VBoxNetFlt-darwin.cpp@ 56266

Last change on this file since 56266 was 56266, checked in by vboxsync, 9 years ago
VBoxNetFlt-darwin: vboxNetFltDarwinSysSockUpcall - log the interface from the net_event_data of the event.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 54.9 KB

Line
1	/* $Id: VBoxNetFlt-darwin.cpp 56266 2015-06-07 00:11:00Z vboxsync $ */
2	/** @file
3	* VBoxNetFlt - Network Filter Driver (Host), Darwin Specific Code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2014 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
90	/*******************************************************************************
91	* Internal Functions *
92	*******************************************************************************/
93	RT_C_DECLS_BEGIN
94	static kern_return_t VBoxNetFltDarwinStart(struct kmod_info pKModInfo, void pvData);
95	static kern_return_t VBoxNetFltDarwinStop(struct kmod_info pKModInfo, void pvData);
96
97	static void vboxNetFltDarwinSysSockUpcall(socket_t pSysSock, void *pvData, int fWait);
98	RT_C_DECLS_END
99
100
101	/*******************************************************************************
102	* Structures and Typedefs *
103	*******************************************************************************/
104	/**
105	* The mbuf tag data.
106	*
107	* We have to associate the ethernet header with each packet we're sending
108	* because things like icmp will inherit the tag it self so the tag along
109	* isn't sufficient to identify our mbufs. For the icmp scenario the ethernet
110	* header naturally changes before the packet is send pack, so let check it.
111	*/
112	typedef struct VBOXNETFLTTAG
113	{
114	/** The ethernet header of the outgoing frame. */
115	RTNETETHERHDR EthHdr;
116	} VBOXNETFLTTAG;
117	/** Pointer to a VBoxNetFlt mbuf tag. */
118	typedef VBOXNETFLTTAG *PVBOXNETFLTTAG;
119	/** Pointer to a const VBoxNetFlt mbuf tag. */
120	typedef VBOXNETFLTTAG const *PCVBOXNETFLTTAG;
121
122
123	/*******************************************************************************
124	* Global Variables *
125	*******************************************************************************/
126	/**
127	* Declare the module stuff.
128	*/
129	RT_C_DECLS_BEGIN
130	extern kern_return_t _start(struct kmod_info pKModInfo, void pvData);
131	extern kern_return_t _stop(struct kmod_info pKModInfo, void pvData);
132
133	KMOD_EXPLICIT_DECL(VBoxNetFlt, VBOX_VERSION_STRING, _start, _stop)
134	DECLHIDDEN(kmod_start_func_t *) _realmain = VBoxNetFltDarwinStart;
135	DECLHIDDEN(kmod_stop_func_t *) _antimain = VBoxNetFltDarwinStop;
136	DECLHIDDEN(int) _kext_apple_cc = __APPLE_CC__;
137	RT_C_DECLS_END
138
139
140	/**
141	* The (common) global data.
142	*/
143	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
144
145	/** The unique tag id for this module.
146	* This is basically a unique string hash that lives on until reboot.
147	* It is used for tagging mbufs. */
148	static mbuf_tag_id_t g_idTag;
149
150	/** The offset of the struct ifnet::if_pcount variable.
151	* @remarks Initial value is valid for Lion and earlier. We adjust it on attach
152	* for later releases. */
153	static unsigned g_offIfNetPCount = sizeof(void ) (1 /if_softc/ + 1 /if_name/ + 2 /if_link/ + 2 /if_addrhead/ + 1 /if_check_multi/)
154	+ sizeof(u_long) /if_refcnt/;
155	/** Macro for accessing ifnet::if_pcount. */
156	#define VBOX_GET_PCOUNT(pIfNet) ( (int )((uintptr_t)pIfNet + g_offIfNetPCount) )
157	/** The size of area of ifnet structure we try to locate if_pcount in. */
158	#define VBOXNETFLT_DARWIN_IFNET_SIZE 256
159	/** Indicates whether g_offIfNetPCount has been adjusted already (no point in
160	* doing it more than once). */
161	static bool g_fNetPCountFound = false;
162
163
164	/**
165	* Change the promiscuous setting and try spot the changed in @a pIfNet.
166	*
167	* @returns Offset of potential p_count field.
168	* @param pIfNet The interface we're attaching to.
169	* @param iPromisc Whether to enable (1) or disable (0) promiscuous mode.
170	*
171	* @note This implementation relies on if_pcount to be aligned on sizeof(int).
172	*/
173	static unsigned vboxNetFltDarwinSetAndDiff(ifnet_t pIfNet, int iPromisc)
174	{
175	int aiSavedState[VBOXNETFLT_DARWIN_IFNET_SIZE / sizeof(int)];
176	memcpy(aiSavedState, pIfNet, sizeof(aiSavedState));
177
178	ifnet_set_promiscuous(pIfNet, iPromisc);
179
180	int const iDiff = iPromisc ? 1 : -1;
181
182	/*
183	* We assume that ifnet structure will never have less members in front of if_pcount
184	* than it used to have in Lion. If this turns out to be false assumption we will
185	* have to start from zero offset.
186	*/
187	for (unsigned i = g_offIfNetPCount / sizeof(int); i < RT_ELEMENTS(aiSavedState); i++)
188	if (((int*)pIfNet)[i] - aiSavedState[i] == iDiff)
189	return i * sizeof(int);
190
191	return 0;
192	}
193
194
195	/**
196	* Detect and adjust the offset of ifnet::if_pcount.
197	*
198	* @param pIfNet The interface we're attaching to.
199	*/
200	static void vboxNetFltDarwinDetectPCountOffset(ifnet_t pIfNet)
201	{
202	if (g_fNetPCountFound)
203	return;
204
205	/*
206	* It would be nice to use locking at this point, but it is not available via KPI.
207	* This is why we try several times. At each attempt we modify if_pcount four times
208	* to rule out false detections.
209	*/
210	unsigned offTry1, offTry2, offTry3, offTry4;
211	for (int iAttempt = 0; iAttempt < 3; iAttempt++)
212	{
213	offTry1 = vboxNetFltDarwinSetAndDiff(pIfNet, 1);
214	offTry2 = vboxNetFltDarwinSetAndDiff(pIfNet, 1);
215	offTry3 = vboxNetFltDarwinSetAndDiff(pIfNet, 0);
216	offTry4 = vboxNetFltDarwinSetAndDiff(pIfNet, 0);
217	if (offTry1 == offTry2 && offTry2 == offTry3 && offTry3 == offTry4)
218	{
219	if (g_offIfNetPCount != offTry1)
220	{
221	Log(("VBoxNetFltDarwinDetectPCountOffset: Adjusted if_pcount offset to %x from %x.\n", offTry1, g_offIfNetPCount));
222	g_offIfNetPCount = offTry1;
223	g_fNetPCountFound = true;
224	}
225	break;
226	}
227	}
228
229	if (g_offIfNetPCount != offTry1)
230	LogRel(("VBoxNetFlt: Failed to detect promiscuous count, all traffic may reach wire (%x != %x).\n", g_offIfNetPCount, offTry1));
231	}
232
233
234	/**
235	* Start the kernel module.
236	*/
237	static kern_return_t VBoxNetFltDarwinStart(struct kmod_info pKModInfo, void pvData)
238	{
239	int rc;
240
241	/*
242	* Initialize IPRT and find our module tag id.
243	* (IPRT is shared with VBoxDrv, it creates the loggers.)
244	*/
245	rc = RTR0Init(0);
246	if (RT_SUCCESS(rc))
247	{
248	Log(("VBoxNetFltDarwinStart\n"));
249	errno_t err = mbuf_tag_id_find("org.VirtualBox.kext.VBoxFltDrv", &g_idTag);
250	if (!err)
251	{
252	/*
253	* Initialize the globals and connect to the support driver.
254	*
255	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
256	* for establishing the connect to the support driver.
257	*/
258	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
259	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
260	if (RT_SUCCESS(rc))
261	{
262	LogRel(("VBoxFltDrv: version " VBOX_VERSION_STRING " r%d\n", VBOX_SVN_REV));
263	return KMOD_RETURN_SUCCESS;
264	}
265
266	LogRel(("VBoxFltDrv: failed to initialize device extension (rc=%d)\n", rc));
267	}
268	else
269	LogRel(("VBoxFltDrv: mbuf_tag_id_find failed, err=%d\n", err));
270	RTR0Term();
271	}
272	else
273	printf("VBoxFltDrv: failed to initialize IPRT (rc=%d)\n", rc);
274
275	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
276	return KMOD_RETURN_FAILURE;
277	}
278
279
280	/**
281	* Stop the kernel module.
282	*/
283	static kern_return_t VBoxNetFltDarwinStop(struct kmod_info pKModInfo, void pvData)
284	{
285	Log(("VBoxNetFltDarwinStop\n"));
286
287	/*
288	* Refuse to unload if anyone is currently using the filter driver.
289	* This is important as I/O kit / xnu will to be able to do usage
290	* tracking for us!
291	*/
292	int rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
293	if (RT_FAILURE(rc))
294	{
295	Log(("VBoxNetFltDarwinStop - failed, busy.\n"));
296	return KMOD_RETURN_FAILURE;
297	}
298
299	/*
300	* Undo the work done during start (in reverse order).
301	*/
302	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
303
304	RTR0Term();
305
306	return KMOD_RETURN_SUCCESS;
307	}
308
309
310	/**
311	* Reads and retains the host interface handle.
312	*
313	* @returns The handle, NULL if detached.
314	* @param pThis
315	*/
316	DECLINLINE(ifnet_t) vboxNetFltDarwinRetainIfNet(PVBOXNETFLTINS pThis)
317	{
318	ifnet_t pIfNet = NULL;
319
320	/*
321	* Be careful here to avoid problems racing the detached callback.
322	*/
323	RTSpinlockAcquire(pThis->hSpinlock);
324	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
325	{
326	pIfNet = ASMAtomicUoReadPtrT(&pThis->u.s.pIfNet, ifnet_t);
327	if (pIfNet)
328	ifnet_reference(pIfNet);
329	}
330	RTSpinlockRelease(pThis->hSpinlock);
331
332	return pIfNet;
333	}
334
335
336	/**
337	* Release the host interface handle previously retained
338	* by vboxNetFltDarwinRetainIfNet.
339	*
340	* @param pThis The instance.
341	* @param pIfNet The vboxNetFltDarwinRetainIfNet return value, NULL is fine.
342	*/
343	DECLINLINE(void) vboxNetFltDarwinReleaseIfNet(PVBOXNETFLTINS pThis, ifnet_t pIfNet)
344	{
345	NOREF(pThis);
346	if (pIfNet)
347	ifnet_release(pIfNet);
348	}
349
350
351	/**
352	* Checks whether this is an mbuf created by vboxNetFltDarwinMBufFromSG,
353	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
354	*
355	* @returns true / false accordingly.
356	* @param pThis The instance.
357	* @param pMBuf The mbuf.
358	* @param pvFrame The frame pointer, optional.
359	*/
360	DECLINLINE(bool) vboxNetFltDarwinMBufIsOur(PVBOXNETFLTINS pThis, mbuf_t pMBuf, void *pvFrame)
361	{
362	NOREF(pThis);
363
364	/*
365	* Lookup the tag set by vboxNetFltDarwinMBufFromSG.
366	*/
367	PCVBOXNETFLTTAG pTagData;
368	size_t cbTagData;
369	errno_t err = mbuf_tag_find(pMBuf, g_idTag, 0 /* type /, &cbTagData, (void *)&pTagData);
370	if (err)
371	return false;
372	AssertReturn(cbTagData == sizeof(*pTagData), false);
373
374	/*
375	* Dig out the ethernet header from the mbuf.
376	*/
377	PCRTNETETHERHDR pEthHdr = (PCRTNETETHERHDR)pvFrame;
378	if (!pEthHdr)
379	pEthHdr = (PCRTNETETHERHDR)mbuf_pkthdr_header(pMBuf);
380	if (!pEthHdr)
381	pEthHdr = (PCRTNETETHERHDR)mbuf_data(pMBuf);
382	/* ASSUMING that there is enough data to work on! */
383	if ( pEthHdr->DstMac.au8[0] != pTagData->EthHdr.DstMac.au8[0]
384	\|\| pEthHdr->DstMac.au8[1] != pTagData->EthHdr.DstMac.au8[1]
385	\|\| pEthHdr->DstMac.au8[2] != pTagData->EthHdr.DstMac.au8[2]
386	\|\| pEthHdr->DstMac.au8[3] != pTagData->EthHdr.DstMac.au8[3]
387	\|\| pEthHdr->DstMac.au8[4] != pTagData->EthHdr.DstMac.au8[4]
388	\|\| pEthHdr->DstMac.au8[5] != pTagData->EthHdr.DstMac.au8[5]
389	\|\| pEthHdr->SrcMac.au8[0] != pTagData->EthHdr.SrcMac.au8[0]
390	\|\| pEthHdr->SrcMac.au8[1] != pTagData->EthHdr.SrcMac.au8[1]
391	\|\| pEthHdr->SrcMac.au8[2] != pTagData->EthHdr.SrcMac.au8[2]
392	\|\| pEthHdr->SrcMac.au8[3] != pTagData->EthHdr.SrcMac.au8[3]
393	\|\| pEthHdr->SrcMac.au8[4] != pTagData->EthHdr.SrcMac.au8[4]
394	\|\| pEthHdr->SrcMac.au8[5] != pTagData->EthHdr.SrcMac.au8[5]
395	\|\| pEthHdr->EtherType != pTagData->EthHdr.EtherType)
396	{
397	Log3(("tagged, but the ethernet header has changed\n"));
398	return false;
399	}
400
401	return true;
402	}
403
404
405	/**
406	* Internal worker that create a darwin mbuf for a (scatter/)gather list.
407	*
408	* @returns Pointer to the mbuf.
409	* @param pThis The instance.
410	* @param pSG The (scatter/)gather list.
411	*/
412	static mbuf_t vboxNetFltDarwinMBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG)
413	{
414	/// @todo future? mbuf_how_t How = preemption enabled ? MBUF_DONTWAIT : MBUF_WAITOK;
415	mbuf_how_t How = MBUF_WAITOK;
416
417	/*
418	* We need some way of getting back to our instance data when
419	* the mbuf is freed, so use pvUserData for this.
420	* -- this is not relevant anylonger! --
421	*/
422	Assert(!pSG->pvUserData \|\| pSG->pvUserData == pThis);
423	Assert(!pSG->pvUserData2);
424	pSG->pvUserData = pThis;
425
426	/*
427	* Allocate a packet and copy over the data.
428	*
429	* Using mbuf_attachcluster() here would've been nice but there are two
430	* issues with it: (1) it's 10.5.x only, and (2) the documentation indicates
431	* that it's not supposed to be used for really external buffers. The 2nd
432	* point might be argued against considering that the only m_clattach user
433	* is mallocs memory for the ext mbuf and not doing what's stated in the docs.
434	* However, it's hard to tell if these m_clattach buffers actually makes it
435	* to the NICs or not, and even if they did, the NIC would need the physical
436	* addresses for the pages they contain and might end up copying the data
437	* to a new mbuf anyway.
438	*
439	* So, in the end it's better to just do it the simple way that will work
440	* 100%, even if it involves some extra work (alloc + copy) we really wished
441	* to avoid.
442	*
443	* Note. We can't make use of the physical addresses on darwin because the
444	* way the mbuf / cluster stuff works (see mbuf_data_to_physical and
445	* mcl_to_paddr).
446	*/
447	mbuf_t pPkt = NULL;
448	errno_t err = mbuf_allocpacket(How, pSG->cbTotal, NULL, &pPkt);
449	if (!err)
450	{
451	/* Skip zero sized memory buffers (paranoia). */
452	mbuf_t pCur = pPkt;
453	while (pCur && !mbuf_maxlen(pCur))
454	pCur = mbuf_next(pCur);
455	Assert(pCur);
456
457	/* Set the required packet header attributes. */
458	mbuf_pkthdr_setlen(pPkt, pSG->cbTotal);
459	mbuf_pkthdr_setheader(pPkt, mbuf_data(pCur));
460
461	/* Special case the single buffer copy. */
462	if ( mbuf_next(pCur)
463	&& mbuf_maxlen(pCur) >= pSG->cbTotal)
464	{
465	mbuf_setlen(pCur, pSG->cbTotal);
466	IntNetSgRead(pSG, mbuf_data(pCur));
467	}
468	else
469	{
470	/* Multi buffer copying. */
471	size_t cbLeft = pSG->cbTotal;
472	size_t offSrc = 0;
473	while (cbLeft > 0 && pCur)
474	{
475	size_t cb = mbuf_maxlen(pCur);
476	if (cb > cbLeft)
477	cb = cbLeft;
478	mbuf_setlen(pCur, cb);
479	IntNetSgReadEx(pSG, offSrc, cb, mbuf_data(pCur));
480
481	/* advance */
482	offSrc += cb;
483	cbLeft -= cb;
484	pCur = mbuf_next(pCur);
485	}
486	Assert(cbLeft == 0);
487	}
488	if (!err)
489	{
490	/*
491	* Tag the packet and return successfully.
492	*/
493	PVBOXNETFLTTAG pTagData;
494	err = mbuf_tag_allocate(pPkt, g_idTag, 0 /* type /, sizeof(VBOXNETFLTTAG) / tag len /, How, (void *)&pTagData);
495	if (!err)
496	{
497	Assert(pSG->aSegs[0].cb >= sizeof(pTagData->EthHdr));
498	memcpy(&pTagData->EthHdr, pSG->aSegs[0].pv, sizeof(pTagData->EthHdr));
499	return pPkt;
500	}
501
502	/* bailout: */
503	AssertMsg(err == ENOMEM \|\| err == EWOULDBLOCK, ("err=%d\n", err));
504	}
505
506	mbuf_freem(pPkt);
507	}
508	else
509	AssertMsg(err == ENOMEM \|\| err == EWOULDBLOCK, ("err=%d\n", err));
510	pSG->pvUserData = NULL;
511
512	return NULL;
513	}
514
515
516	/**
517	* Calculates the number of segments required to represent the mbuf.
518	*
519	* @returns Number of segments.
520	* @param pThis The instance.
521	* @param pMBuf The mbuf.
522	* @param pvFrame The frame pointer, optional.
523	*/
524	DECLINLINE(unsigned) vboxNetFltDarwinMBufCalcSGSegs(PVBOXNETFLTINS pThis, mbuf_t pMBuf, void *pvFrame)
525	{
526	NOREF(pThis);
527
528	/*
529	* Count the buffers in the chain.
530	*/
531	unsigned cSegs = 0;
532	for (mbuf_t pCur = pMBuf; pCur; pCur = mbuf_next(pCur))
533	if (mbuf_len(pCur))
534	cSegs++;
535	else if ( !cSegs
536	&& pvFrame
537	&& (uintptr_t)pvFrame - (uintptr_t)mbuf_datastart(pMBuf) < mbuf_maxlen(pMBuf))
538	cSegs++;
539
540	#ifdef PADD_RUNT_FRAMES_FROM_HOST
541	/*
542	* Add one buffer if the total is less than the ethernet minimum 60 bytes.
543	* This may allocate a segment too much if the ethernet header is separated,
544	* but that shouldn't harm us much.
545	*/
546	if (mbuf_pkthdr_len(pMBuf) < 60)
547	cSegs++;
548	#endif
549
550	#ifdef VBOXNETFLT_DARWIN_TEST_SEG_SIZE
551	/* maximize the number of segments. */
552	cSegs = RT_MAX(VBOXNETFLT_DARWIN_MAX_SEGS - 1, cSegs);
553	#endif
554
555	return cSegs ? cSegs : 1;
556	}
557
558
559	/**
560	* Initializes a SG list from an mbuf.
561	*
562	* @returns Number of segments.
563	* @param pThis The instance.
564	* @param pMBuf The mbuf.
565	* @param pSG The SG.
566	* @param pvFrame The frame pointer, optional.
567	* @param cSegs The number of segments allocated for the SG.
568	* This should match the number in the mbuf exactly!
569	* @param fSrc The source of the frame.
570	*/
571	DECLINLINE(void) vboxNetFltDarwinMBufToSG(PVBOXNETFLTINS pThis, mbuf_t pMBuf, void *pvFrame, PINTNETSG pSG, unsigned cSegs, uint32_t fSrc)
572	{
573	NOREF(pThis);
574
575	/*
576	* Walk the chain and convert the buffers to segments. Works INTNETSG::cbTotal.
577	*/
578	unsigned iSeg = 0;
579	IntNetSgInitTempSegs(pSG, 0 /cbTotal/, cSegs, 0 /cSegsUsed/);
580	for (mbuf_t pCur = pMBuf; pCur; pCur = mbuf_next(pCur))
581	{
582	size_t cbSeg = mbuf_len(pCur);
583	if (cbSeg)
584	{
585	void *pvSeg = mbuf_data(pCur);
586
587	/* deal with pvFrame */
588	if (!iSeg && pvFrame && pvFrame != pvSeg)
589	{
590	void *pvStart = mbuf_datastart(pMBuf);
591	uintptr_t offSeg = (uintptr_t)pvSeg - (uintptr_t)pvStart;
592	uintptr_t offSegEnd = offSeg + cbSeg;
593	Assert(pvStart && pvSeg && offSeg < mbuf_maxlen(pMBuf) && offSegEnd <= mbuf_maxlen(pMBuf)); NOREF(offSegEnd);
594	uintptr_t offFrame = (uintptr_t)pvFrame - (uintptr_t)pvStart;
595	if (RT_LIKELY(offFrame < offSeg))
596	{
597	pvSeg = pvFrame;
598	cbSeg += offSeg - offFrame;
599	}
600	else
601	AssertMsgFailed(("pvFrame=%p pvStart=%p pvSeg=%p offSeg=%p cbSeg=%#zx offSegEnd=%p offFrame=%p maxlen=%#zx\n",
602	pvFrame, pvStart, pvSeg, offSeg, cbSeg, offSegEnd, offFrame, mbuf_maxlen(pMBuf)));
603	pvFrame = NULL;
604	}
605
606	AssertBreak(iSeg < cSegs);
607	pSG->cbTotal += cbSeg;
608	pSG->aSegs[iSeg].cb = cbSeg;
609	pSG->aSegs[iSeg].pv = pvSeg;
610	pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
611	iSeg++;
612	}
613	/* The pvFrame might be in a now empty buffer. */
614	else if ( !iSeg
615	&& pvFrame
616	&& (uintptr_t)pvFrame - (uintptr_t)mbuf_datastart(pMBuf) < mbuf_maxlen(pMBuf))
617	{
618	cbSeg = (uintptr_t)mbuf_datastart(pMBuf) + mbuf_maxlen(pMBuf) - (uintptr_t)pvFrame;
619	pSG->cbTotal += cbSeg;
620	pSG->aSegs[iSeg].cb = cbSeg;
621	pSG->aSegs[iSeg].pv = pvFrame;
622	pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
623	iSeg++;
624	pvFrame = NULL;
625	}
626	}
627
628	Assert(iSeg && iSeg <= cSegs);
629	pSG->cSegsUsed = iSeg;
630
631	#ifdef PADD_RUNT_FRAMES_FROM_HOST
632	/*
633	* Add a trailer if the frame is too small.
634	*
635	* Since we're getting to the packet before it is framed, it has not
636	* yet been padded. The current solution is to add a segment pointing
637	* to a buffer containing all zeros and pray that works for all frames...
638	*/
639	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
640	{
641	AssertReturnVoid(iSeg < cSegs);
642
643	static uint8_t const s_abZero[128] = {0};
644	pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
645	pSG->aSegs[iSeg].pv = (void *)&s_abZero[0];
646	pSG->aSegs[iSeg].cb = 60 - pSG->cbTotal;
647	pSG->cbTotal = 60;
648	pSG->cSegsUsed++;
649	}
650	#endif
651
652	#ifdef VBOXNETFLT_DARWIN_TEST_SEG_SIZE
653	/*
654	* Redistribute the segments.
655	*/
656	if (pSG->cSegsUsed < pSG->cSegsAlloc)
657	{
658	/* copy the segments to the end. */
659	int iSrc = pSG->cSegsUsed;
660	int iDst = pSG->cSegsAlloc;
661	while (iSrc > 0)
662	{
663	iDst--;
664	iSrc--;
665	pSG->aSegs[iDst] = pSG->aSegs[iSrc];
666	}
667
668	/* create small segments from the start. */
669	pSG->cSegsUsed = pSG->cSegsAlloc;
670	iSrc = iDst;
671	iDst = 0;
672	while ( iDst < iSrc
673	&& iDst < pSG->cSegsAlloc)
674	{
675	pSG->aSegs[iDst].Phys = NIL_RTHCPHYS;
676	pSG->aSegs[iDst].pv = pSG->aSegs[iSrc].pv;
677	pSG->aSegs[iDst].cb = RT_MIN(pSG->aSegs[iSrc].cb, VBOXNETFLT_DARWIN_TEST_SEG_SIZE);
678	if (pSG->aSegs[iDst].cb != pSG->aSegs[iSrc].cb)
679	{
680	pSG->aSegs[iSrc].cb -= pSG->aSegs[iDst].cb;
681	pSG->aSegs[iSrc].pv = (uint8_t *)pSG->aSegs[iSrc].pv + pSG->aSegs[iDst].cb;
682	}
683	else if (++iSrc >= pSG->cSegsAlloc)
684	{
685	pSG->cSegsUsed = iDst + 1;
686	break;
687	}
688	iDst++;
689	}
690	}
691	#endif
692
693	AssertMsg(!pvFrame, ("pvFrame=%p pMBuf=%p iSeg=%d\n", pvFrame, pMBuf, iSeg));
694	}
695
696
697	/**
698	* Helper for determining whether the host wants the interface to be
699	* promiscuous.
700	*/
701	static bool vboxNetFltDarwinIsPromiscuous(PVBOXNETFLTINS pThis)
702	{
703	bool fRc = false;
704	ifnet_t pIfNet = vboxNetFltDarwinRetainIfNet(pThis);
705	if (pIfNet)
706	{
707	/* gather the data */
708	uint16_t fIf = ifnet_flags(pIfNet);
709	unsigned cPromisc = VBOX_GET_PCOUNT(pIfNet);
710	bool fSetPromiscuous = ASMAtomicUoReadBool(&pThis->u.s.fSetPromiscuous);
711	vboxNetFltDarwinReleaseIfNet(pThis, pIfNet);
712
713	/* calc the return. */
714	fRc = (fIf & IFF_PROMISC)
715	&& cPromisc > fSetPromiscuous;
716	}
717	return fRc;
718	}
719
720
721
722	/**
723	*
724	* @see iff_detached_func in the darwin kpi.
725	*/
726	static void vboxNetFltDarwinIffDetached(void *pvThis, ifnet_t pIfNet)
727	{
728	PVBOXNETFLTINS pThis = (PVBOXNETFLTINS)pvThis;
729	uint64_t NanoTS = RTTimeSystemNanoTS();
730	LogFlow(("vboxNetFltDarwinIffDetached: pThis=%p NanoTS=%RU64 (%d)\n",
731	pThis, NanoTS, VALID_PTR(pIfNet) ? VBOX_GET_PCOUNT(pIfNet) : -1));
732
733	Assert(!pThis->fDisconnectedFromHost);
734	Assert(!pThis->fRediscoveryPending);
735
736	/*
737	* If we've put it into promiscuous mode, undo that now. If we don't
738	* the if_pcount will go all wrong when it's replugged.
739	*/
740	if (ASMAtomicXchgBool(&pThis->u.s.fSetPromiscuous, false))
741	ifnet_set_promiscuous(pIfNet, 0);
742
743	/*
744	* We carefully take the spinlock and increase the interface reference
745	* behind it in order to avoid problematic races with the detached callback.
746	*/
747	RTSpinlockAcquire(pThis->hSpinlock);
748
749	pIfNet = ASMAtomicUoReadPtrT(&pThis->u.s.pIfNet, ifnet_t);
750	int cPromisc = VALID_PTR(pIfNet) ? VBOX_GET_PCOUNT(pIfNet) : - 1;
751
752	ASMAtomicUoWriteNullPtr(&pThis->u.s.pIfNet);
753	ASMAtomicUoWriteNullPtr(&pThis->u.s.pIfFilter);
754	ASMAtomicWriteBool(&pThis->u.s.fNeedSetPromiscuous, false);
755	pThis->u.s.fSetPromiscuous = false;
756	ASMAtomicUoWriteU64(&pThis->NanoTSLastRediscovery, NanoTS);
757	ASMAtomicUoWriteBool(&pThis->fRediscoveryPending, false);
758	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
759
760	RTSpinlockRelease(pThis->hSpinlock);
761
762	if (pIfNet)
763	ifnet_release(pIfNet);
764	LogRel(("VBoxNetFlt: was detached from '%s' (%d)\n", pThis->szName, cPromisc));
765	}
766
767
768	/**
769	*
770	* @see iff_ioctl_func in the darwin kpi.
771	*/
772	static errno_t vboxNetFltDarwinIffIoCtl(void pvThis, ifnet_t pIfNet, protocol_family_t eProtocol, u_long uCmd, void pvArg)
773	{
774	PVBOXNETFLTINS pThis = (PVBOXNETFLTINS)pvThis;
775	LogFlow(("vboxNetFltDarwinIffIoCtl: pThis=%p uCmd=%lx\n", pThis, uCmd));
776
777	/*
778	* Update fOtherPromiscuous.
779	*/
780	/** @todo we'll have to find the offset of if_pcount to get this right! */
781	//if (uCmd == SIOCSIFFLAGS)
782	//{
783	//
784	//}
785
786	/*
787	* We didn't handle it, continue processing.
788	*/
789	NOREF(pThis);
790	NOREF(eProtocol);
791	NOREF(uCmd);
792	NOREF(pvArg);
793	return EOPNOTSUPP;
794	}
795
796
797	/**
798	*
799	* @see iff_event_func in the darwin kpi.
800	*/
801	static void vboxNetFltDarwinIffEvent(void pvThis, ifnet_t pIfNet, protocol_family_t eProtocol, const struct kev_msg pEvMsg)
802	{
803	PVBOXNETFLTINS pThis = (PVBOXNETFLTINS)pvThis;
804	LogFlow(("vboxNetFltDarwinIffEvent: pThis=%p\n", pThis));
805
806	NOREF(pThis);
807	NOREF(pIfNet);
808	NOREF(eProtocol);
809	NOREF(pEvMsg);
810
811	/*
812	* Watch out for the interface going online / offline.
813	*/
814	if ( VALID_PTR(pThis)
815	&& VALID_PTR(pEvMsg)
816	&& pEvMsg->vendor_code == KEV_VENDOR_APPLE
817	&& pEvMsg->kev_class == KEV_NETWORK_CLASS
818	&& pEvMsg->kev_subclass == KEV_DL_SUBCLASS)
819	{
820	if (pThis->u.s.pIfNet == pIfNet)
821	{
822	if (pEvMsg->event_code == KEV_DL_LINK_ON)
823	{
824	if (ASMAtomicUoReadBool(&pThis->u.s.fNeedSetPromiscuous))
825	{
826	/* failed to bring it online. */
827	errno_t err = ifnet_set_promiscuous(pIfNet, 1);
828	if (!err)
829	{
830	ASMAtomicWriteBool(&pThis->u.s.fSetPromiscuous, true);
831	ASMAtomicWriteBool(&pThis->u.s.fNeedSetPromiscuous, false);
832	Log(("vboxNetFltDarwinIffEvent: enabled promiscuous mode on %s (%d)\n", pThis->szName, VBOX_GET_PCOUNT(pIfNet)));
833	}
834	else
835	Log(("vboxNetFltDarwinIffEvent: ifnet_set_promiscuous failed on %s, err=%d (%d)\n", pThis->szName, err, VBOX_GET_PCOUNT(pIfNet)));
836	}
837	else if ( ASMAtomicUoReadBool(&pThis->u.s.fSetPromiscuous)
838	&& !(ifnet_flags(pIfNet) & IFF_PROMISC))
839	{
840	/* Try fix the inconsistency. */
841	errno_t err = ifnet_set_flags(pIfNet, IFF_PROMISC, IFF_PROMISC);
842	if (!err)
843	err = ifnet_ioctl(pIfNet, 0, SIOCSIFFLAGS, NULL);
844	if (!err && (ifnet_flags(pIfNet) & IFF_PROMISC))
845	Log(("vboxNetFltDarwinIffEvent: fixed IFF_PROMISC on %s (%d)\n", pThis->szName, VBOX_GET_PCOUNT(pIfNet)));
846	else
847	Log(("vboxNetFltDarwinIffEvent: failed to fix IFF_PROMISC on %s, err=%d flags=%#x (%d)\n",
848	pThis->szName, err, ifnet_flags(pIfNet), VBOX_GET_PCOUNT(pIfNet)));
849	}
850	else
851	Log(("vboxNetFltDarwinIffEvent: online, '%s'. flags=%#x (%d)\n", pThis->szName, ifnet_flags(pIfNet), VBOX_GET_PCOUNT(pIfNet)));
852	}
853	else if (pEvMsg->event_code == KEV_DL_LINK_OFF)
854	Log(("vboxNetFltDarwinIffEvent: %s goes down (%d)\n", pThis->szName, VBOX_GET_PCOUNT(pIfNet)));
855	/** @todo KEV_DL_LINK_ADDRESS_CHANGED -> pfnReportMacAddress */
856	/** @todo KEV_DL_SIFFLAGS -> pfnReportPromiscuousMode */
857	}
858	else
859	Log(("vboxNetFltDarwinIffEvent: pThis->u.s.pIfNet=%p pIfNet=%p (%d)\n", pThis->u.s.pIfNet, pIfNet, VALID_PTR(pIfNet) ? VBOX_GET_PCOUNT(pIfNet) : -1));
860	}
861	else if (VALID_PTR(pEvMsg))
862	Log(("vboxNetFltDarwinIffEvent: vendor_code=%#x kev_class=%#x kev_subclass=%#x event_code=%#x\n",
863	pEvMsg->vendor_code, pEvMsg->kev_class, pEvMsg->kev_subclass, pEvMsg->event_code));
864	}
865
866
867	/**
868	* Internal worker for vboxNetFltDarwinIffInput and vboxNetFltDarwinIffOutput,
869	*
870	* @returns 0 or EJUSTRETURN.
871	* @param pThis The instance.
872	* @param pMBuf The mbuf.
873	* @param pvFrame The start of the frame, optional.
874	* @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
875	* @param eProtocol The protocol.
876	*/
877	static errno_t vboxNetFltDarwinIffInputOutputWorker(PVBOXNETFLTINS pThis, mbuf_t pMBuf, void *pvFrame,
878	uint32_t fSrc, protocol_family_t eProtocol)
879	{
880	/*
881	* Drop it immediately?
882	*/
883	Log2(("vboxNetFltDarwinIffInputOutputWorker: pThis=%p pMBuf=%p pvFrame=%p fSrc=%#x cbPkt=%x\n",
884	pThis, pMBuf, pvFrame, fSrc, pMBuf ? mbuf_pkthdr_len(pMBuf) : -1));
885	if (!pMBuf)
886	return 0;
887	#if 0 /* debugging lost icmp packets */
888	if (mbuf_pkthdr_len(pMBuf) > 0x300)
889	{
890	uint8_t pb = (uint8_t )(pvFrame ? pvFrame : mbuf_data(pMBuf));
891	Log3(("D=%.6Rhxs S=%.6Rhxs T=%04x IFF\n", pb, pb + 6, RT_BE2H_U16((uint16_t )(pb + 12))));
892	}
893	#endif
894	if (vboxNetFltDarwinMBufIsOur(pThis, pMBuf, pvFrame))
895	return 0;
896
897	/*
898	* Active? Retain the instance and increment the busy counter.
899	*/
900	if (!vboxNetFltTryRetainBusyActive(pThis))
901	return 0;
902
903	/*
904	* Finalize out-bound packets since the stack puts off finalizing
905	* TCP/IP checksums as long as possible.
906	* ASSUMES this only applies to outbound IP packets.
907	*/
908	if ( (fSrc & INTNETTRUNKDIR_HOST)
909	&& eProtocol == PF_INET)
910	{
911	Assert(!pvFrame);
912	mbuf_outbound_finalize(pMBuf, eProtocol, sizeof(RTNETETHERHDR));
913	}
914
915	/*
916	* Create a (scatter/)gather list for the mbuf and feed it to the internal network.
917	*/
918	bool fDropIt = false;
919	unsigned cSegs = vboxNetFltDarwinMBufCalcSGSegs(pThis, pMBuf, pvFrame);
920	if (cSegs < VBOXNETFLT_DARWIN_MAX_SEGS)
921	{
922	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
923	vboxNetFltDarwinMBufToSG(pThis, pMBuf, pvFrame, pSG, cSegs, fSrc);
924
925	fDropIt = pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
926	if (fDropIt)
927	{
928	/*
929	* Check if this interface is in promiscuous mode. We should not drop
930	* any packets before they get to the driver as it passes them to tap
931	* callbacks in order for BPF to work properly.
932	*/
933	if (vboxNetFltDarwinIsPromiscuous(pThis))
934	fDropIt = false;
935	else
936	mbuf_freem(pMBuf);
937	}
938	}
939
940	vboxNetFltRelease(pThis, true /* fBusy */);
941
942	return fDropIt ? EJUSTRETURN : 0;
943	}
944
945
946	/**
947	* From the host.
948	*
949	* @see iff_output_func in the darwin kpi.
950	*/
951	static errno_t vboxNetFltDarwinIffOutput(void pvThis, ifnet_t pIfNet, protocol_family_t eProtocol, mbuf_t ppMBuf)
952	{
953	/** @todo there was some note about the ethernet header here or something like that... */
954
955	NOREF(eProtocol);
956	NOREF(pIfNet);
957	return vboxNetFltDarwinIffInputOutputWorker((PVBOXNETFLTINS)pvThis, *ppMBuf, NULL, INTNETTRUNKDIR_HOST, eProtocol);
958	}
959
960
961	/**
962	* From the wire.
963	*
964	* @see iff_input_func in the darwin kpi.
965	*/
966	static errno_t vboxNetFltDarwinIffInput(void pvThis, ifnet_t pIfNet, protocol_family_t eProtocol, mbuf_t ppMBuf, char **ppchFrame)
967	{
968	NOREF(eProtocol);
969	NOREF(pIfNet);
970	return vboxNetFltDarwinIffInputOutputWorker((PVBOXNETFLTINS)pvThis, ppMBuf, ppchFrame, INTNETTRUNKDIR_WIRE, eProtocol);
971	}
972
973
974	/** A worker thread for vboxNetFltSendDummy(). */
975	static DECLCALLBACK(int) vboxNetFltSendDummyWorker(RTTHREAD hThreadSelf, void *pvUser)
976	{
977	Assert(pvUser);
978	ifnet_t pIfNet = (ifnet_t)pvUser;
979	return VBoxNetSendDummy(pIfNet);
980	}
981
982
983	/**
984	* Prevent GUI icon freeze issue when VirtualBoxVM process terminates.
985	*
986	* This function is a workaround for stuck-in-dock issue. The idea here is to
987	* send a dummy packet to an interface from the context of a kernel thread.
988	* Therefore, an XNU's receive thread (which is created as a result if we are
989	* the first who is communicating with the interface) will be associated with
990	* the kernel thread instead of VirtualBoxVM process.
991	*
992	* @param pIfNet Interface to be used to send data.
993	*/
994	static void vboxNetFltSendDummy(ifnet_t pIfNet)
995	{
996	RTTHREAD hThread;
997	int rc = RTThreadCreate(&hThread, vboxNetFltSendDummyWorker, (void *)pIfNet, 0,
998	RTTHREADTYPE_DEFAULT, RTTHREADFLAGS_WAITABLE, "DummyThread");
999	if (RT_SUCCESS(rc))
1000	{
1001	RTThreadWait(hThread, RT_INDEFINITE_WAIT, NULL);
1002	LogFlow(("vboxNetFltSendDummy: a dummy packet has been successfully sent in order to prevent stuck-in-dock issue\n"));
1003	}
1004	else
1005	LogFlow(("vboxNetFltSendDummy: unable to send dummy packet in order to prevent stuck-in-dock issue\n"));
1006	}
1007
1008
1009	/**
1010	* Internal worker for vboxNetFltOsInitInstance and vboxNetFltOsMaybeRediscovered.
1011	*
1012	* @returns VBox status code.
1013	* @param pThis The instance.
1014	* @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1015	* flood the release log.
1016	*/
1017	static int vboxNetFltDarwinAttachToInterface(PVBOXNETFLTINS pThis, bool fRediscovery)
1018	{
1019	LogFlow(("vboxNetFltDarwinAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1020
1021	/*
1022	* Locate the interface first.
1023	*
1024	* The pIfNet member is updated before iflt_attach is called and used
1025	* to deal with the hypothetical case where someone rips out the
1026	* interface immediately after our iflt_attach call.
1027	*/
1028	ifnet_t pIfNet = NULL;
1029	errno_t err = ifnet_find_by_name(pThis->szName, &pIfNet);
1030	if (err)
1031	{
1032	Assert(err == ENXIO);
1033	if (!fRediscovery)
1034	LogRel(("VBoxFltDrv: failed to find ifnet '%s' (err=%d)\n", pThis->szName, err));
1035	else
1036	Log(("VBoxFltDrv: failed to find ifnet '%s' (err=%d)\n", pThis->szName, err));
1037	return VERR_INTNET_FLT_IF_NOT_FOUND;
1038	}
1039
1040	RTSpinlockAcquire(pThis->hSpinlock);
1041	ASMAtomicUoWritePtr(&pThis->u.s.pIfNet, pIfNet);
1042	RTSpinlockRelease(pThis->hSpinlock);
1043
1044	/* Adjust g_offIfNetPCount as it varies for different versions of xnu. */
1045	vboxNetFltDarwinDetectPCountOffset(pIfNet);
1046
1047	/* Prevent stuck-in-dock issue by associating interface receive thread with kernel thread. */
1048	vboxNetFltSendDummy(pIfNet);
1049
1050	/*
1051	* Get the mac address while we still have a valid ifnet reference.
1052	*/
1053	err = ifnet_lladdr_copy_bytes(pIfNet, &pThis->u.s.MacAddr, sizeof(pThis->u.s.MacAddr));
1054	if (!err)
1055	{
1056	/*
1057	* Try attach the filter.
1058	*/
1059	struct iff_filter RegRec;
1060	RegRec.iff_cookie = pThis;
1061	RegRec.iff_name = "VBoxNetFlt";
1062	RegRec.iff_protocol = 0;
1063	RegRec.iff_input = vboxNetFltDarwinIffInput;
1064	RegRec.iff_output = vboxNetFltDarwinIffOutput;
1065	RegRec.iff_event = vboxNetFltDarwinIffEvent;
1066	RegRec.iff_ioctl = vboxNetFltDarwinIffIoCtl;
1067	RegRec.iff_detached = vboxNetFltDarwinIffDetached;
1068	interface_filter_t pIfFilter = NULL;
1069	err = iflt_attach(pIfNet, &RegRec, &pIfFilter);
1070	Assert(err \|\| pIfFilter);
1071
1072	RTSpinlockAcquire(pThis->hSpinlock);
1073	pIfNet = ASMAtomicUoReadPtrT(&pThis->u.s.pIfNet, ifnet_t);
1074	if (pIfNet && !err)
1075	{
1076	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1077	ASMAtomicUoWritePtr(&pThis->u.s.pIfFilter, pIfFilter);
1078	pIfNet = NULL; /* don't dereference it */
1079	}
1080	RTSpinlockRelease(pThis->hSpinlock);
1081
1082	/* Report capabilities. */
1083	if ( !pIfNet
1084	&& vboxNetFltTryRetainBusyNotDisconnected(pThis))
1085	{
1086	Assert(pThis->pSwitchPort);
1087	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1088	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltDarwinIsPromiscuous(pThis));
1089	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, 0, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1090	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, 0 /* none */);
1091	vboxNetFltRelease(pThis, true /fBusy/);
1092	}
1093	}
1094
1095	/* Release the interface on failure. */
1096	if (pIfNet)
1097	ifnet_release(pIfNet);
1098
1099	int rc = RTErrConvertFromErrno(err);
1100	if (RT_SUCCESS(rc))
1101	LogRel(("VBoxFltDrv: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1102	else
1103	LogRel(("VBoxFltDrv: failed to attach to ifnet '%s' (err=%d)\n", pThis->szName, err));
1104	return rc;
1105	}
1106
1107
1108	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
1109	{
1110	vboxNetFltDarwinAttachToInterface(pThis, true /* fRediscovery */);
1111	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
1112	}
1113
1114
1115	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
1116	{
1117	NOREF(pvIfData);
1118
1119	int rc = VINF_SUCCESS;
1120	ifnet_t pIfNet = vboxNetFltDarwinRetainIfNet(pThis);
1121	if (pIfNet)
1122	{
1123	/*
1124	* Create a mbuf for the gather list and push it onto the wire.
1125	*
1126	* Note! If the interface is in the promiscuous mode we need to send the
1127	* packet down the stack so it reaches the driver and Berkeley
1128	* Packet Filter (see @bugref{5817}).
1129	*/
1130	if ((fDst & INTNETTRUNKDIR_WIRE) \|\| vboxNetFltDarwinIsPromiscuous(pThis))
1131	{
1132	mbuf_t pMBuf = vboxNetFltDarwinMBufFromSG(pThis, pSG);
1133	if (pMBuf)
1134	{
1135	errno_t err = ifnet_output_raw(pIfNet, PF_LINK, pMBuf);
1136	if (err)
1137	rc = RTErrConvertFromErrno(err);
1138	}
1139	else
1140	rc = VERR_NO_MEMORY;
1141	}
1142
1143	/*
1144	* Create a mbuf for the gather list and push it onto the host stack.
1145	*/
1146	if (fDst & INTNETTRUNKDIR_HOST)
1147	{
1148	mbuf_t pMBuf = vboxNetFltDarwinMBufFromSG(pThis, pSG);
1149	if (pMBuf)
1150	{
1151	/* This is what IONetworkInterface::inputPacket does. */
1152	unsigned const cbEthHdr = 14;
1153	mbuf_pkthdr_setheader(pMBuf, mbuf_data(pMBuf));
1154	mbuf_pkthdr_setlen(pMBuf, mbuf_pkthdr_len(pMBuf) - cbEthHdr);
1155	mbuf_setdata(pMBuf, (uint8_t *)mbuf_data(pMBuf) + cbEthHdr, mbuf_len(pMBuf) - cbEthHdr);
1156	mbuf_pkthdr_setrcvif(pMBuf, pIfNet); /* will crash without this. */
1157
1158	errno_t err = ifnet_input(pIfNet, pMBuf, NULL);
1159	if (err)
1160	rc = RTErrConvertFromErrno(err);
1161	}
1162	else
1163	rc = VERR_NO_MEMORY;
1164	}
1165
1166	vboxNetFltDarwinReleaseIfNet(pThis, pIfNet);
1167	}
1168
1169	return rc;
1170	}
1171
1172
1173	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
1174	{
1175	ifnet_t pIfNet = vboxNetFltDarwinRetainIfNet(pThis);
1176	if (pIfNet)
1177	{
1178	if (pThis->fDisablePromiscuous)
1179	{
1180	/*
1181	* Promiscuous mode should not be used (wireless), we just need to
1182	* make sure the interface is up.
1183	*/
1184	if (fActive)
1185	{
1186	u_int16_t fIf = ifnet_flags(pIfNet);
1187	if ((fIf & (IFF_UP \| IFF_RUNNING)) != (IFF_UP \| IFF_RUNNING))
1188	{
1189	ifnet_set_flags(pIfNet, IFF_UP, IFF_UP);
1190	ifnet_ioctl(pIfNet, 0, SIOCSIFFLAGS, NULL);
1191	}
1192	}
1193	}
1194	else
1195	{
1196	/*
1197	* This api is a bit weird, the best reference is the code.
1198	*
1199	* Also, we have a bit or race conditions wrt the maintenance of
1200	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
1201	*/
1202	unsigned const cPromiscBefore = VBOX_GET_PCOUNT(pIfNet);
1203	u_int16_t fIf;
1204	if (fActive)
1205	{
1206	Assert(!pThis->u.s.fSetPromiscuous);
1207	errno_t err = ENETDOWN;
1208	ASMAtomicWriteBool(&pThis->u.s.fNeedSetPromiscuous, true);
1209
1210	/*
1211	* Try bring the interface up and running if it's down.
1212	*/
1213	fIf = ifnet_flags(pIfNet);
1214	if ((fIf & (IFF_UP \| IFF_RUNNING)) != (IFF_UP \| IFF_RUNNING))
1215	{
1216	err = ifnet_set_flags(pIfNet, IFF_UP, IFF_UP);
1217	errno_t err2 = ifnet_ioctl(pIfNet, 0, SIOCSIFFLAGS, NULL);
1218	if (!err)
1219	err = err2;
1220	fIf = ifnet_flags(pIfNet);
1221	}
1222
1223	/*
1224	* Is it already up? If it isn't, leave it to the link event or
1225	* we'll upset if_pcount (as stated above, ifnet_set_promiscuous is weird).
1226	*/
1227	if ((fIf & (IFF_UP \| IFF_RUNNING)) == (IFF_UP \| IFF_RUNNING))
1228	{
1229	err = ifnet_set_promiscuous(pIfNet, 1);
1230	pThis->u.s.fSetPromiscuous = err == 0;
1231	if (!err)
1232	{
1233	ASMAtomicWriteBool(&pThis->u.s.fNeedSetPromiscuous, false);
1234
1235	/* check if it actually worked, this stuff is not always behaving well. */
1236	if (!(ifnet_flags(pIfNet) & IFF_PROMISC))
1237	{
1238	err = ifnet_set_flags(pIfNet, IFF_PROMISC, IFF_PROMISC);
1239	if (!err)
1240	err = ifnet_ioctl(pIfNet, 0, SIOCSIFFLAGS, NULL);
1241	if (!err)
1242	Log(("vboxNetFlt: fixed IFF_PROMISC on %s (%d->%d)\n", pThis->szName, cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1243	else
1244	Log(("VBoxNetFlt: failed to fix IFF_PROMISC on %s, err=%d (%d->%d)\n",
1245	pThis->szName, err, cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1246	}
1247	}
1248	else
1249	Log(("VBoxNetFlt: ifnet_set_promiscuous -> err=%d grr! (%d->%d)\n", err, cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1250	}
1251	else if (!err)
1252	Log(("VBoxNetFlt: Waiting for the link to come up... (%d->%d)\n", cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1253	if (err)
1254	LogRel(("VBoxNetFlt: Failed to put '%s' into promiscuous mode, err=%d (%d->%d)\n", pThis->szName, err, cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1255	}
1256	else
1257	{
1258	ASMAtomicWriteBool(&pThis->u.s.fNeedSetPromiscuous, false);
1259	if (pThis->u.s.fSetPromiscuous)
1260	{
1261	errno_t err = ifnet_set_promiscuous(pIfNet, 0);
1262	AssertMsg(!err, ("%d\n", err)); NOREF(err);
1263	}
1264	pThis->u.s.fSetPromiscuous = false;
1265
1266	fIf = ifnet_flags(pIfNet);
1267	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, VBOX_GET_PCOUNT(pIfNet)));
1268	}
1269	}
1270
1271	vboxNetFltDarwinReleaseIfNet(pThis, pIfNet);
1272	}
1273	}
1274
1275
1276	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
1277	{
1278	/* Nothing to do here. */
1279	return VINF_SUCCESS;
1280	}
1281
1282
1283	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
1284	{
1285	/* Nothing to do here. */
1286	return VINF_SUCCESS;
1287	}
1288
1289
1290	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
1291	{
1292	interface_filter_t pIfFilter;
1293
1294	/*
1295	* Carefully obtain the interface filter reference and detach it.
1296	*/
1297	RTSpinlockAcquire(pThis->hSpinlock);
1298	pIfFilter = ASMAtomicUoReadPtrT(&pThis->u.s.pIfFilter, interface_filter_t);
1299	if (pIfFilter)
1300	ASMAtomicUoWriteNullPtr(&pThis->u.s.pIfFilter);
1301	RTSpinlockRelease(pThis->hSpinlock);
1302
1303	if (pIfFilter)
1304	iflt_detach(pIfFilter);
1305
1306	if (pThis->u.s.pSysSock != NULL)
1307	{
1308	sock_close(pThis->u.s.pSysSock);
1309	pThis->u.s.pSysSock = NULL;
1310	}
1311	}
1312
1313
1314	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
1315	{
1316	NOREF(pvContext);
1317
1318	int rc = vboxNetFltDarwinAttachToInterface(pThis, false /* fRediscovery */);
1319	if (RT_FAILURE(rc))
1320	return rc;
1321
1322	if (pThis->pSwitchPort->pfnNotifyHostAddress == NULL)
1323	return rc;
1324
1325	/*
1326	* XXX: uwe
1327	*
1328	* Learn host's IP addresses and set up notifications for changes.
1329	* To avoid racing, set up notifications first.
1330	*
1331	* XXX: This should probably be global, since the only thing
1332	* specific to ifnet here is its IPv6 link-local address.
1333	*/
1334	errno_t error;
1335
1336	error = sock_socket(PF_SYSTEM, SOCK_RAW, SYSPROTO_EVENT,
1337	vboxNetFltDarwinSysSockUpcall, pThis,
1338	&pThis->u.s.pSysSock);
1339	if (error != 0)
1340	{
1341	LogRel(("sock_socket(SYSPROTO_EVENT): error %d\n", error));
1342	return rc;
1343	}
1344
1345	int nbio = 1;
1346	error = sock_ioctl(pThis->u.s.pSysSock, FIONBIO, &nbio);
1347	if (error != 0)
1348	{
1349	LogRel(("FIONBIO: error %d\n", error));
1350	sock_close(pThis->u.s.pSysSock);
1351	return rc;
1352	}
1353
1354	if (!sock_isnonblocking(pThis->u.s.pSysSock))
1355	{
1356	LogRel(("FIONBIO ok, but socket is blocking?!\n"));
1357	sock_close(pThis->u.s.pSysSock);
1358	return rc;
1359	}
1360
1361	struct kev_request req;
1362	req.vendor_code = KEV_VENDOR_APPLE;
1363	req.kev_class = KEV_NETWORK_CLASS;
1364	req.kev_subclass = KEV_ANY_SUBCLASS; /* need both INET and INET6, so have to request all */
1365
1366	error = sock_ioctl(pThis->u.s.pSysSock, SIOCSKEVFILT, &req);
1367	if (error != 0)
1368	{
1369	LogRel(("SIOCSKEVFILT: error %d\n", error));
1370	sock_close(pThis->u.s.pSysSock);
1371	return rc;
1372	}
1373
1374	ifnet_t pIfNet = pThis->u.s.pIfNet; /* already retained */
1375
1376	ifaddr_t *pIfAddrList;
1377	error = ifnet_get_address_list(/* all interfaces*/ NULL, &pIfAddrList);
1378	if (error != 0)
1379	{
1380	LogRel(("ifnet_get_address_list: error %d\n", error));
1381	return rc;
1382	}
1383
1384	for (ifaddr_t pIfAddr = pIfAddrList; pIfAddr != NULL; ++pIfAddr)
1385	{
1386	ifaddr_t ifa = *pIfAddr;
1387	sa_family_t family = ifaddr_address_family(ifa);
1388	struct sockaddr_storage ss;
1389
1390	error = ifaddr_address(ifa, (struct sockaddr *)&ss, sizeof(ss));
1391	if (error != 0)
1392	{
1393	LogRel(("getting address family %d: error %d\n", family, error));
1394	continue;
1395	}
1396
1397	if (family == AF_INET)
1398	{
1399	struct sockaddr_in sin = (struct sockaddr_in )&ss;
1400	u_int32_t u32Addr = ntohl(sin->sin_addr.s_addr);
1401
1402	if ((u32Addr >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
1403	continue;
1404
1405	Log(("> inet %RTnaipv4\n", sin->sin_addr.s_addr));
1406	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1407	/* :fAdded */ true, kIntNetAddrType_IPv4, &sin->sin_addr);
1408	}
1409	else if (family == AF_INET6)
1410	{
1411	struct sockaddr_in6 sin6 = (struct sockaddr_in6 )&ss;
1412
1413	if (IN6_IS_ADDR_LOOPBACK(&sin6->sin6_addr))
1414	continue;
1415
1416	/* link-local from other interfaces are out of scope */
1417	if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr) && ifaddr_ifnet(ifa) != pIfNet)
1418	continue;
1419
1420	Log(("> inet6 %RTnaipv6\n", &sin6->sin6_addr));
1421	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1422	/* :fAdded */ true, kIntNetAddrType_IPv6, &sin6->sin6_addr);
1423	}
1424	}
1425
1426	ifnet_free_address_list(pIfAddrList);
1427
1428	/*
1429	* Now that we've got current addresses, check for events that
1430	* might have happened while we were working.
1431	*/
1432	vboxNetFltDarwinSysSockUpcall(pThis->u.s.pSysSock, pThis, MBUF_DONTWAIT);
1433
1434	return rc;
1435	}
1436
1437
1438	static void vboxNetFltDarwinSysSockUpcall(socket_t pSysSock, void *pvData, int fWait)
1439	{
1440	PVBOXNETFLTINS pThis = (PVBOXNETFLTINS)pvData;
1441	errno_t error;
1442
1443	NOREF(fWait);
1444
1445	if (RT_UNLIKELY(pSysSock != pThis->u.s.pSysSock))
1446	{
1447	Log(("vboxNetFltDarwinSysSockUpcall: %p != %p?\n",
1448	pSysSock, pThis->u.s.pSysSock));
1449	return;
1450	}
1451
1452	for (;;) {
1453	mbuf_t m;
1454	size_t len = sizeof(struct kern_event_msg) - sizeof(u_int32_t)
1455	+ sizeof(struct kev_in6_data);
1456
1457	error = sock_receivembuf(pSysSock, NULL, &m, 0, &len);
1458	if (error == EWOULDBLOCK)
1459	{
1460	Log(("vboxNetFltDarwinSysSockUpcall: EWOULDBLOCK - we are done\n"));
1461	error = 0;
1462	break;
1463	}
1464	else if (error != 0)
1465	{
1466	Log(("sock_receivembuf: error %d\n", error));
1467	break;
1468	}
1469
1470	if (len < sizeof(struct kern_event_msg) - sizeof(u_int32_t))
1471	{
1472	Log(("vboxNetFltDarwinSysSockUpcall: %u bytes is too short\n",
1473	(unsigned int)len));
1474	mbuf_freem(m);
1475	return;
1476	}
1477
1478	struct kern_event_msg msg = (struct kern_event_msg )mbuf_data(m);
1479	if (msg->kev_subclass == KEV_INET_SUBCLASS)
1480	{
1481	if (len - (sizeof(struct kern_event_msg) - sizeof(u_int32_t)) < sizeof(struct kev_in_data))
1482	{
1483	Log(("vboxNetFltDarwinSysSockUpcall: %u bytes is too short for KEV_INET_SUBCLASS\n",
1484	(unsigned int)len));
1485	mbuf_freem(m);
1486	return;
1487	}
1488
1489	struct kev_in_data iev = (struct kev_in_data )msg->event_data;
1490	struct net_event_data *link = &iev->link_data;
1491	PCRTNETADDRU pAddr = (PCRTNETADDRU)&iev->ia_addr;
1492	switch (msg->event_code)
1493	{
1494	case KEV_INET_NEW_ADDR:
1495	Log(("KEV_INET_NEW_ADDR %.*s%d: %RTnaipv4\n",
1496	IFNAMSIZ, link->if_name, link->if_unit, pAddr->IPv4));
1497
1498	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1499	/* :fAdded */ true, kIntNetAddrType_IPv4, pAddr);
1500	break;
1501
1502	case KEV_INET_ADDR_DELETED:
1503	Log(("KEV_INET_ADDR_DELETED %.*s%d: %RTnaipv4\n",
1504	IFNAMSIZ, link->if_name, link->if_unit, pAddr->IPv4));
1505
1506	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1507	/* :fAdded */ false, kIntNetAddrType_IPv4, pAddr);
1508	break;
1509
1510	default:
1511	Log(("KEV INET event %u %.*s%d: addr %RTnaipv4\n",
1512	msg->event_code, IFNAMSIZ, link->if_name, link->if_unit, pAddr->IPv4));
1513	break;
1514	}
1515	}
1516	else if (msg->kev_subclass == KEV_INET6_SUBCLASS)
1517	{
1518	if (len - (sizeof(struct kern_event_msg) - sizeof(u_int32_t)) < sizeof(struct kev_in6_data))
1519	{
1520	Log(("vboxNetFltDarwinSysSockUpcall: %u bytes is too short for KEV_INET6_SUBCLASS\n",
1521	(unsigned int)len));
1522	mbuf_freem(m);
1523	return;
1524	}
1525
1526	struct kev_in6_data iev6 = (struct kev_in6_data )msg->event_data;
1527	struct net_event_data *link = &iev6->link_data;
1528	PCRTNETADDRU pAddr = (PCRTNETADDRU)&iev6->ia_addr.sin6_addr;
1529	switch (msg->event_code)
1530	{
1531	case KEV_INET6_NEW_USER_ADDR:
1532	Log(("KEV_INET6_NEW_USER_ADDR %.*s%d: %RTnaipv6\n",
1533	IFNAMSIZ, link->if_name, link->if_unit, pAddr));
1534	goto kev_inet6_new;
1535
1536	case KEV_INET6_NEW_LL_ADDR:
1537	Log(("KEV_INET6_NEW_LL_ADDR %.*s%d: %RTnaipv6\n",
1538	IFNAMSIZ, link->if_name, link->if_unit, pAddr));
1539	/* XXX: uwe: TODO: only interface we are attached to */
1540	goto kev_inet6_new;
1541
1542	case KEV_INET6_NEW_RTADV_ADDR:
1543	Log(("KEV_INET6_NEW_RTADV_ADDR %.*s%d: %RTnaipv6\n",
1544	IFNAMSIZ, link->if_name, link->if_unit, pAddr));
1545	goto kev_inet6_new;
1546
1547	kev_inet6_new:
1548	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1549	/* :fAdded */ true, kIntNetAddrType_IPv6, pAddr);
1550	break;
1551
1552	case KEV_INET6_ADDR_DELETED:
1553	Log(("KEV_INET6_ADDR_DELETED %.*s%d: %RTnaipv6\n",
1554	IFNAMSIZ, link->if_name, link->if_unit, pAddr));
1555
1556	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1557	/* :fAdded */ false, kIntNetAddrType_IPv6, pAddr);
1558	break;
1559
1560	default:
1561	Log(("KEV INET6 event %u %.*s%d: addr %RTnaipv6\n",
1562	msg->event_code, IFNAMSIZ, link->if_name, link->if_unit, pAddr));
1563	break;
1564	}
1565	}
1566	else
1567	{
1568	Log(("vboxNetFltDarwinSysSockUpcall: subclass %u ignored\n",
1569	(unsigned)msg->kev_subclass));
1570	}
1571
1572	mbuf_freem(m);
1573	}
1574	}
1575
1576
1577	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
1578	{
1579	/*
1580	* Init the darwin specific members.
1581	*/
1582	pThis->u.s.pIfNet = NULL;
1583	pThis->u.s.pIfFilter = NULL;
1584	pThis->u.s.fSetPromiscuous = false;
1585	pThis->u.s.fNeedSetPromiscuous = false;
1586	//pThis->u.s.MacAddr = {0};
1587	pThis->u.s.pSysSock = NULL;
1588
1589	return VINF_SUCCESS;
1590	}
1591
1592
1593	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
1594	{
1595	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
1596	}
1597
1598
1599	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *ppvIfData)
1600	{
1601	/* Nothing to do */
1602	NOREF(pThis); NOREF(pvIf); NOREF(ppvIfData);
1603	return VINF_SUCCESS;
1604	}
1605
1606
1607	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
1608	{
1609	/* Nothing to do */
1610	NOREF(pThis); NOREF(pvIfData);
1611	return VINF_SUCCESS;
1612	}
1613

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/darwin/VBoxNetFlt-darwin.cpp@ 56266

Download in other formats: