VBoxNetFlt-linux.c@ 40733

Last change on this file since 40733 was 40733, checked in by vboxsync, 13 years ago
VBoxNetFlt: 3.4-rc1 support (VLAN_ETH_ALEN -> ETH_ALEN)
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 69.7 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 40733 2012-04-01 07:56:35Z vboxsync $ */
2	/** @file
3	* VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2008 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/*******************************************************************************
19	* Header Files *
20	*******************************************************************************/
21	#define LOG_GROUP LOG_GROUP_NET_FLT_DRV
22	#define VBOXNETFLT_LINUX_NO_XMIT_QUEUE
23	#include "the-linux-kernel.h"
24	#include "version-generated.h"
25	#include "product-generated.h"
26	#include <linux/netdevice.h>
27	#include <linux/etherdevice.h>
28	#include <linux/rtnetlink.h>
29	#include <linux/miscdevice.h>
30	#include <linux/ip.h>
31	#include <linux/if_vlan.h>
32
33	#include <VBox/log.h>
34	#include <VBox/err.h>
35	#include <VBox/intnetinline.h>
36	#include <VBox/vmm/pdmnetinline.h>
37	#include <VBox/param.h>
38	#include <iprt/alloca.h>
39	#include <iprt/assert.h>
40	#include <iprt/spinlock.h>
41	#include <iprt/semaphore.h>
42	#include <iprt/initterm.h>
43	#include <iprt/process.h>
44	#include <iprt/mem.h>
45	#include <iprt/net.h>
46	#include <iprt/log.h>
47	#include <iprt/mp.h>
48	#include <iprt/mem.h>
49	#include <iprt/time.h>
50
51	#define VBOXNETFLT_OS_SPECFIC 1
52	#include "../VBoxNetFltInternal.h"
53
54
55	/*******************************************************************************
56	* Defined Constants And Macros *
57	*******************************************************************************/
58	#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
59	#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
60	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
61	# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
62	#endif
63
64	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
65	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
66	# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
67	#else
68	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
69	# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
70	#endif
71
72	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
73	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
74	#else
75	# define CHECKSUM_PARTIAL CHECKSUM_HW
76	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
77	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
78	# else
79	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
80	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
81	# else
82	# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
83	# endif
84	/* Versions prior 2.6.10 use stats for both bstats and qstats */
85	# define bstats stats
86	# define qstats stats
87	# endif
88	#endif
89
90	#ifndef NET_IP_ALIGN
91	# define NET_IP_ALIGN 2
92	#endif
93
94	#if 0
95	/** Create scatter / gather segments for fragments. When not used, we will
96	* linearize the socket buffer before creating the internal networking SG. */
97	# define VBOXNETFLT_SG_SUPPORT 1
98	#endif
99
100	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
101	/** Indicates that the linux kernel may send us GSO frames. */
102	# define VBOXNETFLT_WITH_GSO 1
103
104	/** This enables or disables the transmitting of GSO frame from the internal
105	* network and to the host. */
106	# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
107
108	# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
109	/** This enables or disables the transmitting of GSO frame from the internal
110	* network and to the wire. */
111	# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
112	# endif
113
114	/** This enables or disables the forwarding/flooding of GSO frame from the host
115	* to the internal network. */
116	# define VBOXNETFLT_WITH_GSO_RECV 1
117
118	#endif
119
120	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
121	/** This enables or disables handling of GSO frames coming from the wire (GRO). */
122	# define VBOXNETFLT_WITH_GRO 1
123	#endif
124	/*
125	* GRO support was backported to RHEL 5.4
126	*/
127	#ifdef RHEL_RELEASE_CODE
128	# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
129	# define VBOXNETFLT_WITH_GRO 1
130	# endif
131	#endif
132
133	/*******************************************************************************
134	* Internal Functions *
135	*******************************************************************************/
136	static int VBoxNetFltLinuxInit(void);
137	static void VBoxNetFltLinuxUnload(void);
138	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
139
140
141	/*******************************************************************************
142	* Global Variables *
143	*******************************************************************************/
144	/**
145	* The (common) global data.
146	*/
147	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
148
149	module_init(VBoxNetFltLinuxInit);
150	module_exit(VBoxNetFltLinuxUnload);
151
152	MODULE_AUTHOR(VBOX_VENDOR);
153	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
154	MODULE_LICENSE("GPL");
155	#ifdef MODULE_VERSION
156	MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
157	#endif
158
159
160	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
161	unsigned dev_get_flags(const struct net_device *dev)
162	{
163	unsigned flags;
164
165	flags = (dev->flags & ~(IFF_PROMISC \|
166	IFF_ALLMULTI \|
167	IFF_RUNNING)) \|
168	(dev->gflags & (IFF_PROMISC \|
169	IFF_ALLMULTI));
170
171	if (netif_running(dev) && netif_carrier_ok(dev))
172	flags \|= IFF_RUNNING;
173
174	return flags;
175	}
176	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
177
178
179	/**
180	* Initialize module.
181	*
182	* @returns appropriate status code.
183	*/
184	static int __init VBoxNetFltLinuxInit(void)
185	{
186	int rc;
187	/*
188	* Initialize IPRT.
189	*/
190	rc = RTR0Init(0);
191	if (RT_SUCCESS(rc))
192	{
193	Log(("VBoxNetFltLinuxInit\n"));
194
195	/*
196	* Initialize the globals and connect to the support driver.
197	*
198	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
199	* for establishing the connect to the support driver.
200	*/
201	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
202	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
203	if (RT_SUCCESS(rc))
204	{
205	LogRel(("VBoxNetFlt: Successfully started.\n"));
206	return 0;
207	}
208
209	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
210	RTR0Term();
211	}
212	else
213	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
214
215	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
216	return -RTErrConvertToErrno(rc);
217	}
218
219
220	/**
221	* Unload the module.
222	*
223	* @todo We have to prevent this if we're busy!
224	*/
225	static void __exit VBoxNetFltLinuxUnload(void)
226	{
227	int rc;
228	Log(("VBoxNetFltLinuxUnload\n"));
229	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
230
231	/*
232	* Undo the work done during start (in reverse order).
233	*/
234	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
235	AssertRC(rc); NOREF(rc);
236
237	RTR0Term();
238
239	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
240
241	Log(("VBoxNetFltLinuxUnload - done\n"));
242	}
243
244
245	/**
246	* We filter traffic from the host to the internal network
247	* before it reaches the NIC driver.
248	*
249	* The current code uses a very ugly hack overriding hard_start_xmit
250	* callback in the device structure, but it has been shown to give us a
251	* performance boost of 60-100% though. Eventually we have to find some
252	* less hacky way of getting this job done.
253	*/
254	#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
255
256	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
257
258	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
259
260	# include <linux/ethtool.h>
261
262	typedef struct ethtool_ops OVR_OPSTYPE;
263	# define OVR_OPS ethtool_ops
264	# define OVR_XMIT pfnStartXmit
265
266	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
267
268	typedef struct net_device_ops OVR_OPSTYPE;
269	# define OVR_OPS netdev_ops
270	# define OVR_XMIT pOrgOps->ndo_start_xmit
271
272	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
273
274	/**
275	* The overridden net_device_ops of the device we're attached to.
276	*
277	* As there is no net_device_ops structure in pre-2.6.29 kernels we override
278	* ethtool_ops instead along with hard_start_xmit callback in net_device
279	* structure.
280	*
281	* This is a very dirty hack that was created to explore how much we can improve
282	* the host to guest transfers by not CC'ing the NIC. It turns out to be
283	* the only way to filter outgoing packets for devices without TX queue.
284	*/
285	typedef struct VBoxNetDeviceOpsOverride
286	{
287	/** Our overridden ops. */
288	OVR_OPSTYPE Ops;
289	/** Magic word. */
290	uint32_t u32Magic;
291	/** Pointer to the original ops. */
292	OVR_OPSTYPE const *pOrgOps;
293	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
294	/** Pointer to the original hard_start_xmit function. */
295	int (pfnStartXmit)(struct sk_buff pSkb, struct net_device *pDev);
296	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
297	/** Pointer to the net filter instance. */
298	PVBOXNETFLTINS pVBoxNetFlt;
299	/** The number of filtered packages. */
300	uint64_t cFiltered;
301	/** The total number of packets */
302	uint64_t cTotal;
303	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
304	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
305	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
306
307	/**
308	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
309	* because they belong on the internal network.
310	*
311	* @returns NETDEV_TX_XXX.
312	* @param pSkb The socket buffer to transmit.
313	* @param pDev The net device.
314	*/
315	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
316	{
317	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
318	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
319	PCRTNETETHERHDR pEtherHdr;
320	PINTNETTRUNKSWPORT pSwitchPort;
321	uint32_t cbHdrs;
322
323
324	/*
325	* Validate the override structure.
326	*
327	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
328	* to be production quality code, we would have to be much more
329	* careful here and avoid the race.
330	*/
331	if ( !VALID_PTR(pOverride)
332	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
333	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
334	\|\| !VALID_PTR(pOverride->pOrgOps)
335	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
336	)
337	{
338	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
339	dev_kfree_skb(pSkb);
340	return NETDEV_TX_OK;
341	}
342	pOverride->cTotal++;
343
344	/*
345	* Do the filtering base on the default OUI of our virtual NICs
346	*
347	* Note! In a real solution, we would ask the switch whether the
348	* destination MAC is 100% to be on the internal network and then
349	* drop it.
350	*/
351	cbHdrs = skb_headlen(pSkb);
352	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
353	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
354	if ( pEtherHdr
355	&& VALID_PTR(pOverride->pVBoxNetFlt)
356	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
357	&& VALID_PTR(pSwitchPort)
358	&& cbHdrs >= 6)
359	{
360	INTNETSWDECISION enmDecision;
361
362	/** @todo consider reference counting, etc. */
363	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
364	if (enmDecision == INTNETSWDECISION_INTNET)
365	{
366	dev_kfree_skb(pSkb);
367	pOverride->cFiltered++;
368	return NETDEV_TX_OK;
369	}
370	}
371
372	return pOverride->OVR_XMIT(pSkb, pDev);
373	}
374
375	/**
376	* Hooks the device ndo_start_xmit operation of the device.
377	*
378	* @param pThis The net filter instance.
379	* @param pDev The net device.
380	*/
381	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
382	{
383	PVBOXNETDEVICEOPSOVERRIDE pOverride;
384	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
385
386	/* Cancel override if ethtool_ops is missing (host-only case, #5712) */
387	if (!VALID_PTR(pDev->OVR_OPS))
388	return;
389	pOverride = RTMemAlloc(sizeof(*pOverride));
390	if (!pOverride)
391	return;
392	pOverride->pOrgOps = pDev->OVR_OPS;
393	pOverride->Ops = *pDev->OVR_OPS;
394	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
395	pOverride->pfnStartXmit = pDev->hard_start_xmit;
396	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
397	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
398	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
399	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
400	pOverride->cTotal = 0;
401	pOverride->cFiltered = 0;
402	pOverride->pVBoxNetFlt = pThis;
403
404	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp); /* (this isn't necessary, but so what) */
405	ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
406	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
407	ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
408	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
409	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
410	}
411
412	/**
413	* Undos what vboxNetFltLinuxHookDev did.
414	*
415	* @param pThis The net filter instance.
416	* @param pDev The net device. Can be NULL, in which case
417	* we'll try retrieve it from @a pThis.
418	*/
419	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
420	{
421	PVBOXNETDEVICEOPSOVERRIDE pOverride;
422	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
423
424	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
425	if (!pDev)
426	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
427	if (VALID_PTR(pDev))
428	{
429	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
430	if ( VALID_PTR(pOverride)
431	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
432	&& VALID_PTR(pOverride->pOrgOps)
433	)
434	{
435	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
436	ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
437	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
438	ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
439	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
440	}
441	else
442	pOverride = NULL;
443	}
444	else
445	pOverride = NULL;
446	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
447
448	if (pOverride)
449	{
450	printk("vboxnetflt: dropped %llu out of %llu packets\n", pOverride->cFiltered, pOverride->cTotal);
451	RTMemFree(pOverride);
452	}
453	}
454
455	#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
456
457
458	/**
459	* Reads and retains the host interface handle.
460	*
461	* @returns The handle, NULL if detached.
462	* @param pThis
463	*/
464	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
465	{
466	#if 0
467	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
468	struct net_device *pDev = NULL;
469
470	Log(("vboxNetFltLinuxRetainNetDev\n"));
471	/*
472	* Be careful here to avoid problems racing the detached callback.
473	*/
474	RTSpinlockAcquire(pThis->hSpinlock, &Tmp);
475	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
476	{
477	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
478	if (pDev)
479	{
480	dev_hold(pDev);
481	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
482	pDev, pDev->name,
483	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
484	netdev_refcnt_read(pDev)
485	#else
486	atomic_read(&pDev->refcnt)
487	#endif
488	));
489	}
490	}
491	RTSpinlockRelease(pThis->hSpinlock, &Tmp);
492
493	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
494	return pDev;
495	#else
496	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
497	#endif
498	}
499
500
501	/**
502	* Release the host interface handle previously retained
503	* by vboxNetFltLinuxRetainNetDev.
504	*
505	* @param pThis The instance.
506	* @param pDev The vboxNetFltLinuxRetainNetDev
507	* return value, NULL is fine.
508	*/
509	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
510	{
511	#if 0
512	Log(("vboxNetFltLinuxReleaseNetDev\n"));
513	NOREF(pThis);
514	if (pDev)
515	{
516	dev_put(pDev);
517	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
518	pDev, pDev->name,
519	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
520	netdev_refcnt_read(pDev)
521	#else
522	atomic_read(&pDev->refcnt)
523	#endif
524	));
525	}
526	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
527	#endif
528	}
529
530	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
531	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
532
533	/**
534	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
535	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
536	*
537	* @returns true / false accordingly.
538	* @param pBuf The sk_buff.
539	*/
540	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
541	{
542	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
543	}
544
545
546	/**
547	* Internal worker that create a linux sk_buff for a
548	* (scatter/)gather list.
549	*
550	* @returns Pointer to the sk_buff.
551	* @param pThis The instance.
552	* @param pSG The (scatter/)gather list.
553	* @param fDstWire Set if the destination is the wire.
554	*/
555	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
556	{
557	struct sk_buff *pPkt;
558	struct net_device *pDev;
559	unsigned fGsoType = 0;
560
561	if (pSG->cbTotal == 0)
562	{
563	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
564	return NULL;
565	}
566
567	/** @todo We should use fragments mapping the SG buffers with large packets.
568	* 256 bytes seems to be the a threshold used a lot for this. It
569	* requires some nasty work on the intnet side though... */
570	/*
571	* Allocate a packet and copy over the data.
572	*/
573	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
574	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
575	if (RT_UNLIKELY(!pPkt))
576	{
577	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
578	pSG->pvUserData = NULL;
579	return NULL;
580	}
581	pPkt->dev = pDev;
582	pPkt->ip_summed = CHECKSUM_NONE;
583
584	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
585	skb_reserve(pPkt, NET_IP_ALIGN);
586
587	/* Copy the segments. */
588	skb_put(pPkt, pSG->cbTotal);
589	IntNetSgRead(pSG, pPkt->data);
590
591	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
592	/*
593	* Setup GSO if used by this packet.
594	*/
595	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
596	{
597	default:
598	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
599	/* fall thru */
600	case PDMNETWORKGSOTYPE_INVALID:
601	fGsoType = 0;
602	break;
603	case PDMNETWORKGSOTYPE_IPV4_TCP:
604	fGsoType = SKB_GSO_TCPV4;
605	break;
606	case PDMNETWORKGSOTYPE_IPV4_UDP:
607	fGsoType = SKB_GSO_UDP;
608	break;
609	case PDMNETWORKGSOTYPE_IPV6_TCP:
610	fGsoType = SKB_GSO_TCPV6;
611	break;
612	}
613	if (fGsoType)
614	{
615	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
616
617	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
618	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
619	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
620
621	/*
622	* We need to set checksum fields even if the packet goes to the host
623	* directly as it may be immediately forwarded by IP layer @bugref{5020}.
624	*/
625	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
626	pPkt->ip_summed = CHECKSUM_PARTIAL;
627	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
628	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
629	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
630	pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
631	else
632	pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
633	# else
634	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
635	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
636	pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
637	else
638	pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
639	# endif
640	if (!fDstWire)
641	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
642	}
643	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
644
645	/*
646	* Finish up the socket buffer.
647	*/
648	pPkt->protocol = eth_type_trans(pPkt, pDev);
649	if (fDstWire)
650	{
651	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
652
653	/* Restore ethernet header back. */
654	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
655	VBOX_SKB_RESET_MAC_HDR(pPkt);
656	}
657	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
658
659	return pPkt;
660	}
661
662
663	/**
664	* Initializes a SG list from an sk_buff.
665	*
666	* @returns Number of segments.
667	* @param pThis The instance.
668	* @param pBuf The sk_buff.
669	* @param pSG The SG.
670	* @param pvFrame The frame pointer, optional.
671	* @param cSegs The number of segments allocated for the SG.
672	* This should match the number in the mbuf exactly!
673	* @param fSrc The source of the frame.
674	* @param pGso Pointer to the GSO context if it's a GSO
675	* internal network frame. NULL if regular frame.
676	*/
677	DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
678	unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
679	{
680	int i;
681	NOREF(pThis);
682
683	Assert(!skb_shinfo(pBuf)->frag_list);
684
685	if (!pGsoCtx)
686	IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /cSegsUsed/);
687	else
688	IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /cSegsUsed/, pGsoCtx);
689
690	#ifdef VBOXNETFLT_SG_SUPPORT
691	pSG->aSegs[0].cb = skb_headlen(pBuf);
692	pSG->aSegs[0].pv = pBuf->data;
693	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
694
695	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
696	{
697	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
698	pSG->aSegs[i+1].cb = pFrag->size;
699	pSG->aSegs[i+1].pv = kmap(pFrag->page);
700	printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
701	pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
702	}
703	++i;
704
705	#else
706	pSG->aSegs[0].cb = pBuf->len;
707	pSG->aSegs[0].pv = pBuf->data;
708	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
709	i = 1;
710	#endif
711
712	pSG->cSegsUsed = i;
713
714	#ifdef PADD_RUNT_FRAMES_FROM_HOST
715	/*
716	* Add a trailer if the frame is too small.
717	*
718	* Since we're getting to the packet before it is framed, it has not
719	* yet been padded. The current solution is to add a segment pointing
720	* to a buffer containing all zeros and pray that works for all frames...
721	*/
722	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
723	{
724	static uint8_t const s_abZero[128] = {0};
725
726	AssertReturnVoid(i < cSegs);
727
728	pSG->aSegs[i].Phys = NIL_RTHCPHYS;
729	pSG->aSegs[i].pv = (void *)&s_abZero[0];
730	pSG->aSegs[i].cb = 60 - pSG->cbTotal;
731	pSG->cbTotal = 60;
732	pSG->cSegsUsed++;
733	Assert(i + 1 <= pSG->cSegsAlloc)
734	}
735	#endif
736
737	Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
738	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
739	for (i = 0; i < pSG->cSegsUsed; i++)
740	Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
741	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
742	}
743
744	/**
745	* Packet handler,
746	*
747	* @returns 0 or EJUSTRETURN.
748	* @param pThis The instance.
749	* @param pMBuf The mbuf.
750	* @param pvFrame The start of the frame, optional.
751	* @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
752	* @param eProtocol The protocol.
753	*/
754	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
755	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
756	struct net_device *pSkbDev,
757	struct packet_type *pPacketType,
758	struct net_device *pOrigDev)
759	#else
760	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
761	struct net_device *pSkbDev,
762	struct packet_type *pPacketType)
763	#endif
764	{
765	PVBOXNETFLTINS pThis;
766	struct net_device *pDev;
767	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
768	pBuf, pSkbDev, pPacketType));
769	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
770	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",
771	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));
772	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
773	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
774	# endif
775	#else
776	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",
777	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));
778	#endif
779	/*
780	* Drop it immediately?
781	*/
782	if (!pBuf)
783	return 0;
784
785	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
786	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
787	if (pDev != pSkbDev)
788	{
789	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
790	return 0;
791	}
792
793	Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
794	if (vboxNetFltLinuxSkBufIsOur(pBuf))
795	{
796	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
797	dev_kfree_skb(pBuf);
798	return 0;
799	}
800
801	#ifndef VBOXNETFLT_SG_SUPPORT
802	{
803	/*
804	* Get rid of fragmented packets, they cause too much trouble.
805	*/
806	unsigned int uMacLen = pBuf->mac_len;
807	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
808	kfree_skb(pBuf);
809	if (!pCopy)
810	{
811	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
812	return 0;
813	}
814	pBuf = pCopy;
815	/* Somehow skb_copy ignores mac_len */
816	pBuf->mac_len = uMacLen;
817	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
818	/* Restore VLAN tag stripped by host hardware */
819	if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
820	{
821	uint8_t pMac = (uint8_t)skb_mac_header(pBuf);
822	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )(pMac - VLAN_HLEN);
823	# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
824	memmove(pVHdr, pMac, ETH_ALEN * 2);
825	# else
826	memmove(pVHdr, pMac, VLAN_ETH_ALEN * 2);
827	# endif
828	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
829	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
830	pBuf->mac_header -= VLAN_HLEN;
831	pBuf->mac_len += VLAN_HLEN;
832	}
833	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
834
835	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
836	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",
837	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));
838	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
839	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
840	# endif
841	# else
842	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",
843	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));
844	# endif
845	}
846	#endif
847
848	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
849	/* Forward it to the internal network. */
850	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
851	#else
852	/* Add the packet to transmit queue and schedule the bottom half. */
853	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
854	schedule_work(&pThis->u.s.XmitTask);
855	Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
856	&pThis->u.s.XmitTask, pBuf));
857	#endif
858
859	/* It does not really matter what we return, it is ignored by the kernel. */
860	return 0;
861	}
862
863	/**
864	* Calculate the number of INTNETSEG segments the socket buffer will need.
865	*
866	* @returns Segment count.
867	* @param pBuf The socket buffer.
868	*/
869	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
870	{
871	#ifdef VBOXNETFLT_SG_SUPPORT
872	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
873	#else
874	unsigned cSegs = 1;
875	#endif
876	#ifdef PADD_RUNT_FRAMES_FROM_HOST
877	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
878	if (pBuf->len < 60)
879	cSegs++;
880	#endif
881	return cSegs;
882	}
883
884	/**
885	* Destroy the intnet scatter / gather buffer created by
886	* vboxNetFltLinuxSkBufToSG.
887	*/
888	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
889	{
890	#ifdef VBOXNETFLT_SG_SUPPORT
891	int i;
892
893	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
894	{
895	printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
896	kunmap(pSG->aSegs[i+1].pv);
897	}
898	#endif
899	NOREF(pSG);
900	}
901
902	#ifdef LOG_ENABLED
903	/**
904	* Logging helper.
905	*/
906	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
907	{
908	int i, offSeg;
909	uint8_t pInt, pExt;
910	static int iPacketNo = 1;
911	iPacketNo += iIncrement;
912	if (fEgress)
913	{
914	pExt = pSG->aSegs[0].pv;
915	pInt = pExt + 6;
916	}
917	else
918	{
919	pInt = pSG->aSegs[0].pv;
920	pExt = pInt + 6;
921	}
922	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
923	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
924	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
925	fEgress ? "-->" : "<--", pszWhere,
926	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
927	pSG->cbTotal, iPacketNo));
928	if (pSG->cSegsUsed == 1)
929	{
930	Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
931	}
932	else
933	{
934	for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
935	{
936	Log3(("-- segment %d at 0x%x (%d bytes) --\n%.*Rhxd\n",
937	i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
938	offSeg += pSG->aSegs[i].cb;
939	}
940	}
941
942	}
943	#else
944	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
945	#endif
946
947	#ifdef VBOXNETFLT_WITH_GSO_RECV
948
949	/**
950	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
951	* GSO socket buffer without having to segment it.
952	*
953	* @returns true on success, false if needs segmenting.
954	* @param pThis The net filter instance.
955	* @param pSkb The GSO socket buffer.
956	* @param fSrc The source.
957	* @param pGsoCtx Where to return the GSO context on success.
958	*/
959	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
960	PPDMNETWORKGSO pGsoCtx)
961	{
962	PDMNETWORKGSOTYPE enmGsoType;
963	uint16_t uEtherType;
964	unsigned int cbTransport;
965	unsigned int offTransport;
966	unsigned int cbTransportHdr;
967	unsigned uProtocol;
968	union
969	{
970	RTNETIPV4 IPv4;
971	RTNETIPV6 IPv6;
972	RTNETTCP Tcp;
973	uint8_t ab[40];
974	uint16_t au16[40/2];
975	uint32_t au32[40/4];
976	} Buf;
977
978	/*
979	* Check the GSO properties of the socket buffer and make sure it fits.
980	*/
981	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
982	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP \| SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
983	{
984	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
985	return false;
986	}
987	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
988	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
989	{
990	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
991	return false;
992	}
993	/*
994	* It is possible to receive GSO packets from wire if GRO is enabled.
995	*/
996	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
997	{
998	Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
999	#ifdef VBOXNETFLT_WITH_GRO
1000	/*
1001	* The packet came from the wire and the driver has already consumed
1002	* mac header. We need to restore it back.
1003	*/
1004	pSkb->mac_len = skb_network_header(pSkb) - skb_mac_header(pSkb);
1005	skb_push(pSkb, pSkb->mac_len);
1006	Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1007	pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
1008	#else /* !VBOXNETFLT_WITH_GRO */
1009	/* Older kernels didn't have GRO. */
1010	return false;
1011	#endif /* !VBOXNETFLT_WITH_GRO */
1012	}
1013	else
1014	{
1015	/*
1016	* skb_gso_segment does the following. Do we need to do it as well?
1017	*/
1018	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1019	skb_reset_mac_header(pSkb);
1020	pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1021	#else
1022	pSkb->mac.raw = pSkb->data;
1023	pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1024	#endif
1025	}
1026
1027	/*
1028	* Switch on the ethertype.
1029	*/
1030	uEtherType = pSkb->protocol;
1031	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1032	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1033	{
1034	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1035	if (puEtherType)
1036	uEtherType = *puEtherType;
1037	}
1038	switch (uEtherType)
1039	{
1040	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1041	{
1042	unsigned int cbHdr;
1043	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1044	if (RT_UNLIKELY(!pIPv4))
1045	{
1046	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1047	return false;
1048	}
1049
1050	cbHdr = pIPv4->ip_hl * 4;
1051	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1052	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1053	\|\| cbHdr > cbTransport ))
1054	{
1055	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1056	return false;
1057	}
1058	cbTransport -= cbHdr;
1059	offTransport = pSkb->mac_len + cbHdr;
1060	uProtocol = pIPv4->ip_p;
1061	if (uProtocol == RTNETIPV4_PROT_TCP)
1062	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1063	else if (uProtocol == RTNETIPV4_PROT_UDP)
1064	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1065	else /** @todo IPv6: 4to6 tunneling */
1066	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1067	break;
1068	}
1069
1070	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1071	{
1072	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1073	if (RT_UNLIKELY(!pIPv6))
1074	{
1075	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1076	return false;
1077	}
1078
1079	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1080	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1081	uProtocol = pIPv6->ip6_nxt;
1082	/** @todo IPv6: Dig our way out of the other headers. */
1083	if (uProtocol == RTNETIPV4_PROT_TCP)
1084	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1085	else if (uProtocol == RTNETIPV4_PROT_UDP)
1086	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1087	else
1088	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1089	break;
1090	}
1091
1092	default:
1093	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1094	return false;
1095	}
1096
1097	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1098	{
1099	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1100	return false;
1101	}
1102
1103	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1104	\|\| offTransport + cbTransport > pSkb->len
1105	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1106	{
1107	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1108	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1109	return false;
1110	}
1111
1112	/*
1113	* Check the TCP/UDP bits.
1114	*/
1115	if (uProtocol == RTNETIPV4_PROT_TCP)
1116	{
1117	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1118	if (RT_UNLIKELY(!pTcp))
1119	{
1120	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1121	return false;
1122	}
1123
1124	cbTransportHdr = pTcp->th_off * 4;
1125	pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1126	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1127	\|\| cbTransportHdr > cbTransport
1128	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1129	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1130	{
1131	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1132	return false;
1133	}
1134
1135	}
1136	else
1137	{
1138	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1139	cbTransportHdr = sizeof(RTNETUDP);
1140	pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1141	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1142	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1143	{
1144	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1145	return false;
1146	}
1147	}
1148
1149	/*
1150	* We're good, init the GSO context.
1151	*/
1152	pGsoCtx->u8Type = enmGsoType;
1153	pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1154	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1155	pGsoCtx->offHdr1 = pSkb->mac_len;
1156	pGsoCtx->offHdr2 = offTransport;
1157	pGsoCtx->u8Unused = 0;
1158
1159	return true;
1160	}
1161
1162	/**
1163	* Forward the socket buffer as a GSO internal network frame.
1164	*
1165	* @returns IPRT status code.
1166	* @param pThis The net filter instance.
1167	* @param pSkb The GSO socket buffer.
1168	* @param fSrc The source.
1169	* @param pGsoCtx Where to return the GSO context on success.
1170	*/
1171	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1172	{
1173	int rc;
1174	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1175	if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1176	{
1177	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1178	if (RT_LIKELY(pSG))
1179	{
1180	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1181
1182	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1183	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1184
1185	vboxNetFltLinuxDestroySG(pSG);
1186	rc = VINF_SUCCESS;
1187	}
1188	else
1189	{
1190	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1191	rc = VERR_NO_MEMORY;
1192	}
1193	}
1194	else
1195	{
1196	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1197	rc = VERR_INTERNAL_ERROR_3;
1198	}
1199
1200	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1201	dev_kfree_skb(pSkb);
1202	return rc;
1203	}
1204
1205	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1206
1207	/**
1208	* Worker for vboxNetFltLinuxForwardToIntNet.
1209	*
1210	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1211	* @param pThis The net filter instance.
1212	* @param pBuf The socket buffer.
1213	* @param fSrc The source.
1214	*/
1215	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1216	{
1217	int rc;
1218	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1219	if (cSegs <= MAX_SKB_FRAGS + 1)
1220	{
1221	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1222	if (RT_LIKELY(pSG))
1223	{
1224	if (fSrc & INTNETTRUNKDIR_WIRE)
1225	{
1226	/*
1227	* The packet came from wire, ethernet header was removed by device driver.
1228	* Restore it using mac_len field. This takes into account VLAN headers too.
1229	*/
1230	skb_push(pBuf, pBuf->mac_len);
1231	}
1232
1233	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /pGsoCtx/);
1234
1235	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1236	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1237
1238	vboxNetFltLinuxDestroySG(pSG);
1239	rc = VINF_SUCCESS;
1240	}
1241	else
1242	{
1243	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1244	rc = VERR_NO_MEMORY;
1245	}
1246	}
1247	else
1248	{
1249	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1250	rc = VERR_INTERNAL_ERROR_3;
1251	}
1252
1253	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1254	dev_kfree_skb(pBuf);
1255	return rc;
1256	}
1257
1258	/**
1259	*
1260	* @param pBuf The socket buffer. This is consumed by this function.
1261	*/
1262	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1263	{
1264	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1265
1266	#ifdef VBOXNETFLT_WITH_GSO
1267	if (skb_is_gso(pBuf))
1268	{
1269	PDMNETWORKGSO GsoCtx;
1270	Log3(("vboxNetFltLinuxForwardToIntNet: 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 ip_summed=%d\n",
1271	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, pBuf->ip_summed));
1272	# ifdef VBOXNETFLT_WITH_GSO_RECV
1273	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1274	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1275	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1276	else
1277	# endif
1278	{
1279	/* Need to segment the packet */
1280	struct sk_buff *pNext;
1281	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1282	if (IS_ERR(pSegment))
1283	{
1284	dev_kfree_skb(pBuf);
1285	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1286	return;
1287	}
1288
1289	for (; pSegment; pSegment = pNext)
1290	{
1291	Log3(("vboxNetFltLinuxForwardToIntNet: segment 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",
1292	pSegment->len, pSegment->data_len, pSegment->truesize, pSegment->next, skb_shinfo(pSegment)->nr_frags, skb_shinfo(pSegment)->gso_size, skb_shinfo(pSegment)->gso_segs, skb_shinfo(pSegment)->gso_type, skb_shinfo(pSegment)->frag_list, pSegment->pkt_type));
1293	pNext = pSegment->next;
1294	pSegment->next = 0;
1295	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1296	}
1297	dev_kfree_skb(pBuf);
1298	}
1299	}
1300	else
1301	#endif /* VBOXNETFLT_WITH_GSO */
1302	{
1303	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1304	{
1305	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1306	/*
1307	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1308	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1309	* header length from the header itself and reconstruct 'h' pointer
1310	* to TCP (or whatever) header.
1311	*/
1312	unsigned char *tmp = pBuf->h.raw;
1313	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1314	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1315	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1316	if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1317	{
1318	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1319	dev_kfree_skb(pBuf);
1320	return;
1321	}
1322	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1323	/* Restore the original (wrong) pointer. */
1324	pBuf->h.raw = tmp;
1325	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1326	}
1327	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1328	}
1329	}
1330
1331	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1332	/**
1333	* Work queue handler that forwards the socket buffers queued by
1334	* vboxNetFltLinuxPacketHandler to the internal network.
1335	*
1336	* @param pWork The work queue.
1337	*/
1338	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1339	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1340	# else
1341	static void vboxNetFltLinuxXmitTask(void *pWork)
1342	# endif
1343	{
1344	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1345	struct sk_buff *pBuf;
1346
1347	Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1348
1349	/*
1350	* Active? Retain the instance and increment the busy counter.
1351	*/
1352	if (vboxNetFltTryRetainBusyActive(pThis))
1353	{
1354	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1355	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1356
1357	vboxNetFltRelease(pThis, true /* fBusy */);
1358	}
1359	else
1360	{
1361	/** @todo Shouldn't we just drop the packets here? There is little point in
1362	* making them accumulate when the VM is paused and it'll only waste
1363	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1364	* before start draining the packets (goes for the intnet ring buf
1365	* too)? */
1366	}
1367	}
1368	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1369
1370	/**
1371	* Reports the GSO capabilities of the hardware NIC.
1372	*
1373	* @param pThis The net filter instance. The caller hold a
1374	* reference to this.
1375	*/
1376	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1377	{
1378	#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1379	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1380	{
1381	struct net_device *pDev;
1382	PINTNETTRUNKSWPORT pSwitchPort;
1383	unsigned int fFeatures;
1384	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1385
1386	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1387
1388	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1389	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1390	if (pDev)
1391	fFeatures = pDev->features;
1392	else
1393	fFeatures = 0;
1394
1395	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1396
1397	if (pThis->pSwitchPort)
1398	{
1399	/* Set/update the GSO capabilities of the NIC. */
1400	uint32_t fGsoCapabilites = 0;
1401	if (fFeatures & NETIF_F_TSO)
1402	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1403	if (fFeatures & NETIF_F_TSO6)
1404	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1405	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1406	if (fFeatures & NETIF_F_UFO)
1407	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1408	if (fFeatures & NETIF_F_UFO)
1409	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1410	# endif
1411	Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1412	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1413	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : "",
1414	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)) ? "ufo " : "",
1415	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)) ? "ufo6 " : ""));
1416	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1417	}
1418
1419	vboxNetFltRelease(pThis, true /fBusy/);
1420	}
1421	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1422	}
1423
1424	/**
1425	* Helper that determines whether the host (ignoreing us) is operating the
1426	* interface in promiscuous mode or not.
1427	*/
1428	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1429	{
1430	bool fRc = false;
1431	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1432	if (pDev)
1433	{
1434	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1435	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1436	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1437	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1438	}
1439	return fRc;
1440	}
1441
1442	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36)
1443	/**
1444	* Helper for detecting TAP devices.
1445	*/
1446	static bool vboxNetFltIsTapDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1447	{
1448	if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1449	{
1450	struct ethtool_drvinfo Info;
1451
1452	memset(&Info, 0, sizeof(Info));
1453	Info.cmd = ETHTOOL_GDRVINFO;
1454	pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1455	Log3(("vboxNetFltIsTapDevice: driver=%s version=%s bus_info=%s\n",
1456	Info.driver, Info.version, Info.bus_info));
1457
1458	return !strncmp(Info.driver, "tun", 4)
1459	&& !strncmp(Info.bus_info, "tap", 4);
1460	}
1461
1462	return false;
1463	}
1464
1465	/**
1466	* Helper for updating the link state of TAP devices.
1467	* Only TAP devices are affected.
1468	*/
1469	static void vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1470	{
1471	if (vboxNetFltIsTapDevice(pThis, pDev))
1472	{
1473	Log3(("vboxNetFltSetTapLinkState: bringing %s tap device link state\n",
1474	fLinkUp ? "up" : "down"));
1475	netif_tx_lock_bh(pDev);
1476	if (fLinkUp)
1477	netif_carrier_on(pDev);
1478	else
1479	netif_carrier_off(pDev);
1480	netif_tx_unlock_bh(pDev);
1481	}
1482	}
1483	#else /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1484	DECLINLINE(void) vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1485	{
1486	/* Nothing to do for pre-2.6.36 kernels. */
1487	}
1488	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1489
1490	/**
1491	* Internal worker for vboxNetFltLinuxNotifierCallback.
1492	*
1493	* @returns VBox status code.
1494	* @param pThis The instance.
1495	* @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1496	* flood the release log.
1497	*/
1498	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1499	{
1500	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1501	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1502
1503	/*
1504	* Retain and store the device.
1505	*/
1506	dev_hold(pDev);
1507
1508	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1509	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1510	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1511
1512	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1513	pDev, pDev->name,
1514	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1515	netdev_refcnt_read(pDev)
1516	#else
1517	atomic_read(&pDev->refcnt)
1518	#endif
1519	));
1520	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1521	pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1522
1523	/* Get the mac address while we still have a valid net_device reference. */
1524	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1525
1526	/*
1527	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1528	*/
1529	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1530	pThis->u.s.PacketType.dev = pDev;
1531	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1532	dev_add_pack(&pThis->u.s.PacketType);
1533
1534	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1535	vboxNetFltLinuxHookDev(pThis, pDev);
1536	#endif
1537
1538	/*
1539	* If attaching to TAP interface we need to bring the link state up
1540	* starting from 2.6.36 kernel.
1541	*/
1542	vboxNetFltSetTapLinkState(pThis, pDev, true);
1543
1544	/*
1545	* Set indicators that require the spinlock. Be abit paranoid about racing
1546	* the device notification handle.
1547	*/
1548	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1549	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1550	if (pDev)
1551	{
1552	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1553	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1554	pDev = NULL; /* don't dereference it */
1555	}
1556	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1557	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1558
1559	/*
1560	* If the above succeeded report GSO capabilities, if not undo and
1561	* release the device.
1562	*/
1563	if (!pDev)
1564	{
1565	Assert(pThis->pSwitchPort);
1566	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1567	{
1568	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1569	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1570	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1571	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1572	vboxNetFltRelease(pThis, true /fBusy/);
1573	}
1574	}
1575	else
1576	{
1577	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1578	vboxNetFltLinuxUnhookDev(pThis, pDev);
1579	#endif
1580	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1581	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1582	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1583	dev_put(pDev);
1584	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n",
1585	pDev, pDev->name,
1586	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1587	netdev_refcnt_read(pDev)
1588	#else
1589	atomic_read(&pDev->refcnt)
1590	#endif
1591	));
1592	}
1593
1594	LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1595	return VINF_SUCCESS;
1596	}
1597
1598
1599	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1600	{
1601	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1602
1603	Assert(!pThis->fDisconnectedFromHost);
1604
1605	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1606	vboxNetFltLinuxUnhookDev(pThis, pDev);
1607	#endif
1608
1609	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1610	ASMAtomicWriteBool(&pThis->u.s.fRegistered, false);
1611	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1612	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1613	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1614
1615	dev_remove_pack(&pThis->u.s.PacketType);
1616	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1617	skb_queue_purge(&pThis->u.s.XmitQueue);
1618	#endif
1619	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
1620	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1621	pDev, pDev->name,
1622	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1623	netdev_refcnt_read(pDev)
1624	#else
1625	atomic_read(&pDev->refcnt)
1626	#endif
1627	));
1628	dev_put(pDev);
1629
1630	return NOTIFY_OK;
1631	}
1632
1633	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1634	{
1635	/* Check if we are not suspended and promiscuous mode has not been set. */
1636	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1637	&& !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1638	{
1639	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1640	dev_set_promiscuity(pDev, 1);
1641	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1642	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1643	}
1644	else
1645	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1646	return NOTIFY_OK;
1647	}
1648
1649	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1650	{
1651	/* Undo promiscuous mode if we has set it. */
1652	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1653	{
1654	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1655	dev_set_promiscuity(pDev, -1);
1656	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1657	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1658	}
1659	else
1660	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1661	return NOTIFY_OK;
1662	}
1663
1664	#ifdef LOG_ENABLED
1665	/** Stringify the NETDEV_XXX constants. */
1666	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
1667	{
1668	const char *pszEvent = "NETDRV_<unknown>";
1669	switch (ulEventType)
1670	{
1671	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
1672	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
1673	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
1674	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
1675	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
1676	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
1677	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
1678	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
1679	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
1680	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
1681	# ifdef NETDEV_FEAT_CHANGE
1682	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
1683	# endif
1684	}
1685	return pszEvent;
1686	}
1687	#endif /* LOG_ENABLED */
1688
1689	/**
1690	* Callback for listening to netdevice events.
1691	*
1692	* This works the rediscovery, clean up on unregistration, promiscuity on
1693	* up/down, and GSO feature changes from ethtool.
1694	*
1695	* @returns NOTIFY_OK
1696	* @param self Pointer to our notifier registration block.
1697	* @param ulEventType The event.
1698	* @param ptr Event specific, but it is usually the device it
1699	* relates to.
1700	*/
1701	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
1702
1703	{
1704	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
1705	struct net_device pDev = (struct net_device )ptr;
1706	int rc = NOTIFY_OK;
1707
1708	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1709	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1710	if ( ulEventType == NETDEV_REGISTER
1711	&& !strcmp(pDev->name, pThis->szName))
1712	{
1713	vboxNetFltLinuxAttachToInterface(pThis, pDev);
1714	}
1715	else
1716	{
1717	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1718	if (pDev == ptr)
1719	{
1720	switch (ulEventType)
1721	{
1722	case NETDEV_UNREGISTER:
1723	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
1724	break;
1725	case NETDEV_UP:
1726	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
1727	break;
1728	case NETDEV_GOING_DOWN:
1729	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
1730	break;
1731	case NETDEV_CHANGENAME:
1732	break;
1733	#ifdef NETDEV_FEAT_CHANGE
1734	case NETDEV_FEAT_CHANGE:
1735	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1736	break;
1737	#endif
1738	}
1739	}
1740	}
1741
1742	return rc;
1743	}
1744
1745	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
1746	{
1747	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
1748	}
1749
1750	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
1751	{
1752	struct net_device * pDev;
1753	int err;
1754	int rc = VINF_SUCCESS;
1755	NOREF(pvIfData);
1756
1757	LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
1758
1759	pDev = vboxNetFltLinuxRetainNetDev(pThis);
1760	if (pDev)
1761	{
1762	/*
1763	* Create a sk_buff for the gather list and push it onto the wire.
1764	*/
1765	if (fDst & INTNETTRUNKDIR_WIRE)
1766	{
1767	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
1768	if (pBuf)
1769	{
1770	vboxNetFltDumpPacket(pSG, true, "wire", 1);
1771	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1772	Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
1773	err = dev_queue_xmit(pBuf);
1774	if (err)
1775	rc = RTErrConvertFromErrno(err);
1776	}
1777	else
1778	rc = VERR_NO_MEMORY;
1779	}
1780
1781	/*
1782	* Create a sk_buff for the gather list and push it onto the host stack.
1783	*/
1784	if (fDst & INTNETTRUNKDIR_HOST)
1785	{
1786	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
1787	if (pBuf)
1788	{
1789	vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
1790	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1791	Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
1792	err = netif_rx_ni(pBuf);
1793	if (err)
1794	rc = RTErrConvertFromErrno(err);
1795	}
1796	else
1797	rc = VERR_NO_MEMORY;
1798	}
1799
1800	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1801	}
1802
1803	return rc;
1804	}
1805
1806
1807	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
1808	{
1809	struct net_device * pDev;
1810
1811	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
1812	pThis, pThis->szName, fActive?"true":"false",
1813	pThis->fDisablePromiscuous?"true":"false"));
1814
1815	if (pThis->fDisablePromiscuous)
1816	return;
1817
1818	pDev = vboxNetFltLinuxRetainNetDev(pThis);
1819	if (pDev)
1820	{
1821	/*
1822	* This api is a bit weird, the best reference is the code.
1823	*
1824	* Also, we have a bit or race conditions wrt the maintenance of
1825	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
1826	*/
1827	#ifdef LOG_ENABLED
1828	u_int16_t fIf;
1829	unsigned const cPromiscBefore = pDev->promiscuity;
1830	#endif
1831	if (fActive)
1832	{
1833	Assert(!pThis->u.s.fPromiscuousSet);
1834
1835	rtnl_lock();
1836	dev_set_promiscuity(pDev, 1);
1837	rtnl_unlock();
1838	pThis->u.s.fPromiscuousSet = true;
1839	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1840	}
1841	else
1842	{
1843	if (pThis->u.s.fPromiscuousSet)
1844	{
1845	rtnl_lock();
1846	dev_set_promiscuity(pDev, -1);
1847	rtnl_unlock();
1848	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1849	}
1850	pThis->u.s.fPromiscuousSet = false;
1851
1852	#ifdef LOG_ENABLED
1853	fIf = dev_get_flags(pDev);
1854	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
1855	#endif
1856	}
1857
1858	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1859	}
1860	}
1861
1862
1863	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
1864	{
1865	/*
1866	* Remove packet handler when we get disconnected from internal switch as
1867	* we don't want the handler to forward packets to disconnected switch.
1868	*/
1869	dev_remove_pack(&pThis->u.s.PacketType);
1870	return VINF_SUCCESS;
1871	}
1872
1873
1874	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
1875	{
1876	/*
1877	* Report the GSO capabilities of the host and device (if connected).
1878	* Note! No need to mark ourselves busy here.
1879	*/
1880	/** @todo duplicate work here now? Attach */
1881	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1882	Log3(("vboxNetFltOsConnectIt: reporting host tso tso6 ufo\n"));
1883	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
1884	0
1885	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
1886	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
1887	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
1888	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1889	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
1890	# endif
1891	, INTNETTRUNKDIR_HOST);
1892
1893	#endif
1894	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1895
1896	return VINF_SUCCESS;
1897	}
1898
1899
1900	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
1901	{
1902	struct net_device *pDev;
1903	bool fRegistered;
1904	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1905
1906	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1907	vboxNetFltLinuxUnhookDev(pThis, NULL);
1908	#endif
1909
1910	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
1911	* unlikely, but none the less). Since it doesn't actually update the
1912	* state (just reads it), it is likely to panic in some interesting
1913	* ways. */
1914
1915	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1916	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1917	fRegistered = ASMAtomicUoReadBool(&pThis->u.s.fRegistered);
1918	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1919
1920	if (fRegistered)
1921	{
1922	vboxNetFltSetTapLinkState(pThis, pDev, false);
1923
1924	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1925	skb_queue_purge(&pThis->u.s.XmitQueue);
1926	#endif
1927	Log(("vboxNetFltOsDeleteInstance: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
1928	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
1929	pDev, pDev->name,
1930	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1931	netdev_refcnt_read(pDev)
1932	#else
1933	atomic_read(&pDev->refcnt)
1934	#endif
1935	));
1936	dev_put(pDev);
1937	}
1938	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
1939	unregister_netdevice_notifier(&pThis->u.s.Notifier);
1940	module_put(THIS_MODULE);
1941	}
1942
1943
1944	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
1945	{
1946	int err;
1947	NOREF(pvContext);
1948
1949	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
1950	err = register_netdevice_notifier(&pThis->u.s.Notifier);
1951	if (err)
1952	return VERR_INTNET_FLT_IF_FAILED;
1953	if (!pThis->u.s.fRegistered)
1954	{
1955	unregister_netdevice_notifier(&pThis->u.s.Notifier);
1956	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
1957	return VERR_INTNET_FLT_IF_NOT_FOUND;
1958	}
1959
1960	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
1961	if ( pThis->fDisconnectedFromHost
1962	\|\| !try_module_get(THIS_MODULE))
1963	return VERR_INTNET_FLT_IF_FAILED;
1964
1965	return VINF_SUCCESS;
1966	}
1967
1968	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
1969	{
1970	/*
1971	* Init the linux specific members.
1972	*/
1973	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1974	pThis->u.s.fRegistered = false;
1975	pThis->u.s.fPromiscuousSet = false;
1976	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
1977	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1978	skb_queue_head_init(&pThis->u.s.XmitQueue);
1979	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1980	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
1981	# else
1982	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
1983	# endif
1984	#endif
1985
1986	return VINF_SUCCESS;
1987	}
1988
1989
1990	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
1991	{
1992	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
1993	}
1994
1995
1996	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
1997	{
1998	/* Nothing to do */
1999	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2000	return VINF_SUCCESS;
2001	}
2002
2003
2004	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2005	{
2006	/* Nothing to do */
2007	NOREF(pThis); NOREF(pvIfData);
2008	return VINF_SUCCESS;
2009	}
2010

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 40733

Download in other formats: