VirtualBox

Ticket #8256: VBoxNetFlt-linux.c

File VBoxNetFlt-linux.c, 87.3 KB (added by Adrienn, 14 years ago)

The patched VBoxNetFlt-linux.c

Line 
1/* $Id: VBoxNetFlt-linux.c 69438 2011-01-14 08:46:18Z aleksey $ */
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/vmm/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#define VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
54#ifdef CONFIG_NET_SCHED
55//# define VBOXNETFLT_WITH_QDISC /* Comment this out to disable qdisc support */
56# ifdef VBOXNETFLT_WITH_QDISC
57# include <net/pkt_sched.h>
58# endif /* VBOXNETFLT_WITH_QDISC */
59#endif
60
61
62/*******************************************************************************
63* Defined Constants And Macros *
64*******************************************************************************/
65#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
66#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
67#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
68# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
69#endif
70
71#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
72# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
73# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
74#else
75# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
76# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
77#endif
78
79#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
80# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
81#else
82# define CHECKSUM_PARTIAL CHECKSUM_HW
83# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
84# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
85# else
86# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
87# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
88# else
89# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
90# endif
91/* Versions prior 2.6.10 use stats for both bstats and qstats */
92# define bstats stats
93# define qstats stats
94# endif
95#endif
96
97#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 13)
98static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
99{
100 kfree_skb(skb);
101 sch->stats.drops++;
102
103 return NET_XMIT_DROP;
104}
105#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 13) */
106
107#ifndef NET_IP_ALIGN
108# define NET_IP_ALIGN 2
109#endif
110
111#if 0
112/** Create scatter / gather segments for fragments. When not used, we will
113 * linearize the socket buffer before creating the internal networking SG. */
114# define VBOXNETFLT_SG_SUPPORT 1
115#endif
116
117#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
118/** Indicates that the linux kernel may send us GSO frames. */
119# define VBOXNETFLT_WITH_GSO 1
120
121/** This enables or disables the transmitting of GSO frame from the internal
122 * network and to the host. */
123# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
124
125# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
126/** This enables or disables the transmitting of GSO frame from the internal
127 * network and to the wire. */
128# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
129# endif
130
131/** This enables or disables the forwarding/flooding of GSO frame from the host
132 * to the internal network. */
133# define VBOXNETFLT_WITH_GSO_RECV 1
134
135#endif
136
137#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
138/** This enables or disables handling of GSO frames coming from the wire (GRO). */
139# define VBOXNETFLT_WITH_GRO 1
140#endif
141/*
142 * GRO support was backported to RHEL 5.4
143 */
144#ifdef RHEL_RELEASE_CODE
145# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
146# define VBOXNETFLT_WITH_GRO 1
147# endif
148#endif
149
150/*******************************************************************************
151* Internal Functions *
152*******************************************************************************/
153static int VBoxNetFltLinuxInit(void);
154static void VBoxNetFltLinuxUnload(void);
155static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
156
157
158/*******************************************************************************
159* Global Variables *
160*******************************************************************************/
161/**
162 * The (common) global data.
163 */
164static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
165
166module_init(VBoxNetFltLinuxInit);
167module_exit(VBoxNetFltLinuxUnload);
168
169MODULE_AUTHOR(VBOX_VENDOR);
170MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
171MODULE_LICENSE("GPL");
172#ifdef MODULE_VERSION
173MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
174#endif
175
176
177#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
178unsigned dev_get_flags(const struct net_device *dev)
179{
180 unsigned flags;
181
182 flags = (dev->flags & ~(IFF_PROMISC |
183 IFF_ALLMULTI |
184 IFF_RUNNING)) |
185 (dev->gflags & (IFF_PROMISC |
186 IFF_ALLMULTI));
187
188 if (netif_running(dev) && netif_carrier_ok(dev))
189 flags |= IFF_RUNNING;
190
191 return flags;
192}
193#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
194
195
196#ifdef VBOXNETFLT_WITH_QDISC
197//#define QDISC_LOG(x) printk x
198# define QDISC_LOG(x) do { } while (0)
199
200# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
201# define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops)
202# elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
203# define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops, parent)
204# elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 37)
205# define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, queue, ops, parent)
206# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37) */
207# define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(queue, ops, parent)
208# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37) */
209
210# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
211# define qdisc_dev(qdisc) (qdisc->dev)
212# define qdisc_pkt_len(skb) (skb->len)
213# define QDISC_GET(dev) (dev->qdisc_sleeping)
214# else
215# define QDISC_GET(dev) (netdev_get_tx_queue(dev, 0)->qdisc_sleeping)
216# endif
217
218# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
219# define QDISC_SAVED_NUM(dev) 1
220# elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
221# define QDISC_SAVED_NUM(dev) dev->num_tx_queues
222# else
223# define QDISC_SAVED_NUM(dev) dev->num_tx_queues+1
224# endif
225
226# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
227# define QDISC_IS_BUSY(dev, qdisc) test_bit(__LINK_STATE_SCHED, &dev->state)
228# elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
229# define QDISC_IS_BUSY(dev, qdisc) (test_bit(__QDISC_STATE_RUNNING, &qdisc->state) || \
230 test_bit(__QDISC_STATE_SCHED, &qdisc->state))
231# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36) */
232# define QDISC_IS_BUSY(dev, qdisc) (qdisc_is_running(qdisc) || \
233 test_bit(__QDISC_STATE_SCHED, &qdisc->state))
234# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36) */
235
236struct VBoxNetQDiscPriv
237{
238 /** Pointer to the single child qdisc. */
239 struct Qdisc *pChild;
240 /*
241 * Technically it is possible to have different qdiscs for different TX
242 * queues so we have to save them all.
243 */
244 /** Pointer to the array of saved qdiscs. */
245 struct Qdisc **ppSaved;
246 /** Pointer to the net filter instance. */
247 PVBOXNETFLTINS pVBoxNetFlt;
248};
249typedef struct VBoxNetQDiscPriv *PVBOXNETQDISCPRIV;
250
251//#define VBOXNETFLT_QDISC_ENQUEUE
252static int vboxNetFltQdiscEnqueue(struct sk_buff *skb, struct Qdisc *sch)
253{
254 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
255 int rc;
256
257# ifdef VBOXNETFLT_QDISC_ENQUEUE
258 if (VALID_PTR(pPriv->pVBoxNetFlt))
259 {
260 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
261 PCRTNETETHERHDR pEtherHdr;
262 PINTNETTRUNKSWPORT pSwitchPort;
263 uint32_t cbHdrs = skb_headlen(skb);
264
265 cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
266 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(skb, 0, cbHdrs, &abHdrBuf[0]);
267 if ( pEtherHdr
268 && (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) != NULL
269 && VALID_PTR(pSwitchPort)
270 && cbHdrs >= 6)
271 {
272 /** @todo consider reference counting, etc. */
273 INTNETSWDECISION enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
274 if (enmDecision == INTNETSWDECISION_INTNET)
275 {
276 struct sk_buff *pBuf = skb_copy(skb, GFP_ATOMIC);
277 pBuf->pkt_type = PACKET_OUTGOING;
278 vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
279 qdisc_drop(skb, sch);
280 ++sch->bstats.packets;
281 sch->bstats.bytes += qdisc_pkt_len(skb);
282 return NET_XMIT_SUCCESS;
283 }
284 }
285 }
286# endif /* VBOXNETFLT_QDISC_ENQUEUE */
287 rc = pPriv->pChild->enqueue(skb, pPriv->pChild);
288 if (rc == NET_XMIT_SUCCESS)
289 {
290 ++sch->q.qlen;
291 ++sch->bstats.packets;
292 sch->bstats.bytes += qdisc_pkt_len(skb);
293 }
294 else
295 ++sch->qstats.drops;
296 return rc;
297}
298
299static struct sk_buff *vboxNetFltQdiscDequeue(struct Qdisc *sch)
300{
301 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
302# ifdef VBOXNETFLT_QDISC_ENQUEUE
303 --sch->q.qlen;
304 return pPriv->pChild->dequeue(pPriv->pChild);
305# else /* VBOXNETFLT_QDISC_ENQUEUE */
306 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
307 PCRTNETETHERHDR pEtherHdr;
308 PINTNETTRUNKSWPORT pSwitchPort;
309 struct sk_buff *pSkb;
310
311 QDISC_LOG(("vboxNetFltDequeue: Enter pThis=%p\n", pPriv->pVBoxNetFlt));
312
313 while ((pSkb = pPriv->pChild->dequeue(pPriv->pChild)) != NULL)
314 {
315 struct sk_buff *pBuf;
316 INTNETSWDECISION enmDecision;
317 uint32_t cbHdrs;
318
319 --sch->q.qlen;
320
321 if (!VALID_PTR(pPriv->pVBoxNetFlt))
322 break;
323
324 cbHdrs = skb_headlen(pSkb);
325 cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
326 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
327 if ( !pEtherHdr
328 || (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) == NULL
329 || !VALID_PTR(pSwitchPort)
330 || cbHdrs < 6)
331 break;
332
333 /** @todo consider reference counting, etc. */
334 enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
335 if (enmDecision != INTNETSWDECISION_INTNET)
336 break;
337
338 pBuf = skb_copy(pSkb, GFP_ATOMIC);
339 pBuf->pkt_type = PACKET_OUTGOING;
340 QDISC_LOG(("vboxNetFltDequeue: pThis=%p\n", pPriv->pVBoxNetFlt));
341 vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
342 qdisc_drop(pSkb, sch);
343 QDISC_LOG(("VBoxNetFlt: Packet for %02x:%02x:%02x:%02x:%02x:%02x dropped\n",
344 pSkb->data[0], pSkb->data[1], pSkb->data[2],
345 pSkb->data[3], pSkb->data[4], pSkb->data[5]));
346 }
347
348 return pSkb;
349# endif /* VBOXNETFLT_QDISC_ENQUEUE */
350}
351
352# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
353static int vboxNetFltQdiscRequeue(struct sk_buff *skb, struct Qdisc *sch)
354{
355 int rc;
356 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
357
358 rc = pPriv->pChild->ops->requeue(skb, pPriv->pChild);
359 if (rc == 0)
360 {
361 sch->q.qlen++;
362# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
363 sch->qstats.requeues++;
364# endif
365 }
366
367 return rc;
368}
369# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
370
371static unsigned int vboxNetFltQdiscDrop(struct Qdisc *sch)
372{
373 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
374 unsigned int cbLen;
375
376 if (pPriv->pChild->ops->drop)
377 {
378 cbLen = pPriv->pChild->ops->drop(pPriv->pChild);
379 if (cbLen != 0)
380 {
381 ++sch->qstats.drops;
382 --sch->q.qlen;
383 return cbLen;
384 }
385 }
386
387 return 0;
388}
389
390# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
391static int vboxNetFltQdiscInit(struct Qdisc *sch, struct rtattr *opt)
392# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
393static int vboxNetFltQdiscInit(struct Qdisc *sch, struct nlattr *opt)
394# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
395{
396 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
397 struct net_device *pDev = qdisc_dev(sch);
398
399 pPriv->pVBoxNetFlt = NULL;
400
401 pPriv->ppSaved = kcalloc(QDISC_SAVED_NUM(pDev), sizeof(pPriv->ppSaved[0]),
402 GFP_KERNEL);
403 if (!pPriv->ppSaved)
404 return -ENOMEM;
405
406 pPriv->pChild = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
407 &pfifo_qdisc_ops,
408 TC_H_MAKE(TC_H_MAJ(sch->handle),
409 TC_H_MIN(1)));
410 if (!pPriv->pChild)
411 {
412 kfree(pPriv->ppSaved);
413 pPriv->ppSaved = NULL;
414 return -ENOMEM;
415 }
416
417 return 0;
418}
419
420static void vboxNetFltQdiscReset(struct Qdisc *sch)
421{
422 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
423
424 qdisc_reset(pPriv->pChild);
425 sch->q.qlen = 0;
426 sch->qstats.backlog = 0;
427}
428
429static void vboxNetFltQdiscDestroy(struct Qdisc* sch)
430{
431 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
432 struct net_device *pDev = qdisc_dev(sch);
433
434 qdisc_destroy(pPriv->pChild);
435 pPriv->pChild = NULL;
436
437 if (pPriv->ppSaved)
438 {
439 int i;
440 for (i = 0; i < QDISC_SAVED_NUM(pDev); i++)
441 if (pPriv->ppSaved[i])
442 qdisc_destroy(pPriv->ppSaved[i]);
443 kfree(pPriv->ppSaved);
444 pPriv->ppSaved = NULL;
445 }
446}
447
448static int vboxNetFltClassGraft(struct Qdisc *sch, unsigned long arg, struct Qdisc *pNew,
449 struct Qdisc **ppOld)
450{
451 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
452
453 if (pNew == NULL)
454 pNew = &noop_qdisc;
455
456 sch_tree_lock(sch);
457 *ppOld = pPriv->pChild;
458 pPriv->pChild = pNew;
459# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
460 sch->q.qlen = 0;
461# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
462 qdisc_tree_decrease_qlen(*ppOld, (*ppOld)->q.qlen);
463# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
464 qdisc_reset(*ppOld);
465 sch_tree_unlock(sch);
466
467 return 0;
468}
469
470static struct Qdisc *vboxNetFltClassLeaf(struct Qdisc *sch, unsigned long arg)
471{
472 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
473 return pPriv->pChild;
474}
475
476static unsigned long vboxNetFltClassGet(struct Qdisc *sch, u32 classid)
477{
478 return 1;
479}
480
481static void vboxNetFltClassPut(struct Qdisc *sch, unsigned long arg)
482{
483}
484
485# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
486static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
487 struct rtattr **tca, unsigned long *arg)
488# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
489static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
490 struct nlattr **tca, unsigned long *arg)
491# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
492{
493 return -ENOSYS;
494}
495
496static int vboxNetFltClassDelete(struct Qdisc *sch, unsigned long arg)
497{
498 return -ENOSYS;
499}
500
501static void vboxNetFltClassWalk(struct Qdisc *sch, struct qdisc_walker *walker)
502{
503 if (!walker->stop) {
504 if (walker->count >= walker->skip)
505 if (walker->fn(sch, 1, walker) < 0) {
506 walker->stop = 1;
507 return;
508 }
509 walker->count++;
510 }
511}
512
513# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
514static struct tcf_proto **vboxNetFltClassFindTcf(struct Qdisc *sch, unsigned long cl)
515{
516 return NULL;
517}
518# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32) */
519
520static int vboxNetFltClassDump(struct Qdisc *sch, unsigned long cl,
521 struct sk_buff *skb, struct tcmsg *tcm)
522{
523 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
524
525 if (cl != 1)
526 return -ENOENT;
527
528 tcm->tcm_handle |= TC_H_MIN(1);
529 tcm->tcm_info = pPriv->pChild->handle;
530
531 return 0;
532}
533
534
535static struct Qdisc_class_ops g_VBoxNetFltClassOps =
536{
537 .graft = vboxNetFltClassGraft,
538 .leaf = vboxNetFltClassLeaf,
539 .get = vboxNetFltClassGet,
540 .put = vboxNetFltClassPut,
541 .change = vboxNetFltClassChange,
542 .delete = vboxNetFltClassDelete,
543 .walk = vboxNetFltClassWalk,
544# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
545 .tcf_chain = vboxNetFltClassFindTcf,
546# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32) */
547 .dump = vboxNetFltClassDump,
548};
549
550
551static struct Qdisc_ops g_VBoxNetFltQDiscOps = {
552 .cl_ops = &g_VBoxNetFltClassOps,
553 .id = "vboxnetflt",
554 .priv_size = sizeof(struct VBoxNetQDiscPriv),
555 .enqueue = vboxNetFltQdiscEnqueue,
556 .dequeue = vboxNetFltQdiscDequeue,
557# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
558 .requeue = vboxNetFltQdiscRequeue,
559# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
560 .peek = qdisc_peek_dequeued,
561# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
562 .drop = vboxNetFltQdiscDrop,
563 .init = vboxNetFltQdiscInit,
564 .reset = vboxNetFltQdiscReset,
565 .destroy = vboxNetFltQdiscDestroy,
566 .owner = THIS_MODULE
567};
568
569/*
570 * If our qdisc is already attached to the device (that means the user
571 * installed it from command line with 'tc' command) we simply update
572 * the pointer to vboxnetflt instance in qdisc's private structure.
573 * Otherwise we need to take some additional steps:
574 * - Create our qdisc;
575 * - Save all references to qdiscs;
576 * - Replace our child with the first qdisc reference;
577 * - Replace all references so they point to our qdisc.
578 */
579static void vboxNetFltLinuxQdiscInstall(PVBOXNETFLTINS pThis, struct net_device *pDev)
580{
581# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
582 int i;
583# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
584 PVBOXNETQDISCPRIV pPriv;
585
586 struct Qdisc *pExisting = QDISC_GET(pDev);
587 /* Do not install our qdisc for devices with no TX queues */
588 if (!pExisting->enqueue)
589 return;
590 if (strcmp(pExisting->ops->id, "vboxnetflt"))
591 {
592 /* The existing qdisc is different from ours, let's create new one. */
593 struct Qdisc *pNew = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
594 &g_VBoxNetFltQDiscOps, TC_H_ROOT);
595 if (!pNew)
596 return; // TODO: Error?
597
598 if (!try_module_get(THIS_MODULE))
599 {
600 /*
601 * This may cause a memory leak but calling qdisc_destroy()
602 * is not an option as it will call module_put().
603 */
604 return;
605 }
606 pPriv = qdisc_priv(pNew);
607
608 qdisc_destroy(pPriv->pChild);
609 pPriv->pChild = QDISC_GET(pDev);
610 atomic_inc(&pPriv->pChild->refcnt);
611 /*
612 * There is no need in deactivating the device or acquiring any locks
613 * prior changing qdiscs since we do not destroy the old qdisc.
614 * Atomic replacement of pointers is enough.
615 */
616 /*
617 * No need to change reference counters here as we merely move
618 * the pointer and the reference counter of the newly allocated
619 * qdisc is already 1.
620 */
621# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
622 pPriv->ppSaved[0] = pDev->qdisc_sleeping;
623 ASMAtomicWritePtr(&pDev->qdisc_sleeping, pNew);
624 ASMAtomicWritePtr(&pDev->qdisc, pNew);
625# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
626 for (i = 0; i < pDev->num_tx_queues; i++)
627 {
628 struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
629
630 pPriv->ppSaved[i] = pQueue->qdisc_sleeping;
631 ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pNew);
632 ASMAtomicWritePtr(&pQueue->qdisc, pNew);
633 if (i)
634 atomic_inc(&pNew->refcnt);
635 }
636 /* Newer kernels store root qdisc in netdev structure as well. */
637# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
638 pPriv->ppSaved[pDev->num_tx_queues] = pDev->qdisc;
639 ASMAtomicWritePtr(&pDev->qdisc, pNew);
640 atomic_inc(&pNew->refcnt);
641# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
642# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
643 /* Sync the queue len with our child */
644 pNew->q.qlen = pPriv->pChild->q.qlen;
645 }
646 else
647 {
648 /* We already have vboxnetflt qdisc, let's use it. */
649 pPriv = qdisc_priv(pExisting);
650 }
651 ASMAtomicWritePtr(&pPriv->pVBoxNetFlt, pThis);
652 QDISC_LOG(("vboxNetFltLinuxInstallQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
653}
654
655static void vboxNetFltLinuxQdiscRemove(PVBOXNETFLTINS pThis, struct net_device *pDev)
656{
657# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
658 int i;
659# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
660 PVBOXNETQDISCPRIV pPriv;
661 struct Qdisc *pQdisc, *pChild;
662 if (!pDev)
663 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
664 if (!VALID_PTR(pDev))
665 {
666 printk("VBoxNetFlt: Failed to detach qdisc, invalid device pointer: %p\n",
667 pDev);
668 return; // TODO: Consider returing an error
669 }
670
671
672 pQdisc = QDISC_GET(pDev);
673 if (strcmp(pQdisc->ops->id, "vboxnetflt"))
674 {
675 if (pQdisc->enqueue)
676 {
677 /* Looks like the user has replaced our qdisc manually. */
678 printk("VBoxNetFlt: Failed to detach qdisc, wrong qdisc: %s\n",
679 pQdisc->ops->id);
680 }
681 return; // TODO: Consider returing an error
682 }
683
684 pPriv = qdisc_priv(pQdisc);
685 Assert(pPriv->pVBoxNetFlt == pThis);
686 ASMAtomicWriteNullPtr(&pPriv->pVBoxNetFlt);
687 pChild = ASMAtomicXchgPtrT(&pPriv->pChild, &noop_qdisc, struct Qdisc *);
688 qdisc_destroy(pChild); /* It won't be the last reference. */
689
690 QDISC_LOG(("vboxNetFltLinuxQdiscRemove: refcnt=%d num_tx_queues=%d\n",
691 atomic_read(&pQdisc->refcnt), pDev->num_tx_queues));
692# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
693 /* Play it safe, make sure the qdisc is not being used. */
694 if (pPriv->ppSaved[0])
695 {
696 ASMAtomicWritePtr(&pDev->qdisc_sleeping, pPriv->ppSaved[0]);
697 ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[0]);
698 pPriv->ppSaved[0] = NULL;
699 while (QDISC_IS_BUSY(pDev, pQdisc))
700 yield();
701 qdisc_destroy(pQdisc); /* Destroy reference */
702 }
703# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
704 for (i = 0; i < pDev->num_tx_queues; i++)
705 {
706 struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
707 if (pPriv->ppSaved[i])
708 {
709 Assert(pQueue->qdisc_sleeping == pQdisc);
710 ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pPriv->ppSaved[i]);
711 ASMAtomicWritePtr(&pQueue->qdisc, pPriv->ppSaved[i]);
712 pPriv->ppSaved[i] = NULL;
713 while (QDISC_IS_BUSY(pDev, pQdisc))
714 yield();
715 qdisc_destroy(pQdisc); /* Destroy reference */
716 }
717 }
718 /* Newer kernels store root qdisc in netdev structure as well. */
719# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
720 ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[pDev->num_tx_queues]);
721 pPriv->ppSaved[pDev->num_tx_queues] = NULL;
722 while (QDISC_IS_BUSY(pDev, pQdisc))
723 yield();
724 qdisc_destroy(pQdisc); /* Destroy reference */
725# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
726# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
727
728 /*
729 * At this point all references to our qdisc should be gone
730 * unless the user had installed it manually.
731 */
732 QDISC_LOG(("vboxNetFltLinuxRemoveQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
733}
734
735#endif /* VBOXNETFLT_WITH_QDISC */
736
737
738/**
739 * Initialize module.
740 *
741 * @returns appropriate status code.
742 */
743static int __init VBoxNetFltLinuxInit(void)
744{
745 int rc;
746 /*
747 * Initialize IPRT.
748 */
749 rc = RTR0Init(0);
750 if (RT_SUCCESS(rc))
751 {
752 Log(("VBoxNetFltLinuxInit\n"));
753
754 /*
755 * Initialize the globals and connect to the support driver.
756 *
757 * This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
758 * for establishing the connect to the support driver.
759 */
760 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
761 rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
762 if (RT_SUCCESS(rc))
763 {
764#ifdef VBOXNETFLT_WITH_QDISC
765 /*memcpy(&g_VBoxNetFltQDiscOps, &pfifo_qdisc_ops, sizeof(g_VBoxNetFltQDiscOps));
766 strcpy(g_VBoxNetFltQDiscOps.id, "vboxnetflt");
767 g_VBoxNetFltQDiscOps.owner = THIS_MODULE;*/
768 rc = register_qdisc(&g_VBoxNetFltQDiscOps);
769 if (rc)
770 {
771 LogRel(("VBoxNetFlt: Failed to registered qdisc: %d\n", rc));
772 return rc;
773 }
774#endif /* VBOXNETFLT_WITH_QDISC */
775 LogRel(("VBoxNetFlt: Successfully started.\n"));
776 return 0;
777 }
778
779 LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
780 RTR0Term();
781 }
782 else
783 LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
784
785 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
786 return -RTErrConvertToErrno(rc);
787}
788
789
790/**
791 * Unload the module.
792 *
793 * @todo We have to prevent this if we're busy!
794 */
795static void __exit VBoxNetFltLinuxUnload(void)
796{
797 int rc;
798 Log(("VBoxNetFltLinuxUnload\n"));
799 Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
800
801#ifdef VBOXNETFLT_WITH_QDISC
802 unregister_qdisc(&g_VBoxNetFltQDiscOps);
803#endif /* VBOXNETFLT_WITH_QDISC */
804 /*
805 * Undo the work done during start (in reverse order).
806 */
807 rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
808 AssertRC(rc); NOREF(rc);
809
810 RTR0Term();
811
812 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
813
814 Log(("VBoxNetFltLinuxUnload - done\n"));
815}
816
817
818/**
819 * Experiment where we filter traffic from the host to the internal network
820 * before it reaches the NIC driver.
821 *
822 * The current code uses a very ugly hack and only works on kernels using the
823 * net_device_ops (>= 2.6.29). It has been shown to give us a
824 * performance boost of 60-100% though. So, we have to find some less hacky way
825 * of getting this job done eventually.
826 *
827 * #define VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
828 */
829#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
830
831# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
832
833# include <linux/ethtool.h>
834
835typedef struct ethtool_ops OVR_OPSTYPE;
836# define OVR_OPS ethtool_ops
837# define OVR_XMIT pfnStartXmit
838
839# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
840
841typedef struct net_device_ops OVR_OPSTYPE;
842# define OVR_OPS netdev_ops
843# define OVR_XMIT pOrgOps->ndo_start_xmit
844
845# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
846
847/**
848 * The overridden net_device_ops of the device we're attached to.
849 *
850 * As there is no net_device_ops structure in pre-2.6.29 kernels we override
851 * ethtool_ops instead along with hard_start_xmit callback in net_device
852 * structure.
853 *
854 * This is a very dirty hack that was created to explore how much we can improve
855 * the host to guest transfers by not CC'ing the NIC. It turns out to be
856 * the only way to filter outgoing packets for devices without TX queue.
857 */
858typedef struct VBoxNetDeviceOpsOverride
859{
860 /** Our overridden ops. */
861 OVR_OPSTYPE Ops;
862 /** Magic word. */
863 uint32_t u32Magic;
864 /** Pointer to the original ops. */
865 OVR_OPSTYPE const *pOrgOps;
866# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
867 /** Pointer to the original hard_start_xmit function. */
868 int (*pfnStartXmit)(struct sk_buff *pSkb, struct net_device *pDev);
869# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
870 /** Pointer to the net filter instance. */
871 PVBOXNETFLTINS pVBoxNetFlt;
872 /** The number of filtered packages. */
873 uint64_t cFiltered;
874 /** The total number of packets */
875 uint64_t cTotal;
876} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
877/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
878#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
879
880/**
881 * ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
882 * because they belong on the internal network.
883 *
884 * @returns NETDEV_TX_XXX.
885 * @param pSkb The socket buffer to transmit.
886 * @param pDev The net device.
887 */
888static int vboxNetFltLinuxStartXmitFilter(struct sk_buff *pSkb, struct net_device *pDev)
889{
890 PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
891 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
892 PCRTNETETHERHDR pEtherHdr;
893 PINTNETTRUNKSWPORT pSwitchPort;
894 uint32_t cbHdrs;
895
896
897 /*
898 * Validate the override structure.
899 *
900 * Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
901 * to be production quality code, we would have to be much more
902 * careful here and avoid the race.
903 */
904 if ( !VALID_PTR(pOverride)
905 || pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
906 || !VALID_PTR(pOverride->pOrgOps))
907 {
908 printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
909 dev_kfree_skb(pSkb);
910 return NETDEV_TX_OK;
911 }
912 pOverride->cTotal++;
913
914 /*
915 * Do the filtering base on the default OUI of our virtual NICs
916 *
917 * Note! In a real solution, we would ask the switch whether the
918 * destination MAC is 100% to be on the internal network and then
919 * drop it.
920 */
921 cbHdrs = skb_headlen(pSkb);
922 cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
923 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
924 if ( pEtherHdr
925 && VALID_PTR(pOverride->pVBoxNetFlt)
926 && (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
927 && VALID_PTR(pSwitchPort)
928 && cbHdrs >= 6)
929 {
930 INTNETSWDECISION enmDecision;
931
932 /** @todo consider reference counting, etc. */
933 enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
934 if (enmDecision == INTNETSWDECISION_INTNET)
935 {
936 dev_kfree_skb(pSkb);
937 pOverride->cFiltered++;
938 return NETDEV_TX_OK;
939 }
940 }
941
942 return pOverride->OVR_XMIT(pSkb, pDev);
943}
944
945/**
946 * Hooks the device ndo_start_xmit operation of the device.
947 *
948 * @param pThis The net filter instance.
949 * @param pDev The net device.
950 */
951static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
952{
953 PVBOXNETDEVICEOPSOVERRIDE pOverride;
954 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
955
956 pOverride = RTMemAlloc(sizeof(*pOverride));
957 if (!pOverride)
958 return;
959 pOverride->pOrgOps = pDev->OVR_OPS;
960 pOverride->Ops = *pDev->OVR_OPS;
961# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
962 pOverride->pfnStartXmit = pDev->hard_start_xmit;
963# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
964 pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
965# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
966 pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
967 pOverride->cTotal = 0;
968 pOverride->cFiltered = 0;
969 pOverride->pVBoxNetFlt = pThis;
970
971 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp); /* (this isn't necessary, but so what) */
972 ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
973# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
974 ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
975# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
976 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
977}
978
979/**
980 * Undos what vboxNetFltLinuxHookDev did.
981 *
982 * @param pThis The net filter instance.
983 * @param pDev The net device. Can be NULL, in which case
984 * we'll try retrieve it from @a pThis.
985 */
986static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
987{
988 PVBOXNETDEVICEOPSOVERRIDE pOverride;
989 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
990
991 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
992 if (!pDev)
993 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
994 if (VALID_PTR(pDev))
995 {
996 pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
997 if ( VALID_PTR(pOverride)
998 && pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
999 && VALID_PTR(pOverride->pOrgOps)
1000 )
1001 {
1002# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
1003 ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
1004# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
1005 ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
1006 ASMAtomicWriteU32(&pOverride->u32Magic, 0);
1007 }
1008 else
1009 pOverride = NULL;
1010 }
1011 else
1012 pOverride = NULL;
1013 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1014
1015 if (pOverride)
1016 {
1017 printk("vboxnetflt: dropped %llu out of %llu packets\n", pOverride->cFiltered, pOverride->cTotal);
1018 RTMemFree(pOverride);
1019 }
1020}
1021
1022#endif /* VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT */
1023
1024
1025/**
1026 * Reads and retains the host interface handle.
1027 *
1028 * @returns The handle, NULL if detached.
1029 * @param pThis
1030 */
1031DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
1032{
1033#if 0
1034 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1035 struct net_device *pDev = NULL;
1036
1037 Log(("vboxNetFltLinuxRetainNetDev\n"));
1038 /*
1039 * Be careful here to avoid problems racing the detached callback.
1040 */
1041 RTSpinlockAcquire(pThis->hSpinlock, &Tmp);
1042 if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
1043 {
1044 pDev = (struct net_device *)ASMAtomicUoReadPtr((void * volatile *)&pThis->u.s.pDev);
1045 if (pDev)
1046 {
1047 dev_hold(pDev);
1048 Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
1049 pDev, pDev->name,
1050#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1051 netdev_refcnt_read(pDev)
1052#else
1053 atomic_read(&pDev->refcnt)
1054#endif
1055 ));
1056 }
1057 }
1058 RTSpinlockRelease(pThis->hSpinlock, &Tmp);
1059
1060 Log(("vboxNetFltLinuxRetainNetDev - done\n"));
1061 return pDev;
1062#else
1063 return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1064#endif
1065}
1066
1067
1068/**
1069 * Release the host interface handle previously retained
1070 * by vboxNetFltLinuxRetainNetDev.
1071 *
1072 * @param pThis The instance.
1073 * @param pDev The vboxNetFltLinuxRetainNetDev
1074 * return value, NULL is fine.
1075 */
1076DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
1077{
1078#if 0
1079 Log(("vboxNetFltLinuxReleaseNetDev\n"));
1080 NOREF(pThis);
1081 if (pDev)
1082 {
1083 dev_put(pDev);
1084 Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
1085 pDev, pDev->name,
1086#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1087 netdev_refcnt_read(pDev)
1088#else
1089 atomic_read(&pDev->refcnt)
1090#endif
1091 ));
1092 }
1093 Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
1094#endif
1095}
1096
1097#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 | (skb->dev->ifindex & 0xFFFF))
1098#define VBOXNETFLT_SKB_TAG(skb) (*(uint32_t*)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
1099
1100/**
1101 * Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
1102 * i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
1103 *
1104 * @returns true / false accordingly.
1105 * @param pBuf The sk_buff.
1106 */
1107DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
1108{
1109 return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
1110}
1111
1112
1113/**
1114 * Internal worker that create a linux sk_buff for a
1115 * (scatter/)gather list.
1116 *
1117 * @returns Pointer to the sk_buff.
1118 * @param pThis The instance.
1119 * @param pSG The (scatter/)gather list.
1120 * @param fDstWire Set if the destination is the wire.
1121 */
1122static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
1123{
1124 struct sk_buff *pPkt;
1125 struct net_device *pDev;
1126 unsigned fGsoType = 0;
1127
1128 if (pSG->cbTotal == 0)
1129 {
1130 LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
1131 return NULL;
1132 }
1133
1134 /** @todo We should use fragments mapping the SG buffers with large packets.
1135 * 256 bytes seems to be the a threshold used a lot for this. It
1136 * requires some nasty work on the intnet side though... */
1137 /*
1138 * Allocate a packet and copy over the data.
1139 */
1140 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1141 pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
1142 if (RT_UNLIKELY(!pPkt))
1143 {
1144 Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
1145 pSG->pvUserData = NULL;
1146 return NULL;
1147 }
1148 pPkt->dev = pDev;
1149 pPkt->ip_summed = CHECKSUM_NONE;
1150
1151 /* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
1152 skb_reserve(pPkt, NET_IP_ALIGN);
1153
1154 /* Copy the segments. */
1155 skb_put(pPkt, pSG->cbTotal);
1156 IntNetSgRead(pSG, pPkt->data);
1157
1158#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1159 /*
1160 * Setup GSO if used by this packet.
1161 */
1162 switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
1163 {
1164 default:
1165 AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
1166 /* fall thru */
1167 case PDMNETWORKGSOTYPE_INVALID:
1168 fGsoType = 0;
1169 break;
1170 case PDMNETWORKGSOTYPE_IPV4_TCP:
1171 fGsoType = SKB_GSO_TCPV4;
1172 break;
1173 case PDMNETWORKGSOTYPE_IPV4_UDP:
1174 fGsoType = SKB_GSO_UDP;
1175 break;
1176 case PDMNETWORKGSOTYPE_IPV6_TCP:
1177 fGsoType = SKB_GSO_TCPV6;
1178 break;
1179 }
1180 if (fGsoType)
1181 {
1182 struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
1183
1184 pShInfo->gso_type = fGsoType | SKB_GSO_DODGY;
1185 pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
1186 pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
1187
1188 /*
1189 * We need to set checksum fields even if the packet goes to the host
1190 * directly as it may be immediately forwarded by IP layer @bugref{5020}.
1191 */
1192 Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrs);
1193 pPkt->ip_summed = CHECKSUM_PARTIAL;
1194# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1195 pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
1196 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
1197 pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
1198 else
1199 pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
1200# else
1201 pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
1202 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
1203 pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
1204 else
1205 pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
1206# endif
1207 if (!fDstWire)
1208 PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
1209 }
1210#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE || VBOXNETFLT_WITH_GSO_XMIT_HOST */
1211
1212 /*
1213 * Finish up the socket buffer.
1214 */
1215 pPkt->protocol = eth_type_trans(pPkt, pDev);
1216 if (fDstWire)
1217 {
1218 VBOX_SKB_RESET_NETWORK_HDR(pPkt);
1219
1220 /* Restore ethernet header back. */
1221 skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
1222 VBOX_SKB_RESET_MAC_HDR(pPkt);
1223 }
1224 VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
1225
1226 return pPkt;
1227}
1228
1229
1230/**
1231 * Initializes a SG list from an sk_buff.
1232 *
1233 * @returns Number of segments.
1234 * @param pThis The instance.
1235 * @param pBuf The sk_buff.
1236 * @param pSG The SG.
1237 * @param pvFrame The frame pointer, optional.
1238 * @param cSegs The number of segments allocated for the SG.
1239 * This should match the number in the mbuf exactly!
1240 * @param fSrc The source of the frame.
1241 * @param pGso Pointer to the GSO context if it's a GSO
1242 * internal network frame. NULL if regular frame.
1243 */
1244DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
1245 unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1246{
1247 int i;
1248 NOREF(pThis);
1249
1250 Assert(!skb_shinfo(pBuf)->frag_list);
1251
1252 if (!pGsoCtx)
1253 IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/);
1254 else
1255 IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/, pGsoCtx);
1256
1257#ifdef VBOXNETFLT_SG_SUPPORT
1258 pSG->aSegs[0].cb = skb_headlen(pBuf);
1259 pSG->aSegs[0].pv = pBuf->data;
1260 pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1261
1262 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1263 {
1264 skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
1265 pSG->aSegs[i+1].cb = pFrag->size;
1266 pSG->aSegs[i+1].pv = kmap(pFrag->page);
1267 printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
1268 pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
1269 }
1270 ++i;
1271
1272#else
1273 pSG->aSegs[0].cb = pBuf->len;
1274 pSG->aSegs[0].pv = pBuf->data;
1275 pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1276 i = 1;
1277#endif
1278
1279 pSG->cSegsUsed = i;
1280
1281#ifdef PADD_RUNT_FRAMES_FROM_HOST
1282 /*
1283 * Add a trailer if the frame is too small.
1284 *
1285 * Since we're getting to the packet before it is framed, it has not
1286 * yet been padded. The current solution is to add a segment pointing
1287 * to a buffer containing all zeros and pray that works for all frames...
1288 */
1289 if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
1290 {
1291 static uint8_t const s_abZero[128] = {0};
1292
1293 AssertReturnVoid(i < cSegs);
1294
1295 pSG->aSegs[i].Phys = NIL_RTHCPHYS;
1296 pSG->aSegs[i].pv = (void *)&s_abZero[0];
1297 pSG->aSegs[i].cb = 60 - pSG->cbTotal;
1298 pSG->cbTotal = 60;
1299 pSG->cSegsUsed++;
1300 Assert(i + 1 <= pSG->cSegsAlloc)
1301 }
1302#endif
1303
1304 Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
1305 pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
1306 for (i = 0; i < pSG->cSegsUsed; i++)
1307 Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
1308 i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1309}
1310
1311/**
1312 * Packet handler,
1313 *
1314 * @returns 0 or EJUSTRETURN.
1315 * @param pThis The instance.
1316 * @param pMBuf The mbuf.
1317 * @param pvFrame The start of the frame, optional.
1318 * @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
1319 * @param eProtocol The protocol.
1320 */
1321#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
1322static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1323 struct net_device *pSkbDev,
1324 struct packet_type *pPacketType,
1325 struct net_device *pOrigDev)
1326#else
1327static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1328 struct net_device *pSkbDev,
1329 struct packet_type *pPacketType)
1330#endif
1331{
1332 PVBOXNETFLTINS pThis;
1333 struct net_device *pDev;
1334 LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1335 pBuf, pSkbDev, pPacketType));
1336#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1337 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",
1338 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));
1339 Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1340#else
1341 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",
1342 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));
1343#endif
1344 /*
1345 * Drop it immediately?
1346 */
1347 if (!pBuf)
1348 return 0;
1349
1350 pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1351 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1352 if (pThis->u.s.pDev != pSkbDev)
1353 {
1354 Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1355 return 0;
1356 }
1357
1358 Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1359 if (vboxNetFltLinuxSkBufIsOur(pBuf))
1360 {
1361 Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1362 dev_kfree_skb(pBuf);
1363 return 0;
1364 }
1365
1366#ifndef VBOXNETFLT_SG_SUPPORT
1367 {
1368 /*
1369 * Get rid of fragmented packets, they cause too much trouble.
1370 */
1371 struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1372 kfree_skb(pBuf);
1373 if (!pCopy)
1374 {
1375 LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1376 return 0;
1377 }
1378 pBuf = pCopy;
1379# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1380 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",
1381 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));
1382 Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1383# else
1384 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",
1385 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));
1386# endif
1387 }
1388#endif
1389
1390#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1391 /* Forward it to the internal network. */
1392 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1393#else
1394 /* Add the packet to transmit queue and schedule the bottom half. */
1395 skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1396 schedule_work(&pThis->u.s.XmitTask);
1397 Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1398 &pThis->u.s.XmitTask, pBuf));
1399#endif
1400
1401 /* It does not really matter what we return, it is ignored by the kernel. */
1402 return 0;
1403}
1404
1405/**
1406 * Calculate the number of INTNETSEG segments the socket buffer will need.
1407 *
1408 * @returns Segment count.
1409 * @param pBuf The socket buffer.
1410 */
1411DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
1412{
1413#ifdef VBOXNETFLT_SG_SUPPORT
1414 unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1415#else
1416 unsigned cSegs = 1;
1417#endif
1418#ifdef PADD_RUNT_FRAMES_FROM_HOST
1419 /* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1420 if (pBuf->len < 60)
1421 cSegs++;
1422#endif
1423 return cSegs;
1424}
1425
1426/**
1427 * Destroy the intnet scatter / gather buffer created by
1428 * vboxNetFltLinuxSkBufToSG.
1429 */
1430static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
1431{
1432#ifdef VBOXNETFLT_SG_SUPPORT
1433 int i;
1434
1435 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1436 {
1437 printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
1438 kunmap(pSG->aSegs[i+1].pv);
1439 }
1440#endif
1441 NOREF(pSG);
1442}
1443
1444#ifdef LOG_ENABLED
1445/**
1446 * Logging helper.
1447 */
1448static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1449{
1450 uint8_t *pInt, *pExt;
1451 static int iPacketNo = 1;
1452 iPacketNo += iIncrement;
1453 if (fEgress)
1454 {
1455 pExt = pSG->aSegs[0].pv;
1456 pInt = pExt + 6;
1457 }
1458 else
1459 {
1460 pInt = pSG->aSegs[0].pv;
1461 pExt = pInt + 6;
1462 }
1463 Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1464 " %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1465 pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1466 fEgress ? "-->" : "<--", pszWhere,
1467 pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1468 pSG->cbTotal, iPacketNo));
1469 Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1470}
1471#else
1472# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1473#endif
1474
1475#ifdef VBOXNETFLT_WITH_GSO_RECV
1476
1477/**
1478 * Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1479 * GSO socket buffer without having to segment it.
1480 *
1481 * @returns true on success, false if needs segmenting.
1482 * @param pThis The net filter instance.
1483 * @param pSkb The GSO socket buffer.
1484 * @param fSrc The source.
1485 * @param pGsoCtx Where to return the GSO context on success.
1486 */
1487static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1488 PPDMNETWORKGSO pGsoCtx)
1489{
1490 PDMNETWORKGSOTYPE enmGsoType;
1491 uint16_t uEtherType;
1492 unsigned int cbTransport;
1493 unsigned int offTransport;
1494 unsigned int cbTransportHdr;
1495 unsigned uProtocol;
1496 union
1497 {
1498 RTNETIPV4 IPv4;
1499 RTNETIPV6 IPv6;
1500 RTNETTCP Tcp;
1501 uint8_t ab[40];
1502 uint16_t au16[40/2];
1503 uint32_t au32[40/4];
1504 } Buf;
1505
1506 /*
1507 * Check the GSO properties of the socket buffer and make sure it fits.
1508 */
1509 /** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1510 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP | SKB_GSO_DODGY | SKB_GSO_TCPV6 | SKB_GSO_TCPV4) ))
1511 {
1512 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1513 return false;
1514 }
1515 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1516 || pSkb->len > VBOX_MAX_GSO_SIZE ))
1517 {
1518 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1519 return false;
1520 }
1521 /*
1522 * It is possible to receive GSO packets from wire if GRO is enabled.
1523 */
1524 if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1525 {
1526 Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1527#ifdef VBOXNETFLT_WITH_GRO
1528 /*
1529 * The packet came from the wire and the driver has already consumed
1530 * mac header. We need to restore it back.
1531 */
1532 pSkb->mac_len = skb_network_header(pSkb) - skb_mac_header(pSkb);
1533 skb_push(pSkb, pSkb->mac_len);
1534 Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1535 pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
1536#else /* !VBOXNETFLT_WITH_GRO */
1537 /* Older kernels didn't have GRO. */
1538 return false;
1539#endif /* !VBOXNETFLT_WITH_GRO */
1540 }
1541 else
1542 {
1543 /*
1544 * skb_gso_segment does the following. Do we need to do it as well?
1545 */
1546#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1547 skb_reset_mac_header(pSkb);
1548 pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1549#else
1550 pSkb->mac.raw = pSkb->data;
1551 pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1552#endif
1553 }
1554
1555 /*
1556 * Switch on the ethertype.
1557 */
1558 uEtherType = pSkb->protocol;
1559 if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1560 && pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1561 {
1562 uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1563 if (puEtherType)
1564 uEtherType = *puEtherType;
1565 }
1566 switch (uEtherType)
1567 {
1568 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1569 {
1570 unsigned int cbHdr;
1571 PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1572 if (RT_UNLIKELY(!pIPv4))
1573 {
1574 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1575 return false;
1576 }
1577
1578 cbHdr = pIPv4->ip_hl * 4;
1579 cbTransport = RT_N2H_U16(pIPv4->ip_len);
1580 if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1581 || cbHdr > cbTransport ))
1582 {
1583 Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1584 return false;
1585 }
1586 cbTransport -= cbHdr;
1587 offTransport = pSkb->mac_len + cbHdr;
1588 uProtocol = pIPv4->ip_p;
1589 if (uProtocol == RTNETIPV4_PROT_TCP)
1590 enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1591 else if (uProtocol == RTNETIPV4_PROT_UDP)
1592 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1593 else /** @todo IPv6: 4to6 tunneling */
1594 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1595 break;
1596 }
1597
1598 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1599 {
1600 PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1601 if (RT_UNLIKELY(!pIPv6))
1602 {
1603 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1604 return false;
1605 }
1606
1607 cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1608 offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1609 uProtocol = pIPv6->ip6_nxt;
1610 /** @todo IPv6: Dig our way out of the other headers. */
1611 if (uProtocol == RTNETIPV4_PROT_TCP)
1612 enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1613 else if (uProtocol == RTNETIPV4_PROT_UDP)
1614 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1615 else
1616 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1617 break;
1618 }
1619
1620 default:
1621 Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1622 return false;
1623 }
1624
1625 if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1626 {
1627 Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1628 return false;
1629 }
1630
1631 if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1632 || offTransport + cbTransport > pSkb->len
1633 || cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1634 {
1635 Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1636 offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1637 return false;
1638 }
1639
1640 /*
1641 * Check the TCP/UDP bits.
1642 */
1643 if (uProtocol == RTNETIPV4_PROT_TCP)
1644 {
1645 PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1646 if (RT_UNLIKELY(!pTcp))
1647 {
1648 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1649 return false;
1650 }
1651
1652 cbTransportHdr = pTcp->th_off * 4;
1653 if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1654 || cbTransportHdr > cbTransport
1655 || offTransport + cbTransportHdr >= UINT8_MAX
1656 || offTransport + cbTransportHdr >= pSkb->len ))
1657 {
1658 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1659 return false;
1660 }
1661
1662 }
1663 else
1664 {
1665 Assert(uProtocol == RTNETIPV4_PROT_UDP);
1666 cbTransportHdr = sizeof(RTNETUDP);
1667 if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1668 || offTransport + cbTransportHdr >= pSkb->len ))
1669 {
1670 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1671 return false;
1672 }
1673 }
1674
1675 /*
1676 * We're good, init the GSO context.
1677 */
1678 pGsoCtx->u8Type = enmGsoType;
1679 pGsoCtx->cbHdrs = offTransport + cbTransportHdr;
1680 pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1681 pGsoCtx->offHdr1 = pSkb->mac_len;
1682 pGsoCtx->offHdr2 = offTransport;
1683 pGsoCtx->au8Unused[0] = 0;
1684 pGsoCtx->au8Unused[1] = 0;
1685
1686 return true;
1687}
1688
1689/**
1690 * Forward the socket buffer as a GSO internal network frame.
1691 *
1692 * @returns IPRT status code.
1693 * @param pThis The net filter instance.
1694 * @param pSkb The GSO socket buffer.
1695 * @param fSrc The source.
1696 * @param pGsoCtx Where to return the GSO context on success.
1697 */
1698static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1699{
1700 int rc;
1701 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1702 if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1703 {
1704 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1705 if (RT_LIKELY(pSG))
1706 {
1707 vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1708
1709 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1710 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1711
1712 vboxNetFltLinuxDestroySG(pSG);
1713 rc = VINF_SUCCESS;
1714 }
1715 else
1716 {
1717 Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1718 rc = VERR_NO_MEMORY;
1719 }
1720 }
1721 else
1722 {
1723 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1724 rc = VERR_INTERNAL_ERROR_3;
1725 }
1726
1727 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1728 dev_kfree_skb(pSkb);
1729 return rc;
1730}
1731
1732#endif /* VBOXNETFLT_WITH_GSO_RECV */
1733
1734/**
1735 * Worker for vboxNetFltLinuxForwardToIntNet.
1736 *
1737 * @returns VINF_SUCCESS or VERR_NO_MEMORY.
1738 * @param pThis The net filter instance.
1739 * @param pBuf The socket buffer.
1740 * @param fSrc The source.
1741 */
1742static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1743{
1744 int rc;
1745 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1746 if (cSegs <= MAX_SKB_FRAGS + 1)
1747 {
1748 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1749 if (RT_LIKELY(pSG))
1750 {
1751 if (fSrc & INTNETTRUNKDIR_WIRE)
1752 {
1753 /*
1754 * The packet came from wire, ethernet header was removed by device driver.
1755 * Restore it.
1756 */
1757 skb_push(pBuf, ETH_HLEN);
1758 }
1759
1760 vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /*pGsoCtx*/);
1761
1762 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1763 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1764
1765 vboxNetFltLinuxDestroySG(pSG);
1766 rc = VINF_SUCCESS;
1767 }
1768 else
1769 {
1770 Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1771 rc = VERR_NO_MEMORY;
1772 }
1773 }
1774 else
1775 {
1776 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1777 rc = VERR_INTERNAL_ERROR_3;
1778 }
1779
1780 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1781 dev_kfree_skb(pBuf);
1782 return rc;
1783}
1784
1785/**
1786 *
1787 * @param pBuf The socket buffer. This is consumed by this function.
1788 */
1789static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1790{
1791 uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1792
1793#ifdef VBOXNETFLT_WITH_GSO
1794 if (skb_is_gso(pBuf))
1795 {
1796 PDMNETWORKGSO GsoCtx;
1797 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",
1798 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));
1799# ifdef VBOXNETFLT_WITH_GSO_RECV
1800 if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP | SKB_GSO_TCPV6 | SKB_GSO_TCPV4))
1801 && vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1802 vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1803 else
1804# endif
1805 {
1806 /* Need to segment the packet */
1807 struct sk_buff *pNext;
1808 struct sk_buff *pSegment = skb_gso_segment(pBuf, 0 /*supported features*/);
1809 if (IS_ERR(pSegment))
1810 {
1811 dev_kfree_skb(pBuf);
1812 LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1813 return;
1814 }
1815
1816 for (; pSegment; pSegment = pNext)
1817 {
1818 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",
1819 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));
1820 pNext = pSegment->next;
1821 pSegment->next = 0;
1822 vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1823 }
1824 dev_kfree_skb(pBuf);
1825 }
1826 }
1827 else
1828#endif /* VBOXNETFLT_WITH_GSO */
1829 {
1830 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1831 {
1832#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1833 /*
1834 * Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1835 * and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1836 * header length from the header itself and reconstruct 'h' pointer
1837 * to TCP (or whatever) header.
1838 */
1839 unsigned char *tmp = pBuf->h.raw;
1840 if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1841 pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1842#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1843 if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1844 {
1845 LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1846 dev_kfree_skb(pBuf);
1847 return;
1848 }
1849#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1850 /* Restore the original (wrong) pointer. */
1851 pBuf->h.raw = tmp;
1852#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1853 }
1854 vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1855 }
1856}
1857
1858#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1859/**
1860 * Work queue handler that forwards the socket buffers queued by
1861 * vboxNetFltLinuxPacketHandler to the internal network.
1862 *
1863 * @param pWork The work queue.
1864 */
1865# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1866static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1867# else
1868static void vboxNetFltLinuxXmitTask(void *pWork)
1869# endif
1870{
1871 PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1872 struct sk_buff *pBuf;
1873
1874 Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1875
1876 /*
1877 * Active? Retain the instance and increment the busy counter.
1878 */
1879 if (vboxNetFltTryRetainBusyActive(pThis))
1880 {
1881 while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1882 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1883
1884 vboxNetFltRelease(pThis, true /* fBusy */);
1885 }
1886 else
1887 {
1888 /** @todo Shouldn't we just drop the packets here? There is little point in
1889 * making them accumulate when the VM is paused and it'll only waste
1890 * kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1891 * before start draining the packets (goes for the intnet ring buf
1892 * too)? */
1893 }
1894}
1895#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1896
1897/**
1898 * Reports the GSO capabilities of the hardware NIC.
1899 *
1900 * @param pThis The net filter instance. The caller hold a
1901 * reference to this.
1902 */
1903static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1904{
1905#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1906 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1907 {
1908 struct net_device *pDev;
1909 PINTNETTRUNKSWPORT pSwitchPort;
1910 unsigned int fFeatures;
1911 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1912
1913 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1914
1915 pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1916 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1917 if (pDev)
1918 fFeatures = pDev->features;
1919 else
1920 fFeatures = 0;
1921
1922 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1923
1924 if (pThis->pSwitchPort)
1925 {
1926 /* Set/update the GSO capabilities of the NIC. */
1927 uint32_t fGsoCapabilites = 0;
1928 if (fFeatures & NETIF_F_TSO)
1929 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1930 if (fFeatures & NETIF_F_TSO6)
1931 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1932# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1933 if (fFeatures & NETIF_F_UFO)
1934 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1935 if (fFeatures & NETIF_F_UFO)
1936 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1937# endif
1938 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1939 }
1940
1941 vboxNetFltRelease(pThis, true /*fBusy*/);
1942 }
1943#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1944}
1945
1946/**
1947 * Helper that determines whether the host (ignoreing us) is operating the
1948 * interface in promiscuous mode or not.
1949 */
1950static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1951{
1952 bool fRc = false;
1953 struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1954 if (pDev)
1955 {
1956 fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1957 LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1958 fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1959 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1960 }
1961 return fRc;
1962}
1963
1964#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36)
1965/**
1966 * Helper for detecting TAP devices.
1967 */
1968static bool vboxNetFltIsTapDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1969{
1970 if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1971 {
1972 struct ethtool_drvinfo Info;
1973
1974 memset(&Info, 0, sizeof(Info));
1975 Info.cmd = ETHTOOL_GDRVINFO;
1976 pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1977 Log3(("vboxNetFltIsTapDevice: driver=%s version=%s bus_info=%s\n",
1978 Info.driver, Info.version, Info.bus_info));
1979
1980 return !strncmp(Info.driver, "tun", 4)
1981 && !strncmp(Info.bus_info, "tap", 4);
1982 }
1983
1984 return false;
1985}
1986
1987/**
1988 * Helper for updating the link state of TAP devices.
1989 * Only TAP devices are affected.
1990 */
1991static void vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1992{
1993 if (vboxNetFltIsTapDevice(pThis, pDev))
1994 {
1995 Log3(("vboxNetFltSetTapLinkState: bringing %s tap device link state\n",
1996 fLinkUp ? "up" : "down"));
1997 netif_tx_lock_bh(pDev);
1998 if (fLinkUp)
1999 netif_carrier_on(pDev);
2000 else
2001 netif_carrier_off(pDev);
2002 netif_tx_unlock_bh(pDev);
2003 }
2004}
2005#else /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
2006DECLINLINE(void) vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
2007{
2008 /* Nothing to do for pre-2.6.36 kernels. */
2009}
2010#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
2011
2012/**
2013 * Internal worker for vboxNetFltLinuxNotifierCallback.
2014 *
2015 * @returns VBox status code.
2016 * @param pThis The instance.
2017 * @param fRediscovery If set we're doing a rediscovery attempt, so, don't
2018 * flood the release log.
2019 */
2020static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
2021{
2022 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
2023 LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
2024
2025 /*
2026 * Retain and store the device.
2027 */
2028 dev_hold(pDev);
2029
2030 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2031 ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
2032 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2033
2034 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
2035 pDev, pDev->name,
2036#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
2037 netdev_refcnt_read(pDev)
2038#else
2039 atomic_read(&pDev->refcnt)
2040#endif
2041 ));
2042 Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2043 pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
2044
2045 /* Get the mac address while we still have a valid net_device reference. */
2046 memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
2047
2048 /*
2049 * Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
2050 */
2051 pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
2052 pThis->u.s.PacketType.dev = pDev;
2053 pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
2054 dev_add_pack(&pThis->u.s.PacketType);
2055
2056#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2057 vboxNetFltLinuxHookDev(pThis, pDev);
2058#endif
2059#ifdef VBOXNETFLT_WITH_QDISC
2060 vboxNetFltLinuxQdiscInstall(pThis, pDev);
2061#endif /* VBOXNETFLT_WITH_QDISC */
2062
2063 /*
2064 * If attaching to TAP interface we need to bring the link state up
2065 * starting from 2.6.36 kernel.
2066 */
2067 vboxNetFltSetTapLinkState(pThis, pDev, true);
2068
2069 /*
2070 * Set indicators that require the spinlock. Be abit paranoid about racing
2071 * the device notification handle.
2072 */
2073 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2074 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2075 if (pDev)
2076 {
2077 ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
2078 ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
2079 pDev = NULL; /* don't dereference it */
2080 }
2081 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2082 Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
2083
2084 /*
2085 * If the above succeeded report GSO capabilities, if not undo and
2086 * release the device.
2087 */
2088 if (!pDev)
2089 {
2090 Assert(pThis->pSwitchPort);
2091 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
2092 {
2093 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2094 pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
2095 pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
2096 pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE | INTNETTRUNKDIR_HOST);
2097 vboxNetFltRelease(pThis, true /*fBusy*/);
2098 }
2099 }
2100 else
2101 {
2102#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2103 vboxNetFltLinuxUnhookDev(pThis, pDev);
2104#endif
2105#ifdef VBOXNETFLT_WITH_QDISC
2106 vboxNetFltLinuxQdiscRemove(pThis, pDev);
2107#endif /* VBOXNETFLT_WITH_QDISC */
2108 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2109 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2110 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2111 dev_put(pDev);
2112 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n",
2113 pDev, pDev->name,
2114#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
2115 netdev_refcnt_read(pDev)
2116#else
2117 atomic_read(&pDev->refcnt)
2118#endif
2119 ));
2120 }
2121
2122 LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
2123 return VINF_SUCCESS;
2124}
2125
2126
2127static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
2128{
2129 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
2130
2131 Assert(!pThis->fDisconnectedFromHost);
2132
2133#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2134 vboxNetFltLinuxUnhookDev(pThis, pDev);
2135#endif
2136#ifdef VBOXNETFLT_WITH_QDISC
2137 vboxNetFltLinuxQdiscRemove(pThis, pDev);
2138#endif /* VBOXNETFLT_WITH_QDISC */
2139
2140 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2141 ASMAtomicWriteBool(&pThis->u.s.fRegistered, false);
2142 ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
2143 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2144 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2145
2146 dev_remove_pack(&pThis->u.s.PacketType);
2147#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2148 skb_queue_purge(&pThis->u.s.XmitQueue);
2149#endif
2150 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
2151 Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
2152 pDev, pDev->name,
2153#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
2154 netdev_refcnt_read(pDev)
2155#else
2156 atomic_read(&pDev->refcnt)
2157#endif
2158 ));
2159 dev_put(pDev);
2160
2161 return NOTIFY_OK;
2162}
2163
2164static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
2165{
2166 /* Check if we are not suspended and promiscuous mode has not been set. */
2167 if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
2168 && !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
2169 {
2170 /* Note that there is no need for locking as the kernel got hold of the lock already. */
2171 dev_set_promiscuity(pDev, 1);
2172 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
2173 Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2174 }
2175 else
2176 Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2177 return NOTIFY_OK;
2178}
2179
2180static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
2181{
2182 /* Undo promiscuous mode if we has set it. */
2183 if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
2184 {
2185 /* Note that there is no need for locking as the kernel got hold of the lock already. */
2186 dev_set_promiscuity(pDev, -1);
2187 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
2188 Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2189 }
2190 else
2191 Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2192 return NOTIFY_OK;
2193}
2194
2195#ifdef LOG_ENABLED
2196/** Stringify the NETDEV_XXX constants. */
2197static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2198{
2199 const char *pszEvent = "NETDRV_<unknown>";
2200 switch (ulEventType)
2201 {
2202 case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2203 case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2204 case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2205 case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2206 case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2207 case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2208 case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2209 case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2210 case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2211 case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2212# ifdef NETDEV_FEAT_CHANGE
2213 case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2214# endif
2215 }
2216 return pszEvent;
2217}
2218#endif /* LOG_ENABLED */
2219
2220/**
2221 * Callback for listening to netdevice events.
2222 *
2223 * This works the rediscovery, clean up on unregistration, promiscuity on
2224 * up/down, and GSO feature changes from ethtool.
2225 *
2226 * @returns NOTIFY_OK
2227 * @param self Pointer to our notifier registration block.
2228 * @param ulEventType The event.
2229 * @param ptr Event specific, but it is usually the device it
2230 * relates to.
2231 */
2232static int vboxNetFltLinuxNotifierCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2233
2234{
2235 PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2236 struct net_device *pDev = (struct net_device *)ptr;
2237 int rc = NOTIFY_OK;
2238
2239 Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2240 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
2241 if ( ulEventType == NETDEV_REGISTER
2242 && !strcmp(pDev->name, pThis->szName))
2243 {
2244 vboxNetFltLinuxAttachToInterface(pThis, pDev);
2245 }
2246 else
2247 {
2248 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2249 if (pDev == ptr)
2250 {
2251 switch (ulEventType)
2252 {
2253 case NETDEV_UNREGISTER:
2254 rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2255 break;
2256 case NETDEV_UP:
2257 rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2258 break;
2259 case NETDEV_GOING_DOWN:
2260 rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2261 break;
2262 case NETDEV_CHANGENAME:
2263 break;
2264#ifdef NETDEV_FEAT_CHANGE
2265 case NETDEV_FEAT_CHANGE:
2266 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2267 break;
2268#endif
2269 }
2270 }
2271 }
2272
2273 return rc;
2274}
2275
2276bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2277{
2278 return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2279}
2280
2281int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2282{
2283 struct net_device * pDev;
2284 int err;
2285 int rc = VINF_SUCCESS;
2286 NOREF(pvIfData);
2287
2288 LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2289
2290 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2291 if (pDev)
2292 {
2293 /*
2294 * Create a sk_buff for the gather list and push it onto the wire.
2295 */
2296 if (fDst & INTNETTRUNKDIR_WIRE)
2297 {
2298 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2299 if (pBuf)
2300 {
2301 vboxNetFltDumpPacket(pSG, true, "wire", 1);
2302 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2303 Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2304 err = dev_queue_xmit(pBuf);
2305 if (err)
2306 rc = RTErrConvertFromErrno(err);
2307 }
2308 else
2309 rc = VERR_NO_MEMORY;
2310 }
2311
2312 /*
2313 * Create a sk_buff for the gather list and push it onto the host stack.
2314 */
2315 if (fDst & INTNETTRUNKDIR_HOST)
2316 {
2317 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2318 if (pBuf)
2319 {
2320 vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2321 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2322 Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2323 err = netif_rx_ni(pBuf);
2324 if (err)
2325 rc = RTErrConvertFromErrno(err);
2326 }
2327 else
2328 rc = VERR_NO_MEMORY;
2329 }
2330
2331 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2332 }
2333
2334 return rc;
2335}
2336
2337
2338void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2339{
2340 struct net_device * pDev;
2341
2342 LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
2343 pThis, pThis->szName, fActive?"true":"false",
2344 pThis->fDisablePromiscuous?"true":"false"));
2345
2346 if (pThis->fDisablePromiscuous)
2347 return;
2348
2349 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2350 if (pDev)
2351 {
2352 /*
2353 * This api is a bit weird, the best reference is the code.
2354 *
2355 * Also, we have a bit or race conditions wrt the maintenance of
2356 * host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2357 */
2358#ifdef LOG_ENABLED
2359 u_int16_t fIf;
2360 unsigned const cPromiscBefore = pDev->promiscuity;
2361#endif
2362 if (fActive)
2363 {
2364 Assert(!pThis->u.s.fPromiscuousSet);
2365
2366 rtnl_lock();
2367 dev_set_promiscuity(pDev, 1);
2368 rtnl_unlock();
2369 pThis->u.s.fPromiscuousSet = true;
2370 Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2371 }
2372 else
2373 {
2374 if (pThis->u.s.fPromiscuousSet)
2375 {
2376 rtnl_lock();
2377 dev_set_promiscuity(pDev, -1);
2378 rtnl_unlock();
2379 Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2380 }
2381 pThis->u.s.fPromiscuousSet = false;
2382
2383#ifdef LOG_ENABLED
2384 fIf = dev_get_flags(pDev);
2385 Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2386#endif
2387 }
2388
2389 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2390 }
2391}
2392
2393
2394int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2395{
2396#ifdef VBOXNETFLT_WITH_QDISC
2397 vboxNetFltLinuxQdiscRemove(pThis, NULL);
2398#endif /* VBOXNETFLT_WITH_QDISC */
2399 /*
2400 * Remove packet handler when we get disconnected from internal switch as
2401 * we don't want the handler to forward packets to disconnected switch.
2402 */
2403 dev_remove_pack(&pThis->u.s.PacketType);
2404 return VINF_SUCCESS;
2405}
2406
2407
2408int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2409{
2410 /*
2411 * Report the GSO capabilities of the host and device (if connected).
2412 * Note! No need to mark ourselves busy here.
2413 */
2414 /** @todo duplicate work here now? Attach */
2415#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2416 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2417 0
2418 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2419 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2420# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2421 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2422 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2423# endif
2424 , INTNETTRUNKDIR_HOST);
2425
2426#endif
2427 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2428
2429 return VINF_SUCCESS;
2430}
2431
2432
2433void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2434{
2435 struct net_device *pDev;
2436 bool fRegistered;
2437 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
2438
2439#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2440 vboxNetFltLinuxUnhookDev(pThis, NULL);
2441#endif
2442
2443 /** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2444 * unlikely, but none the less). Since it doesn't actually update the
2445 * state (just reads it), it is likely to panic in some interesting
2446 * ways. */
2447
2448 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2449 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2450 fRegistered = ASMAtomicUoReadBool(&pThis->u.s.fRegistered);
2451 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2452
2453 if (fRegistered)
2454 {
2455 vboxNetFltSetTapLinkState(pThis, pDev, false);
2456
2457#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2458 skb_queue_purge(&pThis->u.s.XmitQueue);
2459#endif
2460 Log(("vboxNetFltOsDeleteInstance: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
2461 Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
2462 pDev, pDev->name,
2463#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
2464 netdev_refcnt_read(pDev)
2465#else
2466 atomic_read(&pDev->refcnt)
2467#endif
2468 ));
2469 dev_put(pDev);
2470 }
2471 Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2472 unregister_netdevice_notifier(&pThis->u.s.Notifier);
2473 module_put(THIS_MODULE);
2474}
2475
2476
2477int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2478{
2479 int err;
2480 NOREF(pvContext);
2481
2482 pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2483 err = register_netdevice_notifier(&pThis->u.s.Notifier);
2484 if (err)
2485 return VERR_INTNET_FLT_IF_FAILED;
2486 if (!pThis->u.s.fRegistered)
2487 {
2488 unregister_netdevice_notifier(&pThis->u.s.Notifier);
2489 LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2490 return VERR_INTNET_FLT_IF_NOT_FOUND;
2491 }
2492
2493 Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2494 if ( pThis->fDisconnectedFromHost
2495 || !try_module_get(THIS_MODULE))
2496 return VERR_INTNET_FLT_IF_FAILED;
2497
2498 return VINF_SUCCESS;
2499}
2500
2501int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2502{
2503 /*
2504 * Init the linux specific members.
2505 */
2506 pThis->u.s.pDev = NULL;
2507 pThis->u.s.fRegistered = false;
2508 pThis->u.s.fPromiscuousSet = false;
2509 memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2510#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2511 skb_queue_head_init(&pThis->u.s.XmitQueue);
2512# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2513 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2514# else
2515 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2516# endif
2517#endif
2518
2519 return VINF_SUCCESS;
2520}
2521
2522
2523void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2524{
2525 NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2526}
2527
2528
2529int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void *pvIf, void **pvIfData)
2530{
2531 /* Nothing to do */
2532 NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2533 return VINF_SUCCESS;
2534}
2535
2536
2537int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2538{
2539 /* Nothing to do */
2540 NOREF(pThis); NOREF(pvIfData);
2541 return VINF_SUCCESS;
2542}
2543

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy