VBoxNetFlt-linux.c@ 55679

Last change on this file since 55679 was 55679, checked in by vboxsync, 10 years ago
VBoxNetFlt/linux: dev_net() appeared in 2.6.26
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 79.2 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 55679 2015-05-06 01:00:14Z vboxsync $ */
2	/** @file
3	* VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2014 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/inetdevice.h>
31	#include <linux/ip.h>
32	#include <linux/if_vlan.h>
33	#include <net/if_inet6.h>
34	#include <net/addrconf.h>
35
36	#include <VBox/log.h>
37	#include <VBox/err.h>
38	#include <VBox/intnetinline.h>
39	#include <VBox/vmm/pdmnetinline.h>
40	#include <VBox/param.h>
41	#include <iprt/alloca.h>
42	#include <iprt/assert.h>
43	#include <iprt/spinlock.h>
44	#include <iprt/semaphore.h>
45	#include <iprt/initterm.h>
46	#include <iprt/process.h>
47	#include <iprt/mem.h>
48	#include <iprt/net.h>
49	#include <iprt/log.h>
50	#include <iprt/mp.h>
51	#include <iprt/mem.h>
52	#include <iprt/time.h>
53
54	#define VBOXNETFLT_OS_SPECFIC 1
55	#include "../VBoxNetFltInternal.h"
56
57	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 20, 0)
58	# define vlan_tx_tag_get(skb) skb_vlan_tag_get(skb)
59	# define vlan_tx_tag_present(skb) skb_vlan_tag_present(skb)
60	#endif
61
62
63	/*******************************************************************************
64	* Defined Constants And Macros *
65	*******************************************************************************/
66	#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
67	#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
68	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
69	# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
70	#endif
71
72	#if 0
73	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
74	# define VBOX_NETDEV_NAME(dev) netdev_name(dev)
75	#else
76	# define VBOX_NETDEV_NAME(dev) ((dev)->reg_state != NETREG_REGISTERED ? "(unregistered net_device)" : (dev)->name)
77	#endif
78	#endif
79
80	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 26)
81	# define VBOX_DEV_NET(dev) dev_net(dev)
82	#else
83	# define VBOX_DEV_NET(dev) ((dev)->nd_net)
84	#endif
85
86	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
87	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
88	# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
89	#else
90	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
91	# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
92	#endif
93
94	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
95	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
96	#else
97	# define CHECKSUM_PARTIAL CHECKSUM_HW
98	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
99	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
100	# else
101	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
102	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
103	# else
104	# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
105	# endif
106	/* Versions prior 2.6.10 use stats for both bstats and qstats */
107	# define bstats stats
108	# define qstats stats
109	# endif
110	#endif
111
112	#ifndef NET_IP_ALIGN
113	# define NET_IP_ALIGN 2
114	#endif
115
116	#if 0
117	/** Create scatter / gather segments for fragments. When not used, we will
118	* linearize the socket buffer before creating the internal networking SG. */
119	# define VBOXNETFLT_SG_SUPPORT 1
120	#endif
121
122	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
123	/** Indicates that the linux kernel may send us GSO frames. */
124	# define VBOXNETFLT_WITH_GSO 1
125
126	/** This enables or disables the transmitting of GSO frame from the internal
127	* network and to the host. */
128	# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
129
130	# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
131	/** This enables or disables the transmitting of GSO frame from the internal
132	* network and to the wire. */
133	# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
134	# endif
135
136	/** This enables or disables the forwarding/flooding of GSO frame from the host
137	* to the internal network. */
138	# define VBOXNETFLT_WITH_GSO_RECV 1
139
140	#endif
141
142	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
143	/** This enables or disables handling of GSO frames coming from the wire (GRO). */
144	# define VBOXNETFLT_WITH_GRO 1
145	#endif
146	/*
147	* GRO support was backported to RHEL 5.4
148	*/
149	#ifdef RHEL_RELEASE_CODE
150	# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
151	# define VBOXNETFLT_WITH_GRO 1
152	# endif
153	#endif
154
155	/*******************************************************************************
156	* Internal Functions *
157	*******************************************************************************/
158	static int VBoxNetFltLinuxInit(void);
159	static void VBoxNetFltLinuxUnload(void);
160	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
161
162
163	/*******************************************************************************
164	* Global Variables *
165	*******************************************************************************/
166	/**
167	* The (common) global data.
168	*/
169	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
170
171	module_init(VBoxNetFltLinuxInit);
172	module_exit(VBoxNetFltLinuxUnload);
173
174	MODULE_AUTHOR(VBOX_VENDOR);
175	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
176	MODULE_LICENSE("GPL");
177	#ifdef MODULE_VERSION
178	MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
179	#endif
180
181
182	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
183	unsigned dev_get_flags(const struct net_device *dev)
184	{
185	unsigned flags;
186
187	flags = (dev->flags & ~(IFF_PROMISC \|
188	IFF_ALLMULTI \|
189	IFF_RUNNING)) \|
190	(dev->gflags & (IFF_PROMISC \|
191	IFF_ALLMULTI));
192
193	if (netif_running(dev) && netif_carrier_ok(dev))
194	flags \|= IFF_RUNNING;
195
196	return flags;
197	}
198	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
199
200
201	/**
202	* Initialize module.
203	*
204	* @returns appropriate status code.
205	*/
206	static int __init VBoxNetFltLinuxInit(void)
207	{
208	int rc;
209	/*
210	* Initialize IPRT.
211	*/
212	rc = RTR0Init(0);
213	if (RT_SUCCESS(rc))
214	{
215	Log(("VBoxNetFltLinuxInit\n"));
216
217	/*
218	* Initialize the globals and connect to the support driver.
219	*
220	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
221	* for establishing the connect to the support driver.
222	*/
223	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
224	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
225	if (RT_SUCCESS(rc))
226	{
227	LogRel(("VBoxNetFlt: Successfully started.\n"));
228	return 0;
229	}
230
231	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
232	RTR0Term();
233	}
234	else
235	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
236
237	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
238	return -RTErrConvertToErrno(rc);
239	}
240
241
242	/**
243	* Unload the module.
244	*
245	* @todo We have to prevent this if we're busy!
246	*/
247	static void __exit VBoxNetFltLinuxUnload(void)
248	{
249	int rc;
250	Log(("VBoxNetFltLinuxUnload\n"));
251	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
252
253	/*
254	* Undo the work done during start (in reverse order).
255	*/
256	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
257	AssertRC(rc); NOREF(rc);
258
259	RTR0Term();
260
261	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
262
263	Log(("VBoxNetFltLinuxUnload - done\n"));
264	}
265
266
267	/**
268	* We filter traffic from the host to the internal network
269	* before it reaches the NIC driver.
270	*
271	* The current code uses a very ugly hack overriding hard_start_xmit
272	* callback in the device structure, but it has been shown to give us a
273	* performance boost of 60-100% though. Eventually we have to find some
274	* less hacky way of getting this job done.
275	*/
276	#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
277
278	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
279
280	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
281
282	# include <linux/ethtool.h>
283
284	typedef struct ethtool_ops OVR_OPSTYPE;
285	# define OVR_OPS ethtool_ops
286	# define OVR_XMIT pfnStartXmit
287
288	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
289
290	typedef struct net_device_ops OVR_OPSTYPE;
291	# define OVR_OPS netdev_ops
292	# define OVR_XMIT pOrgOps->ndo_start_xmit
293
294	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
295
296	/**
297	* The overridden net_device_ops of the device we're attached to.
298	*
299	* As there is no net_device_ops structure in pre-2.6.29 kernels we override
300	* ethtool_ops instead along with hard_start_xmit callback in net_device
301	* structure.
302	*
303	* This is a very dirty hack that was created to explore how much we can improve
304	* the host to guest transfers by not CC'ing the NIC. It turns out to be
305	* the only way to filter outgoing packets for devices without TX queue.
306	*/
307	typedef struct VBoxNetDeviceOpsOverride
308	{
309	/** Our overridden ops. */
310	OVR_OPSTYPE Ops;
311	/** Magic word. */
312	uint32_t u32Magic;
313	/** Pointer to the original ops. */
314	OVR_OPSTYPE const *pOrgOps;
315	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
316	/** Pointer to the original hard_start_xmit function. */
317	int (pfnStartXmit)(struct sk_buff pSkb, struct net_device *pDev);
318	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
319	/** Pointer to the net filter instance. */
320	PVBOXNETFLTINS pVBoxNetFlt;
321	/** The number of filtered packages. */
322	uint64_t cFiltered;
323	/** The total number of packets */
324	uint64_t cTotal;
325	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
326	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
327	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
328
329	/**
330	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
331	* because they belong on the internal network.
332	*
333	* @returns NETDEV_TX_XXX.
334	* @param pSkb The socket buffer to transmit.
335	* @param pDev The net device.
336	*/
337	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
338	{
339	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
340	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
341	PCRTNETETHERHDR pEtherHdr;
342	PINTNETTRUNKSWPORT pSwitchPort;
343	uint32_t cbHdrs;
344
345
346	/*
347	* Validate the override structure.
348	*
349	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
350	* to be production quality code, we would have to be much more
351	* careful here and avoid the race.
352	*/
353	if ( !VALID_PTR(pOverride)
354	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
355	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
356	\|\| !VALID_PTR(pOverride->pOrgOps)
357	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
358	)
359	{
360	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
361	dev_kfree_skb(pSkb);
362	return NETDEV_TX_OK;
363	}
364	pOverride->cTotal++;
365
366	/*
367	* Do the filtering base on the default OUI of our virtual NICs
368	*
369	* Note! In a real solution, we would ask the switch whether the
370	* destination MAC is 100% to be on the internal network and then
371	* drop it.
372	*/
373	cbHdrs = skb_headlen(pSkb);
374	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
375	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
376	if ( pEtherHdr
377	&& VALID_PTR(pOverride->pVBoxNetFlt)
378	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
379	&& VALID_PTR(pSwitchPort)
380	&& cbHdrs >= 6)
381	{
382	INTNETSWDECISION enmDecision;
383
384	/** @todo consider reference counting, etc. */
385	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
386	if (enmDecision == INTNETSWDECISION_INTNET)
387	{
388	dev_kfree_skb(pSkb);
389	pOverride->cFiltered++;
390	return NETDEV_TX_OK;
391	}
392	}
393
394	return pOverride->OVR_XMIT(pSkb, pDev);
395	}
396
397	/**
398	* Hooks the device ndo_start_xmit operation of the device.
399	*
400	* @param pThis The net filter instance.
401	* @param pDev The net device.
402	*/
403	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
404	{
405	PVBOXNETDEVICEOPSOVERRIDE pOverride;
406
407	/* Cancel override if ethtool_ops is missing (host-only case, @bugref{5712}) */
408	if (!VALID_PTR(pDev->OVR_OPS))
409	return;
410	pOverride = RTMemAlloc(sizeof(*pOverride));
411	if (!pOverride)
412	return;
413	pOverride->pOrgOps = pDev->OVR_OPS;
414	pOverride->Ops = *pDev->OVR_OPS;
415	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
416	pOverride->pfnStartXmit = pDev->hard_start_xmit;
417	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
418	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
419	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
420	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
421	pOverride->cTotal = 0;
422	pOverride->cFiltered = 0;
423	pOverride->pVBoxNetFlt = pThis;
424
425	RTSpinlockAcquire(pThis->hSpinlock); /* (this isn't necessary, but so what) */
426	ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
427	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
428	ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
429	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
430	RTSpinlockRelease(pThis->hSpinlock);
431	}
432
433	/**
434	* Undos what vboxNetFltLinuxHookDev did.
435	*
436	* @param pThis The net filter instance.
437	* @param pDev The net device. Can be NULL, in which case
438	* we'll try retrieve it from @a pThis.
439	*/
440	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
441	{
442	PVBOXNETDEVICEOPSOVERRIDE pOverride;
443
444	RTSpinlockAcquire(pThis->hSpinlock);
445	if (!pDev)
446	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
447	if (VALID_PTR(pDev))
448	{
449	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
450	if ( VALID_PTR(pOverride)
451	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
452	&& VALID_PTR(pOverride->pOrgOps)
453	)
454	{
455	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
456	ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
457	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
458	ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
459	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
460	}
461	else
462	pOverride = NULL;
463	}
464	else
465	pOverride = NULL;
466	RTSpinlockRelease(pThis->hSpinlock);
467
468	if (pOverride)
469	{
470	printk("vboxnetflt: %llu out of %llu packets were not sent (directed to host)\n", pOverride->cFiltered, pOverride->cTotal);
471	RTMemFree(pOverride);
472	}
473	}
474
475	#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
476
477
478	/**
479	* Reads and retains the host interface handle.
480	*
481	* @returns The handle, NULL if detached.
482	* @param pThis
483	*/
484	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
485	{
486	#if 0
487	struct net_device *pDev = NULL;
488
489	Log(("vboxNetFltLinuxRetainNetDev\n"));
490	/*
491	* Be careful here to avoid problems racing the detached callback.
492	*/
493	RTSpinlockAcquire(pThis->hSpinlock);
494	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
495	{
496	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
497	if (pDev)
498	{
499	dev_hold(pDev);
500	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
501	pDev, pDev->name,
502	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
503	netdev_refcnt_read(pDev)
504	#else
505	atomic_read(&pDev->refcnt)
506	#endif
507	));
508	}
509	}
510	RTSpinlockRelease(pThis->hSpinlock);
511
512	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
513	return pDev;
514	#else
515	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
516	#endif
517	}
518
519
520	/**
521	* Release the host interface handle previously retained
522	* by vboxNetFltLinuxRetainNetDev.
523	*
524	* @param pThis The instance.
525	* @param pDev The vboxNetFltLinuxRetainNetDev
526	* return value, NULL is fine.
527	*/
528	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
529	{
530	#if 0
531	Log(("vboxNetFltLinuxReleaseNetDev\n"));
532	NOREF(pThis);
533	if (pDev)
534	{
535	dev_put(pDev);
536	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
537	pDev, pDev->name,
538	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
539	netdev_refcnt_read(pDev)
540	#else
541	atomic_read(&pDev->refcnt)
542	#endif
543	));
544	}
545	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
546	#endif
547	}
548
549	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
550	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
551
552	/**
553	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
554	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
555	*
556	* @returns true / false accordingly.
557	* @param pBuf The sk_buff.
558	*/
559	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
560	{
561	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
562	}
563
564
565	/**
566	* Checks whether this SG list contains a GSO packet.
567	*
568	* @returns true / false accordingly.
569	* @param pSG The (scatter/)gather list.
570	*/
571	DECLINLINE(bool) vboxNetFltLinuxIsGso(PINTNETSG pSG)
572	{
573	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
574	return !((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type == PDMNETWORKGSOTYPE_INVALID);
575	#else /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
576	return false;
577	#endif /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
578	}
579
580
581	/**
582	* Find out the frame size (of a single segment in case of GSO frames).
583	*
584	* @returns the frame size.
585	* @param pSG The (scatter/)gather list.
586	*/
587	DECLINLINE(uint32_t) vboxNetFltLinuxFrameSize(PINTNETSG pSG)
588	{
589	uint16_t u16Type = 0;
590	uint32_t cbVlanTag = 0;
591	if (pSG->aSegs[0].cb >= sizeof(RTNETETHERHDR))
592	u16Type = RT_BE2H_U16(((PCRTNETETHERHDR)pSG->aSegs[0].pv)->EtherType);
593	else if (pSG->cbTotal >= sizeof(RTNETETHERHDR))
594	{
595	uint32_t off = RT_OFFSETOF(RTNETETHERHDR, EtherType);
596	uint32_t i;
597	for (i = 0; i < pSG->cSegsUsed; ++i)
598	{
599	if (off <= pSG->aSegs[i].cb)
600	{
601	if (off + sizeof(uint16_t) <= pSG->aSegs[i].cb)
602	u16Type = RT_BE2H_U16((uint16_t )((uintptr_t)pSG->aSegs[i].pv + off));
603	else if (i + 1 < pSG->cSegsUsed)
604	u16Type = RT_BE2H_U16( ((uint16_t)( ((uint8_t *)pSG->aSegs[i].pv)[off] ) << 8)
605	+ (uint8_t )pSG->aSegs[i + 1].pv); /* ASSUMES no empty segments! */
606	/* else: frame is too short. */
607	break;
608	}
609	off -= pSG->aSegs[i].cb;
610	}
611	}
612	if (u16Type == RTNET_ETHERTYPE_VLAN)
613	cbVlanTag = 4;
614	return (vboxNetFltLinuxIsGso(pSG) ? (uint32_t)pSG->GsoCtx.cbMaxSeg + pSG->GsoCtx.cbHdrsTotal : pSG->cbTotal) - cbVlanTag;
615	}
616
617
618	/**
619	* Internal worker that create a linux sk_buff for a
620	* (scatter/)gather list.
621	*
622	* @returns Pointer to the sk_buff.
623	* @param pThis The instance.
624	* @param pSG The (scatter/)gather list.
625	* @param fDstWire Set if the destination is the wire.
626	*/
627	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
628	{
629	struct sk_buff *pPkt;
630	struct net_device *pDev;
631	unsigned fGsoType = 0;
632
633	if (pSG->cbTotal == 0)
634	{
635	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
636	return NULL;
637	}
638	Log5(("VBoxNetFlt: Packet to %s of %d bytes (frame=%d).\n", fDstWire?"wire":"host", pSG->cbTotal, vboxNetFltLinuxFrameSize(pSG)));
639	if (fDstWire && (vboxNetFltLinuxFrameSize(pSG) > ASMAtomicReadU32(&pThis->u.s.cbMtu) + 14))
640	{
641	static bool s_fOnce = true;
642	if (s_fOnce)
643	{
644	s_fOnce = false;
645	printk("VBoxNetFlt: Dropped over-sized packet (%d bytes) coming from internal network.\n", vboxNetFltLinuxFrameSize(pSG));
646	}
647	return NULL;
648	}
649
650	/** @todo We should use fragments mapping the SG buffers with large packets.
651	* 256 bytes seems to be the a threshold used a lot for this. It
652	* requires some nasty work on the intnet side though... */
653	/*
654	* Allocate a packet and copy over the data.
655	*/
656	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
657	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
658	if (RT_UNLIKELY(!pPkt))
659	{
660	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
661	pSG->pvUserData = NULL;
662	return NULL;
663	}
664	pPkt->dev = pDev;
665	pPkt->ip_summed = CHECKSUM_NONE;
666
667	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
668	skb_reserve(pPkt, NET_IP_ALIGN);
669
670	/* Copy the segments. */
671	skb_put(pPkt, pSG->cbTotal);
672	IntNetSgRead(pSG, pPkt->data);
673
674	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
675	/*
676	* Setup GSO if used by this packet.
677	*/
678	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
679	{
680	default:
681	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
682	/* fall thru */
683	case PDMNETWORKGSOTYPE_INVALID:
684	fGsoType = 0;
685	break;
686	case PDMNETWORKGSOTYPE_IPV4_TCP:
687	fGsoType = SKB_GSO_TCPV4;
688	break;
689	case PDMNETWORKGSOTYPE_IPV4_UDP:
690	fGsoType = SKB_GSO_UDP;
691	break;
692	case PDMNETWORKGSOTYPE_IPV6_TCP:
693	fGsoType = SKB_GSO_TCPV6;
694	break;
695	}
696	if (fGsoType)
697	{
698	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
699
700	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
701	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
702	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
703
704	/*
705	* We need to set checksum fields even if the packet goes to the host
706	* directly as it may be immediately forwarded by IP layer @bugref{5020}.
707	*/
708	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
709	pPkt->ip_summed = CHECKSUM_PARTIAL;
710	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
711	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
712	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
713	pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
714	else
715	pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
716	# else
717	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
718	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
719	pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
720	else
721	pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
722	# endif
723	if (!fDstWire)
724	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
725	}
726	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
727
728	/*
729	* Finish up the socket buffer.
730	*/
731	pPkt->protocol = eth_type_trans(pPkt, pDev);
732	if (fDstWire)
733	{
734	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
735
736	/* Restore ethernet header back. */
737	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
738	VBOX_SKB_RESET_MAC_HDR(pPkt);
739	}
740	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
741
742	return pPkt;
743	}
744
745
746	/**
747	* Initializes a SG list from an sk_buff.
748	*
749	* @returns Number of segments.
750	* @param pThis The instance.
751	* @param pBuf The sk_buff.
752	* @param pSG The SG.
753	* @param pvFrame The frame pointer, optional.
754	* @param cSegs The number of segments allocated for the SG.
755	* This should match the number in the mbuf exactly!
756	* @param fSrc The source of the frame.
757	* @param pGso Pointer to the GSO context if it's a GSO
758	* internal network frame. NULL if regular frame.
759	*/
760	DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
761	unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
762	{
763	int i;
764	NOREF(pThis);
765
766	Assert(!skb_shinfo(pBuf)->frag_list);
767
768	if (!pGsoCtx)
769	IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /cSegsUsed/);
770	else
771	IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /cSegsUsed/, pGsoCtx);
772
773	#ifdef VBOXNETFLT_SG_SUPPORT
774	pSG->aSegs[0].cb = skb_headlen(pBuf);
775	pSG->aSegs[0].pv = pBuf->data;
776	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
777
778	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
779	{
780	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
781	pSG->aSegs[i+1].cb = pFrag->size;
782	pSG->aSegs[i+1].pv = kmap(pFrag->page);
783	printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
784	pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
785	}
786	++i;
787
788	#else
789	pSG->aSegs[0].cb = pBuf->len;
790	pSG->aSegs[0].pv = pBuf->data;
791	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
792	i = 1;
793	#endif
794
795	pSG->cSegsUsed = i;
796
797	#ifdef PADD_RUNT_FRAMES_FROM_HOST
798	/*
799	* Add a trailer if the frame is too small.
800	*
801	* Since we're getting to the packet before it is framed, it has not
802	* yet been padded. The current solution is to add a segment pointing
803	* to a buffer containing all zeros and pray that works for all frames...
804	*/
805	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
806	{
807	static uint8_t const s_abZero[128] = {0};
808
809	AssertReturnVoid(i < cSegs);
810
811	pSG->aSegs[i].Phys = NIL_RTHCPHYS;
812	pSG->aSegs[i].pv = (void *)&s_abZero[0];
813	pSG->aSegs[i].cb = 60 - pSG->cbTotal;
814	pSG->cbTotal = 60;
815	pSG->cSegsUsed++;
816	Assert(i + 1 <= pSG->cSegsAlloc)
817	}
818	#endif
819
820	Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
821	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
822	for (i = 0; i < pSG->cSegsUsed; i++)
823	Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
824	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
825	}
826
827	/**
828	* Packet handler,
829	*
830	* @returns 0 or EJUSTRETURN.
831	* @param pThis The instance.
832	* @param pMBuf The mbuf.
833	* @param pvFrame The start of the frame, optional.
834	* @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
835	* @param eProtocol The protocol.
836	*/
837	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
838	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
839	struct net_device *pSkbDev,
840	struct packet_type *pPacketType,
841	struct net_device *pOrigDev)
842	#else
843	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
844	struct net_device *pSkbDev,
845	struct packet_type *pPacketType)
846	#endif
847	{
848	PVBOXNETFLTINS pThis;
849	struct net_device *pDev;
850	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
851	pBuf, pSkbDev, pPacketType));
852	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
853	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",
854	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));
855	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
856	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
857	# endif
858	#else
859	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",
860	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));
861	#endif
862	/*
863	* Drop it immediately?
864	*/
865	if (!pBuf)
866	return 0;
867
868	if (pBuf->pkt_type == PACKET_LOOPBACK)
869	{
870	/*
871	* We are not interested in loopbacked packets as they will always have
872	* another copy going to the wire.
873	*/
874	Log2(("vboxNetFltLinuxPacketHandler: dropped loopback packet (cb=%u)\n", pBuf->len));
875	dev_kfree_skb(pBuf); /* We must 'consume' all packets we get (@bugref{6539})! */
876	return 0;
877	}
878
879	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
880	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
881	if (pDev != pSkbDev)
882	{
883	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
884	kfree_skb(pBuf); /* This is a failure, so we use kfree_skb instead of dev_kfree_skb. */
885	return 0;
886	}
887
888	Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
889	if (vboxNetFltLinuxSkBufIsOur(pBuf))
890	{
891	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
892	dev_kfree_skb(pBuf);
893	return 0;
894	}
895
896	#ifndef VBOXNETFLT_SG_SUPPORT
897	{
898	/*
899	* Get rid of fragmented packets, they cause too much trouble.
900	*/
901	unsigned int uMacLen = pBuf->mac_len;
902	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
903	dev_kfree_skb(pBuf);
904	if (!pCopy)
905	{
906	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
907	return 0;
908	}
909	pBuf = pCopy;
910	/* Somehow skb_copy ignores mac_len */
911	pBuf->mac_len = uMacLen;
912	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
913	/* Restore VLAN tag stripped by host hardware */
914	if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
915	{
916	uint8_t pMac = (uint8_t)skb_mac_header(pBuf);
917	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )(pMac - VLAN_HLEN);
918	# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
919	memmove(pVHdr, pMac, ETH_ALEN * 2);
920	# else
921	memmove(pVHdr, pMac, VLAN_ETH_ALEN * 2);
922	# endif
923	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
924	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
925	pBuf->mac_header -= VLAN_HLEN;
926	pBuf->mac_len += VLAN_HLEN;
927	}
928	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
929
930	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
931	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",
932	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));
933	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
934	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
935	# endif
936	# else
937	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",
938	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));
939	# endif
940	}
941	#endif
942
943	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
944	/* Forward it to the internal network. */
945	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
946	#else
947	/* Add the packet to transmit queue and schedule the bottom half. */
948	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
949	schedule_work(&pThis->u.s.XmitTask);
950	Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
951	&pThis->u.s.XmitTask, pBuf));
952	#endif
953
954	/* It does not really matter what we return, it is ignored by the kernel. */
955	return 0;
956	}
957
958	/**
959	* Calculate the number of INTNETSEG segments the socket buffer will need.
960	*
961	* @returns Segment count.
962	* @param pBuf The socket buffer.
963	*/
964	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
965	{
966	#ifdef VBOXNETFLT_SG_SUPPORT
967	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
968	#else
969	unsigned cSegs = 1;
970	#endif
971	#ifdef PADD_RUNT_FRAMES_FROM_HOST
972	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
973	if (pBuf->len < 60)
974	cSegs++;
975	#endif
976	return cSegs;
977	}
978
979	/**
980	* Destroy the intnet scatter / gather buffer created by
981	* vboxNetFltLinuxSkBufToSG.
982	*/
983	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
984	{
985	#ifdef VBOXNETFLT_SG_SUPPORT
986	int i;
987
988	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
989	{
990	printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
991	kunmap(pSG->aSegs[i+1].pv);
992	}
993	#endif
994	NOREF(pSG);
995	}
996
997	#ifdef LOG_ENABLED
998	/**
999	* Logging helper.
1000	*/
1001	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1002	{
1003	int i, offSeg;
1004	uint8_t pInt, pExt;
1005	static int iPacketNo = 1;
1006	iPacketNo += iIncrement;
1007	if (fEgress)
1008	{
1009	pExt = pSG->aSegs[0].pv;
1010	pInt = pExt + 6;
1011	}
1012	else
1013	{
1014	pInt = pSG->aSegs[0].pv;
1015	pExt = pInt + 6;
1016	}
1017	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1018	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1019	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1020	fEgress ? "-->" : "<--", pszWhere,
1021	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1022	pSG->cbTotal, iPacketNo));
1023	if (pSG->cSegsUsed == 1)
1024	{
1025	Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1026	}
1027	else
1028	{
1029	for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
1030	{
1031	Log3(("-- segment %d at 0x%x (%d bytes) --\n%.*Rhxd\n",
1032	i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1033	offSeg += pSG->aSegs[i].cb;
1034	}
1035	}
1036
1037	}
1038	#else
1039	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1040	#endif
1041
1042	#ifdef VBOXNETFLT_WITH_GSO_RECV
1043
1044	/**
1045	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1046	* GSO socket buffer without having to segment it.
1047	*
1048	* @returns true on success, false if needs segmenting.
1049	* @param pThis The net filter instance.
1050	* @param pSkb The GSO socket buffer.
1051	* @param fSrc The source.
1052	* @param pGsoCtx Where to return the GSO context on success.
1053	*/
1054	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1055	PPDMNETWORKGSO pGsoCtx)
1056	{
1057	PDMNETWORKGSOTYPE enmGsoType;
1058	uint16_t uEtherType;
1059	unsigned int cbTransport;
1060	unsigned int offTransport;
1061	unsigned int cbTransportHdr;
1062	unsigned uProtocol;
1063	union
1064	{
1065	RTNETIPV4 IPv4;
1066	RTNETIPV6 IPv6;
1067	RTNETTCP Tcp;
1068	uint8_t ab[40];
1069	uint16_t au16[40/2];
1070	uint32_t au32[40/4];
1071	} Buf;
1072
1073	/*
1074	* Check the GSO properties of the socket buffer and make sure it fits.
1075	*/
1076	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1077	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP \| SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
1078	{
1079	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1080	return false;
1081	}
1082	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1083	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
1084	{
1085	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1086	return false;
1087	}
1088	/*
1089	* It is possible to receive GSO packets from wire if GRO is enabled.
1090	*/
1091	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1092	{
1093	Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1094	#ifdef VBOXNETFLT_WITH_GRO
1095	/*
1096	* The packet came from the wire and the driver has already consumed
1097	* mac header. We need to restore it back.
1098	*/
1099	pSkb->mac_len = skb_network_header(pSkb) - skb_mac_header(pSkb);
1100	skb_push(pSkb, pSkb->mac_len);
1101	Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1102	pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
1103	#else /* !VBOXNETFLT_WITH_GRO */
1104	/* Older kernels didn't have GRO. */
1105	return false;
1106	#endif /* !VBOXNETFLT_WITH_GRO */
1107	}
1108	else
1109	{
1110	/*
1111	* skb_gso_segment does the following. Do we need to do it as well?
1112	*/
1113	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1114	skb_reset_mac_header(pSkb);
1115	pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1116	#else
1117	pSkb->mac.raw = pSkb->data;
1118	pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1119	#endif
1120	}
1121
1122	/*
1123	* Switch on the ethertype.
1124	*/
1125	uEtherType = pSkb->protocol;
1126	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1127	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1128	{
1129	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1130	if (puEtherType)
1131	uEtherType = *puEtherType;
1132	}
1133	switch (uEtherType)
1134	{
1135	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1136	{
1137	unsigned int cbHdr;
1138	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1139	if (RT_UNLIKELY(!pIPv4))
1140	{
1141	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1142	return false;
1143	}
1144
1145	cbHdr = pIPv4->ip_hl * 4;
1146	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1147	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1148	\|\| cbHdr > cbTransport ))
1149	{
1150	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1151	return false;
1152	}
1153	cbTransport -= cbHdr;
1154	offTransport = pSkb->mac_len + cbHdr;
1155	uProtocol = pIPv4->ip_p;
1156	if (uProtocol == RTNETIPV4_PROT_TCP)
1157	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1158	else if (uProtocol == RTNETIPV4_PROT_UDP)
1159	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1160	else /** @todo IPv6: 4to6 tunneling */
1161	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1162	break;
1163	}
1164
1165	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1166	{
1167	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1168	if (RT_UNLIKELY(!pIPv6))
1169	{
1170	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1171	return false;
1172	}
1173
1174	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1175	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1176	uProtocol = pIPv6->ip6_nxt;
1177	/** @todo IPv6: Dig our way out of the other headers. */
1178	if (uProtocol == RTNETIPV4_PROT_TCP)
1179	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1180	else if (uProtocol == RTNETIPV4_PROT_UDP)
1181	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1182	else
1183	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1184	break;
1185	}
1186
1187	default:
1188	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1189	return false;
1190	}
1191
1192	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1193	{
1194	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1195	return false;
1196	}
1197
1198	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1199	\|\| offTransport + cbTransport > pSkb->len
1200	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1201	{
1202	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1203	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1204	return false;
1205	}
1206
1207	/*
1208	* Check the TCP/UDP bits.
1209	*/
1210	if (uProtocol == RTNETIPV4_PROT_TCP)
1211	{
1212	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1213	if (RT_UNLIKELY(!pTcp))
1214	{
1215	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1216	return false;
1217	}
1218
1219	cbTransportHdr = pTcp->th_off * 4;
1220	pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1221	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1222	\|\| cbTransportHdr > cbTransport
1223	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1224	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1225	{
1226	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1227	return false;
1228	}
1229
1230	}
1231	else
1232	{
1233	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1234	cbTransportHdr = sizeof(RTNETUDP);
1235	pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1236	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1237	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1238	{
1239	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1240	return false;
1241	}
1242	}
1243
1244	/*
1245	* We're good, init the GSO context.
1246	*/
1247	pGsoCtx->u8Type = enmGsoType;
1248	pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1249	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1250	pGsoCtx->offHdr1 = pSkb->mac_len;
1251	pGsoCtx->offHdr2 = offTransport;
1252	pGsoCtx->u8Unused = 0;
1253
1254	return true;
1255	}
1256
1257	/**
1258	* Forward the socket buffer as a GSO internal network frame.
1259	*
1260	* @returns IPRT status code.
1261	* @param pThis The net filter instance.
1262	* @param pSkb The GSO socket buffer.
1263	* @param fSrc The source.
1264	* @param pGsoCtx Where to return the GSO context on success.
1265	*/
1266	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1267	{
1268	int rc;
1269	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1270	if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1271	{
1272	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1273	if (RT_LIKELY(pSG))
1274	{
1275	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1276
1277	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1278	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1279
1280	vboxNetFltLinuxDestroySG(pSG);
1281	rc = VINF_SUCCESS;
1282	}
1283	else
1284	{
1285	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1286	rc = VERR_NO_MEMORY;
1287	}
1288	}
1289	else
1290	{
1291	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1292	rc = VERR_INTERNAL_ERROR_3;
1293	}
1294
1295	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1296	dev_kfree_skb(pSkb);
1297	return rc;
1298	}
1299
1300	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1301
1302	/**
1303	* Worker for vboxNetFltLinuxForwardToIntNet.
1304	*
1305	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1306	* @param pThis The net filter instance.
1307	* @param pBuf The socket buffer.
1308	* @param fSrc The source.
1309	*/
1310	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1311	{
1312	int rc;
1313	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1314	if (cSegs <= MAX_SKB_FRAGS + 1)
1315	{
1316	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1317	if (RT_LIKELY(pSG))
1318	{
1319	if (fSrc & INTNETTRUNKDIR_WIRE)
1320	{
1321	/*
1322	* The packet came from wire, ethernet header was removed by device driver.
1323	* Restore it using mac_len field. This takes into account VLAN headers too.
1324	*/
1325	skb_push(pBuf, pBuf->mac_len);
1326	}
1327
1328	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /pGsoCtx/);
1329
1330	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1331	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1332
1333	vboxNetFltLinuxDestroySG(pSG);
1334	rc = VINF_SUCCESS;
1335	}
1336	else
1337	{
1338	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1339	rc = VERR_NO_MEMORY;
1340	}
1341	}
1342	else
1343	{
1344	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1345	rc = VERR_INTERNAL_ERROR_3;
1346	}
1347
1348	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1349	dev_kfree_skb(pBuf);
1350	return rc;
1351	}
1352
1353	/**
1354	*
1355	* @param pBuf The socket buffer. This is consumed by this function.
1356	*/
1357	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1358	{
1359	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1360
1361	#ifdef VBOXNETFLT_WITH_GSO
1362	if (skb_is_gso(pBuf))
1363	{
1364	PDMNETWORKGSO GsoCtx;
1365	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",
1366	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));
1367	# ifdef VBOXNETFLT_WITH_GSO_RECV
1368	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1369	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1370	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1371	else
1372	# endif
1373	{
1374	/* Need to segment the packet */
1375	struct sk_buff *pNext;
1376	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1377	if (IS_ERR(pSegment))
1378	{
1379	dev_kfree_skb(pBuf);
1380	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1381	return;
1382	}
1383
1384	for (; pSegment; pSegment = pNext)
1385	{
1386	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",
1387	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));
1388	pNext = pSegment->next;
1389	pSegment->next = 0;
1390	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1391	}
1392	dev_kfree_skb(pBuf);
1393	}
1394	}
1395	else
1396	#endif /* VBOXNETFLT_WITH_GSO */
1397	{
1398	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1399	{
1400	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1401	/*
1402	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1403	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1404	* header length from the header itself and reconstruct 'h' pointer
1405	* to TCP (or whatever) header.
1406	*/
1407	unsigned char *tmp = pBuf->h.raw;
1408	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1409	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1410	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1411	if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1412	{
1413	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1414	dev_kfree_skb(pBuf);
1415	return;
1416	}
1417	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1418	/* Restore the original (wrong) pointer. */
1419	pBuf->h.raw = tmp;
1420	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1421	}
1422	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1423	}
1424	}
1425
1426	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1427	/**
1428	* Work queue handler that forwards the socket buffers queued by
1429	* vboxNetFltLinuxPacketHandler to the internal network.
1430	*
1431	* @param pWork The work queue.
1432	*/
1433	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1434	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1435	# else
1436	static void vboxNetFltLinuxXmitTask(void *pWork)
1437	# endif
1438	{
1439	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1440	struct sk_buff *pBuf;
1441
1442	Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1443
1444	/*
1445	* Active? Retain the instance and increment the busy counter.
1446	*/
1447	if (vboxNetFltTryRetainBusyActive(pThis))
1448	{
1449	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1450	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1451
1452	vboxNetFltRelease(pThis, true /* fBusy */);
1453	}
1454	else
1455	{
1456	/** @todo Shouldn't we just drop the packets here? There is little point in
1457	* making them accumulate when the VM is paused and it'll only waste
1458	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1459	* before start draining the packets (goes for the intnet ring buf
1460	* too)? */
1461	}
1462	}
1463	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1464
1465	/**
1466	* Reports the GSO capabilities of the hardware NIC.
1467	*
1468	* @param pThis The net filter instance. The caller hold a
1469	* reference to this.
1470	*/
1471	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1472	{
1473	#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1474	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1475	{
1476	struct net_device *pDev;
1477	PINTNETTRUNKSWPORT pSwitchPort;
1478	unsigned int fFeatures;
1479
1480	RTSpinlockAcquire(pThis->hSpinlock);
1481
1482	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1483	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1484	if (pDev)
1485	fFeatures = pDev->features;
1486	else
1487	fFeatures = 0;
1488
1489	RTSpinlockRelease(pThis->hSpinlock);
1490
1491	if (pThis->pSwitchPort)
1492	{
1493	/* Set/update the GSO capabilities of the NIC. */
1494	uint32_t fGsoCapabilites = 0;
1495	if (fFeatures & NETIF_F_TSO)
1496	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1497	if (fFeatures & NETIF_F_TSO6)
1498	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1499	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1500	if (fFeatures & NETIF_F_UFO)
1501	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1502	if (fFeatures & NETIF_F_UFO)
1503	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1504	# endif
1505	Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1506	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1507	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : "",
1508	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)) ? "ufo " : "",
1509	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)) ? "ufo6 " : ""));
1510	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1511	}
1512
1513	vboxNetFltRelease(pThis, true /fBusy/);
1514	}
1515	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1516	}
1517
1518	/**
1519	* Helper that determines whether the host (ignoreing us) is operating the
1520	* interface in promiscuous mode or not.
1521	*/
1522	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1523	{
1524	bool fRc = false;
1525	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1526	if (pDev)
1527	{
1528	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1529	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1530	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1531	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1532	}
1533	return fRc;
1534	}
1535
1536	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36)
1537	/**
1538	* Helper for detecting TAP devices.
1539	*/
1540	static bool vboxNetFltIsTapDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1541	{
1542	if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1543	{
1544	struct ethtool_drvinfo Info;
1545
1546	memset(&Info, 0, sizeof(Info));
1547	Info.cmd = ETHTOOL_GDRVINFO;
1548	pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1549	Log3(("vboxNetFltIsTapDevice: driver=%s version=%s bus_info=%s\n",
1550	Info.driver, Info.version, Info.bus_info));
1551
1552	return !strncmp(Info.driver, "tun", 4)
1553	&& !strncmp(Info.bus_info, "tap", 4);
1554	}
1555
1556	return false;
1557	}
1558
1559	/**
1560	* Helper for updating the link state of TAP devices.
1561	* Only TAP devices are affected.
1562	*/
1563	static void vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1564	{
1565	if (vboxNetFltIsTapDevice(pThis, pDev))
1566	{
1567	Log3(("vboxNetFltSetTapLinkState: bringing %s tap device link state\n",
1568	fLinkUp ? "up" : "down"));
1569	netif_tx_lock_bh(pDev);
1570	if (fLinkUp)
1571	netif_carrier_on(pDev);
1572	else
1573	netif_carrier_off(pDev);
1574	netif_tx_unlock_bh(pDev);
1575	}
1576	}
1577	#else /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1578	DECLINLINE(void) vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1579	{
1580	/* Nothing to do for pre-2.6.36 kernels. */
1581	}
1582	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1583
1584	/**
1585	* Internal worker for vboxNetFltLinuxNotifierCallback.
1586	*
1587	* @returns VBox status code.
1588	* @param pThis The instance.
1589	* @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1590	* flood the release log.
1591	*/
1592	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1593	{
1594	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1595
1596	/*
1597	* Retain and store the device.
1598	*/
1599	dev_hold(pDev);
1600
1601	RTSpinlockAcquire(pThis->hSpinlock);
1602	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1603	RTSpinlockRelease(pThis->hSpinlock);
1604
1605	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1606	pDev, pDev->name,
1607	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1608	netdev_refcnt_read(pDev)
1609	#else
1610	atomic_read(&pDev->refcnt)
1611	#endif
1612	));
1613	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1614	pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1615
1616	/* Get the mac address while we still have a valid net_device reference. */
1617	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1618	/* Initialize MTU */
1619	pThis->u.s.cbMtu = pDev->mtu;
1620
1621	/*
1622	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1623	*/
1624	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1625	pThis->u.s.PacketType.dev = pDev;
1626	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1627	dev_add_pack(&pThis->u.s.PacketType);
1628	ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1629	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1630
1631	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1632	vboxNetFltLinuxHookDev(pThis, pDev);
1633	#endif
1634
1635	/*
1636	* If attaching to TAP interface we need to bring the link state up
1637	* starting from 2.6.36 kernel.
1638	*/
1639	vboxNetFltSetTapLinkState(pThis, pDev, true);
1640
1641	/*
1642	* Set indicators that require the spinlock. Be abit paranoid about racing
1643	* the device notification handle.
1644	*/
1645	RTSpinlockAcquire(pThis->hSpinlock);
1646	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1647	if (pDev)
1648	{
1649	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1650	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1651	pDev = NULL; /* don't dereference it */
1652	}
1653	RTSpinlockRelease(pThis->hSpinlock);
1654
1655	/*
1656	* If the above succeeded report GSO capabilities, if not undo and
1657	* release the device.
1658	*/
1659	if (!pDev)
1660	{
1661	Assert(pThis->pSwitchPort);
1662	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1663	{
1664	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1665	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1666	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1667	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1668	vboxNetFltRelease(pThis, true /fBusy/);
1669	}
1670	}
1671	else
1672	{
1673	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1674	vboxNetFltLinuxUnhookDev(pThis, pDev);
1675	#endif
1676	RTSpinlockAcquire(pThis->hSpinlock);
1677	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1678	RTSpinlockRelease(pThis->hSpinlock);
1679	dev_put(pDev);
1680	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n",
1681	pDev, pDev->name,
1682	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1683	netdev_refcnt_read(pDev)
1684	#else
1685	atomic_read(&pDev->refcnt)
1686	#endif
1687	));
1688	}
1689
1690	LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1691	return VINF_SUCCESS;
1692	}
1693
1694
1695	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1696	{
1697	bool fRegistered;
1698	Assert(!pThis->fDisconnectedFromHost);
1699
1700	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1701	vboxNetFltLinuxUnhookDev(pThis, pDev);
1702	#endif
1703
1704	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1705	{
1706	dev_remove_pack(&pThis->u.s.PacketType);
1707	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1708	}
1709
1710	RTSpinlockAcquire(pThis->hSpinlock);
1711	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1712	if (fRegistered)
1713	{
1714	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1715	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1716	}
1717	RTSpinlockRelease(pThis->hSpinlock);
1718
1719	if (fRegistered)
1720	{
1721	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1722	skb_queue_purge(&pThis->u.s.XmitQueue);
1723	#endif
1724	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1725	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1726	pDev, pDev->name,
1727	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1728	netdev_refcnt_read(pDev)
1729	#else
1730	atomic_read(&pDev->refcnt)
1731	#endif
1732	));
1733	dev_put(pDev);
1734	}
1735
1736	return NOTIFY_OK;
1737	}
1738
1739	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1740	{
1741	/* Check if we are not suspended and promiscuous mode has not been set. */
1742	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1743	&& !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1744	{
1745	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1746	dev_set_promiscuity(pDev, 1);
1747	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1748	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1749	}
1750	else
1751	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1752	return NOTIFY_OK;
1753	}
1754
1755	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1756	{
1757	/* Undo promiscuous mode if we has set it. */
1758	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1759	{
1760	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1761	dev_set_promiscuity(pDev, -1);
1762	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1763	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1764	}
1765	else
1766	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1767	return NOTIFY_OK;
1768	}
1769
1770	/**
1771	* Callback for listening to MTU change event.
1772	*
1773	* We need to track changes of host's inteface MTU to discard over-sized frames
1774	* coming from the internal network as they may hang the TX queue of host's
1775	* adapter.
1776	*
1777	* @returns NOTIFY_OK
1778	* @param pThis The netfilter instance.
1779	* @param pDev Pointer to device structure of host's interface.
1780	*/
1781	static int vboxNetFltLinuxDeviceMtuChange(PVBOXNETFLTINS pThis, struct net_device *pDev)
1782	{
1783	ASMAtomicWriteU32(&pThis->u.s.cbMtu, pDev->mtu);
1784	Log(("vboxNetFltLinuxDeviceMtuChange: set MTU for %s to %d\n", pThis->szName, pDev->mtu));
1785	return NOTIFY_OK;
1786	}
1787
1788	#ifdef LOG_ENABLED
1789	/** Stringify the NETDEV_XXX constants. */
1790	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
1791	{
1792	const char *pszEvent = "NETDRV_<unknown>";
1793	switch (ulEventType)
1794	{
1795	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
1796	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
1797	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
1798	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
1799	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
1800	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
1801	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
1802	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
1803	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
1804	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
1805	# ifdef NETDEV_FEAT_CHANGE
1806	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
1807	# endif
1808	}
1809	return pszEvent;
1810	}
1811	#endif /* LOG_ENABLED */
1812
1813	/**
1814	* Callback for listening to netdevice events.
1815	*
1816	* This works the rediscovery, clean up on unregistration, promiscuity on
1817	* up/down, and GSO feature changes from ethtool.
1818	*
1819	* @returns NOTIFY_OK
1820	* @param self Pointer to our notifier registration block.
1821	* @param ulEventType The event.
1822	* @param ptr Event specific, but it is usually the device it
1823	* relates to.
1824	*/
1825	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
1826
1827	{
1828	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
1829	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 11, 0)
1830	struct net_device *pDev = netdev_notifier_info_to_dev(ptr);
1831	#else
1832	struct net_device pDev = (struct net_device )ptr;
1833	#endif
1834	int rc = NOTIFY_OK;
1835
1836	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1837	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1838	if ( ulEventType == NETDEV_REGISTER
1839	&& !strcmp(pDev->name, pThis->szName))
1840	{
1841	vboxNetFltLinuxAttachToInterface(pThis, pDev);
1842	}
1843	else
1844	{
1845	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1846	if (pDev == ptr)
1847	{
1848	switch (ulEventType)
1849	{
1850	case NETDEV_UNREGISTER:
1851	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
1852	break;
1853	case NETDEV_UP:
1854	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
1855	break;
1856	case NETDEV_GOING_DOWN:
1857	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
1858	break;
1859	case NETDEV_CHANGEMTU:
1860	rc = vboxNetFltLinuxDeviceMtuChange(pThis, pDev);
1861	break;
1862	case NETDEV_CHANGENAME:
1863	break;
1864	#ifdef NETDEV_FEAT_CHANGE
1865	case NETDEV_FEAT_CHANGE:
1866	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1867	break;
1868	#endif
1869	}
1870	}
1871	}
1872
1873	return rc;
1874	}
1875
1876	#if 0 /* XXX: temporarily disable */
1877	static int vboxNetFltLinuxNotifierIPv4Callback(struct notifier_block self, unsigned long ulEventType, void ptr)
1878	{
1879	PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv4);
1880	struct net_device *pDev;
1881	struct in_ifaddr ifa = (struct in_ifaddr )ptr;
1882	int rc = NOTIFY_OK;
1883
1884	pDev = vboxNetFltLinuxRetainNetDev(pThis);
1885	Log(("VBoxNetFlt: %s: IPv4 event %s(0x%lx): addr %RTnaipv4 mask %RTnaipv4\n",
1886	pDev ? netdev_name(pDev) : "<???>",
1887	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
1888	ifa->ifa_address, ifa->ifa_mask));
1889
1890	if (pDev != NULL)
1891	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1892
1893	if (pThis->pSwitchPort->pfnNotifyHostAddress)
1894	{
1895	bool fAdded;
1896	if (ulEventType == NETDEV_UP)
1897	fAdded = true;
1898	else if (ulEventType == NETDEV_DOWN)
1899	fAdded = false;
1900	else
1901	return NOTIFY_OK;
1902
1903	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
1904	kIntNetAddrType_IPv4, &ifa->ifa_local);
1905	}
1906
1907	return rc;
1908	}
1909
1910
1911	static int vboxNetFltLinuxNotifierIPv6Callback(struct notifier_block self, unsigned long ulEventType, void ptr)
1912	{
1913	PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv6);
1914	struct net_device *pDev;
1915	struct inet6_ifaddr ifa = (struct inet6_ifaddr )ptr;
1916	int rc = NOTIFY_OK;
1917
1918	pDev = vboxNetFltLinuxRetainNetDev(pThis);
1919	Log(("VBoxNetFlt: %s: IPv6 event %s(0x%lx): %RTnaipv6\n",
1920	pDev ? netdev_name(pDev) : "<???>",
1921	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
1922	&ifa->addr));
1923
1924	if (pDev != NULL)
1925	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1926
1927	if (pThis->pSwitchPort->pfnNotifyHostAddress)
1928	{
1929	bool fAdded;
1930	if (ulEventType == NETDEV_UP)
1931	fAdded = true;
1932	else if (ulEventType == NETDEV_DOWN)
1933	fAdded = false;
1934	else
1935	return NOTIFY_OK;
1936
1937	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
1938	kIntNetAddrType_IPv6, &ifa->addr);
1939	}
1940
1941	return rc;
1942	}
1943	#endif /* 0 */
1944
1945
1946	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
1947	{
1948	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
1949	}
1950
1951	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
1952	{
1953	struct net_device * pDev;
1954	int err;
1955	int rc = VINF_SUCCESS;
1956	NOREF(pvIfData);
1957
1958	LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
1959
1960	pDev = vboxNetFltLinuxRetainNetDev(pThis);
1961	if (pDev)
1962	{
1963	/*
1964	* Create a sk_buff for the gather list and push it onto the wire.
1965	*/
1966	if (fDst & INTNETTRUNKDIR_WIRE)
1967	{
1968	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
1969	if (pBuf)
1970	{
1971	vboxNetFltDumpPacket(pSG, true, "wire", 1);
1972	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1973	Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
1974	err = dev_queue_xmit(pBuf);
1975	if (err)
1976	rc = RTErrConvertFromErrno(err);
1977	}
1978	else
1979	rc = VERR_NO_MEMORY;
1980	}
1981
1982	/*
1983	* Create a sk_buff for the gather list and push it onto the host stack.
1984	*/
1985	if (fDst & INTNETTRUNKDIR_HOST)
1986	{
1987	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
1988	if (pBuf)
1989	{
1990	vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
1991	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1992	Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
1993	err = netif_rx_ni(pBuf);
1994	if (err)
1995	rc = RTErrConvertFromErrno(err);
1996	}
1997	else
1998	rc = VERR_NO_MEMORY;
1999	}
2000
2001	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2002	}
2003
2004	return rc;
2005	}
2006
2007
2008	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2009	{
2010	struct net_device * pDev;
2011
2012	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
2013	pThis, pThis->szName, fActive?"true":"false",
2014	pThis->fDisablePromiscuous?"true":"false"));
2015
2016	if (pThis->fDisablePromiscuous)
2017	return;
2018
2019	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2020	if (pDev)
2021	{
2022	/*
2023	* This api is a bit weird, the best reference is the code.
2024	*
2025	* Also, we have a bit or race conditions wrt the maintenance of
2026	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2027	*/
2028	#ifdef LOG_ENABLED
2029	u_int16_t fIf;
2030	unsigned const cPromiscBefore = pDev->promiscuity;
2031	#endif
2032	if (fActive)
2033	{
2034	Assert(!pThis->u.s.fPromiscuousSet);
2035
2036	rtnl_lock();
2037	dev_set_promiscuity(pDev, 1);
2038	rtnl_unlock();
2039	pThis->u.s.fPromiscuousSet = true;
2040	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2041	}
2042	else
2043	{
2044	if (pThis->u.s.fPromiscuousSet)
2045	{
2046	rtnl_lock();
2047	dev_set_promiscuity(pDev, -1);
2048	rtnl_unlock();
2049	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2050	}
2051	pThis->u.s.fPromiscuousSet = false;
2052
2053	#ifdef LOG_ENABLED
2054	fIf = dev_get_flags(pDev);
2055	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2056	#endif
2057	}
2058
2059	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2060	}
2061	}
2062
2063
2064	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2065	{
2066	/*
2067	* Remove packet handler when we get disconnected from internal switch as
2068	* we don't want the handler to forward packets to disconnected switch.
2069	*/
2070	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
2071	{
2072	dev_remove_pack(&pThis->u.s.PacketType);
2073	Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
2074	}
2075	return VINF_SUCCESS;
2076	}
2077
2078
2079	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2080	{
2081	/*
2082	* Report the GSO capabilities of the host and device (if connected).
2083	* Note! No need to mark ourselves busy here.
2084	*/
2085	/** @todo duplicate work here now? Attach */
2086	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2087	Log3(("vboxNetFltOsConnectIt: reporting host tso tso6 ufo\n"));
2088	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2089	0
2090	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2091	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2092	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2093	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2094	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2095	# endif
2096	, INTNETTRUNKDIR_HOST);
2097
2098	#endif
2099	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2100
2101	return VINF_SUCCESS;
2102	}
2103
2104
2105	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2106	{
2107	struct net_device *pDev;
2108	bool fRegistered;
2109
2110	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
2111	vboxNetFltLinuxUnhookDev(pThis, NULL);
2112	#endif
2113
2114	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2115	* unlikely, but none the less). Since it doesn't actually update the
2116	* state (just reads it), it is likely to panic in some interesting
2117	* ways. */
2118
2119	RTSpinlockAcquire(pThis->hSpinlock);
2120	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2121	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
2122	RTSpinlockRelease(pThis->hSpinlock);
2123
2124	if (fRegistered)
2125	{
2126	vboxNetFltSetTapLinkState(pThis, pDev, false);
2127
2128	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2129	skb_queue_purge(&pThis->u.s.XmitQueue);
2130	#endif
2131	Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
2132	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
2133	pDev, pDev->name,
2134	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
2135	netdev_refcnt_read(pDev)
2136	#else
2137	atomic_read(&pDev->refcnt)
2138	#endif
2139	));
2140	dev_put(pDev);
2141	}
2142
2143	unregister_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2144	unregister_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2145
2146	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2147	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2148	module_put(THIS_MODULE);
2149	}
2150
2151
2152	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2153	{
2154	int err;
2155	NOREF(pvContext);
2156
2157	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2158	err = register_netdevice_notifier(&pThis->u.s.Notifier);
2159	if (err)
2160	return VERR_INTNET_FLT_IF_FAILED;
2161	if (!pThis->u.s.fRegistered)
2162	{
2163	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2164	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2165	return VERR_INTNET_FLT_IF_NOT_FOUND;
2166	}
2167
2168	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2169	if ( pThis->fDisconnectedFromHost
2170	\|\| !try_module_get(THIS_MODULE))
2171	return VERR_INTNET_FLT_IF_FAILED;
2172
2173	#if 0 /* XXX: temporarily disable */
2174	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2175	{
2176	struct net *net = VBOX_DEV_NET(pThis->u.s.pDev);
2177	struct net_device *dev;
2178
2179	rcu_read_lock();
2180	for_each_netdev_rcu(net, dev)
2181	{
2182	struct in_device *in_dev;
2183	struct inet6_dev *in6_dev;
2184
2185	/*
2186	* IPv4
2187	*/
2188	in_dev = __in_dev_get_rcu(dev);
2189	if (in_dev != NULL)
2190	{
2191	for_ifa(in_dev) {
2192	if (ifa->ifa_address == htonl(INADDR_LOOPBACK))
2193	goto continue_netdev;
2194
2195	Log(("%s: %s: IPv4: addr %RTnaipv4 mask %RTnaipv4\n",
2196	__FUNCTION__, netdev_name(dev),
2197	ifa->ifa_address, ifa->ifa_mask));
2198
2199	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2200	/* :fAdded */ true, kIntNetAddrType_IPv4, &ifa->ifa_address);
2201	} endfor_ifa(in_dev);
2202	}
2203
2204	/*
2205	* IPv6
2206	*/
2207	in6_dev = __in6_dev_get(dev);
2208	if (in6_dev != NULL)
2209	{
2210	struct inet6_ifaddr *ifa;
2211
2212	read_lock_bh(&in6_dev->lock);
2213	list_for_each_entry(ifa, &in6_dev->addr_list, if_list)
2214	{
2215	Log(("%s: %s: IPv6: addr %RTnaipv6/%u\n",
2216	__FUNCTION__, netdev_name(dev),
2217	&ifa->addr, (unsigned)ifa->prefix_len));
2218
2219	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2220	/* :fAdded */ true, kIntNetAddrType_IPv6, &ifa->addr);
2221	}
2222	read_unlock_bh(&in6_dev->lock);
2223	}
2224
2225	continue_netdev:
2226	/* continue */;
2227	}
2228	rcu_read_unlock();
2229
2230	Log(("%s: pfnNotifyHostAddress is set, register notifiers\n", __FUNCTION__));
2231
2232	pThis->u.s.NotifierIPv4.notifier_call = vboxNetFltLinuxNotifierIPv4Callback;
2233	err = register_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2234	if (err)
2235	LogRel(("%s: failed to register IPv4 notifier: error %d\n",
2236	__FUNCTION__, err));
2237
2238	pThis->u.s.NotifierIPv6.notifier_call = vboxNetFltLinuxNotifierIPv6Callback;
2239	err = register_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2240	if (err)
2241	LogRel(("%s: failed to register IPv6 notifier: error %d\n",
2242	__FUNCTION__, err));
2243	}
2244	else
2245	Log(("%s: uwe: pfnNotifyHostAddress is NULL\n", __FUNCTION__));
2246	#endif
2247
2248	return VINF_SUCCESS;
2249	}
2250
2251	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2252	{
2253	/*
2254	* Init the linux specific members.
2255	*/
2256	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2257	pThis->u.s.fRegistered = false;
2258	pThis->u.s.fPromiscuousSet = false;
2259	pThis->u.s.fPacketHandler = false;
2260	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2261	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2262	skb_queue_head_init(&pThis->u.s.XmitQueue);
2263	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2264	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2265	# else
2266	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2267	# endif
2268	#endif
2269
2270	return VINF_SUCCESS;
2271	}
2272
2273
2274	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2275	{
2276	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2277	}
2278
2279
2280	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
2281	{
2282	/* Nothing to do */
2283	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2284	return VINF_SUCCESS;
2285	}
2286
2287
2288	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2289	{
2290	/* Nothing to do */
2291	NOREF(pThis); NOREF(pvIfData);
2292	return VINF_SUCCESS;
2293	}
2294

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 55679

Download in other formats: