VirtualBox

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 30143

Last change on this file since 30143 was 30143, checked in by vboxsync, 15 years ago

#5020: Prevent flooding host's kernel log with warnings.
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 80.7 KB
Line 
1/* $Id: VBoxNetFlt-linux.c 30143 2010-06-10 10:00:23Z vboxsync $ */
2/** @file
3 * VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4 */
5
6/*
7 * Copyright (C) 2006-2008 Oracle Corporation
8 *
9 * This file is part of VirtualBox Open Source Edition (OSE), as
10 * available from http://www.virtualbox.org. This file is free software;
11 * you can redistribute it and/or modify it under the terms of the GNU
12 * General Public License (GPL) as published by the Free Software
13 * Foundation, in version 2 as it comes in the "COPYING" file of the
14 * VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15 * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16 */
17
18/*******************************************************************************
19* Header Files *
20*******************************************************************************/
21#define LOG_GROUP LOG_GROUP_NET_FLT_DRV
22#define VBOXNETFLT_LINUX_NO_XMIT_QUEUE
23#include "the-linux-kernel.h"
24#include "version-generated.h"
25#include "product-generated.h"
26#include <linux/netdevice.h>
27#include <linux/etherdevice.h>
28#include <linux/rtnetlink.h>
29#include <linux/miscdevice.h>
30#include <linux/ip.h>
31
32#include <VBox/log.h>
33#include <VBox/err.h>
34#include <VBox/intnetinline.h>
35#include <VBox/pdmnetinline.h>
36#include <VBox/param.h>
37#include <iprt/alloca.h>
38#include <iprt/assert.h>
39#include <iprt/spinlock.h>
40#include <iprt/semaphore.h>
41#include <iprt/initterm.h>
42#include <iprt/process.h>
43#include <iprt/mem.h>
44#include <iprt/net.h>
45#include <iprt/log.h>
46#include <iprt/mp.h>
47#include <iprt/mem.h>
48#include <iprt/time.h>
49
50#define VBOXNETFLT_OS_SPECFIC 1
51#include "../VBoxNetFltInternal.h"
52
53/*
54 * Uncomment the following line to enable qdisc support.
55 */
56//#define VBOXNETFLT_WITH_QDISC
57#ifdef VBOXNETFLT_WITH_QDISC
58#include <net/pkt_sched.h>
59#endif /* VBOXNETFLT_WITH_QDISC */
60
61
62/*******************************************************************************
63* Defined Constants And Macros *
64*******************************************************************************/
65#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
66#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
67#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
68# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
69#endif
70
71#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
72# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
73# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
74#else
75# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
76# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
77#endif
78
79#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
80# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
81#else
82# define CHECKSUM_PARTIAL CHECKSUM_HW
83# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
84# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
85# else
86# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
87# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
88# else
89# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
90# endif
91# endif
92#endif
93
94#ifndef NET_IP_ALIGN
95# define NET_IP_ALIGN 2
96#endif
97
98#if 0
99/** Create scatter / gather segments for fragments. When not used, we will
100 * linearize the socket buffer before creating the internal networking SG. */
101# define VBOXNETFLT_SG_SUPPORT 1
102#endif
103
104#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
105/** Indicates that the linux kernel may send us GSO frames. */
106# define VBOXNETFLT_WITH_GSO 1
107
108/** This enables or disables the transmitting of GSO frame from the internal
109 * network and to the host. */
110# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
111
112# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
113/** This enables or disables the transmitting of GSO frame from the internal
114 * network and to the wire. */
115# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
116# endif
117
118/** This enables or disables the forwarding/flooding of GSO frame from the host
119 * to the internal network. */
120# define VBOXNETFLT_WITH_GSO_RECV 1
121
122#endif
123
124
125/*******************************************************************************
126* Internal Functions *
127*******************************************************************************/
128static int VBoxNetFltLinuxInit(void);
129static void VBoxNetFltLinuxUnload(void);
130static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
131
132
133/*******************************************************************************
134* Global Variables *
135*******************************************************************************/
136/**
137 * The (common) global data.
138 */
139static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
140
141module_init(VBoxNetFltLinuxInit);
142module_exit(VBoxNetFltLinuxUnload);
143
144MODULE_AUTHOR(VBOX_VENDOR);
145MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
146MODULE_LICENSE("GPL");
147#ifdef MODULE_VERSION
148MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
149#endif
150
151
152#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
153unsigned dev_get_flags(const struct net_device *dev)
154{
155 unsigned flags;
156
157 flags = (dev->flags & ~(IFF_PROMISC |
158 IFF_ALLMULTI |
159 IFF_RUNNING)) |
160 (dev->gflags & (IFF_PROMISC |
161 IFF_ALLMULTI));
162
163 if (netif_running(dev) && netif_carrier_ok(dev))
164 flags |= IFF_RUNNING;
165
166 return flags;
167}
168#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
169
170
171#ifdef VBOXNETFLT_WITH_QDISC
172//#define QDISC_LOG(x) printk x
173#define QDISC_LOG(x) do { } while (0)
174
175#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
176#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops)
177#elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
178#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, ops, parent)
179#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
180#define QDISC_CREATE(dev, queue, ops, parent) qdisc_create_dflt(dev, queue, ops, parent)
181#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
182
183#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
184#define qdisc_dev(qdisc) (qdisc->dev)
185#define qdisc_pkt_len(skb) (skb->len)
186#define QDISC_GET(dev) (dev->qdisc_sleeping)
187#else
188#define QDISC_GET(dev) (netdev_get_tx_queue(dev, 0)->qdisc_sleeping)
189#endif
190
191#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
192#define QDISC_SAVED_NUM(dev) 1
193#elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
194#define QDISC_SAVED_NUM(dev) dev->num_tx_queues
195#else
196#define QDISC_SAVED_NUM(dev) dev->num_tx_queues+1
197#endif
198
199#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
200#define QDISC_IS_BUSY(dev, qdisc) test_bit(__LINK_STATE_SCHED, &dev->state)
201#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
202#define QDISC_IS_BUSY(dev, qdisc) (test_bit(__QDISC_STATE_RUNNING, &qdisc->state) || \
203 test_bit(__QDISC_STATE_SCHED, &qdisc->state))
204#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
205
206struct VBoxNetQDiscPriv
207{
208 /** Pointer to the single child qdisc. */
209 struct Qdisc *pChild;
210 /*
211 * Technically it is possible to have different qdiscs for different TX
212 * queues so we have to save them all.
213 */
214 /** Pointer to the array of saved qdiscs. */
215 struct Qdisc **ppSaved;
216 /** Pointer to the net filter instance. */
217 PVBOXNETFLTINS pVBoxNetFlt;
218};
219typedef struct VBoxNetQDiscPriv *PVBOXNETQDISCPRIV;
220
221//#define VBOXNETFLT_QDISC_ENQUEUE
222static int vboxNetFltQdiscEnqueue(struct sk_buff *skb, struct Qdisc *sch)
223{
224 int rc;
225 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
226#ifdef VBOXNETFLT_QDISC_ENQUEUE
227 if (VALID_PTR(pPriv->pVBoxNetFlt))
228 {
229 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
230 PCRTNETETHERHDR pEtherHdr;
231 PINTNETTRUNKSWPORT pSwitchPort;
232
233 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(skb, 0, sizeof(abHdrBuf), &abHdrBuf[0]);
234 if ( pEtherHdr
235 && (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) != NULL
236 && VALID_PTR(pSwitchPort)
237 )
238 {
239 struct sk_buff *pBuf;
240 INTNETSWDECISION enmDecision;
241 uint32_t cbHdrs = skb_headlen(skb);
242 cbHdrs = RT_MAX(cbHdrs, sizeof(abHdrBuf));
243
244#if 1
245 /* Test stub due to missing implementation of pfnPreRecv */
246 static uint8_t tmpAddr[] = { 0x08, 0x00, 0x27, 0xBB, 0x1E, 0xA0 };
247 if (memcmp(pEtherHdr, tmpAddr, 6) != 0)
248 enmDecision = INTNETSWDECISION_BROADCAST;
249 else
250 enmDecision = INTNETSWDECISION_INTNET;
251#else
252 enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
253#endif
254 if (enmDecision == INTNETSWDECISION_INTNET)
255 {
256 pBuf = skb_copy(skb, GFP_ATOMIC);
257 pBuf->pkt_type = PACKET_OUTGOING;
258 vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
259 qdisc_drop(skb, sch);
260 ++sch->bstats.packets;
261 sch->bstats.bytes += qdisc_pkt_len(skb);
262 return NET_XMIT_SUCCESS;
263 }
264 }
265 }
266#endif /* VBOXNETFLT_QDISC_ENQUEUE */
267 rc = pPriv->pChild->enqueue(skb, pPriv->pChild);
268 if (rc == NET_XMIT_SUCCESS)
269 {
270 ++sch->q.qlen;
271 ++sch->bstats.packets;
272 sch->bstats.bytes += qdisc_pkt_len(skb);
273 }
274 else
275 ++sch->qstats.drops;
276 return rc;
277}
278
279static struct sk_buff *vboxNetFltQdiscDequeue(struct Qdisc *sch)
280{
281 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
282#ifdef VBOXNETFLT_QDISC_ENQUEUE
283 --sch->q.qlen;
284 return pPriv->pChild->dequeue(pPriv->pChild);
285#else /* VBOXNETFLT_QDISC_ENQUEUE */
286 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
287 PCRTNETETHERHDR pEtherHdr;
288 PINTNETTRUNKSWPORT pSwitchPort;
289 struct sk_buff *pSkb;
290
291 QDISC_LOG(("vboxNetFltDequeue: Enter pThis=%p\n", pPriv->pVBoxNetFlt));
292
293 while ((pSkb = pPriv->pChild->dequeue(pPriv->pChild)) != NULL)
294 {
295 struct sk_buff *pBuf;
296 INTNETSWDECISION enmDecision;
297 uint32_t cbHdrs = skb_headlen(pSkb);
298
299 --sch->q.qlen;
300
301 if (!VALID_PTR(pPriv->pVBoxNetFlt))
302 break;
303
304 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, sizeof(abHdrBuf), &abHdrBuf[0]);
305 if ( !pEtherHdr
306 || (pSwitchPort = pPriv->pVBoxNetFlt->pSwitchPort) == NULL
307 || !VALID_PTR(pSwitchPort)
308 )
309 break;
310
311 cbHdrs = RT_MAX(cbHdrs, sizeof(abHdrBuf));
312
313#if 1
314 /* Test stub due to missing implementation of pfnPreRecv */
315 static uint8_t tmpAddr[] = { 0x08, 0x00, 0x27, 0xBB, 0x1E, 0xA0 };
316 if (memcmp(pEtherHdr, tmpAddr, 6) != 0)
317 enmDecision = INTNETSWDECISION_BROADCAST;
318 else
319 enmDecision = INTNETSWDECISION_INTNET;
320#else
321 enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
322#endif
323 if (enmDecision != INTNETSWDECISION_INTNET)
324 break;
325
326 pBuf = skb_copy(pSkb, GFP_ATOMIC);
327 pBuf->pkt_type = PACKET_OUTGOING;
328 QDISC_LOG(("vboxNetFltDequeue: pThis=%p\n", pPriv->pVBoxNetFlt));
329 vboxNetFltLinuxForwardToIntNet(pPriv->pVBoxNetFlt, pBuf);
330 qdisc_drop(pSkb, sch);
331 QDISC_LOG(("VBoxNetFlt: Packet for %02x:%02x:%02x:%02x:%02x:%02x dropped\n",
332 pSkb->data[0], pSkb->data[1], pSkb->data[2],
333 pSkb->data[3], pSkb->data[4], pSkb->data[5]));
334 }
335
336 return pSkb;
337#endif /* VBOXNETFLT_QDISC_ENQUEUE */
338}
339
340#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
341static int vboxNetFltQdiscRequeue(struct sk_buff *skb, struct Qdisc *sch)
342{
343 int rc;
344 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
345
346 rc = pPriv->pChild->ops->requeue(skb, pPriv->pChild);
347 if (rc == 0)
348 {
349 sch->q.qlen++;
350 sch->qstats.requeues++;
351 }
352
353 return rc;
354}
355#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
356
357static unsigned int vboxNetFltQdiscDrop(struct Qdisc *sch)
358{
359 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
360
361 ++sch->qstats.drops;
362 --sch->q.qlen;
363 if (pPriv->pChild->ops->drop)
364 return pPriv->pChild->ops->drop(pPriv->pChild);
365
366 return 0;
367}
368
369#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
370static int vboxNetFltQdiscInit(struct Qdisc *sch, struct rtattr *opt)
371#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
372static int vboxNetFltQdiscInit(struct Qdisc *sch, struct nlattr *opt)
373#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
374{
375 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
376 struct net_device *pDev = qdisc_dev(sch);
377
378 pPriv->pVBoxNetFlt = NULL;
379
380 pPriv->ppSaved = kcalloc(QDISC_SAVED_NUM(pDev), sizeof(pPriv->ppSaved[0]),
381 GFP_KERNEL);
382 if (!pPriv->ppSaved)
383 return -ENOMEM;
384
385 pPriv->pChild = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
386 &pfifo_qdisc_ops,
387 TC_H_MAKE(TC_H_MAJ(sch->handle),
388 TC_H_MIN(1)));
389 if (!pPriv->pChild)
390 {
391 kfree(pPriv->ppSaved);
392 pPriv->ppSaved = NULL;
393 return -ENOMEM;
394 }
395
396 return 0;
397}
398
399static void vboxNetFltQdiscReset(struct Qdisc *sch)
400{
401 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
402
403 qdisc_reset(pPriv->pChild);
404 sch->q.qlen = 0;
405 sch->qstats.backlog = 0;
406}
407
408static void vboxNetFltQdiscDestroy(struct Qdisc* sch)
409{
410 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
411 struct net_device *pDev = qdisc_dev(sch);
412
413 qdisc_destroy(pPriv->pChild);
414 pPriv->pChild = NULL;
415
416 if (pPriv->ppSaved)
417 {
418 int i;
419 for (i = 0; i < QDISC_SAVED_NUM(pDev); i++)
420 if (pPriv->ppSaved[i])
421 qdisc_destroy(pPriv->ppSaved[i]);
422 kfree(pPriv->ppSaved);
423 pPriv->ppSaved = NULL;
424 }
425}
426
427static int vboxNetFltClassGraft(struct Qdisc *sch, unsigned long arg, struct Qdisc *pNew,
428 struct Qdisc **ppOld)
429{
430 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
431
432 if (pNew == NULL)
433 pNew = &noop_qdisc;
434
435 sch_tree_lock(sch);
436 *ppOld = pPriv->pChild;
437 pPriv->pChild = pNew;
438#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
439 sch->q.qlen = 0;
440#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
441 qdisc_tree_decrease_qlen(*ppOld, (*ppOld)->q.qlen);
442#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) */
443 qdisc_reset(*ppOld);
444 sch_tree_unlock(sch);
445
446 return 0;
447}
448
449static struct Qdisc *vboxNetFltClassLeaf(struct Qdisc *sch, unsigned long arg)
450{
451 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
452 return pPriv->pChild;
453}
454
455static unsigned long vboxNetFltClassGet(struct Qdisc *sch, u32 classid)
456{
457 return 1;
458}
459
460static void vboxNetFltClassPut(struct Qdisc *sch, unsigned long arg)
461{
462}
463
464#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 25)
465static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
466 struct rtattr **tca, unsigned long *arg)
467#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
468static int vboxNetFltClassChange(struct Qdisc *sch, u32 classid, u32 parentid,
469 struct nlattr **tca, unsigned long *arg)
470#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25) */
471{
472 return -ENOSYS;
473}
474
475static int vboxNetFltClassDelete(struct Qdisc *sch, unsigned long arg)
476{
477 return -ENOSYS;
478}
479
480static void vboxNetFltClassWalk(struct Qdisc *sch, struct qdisc_walker *walker)
481{
482 if (!walker->stop) {
483 if (walker->count >= walker->skip)
484 if (walker->fn(sch, 1, walker) < 0) {
485 walker->stop = 1;
486 return;
487 }
488 walker->count++;
489 }
490}
491
492static struct tcf_proto **vboxNetFltClassFindTcf(struct Qdisc *sch, unsigned long cl)
493{
494 return NULL;
495}
496
497static int vboxNetFltClassDump(struct Qdisc *sch, unsigned long cl,
498 struct sk_buff *skb, struct tcmsg *tcm)
499{
500 PVBOXNETQDISCPRIV pPriv = qdisc_priv(sch);
501
502 if (cl != 1)
503 return -ENOENT;
504
505 tcm->tcm_handle |= TC_H_MIN(1);
506 tcm->tcm_info = pPriv->pChild->handle;
507
508 return 0;
509}
510
511
512static struct Qdisc_class_ops g_VBoxNetFltClassOps =
513{
514 .graft = vboxNetFltClassGraft,
515 .leaf = vboxNetFltClassLeaf,
516 .get = vboxNetFltClassGet,
517 .put = vboxNetFltClassPut,
518 .change = vboxNetFltClassChange,
519 .delete = vboxNetFltClassDelete,
520 .walk = vboxNetFltClassWalk,
521 .tcf_chain = vboxNetFltClassFindTcf,
522 .dump = vboxNetFltClassDump,
523};
524
525
526static struct Qdisc_ops g_VBoxNetFltQDiscOps = {
527 .cl_ops = &g_VBoxNetFltClassOps,
528 .id = "vboxnetflt",
529 .priv_size = sizeof(struct VBoxNetQDiscPriv),
530 .enqueue = vboxNetFltQdiscEnqueue,
531 .dequeue = vboxNetFltQdiscDequeue,
532#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
533 .requeue = vboxNetFltQdiscRequeue,
534#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
535 .peek = qdisc_peek_dequeued,
536#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
537 .drop = vboxNetFltQdiscDrop,
538 .init = vboxNetFltQdiscInit,
539 .reset = vboxNetFltQdiscReset,
540 .destroy = vboxNetFltQdiscDestroy,
541 .owner = THIS_MODULE
542};
543
544/*
545 * If our qdisc is already attached to the device (that means the user
546 * installed it from command line with 'tc' command) we simply update
547 * the pointer to vboxnetflt instance in qdisc's private structure.
548 * Otherwise we need to take some additional steps:
549 * - Create our qdisc;
550 * - Save all references to qdiscs;
551 * - Replace our child with the first qdisc reference;
552 * - Replace all references so they point to our qdisc.
553 */
554static void vboxNetFltLinuxQdiscInstall(PVBOXNETFLTINS pThis, struct net_device *pDev)
555{
556#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
557 int i;
558#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
559 PVBOXNETQDISCPRIV pPriv;
560
561 struct Qdisc *pExisting = QDISC_GET(pDev);
562 if (strcmp(pExisting->ops->id, "vboxnetflt"))
563 {
564 /* The existing qdisc is different from ours, let's create new one. */
565 struct Qdisc *pNew = QDISC_CREATE(pDev, netdev_get_tx_queue(pDev, 0),
566 &g_VBoxNetFltQDiscOps, TC_H_ROOT);
567 if (!pNew)
568 return; // TODO: Error?
569
570 if (!try_module_get(THIS_MODULE))
571 {
572 /*
573 * This may cause a memory leak but calling qdisc_destroy()
574 * is not an option as it will call module_put().
575 */
576 return;
577 }
578 pPriv = qdisc_priv(pNew);
579
580 qdisc_destroy(pPriv->pChild);
581 pPriv->pChild = QDISC_GET(pDev);
582 atomic_inc(&pPriv->pChild->refcnt);
583 /*
584 * There is no need in deactivating the device or acquiring any locks
585 * prior changing qdiscs since we do not destroy the old qdisc.
586 * Atomic replacement of pointers is enough.
587 */
588 /*
589 * No need to change reference counters here as we merely move
590 * the pointer and the reference counter of the newly allocated
591 * qdisc is already 1.
592 */
593#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
594 pPriv->ppSaved[0] = pDev->qdisc_sleeping;
595 ASMAtomicWritePtr(&pDev->qdisc_sleeping, pNew);
596 ASMAtomicWritePtr(&pDev->qdisc, pNew);
597#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
598 for (i = 0; i < pDev->num_tx_queues; i++)
599 {
600 struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
601
602 pPriv->ppSaved[i] = pQueue->qdisc_sleeping;
603 ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pNew);
604 ASMAtomicWritePtr(&pQueue->qdisc, pNew);
605 if (i)
606 atomic_inc(&pNew->refcnt);
607 }
608 /* Newer kernels store root qdisc in netdev structure as well. */
609# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
610 pPriv->ppSaved[pDev->num_tx_queues] = pDev->qdisc;
611 ASMAtomicWritePtr(&pDev->qdisc, pNew);
612 atomic_inc(&pNew->refcnt);
613# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
614#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
615 }
616 else
617 {
618 /* We already have vboxnetflt qdisc, let's use it. */
619 pPriv = qdisc_priv(pExisting);
620 }
621 ASMAtomicWritePtr(&pPriv->pVBoxNetFlt, pThis);
622 QDISC_LOG(("vboxNetFltLinuxInstallQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
623}
624
625static void vboxNetFltLinuxQdiscRemove(PVBOXNETFLTINS pThis, struct net_device *pDev)
626{
627#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
628 int i;
629#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
630 PVBOXNETQDISCPRIV pPriv;
631 struct Qdisc *pQdisc;
632 if (!pDev)
633 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
634 if (!VALID_PTR(pDev))
635 {
636 printk("VBoxNetFlt: Failed to detach qdisc, invalid device pointer: %p\n",
637 pDev);
638 return; // TODO: Consider returing an error
639 }
640
641
642 pQdisc = QDISC_GET(pDev);
643 if (strcmp(pQdisc->ops->id, "vboxnetflt"))
644 {
645 /* Looks like the user has replaced our qdisc manually. */
646 printk("VBoxNetFlt: Failed to detach qdisc, wrong qdisc: %s\n",
647 pQdisc->ops->id);
648 return; // TODO: Consider returing an error
649 }
650
651 pPriv = qdisc_priv(pQdisc);
652 Assert(pPriv->pVBoxNetFlt == pThis);
653 ASMAtomicWriteNullPtr(&pPriv->pVBoxNetFlt);
654
655 QDISC_LOG(("vboxNetFltLinuxQdiscRemove: refcnt=%d num_tx_queues=%d\n",
656 atomic_read(&pQdisc->refcnt), pDev->num_tx_queues));
657#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 27)
658 /* Play it safe, make sure the qdisc is not being used. */
659 if (pPriv->ppSaved[0])
660 {
661 ASMAtomicWritePtr(&pDev->qdisc_sleeping, pPriv->ppSaved[0]);
662 ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[0]);
663 pPriv->ppSaved[0] = NULL;
664 while (QDISC_IS_BUSY(pDev, pQdisc))
665 yield();
666 qdisc_destroy(pQdisc); /* Destroy reference */
667 }
668#else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
669 for (i = 0; i < pDev->num_tx_queues; i++)
670 {
671 struct netdev_queue *pQueue = netdev_get_tx_queue(pDev, i);
672 if (pPriv->ppSaved[i])
673 {
674 Assert(pQueue->qdisc_sleeping == pQdisc);
675 ASMAtomicWritePtr(&pQueue->qdisc_sleeping, pPriv->ppSaved[i]);
676 ASMAtomicWritePtr(&pQueue->qdisc, pPriv->ppSaved[i]);
677 pPriv->ppSaved[i] = NULL;
678 while (QDISC_IS_BUSY(pDev, pQdisc))
679 yield();
680 qdisc_destroy(pQdisc); /* Destroy reference */
681 }
682 }
683 /* Newer kernels store root qdisc in netdev structure as well. */
684#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
685 ASMAtomicWritePtr(&pDev->qdisc, pPriv->ppSaved[pDev->num_tx_queues]);
686 pPriv->ppSaved[pDev->num_tx_queues] = NULL;
687 while (QDISC_IS_BUSY(pDev, pQdisc))
688 yield();
689 qdisc_destroy(pQdisc); /* Destroy reference */
690#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) */
691#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
692
693 /*
694 * At this point all references to our qdisc should be gone
695 * unless the user had installed it manually.
696 */
697 QDISC_LOG(("vboxNetFltLinuxRemoveQdisc: pThis=%p\n", pPriv->pVBoxNetFlt));
698}
699
700#endif /* VBOXNETFLT_WITH_QDISC */
701
702
703/**
704 * Initialize module.
705 *
706 * @returns appropriate status code.
707 */
708static int __init VBoxNetFltLinuxInit(void)
709{
710 int rc;
711 /*
712 * Initialize IPRT.
713 */
714 rc = RTR0Init(0);
715 if (RT_SUCCESS(rc))
716 {
717 Log(("VBoxNetFltLinuxInit\n"));
718
719 /*
720 * Initialize the globals and connect to the support driver.
721 *
722 * This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
723 * for establishing the connect to the support driver.
724 */
725 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
726 rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
727 if (RT_SUCCESS(rc))
728 {
729#ifdef VBOXNETFLT_WITH_QDISC
730 /*memcpy(&g_VBoxNetFltQDiscOps, &pfifo_qdisc_ops, sizeof(g_VBoxNetFltQDiscOps));
731 strcpy(g_VBoxNetFltQDiscOps.id, "vboxnetflt");
732 g_VBoxNetFltQDiscOps.owner = THIS_MODULE;*/
733 rc = register_qdisc(&g_VBoxNetFltQDiscOps);
734 if (rc)
735 {
736 LogRel(("VBoxNetFlt: Failed to registed qdisc: %d\n", rc));
737 return rc;
738 }
739#endif /* VBOXNETFLT_WITH_QDISC */
740 LogRel(("VBoxNetFlt: Successfully started.\n"));
741 return 0;
742 }
743
744 LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
745 RTR0Term();
746 }
747 else
748 LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
749
750 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
751 return -RTErrConvertToErrno(rc);
752}
753
754
755/**
756 * Unload the module.
757 *
758 * @todo We have to prevent this if we're busy!
759 */
760static void __exit VBoxNetFltLinuxUnload(void)
761{
762 int rc;
763 Log(("VBoxNetFltLinuxUnload\n"));
764 Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
765
766#ifdef VBOXNETFLT_WITH_QDISC
767 unregister_qdisc(&g_VBoxNetFltQDiscOps);
768#endif /* VBOXNETFLT_WITH_QDISC */
769 /*
770 * Undo the work done during start (in reverse order).
771 */
772 rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
773 AssertRC(rc); NOREF(rc);
774
775 RTR0Term();
776
777 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
778
779 Log(("VBoxNetFltLinuxUnload - done\n"));
780}
781
782
783/**
784 * Experiment where we filter trafic from the host to the internal network
785 * before it reaches the NIC driver.
786 *
787 * The current code uses a very ugly hack and only works on kernels using the
788 * net_device_ops (>= 2.6.29). It has been shown to give us a
789 * performance boost of 60-100% though. So, we have to find some less hacky way
790 * of getting this job done eventually.
791 *
792 * #define VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
793 */
794#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
795
796/**
797 * The overridden net_device_ops of the device we're attached to.
798 *
799 * Requires Linux 2.6.29 or later.
800 *
801 * This is a very dirty hack that was create to explore how much we can improve
802 * the host to guest transfers by not CC'ing the NIC.
803 */
804typedef struct VBoxNetDeviceOpsOverride
805{
806 /** Our overridden ops. */
807 struct net_device_ops Ops;
808 /** Magic word. */
809 uint32_t u32Magic;
810 /** Pointer to the original ops. */
811 struct net_device_ops const *pOrgOps;
812 /** Pointer to the net filter instance. */
813 PVBOXNETFLTINS pVBoxNetFlt;
814 /** The number of filtered packages. */
815 uint64_t cFiltered;
816 /** The total number of packets */
817 uint64_t cTotal;
818} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
819/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
820#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
821
822/**
823 * ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
824 * because they belong on the internal network.
825 *
826 * @returns NETDEV_TX_XXX.
827 * @param pSkb The socket buffer to transmit.
828 * @param pDev The net device.
829 */
830static int vboxNetFltLinuxStartXmitFilter(struct sk_buff *pSkb, struct net_device *pDev)
831{
832 PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->netdev_ops;
833 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
834 PCRTNETETHERHDR pEtherHdr;
835 PINTNETTRUNKSWPORT pSwitchPort;
836
837
838 /*
839 * Validate the override structure.
840 *
841 * Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
842 * to be production quality code, we would have to be much more
843 * careful here and avoid the race.
844 */
845 if ( !VALID_PTR(pOverride)
846 || pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
847 || !VALID_PTR(pOverride->pOrgOps))
848 {
849 printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
850 dev_kfree_skb(pSkb);
851 return NETDEV_TX_OK;
852 }
853 pOverride->cTotal++;
854
855 /*
856 * Do the filtering base on the defaul OUI of our virtual NICs
857 *
858 * Note! In a real solution, we would ask the switch whether the
859 * destination MAC is 100% to be on the internal network and then
860 * drop it.
861 */
862 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, sizeof(abHdrBuf), &abHdrBuf[0]);
863 if ( pEtherHdr
864 && VALID_PTR(pOverride->pVBoxNetFlt)
865 && (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
866 && VALID_PTR(pSwitchPort)
867 )
868 {
869 INTNETSWDECISION enmDecision;
870 uint32_t cbHdrs = skb_headlen(pSkb);
871 cbHdrs = RT_MAX(cbHdrs, sizeof(abHdrBuf));
872
873 /** @todo consider reference counting, etc. */
874 enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
875 if (enmDecision == INTNETSWDECISION_INTNET)
876 {
877 dev_kfree_skb(pSkb);
878 pOverride->cFiltered++;
879 return NETDEV_TX_OK;
880 }
881 }
882
883 return pOverride->pOrgOps->ndo_start_xmit(pSkb, pDev);
884}
885
886/**
887 * Hooks the device ndo_start_xmit operation of the device.
888 *
889 * @param pThis The net filter instance.
890 * @param pDev The net device.
891 */
892static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
893{
894 PVBOXNETDEVICEOPSOVERRIDE pOverride;
895 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
896
897 pOverride = RTMemAlloc(sizeof(*pOverride));
898 if (!pOverride)
899 return;
900 pOverride->pOrgOps = pDev->netdev_ops;
901 pOverride->Ops = *pDev->netdev_ops;
902 pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
903 pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
904 pOverride->cTotal = 0;
905 pOverride->cFiltered = 0;
906 pOverride->pVBoxNetFlt = pThis;
907
908 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp); /* (this isn't necessary, but so what) */
909 ASMAtomicWritePtr((void * volatile *)&pDev->netdev_ops, pOverride);
910 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
911}
912
913/**
914 * Undos what vboxNetFltLinuxHookDev did.
915 *
916 * @param pThis The net filter instance.
917 * @param pDev The net device. Can be NULL, in which case
918 * we'll try retrieve it from @a pThis.
919 */
920static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
921{
922 PVBOXNETDEVICEOPSOVERRIDE pOverride;
923 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
924
925 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
926 if (!pDev)
927 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
928 if (VALID_PTR(pDev))
929 {
930 pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->netdev_ops;
931 if ( VALID_PTR(pOverride)
932 && pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
933 && VALID_PTR(pOverride->pOrgOps)
934 )
935 {
936 ASMAtomicWritePtr((void * volatile *)&pDev->netdev_ops, pOverride->pOrgOps);
937 ASMAtomicWriteU32(&pOverride->u32Magic, 0);
938 }
939 else
940 pOverride = NULL;
941 }
942 else
943 pOverride = NULL;
944 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
945
946 if (pOverride)
947 {
948 printk("vboxnetflt: dropped %llu out of %llu packets\n", pOverride->cFiltered, pOverride->cTotal);
949 RTMemFree(pOverride);
950 }
951}
952
953#endif /* VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT */
954
955
956/**
957 * Reads and retains the host interface handle.
958 *
959 * @returns The handle, NULL if detached.
960 * @param pThis
961 */
962DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
963{
964#if 0
965 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
966 struct net_device *pDev = NULL;
967
968 Log(("vboxNetFltLinuxRetainNetDev\n"));
969 /*
970 * Be careful here to avoid problems racing the detached callback.
971 */
972 RTSpinlockAcquire(pThis->hSpinlock, &Tmp);
973 if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
974 {
975 pDev = (struct net_device *)ASMAtomicUoReadPtr((void * volatile *)&pThis->u.s.pDev);
976 if (pDev)
977 {
978 dev_hold(pDev);
979 Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
980 }
981 }
982 RTSpinlockRelease(pThis->hSpinlock, &Tmp);
983
984 Log(("vboxNetFltLinuxRetainNetDev - done\n"));
985 return pDev;
986#else
987 return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
988#endif
989}
990
991
992/**
993 * Release the host interface handle previously retained
994 * by vboxNetFltLinuxRetainNetDev.
995 *
996 * @param pThis The instance.
997 * @param pDev The vboxNetFltLinuxRetainNetDev
998 * return value, NULL is fine.
999 */
1000DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
1001{
1002#if 0
1003 Log(("vboxNetFltLinuxReleaseNetDev\n"));
1004 NOREF(pThis);
1005 if (pDev)
1006 {
1007 dev_put(pDev);
1008 Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1009 }
1010 Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
1011#endif
1012}
1013
1014#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 | (skb->dev->ifindex & 0xFFFF))
1015#define VBOXNETFLT_SKB_TAG(skb) (*(uint32_t*)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
1016
1017/**
1018 * Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
1019 * i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
1020 *
1021 * @returns true / false accordingly.
1022 * @param pBuf The sk_buff.
1023 */
1024DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
1025{
1026 return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
1027}
1028
1029
1030/**
1031 * Internal worker that create a linux sk_buff for a
1032 * (scatter/)gather list.
1033 *
1034 * @returns Pointer to the sk_buff.
1035 * @param pThis The instance.
1036 * @param pSG The (scatter/)gather list.
1037 * @param fDstWire Set if the destination is the wire.
1038 */
1039static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
1040{
1041 struct sk_buff *pPkt;
1042 struct net_device *pDev;
1043 unsigned fGsoType = 0;
1044
1045 if (pSG->cbTotal == 0)
1046 {
1047 LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
1048 return NULL;
1049 }
1050
1051 /** @todo We should use fragments mapping the SG buffers with large packets.
1052 * 256 bytes seems to be the a threshold used a lot for this. It
1053 * requires some nasty work on the intnet side though... */
1054 /*
1055 * Allocate a packet and copy over the data.
1056 */
1057 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1058 pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
1059 if (RT_UNLIKELY(!pPkt))
1060 {
1061 Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
1062 pSG->pvUserData = NULL;
1063 return NULL;
1064 }
1065 pPkt->dev = pDev;
1066 pPkt->ip_summed = CHECKSUM_NONE;
1067
1068 /* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
1069 skb_reserve(pPkt, NET_IP_ALIGN);
1070
1071 /* Copy the segments. */
1072 skb_put(pPkt, pSG->cbTotal);
1073 IntNetSgRead(pSG, pPkt->data);
1074
1075#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1076 /*
1077 * Setup GSO if used by this packet.
1078 */
1079 switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
1080 {
1081 default:
1082 AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
1083 /* fall thru */
1084 case PDMNETWORKGSOTYPE_INVALID:
1085 fGsoType = 0;
1086 break;
1087 case PDMNETWORKGSOTYPE_IPV4_TCP:
1088 fGsoType = SKB_GSO_TCPV4;
1089 break;
1090 case PDMNETWORKGSOTYPE_IPV4_UDP:
1091 fGsoType = SKB_GSO_UDP;
1092 break;
1093 case PDMNETWORKGSOTYPE_IPV6_TCP:
1094 fGsoType = SKB_GSO_TCPV6;
1095 break;
1096 }
1097 if (fGsoType)
1098 {
1099 struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
1100
1101 pShInfo->gso_type = fGsoType | SKB_GSO_DODGY;
1102 pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
1103 pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
1104
1105 /*
1106 * We need to set checksum fields even if the packet goes to the host
1107 * directly as it may be immediately forwared by IP layer.
1108 */
1109 {
1110 Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrs);
1111 pPkt->ip_summed = CHECKSUM_PARTIAL;
1112# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1113 pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
1114 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
1115 pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
1116 else
1117 pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
1118# else
1119 pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
1120 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
1121 pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
1122 else
1123 pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
1124# endif
1125 }
1126 if (!fDstWire)
1127 PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, false /*fPayloadChecksum*/);
1128 }
1129#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE || VBOXNETFLT_WITH_GSO_XMIT_HOST */
1130
1131 /*
1132 * Finish up the socket buffer.
1133 */
1134 pPkt->protocol = eth_type_trans(pPkt, pDev);
1135 if (fDstWire)
1136 {
1137 VBOX_SKB_RESET_NETWORK_HDR(pPkt);
1138
1139 /* Restore ethernet header back. */
1140 skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
1141 VBOX_SKB_RESET_MAC_HDR(pPkt);
1142 }
1143 VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
1144
1145 return pPkt;
1146}
1147
1148
1149/**
1150 * Initializes a SG list from an sk_buff.
1151 *
1152 * @returns Number of segments.
1153 * @param pThis The instance.
1154 * @param pBuf The sk_buff.
1155 * @param pSG The SG.
1156 * @param pvFrame The frame pointer, optional.
1157 * @param cSegs The number of segments allocated for the SG.
1158 * This should match the number in the mbuf exactly!
1159 * @param fSrc The source of the frame.
1160 * @param pGso Pointer to the GSO context if it's a GSO
1161 * internal network frame. NULL if regular frame.
1162 */
1163DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
1164 unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1165{
1166 int i;
1167 NOREF(pThis);
1168
1169 Assert(!skb_shinfo(pBuf)->frag_list);
1170
1171 if (fSrc & INTNETTRUNKDIR_WIRE)
1172 {
1173 /*
1174 * The packet came from wire, ethernet header was removed by device driver.
1175 * Restore it.
1176 */
1177 skb_push(pBuf, ETH_HLEN);
1178 }
1179
1180 if (!pGsoCtx)
1181 IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/);
1182 else
1183 IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/, pGsoCtx);
1184
1185#ifdef VBOXNETFLT_SG_SUPPORT
1186 pSG->aSegs[0].cb = skb_headlen(pBuf);
1187 pSG->aSegs[0].pv = pBuf->data;
1188 pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1189
1190 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1191 {
1192 skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
1193 pSG->aSegs[i+1].cb = pFrag->size;
1194 pSG->aSegs[i+1].pv = kmap(pFrag->page);
1195 printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
1196 pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
1197 }
1198 ++i;
1199
1200#else
1201 pSG->aSegs[0].cb = pBuf->len;
1202 pSG->aSegs[0].pv = pBuf->data;
1203 pSG->aSegs[0].Phys = NIL_RTHCPHYS;
1204 i = 1;
1205#endif
1206
1207 pSG->cSegsUsed = i;
1208
1209#ifdef PADD_RUNT_FRAMES_FROM_HOST
1210 /*
1211 * Add a trailer if the frame is too small.
1212 *
1213 * Since we're getting to the packet before it is framed, it has not
1214 * yet been padded. The current solution is to add a segment pointing
1215 * to a buffer containing all zeros and pray that works for all frames...
1216 */
1217 if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
1218 {
1219 static uint8_t const s_abZero[128] = {0};
1220
1221 AssertReturnVoid(i < cSegs);
1222
1223 pSG->aSegs[i].Phys = NIL_RTHCPHYS;
1224 pSG->aSegs[i].pv = (void *)&s_abZero[0];
1225 pSG->aSegs[i].cb = 60 - pSG->cbTotal;
1226 pSG->cbTotal = 60;
1227 pSG->cSegsUsed++;
1228 Assert(i + 1 <= pSG->cSegsAlloc)
1229 }
1230#endif
1231
1232 Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
1233 pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
1234 for (i = 0; i < pSG->cSegsUsed; i++)
1235 Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
1236 i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1237}
1238
1239/**
1240 * Packet handler,
1241 *
1242 * @returns 0 or EJUSTRETURN.
1243 * @param pThis The instance.
1244 * @param pMBuf The mbuf.
1245 * @param pvFrame The start of the frame, optional.
1246 * @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
1247 * @param eProtocol The protocol.
1248 */
1249#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
1250static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1251 struct net_device *pSkbDev,
1252 struct packet_type *pPacketType,
1253 struct net_device *pOrigDev)
1254#else
1255static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1256 struct net_device *pSkbDev,
1257 struct packet_type *pPacketType)
1258#endif
1259{
1260 PVBOXNETFLTINS pThis;
1261 struct net_device *pDev;
1262 LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1263 pBuf, pSkbDev, pPacketType));
1264#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1265 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",
1266 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));
1267#else
1268 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",
1269 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));
1270#endif
1271 /*
1272 * Drop it immediately?
1273 */
1274 if (!pBuf)
1275 return 0;
1276
1277 pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1278 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1279 if (pThis->u.s.pDev != pSkbDev)
1280 {
1281 Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1282 return 0;
1283 }
1284
1285 Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1286 if (vboxNetFltLinuxSkBufIsOur(pBuf))
1287 {
1288 Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1289 dev_kfree_skb(pBuf);
1290 return 0;
1291 }
1292
1293#ifndef VBOXNETFLT_SG_SUPPORT
1294 {
1295 /*
1296 * Get rid of fragmented packets, they cause too much trouble.
1297 */
1298 struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1299 kfree_skb(pBuf);
1300 if (!pCopy)
1301 {
1302 LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1303 return 0;
1304 }
1305 pBuf = pCopy;
1306# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1307 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",
1308 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));
1309# else
1310 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",
1311 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));
1312# endif
1313 }
1314#endif
1315
1316#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1317 /* Forward it to the internal network. */
1318 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1319#else
1320 /* Add the packet to transmit queue and schedule the bottom half. */
1321 skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1322 schedule_work(&pThis->u.s.XmitTask);
1323 Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1324 &pThis->u.s.XmitTask, pBuf));
1325#endif
1326
1327 /* It does not really matter what we return, it is ignored by the kernel. */
1328 return 0;
1329}
1330
1331/**
1332 * Calculate the number of INTNETSEG segments the socket buffer will need.
1333 *
1334 * @returns Segment count.
1335 * @param pBuf The socket buffer.
1336 */
1337DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
1338{
1339#ifdef VBOXNETFLT_SG_SUPPORT
1340 unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1341#else
1342 unsigned cSegs = 1;
1343#endif
1344#ifdef PADD_RUNT_FRAMES_FROM_HOST
1345 /* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1346 if (pBuf->len < 60)
1347 cSegs++;
1348#endif
1349 return cSegs;
1350}
1351
1352/**
1353 * Destroy the intnet scatter / gather buffer created by
1354 * vboxNetFltLinuxSkBufToSG.
1355 */
1356static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
1357{
1358#ifdef VBOXNETFLT_SG_SUPPORT
1359 int i;
1360
1361 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1362 {
1363 printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
1364 kunmap(pSG->aSegs[i+1].pv);
1365 }
1366#endif
1367 NOREF(pSG);
1368}
1369
1370#ifdef LOG_ENABLED
1371/**
1372 * Logging helper.
1373 */
1374static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1375{
1376 uint8_t *pInt, *pExt;
1377 static int iPacketNo = 1;
1378 iPacketNo += iIncrement;
1379 if (fEgress)
1380 {
1381 pExt = pSG->aSegs[0].pv;
1382 pInt = pExt + 6;
1383 }
1384 else
1385 {
1386 pInt = pSG->aSegs[0].pv;
1387 pExt = pInt + 6;
1388 }
1389 Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1390 " %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1391 pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1392 fEgress ? "-->" : "<--", pszWhere,
1393 pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1394 pSG->cbTotal, iPacketNo));
1395 Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1396}
1397#else
1398# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1399#endif
1400
1401#ifdef VBOXNETFLT_WITH_GSO_RECV
1402
1403/**
1404 * Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1405 * GSO socket buffer without having to segment it.
1406 *
1407 * @returns true on success, false if needs segmenting.
1408 * @param pThis The net filter instance.
1409 * @param pSkb The GSO socket buffer.
1410 * @param fSrc The source.
1411 * @param pGsoCtx Where to return the GSO context on success.
1412 */
1413static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1414 PPDMNETWORKGSO pGsoCtx)
1415{
1416 PDMNETWORKGSOTYPE enmGsoType;
1417 uint16_t uEtherType;
1418 unsigned int cbTransport;
1419 unsigned int offTransport;
1420 unsigned int cbTransportHdr;
1421 unsigned uProtocol;
1422 union
1423 {
1424 RTNETIPV4 IPv4;
1425 RTNETIPV6 IPv6;
1426 RTNETTCP Tcp;
1427 uint8_t ab[40];
1428 uint16_t au16[40/2];
1429 uint32_t au32[40/4];
1430 } Buf;
1431
1432 /*
1433 * Check the GSO properties of the socket buffer and make sure it fits.
1434 */
1435 /** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1436 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP | SKB_GSO_DODGY | SKB_GSO_TCPV6 | SKB_GSO_TCPV4) ))
1437 {
1438 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1439 return false;
1440 }
1441 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1442 || pSkb->len > VBOX_MAX_GSO_SIZE ))
1443 {
1444 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1445 return false;
1446 }
1447 if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1448 {
1449 Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1450 return false;
1451 }
1452
1453 /*
1454 * skb_gso_segment does the following. Do we need to do it as well?
1455 */
1456#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1457 skb_reset_mac_header(pSkb);
1458 pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1459#else
1460 pSkb->mac.raw = pSkb->data;
1461 pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1462#endif
1463
1464 /*
1465 * Switch on the ethertype.
1466 */
1467 uEtherType = pSkb->protocol;
1468 if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1469 && pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1470 {
1471 uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1472 if (puEtherType)
1473 uEtherType = *puEtherType;
1474 }
1475 switch (uEtherType)
1476 {
1477 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1478 {
1479 unsigned int cbHdr;
1480 PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1481 if (RT_UNLIKELY(!pIPv4))
1482 {
1483 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1484 return false;
1485 }
1486
1487 cbHdr = pIPv4->ip_hl * 4;
1488 cbTransport = RT_N2H_U16(pIPv4->ip_len);
1489 if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1490 || cbHdr > cbTransport ))
1491 {
1492 Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1493 return false;
1494 }
1495 cbTransport -= cbHdr;
1496 offTransport = pSkb->mac_len + cbHdr;
1497 uProtocol = pIPv4->ip_p;
1498 if (uProtocol == RTNETIPV4_PROT_TCP)
1499 enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1500 else if (uProtocol == RTNETIPV4_PROT_UDP)
1501 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1502 else /** @todo IPv6: 4to6 tunneling */
1503 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1504 break;
1505 }
1506
1507 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1508 {
1509 PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1510 if (RT_UNLIKELY(!pIPv6))
1511 {
1512 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1513 return false;
1514 }
1515
1516 cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1517 offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1518 uProtocol = pIPv6->ip6_nxt;
1519 /** @todo IPv6: Dig our way out of the other headers. */
1520 if (uProtocol == RTNETIPV4_PROT_TCP)
1521 enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1522 else if (uProtocol == RTNETIPV4_PROT_UDP)
1523 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1524 else
1525 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1526 break;
1527 }
1528
1529 default:
1530 Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1531 return false;
1532 }
1533
1534 if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1535 {
1536 Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1537 return false;
1538 }
1539
1540 if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1541 || offTransport + cbTransport > pSkb->len
1542 || cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1543 {
1544 Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1545 offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1546 return false;
1547 }
1548
1549 /*
1550 * Check the TCP/UDP bits.
1551 */
1552 if (uProtocol == RTNETIPV4_PROT_TCP)
1553 {
1554 PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1555 if (RT_UNLIKELY(!pTcp))
1556 {
1557 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1558 return false;
1559 }
1560
1561 cbTransportHdr = pTcp->th_off * 4;
1562 if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1563 || cbTransportHdr > cbTransport
1564 || offTransport + cbTransportHdr >= UINT8_MAX
1565 || offTransport + cbTransportHdr >= pSkb->len ))
1566 {
1567 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1568 return false;
1569 }
1570
1571 }
1572 else
1573 {
1574 Assert(uProtocol == RTNETIPV4_PROT_UDP);
1575 cbTransportHdr = sizeof(RTNETUDP);
1576 if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1577 || offTransport + cbTransportHdr >= pSkb->len ))
1578 {
1579 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1580 return false;
1581 }
1582 }
1583
1584 /*
1585 * We're good, init the GSO context.
1586 */
1587 pGsoCtx->u8Type = enmGsoType;
1588 pGsoCtx->cbHdrs = offTransport + cbTransportHdr;
1589 pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1590 pGsoCtx->offHdr1 = pSkb->mac_len;
1591 pGsoCtx->offHdr2 = offTransport;
1592 pGsoCtx->au8Unused[0] = 0;
1593 pGsoCtx->au8Unused[1] = 0;
1594
1595 return true;
1596}
1597
1598/**
1599 * Forward the socket buffer as a GSO internal network frame.
1600 *
1601 * @returns IPRT status code.
1602 * @param pThis The net filter instance.
1603 * @param pSkb The GSO socket buffer.
1604 * @param fSrc The source.
1605 * @param pGsoCtx Where to return the GSO context on success.
1606 */
1607static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1608{
1609 int rc;
1610 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1611 if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1612 {
1613 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1614 if (RT_LIKELY(pSG))
1615 {
1616 vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1617
1618 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1619 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1620
1621 vboxNetFltLinuxDestroySG(pSG);
1622 rc = VINF_SUCCESS;
1623 }
1624 else
1625 {
1626 Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1627 rc = VERR_NO_MEMORY;
1628 }
1629 }
1630 else
1631 {
1632 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1633 rc = VERR_INTERNAL_ERROR_3;
1634 }
1635
1636 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1637 dev_kfree_skb(pSkb);
1638 return rc;
1639}
1640
1641#endif /* VBOXNETFLT_WITH_GSO_RECV */
1642
1643/**
1644 * Worker for vboxNetFltLinuxForwardToIntNet.
1645 *
1646 * @returns VINF_SUCCESS or VERR_NO_MEMORY.
1647 * @param pThis The net filter instance.
1648 * @param pBuf The socket buffer.
1649 * @param fSrc The source.
1650 */
1651static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1652{
1653 int rc;
1654 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1655 if (cSegs <= MAX_SKB_FRAGS + 1)
1656 {
1657 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1658 if (RT_LIKELY(pSG))
1659 {
1660 vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /*pGsoCtx*/);
1661
1662 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1663 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1664
1665 vboxNetFltLinuxDestroySG(pSG);
1666 rc = VINF_SUCCESS;
1667 }
1668 else
1669 {
1670 Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1671 rc = VERR_NO_MEMORY;
1672 }
1673 }
1674 else
1675 {
1676 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1677 rc = VERR_INTERNAL_ERROR_3;
1678 }
1679
1680 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1681 dev_kfree_skb(pBuf);
1682 return rc;
1683}
1684
1685/**
1686 *
1687 * @param pBuf The socket buffer. This is consumed by this function.
1688 */
1689static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1690{
1691 uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1692
1693#ifdef VBOXNETFLT_WITH_GSO
1694 if (skb_is_gso(pBuf))
1695 {
1696 PDMNETWORKGSO GsoCtx;
1697 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",
1698 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));
1699# ifdef VBOXNETFLT_WITH_GSO_RECV
1700 if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP | SKB_GSO_TCPV6 | SKB_GSO_TCPV4))
1701 && vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1702 vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1703 else
1704# endif
1705 {
1706 /* Need to segment the packet */
1707 struct sk_buff *pNext;
1708 struct sk_buff *pSegment = skb_gso_segment(pBuf, 0 /*supported features*/);
1709 if (IS_ERR(pSegment))
1710 {
1711 dev_kfree_skb(pBuf);
1712 LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1713 return;
1714 }
1715
1716 for (; pSegment; pSegment = pNext)
1717 {
1718 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",
1719 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));
1720 pNext = pSegment->next;
1721 pSegment->next = 0;
1722 vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1723 }
1724 dev_kfree_skb(pBuf);
1725 }
1726 }
1727 else
1728#endif /* VBOXNETFLT_WITH_GSO */
1729 {
1730 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1731 {
1732#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1733 /*
1734 * Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1735 * and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1736 * header length from the header itself and reconstruct 'h' pointer
1737 * to TCP (or whatever) header.
1738 */
1739 unsigned char *tmp = pBuf->h.raw;
1740 if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1741 pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1742#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1743 if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1744 {
1745 LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1746 dev_kfree_skb(pBuf);
1747 return;
1748 }
1749#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1750 /* Restore the original (wrong) pointer. */
1751 pBuf->h.raw = tmp;
1752#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1753 }
1754 vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1755 }
1756}
1757
1758#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1759/**
1760 * Work queue handler that forwards the socket buffers queued by
1761 * vboxNetFltLinuxPacketHandler to the internal network.
1762 *
1763 * @param pWork The work queue.
1764 */
1765# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1766static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1767# else
1768static void vboxNetFltLinuxXmitTask(void *pWork)
1769# endif
1770{
1771 PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1772 struct sk_buff *pBuf;
1773
1774 Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1775
1776 /*
1777 * Active? Retain the instance and increment the busy counter.
1778 */
1779 if (vboxNetFltTryRetainBusyActive(pThis))
1780 {
1781 while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1782 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1783
1784 vboxNetFltRelease(pThis, true /* fBusy */);
1785 }
1786 else
1787 {
1788 /** @todo Shouldn't we just drop the packets here? There is little point in
1789 * making them accumulate when the VM is paused and it'll only waste
1790 * kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1791 * before start draining the packets (goes for the intnet ring buf
1792 * too)? */
1793 }
1794}
1795#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1796
1797/**
1798 * Reports the GSO capabilites of the hardware NIC.
1799 *
1800 * @param pThis The net filter instance. The caller hold a
1801 * reference to this.
1802 */
1803static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1804{
1805#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1806 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1807 {
1808 struct net_device *pDev;
1809 PINTNETTRUNKSWPORT pSwitchPort;
1810 unsigned int fFeatures;
1811 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1812
1813 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1814
1815 pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1816 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1817 if (pDev)
1818 fFeatures = pDev->features;
1819 else
1820 fFeatures = 0;
1821
1822 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1823
1824 if (pThis->pSwitchPort)
1825 {
1826 /* Set/update the GSO capabilities of the NIC. */
1827 uint32_t fGsoCapabilites = 0;
1828 if (fFeatures & NETIF_F_TSO)
1829 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1830 if (fFeatures & NETIF_F_TSO6)
1831 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1832# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1833 if (fFeatures & NETIF_F_UFO)
1834 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1835 if (fFeatures & NETIF_F_UFO)
1836 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1837# endif
1838 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1839 }
1840
1841 vboxNetFltRelease(pThis, true /*fBusy*/);
1842 }
1843#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1844}
1845
1846/**
1847 * Helper that determins whether the host (ignoreing us) is operating the
1848 * interface in promiscuous mode or not.
1849 */
1850static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1851{
1852 bool fRc = false;
1853 struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1854 if (pDev)
1855 {
1856 fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1857 LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1858 fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1859 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1860 }
1861 return fRc;
1862}
1863
1864/**
1865 * Internal worker for vboxNetFltLinuxNotifierCallback.
1866 *
1867 * @returns VBox status code.
1868 * @param pThis The instance.
1869 * @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1870 * flood the release log.
1871 */
1872static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1873{
1874 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1875 LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1876
1877 /*
1878 * Retain and store the device.
1879 */
1880 dev_hold(pDev);
1881
1882 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1883 ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1884 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1885
1886 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1887 Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n", pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1888
1889 /* Get the mac address while we still have a valid net_device reference. */
1890 memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1891
1892 /*
1893 * Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1894 */
1895 pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1896 pThis->u.s.PacketType.dev = pDev;
1897 pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1898 dev_add_pack(&pThis->u.s.PacketType);
1899
1900#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1901 vboxNetFltLinuxHookDev(pThis, pDev);
1902#endif
1903#ifdef VBOXNETFLT_WITH_QDISC
1904 vboxNetFltLinuxQdiscInstall(pThis, pDev);
1905#endif /* VBOXNETFLT_WITH_QDISC */
1906
1907 /*
1908 * Set indicators that require the spinlock. Be abit paranoid about racing
1909 * the device notification handle.
1910 */
1911 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1912 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1913 if (pDev)
1914 {
1915 ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1916 ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1917 pDev = NULL; /* don't dereference it */
1918 }
1919 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1920 Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1921
1922 /*
1923 * If the above succeeded report GSO capabilites, if not undo and
1924 * release the device.
1925 */
1926 if (!pDev)
1927 {
1928 Assert(pThis->pSwitchPort);
1929 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1930 {
1931 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1932 pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1933 pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1934 pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE | INTNETTRUNKDIR_HOST);
1935 vboxNetFltRelease(pThis, true /*fBusy*/);
1936 }
1937 }
1938 else
1939 {
1940#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1941 vboxNetFltLinuxUnhookDev(pThis, pDev);
1942#endif
1943#ifdef VBOXNETFLT_WITH_QDISC
1944 vboxNetFltLinuxQdiscRemove(pThis, pDev);
1945#endif /* VBOXNETFLT_WITH_QDISC */
1946 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1947 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1948 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1949 dev_put(pDev);
1950 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1951 }
1952
1953 LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1954 return VINF_SUCCESS;
1955}
1956
1957
1958static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1959{
1960 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1961
1962 Assert(!pThis->fDisconnectedFromHost);
1963
1964#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
1965 vboxNetFltLinuxUnhookDev(pThis, pDev);
1966#endif
1967#ifdef VBOXNETFLT_WITH_QDISC
1968 vboxNetFltLinuxQdiscRemove(pThis, pDev);
1969#endif /* VBOXNETFLT_WITH_QDISC */
1970
1971 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1972 ASMAtomicWriteBool(&pThis->u.s.fRegistered, false);
1973 ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1974 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1975 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1976
1977 dev_remove_pack(&pThis->u.s.PacketType);
1978#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1979 skb_queue_purge(&pThis->u.s.XmitQueue);
1980#endif
1981 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
1982 Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
1983 dev_put(pDev);
1984
1985 return NOTIFY_OK;
1986}
1987
1988static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1989{
1990 /* Check if we are not suspended and promiscuous mode has not been set. */
1991 if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1992 && !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1993 {
1994 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1995 dev_set_promiscuity(pDev, 1);
1996 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1997 Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1998 }
1999 else
2000 Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2001 return NOTIFY_OK;
2002}
2003
2004static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
2005{
2006 /* Undo promiscuous mode if we has set it. */
2007 if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
2008 {
2009 /* Note that there is no need for locking as the kernel got hold of the lock already. */
2010 dev_set_promiscuity(pDev, -1);
2011 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
2012 Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2013 }
2014 else
2015 Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2016 return NOTIFY_OK;
2017}
2018
2019#ifdef LOG_ENABLED
2020/** Stringify the NETDEV_XXX constants. */
2021static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2022{
2023 const char *pszEvent = "NETDRV_<unknown>";
2024 switch (ulEventType)
2025 {
2026 case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2027 case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2028 case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2029 case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2030 case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2031 case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2032 case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2033 case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2034 case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2035 case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2036# ifdef NETDEV_FEAT_CHANGE
2037 case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2038# endif
2039 }
2040 return pszEvent;
2041}
2042#endif /* LOG_ENABLED */
2043
2044/**
2045 * Callback for listening to netdevice events.
2046 *
2047 * This works the rediscovery, clean up on unregistration, promiscuity on
2048 * up/down, and GSO feature changes from ethtool.
2049 *
2050 * @returns NOTIFY_OK
2051 * @param self Pointer to our notifier registration block.
2052 * @param ulEventType The event.
2053 * @param ptr Event specific, but it is usually the device it
2054 * relates to.
2055 */
2056static int vboxNetFltLinuxNotifierCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2057
2058{
2059 PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2060 struct net_device *pDev = (struct net_device *)ptr;
2061 int rc = NOTIFY_OK;
2062
2063 Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2064 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
2065 if ( ulEventType == NETDEV_REGISTER
2066 && !strcmp(pDev->name, pThis->szName))
2067 {
2068 vboxNetFltLinuxAttachToInterface(pThis, pDev);
2069 }
2070 else
2071 {
2072 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2073 if (pDev == ptr)
2074 {
2075 switch (ulEventType)
2076 {
2077 case NETDEV_UNREGISTER:
2078 rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2079 break;
2080 case NETDEV_UP:
2081 rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2082 break;
2083 case NETDEV_GOING_DOWN:
2084 rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2085 break;
2086 case NETDEV_CHANGENAME:
2087 break;
2088#ifdef NETDEV_FEAT_CHANGE
2089 case NETDEV_FEAT_CHANGE:
2090 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2091 break;
2092#endif
2093 }
2094 }
2095 }
2096
2097 return rc;
2098}
2099
2100bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2101{
2102 return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2103}
2104
2105int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2106{
2107 struct net_device * pDev;
2108 int err;
2109 int rc = VINF_SUCCESS;
2110 NOREF(pvIfData);
2111
2112 LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2113
2114 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2115 if (pDev)
2116 {
2117 /*
2118 * Create a sk_buff for the gather list and push it onto the wire.
2119 */
2120 if (fDst & INTNETTRUNKDIR_WIRE)
2121 {
2122 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2123 if (pBuf)
2124 {
2125 vboxNetFltDumpPacket(pSG, true, "wire", 1);
2126 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2127 Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2128 err = dev_queue_xmit(pBuf);
2129 if (err)
2130 rc = RTErrConvertFromErrno(err);
2131 }
2132 else
2133 rc = VERR_NO_MEMORY;
2134 }
2135
2136 /*
2137 * Create a sk_buff for the gather list and push it onto the host stack.
2138 */
2139 if (fDst & INTNETTRUNKDIR_HOST)
2140 {
2141 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2142 if (pBuf)
2143 {
2144 vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2145 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2146 Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2147 err = netif_rx_ni(pBuf);
2148 if (err)
2149 rc = RTErrConvertFromErrno(err);
2150 }
2151 else
2152 rc = VERR_NO_MEMORY;
2153 }
2154
2155 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2156 }
2157
2158 return rc;
2159}
2160
2161
2162void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2163{
2164 struct net_device * pDev;
2165
2166 LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
2167 pThis, pThis->szName, fActive?"true":"false",
2168 pThis->fDisablePromiscuous?"true":"false"));
2169
2170 if (pThis->fDisablePromiscuous)
2171 return;
2172
2173 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2174 if (pDev)
2175 {
2176 /*
2177 * This api is a bit weird, the best reference is the code.
2178 *
2179 * Also, we have a bit or race conditions wrt the maintance of
2180 * host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2181 */
2182#ifdef LOG_ENABLED
2183 u_int16_t fIf;
2184 unsigned const cPromiscBefore = pDev->promiscuity;
2185#endif
2186 if (fActive)
2187 {
2188 Assert(!pThis->u.s.fPromiscuousSet);
2189
2190 rtnl_lock();
2191 dev_set_promiscuity(pDev, 1);
2192 rtnl_unlock();
2193 pThis->u.s.fPromiscuousSet = true;
2194 Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2195 }
2196 else
2197 {
2198 if (pThis->u.s.fPromiscuousSet)
2199 {
2200 rtnl_lock();
2201 dev_set_promiscuity(pDev, -1);
2202 rtnl_unlock();
2203 Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2204 }
2205 pThis->u.s.fPromiscuousSet = false;
2206
2207#ifdef LOG_ENABLED
2208 fIf = dev_get_flags(pDev);
2209 Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2210#endif
2211 }
2212
2213 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2214 }
2215}
2216
2217
2218int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2219{
2220 /* Nothing to do here. */
2221 return VINF_SUCCESS;
2222}
2223
2224
2225int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2226{
2227 /*
2228 * Report the GSO capabilities of the host and device (if connected).
2229 * Note! No need to mark ourselves busy here.
2230 */
2231 /** @todo duplicate work here now? Attach */
2232#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2233 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2234 0
2235 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2236 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2237# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2238 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2239 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2240# endif
2241 , INTNETTRUNKDIR_HOST);
2242
2243#endif
2244 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2245
2246 return VINF_SUCCESS;
2247}
2248
2249
2250void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2251{
2252 struct net_device *pDev;
2253 bool fRegistered;
2254 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
2255
2256#ifdef VBOXNETFLT_WITH_FILTER_HOST2GUEST_SKBS_EXPERIMENT
2257 vboxNetFltLinuxUnhookDev(pThis, NULL);
2258#endif
2259#ifdef VBOXNETFLT_WITH_QDISC
2260 vboxNetFltLinuxQdiscRemove(pThis, NULL);
2261#endif /* VBOXNETFLT_WITH_QDISC */
2262
2263 /** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2264 * unlikely, but none the less). Since it doesn't actually update the
2265 * state (just reads it), it is likely to panic in some interesting
2266 * ways. */
2267
2268 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
2269 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2270 fRegistered = ASMAtomicUoReadBool(&pThis->u.s.fRegistered);
2271 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
2272
2273 if (fRegistered)
2274 {
2275 dev_remove_pack(&pThis->u.s.PacketType);
2276#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2277 skb_queue_purge(&pThis->u.s.XmitQueue);
2278#endif
2279 Log(("vboxNetFltOsDeleteInstance: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
2280 Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n", pDev, pDev->name, atomic_read(&pDev->refcnt)));
2281 dev_put(pDev);
2282 }
2283 Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2284 unregister_netdevice_notifier(&pThis->u.s.Notifier);
2285 module_put(THIS_MODULE);
2286}
2287
2288
2289int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2290{
2291 int err;
2292 NOREF(pvContext);
2293
2294 pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2295 err = register_netdevice_notifier(&pThis->u.s.Notifier);
2296 if (err)
2297 return VERR_INTNET_FLT_IF_FAILED;
2298 if (!pThis->u.s.fRegistered)
2299 {
2300 unregister_netdevice_notifier(&pThis->u.s.Notifier);
2301 LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2302 return VERR_INTNET_FLT_IF_NOT_FOUND;
2303 }
2304
2305 Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2306 if ( pThis->fDisconnectedFromHost
2307 || !try_module_get(THIS_MODULE))
2308 return VERR_INTNET_FLT_IF_FAILED;
2309
2310 return VINF_SUCCESS;
2311}
2312
2313int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2314{
2315 /*
2316 * Init the linux specific members.
2317 */
2318 pThis->u.s.pDev = NULL;
2319 pThis->u.s.fRegistered = false;
2320 pThis->u.s.fPromiscuousSet = false;
2321 memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2322#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2323 skb_queue_head_init(&pThis->u.s.XmitQueue);
2324# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2325 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2326# else
2327 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2328# endif
2329#endif
2330
2331 return VINF_SUCCESS;
2332}
2333
2334
2335void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2336{
2337 NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2338}
2339
2340
2341int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void *pvIf, void **pvIfData)
2342{
2343 /* Nothing to do */
2344 NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2345 return VINF_SUCCESS;
2346}
2347
2348
2349int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2350{
2351 /* Nothing to do */
2352 NOREF(pThis); NOREF(pvIfData);
2353 return VINF_SUCCESS;
2354}
2355
Note: See TracBrowser for help on using the repository browser.

© 2025 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette