netmap_freebsd.c revision 270252
1/*
2 * Copyright (C) 2013-2014 Universita` di Pisa. All rights reserved.
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
6 * are met:
7 *   1. Redistributions of source code must retain the above copyright
8 *      notice, this list of conditions and the following disclaimer.
9 *   2. Redistributions in binary form must reproduce the above copyright
10 *      notice, this list of conditions and the following disclaimer in the
11 *      documentation and/or other materials provided with the distribution.
12 *
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23 * SUCH DAMAGE.
24 */
25
26/* $FreeBSD: stable/10/sys/dev/netmap/netmap_freebsd.c 270252 2014-08-20 23:34:36Z luigi $ */
27
28#include <sys/types.h>
29#include <sys/module.h>
30#include <sys/errno.h>
31#include <sys/param.h>  /* defines used in kernel.h */
32#include <sys/poll.h>  /* POLLIN, POLLOUT */
33#include <sys/kernel.h> /* types used in module initialization */
34#include <sys/conf.h>	/* DEV_MODULE */
35#include <sys/endian.h>
36
37#include <sys/rwlock.h>
38
39#include <vm/vm.h>      /* vtophys */
40#include <vm/pmap.h>    /* vtophys */
41#include <vm/vm_param.h>
42#include <vm/vm_object.h>
43#include <vm/vm_page.h>
44#include <vm/vm_pager.h>
45#include <vm/uma.h>
46
47
48#include <sys/malloc.h>
49#include <sys/socket.h> /* sockaddrs */
50#include <sys/selinfo.h>
51#include <net/if.h>
52#include <net/if_var.h>
53#include <net/if_types.h> /* IFT_ETHER */
54#include <net/ethernet.h> /* ether_ifdetach */
55#include <net/if_dl.h> /* LLADDR */
56#include <machine/bus.h>        /* bus_dmamap_* */
57#include <netinet/in.h>		/* in6_cksum_pseudo() */
58#include <machine/in_cksum.h>  /* in_pseudo(), in_cksum_hdr() */
59
60#include <net/netmap.h>
61#include <dev/netmap/netmap_kern.h>
62#include <dev/netmap/netmap_mem2.h>
63
64
65/* ======================== FREEBSD-SPECIFIC ROUTINES ================== */
66
67rawsum_t
68nm_csum_raw(uint8_t *data, size_t len, rawsum_t cur_sum)
69{
70	/* TODO XXX please use the FreeBSD implementation for this. */
71	uint16_t *words = (uint16_t *)data;
72	int nw = len / 2;
73	int i;
74
75	for (i = 0; i < nw; i++)
76		cur_sum += be16toh(words[i]);
77
78	if (len & 1)
79		cur_sum += (data[len-1] << 8);
80
81	return cur_sum;
82}
83
84/* Fold a raw checksum: 'cur_sum' is in host byte order, while the
85 * return value is in network byte order.
86 */
87uint16_t
88nm_csum_fold(rawsum_t cur_sum)
89{
90	/* TODO XXX please use the FreeBSD implementation for this. */
91	while (cur_sum >> 16)
92		cur_sum = (cur_sum & 0xFFFF) + (cur_sum >> 16);
93
94	return htobe16((~cur_sum) & 0xFFFF);
95}
96
97uint16_t nm_csum_ipv4(struct nm_iphdr *iph)
98{
99#if 0
100	return in_cksum_hdr((void *)iph);
101#else
102	return nm_csum_fold(nm_csum_raw((uint8_t*)iph, sizeof(struct nm_iphdr), 0));
103#endif
104}
105
106void
107nm_csum_tcpudp_ipv4(struct nm_iphdr *iph, void *data,
108					size_t datalen, uint16_t *check)
109{
110#ifdef INET
111	uint16_t pseudolen = datalen + iph->protocol;
112
113	/* Compute and insert the pseudo-header cheksum. */
114	*check = in_pseudo(iph->saddr, iph->daddr,
115				 htobe16(pseudolen));
116	/* Compute the checksum on TCP/UDP header + payload
117	 * (includes the pseudo-header).
118	 */
119	*check = nm_csum_fold(nm_csum_raw(data, datalen, 0));
120#else
121	static int notsupported = 0;
122	if (!notsupported) {
123		notsupported = 1;
124		D("inet4 segmentation not supported");
125	}
126#endif
127}
128
129void
130nm_csum_tcpudp_ipv6(struct nm_ipv6hdr *ip6h, void *data,
131					size_t datalen, uint16_t *check)
132{
133#ifdef INET6
134	*check = in6_cksum_pseudo((void*)ip6h, datalen, ip6h->nexthdr, 0);
135	*check = nm_csum_fold(nm_csum_raw(data, datalen, 0));
136#else
137	static int notsupported = 0;
138	if (!notsupported) {
139		notsupported = 1;
140		D("inet6 segmentation not supported");
141	}
142#endif
143}
144
145
146/*
147 * Intercept the rx routine in the standard device driver.
148 * Second argument is non-zero to intercept, 0 to restore
149 */
150int
151netmap_catch_rx(struct netmap_adapter *na, int intercept)
152{
153	struct netmap_generic_adapter *gna = (struct netmap_generic_adapter *)na;
154	struct ifnet *ifp = na->ifp;
155
156	if (intercept) {
157		if (gna->save_if_input) {
158			D("cannot intercept again");
159			return EINVAL; /* already set */
160		}
161		gna->save_if_input = ifp->if_input;
162		ifp->if_input = generic_rx_handler;
163	} else {
164		if (!gna->save_if_input){
165			D("cannot restore");
166			return EINVAL;  /* not saved */
167		}
168		ifp->if_input = gna->save_if_input;
169		gna->save_if_input = NULL;
170	}
171
172	return 0;
173}
174
175
176/*
177 * Intercept the packet steering routine in the tx path,
178 * so that we can decide which queue is used for an mbuf.
179 * Second argument is non-zero to intercept, 0 to restore.
180 * On freebsd we just intercept if_transmit.
181 */
182void
183netmap_catch_tx(struct netmap_generic_adapter *gna, int enable)
184{
185	struct netmap_adapter *na = &gna->up.up;
186	struct ifnet *ifp = na->ifp;
187
188	if (enable) {
189		na->if_transmit = ifp->if_transmit;
190		ifp->if_transmit = netmap_transmit;
191	} else {
192		ifp->if_transmit = na->if_transmit;
193	}
194}
195
196
197/*
198 * Transmit routine used by generic_netmap_txsync(). Returns 0 on success
199 * and non-zero on error (which may be packet drops or other errors).
200 * addr and len identify the netmap buffer, m is the (preallocated)
201 * mbuf to use for transmissions.
202 *
203 * We should add a reference to the mbuf so the m_freem() at the end
204 * of the transmission does not consume resources.
205 *
206 * On FreeBSD, and on multiqueue cards, we can force the queue using
207 *      if ((m->m_flags & M_FLOWID) != 0)
208 *              i = m->m_pkthdr.flowid % adapter->num_queues;
209 *      else
210 *              i = curcpu % adapter->num_queues;
211 *
212 */
213int
214generic_xmit_frame(struct ifnet *ifp, struct mbuf *m,
215	void *addr, u_int len, u_int ring_nr)
216{
217	int ret;
218
219	/*
220	 * The mbuf should be a cluster from our special pool,
221	 * so we do not need to do an m_copyback but just copy
222	 * (and eventually, just reference the netmap buffer)
223	 */
224
225	if (GET_MBUF_REFCNT(m) != 1) {
226		D("invalid refcnt %d for %p",
227			GET_MBUF_REFCNT(m), m);
228		panic("in generic_xmit_frame");
229	}
230	// XXX the ext_size check is unnecessary if we link the netmap buf
231	if (m->m_ext.ext_size < len) {
232		RD(5, "size %d < len %d", m->m_ext.ext_size, len);
233		len = m->m_ext.ext_size;
234	}
235	if (0) { /* XXX seems to have negligible benefits */
236		m->m_ext.ext_buf = m->m_data = addr;
237	} else {
238		bcopy(addr, m->m_data, len);
239	}
240	m->m_len = m->m_pkthdr.len = len;
241	// inc refcount. All ours, we could skip the atomic
242	atomic_fetchadd_int(PNT_MBUF_REFCNT(m), 1);
243	m->m_flags |= M_FLOWID;
244	m->m_pkthdr.flowid = ring_nr;
245	m->m_pkthdr.rcvif = ifp; /* used for tx notification */
246	ret = NA(ifp)->if_transmit(ifp, m);
247	return ret;
248}
249
250
251#if __FreeBSD_version >= 1100005
252struct netmap_adapter *
253netmap_getna(if_t ifp)
254{
255	return (NA((struct ifnet *)ifp));
256}
257#endif /* __FreeBSD_version >= 1100005 */
258
259/*
260 * The following two functions are empty until we have a generic
261 * way to extract the info from the ifp
262 */
263int
264generic_find_num_desc(struct ifnet *ifp, unsigned int *tx, unsigned int *rx)
265{
266	D("called, in tx %d rx %d", *tx, *rx);
267	return 0;
268}
269
270
271void
272generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq)
273{
274	D("called, in txq %d rxq %d", *txq, *rxq);
275	*txq = netmap_generic_rings;
276	*rxq = netmap_generic_rings;
277}
278
279
280void
281netmap_mitigation_init(struct nm_generic_mit *mit, int idx, struct netmap_adapter *na)
282{
283	ND("called");
284	mit->mit_pending = 0;
285	mit->mit_ring_idx = idx;
286	mit->mit_na = na;
287}
288
289
290void
291netmap_mitigation_start(struct nm_generic_mit *mit)
292{
293	ND("called");
294}
295
296
297void
298netmap_mitigation_restart(struct nm_generic_mit *mit)
299{
300	ND("called");
301}
302
303
304int
305netmap_mitigation_active(struct nm_generic_mit *mit)
306{
307	ND("called");
308	return 0;
309}
310
311
312void
313netmap_mitigation_cleanup(struct nm_generic_mit *mit)
314{
315	ND("called");
316}
317
318static int
319nm_vi_dummy(struct ifnet *ifp, u_long cmd, caddr_t addr)
320{
321	return EINVAL;
322}
323
324static void
325nm_vi_start(struct ifnet *ifp)
326{
327	panic("nm_vi_start() must not be called");
328}
329
330/*
331 * Index manager of persistent virtual interfaces.
332 * It is used to decide the lowest byte of the MAC address.
333 * We use the same algorithm with management of bridge port index.
334 */
335#define NM_VI_MAX	255
336static struct {
337	uint8_t index[NM_VI_MAX]; /* XXX just for a reasonable number */
338	uint8_t active;
339	struct mtx lock;
340} nm_vi_indices;
341
342void
343nm_vi_init_index(void)
344{
345	int i;
346	for (i = 0; i < NM_VI_MAX; i++)
347		nm_vi_indices.index[i] = i;
348	nm_vi_indices.active = 0;
349	mtx_init(&nm_vi_indices.lock, "nm_vi_indices_lock", NULL, MTX_DEF);
350}
351
352/* return -1 if no index available */
353static int
354nm_vi_get_index(void)
355{
356	int ret;
357
358	mtx_lock(&nm_vi_indices.lock);
359	ret = nm_vi_indices.active == NM_VI_MAX ? -1 :
360		nm_vi_indices.index[nm_vi_indices.active++];
361	mtx_unlock(&nm_vi_indices.lock);
362	return ret;
363}
364
365static void
366nm_vi_free_index(uint8_t val)
367{
368	int i, lim;
369
370	mtx_lock(&nm_vi_indices.lock);
371	lim = nm_vi_indices.active;
372	for (i = 0; i < lim; i++) {
373		if (nm_vi_indices.index[i] == val) {
374			/* swap index[lim-1] and j */
375			int tmp = nm_vi_indices.index[lim-1];
376			nm_vi_indices.index[lim-1] = val;
377			nm_vi_indices.index[i] = tmp;
378			nm_vi_indices.active--;
379			break;
380		}
381	}
382	if (lim == nm_vi_indices.active)
383		D("funny, index %u didn't found", val);
384	mtx_unlock(&nm_vi_indices.lock);
385}
386#undef NM_VI_MAX
387
388/*
389 * Implementation of a netmap-capable virtual interface that
390 * registered to the system.
391 * It is based on if_tap.c and ip_fw_log.c in FreeBSD 9.
392 *
393 * Note: Linux sets refcount to 0 on allocation of net_device,
394 * then increments it on registration to the system.
395 * FreeBSD sets refcount to 1 on if_alloc(), and does not
396 * increment this refcount on if_attach().
397 */
398int
399nm_vi_persist(const char *name, struct ifnet **ret)
400{
401	struct ifnet *ifp;
402	u_short macaddr_hi;
403	uint32_t macaddr_mid;
404	u_char eaddr[6];
405	int unit = nm_vi_get_index(); /* just to decide MAC address */
406
407	if (unit < 0)
408		return EBUSY;
409	/*
410	 * We use the same MAC address generation method with tap
411	 * except for the highest octet is 00:be instead of 00:bd
412	 */
413	macaddr_hi = htons(0x00be); /* XXX tap + 1 */
414	macaddr_mid = (uint32_t) ticks;
415	bcopy(&macaddr_hi, eaddr, sizeof(short));
416	bcopy(&macaddr_mid, &eaddr[2], sizeof(uint32_t));
417	eaddr[5] = (uint8_t)unit;
418
419	ifp = if_alloc(IFT_ETHER);
420	if (ifp == NULL) {
421		D("if_alloc failed");
422		return ENOMEM;
423	}
424	if_initname(ifp, name, IF_DUNIT_NONE);
425	ifp->if_mtu = 65536;
426	ifp->if_flags = IFF_UP | IFF_SIMPLEX | IFF_MULTICAST;
427	ifp->if_init = (void *)nm_vi_dummy;
428	ifp->if_ioctl = nm_vi_dummy;
429	ifp->if_start = nm_vi_start;
430	ifp->if_mtu = ETHERMTU;
431	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
432	ifp->if_capabilities |= IFCAP_LINKSTATE;
433	ifp->if_capenable |= IFCAP_LINKSTATE;
434
435	ether_ifattach(ifp, eaddr);
436	*ret = ifp;
437	return 0;
438}
439/* unregister from the system and drop the final refcount */
440void
441nm_vi_detach(struct ifnet *ifp)
442{
443	nm_vi_free_index(((char *)IF_LLADDR(ifp))[5]);
444	ether_ifdetach(ifp);
445	if_free(ifp);
446}
447
448/*
449 * In order to track whether pages are still mapped, we hook into
450 * the standard cdev_pager and intercept the constructor and
451 * destructor.
452 */
453
454struct netmap_vm_handle_t {
455	struct cdev 		*dev;
456	struct netmap_priv_d	*priv;
457};
458
459
460static int
461netmap_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
462    vm_ooffset_t foff, struct ucred *cred, u_short *color)
463{
464	struct netmap_vm_handle_t *vmh = handle;
465
466	if (netmap_verbose)
467		D("handle %p size %jd prot %d foff %jd",
468			handle, (intmax_t)size, prot, (intmax_t)foff);
469	dev_ref(vmh->dev);
470	return 0;
471}
472
473
474static void
475netmap_dev_pager_dtor(void *handle)
476{
477	struct netmap_vm_handle_t *vmh = handle;
478	struct cdev *dev = vmh->dev;
479	struct netmap_priv_d *priv = vmh->priv;
480
481	if (netmap_verbose)
482		D("handle %p", handle);
483	netmap_dtor(priv);
484	free(vmh, M_DEVBUF);
485	dev_rel(dev);
486}
487
488
489static int
490netmap_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
491	int prot, vm_page_t *mres)
492{
493	struct netmap_vm_handle_t *vmh = object->handle;
494	struct netmap_priv_d *priv = vmh->priv;
495	vm_paddr_t paddr;
496	vm_page_t page;
497	vm_memattr_t memattr;
498	vm_pindex_t pidx;
499
500	ND("object %p offset %jd prot %d mres %p",
501			object, (intmax_t)offset, prot, mres);
502	memattr = object->memattr;
503	pidx = OFF_TO_IDX(offset);
504	paddr = netmap_mem_ofstophys(priv->np_mref, offset);
505	if (paddr == 0)
506		return VM_PAGER_FAIL;
507
508	if (((*mres)->flags & PG_FICTITIOUS) != 0) {
509		/*
510		 * If the passed in result page is a fake page, update it with
511		 * the new physical address.
512		 */
513		page = *mres;
514		vm_page_updatefake(page, paddr, memattr);
515	} else {
516		/*
517		 * Replace the passed in reqpage page with our own fake page and
518		 * free up the all of the original pages.
519		 */
520#ifndef VM_OBJECT_WUNLOCK	/* FreeBSD < 10.x */
521#define VM_OBJECT_WUNLOCK VM_OBJECT_UNLOCK
522#define VM_OBJECT_WLOCK	VM_OBJECT_LOCK
523#endif /* VM_OBJECT_WUNLOCK */
524
525		VM_OBJECT_WUNLOCK(object);
526		page = vm_page_getfake(paddr, memattr);
527		VM_OBJECT_WLOCK(object);
528		vm_page_lock(*mres);
529		vm_page_free(*mres);
530		vm_page_unlock(*mres);
531		*mres = page;
532		vm_page_insert(page, object, pidx);
533	}
534	page->valid = VM_PAGE_BITS_ALL;
535	return (VM_PAGER_OK);
536}
537
538
539static struct cdev_pager_ops netmap_cdev_pager_ops = {
540	.cdev_pg_ctor = netmap_dev_pager_ctor,
541	.cdev_pg_dtor = netmap_dev_pager_dtor,
542	.cdev_pg_fault = netmap_dev_pager_fault,
543};
544
545
546static int
547netmap_mmap_single(struct cdev *cdev, vm_ooffset_t *foff,
548	vm_size_t objsize,  vm_object_t *objp, int prot)
549{
550	int error;
551	struct netmap_vm_handle_t *vmh;
552	struct netmap_priv_d *priv;
553	vm_object_t obj;
554
555	if (netmap_verbose)
556		D("cdev %p foff %jd size %jd objp %p prot %d", cdev,
557		    (intmax_t )*foff, (intmax_t )objsize, objp, prot);
558
559	vmh = malloc(sizeof(struct netmap_vm_handle_t), M_DEVBUF,
560			      M_NOWAIT | M_ZERO);
561	if (vmh == NULL)
562		return ENOMEM;
563	vmh->dev = cdev;
564
565	NMG_LOCK();
566	error = devfs_get_cdevpriv((void**)&priv);
567	if (error)
568		goto err_unlock;
569	vmh->priv = priv;
570	priv->np_refcount++;
571	NMG_UNLOCK();
572
573	error = netmap_get_memory(priv);
574	if (error)
575		goto err_deref;
576
577	obj = cdev_pager_allocate(vmh, OBJT_DEVICE,
578		&netmap_cdev_pager_ops, objsize, prot,
579		*foff, NULL);
580	if (obj == NULL) {
581		D("cdev_pager_allocate failed");
582		error = EINVAL;
583		goto err_deref;
584	}
585
586	*objp = obj;
587	return 0;
588
589err_deref:
590	NMG_LOCK();
591	priv->np_refcount--;
592err_unlock:
593	NMG_UNLOCK();
594// err:
595	free(vmh, M_DEVBUF);
596	return error;
597}
598
599
600// XXX can we remove this ?
601static int
602netmap_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
603{
604	if (netmap_verbose)
605		D("dev %p fflag 0x%x devtype %d td %p",
606			dev, fflag, devtype, td);
607	return 0;
608}
609
610
611static int
612netmap_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
613{
614	struct netmap_priv_d *priv;
615	int error;
616
617	(void)dev;
618	(void)oflags;
619	(void)devtype;
620	(void)td;
621
622	// XXX wait or nowait ?
623	priv = malloc(sizeof(struct netmap_priv_d), M_DEVBUF,
624			      M_NOWAIT | M_ZERO);
625	if (priv == NULL)
626		return ENOMEM;
627
628	error = devfs_set_cdevpriv(priv, netmap_dtor);
629	if (error)
630	        return error;
631
632	priv->np_refcount = 1;
633
634	return 0;
635}
636
637/******************** kqueue support ****************/
638
639/*
640 * The OS_selwakeup also needs to issue a KNOTE_UNLOCKED.
641 * We use a non-zero argument to distinguish the call from the one
642 * in kevent_scan() which instead also needs to run netmap_poll().
643 * The knote uses a global mutex for the time being. We might
644 * try to reuse the one in the si, but it is not allocated
645 * permanently so it might be a bit tricky.
646 *
647 * The *kqfilter function registers one or another f_event
648 * depending on read or write mode.
649 * In the call to f_event() td_fpop is NULL so any child function
650 * calling devfs_get_cdevpriv() would fail - and we need it in
651 * netmap_poll(). As a workaround we store priv into kn->kn_hook
652 * and pass it as first argument to netmap_poll(), which then
653 * uses the failure to tell that we are called from f_event()
654 * and do not need the selrecord().
655 */
656
657void freebsd_selwakeup(struct selinfo *si, int pri);
658
659void
660freebsd_selwakeup(struct selinfo *si, int pri)
661{
662	if (netmap_verbose)
663		D("on knote %p", &si->si_note);
664	selwakeuppri(si, pri);
665	/* use a non-zero hint to tell the notification from the
666	 * call done in kqueue_scan() which uses 0
667	 */
668	KNOTE_UNLOCKED(&si->si_note, 0x100 /* notification */);
669}
670
671static void
672netmap_knrdetach(struct knote *kn)
673{
674	struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook;
675	struct selinfo *si = priv->np_rxsi;
676
677	D("remove selinfo %p", si);
678	knlist_remove(&si->si_note, kn, 0);
679}
680
681static void
682netmap_knwdetach(struct knote *kn)
683{
684	struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook;
685	struct selinfo *si = priv->np_txsi;
686
687	D("remove selinfo %p", si);
688	knlist_remove(&si->si_note, kn, 0);
689}
690
691/*
692 * callback from notifies (generated externally) and our
693 * calls to kevent(). The former we just return 1 (ready)
694 * since we do not know better.
695 * In the latter we call netmap_poll and return 0/1 accordingly.
696 */
697static int
698netmap_knrw(struct knote *kn, long hint, int events)
699{
700	struct netmap_priv_d *priv;
701	int revents;
702
703	if (hint != 0) {
704		ND(5, "call from notify");
705		return 1; /* assume we are ready */
706	}
707	priv = kn->kn_hook;
708	/* the notification may come from an external thread,
709	 * in which case we do not want to run the netmap_poll
710	 * This should be filtered above, but check just in case.
711	 */
712	if (curthread != priv->np_td) { /* should not happen */
713		RD(5, "curthread changed %p %p", curthread, priv->np_td);
714		return 1;
715	} else {
716		revents = netmap_poll((void *)priv, events, curthread);
717		return (events & revents) ? 1 : 0;
718	}
719}
720
721static int
722netmap_knread(struct knote *kn, long hint)
723{
724	return netmap_knrw(kn, hint, POLLIN);
725}
726
727static int
728netmap_knwrite(struct knote *kn, long hint)
729{
730	return netmap_knrw(kn, hint, POLLOUT);
731}
732
733static struct filterops netmap_rfiltops = {
734	.f_isfd = 1,
735	.f_detach = netmap_knrdetach,
736	.f_event = netmap_knread,
737};
738
739static struct filterops netmap_wfiltops = {
740	.f_isfd = 1,
741	.f_detach = netmap_knwdetach,
742	.f_event = netmap_knwrite,
743};
744
745
746/*
747 * This is called when a thread invokes kevent() to record
748 * a change in the configuration of the kqueue().
749 * The 'priv' should be the same as in the netmap device.
750 */
751static int
752netmap_kqfilter(struct cdev *dev, struct knote *kn)
753{
754	struct netmap_priv_d *priv;
755	int error;
756	struct netmap_adapter *na;
757	struct selinfo *si;
758	int ev = kn->kn_filter;
759
760	if (ev != EVFILT_READ && ev != EVFILT_WRITE) {
761		D("bad filter request %d", ev);
762		return 1;
763	}
764	error = devfs_get_cdevpriv((void**)&priv);
765	if (error) {
766		D("device not yet setup");
767		return 1;
768	}
769	na = priv->np_na;
770	if (na == NULL) {
771		D("no netmap adapter for this file descriptor");
772		return 1;
773	}
774	/* the si is indicated in the priv */
775	si = (ev == EVFILT_WRITE) ? priv->np_txsi : priv->np_rxsi;
776	// XXX lock(priv) ?
777	kn->kn_fop = (ev == EVFILT_WRITE) ?
778		&netmap_wfiltops : &netmap_rfiltops;
779	kn->kn_hook = priv;
780	knlist_add(&si->si_note, kn, 1);
781	// XXX unlock(priv)
782	ND("register %p %s td %p priv %p kn %p np_nifp %p kn_fp/fpop %s",
783		na, na->ifp->if_xname, curthread, priv, kn,
784		priv->np_nifp,
785		kn->kn_fp == curthread->td_fpop ? "match" : "MISMATCH");
786	return 0;
787}
788
789struct cdevsw netmap_cdevsw = {
790	.d_version = D_VERSION,
791	.d_name = "netmap",
792	.d_open = netmap_open,
793	.d_mmap_single = netmap_mmap_single,
794	.d_ioctl = netmap_ioctl,
795	.d_poll = netmap_poll,
796	.d_kqfilter = netmap_kqfilter,
797	.d_close = netmap_close,
798};
799/*--- end of kqueue support ----*/
800
801/*
802 * Kernel entry point.
803 *
804 * Initialize/finalize the module and return.
805 *
806 * Return 0 on success, errno on failure.
807 */
808static int
809netmap_loader(__unused struct module *module, int event, __unused void *arg)
810{
811	int error = 0;
812
813	switch (event) {
814	case MOD_LOAD:
815		error = netmap_init();
816		break;
817
818	case MOD_UNLOAD:
819		netmap_fini();
820		break;
821
822	default:
823		error = EOPNOTSUPP;
824		break;
825	}
826
827	return (error);
828}
829
830
831DEV_MODULE(netmap, netmap_loader, NULL);
832