VBoxNetFlt-linux.c@ 69496

Last change on this file since 69496 was 69496, checked in by vboxsync, 7 years ago
*: scm --update-copyright-year
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 91.9 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 69496 2017-10-28 14:55:58Z vboxsync $ */
2	/** @file
3	* VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2017 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 LINUX_VERSION_CODE >= KERNEL_VERSION(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 LINUX_VERSION_CODE >= KERNEL_VERSION(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 LINUX_VERSION_CODE >= KERNEL_VERSION(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 LINUX_VERSION_CODE >= KERNEL_VERSION(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 LINUX_VERSION_CODE >= KERNEL_VERSION(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 LINUX_VERSION_CODE >= KERNEL_VERSION(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 LINUX_VERSION_CODE >= KERNEL_VERSION(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 LINUX_VERSION_CODE >= KERNEL_VERSION(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 LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
136	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
137	#else
138	# define CHECKSUM_PARTIAL CHECKSUM_HW
139	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
140	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
141	# else
142	# if LINUX_VERSION_CODE >= KERNEL_VERSION(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 LINUX_VERSION_CODE >= KERNEL_VERSION(3, 20, 0)
154	# define VBOX_HAVE_SKB_VLAN
155	#else
156	# ifdef RHEL_RELEASE_CODE
157	# if (RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(7, 2) && RHEL_RELEASE_CODE < RHEL_RELEASE_VERSION(8, 0)) \|\| \
158	(RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(6, 8) && RHEL_RELEASE_CODE < RHEL_RELEASE_VERSION(7, 0))
159	# define VBOX_HAVE_SKB_VLAN
160	# endif
161	# endif
162	#endif
163
164	#ifdef VBOX_HAVE_SKB_VLAN
165	# define vlan_tx_tag_get(skb) skb_vlan_tag_get(skb)
166	# define vlan_tx_tag_present(skb) skb_vlan_tag_present(skb)
167	#endif
168
169	#ifndef NET_IP_ALIGN
170	# define NET_IP_ALIGN 2
171	#endif
172
173	#if 1
174	/** Create scatter / gather segments for fragments. When not used, we will
175	* linearize the socket buffer before creating the internal networking SG. */
176	# define VBOXNETFLT_SG_SUPPORT 1
177	#endif
178
179	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
180
181	/** Indicates that the linux kernel may send us GSO frames. */
182	# define VBOXNETFLT_WITH_GSO 1
183
184	/** This enables or disables the transmitting of GSO frame from the internal
185	* network and to the host. */
186	# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
187
188	# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
189	/** This enables or disables the transmitting of GSO frame from the internal
190	* network and to the wire. */
191	# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
192	# endif
193
194	/** This enables or disables the forwarding/flooding of GSO frame from the host
195	* to the internal network. */
196	# define VBOXNETFLT_WITH_GSO_RECV 1
197
198	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18) */
199
200	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
201	/** This enables or disables handling of GSO frames coming from the wire (GRO). */
202	# define VBOXNETFLT_WITH_GRO 1
203	#endif
204
205	/*
206	* GRO support was backported to RHEL 5.4
207	*/
208	#ifdef RHEL_RELEASE_CODE
209	# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
210	# define VBOXNETFLT_WITH_GRO 1
211	# endif
212	#endif
213
214
215	/*********************************************************************************************************************************
216	* Internal Functions *
217	*********************************************************************************************************************************/
218	static int VBoxNetFltLinuxInit(void);
219	static void VBoxNetFltLinuxUnload(void);
220	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
221
222
223	/*********************************************************************************************************************************
224	* Global Variables *
225	*********************************************************************************************************************************/
226	/**
227	* The (common) global data.
228	*/
229	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
230
231	module_init(VBoxNetFltLinuxInit);
232	module_exit(VBoxNetFltLinuxUnload);
233
234	MODULE_AUTHOR(VBOX_VENDOR);
235	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
236	MODULE_LICENSE("GPL");
237	#ifdef MODULE_VERSION
238	MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV) " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
239	#endif
240
241
242	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
243	unsigned dev_get_flags(const struct net_device *dev)
244	{
245	unsigned flags;
246
247	flags = (dev->flags & ~(IFF_PROMISC \|
248	IFF_ALLMULTI \|
249	IFF_RUNNING)) \|
250	(dev->gflags & (IFF_PROMISC \|
251	IFF_ALLMULTI));
252
253	if (netif_running(dev) && netif_carrier_ok(dev))
254	flags \|= IFF_RUNNING;
255
256	return flags;
257	}
258	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
259
260
261	/**
262	* Initialize module.
263	*
264	* @returns appropriate status code.
265	*/
266	static int __init VBoxNetFltLinuxInit(void)
267	{
268	int rc;
269	/*
270	* Initialize IPRT.
271	*/
272	rc = RTR0Init(0);
273	if (RT_SUCCESS(rc))
274	{
275	Log(("VBoxNetFltLinuxInit\n"));
276
277	/*
278	* Initialize the globals and connect to the support driver.
279	*
280	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
281	* for establishing the connect to the support driver.
282	*/
283	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
284	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
285	if (RT_SUCCESS(rc))
286	{
287	LogRel(("VBoxNetFlt: Successfully started.\n"));
288	return 0;
289	}
290
291	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
292	RTR0Term();
293	}
294	else
295	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
296
297	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
298	return -RTErrConvertToErrno(rc);
299	}
300
301
302	/**
303	* Unload the module.
304	*
305	* @todo We have to prevent this if we're busy!
306	*/
307	static void __exit VBoxNetFltLinuxUnload(void)
308	{
309	int rc;
310	Log(("VBoxNetFltLinuxUnload\n"));
311	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
312
313	/*
314	* Undo the work done during start (in reverse order).
315	*/
316	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
317	AssertRC(rc); NOREF(rc);
318
319	RTR0Term();
320
321	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
322
323	Log(("VBoxNetFltLinuxUnload - done\n"));
324	}
325
326
327	/**
328	* We filter traffic from the host to the internal network
329	* before it reaches the NIC driver.
330	*
331	* The current code uses a very ugly hack overriding hard_start_xmit
332	* callback in the device structure, but it has been shown to give us a
333	* performance boost of 60-100% though. Eventually we have to find some
334	* less hacky way of getting this job done.
335	*/
336	#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
337
338	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
339
340	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
341
342	# include <linux/ethtool.h>
343
344	typedef struct ethtool_ops OVR_OPSTYPE;
345	# define OVR_OPS ethtool_ops
346	# define OVR_XMIT pfnStartXmit
347
348	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
349
350	typedef struct net_device_ops OVR_OPSTYPE;
351	# define OVR_OPS netdev_ops
352	# define OVR_XMIT pOrgOps->ndo_start_xmit
353
354	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
355
356	/**
357	* The overridden net_device_ops of the device we're attached to.
358	*
359	* As there is no net_device_ops structure in pre-2.6.29 kernels we override
360	* ethtool_ops instead along with hard_start_xmit callback in net_device
361	* structure.
362	*
363	* This is a very dirty hack that was created to explore how much we can improve
364	* the host to guest transfers by not CC'ing the NIC. It turns out to be
365	* the only way to filter outgoing packets for devices without TX queue.
366	*/
367	typedef struct VBoxNetDeviceOpsOverride
368	{
369	/** Our overridden ops. */
370	OVR_OPSTYPE Ops;
371	/** Magic word. */
372	uint32_t u32Magic;
373	/** Pointer to the original ops. */
374	OVR_OPSTYPE const *pOrgOps;
375	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
376	/** Pointer to the original hard_start_xmit function. */
377	int (pfnStartXmit)(struct sk_buff pSkb, struct net_device *pDev);
378	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
379	/** Pointer to the net filter instance. */
380	PVBOXNETFLTINS pVBoxNetFlt;
381	/** The number of filtered packages. */
382	uint64_t cFiltered;
383	/** The total number of packets */
384	uint64_t cTotal;
385	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
386	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
387	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
388
389	/**
390	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
391	* because they belong on the internal network.
392	*
393	* @returns NETDEV_TX_XXX.
394	* @param pSkb The socket buffer to transmit.
395	* @param pDev The net device.
396	*/
397	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
398	{
399	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
400	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
401	PCRTNETETHERHDR pEtherHdr;
402	PINTNETTRUNKSWPORT pSwitchPort;
403	uint32_t cbHdrs;
404
405
406	/*
407	* Validate the override structure.
408	*
409	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
410	* to be production quality code, we would have to be much more
411	* careful here and avoid the race.
412	*/
413	if ( !VALID_PTR(pOverride)
414	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
415	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
416	\|\| !VALID_PTR(pOverride->pOrgOps)
417	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
418	)
419	{
420	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
421	dev_kfree_skb(pSkb);
422	return NETDEV_TX_OK;
423	}
424	pOverride->cTotal++;
425
426	/*
427	* Do the filtering base on the default OUI of our virtual NICs
428	*
429	* Note! In a real solution, we would ask the switch whether the
430	* destination MAC is 100% to be on the internal network and then
431	* drop it.
432	*/
433	cbHdrs = skb_headlen(pSkb);
434	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
435	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
436	if ( pEtherHdr
437	&& VALID_PTR(pOverride->pVBoxNetFlt)
438	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
439	&& VALID_PTR(pSwitchPort)
440	&& cbHdrs >= 6)
441	{
442	INTNETSWDECISION enmDecision;
443
444	/** @todo consider reference counting, etc. */
445	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
446	if (enmDecision == INTNETSWDECISION_INTNET)
447	{
448	dev_kfree_skb(pSkb);
449	pOverride->cFiltered++;
450	return NETDEV_TX_OK;
451	}
452	}
453
454	return pOverride->OVR_XMIT(pSkb, pDev);
455	}
456
457	/**
458	* Hooks the device ndo_start_xmit operation of the device.
459	*
460	* @param pThis The net filter instance.
461	* @param pDev The net device.
462	*/
463	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
464	{
465	PVBOXNETDEVICEOPSOVERRIDE pOverride;
466
467	/* Cancel override if ethtool_ops is missing (host-only case, @bugref{5712}) */
468	if (!VALID_PTR(pDev->OVR_OPS))
469	return;
470	pOverride = RTMemAlloc(sizeof(*pOverride));
471	if (!pOverride)
472	return;
473	pOverride->pOrgOps = pDev->OVR_OPS;
474	pOverride->Ops = *pDev->OVR_OPS;
475	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
476	pOverride->pfnStartXmit = pDev->hard_start_xmit;
477	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
478	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
479	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
480	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
481	pOverride->cTotal = 0;
482	pOverride->cFiltered = 0;
483	pOverride->pVBoxNetFlt = pThis;
484
485	RTSpinlockAcquire(pThis->hSpinlock); /* (this isn't necessary, but so what) */
486	ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
487	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
488	ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
489	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
490	RTSpinlockRelease(pThis->hSpinlock);
491	}
492
493	/**
494	* Undos what vboxNetFltLinuxHookDev did.
495	*
496	* @param pThis The net filter instance.
497	* @param pDev The net device. Can be NULL, in which case
498	* we'll try retrieve it from @a pThis.
499	*/
500	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
501	{
502	PVBOXNETDEVICEOPSOVERRIDE pOverride;
503
504	RTSpinlockAcquire(pThis->hSpinlock);
505	if (!pDev)
506	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
507	if (VALID_PTR(pDev))
508	{
509	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
510	if ( VALID_PTR(pOverride)
511	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
512	&& VALID_PTR(pOverride->pOrgOps)
513	)
514	{
515	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
516	ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
517	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
518	ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
519	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
520	}
521	else
522	pOverride = NULL;
523	}
524	else
525	pOverride = NULL;
526	RTSpinlockRelease(pThis->hSpinlock);
527
528	if (pOverride)
529	{
530	printk("vboxnetflt: %llu out of %llu packets were not sent (directed to host)\n", pOverride->cFiltered, pOverride->cTotal);
531	RTMemFree(pOverride);
532	}
533	}
534
535	#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
536
537
538	/**
539	* Reads and retains the host interface handle.
540	*
541	* @returns The handle, NULL if detached.
542	* @param pThis
543	*/
544	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
545	{
546	#if 0
547	struct net_device *pDev = NULL;
548
549	Log(("vboxNetFltLinuxRetainNetDev\n"));
550	/*
551	* Be careful here to avoid problems racing the detached callback.
552	*/
553	RTSpinlockAcquire(pThis->hSpinlock);
554	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
555	{
556	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
557	if (pDev)
558	{
559	dev_hold(pDev);
560	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
561	pDev, pDev->name,
562	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
563	netdev_refcnt_read(pDev)
564	#else
565	atomic_read(&pDev->refcnt)
566	#endif
567	));
568	}
569	}
570	RTSpinlockRelease(pThis->hSpinlock);
571
572	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
573	return pDev;
574	#else
575	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
576	#endif
577	}
578
579
580	/**
581	* Release the host interface handle previously retained
582	* by vboxNetFltLinuxRetainNetDev.
583	*
584	* @param pThis The instance.
585	* @param pDev The vboxNetFltLinuxRetainNetDev
586	* return value, NULL is fine.
587	*/
588	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
589	{
590	#if 0
591	Log(("vboxNetFltLinuxReleaseNetDev\n"));
592	NOREF(pThis);
593	if (pDev)
594	{
595	dev_put(pDev);
596	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
597	pDev, pDev->name,
598	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
599	netdev_refcnt_read(pDev)
600	#else
601	atomic_read(&pDev->refcnt)
602	#endif
603	));
604	}
605	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
606	#endif
607	}
608
609	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
610	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
611
612	/**
613	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
614	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
615	*
616	* @returns true / false accordingly.
617	* @param pBuf The sk_buff.
618	*/
619	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
620	{
621	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
622	}
623
624
625	/**
626	* Checks whether this SG list contains a GSO packet.
627	*
628	* @returns true / false accordingly.
629	* @param pSG The (scatter/)gather list.
630	*/
631	DECLINLINE(bool) vboxNetFltLinuxIsGso(PINTNETSG pSG)
632	{
633	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
634	return !((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type == PDMNETWORKGSOTYPE_INVALID);
635	#else /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
636	return false;
637	#endif /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
638	}
639
640
641	/**
642	* Find out the frame size (of a single segment in case of GSO frames).
643	*
644	* @returns the frame size.
645	* @param pSG The (scatter/)gather list.
646	*/
647	DECLINLINE(uint32_t) vboxNetFltLinuxFrameSize(PINTNETSG pSG)
648	{
649	uint16_t u16Type = 0;
650	uint32_t cbVlanTag = 0;
651	if (pSG->aSegs[0].cb >= sizeof(RTNETETHERHDR))
652	u16Type = RT_BE2H_U16(((PCRTNETETHERHDR)pSG->aSegs[0].pv)->EtherType);
653	else if (pSG->cbTotal >= sizeof(RTNETETHERHDR))
654	{
655	uint32_t off = RT_OFFSETOF(RTNETETHERHDR, EtherType);
656	uint32_t i;
657	for (i = 0; i < pSG->cSegsUsed; ++i)
658	{
659	if (off <= pSG->aSegs[i].cb)
660	{
661	if (off + sizeof(uint16_t) <= pSG->aSegs[i].cb)
662	u16Type = RT_BE2H_U16((uint16_t )((uintptr_t)pSG->aSegs[i].pv + off));
663	else if (i + 1 < pSG->cSegsUsed)
664	u16Type = RT_BE2H_U16( ((uint16_t)( ((uint8_t *)pSG->aSegs[i].pv)[off] ) << 8)
665	+ (uint8_t )pSG->aSegs[i + 1].pv); /* ASSUMES no empty segments! */
666	/* else: frame is too short. */
667	break;
668	}
669	off -= pSG->aSegs[i].cb;
670	}
671	}
672	if (u16Type == RTNET_ETHERTYPE_VLAN)
673	cbVlanTag = 4;
674	return (vboxNetFltLinuxIsGso(pSG) ? (uint32_t)pSG->GsoCtx.cbMaxSeg + pSG->GsoCtx.cbHdrsTotal : pSG->cbTotal) - cbVlanTag;
675	}
676
677
678	/**
679	* Internal worker that create a linux sk_buff for a
680	* (scatter/)gather list.
681	*
682	* @returns Pointer to the sk_buff.
683	* @param pThis The instance.
684	* @param pSG The (scatter/)gather list.
685	* @param fDstWire Set if the destination is the wire.
686	*/
687	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
688	{
689	struct sk_buff *pPkt;
690	struct net_device *pDev;
691	unsigned fGsoType = 0;
692
693	if (pSG->cbTotal == 0)
694	{
695	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
696	return NULL;
697	}
698	Log5(("VBoxNetFlt: Packet to %s of %d bytes (frame=%d).\n", fDstWire?"wire":"host", pSG->cbTotal, vboxNetFltLinuxFrameSize(pSG)));
699	if (fDstWire && (vboxNetFltLinuxFrameSize(pSG) > ASMAtomicReadU32(&pThis->u.s.cbMtu) + 14))
700	{
701	static bool s_fOnce = true;
702	if (s_fOnce)
703	{
704	s_fOnce = false;
705	printk("VBoxNetFlt: Dropped over-sized packet (%d bytes) coming from internal network.\n", vboxNetFltLinuxFrameSize(pSG));
706	}
707	return NULL;
708	}
709
710	/** @todo We should use fragments mapping the SG buffers with large packets.
711	* 256 bytes seems to be the a threshold used a lot for this. It
712	* requires some nasty work on the intnet side though... */
713	/*
714	* Allocate a packet and copy over the data.
715	*/
716	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
717	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
718	if (RT_UNLIKELY(!pPkt))
719	{
720	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
721	pSG->pvUserData = NULL;
722	return NULL;
723	}
724	pPkt->dev = pDev;
725	pPkt->ip_summed = CHECKSUM_NONE;
726
727	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
728	skb_reserve(pPkt, NET_IP_ALIGN);
729
730	/* Copy the segments. */
731	skb_put(pPkt, pSG->cbTotal);
732	IntNetSgRead(pSG, pPkt->data);
733
734	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
735	/*
736	* Setup GSO if used by this packet.
737	*/
738	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
739	{
740	default:
741	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
742	/* fall thru */
743	case PDMNETWORKGSOTYPE_INVALID:
744	fGsoType = 0;
745	break;
746	case PDMNETWORKGSOTYPE_IPV4_TCP:
747	fGsoType = SKB_GSO_TCPV4;
748	break;
749	case PDMNETWORKGSOTYPE_IPV4_UDP:
750	fGsoType = SKB_GSO_UDP;
751	break;
752	case PDMNETWORKGSOTYPE_IPV6_TCP:
753	fGsoType = SKB_GSO_TCPV6;
754	break;
755	}
756	if (fGsoType)
757	{
758	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
759
760	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
761	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
762	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
763
764	/*
765	* We need to set checksum fields even if the packet goes to the host
766	* directly as it may be immediately forwarded by IP layer @bugref{5020}.
767	*/
768	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
769	pPkt->ip_summed = CHECKSUM_PARTIAL;
770	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
771	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
772	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
773	pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
774	else
775	pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
776	# else
777	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
778	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
779	pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
780	else
781	pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
782	# endif
783	if (!fDstWire)
784	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
785	}
786	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
787
788	/*
789	* Finish up the socket buffer.
790	*/
791	pPkt->protocol = eth_type_trans(pPkt, pDev);
792	if (fDstWire)
793	{
794	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
795
796	/* Restore ethernet header back. */
797	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
798	VBOX_SKB_RESET_MAC_HDR(pPkt);
799	}
800	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
801
802	return pPkt;
803	}
804
805
806	/**
807	* Return the offset where to start checksum computation from.
808	*
809	* @returns the offset relative to pBuf->data.
810	* @param pBuf The socket buffer.
811	*/
812	DECLINLINE(unsigned) vboxNetFltLinuxGetChecksumStartOffset(struct sk_buff *pBuf)
813	{
814	# if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 21)
815	unsigned char *pTransportHdr = pBuf->h.raw;
816	# if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
817	/*
818	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
819	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
820	* header length from the header itself and reconstruct 'h' pointer
821	* to TCP (or whatever) header.
822	*/
823	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
824	pTransportHdr = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
825	# endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
826	return pTransportHdr - pBuf->data;
827	# else /* LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 21) */
828	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 38)
829	return pBuf->csum_start - skb_headroom(pBuf);
830	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 38) */
831	return skb_checksum_start_offset(pBuf);
832	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 38) */
833	# endif /* LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 21) */
834	}
835
836
837	/**
838	* Initializes a SG list from an sk_buff.
839	*
840	* @returns Number of segments.
841	* @param pThis The instance.
842	* @param pBuf The sk_buff.
843	* @param pSG The SG.
844	* @param cbExtra The number of bytes of extra space allocated immediately after the SG.
845	* @param cSegs The number of segments allocated for the SG.
846	* This should match the number in the mbuf exactly!
847	* @param fSrc The source of the frame.
848	* @param pGsoCtx Pointer to the GSO context if it's a GSO
849	* internal network frame. NULL if regular frame.
850	*/
851	static void vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
852	unsigned cbExtra, unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
853	{
854	int i;
855	NOREF(pThis);
856
857	#ifndef VBOXNETFLT_SG_SUPPORT
858	Assert(!skb_shinfo(pBuf)->frag_list);
859	#else /* VBOXNETFLT_SG_SUPPORT */
860	uint8_t pExtra = (uint8_t )&pSG->aSegs[cSegs];
861	unsigned cbConsumed = 0;
862	unsigned cbProduced = 0;
863
864	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
865	/* Restore VLAN tag stripped by host hardware */
866	if (vlan_tx_tag_present(pBuf))
867	{
868	uint8_t *pMac = pBuf->data;
869	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )pExtra;
870	Assert(ETH_ALEN * 2 + VLAN_HLEN <= cbExtra);
871	memmove(pVHdr, pMac, ETH_ALEN * 2);
872	cbConsumed += ETH_ALEN * 2;
873	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
874	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
875	pVHdr->h_vlan_encapsulated_proto = (uint16_t)(pMac + ETH_ALEN * 2);
876	cbProduced += VLAN_ETH_HLEN;
877	}
878	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
879
880	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
881	{
882	unsigned uCsumStartOffset = vboxNetFltLinuxGetChecksumStartOffset(pBuf);
883	unsigned uCsumStoreOffset = uCsumStartOffset + VBOX_SKB_CSUM_OFFSET(pBuf) - cbConsumed;
884	Log3(("cbConsumed=%u cbProduced=%u uCsumStartOffset=%u uCsumStoreOffset=%u\n",
885	cbConsumed, cbProduced, uCsumStartOffset, uCsumStoreOffset));
886	Assert(cbProduced + uCsumStoreOffset + sizeof(uint16_t) <= cbExtra);
887	/*
888	* We assume that the checksum is stored at the very end of the transport header
889	* so we will have all headers in a single fragment. If our assumption is wrong
890	* we may see suboptimal performance.
891	*/
892	memmove(pExtra + cbProduced,
893	pBuf->data + cbConsumed,
894	uCsumStoreOffset);
895	unsigned uChecksum = skb_checksum(pBuf, uCsumStartOffset, pBuf->len - uCsumStartOffset, 0);
896	(uint16_t)(pExtra + cbProduced + uCsumStoreOffset) = csum_fold(uChecksum);
897	cbProduced += uCsumStoreOffset + sizeof(uint16_t);
898	cbConsumed += uCsumStoreOffset + sizeof(uint16_t);
899	}
900	#endif /* VBOXNETFLT_SG_SUPPORT */
901
902	if (!pGsoCtx)
903	IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /cSegsUsed/);
904	else
905	IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /cSegsUsed/, pGsoCtx);
906
907	int iSeg = 0;
908	#ifdef VBOXNETFLT_SG_SUPPORT
909	if (cbProduced)
910	{
911	pSG->aSegs[iSeg].cb = cbProduced;
912	pSG->aSegs[iSeg].pv = pExtra;
913	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
914	}
915	pSG->aSegs[iSeg].cb = skb_headlen(pBuf) - cbConsumed;
916	pSG->aSegs[iSeg].pv = pBuf->data + cbConsumed;
917	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
918	Assert(iSeg <= pSG->cSegsAlloc);
919
920	# ifdef LOG_ENABLED
921	if (pBuf->data_len)
922	Log6((" kmap_atomic:"));
923	# endif /* LOG_ENABLED */
924	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
925	{
926	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
927	pSG->aSegs[iSeg].cb = pFrag->size;
928	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
929	Log6((" %p", pSG->aSegs[iSeg].pv));
930	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
931	Assert(iSeg <= pSG->cSegsAlloc);
932	}
933	struct sk_buff *pFragBuf;
934	for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
935	{
936	pSG->aSegs[iSeg].cb = skb_headlen(pFragBuf);
937	pSG->aSegs[iSeg].pv = pFragBuf->data;
938	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
939	Assert(iSeg <= pSG->cSegsAlloc);
940	for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
941	{
942	skb_frag_t *pFrag = &skb_shinfo(pFragBuf)->frags[i];
943	pSG->aSegs[iSeg].cb = pFrag->size;
944	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
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	#ifdef PADD_RUNT_FRAMES_FROM_HOST
963	/*
964	* Add a trailer if the frame is too small.
965	*
966	* Since we're getting to the packet before it is framed, it has not
967	* yet been padded. The current solution is to add a segment pointing
968	* to a buffer containing all zeros and pray that works for all frames...
969	*/
970	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
971	{
972	Assert(pBuf->data_len == 0); /* Packets with fragments are never small! */
973	static uint8_t const s_abZero[128] = {0};
974
975	AssertReturnVoid(iSeg < cSegs);
976
977	pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
978	pSG->aSegs[iSeg].pv = (void *)&s_abZero[0];
979	pSG->aSegs[iSeg++].cb = 60 - pSG->cbTotal;
980	pSG->cbTotal = 60;
981	pSG->cSegsUsed++;
982	Assert(iSeg <= pSG->cSegsAlloc)
983	}
984	#endif
985
986	Log6(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
987	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
988	for (i = 0; i < pSG->cSegsUsed; i++)
989	Log6(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
990	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
991	}
992
993	/**
994	* Packet handler; not really documented - figure it out yourself.
995	*
996	* @returns 0 or EJUSTRETURN - this is probably copy & pastry and thus wrong.
997	*/
998	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
999	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1000	struct net_device *pSkbDev,
1001	struct packet_type *pPacketType,
1002	struct net_device *pOrigDev)
1003	#else
1004	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1005	struct net_device *pSkbDev,
1006	struct packet_type *pPacketType)
1007	#endif
1008	{
1009	PVBOXNETFLTINS pThis;
1010	struct net_device *pDev;
1011	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1012	pBuf, pSkbDev, pPacketType));
1013	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1014	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",
1015	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));
1016	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1017	Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1018	# endif
1019	#else
1020	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",
1021	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));
1022	#endif
1023	/*
1024	* Drop it immediately?
1025	*/
1026	if (!pBuf)
1027	return 0;
1028
1029	if (pBuf->pkt_type == PACKET_LOOPBACK)
1030	{
1031	/*
1032	* We are not interested in loopbacked packets as they will always have
1033	* another copy going to the wire.
1034	*/
1035	Log2(("vboxNetFltLinuxPacketHandler: dropped loopback packet (cb=%u)\n", pBuf->len));
1036	dev_kfree_skb(pBuf); /* We must 'consume' all packets we get (@bugref{6539})! */
1037	return 0;
1038	}
1039
1040	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1041	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1042	if (pDev != pSkbDev)
1043	{
1044	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1045	kfree_skb(pBuf); /* This is a failure, so we use kfree_skb instead of dev_kfree_skb. */
1046	return 0;
1047	}
1048
1049	Log6(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1050	if (vboxNetFltLinuxSkBufIsOur(pBuf))
1051	{
1052	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1053	dev_kfree_skb(pBuf);
1054	return 0;
1055	}
1056
1057	#ifndef VBOXNETFLT_SG_SUPPORT
1058	{
1059	/*
1060	* Get rid of fragmented packets, they cause too much trouble.
1061	*/
1062	unsigned int uMacLen = pBuf->mac_len;
1063	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1064	dev_kfree_skb(pBuf);
1065	if (!pCopy)
1066	{
1067	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1068	return 0;
1069	}
1070	pBuf = pCopy;
1071	/* Somehow skb_copy ignores mac_len */
1072	pBuf->mac_len = uMacLen;
1073	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
1074	/* Restore VLAN tag stripped by host hardware */
1075	if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
1076	{
1077	uint8_t pMac = (uint8_t)skb_mac_header(pBuf);
1078	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )(pMac - VLAN_HLEN);
1079	memmove(pVHdr, pMac, ETH_ALEN * 2);
1080	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
1081	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
1082	pBuf->mac_header -= VLAN_HLEN;
1083	pBuf->mac_len += VLAN_HLEN;
1084	}
1085	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
1086
1087	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1088	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",
1089	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));
1090	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1091	Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1092	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22) */
1093	# else /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 18) */
1094	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",
1095	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));
1096	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 18) */
1097	}
1098	#endif /* !VBOXNETFLT_SG_SUPPORT */
1099
1100	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1101	/* Forward it to the internal network. */
1102	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1103	#else /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1104	/* Add the packet to transmit queue and schedule the bottom half. */
1105	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1106	schedule_work(&pThis->u.s.XmitTask);
1107	Log6(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1108	&pThis->u.s.XmitTask, pBuf));
1109	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1110
1111	/* It does not really matter what we return, it is ignored by the kernel. */
1112	return 0;
1113	}
1114
1115	/**
1116	* Calculate the number of INTNETSEG segments the socket buffer will need.
1117	*
1118	* @returns Segment count.
1119	* @param pBuf The socket buffer.
1120	* @param pcbTemp Where to store the number of bytes of the part
1121	* of the socket buffer that will be copied to
1122	* a temporary storage.
1123	*/
1124	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff pBuf, unsigned pcbTemp)
1125	{
1126	*pcbTemp = 0;
1127	#ifdef VBOXNETFLT_SG_SUPPORT
1128	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1129	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1130	{
1131	*pcbTemp = vboxNetFltLinuxGetChecksumStartOffset(pBuf) + VBOX_SKB_CSUM_OFFSET(pBuf) + sizeof(uint16_t);
1132	}
1133	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
1134	if (vlan_tx_tag_present(pBuf))
1135	{
1136	if (*pcbTemp)
1137	*pcbTemp += VLAN_HLEN;
1138	else
1139	*pcbTemp = VLAN_ETH_HLEN;
1140	}
1141	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
1142	if (*pcbTemp)
1143	++cSegs;
1144	struct sk_buff *pFrag;
1145	for (pFrag = skb_shinfo(pBuf)->frag_list; pFrag; pFrag = pFrag->next)
1146	{
1147	Log6(("vboxNetFltLinuxCalcSGSegments: frag=%p len=%d data_len=%d frags=%d frag_list=%p next=%p\n",
1148	pFrag, pFrag->len, pFrag->data_len, skb_shinfo(pFrag)->nr_frags, skb_shinfo(pFrag)->frag_list, pFrag->next));
1149	cSegs += 1 + skb_shinfo(pFrag)->nr_frags;
1150	}
1151	#else
1152	unsigned cSegs = 1;
1153	#endif
1154	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1155	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1156	if (pBuf->len < 60)
1157	cSegs++;
1158	#endif
1159	return cSegs;
1160	}
1161
1162
1163	/**
1164	* Destroy the intnet scatter / gather buffer created by
1165	* vboxNetFltLinuxSkBufToSG.
1166	*
1167	* @param pSG The (scatter/)gather list.
1168	* @param pBuf The original socket buffer that was used to create
1169	* the scatter/gather list.
1170	*/
1171	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG, struct sk_buff *pBuf)
1172	{
1173	#ifdef VBOXNETFLT_SG_SUPPORT
1174	int i, iSeg = 1; /* Skip non-paged part of SKB */
1175	/* Check if the extra buffer behind SG structure was used for modified packet header */
1176	if (pBuf->data != pSG->aSegs[0].pv)
1177	++iSeg; /* Skip it as well */
1178	# ifdef LOG_ENABLED
1179	if (pBuf->data_len)
1180	Log6(("kunmap_atomic:"));
1181	# endif /* LOG_ENABLED */
1182	/* iSeg now points to the first mapped fragment if there are any */
1183	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1184	{
1185	Log6((" %p", pSG->aSegs[iSeg].pv));
1186	VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1187	}
1188	struct sk_buff *pFragBuf;
1189	for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
1190	{
1191	++iSeg; /* Non-fragment (unmapped) portion of chained SKB */
1192	for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
1193	{
1194	Log6((" %p", pSG->aSegs[iSeg].pv));
1195	VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1196	}
1197	}
1198	# ifdef LOG_ENABLED
1199	if (pBuf->data_len)
1200	Log6(("\n"));
1201	# endif /* LOG_ENABLED */
1202	#endif
1203	NOREF(pSG);
1204	}
1205
1206	#ifdef LOG_ENABLED
1207	/**
1208	* Logging helper.
1209	*/
1210	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1211	{
1212	int i, offSeg;
1213	uint8_t pInt, pExt;
1214	static int iPacketNo = 1;
1215	iPacketNo += iIncrement;
1216	if (fEgress)
1217	{
1218	pExt = pSG->aSegs[0].pv;
1219	pInt = pExt + 6;
1220	}
1221	else
1222	{
1223	pInt = pSG->aSegs[0].pv;
1224	pExt = pInt + 6;
1225	}
1226	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1227	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1228	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1229	fEgress ? "-->" : "<--", pszWhere,
1230	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1231	pSG->cbTotal, iPacketNo));
1232	if (pSG->cSegsUsed == 1)
1233	{
1234	Log4(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1235	}
1236	else
1237	{
1238	for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
1239	{
1240	Log4(("-- segment %d at 0x%x (%d bytes)\n --\n%.*Rhxd\n",
1241	i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1242	offSeg += pSG->aSegs[i].cb;
1243	}
1244	}
1245	}
1246	#else
1247	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1248	#endif
1249
1250	#ifdef VBOXNETFLT_WITH_GSO_RECV
1251
1252	/**
1253	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1254	* GSO socket buffer without having to segment it.
1255	*
1256	* @returns true on success, false if needs segmenting.
1257	* @param pThis The net filter instance.
1258	* @param pSkb The GSO socket buffer.
1259	* @param fSrc The source.
1260	* @param pGsoCtx Where to return the GSO context on success.
1261	*/
1262	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1263	PPDMNETWORKGSO pGsoCtx)
1264	{
1265	PDMNETWORKGSOTYPE enmGsoType;
1266	uint16_t uEtherType;
1267	unsigned int cbTransport;
1268	unsigned int offTransport;
1269	unsigned int cbTransportHdr;
1270	unsigned uProtocol;
1271	union
1272	{
1273	RTNETIPV4 IPv4;
1274	RTNETIPV6 IPv6;
1275	RTNETTCP Tcp;
1276	uint8_t ab[40];
1277	uint16_t au16[40/2];
1278	uint32_t au32[40/4];
1279	} Buf;
1280
1281	/*
1282	* Check the GSO properties of the socket buffer and make sure it fits.
1283	*/
1284	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1285	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP \| SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
1286	{
1287	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1288	return false;
1289	}
1290	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1291	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
1292	{
1293	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1294	return false;
1295	}
1296
1297	/*
1298	* Switch on the ethertype.
1299	*/
1300	uEtherType = pSkb->protocol;
1301	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1302	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1303	{
1304	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1305	if (puEtherType)
1306	uEtherType = *puEtherType;
1307	}
1308	switch (uEtherType)
1309	{
1310	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1311	{
1312	unsigned int cbHdr;
1313	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1314	if (RT_UNLIKELY(!pIPv4))
1315	{
1316	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1317	return false;
1318	}
1319
1320	cbHdr = pIPv4->ip_hl * 4;
1321	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1322	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1323	\|\| cbHdr > cbTransport ))
1324	{
1325	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1326	return false;
1327	}
1328	cbTransport -= cbHdr;
1329	offTransport = pSkb->mac_len + cbHdr;
1330	uProtocol = pIPv4->ip_p;
1331	if (uProtocol == RTNETIPV4_PROT_TCP)
1332	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1333	else if (uProtocol == RTNETIPV4_PROT_UDP)
1334	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1335	else /** @todo IPv6: 4to6 tunneling */
1336	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1337	break;
1338	}
1339
1340	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1341	{
1342	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1343	if (RT_UNLIKELY(!pIPv6))
1344	{
1345	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1346	return false;
1347	}
1348
1349	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1350	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1351	uProtocol = pIPv6->ip6_nxt;
1352	/** @todo IPv6: Dig our way out of the other headers. */
1353	if (uProtocol == RTNETIPV4_PROT_TCP)
1354	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1355	else if (uProtocol == RTNETIPV4_PROT_UDP)
1356	enmGsoType = PDMNETWORKGSOTYPE_IPV6_UDP;
1357	else
1358	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1359	break;
1360	}
1361
1362	default:
1363	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1364	return false;
1365	}
1366
1367	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1368	{
1369	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1370	return false;
1371	}
1372
1373	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1374	\|\| offTransport + cbTransport > pSkb->len
1375	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1376	{
1377	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1378	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1379	return false;
1380	}
1381
1382	/*
1383	* Check the TCP/UDP bits.
1384	*/
1385	if (uProtocol == RTNETIPV4_PROT_TCP)
1386	{
1387	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1388	if (RT_UNLIKELY(!pTcp))
1389	{
1390	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1391	return false;
1392	}
1393
1394	cbTransportHdr = pTcp->th_off * 4;
1395	pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1396	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1397	\|\| cbTransportHdr > cbTransport
1398	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1399	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1400	{
1401	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1402	return false;
1403	}
1404
1405	}
1406	else
1407	{
1408	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1409	cbTransportHdr = sizeof(RTNETUDP);
1410	pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1411	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1412	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1413	{
1414	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1415	return false;
1416	}
1417	}
1418
1419	/*
1420	* We're good, init the GSO context.
1421	*/
1422	pGsoCtx->u8Type = enmGsoType;
1423	pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1424	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1425	pGsoCtx->offHdr1 = pSkb->mac_len;
1426	pGsoCtx->offHdr2 = offTransport;
1427	pGsoCtx->u8Unused = 0;
1428
1429	return true;
1430	}
1431
1432	/**
1433	* Forward the socket buffer as a GSO internal network frame.
1434	*
1435	* @returns IPRT status code.
1436	* @param pThis The net filter instance.
1437	* @param pSkb The GSO socket buffer.
1438	* @param fSrc The source.
1439	* @param pGsoCtx Where to return the GSO context on success.
1440	*/
1441	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1442	{
1443	int rc;
1444	unsigned cbExtra;
1445	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb, &cbExtra);
1446	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]) + cbExtra);
1447	if (RT_LIKELY(pSG))
1448	{
1449	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cbExtra, cSegs, fSrc, pGsoCtx);
1450
1451	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1452	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1453
1454	vboxNetFltLinuxDestroySG(pSG, pSkb);
1455	rc = VINF_SUCCESS;
1456	}
1457	else
1458	{
1459	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1460	rc = VERR_NO_MEMORY;
1461	}
1462	return rc;
1463	}
1464
1465	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1466
1467	/**
1468	* Worker for vboxNetFltLinuxForwardToIntNet.
1469	*
1470	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1471	* @param pThis The net filter instance.
1472	* @param pBuf The socket buffer.
1473	* @param fSrc The source.
1474	*/
1475	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1476	{
1477	int rc;
1478	unsigned cbExtra;
1479	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf, &cbExtra);
1480	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]) + cbExtra);
1481	if (RT_LIKELY(pSG))
1482	{
1483	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cbExtra, cSegs, fSrc, NULL /pGsoCtx/);
1484
1485	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1486	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1487
1488	vboxNetFltLinuxDestroySG(pSG, pBuf);
1489	rc = VINF_SUCCESS;
1490	}
1491	else
1492	{
1493	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1494	rc = VERR_NO_MEMORY;
1495	}
1496	return rc;
1497	}
1498
1499
1500	/**
1501	* I won't disclose what I do, figure it out yourself, including pThis referencing.
1502	*
1503	* @param pThis The net filter instance.
1504	* @param pBuf The socket buffer.
1505	* @param fSrc Where the packet comes from.
1506	*/
1507	static void vboxNetFltLinuxForwardToIntNetInner(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1508	{
1509	#ifdef VBOXNETFLT_WITH_GSO
1510	if (skb_is_gso(pBuf))
1511	{
1512	PDMNETWORKGSO GsoCtx;
1513	Log6(("vboxNetFltLinuxForwardToIntNetInner: skb len=%u data_len=%u truesize=%u next=%p"
1514	" nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x ip_summed=%d\n",
1515	pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next,
1516	skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->gso_size,
1517	skb_shinfo(pBuf)->gso_segs, skb_shinfo(pBuf)->gso_type,
1518	skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, pBuf->ip_summed));
1519	#ifndef VBOXNETFLT_SG_SUPPORT
1520	if (RT_LIKELY(fSrc & INTNETTRUNKDIR_HOST))
1521	{
1522	/*
1523	* skb_gso_segment does the following. Do we need to do it as well?
1524	*/
1525	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1526	skb_reset_mac_header(pBuf);
1527	pBuf->mac_len = pBuf->network_header - pBuf->mac_header;
1528	# else
1529	pBuf->mac.raw = pBuf->data;
1530	pBuf->mac_len = pBuf->nh.raw - pBuf->data;
1531	# endif
1532	}
1533	#endif /* !VBOXNETFLT_SG_SUPPORT */
1534	# ifdef VBOXNETFLT_WITH_GSO_RECV
1535	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1536	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1537	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1538	else
1539	# endif /* VBOXNETFLT_WITH_GSO_RECV */
1540	{
1541	/* Need to segment the packet */
1542	struct sk_buff *pNext;
1543	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1544	if (IS_ERR(pSegment))
1545	{
1546	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1547	return;
1548	}
1549
1550	for (; pSegment; pSegment = pNext)
1551	{
1552	Log6(("vboxNetFltLinuxForwardToIntNetInner: segment len=%u data_len=%u truesize=%u next=%p"
1553	" nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x\n",
1554	pSegment->len, pSegment->data_len, pSegment->truesize, pSegment->next,
1555	skb_shinfo(pSegment)->nr_frags, skb_shinfo(pSegment)->gso_size,
1556	skb_shinfo(pSegment)->gso_segs, skb_shinfo(pSegment)->gso_type,
1557	skb_shinfo(pSegment)->frag_list, pSegment->pkt_type));
1558	pNext = pSegment->next;
1559	pSegment->next = 0;
1560	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1561	dev_kfree_skb(pSegment);
1562	}
1563	}
1564	}
1565	else
1566	#endif /* VBOXNETFLT_WITH_GSO */
1567	{
1568	Log6(("vboxNetFltLinuxForwardToIntNetInner: ptk_type=%d ip_summed=%d len=%d"
1569	" data_len=%d headroom=%d hdr_len=%d csum_offset=%d\n",
1570	pBuf->pkt_type, pBuf->ip_summed, pBuf->len, pBuf->data_len, skb_headroom(pBuf),
1571	skb_headlen(pBuf), vboxNetFltLinuxGetChecksumStartOffset(pBuf)));
1572	#ifndef VBOXNETFLT_SG_SUPPORT
1573	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1574	{
1575	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1576	/*
1577	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1578	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1579	* header length from the header itself and reconstruct 'h' pointer
1580	* to TCP (or whatever) header.
1581	*/
1582	unsigned char *tmp = pBuf->h.raw;
1583	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1584	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1585	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1586	int rc = VBOX_SKB_CHECKSUM_HELP(pBuf);
1587	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1588	/* Restore the original (wrong) pointer. */
1589	pBuf->h.raw = tmp;
1590	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1591	if (rc) {
1592	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1593	return;
1594	}
1595	}
1596	#endif /* !VBOXNETFLT_SG_SUPPORT */
1597	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1598	}
1599	}
1600
1601
1602	/**
1603	* Temporarily adjust pBuf->data so it always points to the Ethernet header,
1604	* then forward it to the internal network.
1605	*
1606	* @param pThis The net filter instance.
1607	* @param pBuf The socket buffer. This is consumed by this function.
1608	*/
1609	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1610	{
1611	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1612
1613	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1614	{
1615	/*
1616	* The packet came from the wire and the driver has already consumed
1617	* mac header. We need to restore it back. Moreover, after we are
1618	* through with this skb we need to restore its original state!
1619	*/
1620	skb_push(pBuf, pBuf->mac_len);
1621	Log5(("vboxNetFltLinuxForwardToIntNet: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1622	pBuf->mac_len, pBuf->data, skb_mac_header(pBuf), skb_network_header(pBuf)));
1623	}
1624
1625	vboxNetFltLinuxForwardToIntNetInner(pThis, pBuf, fSrc);
1626
1627	/*
1628	* Restore the original state of skb as there are other handlers this skb
1629	* will be provided to.
1630	*/
1631	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1632	skb_pull(pBuf, pBuf->mac_len);
1633
1634	dev_kfree_skb(pBuf);
1635	}
1636
1637
1638	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1639	/**
1640	* Work queue handler that forwards the socket buffers queued by
1641	* vboxNetFltLinuxPacketHandler to the internal network.
1642	*
1643	* @param pWork The work queue.
1644	*/
1645	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1646	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1647	# else
1648	static void vboxNetFltLinuxXmitTask(void *pWork)
1649	# endif
1650	{
1651	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1652	struct sk_buff *pBuf;
1653
1654	Log6(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1655
1656	/*
1657	* Active? Retain the instance and increment the busy counter.
1658	*/
1659	if (vboxNetFltTryRetainBusyActive(pThis))
1660	{
1661	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1662	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1663
1664	vboxNetFltRelease(pThis, true /* fBusy */);
1665	}
1666	else
1667	{
1668	/** @todo Shouldn't we just drop the packets here? There is little point in
1669	* making them accumulate when the VM is paused and it'll only waste
1670	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1671	* before start draining the packets (goes for the intnet ring buf
1672	* too)? */
1673	}
1674	}
1675	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1676
1677	/**
1678	* Reports the GSO capabilities of the hardware NIC.
1679	*
1680	* @param pThis The net filter instance. The caller hold a
1681	* reference to this.
1682	*/
1683	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1684	{
1685	#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1686	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1687	{
1688	struct net_device *pDev;
1689	PINTNETTRUNKSWPORT pSwitchPort;
1690	unsigned int fFeatures;
1691
1692	RTSpinlockAcquire(pThis->hSpinlock);
1693
1694	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1695	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1696	if (pDev)
1697	fFeatures = pDev->features;
1698	else
1699	fFeatures = 0;
1700
1701	RTSpinlockRelease(pThis->hSpinlock);
1702
1703	if (pThis->pSwitchPort)
1704	{
1705	/* Set/update the GSO capabilities of the NIC. */
1706	uint32_t fGsoCapabilites = 0;
1707	if (fFeatures & NETIF_F_TSO)
1708	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1709	if (fFeatures & NETIF_F_TSO6)
1710	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1711	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1712	if (fFeatures & NETIF_F_UFO)
1713	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1714	if (fFeatures & NETIF_F_UFO)
1715	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1716	# endif
1717	Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1718	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1719	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : "",
1720	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)) ? "ufo " : "",
1721	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)) ? "ufo6 " : ""));
1722	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1723	}
1724
1725	vboxNetFltRelease(pThis, true /fBusy/);
1726	}
1727	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1728	}
1729
1730	/**
1731	* Helper that determines whether the host (ignoreing us) is operating the
1732	* interface in promiscuous mode or not.
1733	*/
1734	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1735	{
1736	bool fRc = false;
1737	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1738	if (pDev)
1739	{
1740	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1741	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1742	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1743	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1744	}
1745	return fRc;
1746	}
1747
1748	/**
1749	* Does this device needs link state change signaled?
1750	* Currently we need it for our own VBoxNetAdp and TAP.
1751	*/
1752	static bool vboxNetFltNeedsLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev)
1753	{
1754	if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1755	{
1756	struct ethtool_drvinfo Info;
1757
1758	memset(&Info, 0, sizeof(Info));
1759	Info.cmd = ETHTOOL_GDRVINFO;
1760	pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1761	Log3(("%s: driver=%.s version=%.s bus_info=%.*s\n",
1762	__FUNCTION__,
1763	sizeof(Info.driver), Info.driver,
1764	sizeof(Info.version), Info.version,
1765	sizeof(Info.bus_info), Info.bus_info));
1766
1767	if (!strncmp(Info.driver, "vboxnet", sizeof(Info.driver)))
1768	return true;
1769
1770	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36) /* TAP started doing carrier */
1771	return !strncmp(Info.driver, "tun", 4)
1772	&& !strncmp(Info.bus_info, "tap", 4);
1773	#endif
1774	}
1775
1776	return false;
1777	}
1778
1779	/**
1780	* Some devices need link state change when filter attaches/detaches
1781	* since the filter is their link in a sense.
1782	*/
1783	static void vboxNetFltSetLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1784	{
1785	if (vboxNetFltNeedsLinkState(pThis, pDev))
1786	{
1787	Log3(("%s: bringing device link %s\n",
1788	__FUNCTION__, fLinkUp ? "up" : "down"));
1789	netif_tx_lock_bh(pDev);
1790	if (fLinkUp)
1791	netif_carrier_on(pDev);
1792	else
1793	netif_carrier_off(pDev);
1794	netif_tx_unlock_bh(pDev);
1795	}
1796	}
1797
1798	/**
1799	* Internal worker for vboxNetFltLinuxNotifierCallback.
1800	*
1801	* @returns VBox status code.
1802	* @param pThis The instance.
1803	* @param pDev The device to attach to.
1804	*/
1805	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1806	{
1807	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1808
1809	/*
1810	* Retain and store the device.
1811	*/
1812	dev_hold(pDev);
1813
1814	RTSpinlockAcquire(pThis->hSpinlock);
1815	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1816	RTSpinlockRelease(pThis->hSpinlock);
1817
1818	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1819	pDev, pDev->name,
1820	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1821	netdev_refcnt_read(pDev)
1822	#else
1823	atomic_read(&pDev->refcnt)
1824	#endif
1825	));
1826	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1827	pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1828
1829	/* Get the mac address while we still have a valid net_device reference. */
1830	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1831	/* Initialize MTU */
1832	pThis->u.s.cbMtu = pDev->mtu;
1833
1834	/*
1835	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1836	*/
1837	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1838	pThis->u.s.PacketType.dev = pDev;
1839	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1840	dev_add_pack(&pThis->u.s.PacketType);
1841	ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1842	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1843
1844	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1845	vboxNetFltLinuxHookDev(pThis, pDev);
1846	#endif
1847
1848	/*
1849	* Are we the "carrier" for this device (e.g. vboxnet or tap)?
1850	*/
1851	vboxNetFltSetLinkState(pThis, pDev, true);
1852
1853	/*
1854	* Set indicators that require the spinlock. Be abit paranoid about racing
1855	* the device notification handle.
1856	*/
1857	RTSpinlockAcquire(pThis->hSpinlock);
1858	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1859	if (pDev)
1860	{
1861	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1862	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1863	pDev = NULL; /* don't dereference it */
1864	}
1865	RTSpinlockRelease(pThis->hSpinlock);
1866
1867	/*
1868	* If the above succeeded report GSO capabilities, if not undo and
1869	* release the device.
1870	*/
1871	if (!pDev)
1872	{
1873	Assert(pThis->pSwitchPort);
1874	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1875	{
1876	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1877	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1878	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1879	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1880	vboxNetFltRelease(pThis, true /fBusy/);
1881	}
1882	}
1883	else
1884	{
1885	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1886	vboxNetFltLinuxUnhookDev(pThis, pDev);
1887	#endif
1888	RTSpinlockAcquire(pThis->hSpinlock);
1889	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1890	RTSpinlockRelease(pThis->hSpinlock);
1891	dev_put(pDev);
1892	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n",
1893	pDev, pDev->name,
1894	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1895	netdev_refcnt_read(pDev)
1896	#else
1897	atomic_read(&pDev->refcnt)
1898	#endif
1899	));
1900	}
1901
1902	LogRel(("VBoxNetFlt: attached to '%s' / %RTmac\n", pThis->szName, &pThis->u.s.MacAddr));
1903	return VINF_SUCCESS;
1904	}
1905
1906
1907	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1908	{
1909	bool fRegistered;
1910	Assert(!pThis->fDisconnectedFromHost);
1911
1912	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1913	vboxNetFltLinuxUnhookDev(pThis, pDev);
1914	#endif
1915
1916	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1917	{
1918	dev_remove_pack(&pThis->u.s.PacketType);
1919	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1920	}
1921
1922	RTSpinlockAcquire(pThis->hSpinlock);
1923	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1924	if (fRegistered)
1925	{
1926	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1927	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1928	}
1929	RTSpinlockRelease(pThis->hSpinlock);
1930
1931	if (fRegistered)
1932	{
1933	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1934	skb_queue_purge(&pThis->u.s.XmitQueue);
1935	#endif
1936	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1937	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1938	pDev, pDev->name,
1939	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1940	netdev_refcnt_read(pDev)
1941	#else
1942	atomic_read(&pDev->refcnt)
1943	#endif
1944	));
1945	dev_put(pDev);
1946	}
1947
1948	return NOTIFY_OK;
1949	}
1950
1951	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1952	{
1953	/* Check if we are not suspended and promiscuous mode has not been set. */
1954	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1955	&& !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1956	{
1957	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1958	dev_set_promiscuity(pDev, 1);
1959	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1960	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1961	}
1962	else
1963	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1964	return NOTIFY_OK;
1965	}
1966
1967	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1968	{
1969	/* Undo promiscuous mode if we has set it. */
1970	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1971	{
1972	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1973	dev_set_promiscuity(pDev, -1);
1974	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1975	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1976	}
1977	else
1978	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1979	return NOTIFY_OK;
1980	}
1981
1982	/**
1983	* Callback for listening to MTU change event.
1984	*
1985	* We need to track changes of host's inteface MTU to discard over-sized frames
1986	* coming from the internal network as they may hang the TX queue of host's
1987	* adapter.
1988	*
1989	* @returns NOTIFY_OK
1990	* @param pThis The netfilter instance.
1991	* @param pDev Pointer to device structure of host's interface.
1992	*/
1993	static int vboxNetFltLinuxDeviceMtuChange(PVBOXNETFLTINS pThis, struct net_device *pDev)
1994	{
1995	ASMAtomicWriteU32(&pThis->u.s.cbMtu, pDev->mtu);
1996	Log(("vboxNetFltLinuxDeviceMtuChange: set MTU for %s to %d\n", pThis->szName, pDev->mtu));
1997	return NOTIFY_OK;
1998	}
1999
2000	#ifdef LOG_ENABLED
2001	/** Stringify the NETDEV_XXX constants. */
2002	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2003	{
2004	const char *pszEvent = "NETDEV_<unknown>";
2005	switch (ulEventType)
2006	{
2007	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2008	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2009	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2010	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2011	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2012	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2013	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2014	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2015	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2016	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2017	# ifdef NETDEV_FEAT_CHANGE
2018	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2019	# endif
2020	}
2021	return pszEvent;
2022	}
2023	#endif /* LOG_ENABLED */
2024
2025	/**
2026	* Callback for listening to netdevice events.
2027	*
2028	* This works the rediscovery, clean up on unregistration, promiscuity on
2029	* up/down, and GSO feature changes from ethtool.
2030	*
2031	* @returns NOTIFY_OK
2032	* @param self Pointer to our notifier registration block.
2033	* @param ulEventType The event.
2034	* @param ptr Event specific, but it is usually the device it
2035	* relates to.
2036	*/
2037	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2038
2039	{
2040	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2041	struct net_device pMyDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device );
2042	struct net_device *pDev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2043	int rc = NOTIFY_OK;
2044
2045	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2046	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, pMyDev));
2047
2048	if (ulEventType == NETDEV_REGISTER)
2049	{
2050	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 24) /* cgroups/namespaces introduced */
2051	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 26)
2052	# define VBOX_DEV_NET(dev) dev_net(dev)
2053	# define VBOX_NET_EQ(n1, n2) net_eq((n1), (n2))
2054	# else
2055	# define VBOX_DEV_NET(dev) ((dev)->nd_net)
2056	# define VBOX_NET_EQ(n1, n2) ((n1) == (n2))
2057	# endif
2058	struct net *pMyNet = current->nsproxy->net_ns;
2059	struct net *pDevNet = VBOX_DEV_NET(pDev);
2060
2061	if (VBOX_NET_EQ(pDevNet, pMyNet))
2062	#endif /* namespaces */
2063	{
2064	if (strcmp(pDev->name, pThis->szName) == 0)
2065	{
2066	vboxNetFltLinuxAttachToInterface(pThis, pDev);
2067	}
2068	}
2069	}
2070	else
2071	{
2072	if (pDev == pMyDev)
2073	{
2074	switch (ulEventType)
2075	{
2076	case NETDEV_UNREGISTER:
2077	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2078	break;
2079	case NETDEV_UP:
2080	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2081	break;
2082	case NETDEV_GOING_DOWN:
2083	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2084	break;
2085	case NETDEV_CHANGEMTU:
2086	rc = vboxNetFltLinuxDeviceMtuChange(pThis, pDev);
2087	break;
2088	case NETDEV_CHANGENAME:
2089	break;
2090	#ifdef NETDEV_FEAT_CHANGE
2091	case NETDEV_FEAT_CHANGE:
2092	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2093	break;
2094	#endif
2095	}
2096	}
2097	}
2098
2099	return rc;
2100	}
2101
2102	/*
2103	* Initial enumeration of netdevs. Called with NETDEV_REGISTER by
2104	* register_netdevice_notifier() under rtnl lock.
2105	*/
2106	static int vboxNetFltLinuxEnumeratorCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2107	{
2108	PVBOXNETFLTINS pThis = ((PVBOXNETFLTNOTIFIER)self)->pThis;
2109	struct net_device *dev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2110	struct in_device *in_dev;
2111	struct inet6_dev *in6_dev;
2112
2113	if (ulEventType != NETDEV_REGISTER)
2114	return NOTIFY_OK;
2115
2116	if (RT_UNLIKELY(pThis->pSwitchPort->pfnNotifyHostAddress == NULL))
2117	return NOTIFY_OK;
2118
2119	/*
2120	* IPv4
2121	*/
2122	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
2123	in_dev = __in_dev_get_rtnl(dev);
2124	#else
2125	in_dev = __in_dev_get(dev);
2126	#endif
2127	if (in_dev != NULL)
2128	{
2129	for_ifa(in_dev) {
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	} endfor_ifa(in_dev);
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 LINUX_VERSION_CODE >= KERNEL_VERSION(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) : "<???>",
2192	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2193	pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<???>",
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) : "<???>",
2237	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2238	pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<???>",
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	err = netif_rx_ni(pBuf);
2315	if (err)
2316	rc = RTErrConvertFromErrno(err);
2317	}
2318	else
2319	rc = VERR_NO_MEMORY;
2320	}
2321
2322	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2323	}
2324
2325	IPRT_LINUX_RESTORE_EFL_AC();
2326	return rc;
2327	}
2328
2329
2330	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2331	{
2332	struct net_device *pDev;
2333	IPRT_LINUX_SAVE_EFL_AC();
2334
2335	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%RTbool, fDisablePromiscuous=%RTbool\n",
2336	pThis, pThis->szName, fActive, pThis->fDisablePromiscuous));
2337
2338	if (pThis->fDisablePromiscuous)
2339	return;
2340
2341	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2342	if (pDev)
2343	{
2344	/*
2345	* This api is a bit weird, the best reference is the code.
2346	*
2347	* Also, we have a bit or race conditions wrt the maintenance of
2348	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2349	*/
2350	#ifdef LOG_ENABLED
2351	u_int16_t fIf;
2352	unsigned const cPromiscBefore = pDev->promiscuity;
2353	#endif
2354	if (fActive)
2355	{
2356	Assert(!pThis->u.s.fPromiscuousSet);
2357
2358	rtnl_lock();
2359	dev_set_promiscuity(pDev, 1);
2360	rtnl_unlock();
2361	pThis->u.s.fPromiscuousSet = true;
2362	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2363	}
2364	else
2365	{
2366	if (pThis->u.s.fPromiscuousSet)
2367	{
2368	rtnl_lock();
2369	dev_set_promiscuity(pDev, -1);
2370	rtnl_unlock();
2371	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2372	}
2373	pThis->u.s.fPromiscuousSet = false;
2374
2375	#ifdef LOG_ENABLED
2376	fIf = dev_get_flags(pDev);
2377	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2378	#endif
2379	}
2380
2381	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2382	}
2383	IPRT_LINUX_RESTORE_EFL_AC();
2384	}
2385
2386
2387	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2388	{
2389	/*
2390	* Remove packet handler when we get disconnected from internal switch as
2391	* we don't want the handler to forward packets to disconnected switch.
2392	*/
2393	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
2394	{
2395	IPRT_LINUX_SAVE_EFL_AC();
2396	dev_remove_pack(&pThis->u.s.PacketType);
2397	Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
2398	IPRT_LINUX_RESTORE_EFL_AC();
2399	}
2400	return VINF_SUCCESS;
2401	}
2402
2403
2404	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2405	{
2406	IPRT_LINUX_SAVE_EFL_AC();
2407
2408	/*
2409	* Report the GSO capabilities of the host and device (if connected).
2410	* Note! No need to mark ourselves busy here.
2411	*/
2412	/** @todo duplicate work here now? Attach */
2413	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2414	Log3(("vboxNetFltOsConnectIt: reporting host tso tso6 ufo\n"));
2415	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2416	0
2417	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2418	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2419	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2420	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2421	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2422	# endif
2423	, INTNETTRUNKDIR_HOST);
2424
2425	#endif
2426	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2427
2428	IPRT_LINUX_RESTORE_EFL_AC();
2429	return VINF_SUCCESS;
2430	}
2431
2432
2433	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2434	{
2435	struct net_device *pDev;
2436	bool fRegistered;
2437	IPRT_LINUX_SAVE_EFL_AC();
2438
2439	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
2440	vboxNetFltLinuxUnhookDev(pThis, NULL);
2441	#endif
2442
2443	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2444	* unlikely, but none the less). Since it doesn't actually update the
2445	* state (just reads it), it is likely to panic in some interesting
2446	* ways. */
2447
2448	RTSpinlockAcquire(pThis->hSpinlock);
2449	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2450	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
2451	RTSpinlockRelease(pThis->hSpinlock);
2452
2453	if (fRegistered)
2454	{
2455	vboxNetFltSetLinkState(pThis, pDev, false);
2456
2457	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2458	skb_queue_purge(&pThis->u.s.XmitQueue);
2459	#endif
2460	Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
2461	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
2462	pDev, pDev->name,
2463	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
2464	netdev_refcnt_read(pDev)
2465	#else
2466	atomic_read(&pDev->refcnt)
2467	#endif
2468	));
2469	dev_put(pDev);
2470	}
2471
2472	unregister_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2473	unregister_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2474
2475	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2476	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2477	module_put(THIS_MODULE);
2478
2479	IPRT_LINUX_RESTORE_EFL_AC();
2480	}
2481
2482
2483	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2484	{
2485	int err;
2486	IPRT_LINUX_SAVE_EFL_AC();
2487	NOREF(pvContext);
2488
2489	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2490	err = register_netdevice_notifier(&pThis->u.s.Notifier);
2491	if (err)
2492	{
2493	IPRT_LINUX_RESTORE_EFL_AC();
2494	return VERR_INTNET_FLT_IF_FAILED;
2495	}
2496	if (!pThis->u.s.fRegistered)
2497	{
2498	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2499	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2500	IPRT_LINUX_RESTORE_EFL_AC();
2501	return VERR_INTNET_FLT_IF_NOT_FOUND;
2502	}
2503
2504	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2505	if ( pThis->fDisconnectedFromHost
2506	\|\| !try_module_get(THIS_MODULE))
2507	{
2508	IPRT_LINUX_RESTORE_EFL_AC();
2509	return VERR_INTNET_FLT_IF_FAILED;
2510	}
2511
2512	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2513	{
2514	VBOXNETFLTNOTIFIER Enumerator;
2515
2516	/*
2517	* register_inetaddr_notifier() and register_inet6addr_notifier()
2518	* do not call the callback for existing devices. Enumerating
2519	* all network devices explicitly is a bit of an ifdef mess,
2520	* so co-opt register_netdevice_notifier() to do that for us.
2521	*/
2522	RT_ZERO(Enumerator);
2523	Enumerator.Notifier.notifier_call = vboxNetFltLinuxEnumeratorCallback;
2524	Enumerator.pThis = pThis;
2525
2526	err = register_netdevice_notifier(&Enumerator.Notifier);
2527	if (err)
2528	{
2529	LogRel(("%s: failed to enumerate network devices: error %d\n", __FUNCTION__, err));
2530	IPRT_LINUX_RESTORE_EFL_AC();
2531	return VINF_SUCCESS;
2532	}
2533
2534	unregister_netdevice_notifier(&Enumerator.Notifier);
2535
2536	pThis->u.s.NotifierIPv4.notifier_call = vboxNetFltLinuxNotifierIPv4Callback;
2537	err = register_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2538	if (err)
2539	LogRel(("%s: failed to register IPv4 notifier: error %d\n", __FUNCTION__, err));
2540
2541	pThis->u.s.NotifierIPv6.notifier_call = vboxNetFltLinuxNotifierIPv6Callback;
2542	err = register_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2543	if (err)
2544	LogRel(("%s: failed to register IPv6 notifier: error %d\n", __FUNCTION__, err));
2545	}
2546
2547	IPRT_LINUX_RESTORE_EFL_AC();
2548	return VINF_SUCCESS;
2549	}
2550
2551	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2552	{
2553	IPRT_LINUX_SAVE_EFL_AC();
2554
2555	/*
2556	* Init the linux specific members.
2557	*/
2558	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2559	pThis->u.s.fRegistered = false;
2560	pThis->u.s.fPromiscuousSet = false;
2561	pThis->u.s.fPacketHandler = false;
2562	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2563	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2564	skb_queue_head_init(&pThis->u.s.XmitQueue);
2565	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2566	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2567	# else
2568	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2569	# endif
2570	#endif
2571
2572	IPRT_LINUX_RESTORE_EFL_AC();
2573	return VINF_SUCCESS;
2574	}
2575
2576
2577	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2578	{
2579	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2580	}
2581
2582
2583	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
2584	{
2585	/* Nothing to do */
2586	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2587	return VINF_SUCCESS;
2588	}
2589
2590
2591	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2592	{
2593	/* Nothing to do */
2594	NOREF(pThis); NOREF(pvIfData);
2595	return VINF_SUCCESS;
2596	}
2597

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 69496

Download in other formats: