VBoxNetFlt-linux.c@ 30771

Last change on this file since 30771 was 30771, checked in by vboxsync, 15 years ago
vboxnetflt: RHEL 5 GRO fix
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 81.5 KB

Line
1	/* $Id: VBoxNetFlt-linux.c 30771 2010-07-09 17:26:55Z vboxsync $ */
2	/** @file
3	* VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2008 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/*******************************************************************************
19	* Header Files *
20	*******************************************************************************/
21	#define LOG_GROUP LOG_GROUP_NET_FLT_DRV
22	#define VBOXNETFLT_LINUX_NO_XMIT_QUEUE
23	#include "the-linux-kernel.h"
24	#include "version-generated.h"
25	#include "product-generated.h"
26	#include <linux/netdevice.h>
27	#include <linux/etherdevice.h>
28	#include <linux/rtnetlink.h>
29	#include <linux/miscdevice.h>
30	#include <linux/ip.h>
31
32	#include <VBox/log.h>
33	#include <VBox/err.h>
34	#include <VBox/intnetinline.h>
35	#include <VBox/pdmnetinline.h>
36	#include <VBox/param.h>
37	#include <iprt/alloca.h>
38	#include <iprt/assert.h>
39	#include <iprt/spinlock.h>
40	#include <iprt/semaphore.h>
41	#include <iprt/initterm.h>
42	#include <iprt/process.h>
43	#include <iprt/mem.h>
44	#include <iprt/net.h>
45	#include <iprt/log.h>
46	#include <iprt/mp.h>
47	#include <iprt/mem.h>
48	#include <iprt/time.h>
49
50	#define VBOXNETFLT_OS_SPECFIC 1
51	#include "../VBoxNetFltInternal.h"
52
53	/*
54	* Uncomment the following line to enable qdisc support.
55	*/
56	//#define VBOXNETFLT_WITH_QDISC
57	#ifdef VBOXNETFLT_WITH_QDISC
58	#include <net/pkt_sched.h>
59	#endif /* VBOXNETFLT_WITH_QDISC */
60
61
62	/*******************************************************************************
63	* Defined Constants And Macros *
64	*******************************************************************************/
65	#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
66	#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
67	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
68	# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
69	#endif
70
71	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
72	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
73	# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
74	#else
75	# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
76	# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
77	#endif
78
79	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
80	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
81	#else
82	# define CHECKSUM_PARTIAL CHECKSUM_HW
83	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
84	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
85	# else
86	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
87	# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
88	# else
89	# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
90	# endif
91	# endif
92	#endif
93
94	#ifndef NET_IP_ALIGN
95	# define NET_IP_ALIGN 2
96	#endif
97
98	#if 0
99	/** Create scatter / gather segments for fragments. When not used, we will
100	* linearize the socket buffer before creating the internal networking SG. */
101	# define VBOXNETFLT_SG_SUPPORT 1
102	#endif
103
104	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
105	/** Indicates that the linux kernel may send us GSO frames. */
106	# define VBOXNETFLT_WITH_GSO 1
107
108	/** This enables or disables the transmitting of GSO frame from the internal
109	* network and to the host. */
110	# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
111
112	# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
113	/** This enables or disables the transmitting of GSO frame from the internal
114	* network and to the wire. */
115	# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
116	# endif
117
118	/** This enables or disables the forwarding/flooding of GSO frame from the host
119	* to the internal network. */
120	# define VBOXNETFLT_WITH_GSO_RECV 1
121
122	#endif
123
124	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
125	/** This enables or disables handling of GSO frames coming from the wire (GRO). */
126	# define VBOXNETFLT_WITH_GRO 1
127	#endif
128	/*
129	* GRO support was backported to RHEL 5.4
130	*/
131	#ifdef RHEL_RELEASE_CODE
132	# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
133	# define VBOXNETFLT_WITH_GRO 1
134	# endif
135	#endif
136
137	/*******************************************************************************
138	* Internal Functions *
139	*******************************************************************************/
140	static int VBoxNetFltLinuxInit(void);
141	static void VBoxNetFltLinuxUnload(void);
142	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
143
144
145	/*******************************************************************************
146	* Global Variables *
147	*******************************************************************************/
148	/**
149	* The (common) global data.
150	*/
151	static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
152
153	module_init(VBoxNetFltLinuxInit);
154	module_exit(VBoxNetFltLinuxUnload);
155
156	MODULE_AUTHOR(VBOX_VENDOR);
157	MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
158	MODULE_LICENSE("GPL");
159	#ifdef MODULE_VERSION
160	MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
161	#endif
162
163
164	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
165	unsigned dev_get_flags(const struct net_device *dev)
166	{
167	unsigned flags;
168
169	flags = (dev->flags & ~(IFF_PROMISC \|
170	IFF_ALLMULTI \|
171	IFF_RUNNING)) \|
172	(dev->gflags & (IFF_PROMISC \|
173	IFF_ALLMULTI));
174
175	if (netif_running(dev) && netif_carrier_ok(dev))
176	flags \|= IFF_RUNNING;
177
178	return flags;
179	}
180	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
181
182
183	#ifdef VBOXNETFLT_WITH_QDISC
184	//#define QDISC_LOG(x) printk x
185	#define QDISC_LOG(x) do { } while (0)
186
187	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
188	#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops)
189	#elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
190	#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops, parent)
191	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
192	#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, queue, ops, parent)
193	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
194
195	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
196	#define qdisc_dev(qdisc) (qdisc->dev)
197	#define qdisc_pkt_len(skb) (skb->len)
198	#define QDISC_GET(dev) (dev->qdisc_sleeping)
199	#else
200	#define QDISC_GET(dev) (netdev_get_tx_queue(dev, 0)->qdisc_sleeping)
201	#endif
202
203	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
204	#define QDISC_SAVED_NUM(dev) 1
205	#elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
206	#define QDISC_SAVED_NUM(dev) dev->num_tx_queues
207	#else
208	#define QDISC_SAVED_NUM(dev) dev->num_tx_queues+1
209	#endif
210
211	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
212	#define QDISC_IS_BUSY(dev, qdisc) test_bit(__LINK_STATE_SCHED, &dev->state)
213	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
214	#define QDISC_IS_BUSY(dev, qdisc) (test_bit(__QDISC_STATE_RUNNING, &qdisc->state) \|\| \
215	test_bit(__QDISC_STATE_SCHED, &qdisc->state))
216	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
217
218	struct VBoxNetQDiscPriv
219	{
220	/** Pointer to the single child qdisc. */
221	struct Qdisc *pChild;
222	/*
223	* Technically it is possible to have different qdiscs for different TX
224	* queues so we have to save them all.
225	*/
226	/** Pointer to the array of saved qdiscs. */
227	struct Qdisc **ppSaved;
228	/** Pointer to the net filter instance. */
229	PVBOXNETFLTINS pVBoxNetFlt;
230	};
231	typedef struct VBoxNetQDiscPriv *PVBOXNETQDISCPRIV;
232
233	//#define VBOXNETFLT_QDISC_ENQUEUE
234	static int vboxNetFltQdiscEnqueue(struct sk_buff skb, struct Qdisc sch)
235	{
236	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
237	int rc;
238
239	#ifdef VBOXNETFLT_QDISC_ENQUEUE
240	if (VALID_PTR(pPriv->pVBoxNetFlt))
241	{
242	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
243	PCRTNETETHERHDR pEtherHdr;
244	PINTNETTRUNKSWPORT pSwitchPort;
245	uint32_t cbHdrs = skb_headlen(skb);
246
247	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
248	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(skb, 0, cbHdrs, &abHdrBuf[0]);
249	if ( pEtherHdr
250	&& (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) != NULL
251	&& VALID_PTR(pSwitchPort)
252	&& cbHdrs >= 6)
253	{
254	/** @todo consider reference counting, etc. */
255	INTNETSWDECISION enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
256	if (enmDecision == INTNETSWDECISION_INTNET)
257	{
258	struct sk_buff *pBuf = skb_copy(skb, GFP_ATOMIC);
259	pBuf->pkt_type = PACKET_OUTGOING;
260	vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
261	qdisc_drop(skb, sch);
262	++sch->bstats.packets;
263	sch->bstats.bytes += qdisc_pkt_len(skb);
264	return NET_XMIT_SUCCESS;
265	}
266	}
267	}
268	#endif /* VBOXNETFLT_QDISC_ENQUEUE */
269	rc = pPriv->pChild->enqueue(skb, pPriv->pChild);
270	if (rc == NET_XMIT_SUCCESS)
271	{
272	++sch->q.qlen;
273	++sch->bstats.packets;
274	sch->bstats.bytes += qdisc_pkt_len(skb);
275	}
276	else
277	++sch->qstats.drops;
278	return rc;
279	}
280
281	static struct sk_buff vboxNetFltQdiscDequeue(struct Qdisc sch)
282	{
283	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
284	#ifdef VBOXNETFLT_QDISC_ENQUEUE
285	--sch->q.qlen;
286	return pPriv->pChild->dequeue(pPriv->pChild);
287	#else /* VBOXNETFLT_QDISC_ENQUEUE */
288	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
289	PCRTNETETHERHDR pEtherHdr;
290	PINTNETTRUNKSWPORT pSwitchPort;
291	struct sk_buff *pSkb;
292
293	QDISC_LOG(("vboxNetFltDequeue: Enter pThis=%p\n", pPriv->pVBoxNetFlt));
294
295	while ((pSkb = pPriv->pChild->dequeue(pPriv->pChild)) != NULL)
296	{
297	struct sk_buff *pBuf;
298	INTNETSWDECISION enmDecision;
299	uint32_t cbHdrs;
300
301	--sch->q.qlen;
302
303	if (!VALID_PTR(pPriv->pVBoxNetFlt))
304	break;
305
306	cbHdrs = skb_headlen(pSkb);
307	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
308	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
309	if ( !pEtherHdr
310	\|\| (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) == NULL
311	\|\| !VALID_PTR(pSwitchPort)
312	\|\| cbHdrs < 6)
313	break;
314
315	/** @todo consider reference counting, etc. */
316	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
317	if (enmDecision != INTNETSWDECISION_INTNET)
318	break;
319
320	pBuf = skb_copy(pSkb, GFP_ATOMIC);
321	pBuf->pkt_type = PACKET_OUTGOING;
322	QDISC_LOG(("vboxNetFltDequeue: pThis=%p\n", pPriv->pVBoxNetFlt));
323	vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
324	qdisc_drop(pSkb, sch);
325	QDISC_LOG(("VBoxNetFlt: Packet for %02x:%02x:%02x:%02x:%02x:%02x dropped\n",
326	pSkb->data[0], pSkb->data[1], pSkb->data[2],
327	pSkb->data[3], pSkb->data[4], pSkb->data[5]));
328	}
329
330	return pSkb;
331	#endif /* VBOXNETFLT_QDISC_ENQUEUE */
332	}
333
334	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
335	static int vboxNetFltQdiscRequeue(struct sk_buff skb, struct Qdisc sch)
336	{
337	int rc;
338	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
339
340	rc = pPriv->pChild->ops->requeue(skb, pPriv->pChild);
341	if (rc == 0)
342	{
343	sch->q.qlen++;
344	sch->qstats.requeues++;
345	}
346
347	return rc;
348	}
349	#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
350
351	static unsigned int vboxNetFltQdiscDrop(struct Qdisc *sch)
352	{
353	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
354
355	++sch->qstats.drops;
356	--sch->q.qlen;
357	if (pPriv->pChild->ops->drop)
358	return pPriv->pChild->ops->drop(pPriv->pChild);
359
360	return 0;
361	}
362
363	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
364	static int vboxNetFltQdiscInit(struct Qdisc sch, struct rtattr opt)
365	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
366	static int vboxNetFltQdiscInit(struct Qdisc sch, struct nlattr opt)
367	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
368	{
369	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
370	struct net_device *pDev = qdisc_dev(sch);
371
372	pPriv->pVBoxNetFlt = NULL;
373
374	pPriv->ppSaved = kcalloc(QDISC_SAVED_NUM(pDev), sizeof(pPriv->ppSaved[0]),
375	GFP_KERNEL);
376	if (!pPriv->ppSaved)
377	return -ENOMEM;
378
379	pPriv->pChild = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
380	&pfifo_qdisc_ops,
381	TC_H_MAKE(TC_H_MAJ(sch->handle),
382	TC_H_MIN(1)));
383	if (!pPriv->pChild)
384	{
385	kfree(pPriv->ppSaved);
386	pPriv->ppSaved = NULL;
387	return -ENOMEM;
388	}
389
390	return 0;
391	}
392
393	static void vboxNetFltQdiscReset(struct Qdisc *sch)
394	{
395	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
396
397	qdisc_reset(pPriv->pChild);
398	sch->q.qlen = 0;
399	sch->qstats.backlog = 0;
400	}
401
402	static void vboxNetFltQdiscDestroy(struct Qdisc* sch)
403	{
404	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
405	struct net_device *pDev = qdisc_dev(sch);
406
407	qdisc_destroy(pPriv->pChild);
408	pPriv->pChild = NULL;
409
410	if (pPriv->ppSaved)
411	{
412	int i;
413	for (i = 0; i < QDISC_SAVED_NUM(pDev); i++)
414	if (pPriv->ppSaved[i])
415	qdisc_destroy(pPriv->ppSaved[i]);
416	kfree(pPriv->ppSaved);
417	pPriv->ppSaved = NULL;
418	}
419	}
420
421	static int vboxNetFltClassGraft(struct Qdisc sch, unsigned long arg, struct Qdisc pNew,
422	struct Qdisc **ppOld)
423	{
424	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
425
426	if (pNew == NULL)
427	pNew = &noop_qdisc;
428
429	sch_tree_lock(sch);
430	*ppOld = pPriv->pChild;
431	pPriv->pChild = pNew;
432	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
433	sch->q.qlen = 0;
434	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
435	qdisc_tree_decrease_qlen(ppOld, (ppOld)->q.qlen);
436	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
437	qdisc_reset(*ppOld);
438	sch_tree_unlock(sch);
439
440	return 0;
441	}
442
443	static struct Qdisc vboxNetFltClassLeaf(struct Qdisc sch, unsigned long arg)
444	{
445	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
446	return pPriv->pChild;
447	}
448
449	static unsigned long vboxNetFltClassGet(struct Qdisc *sch, u32 classid)
450	{
451	return 1;
452	}
453
454	static void vboxNetFltClassPut(struct Qdisc *sch, unsigned long arg)
455	{
456	}
457
458	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
459	static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
460	struct rtattr *tca, unsigned long arg)
461	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
462	static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
463	struct nlattr *tca, unsigned long arg)
464	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
465	{
466	return -ENOSYS;
467	}
468
469	static int vboxNetFltClassDelete(struct Qdisc *sch, unsigned long arg)
470	{
471	return -ENOSYS;
472	}
473
474	static void vboxNetFltClassWalk(struct Qdisc sch, struct qdisc_walker walker)
475	{
476	if (!walker->stop) {
477	if (walker->count >= walker->skip)
478	if (walker->fn(sch, 1, walker) < 0) {
479	walker->stop = 1;
480	return;
481	}
482	walker->count++;
483	}
484	}
485
486	static struct tcf_proto *vboxNetFltClassFindTcf(struct Qdisc sch, unsigned long cl)
487	{
488	return NULL;
489	}
490
491	static int vboxNetFltClassDump(struct Qdisc *sch, unsigned long cl,
492	struct sk_buff skb, struct tcmsg tcm)
493	{
494	PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
495
496	if (cl != 1)
497	return -ENOENT;
498
499	tcm->tcm_handle \|= TC_H_MIN(1);
500	tcm->tcm_info = pPriv->pChild->handle;
501
502	return 0;
503	}
504
505
506	static struct Qdisc_class_ops g_VBoxNetFltClassOps =
507	{
508	.graft = vboxNetFltClassGraft,
509	.leaf = vboxNetFltClassLeaf,
510	.get = vboxNetFltClassGet,
511	.put = vboxNetFltClassPut,
512	.change = vboxNetFltClassChange,
513	.delete = vboxNetFltClassDelete,
514	.walk = vboxNetFltClassWalk,
515	.tcf_chain = vboxNetFltClassFindTcf,
516	.dump = vboxNetFltClassDump,
517	};
518
519
520	static struct Qdisc_ops g_VBoxNetFltQDiscOps = {
521	.cl_ops = &g_VBoxNetFltClassOps,
522	.id = "vboxnetflt",
523	.priv_size = sizeof(struct VBoxNetQDiscPriv),
524	.enqueue = vboxNetFltQdiscEnqueue,
525	.dequeue = vboxNetFltQdiscDequeue,
526	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
527	.requeue = vboxNetFltQdiscRequeue,
528	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
529	.peek = qdisc_peek_dequeued,
530	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
531	.drop = vboxNetFltQdiscDrop,
532	.init = vboxNetFltQdiscInit,
533	.reset = vboxNetFltQdiscReset,
534	.destroy = vboxNetFltQdiscDestroy,
535	.owner = THIS_MODULE
536	};
537
538	/*
539	* If our qdisc is already attached to the device (that means the user
540	* installed it from command line with 'tc' command) we simply update
541	* the pointer to vboxnetflt instance in qdisc's private structure.
542	* Otherwise we need to take some additional steps:
543	* - Create our qdisc;
544	* - Save all references to qdiscs;
545	* - Replace our child with the first qdisc reference;
546	* - Replace all references so they point to our qdisc.
547	*/
548	static void vboxNetFltLinuxQdiscInstall(PVBOXNETFLTINS pThis, struct net_device *pDev)
549	{
550	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
551	int i;
552	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
553	PVBOXNETQDISCPRIV pPriv;
554
555	struct Qdisc *pExisting = QDISC_GET(pDev);
556	if (strcmp(pExisting->ops->id, "vboxnetflt"))
557	{
558	/* The existing qdisc is different from ours, let's create new one. */
559	struct Qdisc *pNew = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
560	&g_VBoxNetFltQDiscOps, TC_H_ROOT);
561	if (!pNew)
562	return; // TODO: Error?
563
564	if (!try_module_get(THIS_MODULE))
565	{
566	/*
567	* This may cause a memory leak but calling qdisc_destroy()
568	* is not an option as it will call module_put().
569	*/
570	return;
571	}
572	pPriv = qdisc_priv(pNew);
573
574	qdisc_destroy(pPriv->pChild);
575	pPriv->pChild = QDISC_GET(pDev);
576	atomic_inc(&pPriv->pChild->refcnt);
577	/*
578	* There is no need in deactivating the device or acquiring any locks
579	* prior changing qdiscs since we do not destroy the old qdisc.
580	* Atomic replacement of pointers is enough.
581	*/
582	/*
583	* No need to change reference counters here as we merely move
584	* the pointer and the reference counter of the newly allocated
585	* qdisc is already 1.
586	*/
587	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
588	pPriv->ppSaved[0] = pDev->qdisc_sleeping;
589	ASMAtomicWritePtr(&pDev->qdisc_sleeping, pNew);
590	ASMAtomicWritePtr(&pDev->qdisc, pNew);
591	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
592	for (i = 0; i < pDev->num_tx_queues; i++)
593	{
594	struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
595
596	pPriv->ppSaved[i] = pQueue->qdisc_sleeping;
597	ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pNew);
598	ASMAtomicWritePtr(&pQueue->qdisc, pNew);
599	if (i)
600	atomic_inc(&pNew->refcnt);
601	}
602	/* Newer kernels store root qdisc in netdev structure as well. */
603	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
604	pPriv->ppSaved[pDev->num_tx_queues] = pDev->qdisc;
605	ASMAtomicWritePtr(&pDev->qdisc, pNew);
606	atomic_inc(&pNew->refcnt);
607	# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
608	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
609	}
610	else
611	{
612	/* We already have vboxnetflt qdisc, let's use it. */
613	pPriv = qdisc_priv(pExisting);
614	}
615	ASMAtomicWritePtr(&pPriv->pVBoxNetFlt, pThis);
616	QDISC_LOG(("vboxNetFltLinuxInstallQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
617	}
618
619	static void vboxNetFltLinuxQdiscRemove(PVBOXNETFLTINS pThis, struct net_device *pDev)
620	{
621	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
622	int i;
623	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
624	PVBOXNETQDISCPRIV pPriv;
625	struct Qdisc *pQdisc;
626	if (!pDev)
627	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
628	if (!VALID_PTR(pDev))
629	{
630	printk("VBoxNetFlt: Failed to detach qdisc, invalid device pointer: %p\n",
631	pDev);
632	return; // TODO: Consider returing an error
633	}
634
635
636	pQdisc = QDISC_GET(pDev);
637	if (strcmp(pQdisc->ops->id, "vboxnetflt"))
638	{
639	/* Looks like the user has replaced our qdisc manually. */
640	printk("VBoxNetFlt: Failed to detach qdisc, wrong qdisc: %s\n",
641	pQdisc->ops->id);
642	return; // TODO: Consider returing an error
643	}
644
645	pPriv = qdisc_priv(pQdisc);
646	Assert(pPriv->pVBoxNetFlt == pThis);
647	ASMAtomicWriteNullPtr(&pPriv->pVBoxNetFlt);
648
649	QDISC_LOG(("vboxNetFltLinuxQdiscRemove: refcnt=%d num_tx_queues=%d\n",
650	atomic_read(&pQdisc->refcnt), pDev->num_tx_queues));
651	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
652	/* Play it safe, make sure the qdisc is not being used. */
653	if (pPriv->ppSaved[0])
654	{
655	ASMAtomicWritePtr(&pDev->qdisc_sleeping, pPriv->ppSaved[0]);
656	ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[0]);
657	pPriv->ppSaved[0] = NULL;
658	while (QDISC_IS_BUSY(pDev, pQdisc))
659	yield();
660	qdisc_destroy(pQdisc); /* Destroy reference */
661	}
662	#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
663	for (i = 0; i < pDev->num_tx_queues; i++)
664	{
665	struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
666	if (pPriv->ppSaved[i])
667	{
668	Assert(pQueue->qdisc_sleeping == pQdisc);
669	ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pPriv->ppSaved[i]);
670	ASMAtomicWritePtr(&pQueue->qdisc, pPriv->ppSaved[i]);
671	pPriv->ppSaved[i] = NULL;
672	while (QDISC_IS_BUSY(pDev, pQdisc))
673	yield();
674	qdisc_destroy(pQdisc); /* Destroy reference */
675	}
676	}
677	/* Newer kernels store root qdisc in netdev structure as well. */
678	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
679	ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[pDev->num_tx_queues]);
680	pPriv->ppSaved[pDev->num_tx_queues] = NULL;
681	while (QDISC_IS_BUSY(pDev, pQdisc))
682	yield();
683	qdisc_destroy(pQdisc); /* Destroy reference */
684	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
685	#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
686
687	/*
688	* At this point all references to our qdisc should be gone
689	* unless the user had installed it manually.
690	*/
691	QDISC_LOG(("vboxNetFltLinuxRemoveQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
692	}
693
694	#endif /* VBOXNETFLT_WITH_QDISC */
695
696
697	/**
698	* Initialize module.
699	*
700	* @returns appropriate status code.
701	*/
702	static int __init VBoxNetFltLinuxInit(void)
703	{
704	int rc;
705	/*
706	* Initialize IPRT.
707	*/
708	rc = RTR0Init(0);
709	if (RT_SUCCESS(rc))
710	{
711	Log(("VBoxNetFltLinuxInit\n"));
712
713	/*
714	* Initialize the globals and connect to the support driver.
715	*
716	* This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
717	* for establishing the connect to the support driver.
718	*/
719	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
720	rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
721	if (RT_SUCCESS(rc))
722	{
723	#ifdef VBOXNETFLT_WITH_QDISC
724	/*memcpy(&g_VBoxNetFltQDiscOps, &pfifo_qdisc_ops, sizeof(g_VBoxNetFltQDiscOps));
725	strcpy(g_VBoxNetFltQDiscOps.id, "vboxnetflt");
726	g_VBoxNetFltQDiscOps.owner = THIS_MODULE;*/
727	rc = register_qdisc(&g_VBoxNetFltQDiscOps);
728	if (rc)
729	{
730	LogRel(("VBoxNetFlt: Failed to registed qdisc: %d\n", rc));
731	return rc;
732	}
733	#endif /* VBOXNETFLT_WITH_QDISC */
734	LogRel(("VBoxNetFlt: Successfully started.\n"));
735	return 0;
736	}
737
738	LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
739	RTR0Term();
740	}
741	else
742	LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
743
744	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
745	return -RTErrConvertToErrno(rc);
746	}
747
748
749	/**
750	* Unload the module.
751	*
752	* @todo We have to prevent this if we're busy!
753	*/
754	static void __exit VBoxNetFltLinuxUnload(void)
755	{
756	int rc;
757	Log(("VBoxNetFltLinuxUnload\n"));
758	Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
759
760	#ifdef VBOXNETFLT_WITH_QDISC
761	unregister_qdisc(&g_VBoxNetFltQDiscOps);
762	#endif /* VBOXNETFLT_WITH_QDISC */
763	/*
764	* Undo the work done during start (in reverse order).
765	*/
766	rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
767	AssertRC(rc); NOREF(rc);
768
769	RTR0Term();
770
771	memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
772
773	Log(("VBoxNetFltLinuxUnload - done\n"));
774	}
775
776
777	/**
778	* Experiment where we filter trafic from the host to the internal network
779	* before it reaches the NIC driver.
780	*
781	* The current code uses a very ugly hack and only works on kernels using the
782	* net_device_ops (>= 2.6.29). It has been shown to give us a
783	* performance boost of 60-100% though. So, we have to find some less hacky way
784	* of getting this job done eventually.
785	*
786	* #define VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
787	*/
788	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
789
790	/**
791	* The overridden net_device_ops of the device we're attached to.
792	*
793	* Requires Linux 2.6.29 or later.
794	*
795	* This is a very dirty hack that was create to explore how much we can improve
796	* the host to guest transfers by not CC'ing the NIC.
797	*/
798	typedef struct VBoxNetDeviceOpsOverride
799	{
800	/** Our overridden ops. */
801	struct net_device_ops Ops;
802	/** Magic word. */
803	uint32_t u32Magic;
804	/** Pointer to the original ops. */
805	struct net_device_ops const *pOrgOps;
806	/** Pointer to the net filter instance. */
807	PVBOXNETFLTINS pVBoxNetFlt;
808	/** The number of filtered packages. */
809	uint64_t cFiltered;
810	/** The total number of packets */
811	uint64_t cTotal;
812	} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
813	/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
814	#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
815
816	/**
817	* ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
818	* because they belong on the internal network.
819	*
820	* @returns NETDEV_TX_XXX.
821	* @param pSkb The socket buffer to transmit.
822	* @param pDev The net device.
823	*/
824	static int vboxNetFltLinuxStartXmitFilter(struct sk_buff pSkb, struct net_device pDev)
825	{
826	PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->netdev_ops;
827	uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
828	PCRTNETETHERHDR pEtherHdr;
829	PINTNETTRUNKSWPORT pSwitchPort;
830	uint32_t cbHdrs;
831
832
833	/*
834	* Validate the override structure.
835	*
836	* Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
837	* to be production quality code, we would have to be much more
838	* careful here and avoid the race.
839	*/
840	if ( !VALID_PTR(pOverride)
841	\|\| pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
842	\|\| !VALID_PTR(pOverride->pOrgOps))
843	{
844	printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
845	dev_kfree_skb(pSkb);
846	return NETDEV_TX_OK;
847	}
848	pOverride->cTotal++;
849
850	/*
851	* Do the filtering base on the defaul OUI of our virtual NICs
852	*
853	* Note! In a real solution, we would ask the switch whether the
854	* destination MAC is 100% to be on the internal network and then
855	* drop it.
856	*/
857	cbHdrs = skb_headlen(pSkb);
858	cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
859	pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
860	if ( pEtherHdr
861	&& VALID_PTR(pOverride->pVBoxNetFlt)
862	&& (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
863	&& VALID_PTR(pSwitchPort)
864	&& cbHdrs >= 6)
865	{
866	INTNETSWDECISION enmDecision;
867
868	/** @todo consider reference counting, etc. */
869	enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
870	if (enmDecision == INTNETSWDECISION_INTNET)
871	{
872	dev_kfree_skb(pSkb);
873	pOverride->cFiltered++;
874	return NETDEV_TX_OK;
875	}
876	}
877
878	return pOverride->pOrgOps->ndo_start_xmit(pSkb, pDev);
879	}
880
881	/**
882	* Hooks the device ndo_start_xmit operation of the device.
883	*
884	* @param pThis The net filter instance.
885	* @param pDev The net device.
886	*/
887	static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
888	{
889	PVBOXNETDEVICEOPSOVERRIDE pOverride;
890	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
891
892	pOverride = RTMemAlloc(sizeof(*pOverride));
893	if (!pOverride)
894	return;
895	pOverride->pOrgOps = pDev->netdev_ops;
896	pOverride->Ops = *pDev->netdev_ops;
897	pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
898	pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
899	pOverride->cTotal = 0;
900	pOverride->cFiltered = 0;
901	pOverride->pVBoxNetFlt = pThis;
902
903	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp); /* (this isn't necessary, but so what) */
904	ASMAtomicWritePtr((void * volatile *)&pDev->netdev_ops, pOverride);
905	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
906	}
907
908	/**
909	* Undos what vboxNetFltLinuxHookDev did.
910	*
911	* @param pThis The net filter instance.
912	* @param pDev The net device. Can be NULL, in which case
913	* we'll try retrieve it from @a pThis.
914	*/
915	static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
916	{
917	PVBOXNETDEVICEOPSOVERRIDE pOverride;
918	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
919
920	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
921	if (!pDev)
922	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
923	if (VALID_PTR(pDev))
924	{
925	pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->netdev_ops;
926	if ( VALID_PTR(pOverride)
927	&& pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
928	&& VALID_PTR(pOverride->pOrgOps)
929	)
930	{
931	ASMAtomicWritePtr((void * volatile *)&pDev->netdev_ops, pOverride->pOrgOps);
932	ASMAtomicWriteU32(&pOverride->u32Magic, 0);
933	}
934	else
935	pOverride = NULL;
936	}
937	else
938	pOverride = NULL;
939	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
940
941	if (pOverride)
942	{
943	printk("vboxnetflt: dropped %llu out of %llu packets\n", pOverride->cFiltered, pOverride->cTotal);
944	RTMemFree(pOverride);
945	}
946	}
947
948	#endif /* VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT */
949
950
951	/**
952	* Reads and retains the host interface handle.
953	*
954	* @returns The handle, NULL if detached.
955	* @param pThis
956	*/
957	DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
958	{
959	#if 0
960	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
961	struct net_device *pDev = NULL;
962
963	Log(("vboxNetFltLinuxRetainNetDev\n"));
964	/*
965	* Be careful here to avoid problems racing the detached callback.
966	*/
967	RTSpinlockAcquire(pThis->hSpinlock, &Tmp);
968	if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
969	{
970	pDev = (struct net_device )ASMAtomicUoReadPtr((void volatile *)&pThis->u.s.pDev);
971	if (pDev)
972	{
973	dev_hold(pDev);
974	Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
975	}
976	}
977	RTSpinlockRelease(pThis->hSpinlock, &Tmp);
978
979	Log(("vboxNetFltLinuxRetainNetDev - done\n"));
980	return pDev;
981	#else
982	return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
983	#endif
984	}
985
986
987	/**
988	* Release the host interface handle previously retained
989	* by vboxNetFltLinuxRetainNetDev.
990	*
991	* @param pThis The instance.
992	* @param pDev The vboxNetFltLinuxRetainNetDev
993	* return value, NULL is fine.
994	*/
995	DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
996	{
997	#if 0
998	Log(("vboxNetFltLinuxReleaseNetDev\n"));
999	NOREF(pThis);
1000	if (pDev)
1001	{
1002	dev_put(pDev);
1003	Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1004	}
1005	Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
1006	#endif
1007	}
1008
1009	#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 \| (skb->dev->ifindex & 0xFFFF))
1010	#define VBOXNETFLT_SKB_TAG(skb) ((uint32_t)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
1011
1012	/**
1013	* Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
1014	* i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
1015	*
1016	* @returns true / false accordingly.
1017	* @param pBuf The sk_buff.
1018	*/
1019	DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
1020	{
1021	return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
1022	}
1023
1024
1025	/**
1026	* Internal worker that create a linux sk_buff for a
1027	* (scatter/)gather list.
1028	*
1029	* @returns Pointer to the sk_buff.
1030	* @param pThis The instance.
1031	* @param pSG The (scatter/)gather list.
1032	* @param fDstWire Set if the destination is the wire.
1033	*/
1034	static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
1035	{
1036	struct sk_buff *pPkt;
1037	struct net_device *pDev;
1038	unsigned fGsoType = 0;
1039
1040	if (pSG->cbTotal == 0)
1041	{
1042	LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
1043	return NULL;
1044	}
1045
1046	/** @todo We should use fragments mapping the SG buffers with large packets.
1047	* 256 bytes seems to be the a threshold used a lot for this. It
1048	* requires some nasty work on the intnet side though... */
1049	/*
1050	* Allocate a packet and copy over the data.
1051	*/
1052	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1053	pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
1054	if (RT_UNLIKELY(!pPkt))
1055	{
1056	Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
1057	pSG->pvUserData = NULL;
1058	return NULL;
1059	}
1060	pPkt->dev = pDev;
1061	pPkt->ip_summed = CHECKSUM_NONE;
1062
1063	/* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
1064	skb_reserve(pPkt, NET_IP_ALIGN);
1065
1066	/* Copy the segments. */
1067	skb_put(pPkt, pSG->cbTotal);
1068	IntNetSgRead(pSG, pPkt->data);
1069
1070	#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) \|\| defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1071	/*
1072	* Setup GSO if used by this packet.
1073	*/
1074	switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
1075	{
1076	default:
1077	AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
1078	/* fall thru */
1079	case PDMNETWORKGSOTYPE_INVALID:
1080	fGsoType = 0;
1081	break;
1082	case PDMNETWORKGSOTYPE_IPV4_TCP:
1083	fGsoType = SKB_GSO_TCPV4;
1084	break;
1085	case PDMNETWORKGSOTYPE_IPV4_UDP:
1086	fGsoType = SKB_GSO_UDP;
1087	break;
1088	case PDMNETWORKGSOTYPE_IPV6_TCP:
1089	fGsoType = SKB_GSO_TCPV6;
1090	break;
1091	}
1092	if (fGsoType)
1093	{
1094	struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
1095
1096	pShInfo->gso_type = fGsoType \| SKB_GSO_DODGY;
1097	pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
1098	pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
1099
1100	/*
1101	* We need to set checksum fields even if the packet goes to the host
1102	* directly as it may be immediately forwared by IP layer @bugref{5020}.
1103	*/
1104	Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrs);
1105	pPkt->ip_summed = CHECKSUM_PARTIAL;
1106	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1107	pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
1108	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
1109	pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
1110	else
1111	pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
1112	# else
1113	pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
1114	if (fGsoType & (SKB_GSO_TCPV4 \| SKB_GSO_TCPV6))
1115	pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
1116	else
1117	pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
1118	# endif
1119	if (!fDstWire)
1120	PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, false /fPayloadChecksum/);
1121	}
1122	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE \|\| VBOXNETFLT_WITH_GSO_XMIT_HOST */
1123
1124	/*
1125	* Finish up the socket buffer.
1126	*/
1127	pPkt->protocol = eth_type_trans(pPkt, pDev);
1128	if (fDstWire)
1129	{
1130	VBOX_SKB_RESET_NETWORK_HDR(pPkt);
1131
1132	/* Restore ethernet header back. */
1133	skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
1134	VBOX_SKB_RESET_MAC_HDR(pPkt);
1135	}
1136	VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
1137
1138	return pPkt;
1139	}
1140
1141
1142	/**
1143	* Initializes a SG list from an sk_buff.
1144	*
1145	* @returns Number of segments.
1146	* @param pThis The instance.
1147	* @param pBuf The sk_buff.
1148	* @param pSG The SG.
1149	* @param pvFrame The frame pointer, optional.
1150	* @param cSegs The number of segments allocated for the SG.
1151	* This should match the number in the mbuf exactly!
1152	* @param fSrc The source of the frame.
1153	* @param pGso Pointer to the GSO context if it's a GSO
1154	* internal network frame. NULL if regular frame.
1155	*/
1156	DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
1157	unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1158	{
1159	int i;
1160	NOREF(pThis);
1161
1162	Assert(!skb_shinfo(pBuf)->frag_list);
1163
1164	if (!pGsoCtx)
1165	IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /cSegsUsed/);
1166	else
1167	IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /cSegsUsed/, pGsoCtx);
1168
1169	#ifdef VBOXNETFLT_SG_SUPPORT
1170	pSG->aSegs[0].cb = skb_headlen(pBuf);
1171	pSG->aSegs[0].pv = pBuf->data;
1172	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1173
1174	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1175	{
1176	skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
1177	pSG->aSegs[i+1].cb = pFrag->size;
1178	pSG->aSegs[i+1].pv = kmap(pFrag->page);
1179	printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
1180	pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
1181	}
1182	++i;
1183
1184	#else
1185	pSG->aSegs[0].cb = pBuf->len;
1186	pSG->aSegs[0].pv = pBuf->data;
1187	pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1188	i = 1;
1189	#endif
1190
1191	pSG->cSegsUsed = i;
1192
1193	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1194	/*
1195	* Add a trailer if the frame is too small.
1196	*
1197	* Since we're getting to the packet before it is framed, it has not
1198	* yet been padded. The current solution is to add a segment pointing
1199	* to a buffer containing all zeros and pray that works for all frames...
1200	*/
1201	if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
1202	{
1203	static uint8_t const s_abZero[128] = {0};
1204
1205	AssertReturnVoid(i < cSegs);
1206
1207	pSG->aSegs[i].Phys = NIL_RTHCPHYS;
1208	pSG->aSegs[i].pv = (void *)&s_abZero[0];
1209	pSG->aSegs[i].cb = 60 - pSG->cbTotal;
1210	pSG->cbTotal = 60;
1211	pSG->cSegsUsed++;
1212	Assert(i + 1 <= pSG->cSegsAlloc)
1213	}
1214	#endif
1215
1216	Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
1217	pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
1218	for (i = 0; i < pSG->cSegsUsed; i++)
1219	Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
1220	i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1221	}
1222
1223	/**
1224	* Packet handler,
1225	*
1226	* @returns 0 or EJUSTRETURN.
1227	* @param pThis The instance.
1228	* @param pMBuf The mbuf.
1229	* @param pvFrame The start of the frame, optional.
1230	* @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
1231	* @param eProtocol The protocol.
1232	*/
1233	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
1234	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1235	struct net_device *pSkbDev,
1236	struct packet_type *pPacketType,
1237	struct net_device *pOrigDev)
1238	#else
1239	static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1240	struct net_device *pSkbDev,
1241	struct packet_type *pPacketType)
1242	#endif
1243	{
1244	PVBOXNETFLTINS pThis;
1245	struct net_device *pDev;
1246	LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1247	pBuf, pSkbDev, pPacketType));
1248	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1249	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",
1250	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));
1251	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1252	#else
1253	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",
1254	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));
1255	#endif
1256	/*
1257	* Drop it immediately?
1258	*/
1259	if (!pBuf)
1260	return 0;
1261
1262	pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1263	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1264	if (pThis->u.s.pDev != pSkbDev)
1265	{
1266	Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1267	return 0;
1268	}
1269
1270	Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1271	if (vboxNetFltLinuxSkBufIsOur(pBuf))
1272	{
1273	Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1274	dev_kfree_skb(pBuf);
1275	return 0;
1276	}
1277
1278	#ifndef VBOXNETFLT_SG_SUPPORT
1279	{
1280	/*
1281	* Get rid of fragmented packets, they cause too much trouble.
1282	*/
1283	struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1284	kfree_skb(pBuf);
1285	if (!pCopy)
1286	{
1287	LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1288	return 0;
1289	}
1290	pBuf = pCopy;
1291	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1292	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",
1293	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));
1294	Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1295	# else
1296	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",
1297	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));
1298	# endif
1299	}
1300	#endif
1301
1302	#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1303	/* Forward it to the internal network. */
1304	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1305	#else
1306	/* Add the packet to transmit queue and schedule the bottom half. */
1307	skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1308	schedule_work(&pThis->u.s.XmitTask);
1309	Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1310	&pThis->u.s.XmitTask, pBuf));
1311	#endif
1312
1313	/* It does not really matter what we return, it is ignored by the kernel. */
1314	return 0;
1315	}
1316
1317	/**
1318	* Calculate the number of INTNETSEG segments the socket buffer will need.
1319	*
1320	* @returns Segment count.
1321	* @param pBuf The socket buffer.
1322	*/
1323	DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
1324	{
1325	#ifdef VBOXNETFLT_SG_SUPPORT
1326	unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1327	#else
1328	unsigned cSegs = 1;
1329	#endif
1330	#ifdef PADD_RUNT_FRAMES_FROM_HOST
1331	/* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1332	if (pBuf->len < 60)
1333	cSegs++;
1334	#endif
1335	return cSegs;
1336	}
1337
1338	/**
1339	* Destroy the intnet scatter / gather buffer created by
1340	* vboxNetFltLinuxSkBufToSG.
1341	*/
1342	static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
1343	{
1344	#ifdef VBOXNETFLT_SG_SUPPORT
1345	int i;
1346
1347	for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1348	{
1349	printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
1350	kunmap(pSG->aSegs[i+1].pv);
1351	}
1352	#endif
1353	NOREF(pSG);
1354	}
1355
1356	#ifdef LOG_ENABLED
1357	/**
1358	* Logging helper.
1359	*/
1360	static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1361	{
1362	uint8_t pInt, pExt;
1363	static int iPacketNo = 1;
1364	iPacketNo += iIncrement;
1365	if (fEgress)
1366	{
1367	pExt = pSG->aSegs[0].pv;
1368	pInt = pExt + 6;
1369	}
1370	else
1371	{
1372	pInt = pSG->aSegs[0].pv;
1373	pExt = pInt + 6;
1374	}
1375	Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1376	" %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1377	pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1378	fEgress ? "-->" : "<--", pszWhere,
1379	pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1380	pSG->cbTotal, iPacketNo));
1381	Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1382	}
1383	#else
1384	# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1385	#endif
1386
1387	#ifdef VBOXNETFLT_WITH_GSO_RECV
1388
1389	/**
1390	* Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1391	* GSO socket buffer without having to segment it.
1392	*
1393	* @returns true on success, false if needs segmenting.
1394	* @param pThis The net filter instance.
1395	* @param pSkb The GSO socket buffer.
1396	* @param fSrc The source.
1397	* @param pGsoCtx Where to return the GSO context on success.
1398	*/
1399	static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1400	PPDMNETWORKGSO pGsoCtx)
1401	{
1402	PDMNETWORKGSOTYPE enmGsoType;
1403	uint16_t uEtherType;
1404	unsigned int cbTransport;
1405	unsigned int offTransport;
1406	unsigned int cbTransportHdr;
1407	unsigned uProtocol;
1408	union
1409	{
1410	RTNETIPV4 IPv4;
1411	RTNETIPV6 IPv6;
1412	RTNETTCP Tcp;
1413	uint8_t ab[40];
1414	uint16_t au16[40/2];
1415	uint32_t au32[40/4];
1416	} Buf;
1417
1418	/*
1419	* Check the GSO properties of the socket buffer and make sure it fits.
1420	*/
1421	/** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1422	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP \| SKB_GSO_DODGY \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4) ))
1423	{
1424	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1425	return false;
1426	}
1427	if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1428	\|\| pSkb->len > VBOX_MAX_GSO_SIZE ))
1429	{
1430	Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1431	return false;
1432	}
1433	/*
1434	* It is possible to receive GSO packets from wire if GRO is enabled.
1435	*/
1436	if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1437	{
1438	Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1439	#ifdef VBOXNETFLT_WITH_GRO
1440	/*
1441	* The packet came from the wire and the driver has already consumed
1442	* mac header. We need to restore it back.
1443	*/
1444	pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1445	skb_push(pSkb, pSkb->mac_len);
1446	Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1447	pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
1448	#else /* !VBOXNETFLT_WITH_GRO */
1449	/* Older kernels didn't have GRO. */
1450	return false;
1451	#endif /* !VBOXNETFLT_WITH_GRO */
1452	}
1453	else
1454	{
1455	/*
1456	* skb_gso_segment does the following. Do we need to do it as well?
1457	*/
1458	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1459	skb_reset_mac_header(pSkb);
1460	pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1461	#else
1462	pSkb->mac.raw = pSkb->data;
1463	pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1464	#endif
1465	}
1466
1467	/*
1468	* Switch on the ethertype.
1469	*/
1470	uEtherType = pSkb->protocol;
1471	if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1472	&& pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1473	{
1474	uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1475	if (puEtherType)
1476	uEtherType = *puEtherType;
1477	}
1478	switch (uEtherType)
1479	{
1480	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1481	{
1482	unsigned int cbHdr;
1483	PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1484	if (RT_UNLIKELY(!pIPv4))
1485	{
1486	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1487	return false;
1488	}
1489
1490	cbHdr = pIPv4->ip_hl * 4;
1491	cbTransport = RT_N2H_U16(pIPv4->ip_len);
1492	if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1493	\|\| cbHdr > cbTransport ))
1494	{
1495	Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1496	return false;
1497	}
1498	cbTransport -= cbHdr;
1499	offTransport = pSkb->mac_len + cbHdr;
1500	uProtocol = pIPv4->ip_p;
1501	if (uProtocol == RTNETIPV4_PROT_TCP)
1502	enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1503	else if (uProtocol == RTNETIPV4_PROT_UDP)
1504	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1505	else /** @todo IPv6: 4to6 tunneling */
1506	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1507	break;
1508	}
1509
1510	case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1511	{
1512	PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1513	if (RT_UNLIKELY(!pIPv6))
1514	{
1515	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1516	return false;
1517	}
1518
1519	cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1520	offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1521	uProtocol = pIPv6->ip6_nxt;
1522	/** @todo IPv6: Dig our way out of the other headers. */
1523	if (uProtocol == RTNETIPV4_PROT_TCP)
1524	enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1525	else if (uProtocol == RTNETIPV4_PROT_UDP)
1526	enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1527	else
1528	enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1529	break;
1530	}
1531
1532	default:
1533	Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1534	return false;
1535	}
1536
1537	if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1538	{
1539	Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1540	return false;
1541	}
1542
1543	if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1544	\|\| offTransport + cbTransport > pSkb->len
1545	\|\| cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1546	{
1547	Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1548	offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1549	return false;
1550	}
1551
1552	/*
1553	* Check the TCP/UDP bits.
1554	*/
1555	if (uProtocol == RTNETIPV4_PROT_TCP)
1556	{
1557	PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1558	if (RT_UNLIKELY(!pTcp))
1559	{
1560	Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1561	return false;
1562	}
1563
1564	cbTransportHdr = pTcp->th_off * 4;
1565	if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1566	\|\| cbTransportHdr > cbTransport
1567	\|\| offTransport + cbTransportHdr >= UINT8_MAX
1568	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1569	{
1570	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1571	return false;
1572	}
1573
1574	}
1575	else
1576	{
1577	Assert(uProtocol == RTNETIPV4_PROT_UDP);
1578	cbTransportHdr = sizeof(RTNETUDP);
1579	if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1580	\|\| offTransport + cbTransportHdr >= pSkb->len ))
1581	{
1582	Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1583	return false;
1584	}
1585	}
1586
1587	/*
1588	* We're good, init the GSO context.
1589	*/
1590	pGsoCtx->u8Type = enmGsoType;
1591	pGsoCtx->cbHdrs = offTransport + cbTransportHdr;
1592	pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1593	pGsoCtx->offHdr1 = pSkb->mac_len;
1594	pGsoCtx->offHdr2 = offTransport;
1595	pGsoCtx->au8Unused[0] = 0;
1596	pGsoCtx->au8Unused[1] = 0;
1597
1598	return true;
1599	}
1600
1601	/**
1602	* Forward the socket buffer as a GSO internal network frame.
1603	*
1604	* @returns IPRT status code.
1605	* @param pThis The net filter instance.
1606	* @param pSkb The GSO socket buffer.
1607	* @param fSrc The source.
1608	* @param pGsoCtx Where to return the GSO context on success.
1609	*/
1610	static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1611	{
1612	int rc;
1613	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1614	if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1615	{
1616	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1617	if (RT_LIKELY(pSG))
1618	{
1619	vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1620
1621	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1622	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1623
1624	vboxNetFltLinuxDestroySG(pSG);
1625	rc = VINF_SUCCESS;
1626	}
1627	else
1628	{
1629	Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1630	rc = VERR_NO_MEMORY;
1631	}
1632	}
1633	else
1634	{
1635	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1636	rc = VERR_INTERNAL_ERROR_3;
1637	}
1638
1639	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1640	dev_kfree_skb(pSkb);
1641	return rc;
1642	}
1643
1644	#endif /* VBOXNETFLT_WITH_GSO_RECV */
1645
1646	/**
1647	* Worker for vboxNetFltLinuxForwardToIntNet.
1648	*
1649	* @returns VINF_SUCCESS or VERR_NO_MEMORY.
1650	* @param pThis The net filter instance.
1651	* @param pBuf The socket buffer.
1652	* @param fSrc The source.
1653	*/
1654	static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1655	{
1656	int rc;
1657	unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1658	if (cSegs <= MAX_SKB_FRAGS + 1)
1659	{
1660	PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1661	if (RT_LIKELY(pSG))
1662	{
1663	if (fSrc & INTNETTRUNKDIR_WIRE)
1664	{
1665	/*
1666	* The packet came from wire, ethernet header was removed by device driver.
1667	* Restore it.
1668	*/
1669	skb_push(pBuf, ETH_HLEN);
1670	}
1671
1672	vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /pGsoCtx/);
1673
1674	vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1675	pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1676
1677	vboxNetFltLinuxDestroySG(pSG);
1678	rc = VINF_SUCCESS;
1679	}
1680	else
1681	{
1682	Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1683	rc = VERR_NO_MEMORY;
1684	}
1685	}
1686	else
1687	{
1688	Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1689	rc = VERR_INTERNAL_ERROR_3;
1690	}
1691
1692	Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1693	dev_kfree_skb(pBuf);
1694	return rc;
1695	}
1696
1697	/**
1698	*
1699	* @param pBuf The socket buffer. This is consumed by this function.
1700	*/
1701	static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1702	{
1703	uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1704
1705	#ifdef VBOXNETFLT_WITH_GSO
1706	if (skb_is_gso(pBuf))
1707	{
1708	PDMNETWORKGSO GsoCtx;
1709	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",
1710	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));
1711	# ifdef VBOXNETFLT_WITH_GSO_RECV
1712	if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP \| SKB_GSO_TCPV6 \| SKB_GSO_TCPV4))
1713	&& vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1714	vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1715	else
1716	# endif
1717	{
1718	/* Need to segment the packet */
1719	struct sk_buff *pNext;
1720	struct sk_buff pSegment = skb_gso_segment(pBuf, 0 /supported features*/);
1721	if (IS_ERR(pSegment))
1722	{
1723	dev_kfree_skb(pBuf);
1724	LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1725	return;
1726	}
1727
1728	for (; pSegment; pSegment = pNext)
1729	{
1730	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",
1731	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));
1732	pNext = pSegment->next;
1733	pSegment->next = 0;
1734	vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1735	}
1736	dev_kfree_skb(pBuf);
1737	}
1738	}
1739	else
1740	#endif /* VBOXNETFLT_WITH_GSO */
1741	{
1742	if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1743	{
1744	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1745	/*
1746	* Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1747	* and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1748	* header length from the header itself and reconstruct 'h' pointer
1749	* to TCP (or whatever) header.
1750	*/
1751	unsigned char *tmp = pBuf->h.raw;
1752	if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1753	pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1754	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1755	if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1756	{
1757	LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1758	dev_kfree_skb(pBuf);
1759	return;
1760	}
1761	#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1762	/* Restore the original (wrong) pointer. */
1763	pBuf->h.raw = tmp;
1764	#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1765	}
1766	vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1767	}
1768	}
1769
1770	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1771	/**
1772	* Work queue handler that forwards the socket buffers queued by
1773	* vboxNetFltLinuxPacketHandler to the internal network.
1774	*
1775	* @param pWork The work queue.
1776	*/
1777	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1778	static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1779	# else
1780	static void vboxNetFltLinuxXmitTask(void *pWork)
1781	# endif
1782	{
1783	PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1784	struct sk_buff *pBuf;
1785
1786	Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1787
1788	/*
1789	* Active? Retain the instance and increment the busy counter.
1790	*/
1791	if (vboxNetFltTryRetainBusyActive(pThis))
1792	{
1793	while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1794	vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1795
1796	vboxNetFltRelease(pThis, true /* fBusy */);
1797	}
1798	else
1799	{
1800	/** @todo Shouldn't we just drop the packets here? There is little point in
1801	* making them accumulate when the VM is paused and it'll only waste
1802	* kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1803	* before start draining the packets (goes for the intnet ring buf
1804	* too)? */
1805	}
1806	}
1807	#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1808
1809	/**
1810	* Reports the GSO capabilites of the hardware NIC.
1811	*
1812	* @param pThis The net filter instance. The caller hold a
1813	* reference to this.
1814	*/
1815	static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1816	{
1817	#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1818	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1819	{
1820	struct net_device *pDev;
1821	PINTNETTRUNKSWPORT pSwitchPort;
1822	unsigned int fFeatures;
1823	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1824
1825	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1826
1827	pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1828	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1829	if (pDev)
1830	fFeatures = pDev->features;
1831	else
1832	fFeatures = 0;
1833
1834	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1835
1836	if (pThis->pSwitchPort)
1837	{
1838	/* Set/update the GSO capabilities of the NIC. */
1839	uint32_t fGsoCapabilites = 0;
1840	if (fFeatures & NETIF_F_TSO)
1841	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1842	if (fFeatures & NETIF_F_TSO6)
1843	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1844	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1845	if (fFeatures & NETIF_F_UFO)
1846	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1847	if (fFeatures & NETIF_F_UFO)
1848	fGsoCapabilites \|= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1849	# endif
1850	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1851	}
1852
1853	vboxNetFltRelease(pThis, true /fBusy/);
1854	}
1855	#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1856	}
1857
1858	/**
1859	* Helper that determins whether the host (ignoreing us) is operating the
1860	* interface in promiscuous mode or not.
1861	*/
1862	static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1863	{
1864	bool fRc = false;
1865	struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1866	if (pDev)
1867	{
1868	fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1869	LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1870	fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1871	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1872	}
1873	return fRc;
1874	}
1875
1876	/**
1877	* Internal worker for vboxNetFltLinuxNotifierCallback.
1878	*
1879	* @returns VBox status code.
1880	* @param pThis The instance.
1881	* @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1882	* flood the release log.
1883	*/
1884	static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1885	{
1886	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1887	LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1888
1889	/*
1890	* Retain and store the device.
1891	*/
1892	dev_hold(pDev);
1893
1894	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1895	ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1896	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1897
1898	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1899	Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n", pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1900
1901	/* Get the mac address while we still have a valid net_device reference. */
1902	memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1903
1904	/*
1905	* Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1906	*/
1907	pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1908	pThis->u.s.PacketType.dev = pDev;
1909	pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1910	dev_add_pack(&pThis->u.s.PacketType);
1911
1912	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1913	vboxNetFltLinuxHookDev(pThis, pDev);
1914	#endif
1915	#ifdef VBOXNETFLT_WITH_QDISC
1916	vboxNetFltLinuxQdiscInstall(pThis, pDev);
1917	#endif /* VBOXNETFLT_WITH_QDISC */
1918
1919	/*
1920	* Set indicators that require the spinlock. Be abit paranoid about racing
1921	* the device notification handle.
1922	*/
1923	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1924	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1925	if (pDev)
1926	{
1927	ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1928	ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1929	pDev = NULL; /* don't dereference it */
1930	}
1931	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1932	Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1933
1934	/*
1935	* If the above succeeded report GSO capabilites, if not undo and
1936	* release the device.
1937	*/
1938	if (!pDev)
1939	{
1940	Assert(pThis->pSwitchPort);
1941	if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1942	{
1943	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1944	pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1945	pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1946	pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE \| INTNETTRUNKDIR_HOST);
1947	vboxNetFltRelease(pThis, true /fBusy/);
1948	}
1949	}
1950	else
1951	{
1952	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1953	vboxNetFltLinuxUnhookDev(pThis, pDev);
1954	#endif
1955	#ifdef VBOXNETFLT_WITH_QDISC
1956	vboxNetFltLinuxQdiscRemove(pThis, pDev);
1957	#endif /* VBOXNETFLT_WITH_QDISC */
1958	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1959	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1960	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1961	dev_put(pDev);
1962	Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1963	}
1964
1965	LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1966	return VINF_SUCCESS;
1967	}
1968
1969
1970	static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1971	{
1972	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1973
1974	Assert(!pThis->fDisconnectedFromHost);
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
1983	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1984	ASMAtomicWriteBool(&pThis->u.s.fRegistered, false);
1985	ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1986	ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1987	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1988
1989	dev_remove_pack(&pThis->u.s.PacketType);
1990	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1991	skb_queue_purge(&pThis->u.s.XmitQueue);
1992	#endif
1993	Log(("vboxNetFltLinuxUnregisterDevice: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
1994	Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1995	dev_put(pDev);
1996
1997	return NOTIFY_OK;
1998	}
1999
2000	static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
2001	{
2002	/* Check if we are not suspended and promiscuous mode has not been set. */
2003	if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
2004	&& !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
2005	{
2006	/* Note that there is no need for locking as the kernel got hold of the lock already. */
2007	dev_set_promiscuity(pDev, 1);
2008	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
2009	Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2010	}
2011	else
2012	Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2013	return NOTIFY_OK;
2014	}
2015
2016	static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
2017	{
2018	/* Undo promiscuous mode if we has set it. */
2019	if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
2020	{
2021	/* Note that there is no need for locking as the kernel got hold of the lock already. */
2022	dev_set_promiscuity(pDev, -1);
2023	ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
2024	Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2025	}
2026	else
2027	Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2028	return NOTIFY_OK;
2029	}
2030
2031	#ifdef LOG_ENABLED
2032	/** Stringify the NETDEV_XXX constants. */
2033	static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2034	{
2035	const char *pszEvent = "NETDRV_<unknown>";
2036	switch (ulEventType)
2037	{
2038	case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2039	case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2040	case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2041	case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2042	case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2043	case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2044	case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2045	case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2046	case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2047	case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2048	# ifdef NETDEV_FEAT_CHANGE
2049	case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2050	# endif
2051	}
2052	return pszEvent;
2053	}
2054	#endif /* LOG_ENABLED */
2055
2056	/**
2057	* Callback for listening to netdevice events.
2058	*
2059	* This works the rediscovery, clean up on unregistration, promiscuity on
2060	* up/down, and GSO feature changes from ethtool.
2061	*
2062	* @returns NOTIFY_OK
2063	* @param self Pointer to our notifier registration block.
2064	* @param ulEventType The event.
2065	* @param ptr Event specific, but it is usually the device it
2066	* relates to.
2067	*/
2068	static int vboxNetFltLinuxNotifierCallback(struct notifier_block self, unsigned long ulEventType, void ptr)
2069
2070	{
2071	PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2072	struct net_device pDev = (struct net_device )ptr;
2073	int rc = NOTIFY_OK;
2074
2075	Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2076	vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
2077	if ( ulEventType == NETDEV_REGISTER
2078	&& !strcmp(pDev->name, pThis->szName))
2079	{
2080	vboxNetFltLinuxAttachToInterface(pThis, pDev);
2081	}
2082	else
2083	{
2084	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2085	if (pDev == ptr)
2086	{
2087	switch (ulEventType)
2088	{
2089	case NETDEV_UNREGISTER:
2090	rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2091	break;
2092	case NETDEV_UP:
2093	rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2094	break;
2095	case NETDEV_GOING_DOWN:
2096	rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2097	break;
2098	case NETDEV_CHANGENAME:
2099	break;
2100	#ifdef NETDEV_FEAT_CHANGE
2101	case NETDEV_FEAT_CHANGE:
2102	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2103	break;
2104	#endif
2105	}
2106	}
2107	}
2108
2109	return rc;
2110	}
2111
2112	bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2113	{
2114	return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2115	}
2116
2117	int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2118	{
2119	struct net_device * pDev;
2120	int err;
2121	int rc = VINF_SUCCESS;
2122	NOREF(pvIfData);
2123
2124	LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2125
2126	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2127	if (pDev)
2128	{
2129	/*
2130	* Create a sk_buff for the gather list and push it onto the wire.
2131	*/
2132	if (fDst & INTNETTRUNKDIR_WIRE)
2133	{
2134	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2135	if (pBuf)
2136	{
2137	vboxNetFltDumpPacket(pSG, true, "wire", 1);
2138	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2139	Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2140	err = dev_queue_xmit(pBuf);
2141	if (err)
2142	rc = RTErrConvertFromErrno(err);
2143	}
2144	else
2145	rc = VERR_NO_MEMORY;
2146	}
2147
2148	/*
2149	* Create a sk_buff for the gather list and push it onto the host stack.
2150	*/
2151	if (fDst & INTNETTRUNKDIR_HOST)
2152	{
2153	struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2154	if (pBuf)
2155	{
2156	vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2157	Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2158	Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2159	err = netif_rx_ni(pBuf);
2160	if (err)
2161	rc = RTErrConvertFromErrno(err);
2162	}
2163	else
2164	rc = VERR_NO_MEMORY;
2165	}
2166
2167	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2168	}
2169
2170	return rc;
2171	}
2172
2173
2174	void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2175	{
2176	struct net_device * pDev;
2177
2178	LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
2179	pThis, pThis->szName, fActive?"true":"false",
2180	pThis->fDisablePromiscuous?"true":"false"));
2181
2182	if (pThis->fDisablePromiscuous)
2183	return;
2184
2185	pDev = vboxNetFltLinuxRetainNetDev(pThis);
2186	if (pDev)
2187	{
2188	/*
2189	* This api is a bit weird, the best reference is the code.
2190	*
2191	* Also, we have a bit or race conditions wrt the maintance of
2192	* host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2193	*/
2194	#ifdef LOG_ENABLED
2195	u_int16_t fIf;
2196	unsigned const cPromiscBefore = pDev->promiscuity;
2197	#endif
2198	if (fActive)
2199	{
2200	Assert(!pThis->u.s.fPromiscuousSet);
2201
2202	rtnl_lock();
2203	dev_set_promiscuity(pDev, 1);
2204	rtnl_unlock();
2205	pThis->u.s.fPromiscuousSet = true;
2206	Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2207	}
2208	else
2209	{
2210	if (pThis->u.s.fPromiscuousSet)
2211	{
2212	rtnl_lock();
2213	dev_set_promiscuity(pDev, -1);
2214	rtnl_unlock();
2215	Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2216	}
2217	pThis->u.s.fPromiscuousSet = false;
2218
2219	#ifdef LOG_ENABLED
2220	fIf = dev_get_flags(pDev);
2221	Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2222	#endif
2223	}
2224
2225	vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2226	}
2227	}
2228
2229
2230	int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2231	{
2232	/* Nothing to do here. */
2233	return VINF_SUCCESS;
2234	}
2235
2236
2237	int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2238	{
2239	/*
2240	* Report the GSO capabilities of the host and device (if connected).
2241	* Note! No need to mark ourselves busy here.
2242	*/
2243	/** @todo duplicate work here now? Attach */
2244	#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2245	pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2246	0
2247	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2248	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2249	# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2250	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2251	\| RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2252	# endif
2253	, INTNETTRUNKDIR_HOST);
2254
2255	#endif
2256	vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2257
2258	return VINF_SUCCESS;
2259	}
2260
2261
2262	void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2263	{
2264	struct net_device *pDev;
2265	bool fRegistered;
2266	RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
2267
2268	#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2269	vboxNetFltLinuxUnhookDev(pThis, NULL);
2270	#endif
2271	#ifdef VBOXNETFLT_WITH_QDISC
2272	vboxNetFltLinuxQdiscRemove(pThis, NULL);
2273	#endif /* VBOXNETFLT_WITH_QDISC */
2274
2275	/** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2276	* unlikely, but none the less). Since it doesn't actually update the
2277	* state (just reads it), it is likely to panic in some interesting
2278	* ways. */
2279
2280	RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2281	pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2282	fRegistered = ASMAtomicUoReadBool(&pThis->u.s.fRegistered);
2283	RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2284
2285	if (fRegistered)
2286	{
2287	dev_remove_pack(&pThis->u.s.PacketType);
2288	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2289	skb_queue_purge(&pThis->u.s.XmitQueue);
2290	#endif
2291	Log(("vboxNetFltOsDeleteInstance: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
2292	Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
2293	dev_put(pDev);
2294	}
2295	Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2296	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2297	module_put(THIS_MODULE);
2298	}
2299
2300
2301	int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2302	{
2303	int err;
2304	NOREF(pvContext);
2305
2306	pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2307	err = register_netdevice_notifier(&pThis->u.s.Notifier);
2308	if (err)
2309	return VERR_INTNET_FLT_IF_FAILED;
2310	if (!pThis->u.s.fRegistered)
2311	{
2312	unregister_netdevice_notifier(&pThis->u.s.Notifier);
2313	LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2314	return VERR_INTNET_FLT_IF_NOT_FOUND;
2315	}
2316
2317	Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2318	if ( pThis->fDisconnectedFromHost
2319	\|\| !try_module_get(THIS_MODULE))
2320	return VERR_INTNET_FLT_IF_FAILED;
2321
2322	return VINF_SUCCESS;
2323	}
2324
2325	int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2326	{
2327	/*
2328	* Init the linux specific members.
2329	*/
2330	pThis->u.s.pDev = NULL;
2331	pThis->u.s.fRegistered = false;
2332	pThis->u.s.fPromiscuousSet = false;
2333	memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2334	#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2335	skb_queue_head_init(&pThis->u.s.XmitQueue);
2336	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2337	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2338	# else
2339	INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2340	# endif
2341	#endif
2342
2343	return VINF_SUCCESS;
2344	}
2345
2346
2347	void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2348	{
2349	NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2350	}
2351
2352
2353	int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void pvIf, void *pvIfData)
2354	{
2355	/* Nothing to do */
2356	NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2357	return VINF_SUCCESS;
2358	}
2359
2360
2361	int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2362	{
2363	/* Nothing to do */
2364	NOREF(pThis); NOREF(pvIfData);
2365	return VINF_SUCCESS;
2366	}
2367

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 30771

Download in other formats: