VBoxNetFlt-linux.c@ 85698

Last change on this file since 85698 was 85698, checked in by vboxsync, 4 years ago
IPRT,lnx-kmods: Use new linux kernel version checking macros. Moved them to separate wrapper header.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 90.3 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 85698 2020-08-11 17:05:29Z 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)
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 RTLNX_VER_MIN(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 /* RTLNX_VER_MIN(2,6,18) */
199
200	#if RTLNX_VER_MIN(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	#if defined(CONFIG_SUSE_VERSION)
215	# if CONFIG_SUSE_VERSION == 15 && CONFIG_SUSE_PATCHLEVEL == 2
216	# define OPENSUSE_152
217	# endif
218	#endif
219
220
221	/*********************************************************************************************************************************
222	* Internal Functions *
223	*********************************************************************************************************************************/
224	static int __init VBoxNetFltLinuxInit(void);
225	static void __exit VBoxNetFltLinuxUnload(void);
226	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
227
228
229	/*********************************************************************************************************************************
230	* Global Variables *
231	*********************************************************************************************************************************/
232	/**
233	* The (common) global data.
234	*/
235	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
236
237	module_init(VBoxNetFltLinuxInit);
238	module_exit(VBoxNetFltLinuxUnload);
239
240	MODULE_AUTHOR(VBOX_VENDOR);
241	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
242	MODULE_LICENSE("GPL");
243	#ifdef MODULE_VERSION
244	MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV) " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
245	#endif
246
247
248	#if RTLNX_VER_MAX(2,6,12) && defined(LOG_ENABLED)
249	unsigned dev_get_flags(const struct net_device *dev)
250	{
251	unsigned flags;
252
253	flags = (dev->flags & ~(IFF_PROMISC \|
254	IFF_ALLMULTI \|
255	IFF_RUNNING)) \|
256	(dev->gflags & (IFF_PROMISC \|
257	IFF_ALLMULTI));
258
259	if (netif_running(dev) && netif_carrier_ok(dev))
260	flags \|= IFF_RUNNING;
261
262	return flags;
263	}
264	#endif /* RTLNX_VER_MAX(2,6,12) */
265
266
267	/**
268	* Initialize module.
269	*
270	* @returns appropriate status code.
271	*/
272	static int __init VBoxNetFltLinuxInit(void)
273	{
274	int rc;
275	/*
276	* Initialize IPRT.
277	*/
278	rc = RTR0Init(0);
279	if (RT_SUCCESS(rc))
280	{
281	Log(("VBoxNetFltLinuxInit\n"));
282
283	/*
284	* Initialize the globals and connect to the support driver.
285	*
286	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
287	* for establishing the connect to the support driver.
288	*/
289	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
290	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
291	if (RT_SUCCESS(rc))
292	{
293	LogRel(("VBoxNetFlt: Successfully started.\n"));
294	return 0;
295	}
296
297	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
298	RTR0Term();
299	}
300	else
301	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
302
303	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
304	return -RTErrConvertToErrno(rc);
305	}
306
307
308	/**
309	* Unload the module.
310	*
311	* @todo We have to prevent this if we're busy!
312	*/
313	static void __exit VBoxNetFltLinuxUnload(void)
314	{
315	int rc;
316	Log(("VBoxNetFltLinuxUnload\n"));
317	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
318
319	/*
320	* Undo the work done during start (in reverse order).
321	*/
322	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
323	AssertRC(rc); NOREF(rc);
324
325	RTR0Term();
326
327	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
328
329	Log(("VBoxNetFltLinuxUnload - done\n"));
330	}
331
332
333	/**
334	* We filter traffic from the host to the internal network
335	* before it reaches the NIC driver.
336	*
337	* The current code uses a very ugly hack overriding hard_start_xmit
338	* callback in the device structure, but it has been shown to give us a
339	* performance boost of 60-100% though. Eventually we have to find some
340	* less hacky way of getting this job done.
341	*/
342	#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
343
344	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
345
346	# if RTLNX_VER_MAX(2,6,29)
347
348	# include <linux/ethtool.h>
349
350	typedef struct ethtool_ops OVR_OPSTYPE;
351	# define OVR_OPS ethtool_ops
352	# define OVR_XMIT pfnStartXmit
353
354	# else /* RTLNX_VER_MIN(2,6,29) */
355
356	typedef struct net_device_ops OVR_OPSTYPE;
357	# define OVR_OPS netdev_ops
358	# define OVR_XMIT pOrgOps->ndo_start_xmit
359
360	# endif /* RTLNX_VER_MIN(2,6,29) */
361
362	/**
363	* The overridden net_device_ops of the device we're attached to.
364	*
365	* As there is no net_device_ops structure in pre-2.6.29 kernels we override
366	* ethtool_ops instead along with hard_start_xmit callback in net_device
367	* structure.
368	*
369	* This is a very dirty hack that was created to explore how much we can improve
370	* the host to guest transfers by not CC'ing the NIC. It turns out to be
371	* the only way to filter outgoing packets for devices without TX queue.
372	*/
373	typedef struct VBoxNetDeviceOpsOverride
374	{
375	/** Our overridden ops. */
376	OVR_OPSTYPE Ops;
377	/** Magic word. */
378	uint32_t u32Magic;
379	/** Pointer to the original ops. */
380	OVR_OPSTYPE const *pOrgOps;
381	# if RTLNX_VER_MAX(2,6,29)
382	/** Pointer to the original hard_start_xmit function. */
383	int (pfnStartXmit)(struct sk_buff pSkb, struct net_device *pDev);
384	# endif /* RTLNX_VER_MAX(2,6,29) */
385	/** Pointer to the net filter instance. */
386	PVBOXNETFLTINS pVBoxNetFlt;
387	/** The number of filtered packages. */
388	uint64_t cFiltered;
389	/** The total number of packets */
390	uint64_t cTotal;
391	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
392	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
393	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
394
395	/**
396	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
397	* because they belong on the internal network.
398	*
399	* @returns NETDEV_TX_XXX.
400	* @param pSkb The socket buffer to transmit.
401	* @param pDev The net device.
402	*/
403	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
404	{
405	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
406	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
407	PCRTNETETHERHDR pEtherHdr;
408	PINTNETTRUNKSWPORT pSwitchPort;
409	uint32_t cbHdrs;
410
411
412	/*
413	* Validate the override structure.
414	*
415	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
416	* to be production quality code, we would have to be much more
417	* careful here and avoid the race.
418	*/
419	if ( !VALID_PTR(pOverride)
420	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
421	# if RTLNX_VER_MIN(2,6,29)
422	\|\| !VALID_PTR(pOverride->pOrgOps)
423	# endif /* RTLNX_VER_MIN(2,6,29) */
424	)
425	{
426	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
427	dev_kfree_skb(pSkb);
428	return NETDEV_TX_OK;
429	}
430	pOverride->cTotal++;
431
432	/*
433	* Do the filtering base on the default OUI of our virtual NICs
434	*
435	* Note! In a real solution, we would ask the switch whether the
436	* destination MAC is 100% to be on the internal network and then
437	* drop it.
438	*/
439	cbHdrs = skb_headlen(pSkb);
440	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
441	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
442	if ( pEtherHdr
443	&& VALID_PTR(pOverride->pVBoxNetFlt)
444	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
445	&& VALID_PTR(pSwitchPort)
446	&& cbHdrs >= 6)
447	{
448	INTNETSWDECISION enmDecision;
449
450	/** @todo consider reference counting, etc. */
451	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
452	if (enmDecision == INTNETSWDECISION_INTNET)
453	{
454	dev_kfree_skb(pSkb);
455	pOverride->cFiltered++;
456	return NETDEV_TX_OK;
457	}
458	}
459
460	return pOverride->OVR_XMIT(pSkb, pDev);
461	}
462
463	/**
464	* Hooks the device ndo_start_xmit operation of the device.
465	*
466	* @param pThis The net filter instance.
467	* @param pDev The net device.
468	*/
469	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
470	{
471	PVBOXNETDEVICEOPSOVERRIDE pOverride;
472
473	/* Cancel override if ethtool_ops is missing (host-only case, @bugref{5712}) */
474	if (!VALID_PTR(pDev->OVR_OPS))
475	return;
476	pOverride = RTMemAlloc(sizeof(*pOverride));
477	if (!pOverride)
478	return;
479	pOverride->pOrgOps = pDev->OVR_OPS;
480	pOverride->Ops = *pDev->OVR_OPS;
481	# if RTLNX_VER_MAX(2,6,29)
482	pOverride->pfnStartXmit = pDev->hard_start_xmit;
483	# else /* RTLNX_VER_MIN(2,6,29) */
484	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
485	# endif /* RTLNX_VER_MIN(2,6,29) */
486	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
487	pOverride->cTotal = 0;
488	pOverride->cFiltered = 0;
489	pOverride->pVBoxNetFlt = pThis;
490
491	RTSpinlockAcquire(pThis->hSpinlock); /* (this isn't necessary, but so what) */
492	ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
493	# if RTLNX_VER_MAX(2,6,29)
494	ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
495	# endif /* RTLNX_VER_MAX(2,6,29) */
496	RTSpinlockRelease(pThis->hSpinlock);
497	}
498
499	/**
500	* Undos what vboxNetFltLinuxHookDev did.
501	*
502	* @param pThis The net filter instance.
503	* @param pDev The net device. Can be NULL, in which case
504	* we'll try retrieve it from @a pThis.
505	*/
506	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
507	{
508	PVBOXNETDEVICEOPSOVERRIDE pOverride;
509
510	RTSpinlockAcquire(pThis->hSpinlock);
511	if (!pDev)
512	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
513	if (VALID_PTR(pDev))
514	{
515	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
516	if ( VALID_PTR(pOverride)
517	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
518	&& VALID_PTR(pOverride->pOrgOps)
519	)
520	{
521	# if RTLNX_VER_MAX(2,6,29)
522	ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
523	# endif /* RTLNX_VER_MAX(2,6,29) */
524	ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
525	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
526	}
527	else
528	pOverride = NULL;
529	}
530	else
531	pOverride = NULL;
532	RTSpinlockRelease(pThis->hSpinlock);
533
534	if (pOverride)
535	{
536	printk("vboxnetflt: %llu out of %llu packets were not sent (directed to host)\n", pOverride->cFiltered, pOverride->cTotal);
537	RTMemFree(pOverride);
538	}
539	}
540
541	#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
542
543
544	/**
545	* Reads and retains the host interface handle.
546	*
547	* @returns The handle, NULL if detached.
548	* @param pThis
549	*/
550	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
551	{
552	#if 0
553	struct net_device *pDev = NULL;
554
555	Log(("vboxNetFltLinuxRetainNetDev\n"));
556	/*
557	* Be careful here to avoid problems racing the detached callback.
558	*/
559	RTSpinlockAcquire(pThis->hSpinlock);
560	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
561	{
562	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
563	if (pDev)
564	{
565	dev_hold(pDev);
566	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
567	pDev, pDev->name,
568	#if RTLNX_VER_MIN(2,6,37)
569	netdev_refcnt_read(pDev)
570	#else
571	atomic_read(&pDev->refcnt)
572	#endif
573	));
574	}
575	}
576	RTSpinlockRelease(pThis->hSpinlock);
577
578	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
579	return pDev;
580	#else
581	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
582	#endif
583	}
584
585
586	/**
587	* Release the host interface handle previously retained
588	* by vboxNetFltLinuxRetainNetDev.
589	*
590	* @param pThis The instance.
591	* @param pDev The vboxNetFltLinuxRetainNetDev
592	* return value, NULL is fine.
593	*/
594	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
595	{
596	#if 0
597	Log(("vboxNetFltLinuxReleaseNetDev\n"));
598	NOREF(pThis);
599	if (pDev)
600	{
601	dev_put(pDev);
602	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
603	pDev, pDev->name,
604	#if RTLNX_VER_MIN(2,6,37)
605	netdev_refcnt_read(pDev)
606	#else
607	atomic_read(&pDev->refcnt)
608	#endif
609	));
610	}
611	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
612	#endif
613	}
614
615	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
616	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
617
618	/**
619	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
620	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
621	*
622	* @returns true / false accordingly.
623	* @param pBuf The sk_buff.
624	*/
625	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
626	{
627	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
628	}
629
630
631	/**
632	* Checks whether this SG list contains a GSO packet.
633	*
634	* @returns true / false accordingly.
635	* @param pSG The (scatter/)gather list.
636	*/
637	DECLINLINE(bool) vboxNetFltLinuxIsGso(PINTNETSG pSG)
638	{
639	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
640	return !((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type == PDMNETWORKGSOTYPE_INVALID);
641	#else /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
642	return false;
643	#endif /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
644	}
645
646
647	/**
648	* Find out the frame size (of a single segment in case of GSO frames).
649	*
650	* @returns the frame size.
651	* @param pSG The (scatter/)gather list.
652	*/
653	DECLINLINE(uint32_t) vboxNetFltLinuxFrameSize(PINTNETSG pSG)
654	{
655	uint16_t u16Type = 0;
656	uint32_t cbVlanTag = 0;
657	if (pSG->aSegs[0].cb >= sizeof(RTNETETHERHDR))
658	u16Type = RT_BE2H_U16(((PCRTNETETHERHDR)pSG->aSegs[0].pv)->EtherType);
659	else if (pSG->cbTotal >= sizeof(RTNETETHERHDR))
660	{
661	uint32_t off = RT_UOFFSETOF(RTNETETHERHDR, EtherType);
662	uint32_t i;
663	for (i = 0; i < pSG->cSegsUsed; ++i)
664	{
665	if (off <= pSG->aSegs[i].cb)
666	{
667	if (off + sizeof(uint16_t) <= pSG->aSegs[i].cb)
668	u16Type = RT_BE2H_U16((uint16_t )((uintptr_t)pSG->aSegs[i].pv + off));
669	else if (i + 1 < pSG->cSegsUsed)
670	u16Type = RT_BE2H_U16( ((uint16_t)( ((uint8_t *)pSG->aSegs[i].pv)[off] ) << 8)
671	+ (uint8_t )pSG->aSegs[i + 1].pv); /* ASSUMES no empty segments! */
672	/* else: frame is too short. */
673	break;
674	}
675	off -= pSG->aSegs[i].cb;
676	}
677	}
678	if (u16Type == RTNET_ETHERTYPE_VLAN)
679	cbVlanTag = 4;
680	return (vboxNetFltLinuxIsGso(pSG) ? (uint32_t)pSG->GsoCtx.cbMaxSeg + pSG->GsoCtx.cbHdrsTotal : pSG->cbTotal) - cbVlanTag;
681	}
682
683
684	/**
685	* Internal worker that create a linux sk_buff for a
686	* (scatter/)gather list.
687	*
688	* @returns Pointer to the sk_buff.
689	* @param pThis The instance.
690	* @param pSG The (scatter/)gather list.
691	* @param fDstWire Set if the destination is the wire.
692	*/
693	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
694	{
695	struct sk_buff *pPkt;
696	struct net_device *pDev;
697	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
698	unsigned fGsoType = 0;
699	#endif
700
701	if (pSG->cbTotal == 0)
702	{
703	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
704	return NULL;
705	}
706	Log5(("VBoxNetFlt: Packet to %s of %d bytes (frame=%d).\n", fDstWire?"wire":"host", pSG->cbTotal, vboxNetFltLinuxFrameSize(pSG)));
707	if (fDstWire && (vboxNetFltLinuxFrameSize(pSG) > ASMAtomicReadU32(&pThis->u.s.cbMtu) + 14))
708	{
709	static bool s_fOnce = true;
710	if (s_fOnce)
711	{
712	s_fOnce = false;
713	printk("VBoxNetFlt: Dropped over-sized packet (%d bytes) coming from internal network.\n", vboxNetFltLinuxFrameSize(pSG));
714	}
715	return NULL;
716	}
717
718	/** @todo We should use fragments mapping the SG buffers with large packets.
719	* 256 bytes seems to be the a threshold used a lot for this. It
720	* requires some nasty work on the intnet side though... */
721	/*
722	* Allocate a packet and copy over the data.
723	*/
724	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
725	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
726	if (RT_UNLIKELY(!pPkt))
727	{
728	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
729	pSG->pvUserData = NULL;
730	return NULL;
731	}
732	pPkt->dev = pDev;
733	pPkt->ip_summed = CHECKSUM_NONE;
734
735	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
736	skb_reserve(pPkt, NET_IP_ALIGN);
737
738	/* Copy the segments. */
739	skb_put(pPkt, pSG->cbTotal);
740	IntNetSgRead(pSG, pPkt->data);
741
742	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
743	/*
744	* Setup GSO if used by this packet.
745	*/
746	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
747	{
748	default:
749	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
750	/* fall thru */
751	case PDMNETWORKGSOTYPE_INVALID:
752	fGsoType = 0;
753	break;
754	case PDMNETWORKGSOTYPE_IPV4_TCP:
755	fGsoType = SKB_GSO_TCPV4;
756	break;
757	case PDMNETWORKGSOTYPE_IPV6_TCP:
758	fGsoType = SKB_GSO_TCPV6;
759	break;
760	}
761	if (fGsoType)
762	{
763	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
764
765	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
766	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
767	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
768
769	/*
770	* We need to set checksum fields even if the packet goes to the host
771	* directly as it may be immediately forwarded by IP layer @bugref{5020}.
772	*/
773	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
774	pPkt->ip_summed = CHECKSUM_PARTIAL;
775	# if RTLNX_VER_MIN(2,6,22)
776	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
777	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
778	pPkt->csum_offset = RT_UOFFSETOF(RTNETTCP, th_sum);
779	else
780	pPkt->csum_offset = RT_UOFFSETOF(RTNETUDP, uh_sum);
781	# else
782	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
783	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
784	pPkt->csum = RT_UOFFSETOF(RTNETTCP, th_sum);
785	else
786	pPkt->csum = RT_UOFFSETOF(RTNETUDP, uh_sum);
787	# endif
788	if (!fDstWire)
789	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
790	}
791	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
792
793	/*
794	* Finish up the socket buffer.
795	*/
796	pPkt->protocol = eth_type_trans(pPkt, pDev);
797	if (fDstWire)
798	{
799	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
800
801	/* Restore ethernet header back. */
802	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
803	VBOX_SKB_RESET_MAC_HDR(pPkt);
804	}
805	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
806
807	return pPkt;
808	}
809
810
811	/**
812	* Return the offset where to start checksum computation from.
813	*
814	* @returns the offset relative to pBuf->data.
815	* @param pBuf The socket buffer.
816	*/
817	DECLINLINE(unsigned) vboxNetFltLinuxGetChecksumStartOffset(struct sk_buff *pBuf)
818	{
819	#if RTLNX_VER_MIN(2,6,38)
820	return skb_checksum_start_offset(pBuf);
821	#elif RTLNX_VER_MIN(2,6,22)
822	return pBuf->csum_start - skb_headroom(pBuf);
823	#else
824	unsigned char *pTransportHdr = pBuf->h.raw;
825	# if RTLNX_VER_MAX(2,6,19)
826	/*
827	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
828	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
829	* header length from the header itself and reconstruct 'h' pointer
830	* to TCP (or whatever) header.
831	*/
832	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
833	pTransportHdr = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
834	# endif
835	return pTransportHdr - pBuf->data;
836	#endif
837	}
838
839
840	/**
841	* Initializes a SG list from an sk_buff.
842	*
843	* @returns Number of segments.
844	* @param pThis The instance.
845	* @param pBuf The sk_buff.
846	* @param pSG The SG.
847	* @param cbExtra The number of bytes of extra space allocated immediately after the SG.
848	* @param cSegs The number of segments allocated for the SG.
849	* This should match the number in the mbuf exactly!
850	* @param fSrc The source of the frame.
851	* @param pGsoCtx Pointer to the GSO context if it's a GSO
852	* internal network frame. NULL if regular frame.
853	*/
854	static void vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
855	unsigned cbExtra, unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
856	{
857	int i;
858	NOREF(pThis);
859
860	#ifndef VBOXNETFLT_SG_SUPPORT
861	Assert(!skb_shinfo(pBuf)->frag_list);
862	#else /* VBOXNETFLT_SG_SUPPORT */
863	uint8_t pExtra = (uint8_t )&pSG->aSegs[cSegs];
864	unsigned cbConsumed = 0;
865	unsigned cbProduced = 0;
866
867	# if RTLNX_VER_MIN(2,6,27)
868	/* Restore VLAN tag stripped by host hardware */
869	if (vlan_tx_tag_present(pBuf))
870	{
871	uint8_t *pMac = pBuf->data;
872	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )pExtra;
873	Assert(ETH_ALEN * 2 + VLAN_HLEN <= cbExtra);
874	memmove(pVHdr, pMac, ETH_ALEN * 2);
875	/* Consume whole Ethernet header: 2 addresses + EtherType (see @bugref{8599}) */
876	cbConsumed += ETH_ALEN * 2 + sizeof(uint16_t);
877	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
878	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
879	pVHdr->h_vlan_encapsulated_proto = (uint16_t)(pMac + ETH_ALEN * 2);
880	cbProduced += VLAN_ETH_HLEN;
881	}
882	# endif /* RTLNX_VER_MIN(2,6,27) */
883
884	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
885	{
886	unsigned uCsumStartOffset = vboxNetFltLinuxGetChecksumStartOffset(pBuf);
887	unsigned uCsumStoreOffset = uCsumStartOffset + VBOX_SKB_CSUM_OFFSET(pBuf) - cbConsumed;
888	Log3(("cbConsumed=%u cbProduced=%u uCsumStartOffset=%u uCsumStoreOffset=%u\n",
889	cbConsumed, cbProduced, uCsumStartOffset, uCsumStoreOffset));
890	Assert(cbProduced + uCsumStoreOffset + sizeof(uint16_t) <= cbExtra);
891	/*
892	* We assume that the checksum is stored at the very end of the transport header
893	* so we will have all headers in a single fragment. If our assumption is wrong
894	* we may see suboptimal performance.
895	*/
896	memmove(pExtra + cbProduced,
897	pBuf->data + cbConsumed,
898	uCsumStoreOffset);
899	unsigned uChecksum = skb_checksum(pBuf, uCsumStartOffset, pBuf->len - uCsumStartOffset, 0);
900	(uint16_t)(pExtra + cbProduced + uCsumStoreOffset) = csum_fold(uChecksum);
901	cbProduced += uCsumStoreOffset + sizeof(uint16_t);
902	cbConsumed += uCsumStoreOffset + sizeof(uint16_t);
903	}
904	#endif /* VBOXNETFLT_SG_SUPPORT */
905
906	if (!pGsoCtx)
907	IntNetSgInitTempSegs(pSG, pBuf->len + cbProduced - cbConsumed, cSegs, 0 /cSegsUsed/);
908	else
909	IntNetSgInitTempSegsGso(pSG, pBuf->len + cbProduced - cbConsumed, cSegs, 0 /cSegsUsed/, pGsoCtx);
910
911	int iSeg = 0;
912	#ifdef VBOXNETFLT_SG_SUPPORT
913	if (cbProduced)
914	{
915	pSG->aSegs[iSeg].cb = cbProduced;
916	pSG->aSegs[iSeg].pv = pExtra;
917	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
918	}
919	pSG->aSegs[iSeg].cb = skb_headlen(pBuf) - cbConsumed;
920	pSG->aSegs[iSeg].pv = pBuf->data + cbConsumed;
921	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
922	Assert(iSeg <= pSG->cSegsAlloc);
923
924	# ifdef LOG_ENABLED
925	if (pBuf->data_len)
926	Log6((" kmap_atomic:"));
927	# endif /* LOG_ENABLED */
928	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
929	{
930	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
931	# if RTLNX_VER_MIN(5,4,0) \|\| defined(OPENSUSE_152)
932	pSG->aSegs[iSeg].cb = pFrag->bv_len;
933	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->bv_offset;
934	# else /* < KERNEL_VERSION(5, 4, 0) */
935	pSG->aSegs[iSeg].cb = pFrag->size;
936	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
937	# endif /* >= KERNEL_VERSION(5, 4, 0) */
938	Log6((" %p", pSG->aSegs[iSeg].pv));
939	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
940	Assert(iSeg <= pSG->cSegsAlloc);
941	}
942	struct sk_buff *pFragBuf;
943	for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
944	{
945	pSG->aSegs[iSeg].cb = skb_headlen(pFragBuf);
946	pSG->aSegs[iSeg].pv = pFragBuf->data;
947	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
948	Assert(iSeg <= pSG->cSegsAlloc);
949	for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
950	{
951	skb_frag_t *pFrag = &skb_shinfo(pFragBuf)->frags[i];
952	# if RTLNX_VER_MIN(5,4,0) \|\| defined(OPENSUSE_152)
953	pSG->aSegs[iSeg].cb = pFrag->bv_len;
954	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->bv_offset;
955	# else /* < KERNEL_VERSION(5, 4, 0) */
956	pSG->aSegs[iSeg].cb = pFrag->size;
957	pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
958	# endif /* >= KERNEL_VERSION(5, 4, 0) */
959	Log6((" %p", pSG->aSegs[iSeg].pv));
960	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
961	Assert(iSeg <= pSG->cSegsAlloc);
962	}
963	}
964	# ifdef LOG_ENABLED
965	if (pBuf->data_len)
966	Log6(("\n"));
967	# endif /* LOG_ENABLED */
968	#else
969	pSG->aSegs[iSeg].cb = pBuf->len;
970	pSG->aSegs[iSeg].pv = pBuf->data;
971	pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
972	#endif
973
974	pSG->cSegsUsed = iSeg;
975
976	#if 0
977	if (cbProduced)
978	{
979	LogRel(("vboxNetFltLinuxSkBufToSG: original packet dump:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
980	LogRel(("vboxNetFltLinuxSkBufToSG: cbConsumed=%u cbProduced=%u cbExtra=%u\n", cbConsumed, cbProduced, cbExtra));
981	uint32_t offset = 0;
982	for (i = 0; i < pSG->cSegsUsed; ++i)
983	{
984	LogRel(("vboxNetFltLinuxSkBufToSG: seg#%d (%d bytes, starting at 0x%x):\n%.*Rhxd\n",
985	i, pSG->aSegs[i].cb, offset, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
986	offset += pSG->aSegs[i].cb;
987	}
988	}
989	#endif
990
991	#ifdef PADD_RUNT_FRAMES_FROM_HOST
992	/*
993	* Add a trailer if the frame is too small.
994	*
995	* Since we're getting to the packet before it is framed, it has not
996	* yet been padded. The current solution is to add a segment pointing
997	* to a buffer containing all zeros and pray that works for all frames...
998	*/
999	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
1000	{
1001	Assert(pBuf->data_len == 0); /* Packets with fragments are never small! */
1002	static uint8_t const s_abZero[128] = {0};
1003
1004	AssertReturnVoid(iSeg < cSegs);
1005
1006	pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
1007	pSG->aSegs[iSeg].pv = (void *)&s_abZero[0];
1008	pSG->aSegs[iSeg++].cb = 60 - pSG->cbTotal;
1009	pSG->cbTotal = 60;
1010	pSG->cSegsUsed++;
1011	Assert(iSeg <= pSG->cSegsAlloc)
1012	}
1013	#endif
1014
1015	Log6(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
1016	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
1017	for (i = 0; i < pSG->cSegsUsed; i++)
1018	Log6(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
1019	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1020	}
1021
1022	/**
1023	* Packet handler; not really documented - figure it out yourself.
1024	*
1025	* @returns 0 or EJUSTRETURN - this is probably copy & pastry and thus wrong.
1026	*/
1027	#if RTLNX_VER_MIN(2,6,14)
1028	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1029	struct net_device *pSkbDev,
1030	struct packet_type *pPacketType,
1031	struct net_device *pOrigDev)
1032	#else
1033	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1034	struct net_device *pSkbDev,
1035	struct packet_type *pPacketType)
1036	#endif
1037	{
1038	PVBOXNETFLTINS pThis;
1039	struct net_device *pDev;
1040	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1041	pBuf, pSkbDev, pPacketType));
1042	#if RTLNX_VER_MIN(2,6,18)
1043	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",
1044	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));
1045	# if RTLNX_VER_MIN(2,6,22)
1046	Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1047	# endif
1048	#else
1049	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",
1050	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));
1051	#endif
1052	/*
1053	* Drop it immediately?
1054	*/
1055	if (!pBuf)
1056	return 0;
1057
1058	if (pBuf->pkt_type == PACKET_LOOPBACK)
1059	{
1060	/*
1061	* We are not interested in loopbacked packets as they will always have
1062	* another copy going to the wire.
1063	*/
1064	Log2(("vboxNetFltLinuxPacketHandler: dropped loopback packet (cb=%u)\n", pBuf->len));
1065	dev_kfree_skb(pBuf); /* We must 'consume' all packets we get (@bugref{6539})! */
1066	return 0;
1067	}
1068
1069	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1070	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1071	if (pDev != pSkbDev)
1072	{
1073	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1074	kfree_skb(pBuf); /* This is a failure, so we use kfree_skb instead of dev_kfree_skb. */
1075	return 0;
1076	}
1077
1078	Log6(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1079	if (vboxNetFltLinuxSkBufIsOur(pBuf))
1080	{
1081	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1082	dev_kfree_skb(pBuf);
1083	return 0;
1084	}
1085
1086	#ifndef VBOXNETFLT_SG_SUPPORT
1087	{
1088	/*
1089	* Get rid of fragmented packets, they cause too much trouble.
1090	*/
1091	unsigned int uMacLen = pBuf->mac_len;
1092	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1093	dev_kfree_skb(pBuf);
1094	if (!pCopy)
1095	{
1096	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1097	return 0;
1098	}
1099	pBuf = pCopy;
1100	/* Somehow skb_copy ignores mac_len */
1101	pBuf->mac_len = uMacLen;
1102	# if RTLNX_VER_MIN(2,6,27)
1103	/* Restore VLAN tag stripped by host hardware */
1104	if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
1105	{
1106	uint8_t pMac = (uint8_t)skb_mac_header(pBuf);
1107	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )(pMac - VLAN_HLEN);
1108	memmove(pVHdr, pMac, ETH_ALEN * 2);
1109	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
1110	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
1111	pBuf->mac_header -= VLAN_HLEN;
1112	pBuf->mac_len += VLAN_HLEN;
1113	}
1114	# endif /* RTLNX_VER_MIN(2,6,27) */
1115
1116	# if RTLNX_VER_MIN(2,6,18)
1117	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",
1118	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));
1119	# if RTLNX_VER_MIN(2,6,22)
1120	Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1121	# endif /* RTLNX_VER_MIN(2,6,22) */
1122	# else /* RTLNX_VER_MAX(2,6,18) */
1123	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",
1124	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));
1125	# endif /* RTLNX_VER_MAX(2,6,18) */
1126	}
1127	#endif /* !VBOXNETFLT_SG_SUPPORT */
1128
1129	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1130	/* Forward it to the internal network. */
1131	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1132	#else /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1133	/* Add the packet to transmit queue and schedule the bottom half. */
1134	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1135	schedule_work(&pThis->u.s.XmitTask);
1136	Log6(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1137	&pThis->u.s.XmitTask, pBuf));
1138	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1139
1140	/* It does not really matter what we return, it is ignored by the kernel. */
1141	return 0;
1142	}
1143
1144	/**
1145	* Calculate the number of INTNETSEG segments the socket buffer will need.
1146	*
1147	* @returns Segment count.
1148	* @param pBuf The socket buffer.
1149	* @param pcbTemp Where to store the number of bytes of the part
1150	* of the socket buffer that will be copied to
1151	* a temporary storage.
1152	*/
1153	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff pBuf, unsigned pcbTemp)
1154	{
1155	*pcbTemp = 0;
1156	#ifdef VBOXNETFLT_SG_SUPPORT
1157	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1158	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1159	{
1160	*pcbTemp = vboxNetFltLinuxGetChecksumStartOffset(pBuf) + VBOX_SKB_CSUM_OFFSET(pBuf) + sizeof(uint16_t);
1161	}
1162	# if RTLNX_VER_MIN(2,6,27)
1163	if (vlan_tx_tag_present(pBuf))
1164	{
1165	if (*pcbTemp)
1166	*pcbTemp += VLAN_HLEN;
1167	else
1168	*pcbTemp = VLAN_ETH_HLEN;
1169	}
1170	# endif /* RTLNX_VER_MIN(2,6,27) */
1171	if (*pcbTemp)
1172	++cSegs;
1173	struct sk_buff *pFrag;
1174	for (pFrag = skb_shinfo(pBuf)->frag_list; pFrag; pFrag = pFrag->next)
1175	{
1176	Log6(("vboxNetFltLinuxCalcSGSegments: frag=%p len=%d data_len=%d frags=%d frag_list=%p next=%p\n",
1177	pFrag, pFrag->len, pFrag->data_len, skb_shinfo(pFrag)->nr_frags, skb_shinfo(pFrag)->frag_list, pFrag->next));
1178	cSegs += 1 + skb_shinfo(pFrag)->nr_frags;
1179	}
1180	#else
1181	unsigned cSegs = 1;
1182	#endif
1183	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1184	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1185	if (pBuf->len < 60)
1186	cSegs++;
1187	#endif
1188	return cSegs;
1189	}
1190
1191
1192	/**
1193	* Destroy the intnet scatter / gather buffer created by
1194	* vboxNetFltLinuxSkBufToSG.
1195	*
1196	* @param pSG The (scatter/)gather list.
1197	* @param pBuf The original socket buffer that was used to create
1198	* the scatter/gather list.
1199	*/
1200	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG, struct sk_buff *pBuf)
1201	{
1202	#ifdef VBOXNETFLT_SG_SUPPORT
1203	int i, iSeg = 1; /* Skip non-paged part of SKB */
1204	/* Check if the extra buffer behind SG structure was used for modified packet header */
1205	if (pBuf->data != pSG->aSegs[0].pv)
1206	++iSeg; /* Skip it as well */
1207	# ifdef LOG_ENABLED
1208	if (pBuf->data_len)
1209	Log6(("kunmap_atomic:"));
1210	# endif /* LOG_ENABLED */
1211	/* iSeg now points to the first mapped fragment if there are any */
1212	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1213	{
1214	Log6((" %p", pSG->aSegs[iSeg].pv));
1215	VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1216	}
1217	struct sk_buff *pFragBuf;
1218	for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
1219	{
1220	++iSeg; /* Non-fragment (unmapped) portion of chained SKB */
1221	for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
1222	{
1223	Log6((" %p", pSG->aSegs[iSeg].pv));
1224	VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1225	}
1226	}
1227	# ifdef LOG_ENABLED
1228	if (pBuf->data_len)
1229	Log6(("\n"));
1230	# endif /* LOG_ENABLED */
1231	#endif
1232	NOREF(pSG);
1233	}
1234
1235	#ifdef LOG_ENABLED
1236	/**
1237	* Logging helper.
1238	*/
1239	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1240	{
1241	int i, offSeg;
1242	uint8_t pInt, pExt;
1243	static int iPacketNo = 1;
1244	iPacketNo += iIncrement;
1245	if (fEgress)
1246	{
1247	pExt = pSG->aSegs[0].pv;
1248	pInt = pExt + 6;
1249	}
1250	else
1251	{
1252	pInt = pSG->aSegs[0].pv;
1253	pExt = pInt + 6;
1254	}
1255	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1256	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1257	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1258	fEgress ? "-->" : "<--", pszWhere,
1259	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1260	pSG->cbTotal, iPacketNo));
1261	if (pSG->cSegsUsed == 1)
1262	{
1263	Log4(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1264	}
1265	else
1266	{
1267	for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
1268	{
1269	Log4(("-- segment %d at 0x%x (%d bytes)\n --\n%.*Rhxd\n",
1270	i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1271	offSeg += pSG->aSegs[i].cb;
1272	}
1273	}
1274	}
1275	#else
1276	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1277	#endif
1278
1279	#ifdef VBOXNETFLT_WITH_GSO_RECV
1280
1281	/**
1282	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1283	* GSO socket buffer without having to segment it.
1284	*
1285	* @returns true on success, false if needs segmenting.
1286	* @param pThis The net filter instance.
1287	* @param pSkb The GSO socket buffer.
1288	* @param fSrc The source.
1289	* @param pGsoCtx Where to return the GSO context on success.
1290	*/
1291	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1292	PPDMNETWORKGSO pGsoCtx)
1293	{
1294	PDMNETWORKGSOTYPE enmGsoType;
1295	uint16_t uEtherType;
1296	unsigned int cbTransport;
1297	unsigned int offTransport;
1298	unsigned int cbTransportHdr;
1299	unsigned uProtocol;
1300	union
1301	{
1302	RTNETIPV4 IPv4;
1303	RTNETIPV6 IPv6;
1304	RTNETTCP Tcp;
1305	uint8_t ab[40];
1306	uint16_t au16[40/2];
1307	uint32_t au32[40/4];
1308	} Buf;
1309
1310	/*
1311	* Check the GSO properties of the socket buffer and make sure it fits.
1312	*/
1313	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1314	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
1315	{
1316	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1317	return false;
1318	}
1319	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1320	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
1321	{
1322	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1323	return false;
1324	}
1325
1326	/*
1327	* Switch on the ethertype.
1328	*/
1329	uEtherType = pSkb->protocol;
1330	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1331	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1332	{
1333	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1334	if (puEtherType)
1335	uEtherType = *puEtherType;
1336	}
1337	switch (uEtherType)
1338	{
1339	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1340	{
1341	unsigned int cbHdr;
1342	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1343	if (RT_UNLIKELY(!pIPv4))
1344	{
1345	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1346	return false;
1347	}
1348
1349	cbHdr = pIPv4->ip_hl * 4;
1350	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1351	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1352	\|\| cbHdr > cbTransport ))
1353	{
1354	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1355	return false;
1356	}
1357	cbTransport -= cbHdr;
1358	offTransport = pSkb->mac_len + cbHdr;
1359	uProtocol = pIPv4->ip_p;
1360	if (uProtocol == RTNETIPV4_PROT_TCP)
1361	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1362	else if (uProtocol == RTNETIPV4_PROT_UDP)
1363	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1364	else /** @todo IPv6: 4to6 tunneling */
1365	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1366	break;
1367	}
1368
1369	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1370	{
1371	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1372	if (RT_UNLIKELY(!pIPv6))
1373	{
1374	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1375	return false;
1376	}
1377
1378	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1379	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1380	uProtocol = pIPv6->ip6_nxt;
1381	/** @todo IPv6: Dig our way out of the other headers. */
1382	if (uProtocol == RTNETIPV4_PROT_TCP)
1383	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1384	else if (uProtocol == RTNETIPV4_PROT_UDP)
1385	enmGsoType = PDMNETWORKGSOTYPE_IPV6_UDP;
1386	else
1387	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1388	break;
1389	}
1390
1391	default:
1392	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1393	return false;
1394	}
1395
1396	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1397	{
1398	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1399	return false;
1400	}
1401
1402	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1403	\|\| offTransport + cbTransport > pSkb->len
1404	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1405	{
1406	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1407	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1408	return false;
1409	}
1410
1411	/*
1412	* Check the TCP/UDP bits.
1413	*/
1414	if (uProtocol == RTNETIPV4_PROT_TCP)
1415	{
1416	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1417	if (RT_UNLIKELY(!pTcp))
1418	{
1419	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1420	return false;
1421	}
1422
1423	cbTransportHdr = pTcp->th_off * 4;
1424	pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1425	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1426	\|\| cbTransportHdr > cbTransport
1427	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1428	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1429	{
1430	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1431	return false;
1432	}
1433
1434	}
1435	else
1436	{
1437	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1438	cbTransportHdr = sizeof(RTNETUDP);
1439	pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1440	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1441	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1442	{
1443	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1444	return false;
1445	}
1446	}
1447
1448	/*
1449	* We're good, init the GSO context.
1450	*/
1451	pGsoCtx->u8Type = enmGsoType;
1452	pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1453	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1454	pGsoCtx->offHdr1 = pSkb->mac_len;
1455	pGsoCtx->offHdr2 = offTransport;
1456	pGsoCtx->u8Unused = 0;
1457
1458	return true;
1459	}
1460
1461	/**
1462	* Forward the socket buffer as a GSO internal network frame.
1463	*
1464	* @returns IPRT status code.
1465	* @param pThis The net filter instance.
1466	* @param pSkb The GSO socket buffer.
1467	* @param fSrc The source.
1468	* @param pGsoCtx Where to return the GSO context on success.
1469	*/
1470	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1471	{
1472	int rc;
1473	unsigned cbExtra;
1474	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb, &cbExtra);
1475	PINTNETSG pSG = (PINTNETSG)alloca(RT_UOFFSETOF_DYN(INTNETSG, aSegs[cSegs]) + cbExtra);
1476	if (RT_LIKELY(pSG))
1477	{
1478	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cbExtra, cSegs, fSrc, pGsoCtx);
1479
1480	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1481	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1482
1483	vboxNetFltLinuxDestroySG(pSG, pSkb);
1484	rc = VINF_SUCCESS;
1485	}
1486	else
1487	{
1488	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1489	rc = VERR_NO_MEMORY;
1490	}
1491	return rc;
1492	}
1493
1494	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1495
1496	/**
1497	* Worker for vboxNetFltLinuxForwardToIntNet.
1498	*
1499	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1500	* @param pThis The net filter instance.
1501	* @param pBuf The socket buffer.
1502	* @param fSrc The source.
1503	*/
1504	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1505	{
1506	int rc;
1507	unsigned cbExtra;
1508	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf, &cbExtra);
1509	PINTNETSG pSG = (PINTNETSG)alloca(RT_UOFFSETOF_DYN(INTNETSG, aSegs[cSegs]) + cbExtra);
1510	if (RT_LIKELY(pSG))
1511	{
1512	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cbExtra, cSegs, fSrc, NULL /pGsoCtx/);
1513
1514	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1515	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1516
1517	vboxNetFltLinuxDestroySG(pSG, pBuf);
1518	rc = VINF_SUCCESS;
1519	}
1520	else
1521	{
1522	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1523	rc = VERR_NO_MEMORY;
1524	}
1525	return rc;
1526	}
1527
1528
1529	/**
1530	* I won't disclose what I do, figure it out yourself, including pThis referencing.
1531	*
1532	* @param pThis The net filter instance.
1533	* @param pBuf The socket buffer.
1534	* @param fSrc Where the packet comes from.
1535	*/
1536	static void vboxNetFltLinuxForwardToIntNetInner(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1537	{
1538	#ifdef VBOXNETFLT_WITH_GSO
1539	if (skb_is_gso(pBuf))
1540	{
1541	PDMNETWORKGSO GsoCtx;
1542	Log6(("vboxNetFltLinuxForwardToIntNetInner: skb len=%u data_len=%u truesize=%u next=%p"
1543	" nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x ip_summed=%d\n",
1544	pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next,
1545	skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->gso_size,
1546	skb_shinfo(pBuf)->gso_segs, skb_shinfo(pBuf)->gso_type,
1547	skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, pBuf->ip_summed));
1548	#ifndef VBOXNETFLT_SG_SUPPORT
1549	if (RT_LIKELY(fSrc & INTNETTRUNKDIR_HOST))
1550	{
1551	/*
1552	* skb_gso_segment does the following. Do we need to do it as well?
1553	*/
1554	# if RTLNX_VER_MIN(2,6,22)
1555	skb_reset_mac_header(pBuf);
1556	pBuf->mac_len = pBuf->network_header - pBuf->mac_header;
1557	# else
1558	pBuf->mac.raw = pBuf->data;
1559	pBuf->mac_len = pBuf->nh.raw - pBuf->data;
1560	# endif
1561	}
1562	#endif /* !VBOXNETFLT_SG_SUPPORT */
1563	# ifdef VBOXNETFLT_WITH_GSO_RECV
1564	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1565	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1566	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1567	else
1568	# endif /* VBOXNETFLT_WITH_GSO_RECV */
1569	{
1570	/* Need to segment the packet */
1571	struct sk_buff *pNext;
1572	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1573	if (IS_ERR(pSegment))
1574	{
1575	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1576	return;
1577	}
1578
1579	for (; pSegment; pSegment = pNext)
1580	{
1581	Log6(("vboxNetFltLinuxForwardToIntNetInner: segment len=%u data_len=%u truesize=%u next=%p"
1582	" nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x\n",
1583	pSegment->len, pSegment->data_len, pSegment->truesize, pSegment->next,
1584	skb_shinfo(pSegment)->nr_frags, skb_shinfo(pSegment)->gso_size,
1585	skb_shinfo(pSegment)->gso_segs, skb_shinfo(pSegment)->gso_type,
1586	skb_shinfo(pSegment)->frag_list, pSegment->pkt_type));
1587	pNext = pSegment->next;
1588	pSegment->next = 0;
1589	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1590	dev_kfree_skb(pSegment);
1591	}
1592	}
1593	}
1594	else
1595	#endif /* VBOXNETFLT_WITH_GSO */
1596	{
1597	Log6(("vboxNetFltLinuxForwardToIntNetInner: ptk_type=%d ip_summed=%d len=%d"
1598	" data_len=%d headroom=%d hdr_len=%d csum_offset=%d\n",
1599	pBuf->pkt_type, pBuf->ip_summed, pBuf->len, pBuf->data_len, skb_headroom(pBuf),
1600	skb_headlen(pBuf), vboxNetFltLinuxGetChecksumStartOffset(pBuf)));
1601	#ifndef VBOXNETFLT_SG_SUPPORT
1602	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1603	{
1604	# if RTLNX_VER_MIN(2,6,19)
1605	int rc = VBOX_SKB_CHECKSUM_HELP(pBuf);
1606	# else
1607	/*
1608	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1609	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1610	* header length from the header itself and reconstruct 'h' pointer
1611	* to TCP (or whatever) header.
1612	*/
1613	unsigned char *tmp = pBuf->h.raw;
1614	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1615	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1616	int rc = VBOX_SKB_CHECKSUM_HELP(pBuf);
1617	/* Restore the original (wrong) pointer. */
1618	pBuf->h.raw = tmp;
1619	# endif
1620	if (rc)
1621	{
1622	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1623	return;
1624	}
1625	}
1626	#endif /* !VBOXNETFLT_SG_SUPPORT */
1627	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1628	}
1629	}
1630
1631
1632	/**
1633	* Temporarily adjust pBuf->data so it always points to the Ethernet header,
1634	* then forward it to the internal network.
1635	*
1636	* @param pThis The net filter instance.
1637	* @param pBuf The socket buffer. This is consumed by this function.
1638	*/
1639	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1640	{
1641	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1642
1643	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1644	{
1645	/*
1646	* The packet came from the wire and the driver has already consumed
1647	* mac header. We need to restore it back. Moreover, after we are
1648	* through with this skb we need to restore its original state!
1649	*/
1650	skb_push(pBuf, pBuf->mac_len);
1651	Log5(("vboxNetFltLinuxForwardToIntNet: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1652	pBuf->mac_len, pBuf->data, skb_mac_header(pBuf), skb_network_header(pBuf)));
1653	}
1654
1655	vboxNetFltLinuxForwardToIntNetInner(pThis, pBuf, fSrc);
1656
1657	/*
1658	* Restore the original state of skb as there are other handlers this skb
1659	* will be provided to.
1660	*/
1661	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1662	skb_pull(pBuf, pBuf->mac_len);
1663
1664	dev_kfree_skb(pBuf);
1665	}
1666
1667
1668	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1669	/**
1670	* Work queue handler that forwards the socket buffers queued by
1671	* vboxNetFltLinuxPacketHandler to the internal network.
1672	*
1673	* @param pWork The work queue.
1674	*/
1675	# if RTLNX_VER_MIN(2,6,20)
1676	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1677	# else
1678	static void vboxNetFltLinuxXmitTask(void *pWork)
1679	# endif
1680	{
1681	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1682	struct sk_buff *pBuf;
1683
1684	Log6(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1685
1686	/*
1687	* Active? Retain the instance and increment the busy counter.
1688	*/
1689	if (vboxNetFltTryRetainBusyActive(pThis))
1690	{
1691	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1692	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1693
1694	vboxNetFltRelease(pThis, true /* fBusy */);
1695	}
1696	else
1697	{
1698	/** @todo Shouldn't we just drop the packets here? There is little point in
1699	* making them accumulate when the VM is paused and it'll only waste
1700	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1701	* before start draining the packets (goes for the intnet ring buf
1702	* too)? */
1703	}
1704	}
1705	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1706
1707	/**
1708	* Reports the GSO capabilities of the hardware NIC.
1709	*
1710	* @param pThis The net filter instance. The caller hold a
1711	* reference to this.
1712	*/
1713	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1714	{
1715	#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1716	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1717	{
1718	struct net_device *pDev;
1719	PINTNETTRUNKSWPORT pSwitchPort;
1720	unsigned int fFeatures;
1721
1722	RTSpinlockAcquire(pThis->hSpinlock);
1723
1724	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1725	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1726	if (pDev)
1727	fFeatures = pDev->features;
1728	else
1729	fFeatures = 0;
1730
1731	RTSpinlockRelease(pThis->hSpinlock);
1732
1733	if (pThis->pSwitchPort)
1734	{
1735	/* Set/update the GSO capabilities of the NIC. */
1736	uint32_t fGsoCapabilites = 0;
1737	if (fFeatures & NETIF_F_TSO)
1738	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1739	if (fFeatures & NETIF_F_TSO6)
1740	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1741	Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1742	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1743	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : ""));
1744	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1745	}
1746
1747	vboxNetFltRelease(pThis, true /fBusy/);
1748	}
1749	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1750	}
1751
1752	/**
1753	* Helper that determines whether the host (ignoreing us) is operating the
1754	* interface in promiscuous mode or not.
1755	*/
1756	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1757	{
1758	bool fRc = false;
1759	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1760	if (pDev)
1761	{
1762	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1763	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1764	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1765	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1766	}
1767	return fRc;
1768	}
1769
1770	/**
1771	* Does this device needs link state change signaled?
1772	* Currently we need it for our own VBoxNetAdp and TAP.
1773	*/
1774	static bool vboxNetFltNeedsLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev)
1775	{
1776	if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1777	{
1778	struct ethtool_drvinfo Info;
1779
1780	memset(&Info, 0, sizeof(Info));
1781	Info.cmd = ETHTOOL_GDRVINFO;
1782	pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1783	Log3(("%s: driver=%.s version=%.s bus_info=%.*s\n",
1784	__FUNCTION__,
1785	sizeof(Info.driver), Info.driver,
1786	sizeof(Info.version), Info.version,
1787	sizeof(Info.bus_info), Info.bus_info));
1788
1789	if (!strncmp(Info.driver, "vboxnet", sizeof(Info.driver)))
1790	return true;
1791
1792	#if RTLNX_VER_MIN(2,6,36) /* TAP started doing carrier */
1793	return !strncmp(Info.driver, "tun", 4)
1794	&& !strncmp(Info.bus_info, "tap", 4);
1795	#endif
1796	}
1797
1798	return false;
1799	}
1800
1801	#if RTLNX_VER_MAX(2,6,18)
1802	DECLINLINE(void) netif_tx_lock_bh(struct net_device *pDev)
1803	{
1804	spin_lock_bh(&pDev->xmit_lock);
1805	}
1806
1807	DECLINLINE(void) netif_tx_unlock_bh(struct net_device *pDev)
1808	{
1809	spin_unlock_bh(&pDev->xmit_lock);
1810	}
1811	#endif
1812
1813	/**
1814	* Some devices need link state change when filter attaches/detaches
1815	* since the filter is their link in a sense.
1816	*/
1817	static void vboxNetFltSetLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1818	{
1819	if (vboxNetFltNeedsLinkState(pThis, pDev))
1820	{
1821	Log3(("%s: bringing device link %s\n",
1822	__FUNCTION__, fLinkUp ? "up" : "down"));
1823	netif_tx_lock_bh(pDev);
1824	if (fLinkUp)
1825	netif_carrier_on(pDev);
1826	else
1827	netif_carrier_off(pDev);
1828	netif_tx_unlock_bh(pDev);
1829	}
1830	}
1831
1832	/**
1833	* Internal worker for vboxNetFltLinuxNotifierCallback.
1834	*
1835	* @returns VBox status code.
1836	* @param pThis The instance.
1837	* @param pDev The device to attach to.
1838	*/
1839	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1840	{
1841	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1842
1843	/*
1844	* Retain and store the device.
1845	*/
1846	dev_hold(pDev);
1847
1848	RTSpinlockAcquire(pThis->hSpinlock);
1849	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1850	RTSpinlockRelease(pThis->hSpinlock);
1851
1852	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1853	pDev, pDev->name,
1854	#if RTLNX_VER_MIN(2,6,37)
1855	netdev_refcnt_read(pDev)
1856	#else
1857	atomic_read(&pDev->refcnt)
1858	#endif
1859	));
1860	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1861	pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1862
1863	/* Get the mac address while we still have a valid net_device reference. */
1864	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1865	/* Initialize MTU */
1866	pThis->u.s.cbMtu = pDev->mtu;
1867
1868	/*
1869	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1870	*/
1871	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1872	pThis->u.s.PacketType.dev = pDev;
1873	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1874	dev_add_pack(&pThis->u.s.PacketType);
1875	ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1876	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1877
1878	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1879	vboxNetFltLinuxHookDev(pThis, pDev);
1880	#endif
1881
1882	/*
1883	* Are we the "carrier" for this device (e.g. vboxnet or tap)?
1884	*/
1885	vboxNetFltSetLinkState(pThis, pDev, true);
1886
1887	/*
1888	* Set indicators that require the spinlock. Be abit paranoid about racing
1889	* the device notification handle.
1890	*/
1891	RTSpinlockAcquire(pThis->hSpinlock);
1892	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1893	if (pDev)
1894	{
1895	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1896	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1897	pDev = NULL; /* don't dereference it */
1898	}
1899	RTSpinlockRelease(pThis->hSpinlock);
1900
1901	/*
1902	* Report GSO capabilities
1903	*/
1904	Assert(pThis->pSwitchPort);
1905	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1906	{
1907	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1908	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1909	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1910	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1911	vboxNetFltRelease(pThis, true /fBusy/);
1912	}
1913
1914	LogRel(("VBoxNetFlt: attached to '%s' / %RTmac\n", pThis->szName, &pThis->u.s.MacAddr));
1915	return VINF_SUCCESS;
1916	}
1917
1918
1919	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1920	{
1921	bool fRegistered;
1922	Assert(!pThis->fDisconnectedFromHost);
1923
1924	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1925	vboxNetFltLinuxUnhookDev(pThis, pDev);
1926	#endif
1927
1928	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1929	{
1930	dev_remove_pack(&pThis->u.s.PacketType);
1931	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1932	}
1933
1934	RTSpinlockAcquire(pThis->hSpinlock);
1935	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1936	if (fRegistered)
1937	{
1938	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1939	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1940	}
1941	RTSpinlockRelease(pThis->hSpinlock);
1942
1943	if (fRegistered)
1944	{
1945	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1946	skb_queue_purge(&pThis->u.s.XmitQueue);
1947	#endif
1948	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1949	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1950	pDev, pDev->name,
1951	#if RTLNX_VER_MIN(2,6,37)
1952	netdev_refcnt_read(pDev)
1953	#else
1954	atomic_read(&pDev->refcnt)
1955	#endif
1956	));
1957	dev_put(pDev);
1958	}
1959
1960	return NOTIFY_OK;
1961	}
1962
1963	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1964	{
1965	/* Check if we are not suspended and promiscuous mode has not been set. */
1966	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1967	&& !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, true);
1972	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1973	}
1974	else
1975	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1976	return NOTIFY_OK;
1977	}
1978
1979	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1980	{
1981	/* Undo promiscuous mode if we has set it. */
1982	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1983	{
1984	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1985	dev_set_promiscuity(pDev, -1);
1986	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1987	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1988	}
1989	else
1990	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1991	return NOTIFY_OK;
1992	}
1993
1994	/**
1995	* Callback for listening to MTU change event.
1996	*
1997	* We need to track changes of host's inteface MTU to discard over-sized frames
1998	* coming from the internal network as they may hang the TX queue of host's
1999	* adapter.
2000	*
2001	* @returns NOTIFY_OK
2002	* @param pThis The netfilter instance.
2003	* @param pDev Pointer to device structure of host's interface.
2004	*/
2005	static int vboxNetFltLinuxDeviceMtuChange(PVBOXNETFLTINS pThis, struct net_device *pDev)
2006	{
2007	ASMAtomicWriteU32(&pThis->u.s.cbMtu, pDev->mtu);
2008	Log(("vboxNetFltLinuxDeviceMtuChange: set MTU for %s to %d\n", pThis->szName, pDev->mtu));
2009	return NOTIFY_OK;
2010	}
2011
2012	#ifdef LOG_ENABLED
2013	/** Stringify the NETDEV_XXX constants. */
2014	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2015	{
2016	const char *pszEvent = "NETDEV_<unknown>";
2017	switch (ulEventType)
2018	{
2019	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2020	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2021	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2022	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2023	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2024	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2025	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2026	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2027	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2028	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2029	# ifdef NETDEV_FEAT_CHANGE
2030	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2031	# endif
2032	}
2033	return pszEvent;
2034	}
2035	#endif /* LOG_ENABLED */
2036
2037	/**
2038	* Callback for listening to netdevice events.
2039	*
2040	* This works the rediscovery, clean up on unregistration, promiscuity on
2041	* up/down, and GSO feature changes from ethtool.
2042	*
2043	* @returns NOTIFY_OK
2044	* @param self Pointer to our notifier registration block.
2045	* @param ulEventType The event.
2046	* @param ptr Event specific, but it is usually the device it
2047	* relates to.
2048	*/
2049	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2050
2051	{
2052	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2053	struct net_device pMyDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device );
2054	struct net_device *pDev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2055	int rc = NOTIFY_OK;
2056
2057	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2058	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, pMyDev));
2059
2060	if (ulEventType == NETDEV_REGISTER)
2061	{
2062	#if RTLNX_VER_MIN(2,6,24) /* cgroups/namespaces introduced */
2063	# if RTLNX_VER_MIN(2,6,26)
2064	# define VBOX_DEV_NET(dev) dev_net(dev)
2065	# define VBOX_NET_EQ(n1, n2) net_eq((n1), (n2))
2066	# else
2067	# define VBOX_DEV_NET(dev) ((dev)->nd_net)
2068	# define VBOX_NET_EQ(n1, n2) ((n1) == (n2))
2069	# endif
2070	struct net *pMyNet = current->nsproxy->net_ns;
2071	struct net *pDevNet = VBOX_DEV_NET(pDev);
2072
2073	if (VBOX_NET_EQ(pDevNet, pMyNet))
2074	#endif /* namespaces */
2075	{
2076	if (strcmp(pDev->name, pThis->szName) == 0)
2077	{
2078	vboxNetFltLinuxAttachToInterface(pThis, pDev);
2079	}
2080	}
2081	}
2082	else
2083	{
2084	if (pDev == pMyDev)
2085	{
2086	switch (ulEventType)
2087	{
2088	case NETDEV_UNREGISTER:
2089	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2090	break;
2091	case NETDEV_UP:
2092	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2093	break;
2094	case NETDEV_GOING_DOWN:
2095	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2096	break;
2097	case NETDEV_CHANGEMTU:
2098	rc = vboxNetFltLinuxDeviceMtuChange(pThis, pDev);
2099	break;
2100	case NETDEV_CHANGENAME:
2101	break;
2102	#ifdef NETDEV_FEAT_CHANGE
2103	case NETDEV_FEAT_CHANGE:
2104	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2105	break;
2106	#endif
2107	}
2108	}
2109	}
2110
2111	return rc;
2112	}
2113
2114	/*
2115	* Initial enumeration of netdevs. Called with NETDEV_REGISTER by
2116	* register_netdevice_notifier() under rtnl lock.
2117	*/
2118	static int vboxNetFltLinuxEnumeratorCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2119	{
2120	PVBOXNETFLTINS pThis = ((PVBOXNETFLTNOTIFIER)self)->pThis;
2121	struct net_device *dev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2122	struct in_device *in_dev;
2123	struct inet6_dev *in6_dev;
2124
2125	if (ulEventType != NETDEV_REGISTER)
2126	return NOTIFY_OK;
2127
2128	if (RT_UNLIKELY(pThis->pSwitchPort->pfnNotifyHostAddress == NULL))
2129	return NOTIFY_OK;
2130
2131	/*
2132	* IPv4
2133	*/
2134	#if RTLNX_VER_MIN(2,6,14)
2135	in_dev = __in_dev_get_rtnl(dev);
2136	#else
2137	in_dev = __in_dev_get(dev);
2138	#endif
2139	if (in_dev != NULL)
2140	{
2141	struct in_ifaddr *ifa;
2142
2143	for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
2144	if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2145	return NOTIFY_OK;
2146
2147	if ( dev != pThis->u.s.pDev
2148	&& VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2149	continue;
2150
2151	Log(("%s: %s: IPv4 addr %RTnaipv4 mask %RTnaipv4\n",
2152	__FUNCTION__, VBOX_NETDEV_NAME(dev),
2153	ifa->ifa_address, ifa->ifa_mask));
2154
2155	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2156	/* :fAdded */ true, kIntNetAddrType_IPv4, &ifa->ifa_address);
2157	}
2158	}
2159
2160	/*
2161	* IPv6
2162	*/
2163	in6_dev = __in6_dev_get(dev);
2164	if (in6_dev != NULL)
2165	{
2166	struct inet6_ifaddr *ifa;
2167
2168	read_lock_bh(&in6_dev->lock);
2169	#if RTLNX_VER_MIN(2,6,35)
2170	list_for_each_entry(ifa, &in6_dev->addr_list, if_list)
2171	#else
2172	for (ifa = in6_dev->addr_list; ifa != NULL; ifa = ifa->if_next)
2173	#endif
2174	{
2175	if ( dev != pThis->u.s.pDev
2176	&& ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL \| IPV6_ADDR_LOOPBACK))
2177	continue;
2178
2179	Log(("%s: %s: IPv6 addr %RTnaipv6/%u\n",
2180	__FUNCTION__, VBOX_NETDEV_NAME(dev),
2181	&ifa->addr, (unsigned)ifa->prefix_len));
2182
2183	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2184	/* :fAdded */ true, kIntNetAddrType_IPv6, &ifa->addr);
2185	}
2186	read_unlock_bh(&in6_dev->lock);
2187	}
2188
2189	return NOTIFY_OK;
2190	}
2191
2192
2193	static int vboxNetFltLinuxNotifierIPv4Callback(struct notifier_block self, unsigned long ulEventType, void ptr)
2194	{
2195	PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv4);
2196	struct net_device pDev, pEventDev;
2197	struct in_ifaddr ifa = (struct in_ifaddr )ptr;
2198	bool fMyDev;
2199	int rc = NOTIFY_OK;
2200
2201	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2202	pEventDev = ifa->ifa_dev->dev;
2203	fMyDev = (pDev == pEventDev);
2204	Log(("VBoxNetFlt: %s: IPv4 event %s(0x%lx) %s: addr %RTnaipv4 mask %RTnaipv4\n",
2205	pDev ? VBOX_NETDEV_NAME(pDev) : "<unknown>",
2206	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2207	pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<unknown>",
2208	ifa->ifa_address, ifa->ifa_mask));
2209
2210	if (pDev != NULL)
2211	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2212
2213	if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2214	return NOTIFY_OK;
2215
2216	if ( !fMyDev
2217	&& VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2218	return NOTIFY_OK;
2219
2220	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2221	{
2222	bool fAdded;
2223	if (ulEventType == NETDEV_UP)
2224	fAdded = true;
2225	else if (ulEventType == NETDEV_DOWN)
2226	fAdded = false;
2227	else
2228	return NOTIFY_OK;
2229
2230	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2231	kIntNetAddrType_IPv4, &ifa->ifa_local);
2232	}
2233
2234	return rc;
2235	}
2236
2237
2238	static int vboxNetFltLinuxNotifierIPv6Callback(struct notifier_block self, unsigned long ulEventType, void ptr)
2239	{
2240	PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv6);
2241	struct net_device pDev, pEventDev;
2242	struct inet6_ifaddr ifa = (struct inet6_ifaddr )ptr;
2243	bool fMyDev;
2244	int rc = NOTIFY_OK;
2245
2246	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2247	pEventDev = ifa->idev->dev;
2248	fMyDev = (pDev == pEventDev);
2249	Log(("VBoxNetFlt: %s: IPv6 event %s(0x%lx) %s: %RTnaipv6\n",
2250	pDev ? VBOX_NETDEV_NAME(pDev) : "<unknown>",
2251	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2252	pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<unknown>",
2253	&ifa->addr));
2254
2255	if (pDev != NULL)
2256	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2257
2258	if ( !fMyDev
2259	&& ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL \| IPV6_ADDR_LOOPBACK))
2260	return NOTIFY_OK;
2261
2262	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2263	{
2264	bool fAdded;
2265	if (ulEventType == NETDEV_UP)
2266	fAdded = true;
2267	else if (ulEventType == NETDEV_DOWN)
2268	fAdded = false;
2269	else
2270	return NOTIFY_OK;
2271
2272	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2273	kIntNetAddrType_IPv6, &ifa->addr);
2274	}
2275
2276	return rc;
2277	}
2278
2279
2280	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2281	{
2282	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2283	}
2284
2285	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2286	{
2287	struct net_device * pDev;
2288	int err;
2289	int rc = VINF_SUCCESS;
2290	IPRT_LINUX_SAVE_EFL_AC();
2291	NOREF(pvIfData);
2292
2293	LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2294
2295	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2296	if (pDev)
2297	{
2298	/*
2299	* Create a sk_buff for the gather list and push it onto the wire.
2300	*/
2301	if (fDst & INTNETTRUNKDIR_WIRE)
2302	{
2303	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2304	if (pBuf)
2305	{
2306	vboxNetFltDumpPacket(pSG, true, "wire", 1);
2307	Log6(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2308	Log6(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2309	err = dev_queue_xmit(pBuf);
2310	if (err)
2311	rc = RTErrConvertFromErrno(err);
2312	}
2313	else
2314	rc = VERR_NO_MEMORY;
2315	}
2316
2317	/*
2318	* Create a sk_buff for the gather list and push it onto the host stack.
2319	*/
2320	if (fDst & INTNETTRUNKDIR_HOST)
2321	{
2322	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2323	if (pBuf)
2324	{
2325	vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2326	Log6(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2327	Log6(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2328	err = netif_rx_ni(pBuf);
2329	if (err)
2330	rc = RTErrConvertFromErrno(err);
2331	}
2332	else
2333	rc = VERR_NO_MEMORY;
2334	}
2335
2336	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2337	}
2338
2339	IPRT_LINUX_RESTORE_EFL_AC();
2340	return rc;
2341	}
2342
2343
2344	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2345	{
2346	struct net_device *pDev;
2347	IPRT_LINUX_SAVE_EFL_AC();
2348
2349	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%RTbool, fDisablePromiscuous=%RTbool\n",
2350	pThis, pThis->szName, fActive, pThis->fDisablePromiscuous));
2351
2352	if (pThis->fDisablePromiscuous)
2353	return;
2354
2355	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2356	if (pDev)
2357	{
2358	/*
2359	* This api is a bit weird, the best reference is the code.
2360	*
2361	* Also, we have a bit or race conditions wrt the maintenance of
2362	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2363	*/
2364	#ifdef LOG_ENABLED
2365	u_int16_t fIf;
2366	unsigned const cPromiscBefore = pDev->promiscuity;
2367	#endif
2368	if (fActive)
2369	{
2370	Assert(!pThis->u.s.fPromiscuousSet);
2371
2372	rtnl_lock();
2373	dev_set_promiscuity(pDev, 1);
2374	rtnl_unlock();
2375	pThis->u.s.fPromiscuousSet = true;
2376	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2377	}
2378	else
2379	{
2380	if (pThis->u.s.fPromiscuousSet)
2381	{
2382	rtnl_lock();
2383	dev_set_promiscuity(pDev, -1);
2384	rtnl_unlock();
2385	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2386	}
2387	pThis->u.s.fPromiscuousSet = false;
2388
2389	#ifdef LOG_ENABLED
2390	fIf = dev_get_flags(pDev);
2391	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2392	#endif
2393	}
2394
2395	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2396	}
2397	IPRT_LINUX_RESTORE_EFL_AC();
2398	}
2399
2400
2401	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2402	{
2403	/*
2404	* Remove packet handler when we get disconnected from internal switch as
2405	* we don't want the handler to forward packets to disconnected switch.
2406	*/
2407	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
2408	{
2409	IPRT_LINUX_SAVE_EFL_AC();
2410	dev_remove_pack(&pThis->u.s.PacketType);
2411	Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
2412	IPRT_LINUX_RESTORE_EFL_AC();
2413	}
2414	return VINF_SUCCESS;
2415	}
2416
2417
2418	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2419	{
2420	IPRT_LINUX_SAVE_EFL_AC();
2421
2422	/*
2423	* Report the GSO capabilities of the host and device (if connected).
2424	* Note! No need to mark ourselves busy here.
2425	*/
2426	/** @todo duplicate work here now? Attach */
2427	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2428	Log3(("vboxNetFltOsConnectIt: reporting host tso tso6\n"));
2429	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2430	0
2431	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2432	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2433	, INTNETTRUNKDIR_HOST);
2434
2435	#endif
2436	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2437
2438	IPRT_LINUX_RESTORE_EFL_AC();
2439	return VINF_SUCCESS;
2440	}
2441
2442
2443	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2444	{
2445	struct net_device *pDev;
2446	bool fRegistered;
2447	IPRT_LINUX_SAVE_EFL_AC();
2448
2449	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
2450	vboxNetFltLinuxUnhookDev(pThis, NULL);
2451	#endif
2452
2453	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2454	* unlikely, but none the less). Since it doesn't actually update the
2455	* state (just reads it), it is likely to panic in some interesting
2456	* ways. */
2457
2458	RTSpinlockAcquire(pThis->hSpinlock);
2459	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2460	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
2461	RTSpinlockRelease(pThis->hSpinlock);
2462
2463	if (fRegistered)
2464	{
2465	vboxNetFltSetLinkState(pThis, pDev, false);
2466
2467	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2468	skb_queue_purge(&pThis->u.s.XmitQueue);
2469	#endif
2470	Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
2471	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
2472	pDev, pDev->name,
2473	#if RTLNX_VER_MIN(2,6,37)
2474	netdev_refcnt_read(pDev)
2475	#else
2476	atomic_read(&pDev->refcnt)
2477	#endif
2478	));
2479	dev_put(pDev);
2480	}
2481
2482	unregister_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2483	unregister_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2484
2485	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2486	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2487	module_put(THIS_MODULE);
2488
2489	IPRT_LINUX_RESTORE_EFL_AC();
2490	}
2491
2492
2493	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2494	{
2495	int err;
2496	IPRT_LINUX_SAVE_EFL_AC();
2497	NOREF(pvContext);
2498
2499	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2500	err = register_netdevice_notifier(&pThis->u.s.Notifier);
2501	if (err)
2502	{
2503	IPRT_LINUX_RESTORE_EFL_AC();
2504	return VERR_INTNET_FLT_IF_FAILED;
2505	}
2506	if (!pThis->u.s.fRegistered)
2507	{
2508	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2509	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2510	IPRT_LINUX_RESTORE_EFL_AC();
2511	return VERR_INTNET_FLT_IF_NOT_FOUND;
2512	}
2513
2514	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2515	if ( pThis->fDisconnectedFromHost
2516	\|\| !try_module_get(THIS_MODULE))
2517	{
2518	IPRT_LINUX_RESTORE_EFL_AC();
2519	return VERR_INTNET_FLT_IF_FAILED;
2520	}
2521
2522	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2523	{
2524	VBOXNETFLTNOTIFIER Enumerator;
2525
2526	/*
2527	* register_inetaddr_notifier() and register_inet6addr_notifier()
2528	* do not call the callback for existing devices. Enumerating
2529	* all network devices explicitly is a bit of an ifdef mess,
2530	* so co-opt register_netdevice_notifier() to do that for us.
2531	*/
2532	RT_ZERO(Enumerator);
2533	Enumerator.Notifier.notifier_call = vboxNetFltLinuxEnumeratorCallback;
2534	Enumerator.pThis = pThis;
2535
2536	err = register_netdevice_notifier(&Enumerator.Notifier);
2537	if (err)
2538	{
2539	LogRel(("%s: failed to enumerate network devices: error %d\n", __FUNCTION__, err));
2540	IPRT_LINUX_RESTORE_EFL_AC();
2541	return VINF_SUCCESS;
2542	}
2543
2544	unregister_netdevice_notifier(&Enumerator.Notifier);
2545
2546	pThis->u.s.NotifierIPv4.notifier_call = vboxNetFltLinuxNotifierIPv4Callback;
2547	err = register_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2548	if (err)
2549	LogRel(("%s: failed to register IPv4 notifier: error %d\n", __FUNCTION__, err));
2550
2551	pThis->u.s.NotifierIPv6.notifier_call = vboxNetFltLinuxNotifierIPv6Callback;
2552	err = register_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2553	if (err)
2554	LogRel(("%s: failed to register IPv6 notifier: error %d\n", __FUNCTION__, err));
2555	}
2556
2557	IPRT_LINUX_RESTORE_EFL_AC();
2558	return VINF_SUCCESS;
2559	}
2560
2561	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2562	{
2563	IPRT_LINUX_SAVE_EFL_AC();
2564
2565	/*
2566	* Init the linux specific members.
2567	*/
2568	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2569	pThis->u.s.fRegistered = false;
2570	pThis->u.s.fPromiscuousSet = false;
2571	pThis->u.s.fPacketHandler = false;
2572	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2573	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2574	skb_queue_head_init(&pThis->u.s.XmitQueue);
2575	# if RTLNX_VER_MIN(2,6,20)
2576	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2577	# else
2578	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2579	# endif
2580	#endif
2581
2582	IPRT_LINUX_RESTORE_EFL_AC();
2583	return VINF_SUCCESS;
2584	}
2585
2586
2587	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2588	{
2589	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2590	}
2591
2592
2593	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
2594	{
2595	/* Nothing to do */
2596	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2597	return VINF_SUCCESS;
2598	}
2599
2600
2601	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2602	{
2603	/* Nothing to do */
2604	NOREF(pThis); NOREF(pvIfData);
2605	return VINF_SUCCESS;
2606	}
2607

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 85698

Download in other formats: