VBoxNetFlt-linux.c@ 86001

Last change on this file since 86001 was 85703, checked in by vboxsync, 4 years ago
IPRT,lnx-kmods: s/RTLNX_RHEL_PREREQ/RTLNX_RHEL_MIN/; added RTLNX_RHEL_MAX and RTLNX_RHEL_RANGE. Use them instead of the RHEL_XXXX defines everywhere.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 89.9 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 85703 2020-08-11 18:54:01Z vboxsync $ */
2	/** @file
3	* VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2020 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	#include <linux/etherdevice.h>
42	#include <linux/rtnetlink.h>
43	#include <linux/miscdevice.h>
44	#include <linux/inetdevice.h>
45	#include <linux/in.h>
46	#include <linux/ip.h>
47	#include <linux/if_vlan.h>
48	#if RTLNX_VER_MIN(4,5,0)
49	#include <uapi/linux/pkt_cls.h>
50	#endif
51	#include <net/ipv6.h>
52	#include <net/if_inet6.h>
53	#include <net/addrconf.h>
54
55	#include <VBox/log.h>
56	#include <VBox/err.h>
57	#include <VBox/intnetinline.h>
58	#include <VBox/vmm/pdmnetinline.h>
59	#include <VBox/param.h>
60	#include <iprt/alloca.h>
61	#include <iprt/assert.h>
62	#include <iprt/spinlock.h>
63	#include <iprt/semaphore.h>
64	#include <iprt/initterm.h>
65	#include <iprt/process.h>
66	#include <iprt/mem.h>
67	#include <iprt/net.h>
68	#include <iprt/log.h>
69	#include <iprt/mp.h>
70	#include <iprt/mem.h>
71	#include <iprt/time.h>
72
73	#define VBOXNETFLT_OS_SPECFIC 1
74	#include "../VBoxNetFltInternal.h"
75
76	typedef struct VBOXNETFLTNOTIFIER {
77	struct notifier_block Notifier;
78	PVBOXNETFLTINS pThis;
79	} VBOXNETFLTNOTIFIER;
80	typedef struct VBOXNETFLTNOTIFIER *PVBOXNETFLTNOTIFIER;
81
82
83	/*********************************************************************************************************************************
84	* Defined Constants And Macros *
85	*********************************************************************************************************************************/
86	#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
87	#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
88	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
89	# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
90	#endif
91
92	#if RTLNX_VER_MIN(3,11,0)
93	# define VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr) netdev_notifier_info_to_dev(ptr)
94	#else
95	# define VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr) ((struct net_device *)ptr)
96	#endif
97
98	#if RTLNX_VER_MIN(3,5,0)
99	# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(skb_frag_page(frag))
100	# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr)
101	#else
102	# if RTLNX_VER_MIN(3,2,0)
103	# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(skb_frag_page(frag), KM_SKB_DATA_SOFTIRQ)
104	# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ)
105	# else
106	# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(frag->page, KM_SKB_DATA_SOFTIRQ)
107	# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ)
108	# endif
109	#endif
110
111	#if RTLNX_VER_MIN(2,6,34)
112	# define VBOX_NETDEV_NAME(dev) netdev_name(dev)
113	#else
114	# define VBOX_NETDEV_NAME(dev) ((dev)->reg_state != NETREG_REGISTERED ? "(unregistered net_device)" : (dev)->name)
115	#endif
116
117	#if RTLNX_VER_MIN(2,6,25)
118	# define VBOX_IPV4_IS_LOOPBACK(addr) ipv4_is_loopback(addr)
119	# define VBOX_IPV4_IS_LINKLOCAL_169(addr) ipv4_is_linklocal_169(addr)
120	#else
121	# define VBOX_IPV4_IS_LOOPBACK(addr) ((addr & htonl(IN_CLASSA_NET)) == htonl(0x7f000000))
122	# define VBOX_IPV4_IS_LINKLOCAL_169(addr) ((addr & htonl(IN_CLASSB_NET)) == htonl(0xa9fe0000))
123	#endif
124
125	#if RTLNX_VER_MIN(2,6,22)
126	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
127	# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
128	# define VBOX_SKB_CSUM_OFFSET(skb) skb->csum_offset
129	#else
130	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
131	# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
132	# define VBOX_SKB_CSUM_OFFSET(skb) skb->csum
133	#endif
134
135	#if RTLNX_VER_MIN(2,6,19)
136	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
137	#else
138	# define CHECKSUM_PARTIAL CHECKSUM_HW
139	# if RTLNX_VER_MIN(2,6,10)
140	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
141	# else
142	# if RTLNX_VER_MIN(2,6,7)
143	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
144	# else
145	# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
146	# endif
147	/* Versions prior 2.6.10 use stats for both bstats and qstats */
148	# define bstats stats
149	# define qstats stats
150	# endif
151	#endif
152
153	#if RTLNX_VER_MIN(3,20,0) \|\| RTLNX_RHEL_RANGE(7,2, 8,0) \|\| RTLNX_RHEL_RANGE(6,8, 7,0)
154	# define VBOX_HAVE_SKB_VLAN
155	#endif
156
157	#ifdef VBOX_HAVE_SKB_VLAN
158	# define vlan_tx_tag_get(skb) skb_vlan_tag_get(skb)
159	# define vlan_tx_tag_present(skb) skb_vlan_tag_present(skb)
160	#endif
161
162	#ifndef NET_IP_ALIGN
163	# define NET_IP_ALIGN 2
164	#endif
165
166	#if 1
167	/** Create scatter / gather segments for fragments. When not used, we will
168	* linearize the socket buffer before creating the internal networking SG. */
169	# define VBOXNETFLT_SG_SUPPORT 1
170	#endif
171
172	#if RTLNX_VER_MIN(2,6,18)
173
174	/** Indicates that the linux kernel may send us GSO frames. */
175	# define VBOXNETFLT_WITH_GSO 1
176
177	/** This enables or disables the transmitting of GSO frame from the internal
178	* network and to the host. */
179	# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
180
181	# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
182	/** This enables or disables the transmitting of GSO frame from the internal
183	* network and to the wire. */
184	# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
185	# endif
186
187	/** This enables or disables the forwarding/flooding of GSO frame from the host
188	* to the internal network. */
189	# define VBOXNETFLT_WITH_GSO_RECV 1
190
191	#endif /* RTLNX_VER_MIN(2,6,18) */
192
193	#if RTLNX_VER_MIN(2,6,29)
194	/** This enables or disables handling of GSO frames coming from the wire (GRO). */
195	# define VBOXNETFLT_WITH_GRO 1
196	#endif
197
198	/*
199	* GRO support was backported to RHEL 5.4
200	*/
201	#if RTLNX_RHEL_MAJ_PREREQ(5, 4)
202	# define VBOXNETFLT_WITH_GRO 1
203	#endif
204
205
206	/*********************************************************************************************************************************
207	* Internal Functions *
208	*********************************************************************************************************************************/
209	static int __init VBoxNetFltLinuxInit(void);
210	static void __exit VBoxNetFltLinuxUnload(void);
211	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
212
213
214	/*********************************************************************************************************************************
215	* Global Variables *
216	*********************************************************************************************************************************/
217	/**
218	* The (common) global data.
219	*/
220	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
221
222	module_init(VBoxNetFltLinuxInit);
223	module_exit(VBoxNetFltLinuxUnload);
224
225	MODULE_AUTHOR(VBOX_VENDOR);
226	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
227	MODULE_LICENSE("GPL");
228	#ifdef MODULE_VERSION
229	MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV) " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
230	#endif
231
232
233	#if RTLNX_VER_MAX(2,6,12) && defined(LOG_ENABLED)
234	unsigned dev_get_flags(const struct net_device *dev)
235	{
236	unsigned flags;
237
238	flags = (dev->flags & ~(IFF_PROMISC \|
239	IFF_ALLMULTI \|
240	IFF_RUNNING)) \|
241	(dev->gflags & (IFF_PROMISC \|
242	IFF_ALLMULTI));
243
244	if (netif_running(dev) && netif_carrier_ok(dev))
245	flags \|= IFF_RUNNING;
246
247	return flags;
248	}
249	#endif /* RTLNX_VER_MAX(2,6,12) */
250
251
252	/**
253	* Initialize module.
254	*
255	* @returns appropriate status code.
256	*/
257	static int __init VBoxNetFltLinuxInit(void)
258	{
259	int rc;
260	/*
261	* Initialize IPRT.
262	*/
263	rc = RTR0Init(0);
264	if (RT_SUCCESS(rc))
265	{
266	Log(("VBoxNetFltLinuxInit\n"));
267
268	/*
269	* Initialize the globals and connect to the support driver.
270	*
271	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
272	* for establishing the connect to the support driver.
273	*/
274	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
275	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
276	if (RT_SUCCESS(rc))
277	{
278	LogRel(("VBoxNetFlt: Successfully started.\n"));
279	return 0;
280	}
281
282	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
283	RTR0Term();
284	}
285	else
286	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
287
288	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
289	return -RTErrConvertToErrno(rc);
290	}
291
292
293	/**
294	* Unload the module.
295	*
296	* @todo We have to prevent this if we're busy!
297	*/
298	static void __exit VBoxNetFltLinuxUnload(void)
299	{
300	int rc;
301	Log(("VBoxNetFltLinuxUnload\n"));
302	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
303
304	/*
305	* Undo the work done during start (in reverse order).
306	*/
307	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
308	AssertRC(rc); NOREF(rc);
309
310	RTR0Term();
311
312	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
313
314	Log(("VBoxNetFltLinuxUnload - done\n"));
315	}
316
317
318	/**
319	* We filter traffic from the host to the internal network
320	* before it reaches the NIC driver.
321	*
322	* The current code uses a very ugly hack overriding hard_start_xmit
323	* callback in the device structure, but it has been shown to give us a
324	* performance boost of 60-100% though. Eventually we have to find some
325	* less hacky way of getting this job done.
326	*/
327	#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
328
329	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
330
331	# if RTLNX_VER_MAX(2,6,29)
332
333	# include <linux/ethtool.h>
334
335	typedef struct ethtool_ops OVR_OPSTYPE;
336	# define OVR_OPS ethtool_ops
337	# define OVR_XMIT pfnStartXmit
338
339	# else /* RTLNX_VER_MIN(2,6,29) */
340
341	typedef struct net_device_ops OVR_OPSTYPE;
342	# define OVR_OPS netdev_ops
343	# define OVR_XMIT pOrgOps->ndo_start_xmit
344
345	# endif /* RTLNX_VER_MIN(2,6,29) */
346
347	/**
348	* The overridden net_device_ops of the device we're attached to.
349	*
350	* As there is no net_device_ops structure in pre-2.6.29 kernels we override
351	* ethtool_ops instead along with hard_start_xmit callback in net_device
352	* structure.
353	*
354	* This is a very dirty hack that was created to explore how much we can improve
355	* the host to guest transfers by not CC'ing the NIC. It turns out to be
356	* the only way to filter outgoing packets for devices without TX queue.
357	*/
358	typedef struct VBoxNetDeviceOpsOverride
359	{
360	/** Our overridden ops. */
361	OVR_OPSTYPE Ops;
362	/** Magic word. */
363	uint32_t u32Magic;
364	/** Pointer to the original ops. */
365	OVR_OPSTYPE const *pOrgOps;
366	# if RTLNX_VER_MAX(2,6,29)
367	/** Pointer to the original hard_start_xmit function. */
368	int (pfnStartXmit)(struct sk_buff pSkb, struct net_device *pDev);
369	# endif /* RTLNX_VER_MAX(2,6,29) */
370	/** Pointer to the net filter instance. */
371	PVBOXNETFLTINS pVBoxNetFlt;
372	/** The number of filtered packages. */
373	uint64_t cFiltered;
374	/** The total number of packets */
375	uint64_t cTotal;
376	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
377	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
378	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
379
380	/**
381	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
382	* because they belong on the internal network.
383	*
384	* @returns NETDEV_TX_XXX.
385	* @param pSkb The socket buffer to transmit.
386	* @param pDev The net device.
387	*/
388	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
389	{
390	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
391	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
392	PCRTNETETHERHDR pEtherHdr;
393	PINTNETTRUNKSWPORT pSwitchPort;
394	uint32_t cbHdrs;
395
396
397	/*
398	* Validate the override structure.
399	*
400	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
401	* to be production quality code, we would have to be much more
402	* careful here and avoid the race.
403	*/
404	if ( !VALID_PTR(pOverride)
405	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
406	# if RTLNX_VER_MIN(2,6,29)
407	\|\| !VALID_PTR(pOverride->pOrgOps)
408	# endif /* RTLNX_VER_MIN(2,6,29) */
409	)
410	{
411	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
412	dev_kfree_skb(pSkb);
413	return NETDEV_TX_OK;
414	}
415	pOverride->cTotal++;
416
417	/*
418	* Do the filtering base on the default OUI of our virtual NICs
419	*
420	* Note! In a real solution, we would ask the switch whether the
421	* destination MAC is 100% to be on the internal network and then
422	* drop it.
423	*/
424	cbHdrs = skb_headlen(pSkb);
425	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
426	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
427	if ( pEtherHdr
428	&& VALID_PTR(pOverride->pVBoxNetFlt)
429	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
430	&& VALID_PTR(pSwitchPort)
431	&& cbHdrs >= 6)
432	{
433	INTNETSWDECISION enmDecision;
434
435	/** @todo consider reference counting, etc. */
436	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
437	if (enmDecision == INTNETSWDECISION_INTNET)
438	{
439	dev_kfree_skb(pSkb);
440	pOverride->cFiltered++;
441	return NETDEV_TX_OK;
442	}
443	}
444
445	return pOverride->OVR_XMIT(pSkb, pDev);
446	}
447
448	/**
449	* Hooks the device ndo_start_xmit operation of the device.
450	*
451	* @param pThis The net filter instance.
452	* @param pDev The net device.
453	*/
454	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
455	{
456	PVBOXNETDEVICEOPSOVERRIDE pOverride;
457
458	/* Cancel override if ethtool_ops is missing (host-only case, @bugref{5712}) */
459	if (!VALID_PTR(pDev->OVR_OPS))
460	return;
461	pOverride = RTMemAlloc(sizeof(*pOverride));
462	if (!pOverride)
463	return;
464	pOverride->pOrgOps = pDev->OVR_OPS;
465	pOverride->Ops = *pDev->OVR_OPS;
466	# if RTLNX_VER_MAX(2,6,29)
467	pOverride->pfnStartXmit = pDev->hard_start_xmit;
468	# else /* RTLNX_VER_MIN(2,6,29) */
469	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
470	# endif /* RTLNX_VER_MIN(2,6,29) */
471	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
472	pOverride->cTotal = 0;
473	pOverride->cFiltered = 0;
474	pOverride->pVBoxNetFlt = pThis;
475
476	RTSpinlockAcquire(pThis->hSpinlock); /* (this isn't necessary, but so what) */
477	ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
478	# if RTLNX_VER_MAX(2,6,29)
479	ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
480	# endif /* RTLNX_VER_MAX(2,6,29) */
481	RTSpinlockRelease(pThis->hSpinlock);
482	}
483
484	/**
485	* Undos what vboxNetFltLinuxHookDev did.
486	*
487	* @param pThis The net filter instance.
488	* @param pDev The net device. Can be NULL, in which case
489	* we'll try retrieve it from @a pThis.
490	*/
491	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
492	{
493	PVBOXNETDEVICEOPSOVERRIDE pOverride;
494
495	RTSpinlockAcquire(pThis->hSpinlock);
496	if (!pDev)
497	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
498	if (VALID_PTR(pDev))
499	{
500	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
501	if ( VALID_PTR(pOverride)
502	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
503	&& VALID_PTR(pOverride->pOrgOps)
504	)
505	{
506	# if RTLNX_VER_MAX(2,6,29)
507	ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
508	# endif /* RTLNX_VER_MAX(2,6,29) */
509	ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
510	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
511	}
512	else
513	pOverride = NULL;
514	}
515	else
516	pOverride = NULL;
517	RTSpinlockRelease(pThis->hSpinlock);
518
519	if (pOverride)
520	{
521	printk("vboxnetflt: %llu out of %llu packets were not sent (directed to host)\n", pOverride->cFiltered, pOverride->cTotal);
522	RTMemFree(pOverride);
523	}
524	}
525
526	#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
527
528
529	/**
530	* Reads and retains the host interface handle.
531	*
532	* @returns The handle, NULL if detached.
533	* @param pThis
534	*/
535	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
536	{
537	#if 0
538	struct net_device *pDev = NULL;
539
540	Log(("vboxNetFltLinuxRetainNetDev\n"));
541	/*
542	* Be careful here to avoid problems racing the detached callback.
543	*/
544	RTSpinlockAcquire(pThis->hSpinlock);
545	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
546	{
547	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
548	if (pDev)
549	{
550	dev_hold(pDev);
551	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
552	pDev, pDev->name,
553	#if RTLNX_VER_MIN(2,6,37)
554	netdev_refcnt_read(pDev)
555	#else
556	atomic_read(&pDev->refcnt)
557	#endif
558	));
559	}
560	}
561	RTSpinlockRelease(pThis->hSpinlock);
562
563	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
564	return pDev;
565	#else
566	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
567	#endif
568	}
569
570
571	/**
572	* Release the host interface handle previously retained
573	* by vboxNetFltLinuxRetainNetDev.
574	*
575	* @param pThis The instance.
576	* @param pDev The vboxNetFltLinuxRetainNetDev
577	* return value, NULL is fine.
578	*/
579	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
580	{
581	#if 0
582	Log(("vboxNetFltLinuxReleaseNetDev\n"));
583	NOREF(pThis);
584	if (pDev)
585	{
586	dev_put(pDev);
587	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
588	pDev, pDev->name,
589	#if RTLNX_VER_MIN(2,6,37)
590	netdev_refcnt_read(pDev)
591	#else
592	atomic_read(&pDev->refcnt)
593	#endif
594	));
595	}
596	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
597	#endif
598	}
599
600	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
601	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
602
603	/**
604	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
605	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
606	*
607	* @returns true / false accordingly.
608	* @param pBuf The sk_buff.
609	*/
610	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
611	{
612	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
613	}
614
615
616	/**
617	* Checks whether this SG list contains a GSO packet.
618	*
619	* @returns true / false accordingly.
620	* @param pSG The (scatter/)gather list.
621	*/
622	DECLINLINE(bool) vboxNetFltLinuxIsGso(PINTNETSG pSG)
623	{
624	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
625	return !((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type == PDMNETWORKGSOTYPE_INVALID);
626	#else /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
627	return false;
628	#endif /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
629	}
630
631
632	/**
633	* Find out the frame size (of a single segment in case of GSO frames).
634	*
635	* @returns the frame size.
636	* @param pSG The (scatter/)gather list.
637	*/
638	DECLINLINE(uint32_t) vboxNetFltLinuxFrameSize(PINTNETSG pSG)
639	{
640	uint16_t u16Type = 0;
641	uint32_t cbVlanTag = 0;
642	if (pSG->aSegs[0].cb >= sizeof(RTNETETHERHDR))
643	u16Type = RT_BE2H_U16(((PCRTNETETHERHDR)pSG->aSegs[0].pv)->EtherType);
644	else if (pSG->cbTotal >= sizeof(RTNETETHERHDR))
645	{
646	uint32_t off = RT_UOFFSETOF(RTNETETHERHDR, EtherType);
647	uint32_t i;
648	for (i = 0; i < pSG->cSegsUsed; ++i)
649	{
650	if (off <= pSG->aSegs[i].cb)
651	{
652	if (off + sizeof(uint16_t) <= pSG->aSegs[i].cb)
653	u16Type = RT_BE2H_U16((uint16_t )((uintptr_t)pSG->aSegs[i].pv + off));
654	else if (i + 1 < pSG->cSegsUsed)
655	u16Type = RT_BE2H_U16( ((uint16_t)( ((uint8_t *)pSG->aSegs[i].pv)[off] ) << 8)
656	+ (uint8_t )pSG->aSegs[i + 1].pv); /* ASSUMES no empty segments! */
657	/* else: frame is too short. */
658	break;
659	}
660	off -= pSG->aSegs[i].cb;
661	}
662	}
663	if (u16Type == RTNET_ETHERTYPE_VLAN)
664	cbVlanTag = 4;
665	return (vboxNetFltLinuxIsGso(pSG) ? (uint32_t)pSG->GsoCtx.cbMaxSeg + pSG->GsoCtx.cbHdrsTotal : pSG->cbTotal) - cbVlanTag;
666	}
667
668
669	/**
670	* Internal worker that create a linux sk_buff for a
671	* (scatter/)gather list.
672	*
673	* @returns Pointer to the sk_buff.
674	* @param pThis The instance.
675	* @param pSG The (scatter/)gather list.
676	* @param fDstWire Set if the destination is the wire.
677	*/
678	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
679	{
680	struct sk_buff *pPkt;
681	struct net_device *pDev;
682	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
683	unsigned fGsoType = 0;
684	#endif
685
686	if (pSG->cbTotal == 0)
687	{
688	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
689	return NULL;
690	}
691	Log5(("VBoxNetFlt: Packet to %s of %d bytes (frame=%d).\n", fDstWire?"wire":"host", pSG->cbTotal, vboxNetFltLinuxFrameSize(pSG)));
692	if (fDstWire && (vboxNetFltLinuxFrameSize(pSG) > ASMAtomicReadU32(&pThis->u.s.cbMtu) + 14))
693	{
694	static bool s_fOnce = true;
695	if (s_fOnce)
696	{
697	s_fOnce = false;
698	printk("VBoxNetFlt: Dropped over-sized packet (%d bytes) coming from internal network.\n", vboxNetFltLinuxFrameSize(pSG));
699	}
700	return NULL;
701	}
702
703	/** @todo We should use fragments mapping the SG buffers with large packets.
704	* 256 bytes seems to be the a threshold used a lot for this. It
705	* requires some nasty work on the intnet side though... */
706	/*
707	* Allocate a packet and copy over the data.
708	*/
709	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
710	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
711	if (RT_UNLIKELY(!pPkt))
712	{
713	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
714	pSG->pvUserData = NULL;
715	return NULL;
716	}
717	pPkt->dev = pDev;
718	pPkt->ip_summed = CHECKSUM_NONE;
719
720	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
721	skb_reserve(pPkt, NET_IP_ALIGN);
722
723	/* Copy the segments. */
724	skb_put(pPkt, pSG->cbTotal);
725	IntNetSgRead(pSG, pPkt->data);
726
727	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
728	/*
729	* Setup GSO if used by this packet.
730	*/
731	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
732	{
733	default:
734	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
735	/* fall thru */
736	case PDMNETWORKGSOTYPE_INVALID:
737	fGsoType = 0;
738	break;
739	case PDMNETWORKGSOTYPE_IPV4_TCP:
740	fGsoType = SKB_GSO_TCPV4;
741	break;
742	case PDMNETWORKGSOTYPE_IPV6_TCP:
743	fGsoType = SKB_GSO_TCPV6;
744	break;
745	}
746	if (fGsoType)
747	{
748	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
749
750	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
751	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
752	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
753
754	/*
755	* We need to set checksum fields even if the packet goes to the host
756	* directly as it may be immediately forwarded by IP layer @bugref{5020}.
757	*/
758	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
759	pPkt->ip_summed = CHECKSUM_PARTIAL;
760	# if RTLNX_VER_MIN(2,6,22)
761	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
762	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
763	pPkt->csum_offset = RT_UOFFSETOF(RTNETTCP, th_sum);
764	else
765	pPkt->csum_offset = RT_UOFFSETOF(RTNETUDP, uh_sum);
766	# else
767	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
768	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
769	pPkt->csum = RT_UOFFSETOF(RTNETTCP, th_sum);
770	else
771	pPkt->csum = RT_UOFFSETOF(RTNETUDP, uh_sum);
772	# endif
773	if (!fDstWire)
774	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
775	}
776	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
777
778	/*
779	* Finish up the socket buffer.
780	*/
781	pPkt->protocol = eth_type_trans(pPkt, pDev);
782	if (fDstWire)
783	{
784	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
785
786	/* Restore ethernet header back. */
787	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
788	VBOX_SKB_RESET_MAC_HDR(pPkt);
789	}
790	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
791
792	return pPkt;
793	}
794
795
796	/**
797	* Return the offset where to start checksum computation from.
798	*
799	* @returns the offset relative to pBuf->data.
800	* @param pBuf The socket buffer.
801	*/
802	DECLINLINE(unsigned) vboxNetFltLinuxGetChecksumStartOffset(struct sk_buff *pBuf)
803	{
804	#if RTLNX_VER_MIN(2,6,38)
805	return skb_checksum_start_offset(pBuf);
806	#elif RTLNX_VER_MIN(2,6,22)
807	return pBuf->csum_start - skb_headroom(pBuf);
808	#else
809	unsigned char *pTransportHdr = pBuf->h.raw;
810	# if RTLNX_VER_MAX(2,6,19)
811	/*
812	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
813	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
814	* header length from the header itself and reconstruct 'h' pointer
815	* to TCP (or whatever) header.
816	*/
817	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
818	pTransportHdr = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
819	# endif
820	return pTransportHdr - pBuf->data;
821	#endif
822	}
823
824
825	/**
826	* Initializes a SG list from an sk_buff.
827	*
828	* @returns Number of segments.
829	* @param pThis The instance.
830	* @param pBuf The sk_buff.
831	* @param pSG The SG.
832	* @param cbExtra The number of bytes of extra space allocated immediately after the SG.
833	* @param cSegs The number of segments allocated for the SG.
834	* This should match the number in the mbuf exactly!
835	* @param fSrc The source of the frame.
836	* @param pGsoCtx Pointer to the GSO context if it's a GSO
837	* internal network frame. NULL if regular frame.
838	*/
839	static void vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
840	unsigned cbExtra, unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
841	{
842	int i;
843	NOREF(pThis);
844
845	#ifndef VBOXNETFLT_SG_SUPPORT
846	Assert(!skb_shinfo(pBuf)->frag_list);
847	#else /* VBOXNETFLT_SG_SUPPORT */
848	uint8_t pExtra = (uint8_t )&pSG->aSegs[cSegs];
849	unsigned cbConsumed = 0;
850	unsigned cbProduced = 0;
851
852	# if RTLNX_VER_MIN(2,6,27)
853	/* Restore VLAN tag stripped by host hardware */
854	if (vlan_tx_tag_present(pBuf))
855	{
856	uint8_t *pMac = pBuf->data;
857	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )pExtra;
858	Assert(ETH_ALEN * 2 + VLAN_HLEN <= cbExtra);
859	memmove(pVHdr, pMac, ETH_ALEN * 2);
860	/* Consume whole Ethernet header: 2 addresses + EtherType (see @bugref{8599}) */
861	cbConsumed += ETH_ALEN * 2 + sizeof(uint16_t);
862	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
863	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
864	pVHdr->h_vlan_encapsulated_proto = (uint16_t)(pMac + ETH_ALEN * 2);
865	cbProduced += VLAN_ETH_HLEN;
866	}
867	# endif /* RTLNX_VER_MIN(2,6,27) */
868
869	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
870	{
871	unsigned uCsumStartOffset = vboxNetFltLinuxGetChecksumStartOffset(pBuf);
872	unsigned uCsumStoreOffset = uCsumStartOffset + VBOX_SKB_CSUM_OFFSET(pBuf) - cbConsumed;
873	Log3(("cbConsumed=%u cbProduced=%u uCsumStartOffset=%u uCsumStoreOffset=%u\n",
874	cbConsumed, cbProduced, uCsumStartOffset, uCsumStoreOffset));
875	Assert(cbProduced + uCsumStoreOffset + sizeof(uint16_t) <= cbExtra);
876	/*
877	* We assume that the checksum is stored at the very end of the transport header
878	* so we will have all headers in a single fragment. If our assumption is wrong
879	* we may see suboptimal performance.
880	*/
881	memmove(pExtra + cbProduced,
882	pBuf->data + cbConsumed,
883	uCsumStoreOffset);
884	unsigned uChecksum = skb_checksum(pBuf, uCsumStartOffset, pBuf->len - uCsumStartOffset, 0);
885	(uint16_t)(pExtra + cbProduced + uCsumStoreOffset) = csum_fold(uChecksum);
886	cbProduced += uCsumStoreOffset + sizeof(uint16_t);
887	cbConsumed += uCsumStoreOffset + sizeof(uint16_t);
888	}
889	#endif /* VBOXNETFLT_SG_SUPPORT */
890
891	if (!pGsoCtx)
892	IntNetSgInitTempSegs(pSG, pBuf->len + cbProduced - cbConsumed, cSegs, 0 /cSegsUsed/);
893	else
894	IntNetSgInitTempSegsGso(pSG, pBuf->len + cbProduced - cbConsumed, cSegs, 0 /cSegsUsed/, pGsoCtx);
895
896	int iSeg = 0;
897	#ifdef VBOXNETFLT_SG_SUPPORT
898	if (cbProduced)
899	{
900	pSG->aSegs[iSeg].cb = cbProduced;
901	pSG->aSegs[iSeg].pv = pExtra;
902	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
903	}
904	pSG->aSegs[iSeg].cb = skb_headlen(pBuf) - cbConsumed;
905	pSG->aSegs[iSeg].pv = pBuf->data + cbConsumed;
906	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
907	Assert(iSeg <= pSG->cSegsAlloc);
908
909	# ifdef LOG_ENABLED
910	if (pBuf->data_len)
911	Log6((" kmap_atomic:"));
912	# endif /* LOG_ENABLED */
913	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
914	{
915	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
916	# if RTLNX_VER_MIN(5,4,0) \|\| RTLNX_SUSE_MAJ_PREREQ(15, 2)
917	pSG->aSegs[iSeg].cb = pFrag->bv_len;
918	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->bv_offset;
919	# else /* < KERNEL_VERSION(5, 4, 0) */
920	pSG->aSegs[iSeg].cb = pFrag->size;
921	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
922	# endif /* >= KERNEL_VERSION(5, 4, 0) */
923	Log6((" %p", pSG->aSegs[iSeg].pv));
924	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
925	Assert(iSeg <= pSG->cSegsAlloc);
926	}
927	struct sk_buff *pFragBuf;
928	for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
929	{
930	pSG->aSegs[iSeg].cb = skb_headlen(pFragBuf);
931	pSG->aSegs[iSeg].pv = pFragBuf->data;
932	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
933	Assert(iSeg <= pSG->cSegsAlloc);
934	for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
935	{
936	skb_frag_t *pFrag = &skb_shinfo(pFragBuf)->frags[i];
937	# if RTLNX_VER_MIN(5,4,0) \|\| RTLNX_SUSE_MAJ_PREREQ(15, 2)
938	pSG->aSegs[iSeg].cb = pFrag->bv_len;
939	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->bv_offset;
940	# else /* < KERNEL_VERSION(5, 4, 0) */
941	pSG->aSegs[iSeg].cb = pFrag->size;
942	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
943	# endif /* >= KERNEL_VERSION(5, 4, 0) */
944	Log6((" %p", pSG->aSegs[iSeg].pv));
945	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
946	Assert(iSeg <= pSG->cSegsAlloc);
947	}
948	}
949	# ifdef LOG_ENABLED
950	if (pBuf->data_len)
951	Log6(("\n"));
952	# endif /* LOG_ENABLED */
953	#else
954	pSG->aSegs[iSeg].cb = pBuf->len;
955	pSG->aSegs[iSeg].pv = pBuf->data;
956	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
957	#endif
958
959	pSG->cSegsUsed = iSeg;
960
961	#if 0
962	if (cbProduced)
963	{
964	LogRel(("vboxNetFltLinuxSkBufToSG: original packet dump:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
965	LogRel(("vboxNetFltLinuxSkBufToSG: cbConsumed=%u cbProduced=%u cbExtra=%u\n", cbConsumed, cbProduced, cbExtra));
966	uint32_t offset = 0;
967	for (i = 0; i < pSG->cSegsUsed; ++i)
968	{
969	LogRel(("vboxNetFltLinuxSkBufToSG: seg#%d (%d bytes, starting at 0x%x):\n%.*Rhxd\n",
970	i, pSG->aSegs[i].cb, offset, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
971	offset += pSG->aSegs[i].cb;
972	}
973	}
974	#endif
975
976	#ifdef PADD_RUNT_FRAMES_FROM_HOST
977	/*
978	* Add a trailer if the frame is too small.
979	*
980	* Since we're getting to the packet before it is framed, it has not
981	* yet been padded. The current solution is to add a segment pointing
982	* to a buffer containing all zeros and pray that works for all frames...
983	*/
984	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
985	{
986	Assert(pBuf->data_len == 0); /* Packets with fragments are never small! */
987	static uint8_t const s_abZero[128] = {0};
988
989	AssertReturnVoid(iSeg < cSegs);
990
991	pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
992	pSG->aSegs[iSeg].pv = (void *)&s_abZero[0];
993	pSG->aSegs[iSeg++].cb = 60 - pSG->cbTotal;
994	pSG->cbTotal = 60;
995	pSG->cSegsUsed++;
996	Assert(iSeg <= pSG->cSegsAlloc)
997	}
998	#endif
999
1000	Log6(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
1001	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
1002	for (i = 0; i < pSG->cSegsUsed; i++)
1003	Log6(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
1004	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1005	}
1006
1007	/**
1008	* Packet handler; not really documented - figure it out yourself.
1009	*
1010	* @returns 0 or EJUSTRETURN - this is probably copy & pastry and thus wrong.
1011	*/
1012	#if RTLNX_VER_MIN(2,6,14)
1013	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1014	struct net_device *pSkbDev,
1015	struct packet_type *pPacketType,
1016	struct net_device *pOrigDev)
1017	#else
1018	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1019	struct net_device *pSkbDev,
1020	struct packet_type *pPacketType)
1021	#endif
1022	{
1023	PVBOXNETFLTINS pThis;
1024	struct net_device *pDev;
1025	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1026	pBuf, pSkbDev, pPacketType));
1027	#if RTLNX_VER_MIN(2,6,18)
1028	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",
1029	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));
1030	# if RTLNX_VER_MIN(2,6,22)
1031	Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1032	# endif
1033	#else
1034	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",
1035	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));
1036	#endif
1037	/*
1038	* Drop it immediately?
1039	*/
1040	if (!pBuf)
1041	return 0;
1042
1043	if (pBuf->pkt_type == PACKET_LOOPBACK)
1044	{
1045	/*
1046	* We are not interested in loopbacked packets as they will always have
1047	* another copy going to the wire.
1048	*/
1049	Log2(("vboxNetFltLinuxPacketHandler: dropped loopback packet (cb=%u)\n", pBuf->len));
1050	dev_kfree_skb(pBuf); /* We must 'consume' all packets we get (@bugref{6539})! */
1051	return 0;
1052	}
1053
1054	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1055	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1056	if (pDev != pSkbDev)
1057	{
1058	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1059	kfree_skb(pBuf); /* This is a failure, so we use kfree_skb instead of dev_kfree_skb. */
1060	return 0;
1061	}
1062
1063	Log6(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1064	if (vboxNetFltLinuxSkBufIsOur(pBuf))
1065	{
1066	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1067	dev_kfree_skb(pBuf);
1068	return 0;
1069	}
1070
1071	#ifndef VBOXNETFLT_SG_SUPPORT
1072	{
1073	/*
1074	* Get rid of fragmented packets, they cause too much trouble.
1075	*/
1076	unsigned int uMacLen = pBuf->mac_len;
1077	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1078	dev_kfree_skb(pBuf);
1079	if (!pCopy)
1080	{
1081	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1082	return 0;
1083	}
1084	pBuf = pCopy;
1085	/* Somehow skb_copy ignores mac_len */
1086	pBuf->mac_len = uMacLen;
1087	# if RTLNX_VER_MIN(2,6,27)
1088	/* Restore VLAN tag stripped by host hardware */
1089	if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
1090	{
1091	uint8_t pMac = (uint8_t)skb_mac_header(pBuf);
1092	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )(pMac - VLAN_HLEN);
1093	memmove(pVHdr, pMac, ETH_ALEN * 2);
1094	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
1095	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
1096	pBuf->mac_header -= VLAN_HLEN;
1097	pBuf->mac_len += VLAN_HLEN;
1098	}
1099	# endif /* RTLNX_VER_MIN(2,6,27) */
1100
1101	# if RTLNX_VER_MIN(2,6,18)
1102	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",
1103	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));
1104	# if RTLNX_VER_MIN(2,6,22)
1105	Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1106	# endif /* RTLNX_VER_MIN(2,6,22) */
1107	# else /* RTLNX_VER_MAX(2,6,18) */
1108	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",
1109	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));
1110	# endif /* RTLNX_VER_MAX(2,6,18) */
1111	}
1112	#endif /* !VBOXNETFLT_SG_SUPPORT */
1113
1114	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1115	/* Forward it to the internal network. */
1116	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1117	#else /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1118	/* Add the packet to transmit queue and schedule the bottom half. */
1119	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1120	schedule_work(&pThis->u.s.XmitTask);
1121	Log6(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1122	&pThis->u.s.XmitTask, pBuf));
1123	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1124
1125	/* It does not really matter what we return, it is ignored by the kernel. */
1126	return 0;
1127	}
1128
1129	/**
1130	* Calculate the number of INTNETSEG segments the socket buffer will need.
1131	*
1132	* @returns Segment count.
1133	* @param pBuf The socket buffer.
1134	* @param pcbTemp Where to store the number of bytes of the part
1135	* of the socket buffer that will be copied to
1136	* a temporary storage.
1137	*/
1138	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff pBuf, unsigned pcbTemp)
1139	{
1140	*pcbTemp = 0;
1141	#ifdef VBOXNETFLT_SG_SUPPORT
1142	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1143	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1144	{
1145	*pcbTemp = vboxNetFltLinuxGetChecksumStartOffset(pBuf) + VBOX_SKB_CSUM_OFFSET(pBuf) + sizeof(uint16_t);
1146	}
1147	# if RTLNX_VER_MIN(2,6,27)
1148	if (vlan_tx_tag_present(pBuf))
1149	{
1150	if (*pcbTemp)
1151	*pcbTemp += VLAN_HLEN;
1152	else
1153	*pcbTemp = VLAN_ETH_HLEN;
1154	}
1155	# endif /* RTLNX_VER_MIN(2,6,27) */
1156	if (*pcbTemp)
1157	++cSegs;
1158	struct sk_buff *pFrag;
1159	for (pFrag = skb_shinfo(pBuf)->frag_list; pFrag; pFrag = pFrag->next)
1160	{
1161	Log6(("vboxNetFltLinuxCalcSGSegments: frag=%p len=%d data_len=%d frags=%d frag_list=%p next=%p\n",
1162	pFrag, pFrag->len, pFrag->data_len, skb_shinfo(pFrag)->nr_frags, skb_shinfo(pFrag)->frag_list, pFrag->next));
1163	cSegs += 1 + skb_shinfo(pFrag)->nr_frags;
1164	}
1165	#else
1166	unsigned cSegs = 1;
1167	#endif
1168	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1169	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1170	if (pBuf->len < 60)
1171	cSegs++;
1172	#endif
1173	return cSegs;
1174	}
1175
1176
1177	/**
1178	* Destroy the intnet scatter / gather buffer created by
1179	* vboxNetFltLinuxSkBufToSG.
1180	*
1181	* @param pSG The (scatter/)gather list.
1182	* @param pBuf The original socket buffer that was used to create
1183	* the scatter/gather list.
1184	*/
1185	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG, struct sk_buff *pBuf)
1186	{
1187	#ifdef VBOXNETFLT_SG_SUPPORT
1188	int i, iSeg = 1; /* Skip non-paged part of SKB */
1189	/* Check if the extra buffer behind SG structure was used for modified packet header */
1190	if (pBuf->data != pSG->aSegs[0].pv)
1191	++iSeg; /* Skip it as well */
1192	# ifdef LOG_ENABLED
1193	if (pBuf->data_len)
1194	Log6(("kunmap_atomic:"));
1195	# endif /* LOG_ENABLED */
1196	/* iSeg now points to the first mapped fragment if there are any */
1197	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1198	{
1199	Log6((" %p", pSG->aSegs[iSeg].pv));
1200	VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1201	}
1202	struct sk_buff *pFragBuf;
1203	for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
1204	{
1205	++iSeg; /* Non-fragment (unmapped) portion of chained SKB */
1206	for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
1207	{
1208	Log6((" %p", pSG->aSegs[iSeg].pv));
1209	VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1210	}
1211	}
1212	# ifdef LOG_ENABLED
1213	if (pBuf->data_len)
1214	Log6(("\n"));
1215	# endif /* LOG_ENABLED */
1216	#endif
1217	NOREF(pSG);
1218	}
1219
1220	#ifdef LOG_ENABLED
1221	/**
1222	* Logging helper.
1223	*/
1224	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1225	{
1226	int i, offSeg;
1227	uint8_t pInt, pExt;
1228	static int iPacketNo = 1;
1229	iPacketNo += iIncrement;
1230	if (fEgress)
1231	{
1232	pExt = pSG->aSegs[0].pv;
1233	pInt = pExt + 6;
1234	}
1235	else
1236	{
1237	pInt = pSG->aSegs[0].pv;
1238	pExt = pInt + 6;
1239	}
1240	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1241	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1242	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1243	fEgress ? "-->" : "<--", pszWhere,
1244	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1245	pSG->cbTotal, iPacketNo));
1246	if (pSG->cSegsUsed == 1)
1247	{
1248	Log4(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1249	}
1250	else
1251	{
1252	for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
1253	{
1254	Log4(("-- segment %d at 0x%x (%d bytes)\n --\n%.*Rhxd\n",
1255	i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1256	offSeg += pSG->aSegs[i].cb;
1257	}
1258	}
1259	}
1260	#else
1261	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1262	#endif
1263
1264	#ifdef VBOXNETFLT_WITH_GSO_RECV
1265
1266	/**
1267	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1268	* GSO socket buffer without having to segment it.
1269	*
1270	* @returns true on success, false if needs segmenting.
1271	* @param pThis The net filter instance.
1272	* @param pSkb The GSO socket buffer.
1273	* @param fSrc The source.
1274	* @param pGsoCtx Where to return the GSO context on success.
1275	*/
1276	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1277	PPDMNETWORKGSO pGsoCtx)
1278	{
1279	PDMNETWORKGSOTYPE enmGsoType;
1280	uint16_t uEtherType;
1281	unsigned int cbTransport;
1282	unsigned int offTransport;
1283	unsigned int cbTransportHdr;
1284	unsigned uProtocol;
1285	union
1286	{
1287	RTNETIPV4 IPv4;
1288	RTNETIPV6 IPv6;
1289	RTNETTCP Tcp;
1290	uint8_t ab[40];
1291	uint16_t au16[40/2];
1292	uint32_t au32[40/4];
1293	} Buf;
1294
1295	/*
1296	* Check the GSO properties of the socket buffer and make sure it fits.
1297	*/
1298	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1299	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
1300	{
1301	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1302	return false;
1303	}
1304	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1305	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
1306	{
1307	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1308	return false;
1309	}
1310
1311	/*
1312	* Switch on the ethertype.
1313	*/
1314	uEtherType = pSkb->protocol;
1315	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1316	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1317	{
1318	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1319	if (puEtherType)
1320	uEtherType = *puEtherType;
1321	}
1322	switch (uEtherType)
1323	{
1324	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1325	{
1326	unsigned int cbHdr;
1327	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1328	if (RT_UNLIKELY(!pIPv4))
1329	{
1330	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1331	return false;
1332	}
1333
1334	cbHdr = pIPv4->ip_hl * 4;
1335	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1336	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1337	\|\| cbHdr > cbTransport ))
1338	{
1339	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1340	return false;
1341	}
1342	cbTransport -= cbHdr;
1343	offTransport = pSkb->mac_len + cbHdr;
1344	uProtocol = pIPv4->ip_p;
1345	if (uProtocol == RTNETIPV4_PROT_TCP)
1346	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1347	else if (uProtocol == RTNETIPV4_PROT_UDP)
1348	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1349	else /** @todo IPv6: 4to6 tunneling */
1350	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1351	break;
1352	}
1353
1354	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1355	{
1356	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1357	if (RT_UNLIKELY(!pIPv6))
1358	{
1359	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1360	return false;
1361	}
1362
1363	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1364	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1365	uProtocol = pIPv6->ip6_nxt;
1366	/** @todo IPv6: Dig our way out of the other headers. */
1367	if (uProtocol == RTNETIPV4_PROT_TCP)
1368	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1369	else if (uProtocol == RTNETIPV4_PROT_UDP)
1370	enmGsoType = PDMNETWORKGSOTYPE_IPV6_UDP;
1371	else
1372	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1373	break;
1374	}
1375
1376	default:
1377	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1378	return false;
1379	}
1380
1381	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1382	{
1383	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1384	return false;
1385	}
1386
1387	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1388	\|\| offTransport + cbTransport > pSkb->len
1389	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1390	{
1391	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1392	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1393	return false;
1394	}
1395
1396	/*
1397	* Check the TCP/UDP bits.
1398	*/
1399	if (uProtocol == RTNETIPV4_PROT_TCP)
1400	{
1401	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1402	if (RT_UNLIKELY(!pTcp))
1403	{
1404	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1405	return false;
1406	}
1407
1408	cbTransportHdr = pTcp->th_off * 4;
1409	pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1410	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1411	\|\| cbTransportHdr > cbTransport
1412	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1413	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1414	{
1415	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1416	return false;
1417	}
1418
1419	}
1420	else
1421	{
1422	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1423	cbTransportHdr = sizeof(RTNETUDP);
1424	pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1425	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1426	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1427	{
1428	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1429	return false;
1430	}
1431	}
1432
1433	/*
1434	* We're good, init the GSO context.
1435	*/
1436	pGsoCtx->u8Type = enmGsoType;
1437	pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1438	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1439	pGsoCtx->offHdr1 = pSkb->mac_len;
1440	pGsoCtx->offHdr2 = offTransport;
1441	pGsoCtx->u8Unused = 0;
1442
1443	return true;
1444	}
1445
1446	/**
1447	* Forward the socket buffer as a GSO internal network frame.
1448	*
1449	* @returns IPRT status code.
1450	* @param pThis The net filter instance.
1451	* @param pSkb The GSO socket buffer.
1452	* @param fSrc The source.
1453	* @param pGsoCtx Where to return the GSO context on success.
1454	*/
1455	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1456	{
1457	int rc;
1458	unsigned cbExtra;
1459	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb, &cbExtra);
1460	PINTNETSG pSG = (PINTNETSG)alloca(RT_UOFFSETOF_DYN(INTNETSG, aSegs[cSegs]) + cbExtra);
1461	if (RT_LIKELY(pSG))
1462	{
1463	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cbExtra, cSegs, fSrc, pGsoCtx);
1464
1465	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1466	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1467
1468	vboxNetFltLinuxDestroySG(pSG, pSkb);
1469	rc = VINF_SUCCESS;
1470	}
1471	else
1472	{
1473	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1474	rc = VERR_NO_MEMORY;
1475	}
1476	return rc;
1477	}
1478
1479	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1480
1481	/**
1482	* Worker for vboxNetFltLinuxForwardToIntNet.
1483	*
1484	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1485	* @param pThis The net filter instance.
1486	* @param pBuf The socket buffer.
1487	* @param fSrc The source.
1488	*/
1489	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1490	{
1491	int rc;
1492	unsigned cbExtra;
1493	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf, &cbExtra);
1494	PINTNETSG pSG = (PINTNETSG)alloca(RT_UOFFSETOF_DYN(INTNETSG, aSegs[cSegs]) + cbExtra);
1495	if (RT_LIKELY(pSG))
1496	{
1497	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cbExtra, cSegs, fSrc, NULL /pGsoCtx/);
1498
1499	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1500	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1501
1502	vboxNetFltLinuxDestroySG(pSG, pBuf);
1503	rc = VINF_SUCCESS;
1504	}
1505	else
1506	{
1507	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1508	rc = VERR_NO_MEMORY;
1509	}
1510	return rc;
1511	}
1512
1513
1514	/**
1515	* I won't disclose what I do, figure it out yourself, including pThis referencing.
1516	*
1517	* @param pThis The net filter instance.
1518	* @param pBuf The socket buffer.
1519	* @param fSrc Where the packet comes from.
1520	*/
1521	static void vboxNetFltLinuxForwardToIntNetInner(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1522	{
1523	#ifdef VBOXNETFLT_WITH_GSO
1524	if (skb_is_gso(pBuf))
1525	{
1526	PDMNETWORKGSO GsoCtx;
1527	Log6(("vboxNetFltLinuxForwardToIntNetInner: skb len=%u data_len=%u truesize=%u next=%p"
1528	" nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x ip_summed=%d\n",
1529	pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next,
1530	skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->gso_size,
1531	skb_shinfo(pBuf)->gso_segs, skb_shinfo(pBuf)->gso_type,
1532	skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, pBuf->ip_summed));
1533	#ifndef VBOXNETFLT_SG_SUPPORT
1534	if (RT_LIKELY(fSrc & INTNETTRUNKDIR_HOST))
1535	{
1536	/*
1537	* skb_gso_segment does the following. Do we need to do it as well?
1538	*/
1539	# if RTLNX_VER_MIN(2,6,22)
1540	skb_reset_mac_header(pBuf);
1541	pBuf->mac_len = pBuf->network_header - pBuf->mac_header;
1542	# else
1543	pBuf->mac.raw = pBuf->data;
1544	pBuf->mac_len = pBuf->nh.raw - pBuf->data;
1545	# endif
1546	}
1547	#endif /* !VBOXNETFLT_SG_SUPPORT */
1548	# ifdef VBOXNETFLT_WITH_GSO_RECV
1549	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1550	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1551	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1552	else
1553	# endif /* VBOXNETFLT_WITH_GSO_RECV */
1554	{
1555	/* Need to segment the packet */
1556	struct sk_buff *pNext;
1557	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1558	if (IS_ERR(pSegment))
1559	{
1560	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1561	return;
1562	}
1563
1564	for (; pSegment; pSegment = pNext)
1565	{
1566	Log6(("vboxNetFltLinuxForwardToIntNetInner: segment len=%u data_len=%u truesize=%u next=%p"
1567	" nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x\n",
1568	pSegment->len, pSegment->data_len, pSegment->truesize, pSegment->next,
1569	skb_shinfo(pSegment)->nr_frags, skb_shinfo(pSegment)->gso_size,
1570	skb_shinfo(pSegment)->gso_segs, skb_shinfo(pSegment)->gso_type,
1571	skb_shinfo(pSegment)->frag_list, pSegment->pkt_type));
1572	pNext = pSegment->next;
1573	pSegment->next = 0;
1574	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1575	dev_kfree_skb(pSegment);
1576	}
1577	}
1578	}
1579	else
1580	#endif /* VBOXNETFLT_WITH_GSO */
1581	{
1582	Log6(("vboxNetFltLinuxForwardToIntNetInner: ptk_type=%d ip_summed=%d len=%d"
1583	" data_len=%d headroom=%d hdr_len=%d csum_offset=%d\n",
1584	pBuf->pkt_type, pBuf->ip_summed, pBuf->len, pBuf->data_len, skb_headroom(pBuf),
1585	skb_headlen(pBuf), vboxNetFltLinuxGetChecksumStartOffset(pBuf)));
1586	#ifndef VBOXNETFLT_SG_SUPPORT
1587	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1588	{
1589	# if RTLNX_VER_MIN(2,6,19)
1590	int rc = VBOX_SKB_CHECKSUM_HELP(pBuf);
1591	# else
1592	/*
1593	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1594	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1595	* header length from the header itself and reconstruct 'h' pointer
1596	* to TCP (or whatever) header.
1597	*/
1598	unsigned char *tmp = pBuf->h.raw;
1599	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1600	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1601	int rc = VBOX_SKB_CHECKSUM_HELP(pBuf);
1602	/* Restore the original (wrong) pointer. */
1603	pBuf->h.raw = tmp;
1604	# endif
1605	if (rc)
1606	{
1607	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1608	return;
1609	}
1610	}
1611	#endif /* !VBOXNETFLT_SG_SUPPORT */
1612	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1613	}
1614	}
1615
1616
1617	/**
1618	* Temporarily adjust pBuf->data so it always points to the Ethernet header,
1619	* then forward it to the internal network.
1620	*
1621	* @param pThis The net filter instance.
1622	* @param pBuf The socket buffer. This is consumed by this function.
1623	*/
1624	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1625	{
1626	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1627
1628	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1629	{
1630	/*
1631	* The packet came from the wire and the driver has already consumed
1632	* mac header. We need to restore it back. Moreover, after we are
1633	* through with this skb we need to restore its original state!
1634	*/
1635	skb_push(pBuf, pBuf->mac_len);
1636	Log5(("vboxNetFltLinuxForwardToIntNet: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1637	pBuf->mac_len, pBuf->data, skb_mac_header(pBuf), skb_network_header(pBuf)));
1638	}
1639
1640	vboxNetFltLinuxForwardToIntNetInner(pThis, pBuf, fSrc);
1641
1642	/*
1643	* Restore the original state of skb as there are other handlers this skb
1644	* will be provided to.
1645	*/
1646	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1647	skb_pull(pBuf, pBuf->mac_len);
1648
1649	dev_kfree_skb(pBuf);
1650	}
1651
1652
1653	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1654	/**
1655	* Work queue handler that forwards the socket buffers queued by
1656	* vboxNetFltLinuxPacketHandler to the internal network.
1657	*
1658	* @param pWork The work queue.
1659	*/
1660	# if RTLNX_VER_MIN(2,6,20)
1661	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1662	# else
1663	static void vboxNetFltLinuxXmitTask(void *pWork)
1664	# endif
1665	{
1666	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1667	struct sk_buff *pBuf;
1668
1669	Log6(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1670
1671	/*
1672	* Active? Retain the instance and increment the busy counter.
1673	*/
1674	if (vboxNetFltTryRetainBusyActive(pThis))
1675	{
1676	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1677	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1678
1679	vboxNetFltRelease(pThis, true /* fBusy */);
1680	}
1681	else
1682	{
1683	/** @todo Shouldn't we just drop the packets here? There is little point in
1684	* making them accumulate when the VM is paused and it'll only waste
1685	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1686	* before start draining the packets (goes for the intnet ring buf
1687	* too)? */
1688	}
1689	}
1690	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1691
1692	/**
1693	* Reports the GSO capabilities of the hardware NIC.
1694	*
1695	* @param pThis The net filter instance. The caller hold a
1696	* reference to this.
1697	*/
1698	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1699	{
1700	#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1701	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1702	{
1703	struct net_device *pDev;
1704	PINTNETTRUNKSWPORT pSwitchPort;
1705	unsigned int fFeatures;
1706
1707	RTSpinlockAcquire(pThis->hSpinlock);
1708
1709	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1710	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1711	if (pDev)
1712	fFeatures = pDev->features;
1713	else
1714	fFeatures = 0;
1715
1716	RTSpinlockRelease(pThis->hSpinlock);
1717
1718	if (pThis->pSwitchPort)
1719	{
1720	/* Set/update the GSO capabilities of the NIC. */
1721	uint32_t fGsoCapabilites = 0;
1722	if (fFeatures & NETIF_F_TSO)
1723	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1724	if (fFeatures & NETIF_F_TSO6)
1725	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1726	Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1727	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1728	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : ""));
1729	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1730	}
1731
1732	vboxNetFltRelease(pThis, true /fBusy/);
1733	}
1734	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1735	}
1736
1737	/**
1738	* Helper that determines whether the host (ignoreing us) is operating the
1739	* interface in promiscuous mode or not.
1740	*/
1741	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1742	{
1743	bool fRc = false;
1744	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1745	if (pDev)
1746	{
1747	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1748	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1749	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1750	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1751	}
1752	return fRc;
1753	}
1754
1755	/**
1756	* Does this device needs link state change signaled?
1757	* Currently we need it for our own VBoxNetAdp and TAP.
1758	*/
1759	static bool vboxNetFltNeedsLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev)
1760	{
1761	if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1762	{
1763	struct ethtool_drvinfo Info;
1764
1765	memset(&Info, 0, sizeof(Info));
1766	Info.cmd = ETHTOOL_GDRVINFO;
1767	pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1768	Log3(("%s: driver=%.s version=%.s bus_info=%.*s\n",
1769	__FUNCTION__,
1770	sizeof(Info.driver), Info.driver,
1771	sizeof(Info.version), Info.version,
1772	sizeof(Info.bus_info), Info.bus_info));
1773
1774	if (!strncmp(Info.driver, "vboxnet", sizeof(Info.driver)))
1775	return true;
1776
1777	#if RTLNX_VER_MIN(2,6,36) /* TAP started doing carrier */
1778	return !strncmp(Info.driver, "tun", 4)
1779	&& !strncmp(Info.bus_info, "tap", 4);
1780	#endif
1781	}
1782
1783	return false;
1784	}
1785
1786	#if RTLNX_VER_MAX(2,6,18)
1787	DECLINLINE(void) netif_tx_lock_bh(struct net_device *pDev)
1788	{
1789	spin_lock_bh(&pDev->xmit_lock);
1790	}
1791
1792	DECLINLINE(void) netif_tx_unlock_bh(struct net_device *pDev)
1793	{
1794	spin_unlock_bh(&pDev->xmit_lock);
1795	}
1796	#endif
1797
1798	/**
1799	* Some devices need link state change when filter attaches/detaches
1800	* since the filter is their link in a sense.
1801	*/
1802	static void vboxNetFltSetLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1803	{
1804	if (vboxNetFltNeedsLinkState(pThis, pDev))
1805	{
1806	Log3(("%s: bringing device link %s\n",
1807	__FUNCTION__, fLinkUp ? "up" : "down"));
1808	netif_tx_lock_bh(pDev);
1809	if (fLinkUp)
1810	netif_carrier_on(pDev);
1811	else
1812	netif_carrier_off(pDev);
1813	netif_tx_unlock_bh(pDev);
1814	}
1815	}
1816
1817	/**
1818	* Internal worker for vboxNetFltLinuxNotifierCallback.
1819	*
1820	* @returns VBox status code.
1821	* @param pThis The instance.
1822	* @param pDev The device to attach to.
1823	*/
1824	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1825	{
1826	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1827
1828	/*
1829	* Retain and store the device.
1830	*/
1831	dev_hold(pDev);
1832
1833	RTSpinlockAcquire(pThis->hSpinlock);
1834	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1835	RTSpinlockRelease(pThis->hSpinlock);
1836
1837	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1838	pDev, pDev->name,
1839	#if RTLNX_VER_MIN(2,6,37)
1840	netdev_refcnt_read(pDev)
1841	#else
1842	atomic_read(&pDev->refcnt)
1843	#endif
1844	));
1845	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1846	pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1847
1848	/* Get the mac address while we still have a valid net_device reference. */
1849	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1850	/* Initialize MTU */
1851	pThis->u.s.cbMtu = pDev->mtu;
1852
1853	/*
1854	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1855	*/
1856	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1857	pThis->u.s.PacketType.dev = pDev;
1858	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1859	dev_add_pack(&pThis->u.s.PacketType);
1860	ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1861	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1862
1863	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1864	vboxNetFltLinuxHookDev(pThis, pDev);
1865	#endif
1866
1867	/*
1868	* Are we the "carrier" for this device (e.g. vboxnet or tap)?
1869	*/
1870	vboxNetFltSetLinkState(pThis, pDev, true);
1871
1872	/*
1873	* Set indicators that require the spinlock. Be abit paranoid about racing
1874	* the device notification handle.
1875	*/
1876	RTSpinlockAcquire(pThis->hSpinlock);
1877	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1878	if (pDev)
1879	{
1880	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1881	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1882	pDev = NULL; /* don't dereference it */
1883	}
1884	RTSpinlockRelease(pThis->hSpinlock);
1885
1886	/*
1887	* Report GSO capabilities
1888	*/
1889	Assert(pThis->pSwitchPort);
1890	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1891	{
1892	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1893	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1894	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1895	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1896	vboxNetFltRelease(pThis, true /fBusy/);
1897	}
1898
1899	LogRel(("VBoxNetFlt: attached to '%s' / %RTmac\n", pThis->szName, &pThis->u.s.MacAddr));
1900	return VINF_SUCCESS;
1901	}
1902
1903
1904	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1905	{
1906	bool fRegistered;
1907	Assert(!pThis->fDisconnectedFromHost);
1908
1909	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1910	vboxNetFltLinuxUnhookDev(pThis, pDev);
1911	#endif
1912
1913	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1914	{
1915	dev_remove_pack(&pThis->u.s.PacketType);
1916	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1917	}
1918
1919	RTSpinlockAcquire(pThis->hSpinlock);
1920	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1921	if (fRegistered)
1922	{
1923	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1924	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1925	}
1926	RTSpinlockRelease(pThis->hSpinlock);
1927
1928	if (fRegistered)
1929	{
1930	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1931	skb_queue_purge(&pThis->u.s.XmitQueue);
1932	#endif
1933	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1934	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1935	pDev, pDev->name,
1936	#if RTLNX_VER_MIN(2,6,37)
1937	netdev_refcnt_read(pDev)
1938	#else
1939	atomic_read(&pDev->refcnt)
1940	#endif
1941	));
1942	dev_put(pDev);
1943	}
1944
1945	return NOTIFY_OK;
1946	}
1947
1948	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1949	{
1950	/* Check if we are not suspended and promiscuous mode has not been set. */
1951	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1952	&& !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1953	{
1954	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1955	dev_set_promiscuity(pDev, 1);
1956	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1957	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1958	}
1959	else
1960	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1961	return NOTIFY_OK;
1962	}
1963
1964	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1965	{
1966	/* Undo promiscuous mode if we has set it. */
1967	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1968	{
1969	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1970	dev_set_promiscuity(pDev, -1);
1971	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1972	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1973	}
1974	else
1975	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1976	return NOTIFY_OK;
1977	}
1978
1979	/**
1980	* Callback for listening to MTU change event.
1981	*
1982	* We need to track changes of host's inteface MTU to discard over-sized frames
1983	* coming from the internal network as they may hang the TX queue of host's
1984	* adapter.
1985	*
1986	* @returns NOTIFY_OK
1987	* @param pThis The netfilter instance.
1988	* @param pDev Pointer to device structure of host's interface.
1989	*/
1990	static int vboxNetFltLinuxDeviceMtuChange(PVBOXNETFLTINS pThis, struct net_device *pDev)
1991	{
1992	ASMAtomicWriteU32(&pThis->u.s.cbMtu, pDev->mtu);
1993	Log(("vboxNetFltLinuxDeviceMtuChange: set MTU for %s to %d\n", pThis->szName, pDev->mtu));
1994	return NOTIFY_OK;
1995	}
1996
1997	#ifdef LOG_ENABLED
1998	/** Stringify the NETDEV_XXX constants. */
1999	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2000	{
2001	const char *pszEvent = "NETDEV_<unknown>";
2002	switch (ulEventType)
2003	{
2004	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2005	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2006	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2007	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2008	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2009	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2010	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2011	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2012	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2013	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2014	# ifdef NETDEV_FEAT_CHANGE
2015	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2016	# endif
2017	}
2018	return pszEvent;
2019	}
2020	#endif /* LOG_ENABLED */
2021
2022	/**
2023	* Callback for listening to netdevice events.
2024	*
2025	* This works the rediscovery, clean up on unregistration, promiscuity on
2026	* up/down, and GSO feature changes from ethtool.
2027	*
2028	* @returns NOTIFY_OK
2029	* @param self Pointer to our notifier registration block.
2030	* @param ulEventType The event.
2031	* @param ptr Event specific, but it is usually the device it
2032	* relates to.
2033	*/
2034	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2035
2036	{
2037	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2038	struct net_device pMyDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device );
2039	struct net_device *pDev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2040	int rc = NOTIFY_OK;
2041
2042	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2043	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, pMyDev));
2044
2045	if (ulEventType == NETDEV_REGISTER)
2046	{
2047	#if RTLNX_VER_MIN(2,6,24) /* cgroups/namespaces introduced */
2048	# if RTLNX_VER_MIN(2,6,26)
2049	# define VBOX_DEV_NET(dev) dev_net(dev)
2050	# define VBOX_NET_EQ(n1, n2) net_eq((n1), (n2))
2051	# else
2052	# define VBOX_DEV_NET(dev) ((dev)->nd_net)
2053	# define VBOX_NET_EQ(n1, n2) ((n1) == (n2))
2054	# endif
2055	struct net *pMyNet = current->nsproxy->net_ns;
2056	struct net *pDevNet = VBOX_DEV_NET(pDev);
2057
2058	if (VBOX_NET_EQ(pDevNet, pMyNet))
2059	#endif /* namespaces */
2060	{
2061	if (strcmp(pDev->name, pThis->szName) == 0)
2062	{
2063	vboxNetFltLinuxAttachToInterface(pThis, pDev);
2064	}
2065	}
2066	}
2067	else
2068	{
2069	if (pDev == pMyDev)
2070	{
2071	switch (ulEventType)
2072	{
2073	case NETDEV_UNREGISTER:
2074	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2075	break;
2076	case NETDEV_UP:
2077	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2078	break;
2079	case NETDEV_GOING_DOWN:
2080	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2081	break;
2082	case NETDEV_CHANGEMTU:
2083	rc = vboxNetFltLinuxDeviceMtuChange(pThis, pDev);
2084	break;
2085	case NETDEV_CHANGENAME:
2086	break;
2087	#ifdef NETDEV_FEAT_CHANGE
2088	case NETDEV_FEAT_CHANGE:
2089	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2090	break;
2091	#endif
2092	}
2093	}
2094	}
2095
2096	return rc;
2097	}
2098
2099	/*
2100	* Initial enumeration of netdevs. Called with NETDEV_REGISTER by
2101	* register_netdevice_notifier() under rtnl lock.
2102	*/
2103	static int vboxNetFltLinuxEnumeratorCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2104	{
2105	PVBOXNETFLTINS pThis = ((PVBOXNETFLTNOTIFIER)self)->pThis;
2106	struct net_device *dev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2107	struct in_device *in_dev;
2108	struct inet6_dev *in6_dev;
2109
2110	if (ulEventType != NETDEV_REGISTER)
2111	return NOTIFY_OK;
2112
2113	if (RT_UNLIKELY(pThis->pSwitchPort->pfnNotifyHostAddress == NULL))
2114	return NOTIFY_OK;
2115
2116	/*
2117	* IPv4
2118	*/
2119	#if RTLNX_VER_MIN(2,6,14)
2120	in_dev = __in_dev_get_rtnl(dev);
2121	#else
2122	in_dev = __in_dev_get(dev);
2123	#endif
2124	if (in_dev != NULL)
2125	{
2126	struct in_ifaddr *ifa;
2127
2128	for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
2129	if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2130	return NOTIFY_OK;
2131
2132	if ( dev != pThis->u.s.pDev
2133	&& VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2134	continue;
2135
2136	Log(("%s: %s: IPv4 addr %RTnaipv4 mask %RTnaipv4\n",
2137	__FUNCTION__, VBOX_NETDEV_NAME(dev),
2138	ifa->ifa_address, ifa->ifa_mask));
2139
2140	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2141	/* :fAdded */ true, kIntNetAddrType_IPv4, &ifa->ifa_address);
2142	}
2143	}
2144
2145	/*
2146	* IPv6
2147	*/
2148	in6_dev = __in6_dev_get(dev);
2149	if (in6_dev != NULL)
2150	{
2151	struct inet6_ifaddr *ifa;
2152
2153	read_lock_bh(&in6_dev->lock);
2154	#if RTLNX_VER_MIN(2,6,35)
2155	list_for_each_entry(ifa, &in6_dev->addr_list, if_list)
2156	#else
2157	for (ifa = in6_dev->addr_list; ifa != NULL; ifa = ifa->if_next)
2158	#endif
2159	{
2160	if ( dev != pThis->u.s.pDev
2161	&& ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL \| IPV6_ADDR_LOOPBACK))
2162	continue;
2163
2164	Log(("%s: %s: IPv6 addr %RTnaipv6/%u\n",
2165	__FUNCTION__, VBOX_NETDEV_NAME(dev),
2166	&ifa->addr, (unsigned)ifa->prefix_len));
2167
2168	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2169	/* :fAdded */ true, kIntNetAddrType_IPv6, &ifa->addr);
2170	}
2171	read_unlock_bh(&in6_dev->lock);
2172	}
2173
2174	return NOTIFY_OK;
2175	}
2176
2177
2178	static int vboxNetFltLinuxNotifierIPv4Callback(struct notifier_block self, unsigned long ulEventType, void ptr)
2179	{
2180	PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv4);
2181	struct net_device pDev, pEventDev;
2182	struct in_ifaddr ifa = (struct in_ifaddr )ptr;
2183	bool fMyDev;
2184	int rc = NOTIFY_OK;
2185
2186	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2187	pEventDev = ifa->ifa_dev->dev;
2188	fMyDev = (pDev == pEventDev);
2189	Log(("VBoxNetFlt: %s: IPv4 event %s(0x%lx) %s: addr %RTnaipv4 mask %RTnaipv4\n",
2190	pDev ? VBOX_NETDEV_NAME(pDev) : "<unknown>",
2191	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2192	pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<unknown>",
2193	ifa->ifa_address, ifa->ifa_mask));
2194
2195	if (pDev != NULL)
2196	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2197
2198	if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2199	return NOTIFY_OK;
2200
2201	if ( !fMyDev
2202	&& VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2203	return NOTIFY_OK;
2204
2205	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2206	{
2207	bool fAdded;
2208	if (ulEventType == NETDEV_UP)
2209	fAdded = true;
2210	else if (ulEventType == NETDEV_DOWN)
2211	fAdded = false;
2212	else
2213	return NOTIFY_OK;
2214
2215	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2216	kIntNetAddrType_IPv4, &ifa->ifa_local);
2217	}
2218
2219	return rc;
2220	}
2221
2222
2223	static int vboxNetFltLinuxNotifierIPv6Callback(struct notifier_block self, unsigned long ulEventType, void ptr)
2224	{
2225	PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv6);
2226	struct net_device pDev, pEventDev;
2227	struct inet6_ifaddr ifa = (struct inet6_ifaddr )ptr;
2228	bool fMyDev;
2229	int rc = NOTIFY_OK;
2230
2231	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2232	pEventDev = ifa->idev->dev;
2233	fMyDev = (pDev == pEventDev);
2234	Log(("VBoxNetFlt: %s: IPv6 event %s(0x%lx) %s: %RTnaipv6\n",
2235	pDev ? VBOX_NETDEV_NAME(pDev) : "<unknown>",
2236	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2237	pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<unknown>",
2238	&ifa->addr));
2239
2240	if (pDev != NULL)
2241	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2242
2243	if ( !fMyDev
2244	&& ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL \| IPV6_ADDR_LOOPBACK))
2245	return NOTIFY_OK;
2246
2247	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2248	{
2249	bool fAdded;
2250	if (ulEventType == NETDEV_UP)
2251	fAdded = true;
2252	else if (ulEventType == NETDEV_DOWN)
2253	fAdded = false;
2254	else
2255	return NOTIFY_OK;
2256
2257	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2258	kIntNetAddrType_IPv6, &ifa->addr);
2259	}
2260
2261	return rc;
2262	}
2263
2264
2265	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2266	{
2267	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2268	}
2269
2270	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2271	{
2272	struct net_device * pDev;
2273	int err;
2274	int rc = VINF_SUCCESS;
2275	IPRT_LINUX_SAVE_EFL_AC();
2276	NOREF(pvIfData);
2277
2278	LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2279
2280	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2281	if (pDev)
2282	{
2283	/*
2284	* Create a sk_buff for the gather list and push it onto the wire.
2285	*/
2286	if (fDst & INTNETTRUNKDIR_WIRE)
2287	{
2288	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2289	if (pBuf)
2290	{
2291	vboxNetFltDumpPacket(pSG, true, "wire", 1);
2292	Log6(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2293	Log6(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2294	err = dev_queue_xmit(pBuf);
2295	if (err)
2296	rc = RTErrConvertFromErrno(err);
2297	}
2298	else
2299	rc = VERR_NO_MEMORY;
2300	}
2301
2302	/*
2303	* Create a sk_buff for the gather list and push it onto the host stack.
2304	*/
2305	if (fDst & INTNETTRUNKDIR_HOST)
2306	{
2307	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2308	if (pBuf)
2309	{
2310	vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2311	Log6(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2312	Log6(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2313	err = netif_rx_ni(pBuf);
2314	if (err)
2315	rc = RTErrConvertFromErrno(err);
2316	}
2317	else
2318	rc = VERR_NO_MEMORY;
2319	}
2320
2321	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2322	}
2323
2324	IPRT_LINUX_RESTORE_EFL_AC();
2325	return rc;
2326	}
2327
2328
2329	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2330	{
2331	struct net_device *pDev;
2332	IPRT_LINUX_SAVE_EFL_AC();
2333
2334	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%RTbool, fDisablePromiscuous=%RTbool\n",
2335	pThis, pThis->szName, fActive, pThis->fDisablePromiscuous));
2336
2337	if (pThis->fDisablePromiscuous)
2338	return;
2339
2340	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2341	if (pDev)
2342	{
2343	/*
2344	* This api is a bit weird, the best reference is the code.
2345	*
2346	* Also, we have a bit or race conditions wrt the maintenance of
2347	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2348	*/
2349	#ifdef LOG_ENABLED
2350	u_int16_t fIf;
2351	unsigned const cPromiscBefore = pDev->promiscuity;
2352	#endif
2353	if (fActive)
2354	{
2355	Assert(!pThis->u.s.fPromiscuousSet);
2356
2357	rtnl_lock();
2358	dev_set_promiscuity(pDev, 1);
2359	rtnl_unlock();
2360	pThis->u.s.fPromiscuousSet = true;
2361	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2362	}
2363	else
2364	{
2365	if (pThis->u.s.fPromiscuousSet)
2366	{
2367	rtnl_lock();
2368	dev_set_promiscuity(pDev, -1);
2369	rtnl_unlock();
2370	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2371	}
2372	pThis->u.s.fPromiscuousSet = false;
2373
2374	#ifdef LOG_ENABLED
2375	fIf = dev_get_flags(pDev);
2376	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2377	#endif
2378	}
2379
2380	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2381	}
2382	IPRT_LINUX_RESTORE_EFL_AC();
2383	}
2384
2385
2386	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2387	{
2388	/*
2389	* Remove packet handler when we get disconnected from internal switch as
2390	* we don't want the handler to forward packets to disconnected switch.
2391	*/
2392	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
2393	{
2394	IPRT_LINUX_SAVE_EFL_AC();
2395	dev_remove_pack(&pThis->u.s.PacketType);
2396	Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
2397	IPRT_LINUX_RESTORE_EFL_AC();
2398	}
2399	return VINF_SUCCESS;
2400	}
2401
2402
2403	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2404	{
2405	IPRT_LINUX_SAVE_EFL_AC();
2406
2407	/*
2408	* Report the GSO capabilities of the host and device (if connected).
2409	* Note! No need to mark ourselves busy here.
2410	*/
2411	/** @todo duplicate work here now? Attach */
2412	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2413	Log3(("vboxNetFltOsConnectIt: reporting host tso tso6\n"));
2414	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2415	0
2416	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2417	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2418	, INTNETTRUNKDIR_HOST);
2419
2420	#endif
2421	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2422
2423	IPRT_LINUX_RESTORE_EFL_AC();
2424	return VINF_SUCCESS;
2425	}
2426
2427
2428	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2429	{
2430	struct net_device *pDev;
2431	bool fRegistered;
2432	IPRT_LINUX_SAVE_EFL_AC();
2433
2434	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
2435	vboxNetFltLinuxUnhookDev(pThis, NULL);
2436	#endif
2437
2438	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2439	* unlikely, but none the less). Since it doesn't actually update the
2440	* state (just reads it), it is likely to panic in some interesting
2441	* ways. */
2442
2443	RTSpinlockAcquire(pThis->hSpinlock);
2444	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2445	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
2446	RTSpinlockRelease(pThis->hSpinlock);
2447
2448	if (fRegistered)
2449	{
2450	vboxNetFltSetLinkState(pThis, pDev, false);
2451
2452	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2453	skb_queue_purge(&pThis->u.s.XmitQueue);
2454	#endif
2455	Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
2456	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
2457	pDev, pDev->name,
2458	#if RTLNX_VER_MIN(2,6,37)
2459	netdev_refcnt_read(pDev)
2460	#else
2461	atomic_read(&pDev->refcnt)
2462	#endif
2463	));
2464	dev_put(pDev);
2465	}
2466
2467	unregister_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2468	unregister_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2469
2470	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2471	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2472	module_put(THIS_MODULE);
2473
2474	IPRT_LINUX_RESTORE_EFL_AC();
2475	}
2476
2477
2478	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2479	{
2480	int err;
2481	IPRT_LINUX_SAVE_EFL_AC();
2482	NOREF(pvContext);
2483
2484	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2485	err = register_netdevice_notifier(&pThis->u.s.Notifier);
2486	if (err)
2487	{
2488	IPRT_LINUX_RESTORE_EFL_AC();
2489	return VERR_INTNET_FLT_IF_FAILED;
2490	}
2491	if (!pThis->u.s.fRegistered)
2492	{
2493	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2494	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2495	IPRT_LINUX_RESTORE_EFL_AC();
2496	return VERR_INTNET_FLT_IF_NOT_FOUND;
2497	}
2498
2499	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2500	if ( pThis->fDisconnectedFromHost
2501	\|\| !try_module_get(THIS_MODULE))
2502	{
2503	IPRT_LINUX_RESTORE_EFL_AC();
2504	return VERR_INTNET_FLT_IF_FAILED;
2505	}
2506
2507	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2508	{
2509	VBOXNETFLTNOTIFIER Enumerator;
2510
2511	/*
2512	* register_inetaddr_notifier() and register_inet6addr_notifier()
2513	* do not call the callback for existing devices. Enumerating
2514	* all network devices explicitly is a bit of an ifdef mess,
2515	* so co-opt register_netdevice_notifier() to do that for us.
2516	*/
2517	RT_ZERO(Enumerator);
2518	Enumerator.Notifier.notifier_call = vboxNetFltLinuxEnumeratorCallback;
2519	Enumerator.pThis = pThis;
2520
2521	err = register_netdevice_notifier(&Enumerator.Notifier);
2522	if (err)
2523	{
2524	LogRel(("%s: failed to enumerate network devices: error %d\n", __FUNCTION__, err));
2525	IPRT_LINUX_RESTORE_EFL_AC();
2526	return VINF_SUCCESS;
2527	}
2528
2529	unregister_netdevice_notifier(&Enumerator.Notifier);
2530
2531	pThis->u.s.NotifierIPv4.notifier_call = vboxNetFltLinuxNotifierIPv4Callback;
2532	err = register_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2533	if (err)
2534	LogRel(("%s: failed to register IPv4 notifier: error %d\n", __FUNCTION__, err));
2535
2536	pThis->u.s.NotifierIPv6.notifier_call = vboxNetFltLinuxNotifierIPv6Callback;
2537	err = register_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2538	if (err)
2539	LogRel(("%s: failed to register IPv6 notifier: error %d\n", __FUNCTION__, err));
2540	}
2541
2542	IPRT_LINUX_RESTORE_EFL_AC();
2543	return VINF_SUCCESS;
2544	}
2545
2546	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2547	{
2548	IPRT_LINUX_SAVE_EFL_AC();
2549
2550	/*
2551	* Init the linux specific members.
2552	*/
2553	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2554	pThis->u.s.fRegistered = false;
2555	pThis->u.s.fPromiscuousSet = false;
2556	pThis->u.s.fPacketHandler = false;
2557	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2558	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2559	skb_queue_head_init(&pThis->u.s.XmitQueue);
2560	# if RTLNX_VER_MIN(2,6,20)
2561	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2562	# else
2563	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2564	# endif
2565	#endif
2566
2567	IPRT_LINUX_RESTORE_EFL_AC();
2568	return VINF_SUCCESS;
2569	}
2570
2571
2572	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2573	{
2574	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2575	}
2576
2577
2578	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
2579	{
2580	/* Nothing to do */
2581	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2582	return VINF_SUCCESS;
2583	}
2584
2585
2586	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2587	{
2588	/* Nothing to do */
2589	NOREF(pThis); NOREF(pvIfData);
2590	return VINF_SUCCESS;
2591	}
2592

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 86001

Download in other formats: