VBoxNetFlt-linux.c@ 31675

Last change on this file since 31675 was 31675, checked in by vboxsync, 14 years ago
vboxnetflt: pre 2.6.18 kernel compilation fixes
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 82.3 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 31675 2010-08-13 18:38:01Z 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	* Comment out the following line to disable 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	/* Versions prior 2.6.10 use stats for both bstats and qstats */
92	# define bstats stats
93	# define qstats stats
94	# endif
95	#endif
96
97	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 13)
98	static inline int qdisc_drop(struct sk_buff skb, struct Qdisc sch)
99	{
100	kfree_skb(skb);
101	sch->stats.drops++;
102
103	return NET_XMIT_DROP;
104	}
105	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 13) */
106
107	#ifndef NET_IP_ALIGN
108	# define NET_IP_ALIGN 2
109	#endif
110
111	#if 0
112	/** Create scatter / gather segments for fragments. When not used, we will
113	* linearize the socket buffer before creating the internal networking SG. */
114	# define VBOXNETFLT_SG_SUPPORT 1
115	#endif
116
117	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
118	/** Indicates that the linux kernel may send us GSO frames. */
119	# define VBOXNETFLT_WITH_GSO 1
120
121	/** This enables or disables the transmitting of GSO frame from the internal
122	* network and to the host. */
123	# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
124
125	# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
126	/** This enables or disables the transmitting of GSO frame from the internal
127	* network and to the wire. */
128	# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
129	# endif
130
131	/** This enables or disables the forwarding/flooding of GSO frame from the host
132	* to the internal network. */
133	# define VBOXNETFLT_WITH_GSO_RECV 1
134
135	#endif
136
137	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
138	/** This enables or disables handling of GSO frames coming from the wire (GRO). */
139	# define VBOXNETFLT_WITH_GRO 1
140	#endif
141	/*
142	* GRO support was backported to RHEL 5.4
143	*/
144	#ifdef RHEL_RELEASE_CODE
145	# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
146	# define VBOXNETFLT_WITH_GRO 1
147	# endif
148	#endif
149
150	/*******************************************************************************
151	* Internal Functions *
152	*******************************************************************************/
153	static int VBoxNetFltLinuxInit(void);
154	static void VBoxNetFltLinuxUnload(void);
155	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
156
157
158	/*******************************************************************************
159	* Global Variables *
160	*******************************************************************************/
161	/**
162	* The (common) global data.
163	*/
164	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
165
166	module_init(VBoxNetFltLinuxInit);
167	module_exit(VBoxNetFltLinuxUnload);
168
169	MODULE_AUTHOR(VBOX_VENDOR);
170	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
171	MODULE_LICENSE("GPL");
172	#ifdef MODULE_VERSION
173	MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
174	#endif
175
176
177	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
178	unsigned dev_get_flags(const struct net_device *dev)
179	{
180	unsigned flags;
181
182	flags = (dev->flags & ~(IFF_PROMISC \|
183	IFF_ALLMULTI \|
184	IFF_RUNNING)) \|
185	(dev->gflags & (IFF_PROMISC \|
186	IFF_ALLMULTI));
187
188	if (netif_running(dev) && netif_carrier_ok(dev))
189	flags \|= IFF_RUNNING;
190
191	return flags;
192	}
193	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
194
195
196	#ifdef VBOXNETFLT_WITH_QDISC
197	//#define QDISC_LOG(x) printk x
198	#define QDISC_LOG(x) do { } while (0)
199
200	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
201	#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops)
202	#elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
203	#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops, parent)
204	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
205	#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, queue, ops, parent)
206	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
207
208	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
209	#define qdisc_dev(qdisc) (qdisc->dev)
210	#define qdisc_pkt_len(skb) (skb->len)
211	#define QDISC_GET(dev) (dev->qdisc_sleeping)
212	#else
213	#define QDISC_GET(dev) (netdev_get_tx_queue(dev, 0)->qdisc_sleeping)
214	#endif
215
216	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
217	#define QDISC_SAVED_NUM(dev) 1
218	#elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
219	#define QDISC_SAVED_NUM(dev) dev->num_tx_queues
220	#else
221	#define QDISC_SAVED_NUM(dev) dev->num_tx_queues+1
222	#endif
223
224	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
225	#define QDISC_IS_BUSY(dev, qdisc) test_bit(__LINK_STATE_SCHED, &dev->state)
226	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
227	#define QDISC_IS_BUSY(dev, qdisc) (test_bit(__QDISC_STATE_RUNNING, &qdisc->state) \|\| \
228	test_bit(__QDISC_STATE_SCHED, &qdisc->state))
229	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
230
231	struct VBoxNetQDiscPriv
232	{
233	/** Pointer to the single child qdisc. */
234	struct Qdisc *pChild;
235	/*
236	* Technically it is possible to have different qdiscs for different TX
237	* queues so we have to save them all.
238	*/
239	/** Pointer to the array of saved qdiscs. */
240	struct Qdisc **ppSaved;
241	/** Pointer to the net filter instance. */
242	PVBOXNETFLTINS pVBoxNetFlt;
243	};
244	typedef struct VBoxNetQDiscPriv *PVBOXNETQDISCPRIV;
245
246	//#define VBOXNETFLT_QDISC_ENQUEUE
247	static int vboxNetFltQdiscEnqueue(struct sk_buff skb, struct Qdisc sch)
248	{
249	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
250	int rc;
251
252	#ifdef VBOXNETFLT_QDISC_ENQUEUE
253	if (VALID_PTR(pPriv->pVBoxNetFlt))
254	{
255	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
256	PCRTNETETHERHDR pEtherHdr;
257	PINTNETTRUNKSWPORT pSwitchPort;
258	uint32_t cbHdrs = skb_headlen(skb);
259
260	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
261	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(skb, 0, cbHdrs, &abHdrBuf[0]);
262	if ( pEtherHdr
263	&& (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) != NULL
264	&& VALID_PTR(pSwitchPort)
265	&& cbHdrs >= 6)
266	{
267	/** @todo consider reference counting, etc. */
268	INTNETSWDECISION enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
269	if (enmDecision == INTNETSWDECISION_INTNET)
270	{
271	struct sk_buff *pBuf = skb_copy(skb, GFP_ATOMIC);
272	pBuf->pkt_type = PACKET_OUTGOING;
273	vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
274	qdisc_drop(skb, sch);
275	++sch->bstats.packets;
276	sch->bstats.bytes += qdisc_pkt_len(skb);
277	return NET_XMIT_SUCCESS;
278	}
279	}
280	}
281	#endif /* VBOXNETFLT_QDISC_ENQUEUE */
282	rc = pPriv->pChild->enqueue(skb, pPriv->pChild);
283	if (rc == NET_XMIT_SUCCESS)
284	{
285	++sch->q.qlen;
286	++sch->bstats.packets;
287	sch->bstats.bytes += qdisc_pkt_len(skb);
288	}
289	else
290	++sch->qstats.drops;
291	return rc;
292	}
293
294	static struct sk_buff vboxNetFltQdiscDequeue(struct Qdisc sch)
295	{
296	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
297	#ifdef VBOXNETFLT_QDISC_ENQUEUE
298	--sch->q.qlen;
299	return pPriv->pChild->dequeue(pPriv->pChild);
300	#else /* VBOXNETFLT_QDISC_ENQUEUE */
301	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
302	PCRTNETETHERHDR pEtherHdr;
303	PINTNETTRUNKSWPORT pSwitchPort;
304	struct sk_buff *pSkb;
305
306	QDISC_LOG(("vboxNetFltDequeue: Enter pThis=%p\n", pPriv->pVBoxNetFlt));
307
308	while ((pSkb = pPriv->pChild->dequeue(pPriv->pChild)) != NULL)
309	{
310	struct sk_buff *pBuf;
311	INTNETSWDECISION enmDecision;
312	uint32_t cbHdrs;
313
314	--sch->q.qlen;
315
316	if (!VALID_PTR(pPriv->pVBoxNetFlt))
317	break;
318
319	cbHdrs = skb_headlen(pSkb);
320	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
321	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
322	if ( !pEtherHdr
323	\|\| (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) == NULL
324	\|\| !VALID_PTR(pSwitchPort)
325	\|\| cbHdrs < 6)
326	break;
327
328	/** @todo consider reference counting, etc. */
329	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
330	if (enmDecision != INTNETSWDECISION_INTNET)
331	break;
332
333	pBuf = skb_copy(pSkb, GFP_ATOMIC);
334	pBuf->pkt_type = PACKET_OUTGOING;
335	QDISC_LOG(("vboxNetFltDequeue: pThis=%p\n", pPriv->pVBoxNetFlt));
336	vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
337	qdisc_drop(pSkb, sch);
338	QDISC_LOG(("VBoxNetFlt: Packet for %02x:%02x:%02x:%02x:%02x:%02x dropped\n",
339	pSkb->data[0], pSkb->data[1], pSkb->data[2],
340	pSkb->data[3], pSkb->data[4], pSkb->data[5]));
341	}
342
343	return pSkb;
344	#endif /* VBOXNETFLT_QDISC_ENQUEUE */
345	}
346
347	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
348	static int vboxNetFltQdiscRequeue(struct sk_buff skb, struct Qdisc sch)
349	{
350	int rc;
351	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
352
353	rc = pPriv->pChild->ops->requeue(skb, pPriv->pChild);
354	if (rc == 0)
355	{
356	sch->q.qlen++;
357	sch->qstats.requeues++;
358	}
359
360	return rc;
361	}
362	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
363
364	static unsigned int vboxNetFltQdiscDrop(struct Qdisc *sch)
365	{
366	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
367	unsigned int cbLen;
368
369	if (pPriv->pChild->ops->drop)
370	{
371	cbLen = pPriv->pChild->ops->drop(pPriv->pChild);
372	if (cbLen != 0)
373	{
374	++sch->qstats.drops;
375	--sch->q.qlen;
376	return cbLen;
377	}
378	}
379
380	return 0;
381	}
382
383	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
384	static int vboxNetFltQdiscInit(struct Qdisc sch, struct rtattr opt)
385	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
386	static int vboxNetFltQdiscInit(struct Qdisc sch, struct nlattr opt)
387	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
388	{
389	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
390	struct net_device *pDev = qdisc_dev(sch);
391
392	pPriv->pVBoxNetFlt = NULL;
393
394	pPriv->ppSaved = kcalloc(QDISC_SAVED_NUM(pDev), sizeof(pPriv->ppSaved[0]),
395	GFP_KERNEL);
396	if (!pPriv->ppSaved)
397	return -ENOMEM;
398
399	pPriv->pChild = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
400	&pfifo_qdisc_ops,
401	TC_H_MAKE(TC_H_MAJ(sch->handle),
402	TC_H_MIN(1)));
403	if (!pPriv->pChild)
404	{
405	kfree(pPriv->ppSaved);
406	pPriv->ppSaved = NULL;
407	return -ENOMEM;
408	}
409
410	return 0;
411	}
412
413	static void vboxNetFltQdiscReset(struct Qdisc *sch)
414	{
415	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
416
417	qdisc_reset(pPriv->pChild);
418	sch->q.qlen = 0;
419	sch->qstats.backlog = 0;
420	}
421
422	static void vboxNetFltQdiscDestroy(struct Qdisc* sch)
423	{
424	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
425	struct net_device *pDev = qdisc_dev(sch);
426
427	qdisc_destroy(pPriv->pChild);
428	pPriv->pChild = NULL;
429
430	if (pPriv->ppSaved)
431	{
432	int i;
433	for (i = 0; i < QDISC_SAVED_NUM(pDev); i++)
434	if (pPriv->ppSaved[i])
435	qdisc_destroy(pPriv->ppSaved[i]);
436	kfree(pPriv->ppSaved);
437	pPriv->ppSaved = NULL;
438	}
439	}
440
441	static int vboxNetFltClassGraft(struct Qdisc sch, unsigned long arg, struct Qdisc pNew,
442	struct Qdisc **ppOld)
443	{
444	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
445
446	if (pNew == NULL)
447	pNew = &noop_qdisc;
448
449	sch_tree_lock(sch);
450	*ppOld = pPriv->pChild;
451	pPriv->pChild = pNew;
452	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
453	sch->q.qlen = 0;
454	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
455	qdisc_tree_decrease_qlen(ppOld, (ppOld)->q.qlen);
456	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
457	qdisc_reset(*ppOld);
458	sch_tree_unlock(sch);
459
460	return 0;
461	}
462
463	static struct Qdisc vboxNetFltClassLeaf(struct Qdisc sch, unsigned long arg)
464	{
465	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
466	return pPriv->pChild;
467	}
468
469	static unsigned long vboxNetFltClassGet(struct Qdisc *sch, u32 classid)
470	{
471	return 1;
472	}
473
474	static void vboxNetFltClassPut(struct Qdisc *sch, unsigned long arg)
475	{
476	}
477
478	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
479	static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
480	struct rtattr *tca, unsigned long arg)
481	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
482	static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
483	struct nlattr *tca, unsigned long arg)
484	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
485	{
486	return -ENOSYS;
487	}
488
489	static int vboxNetFltClassDelete(struct Qdisc *sch, unsigned long arg)
490	{
491	return -ENOSYS;
492	}
493
494	static void vboxNetFltClassWalk(struct Qdisc sch, struct qdisc_walker walker)
495	{
496	if (!walker->stop) {
497	if (walker->count >= walker->skip)
498	if (walker->fn(sch, 1, walker) < 0) {
499	walker->stop = 1;
500	return;
501	}
502	walker->count++;
503	}
504	}
505
506	static struct tcf_proto *vboxNetFltClassFindTcf(struct Qdisc sch, unsigned long cl)
507	{
508	return NULL;
509	}
510
511	static int vboxNetFltClassDump(struct Qdisc *sch, unsigned long cl,
512	struct sk_buff skb, struct tcmsg tcm)
513	{
514	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
515
516	if (cl != 1)
517	return -ENOENT;
518
519	tcm->tcm_handle \|= TC_H_MIN(1);
520	tcm->tcm_info = pPriv->pChild->handle;
521
522	return 0;
523	}
524
525
526	static struct Qdisc_class_ops g_VBoxNetFltClassOps =
527	{
528	.graft = vboxNetFltClassGraft,
529	.leaf = vboxNetFltClassLeaf,
530	.get = vboxNetFltClassGet,
531	.put = vboxNetFltClassPut,
532	.change = vboxNetFltClassChange,
533	.delete = vboxNetFltClassDelete,
534	.walk = vboxNetFltClassWalk,
535	.tcf_chain = vboxNetFltClassFindTcf,
536	.dump = vboxNetFltClassDump,
537	};
538
539
540	static struct Qdisc_ops g_VBoxNetFltQDiscOps = {
541	.cl_ops = &g_VBoxNetFltClassOps,
542	.id = "vboxnetflt",
543	.priv_size = sizeof(struct VBoxNetQDiscPriv),
544	.enqueue = vboxNetFltQdiscEnqueue,
545	.dequeue = vboxNetFltQdiscDequeue,
546	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
547	.requeue = vboxNetFltQdiscRequeue,
548	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
549	.peek = qdisc_peek_dequeued,
550	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
551	.drop = vboxNetFltQdiscDrop,
552	.init = vboxNetFltQdiscInit,
553	.reset = vboxNetFltQdiscReset,
554	.destroy = vboxNetFltQdiscDestroy,
555	.owner = THIS_MODULE
556	};
557
558	/*
559	* If our qdisc is already attached to the device (that means the user
560	* installed it from command line with 'tc' command) we simply update
561	* the pointer to vboxnetflt instance in qdisc's private structure.
562	* Otherwise we need to take some additional steps:
563	* - Create our qdisc;
564	* - Save all references to qdiscs;
565	* - Replace our child with the first qdisc reference;
566	* - Replace all references so they point to our qdisc.
567	*/
568	static void vboxNetFltLinuxQdiscInstall(PVBOXNETFLTINS pThis, struct net_device *pDev)
569	{
570	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
571	int i;
572	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
573	PVBOXNETQDISCPRIV pPriv;
574
575	struct Qdisc *pExisting = QDISC_GET(pDev);
576	if (strcmp(pExisting->ops->id, "vboxnetflt"))
577	{
578	/* The existing qdisc is different from ours, let's create new one. */
579	struct Qdisc *pNew = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
580	&g_VBoxNetFltQDiscOps, TC_H_ROOT);
581	if (!pNew)
582	return; // TODO: Error?
583
584	if (!try_module_get(THIS_MODULE))
585	{
586	/*
587	* This may cause a memory leak but calling qdisc_destroy()
588	* is not an option as it will call module_put().
589	*/
590	return;
591	}
592	pPriv = qdisc_priv(pNew);
593
594	qdisc_destroy(pPriv->pChild);
595	pPriv->pChild = QDISC_GET(pDev);
596	atomic_inc(&pPriv->pChild->refcnt);
597	/*
598	* There is no need in deactivating the device or acquiring any locks
599	* prior changing qdiscs since we do not destroy the old qdisc.
600	* Atomic replacement of pointers is enough.
601	*/
602	/*
603	* No need to change reference counters here as we merely move
604	* the pointer and the reference counter of the newly allocated
605	* qdisc is already 1.
606	*/
607	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
608	pPriv->ppSaved[0] = pDev->qdisc_sleeping;
609	ASMAtomicWritePtr(&pDev->qdisc_sleeping, pNew);
610	ASMAtomicWritePtr(&pDev->qdisc, pNew);
611	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
612	for (i = 0; i < pDev->num_tx_queues; i++)
613	{
614	struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
615
616	pPriv->ppSaved[i] = pQueue->qdisc_sleeping;
617	ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pNew);
618	ASMAtomicWritePtr(&pQueue->qdisc, pNew);
619	if (i)
620	atomic_inc(&pNew->refcnt);
621	}
622	/* Newer kernels store root qdisc in netdev structure as well. */
623	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
624	pPriv->ppSaved[pDev->num_tx_queues] = pDev->qdisc;
625	ASMAtomicWritePtr(&pDev->qdisc, pNew);
626	atomic_inc(&pNew->refcnt);
627	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
628	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
629	/* Synch the queue len with our child */
630	pNew->q.qlen = pPriv->pChild->q.qlen;
631	}
632	else
633	{
634	/* We already have vboxnetflt qdisc, let's use it. */
635	pPriv = qdisc_priv(pExisting);
636	}
637	ASMAtomicWritePtr(&pPriv->pVBoxNetFlt, pThis);
638	QDISC_LOG(("vboxNetFltLinuxInstallQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
639	}
640
641	static void vboxNetFltLinuxQdiscRemove(PVBOXNETFLTINS pThis, struct net_device *pDev)
642	{
643	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
644	int i;
645	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
646	PVBOXNETQDISCPRIV pPriv;
647	struct Qdisc pQdisc, pChild;
648	if (!pDev)
649	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
650	if (!VALID_PTR(pDev))
651	{
652	printk("VBoxNetFlt: Failed to detach qdisc, invalid device pointer: %p\n",
653	pDev);
654	return; // TODO: Consider returing an error
655	}
656
657
658	pQdisc = QDISC_GET(pDev);
659	if (strcmp(pQdisc->ops->id, "vboxnetflt"))
660	{
661	/* Looks like the user has replaced our qdisc manually. */
662	printk("VBoxNetFlt: Failed to detach qdisc, wrong qdisc: %s\n",
663	pQdisc->ops->id);
664	return; // TODO: Consider returing an error
665	}
666
667	pPriv = qdisc_priv(pQdisc);
668	Assert(pPriv->pVBoxNetFlt == pThis);
669	ASMAtomicWriteNullPtr(&pPriv->pVBoxNetFlt);
670	pChild = ASMAtomicXchgPtrT(&pPriv->pChild, &noop_qdisc, struct Qdisc *);
671	qdisc_destroy(pChild); /* It won't be the last reference. */
672
673	QDISC_LOG(("vboxNetFltLinuxQdiscRemove: refcnt=%d num_tx_queues=%d\n",
674	atomic_read(&pQdisc->refcnt), pDev->num_tx_queues));
675	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
676	/* Play it safe, make sure the qdisc is not being used. */
677	if (pPriv->ppSaved[0])
678	{
679	ASMAtomicWritePtr(&pDev->qdisc_sleeping, pPriv->ppSaved[0]);
680	ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[0]);
681	pPriv->ppSaved[0] = NULL;
682	while (QDISC_IS_BUSY(pDev, pQdisc))
683	yield();
684	qdisc_destroy(pQdisc); /* Destroy reference */
685	}
686	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
687	for (i = 0; i < pDev->num_tx_queues; i++)
688	{
689	struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
690	if (pPriv->ppSaved[i])
691	{
692	Assert(pQueue->qdisc_sleeping == pQdisc);
693	ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pPriv->ppSaved[i]);
694	ASMAtomicWritePtr(&pQueue->qdisc, pPriv->ppSaved[i]);
695	pPriv->ppSaved[i] = NULL;
696	while (QDISC_IS_BUSY(pDev, pQdisc))
697	yield();
698	qdisc_destroy(pQdisc); /* Destroy reference */
699	}
700	}
701	/* Newer kernels store root qdisc in netdev structure as well. */
702	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
703	ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[pDev->num_tx_queues]);
704	pPriv->ppSaved[pDev->num_tx_queues] = NULL;
705	while (QDISC_IS_BUSY(pDev, pQdisc))
706	yield();
707	qdisc_destroy(pQdisc); /* Destroy reference */
708	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
709	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
710
711	/*
712	* At this point all references to our qdisc should be gone
713	* unless the user had installed it manually.
714	*/
715	QDISC_LOG(("vboxNetFltLinuxRemoveQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
716	}
717
718	#endif /* VBOXNETFLT_WITH_QDISC */
719
720
721	/**
722	* Initialize module.
723	*
724	* @returns appropriate status code.
725	*/
726	static int __init VBoxNetFltLinuxInit(void)
727	{
728	int rc;
729	/*
730	* Initialize IPRT.
731	*/
732	rc = RTR0Init(0);
733	if (RT_SUCCESS(rc))
734	{
735	Log(("VBoxNetFltLinuxInit\n"));
736
737	/*
738	* Initialize the globals and connect to the support driver.
739	*
740	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
741	* for establishing the connect to the support driver.
742	*/
743	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
744	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
745	if (RT_SUCCESS(rc))
746	{
747	#ifdef VBOXNETFLT_WITH_QDISC
748	/*memcpy(&g_VBoxNetFltQDiscOps, &pfifo_qdisc_ops, sizeof(g_VBoxNetFltQDiscOps));
749	strcpy(g_VBoxNetFltQDiscOps.id, "vboxnetflt");
750	g_VBoxNetFltQDiscOps.owner = THIS_MODULE;*/
751	rc = register_qdisc(&g_VBoxNetFltQDiscOps);
752	if (rc)
753	{
754	LogRel(("VBoxNetFlt: Failed to registed qdisc: %d\n", rc));
755	return rc;
756	}
757	#endif /* VBOXNETFLT_WITH_QDISC */
758	LogRel(("VBoxNetFlt: Successfully started.\n"));
759	return 0;
760	}
761
762	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
763	RTR0Term();
764	}
765	else
766	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
767
768	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
769	return -RTErrConvertToErrno(rc);
770	}
771
772
773	/**
774	* Unload the module.
775	*
776	* @todo We have to prevent this if we're busy!
777	*/
778	static void __exit VBoxNetFltLinuxUnload(void)
779	{
780	int rc;
781	Log(("VBoxNetFltLinuxUnload\n"));
782	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
783
784	#ifdef VBOXNETFLT_WITH_QDISC
785	unregister_qdisc(&g_VBoxNetFltQDiscOps);
786	#endif /* VBOXNETFLT_WITH_QDISC */
787	/*
788	* Undo the work done during start (in reverse order).
789	*/
790	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
791	AssertRC(rc); NOREF(rc);
792
793	RTR0Term();
794
795	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
796
797	Log(("VBoxNetFltLinuxUnload - done\n"));
798	}
799
800
801	/**
802	* Experiment where we filter trafic from the host to the internal network
803	* before it reaches the NIC driver.
804	*
805	* The current code uses a very ugly hack and only works on kernels using the
806	* net_device_ops (>= 2.6.29). It has been shown to give us a
807	* performance boost of 60-100% though. So, we have to find some less hacky way
808	* of getting this job done eventually.
809	*
810	* #define VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
811	*/
812	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
813
814	/**
815	* The overridden net_device_ops of the device we're attached to.
816	*
817	* Requires Linux 2.6.29 or later.
818	*
819	* This is a very dirty hack that was create to explore how much we can improve
820	* the host to guest transfers by not CC'ing the NIC.
821	*/
822	typedef struct VBoxNetDeviceOpsOverride
823	{
824	/** Our overridden ops. */
825	struct net_device_ops Ops;
826	/** Magic word. */
827	uint32_t u32Magic;
828	/** Pointer to the original ops. */
829	struct net_device_ops const *pOrgOps;
830	/** Pointer to the net filter instance. */
831	PVBOXNETFLTINS pVBoxNetFlt;
832	/** The number of filtered packages. */
833	uint64_t cFiltered;
834	/** The total number of packets */
835	uint64_t cTotal;
836	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
837	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
838	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
839
840	/**
841	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
842	* because they belong on the internal network.
843	*
844	* @returns NETDEV_TX_XXX.
845	* @param pSkb The socket buffer to transmit.
846	* @param pDev The net device.
847	*/
848	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
849	{
850	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->netdev_ops;
851	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
852	PCRTNETETHERHDR pEtherHdr;
853	PINTNETTRUNKSWPORT pSwitchPort;
854	uint32_t cbHdrs;
855
856
857	/*
858	* Validate the override structure.
859	*
860	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
861	* to be production quality code, we would have to be much more
862	* careful here and avoid the race.
863	*/
864	if ( !VALID_PTR(pOverride)
865	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
866	\|\| !VALID_PTR(pOverride->pOrgOps))
867	{
868	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
869	dev_kfree_skb(pSkb);
870	return NETDEV_TX_OK;
871	}
872	pOverride->cTotal++;
873
874	/*
875	* Do the filtering base on the defaul OUI of our virtual NICs
876	*
877	* Note! In a real solution, we would ask the switch whether the
878	* destination MAC is 100% to be on the internal network and then
879	* drop it.
880	*/
881	cbHdrs = skb_headlen(pSkb);
882	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
883	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
884	if ( pEtherHdr
885	&& VALID_PTR(pOverride->pVBoxNetFlt)
886	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
887	&& VALID_PTR(pSwitchPort)
888	&& cbHdrs >= 6)
889	{
890	INTNETSWDECISION enmDecision;
891
892	/** @todo consider reference counting, etc. */
893	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
894	if (enmDecision == INTNETSWDECISION_INTNET)
895	{
896	dev_kfree_skb(pSkb);
897	pOverride->cFiltered++;
898	return NETDEV_TX_OK;
899	}
900	}
901
902	return pOverride->pOrgOps->ndo_start_xmit(pSkb, pDev);
903	}
904
905	/**
906	* Hooks the device ndo_start_xmit operation of the device.
907	*
908	* @param pThis The net filter instance.
909	* @param pDev The net device.
910	*/
911	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
912	{
913	PVBOXNETDEVICEOPSOVERRIDE pOverride;
914	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
915
916	pOverride = RTMemAlloc(sizeof(*pOverride));
917	if (!pOverride)
918	return;
919	pOverride->pOrgOps = pDev->netdev_ops;
920	pOverride->Ops = *pDev->netdev_ops;
921	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
922	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
923	pOverride->cTotal = 0;
924	pOverride->cFiltered = 0;
925	pOverride->pVBoxNetFlt = pThis;
926
927	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp); /* (this isn't necessary, but so what) */
928	ASMAtomicWritePtr((void * volatile *)&pDev->netdev_ops, pOverride);
929	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
930	}
931
932	/**
933	* Undos what vboxNetFltLinuxHookDev did.
934	*
935	* @param pThis The net filter instance.
936	* @param pDev The net device. Can be NULL, in which case
937	* we'll try retrieve it from @a pThis.
938	*/
939	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
940	{
941	PVBOXNETDEVICEOPSOVERRIDE pOverride;
942	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
943
944	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
945	if (!pDev)
946	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
947	if (VALID_PTR(pDev))
948	{
949	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->netdev_ops;
950	if ( VALID_PTR(pOverride)
951	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
952	&& VALID_PTR(pOverride->pOrgOps)
953	)
954	{
955	ASMAtomicWritePtr((void * volatile *)&pDev->netdev_ops, pOverride->pOrgOps);
956	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
957	}
958	else
959	pOverride = NULL;
960	}
961	else
962	pOverride = NULL;
963	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
964
965	if (pOverride)
966	{
967	printk("vboxnetflt: dropped %llu out of %llu packets\n", pOverride->cFiltered, pOverride->cTotal);
968	RTMemFree(pOverride);
969	}
970	}
971
972	#endif /* VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT */
973
974
975	/**
976	* Reads and retains the host interface handle.
977	*
978	* @returns The handle, NULL if detached.
979	* @param pThis
980	*/
981	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
982	{
983	#if 0
984	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
985	struct net_device *pDev = NULL;
986
987	Log(("vboxNetFltLinuxRetainNetDev\n"));
988	/*
989	* Be careful here to avoid problems racing the detached callback.
990	*/
991	RTSpinlockAcquire(pThis->hSpinlock, &Tmp);
992	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
993	{
994	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
995	if (pDev)
996	{
997	dev_hold(pDev);
998	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
999	}
1000	}
1001	RTSpinlockRelease(pThis->hSpinlock, &Tmp);
1002
1003	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
1004	return pDev;
1005	#else
1006	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1007	#endif
1008	}
1009
1010
1011	/**
1012	* Release the host interface handle previously retained
1013	* by vboxNetFltLinuxRetainNetDev.
1014	*
1015	* @param pThis The instance.
1016	* @param pDev The vboxNetFltLinuxRetainNetDev
1017	* return value, NULL is fine.
1018	*/
1019	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
1020	{
1021	#if 0
1022	Log(("vboxNetFltLinuxReleaseNetDev\n"));
1023	NOREF(pThis);
1024	if (pDev)
1025	{
1026	dev_put(pDev);
1027	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1028	}
1029	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
1030	#endif
1031	}
1032
1033	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
1034	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
1035
1036	/**
1037	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
1038	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
1039	*
1040	* @returns true / false accordingly.
1041	* @param pBuf The sk_buff.
1042	*/
1043	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
1044	{
1045	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
1046	}
1047
1048
1049	/**
1050	* Internal worker that create a linux sk_buff for a
1051	* (scatter/)gather list.
1052	*
1053	* @returns Pointer to the sk_buff.
1054	* @param pThis The instance.
1055	* @param pSG The (scatter/)gather list.
1056	* @param fDstWire Set if the destination is the wire.
1057	*/
1058	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
1059	{
1060	struct sk_buff *pPkt;
1061	struct net_device *pDev;
1062	unsigned fGsoType = 0;
1063
1064	if (pSG->cbTotal == 0)
1065	{
1066	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
1067	return NULL;
1068	}
1069
1070	/** @todo We should use fragments mapping the SG buffers with large packets.
1071	* 256 bytes seems to be the a threshold used a lot for this. It
1072	* requires some nasty work on the intnet side though... */
1073	/*
1074	* Allocate a packet and copy over the data.
1075	*/
1076	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1077	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
1078	if (RT_UNLIKELY(!pPkt))
1079	{
1080	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
1081	pSG->pvUserData = NULL;
1082	return NULL;
1083	}
1084	pPkt->dev = pDev;
1085	pPkt->ip_summed = CHECKSUM_NONE;
1086
1087	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
1088	skb_reserve(pPkt, NET_IP_ALIGN);
1089
1090	/* Copy the segments. */
1091	skb_put(pPkt, pSG->cbTotal);
1092	IntNetSgRead(pSG, pPkt->data);
1093
1094	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1095	/*
1096	* Setup GSO if used by this packet.
1097	*/
1098	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
1099	{
1100	default:
1101	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
1102	/* fall thru */
1103	case PDMNETWORKGSOTYPE_INVALID:
1104	fGsoType = 0;
1105	break;
1106	case PDMNETWORKGSOTYPE_IPV4_TCP:
1107	fGsoType = SKB_GSO_TCPV4;
1108	break;
1109	case PDMNETWORKGSOTYPE_IPV4_UDP:
1110	fGsoType = SKB_GSO_UDP;
1111	break;
1112	case PDMNETWORKGSOTYPE_IPV6_TCP:
1113	fGsoType = SKB_GSO_TCPV6;
1114	break;
1115	}
1116	if (fGsoType)
1117	{
1118	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
1119
1120	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
1121	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
1122	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
1123
1124	/*
1125	* We need to set checksum fields even if the packet goes to the host
1126	* directly as it may be immediately forwared by IP layer @bugref{5020}.
1127	*/
1128	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrs);
1129	pPkt->ip_summed = CHECKSUM_PARTIAL;
1130	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1131	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
1132	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
1133	pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
1134	else
1135	pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
1136	# else
1137	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
1138	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
1139	pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
1140	else
1141	pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
1142	# endif
1143	if (!fDstWire)
1144	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
1145	}
1146	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
1147
1148	/*
1149	* Finish up the socket buffer.
1150	*/
1151	pPkt->protocol = eth_type_trans(pPkt, pDev);
1152	if (fDstWire)
1153	{
1154	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
1155
1156	/* Restore ethernet header back. */
1157	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
1158	VBOX_SKB_RESET_MAC_HDR(pPkt);
1159	}
1160	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
1161
1162	return pPkt;
1163	}
1164
1165
1166	/**
1167	* Initializes a SG list from an sk_buff.
1168	*
1169	* @returns Number of segments.
1170	* @param pThis The instance.
1171	* @param pBuf The sk_buff.
1172	* @param pSG The SG.
1173	* @param pvFrame The frame pointer, optional.
1174	* @param cSegs The number of segments allocated for the SG.
1175	* This should match the number in the mbuf exactly!
1176	* @param fSrc The source of the frame.
1177	* @param pGso Pointer to the GSO context if it's a GSO
1178	* internal network frame. NULL if regular frame.
1179	*/
1180	DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
1181	unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1182	{
1183	int i;
1184	NOREF(pThis);
1185
1186	Assert(!skb_shinfo(pBuf)->frag_list);
1187
1188	if (!pGsoCtx)
1189	IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /cSegsUsed/);
1190	else
1191	IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /cSegsUsed/, pGsoCtx);
1192
1193	#ifdef VBOXNETFLT_SG_SUPPORT
1194	pSG->aSegs[0].cb = skb_headlen(pBuf);
1195	pSG->aSegs[0].pv = pBuf->data;
1196	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1197
1198	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1199	{
1200	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
1201	pSG->aSegs[i+1].cb = pFrag->size;
1202	pSG->aSegs[i+1].pv = kmap(pFrag->page);
1203	printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
1204	pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
1205	}
1206	++i;
1207
1208	#else
1209	pSG->aSegs[0].cb = pBuf->len;
1210	pSG->aSegs[0].pv = pBuf->data;
1211	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1212	i = 1;
1213	#endif
1214
1215	pSG->cSegsUsed = i;
1216
1217	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1218	/*
1219	* Add a trailer if the frame is too small.
1220	*
1221	* Since we're getting to the packet before it is framed, it has not
1222	* yet been padded. The current solution is to add a segment pointing
1223	* to a buffer containing all zeros and pray that works for all frames...
1224	*/
1225	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
1226	{
1227	static uint8_t const s_abZero[128] = {0};
1228
1229	AssertReturnVoid(i < cSegs);
1230
1231	pSG->aSegs[i].Phys = NIL_RTHCPHYS;
1232	pSG->aSegs[i].pv = (void *)&s_abZero[0];
1233	pSG->aSegs[i].cb = 60 - pSG->cbTotal;
1234	pSG->cbTotal = 60;
1235	pSG->cSegsUsed++;
1236	Assert(i + 1 <= pSG->cSegsAlloc)
1237	}
1238	#endif
1239
1240	Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
1241	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
1242	for (i = 0; i < pSG->cSegsUsed; i++)
1243	Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
1244	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1245	}
1246
1247	/**
1248	* Packet handler,
1249	*
1250	* @returns 0 or EJUSTRETURN.
1251	* @param pThis The instance.
1252	* @param pMBuf The mbuf.
1253	* @param pvFrame The start of the frame, optional.
1254	* @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
1255	* @param eProtocol The protocol.
1256	*/
1257	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
1258	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1259	struct net_device *pSkbDev,
1260	struct packet_type *pPacketType,
1261	struct net_device *pOrigDev)
1262	#else
1263	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1264	struct net_device *pSkbDev,
1265	struct packet_type *pPacketType)
1266	#endif
1267	{
1268	PVBOXNETFLTINS pThis;
1269	struct net_device *pDev;
1270	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1271	pBuf, pSkbDev, pPacketType));
1272	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1273	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",
1274	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));
1275	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1276	#else
1277	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",
1278	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));
1279	#endif
1280	/*
1281	* Drop it immediately?
1282	*/
1283	if (!pBuf)
1284	return 0;
1285
1286	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1287	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1288	if (pThis->u.s.pDev != pSkbDev)
1289	{
1290	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1291	return 0;
1292	}
1293
1294	Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1295	if (vboxNetFltLinuxSkBufIsOur(pBuf))
1296	{
1297	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1298	dev_kfree_skb(pBuf);
1299	return 0;
1300	}
1301
1302	#ifndef VBOXNETFLT_SG_SUPPORT
1303	{
1304	/*
1305	* Get rid of fragmented packets, they cause too much trouble.
1306	*/
1307	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1308	kfree_skb(pBuf);
1309	if (!pCopy)
1310	{
1311	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1312	return 0;
1313	}
1314	pBuf = pCopy;
1315	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1316	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",
1317	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));
1318	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1319	# else
1320	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",
1321	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));
1322	# endif
1323	}
1324	#endif
1325
1326	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1327	/* Forward it to the internal network. */
1328	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1329	#else
1330	/* Add the packet to transmit queue and schedule the bottom half. */
1331	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1332	schedule_work(&pThis->u.s.XmitTask);
1333	Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1334	&pThis->u.s.XmitTask, pBuf));
1335	#endif
1336
1337	/* It does not really matter what we return, it is ignored by the kernel. */
1338	return 0;
1339	}
1340
1341	/**
1342	* Calculate the number of INTNETSEG segments the socket buffer will need.
1343	*
1344	* @returns Segment count.
1345	* @param pBuf The socket buffer.
1346	*/
1347	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
1348	{
1349	#ifdef VBOXNETFLT_SG_SUPPORT
1350	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1351	#else
1352	unsigned cSegs = 1;
1353	#endif
1354	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1355	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1356	if (pBuf->len < 60)
1357	cSegs++;
1358	#endif
1359	return cSegs;
1360	}
1361
1362	/**
1363	* Destroy the intnet scatter / gather buffer created by
1364	* vboxNetFltLinuxSkBufToSG.
1365	*/
1366	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
1367	{
1368	#ifdef VBOXNETFLT_SG_SUPPORT
1369	int i;
1370
1371	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1372	{
1373	printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
1374	kunmap(pSG->aSegs[i+1].pv);
1375	}
1376	#endif
1377	NOREF(pSG);
1378	}
1379
1380	#ifdef LOG_ENABLED
1381	/**
1382	* Logging helper.
1383	*/
1384	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1385	{
1386	uint8_t pInt, pExt;
1387	static int iPacketNo = 1;
1388	iPacketNo += iIncrement;
1389	if (fEgress)
1390	{
1391	pExt = pSG->aSegs[0].pv;
1392	pInt = pExt + 6;
1393	}
1394	else
1395	{
1396	pInt = pSG->aSegs[0].pv;
1397	pExt = pInt + 6;
1398	}
1399	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1400	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1401	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1402	fEgress ? "-->" : "<--", pszWhere,
1403	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1404	pSG->cbTotal, iPacketNo));
1405	Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1406	}
1407	#else
1408	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1409	#endif
1410
1411	#ifdef VBOXNETFLT_WITH_GSO_RECV
1412
1413	/**
1414	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1415	* GSO socket buffer without having to segment it.
1416	*
1417	* @returns true on success, false if needs segmenting.
1418	* @param pThis The net filter instance.
1419	* @param pSkb The GSO socket buffer.
1420	* @param fSrc The source.
1421	* @param pGsoCtx Where to return the GSO context on success.
1422	*/
1423	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1424	PPDMNETWORKGSO pGsoCtx)
1425	{
1426	PDMNETWORKGSOTYPE enmGsoType;
1427	uint16_t uEtherType;
1428	unsigned int cbTransport;
1429	unsigned int offTransport;
1430	unsigned int cbTransportHdr;
1431	unsigned uProtocol;
1432	union
1433	{
1434	RTNETIPV4 IPv4;
1435	RTNETIPV6 IPv6;
1436	RTNETTCP Tcp;
1437	uint8_t ab[40];
1438	uint16_t au16[40/2];
1439	uint32_t au32[40/4];
1440	} Buf;
1441
1442	/*
1443	* Check the GSO properties of the socket buffer and make sure it fits.
1444	*/
1445	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1446	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP \| SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
1447	{
1448	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1449	return false;
1450	}
1451	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1452	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
1453	{
1454	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1455	return false;
1456	}
1457	/*
1458	* It is possible to receive GSO packets from wire if GRO is enabled.
1459	*/
1460	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1461	{
1462	Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1463	#ifdef VBOXNETFLT_WITH_GRO
1464	/*
1465	* The packet came from the wire and the driver has already consumed
1466	* mac header. We need to restore it back.
1467	*/
1468	pSkb->mac_len = skb_network_header(pSkb) - skb_mac_header(pSkb);
1469	skb_push(pSkb, pSkb->mac_len);
1470	Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1471	pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
1472	#else /* !VBOXNETFLT_WITH_GRO */
1473	/* Older kernels didn't have GRO. */
1474	return false;
1475	#endif /* !VBOXNETFLT_WITH_GRO */
1476	}
1477	else
1478	{
1479	/*
1480	* skb_gso_segment does the following. Do we need to do it as well?
1481	*/
1482	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1483	skb_reset_mac_header(pSkb);
1484	pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1485	#else
1486	pSkb->mac.raw = pSkb->data;
1487	pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1488	#endif
1489	}
1490
1491	/*
1492	* Switch on the ethertype.
1493	*/
1494	uEtherType = pSkb->protocol;
1495	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1496	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1497	{
1498	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1499	if (puEtherType)
1500	uEtherType = *puEtherType;
1501	}
1502	switch (uEtherType)
1503	{
1504	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1505	{
1506	unsigned int cbHdr;
1507	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1508	if (RT_UNLIKELY(!pIPv4))
1509	{
1510	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1511	return false;
1512	}
1513
1514	cbHdr = pIPv4->ip_hl * 4;
1515	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1516	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1517	\|\| cbHdr > cbTransport ))
1518	{
1519	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1520	return false;
1521	}
1522	cbTransport -= cbHdr;
1523	offTransport = pSkb->mac_len + cbHdr;
1524	uProtocol = pIPv4->ip_p;
1525	if (uProtocol == RTNETIPV4_PROT_TCP)
1526	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1527	else if (uProtocol == RTNETIPV4_PROT_UDP)
1528	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1529	else /** @todo IPv6: 4to6 tunneling */
1530	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1531	break;
1532	}
1533
1534	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1535	{
1536	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1537	if (RT_UNLIKELY(!pIPv6))
1538	{
1539	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1540	return false;
1541	}
1542
1543	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1544	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1545	uProtocol = pIPv6->ip6_nxt;
1546	/** @todo IPv6: Dig our way out of the other headers. */
1547	if (uProtocol == RTNETIPV4_PROT_TCP)
1548	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1549	else if (uProtocol == RTNETIPV4_PROT_UDP)
1550	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1551	else
1552	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1553	break;
1554	}
1555
1556	default:
1557	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1558	return false;
1559	}
1560
1561	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1562	{
1563	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1564	return false;
1565	}
1566
1567	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1568	\|\| offTransport + cbTransport > pSkb->len
1569	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1570	{
1571	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1572	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1573	return false;
1574	}
1575
1576	/*
1577	* Check the TCP/UDP bits.
1578	*/
1579	if (uProtocol == RTNETIPV4_PROT_TCP)
1580	{
1581	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1582	if (RT_UNLIKELY(!pTcp))
1583	{
1584	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1585	return false;
1586	}
1587
1588	cbTransportHdr = pTcp->th_off * 4;
1589	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1590	\|\| cbTransportHdr > cbTransport
1591	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1592	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1593	{
1594	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1595	return false;
1596	}
1597
1598	}
1599	else
1600	{
1601	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1602	cbTransportHdr = sizeof(RTNETUDP);
1603	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1604	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1605	{
1606	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1607	return false;
1608	}
1609	}
1610
1611	/*
1612	* We're good, init the GSO context.
1613	*/
1614	pGsoCtx->u8Type = enmGsoType;
1615	pGsoCtx->cbHdrs = offTransport + cbTransportHdr;
1616	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1617	pGsoCtx->offHdr1 = pSkb->mac_len;
1618	pGsoCtx->offHdr2 = offTransport;
1619	pGsoCtx->au8Unused[0] = 0;
1620	pGsoCtx->au8Unused[1] = 0;
1621
1622	return true;
1623	}
1624
1625	/**
1626	* Forward the socket buffer as a GSO internal network frame.
1627	*
1628	* @returns IPRT status code.
1629	* @param pThis The net filter instance.
1630	* @param pSkb The GSO socket buffer.
1631	* @param fSrc The source.
1632	* @param pGsoCtx Where to return the GSO context on success.
1633	*/
1634	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1635	{
1636	int rc;
1637	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1638	if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1639	{
1640	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1641	if (RT_LIKELY(pSG))
1642	{
1643	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1644
1645	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1646	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1647
1648	vboxNetFltLinuxDestroySG(pSG);
1649	rc = VINF_SUCCESS;
1650	}
1651	else
1652	{
1653	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1654	rc = VERR_NO_MEMORY;
1655	}
1656	}
1657	else
1658	{
1659	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1660	rc = VERR_INTERNAL_ERROR_3;
1661	}
1662
1663	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1664	dev_kfree_skb(pSkb);
1665	return rc;
1666	}
1667
1668	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1669
1670	/**
1671	* Worker for vboxNetFltLinuxForwardToIntNet.
1672	*
1673	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1674	* @param pThis The net filter instance.
1675	* @param pBuf The socket buffer.
1676	* @param fSrc The source.
1677	*/
1678	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1679	{
1680	int rc;
1681	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1682	if (cSegs <= MAX_SKB_FRAGS + 1)
1683	{
1684	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1685	if (RT_LIKELY(pSG))
1686	{
1687	if (fSrc & INTNETTRUNKDIR_WIRE)
1688	{
1689	/*
1690	* The packet came from wire, ethernet header was removed by device driver.
1691	* Restore it.
1692	*/
1693	skb_push(pBuf, ETH_HLEN);
1694	}
1695
1696	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /pGsoCtx/);
1697
1698	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1699	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1700
1701	vboxNetFltLinuxDestroySG(pSG);
1702	rc = VINF_SUCCESS;
1703	}
1704	else
1705	{
1706	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1707	rc = VERR_NO_MEMORY;
1708	}
1709	}
1710	else
1711	{
1712	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1713	rc = VERR_INTERNAL_ERROR_3;
1714	}
1715
1716	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1717	dev_kfree_skb(pBuf);
1718	return rc;
1719	}
1720
1721	/**
1722	*
1723	* @param pBuf The socket buffer. This is consumed by this function.
1724	*/
1725	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1726	{
1727	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1728
1729	#ifdef VBOXNETFLT_WITH_GSO
1730	if (skb_is_gso(pBuf))
1731	{
1732	PDMNETWORKGSO GsoCtx;
1733	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",
1734	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));
1735	# ifdef VBOXNETFLT_WITH_GSO_RECV
1736	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1737	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1738	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1739	else
1740	# endif
1741	{
1742	/* Need to segment the packet */
1743	struct sk_buff *pNext;
1744	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1745	if (IS_ERR(pSegment))
1746	{
1747	dev_kfree_skb(pBuf);
1748	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1749	return;
1750	}
1751
1752	for (; pSegment; pSegment = pNext)
1753	{
1754	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",
1755	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));
1756	pNext = pSegment->next;
1757	pSegment->next = 0;
1758	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1759	}
1760	dev_kfree_skb(pBuf);
1761	}
1762	}
1763	else
1764	#endif /* VBOXNETFLT_WITH_GSO */
1765	{
1766	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1767	{
1768	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1769	/*
1770	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1771	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1772	* header length from the header itself and reconstruct 'h' pointer
1773	* to TCP (or whatever) header.
1774	*/
1775	unsigned char *tmp = pBuf->h.raw;
1776	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1777	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1778	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1779	if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1780	{
1781	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1782	dev_kfree_skb(pBuf);
1783	return;
1784	}
1785	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1786	/* Restore the original (wrong) pointer. */
1787	pBuf->h.raw = tmp;
1788	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1789	}
1790	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1791	}
1792	}
1793
1794	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1795	/**
1796	* Work queue handler that forwards the socket buffers queued by
1797	* vboxNetFltLinuxPacketHandler to the internal network.
1798	*
1799	* @param pWork The work queue.
1800	*/
1801	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1802	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1803	# else
1804	static void vboxNetFltLinuxXmitTask(void *pWork)
1805	# endif
1806	{
1807	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1808	struct sk_buff *pBuf;
1809
1810	Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1811
1812	/*
1813	* Active? Retain the instance and increment the busy counter.
1814	*/
1815	if (vboxNetFltTryRetainBusyActive(pThis))
1816	{
1817	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1818	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1819
1820	vboxNetFltRelease(pThis, true /* fBusy */);
1821	}
1822	else
1823	{
1824	/** @todo Shouldn't we just drop the packets here? There is little point in
1825	* making them accumulate when the VM is paused and it'll only waste
1826	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1827	* before start draining the packets (goes for the intnet ring buf
1828	* too)? */
1829	}
1830	}
1831	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1832
1833	/**
1834	* Reports the GSO capabilites of the hardware NIC.
1835	*
1836	* @param pThis The net filter instance. The caller hold a
1837	* reference to this.
1838	*/
1839	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1840	{
1841	#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1842	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1843	{
1844	struct net_device *pDev;
1845	PINTNETTRUNKSWPORT pSwitchPort;
1846	unsigned int fFeatures;
1847	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1848
1849	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1850
1851	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1852	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1853	if (pDev)
1854	fFeatures = pDev->features;
1855	else
1856	fFeatures = 0;
1857
1858	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1859
1860	if (pThis->pSwitchPort)
1861	{
1862	/* Set/update the GSO capabilities of the NIC. */
1863	uint32_t fGsoCapabilites = 0;
1864	if (fFeatures & NETIF_F_TSO)
1865	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1866	if (fFeatures & NETIF_F_TSO6)
1867	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1868	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1869	if (fFeatures & NETIF_F_UFO)
1870	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1871	if (fFeatures & NETIF_F_UFO)
1872	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1873	# endif
1874	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1875	}
1876
1877	vboxNetFltRelease(pThis, true /fBusy/);
1878	}
1879	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1880	}
1881
1882	/**
1883	* Helper that determins whether the host (ignoreing us) is operating the
1884	* interface in promiscuous mode or not.
1885	*/
1886	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1887	{
1888	bool fRc = false;
1889	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1890	if (pDev)
1891	{
1892	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1893	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1894	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1895	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1896	}
1897	return fRc;
1898	}
1899
1900	/**
1901	* Internal worker for vboxNetFltLinuxNotifierCallback.
1902	*
1903	* @returns VBox status code.
1904	* @param pThis The instance.
1905	* @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1906	* flood the release log.
1907	*/
1908	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1909	{
1910	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1911	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1912
1913	/*
1914	* Retain and store the device.
1915	*/
1916	dev_hold(pDev);
1917
1918	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1919	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1920	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1921
1922	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1923	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n", pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1924
1925	/* Get the mac address while we still have a valid net_device reference. */
1926	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1927
1928	/*
1929	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1930	*/
1931	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1932	pThis->u.s.PacketType.dev = pDev;
1933	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1934	dev_add_pack(&pThis->u.s.PacketType);
1935
1936	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1937	vboxNetFltLinuxHookDev(pThis, pDev);
1938	#endif
1939	#ifdef VBOXNETFLT_WITH_QDISC
1940	vboxNetFltLinuxQdiscInstall(pThis, pDev);
1941	#endif /* VBOXNETFLT_WITH_QDISC */
1942
1943	/*
1944	* Set indicators that require the spinlock. Be abit paranoid about racing
1945	* the device notification handle.
1946	*/
1947	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1948	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1949	if (pDev)
1950	{
1951	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1952	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1953	pDev = NULL; /* don't dereference it */
1954	}
1955	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1956	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1957
1958	/*
1959	* If the above succeeded report GSO capabilites, if not undo and
1960	* release the device.
1961	*/
1962	if (!pDev)
1963	{
1964	Assert(pThis->pSwitchPort);
1965	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1966	{
1967	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1968	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1969	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1970	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1971	vboxNetFltRelease(pThis, true /fBusy/);
1972	}
1973	}
1974	else
1975	{
1976	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1977	vboxNetFltLinuxUnhookDev(pThis, pDev);
1978	#endif
1979	#ifdef VBOXNETFLT_WITH_QDISC
1980	vboxNetFltLinuxQdiscRemove(pThis, pDev);
1981	#endif /* VBOXNETFLT_WITH_QDISC */
1982	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1983	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1984	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1985	dev_put(pDev);
1986	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1987	}
1988
1989	LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1990	return VINF_SUCCESS;
1991	}
1992
1993
1994	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1995	{
1996	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1997
1998	Assert(!pThis->fDisconnectedFromHost);
1999
2000	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2001	vboxNetFltLinuxUnhookDev(pThis, pDev);
2002	#endif
2003	#ifdef VBOXNETFLT_WITH_QDISC
2004	vboxNetFltLinuxQdiscRemove(pThis, pDev);
2005	#endif /* VBOXNETFLT_WITH_QDISC */
2006
2007	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2008	ASMAtomicWriteBool(&pThis->u.s.fRegistered, false);
2009	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
2010	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2011	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2012
2013	dev_remove_pack(&pThis->u.s.PacketType);
2014	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2015	skb_queue_purge(&pThis->u.s.XmitQueue);
2016	#endif
2017	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
2018	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
2019	dev_put(pDev);
2020
2021	return NOTIFY_OK;
2022	}
2023
2024	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
2025	{
2026	/* Check if we are not suspended and promiscuous mode has not been set. */
2027	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
2028	&& !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
2029	{
2030	/* Note that there is no need for locking as the kernel got hold of the lock already. */
2031	dev_set_promiscuity(pDev, 1);
2032	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
2033	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2034	}
2035	else
2036	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2037	return NOTIFY_OK;
2038	}
2039
2040	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
2041	{
2042	/* Undo promiscuous mode if we has set it. */
2043	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
2044	{
2045	/* Note that there is no need for locking as the kernel got hold of the lock already. */
2046	dev_set_promiscuity(pDev, -1);
2047	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
2048	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2049	}
2050	else
2051	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2052	return NOTIFY_OK;
2053	}
2054
2055	#ifdef LOG_ENABLED
2056	/** Stringify the NETDEV_XXX constants. */
2057	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2058	{
2059	const char *pszEvent = "NETDRV_<unknown>";
2060	switch (ulEventType)
2061	{
2062	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2063	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2064	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2065	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2066	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2067	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2068	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2069	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2070	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2071	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2072	# ifdef NETDEV_FEAT_CHANGE
2073	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2074	# endif
2075	}
2076	return pszEvent;
2077	}
2078	#endif /* LOG_ENABLED */
2079
2080	/**
2081	* Callback for listening to netdevice events.
2082	*
2083	* This works the rediscovery, clean up on unregistration, promiscuity on
2084	* up/down, and GSO feature changes from ethtool.
2085	*
2086	* @returns NOTIFY_OK
2087	* @param self Pointer to our notifier registration block.
2088	* @param ulEventType The event.
2089	* @param ptr Event specific, but it is usually the device it
2090	* relates to.
2091	*/
2092	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2093
2094	{
2095	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2096	struct net_device pDev = (struct net_device )ptr;
2097	int rc = NOTIFY_OK;
2098
2099	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2100	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
2101	if ( ulEventType == NETDEV_REGISTER
2102	&& !strcmp(pDev->name, pThis->szName))
2103	{
2104	vboxNetFltLinuxAttachToInterface(pThis, pDev);
2105	}
2106	else
2107	{
2108	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2109	if (pDev == ptr)
2110	{
2111	switch (ulEventType)
2112	{
2113	case NETDEV_UNREGISTER:
2114	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2115	break;
2116	case NETDEV_UP:
2117	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2118	break;
2119	case NETDEV_GOING_DOWN:
2120	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2121	break;
2122	case NETDEV_CHANGENAME:
2123	break;
2124	#ifdef NETDEV_FEAT_CHANGE
2125	case NETDEV_FEAT_CHANGE:
2126	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2127	break;
2128	#endif
2129	}
2130	}
2131	}
2132
2133	return rc;
2134	}
2135
2136	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2137	{
2138	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2139	}
2140
2141	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2142	{
2143	struct net_device * pDev;
2144	int err;
2145	int rc = VINF_SUCCESS;
2146	NOREF(pvIfData);
2147
2148	LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2149
2150	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2151	if (pDev)
2152	{
2153	/*
2154	* Create a sk_buff for the gather list and push it onto the wire.
2155	*/
2156	if (fDst & INTNETTRUNKDIR_WIRE)
2157	{
2158	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2159	if (pBuf)
2160	{
2161	vboxNetFltDumpPacket(pSG, true, "wire", 1);
2162	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2163	Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2164	err = dev_queue_xmit(pBuf);
2165	if (err)
2166	rc = RTErrConvertFromErrno(err);
2167	}
2168	else
2169	rc = VERR_NO_MEMORY;
2170	}
2171
2172	/*
2173	* Create a sk_buff for the gather list and push it onto the host stack.
2174	*/
2175	if (fDst & INTNETTRUNKDIR_HOST)
2176	{
2177	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2178	if (pBuf)
2179	{
2180	vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2181	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2182	Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2183	err = netif_rx_ni(pBuf);
2184	if (err)
2185	rc = RTErrConvertFromErrno(err);
2186	}
2187	else
2188	rc = VERR_NO_MEMORY;
2189	}
2190
2191	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2192	}
2193
2194	return rc;
2195	}
2196
2197
2198	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2199	{
2200	struct net_device * pDev;
2201
2202	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
2203	pThis, pThis->szName, fActive?"true":"false",
2204	pThis->fDisablePromiscuous?"true":"false"));
2205
2206	if (pThis->fDisablePromiscuous)
2207	return;
2208
2209	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2210	if (pDev)
2211	{
2212	/*
2213	* This api is a bit weird, the best reference is the code.
2214	*
2215	* Also, we have a bit or race conditions wrt the maintance of
2216	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2217	*/
2218	#ifdef LOG_ENABLED
2219	u_int16_t fIf;
2220	unsigned const cPromiscBefore = pDev->promiscuity;
2221	#endif
2222	if (fActive)
2223	{
2224	Assert(!pThis->u.s.fPromiscuousSet);
2225
2226	rtnl_lock();
2227	dev_set_promiscuity(pDev, 1);
2228	rtnl_unlock();
2229	pThis->u.s.fPromiscuousSet = true;
2230	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2231	}
2232	else
2233	{
2234	if (pThis->u.s.fPromiscuousSet)
2235	{
2236	rtnl_lock();
2237	dev_set_promiscuity(pDev, -1);
2238	rtnl_unlock();
2239	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2240	}
2241	pThis->u.s.fPromiscuousSet = false;
2242
2243	#ifdef LOG_ENABLED
2244	fIf = dev_get_flags(pDev);
2245	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2246	#endif
2247	}
2248
2249	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2250	}
2251	}
2252
2253
2254	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2255	{
2256	#ifdef VBOXNETFLT_WITH_QDISC
2257	vboxNetFltLinuxQdiscRemove(pThis, NULL);
2258	#endif /* VBOXNETFLT_WITH_QDISC */
2259	/*
2260	* Remove packet handler when we get disconnected from internal switch as
2261	* we don't want the handler to forward packets to disconnected switch.
2262	*/
2263	dev_remove_pack(&pThis->u.s.PacketType);
2264	return VINF_SUCCESS;
2265	}
2266
2267
2268	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2269	{
2270	/*
2271	* Report the GSO capabilities of the host and device (if connected).
2272	* Note! No need to mark ourselves busy here.
2273	*/
2274	/** @todo duplicate work here now? Attach */
2275	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2276	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2277	0
2278	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2279	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2280	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2281	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2282	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2283	# endif
2284	, INTNETTRUNKDIR_HOST);
2285
2286	#endif
2287	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2288
2289	return VINF_SUCCESS;
2290	}
2291
2292
2293	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2294	{
2295	struct net_device *pDev;
2296	bool fRegistered;
2297	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
2298
2299	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2300	vboxNetFltLinuxUnhookDev(pThis, NULL);
2301	#endif
2302
2303	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2304	* unlikely, but none the less). Since it doesn't actually update the
2305	* state (just reads it), it is likely to panic in some interesting
2306	* ways. */
2307
2308	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2309	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2310	fRegistered = ASMAtomicUoReadBool(&pThis->u.s.fRegistered);
2311	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2312
2313	if (fRegistered)
2314	{
2315	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2316	skb_queue_purge(&pThis->u.s.XmitQueue);
2317	#endif
2318	Log(("vboxNetFltOsDeleteInstance: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
2319	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
2320	dev_put(pDev);
2321	}
2322	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2323	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2324	module_put(THIS_MODULE);
2325	}
2326
2327
2328	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2329	{
2330	int err;
2331	NOREF(pvContext);
2332
2333	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2334	err = register_netdevice_notifier(&pThis->u.s.Notifier);
2335	if (err)
2336	return VERR_INTNET_FLT_IF_FAILED;
2337	if (!pThis->u.s.fRegistered)
2338	{
2339	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2340	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2341	return VERR_INTNET_FLT_IF_NOT_FOUND;
2342	}
2343
2344	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2345	if ( pThis->fDisconnectedFromHost
2346	\|\| !try_module_get(THIS_MODULE))
2347	return VERR_INTNET_FLT_IF_FAILED;
2348
2349	return VINF_SUCCESS;
2350	}
2351
2352	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2353	{
2354	/*
2355	* Init the linux specific members.
2356	*/
2357	pThis->u.s.pDev = NULL;
2358	pThis->u.s.fRegistered = false;
2359	pThis->u.s.fPromiscuousSet = false;
2360	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2361	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2362	skb_queue_head_init(&pThis->u.s.XmitQueue);
2363	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2364	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2365	# else
2366	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2367	# endif
2368	#endif
2369
2370	return VINF_SUCCESS;
2371	}
2372
2373
2374	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2375	{
2376	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2377	}
2378
2379
2380	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
2381	{
2382	/* Nothing to do */
2383	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2384	return VINF_SUCCESS;
2385	}
2386
2387
2388	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2389	{
2390	/* Nothing to do */
2391	NOREF(pThis); NOREF(pvIfData);
2392	return VINF_SUCCESS;
2393	}
2394

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 31675

Download in other formats: