d->req.nr_tx_rings = parent->req.nr_tx_rings;
845 d->req.nr_rx_rings = parent->req.nr_rx_rings;
847 if (new_flags & NM_OPEN_IFNAME) {
848 D("overriding ifname %s ringid 0x%x flags 0x%x",
849 parent->req.nr_name, parent->req.nr_ringid,
850 parent->req.nr_flags);
851 memcpy(d->req.nr_name, parent->req.nr_name,
852 sizeof(d->req.nr_name));
853 d->req.nr_ringid = parent->req.nr_ringid;
854 d->req.nr_flags = parent->req.nr_flags;
857 /* add the *XPOLL flags */
858 d->req.nr_ringid |= new_flags & (NETMAP_NO_TX_POLL | NETMAP_DO_RX_POLL);
860 if (ioctl(d->fd, NIOCREGIF, &d->req)) {
861 snprintf(errmsg, MAXERRMSG, "NIOCREGIF failed: %s", strerror(errno));
865 /* if parent is defined, do nm_mmap() even if NM_OPEN_NO_MMAP is set */
866 if ((!(new_flags & NM_OPEN_NO_MMAP) || parent) && nm_mmap(d, parent)) {
867 snprintf(errmsg, MAXERRMSG, "mmap failed: %s", strerror(errno));
871 nr_reg = d->req.nr_flags & NR_REG_MASK;
873 if (nr_reg == NR_REG_SW) { /* host stack */
874 d->first_tx_ring = d->last_tx_ring = d->req.nr_tx_rings;
875 d->first_rx_ring = d->last_rx_ring = d->req.nr_rx_rings;
876 } else if (nr_reg == NR_REG_ALL_NIC) { /* only nic */
877 d->first_tx_ring = 0;
878 d->first_rx_ring = 0;
879 d->last_tx_ring = d->req.nr_tx_rings - 1;
880 d->last_rx_ring = d->req.nr_rx_rings - 1;
881 } else if (nr_reg == NR_REG_NIC_SW) {
882 d->first_tx_ring = 0;
883 d->first_rx_ring = 0;
884 d->last_tx_ring = d->req.nr_tx_rings;
885 d->last_rx_ring = d->req.nr_rx_rings;
886 } else if (nr_reg == NR_REG_ONE_NIC) {
887 /* XXX check validity */
888 d->first_tx_ring = d->last_tx_ring =
889 d->first_rx_ring = d->last_rx_ring = d->req.nr_ringid & NETMAP_RING_MASK;
891 d->first_tx_ring = d->last_tx_ring = 0;
892 d->first_rx_ring = d->last_rx_ring = 0;
895 #ifdef DEBUG_NETMAP_USER
896 { /* debugging code */
899 D("%s tx %d .. %d %d rx %d .. %d %d", ifname,
900 d->first_tx_ring, d->last_tx_ring, d->req.nr_tx_rings,
901 d->first_rx_ring, d->last_rx_ring, d->req.nr_rx_rings);
902 for (i = 0; i <= d->req.nr_tx_rings; i++) {
903 struct netmap_ring *r = NETMAP_TXRING(d->nifp, i);
904 D("TX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
906 for (i = 0; i <= d->req.nr_rx_rings; i++) {
907 struct netmap_ring *r = NETMAP_RXRING(d->nifp, i);
908 D("RX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
911 #endif /* debugging */
913 d->cur_tx_ring = d->first_tx_ring;
914 d->cur_rx_ring = d->first_rx_ring;
920 D("%s %s", errmsg, ifname);
928 nm_close(struct nm_desc *d)
931 * ugly trick to avoid unused warnings
933 static void *__xxzt[] __attribute__ ((unused)) =
934 { (void *)nm_open, (void *)nm_inject,
935 (void *)nm_dispatch, (void *)nm_nextpkt } ;
937 if (d == NULL || d->self != d)
939 if (d->done_mmap && d->mem)
940 munmap(d->mem, d->memsize);
945 bzero(d, sizeof(*d));
952 nm_mmap(struct nm_desc *d, const struct nm_desc *parent)
954 //XXX TODO: check if mmap is already done
956 if (IS_NETMAP_DESC(parent) && parent->mem &&
957 parent->req.nr_arg2 == d->req.nr_arg2) {
958 /* do not mmap, inherit from parent */
959 D("do not mmap, inherit from parent");
960 d->memsize = parent->memsize;
961 d->mem = parent->mem;
963 /* XXX TODO: check if memsize is too large (or there is overflow) */
964 d->memsize = d->req.nr_memsize;
965 d->mem = mmap(0, d->memsize, PROT_WRITE | PROT_READ, MAP_SHARED,
967 if (d->mem == MAP_FAILED) {
973 struct netmap_if *nifp = NETMAP_IF(d->mem, d->req.nr_offset);
974 struct netmap_ring *r = NETMAP_RXRING(nifp, );
976 *(struct netmap_if **)(uintptr_t)&(d->nifp) = nifp;
977 *(struct netmap_ring **)(uintptr_t)&d->some_ring = r;
978 *(void **)(uintptr_t)&d->buf_start = NETMAP_BUF(r, 0);
979 *(void **)(uintptr_t)&d->buf_end =
980 (char *)d->mem + d->memsize;
990 * Same prototype as pcap_inject(), only need to cast.
993 nm_inject(struct nm_desc *d, const void *buf, size_t size)
995 u_int c, n = d->last_tx_ring - d->first_tx_ring + 1;
997 for (c = 0; c < n ; c++) {
998 /* compute current ring to use */
999 struct netmap_ring *ring;
1001 uint32_t ri = d->cur_tx_ring + c;
1003 if (ri > d->last_tx_ring)
1004 ri = d->first_tx_ring;
1005 ring = NETMAP_TXRING(d->nifp, ri);
1006 if (nm_ring_empty(ring)) {
1010 idx = ring->slot[i].buf_idx;
1011 ring->slot[i].len = size;
1012 nm_pkt_copy(buf, NETMAP_BUF(ring, idx), size);
1013 d->cur_tx_ring = ri;
1014 ring->head = ring->cur = nm_ring_next(ring, i);
1017 return 0; /* fail */
1022 * Same prototype as pcap_dispatch(), only need to cast.
1025 nm_dispatch(struct nm_desc *d, int cnt, nm_cb_t cb, u_char *arg)
1027 int n = d->last_rx_ring - d->first_rx_ring + 1;
1028 int c, got = 0, ri = d->cur_rx_ring;
1030 d->hdr.flags = NM_MORE_PKTS;
1035 /* cnt == -1 means infinite, but rings have a finite amount
1036 * of buffers and the int is large enough that we never wrap,
1037 * so we can omit checking for -1
1039 for (c=0; c < n && cnt != got; c++) {
1040 /* compute current ring to use */
1041 struct netmap_ring *ring;
1043 ri = d->cur_rx_ring + c;
1044 if (ri > d->last_rx_ring)
1045 ri = d->first_rx_ring;
1046 ring = NETMAP_RXRING(d->nifp, ri);
1047 for ( ; !nm_ring_empty(ring) && cnt != got; got++) {
1049 if (d->hdr.buf) { /* from previous round */
1050 cb(arg, &d->hdr, d->hdr.buf);
1053 idx = ring->slot[i].buf_idx;
1054 d->hdr.slot = &ring->slot[i];
1055 d->hdr.buf = (u_char *)NETMAP_BUF(ring, idx);
1056 // __builtin_prefetch(buf);
1057 d->hdr.len = d->hdr.caplen = ring->slot[i].len;
1058 d->hdr.ts = ring->ts;
1059 ring->head = ring->cur = nm_ring_next(ring, i);
1062 if (d->hdr.buf) { /* from previous round */
1064 cb(arg, &d->hdr, d->hdr.buf);
1066 d->cur_rx_ring = ri;
1071 nm_nextpkt(struct nm_desc *d, struct nm_pkthdr *hdr)
1073 int ri = d->cur_rx_ring;
1076 /* compute current ring to use */
1077 struct netmap_ring *ring = NETMAP_RXRING(d->nifp, ri);
1078 if (!nm_ring_empty(ring)) {
1079 u_int i = ring->cur;
1080 u_int idx = ring->slot[i].buf_idx;
1081 u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
1083 // __builtin_prefetch(buf);
1085 hdr->len = hdr->caplen = ring->slot[i].len;
1086 ring->cur = nm_ring_next(ring, i);
1087 /* we could postpone advancing head if we want
1088 * to hold the buffer. This can be supported in
1091 ring->head = ring->cur;
1092 d->cur_rx_ring = ri;
1096 if (ri > d->last_rx_ring)
1097 ri = d->first_rx_ring;
1098 } while (ri != d->cur_rx_ring);
1099 return NULL; /* nothing found */
1102 #endif /* !HAVE_NETMAP_WITH_LIBS */
1104 #endif /* NETMAP_WITH_LIBS */
1106 #endif /* _NET_NETMAP_USER_H_ */