VBoxNetFlt-darwin.cpp@ 56743

Last change on this file since 56743 was 56723, checked in by vboxsync, 10 years ago
VBoxNetFlt: use RTmac instead of Rhxs format specifier when logging the MAC address to the release log
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 56.0 KB

Line
1	/* $Id: VBoxNetFlt-darwin.cpp 56723 2015-07-01 07:49:13Z vboxsync $ */
2	/** @file
3	* VBoxNetFlt - Network Filter Driver (Host), Darwin Specific Code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2015 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/*******************************************************************************
19	* Header Files *
20	*******************************************************************************/
21	/*
22	* Deal with conflicts first.
23	* PVM - BSD mess, that FreeBSD has correct a long time ago.
24	* iprt/types.h before sys/param.h - prevents UINT32_C and friends.
25	*/
26	#include <iprt/types.h>
27	#include <sys/param.h>
28	#undef PVM
29
30	#include <IOKit/IOLib.h> /* Assert as function */
31
32	#define LOG_GROUP LOG_GROUP_NET_FLT_DRV
33	#include <VBox/log.h>
34	#include <VBox/err.h>
35	#include <VBox/intnetinline.h>
36	#include <VBox/version.h>
37	#include <iprt/initterm.h>
38	#include <iprt/assert.h>
39	#include <iprt/spinlock.h>
40	#include <iprt/semaphore.h>
41	#include <iprt/process.h>
42	#include <iprt/alloc.h>
43	#include <iprt/alloca.h>
44	#include <iprt/time.h>
45	#include <iprt/net.h>
46	#include <iprt/thread.h>
47
48	#include "../../darwin/VBoxNetSend.h"
49
50	#include <mach/kmod.h>
51	#include <sys/conf.h>
52	#include <sys/errno.h>
53	#include <sys/ioccom.h>
54	#include <sys/filio.h>
55	#include <sys/malloc.h>
56	#include <sys/proc.h>
57	#include <sys/socket.h>
58	#include <sys/sockio.h>
59	#include <sys/kern_event.h>
60	#include <net/kpi_interface.h>
61	RT_C_DECLS_BEGIN /* Buggy 10.4 headers, fixed in 10.5. */
62	#include <sys/kpi_mbuf.h>
63	#include <net/kpi_interfacefilter.h>
64	RT_C_DECLS_END
65
66	#include <sys/kpi_socket.h>
67	#include <net/if.h>
68	#include <net/if_var.h>
69	#include <netinet/in.h>
70	#include <netinet/in_var.h>
71	#include <netinet6/in6_var.h>
72
73	#define VBOXNETFLT_OS_SPECFIC 1
74	#include "../VBoxNetFltInternal.h"
75
76
77	/*******************************************************************************
78	* Defined Constants And Macros *
79	*******************************************************************************/
80	/** The maximum number of SG segments.
81	* Used to prevent stack overflow and similar bad stuff. */
82	#define VBOXNETFLT_DARWIN_MAX_SEGS 32
83
84	#if 0
85	/** For testing extremely segmented frames. */
86	#define VBOXNETFLT_DARWIN_TEST_SEG_SIZE 14
87	#endif
88
89	/* XXX: hidden undef #ifdef __APPLE__ */
90	#define VBOX_IN_LOOPBACK(addr) (((addr) & IN_CLASSA_NET) == 0x7f000000)
91	#define VBOX_IN_LINKLOCAL(addr) (((addr) & IN_CLASSB_NET) == 0xa9fe0000)
92
93
94
95	/*******************************************************************************
96	* Internal Functions *
97	*******************************************************************************/
98	RT_C_DECLS_BEGIN
99	static kern_return_t VBoxNetFltDarwinStart(struct kmod_info pKModInfo, void pvData);
100	static kern_return_t VBoxNetFltDarwinStop(struct kmod_info pKModInfo, void pvData);
101
102	static void vboxNetFltDarwinSysSockUpcall(socket_t pSysSock, void *pvData, int fWait);
103	RT_C_DECLS_END
104
105
106	/*******************************************************************************
107	* Structures and Typedefs *
108	*******************************************************************************/
109	/**
110	* The mbuf tag data.
111	*
112	* We have to associate the ethernet header with each packet we're sending
113	* because things like icmp will inherit the tag it self so the tag along
114	* isn't sufficient to identify our mbufs. For the icmp scenario the ethernet
115	* header naturally changes before the packet is send pack, so let check it.
116	*/
117	typedef struct VBOXNETFLTTAG
118	{
119	/** The ethernet header of the outgoing frame. */
120	RTNETETHERHDR EthHdr;
121	} VBOXNETFLTTAG;
122	/** Pointer to a VBoxNetFlt mbuf tag. */
123	typedef VBOXNETFLTTAG *PVBOXNETFLTTAG;
124	/** Pointer to a const VBoxNetFlt mbuf tag. */
125	typedef VBOXNETFLTTAG const *PCVBOXNETFLTTAG;
126
127
128	/*******************************************************************************
129	* Global Variables *
130	*******************************************************************************/
131	/**
132	* Declare the module stuff.
133	*/
134	RT_C_DECLS_BEGIN
135	extern kern_return_t _start(struct kmod_info pKModInfo, void pvData);
136	extern kern_return_t _stop(struct kmod_info pKModInfo, void pvData);
137
138	KMOD_EXPLICIT_DECL(VBoxNetFlt, VBOX_VERSION_STRING, _start, _stop)
139	DECLHIDDEN(kmod_start_func_t *) _realmain = VBoxNetFltDarwinStart;
140	DECLHIDDEN(kmod_stop_func_t *) _antimain = VBoxNetFltDarwinStop;
141	DECLHIDDEN(int) _kext_apple_cc = __APPLE_CC__;
142	RT_C_DECLS_END
143
144
145	/**
146	* The (common) global data.
147	*/
148	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
149
150	/** The unique tag id for this module.
151	* This is basically a unique string hash that lives on until reboot.
152	* It is used for tagging mbufs. */
153	static mbuf_tag_id_t g_idTag;
154
155	/** The offset of the struct ifnet::if_pcount variable.
156	* @remarks Initial value is valid for Lion and earlier. We adjust it on attach
157	* for later releases. */
158	static unsigned g_offIfNetPCount = sizeof(void ) (1 /if_softc/ + 1 /if_name/ + 2 /if_link/ + 2 /if_addrhead/ + 1 /if_check_multi/)
159	+ sizeof(u_long) /if_refcnt/;
160	/** Macro for accessing ifnet::if_pcount. */
161	#define VBOX_GET_PCOUNT(pIfNet) ( (int )((uintptr_t)pIfNet + g_offIfNetPCount) )
162	/** The size of area of ifnet structure we try to locate if_pcount in. */
163	#define VBOXNETFLT_DARWIN_IFNET_SIZE 256
164	/** Indicates whether g_offIfNetPCount has been adjusted already (no point in
165	* doing it more than once). */
166	static bool g_fNetPCountFound = false;
167
168
169	/**
170	* Change the promiscuous setting and try spot the changed in @a pIfNet.
171	*
172	* @returns Offset of potential p_count field.
173	* @param pIfNet The interface we're attaching to.
174	* @param iPromisc Whether to enable (1) or disable (0) promiscuous mode.
175	*
176	* @note This implementation relies on if_pcount to be aligned on sizeof(int).
177	*/
178	static unsigned vboxNetFltDarwinSetAndDiff(ifnet_t pIfNet, int iPromisc)
179	{
180	int aiSavedState[VBOXNETFLT_DARWIN_IFNET_SIZE / sizeof(int)];
181	memcpy(aiSavedState, pIfNet, sizeof(aiSavedState));
182
183	ifnet_set_promiscuous(pIfNet, iPromisc);
184
185	int const iDiff = iPromisc ? 1 : -1;
186
187	/*
188	* We assume that ifnet structure will never have less members in front of if_pcount
189	* than it used to have in Lion. If this turns out to be false assumption we will
190	* have to start from zero offset.
191	*/
192	for (unsigned i = g_offIfNetPCount / sizeof(int); i < RT_ELEMENTS(aiSavedState); i++)
193	if (((int*)pIfNet)[i] - aiSavedState[i] == iDiff)
194	return i * sizeof(int);
195
196	return 0;
197	}
198
199
200	/**
201	* Detect and adjust the offset of ifnet::if_pcount.
202	*
203	* @param pIfNet The interface we're attaching to.
204	*/
205	static void vboxNetFltDarwinDetectPCountOffset(ifnet_t pIfNet)
206	{
207	if (g_fNetPCountFound)
208	return;
209
210	/*
211	* It would be nice to use locking at this point, but it is not available via KPI.
212	* This is why we try several times. At each attempt we modify if_pcount four times
213	* to rule out false detections.
214	*/
215	unsigned offTry1, offTry2, offTry3, offTry4;
216	for (int iAttempt = 0; iAttempt < 3; iAttempt++)
217	{
218	offTry1 = vboxNetFltDarwinSetAndDiff(pIfNet, 1);
219	offTry2 = vboxNetFltDarwinSetAndDiff(pIfNet, 1);
220	offTry3 = vboxNetFltDarwinSetAndDiff(pIfNet, 0);
221	offTry4 = vboxNetFltDarwinSetAndDiff(pIfNet, 0);
222	if (offTry1 == offTry2 && offTry2 == offTry3 && offTry3 == offTry4)
223	{
224	if (g_offIfNetPCount != offTry1)
225	{
226	Log(("VBoxNetFltDarwinDetectPCountOffset: Adjusted if_pcount offset to %x from %x.\n", offTry1, g_offIfNetPCount));
227	g_offIfNetPCount = offTry1;
228	g_fNetPCountFound = true;
229	}
230	break;
231	}
232	}
233
234	if (g_offIfNetPCount != offTry1)
235	LogRel(("VBoxNetFlt: Failed to detect promiscuous count, all traffic may reach wire (%x != %x).\n", g_offIfNetPCount, offTry1));
236	}
237
238
239	/**
240	* Start the kernel module.
241	*/
242	static kern_return_t VBoxNetFltDarwinStart(struct kmod_info pKModInfo, void pvData)
243	{
244	int rc;
245
246	/*
247	* Initialize IPRT and find our module tag id.
248	* (IPRT is shared with VBoxDrv, it creates the loggers.)
249	*/
250	rc = RTR0Init(0);
251	if (RT_SUCCESS(rc))
252	{
253	Log(("VBoxNetFltDarwinStart\n"));
254	errno_t err = mbuf_tag_id_find("org.VirtualBox.kext.VBoxFltDrv", &g_idTag);
255	if (!err)
256	{
257	/*
258	* Initialize the globals and connect to the support driver.
259	*
260	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
261	* for establishing the connect to the support driver.
262	*/
263	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
264	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
265	if (RT_SUCCESS(rc))
266	{
267	LogRel(("VBoxFltDrv: version " VBOX_VERSION_STRING " r%d\n", VBOX_SVN_REV));
268	return KMOD_RETURN_SUCCESS;
269	}
270
271	LogRel(("VBoxFltDrv: failed to initialize device extension (rc=%d)\n", rc));
272	}
273	else
274	LogRel(("VBoxFltDrv: mbuf_tag_id_find failed, err=%d\n", err));
275	RTR0Term();
276	}
277	else
278	printf("VBoxFltDrv: failed to initialize IPRT (rc=%d)\n", rc);
279
280	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
281	return KMOD_RETURN_FAILURE;
282	}
283
284
285	/**
286	* Stop the kernel module.
287	*/
288	static kern_return_t VBoxNetFltDarwinStop(struct kmod_info pKModInfo, void pvData)
289	{
290	Log(("VBoxNetFltDarwinStop\n"));
291
292	/*
293	* Refuse to unload if anyone is currently using the filter driver.
294	* This is important as I/O kit / xnu will to be able to do usage
295	* tracking for us!
296	*/
297	int rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
298	if (RT_FAILURE(rc))
299	{
300	Log(("VBoxNetFltDarwinStop - failed, busy.\n"));
301	return KMOD_RETURN_FAILURE;
302	}
303
304	/*
305	* Undo the work done during start (in reverse order).
306	*/
307	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
308
309	RTR0Term();
310
311	return KMOD_RETURN_SUCCESS;
312	}
313
314
315	/**
316	* Reads and retains the host interface handle.
317	*
318	* @returns The handle, NULL if detached.
319	* @param pThis
320	*/
321	DECLINLINE(ifnet_t) vboxNetFltDarwinRetainIfNet(PVBOXNETFLTINS pThis)
322	{
323	ifnet_t pIfNet = NULL;
324
325	/*
326	* Be careful here to avoid problems racing the detached callback.
327	*/
328	RTSpinlockAcquire(pThis->hSpinlock);
329	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
330	{
331	pIfNet = ASMAtomicUoReadPtrT(&pThis->u.s.pIfNet, ifnet_t);
332	if (pIfNet)
333	ifnet_reference(pIfNet);
334	}
335	RTSpinlockRelease(pThis->hSpinlock);
336
337	return pIfNet;
338	}
339
340
341	/**
342	* Release the host interface handle previously retained
343	* by vboxNetFltDarwinRetainIfNet.
344	*
345	* @param pThis The instance.
346	* @param pIfNet The vboxNetFltDarwinRetainIfNet return value, NULL is fine.
347	*/
348	DECLINLINE(void) vboxNetFltDarwinReleaseIfNet(PVBOXNETFLTINS pThis, ifnet_t pIfNet)
349	{
350	NOREF(pThis);
351	if (pIfNet)
352	ifnet_release(pIfNet);
353	}
354
355
356	/**
357	* Checks whether this is an mbuf created by vboxNetFltDarwinMBufFromSG,
358	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
359	*
360	* @returns true / false accordingly.
361	* @param pThis The instance.
362	* @param pMBuf The mbuf.
363	* @param pvFrame The frame pointer, optional.
364	*/
365	DECLINLINE(bool) vboxNetFltDarwinMBufIsOur(PVBOXNETFLTINS pThis, mbuf_t pMBuf, void *pvFrame)
366	{
367	NOREF(pThis);
368
369	/*
370	* Lookup the tag set by vboxNetFltDarwinMBufFromSG.
371	*/
372	PCVBOXNETFLTTAG pTagData;
373	size_t cbTagData;
374	errno_t err = mbuf_tag_find(pMBuf, g_idTag, 0 /* type /, &cbTagData, (void *)&pTagData);
375	if (err)
376	return false;
377	AssertReturn(cbTagData == sizeof(*pTagData), false);
378
379	/*
380	* Dig out the ethernet header from the mbuf.
381	*/
382	PCRTNETETHERHDR pEthHdr = (PCRTNETETHERHDR)pvFrame;
383	if (!pEthHdr)
384	pEthHdr = (PCRTNETETHERHDR)mbuf_pkthdr_header(pMBuf);
385	if (!pEthHdr)
386	pEthHdr = (PCRTNETETHERHDR)mbuf_data(pMBuf);
387	/* ASSUMING that there is enough data to work on! */
388	if ( pEthHdr->DstMac.au8[0] != pTagData->EthHdr.DstMac.au8[0]
389	\|\| pEthHdr->DstMac.au8[1] != pTagData->EthHdr.DstMac.au8[1]
390	\|\| pEthHdr->DstMac.au8[2] != pTagData->EthHdr.DstMac.au8[2]
391	\|\| pEthHdr->DstMac.au8[3] != pTagData->EthHdr.DstMac.au8[3]
392	\|\| pEthHdr->DstMac.au8[4] != pTagData->EthHdr.DstMac.au8[4]
393	\|\| pEthHdr->DstMac.au8[5] != pTagData->EthHdr.DstMac.au8[5]
394	\|\| pEthHdr->SrcMac.au8[0] != pTagData->EthHdr.SrcMac.au8[0]
395	\|\| pEthHdr->SrcMac.au8[1] != pTagData->EthHdr.SrcMac.au8[1]
396	\|\| pEthHdr->SrcMac.au8[2] != pTagData->EthHdr.SrcMac.au8[2]
397	\|\| pEthHdr->SrcMac.au8[3] != pTagData->EthHdr.SrcMac.au8[3]
398	\|\| pEthHdr->SrcMac.au8[4] != pTagData->EthHdr.SrcMac.au8[4]
399	\|\| pEthHdr->SrcMac.au8[5] != pTagData->EthHdr.SrcMac.au8[5]
400	\|\| pEthHdr->EtherType != pTagData->EthHdr.EtherType)
401	{
402	Log3(("tagged, but the ethernet header has changed\n"));
403	return false;
404	}
405
406	return true;
407	}
408
409
410	/**
411	* Internal worker that create a darwin mbuf for a (scatter/)gather list.
412	*
413	* @returns Pointer to the mbuf.
414	* @param pThis The instance.
415	* @param pSG The (scatter/)gather list.
416	*/
417	static mbuf_t vboxNetFltDarwinMBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG)
418	{
419	/// @todo future? mbuf_how_t How = preemption enabled ? MBUF_DONTWAIT : MBUF_WAITOK;
420	mbuf_how_t How = MBUF_WAITOK;
421
422	/*
423	* We need some way of getting back to our instance data when
424	* the mbuf is freed, so use pvUserData for this.
425	* -- this is not relevant anylonger! --
426	*/
427	Assert(!pSG->pvUserData \|\| pSG->pvUserData == pThis);
428	Assert(!pSG->pvUserData2);
429	pSG->pvUserData = pThis;
430
431	/*
432	* Allocate a packet and copy over the data.
433	*
434	* Using mbuf_attachcluster() here would've been nice but there are two
435	* issues with it: (1) it's 10.5.x only, and (2) the documentation indicates
436	* that it's not supposed to be used for really external buffers. The 2nd
437	* point might be argued against considering that the only m_clattach user
438	* is mallocs memory for the ext mbuf and not doing what's stated in the docs.
439	* However, it's hard to tell if these m_clattach buffers actually makes it
440	* to the NICs or not, and even if they did, the NIC would need the physical
441	* addresses for the pages they contain and might end up copying the data
442	* to a new mbuf anyway.
443	*
444	* So, in the end it's better to just do it the simple way that will work
445	* 100%, even if it involves some extra work (alloc + copy) we really wished
446	* to avoid.
447	*
448	* Note. We can't make use of the physical addresses on darwin because the
449	* way the mbuf / cluster stuff works (see mbuf_data_to_physical and
450	* mcl_to_paddr).
451	*/
452	mbuf_t pPkt = NULL;
453	errno_t err = mbuf_allocpacket(How, pSG->cbTotal, NULL, &pPkt);
454	if (!err)
455	{
456	/* Skip zero sized memory buffers (paranoia). */
457	mbuf_t pCur = pPkt;
458	while (pCur && !mbuf_maxlen(pCur))
459	pCur = mbuf_next(pCur);
460	Assert(pCur);
461
462	/* Set the required packet header attributes. */
463	mbuf_pkthdr_setlen(pPkt, pSG->cbTotal);
464	mbuf_pkthdr_setheader(pPkt, mbuf_data(pCur));
465
466	/* Special case the single buffer copy. */
467	if ( mbuf_next(pCur)
468	&& mbuf_maxlen(pCur) >= pSG->cbTotal)
469	{
470	mbuf_setlen(pCur, pSG->cbTotal);
471	IntNetSgRead(pSG, mbuf_data(pCur));
472	}
473	else
474	{
475	/* Multi buffer copying. */
476	size_t cbLeft = pSG->cbTotal;
477	size_t offSrc = 0;
478	while (cbLeft > 0 && pCur)
479	{
480	size_t cb = mbuf_maxlen(pCur);
481	if (cb > cbLeft)
482	cb = cbLeft;
483	mbuf_setlen(pCur, cb);
484	IntNetSgReadEx(pSG, offSrc, cb, mbuf_data(pCur));
485
486	/* advance */
487	offSrc += cb;
488	cbLeft -= cb;
489	pCur = mbuf_next(pCur);
490	}
491	Assert(cbLeft == 0);
492	}
493	if (!err)
494	{
495	/*
496	* Tag the packet and return successfully.
497	*/
498	PVBOXNETFLTTAG pTagData;
499	err = mbuf_tag_allocate(pPkt, g_idTag, 0 /* type /, sizeof(VBOXNETFLTTAG) / tag len /, How, (void *)&pTagData);
500	if (!err)
501	{
502	Assert(pSG->aSegs[0].cb >= sizeof(pTagData->EthHdr));
503	memcpy(&pTagData->EthHdr, pSG->aSegs[0].pv, sizeof(pTagData->EthHdr));
504	return pPkt;
505	}
506
507	/* bailout: */
508	AssertMsg(err == ENOMEM \|\| err == EWOULDBLOCK, ("err=%d\n", err));
509	}
510
511	mbuf_freem(pPkt);
512	}
513	else
514	AssertMsg(err == ENOMEM \|\| err == EWOULDBLOCK, ("err=%d\n", err));
515	pSG->pvUserData = NULL;
516
517	return NULL;
518	}
519
520
521	/**
522	* Calculates the number of segments required to represent the mbuf.
523	*
524	* @returns Number of segments.
525	* @param pThis The instance.
526	* @param pMBuf The mbuf.
527	* @param pvFrame The frame pointer, optional.
528	*/
529	DECLINLINE(unsigned) vboxNetFltDarwinMBufCalcSGSegs(PVBOXNETFLTINS pThis, mbuf_t pMBuf, void *pvFrame)
530	{
531	NOREF(pThis);
532
533	/*
534	* Count the buffers in the chain.
535	*/
536	unsigned cSegs = 0;
537	for (mbuf_t pCur = pMBuf; pCur; pCur = mbuf_next(pCur))
538	if (mbuf_len(pCur))
539	cSegs++;
540	else if ( !cSegs
541	&& pvFrame
542	&& (uintptr_t)pvFrame - (uintptr_t)mbuf_datastart(pMBuf) < mbuf_maxlen(pMBuf))
543	cSegs++;
544
545	#ifdef PADD_RUNT_FRAMES_FROM_HOST
546	/*
547	* Add one buffer if the total is less than the ethernet minimum 60 bytes.
548	* This may allocate a segment too much if the ethernet header is separated,
549	* but that shouldn't harm us much.
550	*/
551	if (mbuf_pkthdr_len(pMBuf) < 60)
552	cSegs++;
553	#endif
554
555	#ifdef VBOXNETFLT_DARWIN_TEST_SEG_SIZE
556	/* maximize the number of segments. */
557	cSegs = RT_MAX(VBOXNETFLT_DARWIN_MAX_SEGS - 1, cSegs);
558	#endif
559
560	return cSegs ? cSegs : 1;
561	}
562
563
564	/**
565	* Initializes a SG list from an mbuf.
566	*
567	* @returns Number of segments.
568	* @param pThis The instance.
569	* @param pMBuf The mbuf.
570	* @param pSG The SG.
571	* @param pvFrame The frame pointer, optional.
572	* @param cSegs The number of segments allocated for the SG.
573	* This should match the number in the mbuf exactly!
574	* @param fSrc The source of the frame.
575	*/
576	DECLINLINE(void) vboxNetFltDarwinMBufToSG(PVBOXNETFLTINS pThis, mbuf_t pMBuf, void *pvFrame, PINTNETSG pSG, unsigned cSegs, uint32_t fSrc)
577	{
578	NOREF(pThis);
579
580	/*
581	* Walk the chain and convert the buffers to segments. Works INTNETSG::cbTotal.
582	*/
583	unsigned iSeg = 0;
584	IntNetSgInitTempSegs(pSG, 0 /cbTotal/, cSegs, 0 /cSegsUsed/);
585	for (mbuf_t pCur = pMBuf; pCur; pCur = mbuf_next(pCur))
586	{
587	size_t cbSeg = mbuf_len(pCur);
588	if (cbSeg)
589	{
590	void *pvSeg = mbuf_data(pCur);
591
592	/* deal with pvFrame */
593	if (!iSeg && pvFrame && pvFrame != pvSeg)
594	{
595	void *pvStart = mbuf_datastart(pMBuf);
596	uintptr_t offSeg = (uintptr_t)pvSeg - (uintptr_t)pvStart;
597	uintptr_t offSegEnd = offSeg + cbSeg;
598	Assert(pvStart && pvSeg && offSeg < mbuf_maxlen(pMBuf) && offSegEnd <= mbuf_maxlen(pMBuf)); NOREF(offSegEnd);
599	uintptr_t offFrame = (uintptr_t)pvFrame - (uintptr_t)pvStart;
600	if (RT_LIKELY(offFrame < offSeg))
601	{
602	pvSeg = pvFrame;
603	cbSeg += offSeg - offFrame;
604	}
605	else
606	AssertMsgFailed(("pvFrame=%p pvStart=%p pvSeg=%p offSeg=%p cbSeg=%#zx offSegEnd=%p offFrame=%p maxlen=%#zx\n",
607	pvFrame, pvStart, pvSeg, offSeg, cbSeg, offSegEnd, offFrame, mbuf_maxlen(pMBuf)));
608	pvFrame = NULL;
609	}
610
611	AssertBreak(iSeg < cSegs);
612	pSG->cbTotal += cbSeg;
613	pSG->aSegs[iSeg].cb = cbSeg;
614	pSG->aSegs[iSeg].pv = pvSeg;
615	pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
616	iSeg++;
617	}
618	/* The pvFrame might be in a now empty buffer. */
619	else if ( !iSeg
620	&& pvFrame
621	&& (uintptr_t)pvFrame - (uintptr_t)mbuf_datastart(pMBuf) < mbuf_maxlen(pMBuf))
622	{
623	cbSeg = (uintptr_t)mbuf_datastart(pMBuf) + mbuf_maxlen(pMBuf) - (uintptr_t)pvFrame;
624	pSG->cbTotal += cbSeg;
625	pSG->aSegs[iSeg].cb = cbSeg;
626	pSG->aSegs[iSeg].pv = pvFrame;
627	pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
628	iSeg++;
629	pvFrame = NULL;
630	}
631	}
632
633	Assert(iSeg && iSeg <= cSegs);
634	pSG->cSegsUsed = iSeg;
635
636	#ifdef PADD_RUNT_FRAMES_FROM_HOST
637	/*
638	* Add a trailer if the frame is too small.
639	*
640	* Since we're getting to the packet before it is framed, it has not
641	* yet been padded. The current solution is to add a segment pointing
642	* to a buffer containing all zeros and pray that works for all frames...
643	*/
644	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
645	{
646	AssertReturnVoid(iSeg < cSegs);
647
648	static uint8_t const s_abZero[128] = {0};
649	pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
650	pSG->aSegs[iSeg].pv = (void *)&s_abZero[0];
651	pSG->aSegs[iSeg].cb = 60 - pSG->cbTotal;
652	pSG->cbTotal = 60;
653	pSG->cSegsUsed++;
654	}
655	#endif
656
657	#ifdef VBOXNETFLT_DARWIN_TEST_SEG_SIZE
658	/*
659	* Redistribute the segments.
660	*/
661	if (pSG->cSegsUsed < pSG->cSegsAlloc)
662	{
663	/* copy the segments to the end. */
664	int iSrc = pSG->cSegsUsed;
665	int iDst = pSG->cSegsAlloc;
666	while (iSrc > 0)
667	{
668	iDst--;
669	iSrc--;
670	pSG->aSegs[iDst] = pSG->aSegs[iSrc];
671	}
672
673	/* create small segments from the start. */
674	pSG->cSegsUsed = pSG->cSegsAlloc;
675	iSrc = iDst;
676	iDst = 0;
677	while ( iDst < iSrc
678	&& iDst < pSG->cSegsAlloc)
679	{
680	pSG->aSegs[iDst].Phys = NIL_RTHCPHYS;
681	pSG->aSegs[iDst].pv = pSG->aSegs[iSrc].pv;
682	pSG->aSegs[iDst].cb = RT_MIN(pSG->aSegs[iSrc].cb, VBOXNETFLT_DARWIN_TEST_SEG_SIZE);
683	if (pSG->aSegs[iDst].cb != pSG->aSegs[iSrc].cb)
684	{
685	pSG->aSegs[iSrc].cb -= pSG->aSegs[iDst].cb;
686	pSG->aSegs[iSrc].pv = (uint8_t *)pSG->aSegs[iSrc].pv + pSG->aSegs[iDst].cb;
687	}
688	else if (++iSrc >= pSG->cSegsAlloc)
689	{
690	pSG->cSegsUsed = iDst + 1;
691	break;
692	}
693	iDst++;
694	}
695	}
696	#endif
697
698	AssertMsg(!pvFrame, ("pvFrame=%p pMBuf=%p iSeg=%d\n", pvFrame, pMBuf, iSeg));
699	}
700
701
702	/**
703	* Helper for determining whether the host wants the interface to be
704	* promiscuous.
705	*/
706	static bool vboxNetFltDarwinIsPromiscuous(PVBOXNETFLTINS pThis)
707	{
708	bool fRc = false;
709	ifnet_t pIfNet = vboxNetFltDarwinRetainIfNet(pThis);
710	if (pIfNet)
711	{
712	/* gather the data */
713	uint16_t fIf = ifnet_flags(pIfNet);
714	unsigned cPromisc = VBOX_GET_PCOUNT(pIfNet);
715	bool fSetPromiscuous = ASMAtomicUoReadBool(&pThis->u.s.fSetPromiscuous);
716	vboxNetFltDarwinReleaseIfNet(pThis, pIfNet);
717
718	/* calc the return. */
719	fRc = (fIf & IFF_PROMISC)
720	&& cPromisc > fSetPromiscuous;
721	}
722	return fRc;
723	}
724
725
726
727	/**
728	*
729	* @see iff_detached_func in the darwin kpi.
730	*/
731	static void vboxNetFltDarwinIffDetached(void *pvThis, ifnet_t pIfNet)
732	{
733	PVBOXNETFLTINS pThis = (PVBOXNETFLTINS)pvThis;
734	uint64_t NanoTS = RTTimeSystemNanoTS();
735	LogFlow(("vboxNetFltDarwinIffDetached: pThis=%p NanoTS=%RU64 (%d)\n",
736	pThis, NanoTS, VALID_PTR(pIfNet) ? VBOX_GET_PCOUNT(pIfNet) : -1));
737
738	Assert(!pThis->fDisconnectedFromHost);
739	Assert(!pThis->fRediscoveryPending);
740
741	/*
742	* If we've put it into promiscuous mode, undo that now. If we don't
743	* the if_pcount will go all wrong when it's replugged.
744	*/
745	if (ASMAtomicXchgBool(&pThis->u.s.fSetPromiscuous, false))
746	ifnet_set_promiscuous(pIfNet, 0);
747
748	/*
749	* We carefully take the spinlock and increase the interface reference
750	* behind it in order to avoid problematic races with the detached callback.
751	*/
752	RTSpinlockAcquire(pThis->hSpinlock);
753
754	pIfNet = ASMAtomicUoReadPtrT(&pThis->u.s.pIfNet, ifnet_t);
755	int cPromisc = VALID_PTR(pIfNet) ? VBOX_GET_PCOUNT(pIfNet) : - 1;
756
757	ASMAtomicUoWriteNullPtr(&pThis->u.s.pIfNet);
758	ASMAtomicUoWriteNullPtr(&pThis->u.s.pIfFilter);
759	ASMAtomicWriteBool(&pThis->u.s.fNeedSetPromiscuous, false);
760	pThis->u.s.fSetPromiscuous = false;
761	ASMAtomicUoWriteU64(&pThis->NanoTSLastRediscovery, NanoTS);
762	ASMAtomicUoWriteBool(&pThis->fRediscoveryPending, false);
763	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
764
765	RTSpinlockRelease(pThis->hSpinlock);
766
767	if (pIfNet)
768	ifnet_release(pIfNet);
769	LogRel(("VBoxNetFlt: was detached from '%s' (%d)\n", pThis->szName, cPromisc));
770	}
771
772
773	/**
774	*
775	* @see iff_ioctl_func in the darwin kpi.
776	*/
777	static errno_t vboxNetFltDarwinIffIoCtl(void pvThis, ifnet_t pIfNet, protocol_family_t eProtocol, u_long uCmd, void pvArg)
778	{
779	PVBOXNETFLTINS pThis = (PVBOXNETFLTINS)pvThis;
780	LogFlow(("vboxNetFltDarwinIffIoCtl: pThis=%p uCmd=%lx\n", pThis, uCmd));
781
782	/*
783	* Update fOtherPromiscuous.
784	*/
785	/** @todo we'll have to find the offset of if_pcount to get this right! */
786	//if (uCmd == SIOCSIFFLAGS)
787	//{
788	//
789	//}
790
791	/*
792	* We didn't handle it, continue processing.
793	*/
794	NOREF(pThis);
795	NOREF(eProtocol);
796	NOREF(uCmd);
797	NOREF(pvArg);
798	return EOPNOTSUPP;
799	}
800
801
802	/**
803	*
804	* @see iff_event_func in the darwin kpi.
805	*/
806	static void vboxNetFltDarwinIffEvent(void pvThis, ifnet_t pIfNet, protocol_family_t eProtocol, const struct kev_msg pEvMsg)
807	{
808	PVBOXNETFLTINS pThis = (PVBOXNETFLTINS)pvThis;
809	LogFlow(("vboxNetFltDarwinIffEvent: pThis=%p\n", pThis));
810
811	NOREF(pThis);
812	NOREF(pIfNet);
813	NOREF(eProtocol);
814	NOREF(pEvMsg);
815
816	/*
817	* Watch out for the interface going online / offline.
818	*/
819	if ( VALID_PTR(pThis)
820	&& VALID_PTR(pEvMsg)
821	&& pEvMsg->vendor_code == KEV_VENDOR_APPLE
822	&& pEvMsg->kev_class == KEV_NETWORK_CLASS
823	&& pEvMsg->kev_subclass == KEV_DL_SUBCLASS)
824	{
825	if (pThis->u.s.pIfNet == pIfNet)
826	{
827	if (pEvMsg->event_code == KEV_DL_LINK_ON)
828	{
829	if (ASMAtomicUoReadBool(&pThis->u.s.fNeedSetPromiscuous))
830	{
831	/* failed to bring it online. */
832	errno_t err = ifnet_set_promiscuous(pIfNet, 1);
833	if (!err)
834	{
835	ASMAtomicWriteBool(&pThis->u.s.fSetPromiscuous, true);
836	ASMAtomicWriteBool(&pThis->u.s.fNeedSetPromiscuous, false);
837	Log(("vboxNetFltDarwinIffEvent: enabled promiscuous mode on %s (%d)\n", pThis->szName, VBOX_GET_PCOUNT(pIfNet)));
838	}
839	else
840	Log(("vboxNetFltDarwinIffEvent: ifnet_set_promiscuous failed on %s, err=%d (%d)\n", pThis->szName, err, VBOX_GET_PCOUNT(pIfNet)));
841	}
842	else if ( ASMAtomicUoReadBool(&pThis->u.s.fSetPromiscuous)
843	&& !(ifnet_flags(pIfNet) & IFF_PROMISC))
844	{
845	/* Try fix the inconsistency. */
846	errno_t err = ifnet_set_flags(pIfNet, IFF_PROMISC, IFF_PROMISC);
847	if (!err)
848	err = ifnet_ioctl(pIfNet, 0, SIOCSIFFLAGS, NULL);
849	if (!err && (ifnet_flags(pIfNet) & IFF_PROMISC))
850	Log(("vboxNetFltDarwinIffEvent: fixed IFF_PROMISC on %s (%d)\n", pThis->szName, VBOX_GET_PCOUNT(pIfNet)));
851	else
852	Log(("vboxNetFltDarwinIffEvent: failed to fix IFF_PROMISC on %s, err=%d flags=%#x (%d)\n",
853	pThis->szName, err, ifnet_flags(pIfNet), VBOX_GET_PCOUNT(pIfNet)));
854	}
855	else
856	Log(("vboxNetFltDarwinIffEvent: online, '%s'. flags=%#x (%d)\n", pThis->szName, ifnet_flags(pIfNet), VBOX_GET_PCOUNT(pIfNet)));
857	}
858	else if (pEvMsg->event_code == KEV_DL_LINK_OFF)
859	Log(("vboxNetFltDarwinIffEvent: %s goes down (%d)\n", pThis->szName, VBOX_GET_PCOUNT(pIfNet)));
860	/** @todo KEV_DL_LINK_ADDRESS_CHANGED -> pfnReportMacAddress */
861	/** @todo KEV_DL_SIFFLAGS -> pfnReportPromiscuousMode */
862	}
863	else
864	Log(("vboxNetFltDarwinIffEvent: pThis->u.s.pIfNet=%p pIfNet=%p (%d)\n", pThis->u.s.pIfNet, pIfNet, VALID_PTR(pIfNet) ? VBOX_GET_PCOUNT(pIfNet) : -1));
865	}
866	else if (VALID_PTR(pEvMsg))
867	Log(("vboxNetFltDarwinIffEvent: vendor_code=%#x kev_class=%#x kev_subclass=%#x event_code=%#x\n",
868	pEvMsg->vendor_code, pEvMsg->kev_class, pEvMsg->kev_subclass, pEvMsg->event_code));
869	}
870
871
872	/**
873	* Internal worker for vboxNetFltDarwinIffInput and vboxNetFltDarwinIffOutput,
874	*
875	* @returns 0 or EJUSTRETURN.
876	* @param pThis The instance.
877	* @param pMBuf The mbuf.
878	* @param pvFrame The start of the frame, optional.
879	* @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
880	* @param eProtocol The protocol.
881	*/
882	static errno_t vboxNetFltDarwinIffInputOutputWorker(PVBOXNETFLTINS pThis, mbuf_t pMBuf, void *pvFrame,
883	uint32_t fSrc, protocol_family_t eProtocol)
884	{
885	/*
886	* Drop it immediately?
887	*/
888	Log2(("vboxNetFltDarwinIffInputOutputWorker: pThis=%p pMBuf=%p pvFrame=%p fSrc=%#x cbPkt=%x\n",
889	pThis, pMBuf, pvFrame, fSrc, pMBuf ? mbuf_pkthdr_len(pMBuf) : -1));
890	if (!pMBuf)
891	return 0;
892	#if 0 /* debugging lost icmp packets */
893	if (mbuf_pkthdr_len(pMBuf) > 0x300)
894	{
895	uint8_t pb = (uint8_t )(pvFrame ? pvFrame : mbuf_data(pMBuf));
896	Log3(("D=%.6Rhxs S=%.6Rhxs T=%04x IFF\n", pb, pb + 6, RT_BE2H_U16((uint16_t )(pb + 12))));
897	}
898	#endif
899	if (vboxNetFltDarwinMBufIsOur(pThis, pMBuf, pvFrame))
900	return 0;
901
902	/*
903	* Active? Retain the instance and increment the busy counter.
904	*/
905	if (!vboxNetFltTryRetainBusyActive(pThis))
906	return 0;
907
908	/*
909	* Finalize out-bound packets since the stack puts off finalizing
910	* TCP/IP checksums as long as possible.
911	* ASSUMES this only applies to outbound IP packets.
912	*/
913	if ( (fSrc & INTNETTRUNKDIR_HOST)
914	&& eProtocol == PF_INET)
915	{
916	Assert(!pvFrame);
917	mbuf_outbound_finalize(pMBuf, eProtocol, sizeof(RTNETETHERHDR));
918	}
919
920	/*
921	* Create a (scatter/)gather list for the mbuf and feed it to the internal network.
922	*/
923	bool fDropIt = false;
924	unsigned cSegs = vboxNetFltDarwinMBufCalcSGSegs(pThis, pMBuf, pvFrame);
925	if (cSegs < VBOXNETFLT_DARWIN_MAX_SEGS)
926	{
927	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
928	vboxNetFltDarwinMBufToSG(pThis, pMBuf, pvFrame, pSG, cSegs, fSrc);
929
930	fDropIt = pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
931	if (fDropIt)
932	{
933	/*
934	* Check if this interface is in promiscuous mode. We should not drop
935	* any packets before they get to the driver as it passes them to tap
936	* callbacks in order for BPF to work properly.
937	*/
938	if (vboxNetFltDarwinIsPromiscuous(pThis))
939	fDropIt = false;
940	else
941	mbuf_freem(pMBuf);
942	}
943	}
944
945	vboxNetFltRelease(pThis, true /* fBusy */);
946
947	return fDropIt ? EJUSTRETURN : 0;
948	}
949
950
951	/**
952	* From the host.
953	*
954	* @see iff_output_func in the darwin kpi.
955	*/
956	static errno_t vboxNetFltDarwinIffOutput(void pvThis, ifnet_t pIfNet, protocol_family_t eProtocol, mbuf_t ppMBuf)
957	{
958	/** @todo there was some note about the ethernet header here or something like that... */
959
960	NOREF(eProtocol);
961	NOREF(pIfNet);
962	return vboxNetFltDarwinIffInputOutputWorker((PVBOXNETFLTINS)pvThis, *ppMBuf, NULL, INTNETTRUNKDIR_HOST, eProtocol);
963	}
964
965
966	/**
967	* From the wire.
968	*
969	* @see iff_input_func in the darwin kpi.
970	*/
971	static errno_t vboxNetFltDarwinIffInput(void pvThis, ifnet_t pIfNet, protocol_family_t eProtocol, mbuf_t ppMBuf, char **ppchFrame)
972	{
973	NOREF(eProtocol);
974	NOREF(pIfNet);
975	return vboxNetFltDarwinIffInputOutputWorker((PVBOXNETFLTINS)pvThis, ppMBuf, ppchFrame, INTNETTRUNKDIR_WIRE, eProtocol);
976	}
977
978
979	/** A worker thread for vboxNetFltSendDummy(). */
980	static DECLCALLBACK(int) vboxNetFltSendDummyWorker(RTTHREAD hThreadSelf, void *pvUser)
981	{
982	Assert(pvUser);
983	ifnet_t pIfNet = (ifnet_t)pvUser;
984	return VBoxNetSendDummy(pIfNet);
985	}
986
987
988	/**
989	* Prevent GUI icon freeze issue when VirtualBoxVM process terminates.
990	*
991	* This function is a workaround for stuck-in-dock issue. The idea here is to
992	* send a dummy packet to an interface from the context of a kernel thread.
993	* Therefore, an XNU's receive thread (which is created as a result if we are
994	* the first who is communicating with the interface) will be associated with
995	* the kernel thread instead of VirtualBoxVM process.
996	*
997	* @param pIfNet Interface to be used to send data.
998	*/
999	static void vboxNetFltSendDummy(ifnet_t pIfNet)
1000	{
1001	RTTHREAD hThread;
1002	int rc = RTThreadCreate(&hThread, vboxNetFltSendDummyWorker, (void *)pIfNet, 0,
1003	RTTHREADTYPE_DEFAULT, RTTHREADFLAGS_WAITABLE, "DummyThread");
1004	if (RT_SUCCESS(rc))
1005	{
1006	RTThreadWait(hThread, RT_INDEFINITE_WAIT, NULL);
1007	LogFlow(("vboxNetFltSendDummy: a dummy packet has been successfully sent in order to prevent stuck-in-dock issue\n"));
1008	}
1009	else
1010	LogFlow(("vboxNetFltSendDummy: unable to send dummy packet in order to prevent stuck-in-dock issue\n"));
1011	}
1012
1013
1014	/**
1015	* Internal worker for vboxNetFltOsInitInstance and vboxNetFltOsMaybeRediscovered.
1016	*
1017	* @returns VBox status code.
1018	* @param pThis The instance.
1019	* @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1020	* flood the release log.
1021	*/
1022	static int vboxNetFltDarwinAttachToInterface(PVBOXNETFLTINS pThis, bool fRediscovery)
1023	{
1024	LogFlow(("vboxNetFltDarwinAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1025
1026	/*
1027	* Locate the interface first.
1028	*
1029	* The pIfNet member is updated before iflt_attach is called and used
1030	* to deal with the hypothetical case where someone rips out the
1031	* interface immediately after our iflt_attach call.
1032	*/
1033	ifnet_t pIfNet = NULL;
1034	errno_t err = ifnet_find_by_name(pThis->szName, &pIfNet);
1035	if (err)
1036	{
1037	Assert(err == ENXIO);
1038	if (!fRediscovery)
1039	LogRel(("VBoxFltDrv: failed to find ifnet '%s' (err=%d)\n", pThis->szName, err));
1040	else
1041	Log(("VBoxFltDrv: failed to find ifnet '%s' (err=%d)\n", pThis->szName, err));
1042	return VERR_INTNET_FLT_IF_NOT_FOUND;
1043	}
1044
1045	RTSpinlockAcquire(pThis->hSpinlock);
1046	ASMAtomicUoWritePtr(&pThis->u.s.pIfNet, pIfNet);
1047	RTSpinlockRelease(pThis->hSpinlock);
1048
1049	/* Adjust g_offIfNetPCount as it varies for different versions of xnu. */
1050	vboxNetFltDarwinDetectPCountOffset(pIfNet);
1051
1052	/* Prevent stuck-in-dock issue by associating interface receive thread with kernel thread. */
1053	vboxNetFltSendDummy(pIfNet);
1054
1055	/*
1056	* Get the mac address while we still have a valid ifnet reference.
1057	*/
1058	err = ifnet_lladdr_copy_bytes(pIfNet, &pThis->u.s.MacAddr, sizeof(pThis->u.s.MacAddr));
1059	if (!err)
1060	{
1061	/*
1062	* Try attach the filter.
1063	*/
1064	struct iff_filter RegRec;
1065	RegRec.iff_cookie = pThis;
1066	RegRec.iff_name = "VBoxNetFlt";
1067	RegRec.iff_protocol = 0;
1068	RegRec.iff_input = vboxNetFltDarwinIffInput;
1069	RegRec.iff_output = vboxNetFltDarwinIffOutput;
1070	RegRec.iff_event = vboxNetFltDarwinIffEvent;
1071	RegRec.iff_ioctl = vboxNetFltDarwinIffIoCtl;
1072	RegRec.iff_detached = vboxNetFltDarwinIffDetached;
1073	interface_filter_t pIfFilter = NULL;
1074	err = iflt_attach(pIfNet, &RegRec, &pIfFilter);
1075	Assert(err \|\| pIfFilter);
1076
1077	RTSpinlockAcquire(pThis->hSpinlock);
1078	pIfNet = ASMAtomicUoReadPtrT(&pThis->u.s.pIfNet, ifnet_t);
1079	if (pIfNet && !err)
1080	{
1081	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1082	ASMAtomicUoWritePtr(&pThis->u.s.pIfFilter, pIfFilter);
1083	pIfNet = NULL; /* don't dereference it */
1084	}
1085	RTSpinlockRelease(pThis->hSpinlock);
1086
1087	/* Report capabilities. */
1088	if ( !pIfNet
1089	&& vboxNetFltTryRetainBusyNotDisconnected(pThis))
1090	{
1091	Assert(pThis->pSwitchPort);
1092	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1093	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltDarwinIsPromiscuous(pThis));
1094	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, 0, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1095	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, 0 /* none */);
1096	vboxNetFltRelease(pThis, true /fBusy/);
1097	}
1098	}
1099
1100	/* Release the interface on failure. */
1101	if (pIfNet)
1102	ifnet_release(pIfNet);
1103
1104	int rc = RTErrConvertFromErrno(err);
1105	if (RT_SUCCESS(rc))
1106	LogRel(("VBoxFltDrv: attached to '%s' / %RTmac\n", pThis->szName, &pThis->u.s.MacAddr));
1107	else
1108	LogRel(("VBoxFltDrv: failed to attach to ifnet '%s' (err=%d)\n", pThis->szName, err));
1109	return rc;
1110	}
1111
1112
1113	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
1114	{
1115	vboxNetFltDarwinAttachToInterface(pThis, true /* fRediscovery */);
1116	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
1117	}
1118
1119
1120	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
1121	{
1122	NOREF(pvIfData);
1123
1124	int rc = VINF_SUCCESS;
1125	ifnet_t pIfNet = vboxNetFltDarwinRetainIfNet(pThis);
1126	if (pIfNet)
1127	{
1128	/*
1129	* Create a mbuf for the gather list and push it onto the wire.
1130	*
1131	* Note! If the interface is in the promiscuous mode we need to send the
1132	* packet down the stack so it reaches the driver and Berkeley
1133	* Packet Filter (see @bugref{5817}).
1134	*/
1135	if ((fDst & INTNETTRUNKDIR_WIRE) \|\| vboxNetFltDarwinIsPromiscuous(pThis))
1136	{
1137	mbuf_t pMBuf = vboxNetFltDarwinMBufFromSG(pThis, pSG);
1138	if (pMBuf)
1139	{
1140	errno_t err = ifnet_output_raw(pIfNet, PF_LINK, pMBuf);
1141	if (err)
1142	rc = RTErrConvertFromErrno(err);
1143	}
1144	else
1145	rc = VERR_NO_MEMORY;
1146	}
1147
1148	/*
1149	* Create a mbuf for the gather list and push it onto the host stack.
1150	*/
1151	if (fDst & INTNETTRUNKDIR_HOST)
1152	{
1153	mbuf_t pMBuf = vboxNetFltDarwinMBufFromSG(pThis, pSG);
1154	if (pMBuf)
1155	{
1156	/* This is what IONetworkInterface::inputPacket does. */
1157	unsigned const cbEthHdr = 14;
1158	mbuf_pkthdr_setheader(pMBuf, mbuf_data(pMBuf));
1159	mbuf_pkthdr_setlen(pMBuf, mbuf_pkthdr_len(pMBuf) - cbEthHdr);
1160	mbuf_setdata(pMBuf, (uint8_t *)mbuf_data(pMBuf) + cbEthHdr, mbuf_len(pMBuf) - cbEthHdr);
1161	mbuf_pkthdr_setrcvif(pMBuf, pIfNet); /* will crash without this. */
1162
1163	errno_t err = ifnet_input(pIfNet, pMBuf, NULL);
1164	if (err)
1165	rc = RTErrConvertFromErrno(err);
1166	}
1167	else
1168	rc = VERR_NO_MEMORY;
1169	}
1170
1171	vboxNetFltDarwinReleaseIfNet(pThis, pIfNet);
1172	}
1173
1174	return rc;
1175	}
1176
1177
1178	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
1179	{
1180	ifnet_t pIfNet = vboxNetFltDarwinRetainIfNet(pThis);
1181	if (pIfNet)
1182	{
1183	if (pThis->fDisablePromiscuous)
1184	{
1185	/*
1186	* Promiscuous mode should not be used (wireless), we just need to
1187	* make sure the interface is up.
1188	*/
1189	if (fActive)
1190	{
1191	u_int16_t fIf = ifnet_flags(pIfNet);
1192	if ((fIf & (IFF_UP \| IFF_RUNNING)) != (IFF_UP \| IFF_RUNNING))
1193	{
1194	ifnet_set_flags(pIfNet, IFF_UP, IFF_UP);
1195	ifnet_ioctl(pIfNet, 0, SIOCSIFFLAGS, NULL);
1196	}
1197	}
1198	}
1199	else
1200	{
1201	/*
1202	* This api is a bit weird, the best reference is the code.
1203	*
1204	* Also, we have a bit or race conditions wrt the maintenance of
1205	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
1206	*/
1207	unsigned const cPromiscBefore = VBOX_GET_PCOUNT(pIfNet);
1208	u_int16_t fIf;
1209	if (fActive)
1210	{
1211	Assert(!pThis->u.s.fSetPromiscuous);
1212	errno_t err = ENETDOWN;
1213	ASMAtomicWriteBool(&pThis->u.s.fNeedSetPromiscuous, true);
1214
1215	/*
1216	* Try bring the interface up and running if it's down.
1217	*/
1218	fIf = ifnet_flags(pIfNet);
1219	if ((fIf & (IFF_UP \| IFF_RUNNING)) != (IFF_UP \| IFF_RUNNING))
1220	{
1221	err = ifnet_set_flags(pIfNet, IFF_UP, IFF_UP);
1222	errno_t err2 = ifnet_ioctl(pIfNet, 0, SIOCSIFFLAGS, NULL);
1223	if (!err)
1224	err = err2;
1225	fIf = ifnet_flags(pIfNet);
1226	}
1227
1228	/*
1229	* Is it already up? If it isn't, leave it to the link event or
1230	* we'll upset if_pcount (as stated above, ifnet_set_promiscuous is weird).
1231	*/
1232	if ((fIf & (IFF_UP \| IFF_RUNNING)) == (IFF_UP \| IFF_RUNNING))
1233	{
1234	err = ifnet_set_promiscuous(pIfNet, 1);
1235	pThis->u.s.fSetPromiscuous = err == 0;
1236	if (!err)
1237	{
1238	ASMAtomicWriteBool(&pThis->u.s.fNeedSetPromiscuous, false);
1239
1240	/* check if it actually worked, this stuff is not always behaving well. */
1241	if (!(ifnet_flags(pIfNet) & IFF_PROMISC))
1242	{
1243	err = ifnet_set_flags(pIfNet, IFF_PROMISC, IFF_PROMISC);
1244	if (!err)
1245	err = ifnet_ioctl(pIfNet, 0, SIOCSIFFLAGS, NULL);
1246	if (!err)
1247	Log(("vboxNetFlt: fixed IFF_PROMISC on %s (%d->%d)\n", pThis->szName, cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1248	else
1249	Log(("VBoxNetFlt: failed to fix IFF_PROMISC on %s, err=%d (%d->%d)\n",
1250	pThis->szName, err, cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1251	}
1252	}
1253	else
1254	Log(("VBoxNetFlt: ifnet_set_promiscuous -> err=%d grr! (%d->%d)\n", err, cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1255	}
1256	else if (!err)
1257	Log(("VBoxNetFlt: Waiting for the link to come up... (%d->%d)\n", cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1258	if (err)
1259	LogRel(("VBoxNetFlt: Failed to put '%s' into promiscuous mode, err=%d (%d->%d)\n", pThis->szName, err, cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1260	}
1261	else
1262	{
1263	ASMAtomicWriteBool(&pThis->u.s.fNeedSetPromiscuous, false);
1264	if (pThis->u.s.fSetPromiscuous)
1265	{
1266	errno_t err = ifnet_set_promiscuous(pIfNet, 0);
1267	AssertMsg(!err, ("%d\n", err)); NOREF(err);
1268	}
1269	pThis->u.s.fSetPromiscuous = false;
1270
1271	fIf = ifnet_flags(pIfNet);
1272	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1273	}
1274	}
1275
1276	vboxNetFltDarwinReleaseIfNet(pThis, pIfNet);
1277	}
1278	}
1279
1280
1281	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
1282	{
1283	/* Nothing to do here. */
1284	return VINF_SUCCESS;
1285	}
1286
1287
1288	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
1289	{
1290	/* Nothing to do here. */
1291	return VINF_SUCCESS;
1292	}
1293
1294
1295	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
1296	{
1297	interface_filter_t pIfFilter;
1298
1299	/*
1300	* Carefully obtain the interface filter reference and detach it.
1301	*/
1302	RTSpinlockAcquire(pThis->hSpinlock);
1303	pIfFilter = ASMAtomicUoReadPtrT(&pThis->u.s.pIfFilter, interface_filter_t);
1304	if (pIfFilter)
1305	ASMAtomicUoWriteNullPtr(&pThis->u.s.pIfFilter);
1306	RTSpinlockRelease(pThis->hSpinlock);
1307
1308	if (pIfFilter)
1309	iflt_detach(pIfFilter);
1310
1311	if (pThis->u.s.pSysSock != NULL)
1312	{
1313	sock_close(pThis->u.s.pSysSock);
1314	pThis->u.s.pSysSock = NULL;
1315	}
1316	}
1317
1318
1319	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
1320	{
1321	NOREF(pvContext);
1322
1323	int rc = vboxNetFltDarwinAttachToInterface(pThis, false /* fRediscovery */);
1324	if (RT_FAILURE(rc))
1325	return rc;
1326
1327	if (pThis->pSwitchPort->pfnNotifyHostAddress == NULL)
1328	return rc;
1329
1330	/*
1331	* XXX: uwe
1332	*
1333	* Learn host's IP addresses and set up notifications for changes.
1334	* To avoid racing, set up notifications first.
1335	*
1336	* XXX: This should probably be global, since the only thing
1337	* specific to ifnet here is its IPv6 link-local address.
1338	*/
1339	errno_t error;
1340
1341	error = sock_socket(PF_SYSTEM, SOCK_RAW, SYSPROTO_EVENT,
1342	vboxNetFltDarwinSysSockUpcall, pThis,
1343	&pThis->u.s.pSysSock);
1344	if (error != 0)
1345	{
1346	LogRel(("sock_socket(SYSPROTO_EVENT): error %d\n", error));
1347	return rc;
1348	}
1349
1350	int nbio = 1;
1351	error = sock_ioctl(pThis->u.s.pSysSock, FIONBIO, &nbio);
1352	if (error != 0)
1353	{
1354	LogRel(("FIONBIO: error %d\n", error));
1355	sock_close(pThis->u.s.pSysSock);
1356	return rc;
1357	}
1358
1359	if (!sock_isnonblocking(pThis->u.s.pSysSock))
1360	{
1361	LogRel(("FIONBIO ok, but socket is blocking?!\n"));
1362	sock_close(pThis->u.s.pSysSock);
1363	return rc;
1364	}
1365
1366	struct kev_request req;
1367	req.vendor_code = KEV_VENDOR_APPLE;
1368	req.kev_class = KEV_NETWORK_CLASS;
1369	req.kev_subclass = KEV_ANY_SUBCLASS; /* need both INET and INET6, so have to request all */
1370
1371	error = sock_ioctl(pThis->u.s.pSysSock, SIOCSKEVFILT, &req);
1372	if (error != 0)
1373	{
1374	LogRel(("SIOCSKEVFILT: error %d\n", error));
1375	sock_close(pThis->u.s.pSysSock);
1376	return rc;
1377	}
1378
1379	ifnet_t pIfNet = pThis->u.s.pIfNet; /* already retained */
1380
1381	ifaddr_t *pIfAddrList;
1382	error = ifnet_get_address_list(/* all interfaces*/ NULL, &pIfAddrList);
1383	if (error != 0)
1384	{
1385	LogRel(("ifnet_get_address_list: error %d\n", error));
1386	return rc;
1387	}
1388
1389	for (ifaddr_t pIfAddr = pIfAddrList; pIfAddr != NULL; ++pIfAddr)
1390	{
1391	ifaddr_t ifa = *pIfAddr;
1392	sa_family_t family = ifaddr_address_family(ifa);
1393	struct sockaddr_storage ss;
1394
1395	error = ifaddr_address(ifa, (struct sockaddr *)&ss, sizeof(ss));
1396	if (error != 0)
1397	{
1398	LogRel(("getting address family %d: error %d\n", family, error));
1399	continue;
1400	}
1401
1402	if (family == AF_INET)
1403	{
1404	struct sockaddr_in sin = (struct sockaddr_in )&ss;
1405	u_int32_t u32Addr = ntohl(sin->sin_addr.s_addr);
1406
1407	if (VBOX_IN_LOOPBACK(u32Addr))
1408	continue;
1409
1410	if (ifaddr_ifnet(ifa) != pIfNet && VBOX_IN_LINKLOCAL(u32Addr))
1411	continue;
1412
1413	Log(("> inet %RTnaipv4\n", sin->sin_addr.s_addr));
1414	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1415	/* :fAdded */ true, kIntNetAddrType_IPv4, &sin->sin_addr);
1416	}
1417	else if (family == AF_INET6)
1418	{
1419	struct sockaddr_in6 sin6 = (struct sockaddr_in6 )&ss;
1420
1421	if (IN6_IS_ADDR_LOOPBACK(&sin6->sin6_addr))
1422	continue;
1423
1424	/* link-local from other interfaces are out of scope */
1425	if (ifaddr_ifnet(ifa) != pIfNet && IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr))
1426	continue;
1427
1428	Log(("> inet6 %RTnaipv6\n", &sin6->sin6_addr));
1429	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1430	/* :fAdded */ true, kIntNetAddrType_IPv6, &sin6->sin6_addr);
1431	}
1432	}
1433
1434	ifnet_free_address_list(pIfAddrList);
1435
1436	/*
1437	* Now that we've got current addresses, check for events that
1438	* might have happened while we were working.
1439	*/
1440	vboxNetFltDarwinSysSockUpcall(pThis->u.s.pSysSock, pThis, MBUF_DONTWAIT);
1441
1442	return rc;
1443	}
1444
1445
1446	static void vboxNetFltDarwinSysSockUpcall(socket_t pSysSock, void *pvData, int fWait)
1447	{
1448	PVBOXNETFLTINS pThis = (PVBOXNETFLTINS)pvData;
1449	errno_t error;
1450
1451	NOREF(fWait);
1452
1453	if (RT_UNLIKELY(pSysSock != pThis->u.s.pSysSock))
1454	{
1455	Log(("vboxNetFltDarwinSysSockUpcall: %p != %p?\n",
1456	pSysSock, pThis->u.s.pSysSock));
1457	return;
1458	}
1459
1460	struct net_event_data my_link;
1461	ifnet_t pIfNet = pThis->u.s.pIfNet; /* XXX: retain? */
1462	ifnet_family_t if_family = ifnet_family(pIfNet);
1463	u_int32_t if_unit = ifnet_unit(pIfNet);
1464
1465	for (;;) {
1466	mbuf_t m;
1467	size_t len = sizeof(struct kern_event_msg) - sizeof(u_int32_t)
1468	+ sizeof(struct kev_in6_data);
1469
1470	error = sock_receivembuf(pSysSock, NULL, &m, 0, &len);
1471	if (error == EWOULDBLOCK)
1472	{
1473	Log(("vboxNetFltDarwinSysSockUpcall: EWOULDBLOCK - we are done\n"));
1474	error = 0;
1475	break;
1476	}
1477	else if (error != 0)
1478	{
1479	Log(("sock_receivembuf: error %d\n", error));
1480	break;
1481	}
1482
1483	if (len < sizeof(struct kern_event_msg) - sizeof(u_int32_t))
1484	{
1485	Log(("vboxNetFltDarwinSysSockUpcall: %u bytes is too short\n",
1486	(unsigned int)len));
1487	mbuf_freem(m);
1488	return;
1489	}
1490
1491	struct kern_event_msg msg = (struct kern_event_msg )mbuf_data(m);
1492	if (msg->kev_subclass == KEV_INET_SUBCLASS)
1493	{
1494	if (len - (sizeof(struct kern_event_msg) - sizeof(u_int32_t)) < sizeof(struct kev_in_data))
1495	{
1496	Log(("vboxNetFltDarwinSysSockUpcall: %u bytes is too short for KEV_INET_SUBCLASS\n",
1497	(unsigned int)len));
1498	mbuf_freem(m);
1499	return;
1500	}
1501
1502	struct kev_in_data iev = (struct kev_in_data )msg->event_data;
1503	struct net_event_data *link = &iev->link_data;
1504	PCRTNETADDRU pAddr = (PCRTNETADDRU)&iev->ia_addr;
1505	u_int32_t u32Addr = ntohl(pAddr->IPv4.u);
1506
1507	if (VBOX_IN_LOOPBACK(u32Addr))
1508	{
1509	mbuf_freem(m);
1510	continue;
1511	}
1512
1513	if ( (link->if_family != if_family \|\| link->if_unit != if_unit)
1514	&& VBOX_IN_LINKLOCAL(u32Addr))
1515	{
1516	mbuf_freem(m);
1517	continue;
1518	}
1519
1520	switch (msg->event_code)
1521	{
1522	case KEV_INET_NEW_ADDR:
1523	Log(("KEV_INET_NEW_ADDR %.*s%d: %RTnaipv4\n",
1524	IFNAMSIZ, link->if_name, link->if_unit, pAddr->IPv4.u));
1525
1526	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1527	/* :fAdded */ true, kIntNetAddrType_IPv4, pAddr);
1528	break;
1529
1530	case KEV_INET_ADDR_DELETED:
1531	Log(("KEV_INET_ADDR_DELETED %.*s%d: %RTnaipv4\n",
1532	IFNAMSIZ, link->if_name, link->if_unit, pAddr->IPv4.u));
1533
1534	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1535	/* :fAdded */ false, kIntNetAddrType_IPv4, pAddr);
1536	break;
1537
1538	default:
1539	Log(("KEV INET event %u %.*s%d: addr %RTnaipv4\n",
1540	msg->event_code, IFNAMSIZ, link->if_name, link->if_unit, pAddr->IPv4.u));
1541	break;
1542	}
1543	}
1544	else if (msg->kev_subclass == KEV_INET6_SUBCLASS)
1545	{
1546	if (len - (sizeof(struct kern_event_msg) - sizeof(u_int32_t)) < sizeof(struct kev_in6_data))
1547	{
1548	Log(("vboxNetFltDarwinSysSockUpcall: %u bytes is too short for KEV_INET6_SUBCLASS\n",
1549	(unsigned int)len));
1550	mbuf_freem(m);
1551	return;
1552	}
1553
1554	struct kev_in6_data iev6 = (struct kev_in6_data )msg->event_data;
1555	struct net_event_data *link = &iev6->link_data;
1556	PCRTNETADDRU pAddr = (PCRTNETADDRU)&iev6->ia_addr.sin6_addr;
1557
1558	if (IN6_IS_ADDR_LOOPBACK(&iev6->ia_addr.sin6_addr))
1559	{
1560	mbuf_freem(m);
1561	continue;
1562	}
1563
1564	if ( (link->if_family != if_family \|\| link->if_unit != if_unit)
1565	&& IN6_IS_ADDR_LINKLOCAL(&iev6->ia_addr.sin6_addr))
1566	{
1567	mbuf_freem(m);
1568	continue;
1569	}
1570
1571
1572
1573	switch (msg->event_code)
1574	{
1575	case KEV_INET6_NEW_USER_ADDR:
1576	Log(("KEV_INET6_NEW_USER_ADDR %.*s%d: %RTnaipv6\n",
1577	IFNAMSIZ, link->if_name, link->if_unit, pAddr));
1578	goto kev_inet6_new;
1579
1580	case KEV_INET6_NEW_LL_ADDR:
1581	Log(("KEV_INET6_NEW_LL_ADDR %.*s%d: %RTnaipv6\n",
1582	IFNAMSIZ, link->if_name, link->if_unit, pAddr));
1583	goto kev_inet6_new;
1584
1585	case KEV_INET6_NEW_RTADV_ADDR:
1586	Log(("KEV_INET6_NEW_RTADV_ADDR %.*s%d: %RTnaipv6\n",
1587	IFNAMSIZ, link->if_name, link->if_unit, pAddr));
1588	goto kev_inet6_new;
1589
1590	kev_inet6_new:
1591	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1592	/* :fAdded */ true, kIntNetAddrType_IPv6, pAddr);
1593	break;
1594
1595	case KEV_INET6_ADDR_DELETED:
1596	Log(("KEV_INET6_ADDR_DELETED %.*s%d: %RTnaipv6\n",
1597	IFNAMSIZ, link->if_name, link->if_unit, pAddr));
1598
1599	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1600	/* :fAdded */ false, kIntNetAddrType_IPv6, pAddr);
1601	break;
1602
1603	default:
1604	Log(("KEV INET6 event %u %.*s%d: addr %RTnaipv6\n",
1605	msg->event_code, IFNAMSIZ, link->if_name, link->if_unit, pAddr));
1606	break;
1607	}
1608	}
1609	else
1610	{
1611	Log(("vboxNetFltDarwinSysSockUpcall: subclass %u ignored\n",
1612	(unsigned)msg->kev_subclass));
1613	}
1614
1615	mbuf_freem(m);
1616	}
1617	}
1618
1619
1620	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
1621	{
1622	/*
1623	* Init the darwin specific members.
1624	*/
1625	pThis->u.s.pIfNet = NULL;
1626	pThis->u.s.pIfFilter = NULL;
1627	pThis->u.s.fSetPromiscuous = false;
1628	pThis->u.s.fNeedSetPromiscuous = false;
1629	//pThis->u.s.MacAddr = {0};
1630	pThis->u.s.pSysSock = NULL;
1631
1632	return VINF_SUCCESS;
1633	}
1634
1635
1636	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
1637	{
1638	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
1639	}
1640
1641
1642	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *ppvIfData)
1643	{
1644	/* Nothing to do */
1645	NOREF(pThis); NOREF(pvIf); NOREF(ppvIfData);
1646	return VINF_SUCCESS;
1647	}
1648
1649
1650	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
1651	{
1652	/* Nothing to do */
1653	NOREF(pThis); NOREF(pvIfData);
1654	return VINF_SUCCESS;
1655	}
1656

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/darwin/VBoxNetFlt-darwin.cpp@ 56743

Download in other formats: