VBoxNetFlt-linux.c@ 94991

Last change on this file since 94991 was 94967, checked in by vboxsync, 3 years ago
HostDrivers/{VBoxNetAdp,VBoxNetFlt}: Advertise TSO support to increase performance, bugref:9723
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 90.1 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 94967 2022-05-09 16:58:18Z vboxsync $ */
2	/** @file
3	* VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2022 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	* The contents of this file may alternatively be used under the terms
18	* of the Common Development and Distribution License Version 1.0
19	* (CDDL) only, as it comes in the "COPYING.CDDL" file of the
20	* VirtualBox OSE distribution, in which case the provisions of the
21	* CDDL are applicable instead of those of the GPL.
22	*
23	* You may elect to license modified versions of this file under the
24	* terms and conditions of either the GPL or the CDDL or both.
25	*/
26
27
28	/*********************************************************************************************************************************
29	* Header Files *
30	*********************************************************************************************************************************/
31	#define LOG_GROUP LOG_GROUP_NET_FLT_DRV
32	#define VBOXNETFLT_LINUX_NO_XMIT_QUEUE
33	#include "the-linux-kernel.h"
34	#include "version-generated.h"
35	#include "revision-generated.h"
36	#include "product-generated.h"
37	#if RTLNX_VER_MIN(2,6,24)
38	# include <linux/nsproxy.h>
39	#endif
40	#include <linux/netdevice.h>
41	#if RTLNX_VER_MAX(2,6,29) \|\| RTLNX_VER_MIN(5,11,0)
42	# include <linux/ethtool.h>
43	#endif
44	#include <linux/etherdevice.h>
45	#include <linux/rtnetlink.h>
46	#include <linux/miscdevice.h>
47	#include <linux/inetdevice.h>
48	#include <linux/in.h>
49	#include <linux/ip.h>
50	#include <linux/if_vlan.h>
51	#if RTLNX_VER_MIN(4,5,0)
52	# include <uapi/linux/pkt_cls.h>
53	#endif
54	#include <net/ipv6.h>
55	#include <net/if_inet6.h>
56	#include <net/addrconf.h>
57
58	#include <VBox/log.h>
59	#include <VBox/err.h>
60	#include <VBox/intnetinline.h>
61	#include <VBox/vmm/pdmnetinline.h>
62	#include <VBox/param.h>
63	#include <iprt/alloca.h>
64	#include <iprt/assert.h>
65	#include <iprt/spinlock.h>
66	#include <iprt/semaphore.h>
67	#include <iprt/initterm.h>
68	#include <iprt/process.h>
69	#include <iprt/mem.h>
70	#include <iprt/net.h>
71	#include <iprt/log.h>
72	#include <iprt/mp.h>
73	#include <iprt/mem.h>
74	#include <iprt/time.h>
75
76	#define VBOXNETFLT_OS_SPECFIC 1
77	#include "../VBoxNetFltInternal.h"
78
79	typedef struct VBOXNETFLTNOTIFIER {
80	struct notifier_block Notifier;
81	PVBOXNETFLTINS pThis;
82	} VBOXNETFLTNOTIFIER;
83	typedef struct VBOXNETFLTNOTIFIER *PVBOXNETFLTNOTIFIER;
84
85
86	/*********************************************************************************************************************************
87	* Defined Constants And Macros *
88	*********************************************************************************************************************************/
89	#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
90	#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
91	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
92	# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
93	#endif
94
95	#if RTLNX_VER_MIN(3,11,0)
96	# define VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr) netdev_notifier_info_to_dev(ptr)
97	#else
98	# define VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr) ((struct net_device *)ptr)
99	#endif
100
101	#if RTLNX_VER_MIN(3,5,0)
102	# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(skb_frag_page(frag))
103	# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr)
104	#else
105	# if RTLNX_VER_MIN(3,2,0)
106	# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(skb_frag_page(frag), KM_SKB_DATA_SOFTIRQ)
107	# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ)
108	# else
109	# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(frag->page, KM_SKB_DATA_SOFTIRQ)
110	# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ)
111	# endif
112	#endif
113
114	#if RTLNX_VER_MIN(2,6,34)
115	# define VBOX_NETDEV_NAME(dev) netdev_name(dev)
116	#else
117	# define VBOX_NETDEV_NAME(dev) ((dev)->reg_state != NETREG_REGISTERED ? "(unregistered net_device)" : (dev)->name)
118	#endif
119
120	#if RTLNX_VER_MIN(2,6,25)
121	# define VBOX_IPV4_IS_LOOPBACK(addr) ipv4_is_loopback(addr)
122	# define VBOX_IPV4_IS_LINKLOCAL_169(addr) ipv4_is_linklocal_169(addr)
123	#else
124	# define VBOX_IPV4_IS_LOOPBACK(addr) ((addr & htonl(IN_CLASSA_NET)) == htonl(0x7f000000))
125	# define VBOX_IPV4_IS_LINKLOCAL_169(addr) ((addr & htonl(IN_CLASSB_NET)) == htonl(0xa9fe0000))
126	#endif
127
128	#if RTLNX_VER_MIN(2,6,22)
129	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
130	# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
131	# define VBOX_SKB_CSUM_OFFSET(skb) skb->csum_offset
132	#else
133	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
134	# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
135	# define VBOX_SKB_CSUM_OFFSET(skb) skb->csum
136	#endif
137
138	#if RTLNX_VER_MIN(2,6,19)
139	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
140	#else
141	# define CHECKSUM_PARTIAL CHECKSUM_HW
142	# if RTLNX_VER_MIN(2,6,10)
143	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
144	# else
145	# if RTLNX_VER_MIN(2,6,7)
146	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
147	# else
148	# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
149	# endif
150	/* Versions prior 2.6.10 use stats for both bstats and qstats */
151	# define bstats stats
152	# define qstats stats
153	# endif
154	#endif
155
156	#if RTLNX_VER_MIN(3,20,0) \|\| RTLNX_RHEL_RANGE(7,2, 8,0) \|\| RTLNX_RHEL_RANGE(6,8, 7,0)
157	# define VBOX_HAVE_SKB_VLAN
158	#endif
159
160	#ifdef VBOX_HAVE_SKB_VLAN
161	# define vlan_tx_tag_get(skb) skb_vlan_tag_get(skb)
162	# define vlan_tx_tag_present(skb) skb_vlan_tag_present(skb)
163	#endif
164
165	#ifndef NET_IP_ALIGN
166	# define NET_IP_ALIGN 2
167	#endif
168
169	#if 1
170	/** Create scatter / gather segments for fragments. When not used, we will
171	* linearize the socket buffer before creating the internal networking SG. */
172	# define VBOXNETFLT_SG_SUPPORT 1
173	#endif
174
175	#if RTLNX_VER_MIN(2,6,18)
176
177	/** Indicates that the linux kernel may send us GSO frames. */
178	# define VBOXNETFLT_WITH_GSO 1
179
180	/** This enables or disables the transmitting of GSO frame from the internal
181	* network and to the host. */
182	# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
183
184	# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
185	/** This enables or disables the transmitting of GSO frame from the internal
186	* network and to the wire. */
187	# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
188	# endif
189
190	/** This enables or disables the forwarding/flooding of GSO frame from the host
191	* to the internal network. */
192	# define VBOXNETFLT_WITH_GSO_RECV 1
193
194	#endif /* RTLNX_VER_MIN(2,6,18) */
195
196	#if RTLNX_VER_MIN(2,6,29)
197	/** This enables or disables handling of GSO frames coming from the wire (GRO). */
198	# define VBOXNETFLT_WITH_GRO 1
199	#endif
200
201	/*
202	* GRO support was backported to RHEL 5.4
203	*/
204	#if RTLNX_RHEL_MAJ_PREREQ(5, 4)
205	# define VBOXNETFLT_WITH_GRO 1
206	#endif
207
208
209	/*********************************************************************************************************************************
210	* Internal Functions *
211	*********************************************************************************************************************************/
212	static int __init VBoxNetFltLinuxInit(void);
213	static void __exit VBoxNetFltLinuxUnload(void);
214	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
215
216
217	/*********************************************************************************************************************************
218	* Global Variables *
219	*********************************************************************************************************************************/
220	/**
221	* The (common) global data.
222	*/
223	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
224
225	module_init(VBoxNetFltLinuxInit);
226	module_exit(VBoxNetFltLinuxUnload);
227
228	MODULE_AUTHOR(VBOX_VENDOR);
229	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
230	MODULE_LICENSE("GPL");
231	#ifdef MODULE_VERSION
232	MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV) " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
233	#endif
234
235
236	#if RTLNX_VER_MAX(2,6,12) && defined(LOG_ENABLED)
237	unsigned dev_get_flags(const struct net_device *dev)
238	{
239	unsigned flags;
240
241	flags = (dev->flags & ~(IFF_PROMISC \|
242	IFF_ALLMULTI \|
243	IFF_RUNNING)) \|
244	(dev->gflags & (IFF_PROMISC \|
245	IFF_ALLMULTI));
246
247	if (netif_running(dev) && netif_carrier_ok(dev))
248	flags \|= IFF_RUNNING;
249
250	return flags;
251	}
252	#endif /* RTLNX_VER_MAX(2,6,12) */
253
254
255	/**
256	* Initialize module.
257	*
258	* @returns appropriate status code.
259	*/
260	static int __init VBoxNetFltLinuxInit(void)
261	{
262	int rc;
263	/*
264	* Initialize IPRT.
265	*/
266	rc = RTR0Init(0);
267	if (RT_SUCCESS(rc))
268	{
269	Log(("VBoxNetFltLinuxInit\n"));
270
271	/*
272	* Initialize the globals and connect to the support driver.
273	*
274	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
275	* for establishing the connect to the support driver.
276	*/
277	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
278	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
279	if (RT_SUCCESS(rc))
280	{
281	LogRel(("VBoxNetFlt: Successfully started.\n"));
282	return 0;
283	}
284
285	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
286	RTR0Term();
287	}
288	else
289	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
290
291	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
292	return -RTErrConvertToErrno(rc);
293	}
294
295
296	/**
297	* Unload the module.
298	*
299	* @todo We have to prevent this if we're busy!
300	*/
301	static void __exit VBoxNetFltLinuxUnload(void)
302	{
303	int rc;
304	Log(("VBoxNetFltLinuxUnload\n"));
305	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
306
307	/*
308	* Undo the work done during start (in reverse order).
309	*/
310	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
311	AssertRC(rc); NOREF(rc);
312
313	RTR0Term();
314
315	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
316
317	Log(("VBoxNetFltLinuxUnload - done\n"));
318	}
319
320
321	/**
322	* We filter traffic from the host to the internal network
323	* before it reaches the NIC driver.
324	*
325	* The current code uses a very ugly hack overriding hard_start_xmit
326	* callback in the device structure, but it has been shown to give us a
327	* performance boost of 60-100% though. Eventually we have to find some
328	* less hacky way of getting this job done.
329	*/
330	#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
331
332	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
333
334	# if RTLNX_VER_MAX(2,6,29)
335
336	typedef struct ethtool_ops OVR_OPSTYPE;
337	# define OVR_OPS ethtool_ops
338	# define OVR_XMIT pfnStartXmit
339
340	# else /* RTLNX_VER_MIN(2,6,29) */
341
342	typedef struct net_device_ops OVR_OPSTYPE;
343	# define OVR_OPS netdev_ops
344	# define OVR_XMIT pOrgOps->ndo_start_xmit
345
346	# endif /* RTLNX_VER_MIN(2,6,29) */
347
348	/**
349	* The overridden net_device_ops of the device we're attached to.
350	*
351	* As there is no net_device_ops structure in pre-2.6.29 kernels we override
352	* ethtool_ops instead along with hard_start_xmit callback in net_device
353	* structure.
354	*
355	* This is a very dirty hack that was created to explore how much we can improve
356	* the host to guest transfers by not CC'ing the NIC. It turns out to be
357	* the only way to filter outgoing packets for devices without TX queue.
358	*/
359	typedef struct VBoxNetDeviceOpsOverride
360	{
361	/** Our overridden ops. */
362	OVR_OPSTYPE Ops;
363	/** Magic word. */
364	uint32_t u32Magic;
365	/** Pointer to the original ops. */
366	OVR_OPSTYPE const *pOrgOps;
367	# if RTLNX_VER_MAX(2,6,29)
368	/** Pointer to the original hard_start_xmit function. */
369	int (pfnStartXmit)(struct sk_buff pSkb, struct net_device *pDev);
370	# endif /* RTLNX_VER_MAX(2,6,29) */
371	/** Pointer to the net filter instance. */
372	PVBOXNETFLTINS pVBoxNetFlt;
373	/** The number of filtered packages. */
374	uint64_t cFiltered;
375	/** The total number of packets */
376	uint64_t cTotal;
377	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
378	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
379	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
380
381	/**
382	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
383	* because they belong on the internal network.
384	*
385	* @returns NETDEV_TX_XXX.
386	* @param pSkb The socket buffer to transmit.
387	* @param pDev The net device.
388	*/
389	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
390	{
391	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
392	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
393	PCRTNETETHERHDR pEtherHdr;
394	PINTNETTRUNKSWPORT pSwitchPort;
395	uint32_t cbHdrs;
396
397
398	/*
399	* Validate the override structure.
400	*
401	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
402	* to be production quality code, we would have to be much more
403	* careful here and avoid the race.
404	*/
405	if ( !RT_VALID_PTR(pOverride)
406	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
407	# if RTLNX_VER_MIN(2,6,29)
408	\|\| !RT_VALID_PTR(pOverride->pOrgOps)
409	# endif
410	)
411	{
412	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
413	dev_kfree_skb(pSkb);
414	return NETDEV_TX_OK;
415	}
416	pOverride->cTotal++;
417
418	/*
419	* Do the filtering base on the default OUI of our virtual NICs
420	*
421	* Note! In a real solution, we would ask the switch whether the
422	* destination MAC is 100% to be on the internal network and then
423	* drop it.
424	*/
425	cbHdrs = skb_headlen(pSkb);
426	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
427	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
428	if ( pEtherHdr
429	&& RT_VALID_PTR(pOverride->pVBoxNetFlt)
430	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
431	&& RT_VALID_PTR(pSwitchPort)
432	&& cbHdrs >= 6)
433	{
434	INTNETSWDECISION enmDecision;
435
436	/** @todo consider reference counting, etc. */
437	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
438	if (enmDecision == INTNETSWDECISION_INTNET)
439	{
440	dev_kfree_skb(pSkb);
441	pOverride->cFiltered++;
442	return NETDEV_TX_OK;
443	}
444	}
445
446	return pOverride->OVR_XMIT(pSkb, pDev);
447	}
448
449	/**
450	* Hooks the device ndo_start_xmit operation of the device.
451	*
452	* @param pThis The net filter instance.
453	* @param pDev The net device.
454	*/
455	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
456	{
457	PVBOXNETDEVICEOPSOVERRIDE pOverride;
458
459	/* Cancel override if ethtool_ops is missing (host-only case, @bugref{5712}) */
460	if (!RT_VALID_PTR(pDev->OVR_OPS))
461	return;
462	pOverride = RTMemAlloc(sizeof(*pOverride));
463	if (!pOverride)
464	return;
465	pOverride->pOrgOps = pDev->OVR_OPS;
466	pOverride->Ops = *pDev->OVR_OPS;
467	# if RTLNX_VER_MAX(2,6,29)
468	pOverride->pfnStartXmit = pDev->hard_start_xmit;
469	# else /* RTLNX_VER_MIN(2,6,29) */
470	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
471	# endif /* RTLNX_VER_MIN(2,6,29) */
472	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
473	pOverride->cTotal = 0;
474	pOverride->cFiltered = 0;
475	pOverride->pVBoxNetFlt = pThis;
476
477	RTSpinlockAcquire(pThis->hSpinlock); /* (this isn't necessary, but so what) */
478	ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
479	# if RTLNX_VER_MAX(2,6,29)
480	ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
481	# endif /* RTLNX_VER_MAX(2,6,29) */
482	RTSpinlockRelease(pThis->hSpinlock);
483	}
484
485	/**
486	* Undos what vboxNetFltLinuxHookDev did.
487	*
488	* @param pThis The net filter instance.
489	* @param pDev The net device. Can be NULL, in which case
490	* we'll try retrieve it from @a pThis.
491	*/
492	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
493	{
494	PVBOXNETDEVICEOPSOVERRIDE pOverride;
495
496	RTSpinlockAcquire(pThis->hSpinlock);
497	if (!pDev)
498	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
499	if (RT_VALID_PTR(pDev))
500	{
501	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
502	if ( RT_VALID_PTR(pOverride)
503	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
504	&& RT_VALID_PTR(pOverride->pOrgOps)
505	)
506	{
507	# if RTLNX_VER_MAX(2,6,29)
508	ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
509	# endif /* RTLNX_VER_MAX(2,6,29) */
510	ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
511	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
512	}
513	else
514	pOverride = NULL;
515	}
516	else
517	pOverride = NULL;
518	RTSpinlockRelease(pThis->hSpinlock);
519
520	if (pOverride)
521	{
522	printk("vboxnetflt: %llu out of %llu packets were not sent (directed to host)\n", pOverride->cFiltered, pOverride->cTotal);
523	RTMemFree(pOverride);
524	}
525	}
526
527	#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
528
529
530	/**
531	* Reads and retains the host interface handle.
532	*
533	* @returns The handle, NULL if detached.
534	* @param pThis
535	*/
536	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
537	{
538	#if 0
539	struct net_device *pDev = NULL;
540
541	Log(("vboxNetFltLinuxRetainNetDev\n"));
542	/*
543	* Be careful here to avoid problems racing the detached callback.
544	*/
545	RTSpinlockAcquire(pThis->hSpinlock);
546	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
547	{
548	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
549	if (pDev)
550	{
551	dev_hold(pDev);
552	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
553	pDev, pDev->name,
554	#if RTLNX_VER_MIN(2,6,37)
555	netdev_refcnt_read(pDev)
556	#else
557	atomic_read(&pDev->refcnt)
558	#endif
559	));
560	}
561	}
562	RTSpinlockRelease(pThis->hSpinlock);
563
564	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
565	return pDev;
566	#else
567	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
568	#endif
569	}
570
571
572	/**
573	* Release the host interface handle previously retained
574	* by vboxNetFltLinuxRetainNetDev.
575	*
576	* @param pThis The instance.
577	* @param pDev The vboxNetFltLinuxRetainNetDev
578	* return value, NULL is fine.
579	*/
580	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
581	{
582	#if 0
583	Log(("vboxNetFltLinuxReleaseNetDev\n"));
584	NOREF(pThis);
585	if (pDev)
586	{
587	dev_put(pDev);
588	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
589	pDev, pDev->name,
590	#if RTLNX_VER_MIN(2,6,37)
591	netdev_refcnt_read(pDev)
592	#else
593	atomic_read(&pDev->refcnt)
594	#endif
595	));
596	}
597	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
598	#endif
599	}
600
601	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
602	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
603
604	/**
605	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
606	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
607	*
608	* @returns true / false accordingly.
609	* @param pBuf The sk_buff.
610	*/
611	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
612	{
613	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
614	}
615
616
617	/**
618	* Checks whether this SG list contains a GSO packet.
619	*
620	* @returns true / false accordingly.
621	* @param pSG The (scatter/)gather list.
622	*/
623	DECLINLINE(bool) vboxNetFltLinuxIsGso(PINTNETSG pSG)
624	{
625	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
626	return !((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type == PDMNETWORKGSOTYPE_INVALID);
627	#else /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
628	return false;
629	#endif /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
630	}
631
632
633	/**
634	* Find out the frame size (of a single segment in case of GSO frames).
635	*
636	* @returns the frame size.
637	* @param pSG The (scatter/)gather list.
638	*/
639	DECLINLINE(uint32_t) vboxNetFltLinuxFrameSize(PINTNETSG pSG)
640	{
641	uint16_t u16Type = 0;
642	uint32_t cbVlanTag = 0;
643	if (pSG->aSegs[0].cb >= sizeof(RTNETETHERHDR))
644	u16Type = RT_BE2H_U16(((PCRTNETETHERHDR)pSG->aSegs[0].pv)->EtherType);
645	else if (pSG->cbTotal >= sizeof(RTNETETHERHDR))
646	{
647	uint32_t off = RT_UOFFSETOF(RTNETETHERHDR, EtherType);
648	uint32_t i;
649	for (i = 0; i < pSG->cSegsUsed; ++i)
650	{
651	if (off <= pSG->aSegs[i].cb)
652	{
653	if (off + sizeof(uint16_t) <= pSG->aSegs[i].cb)
654	u16Type = RT_BE2H_U16((uint16_t )((uintptr_t)pSG->aSegs[i].pv + off));
655	else if (i + 1 < pSG->cSegsUsed)
656	u16Type = RT_BE2H_U16( ((uint16_t)( ((uint8_t *)pSG->aSegs[i].pv)[off] ) << 8)
657	+ (uint8_t )pSG->aSegs[i + 1].pv); /* ASSUMES no empty segments! */
658	/* else: frame is too short. */
659	break;
660	}
661	off -= pSG->aSegs[i].cb;
662	}
663	}
664	if (u16Type == RTNET_ETHERTYPE_VLAN)
665	cbVlanTag = 4;
666	return (vboxNetFltLinuxIsGso(pSG) ? (uint32_t)pSG->GsoCtx.cbMaxSeg + pSG->GsoCtx.cbHdrsTotal : pSG->cbTotal) - cbVlanTag;
667	}
668
669
670	/**
671	* Internal worker that create a linux sk_buff for a
672	* (scatter/)gather list.
673	*
674	* @returns Pointer to the sk_buff.
675	* @param pThis The instance.
676	* @param pSG The (scatter/)gather list.
677	* @param fDstWire Set if the destination is the wire.
678	*/
679	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
680	{
681	struct sk_buff *pPkt;
682	struct net_device *pDev;
683	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
684	unsigned fGsoType = 0;
685	#endif
686
687	if (pSG->cbTotal == 0)
688	{
689	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
690	return NULL;
691	}
692	Log5(("VBoxNetFlt: Packet to %s of %d bytes (frame=%d).\n", fDstWire?"wire":"host", pSG->cbTotal, vboxNetFltLinuxFrameSize(pSG)));
693	if (fDstWire && (vboxNetFltLinuxFrameSize(pSG) > ASMAtomicReadU32(&pThis->u.s.cbMtu) + 14))
694	{
695	static bool s_fOnce = true;
696	if (s_fOnce)
697	{
698	s_fOnce = false;
699	printk("VBoxNetFlt: Dropped over-sized packet (%d bytes) coming from internal network.\n", vboxNetFltLinuxFrameSize(pSG));
700	}
701	return NULL;
702	}
703
704	/** @todo We should use fragments mapping the SG buffers with large packets.
705	* 256 bytes seems to be the a threshold used a lot for this. It
706	* requires some nasty work on the intnet side though... */
707	/*
708	* Allocate a packet and copy over the data.
709	*/
710	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
711	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
712	if (RT_UNLIKELY(!pPkt))
713	{
714	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
715	pSG->pvUserData = NULL;
716	return NULL;
717	}
718	pPkt->dev = pDev;
719	pPkt->ip_summed = CHECKSUM_NONE;
720
721	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
722	skb_reserve(pPkt, NET_IP_ALIGN);
723
724	/* Copy the segments. */
725	skb_put(pPkt, pSG->cbTotal);
726	IntNetSgRead(pSG, pPkt->data);
727
728	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
729	/*
730	* Setup GSO if used by this packet.
731	*/
732	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
733	{
734	default:
735	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
736	RT_FALL_THRU();
737	case PDMNETWORKGSOTYPE_INVALID:
738	fGsoType = 0;
739	break;
740	case PDMNETWORKGSOTYPE_IPV4_TCP:
741	fGsoType = SKB_GSO_TCPV4;
742	break;
743	case PDMNETWORKGSOTYPE_IPV6_TCP:
744	fGsoType = SKB_GSO_TCPV6;
745	break;
746	}
747	if (fGsoType)
748	{
749	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
750
751	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
752	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
753	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
754
755	/*
756	* We need to set checksum fields even if the packet goes to the host
757	* directly as it may be immediately forwarded by IP layer @bugref{5020}.
758	*/
759	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
760	pPkt->ip_summed = CHECKSUM_PARTIAL;
761	# if RTLNX_VER_MIN(2,6,22)
762	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
763	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
764	pPkt->csum_offset = RT_UOFFSETOF(RTNETTCP, th_sum);
765	else
766	pPkt->csum_offset = RT_UOFFSETOF(RTNETUDP, uh_sum);
767	# else
768	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
769	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
770	pPkt->csum = RT_UOFFSETOF(RTNETTCP, th_sum);
771	else
772	pPkt->csum = RT_UOFFSETOF(RTNETUDP, uh_sum);
773	# endif
774	if (!fDstWire)
775	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
776	}
777	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
778
779	/*
780	* Finish up the socket buffer.
781	*/
782	pPkt->protocol = eth_type_trans(pPkt, pDev);
783	if (fDstWire)
784	{
785	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
786
787	/* Restore ethernet header back. */
788	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
789	VBOX_SKB_RESET_MAC_HDR(pPkt);
790	}
791	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
792
793	return pPkt;
794	}
795
796
797	/**
798	* Return the offset where to start checksum computation from.
799	*
800	* @returns the offset relative to pBuf->data.
801	* @param pBuf The socket buffer.
802	*/
803	DECLINLINE(unsigned) vboxNetFltLinuxGetChecksumStartOffset(struct sk_buff *pBuf)
804	{
805	#if RTLNX_VER_MIN(2,6,38)
806	return skb_checksum_start_offset(pBuf);
807	#elif RTLNX_VER_MIN(2,6,22)
808	return pBuf->csum_start - skb_headroom(pBuf);
809	#else
810	unsigned char *pTransportHdr = pBuf->h.raw;
811	# if RTLNX_VER_MAX(2,6,19)
812	/*
813	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
814	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
815	* header length from the header itself and reconstruct 'h' pointer
816	* to TCP (or whatever) header.
817	*/
818	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
819	pTransportHdr = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
820	# endif
821	return pTransportHdr - pBuf->data;
822	#endif
823	}
824
825
826	/**
827	* Initializes a SG list from an sk_buff.
828	*
829	* @returns Number of segments.
830	* @param pThis The instance.
831	* @param pBuf The sk_buff.
832	* @param pSG The SG.
833	* @param cbExtra The number of bytes of extra space allocated immediately after the SG.
834	* @param cSegs The number of segments allocated for the SG.
835	* This should match the number in the mbuf exactly!
836	* @param fSrc The source of the frame.
837	* @param pGsoCtx Pointer to the GSO context if it's a GSO
838	* internal network frame. NULL if regular frame.
839	*/
840	static void vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
841	unsigned cbExtra, unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
842	{
843	int i;
844	NOREF(pThis);
845
846	#ifndef VBOXNETFLT_SG_SUPPORT
847	Assert(!skb_shinfo(pBuf)->frag_list);
848	#else /* VBOXNETFLT_SG_SUPPORT */
849	uint8_t pExtra = (uint8_t )&pSG->aSegs[cSegs];
850	unsigned cbConsumed = 0;
851	unsigned cbProduced = 0;
852
853	# if RTLNX_VER_MIN(2,6,27)
854	/* Restore VLAN tag stripped by host hardware */
855	if (vlan_tx_tag_present(pBuf))
856	{
857	uint8_t *pMac = pBuf->data;
858	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )pExtra;
859	Assert(ETH_ALEN * 2 + VLAN_HLEN <= cbExtra);
860	memmove(pVHdr, pMac, ETH_ALEN * 2);
861	/* Consume whole Ethernet header: 2 addresses + EtherType (see @bugref{8599}) */
862	cbConsumed += ETH_ALEN * 2 + sizeof(uint16_t);
863	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
864	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
865	pVHdr->h_vlan_encapsulated_proto = (uint16_t)(pMac + ETH_ALEN * 2);
866	cbProduced += VLAN_ETH_HLEN;
867	}
868	# endif /* RTLNX_VER_MIN(2,6,27) */
869
870	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
871	{
872	unsigned uCsumStartOffset = vboxNetFltLinuxGetChecksumStartOffset(pBuf);
873	unsigned uCsumStoreOffset = uCsumStartOffset + VBOX_SKB_CSUM_OFFSET(pBuf) - cbConsumed;
874	Log3(("cbConsumed=%u cbProduced=%u uCsumStartOffset=%u uCsumStoreOffset=%u\n",
875	cbConsumed, cbProduced, uCsumStartOffset, uCsumStoreOffset));
876	Assert(cbProduced + uCsumStoreOffset + sizeof(uint16_t) <= cbExtra);
877	/*
878	* We assume that the checksum is stored at the very end of the transport header
879	* so we will have all headers in a single fragment. If our assumption is wrong
880	* we may see suboptimal performance.
881	*/
882	memmove(pExtra + cbProduced,
883	pBuf->data + cbConsumed,
884	uCsumStoreOffset);
885	unsigned uChecksum = skb_checksum(pBuf, uCsumStartOffset, pBuf->len - uCsumStartOffset, 0);
886	(uint16_t)(pExtra + cbProduced + uCsumStoreOffset) = csum_fold(uChecksum);
887	cbProduced += uCsumStoreOffset + sizeof(uint16_t);
888	cbConsumed += uCsumStoreOffset + sizeof(uint16_t);
889	}
890	#endif /* VBOXNETFLT_SG_SUPPORT */
891
892	if (!pGsoCtx)
893	IntNetSgInitTempSegs(pSG, pBuf->len + cbProduced - cbConsumed, cSegs, 0 /cSegsUsed/);
894	else
895	IntNetSgInitTempSegsGso(pSG, pBuf->len + cbProduced - cbConsumed, cSegs, 0 /cSegsUsed/, pGsoCtx);
896
897	int iSeg = 0;
898	#ifdef VBOXNETFLT_SG_SUPPORT
899	if (cbProduced)
900	{
901	pSG->aSegs[iSeg].cb = cbProduced;
902	pSG->aSegs[iSeg].pv = pExtra;
903	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
904	}
905	pSG->aSegs[iSeg].cb = skb_headlen(pBuf) - cbConsumed;
906	pSG->aSegs[iSeg].pv = pBuf->data + cbConsumed;
907	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
908	Assert(iSeg <= pSG->cSegsAlloc);
909
910	# ifdef LOG_ENABLED
911	if (pBuf->data_len)
912	Log6((" kmap_atomic:"));
913	# endif /* LOG_ENABLED */
914	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
915	{
916	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
917	# if RTLNX_VER_MIN(5,4,0) \|\| RTLNX_SUSE_MAJ_PREREQ(15, 2)
918	pSG->aSegs[iSeg].cb = pFrag->bv_len;
919	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->bv_offset;
920	# else /* < KERNEL_VERSION(5, 4, 0) */
921	pSG->aSegs[iSeg].cb = pFrag->size;
922	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
923	# endif /* >= KERNEL_VERSION(5, 4, 0) */
924	Log6((" %p", pSG->aSegs[iSeg].pv));
925	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
926	Assert(iSeg <= pSG->cSegsAlloc);
927	}
928	struct sk_buff *pFragBuf;
929	for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
930	{
931	pSG->aSegs[iSeg].cb = skb_headlen(pFragBuf);
932	pSG->aSegs[iSeg].pv = pFragBuf->data;
933	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
934	Assert(iSeg <= pSG->cSegsAlloc);
935	for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
936	{
937	skb_frag_t *pFrag = &skb_shinfo(pFragBuf)->frags[i];
938	# if RTLNX_VER_MIN(5,4,0) \|\| RTLNX_SUSE_MAJ_PREREQ(15, 2)
939	pSG->aSegs[iSeg].cb = pFrag->bv_len;
940	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->bv_offset;
941	# else /* < KERNEL_VERSION(5, 4, 0) */
942	pSG->aSegs[iSeg].cb = pFrag->size;
943	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
944	# endif /* >= KERNEL_VERSION(5, 4, 0) */
945	Log6((" %p", pSG->aSegs[iSeg].pv));
946	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
947	Assert(iSeg <= pSG->cSegsAlloc);
948	}
949	}
950	# ifdef LOG_ENABLED
951	if (pBuf->data_len)
952	Log6(("\n"));
953	# endif /* LOG_ENABLED */
954	#else
955	pSG->aSegs[iSeg].cb = pBuf->len;
956	pSG->aSegs[iSeg].pv = pBuf->data;
957	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
958	#endif
959
960	pSG->cSegsUsed = iSeg;
961
962	#if 0
963	if (cbProduced)
964	{
965	LogRel(("vboxNetFltLinuxSkBufToSG: original packet dump:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
966	LogRel(("vboxNetFltLinuxSkBufToSG: cbConsumed=%u cbProduced=%u cbExtra=%u\n", cbConsumed, cbProduced, cbExtra));
967	uint32_t offset = 0;
968	for (i = 0; i < pSG->cSegsUsed; ++i)
969	{
970	LogRel(("vboxNetFltLinuxSkBufToSG: seg#%d (%d bytes, starting at 0x%x):\n%.*Rhxd\n",
971	i, pSG->aSegs[i].cb, offset, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
972	offset += pSG->aSegs[i].cb;
973	}
974	}
975	#endif
976
977	#ifdef PADD_RUNT_FRAMES_FROM_HOST
978	/*
979	* Add a trailer if the frame is too small.
980	*
981	* Since we're getting to the packet before it is framed, it has not
982	* yet been padded. The current solution is to add a segment pointing
983	* to a buffer containing all zeros and pray that works for all frames...
984	*/
985	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
986	{
987	Assert(pBuf->data_len == 0); /* Packets with fragments are never small! */
988	static uint8_t const s_abZero[128] = {0};
989
990	AssertReturnVoid(iSeg < cSegs);
991
992	pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
993	pSG->aSegs[iSeg].pv = (void *)&s_abZero[0];
994	pSG->aSegs[iSeg++].cb = 60 - pSG->cbTotal;
995	pSG->cbTotal = 60;
996	pSG->cSegsUsed++;
997	Assert(iSeg <= pSG->cSegsAlloc)
998	}
999	#endif
1000
1001	Log6(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
1002	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
1003	for (i = 0; i < pSG->cSegsUsed; i++)
1004	Log6(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
1005	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1006	}
1007
1008	/**
1009	* Packet handler; not really documented - figure it out yourself.
1010	*
1011	* @returns 0 or EJUSTRETURN - this is probably copy & pastry and thus wrong.
1012	*/
1013	#if RTLNX_VER_MIN(2,6,14)
1014	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1015	struct net_device *pSkbDev,
1016	struct packet_type *pPacketType,
1017	struct net_device *pOrigDev)
1018	#else
1019	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1020	struct net_device *pSkbDev,
1021	struct packet_type *pPacketType)
1022	#endif
1023	{
1024	PVBOXNETFLTINS pThis;
1025	struct net_device *pDev;
1026	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1027	pBuf, pSkbDev, pPacketType));
1028	#if RTLNX_VER_MIN(2,6,18)
1029	Log3(("vboxNetFltLinuxPacketHandler: skb len=%u data_len=%u truesize=%u next=%p nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x\n",
1030	pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next, skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->gso_size, skb_shinfo(pBuf)->gso_segs, skb_shinfo(pBuf)->gso_type, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type));
1031	# if RTLNX_VER_MIN(2,6,22)
1032	Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1033	# endif
1034	#else
1035	Log3(("vboxNetFltLinuxPacketHandler: skb len=%u data_len=%u truesize=%u next=%p nr_frags=%u tso_size=%u tso_seqs=%u frag_list=%p pkt_type=%x\n",
1036	pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next, skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->tso_size, skb_shinfo(pBuf)->tso_segs, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type));
1037	#endif
1038	/*
1039	* Drop it immediately?
1040	*/
1041	if (!pBuf)
1042	return 0;
1043
1044	if (pBuf->pkt_type == PACKET_LOOPBACK)
1045	{
1046	/*
1047	* We are not interested in loopbacked packets as they will always have
1048	* another copy going to the wire.
1049	*/
1050	Log2(("vboxNetFltLinuxPacketHandler: dropped loopback packet (cb=%u)\n", pBuf->len));
1051	dev_kfree_skb(pBuf); /* We must 'consume' all packets we get (@bugref{6539})! */
1052	return 0;
1053	}
1054
1055	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1056	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1057	if (pDev != pSkbDev)
1058	{
1059	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1060	kfree_skb(pBuf); /* This is a failure, so we use kfree_skb instead of dev_kfree_skb. */
1061	return 0;
1062	}
1063
1064	Log6(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1065	if (vboxNetFltLinuxSkBufIsOur(pBuf))
1066	{
1067	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1068	dev_kfree_skb(pBuf);
1069	return 0;
1070	}
1071
1072	#ifndef VBOXNETFLT_SG_SUPPORT
1073	{
1074	/*
1075	* Get rid of fragmented packets, they cause too much trouble.
1076	*/
1077	unsigned int uMacLen = pBuf->mac_len;
1078	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1079	dev_kfree_skb(pBuf);
1080	if (!pCopy)
1081	{
1082	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1083	return 0;
1084	}
1085	pBuf = pCopy;
1086	/* Somehow skb_copy ignores mac_len */
1087	pBuf->mac_len = uMacLen;
1088	# if RTLNX_VER_MIN(2,6,27)
1089	/* Restore VLAN tag stripped by host hardware */
1090	if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
1091	{
1092	uint8_t pMac = (uint8_t)skb_mac_header(pBuf);
1093	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )(pMac - VLAN_HLEN);
1094	memmove(pVHdr, pMac, ETH_ALEN * 2);
1095	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
1096	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
1097	pBuf->mac_header -= VLAN_HLEN;
1098	pBuf->mac_len += VLAN_HLEN;
1099	}
1100	# endif /* RTLNX_VER_MIN(2,6,27) */
1101
1102	# if RTLNX_VER_MIN(2,6,18)
1103	Log3(("vboxNetFltLinuxPacketHandler: skb copy len=%u data_len=%u truesize=%u next=%p nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x\n",
1104	pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next, skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->gso_size, skb_shinfo(pBuf)->gso_segs, skb_shinfo(pBuf)->gso_type, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type));
1105	# if RTLNX_VER_MIN(2,6,22)
1106	Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1107	# endif /* RTLNX_VER_MIN(2,6,22) */
1108	# else /* RTLNX_VER_MAX(2,6,18) */
1109	Log3(("vboxNetFltLinuxPacketHandler: skb copy len=%u data_len=%u truesize=%u next=%p nr_frags=%u tso_size=%u tso_seqs=%u frag_list=%p pkt_type=%x\n",
1110	pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next, skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->tso_size, skb_shinfo(pBuf)->tso_segs, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type));
1111	# endif /* RTLNX_VER_MAX(2,6,18) */
1112	}
1113	#endif /* !VBOXNETFLT_SG_SUPPORT */
1114
1115	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1116	/* Forward it to the internal network. */
1117	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1118	#else /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1119	/* Add the packet to transmit queue and schedule the bottom half. */
1120	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1121	schedule_work(&pThis->u.s.XmitTask);
1122	Log6(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1123	&pThis->u.s.XmitTask, pBuf));
1124	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1125
1126	/* It does not really matter what we return, it is ignored by the kernel. */
1127	return 0;
1128	}
1129
1130	/**
1131	* Calculate the number of INTNETSEG segments the socket buffer will need.
1132	*
1133	* @returns Segment count.
1134	* @param pBuf The socket buffer.
1135	* @param pcbTemp Where to store the number of bytes of the part
1136	* of the socket buffer that will be copied to
1137	* a temporary storage.
1138	*/
1139	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff pBuf, unsigned pcbTemp)
1140	{
1141	*pcbTemp = 0;
1142	#ifdef VBOXNETFLT_SG_SUPPORT
1143	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1144	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1145	{
1146	*pcbTemp = vboxNetFltLinuxGetChecksumStartOffset(pBuf) + VBOX_SKB_CSUM_OFFSET(pBuf) + sizeof(uint16_t);
1147	}
1148	# if RTLNX_VER_MIN(2,6,27)
1149	if (vlan_tx_tag_present(pBuf))
1150	{
1151	if (*pcbTemp)
1152	*pcbTemp += VLAN_HLEN;
1153	else
1154	*pcbTemp = VLAN_ETH_HLEN;
1155	}
1156	# endif /* RTLNX_VER_MIN(2,6,27) */
1157	if (*pcbTemp)
1158	++cSegs;
1159	struct sk_buff *pFrag;
1160	for (pFrag = skb_shinfo(pBuf)->frag_list; pFrag; pFrag = pFrag->next)
1161	{
1162	Log6(("vboxNetFltLinuxCalcSGSegments: frag=%p len=%d data_len=%d frags=%d frag_list=%p next=%p\n",
1163	pFrag, pFrag->len, pFrag->data_len, skb_shinfo(pFrag)->nr_frags, skb_shinfo(pFrag)->frag_list, pFrag->next));
1164	cSegs += 1 + skb_shinfo(pFrag)->nr_frags;
1165	}
1166	#else
1167	unsigned cSegs = 1;
1168	#endif
1169	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1170	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1171	if (pBuf->len < 60)
1172	cSegs++;
1173	#endif
1174	return cSegs;
1175	}
1176
1177
1178	/**
1179	* Destroy the intnet scatter / gather buffer created by
1180	* vboxNetFltLinuxSkBufToSG.
1181	*
1182	* @param pSG The (scatter/)gather list.
1183	* @param pBuf The original socket buffer that was used to create
1184	* the scatter/gather list.
1185	*/
1186	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG, struct sk_buff *pBuf)
1187	{
1188	#ifdef VBOXNETFLT_SG_SUPPORT
1189	int i, iSeg = 1; /* Skip non-paged part of SKB */
1190	/* Check if the extra buffer behind SG structure was used for modified packet header */
1191	if (pBuf->data != pSG->aSegs[0].pv)
1192	++iSeg; /* Skip it as well */
1193	# ifdef LOG_ENABLED
1194	if (pBuf->data_len)
1195	Log6(("kunmap_atomic:"));
1196	# endif /* LOG_ENABLED */
1197	/* iSeg now points to the first mapped fragment if there are any */
1198	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1199	{
1200	Log6((" %p", pSG->aSegs[iSeg].pv));
1201	VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1202	}
1203	struct sk_buff *pFragBuf;
1204	for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
1205	{
1206	++iSeg; /* Non-fragment (unmapped) portion of chained SKB */
1207	for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
1208	{
1209	Log6((" %p", pSG->aSegs[iSeg].pv));
1210	VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1211	}
1212	}
1213	# ifdef LOG_ENABLED
1214	if (pBuf->data_len)
1215	Log6(("\n"));
1216	# endif /* LOG_ENABLED */
1217	#endif
1218	NOREF(pSG);
1219	}
1220
1221	#ifdef LOG_ENABLED
1222	/**
1223	* Logging helper.
1224	*/
1225	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1226	{
1227	int i, offSeg;
1228	uint8_t pInt, pExt;
1229	static int iPacketNo = 1;
1230	iPacketNo += iIncrement;
1231	if (fEgress)
1232	{
1233	pExt = pSG->aSegs[0].pv;
1234	pInt = pExt + 6;
1235	}
1236	else
1237	{
1238	pInt = pSG->aSegs[0].pv;
1239	pExt = pInt + 6;
1240	}
1241	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1242	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1243	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1244	fEgress ? "-->" : "<--", pszWhere,
1245	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1246	pSG->cbTotal, iPacketNo));
1247	if (pSG->cSegsUsed == 1)
1248	{
1249	Log4(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1250	}
1251	else
1252	{
1253	for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
1254	{
1255	Log4(("-- segment %d at 0x%x (%d bytes)\n --\n%.*Rhxd\n",
1256	i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1257	offSeg += pSG->aSegs[i].cb;
1258	}
1259	}
1260	}
1261	#else
1262	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1263	#endif
1264
1265	#ifdef VBOXNETFLT_WITH_GSO_RECV
1266
1267	/**
1268	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1269	* GSO socket buffer without having to segment it.
1270	*
1271	* @returns true on success, false if needs segmenting.
1272	* @param pThis The net filter instance.
1273	* @param pSkb The GSO socket buffer.
1274	* @param fSrc The source.
1275	* @param pGsoCtx Where to return the GSO context on success.
1276	*/
1277	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1278	PPDMNETWORKGSO pGsoCtx)
1279	{
1280	PDMNETWORKGSOTYPE enmGsoType;
1281	uint16_t uEtherType;
1282	unsigned int cbTransport;
1283	unsigned int offTransport;
1284	unsigned int cbTransportHdr;
1285	unsigned uProtocol;
1286	union
1287	{
1288	RTNETIPV4 IPv4;
1289	RTNETIPV6 IPv6;
1290	RTNETTCP Tcp;
1291	uint8_t ab[40];
1292	uint16_t au16[40/2];
1293	uint32_t au32[40/4];
1294	} Buf;
1295
1296	/*
1297	* Check the GSO properties of the socket buffer and make sure it fits.
1298	*/
1299	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1300	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
1301	{
1302	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1303	return false;
1304	}
1305	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1306	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
1307	{
1308	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1309	return false;
1310	}
1311
1312	/*
1313	* Switch on the ethertype.
1314	*/
1315	uEtherType = pSkb->protocol;
1316	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1317	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1318	{
1319	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1320	if (puEtherType)
1321	uEtherType = *puEtherType;
1322	}
1323	switch (uEtherType)
1324	{
1325	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1326	{
1327	unsigned int cbHdr;
1328	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1329	if (RT_UNLIKELY(!pIPv4))
1330	{
1331	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1332	return false;
1333	}
1334
1335	cbHdr = pIPv4->ip_hl * 4;
1336	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1337	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1338	\|\| cbHdr > cbTransport ))
1339	{
1340	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1341	return false;
1342	}
1343	cbTransport -= cbHdr;
1344	offTransport = pSkb->mac_len + cbHdr;
1345	uProtocol = pIPv4->ip_p;
1346	if (uProtocol == RTNETIPV4_PROT_TCP)
1347	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1348	else if (uProtocol == RTNETIPV4_PROT_UDP)
1349	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1350	else /** @todo IPv6: 4to6 tunneling */
1351	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1352	break;
1353	}
1354
1355	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1356	{
1357	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1358	if (RT_UNLIKELY(!pIPv6))
1359	{
1360	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1361	return false;
1362	}
1363
1364	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1365	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1366	uProtocol = pIPv6->ip6_nxt;
1367	/** @todo IPv6: Dig our way out of the other headers. */
1368	if (uProtocol == RTNETIPV4_PROT_TCP)
1369	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1370	else if (uProtocol == RTNETIPV4_PROT_UDP)
1371	enmGsoType = PDMNETWORKGSOTYPE_IPV6_UDP;
1372	else
1373	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1374	break;
1375	}
1376
1377	default:
1378	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1379	return false;
1380	}
1381
1382	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1383	{
1384	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1385	return false;
1386	}
1387
1388	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1389	\|\| offTransport + cbTransport > pSkb->len
1390	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1391	{
1392	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1393	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1394	return false;
1395	}
1396
1397	/*
1398	* Check the TCP/UDP bits.
1399	*/
1400	if (uProtocol == RTNETIPV4_PROT_TCP)
1401	{
1402	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1403	if (RT_UNLIKELY(!pTcp))
1404	{
1405	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1406	return false;
1407	}
1408
1409	cbTransportHdr = pTcp->th_off * 4;
1410	pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1411	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1412	\|\| cbTransportHdr > cbTransport
1413	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1414	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1415	{
1416	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1417	return false;
1418	}
1419
1420	}
1421	else
1422	{
1423	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1424	cbTransportHdr = sizeof(RTNETUDP);
1425	pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1426	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1427	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1428	{
1429	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1430	return false;
1431	}
1432	}
1433
1434	/*
1435	* We're good, init the GSO context.
1436	*/
1437	pGsoCtx->u8Type = enmGsoType;
1438	pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1439	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1440	pGsoCtx->offHdr1 = pSkb->mac_len;
1441	pGsoCtx->offHdr2 = offTransport;
1442	pGsoCtx->u8Unused = 0;
1443
1444	return true;
1445	}
1446
1447	/**
1448	* Forward the socket buffer as a GSO internal network frame.
1449	*
1450	* @returns IPRT status code.
1451	* @param pThis The net filter instance.
1452	* @param pSkb The GSO socket buffer.
1453	* @param fSrc The source.
1454	* @param pGsoCtx Where to return the GSO context on success.
1455	*/
1456	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1457	{
1458	int rc;
1459	unsigned cbExtra;
1460	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb, &cbExtra);
1461	PINTNETSG pSG = (PINTNETSG)alloca(RT_UOFFSETOF_DYN(INTNETSG, aSegs[cSegs]) + cbExtra);
1462	if (RT_LIKELY(pSG))
1463	{
1464	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cbExtra, cSegs, fSrc, pGsoCtx);
1465
1466	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1467	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1468
1469	vboxNetFltLinuxDestroySG(pSG, pSkb);
1470	rc = VINF_SUCCESS;
1471	}
1472	else
1473	{
1474	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1475	rc = VERR_NO_MEMORY;
1476	}
1477	return rc;
1478	}
1479
1480	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1481
1482	/**
1483	* Worker for vboxNetFltLinuxForwardToIntNet.
1484	*
1485	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1486	* @param pThis The net filter instance.
1487	* @param pBuf The socket buffer.
1488	* @param fSrc The source.
1489	*/
1490	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1491	{
1492	int rc;
1493	unsigned cbExtra;
1494	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf, &cbExtra);
1495	PINTNETSG pSG = (PINTNETSG)alloca(RT_UOFFSETOF_DYN(INTNETSG, aSegs[cSegs]) + cbExtra);
1496	if (RT_LIKELY(pSG))
1497	{
1498	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cbExtra, cSegs, fSrc, NULL /pGsoCtx/);
1499
1500	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1501	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1502
1503	vboxNetFltLinuxDestroySG(pSG, pBuf);
1504	rc = VINF_SUCCESS;
1505	}
1506	else
1507	{
1508	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1509	rc = VERR_NO_MEMORY;
1510	}
1511	return rc;
1512	}
1513
1514
1515	/**
1516	* I won't disclose what I do, figure it out yourself, including pThis referencing.
1517	*
1518	* @param pThis The net filter instance.
1519	* @param pBuf The socket buffer.
1520	* @param fSrc Where the packet comes from.
1521	*/
1522	static void vboxNetFltLinuxForwardToIntNetInner(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1523	{
1524	#ifdef VBOXNETFLT_WITH_GSO
1525	if (skb_is_gso(pBuf))
1526	{
1527	PDMNETWORKGSO GsoCtx;
1528	Log6(("vboxNetFltLinuxForwardToIntNetInner: skb len=%u data_len=%u truesize=%u next=%p"
1529	" nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x ip_summed=%d\n",
1530	pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next,
1531	skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->gso_size,
1532	skb_shinfo(pBuf)->gso_segs, skb_shinfo(pBuf)->gso_type,
1533	skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, pBuf->ip_summed));
1534
1535	if (RT_LIKELY(fSrc & INTNETTRUNKDIR_HOST))
1536	{
1537	/*
1538	* skb_gso_segment does the following. Do we need to do it as well?
1539	*/
1540	# if RTLNX_VER_MIN(2,6,22)
1541	skb_reset_mac_header(pBuf);
1542	pBuf->mac_len = pBuf->network_header - pBuf->mac_header;
1543	# else
1544	pBuf->mac.raw = pBuf->data;
1545	pBuf->mac_len = pBuf->nh.raw - pBuf->data;
1546	# endif
1547	}
1548
1549	# ifdef VBOXNETFLT_WITH_GSO_RECV
1550	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1551	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1552	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1553	else
1554	# endif /* VBOXNETFLT_WITH_GSO_RECV */
1555	{
1556	/* Need to segment the packet */
1557	struct sk_buff *pNext;
1558	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1559	if (IS_ERR(pSegment))
1560	{
1561	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1562	return;
1563	}
1564
1565	for (; pSegment; pSegment = pNext)
1566	{
1567	Log6(("vboxNetFltLinuxForwardToIntNetInner: segment len=%u data_len=%u truesize=%u next=%p"
1568	" nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x\n",
1569	pSegment->len, pSegment->data_len, pSegment->truesize, pSegment->next,
1570	skb_shinfo(pSegment)->nr_frags, skb_shinfo(pSegment)->gso_size,
1571	skb_shinfo(pSegment)->gso_segs, skb_shinfo(pSegment)->gso_type,
1572	skb_shinfo(pSegment)->frag_list, pSegment->pkt_type));
1573	pNext = pSegment->next;
1574	pSegment->next = 0;
1575	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1576	dev_kfree_skb(pSegment);
1577	}
1578	}
1579	}
1580	else
1581	#endif /* VBOXNETFLT_WITH_GSO */
1582	{
1583	Log6(("vboxNetFltLinuxForwardToIntNetInner: ptk_type=%d ip_summed=%d len=%d"
1584	" data_len=%d headroom=%d hdr_len=%d csum_offset=%d\n",
1585	pBuf->pkt_type, pBuf->ip_summed, pBuf->len, pBuf->data_len, skb_headroom(pBuf),
1586	skb_headlen(pBuf), vboxNetFltLinuxGetChecksumStartOffset(pBuf)));
1587	#ifndef VBOXNETFLT_SG_SUPPORT
1588	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1589	{
1590	# if RTLNX_VER_MIN(2,6,19)
1591	int rc = VBOX_SKB_CHECKSUM_HELP(pBuf);
1592	# else
1593	/*
1594	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1595	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1596	* header length from the header itself and reconstruct 'h' pointer
1597	* to TCP (or whatever) header.
1598	*/
1599	unsigned char *tmp = pBuf->h.raw;
1600	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1601	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1602	int rc = VBOX_SKB_CHECKSUM_HELP(pBuf);
1603	/* Restore the original (wrong) pointer. */
1604	pBuf->h.raw = tmp;
1605	# endif
1606	if (rc)
1607	{
1608	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1609	return;
1610	}
1611	}
1612	#endif /* !VBOXNETFLT_SG_SUPPORT */
1613	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1614	}
1615	}
1616
1617
1618	/**
1619	* Temporarily adjust pBuf->data so it always points to the Ethernet header,
1620	* then forward it to the internal network.
1621	*
1622	* @param pThis The net filter instance.
1623	* @param pBuf The socket buffer. This is consumed by this function.
1624	*/
1625	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1626	{
1627	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1628
1629	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1630	{
1631	/*
1632	* The packet came from the wire and the driver has already consumed
1633	* mac header. We need to restore it back. Moreover, after we are
1634	* through with this skb we need to restore its original state!
1635	*/
1636	skb_push(pBuf, pBuf->mac_len);
1637	Log5(("vboxNetFltLinuxForwardToIntNet: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1638	pBuf->mac_len, pBuf->data, skb_mac_header(pBuf), skb_network_header(pBuf)));
1639	}
1640
1641	vboxNetFltLinuxForwardToIntNetInner(pThis, pBuf, fSrc);
1642
1643	/*
1644	* Restore the original state of skb as there are other handlers this skb
1645	* will be provided to.
1646	*/
1647	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1648	skb_pull(pBuf, pBuf->mac_len);
1649
1650	dev_kfree_skb(pBuf);
1651	}
1652
1653
1654	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1655	/**
1656	* Work queue handler that forwards the socket buffers queued by
1657	* vboxNetFltLinuxPacketHandler to the internal network.
1658	*
1659	* @param pWork The work queue.
1660	*/
1661	# if RTLNX_VER_MIN(2,6,20)
1662	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1663	# else
1664	static void vboxNetFltLinuxXmitTask(void *pWork)
1665	# endif
1666	{
1667	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1668	struct sk_buff *pBuf;
1669
1670	Log6(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1671
1672	/*
1673	* Active? Retain the instance and increment the busy counter.
1674	*/
1675	if (vboxNetFltTryRetainBusyActive(pThis))
1676	{
1677	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1678	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1679
1680	vboxNetFltRelease(pThis, true /* fBusy */);
1681	}
1682	else
1683	{
1684	/** @todo Shouldn't we just drop the packets here? There is little point in
1685	* making them accumulate when the VM is paused and it'll only waste
1686	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1687	* before start draining the packets (goes for the intnet ring buf
1688	* too)? */
1689	}
1690	}
1691	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1692
1693	/**
1694	* Reports the GSO capabilities of the hardware NIC.
1695	*
1696	* @param pThis The net filter instance. The caller hold a
1697	* reference to this.
1698	*/
1699	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1700	{
1701	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1702	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1703	{
1704	struct net_device *pDev;
1705	PINTNETTRUNKSWPORT pSwitchPort;
1706	unsigned int fFeatures;
1707
1708	RTSpinlockAcquire(pThis->hSpinlock);
1709
1710	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1711	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1712	if (pDev)
1713	fFeatures = pDev->features;
1714	else
1715	fFeatures = 0;
1716
1717	RTSpinlockRelease(pThis->hSpinlock);
1718
1719	if (pThis->pSwitchPort)
1720	{
1721	/* Set/update the GSO capabilities of the NIC. */
1722	uint32_t fGsoCapabilites = 0;
1723	if (fFeatures & NETIF_F_TSO)
1724	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1725	if (fFeatures & NETIF_F_TSO6)
1726	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1727	Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s\n",
1728	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1729	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : ""));
1730	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1731	}
1732
1733	vboxNetFltRelease(pThis, true /fBusy/);
1734	}
1735	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
1736	}
1737
1738	/**
1739	* Helper that determines whether the host (ignoreing us) is operating the
1740	* interface in promiscuous mode or not.
1741	*/
1742	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1743	{
1744	bool fRc = false;
1745	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1746	if (pDev)
1747	{
1748	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1749	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1750	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1751	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1752	}
1753	return fRc;
1754	}
1755
1756	/**
1757	* Does this device needs link state change signaled?
1758	* Currently we need it for our own VBoxNetAdp and TAP.
1759	*/
1760	static bool vboxNetFltNeedsLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev)
1761	{
1762	if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1763	{
1764	struct ethtool_drvinfo Info;
1765
1766	memset(&Info, 0, sizeof(Info));
1767	Info.cmd = ETHTOOL_GDRVINFO;
1768	pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1769	Log3(("%s: driver=%.s version=%.s bus_info=%.*s\n",
1770	__FUNCTION__,
1771	sizeof(Info.driver), Info.driver,
1772	sizeof(Info.version), Info.version,
1773	sizeof(Info.bus_info), Info.bus_info));
1774
1775	if (!strncmp(Info.driver, "vboxnet", sizeof(Info.driver)))
1776	return true;
1777
1778	#if RTLNX_VER_MIN(2,6,36) /* TAP started doing carrier */
1779	return !strncmp(Info.driver, "tun", 4)
1780	&& !strncmp(Info.bus_info, "tap", 4);
1781	#endif
1782	}
1783
1784	return false;
1785	}
1786
1787	#if RTLNX_VER_MAX(2,6,18)
1788	DECLINLINE(void) netif_tx_lock_bh(struct net_device *pDev)
1789	{
1790	spin_lock_bh(&pDev->xmit_lock);
1791	}
1792
1793	DECLINLINE(void) netif_tx_unlock_bh(struct net_device *pDev)
1794	{
1795	spin_unlock_bh(&pDev->xmit_lock);
1796	}
1797	#endif
1798
1799	/**
1800	* Some devices need link state change when filter attaches/detaches
1801	* since the filter is their link in a sense.
1802	*/
1803	static void vboxNetFltSetLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1804	{
1805	if (vboxNetFltNeedsLinkState(pThis, pDev))
1806	{
1807	Log3(("%s: bringing device link %s\n",
1808	__FUNCTION__, fLinkUp ? "up" : "down"));
1809	netif_tx_lock_bh(pDev);
1810	if (fLinkUp)
1811	netif_carrier_on(pDev);
1812	else
1813	netif_carrier_off(pDev);
1814	netif_tx_unlock_bh(pDev);
1815	}
1816	}
1817
1818	/**
1819	* Internal worker for vboxNetFltLinuxNotifierCallback.
1820	*
1821	* @returns VBox status code.
1822	* @param pThis The instance.
1823	* @param pDev The device to attach to.
1824	*/
1825	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1826	{
1827	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1828
1829	/*
1830	* Retain and store the device.
1831	*/
1832	dev_hold(pDev);
1833
1834	RTSpinlockAcquire(pThis->hSpinlock);
1835	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1836	RTSpinlockRelease(pThis->hSpinlock);
1837
1838	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1839	pDev, pDev->name,
1840	#if RTLNX_VER_MIN(2,6,37)
1841	netdev_refcnt_read(pDev)
1842	#else
1843	atomic_read(&pDev->refcnt)
1844	#endif
1845	));
1846	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1847	pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1848
1849	/* Get the mac address while we still have a valid net_device reference. */
1850	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1851	/* Initialize MTU */
1852	pThis->u.s.cbMtu = pDev->mtu;
1853
1854	/*
1855	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1856	*/
1857	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1858	pThis->u.s.PacketType.dev = pDev;
1859	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1860	dev_add_pack(&pThis->u.s.PacketType);
1861	ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1862	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1863
1864	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1865	vboxNetFltLinuxHookDev(pThis, pDev);
1866	#endif
1867
1868	/*
1869	* Are we the "carrier" for this device (e.g. vboxnet or tap)?
1870	*/
1871	vboxNetFltSetLinkState(pThis, pDev, true);
1872
1873	/*
1874	* Set indicators that require the spinlock. Be abit paranoid about racing
1875	* the device notification handle.
1876	*/
1877	RTSpinlockAcquire(pThis->hSpinlock);
1878	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1879	if (pDev)
1880	{
1881	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1882	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1883	pDev = NULL; /* don't dereference it */
1884	}
1885	RTSpinlockRelease(pThis->hSpinlock);
1886
1887	/*
1888	* Report GSO capabilities
1889	*/
1890	Assert(pThis->pSwitchPort);
1891	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1892	{
1893	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1894	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1895	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1896	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1897	vboxNetFltRelease(pThis, true /fBusy/);
1898	}
1899
1900	LogRel(("VBoxNetFlt: attached to '%s' / %RTmac\n", pThis->szName, &pThis->u.s.MacAddr));
1901	return VINF_SUCCESS;
1902	}
1903
1904
1905	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1906	{
1907	bool fRegistered;
1908	Assert(!pThis->fDisconnectedFromHost);
1909
1910	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1911	vboxNetFltLinuxUnhookDev(pThis, pDev);
1912	#endif
1913
1914	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1915	{
1916	dev_remove_pack(&pThis->u.s.PacketType);
1917	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1918	}
1919
1920	RTSpinlockAcquire(pThis->hSpinlock);
1921	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1922	if (fRegistered)
1923	{
1924	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1925	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1926	}
1927	RTSpinlockRelease(pThis->hSpinlock);
1928
1929	if (fRegistered)
1930	{
1931	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1932	skb_queue_purge(&pThis->u.s.XmitQueue);
1933	#endif
1934	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1935	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1936	pDev, pDev->name,
1937	#if RTLNX_VER_MIN(2,6,37)
1938	netdev_refcnt_read(pDev)
1939	#else
1940	atomic_read(&pDev->refcnt)
1941	#endif
1942	));
1943	dev_put(pDev);
1944	}
1945
1946	return NOTIFY_OK;
1947	}
1948
1949	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1950	{
1951	/* Check if we are not suspended and promiscuous mode has not been set. */
1952	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1953	&& !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1954	{
1955	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1956	dev_set_promiscuity(pDev, 1);
1957	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1958	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1959	}
1960	else
1961	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1962	return NOTIFY_OK;
1963	}
1964
1965	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1966	{
1967	/* Undo promiscuous mode if we has set it. */
1968	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1969	{
1970	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1971	dev_set_promiscuity(pDev, -1);
1972	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1973	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1974	}
1975	else
1976	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1977	return NOTIFY_OK;
1978	}
1979
1980	/**
1981	* Callback for listening to MTU change event.
1982	*
1983	* We need to track changes of host's inteface MTU to discard over-sized frames
1984	* coming from the internal network as they may hang the TX queue of host's
1985	* adapter.
1986	*
1987	* @returns NOTIFY_OK
1988	* @param pThis The netfilter instance.
1989	* @param pDev Pointer to device structure of host's interface.
1990	*/
1991	static int vboxNetFltLinuxDeviceMtuChange(PVBOXNETFLTINS pThis, struct net_device *pDev)
1992	{
1993	ASMAtomicWriteU32(&pThis->u.s.cbMtu, pDev->mtu);
1994	Log(("vboxNetFltLinuxDeviceMtuChange: set MTU for %s to %d\n", pThis->szName, pDev->mtu));
1995	return NOTIFY_OK;
1996	}
1997
1998	#ifdef LOG_ENABLED
1999	/** Stringify the NETDEV_XXX constants. */
2000	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2001	{
2002	const char *pszEvent = "NETDEV_<unknown>";
2003	switch (ulEventType)
2004	{
2005	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2006	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2007	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2008	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2009	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2010	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2011	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2012	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2013	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2014	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2015	# ifdef NETDEV_FEAT_CHANGE
2016	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2017	# endif
2018	}
2019	return pszEvent;
2020	}
2021	#endif /* LOG_ENABLED */
2022
2023	/**
2024	* Callback for listening to netdevice events.
2025	*
2026	* This works the rediscovery, clean up on unregistration, promiscuity on
2027	* up/down, and GSO feature changes from ethtool.
2028	*
2029	* @returns NOTIFY_OK
2030	* @param self Pointer to our notifier registration block.
2031	* @param ulEventType The event.
2032	* @param ptr Event specific, but it is usually the device it
2033	* relates to.
2034	*/
2035	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2036
2037	{
2038	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2039	struct net_device pMyDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device );
2040	struct net_device *pDev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2041	int rc = NOTIFY_OK;
2042
2043	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2044	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, pMyDev));
2045
2046	if (ulEventType == NETDEV_REGISTER)
2047	{
2048	#if RTLNX_VER_MIN(2,6,24) /* cgroups/namespaces introduced */
2049	# if RTLNX_VER_MIN(2,6,26)
2050	# define VBOX_DEV_NET(dev) dev_net(dev)
2051	# define VBOX_NET_EQ(n1, n2) net_eq((n1), (n2))
2052	# else
2053	# define VBOX_DEV_NET(dev) ((dev)->nd_net)
2054	# define VBOX_NET_EQ(n1, n2) ((n1) == (n2))
2055	# endif
2056	struct net *pMyNet = current->nsproxy->net_ns;
2057	struct net *pDevNet = VBOX_DEV_NET(pDev);
2058
2059	if (VBOX_NET_EQ(pDevNet, pMyNet))
2060	#endif /* namespaces */
2061	{
2062	if (strcmp(pDev->name, pThis->szName) == 0)
2063	{
2064	vboxNetFltLinuxAttachToInterface(pThis, pDev);
2065	}
2066	}
2067	}
2068	else
2069	{
2070	if (pDev == pMyDev)
2071	{
2072	switch (ulEventType)
2073	{
2074	case NETDEV_UNREGISTER:
2075	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2076	break;
2077	case NETDEV_UP:
2078	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2079	break;
2080	case NETDEV_GOING_DOWN:
2081	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2082	break;
2083	case NETDEV_CHANGEMTU:
2084	rc = vboxNetFltLinuxDeviceMtuChange(pThis, pDev);
2085	break;
2086	case NETDEV_CHANGENAME:
2087	break;
2088	#ifdef NETDEV_FEAT_CHANGE
2089	case NETDEV_FEAT_CHANGE:
2090	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2091	break;
2092	#endif
2093	}
2094	}
2095	}
2096
2097	return rc;
2098	}
2099
2100	/*
2101	* Initial enumeration of netdevs. Called with NETDEV_REGISTER by
2102	* register_netdevice_notifier() under rtnl lock.
2103	*/
2104	static int vboxNetFltLinuxEnumeratorCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2105	{
2106	PVBOXNETFLTINS pThis = ((PVBOXNETFLTNOTIFIER)self)->pThis;
2107	struct net_device *dev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2108	struct in_device *in_dev;
2109	struct inet6_dev *in6_dev;
2110
2111	if (ulEventType != NETDEV_REGISTER)
2112	return NOTIFY_OK;
2113
2114	if (RT_UNLIKELY(pThis->pSwitchPort->pfnNotifyHostAddress == NULL))
2115	return NOTIFY_OK;
2116
2117	/*
2118	* IPv4
2119	*/
2120	#if RTLNX_VER_MIN(2,6,14)
2121	in_dev = __in_dev_get_rtnl(dev);
2122	#else
2123	in_dev = __in_dev_get(dev);
2124	#endif
2125	if (in_dev != NULL)
2126	{
2127	struct in_ifaddr *ifa;
2128
2129	for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
2130	if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2131	return NOTIFY_OK;
2132
2133	if ( dev != pThis->u.s.pDev
2134	&& VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2135	continue;
2136
2137	Log(("%s: %s: IPv4 addr %RTnaipv4 mask %RTnaipv4\n",
2138	__FUNCTION__, VBOX_NETDEV_NAME(dev),
2139	ifa->ifa_address, ifa->ifa_mask));
2140
2141	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2142	/* :fAdded */ true, kIntNetAddrType_IPv4, &ifa->ifa_address);
2143	}
2144	}
2145
2146	/*
2147	* IPv6
2148	*/
2149	in6_dev = __in6_dev_get(dev);
2150	if (in6_dev != NULL)
2151	{
2152	struct inet6_ifaddr *ifa;
2153
2154	read_lock_bh(&in6_dev->lock);
2155	#if RTLNX_VER_MIN(2,6,35)
2156	list_for_each_entry(ifa, &in6_dev->addr_list, if_list)
2157	#else
2158	for (ifa = in6_dev->addr_list; ifa != NULL; ifa = ifa->if_next)
2159	#endif
2160	{
2161	if ( dev != pThis->u.s.pDev
2162	&& ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL \| IPV6_ADDR_LOOPBACK))
2163	continue;
2164
2165	Log(("%s: %s: IPv6 addr %RTnaipv6/%u\n",
2166	__FUNCTION__, VBOX_NETDEV_NAME(dev),
2167	&ifa->addr, (unsigned)ifa->prefix_len));
2168
2169	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2170	/* :fAdded */ true, kIntNetAddrType_IPv6, &ifa->addr);
2171	}
2172	read_unlock_bh(&in6_dev->lock);
2173	}
2174
2175	return NOTIFY_OK;
2176	}
2177
2178
2179	static int vboxNetFltLinuxNotifierIPv4Callback(struct notifier_block self, unsigned long ulEventType, void ptr)
2180	{
2181	PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv4);
2182	struct net_device pDev, pEventDev;
2183	struct in_ifaddr ifa = (struct in_ifaddr )ptr;
2184	bool fMyDev;
2185	int rc = NOTIFY_OK;
2186
2187	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2188	pEventDev = ifa->ifa_dev->dev;
2189	fMyDev = (pDev == pEventDev);
2190	Log(("VBoxNetFlt: %s: IPv4 event %s(0x%lx) %s: addr %RTnaipv4 mask %RTnaipv4\n",
2191	pDev ? VBOX_NETDEV_NAME(pDev) : "<unknown>",
2192	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2193	pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<unknown>",
2194	ifa->ifa_address, ifa->ifa_mask));
2195
2196	if (pDev != NULL)
2197	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2198
2199	if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2200	return NOTIFY_OK;
2201
2202	if ( !fMyDev
2203	&& VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2204	return NOTIFY_OK;
2205
2206	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2207	{
2208	bool fAdded;
2209	if (ulEventType == NETDEV_UP)
2210	fAdded = true;
2211	else if (ulEventType == NETDEV_DOWN)
2212	fAdded = false;
2213	else
2214	return NOTIFY_OK;
2215
2216	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2217	kIntNetAddrType_IPv4, &ifa->ifa_local);
2218	}
2219
2220	return rc;
2221	}
2222
2223
2224	static int vboxNetFltLinuxNotifierIPv6Callback(struct notifier_block self, unsigned long ulEventType, void ptr)
2225	{
2226	PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv6);
2227	struct net_device pDev, pEventDev;
2228	struct inet6_ifaddr ifa = (struct inet6_ifaddr )ptr;
2229	bool fMyDev;
2230	int rc = NOTIFY_OK;
2231
2232	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2233	pEventDev = ifa->idev->dev;
2234	fMyDev = (pDev == pEventDev);
2235	Log(("VBoxNetFlt: %s: IPv6 event %s(0x%lx) %s: %RTnaipv6\n",
2236	pDev ? VBOX_NETDEV_NAME(pDev) : "<unknown>",
2237	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2238	pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<unknown>",
2239	&ifa->addr));
2240
2241	if (pDev != NULL)
2242	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2243
2244	if ( !fMyDev
2245	&& ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL \| IPV6_ADDR_LOOPBACK))
2246	return NOTIFY_OK;
2247
2248	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2249	{
2250	bool fAdded;
2251	if (ulEventType == NETDEV_UP)
2252	fAdded = true;
2253	else if (ulEventType == NETDEV_DOWN)
2254	fAdded = false;
2255	else
2256	return NOTIFY_OK;
2257
2258	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2259	kIntNetAddrType_IPv6, &ifa->addr);
2260	}
2261
2262	return rc;
2263	}
2264
2265
2266	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2267	{
2268	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2269	}
2270
2271	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2272	{
2273	struct net_device * pDev;
2274	int err;
2275	int rc = VINF_SUCCESS;
2276	IPRT_LINUX_SAVE_EFL_AC();
2277	NOREF(pvIfData);
2278
2279	LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2280
2281	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2282	if (pDev)
2283	{
2284	/*
2285	* Create a sk_buff for the gather list and push it onto the wire.
2286	*/
2287	if (fDst & INTNETTRUNKDIR_WIRE)
2288	{
2289	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2290	if (pBuf)
2291	{
2292	vboxNetFltDumpPacket(pSG, true, "wire", 1);
2293	Log6(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2294	Log6(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2295	err = dev_queue_xmit(pBuf);
2296	if (err)
2297	rc = RTErrConvertFromErrno(err);
2298	}
2299	else
2300	rc = VERR_NO_MEMORY;
2301	}
2302
2303	/*
2304	* Create a sk_buff for the gather list and push it onto the host stack.
2305	*/
2306	if (fDst & INTNETTRUNKDIR_HOST)
2307	{
2308	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2309	if (pBuf)
2310	{
2311	vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2312	Log6(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2313	Log6(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2314	#if RTLNX_VER_MIN(5,18,0)
2315	local_bh_disable();
2316	err = netif_rx(pBuf);
2317	local_bh_enable();
2318	#else
2319	err = netif_rx_ni(pBuf);
2320	#endif
2321	if (err)
2322	rc = RTErrConvertFromErrno(err);
2323	}
2324	else
2325	rc = VERR_NO_MEMORY;
2326	}
2327
2328	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2329	}
2330
2331	IPRT_LINUX_RESTORE_EFL_AC();
2332	return rc;
2333	}
2334
2335
2336	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2337	{
2338	struct net_device *pDev;
2339	IPRT_LINUX_SAVE_EFL_AC();
2340
2341	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%RTbool, fDisablePromiscuous=%RTbool\n",
2342	pThis, pThis->szName, fActive, pThis->fDisablePromiscuous));
2343
2344	if (pThis->fDisablePromiscuous)
2345	return;
2346
2347	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2348	if (pDev)
2349	{
2350	/*
2351	* This api is a bit weird, the best reference is the code.
2352	*
2353	* Also, we have a bit or race conditions wrt the maintenance of
2354	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2355	*/
2356	#ifdef LOG_ENABLED
2357	u_int16_t fIf;
2358	unsigned const cPromiscBefore = pDev->promiscuity;
2359	#endif
2360	if (fActive)
2361	{
2362	Assert(!pThis->u.s.fPromiscuousSet);
2363
2364	rtnl_lock();
2365	dev_set_promiscuity(pDev, 1);
2366	rtnl_unlock();
2367	pThis->u.s.fPromiscuousSet = true;
2368	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2369	}
2370	else
2371	{
2372	if (pThis->u.s.fPromiscuousSet)
2373	{
2374	rtnl_lock();
2375	dev_set_promiscuity(pDev, -1);
2376	rtnl_unlock();
2377	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2378	}
2379	pThis->u.s.fPromiscuousSet = false;
2380
2381	#ifdef LOG_ENABLED
2382	fIf = dev_get_flags(pDev);
2383	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2384	#endif
2385	}
2386
2387	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2388	}
2389	IPRT_LINUX_RESTORE_EFL_AC();
2390	}
2391
2392
2393	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2394	{
2395	/*
2396	* Remove packet handler when we get disconnected from internal switch as
2397	* we don't want the handler to forward packets to disconnected switch.
2398	*/
2399	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
2400	{
2401	IPRT_LINUX_SAVE_EFL_AC();
2402	dev_remove_pack(&pThis->u.s.PacketType);
2403	Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
2404	IPRT_LINUX_RESTORE_EFL_AC();
2405	}
2406	return VINF_SUCCESS;
2407	}
2408
2409
2410	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2411	{
2412	IPRT_LINUX_SAVE_EFL_AC();
2413
2414	/*
2415	* Report the GSO capabilities of the host and device (if connected).
2416	* Note! No need to mark ourselves busy here.
2417	*/
2418	/** @todo duplicate work here now? Attach */
2419	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2420	Log3(("vboxNetFltOsConnectIt: reporting host tso tso6\n"));
2421	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2422	0
2423	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2424	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2425	, INTNETTRUNKDIR_HOST);
2426
2427	#endif
2428	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2429
2430	IPRT_LINUX_RESTORE_EFL_AC();
2431	return VINF_SUCCESS;
2432	}
2433
2434
2435	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2436	{
2437	struct net_device *pDev;
2438	bool fRegistered;
2439	IPRT_LINUX_SAVE_EFL_AC();
2440
2441	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
2442	vboxNetFltLinuxUnhookDev(pThis, NULL);
2443	#endif
2444
2445	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2446	* unlikely, but none the less). Since it doesn't actually update the
2447	* state (just reads it), it is likely to panic in some interesting
2448	* ways. */
2449
2450	RTSpinlockAcquire(pThis->hSpinlock);
2451	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2452	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
2453	RTSpinlockRelease(pThis->hSpinlock);
2454
2455	if (fRegistered)
2456	{
2457	vboxNetFltSetLinkState(pThis, pDev, false);
2458
2459	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2460	skb_queue_purge(&pThis->u.s.XmitQueue);
2461	#endif
2462	Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
2463	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
2464	pDev, pDev->name,
2465	#if RTLNX_VER_MIN(2,6,37)
2466	netdev_refcnt_read(pDev)
2467	#else
2468	atomic_read(&pDev->refcnt)
2469	#endif
2470	));
2471	dev_put(pDev);
2472	}
2473
2474	unregister_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2475	unregister_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2476
2477	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2478	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2479	module_put(THIS_MODULE);
2480
2481	IPRT_LINUX_RESTORE_EFL_AC();
2482	}
2483
2484
2485	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2486	{
2487	int err;
2488	IPRT_LINUX_SAVE_EFL_AC();
2489	NOREF(pvContext);
2490
2491	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2492	err = register_netdevice_notifier(&pThis->u.s.Notifier);
2493	if (err)
2494	{
2495	IPRT_LINUX_RESTORE_EFL_AC();
2496	return VERR_INTNET_FLT_IF_FAILED;
2497	}
2498	if (!pThis->u.s.fRegistered)
2499	{
2500	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2501	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2502	IPRT_LINUX_RESTORE_EFL_AC();
2503	return VERR_INTNET_FLT_IF_NOT_FOUND;
2504	}
2505
2506	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2507	if ( pThis->fDisconnectedFromHost
2508	\|\| !try_module_get(THIS_MODULE))
2509	{
2510	IPRT_LINUX_RESTORE_EFL_AC();
2511	return VERR_INTNET_FLT_IF_FAILED;
2512	}
2513
2514	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2515	{
2516	VBOXNETFLTNOTIFIER Enumerator;
2517
2518	/*
2519	* register_inetaddr_notifier() and register_inet6addr_notifier()
2520	* do not call the callback for existing devices. Enumerating
2521	* all network devices explicitly is a bit of an ifdef mess,
2522	* so co-opt register_netdevice_notifier() to do that for us.
2523	*/
2524	RT_ZERO(Enumerator);
2525	Enumerator.Notifier.notifier_call = vboxNetFltLinuxEnumeratorCallback;
2526	Enumerator.pThis = pThis;
2527
2528	err = register_netdevice_notifier(&Enumerator.Notifier);
2529	if (err)
2530	{
2531	LogRel(("%s: failed to enumerate network devices: error %d\n", __FUNCTION__, err));
2532	IPRT_LINUX_RESTORE_EFL_AC();
2533	return VINF_SUCCESS;
2534	}
2535
2536	unregister_netdevice_notifier(&Enumerator.Notifier);
2537
2538	pThis->u.s.NotifierIPv4.notifier_call = vboxNetFltLinuxNotifierIPv4Callback;
2539	err = register_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2540	if (err)
2541	LogRel(("%s: failed to register IPv4 notifier: error %d\n", __FUNCTION__, err));
2542
2543	pThis->u.s.NotifierIPv6.notifier_call = vboxNetFltLinuxNotifierIPv6Callback;
2544	err = register_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2545	if (err)
2546	LogRel(("%s: failed to register IPv6 notifier: error %d\n", __FUNCTION__, err));
2547	}
2548
2549	IPRT_LINUX_RESTORE_EFL_AC();
2550	return VINF_SUCCESS;
2551	}
2552
2553	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2554	{
2555	IPRT_LINUX_SAVE_EFL_AC();
2556
2557	/*
2558	* Init the linux specific members.
2559	*/
2560	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2561	pThis->u.s.fRegistered = false;
2562	pThis->u.s.fPromiscuousSet = false;
2563	pThis->u.s.fPacketHandler = false;
2564	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2565	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2566	skb_queue_head_init(&pThis->u.s.XmitQueue);
2567	# if RTLNX_VER_MIN(2,6,20)
2568	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2569	# else
2570	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2571	# endif
2572	#endif
2573
2574	IPRT_LINUX_RESTORE_EFL_AC();
2575	return VINF_SUCCESS;
2576	}
2577
2578
2579	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2580	{
2581	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2582	}
2583
2584
2585	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
2586	{
2587	/* Nothing to do */
2588	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2589	return VINF_SUCCESS;
2590	}
2591
2592
2593	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2594	{
2595	/* Nothing to do */
2596	NOREF(pThis); NOREF(pvIfData);
2597	return VINF_SUCCESS;
2598	}
2599

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 94991

Download in other formats: