VBoxNetFlt-linux.c@ 30111

Last change on this file since 30111 was 30111, checked in by vboxsync, 15 years ago
iprt/asm.h,: Revised the ASMAtomicPtr functions and macros. The new saves lots of unsafe (void * volatile *) casts as well as adding some type safety when using GCC (typeof rulez).
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 80.7 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 30111 2010-06-09 12:14:59Z vboxsync $ */
2	/** @file
3	* VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2008 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/*******************************************************************************
19	* Header Files *
20	*******************************************************************************/
21	#define LOG_GROUP LOG_GROUP_NET_FLT_DRV
22	#define VBOXNETFLT_LINUX_NO_XMIT_QUEUE
23	#include "the-linux-kernel.h"
24	#include "version-generated.h"
25	#include "product-generated.h"
26	#include <linux/netdevice.h>
27	#include <linux/etherdevice.h>
28	#include <linux/rtnetlink.h>
29	#include <linux/miscdevice.h>
30	#include <linux/ip.h>
31
32	#include <VBox/log.h>
33	#include <VBox/err.h>
34	#include <VBox/intnetinline.h>
35	#include <VBox/pdmnetinline.h>
36	#include <VBox/param.h>
37	#include <iprt/alloca.h>
38	#include <iprt/assert.h>
39	#include <iprt/spinlock.h>
40	#include <iprt/semaphore.h>
41	#include <iprt/initterm.h>
42	#include <iprt/process.h>
43	#include <iprt/mem.h>
44	#include <iprt/net.h>
45	#include <iprt/log.h>
46	#include <iprt/mp.h>
47	#include <iprt/mem.h>
48	#include <iprt/time.h>
49
50	#define VBOXNETFLT_OS_SPECFIC 1
51	#include "../VBoxNetFltInternal.h"
52
53	/*
54	* Uncomment the following line to enable qdisc support.
55	*/
56	//#define VBOXNETFLT_WITH_QDISC
57	#ifdef VBOXNETFLT_WITH_QDISC
58	#include <net/pkt_sched.h>
59	#endif /* VBOXNETFLT_WITH_QDISC */
60
61
62	/*******************************************************************************
63	* Defined Constants And Macros *
64	*******************************************************************************/
65	#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
66	#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
67	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
68	# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
69	#endif
70
71	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
72	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
73	# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
74	#else
75	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
76	# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
77	#endif
78
79	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
80	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
81	#else
82	# define CHECKSUM_PARTIAL CHECKSUM_HW
83	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
84	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
85	# else
86	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
87	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
88	# else
89	# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
90	# endif
91	# endif
92	#endif
93
94	#ifndef NET_IP_ALIGN
95	# define NET_IP_ALIGN 2
96	#endif
97
98	#if 0
99	/** Create scatter / gather segments for fragments. When not used, we will
100	* linearize the socket buffer before creating the internal networking SG. */
101	# define VBOXNETFLT_SG_SUPPORT 1
102	#endif
103
104	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
105	/** Indicates that the linux kernel may send us GSO frames. */
106	# define VBOXNETFLT_WITH_GSO 1
107
108	/** This enables or disables the transmitting of GSO frame from the internal
109	* network and to the host. */
110	# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
111
112	# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
113	/** This enables or disables the transmitting of GSO frame from the internal
114	* network and to the wire. */
115	# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
116	# endif
117
118	/** This enables or disables the forwarding/flooding of GSO frame from the host
119	* to the internal network. */
120	# define VBOXNETFLT_WITH_GSO_RECV 1
121
122	#endif
123
124
125	/*******************************************************************************
126	* Internal Functions *
127	*******************************************************************************/
128	static int VBoxNetFltLinuxInit(void);
129	static void VBoxNetFltLinuxUnload(void);
130	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
131
132
133	/*******************************************************************************
134	* Global Variables *
135	*******************************************************************************/
136	/**
137	* The (common) global data.
138	*/
139	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
140
141	module_init(VBoxNetFltLinuxInit);
142	module_exit(VBoxNetFltLinuxUnload);
143
144	MODULE_AUTHOR(VBOX_VENDOR);
145	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
146	MODULE_LICENSE("GPL");
147	#ifdef MODULE_VERSION
148	MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
149	#endif
150
151
152	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
153	unsigned dev_get_flags(const struct net_device *dev)
154	{
155	unsigned flags;
156
157	flags = (dev->flags & ~(IFF_PROMISC \|
158	IFF_ALLMULTI \|
159	IFF_RUNNING)) \|
160	(dev->gflags & (IFF_PROMISC \|
161	IFF_ALLMULTI));
162
163	if (netif_running(dev) && netif_carrier_ok(dev))
164	flags \|= IFF_RUNNING;
165
166	return flags;
167	}
168	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
169
170
171	#ifdef VBOXNETFLT_WITH_QDISC
172	//#define QDISC_LOG(x) printk x
173	#define QDISC_LOG(x) do { } while (0)
174
175	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
176	#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops)
177	#elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
178	#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops, parent)
179	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
180	#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, queue, ops, parent)
181	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
182
183	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
184	#define qdisc_dev(qdisc) (qdisc->dev)
185	#define qdisc_pkt_len(skb) (skb->len)
186	#define QDISC_GET(dev) (dev->qdisc_sleeping)
187	#else
188	#define QDISC_GET(dev) (netdev_get_tx_queue(dev, 0)->qdisc_sleeping)
189	#endif
190
191	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
192	#define QDISC_SAVED_NUM(dev) 1
193	#elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
194	#define QDISC_SAVED_NUM(dev) dev->num_tx_queues
195	#else
196	#define QDISC_SAVED_NUM(dev) dev->num_tx_queues+1
197	#endif
198
199	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
200	#define QDISC_IS_BUSY(dev, qdisc) test_bit(__LINK_STATE_SCHED, &dev->state)
201	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
202	#define QDISC_IS_BUSY(dev, qdisc) (test_bit(__QDISC_STATE_RUNNING, &qdisc->state) \|\| \
203	test_bit(__QDISC_STATE_SCHED, &qdisc->state))
204	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
205
206	struct VBoxNetQDiscPriv
207	{
208	/** Pointer to the single child qdisc. */
209	struct Qdisc *pChild;
210	/*
211	* Technically it is possible to have different qdiscs for different TX
212	* queues so we have to save them all.
213	*/
214	/** Pointer to the array of saved qdiscs. */
215	struct Qdisc **ppSaved;
216	/** Pointer to the net filter instance. */
217	PVBOXNETFLTINS pVBoxNetFlt;
218	};
219	typedef struct VBoxNetQDiscPriv *PVBOXNETQDISCPRIV;
220
221	//#define VBOXNETFLT_QDISC_ENQUEUE
222	static int vboxNetFltQdiscEnqueue(struct sk_buff skb, struct Qdisc sch)
223	{
224	int rc;
225	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
226	#ifdef VBOXNETFLT_QDISC_ENQUEUE
227	if (VALID_PTR(pPriv->pVBoxNetFlt))
228	{
229	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
230	PCRTNETETHERHDR pEtherHdr;
231	PINTNETTRUNKSWPORT pSwitchPort;
232
233	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(skb, 0, sizeof(abHdrBuf), &abHdrBuf[0]);
234	if ( pEtherHdr
235	&& (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) != NULL
236	&& VALID_PTR(pSwitchPort)
237	)
238	{
239	struct sk_buff *pBuf;
240	INTNETSWDECISION enmDecision;
241	uint32_t cbHdrs = skb_headlen(skb);
242	cbHdrs = RT_MAX(cbHdrs, sizeof(abHdrBuf));
243
244	#if 1
245	/* Test stub due to missing implementation of pfnPreRecv */
246	static uint8_t tmpAddr[] = { 0x08, 0x00, 0x27, 0xBB, 0x1E, 0xA0 };
247	if (memcmp(pEtherHdr, tmpAddr, 6) != 0)
248	enmDecision = INTNETSWDECISION_BROADCAST;
249	else
250	enmDecision = INTNETSWDECISION_INTNET;
251	#else
252	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
253	#endif
254	if (enmDecision == INTNETSWDECISION_INTNET)
255	{
256	pBuf = skb_copy(skb, GFP_ATOMIC);
257	pBuf->pkt_type = PACKET_OUTGOING;
258	vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
259	qdisc_drop(skb, sch);
260	++sch->bstats.packets;
261	sch->bstats.bytes += qdisc_pkt_len(skb);
262	return NET_XMIT_SUCCESS;
263	}
264	}
265	}
266	#endif /* VBOXNETFLT_QDISC_ENQUEUE */
267	rc = pPriv->pChild->enqueue(skb, pPriv->pChild);
268	if (rc == NET_XMIT_SUCCESS)
269	{
270	++sch->q.qlen;
271	++sch->bstats.packets;
272	sch->bstats.bytes += qdisc_pkt_len(skb);
273	}
274	else
275	++sch->qstats.drops;
276	return rc;
277	}
278
279	static struct sk_buff vboxNetFltQdiscDequeue(struct Qdisc sch)
280	{
281	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
282	#ifdef VBOXNETFLT_QDISC_ENQUEUE
283	--sch->q.qlen;
284	return pPriv->pChild->dequeue(pPriv->pChild);
285	#else /* VBOXNETFLT_QDISC_ENQUEUE */
286	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
287	PCRTNETETHERHDR pEtherHdr;
288	PINTNETTRUNKSWPORT pSwitchPort;
289	struct sk_buff *pSkb;
290
291	QDISC_LOG(("vboxNetFltDequeue: Enter pThis=%p\n", pPriv->pVBoxNetFlt));
292
293	while ((pSkb = pPriv->pChild->dequeue(pPriv->pChild)) != NULL)
294	{
295	struct sk_buff *pBuf;
296	INTNETSWDECISION enmDecision;
297	uint32_t cbHdrs = skb_headlen(pSkb);
298
299	--sch->q.qlen;
300
301	if (!VALID_PTR(pPriv->pVBoxNetFlt))
302	break;
303
304	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, sizeof(abHdrBuf), &abHdrBuf[0]);
305	if ( !pEtherHdr
306	\|\| (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) == NULL
307	\|\| !VALID_PTR(pSwitchPort)
308	)
309	break;
310
311	cbHdrs = RT_MAX(cbHdrs, sizeof(abHdrBuf));
312
313	#if 1
314	/* Test stub due to missing implementation of pfnPreRecv */
315	static uint8_t tmpAddr[] = { 0x08, 0x00, 0x27, 0xBB, 0x1E, 0xA0 };
316	if (memcmp(pEtherHdr, tmpAddr, 6) != 0)
317	enmDecision = INTNETSWDECISION_BROADCAST;
318	else
319	enmDecision = INTNETSWDECISION_INTNET;
320	#else
321	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
322	#endif
323	if (enmDecision != INTNETSWDECISION_INTNET)
324	break;
325
326	pBuf = skb_copy(pSkb, GFP_ATOMIC);
327	pBuf->pkt_type = PACKET_OUTGOING;
328	QDISC_LOG(("vboxNetFltDequeue: pThis=%p\n", pPriv->pVBoxNetFlt));
329	vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
330	qdisc_drop(pSkb, sch);
331	QDISC_LOG(("VBoxNetFlt: Packet for %02x:%02x:%02x:%02x:%02x:%02x dropped\n",
332	pSkb->data[0], pSkb->data[1], pSkb->data[2],
333	pSkb->data[3], pSkb->data[4], pSkb->data[5]));
334	}
335
336	return pSkb;
337	#endif /* VBOXNETFLT_QDISC_ENQUEUE */
338	}
339
340	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
341	static int vboxNetFltQdiscRequeue(struct sk_buff skb, struct Qdisc sch)
342	{
343	int rc;
344	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
345
346	rc = pPriv->pChild->ops->requeue(skb, pPriv->pChild);
347	if (rc == 0)
348	{
349	sch->q.qlen++;
350	sch->qstats.requeues++;
351	}
352
353	return rc;
354	}
355	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
356
357	static unsigned int vboxNetFltQdiscDrop(struct Qdisc *sch)
358	{
359	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
360
361	++sch->qstats.drops;
362	--sch->q.qlen;
363	if (pPriv->pChild->ops->drop)
364	return pPriv->pChild->ops->drop(pPriv->pChild);
365
366	return 0;
367	}
368
369	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
370	static int vboxNetFltQdiscInit(struct Qdisc sch, struct rtattr opt)
371	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
372	static int vboxNetFltQdiscInit(struct Qdisc sch, struct nlattr opt)
373	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
374	{
375	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
376	struct net_device *pDev = qdisc_dev(sch);
377
378	pPriv->pVBoxNetFlt = NULL;
379
380	pPriv->ppSaved = kcalloc(QDISC_SAVED_NUM(pDev), sizeof(pPriv->ppSaved[0]),
381	GFP_KERNEL);
382	if (!pPriv->ppSaved)
383	return -ENOMEM;
384
385	pPriv->pChild = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
386	&pfifo_qdisc_ops,
387	TC_H_MAKE(TC_H_MAJ(sch->handle),
388	TC_H_MIN(1)));
389	if (!pPriv->pChild)
390	{
391	kfree(pPriv->ppSaved);
392	pPriv->ppSaved = NULL;
393	return -ENOMEM;
394	}
395
396	return 0;
397	}
398
399	static void vboxNetFltQdiscReset(struct Qdisc *sch)
400	{
401	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
402
403	qdisc_reset(pPriv->pChild);
404	sch->q.qlen = 0;
405	sch->qstats.backlog = 0;
406	}
407
408	static void vboxNetFltQdiscDestroy(struct Qdisc* sch)
409	{
410	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
411	struct net_device *pDev = qdisc_dev(sch);
412
413	qdisc_destroy(pPriv->pChild);
414	pPriv->pChild = NULL;
415
416	if (pPriv->ppSaved)
417	{
418	int i;
419	for (i = 0; i < QDISC_SAVED_NUM(pDev); i++)
420	if (pPriv->ppSaved[i])
421	qdisc_destroy(pPriv->ppSaved[i]);
422	kfree(pPriv->ppSaved);
423	pPriv->ppSaved = NULL;
424	}
425	}
426
427	static int vboxNetFltClassGraft(struct Qdisc sch, unsigned long arg, struct Qdisc pNew,
428	struct Qdisc **ppOld)
429	{
430	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
431
432	if (pNew == NULL)
433	pNew = &noop_qdisc;
434
435	sch_tree_lock(sch);
436	*ppOld = pPriv->pChild;
437	pPriv->pChild = pNew;
438	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
439	sch->q.qlen = 0;
440	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
441	qdisc_tree_decrease_qlen(ppOld, (ppOld)->q.qlen);
442	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
443	qdisc_reset(*ppOld);
444	sch_tree_unlock(sch);
445
446	return 0;
447	}
448
449	static struct Qdisc vboxNetFltClassLeaf(struct Qdisc sch, unsigned long arg)
450	{
451	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
452	return pPriv->pChild;
453	}
454
455	static unsigned long vboxNetFltClassGet(struct Qdisc *sch, u32 classid)
456	{
457	return 1;
458	}
459
460	static void vboxNetFltClassPut(struct Qdisc *sch, unsigned long arg)
461	{
462	}
463
464	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
465	static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
466	struct rtattr *tca, unsigned long arg)
467	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
468	static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
469	struct nlattr *tca, unsigned long arg)
470	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
471	{
472	return -ENOSYS;
473	}
474
475	static int vboxNetFltClassDelete(struct Qdisc *sch, unsigned long arg)
476	{
477	return -ENOSYS;
478	}
479
480	static void vboxNetFltClassWalk(struct Qdisc sch, struct qdisc_walker walker)
481	{
482	if (!walker->stop) {
483	if (walker->count >= walker->skip)
484	if (walker->fn(sch, 1, walker) < 0) {
485	walker->stop = 1;
486	return;
487	}
488	walker->count++;
489	}
490	}
491
492	static struct tcf_proto *vboxNetFltClassFindTcf(struct Qdisc sch, unsigned long cl)
493	{
494	return NULL;
495	}
496
497	static int vboxNetFltClassDump(struct Qdisc *sch, unsigned long cl,
498	struct sk_buff skb, struct tcmsg tcm)
499	{
500	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
501
502	if (cl != 1)
503	return -ENOENT;
504
505	tcm->tcm_handle \|= TC_H_MIN(1);
506	tcm->tcm_info = pPriv->pChild->handle;
507
508	return 0;
509	}
510
511
512	static struct Qdisc_class_ops g_VBoxNetFltClassOps =
513	{
514	.graft = vboxNetFltClassGraft,
515	.leaf = vboxNetFltClassLeaf,
516	.get = vboxNetFltClassGet,
517	.put = vboxNetFltClassPut,
518	.change = vboxNetFltClassChange,
519	.delete = vboxNetFltClassDelete,
520	.walk = vboxNetFltClassWalk,
521	.tcf_chain = vboxNetFltClassFindTcf,
522	.dump = vboxNetFltClassDump,
523	};
524
525
526	static struct Qdisc_ops g_VBoxNetFltQDiscOps = {
527	.cl_ops = &g_VBoxNetFltClassOps,
528	.id = "vboxnetflt",
529	.priv_size = sizeof(struct VBoxNetQDiscPriv),
530	.enqueue = vboxNetFltQdiscEnqueue,
531	.dequeue = vboxNetFltQdiscDequeue,
532	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
533	.requeue = vboxNetFltQdiscRequeue,
534	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
535	.peek = qdisc_peek_dequeued,
536	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
537	.drop = vboxNetFltQdiscDrop,
538	.init = vboxNetFltQdiscInit,
539	.reset = vboxNetFltQdiscReset,
540	.destroy = vboxNetFltQdiscDestroy,
541	.owner = THIS_MODULE
542	};
543
544	/*
545	* If our qdisc is already attached to the device (that means the user
546	* installed it from command line with 'tc' command) we simply update
547	* the pointer to vboxnetflt instance in qdisc's private structure.
548	* Otherwise we need to take some additional steps:
549	* - Create our qdisc;
550	* - Save all references to qdiscs;
551	* - Replace our child with the first qdisc reference;
552	* - Replace all references so they point to our qdisc.
553	*/
554	static void vboxNetFltLinuxQdiscInstall(PVBOXNETFLTINS pThis, struct net_device *pDev)
555	{
556	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
557	int i;
558	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
559	PVBOXNETQDISCPRIV pPriv;
560
561	struct Qdisc *pExisting = QDISC_GET(pDev);
562	if (strcmp(pExisting->ops->id, "vboxnetflt"))
563	{
564	/* The existing qdisc is different from ours, let's create new one. */
565	struct Qdisc *pNew = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
566	&g_VBoxNetFltQDiscOps, TC_H_ROOT);
567	if (!pNew)
568	return; // TODO: Error?
569
570	if (!try_module_get(THIS_MODULE))
571	{
572	/*
573	* This may cause a memory leak but calling qdisc_destroy()
574	* is not an option as it will call module_put().
575	*/
576	return;
577	}
578	pPriv = qdisc_priv(pNew);
579
580	qdisc_destroy(pPriv->pChild);
581	pPriv->pChild = QDISC_GET(pDev);
582	atomic_inc(&pPriv->pChild->refcnt);
583	/*
584	* There is no need in deactivating the device or acquiring any locks
585	* prior changing qdiscs since we do not destroy the old qdisc.
586	* Atomic replacement of pointers is enough.
587	*/
588	/*
589	* No need to change reference counters here as we merely move
590	* the pointer and the reference counter of the newly allocated
591	* qdisc is already 1.
592	*/
593	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
594	pPriv->ppSaved[0] = pDev->qdisc_sleeping;
595	ASMAtomicWritePtr(&pDev->qdisc_sleeping, pNew);
596	ASMAtomicWritePtr(&pDev->qdisc, pNew);
597	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
598	for (i = 0; i < pDev->num_tx_queues; i++)
599	{
600	struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
601
602	pPriv->ppSaved[i] = pQueue->qdisc_sleeping;
603	ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pNew);
604	ASMAtomicWritePtr(&pQueue->qdisc, pNew);
605	if (i)
606	atomic_inc(&pNew->refcnt);
607	}
608	/* Newer kernels store root qdisc in netdev structure as well. */
609	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
610	pPriv->ppSaved[pDev->num_tx_queues] = pDev->qdisc;
611	ASMAtomicWritePtr(&pDev->qdisc, pNew);
612	atomic_inc(&pNew->refcnt);
613	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
614	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
615	}
616	else
617	{
618	/* We already have vboxnetflt qdisc, let's use it. */
619	pPriv = qdisc_priv(pExisting);
620	}
621	ASMAtomicWritePtr(&pPriv->pVBoxNetFlt, pThis);
622	QDISC_LOG(("vboxNetFltLinuxInstallQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
623	}
624
625	static void vboxNetFltLinuxQdiscRemove(PVBOXNETFLTINS pThis, struct net_device *pDev)
626	{
627	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
628	int i;
629	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
630	PVBOXNETQDISCPRIV pPriv;
631	struct Qdisc *pQdisc;
632	if (!pDev)
633	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
634	if (!VALID_PTR(pDev))
635	{
636	printk("VBoxNetFlt: Failed to detach qdisc, invalid device pointer: %p\n",
637	pDev);
638	return; // TODO: Consider returing an error
639	}
640
641
642	pQdisc = QDISC_GET(pDev);
643	if (strcmp(pQdisc->ops->id, "vboxnetflt"))
644	{
645	/* Looks like the user has replaced our qdisc manually. */
646	printk("VBoxNetFlt: Failed to detach qdisc, wrong qdisc: %s\n",
647	pQdisc->ops->id);
648	return; // TODO: Consider returing an error
649	}
650
651	pPriv = qdisc_priv(pQdisc);
652	Assert(pPriv->pVBoxNetFlt == pThis);
653	ASMAtomicWritePtr(&pPriv->pVBoxNetFlt, NULL);
654
655	QDISC_LOG(("vboxNetFltLinuxQdiscRemove: refcnt=%d num_tx_queues=%d\n",
656	atomic_read(&pQdisc->refcnt), pDev->num_tx_queues));
657	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
658	/* Play it safe, make sure the qdisc is not being used. */
659	if (pPriv->ppSaved[0])
660	{
661	ASMAtomicWritePtr(&pDev->qdisc_sleeping, pPriv->ppSaved[0]);
662	ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[0]);
663	pPriv->ppSaved[0] = NULL;
664	while (QDISC_IS_BUSY(pDev, pQdisc))
665	yield();
666	qdisc_destroy(pQdisc); /* Destroy reference */
667	}
668	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
669	for (i = 0; i < pDev->num_tx_queues; i++)
670	{
671	struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
672	if (pPriv->ppSaved[i])
673	{
674	Assert(pQueue->qdisc_sleeping == pQdisc);
675	ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pPriv->ppSaved[i]);
676	ASMAtomicWritePtr(&pQueue->qdisc, pPriv->ppSaved[i]);
677	pPriv->ppSaved[i] = NULL;
678	while (QDISC_IS_BUSY(pDev, pQdisc))
679	yield();
680	qdisc_destroy(pQdisc); /* Destroy reference */
681	}
682	}
683	/* Newer kernels store root qdisc in netdev structure as well. */
684	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
685	ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[pDev->num_tx_queues]);
686	pPriv->ppSaved[pDev->num_tx_queues] = NULL;
687	while (QDISC_IS_BUSY(pDev, pQdisc))
688	yield();
689	qdisc_destroy(pQdisc); /* Destroy reference */
690	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
691	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
692
693	/*
694	* At this point all references to our qdisc should be gone
695	* unless the user had installed it manually.
696	*/
697	QDISC_LOG(("vboxNetFltLinuxRemoveQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
698	}
699
700	#endif /* VBOXNETFLT_WITH_QDISC */
701
702
703	/**
704	* Initialize module.
705	*
706	* @returns appropriate status code.
707	*/
708	static int __init VBoxNetFltLinuxInit(void)
709	{
710	int rc;
711	/*
712	* Initialize IPRT.
713	*/
714	rc = RTR0Init(0);
715	if (RT_SUCCESS(rc))
716	{
717	Log(("VBoxNetFltLinuxInit\n"));
718
719	/*
720	* Initialize the globals and connect to the support driver.
721	*
722	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
723	* for establishing the connect to the support driver.
724	*/
725	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
726	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
727	if (RT_SUCCESS(rc))
728	{
729	#ifdef VBOXNETFLT_WITH_QDISC
730	/*memcpy(&g_VBoxNetFltQDiscOps, &pfifo_qdisc_ops, sizeof(g_VBoxNetFltQDiscOps));
731	strcpy(g_VBoxNetFltQDiscOps.id, "vboxnetflt");
732	g_VBoxNetFltQDiscOps.owner = THIS_MODULE;*/
733	rc = register_qdisc(&g_VBoxNetFltQDiscOps);
734	if (rc)
735	{
736	LogRel(("VBoxNetFlt: Failed to registed qdisc: %d\n", rc));
737	return rc;
738	}
739	#endif /* VBOXNETFLT_WITH_QDISC */
740	LogRel(("VBoxNetFlt: Successfully started.\n"));
741	return 0;
742	}
743
744	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
745	RTR0Term();
746	}
747	else
748	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
749
750	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
751	return -RTErrConvertToErrno(rc);
752	}
753
754
755	/**
756	* Unload the module.
757	*
758	* @todo We have to prevent this if we're busy!
759	*/
760	static void __exit VBoxNetFltLinuxUnload(void)
761	{
762	int rc;
763	Log(("VBoxNetFltLinuxUnload\n"));
764	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
765
766	#ifdef VBOXNETFLT_WITH_QDISC
767	unregister_qdisc(&g_VBoxNetFltQDiscOps);
768	#endif /* VBOXNETFLT_WITH_QDISC */
769	/*
770	* Undo the work done during start (in reverse order).
771	*/
772	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
773	AssertRC(rc); NOREF(rc);
774
775	RTR0Term();
776
777	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
778
779	Log(("VBoxNetFltLinuxUnload - done\n"));
780	}
781
782
783	/**
784	* Experiment where we filter trafic from the host to the internal network
785	* before it reaches the NIC driver.
786	*
787	* The current code uses a very ugly hack and only works on kernels using the
788	* net_device_ops (>= 2.6.29). It has been shown to give us a
789	* performance boost of 60-100% though. So, we have to find some less hacky way
790	* of getting this job done eventually.
791	*
792	* #define VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
793	*/
794	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
795
796	/**
797	* The overridden net_device_ops of the device we're attached to.
798	*
799	* Requires Linux 2.6.29 or later.
800	*
801	* This is a very dirty hack that was create to explore how much we can improve
802	* the host to guest transfers by not CC'ing the NIC.
803	*/
804	typedef struct VBoxNetDeviceOpsOverride
805	{
806	/** Our overridden ops. */
807	struct net_device_ops Ops;
808	/** Magic word. */
809	uint32_t u32Magic;
810	/** Pointer to the original ops. */
811	struct net_device_ops const *pOrgOps;
812	/** Pointer to the net filter instance. */
813	PVBOXNETFLTINS pVBoxNetFlt;
814	/** The number of filtered packages. */
815	uint64_t cFiltered;
816	/** The total number of packets */
817	uint64_t cTotal;
818	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
819	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
820	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
821
822	/**
823	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
824	* because they belong on the internal network.
825	*
826	* @returns NETDEV_TX_XXX.
827	* @param pSkb The socket buffer to transmit.
828	* @param pDev The net device.
829	*/
830	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
831	{
832	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->netdev_ops;
833	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
834	PCRTNETETHERHDR pEtherHdr;
835	PINTNETTRUNKSWPORT pSwitchPort;
836
837
838	/*
839	* Validate the override structure.
840	*
841	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
842	* to be production quality code, we would have to be much more
843	* careful here and avoid the race.
844	*/
845	if ( !VALID_PTR(pOverride)
846	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
847	\|\| !VALID_PTR(pOverride->pOrgOps))
848	{
849	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
850	dev_kfree_skb(pSkb);
851	return NETDEV_TX_OK;
852	}
853	pOverride->cTotal++;
854
855	/*
856	* Do the filtering base on the defaul OUI of our virtual NICs
857	*
858	* Note! In a real solution, we would ask the switch whether the
859	* destination MAC is 100% to be on the internal network and then
860	* drop it.
861	*/
862	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, sizeof(abHdrBuf), &abHdrBuf[0]);
863	if ( pEtherHdr
864	&& VALID_PTR(pOverride->pVBoxNetFlt)
865	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
866	&& VALID_PTR(pSwitchPort)
867	)
868	{
869	INTNETSWDECISION enmDecision;
870	uint32_t cbHdrs = skb_headlen(pSkb);
871	cbHdrs = RT_MAX(cbHdrs, sizeof(abHdrBuf));
872
873	/** @todo consider reference counting, etc. */
874	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
875	if (enmDecision == INTNETSWDECISION_INTNET)
876	{
877	dev_kfree_skb(pSkb);
878	pOverride->cFiltered++;
879	return NETDEV_TX_OK;
880	}
881	}
882
883	return pOverride->pOrgOps->ndo_start_xmit(pSkb, pDev);
884	}
885
886	/**
887	* Hooks the device ndo_start_xmit operation of the device.
888	*
889	* @param pThis The net filter instance.
890	* @param pDev The net device.
891	*/
892	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
893	{
894	PVBOXNETDEVICEOPSOVERRIDE pOverride;
895	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
896
897	pOverride = RTMemAlloc(sizeof(*pOverride));
898	if (!pOverride)
899	return;
900	pOverride->pOrgOps = pDev->netdev_ops;
901	pOverride->Ops = *pDev->netdev_ops;
902	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
903	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
904	pOverride->cTotal = 0;
905	pOverride->cFiltered = 0;
906	pOverride->pVBoxNetFlt = pThis;
907
908	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp); /* (this isn't necessary, but so what) */
909	ASMAtomicWritePtr((void * volatile *)&pDev->netdev_ops, pOverride);
910	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
911	}
912
913	/**
914	* Undos what vboxNetFltLinuxHookDev did.
915	*
916	* @param pThis The net filter instance.
917	* @param pDev The net device. Can be NULL, in which case
918	* we'll try retrieve it from @a pThis.
919	*/
920	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
921	{
922	PVBOXNETDEVICEOPSOVERRIDE pOverride;
923	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
924
925	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
926	if (!pDev)
927	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
928	if (VALID_PTR(pDev))
929	{
930	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->netdev_ops;
931	if ( VALID_PTR(pOverride)
932	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
933	&& VALID_PTR(pOverride->pOrgOps)
934	)
935	{
936	ASMAtomicWritePtr((void * volatile *)&pDev->netdev_ops, pOverride->pOrgOps);
937	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
938	}
939	else
940	pOverride = NULL;
941	}
942	else
943	pOverride = NULL;
944	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
945
946	if (pOverride)
947	{
948	printk("vboxnetflt: dropped %llu out of %llu packets\n", pOverride->cFiltered, pOverride->cTotal);
949	RTMemFree(pOverride);
950	}
951	}
952
953	#endif /* VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT */
954
955
956	/**
957	* Reads and retains the host interface handle.
958	*
959	* @returns The handle, NULL if detached.
960	* @param pThis
961	*/
962	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
963	{
964	#if 0
965	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
966	struct net_device *pDev = NULL;
967
968	Log(("vboxNetFltLinuxRetainNetDev\n"));
969	/*
970	* Be careful here to avoid problems racing the detached callback.
971	*/
972	RTSpinlockAcquire(pThis->hSpinlock, &Tmp);
973	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
974	{
975	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
976	if (pDev)
977	{
978	dev_hold(pDev);
979	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
980	}
981	}
982	RTSpinlockRelease(pThis->hSpinlock, &Tmp);
983
984	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
985	return pDev;
986	#else
987	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
988	#endif
989	}
990
991
992	/**
993	* Release the host interface handle previously retained
994	* by vboxNetFltLinuxRetainNetDev.
995	*
996	* @param pThis The instance.
997	* @param pDev The vboxNetFltLinuxRetainNetDev
998	* return value, NULL is fine.
999	*/
1000	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
1001	{
1002	#if 0
1003	Log(("vboxNetFltLinuxReleaseNetDev\n"));
1004	NOREF(pThis);
1005	if (pDev)
1006	{
1007	dev_put(pDev);
1008	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1009	}
1010	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
1011	#endif
1012	}
1013
1014	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
1015	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
1016
1017	/**
1018	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
1019	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
1020	*
1021	* @returns true / false accordingly.
1022	* @param pBuf The sk_buff.
1023	*/
1024	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
1025	{
1026	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
1027	}
1028
1029
1030	/**
1031	* Internal worker that create a linux sk_buff for a
1032	* (scatter/)gather list.
1033	*
1034	* @returns Pointer to the sk_buff.
1035	* @param pThis The instance.
1036	* @param pSG The (scatter/)gather list.
1037	* @param fDstWire Set if the destination is the wire.
1038	*/
1039	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
1040	{
1041	struct sk_buff *pPkt;
1042	struct net_device *pDev;
1043	unsigned fGsoType = 0;
1044
1045	if (pSG->cbTotal == 0)
1046	{
1047	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
1048	return NULL;
1049	}
1050
1051	/** @todo We should use fragments mapping the SG buffers with large packets.
1052	* 256 bytes seems to be the a threshold used a lot for this. It
1053	* requires some nasty work on the intnet side though... */
1054	/*
1055	* Allocate a packet and copy over the data.
1056	*/
1057	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1058	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
1059	if (RT_UNLIKELY(!pPkt))
1060	{
1061	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
1062	pSG->pvUserData = NULL;
1063	return NULL;
1064	}
1065	pPkt->dev = pDev;
1066	pPkt->ip_summed = CHECKSUM_NONE;
1067
1068	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
1069	skb_reserve(pPkt, NET_IP_ALIGN);
1070
1071	/* Copy the segments. */
1072	skb_put(pPkt, pSG->cbTotal);
1073	IntNetSgRead(pSG, pPkt->data);
1074
1075	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1076	/*
1077	* Setup GSO if used by this packet.
1078	*/
1079	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
1080	{
1081	default:
1082	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
1083	/* fall thru */
1084	case PDMNETWORKGSOTYPE_INVALID:
1085	fGsoType = 0;
1086	break;
1087	case PDMNETWORKGSOTYPE_IPV4_TCP:
1088	fGsoType = SKB_GSO_TCPV4;
1089	break;
1090	case PDMNETWORKGSOTYPE_IPV4_UDP:
1091	fGsoType = SKB_GSO_UDP;
1092	break;
1093	case PDMNETWORKGSOTYPE_IPV6_TCP:
1094	fGsoType = SKB_GSO_TCPV6;
1095	break;
1096	}
1097	if (fGsoType)
1098	{
1099	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
1100
1101	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
1102	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
1103	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
1104
1105	if (fDstWire)
1106	{
1107	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrs);
1108	pPkt->ip_summed = CHECKSUM_PARTIAL;
1109	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1110	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
1111	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
1112	pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
1113	else
1114	pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
1115	# else
1116	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
1117	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
1118	pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
1119	else
1120	pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
1121	# endif
1122	}
1123	else
1124	{
1125	pPkt->ip_summed = CHECKSUM_UNNECESSARY;
1126	pPkt->csum = 0;
1127	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, false /fPayloadChecksum/);
1128	}
1129	}
1130	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
1131
1132	/*
1133	* Finish up the socket buffer.
1134	*/
1135	pPkt->protocol = eth_type_trans(pPkt, pDev);
1136	if (fDstWire)
1137	{
1138	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
1139
1140	/* Restore ethernet header back. */
1141	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
1142	VBOX_SKB_RESET_MAC_HDR(pPkt);
1143	}
1144	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
1145
1146	return pPkt;
1147	}
1148
1149
1150	/**
1151	* Initializes a SG list from an sk_buff.
1152	*
1153	* @returns Number of segments.
1154	* @param pThis The instance.
1155	* @param pBuf The sk_buff.
1156	* @param pSG The SG.
1157	* @param pvFrame The frame pointer, optional.
1158	* @param cSegs The number of segments allocated for the SG.
1159	* This should match the number in the mbuf exactly!
1160	* @param fSrc The source of the frame.
1161	* @param pGso Pointer to the GSO context if it's a GSO
1162	* internal network frame. NULL if regular frame.
1163	*/
1164	DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
1165	unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1166	{
1167	int i;
1168	NOREF(pThis);
1169
1170	Assert(!skb_shinfo(pBuf)->frag_list);
1171
1172	if (fSrc & INTNETTRUNKDIR_WIRE)
1173	{
1174	/*
1175	* The packet came from wire, ethernet header was removed by device driver.
1176	* Restore it.
1177	*/
1178	skb_push(pBuf, ETH_HLEN);
1179	}
1180
1181	if (!pGsoCtx)
1182	IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /cSegsUsed/);
1183	else
1184	IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /cSegsUsed/, pGsoCtx);
1185
1186	#ifdef VBOXNETFLT_SG_SUPPORT
1187	pSG->aSegs[0].cb = skb_headlen(pBuf);
1188	pSG->aSegs[0].pv = pBuf->data;
1189	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1190
1191	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1192	{
1193	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
1194	pSG->aSegs[i+1].cb = pFrag->size;
1195	pSG->aSegs[i+1].pv = kmap(pFrag->page);
1196	printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
1197	pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
1198	}
1199	++i;
1200
1201	#else
1202	pSG->aSegs[0].cb = pBuf->len;
1203	pSG->aSegs[0].pv = pBuf->data;
1204	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1205	i = 1;
1206	#endif
1207
1208	pSG->cSegsUsed = i;
1209
1210	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1211	/*
1212	* Add a trailer if the frame is too small.
1213	*
1214	* Since we're getting to the packet before it is framed, it has not
1215	* yet been padded. The current solution is to add a segment pointing
1216	* to a buffer containing all zeros and pray that works for all frames...
1217	*/
1218	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
1219	{
1220	static uint8_t const s_abZero[128] = {0};
1221
1222	AssertReturnVoid(i < cSegs);
1223
1224	pSG->aSegs[i].Phys = NIL_RTHCPHYS;
1225	pSG->aSegs[i].pv = (void *)&s_abZero[0];
1226	pSG->aSegs[i].cb = 60 - pSG->cbTotal;
1227	pSG->cbTotal = 60;
1228	pSG->cSegsUsed++;
1229	Assert(i + 1 <= pSG->cSegsAlloc)
1230	}
1231	#endif
1232
1233	Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
1234	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
1235	for (i = 0; i < pSG->cSegsUsed; i++)
1236	Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
1237	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1238	}
1239
1240	/**
1241	* Packet handler,
1242	*
1243	* @returns 0 or EJUSTRETURN.
1244	* @param pThis The instance.
1245	* @param pMBuf The mbuf.
1246	* @param pvFrame The start of the frame, optional.
1247	* @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
1248	* @param eProtocol The protocol.
1249	*/
1250	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
1251	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1252	struct net_device *pSkbDev,
1253	struct packet_type *pPacketType,
1254	struct net_device *pOrigDev)
1255	#else
1256	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1257	struct net_device *pSkbDev,
1258	struct packet_type *pPacketType)
1259	#endif
1260	{
1261	PVBOXNETFLTINS pThis;
1262	struct net_device *pDev;
1263	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1264	pBuf, pSkbDev, pPacketType));
1265	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1266	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",
1267	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));
1268	#else
1269	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",
1270	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));
1271	#endif
1272	/*
1273	* Drop it immediately?
1274	*/
1275	if (!pBuf)
1276	return 0;
1277
1278	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1279	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1280	if (pThis->u.s.pDev != pSkbDev)
1281	{
1282	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1283	return 0;
1284	}
1285
1286	Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1287	if (vboxNetFltLinuxSkBufIsOur(pBuf))
1288	{
1289	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1290	dev_kfree_skb(pBuf);
1291	return 0;
1292	}
1293
1294	#ifndef VBOXNETFLT_SG_SUPPORT
1295	{
1296	/*
1297	* Get rid of fragmented packets, they cause too much trouble.
1298	*/
1299	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1300	kfree_skb(pBuf);
1301	if (!pCopy)
1302	{
1303	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1304	return 0;
1305	}
1306	pBuf = pCopy;
1307	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1308	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",
1309	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));
1310	# else
1311	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",
1312	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));
1313	# endif
1314	}
1315	#endif
1316
1317	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1318	/* Forward it to the internal network. */
1319	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1320	#else
1321	/* Add the packet to transmit queue and schedule the bottom half. */
1322	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1323	schedule_work(&pThis->u.s.XmitTask);
1324	Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1325	&pThis->u.s.XmitTask, pBuf));
1326	#endif
1327
1328	/* It does not really matter what we return, it is ignored by the kernel. */
1329	return 0;
1330	}
1331
1332	/**
1333	* Calculate the number of INTNETSEG segments the socket buffer will need.
1334	*
1335	* @returns Segment count.
1336	* @param pBuf The socket buffer.
1337	*/
1338	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
1339	{
1340	#ifdef VBOXNETFLT_SG_SUPPORT
1341	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1342	#else
1343	unsigned cSegs = 1;
1344	#endif
1345	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1346	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1347	if (pBuf->len < 60)
1348	cSegs++;
1349	#endif
1350	return cSegs;
1351	}
1352
1353	/**
1354	* Destroy the intnet scatter / gather buffer created by
1355	* vboxNetFltLinuxSkBufToSG.
1356	*/
1357	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
1358	{
1359	#ifdef VBOXNETFLT_SG_SUPPORT
1360	int i;
1361
1362	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1363	{
1364	printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
1365	kunmap(pSG->aSegs[i+1].pv);
1366	}
1367	#endif
1368	NOREF(pSG);
1369	}
1370
1371	#ifdef LOG_ENABLED
1372	/**
1373	* Logging helper.
1374	*/
1375	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1376	{
1377	uint8_t pInt, pExt;
1378	static int iPacketNo = 1;
1379	iPacketNo += iIncrement;
1380	if (fEgress)
1381	{
1382	pExt = pSG->aSegs[0].pv;
1383	pInt = pExt + 6;
1384	}
1385	else
1386	{
1387	pInt = pSG->aSegs[0].pv;
1388	pExt = pInt + 6;
1389	}
1390	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1391	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1392	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1393	fEgress ? "-->" : "<--", pszWhere,
1394	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1395	pSG->cbTotal, iPacketNo));
1396	Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1397	}
1398	#else
1399	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1400	#endif
1401
1402	#ifdef VBOXNETFLT_WITH_GSO_RECV
1403
1404	/**
1405	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1406	* GSO socket buffer without having to segment it.
1407	*
1408	* @returns true on success, false if needs segmenting.
1409	* @param pThis The net filter instance.
1410	* @param pSkb The GSO socket buffer.
1411	* @param fSrc The source.
1412	* @param pGsoCtx Where to return the GSO context on success.
1413	*/
1414	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1415	PPDMNETWORKGSO pGsoCtx)
1416	{
1417	PDMNETWORKGSOTYPE enmGsoType;
1418	uint16_t uEtherType;
1419	unsigned int cbTransport;
1420	unsigned int offTransport;
1421	unsigned int cbTransportHdr;
1422	unsigned uProtocol;
1423	union
1424	{
1425	RTNETIPV4 IPv4;
1426	RTNETIPV6 IPv6;
1427	RTNETTCP Tcp;
1428	uint8_t ab[40];
1429	uint16_t au16[40/2];
1430	uint32_t au32[40/4];
1431	} Buf;
1432
1433	/*
1434	* Check the GSO properties of the socket buffer and make sure it fits.
1435	*/
1436	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1437	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP \| SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
1438	{
1439	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1440	return false;
1441	}
1442	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1443	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
1444	{
1445	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1446	return false;
1447	}
1448	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1449	{
1450	Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1451	return false;
1452	}
1453
1454	/*
1455	* skb_gso_segment does the following. Do we need to do it as well?
1456	*/
1457	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1458	skb_reset_mac_header(pSkb);
1459	pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1460	#else
1461	pSkb->mac.raw = pSkb->data;
1462	pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1463	#endif
1464
1465	/*
1466	* Switch on the ethertype.
1467	*/
1468	uEtherType = pSkb->protocol;
1469	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1470	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1471	{
1472	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1473	if (puEtherType)
1474	uEtherType = *puEtherType;
1475	}
1476	switch (uEtherType)
1477	{
1478	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1479	{
1480	unsigned int cbHdr;
1481	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1482	if (RT_UNLIKELY(!pIPv4))
1483	{
1484	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1485	return false;
1486	}
1487
1488	cbHdr = pIPv4->ip_hl * 4;
1489	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1490	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1491	\|\| cbHdr > cbTransport ))
1492	{
1493	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1494	return false;
1495	}
1496	cbTransport -= cbHdr;
1497	offTransport = pSkb->mac_len + cbHdr;
1498	uProtocol = pIPv4->ip_p;
1499	if (uProtocol == RTNETIPV4_PROT_TCP)
1500	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1501	else if (uProtocol == RTNETIPV4_PROT_UDP)
1502	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1503	else /** @todo IPv6: 4to6 tunneling */
1504	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1505	break;
1506	}
1507
1508	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1509	{
1510	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1511	if (RT_UNLIKELY(!pIPv6))
1512	{
1513	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1514	return false;
1515	}
1516
1517	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1518	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1519	uProtocol = pIPv6->ip6_nxt;
1520	/** @todo IPv6: Dig our way out of the other headers. */
1521	if (uProtocol == RTNETIPV4_PROT_TCP)
1522	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1523	else if (uProtocol == RTNETIPV4_PROT_UDP)
1524	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1525	else
1526	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1527	break;
1528	}
1529
1530	default:
1531	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1532	return false;
1533	}
1534
1535	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1536	{
1537	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1538	return false;
1539	}
1540
1541	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1542	\|\| offTransport + cbTransport > pSkb->len
1543	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1544	{
1545	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1546	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1547	return false;
1548	}
1549
1550	/*
1551	* Check the TCP/UDP bits.
1552	*/
1553	if (uProtocol == RTNETIPV4_PROT_TCP)
1554	{
1555	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1556	if (RT_UNLIKELY(!pTcp))
1557	{
1558	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1559	return false;
1560	}
1561
1562	cbTransportHdr = pTcp->th_off * 4;
1563	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1564	\|\| cbTransportHdr > cbTransport
1565	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1566	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1567	{
1568	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1569	return false;
1570	}
1571
1572	}
1573	else
1574	{
1575	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1576	cbTransportHdr = sizeof(RTNETUDP);
1577	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1578	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1579	{
1580	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1581	return false;
1582	}
1583	}
1584
1585	/*
1586	* We're good, init the GSO context.
1587	*/
1588	pGsoCtx->u8Type = enmGsoType;
1589	pGsoCtx->cbHdrs = offTransport + cbTransportHdr;
1590	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1591	pGsoCtx->offHdr1 = pSkb->mac_len;
1592	pGsoCtx->offHdr2 = offTransport;
1593	pGsoCtx->au8Unused[0] = 0;
1594	pGsoCtx->au8Unused[1] = 0;
1595
1596	return true;
1597	}
1598
1599	/**
1600	* Forward the socket buffer as a GSO internal network frame.
1601	*
1602	* @returns IPRT status code.
1603	* @param pThis The net filter instance.
1604	* @param pSkb The GSO socket buffer.
1605	* @param fSrc The source.
1606	* @param pGsoCtx Where to return the GSO context on success.
1607	*/
1608	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1609	{
1610	int rc;
1611	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1612	if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1613	{
1614	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1615	if (RT_LIKELY(pSG))
1616	{
1617	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1618
1619	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1620	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1621
1622	vboxNetFltLinuxDestroySG(pSG);
1623	rc = VINF_SUCCESS;
1624	}
1625	else
1626	{
1627	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1628	rc = VERR_NO_MEMORY;
1629	}
1630	}
1631	else
1632	{
1633	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1634	rc = VERR_INTERNAL_ERROR_3;
1635	}
1636
1637	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1638	dev_kfree_skb(pSkb);
1639	return rc;
1640	}
1641
1642	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1643
1644	/**
1645	* Worker for vboxNetFltLinuxForwardToIntNet.
1646	*
1647	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1648	* @param pThis The net filter instance.
1649	* @param pBuf The socket buffer.
1650	* @param fSrc The source.
1651	*/
1652	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1653	{
1654	int rc;
1655	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1656	if (cSegs <= MAX_SKB_FRAGS + 1)
1657	{
1658	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1659	if (RT_LIKELY(pSG))
1660	{
1661	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /pGsoCtx/);
1662
1663	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1664	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1665
1666	vboxNetFltLinuxDestroySG(pSG);
1667	rc = VINF_SUCCESS;
1668	}
1669	else
1670	{
1671	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1672	rc = VERR_NO_MEMORY;
1673	}
1674	}
1675	else
1676	{
1677	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1678	rc = VERR_INTERNAL_ERROR_3;
1679	}
1680
1681	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1682	dev_kfree_skb(pBuf);
1683	return rc;
1684	}
1685
1686	/**
1687	*
1688	* @param pBuf The socket buffer. This is consumed by this function.
1689	*/
1690	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1691	{
1692	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1693
1694	#ifdef VBOXNETFLT_WITH_GSO
1695	if (skb_is_gso(pBuf))
1696	{
1697	PDMNETWORKGSO GsoCtx;
1698	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",
1699	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));
1700	# ifdef VBOXNETFLT_WITH_GSO_RECV
1701	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1702	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1703	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1704	else
1705	# endif
1706	{
1707	/* Need to segment the packet */
1708	struct sk_buff *pNext;
1709	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1710	if (IS_ERR(pSegment))
1711	{
1712	dev_kfree_skb(pBuf);
1713	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1714	return;
1715	}
1716
1717	for (; pSegment; pSegment = pNext)
1718	{
1719	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",
1720	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));
1721	pNext = pSegment->next;
1722	pSegment->next = 0;
1723	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1724	}
1725	dev_kfree_skb(pBuf);
1726	}
1727	}
1728	else
1729	#endif /* VBOXNETFLT_WITH_GSO */
1730	{
1731	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1732	{
1733	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1734	/*
1735	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1736	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1737	* header length from the header itself and reconstruct 'h' pointer
1738	* to TCP (or whatever) header.
1739	*/
1740	unsigned char *tmp = pBuf->h.raw;
1741	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1742	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1743	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1744	if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1745	{
1746	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1747	dev_kfree_skb(pBuf);
1748	return;
1749	}
1750	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1751	/* Restore the original (wrong) pointer. */
1752	pBuf->h.raw = tmp;
1753	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1754	}
1755	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1756	}
1757	}
1758
1759	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1760	/**
1761	* Work queue handler that forwards the socket buffers queued by
1762	* vboxNetFltLinuxPacketHandler to the internal network.
1763	*
1764	* @param pWork The work queue.
1765	*/
1766	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1767	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1768	# else
1769	static void vboxNetFltLinuxXmitTask(void *pWork)
1770	# endif
1771	{
1772	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1773	struct sk_buff *pBuf;
1774
1775	Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1776
1777	/*
1778	* Active? Retain the instance and increment the busy counter.
1779	*/
1780	if (vboxNetFltTryRetainBusyActive(pThis))
1781	{
1782	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1783	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1784
1785	vboxNetFltRelease(pThis, true /* fBusy */);
1786	}
1787	else
1788	{
1789	/** @todo Shouldn't we just drop the packets here? There is little point in
1790	* making them accumulate when the VM is paused and it'll only waste
1791	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1792	* before start draining the packets (goes for the intnet ring buf
1793	* too)? */
1794	}
1795	}
1796	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1797
1798	/**
1799	* Reports the GSO capabilites of the hardware NIC.
1800	*
1801	* @param pThis The net filter instance. The caller hold a
1802	* reference to this.
1803	*/
1804	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1805	{
1806	#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1807	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1808	{
1809	struct net_device *pDev;
1810	PINTNETTRUNKSWPORT pSwitchPort;
1811	unsigned int fFeatures;
1812	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1813
1814	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1815
1816	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1817	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1818	if (pDev)
1819	fFeatures = pDev->features;
1820	else
1821	fFeatures = 0;
1822
1823	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1824
1825	if (pThis->pSwitchPort)
1826	{
1827	/* Set/update the GSO capabilities of the NIC. */
1828	uint32_t fGsoCapabilites = 0;
1829	if (fFeatures & NETIF_F_TSO)
1830	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1831	if (fFeatures & NETIF_F_TSO6)
1832	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1833	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1834	if (fFeatures & NETIF_F_UFO)
1835	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1836	if (fFeatures & NETIF_F_UFO)
1837	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1838	# endif
1839	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1840	}
1841
1842	vboxNetFltRelease(pThis, true /fBusy/);
1843	}
1844	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1845	}
1846
1847	/**
1848	* Helper that determins whether the host (ignoreing us) is operating the
1849	* interface in promiscuous mode or not.
1850	*/
1851	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1852	{
1853	bool fRc = false;
1854	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1855	if (pDev)
1856	{
1857	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1858	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1859	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1860	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1861	}
1862	return fRc;
1863	}
1864
1865	/**
1866	* Internal worker for vboxNetFltLinuxNotifierCallback.
1867	*
1868	* @returns VBox status code.
1869	* @param pThis The instance.
1870	* @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1871	* flood the release log.
1872	*/
1873	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1874	{
1875	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1876	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1877
1878	/*
1879	* Retain and store the device.
1880	*/
1881	dev_hold(pDev);
1882
1883	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1884	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1885	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1886
1887	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1888	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n", pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1889
1890	/* Get the mac address while we still have a valid net_device reference. */
1891	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1892
1893	/*
1894	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1895	*/
1896	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1897	pThis->u.s.PacketType.dev = pDev;
1898	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1899	dev_add_pack(&pThis->u.s.PacketType);
1900
1901	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1902	vboxNetFltLinuxHookDev(pThis, pDev);
1903	#endif
1904	#ifdef VBOXNETFLT_WITH_QDISC
1905	vboxNetFltLinuxQdiscInstall(pThis, pDev);
1906	#endif /* VBOXNETFLT_WITH_QDISC */
1907
1908	/*
1909	* Set indicators that require the spinlock. Be abit paranoid about racing
1910	* the device notification handle.
1911	*/
1912	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1913	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1914	if (pDev)
1915	{
1916	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1917	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1918	pDev = NULL; /* don't dereference it */
1919	}
1920	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1921	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1922
1923	/*
1924	* If the above succeeded report GSO capabilites, if not undo and
1925	* release the device.
1926	*/
1927	if (!pDev)
1928	{
1929	Assert(pThis->pSwitchPort);
1930	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1931	{
1932	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1933	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1934	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1935	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1936	vboxNetFltRelease(pThis, true /fBusy/);
1937	}
1938	}
1939	else
1940	{
1941	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1942	vboxNetFltLinuxUnhookDev(pThis, pDev);
1943	#endif
1944	#ifdef VBOXNETFLT_WITH_QDISC
1945	vboxNetFltLinuxQdiscRemove(pThis, pDev);
1946	#endif /* VBOXNETFLT_WITH_QDISC */
1947	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1948	ASMAtomicUoWritePtr(&pThis->u.s.pDev, NULL);
1949	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1950	dev_put(pDev);
1951	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1952	}
1953
1954	LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1955	return VINF_SUCCESS;
1956	}
1957
1958
1959	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1960	{
1961	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1962
1963	Assert(!pThis->fDisconnectedFromHost);
1964
1965	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1966	vboxNetFltLinuxUnhookDev(pThis, pDev);
1967	#endif
1968	#ifdef VBOXNETFLT_WITH_QDISC
1969	vboxNetFltLinuxQdiscRemove(pThis, pDev);
1970	#endif /* VBOXNETFLT_WITH_QDISC */
1971
1972	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1973	ASMAtomicWriteBool(&pThis->u.s.fRegistered, false);
1974	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1975	ASMAtomicUoWritePtr(&pThis->u.s.pDev, NULL);
1976	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1977
1978	dev_remove_pack(&pThis->u.s.PacketType);
1979	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1980	skb_queue_purge(&pThis->u.s.XmitQueue);
1981	#endif
1982	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
1983	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1984	dev_put(pDev);
1985
1986	return NOTIFY_OK;
1987	}
1988
1989	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1990	{
1991	/* Check if we are not suspended and promiscuous mode has not been set. */
1992	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1993	&& !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1994	{
1995	/* Note that there is no need for locking as the kernel got hold of the lock already. */
1996	dev_set_promiscuity(pDev, 1);
1997	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1998	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1999	}
2000	else
2001	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2002	return NOTIFY_OK;
2003	}
2004
2005	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
2006	{
2007	/* Undo promiscuous mode if we has set it. */
2008	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
2009	{
2010	/* Note that there is no need for locking as the kernel got hold of the lock already. */
2011	dev_set_promiscuity(pDev, -1);
2012	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
2013	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2014	}
2015	else
2016	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2017	return NOTIFY_OK;
2018	}
2019
2020	#ifdef LOG_ENABLED
2021	/** Stringify the NETDEV_XXX constants. */
2022	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2023	{
2024	const char *pszEvent = "NETDRV_<unknown>";
2025	switch (ulEventType)
2026	{
2027	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2028	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2029	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2030	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2031	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2032	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2033	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2034	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2035	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2036	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2037	# ifdef NETDEV_FEAT_CHANGE
2038	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2039	# endif
2040	}
2041	return pszEvent;
2042	}
2043	#endif /* LOG_ENABLED */
2044
2045	/**
2046	* Callback for listening to netdevice events.
2047	*
2048	* This works the rediscovery, clean up on unregistration, promiscuity on
2049	* up/down, and GSO feature changes from ethtool.
2050	*
2051	* @returns NOTIFY_OK
2052	* @param self Pointer to our notifier registration block.
2053	* @param ulEventType The event.
2054	* @param ptr Event specific, but it is usually the device it
2055	* relates to.
2056	*/
2057	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2058
2059	{
2060	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2061	struct net_device pDev = (struct net_device )ptr;
2062	int rc = NOTIFY_OK;
2063
2064	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2065	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
2066	if ( ulEventType == NETDEV_REGISTER
2067	&& !strcmp(pDev->name, pThis->szName))
2068	{
2069	vboxNetFltLinuxAttachToInterface(pThis, pDev);
2070	}
2071	else
2072	{
2073	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2074	if (pDev == ptr)
2075	{
2076	switch (ulEventType)
2077	{
2078	case NETDEV_UNREGISTER:
2079	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2080	break;
2081	case NETDEV_UP:
2082	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2083	break;
2084	case NETDEV_GOING_DOWN:
2085	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2086	break;
2087	case NETDEV_CHANGENAME:
2088	break;
2089	#ifdef NETDEV_FEAT_CHANGE
2090	case NETDEV_FEAT_CHANGE:
2091	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2092	break;
2093	#endif
2094	}
2095	}
2096	}
2097
2098	return rc;
2099	}
2100
2101	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2102	{
2103	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2104	}
2105
2106	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2107	{
2108	struct net_device * pDev;
2109	int err;
2110	int rc = VINF_SUCCESS;
2111	NOREF(pvIfData);
2112
2113	LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2114
2115	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2116	if (pDev)
2117	{
2118	/*
2119	* Create a sk_buff for the gather list and push it onto the wire.
2120	*/
2121	if (fDst & INTNETTRUNKDIR_WIRE)
2122	{
2123	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2124	if (pBuf)
2125	{
2126	vboxNetFltDumpPacket(pSG, true, "wire", 1);
2127	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2128	Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2129	err = dev_queue_xmit(pBuf);
2130	if (err)
2131	rc = RTErrConvertFromErrno(err);
2132	}
2133	else
2134	rc = VERR_NO_MEMORY;
2135	}
2136
2137	/*
2138	* Create a sk_buff for the gather list and push it onto the host stack.
2139	*/
2140	if (fDst & INTNETTRUNKDIR_HOST)
2141	{
2142	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2143	if (pBuf)
2144	{
2145	vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2146	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2147	Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2148	err = netif_rx_ni(pBuf);
2149	if (err)
2150	rc = RTErrConvertFromErrno(err);
2151	}
2152	else
2153	rc = VERR_NO_MEMORY;
2154	}
2155
2156	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2157	}
2158
2159	return rc;
2160	}
2161
2162
2163	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2164	{
2165	struct net_device * pDev;
2166
2167	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
2168	pThis, pThis->szName, fActive?"true":"false",
2169	pThis->fDisablePromiscuous?"true":"false"));
2170
2171	if (pThis->fDisablePromiscuous)
2172	return;
2173
2174	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2175	if (pDev)
2176	{
2177	/*
2178	* This api is a bit weird, the best reference is the code.
2179	*
2180	* Also, we have a bit or race conditions wrt the maintance of
2181	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2182	*/
2183	#ifdef LOG_ENABLED
2184	u_int16_t fIf;
2185	unsigned const cPromiscBefore = pDev->promiscuity;
2186	#endif
2187	if (fActive)
2188	{
2189	Assert(!pThis->u.s.fPromiscuousSet);
2190
2191	rtnl_lock();
2192	dev_set_promiscuity(pDev, 1);
2193	rtnl_unlock();
2194	pThis->u.s.fPromiscuousSet = true;
2195	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2196	}
2197	else
2198	{
2199	if (pThis->u.s.fPromiscuousSet)
2200	{
2201	rtnl_lock();
2202	dev_set_promiscuity(pDev, -1);
2203	rtnl_unlock();
2204	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2205	}
2206	pThis->u.s.fPromiscuousSet = false;
2207
2208	#ifdef LOG_ENABLED
2209	fIf = dev_get_flags(pDev);
2210	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2211	#endif
2212	}
2213
2214	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2215	}
2216	}
2217
2218
2219	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2220	{
2221	/* Nothing to do here. */
2222	return VINF_SUCCESS;
2223	}
2224
2225
2226	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2227	{
2228	/*
2229	* Report the GSO capabilities of the host and device (if connected).
2230	* Note! No need to mark ourselves busy here.
2231	*/
2232	/** @todo duplicate work here now? Attach */
2233	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2234	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2235	0
2236	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2237	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2238	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2239	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2240	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2241	# endif
2242	, INTNETTRUNKDIR_HOST);
2243
2244	#endif
2245	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2246
2247	return VINF_SUCCESS;
2248	}
2249
2250
2251	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2252	{
2253	struct net_device *pDev;
2254	bool fRegistered;
2255	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
2256
2257	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2258	vboxNetFltLinuxUnhookDev(pThis, NULL);
2259	#endif
2260	#ifdef VBOXNETFLT_WITH_QDISC
2261	vboxNetFltLinuxQdiscRemove(pThis, NULL);
2262	#endif /* VBOXNETFLT_WITH_QDISC */
2263
2264	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2265	* unlikely, but none the less). Since it doesn't actually update the
2266	* state (just reads it), it is likely to panic in some interesting
2267	* ways. */
2268
2269	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2270	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2271	fRegistered = ASMAtomicUoReadBool(&pThis->u.s.fRegistered);
2272	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2273
2274	if (fRegistered)
2275	{
2276	dev_remove_pack(&pThis->u.s.PacketType);
2277	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2278	skb_queue_purge(&pThis->u.s.XmitQueue);
2279	#endif
2280	Log(("vboxNetFltOsDeleteInstance: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
2281	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
2282	dev_put(pDev);
2283	}
2284	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2285	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2286	module_put(THIS_MODULE);
2287	}
2288
2289
2290	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2291	{
2292	int err;
2293	NOREF(pvContext);
2294
2295	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2296	err = register_netdevice_notifier(&pThis->u.s.Notifier);
2297	if (err)
2298	return VERR_INTNET_FLT_IF_FAILED;
2299	if (!pThis->u.s.fRegistered)
2300	{
2301	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2302	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2303	return VERR_INTNET_FLT_IF_NOT_FOUND;
2304	}
2305
2306	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2307	if ( pThis->fDisconnectedFromHost
2308	\|\| !try_module_get(THIS_MODULE))
2309	return VERR_INTNET_FLT_IF_FAILED;
2310
2311	return VINF_SUCCESS;
2312	}
2313
2314	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2315	{
2316	/*
2317	* Init the linux specific members.
2318	*/
2319	pThis->u.s.pDev = NULL;
2320	pThis->u.s.fRegistered = false;
2321	pThis->u.s.fPromiscuousSet = false;
2322	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2323	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2324	skb_queue_head_init(&pThis->u.s.XmitQueue);
2325	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2326	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2327	# else
2328	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2329	# endif
2330	#endif
2331
2332	return VINF_SUCCESS;
2333	}
2334
2335
2336	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2337	{
2338	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2339	}
2340
2341
2342	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
2343	{
2344	/* Nothing to do */
2345	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2346	return VINF_SUCCESS;
2347	}
2348
2349
2350	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2351	{
2352	/* Nothing to do */
2353	NOREF(pThis); NOREF(pvIfData);
2354	return VINF_SUCCESS;
2355	}
2356

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 30111

Download in other formats: