VBoxNetFlt-linux.c@ 56143

Last change on this file since 56143 was 56143, checked in by vboxsync, 10 years ago
VBoxNetFlt/linux: Instead of fighting an uphill battle with ifdefs to loop over all net devices, abuse/co-opt register_netdevice_notifier() to do that for us.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 80.0 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 56143 2015-05-28 19:47:00Z 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	typedef struct VBOXNETFLTNOTIFIER {
58	struct notifier_block Notifier;
59	PVBOXNETFLTINS pThis;
60	} VBOXNETFLTNOTIFIER;
61	typedef struct VBOXNETFLTNOTIFIER *PVBOXNETFLTNOTIFIER;
62
63
64	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 20, 0)
65	# define vlan_tx_tag_get(skb) skb_vlan_tag_get(skb)
66	# define vlan_tx_tag_present(skb) skb_vlan_tag_present(skb)
67	#endif
68
69
70	/*******************************************************************************
71	* Defined Constants And Macros *
72	*******************************************************************************/
73	#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
74	#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
75	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
76	# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
77	#endif
78
79	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 11, 0)
80	# define VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr) netdev_notifier_info_to_dev(ptr)
81	#else
82	# define VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr) ((struct net_device *)ptr)
83	#endif
84
85	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 34)
86	# define VBOX_NETDEV_NAME(dev) netdev_name(dev)
87	#else
88	# define VBOX_NETDEV_NAME(dev) ((dev)->reg_state != NETREG_REGISTERED ? "(unregistered net_device)" : (dev)->name)
89	#endif
90
91	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
92	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
93	# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
94	#else
95	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
96	# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
97	#endif
98
99	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
100	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
101	#else
102	# define CHECKSUM_PARTIAL CHECKSUM_HW
103	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
104	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
105	# else
106	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
107	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
108	# else
109	# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
110	# endif
111	/* Versions prior 2.6.10 use stats for both bstats and qstats */
112	# define bstats stats
113	# define qstats stats
114	# endif
115	#endif
116
117	#ifndef NET_IP_ALIGN
118	# define NET_IP_ALIGN 2
119	#endif
120
121	#if 0
122	/** Create scatter / gather segments for fragments. When not used, we will
123	* linearize the socket buffer before creating the internal networking SG. */
124	# define VBOXNETFLT_SG_SUPPORT 1
125	#endif
126
127	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
128	/** Indicates that the linux kernel may send us GSO frames. */
129	# define VBOXNETFLT_WITH_GSO 1
130
131	/** This enables or disables the transmitting of GSO frame from the internal
132	* network and to the host. */
133	# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
134
135	# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
136	/** This enables or disables the transmitting of GSO frame from the internal
137	* network and to the wire. */
138	# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
139	# endif
140
141	/** This enables or disables the forwarding/flooding of GSO frame from the host
142	* to the internal network. */
143	# define VBOXNETFLT_WITH_GSO_RECV 1
144
145	#endif
146
147	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
148	/** This enables or disables handling of GSO frames coming from the wire (GRO). */
149	# define VBOXNETFLT_WITH_GRO 1
150	#endif
151	/*
152	* GRO support was backported to RHEL 5.4
153	*/
154	#ifdef RHEL_RELEASE_CODE
155	# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
156	# define VBOXNETFLT_WITH_GRO 1
157	# endif
158	#endif
159
160	/*******************************************************************************
161	* Internal Functions *
162	*******************************************************************************/
163	static int VBoxNetFltLinuxInit(void);
164	static void VBoxNetFltLinuxUnload(void);
165	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
166
167
168	/*******************************************************************************
169	* Global Variables *
170	*******************************************************************************/
171	/**
172	* The (common) global data.
173	*/
174	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
175
176	module_init(VBoxNetFltLinuxInit);
177	module_exit(VBoxNetFltLinuxUnload);
178
179	MODULE_AUTHOR(VBOX_VENDOR);
180	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
181	MODULE_LICENSE("GPL");
182	#ifdef MODULE_VERSION
183	MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
184	#endif
185
186
187	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
188	unsigned dev_get_flags(const struct net_device *dev)
189	{
190	unsigned flags;
191
192	flags = (dev->flags & ~(IFF_PROMISC \|
193	IFF_ALLMULTI \|
194	IFF_RUNNING)) \|
195	(dev->gflags & (IFF_PROMISC \|
196	IFF_ALLMULTI));
197
198	if (netif_running(dev) && netif_carrier_ok(dev))
199	flags \|= IFF_RUNNING;
200
201	return flags;
202	}
203	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
204
205
206	/**
207	* Initialize module.
208	*
209	* @returns appropriate status code.
210	*/
211	static int __init VBoxNetFltLinuxInit(void)
212	{
213	int rc;
214	/*
215	* Initialize IPRT.
216	*/
217	rc = RTR0Init(0);
218	if (RT_SUCCESS(rc))
219	{
220	Log(("VBoxNetFltLinuxInit\n"));
221
222	/*
223	* Initialize the globals and connect to the support driver.
224	*
225	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
226	* for establishing the connect to the support driver.
227	*/
228	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
229	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
230	if (RT_SUCCESS(rc))
231	{
232	LogRel(("VBoxNetFlt: Successfully started.\n"));
233	return 0;
234	}
235
236	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
237	RTR0Term();
238	}
239	else
240	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
241
242	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
243	return -RTErrConvertToErrno(rc);
244	}
245
246
247	/**
248	* Unload the module.
249	*
250	* @todo We have to prevent this if we're busy!
251	*/
252	static void __exit VBoxNetFltLinuxUnload(void)
253	{
254	int rc;
255	Log(("VBoxNetFltLinuxUnload\n"));
256	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
257
258	/*
259	* Undo the work done during start (in reverse order).
260	*/
261	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
262	AssertRC(rc); NOREF(rc);
263
264	RTR0Term();
265
266	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
267
268	Log(("VBoxNetFltLinuxUnload - done\n"));
269	}
270
271
272	/**
273	* We filter traffic from the host to the internal network
274	* before it reaches the NIC driver.
275	*
276	* The current code uses a very ugly hack overriding hard_start_xmit
277	* callback in the device structure, but it has been shown to give us a
278	* performance boost of 60-100% though. Eventually we have to find some
279	* less hacky way of getting this job done.
280	*/
281	#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
282
283	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
284
285	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
286
287	# include <linux/ethtool.h>
288
289	typedef struct ethtool_ops OVR_OPSTYPE;
290	# define OVR_OPS ethtool_ops
291	# define OVR_XMIT pfnStartXmit
292
293	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
294
295	typedef struct net_device_ops OVR_OPSTYPE;
296	# define OVR_OPS netdev_ops
297	# define OVR_XMIT pOrgOps->ndo_start_xmit
298
299	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
300
301	/**
302	* The overridden net_device_ops of the device we're attached to.
303	*
304	* As there is no net_device_ops structure in pre-2.6.29 kernels we override
305	* ethtool_ops instead along with hard_start_xmit callback in net_device
306	* structure.
307	*
308	* This is a very dirty hack that was created to explore how much we can improve
309	* the host to guest transfers by not CC'ing the NIC. It turns out to be
310	* the only way to filter outgoing packets for devices without TX queue.
311	*/
312	typedef struct VBoxNetDeviceOpsOverride
313	{
314	/** Our overridden ops. */
315	OVR_OPSTYPE Ops;
316	/** Magic word. */
317	uint32_t u32Magic;
318	/** Pointer to the original ops. */
319	OVR_OPSTYPE const *pOrgOps;
320	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
321	/** Pointer to the original hard_start_xmit function. */
322	int (pfnStartXmit)(struct sk_buff pSkb, struct net_device *pDev);
323	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
324	/** Pointer to the net filter instance. */
325	PVBOXNETFLTINS pVBoxNetFlt;
326	/** The number of filtered packages. */
327	uint64_t cFiltered;
328	/** The total number of packets */
329	uint64_t cTotal;
330	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
331	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
332	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
333
334	/**
335	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
336	* because they belong on the internal network.
337	*
338	* @returns NETDEV_TX_XXX.
339	* @param pSkb The socket buffer to transmit.
340	* @param pDev The net device.
341	*/
342	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
343	{
344	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
345	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
346	PCRTNETETHERHDR pEtherHdr;
347	PINTNETTRUNKSWPORT pSwitchPort;
348	uint32_t cbHdrs;
349
350
351	/*
352	* Validate the override structure.
353	*
354	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
355	* to be production quality code, we would have to be much more
356	* careful here and avoid the race.
357	*/
358	if ( !VALID_PTR(pOverride)
359	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
360	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
361	\|\| !VALID_PTR(pOverride->pOrgOps)
362	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
363	)
364	{
365	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
366	dev_kfree_skb(pSkb);
367	return NETDEV_TX_OK;
368	}
369	pOverride->cTotal++;
370
371	/*
372	* Do the filtering base on the default OUI of our virtual NICs
373	*
374	* Note! In a real solution, we would ask the switch whether the
375	* destination MAC is 100% to be on the internal network and then
376	* drop it.
377	*/
378	cbHdrs = skb_headlen(pSkb);
379	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
380	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
381	if ( pEtherHdr
382	&& VALID_PTR(pOverride->pVBoxNetFlt)
383	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
384	&& VALID_PTR(pSwitchPort)
385	&& cbHdrs >= 6)
386	{
387	INTNETSWDECISION enmDecision;
388
389	/** @todo consider reference counting, etc. */
390	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
391	if (enmDecision == INTNETSWDECISION_INTNET)
392	{
393	dev_kfree_skb(pSkb);
394	pOverride->cFiltered++;
395	return NETDEV_TX_OK;
396	}
397	}
398
399	return pOverride->OVR_XMIT(pSkb, pDev);
400	}
401
402	/**
403	* Hooks the device ndo_start_xmit operation of the device.
404	*
405	* @param pThis The net filter instance.
406	* @param pDev The net device.
407	*/
408	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
409	{
410	PVBOXNETDEVICEOPSOVERRIDE pOverride;
411
412	/* Cancel override if ethtool_ops is missing (host-only case, @bugref{5712}) */
413	if (!VALID_PTR(pDev->OVR_OPS))
414	return;
415	pOverride = RTMemAlloc(sizeof(*pOverride));
416	if (!pOverride)
417	return;
418	pOverride->pOrgOps = pDev->OVR_OPS;
419	pOverride->Ops = *pDev->OVR_OPS;
420	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
421	pOverride->pfnStartXmit = pDev->hard_start_xmit;
422	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
423	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
424	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
425	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
426	pOverride->cTotal = 0;
427	pOverride->cFiltered = 0;
428	pOverride->pVBoxNetFlt = pThis;
429
430	RTSpinlockAcquire(pThis->hSpinlock); /* (this isn't necessary, but so what) */
431	ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
432	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
433	ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
434	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
435	RTSpinlockRelease(pThis->hSpinlock);
436	}
437
438	/**
439	* Undos what vboxNetFltLinuxHookDev did.
440	*
441	* @param pThis The net filter instance.
442	* @param pDev The net device. Can be NULL, in which case
443	* we'll try retrieve it from @a pThis.
444	*/
445	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
446	{
447	PVBOXNETDEVICEOPSOVERRIDE pOverride;
448
449	RTSpinlockAcquire(pThis->hSpinlock);
450	if (!pDev)
451	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
452	if (VALID_PTR(pDev))
453	{
454	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
455	if ( VALID_PTR(pOverride)
456	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
457	&& VALID_PTR(pOverride->pOrgOps)
458	)
459	{
460	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
461	ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
462	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
463	ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
464	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
465	}
466	else
467	pOverride = NULL;
468	}
469	else
470	pOverride = NULL;
471	RTSpinlockRelease(pThis->hSpinlock);
472
473	if (pOverride)
474	{
475	printk("vboxnetflt: %llu out of %llu packets were not sent (directed to host)\n", pOverride->cFiltered, pOverride->cTotal);
476	RTMemFree(pOverride);
477	}
478	}
479
480	#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
481
482
483	/**
484	* Reads and retains the host interface handle.
485	*
486	* @returns The handle, NULL if detached.
487	* @param pThis
488	*/
489	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
490	{
491	#if 0
492	struct net_device *pDev = NULL;
493
494	Log(("vboxNetFltLinuxRetainNetDev\n"));
495	/*
496	* Be careful here to avoid problems racing the detached callback.
497	*/
498	RTSpinlockAcquire(pThis->hSpinlock);
499	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
500	{
501	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
502	if (pDev)
503	{
504	dev_hold(pDev);
505	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
506	pDev, pDev->name,
507	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
508	netdev_refcnt_read(pDev)
509	#else
510	atomic_read(&pDev->refcnt)
511	#endif
512	));
513	}
514	}
515	RTSpinlockRelease(pThis->hSpinlock);
516
517	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
518	return pDev;
519	#else
520	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
521	#endif
522	}
523
524
525	/**
526	* Release the host interface handle previously retained
527	* by vboxNetFltLinuxRetainNetDev.
528	*
529	* @param pThis The instance.
530	* @param pDev The vboxNetFltLinuxRetainNetDev
531	* return value, NULL is fine.
532	*/
533	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
534	{
535	#if 0
536	Log(("vboxNetFltLinuxReleaseNetDev\n"));
537	NOREF(pThis);
538	if (pDev)
539	{
540	dev_put(pDev);
541	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
542	pDev, pDev->name,
543	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
544	netdev_refcnt_read(pDev)
545	#else
546	atomic_read(&pDev->refcnt)
547	#endif
548	));
549	}
550	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
551	#endif
552	}
553
554	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
555	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
556
557	/**
558	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
559	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
560	*
561	* @returns true / false accordingly.
562	* @param pBuf The sk_buff.
563	*/
564	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
565	{
566	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
567	}
568
569
570	/**
571	* Checks whether this SG list contains a GSO packet.
572	*
573	* @returns true / false accordingly.
574	* @param pSG The (scatter/)gather list.
575	*/
576	DECLINLINE(bool) vboxNetFltLinuxIsGso(PINTNETSG pSG)
577	{
578	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
579	return !((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type == PDMNETWORKGSOTYPE_INVALID);
580	#else /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
581	return false;
582	#endif /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
583	}
584
585
586	/**
587	* Find out the frame size (of a single segment in case of GSO frames).
588	*
589	* @returns the frame size.
590	* @param pSG The (scatter/)gather list.
591	*/
592	DECLINLINE(uint32_t) vboxNetFltLinuxFrameSize(PINTNETSG pSG)
593	{
594	uint16_t u16Type = 0;
595	uint32_t cbVlanTag = 0;
596	if (pSG->aSegs[0].cb >= sizeof(RTNETETHERHDR))
597	u16Type = RT_BE2H_U16(((PCRTNETETHERHDR)pSG->aSegs[0].pv)->EtherType);
598	else if (pSG->cbTotal >= sizeof(RTNETETHERHDR))
599	{
600	uint32_t off = RT_OFFSETOF(RTNETETHERHDR, EtherType);
601	uint32_t i;
602	for (i = 0; i < pSG->cSegsUsed; ++i)
603	{
604	if (off <= pSG->aSegs[i].cb)
605	{
606	if (off + sizeof(uint16_t) <= pSG->aSegs[i].cb)
607	u16Type = RT_BE2H_U16((uint16_t )((uintptr_t)pSG->aSegs[i].pv + off));
608	else if (i + 1 < pSG->cSegsUsed)
609	u16Type = RT_BE2H_U16( ((uint16_t)( ((uint8_t *)pSG->aSegs[i].pv)[off] ) << 8)
610	+ (uint8_t )pSG->aSegs[i + 1].pv); /* ASSUMES no empty segments! */
611	/* else: frame is too short. */
612	break;
613	}
614	off -= pSG->aSegs[i].cb;
615	}
616	}
617	if (u16Type == RTNET_ETHERTYPE_VLAN)
618	cbVlanTag = 4;
619	return (vboxNetFltLinuxIsGso(pSG) ? (uint32_t)pSG->GsoCtx.cbMaxSeg + pSG->GsoCtx.cbHdrsTotal : pSG->cbTotal) - cbVlanTag;
620	}
621
622
623	/**
624	* Internal worker that create a linux sk_buff for a
625	* (scatter/)gather list.
626	*
627	* @returns Pointer to the sk_buff.
628	* @param pThis The instance.
629	* @param pSG The (scatter/)gather list.
630	* @param fDstWire Set if the destination is the wire.
631	*/
632	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
633	{
634	struct sk_buff *pPkt;
635	struct net_device *pDev;
636	unsigned fGsoType = 0;
637
638	if (pSG->cbTotal == 0)
639	{
640	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
641	return NULL;
642	}
643	Log5(("VBoxNetFlt: Packet to %s of %d bytes (frame=%d).\n", fDstWire?"wire":"host", pSG->cbTotal, vboxNetFltLinuxFrameSize(pSG)));
644	if (fDstWire && (vboxNetFltLinuxFrameSize(pSG) > ASMAtomicReadU32(&pThis->u.s.cbMtu) + 14))
645	{
646	static bool s_fOnce = true;
647	if (s_fOnce)
648	{
649	s_fOnce = false;
650	printk("VBoxNetFlt: Dropped over-sized packet (%d bytes) coming from internal network.\n", vboxNetFltLinuxFrameSize(pSG));
651	}
652	return NULL;
653	}
654
655	/** @todo We should use fragments mapping the SG buffers with large packets.
656	* 256 bytes seems to be the a threshold used a lot for this. It
657	* requires some nasty work on the intnet side though... */
658	/*
659	* Allocate a packet and copy over the data.
660	*/
661	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
662	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
663	if (RT_UNLIKELY(!pPkt))
664	{
665	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
666	pSG->pvUserData = NULL;
667	return NULL;
668	}
669	pPkt->dev = pDev;
670	pPkt->ip_summed = CHECKSUM_NONE;
671
672	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
673	skb_reserve(pPkt, NET_IP_ALIGN);
674
675	/* Copy the segments. */
676	skb_put(pPkt, pSG->cbTotal);
677	IntNetSgRead(pSG, pPkt->data);
678
679	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
680	/*
681	* Setup GSO if used by this packet.
682	*/
683	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
684	{
685	default:
686	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
687	/* fall thru */
688	case PDMNETWORKGSOTYPE_INVALID:
689	fGsoType = 0;
690	break;
691	case PDMNETWORKGSOTYPE_IPV4_TCP:
692	fGsoType = SKB_GSO_TCPV4;
693	break;
694	case PDMNETWORKGSOTYPE_IPV4_UDP:
695	fGsoType = SKB_GSO_UDP;
696	break;
697	case PDMNETWORKGSOTYPE_IPV6_TCP:
698	fGsoType = SKB_GSO_TCPV6;
699	break;
700	}
701	if (fGsoType)
702	{
703	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
704
705	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
706	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
707	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
708
709	/*
710	* We need to set checksum fields even if the packet goes to the host
711	* directly as it may be immediately forwarded by IP layer @bugref{5020}.
712	*/
713	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
714	pPkt->ip_summed = CHECKSUM_PARTIAL;
715	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
716	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
717	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
718	pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
719	else
720	pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
721	# else
722	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
723	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
724	pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
725	else
726	pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
727	# endif
728	if (!fDstWire)
729	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
730	}
731	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
732
733	/*
734	* Finish up the socket buffer.
735	*/
736	pPkt->protocol = eth_type_trans(pPkt, pDev);
737	if (fDstWire)
738	{
739	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
740
741	/* Restore ethernet header back. */
742	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
743	VBOX_SKB_RESET_MAC_HDR(pPkt);
744	}
745	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
746
747	return pPkt;
748	}
749
750
751	/**
752	* Initializes a SG list from an sk_buff.
753	*
754	* @returns Number of segments.
755	* @param pThis The instance.
756	* @param pBuf The sk_buff.
757	* @param pSG The SG.
758	* @param pvFrame The frame pointer, optional.
759	* @param cSegs The number of segments allocated for the SG.
760	* This should match the number in the mbuf exactly!
761	* @param fSrc The source of the frame.
762	* @param pGso Pointer to the GSO context if it's a GSO
763	* internal network frame. NULL if regular frame.
764	*/
765	DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
766	unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
767	{
768	int i;
769	NOREF(pThis);
770
771	Assert(!skb_shinfo(pBuf)->frag_list);
772
773	if (!pGsoCtx)
774	IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /cSegsUsed/);
775	else
776	IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /cSegsUsed/, pGsoCtx);
777
778	#ifdef VBOXNETFLT_SG_SUPPORT
779	pSG->aSegs[0].cb = skb_headlen(pBuf);
780	pSG->aSegs[0].pv = pBuf->data;
781	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
782
783	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
784	{
785	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
786	pSG->aSegs[i+1].cb = pFrag->size;
787	pSG->aSegs[i+1].pv = kmap(pFrag->page);
788	printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
789	pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
790	}
791	++i;
792
793	#else
794	pSG->aSegs[0].cb = pBuf->len;
795	pSG->aSegs[0].pv = pBuf->data;
796	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
797	i = 1;
798	#endif
799
800	pSG->cSegsUsed = i;
801
802	#ifdef PADD_RUNT_FRAMES_FROM_HOST
803	/*
804	* Add a trailer if the frame is too small.
805	*
806	* Since we're getting to the packet before it is framed, it has not
807	* yet been padded. The current solution is to add a segment pointing
808	* to a buffer containing all zeros and pray that works for all frames...
809	*/
810	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
811	{
812	static uint8_t const s_abZero[128] = {0};
813
814	AssertReturnVoid(i < cSegs);
815
816	pSG->aSegs[i].Phys = NIL_RTHCPHYS;
817	pSG->aSegs[i].pv = (void *)&s_abZero[0];
818	pSG->aSegs[i].cb = 60 - pSG->cbTotal;
819	pSG->cbTotal = 60;
820	pSG->cSegsUsed++;
821	Assert(i + 1 <= pSG->cSegsAlloc)
822	}
823	#endif
824
825	Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
826	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
827	for (i = 0; i < pSG->cSegsUsed; i++)
828	Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
829	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
830	}
831
832	/**
833	* Packet handler,
834	*
835	* @returns 0 or EJUSTRETURN.
836	* @param pThis The instance.
837	* @param pMBuf The mbuf.
838	* @param pvFrame The start of the frame, optional.
839	* @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
840	* @param eProtocol The protocol.
841	*/
842	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
843	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
844	struct net_device *pSkbDev,
845	struct packet_type *pPacketType,
846	struct net_device *pOrigDev)
847	#else
848	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
849	struct net_device *pSkbDev,
850	struct packet_type *pPacketType)
851	#endif
852	{
853	PVBOXNETFLTINS pThis;
854	struct net_device *pDev;
855	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
856	pBuf, pSkbDev, pPacketType));
857	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
858	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",
859	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));
860	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
861	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
862	# endif
863	#else
864	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",
865	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));
866	#endif
867	/*
868	* Drop it immediately?
869	*/
870	if (!pBuf)
871	return 0;
872
873	if (pBuf->pkt_type == PACKET_LOOPBACK)
874	{
875	/*
876	* We are not interested in loopbacked packets as they will always have
877	* another copy going to the wire.
878	*/
879	Log2(("vboxNetFltLinuxPacketHandler: dropped loopback packet (cb=%u)\n", pBuf->len));
880	dev_kfree_skb(pBuf); /* We must 'consume' all packets we get (@bugref{6539})! */
881	return 0;
882	}
883
884	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
885	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
886	if (pDev != pSkbDev)
887	{
888	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
889	kfree_skb(pBuf); /* This is a failure, so we use kfree_skb instead of dev_kfree_skb. */
890	return 0;
891	}
892
893	Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
894	if (vboxNetFltLinuxSkBufIsOur(pBuf))
895	{
896	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
897	dev_kfree_skb(pBuf);
898	return 0;
899	}
900
901	#ifndef VBOXNETFLT_SG_SUPPORT
902	{
903	/*
904	* Get rid of fragmented packets, they cause too much trouble.
905	*/
906	unsigned int uMacLen = pBuf->mac_len;
907	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
908	dev_kfree_skb(pBuf);
909	if (!pCopy)
910	{
911	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
912	return 0;
913	}
914	pBuf = pCopy;
915	/* Somehow skb_copy ignores mac_len */
916	pBuf->mac_len = uMacLen;
917	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
918	/* Restore VLAN tag stripped by host hardware */
919	if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
920	{
921	uint8_t pMac = (uint8_t)skb_mac_header(pBuf);
922	struct vlan_ethhdr pVHdr = (struct vlan_ethhdr )(pMac - VLAN_HLEN);
923	# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
924	memmove(pVHdr, pMac, ETH_ALEN * 2);
925	# else
926	memmove(pVHdr, pMac, VLAN_ETH_ALEN * 2);
927	# endif
928	pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
929	pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
930	pBuf->mac_header -= VLAN_HLEN;
931	pBuf->mac_len += VLAN_HLEN;
932	}
933	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
934
935	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
936	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",
937	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));
938	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
939	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
940	# endif
941	# else
942	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",
943	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));
944	# endif
945	}
946	#endif
947
948	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
949	/* Forward it to the internal network. */
950	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
951	#else
952	/* Add the packet to transmit queue and schedule the bottom half. */
953	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
954	schedule_work(&pThis->u.s.XmitTask);
955	Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
956	&pThis->u.s.XmitTask, pBuf));
957	#endif
958
959	/* It does not really matter what we return, it is ignored by the kernel. */
960	return 0;
961	}
962
963	/**
964	* Calculate the number of INTNETSEG segments the socket buffer will need.
965	*
966	* @returns Segment count.
967	* @param pBuf The socket buffer.
968	*/
969	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
970	{
971	#ifdef VBOXNETFLT_SG_SUPPORT
972	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
973	#else
974	unsigned cSegs = 1;
975	#endif
976	#ifdef PADD_RUNT_FRAMES_FROM_HOST
977	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
978	if (pBuf->len < 60)
979	cSegs++;
980	#endif
981	return cSegs;
982	}
983
984	/**
985	* Destroy the intnet scatter / gather buffer created by
986	* vboxNetFltLinuxSkBufToSG.
987	*/
988	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
989	{
990	#ifdef VBOXNETFLT_SG_SUPPORT
991	int i;
992
993	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
994	{
995	printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
996	kunmap(pSG->aSegs[i+1].pv);
997	}
998	#endif
999	NOREF(pSG);
1000	}
1001
1002	#ifdef LOG_ENABLED
1003	/**
1004	* Logging helper.
1005	*/
1006	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1007	{
1008	int i, offSeg;
1009	uint8_t pInt, pExt;
1010	static int iPacketNo = 1;
1011	iPacketNo += iIncrement;
1012	if (fEgress)
1013	{
1014	pExt = pSG->aSegs[0].pv;
1015	pInt = pExt + 6;
1016	}
1017	else
1018	{
1019	pInt = pSG->aSegs[0].pv;
1020	pExt = pInt + 6;
1021	}
1022	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1023	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1024	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1025	fEgress ? "-->" : "<--", pszWhere,
1026	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1027	pSG->cbTotal, iPacketNo));
1028	if (pSG->cSegsUsed == 1)
1029	{
1030	Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1031	}
1032	else
1033	{
1034	for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
1035	{
1036	Log3(("-- segment %d at 0x%x (%d bytes) --\n%.*Rhxd\n",
1037	i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1038	offSeg += pSG->aSegs[i].cb;
1039	}
1040	}
1041
1042	}
1043	#else
1044	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1045	#endif
1046
1047	#ifdef VBOXNETFLT_WITH_GSO_RECV
1048
1049	/**
1050	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1051	* GSO socket buffer without having to segment it.
1052	*
1053	* @returns true on success, false if needs segmenting.
1054	* @param pThis The net filter instance.
1055	* @param pSkb The GSO socket buffer.
1056	* @param fSrc The source.
1057	* @param pGsoCtx Where to return the GSO context on success.
1058	*/
1059	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1060	PPDMNETWORKGSO pGsoCtx)
1061	{
1062	PDMNETWORKGSOTYPE enmGsoType;
1063	uint16_t uEtherType;
1064	unsigned int cbTransport;
1065	unsigned int offTransport;
1066	unsigned int cbTransportHdr;
1067	unsigned uProtocol;
1068	union
1069	{
1070	RTNETIPV4 IPv4;
1071	RTNETIPV6 IPv6;
1072	RTNETTCP Tcp;
1073	uint8_t ab[40];
1074	uint16_t au16[40/2];
1075	uint32_t au32[40/4];
1076	} Buf;
1077
1078	/*
1079	* Check the GSO properties of the socket buffer and make sure it fits.
1080	*/
1081	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1082	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP \| SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
1083	{
1084	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1085	return false;
1086	}
1087	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1088	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
1089	{
1090	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1091	return false;
1092	}
1093	/*
1094	* It is possible to receive GSO packets from wire if GRO is enabled.
1095	*/
1096	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1097	{
1098	Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1099	#ifdef VBOXNETFLT_WITH_GRO
1100	/*
1101	* The packet came from the wire and the driver has already consumed
1102	* mac header. We need to restore it back.
1103	*/
1104	pSkb->mac_len = skb_network_header(pSkb) - skb_mac_header(pSkb);
1105	skb_push(pSkb, pSkb->mac_len);
1106	Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1107	pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
1108	#else /* !VBOXNETFLT_WITH_GRO */
1109	/* Older kernels didn't have GRO. */
1110	return false;
1111	#endif /* !VBOXNETFLT_WITH_GRO */
1112	}
1113	else
1114	{
1115	/*
1116	* skb_gso_segment does the following. Do we need to do it as well?
1117	*/
1118	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1119	skb_reset_mac_header(pSkb);
1120	pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1121	#else
1122	pSkb->mac.raw = pSkb->data;
1123	pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1124	#endif
1125	}
1126
1127	/*
1128	* Switch on the ethertype.
1129	*/
1130	uEtherType = pSkb->protocol;
1131	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1132	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1133	{
1134	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1135	if (puEtherType)
1136	uEtherType = *puEtherType;
1137	}
1138	switch (uEtherType)
1139	{
1140	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1141	{
1142	unsigned int cbHdr;
1143	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1144	if (RT_UNLIKELY(!pIPv4))
1145	{
1146	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1147	return false;
1148	}
1149
1150	cbHdr = pIPv4->ip_hl * 4;
1151	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1152	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1153	\|\| cbHdr > cbTransport ))
1154	{
1155	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1156	return false;
1157	}
1158	cbTransport -= cbHdr;
1159	offTransport = pSkb->mac_len + cbHdr;
1160	uProtocol = pIPv4->ip_p;
1161	if (uProtocol == RTNETIPV4_PROT_TCP)
1162	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1163	else if (uProtocol == RTNETIPV4_PROT_UDP)
1164	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1165	else /** @todo IPv6: 4to6 tunneling */
1166	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1167	break;
1168	}
1169
1170	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1171	{
1172	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1173	if (RT_UNLIKELY(!pIPv6))
1174	{
1175	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1176	return false;
1177	}
1178
1179	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1180	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1181	uProtocol = pIPv6->ip6_nxt;
1182	/** @todo IPv6: Dig our way out of the other headers. */
1183	if (uProtocol == RTNETIPV4_PROT_TCP)
1184	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1185	else if (uProtocol == RTNETIPV4_PROT_UDP)
1186	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1187	else
1188	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1189	break;
1190	}
1191
1192	default:
1193	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1194	return false;
1195	}
1196
1197	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1198	{
1199	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1200	return false;
1201	}
1202
1203	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1204	\|\| offTransport + cbTransport > pSkb->len
1205	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1206	{
1207	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1208	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1209	return false;
1210	}
1211
1212	/*
1213	* Check the TCP/UDP bits.
1214	*/
1215	if (uProtocol == RTNETIPV4_PROT_TCP)
1216	{
1217	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1218	if (RT_UNLIKELY(!pTcp))
1219	{
1220	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1221	return false;
1222	}
1223
1224	cbTransportHdr = pTcp->th_off * 4;
1225	pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1226	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1227	\|\| cbTransportHdr > cbTransport
1228	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1229	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1230	{
1231	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1232	return false;
1233	}
1234
1235	}
1236	else
1237	{
1238	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1239	cbTransportHdr = sizeof(RTNETUDP);
1240	pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1241	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1242	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1243	{
1244	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1245	return false;
1246	}
1247	}
1248
1249	/*
1250	* We're good, init the GSO context.
1251	*/
1252	pGsoCtx->u8Type = enmGsoType;
1253	pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1254	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1255	pGsoCtx->offHdr1 = pSkb->mac_len;
1256	pGsoCtx->offHdr2 = offTransport;
1257	pGsoCtx->u8Unused = 0;
1258
1259	return true;
1260	}
1261
1262	/**
1263	* Forward the socket buffer as a GSO internal network frame.
1264	*
1265	* @returns IPRT status code.
1266	* @param pThis The net filter instance.
1267	* @param pSkb The GSO socket buffer.
1268	* @param fSrc The source.
1269	* @param pGsoCtx Where to return the GSO context on success.
1270	*/
1271	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1272	{
1273	int rc;
1274	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1275	if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1276	{
1277	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1278	if (RT_LIKELY(pSG))
1279	{
1280	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1281
1282	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1283	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1284
1285	vboxNetFltLinuxDestroySG(pSG);
1286	rc = VINF_SUCCESS;
1287	}
1288	else
1289	{
1290	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1291	rc = VERR_NO_MEMORY;
1292	}
1293	}
1294	else
1295	{
1296	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1297	rc = VERR_INTERNAL_ERROR_3;
1298	}
1299
1300	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1301	dev_kfree_skb(pSkb);
1302	return rc;
1303	}
1304
1305	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1306
1307	/**
1308	* Worker for vboxNetFltLinuxForwardToIntNet.
1309	*
1310	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1311	* @param pThis The net filter instance.
1312	* @param pBuf The socket buffer.
1313	* @param fSrc The source.
1314	*/
1315	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1316	{
1317	int rc;
1318	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1319	if (cSegs <= MAX_SKB_FRAGS + 1)
1320	{
1321	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1322	if (RT_LIKELY(pSG))
1323	{
1324	if (fSrc & INTNETTRUNKDIR_WIRE)
1325	{
1326	/*
1327	* The packet came from wire, ethernet header was removed by device driver.
1328	* Restore it using mac_len field. This takes into account VLAN headers too.
1329	*/
1330	skb_push(pBuf, pBuf->mac_len);
1331	}
1332
1333	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /pGsoCtx/);
1334
1335	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1336	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1337
1338	vboxNetFltLinuxDestroySG(pSG);
1339	rc = VINF_SUCCESS;
1340	}
1341	else
1342	{
1343	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1344	rc = VERR_NO_MEMORY;
1345	}
1346	}
1347	else
1348	{
1349	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1350	rc = VERR_INTERNAL_ERROR_3;
1351	}
1352
1353	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1354	dev_kfree_skb(pBuf);
1355	return rc;
1356	}
1357
1358	/**
1359	*
1360	* @param pBuf The socket buffer. This is consumed by this function.
1361	*/
1362	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1363	{
1364	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1365
1366	#ifdef VBOXNETFLT_WITH_GSO
1367	if (skb_is_gso(pBuf))
1368	{
1369	PDMNETWORKGSO GsoCtx;
1370	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",
1371	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));
1372	# ifdef VBOXNETFLT_WITH_GSO_RECV
1373	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1374	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1375	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1376	else
1377	# endif
1378	{
1379	/* Need to segment the packet */
1380	struct sk_buff *pNext;
1381	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1382	if (IS_ERR(pSegment))
1383	{
1384	dev_kfree_skb(pBuf);
1385	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1386	return;
1387	}
1388
1389	for (; pSegment; pSegment = pNext)
1390	{
1391	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",
1392	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));
1393	pNext = pSegment->next;
1394	pSegment->next = 0;
1395	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1396	}
1397	dev_kfree_skb(pBuf);
1398	}
1399	}
1400	else
1401	#endif /* VBOXNETFLT_WITH_GSO */
1402	{
1403	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1404	{
1405	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1406	/*
1407	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1408	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1409	* header length from the header itself and reconstruct 'h' pointer
1410	* to TCP (or whatever) header.
1411	*/
1412	unsigned char *tmp = pBuf->h.raw;
1413	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1414	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1415	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1416	if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1417	{
1418	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1419	dev_kfree_skb(pBuf);
1420	return;
1421	}
1422	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1423	/* Restore the original (wrong) pointer. */
1424	pBuf->h.raw = tmp;
1425	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1426	}
1427	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1428	}
1429	}
1430
1431	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1432	/**
1433	* Work queue handler that forwards the socket buffers queued by
1434	* vboxNetFltLinuxPacketHandler to the internal network.
1435	*
1436	* @param pWork The work queue.
1437	*/
1438	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1439	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1440	# else
1441	static void vboxNetFltLinuxXmitTask(void *pWork)
1442	# endif
1443	{
1444	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1445	struct sk_buff *pBuf;
1446
1447	Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1448
1449	/*
1450	* Active? Retain the instance and increment the busy counter.
1451	*/
1452	if (vboxNetFltTryRetainBusyActive(pThis))
1453	{
1454	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1455	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1456
1457	vboxNetFltRelease(pThis, true /* fBusy */);
1458	}
1459	else
1460	{
1461	/** @todo Shouldn't we just drop the packets here? There is little point in
1462	* making them accumulate when the VM is paused and it'll only waste
1463	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1464	* before start draining the packets (goes for the intnet ring buf
1465	* too)? */
1466	}
1467	}
1468	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1469
1470	/**
1471	* Reports the GSO capabilities of the hardware NIC.
1472	*
1473	* @param pThis The net filter instance. The caller hold a
1474	* reference to this.
1475	*/
1476	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1477	{
1478	#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1479	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1480	{
1481	struct net_device *pDev;
1482	PINTNETTRUNKSWPORT pSwitchPort;
1483	unsigned int fFeatures;
1484
1485	RTSpinlockAcquire(pThis->hSpinlock);
1486
1487	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1488	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1489	if (pDev)
1490	fFeatures = pDev->features;
1491	else
1492	fFeatures = 0;
1493
1494	RTSpinlockRelease(pThis->hSpinlock);
1495
1496	if (pThis->pSwitchPort)
1497	{
1498	/* Set/update the GSO capabilities of the NIC. */
1499	uint32_t fGsoCapabilites = 0;
1500	if (fFeatures & NETIF_F_TSO)
1501	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1502	if (fFeatures & NETIF_F_TSO6)
1503	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1504	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1505	if (fFeatures & NETIF_F_UFO)
1506	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1507	if (fFeatures & NETIF_F_UFO)
1508	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1509	# endif
1510	Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1511	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1512	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : "",
1513	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)) ? "ufo " : "",
1514	(fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)) ? "ufo6 " : ""));
1515	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1516	}
1517
1518	vboxNetFltRelease(pThis, true /fBusy/);
1519	}
1520	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1521	}
1522
1523	/**
1524	* Helper that determines whether the host (ignoreing us) is operating the
1525	* interface in promiscuous mode or not.
1526	*/
1527	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1528	{
1529	bool fRc = false;
1530	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1531	if (pDev)
1532	{
1533	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1534	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1535	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1536	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1537	}
1538	return fRc;
1539	}
1540
1541	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36)
1542	/**
1543	* Helper for detecting TAP devices.
1544	*/
1545	static bool vboxNetFltIsTapDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1546	{
1547	if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1548	{
1549	struct ethtool_drvinfo Info;
1550
1551	memset(&Info, 0, sizeof(Info));
1552	Info.cmd = ETHTOOL_GDRVINFO;
1553	pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1554	Log3(("vboxNetFltIsTapDevice: driver=%s version=%s bus_info=%s\n",
1555	Info.driver, Info.version, Info.bus_info));
1556
1557	return !strncmp(Info.driver, "tun", 4)
1558	&& !strncmp(Info.bus_info, "tap", 4);
1559	}
1560
1561	return false;
1562	}
1563
1564	/**
1565	* Helper for updating the link state of TAP devices.
1566	* Only TAP devices are affected.
1567	*/
1568	static void vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1569	{
1570	if (vboxNetFltIsTapDevice(pThis, pDev))
1571	{
1572	Log3(("vboxNetFltSetTapLinkState: bringing %s tap device link state\n",
1573	fLinkUp ? "up" : "down"));
1574	netif_tx_lock_bh(pDev);
1575	if (fLinkUp)
1576	netif_carrier_on(pDev);
1577	else
1578	netif_carrier_off(pDev);
1579	netif_tx_unlock_bh(pDev);
1580	}
1581	}
1582	#else /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1583	DECLINLINE(void) vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1584	{
1585	/* Nothing to do for pre-2.6.36 kernels. */
1586	}
1587	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1588
1589	/**
1590	* Internal worker for vboxNetFltLinuxNotifierCallback.
1591	*
1592	* @returns VBox status code.
1593	* @param pThis The instance.
1594	* @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1595	* flood the release log.
1596	*/
1597	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1598	{
1599	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1600
1601	/*
1602	* Retain and store the device.
1603	*/
1604	dev_hold(pDev);
1605
1606	RTSpinlockAcquire(pThis->hSpinlock);
1607	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1608	RTSpinlockRelease(pThis->hSpinlock);
1609
1610	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1611	pDev, pDev->name,
1612	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1613	netdev_refcnt_read(pDev)
1614	#else
1615	atomic_read(&pDev->refcnt)
1616	#endif
1617	));
1618	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1619	pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1620
1621	/* Get the mac address while we still have a valid net_device reference. */
1622	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1623	/* Initialize MTU */
1624	pThis->u.s.cbMtu = pDev->mtu;
1625
1626	/*
1627	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1628	*/
1629	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1630	pThis->u.s.PacketType.dev = pDev;
1631	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1632	dev_add_pack(&pThis->u.s.PacketType);
1633	ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1634	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1635
1636	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1637	vboxNetFltLinuxHookDev(pThis, pDev);
1638	#endif
1639
1640	/*
1641	* If attaching to TAP interface we need to bring the link state up
1642	* starting from 2.6.36 kernel.
1643	*/
1644	vboxNetFltSetTapLinkState(pThis, pDev, true);
1645
1646	/*
1647	* Set indicators that require the spinlock. Be abit paranoid about racing
1648	* the device notification handle.
1649	*/
1650	RTSpinlockAcquire(pThis->hSpinlock);
1651	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1652	if (pDev)
1653	{
1654	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1655	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1656	pDev = NULL; /* don't dereference it */
1657	}
1658	RTSpinlockRelease(pThis->hSpinlock);
1659
1660	/*
1661	* If the above succeeded report GSO capabilities, if not undo and
1662	* release the device.
1663	*/
1664	if (!pDev)
1665	{
1666	Assert(pThis->pSwitchPort);
1667	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1668	{
1669	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1670	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1671	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1672	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1673	vboxNetFltRelease(pThis, true /fBusy/);
1674	}
1675	}
1676	else
1677	{
1678	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1679	vboxNetFltLinuxUnhookDev(pThis, pDev);
1680	#endif
1681	RTSpinlockAcquire(pThis->hSpinlock);
1682	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1683	RTSpinlockRelease(pThis->hSpinlock);
1684	dev_put(pDev);
1685	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n",
1686	pDev, pDev->name,
1687	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1688	netdev_refcnt_read(pDev)
1689	#else
1690	atomic_read(&pDev->refcnt)
1691	#endif
1692	));
1693	}
1694
1695	LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1696	return VINF_SUCCESS;
1697	}
1698
1699
1700	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1701	{
1702	bool fRegistered;
1703	Assert(!pThis->fDisconnectedFromHost);
1704
1705	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1706	vboxNetFltLinuxUnhookDev(pThis, pDev);
1707	#endif
1708
1709	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1710	{
1711	dev_remove_pack(&pThis->u.s.PacketType);
1712	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1713	}
1714
1715	RTSpinlockAcquire(pThis->hSpinlock);
1716	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1717	if (fRegistered)
1718	{
1719	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1720	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1721	}
1722	RTSpinlockRelease(pThis->hSpinlock);
1723
1724	if (fRegistered)
1725	{
1726	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1727	skb_queue_purge(&pThis->u.s.XmitQueue);
1728	#endif
1729	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1730	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1731	pDev, pDev->name,
1732	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1733	netdev_refcnt_read(pDev)
1734	#else
1735	atomic_read(&pDev->refcnt)
1736	#endif
1737	));
1738	dev_put(pDev);
1739	}
1740
1741	return NOTIFY_OK;
1742	}
1743
1744	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1745	{
1746	/* Check if we are not suspended and promiscuous mode has not been set. */
1747	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1748	&& !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1749	{
1750	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1751	dev_set_promiscuity(pDev, 1);
1752	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1753	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1754	}
1755	else
1756	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1757	return NOTIFY_OK;
1758	}
1759
1760	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1761	{
1762	/* Undo promiscuous mode if we has set it. */
1763	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1764	{
1765	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1766	dev_set_promiscuity(pDev, -1);
1767	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1768	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1769	}
1770	else
1771	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1772	return NOTIFY_OK;
1773	}
1774
1775	/**
1776	* Callback for listening to MTU change event.
1777	*
1778	* We need to track changes of host's inteface MTU to discard over-sized frames
1779	* coming from the internal network as they may hang the TX queue of host's
1780	* adapter.
1781	*
1782	* @returns NOTIFY_OK
1783	* @param pThis The netfilter instance.
1784	* @param pDev Pointer to device structure of host's interface.
1785	*/
1786	static int vboxNetFltLinuxDeviceMtuChange(PVBOXNETFLTINS pThis, struct net_device *pDev)
1787	{
1788	ASMAtomicWriteU32(&pThis->u.s.cbMtu, pDev->mtu);
1789	Log(("vboxNetFltLinuxDeviceMtuChange: set MTU for %s to %d\n", pThis->szName, pDev->mtu));
1790	return NOTIFY_OK;
1791	}
1792
1793	#ifdef LOG_ENABLED
1794	/** Stringify the NETDEV_XXX constants. */
1795	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
1796	{
1797	const char *pszEvent = "NETDRV_<unknown>";
1798	switch (ulEventType)
1799	{
1800	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
1801	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
1802	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
1803	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
1804	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
1805	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
1806	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
1807	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
1808	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
1809	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
1810	# ifdef NETDEV_FEAT_CHANGE
1811	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
1812	# endif
1813	}
1814	return pszEvent;
1815	}
1816	#endif /* LOG_ENABLED */
1817
1818	/**
1819	* Callback for listening to netdevice events.
1820	*
1821	* This works the rediscovery, clean up on unregistration, promiscuity on
1822	* up/down, and GSO feature changes from ethtool.
1823	*
1824	* @returns NOTIFY_OK
1825	* @param self Pointer to our notifier registration block.
1826	* @param ulEventType The event.
1827	* @param ptr Event specific, but it is usually the device it
1828	* relates to.
1829	*/
1830	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
1831
1832	{
1833	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
1834	struct net_device pMyDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device );
1835	struct net_device *pDev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
1836	int rc = NOTIFY_OK;
1837
1838	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1839	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, pMyDev));
1840
1841	if ( ulEventType == NETDEV_REGISTER
1842	&& !strcmp(pDev->name, pThis->szName))
1843	{
1844	vboxNetFltLinuxAttachToInterface(pThis, pDev);
1845	}
1846	else
1847	{
1848	if (pDev == pMyDev)
1849	{
1850	switch (ulEventType)
1851	{
1852	case NETDEV_UNREGISTER:
1853	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
1854	break;
1855	case NETDEV_UP:
1856	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
1857	break;
1858	case NETDEV_GOING_DOWN:
1859	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
1860	break;
1861	case NETDEV_CHANGEMTU:
1862	rc = vboxNetFltLinuxDeviceMtuChange(pThis, pDev);
1863	break;
1864	case NETDEV_CHANGENAME:
1865	break;
1866	#ifdef NETDEV_FEAT_CHANGE
1867	case NETDEV_FEAT_CHANGE:
1868	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1869	break;
1870	#endif
1871	}
1872	}
1873	}
1874
1875	return rc;
1876	}
1877
1878	static int vboxNetFltLinuxEnumeratorCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
1879	{
1880	PVBOXNETFLTINS pThis = ((PVBOXNETFLTNOTIFIER)self)->pThis;
1881	struct net_device *dev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
1882	struct in_device *in_dev;
1883	struct inet6_dev *in6_dev;
1884
1885	if (ulEventType != NETDEV_REGISTER)
1886	return NOTIFY_OK;
1887
1888	if (RT_UNLIKELY(pThis->pSwitchPort->pfnNotifyHostAddress == NULL))
1889	return NOTIFY_OK;
1890
1891	/*
1892	* IPv4
1893	*/
1894	in_dev = __in_dev_get_rcu(dev);
1895	if (in_dev != NULL)
1896	{
1897	for_ifa(in_dev) {
1898	if (IN_LOOPBACK(ntohl(ifa->ifa_address)))
1899	return NOTIFY_OK;
1900
1901	Log(("%s: %s: IPv4 addr %RTnaipv4 mask %RTnaipv4\n",
1902	__FUNCTION__, VBOX_NETDEV_NAME(dev),
1903	ifa->ifa_address, ifa->ifa_mask));
1904
1905	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1906	/* :fAdded */ true, kIntNetAddrType_IPv4, &ifa->ifa_address);
1907	} endfor_ifa(in_dev);
1908	}
1909
1910	/*
1911	* IPv6
1912	*/
1913	in6_dev = __in6_dev_get(dev);
1914	if (in6_dev != NULL)
1915	{
1916	struct inet6_ifaddr *ifa;
1917
1918	read_lock_bh(&in6_dev->lock);
1919	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 35)
1920	list_for_each_entry(ifa, &in6_dev->addr_list, if_list)
1921	#else
1922	for (ifa = in6_dev->addr_list; ifa != NULL; ifa = ifa->if_next)
1923	#endif
1924	{
1925	if ( dev != pThis->u.s.pDev
1926	&& ipv6_addr_src_scope(&ifa->addr) <= IPV6_ADDR_SCOPE_LINKLOCAL)
1927	continue;
1928
1929	Log(("%s: %s: IPv6 addr %RTnaipv6/%u\n",
1930	__FUNCTION__, VBOX_NETDEV_NAME(dev),
1931	&ifa->addr, (unsigned)ifa->prefix_len));
1932
1933	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1934	/* :fAdded */ true, kIntNetAddrType_IPv6, &ifa->addr);
1935	}
1936	read_unlock_bh(&in6_dev->lock);
1937	}
1938
1939	return NOTIFY_OK;
1940	}
1941
1942
1943	static int vboxNetFltLinuxNotifierIPv4Callback(struct notifier_block self, unsigned long ulEventType, void ptr)
1944	{
1945	PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv4);
1946	struct net_device *pDev;
1947	struct in_ifaddr ifa = (struct in_ifaddr )ptr;
1948	int rc = NOTIFY_OK;
1949
1950	pDev = vboxNetFltLinuxRetainNetDev(pThis);
1951	Log(("VBoxNetFlt: %s: IPv4 event %s(0x%lx): addr %RTnaipv4 mask %RTnaipv4\n",
1952	pDev ? VBOX_NETDEV_NAME(pDev) : "<???>",
1953	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
1954	ifa->ifa_address, ifa->ifa_mask));
1955
1956	if (pDev != NULL)
1957	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1958
1959	if (pThis->pSwitchPort->pfnNotifyHostAddress)
1960	{
1961	bool fAdded;
1962	if (ulEventType == NETDEV_UP)
1963	fAdded = true;
1964	else if (ulEventType == NETDEV_DOWN)
1965	fAdded = false;
1966	else
1967	return NOTIFY_OK;
1968
1969	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
1970	kIntNetAddrType_IPv4, &ifa->ifa_local);
1971	}
1972
1973	return rc;
1974	}
1975
1976
1977	static int vboxNetFltLinuxNotifierIPv6Callback(struct notifier_block self, unsigned long ulEventType, void ptr)
1978	{
1979	PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv6);
1980	struct net_device *pDev;
1981	struct inet6_ifaddr ifa = (struct inet6_ifaddr )ptr;
1982	int rc = NOTIFY_OK;
1983
1984	pDev = vboxNetFltLinuxRetainNetDev(pThis);
1985	Log(("VBoxNetFlt: %s: IPv6 event %s(0x%lx): %RTnaipv6\n",
1986	pDev ? VBOX_NETDEV_NAME(pDev) : "<???>",
1987	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
1988	&ifa->addr));
1989
1990	if (pDev != NULL)
1991	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1992
1993	if (pThis->pSwitchPort->pfnNotifyHostAddress)
1994	{
1995	bool fAdded;
1996	if (ulEventType == NETDEV_UP)
1997	fAdded = true;
1998	else if (ulEventType == NETDEV_DOWN)
1999	fAdded = false;
2000	else
2001	return NOTIFY_OK;
2002
2003	pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2004	kIntNetAddrType_IPv6, &ifa->addr);
2005	}
2006
2007	return rc;
2008	}
2009
2010
2011	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2012	{
2013	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2014	}
2015
2016	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2017	{
2018	struct net_device * pDev;
2019	int err;
2020	int rc = VINF_SUCCESS;
2021	NOREF(pvIfData);
2022
2023	LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2024
2025	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2026	if (pDev)
2027	{
2028	/*
2029	* Create a sk_buff for the gather list and push it onto the wire.
2030	*/
2031	if (fDst & INTNETTRUNKDIR_WIRE)
2032	{
2033	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2034	if (pBuf)
2035	{
2036	vboxNetFltDumpPacket(pSG, true, "wire", 1);
2037	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2038	Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2039	err = dev_queue_xmit(pBuf);
2040	if (err)
2041	rc = RTErrConvertFromErrno(err);
2042	}
2043	else
2044	rc = VERR_NO_MEMORY;
2045	}
2046
2047	/*
2048	* Create a sk_buff for the gather list and push it onto the host stack.
2049	*/
2050	if (fDst & INTNETTRUNKDIR_HOST)
2051	{
2052	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2053	if (pBuf)
2054	{
2055	vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2056	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2057	Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2058	err = netif_rx_ni(pBuf);
2059	if (err)
2060	rc = RTErrConvertFromErrno(err);
2061	}
2062	else
2063	rc = VERR_NO_MEMORY;
2064	}
2065
2066	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2067	}
2068
2069	return rc;
2070	}
2071
2072
2073	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2074	{
2075	struct net_device * pDev;
2076
2077	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
2078	pThis, pThis->szName, fActive?"true":"false",
2079	pThis->fDisablePromiscuous?"true":"false"));
2080
2081	if (pThis->fDisablePromiscuous)
2082	return;
2083
2084	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2085	if (pDev)
2086	{
2087	/*
2088	* This api is a bit weird, the best reference is the code.
2089	*
2090	* Also, we have a bit or race conditions wrt the maintenance of
2091	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2092	*/
2093	#ifdef LOG_ENABLED
2094	u_int16_t fIf;
2095	unsigned const cPromiscBefore = pDev->promiscuity;
2096	#endif
2097	if (fActive)
2098	{
2099	Assert(!pThis->u.s.fPromiscuousSet);
2100
2101	rtnl_lock();
2102	dev_set_promiscuity(pDev, 1);
2103	rtnl_unlock();
2104	pThis->u.s.fPromiscuousSet = true;
2105	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2106	}
2107	else
2108	{
2109	if (pThis->u.s.fPromiscuousSet)
2110	{
2111	rtnl_lock();
2112	dev_set_promiscuity(pDev, -1);
2113	rtnl_unlock();
2114	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2115	}
2116	pThis->u.s.fPromiscuousSet = false;
2117
2118	#ifdef LOG_ENABLED
2119	fIf = dev_get_flags(pDev);
2120	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2121	#endif
2122	}
2123
2124	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2125	}
2126	}
2127
2128
2129	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2130	{
2131	/*
2132	* Remove packet handler when we get disconnected from internal switch as
2133	* we don't want the handler to forward packets to disconnected switch.
2134	*/
2135	if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
2136	{
2137	dev_remove_pack(&pThis->u.s.PacketType);
2138	Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
2139	}
2140	return VINF_SUCCESS;
2141	}
2142
2143
2144	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2145	{
2146	/*
2147	* Report the GSO capabilities of the host and device (if connected).
2148	* Note! No need to mark ourselves busy here.
2149	*/
2150	/** @todo duplicate work here now? Attach */
2151	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2152	Log3(("vboxNetFltOsConnectIt: reporting host tso tso6 ufo\n"));
2153	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2154	0
2155	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2156	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2157	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2158	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2159	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2160	# endif
2161	, INTNETTRUNKDIR_HOST);
2162
2163	#endif
2164	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2165
2166	return VINF_SUCCESS;
2167	}
2168
2169
2170	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2171	{
2172	struct net_device *pDev;
2173	bool fRegistered;
2174
2175	#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
2176	vboxNetFltLinuxUnhookDev(pThis, NULL);
2177	#endif
2178
2179	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2180	* unlikely, but none the less). Since it doesn't actually update the
2181	* state (just reads it), it is likely to panic in some interesting
2182	* ways. */
2183
2184	RTSpinlockAcquire(pThis->hSpinlock);
2185	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2186	fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
2187	RTSpinlockRelease(pThis->hSpinlock);
2188
2189	if (fRegistered)
2190	{
2191	vboxNetFltSetTapLinkState(pThis, pDev, false);
2192
2193	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2194	skb_queue_purge(&pThis->u.s.XmitQueue);
2195	#endif
2196	Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
2197	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
2198	pDev, pDev->name,
2199	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
2200	netdev_refcnt_read(pDev)
2201	#else
2202	atomic_read(&pDev->refcnt)
2203	#endif
2204	));
2205	dev_put(pDev);
2206	}
2207
2208	unregister_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2209	unregister_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2210
2211	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2212	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2213	module_put(THIS_MODULE);
2214	}
2215
2216
2217	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2218	{
2219	int err;
2220	NOREF(pvContext);
2221
2222	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2223	err = register_netdevice_notifier(&pThis->u.s.Notifier);
2224	if (err)
2225	return VERR_INTNET_FLT_IF_FAILED;
2226	if (!pThis->u.s.fRegistered)
2227	{
2228	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2229	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2230	return VERR_INTNET_FLT_IF_NOT_FOUND;
2231	}
2232
2233	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2234	if ( pThis->fDisconnectedFromHost
2235	\|\| !try_module_get(THIS_MODULE))
2236	return VERR_INTNET_FLT_IF_FAILED;
2237
2238	if (pThis->pSwitchPort->pfnNotifyHostAddress)
2239	{
2240	VBOXNETFLTNOTIFIER Enumerator;
2241
2242	/*
2243	* register_inetaddr_notifier() and register_inet6addr_notifier()
2244	* do not call the callback for existing devices. Enumerating
2245	* all network devices explicitly is a bit of an ifdef mess,
2246	* so co-opt register_netdevice_notifier() to do that for us.
2247	*/
2248	RT_ZERO(Enumerator);
2249	Enumerator.Notifier.notifier_call = vboxNetFltLinuxEnumeratorCallback;
2250	Enumerator.pThis = pThis;
2251
2252	err = register_netdevice_notifier(&Enumerator.Notifier);
2253	if (err)
2254	{
2255	LogRel(("%s: failed to enumerate network devices: error %d\n",
2256	__FUNCTION__, err));
2257	return VINF_SUCCESS;
2258	}
2259
2260	unregister_netdevice_notifier(&Enumerator.Notifier);
2261
2262	pThis->u.s.NotifierIPv4.notifier_call = vboxNetFltLinuxNotifierIPv4Callback;
2263	err = register_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2264	if (err)
2265	LogRel(("%s: failed to register IPv4 notifier: error %d\n",
2266	__FUNCTION__, err));
2267
2268	pThis->u.s.NotifierIPv6.notifier_call = vboxNetFltLinuxNotifierIPv6Callback;
2269	err = register_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2270	if (err)
2271	LogRel(("%s: failed to register IPv6 notifier: error %d\n",
2272	__FUNCTION__, err));
2273	}
2274
2275	return VINF_SUCCESS;
2276	}
2277
2278	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2279	{
2280	/*
2281	* Init the linux specific members.
2282	*/
2283	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2284	pThis->u.s.fRegistered = false;
2285	pThis->u.s.fPromiscuousSet = false;
2286	pThis->u.s.fPacketHandler = false;
2287	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2288	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2289	skb_queue_head_init(&pThis->u.s.XmitQueue);
2290	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2291	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2292	# else
2293	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2294	# endif
2295	#endif
2296
2297	return VINF_SUCCESS;
2298	}
2299
2300
2301	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2302	{
2303	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2304	}
2305
2306
2307	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
2308	{
2309	/* Nothing to do */
2310	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2311	return VINF_SUCCESS;
2312	}
2313
2314
2315	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2316	{
2317	/* Nothing to do */
2318	NOREF(pThis); NOREF(pvIfData);
2319	return VINF_SUCCESS;
2320	}
2321

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 56143

Download in other formats: