tcp: Cache align ACK queue header.
[dragonfly.git] / sys / net / netmap_user.h
blob819f9e541890f0ef322081651d561b4a222d834b
1 /*
2 * Copyright (C) 2011 Matteo Landi, Luigi Rizzo. All rights reserved.
3 * Copyright (C) 2013 Universita` di Pisa
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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
25 * SUCH DAMAGE.
29 * $FreeBSD$
31 * This header contains the macros used to manipulate netmap structures
32 * and packets 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)
41 * From there:
42 * struct netmap_ring *NETMAP_TXRING(nifp, index)
43 * struct netmap_ring *NETMAP_RXRING(nifp, index)
44 * we can access ring->nr_cur, ring->nr_avail, ring->nr_flags
46 * ring->slot[i] gives us the i-th slot (we can access
47 * directly plen, flags, bufindex)
49 * char *buf = NETMAP_BUF(ring, x) returns a pointer to
50 * the buffer numbered x
52 * Since rings are circular, we have macros to compute the next index
53 * i = NETMAP_RING_NEXT(ring, i);
55 * To ease porting apps from pcap to netmap we supply a few fuctions
56 * that can be called to open, close and read from netmap in a way
57 * similar to libpcap.
59 * In order to use these, include #define NETMAP_WITH_LIBS
60 * in the source file that invokes these functions.
63 #ifndef _NET_NETMAP_USER_H_
64 #define _NET_NETMAP_USER_H_
66 #include <stdint.h>
67 #include <net/if.h> /* IFNAMSIZ */
68 #include <net/netmap.h>
70 #define _NETMAP_OFFSET(type, ptr, offset) \
71 ((type)(void *)((char *)(ptr) + (offset)))
73 #define NETMAP_IF(b, o) _NETMAP_OFFSET(struct netmap_if *, b, o)
75 #define NETMAP_TXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
76 nifp, (nifp)->ring_ofs[index] )
78 #define NETMAP_RXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
79 nifp, (nifp)->ring_ofs[index + (nifp)->ni_tx_rings + 1] )
81 #define NETMAP_BUF(ring, index) \
82 ((char *)(ring) + (ring)->buf_ofs + ((index)*(ring)->nr_buf_size))
84 #define NETMAP_BUF_IDX(ring, buf) \
85 ( ((char *)(buf) - ((char *)(ring) + (ring)->buf_ofs) ) / \
86 (ring)->nr_buf_size )
88 #define NETMAP_RING_NEXT(r, i) \
89 ((i)+1 == (r)->num_slots ? 0 : (i) + 1 )
91 #define NETMAP_RING_FIRST_RESERVED(r) \
92 ( (r)->cur < (r)->reserved ? \
93 (r)->cur + (r)->num_slots - (r)->reserved : \
94 (r)->cur - (r)->reserved )
97 * Return 1 if the given tx ring is empty.
99 #define NETMAP_TX_RING_EMPTY(r) ((r)->avail >= (r)->num_slots - 1)
101 #ifdef NETMAP_WITH_LIBS
103 * Support for simple I/O libraries.
104 * Include other system headers required for compiling this.
107 #ifndef HAVE_NETMAP_WITH_LIBS
108 #define HAVE_NETMAP_WITH_LIBS
110 #include <sys/time.h>
111 #include <sys/mman.h>
112 #include <string.h> /* memset */
113 #include <sys/ioctl.h>
114 #include <sys/errno.h> /* EINVAL */
115 #include <fcntl.h> /* O_RDWR */
116 #include <malloc.h>
118 struct nm_hdr_t { /* same as pcap_pkthdr */
119 struct timeval ts;
120 uint32_t caplen;
121 uint32_t len;
124 struct nm_desc_t {
125 struct nm_desc_t *self;
126 int fd;
127 void *mem;
128 int memsize;
129 struct netmap_if *nifp;
130 uint16_t first_ring, last_ring, cur_ring;
131 struct nmreq req;
132 struct nm_hdr_t hdr;
136 * when the descriptor is open correctly, d->self == d
138 #define P2NMD(p) ((struct nm_desc_t *)(p))
139 #define IS_NETMAP_DESC(d) (P2NMD(d)->self == P2NMD(d))
140 #define NETMAP_FD(d) (P2NMD(d)->fd)
143 * The callback, invoked on each received packet. Same as libpcap
145 typedef void (*nm_cb_t)(u_char *, const struct nm_hdr_t *, const u_char *d);
148 * The open routine accepts an ifname (netmap:foo or vale:foo) and
149 * optionally a second (string) argument indicating the ring number
150 * to open. If successful, t opens the fd and maps the memory.
152 static struct nm_desc_t *nm_open(const char *ifname,
153 const char *ring_no, int flags, int ring_flags);
156 * nm_dispatch() is the same as pcap_dispatch()
157 * nm_next() is the same as pcap_next()
159 static int nm_dispatch(struct nm_desc_t *, int, nm_cb_t, u_char *);
160 static u_char *nm_next(struct nm_desc_t *, struct nm_hdr_t *);
163 * unmap memory, close file descriptor and free the descriptor.
165 static int nm_close(struct nm_desc_t *);
169 * Try to open, return descriptor if successful, NULL otherwise.
170 * An invalid netmap name will return errno = 0;
172 static struct nm_desc_t *
173 nm_open(const char *ifname, const char *ring_name, int flags, int ring_flags)
175 struct nm_desc_t *d;
176 u_int n;
178 if (strncmp(ifname, "netmap:", 7) && strncmp(ifname, "vale", 4)) {
179 errno = 0; /* name not recognised */
180 return NULL;
182 if (ifname[0] == 'n')
183 ifname += 7;
184 d = (struct nm_desc_t *)calloc(1, sizeof(*d));
185 if (d == NULL) {
186 errno = ENOMEM;
187 return NULL;
189 d->self = d; /* set this early so nm_close() works */
190 d->fd = open("/dev/netmap", O_RDWR);
191 if (d->fd < 0)
192 goto fail;
194 if (flags & NETMAP_SW_RING) {
195 d->req.nr_ringid = NETMAP_SW_RING;
196 } else {
197 u_int r;
198 if (flags & NETMAP_HW_RING) /* interpret ring as int */
199 r = (uintptr_t)ring_name;
200 else /* interpret ring as numeric string */
201 r = ring_name ? atoi(ring_name) : ~0;
202 r = (r < NETMAP_RING_MASK) ? (r | NETMAP_HW_RING) : 0;
203 d->req.nr_ringid = r; /* set the ring */
205 d->req.nr_ringid |= (flags & ~NETMAP_RING_MASK);
206 d->req.nr_version = NETMAP_API;
207 strncpy(d->req.nr_name, ifname, sizeof(d->req.nr_name));
208 if (ioctl(d->fd, NIOCREGIF, &d->req))
209 goto fail;
211 d->memsize = d->req.nr_memsize;
212 d->mem = mmap(0, d->memsize, PROT_WRITE | PROT_READ, MAP_SHARED,
213 d->fd, 0);
214 if (d->mem == NULL)
215 goto fail;
216 d->nifp = NETMAP_IF(d->mem, d->req.nr_offset);
217 if (d->req.nr_ringid & NETMAP_SW_RING) {
218 d->first_ring = d->last_ring = d->req.nr_rx_rings;
219 } else if (d->req.nr_ringid & NETMAP_HW_RING) {
220 d->first_ring = d->last_ring =
221 d->req.nr_ringid & NETMAP_RING_MASK;
222 } else {
223 d->first_ring = 0;
224 d->last_ring = d->req.nr_rx_rings - 1;
226 d->cur_ring = d->first_ring;
227 for (n = d->first_ring; n <= d->last_ring; n++) {
228 struct netmap_ring *ring = NETMAP_RXRING(d->nifp, n);
229 ring->flags |= ring_flags;
231 return d;
233 fail:
234 nm_close(d);
235 errno = EINVAL;
236 return NULL;
240 static int
241 nm_close(struct nm_desc_t *d)
243 if (d == NULL || d->self != d)
244 return EINVAL;
245 if (d->mem)
246 munmap(d->mem, d->memsize);
247 if (d->fd)
248 close(d->fd);
249 bzero(d, sizeof(*d));
250 free(d);
251 return 0;
256 * Same prototype as pcap_dispatch(), only need to cast.
258 inline /* not really, but disable unused warnings */
259 static int
260 nm_dispatch(struct nm_desc_t *d, int cnt, nm_cb_t cb, u_char *arg)
262 int n = d->last_ring - d->first_ring + 1;
263 int c, got = 0, ri = d->cur_ring;
265 if (cnt == 0)
266 cnt = -1;
267 /* cnt == -1 means infinite, but rings have a finite amount
268 * of buffers and the int is large enough that we never wrap,
269 * so we can omit checking for -1
271 for (c=0; c < n && cnt != got; c++) {
272 /* compute current ring to use */
273 struct netmap_ring *ring;
275 ri = d->cur_ring + c;
276 if (ri > d->last_ring)
277 ri = d->first_ring;
278 ring = NETMAP_RXRING(d->nifp, ri);
279 for ( ; ring->avail > 0 && cnt != got; got++) {
280 u_int i = ring->cur;
281 u_int idx = ring->slot[i].buf_idx;
282 u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
283 // XXX should check valid buf
284 // prefetch(buf);
285 d->hdr.len = d->hdr.caplen = ring->slot[i].len;
286 d->hdr.ts = ring->ts;
287 cb(arg, &d->hdr, buf);
288 ring->cur = NETMAP_RING_NEXT(ring, i);
289 ring->avail--;
292 d->cur_ring = ri;
293 return got;
296 inline /* not really, but disable unused warnings */
297 static u_char *
298 nm_next(struct nm_desc_t *d, struct nm_hdr_t *hdr)
300 int ri = d->cur_ring;
302 do {
303 /* compute current ring to use */
304 struct netmap_ring *ring = NETMAP_RXRING(d->nifp, ri);
305 if (ring->avail > 0) {
306 u_int i = ring->cur;
307 u_int idx = ring->slot[i].buf_idx;
308 u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
309 // XXX should check valid buf
310 // prefetch(buf);
311 hdr->ts = ring->ts;
312 hdr->len = hdr->caplen = ring->slot[i].len;
313 ring->cur = NETMAP_RING_NEXT(ring, i);
314 ring->avail--;
315 d->cur_ring = ri;
316 return buf;
318 ri++;
319 if (ri > d->last_ring)
320 ri = d->first_ring;
321 } while (ri != d->cur_ring);
322 return NULL; /* nothing found */
325 #endif /* !HAVE_NETMAP_WITH_LIBS */
327 #endif /* NETMAP_WITH_LIBS */
329 #endif /* _NET_NETMAP_USER_H_ */