VirtualBox

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 39080

Last change on this file since 39080 was 39019, checked in by vboxsync, 14 years ago

netflt: renamed host-to-wire filter conditionals, updated comments
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 68.7 KB
Line 
1/* $Id: VBoxNetFlt-linux.c 39019 2011-10-18 14:11:56Z 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/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
54/*******************************************************************************
55* Defined Constants And Macros *
56*******************************************************************************/
57#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
58#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
59#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
60# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
61#endif
62
63#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
64# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
65# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
66#else
67# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
68# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
69#endif
70
71#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
72# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
73#else
74# define CHECKSUM_PARTIAL CHECKSUM_HW
75# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
76# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
77# else
78# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
79# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
80# else
81# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
82# endif
83/* Versions prior 2.6.10 use stats for both bstats and qstats */
84# define bstats stats
85# define qstats stats
86# endif
87#endif
88
89#ifndef NET_IP_ALIGN
90# define NET_IP_ALIGN 2
91#endif
92
93#if 0
94/** Create scatter / gather segments for fragments. When not used, we will
95 * linearize the socket buffer before creating the internal networking SG. */
96# define VBOXNETFLT_SG_SUPPORT 1
97#endif
98
99#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
100/** Indicates that the linux kernel may send us GSO frames. */
101# define VBOXNETFLT_WITH_GSO 1
102
103/** This enables or disables the transmitting of GSO frame from the internal
104 * network and to the host. */
105# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
106
107# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
108/** This enables or disables the transmitting of GSO frame from the internal
109 * network and to the wire. */
110# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
111# endif
112
113/** This enables or disables the forwarding/flooding of GSO frame from the host
114 * to the internal network. */
115# define VBOXNETFLT_WITH_GSO_RECV 1
116
117#endif
118
119#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
120/** This enables or disables handling of GSO frames coming from the wire (GRO). */
121# define VBOXNETFLT_WITH_GRO 1
122#endif
123/*
124 * GRO support was backported to RHEL 5.4
125 */
126#ifdef RHEL_RELEASE_CODE
127# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
128# define VBOXNETFLT_WITH_GRO 1
129# endif
130#endif
131
132/*******************************************************************************
133* Internal Functions *
134*******************************************************************************/
135static int VBoxNetFltLinuxInit(void);
136static void VBoxNetFltLinuxUnload(void);
137static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
138
139
140/*******************************************************************************
141* Global Variables *
142*******************************************************************************/
143/**
144 * The (common) global data.
145 */
146static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
147
148module_init(VBoxNetFltLinuxInit);
149module_exit(VBoxNetFltLinuxUnload);
150
151MODULE_AUTHOR(VBOX_VENDOR);
152MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
153MODULE_LICENSE("GPL");
154#ifdef MODULE_VERSION
155MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
156#endif
157
158
159#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
160unsigned dev_get_flags(const struct net_device *dev)
161{
162 unsigned flags;
163
164 flags = (dev->flags & ~(IFF_PROMISC |
165 IFF_ALLMULTI |
166 IFF_RUNNING)) |
167 (dev->gflags & (IFF_PROMISC |
168 IFF_ALLMULTI));
169
170 if (netif_running(dev) && netif_carrier_ok(dev))
171 flags |= IFF_RUNNING;
172
173 return flags;
174}
175#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
176
177
178/**
179 * Initialize module.
180 *
181 * @returns appropriate status code.
182 */
183static int __init VBoxNetFltLinuxInit(void)
184{
185 int rc;
186 /*
187 * Initialize IPRT.
188 */
189 rc = RTR0Init(0);
190 if (RT_SUCCESS(rc))
191 {
192 Log(("VBoxNetFltLinuxInit\n"));
193
194 /*
195 * Initialize the globals and connect to the support driver.
196 *
197 * This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
198 * for establishing the connect to the support driver.
199 */
200 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
201 rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
202 if (RT_SUCCESS(rc))
203 {
204 LogRel(("VBoxNetFlt: Successfully started.\n"));
205 return 0;
206 }
207
208 LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
209 RTR0Term();
210 }
211 else
212 LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
213
214 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
215 return -RTErrConvertToErrno(rc);
216}
217
218
219/**
220 * Unload the module.
221 *
222 * @todo We have to prevent this if we're busy!
223 */
224static void __exit VBoxNetFltLinuxUnload(void)
225{
226 int rc;
227 Log(("VBoxNetFltLinuxUnload\n"));
228 Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
229
230 /*
231 * Undo the work done during start (in reverse order).
232 */
233 rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
234 AssertRC(rc); NOREF(rc);
235
236 RTR0Term();
237
238 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
239
240 Log(("VBoxNetFltLinuxUnload - done\n"));
241}
242
243
244/**
245 * We filter traffic from the host to the internal network
246 * before it reaches the NIC driver.
247 *
248 * The current code uses a very ugly hack overriding hard_start_xmit
249 * callback in the device structure, but it has been shown to give us a
250 * performance boost of 60-100% though. Eventually we have to find some
251 * less hacky way of getting this job done.
252 */
253#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
254
255#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
256
257# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
258
259# include <linux/ethtool.h>
260
261typedef struct ethtool_ops OVR_OPSTYPE;
262# define OVR_OPS ethtool_ops
263# define OVR_XMIT pfnStartXmit
264
265# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
266
267typedef struct net_device_ops OVR_OPSTYPE;
268# define OVR_OPS netdev_ops
269# define OVR_XMIT pOrgOps->ndo_start_xmit
270
271# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
272
273/**
274 * The overridden net_device_ops of the device we're attached to.
275 *
276 * As there is no net_device_ops structure in pre-2.6.29 kernels we override
277 * ethtool_ops instead along with hard_start_xmit callback in net_device
278 * structure.
279 *
280 * This is a very dirty hack that was created to explore how much we can improve
281 * the host to guest transfers by not CC'ing the NIC. It turns out to be
282 * the only way to filter outgoing packets for devices without TX queue.
283 */
284typedef struct VBoxNetDeviceOpsOverride
285{
286 /** Our overridden ops. */
287 OVR_OPSTYPE Ops;
288 /** Magic word. */
289 uint32_t u32Magic;
290 /** Pointer to the original ops. */
291 OVR_OPSTYPE const *pOrgOps;
292# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
293 /** Pointer to the original hard_start_xmit function. */
294 int (*pfnStartXmit)(struct sk_buff *pSkb, struct net_device *pDev);
295# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
296 /** Pointer to the net filter instance. */
297 PVBOXNETFLTINS pVBoxNetFlt;
298 /** The number of filtered packages. */
299 uint64_t cFiltered;
300 /** The total number of packets */
301 uint64_t cTotal;
302} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
303/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
304#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
305
306/**
307 * ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
308 * because they belong on the internal network.
309 *
310 * @returns NETDEV_TX_XXX.
311 * @param pSkb The socket buffer to transmit.
312 * @param pDev The net device.
313 */
314static int vboxNetFltLinuxStartXmitFilter(struct sk_buff *pSkb, struct net_device *pDev)
315{
316 PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
317 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
318 PCRTNETETHERHDR pEtherHdr;
319 PINTNETTRUNKSWPORT pSwitchPort;
320 uint32_t cbHdrs;
321
322
323 /*
324 * Validate the override structure.
325 *
326 * Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
327 * to be production quality code, we would have to be much more
328 * careful here and avoid the race.
329 */
330 if ( !VALID_PTR(pOverride)
331 || pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
332# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
333 || !VALID_PTR(pOverride->pOrgOps)
334# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
335 )
336 {
337 printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
338 dev_kfree_skb(pSkb);
339 return NETDEV_TX_OK;
340 }
341 pOverride->cTotal++;
342
343 /*
344 * Do the filtering base on the default OUI of our virtual NICs
345 *
346 * Note! In a real solution, we would ask the switch whether the
347 * destination MAC is 100% to be on the internal network and then
348 * drop it.
349 */
350 cbHdrs = skb_headlen(pSkb);
351 cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
352 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
353 if ( pEtherHdr
354 && VALID_PTR(pOverride->pVBoxNetFlt)
355 && (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
356 && VALID_PTR(pSwitchPort)
357 && cbHdrs >= 6)
358 {
359 INTNETSWDECISION enmDecision;
360
361 /** @todo consider reference counting, etc. */
362 enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
363 if (enmDecision == INTNETSWDECISION_INTNET)
364 {
365 dev_kfree_skb(pSkb);
366 pOverride->cFiltered++;
367 return NETDEV_TX_OK;
368 }
369 }
370
371 return pOverride->OVR_XMIT(pSkb, pDev);
372}
373
374/**
375 * Hooks the device ndo_start_xmit operation of the device.
376 *
377 * @param pThis The net filter instance.
378 * @param pDev The net device.
379 */
380static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
381{
382 PVBOXNETDEVICEOPSOVERRIDE pOverride;
383 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
384
385 /* Cancel override if ethtool_ops is missing (host-only case, #5712) */
386 if (!VALID_PTR(pDev->OVR_OPS))
387 return;
388 pOverride = RTMemAlloc(sizeof(*pOverride));
389 if (!pOverride)
390 return;
391 pOverride->pOrgOps = pDev->OVR_OPS;
392 pOverride->Ops = *pDev->OVR_OPS;
393# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
394 pOverride->pfnStartXmit = pDev->hard_start_xmit;
395# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
396 pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
397# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
398 pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
399 pOverride->cTotal = 0;
400 pOverride->cFiltered = 0;
401 pOverride->pVBoxNetFlt = pThis;
402
403 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp); /* (this isn't necessary, but so what) */
404 ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
405# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
406 ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
407# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
408 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
409}
410
411/**
412 * Undos what vboxNetFltLinuxHookDev did.
413 *
414 * @param pThis The net filter instance.
415 * @param pDev The net device. Can be NULL, in which case
416 * we'll try retrieve it from @a pThis.
417 */
418static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
419{
420 PVBOXNETDEVICEOPSOVERRIDE pOverride;
421 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
422
423 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
424 if (!pDev)
425 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
426 if (VALID_PTR(pDev))
427 {
428 pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
429 if ( VALID_PTR(pOverride)
430 && pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
431 && VALID_PTR(pOverride->pOrgOps)
432 )
433 {
434# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
435 ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
436# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
437 ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
438 ASMAtomicWriteU32(&pOverride->u32Magic, 0);
439 }
440 else
441 pOverride = NULL;
442 }
443 else
444 pOverride = NULL;
445 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
446
447 if (pOverride)
448 {
449 printk("vboxnetflt: dropped %llu out of %llu packets\n", pOverride->cFiltered, pOverride->cTotal);
450 RTMemFree(pOverride);
451 }
452}
453
454#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
455
456
457/**
458 * Reads and retains the host interface handle.
459 *
460 * @returns The handle, NULL if detached.
461 * @param pThis
462 */
463DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
464{
465#if 0
466 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
467 struct net_device *pDev = NULL;
468
469 Log(("vboxNetFltLinuxRetainNetDev\n"));
470 /*
471 * Be careful here to avoid problems racing the detached callback.
472 */
473 RTSpinlockAcquire(pThis->hSpinlock, &Tmp);
474 if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
475 {
476 pDev = (struct net_device *)ASMAtomicUoReadPtr((void * volatile *)&pThis->u.s.pDev);
477 if (pDev)
478 {
479 dev_hold(pDev);
480 Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
481 pDev, pDev->name,
482#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
483 netdev_refcnt_read(pDev)
484#else
485 atomic_read(&pDev->refcnt)
486#endif
487 ));
488 }
489 }
490 RTSpinlockRelease(pThis->hSpinlock, &Tmp);
491
492 Log(("vboxNetFltLinuxRetainNetDev - done\n"));
493 return pDev;
494#else
495 return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
496#endif
497}
498
499
500/**
501 * Release the host interface handle previously retained
502 * by vboxNetFltLinuxRetainNetDev.
503 *
504 * @param pThis The instance.
505 * @param pDev The vboxNetFltLinuxRetainNetDev
506 * return value, NULL is fine.
507 */
508DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
509{
510#if 0
511 Log(("vboxNetFltLinuxReleaseNetDev\n"));
512 NOREF(pThis);
513 if (pDev)
514 {
515 dev_put(pDev);
516 Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
517 pDev, pDev->name,
518#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
519 netdev_refcnt_read(pDev)
520#else
521 atomic_read(&pDev->refcnt)
522#endif
523 ));
524 }
525 Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
526#endif
527}
528
529#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 | (skb->dev->ifindex & 0xFFFF))
530#define VBOXNETFLT_SKB_TAG(skb) (*(uint32_t*)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
531
532/**
533 * Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
534 * i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
535 *
536 * @returns true / false accordingly.
537 * @param pBuf The sk_buff.
538 */
539DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
540{
541 return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
542}
543
544
545/**
546 * Internal worker that create a linux sk_buff for a
547 * (scatter/)gather list.
548 *
549 * @returns Pointer to the sk_buff.
550 * @param pThis The instance.
551 * @param pSG The (scatter/)gather list.
552 * @param fDstWire Set if the destination is the wire.
553 */
554static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
555{
556 struct sk_buff *pPkt;
557 struct net_device *pDev;
558 unsigned fGsoType = 0;
559
560 if (pSG->cbTotal == 0)
561 {
562 LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
563 return NULL;
564 }
565
566 /** @todo We should use fragments mapping the SG buffers with large packets.
567 * 256 bytes seems to be the a threshold used a lot for this. It
568 * requires some nasty work on the intnet side though... */
569 /*
570 * Allocate a packet and copy over the data.
571 */
572 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
573 pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
574 if (RT_UNLIKELY(!pPkt))
575 {
576 Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
577 pSG->pvUserData = NULL;
578 return NULL;
579 }
580 pPkt->dev = pDev;
581 pPkt->ip_summed = CHECKSUM_NONE;
582
583 /* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
584 skb_reserve(pPkt, NET_IP_ALIGN);
585
586 /* Copy the segments. */
587 skb_put(pPkt, pSG->cbTotal);
588 IntNetSgRead(pSG, pPkt->data);
589
590#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
591 /*
592 * Setup GSO if used by this packet.
593 */
594 switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
595 {
596 default:
597 AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
598 /* fall thru */
599 case PDMNETWORKGSOTYPE_INVALID:
600 fGsoType = 0;
601 break;
602 case PDMNETWORKGSOTYPE_IPV4_TCP:
603 fGsoType = SKB_GSO_TCPV4;
604 break;
605 case PDMNETWORKGSOTYPE_IPV4_UDP:
606 fGsoType = SKB_GSO_UDP;
607 break;
608 case PDMNETWORKGSOTYPE_IPV6_TCP:
609 fGsoType = SKB_GSO_TCPV6;
610 break;
611 }
612 if (fGsoType)
613 {
614 struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
615
616 pShInfo->gso_type = fGsoType | SKB_GSO_DODGY;
617 pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
618 pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
619
620 /*
621 * We need to set checksum fields even if the packet goes to the host
622 * directly as it may be immediately forwarded by IP layer @bugref{5020}.
623 */
624 Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
625 pPkt->ip_summed = CHECKSUM_PARTIAL;
626# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
627 pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
628 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
629 pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
630 else
631 pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
632# else
633 pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
634 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
635 pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
636 else
637 pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
638# endif
639 if (!fDstWire)
640 PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
641 }
642#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE || VBOXNETFLT_WITH_GSO_XMIT_HOST */
643
644 /*
645 * Finish up the socket buffer.
646 */
647 pPkt->protocol = eth_type_trans(pPkt, pDev);
648 if (fDstWire)
649 {
650 VBOX_SKB_RESET_NETWORK_HDR(pPkt);
651
652 /* Restore ethernet header back. */
653 skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
654 VBOX_SKB_RESET_MAC_HDR(pPkt);
655 }
656 VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
657
658 return pPkt;
659}
660
661
662/**
663 * Initializes a SG list from an sk_buff.
664 *
665 * @returns Number of segments.
666 * @param pThis The instance.
667 * @param pBuf The sk_buff.
668 * @param pSG The SG.
669 * @param pvFrame The frame pointer, optional.
670 * @param cSegs The number of segments allocated for the SG.
671 * This should match the number in the mbuf exactly!
672 * @param fSrc The source of the frame.
673 * @param pGso Pointer to the GSO context if it's a GSO
674 * internal network frame. NULL if regular frame.
675 */
676DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
677 unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
678{
679 int i;
680 NOREF(pThis);
681
682 Assert(!skb_shinfo(pBuf)->frag_list);
683
684 if (!pGsoCtx)
685 IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/);
686 else
687 IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/, pGsoCtx);
688
689#ifdef VBOXNETFLT_SG_SUPPORT
690 pSG->aSegs[0].cb = skb_headlen(pBuf);
691 pSG->aSegs[0].pv = pBuf->data;
692 pSG->aSegs[0].Phys = NIL_RTHCPHYS;
693
694 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
695 {
696 skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
697 pSG->aSegs[i+1].cb = pFrag->size;
698 pSG->aSegs[i+1].pv = kmap(pFrag->page);
699 printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
700 pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
701 }
702 ++i;
703
704#else
705 pSG->aSegs[0].cb = pBuf->len;
706 pSG->aSegs[0].pv = pBuf->data;
707 pSG->aSegs[0].Phys = NIL_RTHCPHYS;
708 i = 1;
709#endif
710
711 pSG->cSegsUsed = i;
712
713#ifdef PADD_RUNT_FRAMES_FROM_HOST
714 /*
715 * Add a trailer if the frame is too small.
716 *
717 * Since we're getting to the packet before it is framed, it has not
718 * yet been padded. The current solution is to add a segment pointing
719 * to a buffer containing all zeros and pray that works for all frames...
720 */
721 if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
722 {
723 static uint8_t const s_abZero[128] = {0};
724
725 AssertReturnVoid(i < cSegs);
726
727 pSG->aSegs[i].Phys = NIL_RTHCPHYS;
728 pSG->aSegs[i].pv = (void *)&s_abZero[0];
729 pSG->aSegs[i].cb = 60 - pSG->cbTotal;
730 pSG->cbTotal = 60;
731 pSG->cSegsUsed++;
732 Assert(i + 1 <= pSG->cSegsAlloc)
733 }
734#endif
735
736 Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
737 pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
738 for (i = 0; i < pSG->cSegsUsed; i++)
739 Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
740 i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
741}
742
743/**
744 * Packet handler,
745 *
746 * @returns 0 or EJUSTRETURN.
747 * @param pThis The instance.
748 * @param pMBuf The mbuf.
749 * @param pvFrame The start of the frame, optional.
750 * @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
751 * @param eProtocol The protocol.
752 */
753#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
754static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
755 struct net_device *pSkbDev,
756 struct packet_type *pPacketType,
757 struct net_device *pOrigDev)
758#else
759static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
760 struct net_device *pSkbDev,
761 struct packet_type *pPacketType)
762#endif
763{
764 PVBOXNETFLTINS pThis;
765 struct net_device *pDev;
766 LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
767 pBuf, pSkbDev, pPacketType));
768#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
769 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",
770 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));
771# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
772 Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
773# endif
774#else
775 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",
776 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));
777#endif
778 /*
779 * Drop it immediately?
780 */
781 if (!pBuf)
782 return 0;
783
784 pThis = VBOX_FLT_PT_TO_INST(pPacketType);
785 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
786 if (pDev != pSkbDev)
787 {
788 Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
789 return 0;
790 }
791
792 Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
793 if (vboxNetFltLinuxSkBufIsOur(pBuf))
794 {
795 Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
796 dev_kfree_skb(pBuf);
797 return 0;
798 }
799
800#ifndef VBOXNETFLT_SG_SUPPORT
801 {
802 /*
803 * Get rid of fragmented packets, they cause too much trouble.
804 */
805 struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
806 kfree_skb(pBuf);
807 if (!pCopy)
808 {
809 LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
810 return 0;
811 }
812 pBuf = pCopy;
813# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
814 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",
815 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));
816# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
817 Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
818# endif
819# else
820 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",
821 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));
822# endif
823 }
824#endif
825
826#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
827 /* Forward it to the internal network. */
828 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
829#else
830 /* Add the packet to transmit queue and schedule the bottom half. */
831 skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
832 schedule_work(&pThis->u.s.XmitTask);
833 Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
834 &pThis->u.s.XmitTask, pBuf));
835#endif
836
837 /* It does not really matter what we return, it is ignored by the kernel. */
838 return 0;
839}
840
841/**
842 * Calculate the number of INTNETSEG segments the socket buffer will need.
843 *
844 * @returns Segment count.
845 * @param pBuf The socket buffer.
846 */
847DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
848{
849#ifdef VBOXNETFLT_SG_SUPPORT
850 unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
851#else
852 unsigned cSegs = 1;
853#endif
854#ifdef PADD_RUNT_FRAMES_FROM_HOST
855 /* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
856 if (pBuf->len < 60)
857 cSegs++;
858#endif
859 return cSegs;
860}
861
862/**
863 * Destroy the intnet scatter / gather buffer created by
864 * vboxNetFltLinuxSkBufToSG.
865 */
866static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
867{
868#ifdef VBOXNETFLT_SG_SUPPORT
869 int i;
870
871 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
872 {
873 printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
874 kunmap(pSG->aSegs[i+1].pv);
875 }
876#endif
877 NOREF(pSG);
878}
879
880#ifdef LOG_ENABLED
881/**
882 * Logging helper.
883 */
884static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
885{
886 int i, offSeg;
887 uint8_t *pInt, *pExt;
888 static int iPacketNo = 1;
889 iPacketNo += iIncrement;
890 if (fEgress)
891 {
892 pExt = pSG->aSegs[0].pv;
893 pInt = pExt + 6;
894 }
895 else
896 {
897 pInt = pSG->aSegs[0].pv;
898 pExt = pInt + 6;
899 }
900 Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
901 " %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
902 pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
903 fEgress ? "-->" : "<--", pszWhere,
904 pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
905 pSG->cbTotal, iPacketNo));
906 if (pSG->cSegsUsed == 1)
907 {
908 Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
909 }
910 else
911 {
912 for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
913 {
914 Log3(("-- segment %d at 0x%x (%d bytes) --\n%.*Rhxd\n",
915 i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
916 offSeg += pSG->aSegs[i].cb;
917 }
918 }
919
920}
921#else
922# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
923#endif
924
925#ifdef VBOXNETFLT_WITH_GSO_RECV
926
927/**
928 * Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
929 * GSO socket buffer without having to segment it.
930 *
931 * @returns true on success, false if needs segmenting.
932 * @param pThis The net filter instance.
933 * @param pSkb The GSO socket buffer.
934 * @param fSrc The source.
935 * @param pGsoCtx Where to return the GSO context on success.
936 */
937static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
938 PPDMNETWORKGSO pGsoCtx)
939{
940 PDMNETWORKGSOTYPE enmGsoType;
941 uint16_t uEtherType;
942 unsigned int cbTransport;
943 unsigned int offTransport;
944 unsigned int cbTransportHdr;
945 unsigned uProtocol;
946 union
947 {
948 RTNETIPV4 IPv4;
949 RTNETIPV6 IPv6;
950 RTNETTCP Tcp;
951 uint8_t ab[40];
952 uint16_t au16[40/2];
953 uint32_t au32[40/4];
954 } Buf;
955
956 /*
957 * Check the GSO properties of the socket buffer and make sure it fits.
958 */
959 /** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
960 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP | SKB_GSO_DODGY | SKB_GSO_TCPV6 | SKB_GSO_TCPV4) ))
961 {
962 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
963 return false;
964 }
965 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
966 || pSkb->len > VBOX_MAX_GSO_SIZE ))
967 {
968 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
969 return false;
970 }
971 /*
972 * It is possible to receive GSO packets from wire if GRO is enabled.
973 */
974 if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
975 {
976 Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
977#ifdef VBOXNETFLT_WITH_GRO
978 /*
979 * The packet came from the wire and the driver has already consumed
980 * mac header. We need to restore it back.
981 */
982 pSkb->mac_len = skb_network_header(pSkb) - skb_mac_header(pSkb);
983 skb_push(pSkb, pSkb->mac_len);
984 Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
985 pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
986#else /* !VBOXNETFLT_WITH_GRO */
987 /* Older kernels didn't have GRO. */
988 return false;
989#endif /* !VBOXNETFLT_WITH_GRO */
990 }
991 else
992 {
993 /*
994 * skb_gso_segment does the following. Do we need to do it as well?
995 */
996#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
997 skb_reset_mac_header(pSkb);
998 pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
999#else
1000 pSkb->mac.raw = pSkb->data;
1001 pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1002#endif
1003 }
1004
1005 /*
1006 * Switch on the ethertype.
1007 */
1008 uEtherType = pSkb->protocol;
1009 if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1010 && pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1011 {
1012 uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1013 if (puEtherType)
1014 uEtherType = *puEtherType;
1015 }
1016 switch (uEtherType)
1017 {
1018 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1019 {
1020 unsigned int cbHdr;
1021 PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1022 if (RT_UNLIKELY(!pIPv4))
1023 {
1024 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1025 return false;
1026 }
1027
1028 cbHdr = pIPv4->ip_hl * 4;
1029 cbTransport = RT_N2H_U16(pIPv4->ip_len);
1030 if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1031 || cbHdr > cbTransport ))
1032 {
1033 Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1034 return false;
1035 }
1036 cbTransport -= cbHdr;
1037 offTransport = pSkb->mac_len + cbHdr;
1038 uProtocol = pIPv4->ip_p;
1039 if (uProtocol == RTNETIPV4_PROT_TCP)
1040 enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1041 else if (uProtocol == RTNETIPV4_PROT_UDP)
1042 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1043 else /** @todo IPv6: 4to6 tunneling */
1044 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1045 break;
1046 }
1047
1048 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1049 {
1050 PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1051 if (RT_UNLIKELY(!pIPv6))
1052 {
1053 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1054 return false;
1055 }
1056
1057 cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1058 offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1059 uProtocol = pIPv6->ip6_nxt;
1060 /** @todo IPv6: Dig our way out of the other headers. */
1061 if (uProtocol == RTNETIPV4_PROT_TCP)
1062 enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1063 else if (uProtocol == RTNETIPV4_PROT_UDP)
1064 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1065 else
1066 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1067 break;
1068 }
1069
1070 default:
1071 Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1072 return false;
1073 }
1074
1075 if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1076 {
1077 Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1078 return false;
1079 }
1080
1081 if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1082 || offTransport + cbTransport > pSkb->len
1083 || cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1084 {
1085 Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1086 offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1087 return false;
1088 }
1089
1090 /*
1091 * Check the TCP/UDP bits.
1092 */
1093 if (uProtocol == RTNETIPV4_PROT_TCP)
1094 {
1095 PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1096 if (RT_UNLIKELY(!pTcp))
1097 {
1098 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1099 return false;
1100 }
1101
1102 cbTransportHdr = pTcp->th_off * 4;
1103 pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1104 if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1105 || cbTransportHdr > cbTransport
1106 || offTransport + cbTransportHdr >= UINT8_MAX
1107 || offTransport + cbTransportHdr >= pSkb->len ))
1108 {
1109 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1110 return false;
1111 }
1112
1113 }
1114 else
1115 {
1116 Assert(uProtocol == RTNETIPV4_PROT_UDP);
1117 cbTransportHdr = sizeof(RTNETUDP);
1118 pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1119 if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1120 || offTransport + cbTransportHdr >= pSkb->len ))
1121 {
1122 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1123 return false;
1124 }
1125 }
1126
1127 /*
1128 * We're good, init the GSO context.
1129 */
1130 pGsoCtx->u8Type = enmGsoType;
1131 pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1132 pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1133 pGsoCtx->offHdr1 = pSkb->mac_len;
1134 pGsoCtx->offHdr2 = offTransport;
1135 pGsoCtx->u8Unused = 0;
1136
1137 return true;
1138}
1139
1140/**
1141 * Forward the socket buffer as a GSO internal network frame.
1142 *
1143 * @returns IPRT status code.
1144 * @param pThis The net filter instance.
1145 * @param pSkb The GSO socket buffer.
1146 * @param fSrc The source.
1147 * @param pGsoCtx Where to return the GSO context on success.
1148 */
1149static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1150{
1151 int rc;
1152 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1153 if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1154 {
1155 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1156 if (RT_LIKELY(pSG))
1157 {
1158 vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1159
1160 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1161 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1162
1163 vboxNetFltLinuxDestroySG(pSG);
1164 rc = VINF_SUCCESS;
1165 }
1166 else
1167 {
1168 Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1169 rc = VERR_NO_MEMORY;
1170 }
1171 }
1172 else
1173 {
1174 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1175 rc = VERR_INTERNAL_ERROR_3;
1176 }
1177
1178 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1179 dev_kfree_skb(pSkb);
1180 return rc;
1181}
1182
1183#endif /* VBOXNETFLT_WITH_GSO_RECV */
1184
1185/**
1186 * Worker for vboxNetFltLinuxForwardToIntNet.
1187 *
1188 * @returns VINF_SUCCESS or VERR_NO_MEMORY.
1189 * @param pThis The net filter instance.
1190 * @param pBuf The socket buffer.
1191 * @param fSrc The source.
1192 */
1193static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1194{
1195 int rc;
1196 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1197 if (cSegs <= MAX_SKB_FRAGS + 1)
1198 {
1199 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1200 if (RT_LIKELY(pSG))
1201 {
1202 if (fSrc & INTNETTRUNKDIR_WIRE)
1203 {
1204 /*
1205 * The packet came from wire, ethernet header was removed by device driver.
1206 * Restore it.
1207 */
1208 skb_push(pBuf, ETH_HLEN);
1209 }
1210
1211 vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /*pGsoCtx*/);
1212
1213 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1214 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1215
1216 vboxNetFltLinuxDestroySG(pSG);
1217 rc = VINF_SUCCESS;
1218 }
1219 else
1220 {
1221 Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1222 rc = VERR_NO_MEMORY;
1223 }
1224 }
1225 else
1226 {
1227 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1228 rc = VERR_INTERNAL_ERROR_3;
1229 }
1230
1231 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1232 dev_kfree_skb(pBuf);
1233 return rc;
1234}
1235
1236/**
1237 *
1238 * @param pBuf The socket buffer. This is consumed by this function.
1239 */
1240static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1241{
1242 uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1243
1244#ifdef VBOXNETFLT_WITH_GSO
1245 if (skb_is_gso(pBuf))
1246 {
1247 PDMNETWORKGSO GsoCtx;
1248 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",
1249 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));
1250# ifdef VBOXNETFLT_WITH_GSO_RECV
1251 if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP | SKB_GSO_TCPV6 | SKB_GSO_TCPV4))
1252 && vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1253 vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1254 else
1255# endif
1256 {
1257 /* Need to segment the packet */
1258 struct sk_buff *pNext;
1259 struct sk_buff *pSegment = skb_gso_segment(pBuf, 0 /*supported features*/);
1260 if (IS_ERR(pSegment))
1261 {
1262 dev_kfree_skb(pBuf);
1263 LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1264 return;
1265 }
1266
1267 for (; pSegment; pSegment = pNext)
1268 {
1269 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",
1270 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));
1271 pNext = pSegment->next;
1272 pSegment->next = 0;
1273 vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1274 }
1275 dev_kfree_skb(pBuf);
1276 }
1277 }
1278 else
1279#endif /* VBOXNETFLT_WITH_GSO */
1280 {
1281 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1282 {
1283#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1284 /*
1285 * Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1286 * and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1287 * header length from the header itself and reconstruct 'h' pointer
1288 * to TCP (or whatever) header.
1289 */
1290 unsigned char *tmp = pBuf->h.raw;
1291 if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1292 pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1293#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1294 if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1295 {
1296 LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1297 dev_kfree_skb(pBuf);
1298 return;
1299 }
1300#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1301 /* Restore the original (wrong) pointer. */
1302 pBuf->h.raw = tmp;
1303#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1304 }
1305 vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1306 }
1307}
1308
1309#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1310/**
1311 * Work queue handler that forwards the socket buffers queued by
1312 * vboxNetFltLinuxPacketHandler to the internal network.
1313 *
1314 * @param pWork The work queue.
1315 */
1316# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1317static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1318# else
1319static void vboxNetFltLinuxXmitTask(void *pWork)
1320# endif
1321{
1322 PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1323 struct sk_buff *pBuf;
1324
1325 Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1326
1327 /*
1328 * Active? Retain the instance and increment the busy counter.
1329 */
1330 if (vboxNetFltTryRetainBusyActive(pThis))
1331 {
1332 while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1333 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1334
1335 vboxNetFltRelease(pThis, true /* fBusy */);
1336 }
1337 else
1338 {
1339 /** @todo Shouldn't we just drop the packets here? There is little point in
1340 * making them accumulate when the VM is paused and it'll only waste
1341 * kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1342 * before start draining the packets (goes for the intnet ring buf
1343 * too)? */
1344 }
1345}
1346#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1347
1348/**
1349 * Reports the GSO capabilities of the hardware NIC.
1350 *
1351 * @param pThis The net filter instance. The caller hold a
1352 * reference to this.
1353 */
1354static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1355{
1356#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1357 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1358 {
1359 struct net_device *pDev;
1360 PINTNETTRUNKSWPORT pSwitchPort;
1361 unsigned int fFeatures;
1362 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1363
1364 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1365
1366 pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1367 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1368 if (pDev)
1369 fFeatures = pDev->features;
1370 else
1371 fFeatures = 0;
1372
1373 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1374
1375 if (pThis->pSwitchPort)
1376 {
1377 /* Set/update the GSO capabilities of the NIC. */
1378 uint32_t fGsoCapabilites = 0;
1379 if (fFeatures & NETIF_F_TSO)
1380 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1381 if (fFeatures & NETIF_F_TSO6)
1382 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1383# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1384 if (fFeatures & NETIF_F_UFO)
1385 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1386 if (fFeatures & NETIF_F_UFO)
1387 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1388# endif
1389 Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1390 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1391 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : "",
1392 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)) ? "ufo " : "",
1393 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)) ? "ufo6 " : ""));
1394 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1395 }
1396
1397 vboxNetFltRelease(pThis, true /*fBusy*/);
1398 }
1399#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1400}
1401
1402/**
1403 * Helper that determines whether the host (ignoreing us) is operating the
1404 * interface in promiscuous mode or not.
1405 */
1406static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1407{
1408 bool fRc = false;
1409 struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1410 if (pDev)
1411 {
1412 fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1413 LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1414 fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1415 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1416 }
1417 return fRc;
1418}
1419
1420#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36)
1421/**
1422 * Helper for detecting TAP devices.
1423 */
1424static bool vboxNetFltIsTapDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1425{
1426 if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1427 {
1428 struct ethtool_drvinfo Info;
1429
1430 memset(&Info, 0, sizeof(Info));
1431 Info.cmd = ETHTOOL_GDRVINFO;
1432 pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1433 Log3(("vboxNetFltIsTapDevice: driver=%s version=%s bus_info=%s\n",
1434 Info.driver, Info.version, Info.bus_info));
1435
1436 return !strncmp(Info.driver, "tun", 4)
1437 && !strncmp(Info.bus_info, "tap", 4);
1438 }
1439
1440 return false;
1441}
1442
1443/**
1444 * Helper for updating the link state of TAP devices.
1445 * Only TAP devices are affected.
1446 */
1447static void vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1448{
1449 if (vboxNetFltIsTapDevice(pThis, pDev))
1450 {
1451 Log3(("vboxNetFltSetTapLinkState: bringing %s tap device link state\n",
1452 fLinkUp ? "up" : "down"));
1453 netif_tx_lock_bh(pDev);
1454 if (fLinkUp)
1455 netif_carrier_on(pDev);
1456 else
1457 netif_carrier_off(pDev);
1458 netif_tx_unlock_bh(pDev);
1459 }
1460}
1461#else /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1462DECLINLINE(void) vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1463{
1464 /* Nothing to do for pre-2.6.36 kernels. */
1465}
1466#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1467
1468/**
1469 * Internal worker for vboxNetFltLinuxNotifierCallback.
1470 *
1471 * @returns VBox status code.
1472 * @param pThis The instance.
1473 * @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1474 * flood the release log.
1475 */
1476static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1477{
1478 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1479 LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1480
1481 /*
1482 * Retain and store the device.
1483 */
1484 dev_hold(pDev);
1485
1486 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1487 ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1488 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1489
1490 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1491 pDev, pDev->name,
1492#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1493 netdev_refcnt_read(pDev)
1494#else
1495 atomic_read(&pDev->refcnt)
1496#endif
1497 ));
1498 Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1499 pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1500
1501 /* Get the mac address while we still have a valid net_device reference. */
1502 memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1503
1504 /*
1505 * Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1506 */
1507 pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1508 pThis->u.s.PacketType.dev = pDev;
1509 pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1510 dev_add_pack(&pThis->u.s.PacketType);
1511
1512#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1513 vboxNetFltLinuxHookDev(pThis, pDev);
1514#endif
1515
1516 /*
1517 * If attaching to TAP interface we need to bring the link state up
1518 * starting from 2.6.36 kernel.
1519 */
1520 vboxNetFltSetTapLinkState(pThis, pDev, true);
1521
1522 /*
1523 * Set indicators that require the spinlock. Be abit paranoid about racing
1524 * the device notification handle.
1525 */
1526 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1527 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1528 if (pDev)
1529 {
1530 ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1531 ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1532 pDev = NULL; /* don't dereference it */
1533 }
1534 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1535 Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1536
1537 /*
1538 * If the above succeeded report GSO capabilities, if not undo and
1539 * release the device.
1540 */
1541 if (!pDev)
1542 {
1543 Assert(pThis->pSwitchPort);
1544 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1545 {
1546 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1547 pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1548 pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1549 pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE | INTNETTRUNKDIR_HOST);
1550 vboxNetFltRelease(pThis, true /*fBusy*/);
1551 }
1552 }
1553 else
1554 {
1555#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1556 vboxNetFltLinuxUnhookDev(pThis, pDev);
1557#endif
1558 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1559 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1560 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1561 dev_put(pDev);
1562 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n",
1563 pDev, pDev->name,
1564#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1565 netdev_refcnt_read(pDev)
1566#else
1567 atomic_read(&pDev->refcnt)
1568#endif
1569 ));
1570 }
1571
1572 LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1573 return VINF_SUCCESS;
1574}
1575
1576
1577static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1578{
1579 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1580
1581 Assert(!pThis->fDisconnectedFromHost);
1582
1583#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1584 vboxNetFltLinuxUnhookDev(pThis, pDev);
1585#endif
1586
1587 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1588 ASMAtomicWriteBool(&pThis->u.s.fRegistered, false);
1589 ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1590 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1591 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1592
1593 dev_remove_pack(&pThis->u.s.PacketType);
1594#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1595 skb_queue_purge(&pThis->u.s.XmitQueue);
1596#endif
1597 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
1598 Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1599 pDev, pDev->name,
1600#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1601 netdev_refcnt_read(pDev)
1602#else
1603 atomic_read(&pDev->refcnt)
1604#endif
1605 ));
1606 dev_put(pDev);
1607
1608 return NOTIFY_OK;
1609}
1610
1611static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1612{
1613 /* Check if we are not suspended and promiscuous mode has not been set. */
1614 if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1615 && !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1616 {
1617 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1618 dev_set_promiscuity(pDev, 1);
1619 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1620 Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1621 }
1622 else
1623 Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1624 return NOTIFY_OK;
1625}
1626
1627static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1628{
1629 /* Undo promiscuous mode if we has set it. */
1630 if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1631 {
1632 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1633 dev_set_promiscuity(pDev, -1);
1634 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1635 Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1636 }
1637 else
1638 Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1639 return NOTIFY_OK;
1640}
1641
1642#ifdef LOG_ENABLED
1643/** Stringify the NETDEV_XXX constants. */
1644static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
1645{
1646 const char *pszEvent = "NETDRV_<unknown>";
1647 switch (ulEventType)
1648 {
1649 case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
1650 case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
1651 case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
1652 case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
1653 case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
1654 case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
1655 case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
1656 case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
1657 case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
1658 case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
1659# ifdef NETDEV_FEAT_CHANGE
1660 case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
1661# endif
1662 }
1663 return pszEvent;
1664}
1665#endif /* LOG_ENABLED */
1666
1667/**
1668 * Callback for listening to netdevice events.
1669 *
1670 * This works the rediscovery, clean up on unregistration, promiscuity on
1671 * up/down, and GSO feature changes from ethtool.
1672 *
1673 * @returns NOTIFY_OK
1674 * @param self Pointer to our notifier registration block.
1675 * @param ulEventType The event.
1676 * @param ptr Event specific, but it is usually the device it
1677 * relates to.
1678 */
1679static int vboxNetFltLinuxNotifierCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
1680
1681{
1682 PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
1683 struct net_device *pDev = (struct net_device *)ptr;
1684 int rc = NOTIFY_OK;
1685
1686 Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1687 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1688 if ( ulEventType == NETDEV_REGISTER
1689 && !strcmp(pDev->name, pThis->szName))
1690 {
1691 vboxNetFltLinuxAttachToInterface(pThis, pDev);
1692 }
1693 else
1694 {
1695 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1696 if (pDev == ptr)
1697 {
1698 switch (ulEventType)
1699 {
1700 case NETDEV_UNREGISTER:
1701 rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
1702 break;
1703 case NETDEV_UP:
1704 rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
1705 break;
1706 case NETDEV_GOING_DOWN:
1707 rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
1708 break;
1709 case NETDEV_CHANGENAME:
1710 break;
1711#ifdef NETDEV_FEAT_CHANGE
1712 case NETDEV_FEAT_CHANGE:
1713 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1714 break;
1715#endif
1716 }
1717 }
1718 }
1719
1720 return rc;
1721}
1722
1723bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
1724{
1725 return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
1726}
1727
1728int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
1729{
1730 struct net_device * pDev;
1731 int err;
1732 int rc = VINF_SUCCESS;
1733 NOREF(pvIfData);
1734
1735 LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
1736
1737 pDev = vboxNetFltLinuxRetainNetDev(pThis);
1738 if (pDev)
1739 {
1740 /*
1741 * Create a sk_buff for the gather list and push it onto the wire.
1742 */
1743 if (fDst & INTNETTRUNKDIR_WIRE)
1744 {
1745 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
1746 if (pBuf)
1747 {
1748 vboxNetFltDumpPacket(pSG, true, "wire", 1);
1749 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1750 Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
1751 err = dev_queue_xmit(pBuf);
1752 if (err)
1753 rc = RTErrConvertFromErrno(err);
1754 }
1755 else
1756 rc = VERR_NO_MEMORY;
1757 }
1758
1759 /*
1760 * Create a sk_buff for the gather list and push it onto the host stack.
1761 */
1762 if (fDst & INTNETTRUNKDIR_HOST)
1763 {
1764 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
1765 if (pBuf)
1766 {
1767 vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
1768 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1769 Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
1770 err = netif_rx_ni(pBuf);
1771 if (err)
1772 rc = RTErrConvertFromErrno(err);
1773 }
1774 else
1775 rc = VERR_NO_MEMORY;
1776 }
1777
1778 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1779 }
1780
1781 return rc;
1782}
1783
1784
1785void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
1786{
1787 struct net_device * pDev;
1788
1789 LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
1790 pThis, pThis->szName, fActive?"true":"false",
1791 pThis->fDisablePromiscuous?"true":"false"));
1792
1793 if (pThis->fDisablePromiscuous)
1794 return;
1795
1796 pDev = vboxNetFltLinuxRetainNetDev(pThis);
1797 if (pDev)
1798 {
1799 /*
1800 * This api is a bit weird, the best reference is the code.
1801 *
1802 * Also, we have a bit or race conditions wrt the maintenance of
1803 * host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
1804 */
1805#ifdef LOG_ENABLED
1806 u_int16_t fIf;
1807 unsigned const cPromiscBefore = pDev->promiscuity;
1808#endif
1809 if (fActive)
1810 {
1811 Assert(!pThis->u.s.fPromiscuousSet);
1812
1813 rtnl_lock();
1814 dev_set_promiscuity(pDev, 1);
1815 rtnl_unlock();
1816 pThis->u.s.fPromiscuousSet = true;
1817 Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1818 }
1819 else
1820 {
1821 if (pThis->u.s.fPromiscuousSet)
1822 {
1823 rtnl_lock();
1824 dev_set_promiscuity(pDev, -1);
1825 rtnl_unlock();
1826 Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1827 }
1828 pThis->u.s.fPromiscuousSet = false;
1829
1830#ifdef LOG_ENABLED
1831 fIf = dev_get_flags(pDev);
1832 Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
1833#endif
1834 }
1835
1836 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1837 }
1838}
1839
1840
1841int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
1842{
1843 /*
1844 * Remove packet handler when we get disconnected from internal switch as
1845 * we don't want the handler to forward packets to disconnected switch.
1846 */
1847 dev_remove_pack(&pThis->u.s.PacketType);
1848 return VINF_SUCCESS;
1849}
1850
1851
1852int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
1853{
1854 /*
1855 * Report the GSO capabilities of the host and device (if connected).
1856 * Note! No need to mark ourselves busy here.
1857 */
1858 /** @todo duplicate work here now? Attach */
1859#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1860 Log3(("vboxNetFltOsConnectIt: reporting host tso tso6 ufo\n"));
1861 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
1862 0
1863 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
1864 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
1865 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
1866# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1867 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
1868# endif
1869 , INTNETTRUNKDIR_HOST);
1870
1871#endif
1872 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1873
1874 return VINF_SUCCESS;
1875}
1876
1877
1878void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
1879{
1880 struct net_device *pDev;
1881 bool fRegistered;
1882 RTSPINLOCKTMP Tmp = RTSPINLOCKTMP_INITIALIZER;
1883
1884#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1885 vboxNetFltLinuxUnhookDev(pThis, NULL);
1886#endif
1887
1888 /** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
1889 * unlikely, but none the less). Since it doesn't actually update the
1890 * state (just reads it), it is likely to panic in some interesting
1891 * ways. */
1892
1893 RTSpinlockAcquireNoInts(pThis->hSpinlock, &Tmp);
1894 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1895 fRegistered = ASMAtomicUoReadBool(&pThis->u.s.fRegistered);
1896 RTSpinlockReleaseNoInts(pThis->hSpinlock, &Tmp);
1897
1898 if (fRegistered)
1899 {
1900 vboxNetFltSetTapLinkState(pThis, pDev, false);
1901
1902#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1903 skb_queue_purge(&pThis->u.s.XmitQueue);
1904#endif
1905 Log(("vboxNetFltOsDeleteInstance: this=%p: Packet handler removed, xmit queue purged.\n", pThis));
1906 Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
1907 pDev, pDev->name,
1908#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1909 netdev_refcnt_read(pDev)
1910#else
1911 atomic_read(&pDev->refcnt)
1912#endif
1913 ));
1914 dev_put(pDev);
1915 }
1916 Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
1917 unregister_netdevice_notifier(&pThis->u.s.Notifier);
1918 module_put(THIS_MODULE);
1919}
1920
1921
1922int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
1923{
1924 int err;
1925 NOREF(pvContext);
1926
1927 pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
1928 err = register_netdevice_notifier(&pThis->u.s.Notifier);
1929 if (err)
1930 return VERR_INTNET_FLT_IF_FAILED;
1931 if (!pThis->u.s.fRegistered)
1932 {
1933 unregister_netdevice_notifier(&pThis->u.s.Notifier);
1934 LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
1935 return VERR_INTNET_FLT_IF_NOT_FOUND;
1936 }
1937
1938 Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
1939 if ( pThis->fDisconnectedFromHost
1940 || !try_module_get(THIS_MODULE))
1941 return VERR_INTNET_FLT_IF_FAILED;
1942
1943 return VINF_SUCCESS;
1944}
1945
1946int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
1947{
1948 /*
1949 * Init the linux specific members.
1950 */
1951 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1952 pThis->u.s.fRegistered = false;
1953 pThis->u.s.fPromiscuousSet = false;
1954 memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
1955#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1956 skb_queue_head_init(&pThis->u.s.XmitQueue);
1957# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1958 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
1959# else
1960 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
1961# endif
1962#endif
1963
1964 return VINF_SUCCESS;
1965}
1966
1967
1968void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
1969{
1970 NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
1971}
1972
1973
1974int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void *pvIf, void **pvIfData)
1975{
1976 /* Nothing to do */
1977 NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
1978 return VINF_SUCCESS;
1979}
1980
1981
1982int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
1983{
1984 /* Nothing to do */
1985 NOREF(pThis); NOREF(pvIfData);
1986 return VINF_SUCCESS;
1987}
1988
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