VBoxNetFlt-linux.c@ 31705

Last change on this file since 31705 was 31705, checked in by vboxsync, 14 years ago
vboxnetflt: indentation
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 82.7 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 31705 2010-08-16 15:24:39Z 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	#ifdef CONFIG_NET_SCHED
54	# define VBOXNETFLT_WITH_QDISC /* Comment this out to disable qdisc support */
55	# ifdef VBOXNETFLT_WITH_QDISC
56	# include <net/pkt_sched.h>
57	# endif /* VBOXNETFLT_WITH_QDISC */
58	#endif
59
60
61	/*******************************************************************************
62	* Defined Constants And Macros *
63	*******************************************************************************/
64	#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
65	#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
66	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
67	# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
68	#endif
69
70	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
71	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
72	# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
73	#else
74	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
75	# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
76	#endif
77
78	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
79	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
80	#else
81	# define CHECKSUM_PARTIAL CHECKSUM_HW
82	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
83	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
84	# else
85	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
86	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
87	# else
88	# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
89	# endif
90	/* Versions prior 2.6.10 use stats for both bstats and qstats */
91	# define bstats stats
92	# define qstats stats
93	# endif
94	#endif
95
96	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 13)
97	static inline int qdisc_drop(struct sk_buff skb, struct Qdisc sch)
98	{
99	kfree_skb(skb);
100	sch->stats.drops++;
101
102	return NET_XMIT_DROP;
103	}
104	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 13) */
105
106	#ifndef NET_IP_ALIGN
107	# define NET_IP_ALIGN 2
108	#endif
109
110	#if 0
111	/** Create scatter / gather segments for fragments. When not used, we will
112	* linearize the socket buffer before creating the internal networking SG. */
113	# define VBOXNETFLT_SG_SUPPORT 1
114	#endif
115
116	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
117	/** Indicates that the linux kernel may send us GSO frames. */
118	# define VBOXNETFLT_WITH_GSO 1
119
120	/** This enables or disables the transmitting of GSO frame from the internal
121	* network and to the host. */
122	# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
123
124	# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
125	/** This enables or disables the transmitting of GSO frame from the internal
126	* network and to the wire. */
127	# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
128	# endif
129
130	/** This enables or disables the forwarding/flooding of GSO frame from the host
131	* to the internal network. */
132	# define VBOXNETFLT_WITH_GSO_RECV 1
133
134	#endif
135
136	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
137	/** This enables or disables handling of GSO frames coming from the wire (GRO). */
138	# define VBOXNETFLT_WITH_GRO 1
139	#endif
140	/*
141	* GRO support was backported to RHEL 5.4
142	*/
143	#ifdef RHEL_RELEASE_CODE
144	# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
145	# define VBOXNETFLT_WITH_GRO 1
146	# endif
147	#endif
148
149	/*******************************************************************************
150	* Internal Functions *
151	*******************************************************************************/
152	static int VBoxNetFltLinuxInit(void);
153	static void VBoxNetFltLinuxUnload(void);
154	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
155
156
157	/*******************************************************************************
158	* Global Variables *
159	*******************************************************************************/
160	/**
161	* The (common) global data.
162	*/
163	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
164
165	module_init(VBoxNetFltLinuxInit);
166	module_exit(VBoxNetFltLinuxUnload);
167
168	MODULE_AUTHOR(VBOX_VENDOR);
169	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
170	MODULE_LICENSE("GPL");
171	#ifdef MODULE_VERSION
172	MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
173	#endif
174
175
176	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
177	unsigned dev_get_flags(const struct net_device *dev)
178	{
179	unsigned flags;
180
181	flags = (dev->flags & ~(IFF_PROMISC \|
182	IFF_ALLMULTI \|
183	IFF_RUNNING)) \|
184	(dev->gflags & (IFF_PROMISC \|
185	IFF_ALLMULTI));
186
187	if (netif_running(dev) && netif_carrier_ok(dev))
188	flags \|= IFF_RUNNING;
189
190	return flags;
191	}
192	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
193
194
195	#ifdef VBOXNETFLT_WITH_QDISC
196	//#define QDISC_LOG(x) printk x
197	# define QDISC_LOG(x) do { } while (0)
198
199	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
200	# define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops)
201	# elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
202	# define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops, parent)
203	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
204	# define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, queue, ops, parent)
205	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
206
207	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
208	# define qdisc_dev(qdisc) (qdisc->dev)
209	# define qdisc_pkt_len(skb) (skb->len)
210	# define QDISC_GET(dev) (dev->qdisc_sleeping)
211	# else
212	# define QDISC_GET(dev) (netdev_get_tx_queue(dev, 0)->qdisc_sleeping)
213	# endif
214
215	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
216	# define QDISC_SAVED_NUM(dev) 1
217	# elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
218	# define QDISC_SAVED_NUM(dev) dev->num_tx_queues
219	# else
220	# define QDISC_SAVED_NUM(dev) dev->num_tx_queues+1
221	# endif
222
223	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
224	# define QDISC_IS_BUSY(dev, qdisc) test_bit(__LINK_STATE_SCHED, &dev->state)
225	# elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
226	# define QDISC_IS_BUSY(dev, qdisc) (test_bit(__QDISC_STATE_RUNNING, &qdisc->state) \|\| \
227	test_bit(__QDISC_STATE_SCHED, &qdisc->state))
228	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36) */
229	# define QDISC_IS_BUSY(dev, qdisc) (qdisc_is_running(qdisc) \|\| \
230	test_bit(__QDISC_STATE_SCHED, &qdisc->state))
231	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36) */
232
233	struct VBoxNetQDiscPriv
234	{
235	/** Pointer to the single child qdisc. */
236	struct Qdisc *pChild;
237	/*
238	* Technically it is possible to have different qdiscs for different TX
239	* queues so we have to save them all.
240	*/
241	/** Pointer to the array of saved qdiscs. */
242	struct Qdisc **ppSaved;
243	/** Pointer to the net filter instance. */
244	PVBOXNETFLTINS pVBoxNetFlt;
245	};
246	typedef struct VBoxNetQDiscPriv *PVBOXNETQDISCPRIV;
247
248	//#define VBOXNETFLT_QDISC_ENQUEUE
249	static int vboxNetFltQdiscEnqueue(struct sk_buff skb, struct Qdisc sch)
250	{
251	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
252	int rc;
253
254	# ifdef VBOXNETFLT_QDISC_ENQUEUE
255	if (VALID_PTR(pPriv->pVBoxNetFlt))
256	{
257	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
258	PCRTNETETHERHDR pEtherHdr;
259	PINTNETTRUNKSWPORT pSwitchPort;
260	uint32_t cbHdrs = skb_headlen(skb);
261
262	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
263	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(skb, 0, cbHdrs, &abHdrBuf[0]);
264	if ( pEtherHdr
265	&& (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) != NULL
266	&& VALID_PTR(pSwitchPort)
267	&& cbHdrs >= 6)
268	{
269	/** @todo consider reference counting, etc. */
270	INTNETSWDECISION enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
271	if (enmDecision == INTNETSWDECISION_INTNET)
272	{
273	struct sk_buff *pBuf = skb_copy(skb, GFP_ATOMIC);
274	pBuf->pkt_type = PACKET_OUTGOING;
275	vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
276	qdisc_drop(skb, sch);
277	++sch->bstats.packets;
278	sch->bstats.bytes += qdisc_pkt_len(skb);
279	return NET_XMIT_SUCCESS;
280	}
281	}
282	}
283	# endif /* VBOXNETFLT_QDISC_ENQUEUE */
284	rc = pPriv->pChild->enqueue(skb, pPriv->pChild);
285	if (rc == NET_XMIT_SUCCESS)
286	{
287	++sch->q.qlen;
288	++sch->bstats.packets;
289	sch->bstats.bytes += qdisc_pkt_len(skb);
290	}
291	else
292	++sch->qstats.drops;
293	return rc;
294	}
295
296	static struct sk_buff vboxNetFltQdiscDequeue(struct Qdisc sch)
297	{
298	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
299	# ifdef VBOXNETFLT_QDISC_ENQUEUE
300	--sch->q.qlen;
301	return pPriv->pChild->dequeue(pPriv->pChild);
302	# else /* VBOXNETFLT_QDISC_ENQUEUE */
303	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
304	PCRTNETETHERHDR pEtherHdr;
305	PINTNETTRUNKSWPORT pSwitchPort;
306	struct sk_buff *pSkb;
307
308	QDISC_LOG(("vboxNetFltDequeue: Enter pThis=%p\n", pPriv->pVBoxNetFlt));
309
310	while ((pSkb = pPriv->pChild->dequeue(pPriv->pChild)) != NULL)
311	{
312	struct sk_buff *pBuf;
313	INTNETSWDECISION enmDecision;
314	uint32_t cbHdrs;
315
316	--sch->q.qlen;
317
318	if (!VALID_PTR(pPriv->pVBoxNetFlt))
319	break;
320
321	cbHdrs = skb_headlen(pSkb);
322	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
323	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
324	if ( !pEtherHdr
325	\|\| (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) == NULL
326	\|\| !VALID_PTR(pSwitchPort)
327	\|\| cbHdrs < 6)
328	break;
329
330	/** @todo consider reference counting, etc. */
331	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
332	if (enmDecision != INTNETSWDECISION_INTNET)
333	break;
334
335	pBuf = skb_copy(pSkb, GFP_ATOMIC);
336	pBuf->pkt_type = PACKET_OUTGOING;
337	QDISC_LOG(("vboxNetFltDequeue: pThis=%p\n", pPriv->pVBoxNetFlt));
338	vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
339	qdisc_drop(pSkb, sch);
340	QDISC_LOG(("VBoxNetFlt: Packet for %02x:%02x:%02x:%02x:%02x:%02x dropped\n",
341	pSkb->data[0], pSkb->data[1], pSkb->data[2],
342	pSkb->data[3], pSkb->data[4], pSkb->data[5]));
343	}
344
345	return pSkb;
346	# endif /* VBOXNETFLT_QDISC_ENQUEUE */
347	}
348
349	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
350	static int vboxNetFltQdiscRequeue(struct sk_buff skb, struct Qdisc sch)
351	{
352	int rc;
353	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
354
355	rc = pPriv->pChild->ops->requeue(skb, pPriv->pChild);
356	if (rc == 0)
357	{
358	sch->q.qlen++;
359	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
360	sch->qstats.requeues++;
361	# endif
362	}
363
364	return rc;
365	}
366	# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
367
368	static unsigned int vboxNetFltQdiscDrop(struct Qdisc *sch)
369	{
370	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
371	unsigned int cbLen;
372
373	if (pPriv->pChild->ops->drop)
374	{
375	cbLen = pPriv->pChild->ops->drop(pPriv->pChild);
376	if (cbLen != 0)
377	{
378	++sch->qstats.drops;
379	--sch->q.qlen;
380	return cbLen;
381	}
382	}
383
384	return 0;
385	}
386
387	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
388	static int vboxNetFltQdiscInit(struct Qdisc sch, struct rtattr opt)
389	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
390	static int vboxNetFltQdiscInit(struct Qdisc sch, struct nlattr opt)
391	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
392	{
393	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
394	struct net_device *pDev = qdisc_dev(sch);
395
396	pPriv->pVBoxNetFlt = NULL;
397
398	pPriv->ppSaved = kcalloc(QDISC_SAVED_NUM(pDev), sizeof(pPriv->ppSaved[0]),
399	GFP_KERNEL);
400	if (!pPriv->ppSaved)
401	return -ENOMEM;
402
403	pPriv->pChild = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
404	&pfifo_qdisc_ops,
405	TC_H_MAKE(TC_H_MAJ(sch->handle),
406	TC_H_MIN(1)));
407	if (!pPriv->pChild)
408	{
409	kfree(pPriv->ppSaved);
410	pPriv->ppSaved = NULL;
411	return -ENOMEM;
412	}
413
414	return 0;
415	}
416
417	static void vboxNetFltQdiscReset(struct Qdisc *sch)
418	{
419	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
420
421	qdisc_reset(pPriv->pChild);
422	sch->q.qlen = 0;
423	sch->qstats.backlog = 0;
424	}
425
426	static void vboxNetFltQdiscDestroy(struct Qdisc* sch)
427	{
428	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
429	struct net_device *pDev = qdisc_dev(sch);
430
431	qdisc_destroy(pPriv->pChild);
432	pPriv->pChild = NULL;
433
434	if (pPriv->ppSaved)
435	{
436	int i;
437	for (i = 0; i < QDISC_SAVED_NUM(pDev); i++)
438	if (pPriv->ppSaved[i])
439	qdisc_destroy(pPriv->ppSaved[i]);
440	kfree(pPriv->ppSaved);
441	pPriv->ppSaved = NULL;
442	}
443	}
444
445	static int vboxNetFltClassGraft(struct Qdisc sch, unsigned long arg, struct Qdisc pNew,
446	struct Qdisc **ppOld)
447	{
448	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
449
450	if (pNew == NULL)
451	pNew = &noop_qdisc;
452
453	sch_tree_lock(sch);
454	*ppOld = pPriv->pChild;
455	pPriv->pChild = pNew;
456	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
457	sch->q.qlen = 0;
458	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
459	qdisc_tree_decrease_qlen(ppOld, (ppOld)->q.qlen);
460	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
461	qdisc_reset(*ppOld);
462	sch_tree_unlock(sch);
463
464	return 0;
465	}
466
467	static struct Qdisc vboxNetFltClassLeaf(struct Qdisc sch, unsigned long arg)
468	{
469	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
470	return pPriv->pChild;
471	}
472
473	static unsigned long vboxNetFltClassGet(struct Qdisc *sch, u32 classid)
474	{
475	return 1;
476	}
477
478	static void vboxNetFltClassPut(struct Qdisc *sch, unsigned long arg)
479	{
480	}
481
482	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
483	static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
484	struct rtattr *tca, unsigned long arg)
485	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
486	static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
487	struct nlattr *tca, unsigned long arg)
488	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
489	{
490	return -ENOSYS;
491	}
492
493	static int vboxNetFltClassDelete(struct Qdisc *sch, unsigned long arg)
494	{
495	return -ENOSYS;
496	}
497
498	static void vboxNetFltClassWalk(struct Qdisc sch, struct qdisc_walker walker)
499	{
500	if (!walker->stop) {
501	if (walker->count >= walker->skip)
502	if (walker->fn(sch, 1, walker) < 0) {
503	walker->stop = 1;
504	return;
505	}
506	walker->count++;
507	}
508	}
509
510	static struct tcf_proto *vboxNetFltClassFindTcf(struct Qdisc sch, unsigned long cl)
511	{
512	return NULL;
513	}
514
515	static int vboxNetFltClassDump(struct Qdisc *sch, unsigned long cl,
516	struct sk_buff skb, struct tcmsg tcm)
517	{
518	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
519
520	if (cl != 1)
521	return -ENOENT;
522
523	tcm->tcm_handle \|= TC_H_MIN(1);
524	tcm->tcm_info = pPriv->pChild->handle;
525
526	return 0;
527	}
528
529
530	static struct Qdisc_class_ops g_VBoxNetFltClassOps =
531	{
532	.graft = vboxNetFltClassGraft,
533	.leaf = vboxNetFltClassLeaf,
534	.get = vboxNetFltClassGet,
535	.put = vboxNetFltClassPut,
536	.change = vboxNetFltClassChange,
537	.delete = vboxNetFltClassDelete,
538	.walk = vboxNetFltClassWalk,
539	.tcf_chain = vboxNetFltClassFindTcf,
540	.dump = vboxNetFltClassDump,
541	};
542
543
544	static struct Qdisc_ops g_VBoxNetFltQDiscOps = {
545	.cl_ops = &g_VBoxNetFltClassOps,
546	.id = "vboxnetflt",
547	.priv_size = sizeof(struct VBoxNetQDiscPriv),
548	.enqueue = vboxNetFltQdiscEnqueue,
549	.dequeue = vboxNetFltQdiscDequeue,
550	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
551	.requeue = vboxNetFltQdiscRequeue,
552	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
553	.peek = qdisc_peek_dequeued,
554	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
555	.drop = vboxNetFltQdiscDrop,
556	.init = vboxNetFltQdiscInit,
557	.reset = vboxNetFltQdiscReset,
558	.destroy = vboxNetFltQdiscDestroy,
559	.owner = THIS_MODULE
560	};
561
562	/*
563	* If our qdisc is already attached to the device (that means the user
564	* installed it from command line with 'tc' command) we simply update
565	* the pointer to vboxnetflt instance in qdisc's private structure.
566	* Otherwise we need to take some additional steps:
567	* - Create our qdisc;
568	* - Save all references to qdiscs;
569	* - Replace our child with the first qdisc reference;
570	* - Replace all references so they point to our qdisc.
571	*/
572	static void vboxNetFltLinuxQdiscInstall(PVBOXNETFLTINS pThis, struct net_device *pDev)
573	{
574	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
575	int i;
576	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
577	PVBOXNETQDISCPRIV pPriv;
578
579	struct Qdisc *pExisting = QDISC_GET(pDev);
580	if (strcmp(pExisting->ops->id, "vboxnetflt"))
581	{
582	/* The existing qdisc is different from ours, let's create new one. */
583	struct Qdisc *pNew = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
584	&g_VBoxNetFltQDiscOps, TC_H_ROOT);
585	if (!pNew)
586	return; // TODO: Error?
587
588	if (!try_module_get(THIS_MODULE))
589	{
590	/*
591	* This may cause a memory leak but calling qdisc_destroy()
592	* is not an option as it will call module_put().
593	*/
594	return;
595	}
596	pPriv = qdisc_priv(pNew);
597
598	qdisc_destroy(pPriv->pChild);
599	pPriv->pChild = QDISC_GET(pDev);
600	atomic_inc(&pPriv->pChild->refcnt);
601	/*
602	* There is no need in deactivating the device or acquiring any locks
603	* prior changing qdiscs since we do not destroy the old qdisc.
604	* Atomic replacement of pointers is enough.
605	*/
606	/*
607	* No need to change reference counters here as we merely move
608	* the pointer and the reference counter of the newly allocated
609	* qdisc is already 1.
610	*/
611	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
612	pPriv->ppSaved[0] = pDev->qdisc_sleeping;
613	ASMAtomicWritePtr(&pDev->qdisc_sleeping, pNew);
614	ASMAtomicWritePtr(&pDev->qdisc, pNew);
615	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
616	for (i = 0; i < pDev->num_tx_queues; i++)
617	{
618	struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
619
620	pPriv->ppSaved[i] = pQueue->qdisc_sleeping;
621	ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pNew);
622	ASMAtomicWritePtr(&pQueue->qdisc, pNew);
623	if (i)
624	atomic_inc(&pNew->refcnt);
625	}
626	/* Newer kernels store root qdisc in netdev structure as well. */
627	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
628	pPriv->ppSaved[pDev->num_tx_queues] = pDev->qdisc;
629	ASMAtomicWritePtr(&pDev->qdisc, pNew);
630	atomic_inc(&pNew->refcnt);
631	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
632	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
633	/* Synch the queue len with our child */
634	pNew->q.qlen = pPriv->pChild->q.qlen;
635	}
636	else
637	{
638	/* We already have vboxnetflt qdisc, let's use it. */
639	pPriv = qdisc_priv(pExisting);
640	}
641	ASMAtomicWritePtr(&pPriv->pVBoxNetFlt, pThis);
642	QDISC_LOG(("vboxNetFltLinuxInstallQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
643	}
644
645	static void vboxNetFltLinuxQdiscRemove(PVBOXNETFLTINS pThis, struct net_device *pDev)
646	{
647	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
648	int i;
649	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
650	PVBOXNETQDISCPRIV pPriv;
651	struct Qdisc pQdisc, pChild;
652	if (!pDev)
653	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
654	if (!VALID_PTR(pDev))
655	{
656	printk("VBoxNetFlt: Failed to detach qdisc, invalid device pointer: %p\n",
657	pDev);
658	return; // TODO: Consider returing an error
659	}
660
661
662	pQdisc = QDISC_GET(pDev);
663	if (strcmp(pQdisc->ops->id, "vboxnetflt"))
664	{
665	/* Looks like the user has replaced our qdisc manually. */
666	printk("VBoxNetFlt: Failed to detach qdisc, wrong qdisc: %s\n",
667	pQdisc->ops->id);
668	return; // TODO: Consider returing an error
669	}
670
671	pPriv = qdisc_priv(pQdisc);
672	Assert(pPriv->pVBoxNetFlt == pThis);
673	ASMAtomicWriteNullPtr(&pPriv->pVBoxNetFlt);
674	pChild = ASMAtomicXchgPtrT(&pPriv->pChild, &noop_qdisc, struct Qdisc *);
675	qdisc_destroy(pChild); /* It won't be the last reference. */
676
677	QDISC_LOG(("vboxNetFltLinuxQdiscRemove: refcnt=%d num_tx_queues=%d\n",
678	atomic_read(&pQdisc->refcnt), pDev->num_tx_queues));
679	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
680	/* Play it safe, make sure the qdisc is not being used. */
681	if (pPriv->ppSaved[0])
682	{
683	ASMAtomicWritePtr(&pDev->qdisc_sleeping, pPriv->ppSaved[0]);
684	ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[0]);
685	pPriv->ppSaved[0] = NULL;
686	while (QDISC_IS_BUSY(pDev, pQdisc))
687	yield();
688	qdisc_destroy(pQdisc); /* Destroy reference */
689	}
690	# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
691	for (i = 0; i < pDev->num_tx_queues; i++)
692	{
693	struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
694	if (pPriv->ppSaved[i])
695	{
696	Assert(pQueue->qdisc_sleeping == pQdisc);
697	ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pPriv->ppSaved[i]);
698	ASMAtomicWritePtr(&pQueue->qdisc, pPriv->ppSaved[i]);
699	pPriv->ppSaved[i] = NULL;
700	while (QDISC_IS_BUSY(pDev, pQdisc))
701	yield();
702	qdisc_destroy(pQdisc); /* Destroy reference */
703	}
704	}
705	/* Newer kernels store root qdisc in netdev structure as well. */
706	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
707	ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[pDev->num_tx_queues]);
708	pPriv->ppSaved[pDev->num_tx_queues] = NULL;
709	while (QDISC_IS_BUSY(pDev, pQdisc))
710	yield();
711	qdisc_destroy(pQdisc); /* Destroy reference */
712	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
713	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
714
715	/*
716	* At this point all references to our qdisc should be gone
717	* unless the user had installed it manually.
718	*/
719	QDISC_LOG(("vboxNetFltLinuxRemoveQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
720	}
721
722	#endif /* VBOXNETFLT_WITH_QDISC */
723
724
725	/**
726	* Initialize module.
727	*
728	* @returns appropriate status code.
729	*/
730	static int __init VBoxNetFltLinuxInit(void)
731	{
732	int rc;
733	/*
734	* Initialize IPRT.
735	*/
736	rc = RTR0Init(0);
737	if (RT_SUCCESS(rc))
738	{
739	Log(("VBoxNetFltLinuxInit\n"));
740
741	/*
742	* Initialize the globals and connect to the support driver.
743	*
744	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
745	* for establishing the connect to the support driver.
746	*/
747	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
748	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
749	if (RT_SUCCESS(rc))
750	{
751	#ifdef VBOXNETFLT_WITH_QDISC
752	/*memcpy(&g_VBoxNetFltQDiscOps, &pfifo_qdisc_ops, sizeof(g_VBoxNetFltQDiscOps));
753	strcpy(g_VBoxNetFltQDiscOps.id, "vboxnetflt");
754	g_VBoxNetFltQDiscOps.owner = THIS_MODULE;*/
755	rc = register_qdisc(&g_VBoxNetFltQDiscOps);
756	if (rc)
757	{
758	LogRel(("VBoxNetFlt: Failed to registed qdisc: %d\n", rc));
759	return rc;
760	}
761	#endif /* VBOXNETFLT_WITH_QDISC */
762	LogRel(("VBoxNetFlt: Successfully started.\n"));
763	return 0;
764	}
765
766	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
767	RTR0Term();
768	}
769	else
770	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
771
772	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
773	return -RTErrConvertToErrno(rc);
774	}
775
776
777	/**
778	* Unload the module.
779	*
780	* @todo We have to prevent this if we're busy!
781	*/
782	static void __exit VBoxNetFltLinuxUnload(void)
783	{
784	int rc;
785	Log(("VBoxNetFltLinuxUnload\n"));
786	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
787
788	#ifdef VBOXNETFLT_WITH_QDISC
789	unregister_qdisc(&g_VBoxNetFltQDiscOps);
790	#endif /* VBOXNETFLT_WITH_QDISC */
791	/*
792	* Undo the work done during start (in reverse order).
793	*/
794	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
795	AssertRC(rc); NOREF(rc);
796
797	RTR0Term();
798
799	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
800
801	Log(("VBoxNetFltLinuxUnload - done\n"));
802	}
803
804
805	/**
806	* Experiment where we filter trafic from the host to the internal network
807	* before it reaches the NIC driver.
808	*
809	* The current code uses a very ugly hack and only works on kernels using the
810	* net_device_ops (>= 2.6.29). It has been shown to give us a
811	* performance boost of 60-100% though. So, we have to find some less hacky way
812	* of getting this job done eventually.
813	*
814	* #define VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
815	*/
816	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
817
818	/**
819	* The overridden net_device_ops of the device we're attached to.
820	*
821	* Requires Linux 2.6.29 or later.
822	*
823	* This is a very dirty hack that was create to explore how much we can improve
824	* the host to guest transfers by not CC'ing the NIC.
825	*/
826	typedef struct VBoxNetDeviceOpsOverride
827	{
828	/** Our overridden ops. */
829	struct net_device_ops Ops;
830	/** Magic word. */
831	uint32_t u32Magic;
832	/** Pointer to the original ops. */
833	struct net_device_ops const *pOrgOps;
834	/** Pointer to the net filter instance. */
835	PVBOXNETFLTINS pVBoxNetFlt;
836	/** The number of filtered packages. */
837	uint64_t cFiltered;
838	/** The total number of packets */
839	uint64_t cTotal;
840	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
841	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
842	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
843
844	/**
845	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
846	* because they belong on the internal network.
847	*
848	* @returns NETDEV_TX_XXX.
849	* @param pSkb The socket buffer to transmit.
850	* @param pDev The net device.
851	*/
852	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
853	{
854	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->netdev_ops;
855	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
856	PCRTNETETHERHDR pEtherHdr;
857	PINTNETTRUNKSWPORT pSwitchPort;
858	uint32_t cbHdrs;
859
860
861	/*
862	* Validate the override structure.
863	*
864	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
865	* to be production quality code, we would have to be much more
866	* careful here and avoid the race.
867	*/
868	if ( !VALID_PTR(pOverride)
869	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
870	\|\| !VALID_PTR(pOverride->pOrgOps))
871	{
872	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
873	dev_kfree_skb(pSkb);
874	return NETDEV_TX_OK;
875	}
876	pOverride->cTotal++;
877
878	/*
879	* Do the filtering base on the defaul OUI of our virtual NICs
880	*
881	* Note! In a real solution, we would ask the switch whether the
882	* destination MAC is 100% to be on the internal network and then
883	* drop it.
884	*/
885	cbHdrs = skb_headlen(pSkb);
886	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
887	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
888	if ( pEtherHdr
889	&& VALID_PTR(pOverride->pVBoxNetFlt)
890	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
891	&& VALID_PTR(pSwitchPort)
892	&& cbHdrs >= 6)
893	{
894	INTNETSWDECISION enmDecision;
895
896	/** @todo consider reference counting, etc. */
897	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
898	if (enmDecision == INTNETSWDECISION_INTNET)
899	{
900	dev_kfree_skb(pSkb);
901	pOverride->cFiltered++;
902	return NETDEV_TX_OK;
903	}
904	}
905
906	return pOverride->pOrgOps->ndo_start_xmit(pSkb, pDev);
907	}
908
909	/**
910	* Hooks the device ndo_start_xmit operation of the device.
911	*
912	* @param pThis The net filter instance.
913	* @param pDev The net device.
914	*/
915	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
916	{
917	PVBOXNETDEVICEOPSOVERRIDE pOverride;
918	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
919
920	pOverride = RTMemAlloc(sizeof(*pOverride));
921	if (!pOverride)
922	return;
923	pOverride->pOrgOps = pDev->netdev_ops;
924	pOverride->Ops = *pDev->netdev_ops;
925	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
926	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
927	pOverride->cTotal = 0;
928	pOverride->cFiltered = 0;
929	pOverride->pVBoxNetFlt = pThis;
930
931	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp); /* (this isn't necessary, but so what) */
932	ASMAtomicWritePtr((void * volatile *)&pDev->netdev_ops, pOverride);
933	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
934	}
935
936	/**
937	* Undos what vboxNetFltLinuxHookDev did.
938	*
939	* @param pThis The net filter instance.
940	* @param pDev The net device. Can be NULL, in which case
941	* we'll try retrieve it from @a pThis.
942	*/
943	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
944	{
945	PVBOXNETDEVICEOPSOVERRIDE pOverride;
946	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
947
948	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
949	if (!pDev)
950	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
951	if (VALID_PTR(pDev))
952	{
953	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->netdev_ops;
954	if ( VALID_PTR(pOverride)
955	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
956	&& VALID_PTR(pOverride->pOrgOps)
957	)
958	{
959	ASMAtomicWritePtr((void * volatile *)&pDev->netdev_ops, pOverride->pOrgOps);
960	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
961	}
962	else
963	pOverride = NULL;
964	}
965	else
966	pOverride = NULL;
967	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
968
969	if (pOverride)
970	{
971	printk("vboxnetflt: dropped %llu out of %llu packets\n", pOverride->cFiltered, pOverride->cTotal);
972	RTMemFree(pOverride);
973	}
974	}
975
976	#endif /* VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT */
977
978
979	/**
980	* Reads and retains the host interface handle.
981	*
982	* @returns The handle, NULL if detached.
983	* @param pThis
984	*/
985	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
986	{
987	#if 0
988	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
989	struct net_device *pDev = NULL;
990
991	Log(("vboxNetFltLinuxRetainNetDev\n"));
992	/*
993	* Be careful here to avoid problems racing the detached callback.
994	*/
995	RTSpinlockAcquire(pThis->hSpinlock, &Tmp);
996	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
997	{
998	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
999	if (pDev)
1000	{
1001	dev_hold(pDev);
1002	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1003	}
1004	}
1005	RTSpinlockRelease(pThis->hSpinlock, &Tmp);
1006
1007	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
1008	return pDev;
1009	#else
1010	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1011	#endif
1012	}
1013
1014
1015	/**
1016	* Release the host interface handle previously retained
1017	* by vboxNetFltLinuxRetainNetDev.
1018	*
1019	* @param pThis The instance.
1020	* @param pDev The vboxNetFltLinuxRetainNetDev
1021	* return value, NULL is fine.
1022	*/
1023	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
1024	{
1025	#if 0
1026	Log(("vboxNetFltLinuxReleaseNetDev\n"));
1027	NOREF(pThis);
1028	if (pDev)
1029	{
1030	dev_put(pDev);
1031	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1032	}
1033	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
1034	#endif
1035	}
1036
1037	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
1038	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
1039
1040	/**
1041	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
1042	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
1043	*
1044	* @returns true / false accordingly.
1045	* @param pBuf The sk_buff.
1046	*/
1047	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
1048	{
1049	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
1050	}
1051
1052
1053	/**
1054	* Internal worker that create a linux sk_buff for a
1055	* (scatter/)gather list.
1056	*
1057	* @returns Pointer to the sk_buff.
1058	* @param pThis The instance.
1059	* @param pSG The (scatter/)gather list.
1060	* @param fDstWire Set if the destination is the wire.
1061	*/
1062	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
1063	{
1064	struct sk_buff *pPkt;
1065	struct net_device *pDev;
1066	unsigned fGsoType = 0;
1067
1068	if (pSG->cbTotal == 0)
1069	{
1070	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
1071	return NULL;
1072	}
1073
1074	/** @todo We should use fragments mapping the SG buffers with large packets.
1075	* 256 bytes seems to be the a threshold used a lot for this. It
1076	* requires some nasty work on the intnet side though... */
1077	/*
1078	* Allocate a packet and copy over the data.
1079	*/
1080	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1081	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
1082	if (RT_UNLIKELY(!pPkt))
1083	{
1084	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
1085	pSG->pvUserData = NULL;
1086	return NULL;
1087	}
1088	pPkt->dev = pDev;
1089	pPkt->ip_summed = CHECKSUM_NONE;
1090
1091	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
1092	skb_reserve(pPkt, NET_IP_ALIGN);
1093
1094	/* Copy the segments. */
1095	skb_put(pPkt, pSG->cbTotal);
1096	IntNetSgRead(pSG, pPkt->data);
1097
1098	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1099	/*
1100	* Setup GSO if used by this packet.
1101	*/
1102	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
1103	{
1104	default:
1105	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
1106	/* fall thru */
1107	case PDMNETWORKGSOTYPE_INVALID:
1108	fGsoType = 0;
1109	break;
1110	case PDMNETWORKGSOTYPE_IPV4_TCP:
1111	fGsoType = SKB_GSO_TCPV4;
1112	break;
1113	case PDMNETWORKGSOTYPE_IPV4_UDP:
1114	fGsoType = SKB_GSO_UDP;
1115	break;
1116	case PDMNETWORKGSOTYPE_IPV6_TCP:
1117	fGsoType = SKB_GSO_TCPV6;
1118	break;
1119	}
1120	if (fGsoType)
1121	{
1122	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
1123
1124	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
1125	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
1126	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
1127
1128	/*
1129	* We need to set checksum fields even if the packet goes to the host
1130	* directly as it may be immediately forwared by IP layer @bugref{5020}.
1131	*/
1132	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrs);
1133	pPkt->ip_summed = CHECKSUM_PARTIAL;
1134	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1135	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
1136	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
1137	pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
1138	else
1139	pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
1140	# else
1141	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
1142	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
1143	pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
1144	else
1145	pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
1146	# endif
1147	if (!fDstWire)
1148	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
1149	}
1150	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
1151
1152	/*
1153	* Finish up the socket buffer.
1154	*/
1155	pPkt->protocol = eth_type_trans(pPkt, pDev);
1156	if (fDstWire)
1157	{
1158	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
1159
1160	/* Restore ethernet header back. */
1161	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
1162	VBOX_SKB_RESET_MAC_HDR(pPkt);
1163	}
1164	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
1165
1166	return pPkt;
1167	}
1168
1169
1170	/**
1171	* Initializes a SG list from an sk_buff.
1172	*
1173	* @returns Number of segments.
1174	* @param pThis The instance.
1175	* @param pBuf The sk_buff.
1176	* @param pSG The SG.
1177	* @param pvFrame The frame pointer, optional.
1178	* @param cSegs The number of segments allocated for the SG.
1179	* This should match the number in the mbuf exactly!
1180	* @param fSrc The source of the frame.
1181	* @param pGso Pointer to the GSO context if it's a GSO
1182	* internal network frame. NULL if regular frame.
1183	*/
1184	DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
1185	unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1186	{
1187	int i;
1188	NOREF(pThis);
1189
1190	Assert(!skb_shinfo(pBuf)->frag_list);
1191
1192	if (!pGsoCtx)
1193	IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /cSegsUsed/);
1194	else
1195	IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /cSegsUsed/, pGsoCtx);
1196
1197	#ifdef VBOXNETFLT_SG_SUPPORT
1198	pSG->aSegs[0].cb = skb_headlen(pBuf);
1199	pSG->aSegs[0].pv = pBuf->data;
1200	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1201
1202	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1203	{
1204	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
1205	pSG->aSegs[i+1].cb = pFrag->size;
1206	pSG->aSegs[i+1].pv = kmap(pFrag->page);
1207	printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
1208	pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
1209	}
1210	++i;
1211
1212	#else
1213	pSG->aSegs[0].cb = pBuf->len;
1214	pSG->aSegs[0].pv = pBuf->data;
1215	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1216	i = 1;
1217	#endif
1218
1219	pSG->cSegsUsed = i;
1220
1221	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1222	/*
1223	* Add a trailer if the frame is too small.
1224	*
1225	* Since we're getting to the packet before it is framed, it has not
1226	* yet been padded. The current solution is to add a segment pointing
1227	* to a buffer containing all zeros and pray that works for all frames...
1228	*/
1229	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
1230	{
1231	static uint8_t const s_abZero[128] = {0};
1232
1233	AssertReturnVoid(i < cSegs);
1234
1235	pSG->aSegs[i].Phys = NIL_RTHCPHYS;
1236	pSG->aSegs[i].pv = (void *)&s_abZero[0];
1237	pSG->aSegs[i].cb = 60 - pSG->cbTotal;
1238	pSG->cbTotal = 60;
1239	pSG->cSegsUsed++;
1240	Assert(i + 1 <= pSG->cSegsAlloc)
1241	}
1242	#endif
1243
1244	Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
1245	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
1246	for (i = 0; i < pSG->cSegsUsed; i++)
1247	Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
1248	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1249	}
1250
1251	/**
1252	* Packet handler,
1253	*
1254	* @returns 0 or EJUSTRETURN.
1255	* @param pThis The instance.
1256	* @param pMBuf The mbuf.
1257	* @param pvFrame The start of the frame, optional.
1258	* @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
1259	* @param eProtocol The protocol.
1260	*/
1261	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
1262	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1263	struct net_device *pSkbDev,
1264	struct packet_type *pPacketType,
1265	struct net_device *pOrigDev)
1266	#else
1267	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1268	struct net_device *pSkbDev,
1269	struct packet_type *pPacketType)
1270	#endif
1271	{
1272	PVBOXNETFLTINS pThis;
1273	struct net_device *pDev;
1274	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1275	pBuf, pSkbDev, pPacketType));
1276	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1277	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",
1278	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));
1279	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1280	#else
1281	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",
1282	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));
1283	#endif
1284	/*
1285	* Drop it immediately?
1286	*/
1287	if (!pBuf)
1288	return 0;
1289
1290	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1291	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1292	if (pThis->u.s.pDev != pSkbDev)
1293	{
1294	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1295	return 0;
1296	}
1297
1298	Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1299	if (vboxNetFltLinuxSkBufIsOur(pBuf))
1300	{
1301	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1302	dev_kfree_skb(pBuf);
1303	return 0;
1304	}
1305
1306	#ifndef VBOXNETFLT_SG_SUPPORT
1307	{
1308	/*
1309	* Get rid of fragmented packets, they cause too much trouble.
1310	*/
1311	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1312	kfree_skb(pBuf);
1313	if (!pCopy)
1314	{
1315	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1316	return 0;
1317	}
1318	pBuf = pCopy;
1319	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1320	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",
1321	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));
1322	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1323	# else
1324	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",
1325	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));
1326	# endif
1327	}
1328	#endif
1329
1330	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1331	/* Forward it to the internal network. */
1332	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1333	#else
1334	/* Add the packet to transmit queue and schedule the bottom half. */
1335	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1336	schedule_work(&pThis->u.s.XmitTask);
1337	Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1338	&pThis->u.s.XmitTask, pBuf));
1339	#endif
1340
1341	/* It does not really matter what we return, it is ignored by the kernel. */
1342	return 0;
1343	}
1344
1345	/**
1346	* Calculate the number of INTNETSEG segments the socket buffer will need.
1347	*
1348	* @returns Segment count.
1349	* @param pBuf The socket buffer.
1350	*/
1351	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
1352	{
1353	#ifdef VBOXNETFLT_SG_SUPPORT
1354	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1355	#else
1356	unsigned cSegs = 1;
1357	#endif
1358	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1359	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1360	if (pBuf->len < 60)
1361	cSegs++;
1362	#endif
1363	return cSegs;
1364	}
1365
1366	/**
1367	* Destroy the intnet scatter / gather buffer created by
1368	* vboxNetFltLinuxSkBufToSG.
1369	*/
1370	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
1371	{
1372	#ifdef VBOXNETFLT_SG_SUPPORT
1373	int i;
1374
1375	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1376	{
1377	printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
1378	kunmap(pSG->aSegs[i+1].pv);
1379	}
1380	#endif
1381	NOREF(pSG);
1382	}
1383
1384	#ifdef LOG_ENABLED
1385	/**
1386	* Logging helper.
1387	*/
1388	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1389	{
1390	uint8_t pInt, pExt;
1391	static int iPacketNo = 1;
1392	iPacketNo += iIncrement;
1393	if (fEgress)
1394	{
1395	pExt = pSG->aSegs[0].pv;
1396	pInt = pExt + 6;
1397	}
1398	else
1399	{
1400	pInt = pSG->aSegs[0].pv;
1401	pExt = pInt + 6;
1402	}
1403	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1404	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1405	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1406	fEgress ? "-->" : "<--", pszWhere,
1407	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1408	pSG->cbTotal, iPacketNo));
1409	Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1410	}
1411	#else
1412	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1413	#endif
1414
1415	#ifdef VBOXNETFLT_WITH_GSO_RECV
1416
1417	/**
1418	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1419	* GSO socket buffer without having to segment it.
1420	*
1421	* @returns true on success, false if needs segmenting.
1422	* @param pThis The net filter instance.
1423	* @param pSkb The GSO socket buffer.
1424	* @param fSrc The source.
1425	* @param pGsoCtx Where to return the GSO context on success.
1426	*/
1427	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1428	PPDMNETWORKGSO pGsoCtx)
1429	{
1430	PDMNETWORKGSOTYPE enmGsoType;
1431	uint16_t uEtherType;
1432	unsigned int cbTransport;
1433	unsigned int offTransport;
1434	unsigned int cbTransportHdr;
1435	unsigned uProtocol;
1436	union
1437	{
1438	RTNETIPV4 IPv4;
1439	RTNETIPV6 IPv6;
1440	RTNETTCP Tcp;
1441	uint8_t ab[40];
1442	uint16_t au16[40/2];
1443	uint32_t au32[40/4];
1444	} Buf;
1445
1446	/*
1447	* Check the GSO properties of the socket buffer and make sure it fits.
1448	*/
1449	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1450	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP \| SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
1451	{
1452	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1453	return false;
1454	}
1455	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1456	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
1457	{
1458	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1459	return false;
1460	}
1461	/*
1462	* It is possible to receive GSO packets from wire if GRO is enabled.
1463	*/
1464	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1465	{
1466	Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1467	#ifdef VBOXNETFLT_WITH_GRO
1468	/*
1469	* The packet came from the wire and the driver has already consumed
1470	* mac header. We need to restore it back.
1471	*/
1472	pSkb->mac_len = skb_network_header(pSkb) - skb_mac_header(pSkb);
1473	skb_push(pSkb, pSkb->mac_len);
1474	Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1475	pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
1476	#else /* !VBOXNETFLT_WITH_GRO */
1477	/* Older kernels didn't have GRO. */
1478	return false;
1479	#endif /* !VBOXNETFLT_WITH_GRO */
1480	}
1481	else
1482	{
1483	/*
1484	* skb_gso_segment does the following. Do we need to do it as well?
1485	*/
1486	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1487	skb_reset_mac_header(pSkb);
1488	pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1489	#else
1490	pSkb->mac.raw = pSkb->data;
1491	pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1492	#endif
1493	}
1494
1495	/*
1496	* Switch on the ethertype.
1497	*/
1498	uEtherType = pSkb->protocol;
1499	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1500	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1501	{
1502	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1503	if (puEtherType)
1504	uEtherType = *puEtherType;
1505	}
1506	switch (uEtherType)
1507	{
1508	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1509	{
1510	unsigned int cbHdr;
1511	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1512	if (RT_UNLIKELY(!pIPv4))
1513	{
1514	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1515	return false;
1516	}
1517
1518	cbHdr = pIPv4->ip_hl * 4;
1519	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1520	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1521	\|\| cbHdr > cbTransport ))
1522	{
1523	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1524	return false;
1525	}
1526	cbTransport -= cbHdr;
1527	offTransport = pSkb->mac_len + cbHdr;
1528	uProtocol = pIPv4->ip_p;
1529	if (uProtocol == RTNETIPV4_PROT_TCP)
1530	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1531	else if (uProtocol == RTNETIPV4_PROT_UDP)
1532	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1533	else /** @todo IPv6: 4to6 tunneling */
1534	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1535	break;
1536	}
1537
1538	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1539	{
1540	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1541	if (RT_UNLIKELY(!pIPv6))
1542	{
1543	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1544	return false;
1545	}
1546
1547	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1548	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1549	uProtocol = pIPv6->ip6_nxt;
1550	/** @todo IPv6: Dig our way out of the other headers. */
1551	if (uProtocol == RTNETIPV4_PROT_TCP)
1552	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1553	else if (uProtocol == RTNETIPV4_PROT_UDP)
1554	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1555	else
1556	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1557	break;
1558	}
1559
1560	default:
1561	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1562	return false;
1563	}
1564
1565	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1566	{
1567	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1568	return false;
1569	}
1570
1571	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1572	\|\| offTransport + cbTransport > pSkb->len
1573	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1574	{
1575	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1576	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1577	return false;
1578	}
1579
1580	/*
1581	* Check the TCP/UDP bits.
1582	*/
1583	if (uProtocol == RTNETIPV4_PROT_TCP)
1584	{
1585	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1586	if (RT_UNLIKELY(!pTcp))
1587	{
1588	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1589	return false;
1590	}
1591
1592	cbTransportHdr = pTcp->th_off * 4;
1593	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1594	\|\| cbTransportHdr > cbTransport
1595	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1596	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1597	{
1598	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1599	return false;
1600	}
1601
1602	}
1603	else
1604	{
1605	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1606	cbTransportHdr = sizeof(RTNETUDP);
1607	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1608	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1609	{
1610	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1611	return false;
1612	}
1613	}
1614
1615	/*
1616	* We're good, init the GSO context.
1617	*/
1618	pGsoCtx->u8Type = enmGsoType;
1619	pGsoCtx->cbHdrs = offTransport + cbTransportHdr;
1620	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1621	pGsoCtx->offHdr1 = pSkb->mac_len;
1622	pGsoCtx->offHdr2 = offTransport;
1623	pGsoCtx->au8Unused[0] = 0;
1624	pGsoCtx->au8Unused[1] = 0;
1625
1626	return true;
1627	}
1628
1629	/**
1630	* Forward the socket buffer as a GSO internal network frame.
1631	*
1632	* @returns IPRT status code.
1633	* @param pThis The net filter instance.
1634	* @param pSkb The GSO socket buffer.
1635	* @param fSrc The source.
1636	* @param pGsoCtx Where to return the GSO context on success.
1637	*/
1638	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1639	{
1640	int rc;
1641	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1642	if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1643	{
1644	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1645	if (RT_LIKELY(pSG))
1646	{
1647	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1648
1649	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1650	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1651
1652	vboxNetFltLinuxDestroySG(pSG);
1653	rc = VINF_SUCCESS;
1654	}
1655	else
1656	{
1657	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1658	rc = VERR_NO_MEMORY;
1659	}
1660	}
1661	else
1662	{
1663	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1664	rc = VERR_INTERNAL_ERROR_3;
1665	}
1666
1667	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1668	dev_kfree_skb(pSkb);
1669	return rc;
1670	}
1671
1672	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1673
1674	/**
1675	* Worker for vboxNetFltLinuxForwardToIntNet.
1676	*
1677	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1678	* @param pThis The net filter instance.
1679	* @param pBuf The socket buffer.
1680	* @param fSrc The source.
1681	*/
1682	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1683	{
1684	int rc;
1685	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1686	if (cSegs <= MAX_SKB_FRAGS + 1)
1687	{
1688	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1689	if (RT_LIKELY(pSG))
1690	{
1691	if (fSrc & INTNETTRUNKDIR_WIRE)
1692	{
1693	/*
1694	* The packet came from wire, ethernet header was removed by device driver.
1695	* Restore it.
1696	*/
1697	skb_push(pBuf, ETH_HLEN);
1698	}
1699
1700	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /pGsoCtx/);
1701
1702	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1703	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1704
1705	vboxNetFltLinuxDestroySG(pSG);
1706	rc = VINF_SUCCESS;
1707	}
1708	else
1709	{
1710	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1711	rc = VERR_NO_MEMORY;
1712	}
1713	}
1714	else
1715	{
1716	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1717	rc = VERR_INTERNAL_ERROR_3;
1718	}
1719
1720	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1721	dev_kfree_skb(pBuf);
1722	return rc;
1723	}
1724
1725	/**
1726	*
1727	* @param pBuf The socket buffer. This is consumed by this function.
1728	*/
1729	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1730	{
1731	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1732
1733	#ifdef VBOXNETFLT_WITH_GSO
1734	if (skb_is_gso(pBuf))
1735	{
1736	PDMNETWORKGSO GsoCtx;
1737	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",
1738	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));
1739	# ifdef VBOXNETFLT_WITH_GSO_RECV
1740	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1741	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1742	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1743	else
1744	# endif
1745	{
1746	/* Need to segment the packet */
1747	struct sk_buff *pNext;
1748	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1749	if (IS_ERR(pSegment))
1750	{
1751	dev_kfree_skb(pBuf);
1752	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1753	return;
1754	}
1755
1756	for (; pSegment; pSegment = pNext)
1757	{
1758	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",
1759	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));
1760	pNext = pSegment->next;
1761	pSegment->next = 0;
1762	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1763	}
1764	dev_kfree_skb(pBuf);
1765	}
1766	}
1767	else
1768	#endif /* VBOXNETFLT_WITH_GSO */
1769	{
1770	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1771	{
1772	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1773	/*
1774	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1775	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1776	* header length from the header itself and reconstruct 'h' pointer
1777	* to TCP (or whatever) header.
1778	*/
1779	unsigned char *tmp = pBuf->h.raw;
1780	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1781	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1782	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1783	if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1784	{
1785	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1786	dev_kfree_skb(pBuf);
1787	return;
1788	}
1789	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1790	/* Restore the original (wrong) pointer. */
1791	pBuf->h.raw = tmp;
1792	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1793	}
1794	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1795	}
1796	}
1797
1798	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1799	/**
1800	* Work queue handler that forwards the socket buffers queued by
1801	* vboxNetFltLinuxPacketHandler to the internal network.
1802	*
1803	* @param pWork The work queue.
1804	*/
1805	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1806	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1807	# else
1808	static void vboxNetFltLinuxXmitTask(void *pWork)
1809	# endif
1810	{
1811	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1812	struct sk_buff *pBuf;
1813
1814	Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1815
1816	/*
1817	* Active? Retain the instance and increment the busy counter.
1818	*/
1819	if (vboxNetFltTryRetainBusyActive(pThis))
1820	{
1821	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1822	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1823
1824	vboxNetFltRelease(pThis, true /* fBusy */);
1825	}
1826	else
1827	{
1828	/** @todo Shouldn't we just drop the packets here? There is little point in
1829	* making them accumulate when the VM is paused and it'll only waste
1830	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1831	* before start draining the packets (goes for the intnet ring buf
1832	* too)? */
1833	}
1834	}
1835	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1836
1837	/**
1838	* Reports the GSO capabilites of the hardware NIC.
1839	*
1840	* @param pThis The net filter instance. The caller hold a
1841	* reference to this.
1842	*/
1843	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1844	{
1845	#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1846	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1847	{
1848	struct net_device *pDev;
1849	PINTNETTRUNKSWPORT pSwitchPort;
1850	unsigned int fFeatures;
1851	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1852
1853	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1854
1855	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1856	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1857	if (pDev)
1858	fFeatures = pDev->features;
1859	else
1860	fFeatures = 0;
1861
1862	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1863
1864	if (pThis->pSwitchPort)
1865	{
1866	/* Set/update the GSO capabilities of the NIC. */
1867	uint32_t fGsoCapabilites = 0;
1868	if (fFeatures & NETIF_F_TSO)
1869	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1870	if (fFeatures & NETIF_F_TSO6)
1871	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1872	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1873	if (fFeatures & NETIF_F_UFO)
1874	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1875	if (fFeatures & NETIF_F_UFO)
1876	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1877	# endif
1878	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1879	}
1880
1881	vboxNetFltRelease(pThis, true /fBusy/);
1882	}
1883	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1884	}
1885
1886	/**
1887	* Helper that determins whether the host (ignoreing us) is operating the
1888	* interface in promiscuous mode or not.
1889	*/
1890	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1891	{
1892	bool fRc = false;
1893	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1894	if (pDev)
1895	{
1896	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1897	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1898	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1899	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1900	}
1901	return fRc;
1902	}
1903
1904	/**
1905	* Internal worker for vboxNetFltLinuxNotifierCallback.
1906	*
1907	* @returns VBox status code.
1908	* @param pThis The instance.
1909	* @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1910	* flood the release log.
1911	*/
1912	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1913	{
1914	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1915	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1916
1917	/*
1918	* Retain and store the device.
1919	*/
1920	dev_hold(pDev);
1921
1922	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1923	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1924	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1925
1926	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1927	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n", pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1928
1929	/* Get the mac address while we still have a valid net_device reference. */
1930	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1931
1932	/*
1933	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1934	*/
1935	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1936	pThis->u.s.PacketType.dev = pDev;
1937	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1938	dev_add_pack(&pThis->u.s.PacketType);
1939
1940	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1941	vboxNetFltLinuxHookDev(pThis, pDev);
1942	#endif
1943	#ifdef VBOXNETFLT_WITH_QDISC
1944	vboxNetFltLinuxQdiscInstall(pThis, pDev);
1945	#endif /* VBOXNETFLT_WITH_QDISC */
1946
1947	/*
1948	* Set indicators that require the spinlock. Be abit paranoid about racing
1949	* the device notification handle.
1950	*/
1951	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1952	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1953	if (pDev)
1954	{
1955	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1956	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1957	pDev = NULL; /* don't dereference it */
1958	}
1959	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1960	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1961
1962	/*
1963	* If the above succeeded report GSO capabilites, if not undo and
1964	* release the device.
1965	*/
1966	if (!pDev)
1967	{
1968	Assert(pThis->pSwitchPort);
1969	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1970	{
1971	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1972	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1973	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1974	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1975	vboxNetFltRelease(pThis, true /fBusy/);
1976	}
1977	}
1978	else
1979	{
1980	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1981	vboxNetFltLinuxUnhookDev(pThis, pDev);
1982	#endif
1983	#ifdef VBOXNETFLT_WITH_QDISC
1984	vboxNetFltLinuxQdiscRemove(pThis, pDev);
1985	#endif /* VBOXNETFLT_WITH_QDISC */
1986	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1987	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1988	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1989	dev_put(pDev);
1990	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1991	}
1992
1993	LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1994	return VINF_SUCCESS;
1995	}
1996
1997
1998	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1999	{
2000	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
2001
2002	Assert(!pThis->fDisconnectedFromHost);
2003
2004	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2005	vboxNetFltLinuxUnhookDev(pThis, pDev);
2006	#endif
2007	#ifdef VBOXNETFLT_WITH_QDISC
2008	vboxNetFltLinuxQdiscRemove(pThis, pDev);
2009	#endif /* VBOXNETFLT_WITH_QDISC */
2010
2011	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2012	ASMAtomicWriteBool(&pThis->u.s.fRegistered, false);
2013	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
2014	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2015	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2016
2017	dev_remove_pack(&pThis->u.s.PacketType);
2018	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2019	skb_queue_purge(&pThis->u.s.XmitQueue);
2020	#endif
2021	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
2022	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
2023	dev_put(pDev);
2024
2025	return NOTIFY_OK;
2026	}
2027
2028	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
2029	{
2030	/* Check if we are not suspended and promiscuous mode has not been set. */
2031	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
2032	&& !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
2033	{
2034	/* Note that there is no need for locking as the kernel got hold of the lock already. */
2035	dev_set_promiscuity(pDev, 1);
2036	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
2037	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2038	}
2039	else
2040	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2041	return NOTIFY_OK;
2042	}
2043
2044	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
2045	{
2046	/* Undo promiscuous mode if we has set it. */
2047	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
2048	{
2049	/* Note that there is no need for locking as the kernel got hold of the lock already. */
2050	dev_set_promiscuity(pDev, -1);
2051	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
2052	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2053	}
2054	else
2055	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2056	return NOTIFY_OK;
2057	}
2058
2059	#ifdef LOG_ENABLED
2060	/** Stringify the NETDEV_XXX constants. */
2061	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2062	{
2063	const char *pszEvent = "NETDRV_<unknown>";
2064	switch (ulEventType)
2065	{
2066	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2067	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2068	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2069	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2070	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2071	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2072	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2073	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2074	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2075	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2076	# ifdef NETDEV_FEAT_CHANGE
2077	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2078	# endif
2079	}
2080	return pszEvent;
2081	}
2082	#endif /* LOG_ENABLED */
2083
2084	/**
2085	* Callback for listening to netdevice events.
2086	*
2087	* This works the rediscovery, clean up on unregistration, promiscuity on
2088	* up/down, and GSO feature changes from ethtool.
2089	*
2090	* @returns NOTIFY_OK
2091	* @param self Pointer to our notifier registration block.
2092	* @param ulEventType The event.
2093	* @param ptr Event specific, but it is usually the device it
2094	* relates to.
2095	*/
2096	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2097
2098	{
2099	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2100	struct net_device pDev = (struct net_device )ptr;
2101	int rc = NOTIFY_OK;
2102
2103	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2104	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
2105	if ( ulEventType == NETDEV_REGISTER
2106	&& !strcmp(pDev->name, pThis->szName))
2107	{
2108	vboxNetFltLinuxAttachToInterface(pThis, pDev);
2109	}
2110	else
2111	{
2112	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2113	if (pDev == ptr)
2114	{
2115	switch (ulEventType)
2116	{
2117	case NETDEV_UNREGISTER:
2118	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2119	break;
2120	case NETDEV_UP:
2121	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2122	break;
2123	case NETDEV_GOING_DOWN:
2124	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2125	break;
2126	case NETDEV_CHANGENAME:
2127	break;
2128	#ifdef NETDEV_FEAT_CHANGE
2129	case NETDEV_FEAT_CHANGE:
2130	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2131	break;
2132	#endif
2133	}
2134	}
2135	}
2136
2137	return rc;
2138	}
2139
2140	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2141	{
2142	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2143	}
2144
2145	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2146	{
2147	struct net_device * pDev;
2148	int err;
2149	int rc = VINF_SUCCESS;
2150	NOREF(pvIfData);
2151
2152	LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2153
2154	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2155	if (pDev)
2156	{
2157	/*
2158	* Create a sk_buff for the gather list and push it onto the wire.
2159	*/
2160	if (fDst & INTNETTRUNKDIR_WIRE)
2161	{
2162	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2163	if (pBuf)
2164	{
2165	vboxNetFltDumpPacket(pSG, true, "wire", 1);
2166	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2167	Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2168	err = dev_queue_xmit(pBuf);
2169	if (err)
2170	rc = RTErrConvertFromErrno(err);
2171	}
2172	else
2173	rc = VERR_NO_MEMORY;
2174	}
2175
2176	/*
2177	* Create a sk_buff for the gather list and push it onto the host stack.
2178	*/
2179	if (fDst & INTNETTRUNKDIR_HOST)
2180	{
2181	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2182	if (pBuf)
2183	{
2184	vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2185	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2186	Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2187	err = netif_rx_ni(pBuf);
2188	if (err)
2189	rc = RTErrConvertFromErrno(err);
2190	}
2191	else
2192	rc = VERR_NO_MEMORY;
2193	}
2194
2195	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2196	}
2197
2198	return rc;
2199	}
2200
2201
2202	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2203	{
2204	struct net_device * pDev;
2205
2206	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
2207	pThis, pThis->szName, fActive?"true":"false",
2208	pThis->fDisablePromiscuous?"true":"false"));
2209
2210	if (pThis->fDisablePromiscuous)
2211	return;
2212
2213	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2214	if (pDev)
2215	{
2216	/*
2217	* This api is a bit weird, the best reference is the code.
2218	*
2219	* Also, we have a bit or race conditions wrt the maintance of
2220	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2221	*/
2222	#ifdef LOG_ENABLED
2223	u_int16_t fIf;
2224	unsigned const cPromiscBefore = pDev->promiscuity;
2225	#endif
2226	if (fActive)
2227	{
2228	Assert(!pThis->u.s.fPromiscuousSet);
2229
2230	rtnl_lock();
2231	dev_set_promiscuity(pDev, 1);
2232	rtnl_unlock();
2233	pThis->u.s.fPromiscuousSet = true;
2234	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2235	}
2236	else
2237	{
2238	if (pThis->u.s.fPromiscuousSet)
2239	{
2240	rtnl_lock();
2241	dev_set_promiscuity(pDev, -1);
2242	rtnl_unlock();
2243	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2244	}
2245	pThis->u.s.fPromiscuousSet = false;
2246
2247	#ifdef LOG_ENABLED
2248	fIf = dev_get_flags(pDev);
2249	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2250	#endif
2251	}
2252
2253	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2254	}
2255	}
2256
2257
2258	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2259	{
2260	#ifdef VBOXNETFLT_WITH_QDISC
2261	vboxNetFltLinuxQdiscRemove(pThis, NULL);
2262	#endif /* VBOXNETFLT_WITH_QDISC */
2263	/*
2264	* Remove packet handler when we get disconnected from internal switch as
2265	* we don't want the handler to forward packets to disconnected switch.
2266	*/
2267	dev_remove_pack(&pThis->u.s.PacketType);
2268	return VINF_SUCCESS;
2269	}
2270
2271
2272	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2273	{
2274	/*
2275	* Report the GSO capabilities of the host and device (if connected).
2276	* Note! No need to mark ourselves busy here.
2277	*/
2278	/** @todo duplicate work here now? Attach */
2279	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2280	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2281	0
2282	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2283	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2284	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2285	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2286	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2287	# endif
2288	, INTNETTRUNKDIR_HOST);
2289
2290	#endif
2291	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2292
2293	return VINF_SUCCESS;
2294	}
2295
2296
2297	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2298	{
2299	struct net_device *pDev;
2300	bool fRegistered;
2301	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
2302
2303	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2304	vboxNetFltLinuxUnhookDev(pThis, NULL);
2305	#endif
2306
2307	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2308	* unlikely, but none the less). Since it doesn't actually update the
2309	* state (just reads it), it is likely to panic in some interesting
2310	* ways. */
2311
2312	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2313	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2314	fRegistered = ASMAtomicUoReadBool(&pThis->u.s.fRegistered);
2315	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2316
2317	if (fRegistered)
2318	{
2319	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2320	skb_queue_purge(&pThis->u.s.XmitQueue);
2321	#endif
2322	Log(("vboxNetFltOsDeleteInstance: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
2323	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
2324	dev_put(pDev);
2325	}
2326	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2327	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2328	module_put(THIS_MODULE);
2329	}
2330
2331
2332	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2333	{
2334	int err;
2335	NOREF(pvContext);
2336
2337	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2338	err = register_netdevice_notifier(&pThis->u.s.Notifier);
2339	if (err)
2340	return VERR_INTNET_FLT_IF_FAILED;
2341	if (!pThis->u.s.fRegistered)
2342	{
2343	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2344	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2345	return VERR_INTNET_FLT_IF_NOT_FOUND;
2346	}
2347
2348	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2349	if ( pThis->fDisconnectedFromHost
2350	\|\| !try_module_get(THIS_MODULE))
2351	return VERR_INTNET_FLT_IF_FAILED;
2352
2353	return VINF_SUCCESS;
2354	}
2355
2356	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2357	{
2358	/*
2359	* Init the linux specific members.
2360	*/
2361	pThis->u.s.pDev = NULL;
2362	pThis->u.s.fRegistered = false;
2363	pThis->u.s.fPromiscuousSet = false;
2364	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2365	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2366	skb_queue_head_init(&pThis->u.s.XmitQueue);
2367	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2368	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2369	# else
2370	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2371	# endif
2372	#endif
2373
2374	return VINF_SUCCESS;
2375	}
2376
2377
2378	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2379	{
2380	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2381	}
2382
2383
2384	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
2385	{
2386	/* Nothing to do */
2387	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2388	return VINF_SUCCESS;
2389	}
2390
2391
2392	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2393	{
2394	/* Nothing to do */
2395	NOREF(pThis); NOREF(pvIfData);
2396	return VINF_SUCCESS;
2397	}
2398

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 31705

Download in other formats: