VirtualBox

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 44528

Last change on this file since 44528 was 44528, checked in by vboxsync, 12 years ago

header (C) fixes
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 70.2 KB
Line 
1/* $Id: VBoxNetFlt-linux.c 44528 2013-02-04 14:27:54Z vboxsync $ */
2/** @file
3 * VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4 */
5
6/*
7 * Copyright (C) 2006-2013 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#include <linux/if_vlan.h>
32
33#include <VBox/log.h>
34#include <VBox/err.h>
35#include <VBox/intnetinline.h>
36#include <VBox/vmm/pdmnetinline.h>
37#include <VBox/param.h>
38#include <iprt/alloca.h>
39#include <iprt/assert.h>
40#include <iprt/spinlock.h>
41#include <iprt/semaphore.h>
42#include <iprt/initterm.h>
43#include <iprt/process.h>
44#include <iprt/mem.h>
45#include <iprt/net.h>
46#include <iprt/log.h>
47#include <iprt/mp.h>
48#include <iprt/mem.h>
49#include <iprt/time.h>
50
51#define VBOXNETFLT_OS_SPECFIC 1
52#include "../VBoxNetFltInternal.h"
53
54
55/*******************************************************************************
56* Defined Constants And Macros *
57*******************************************************************************/
58#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
59#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
60#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
61# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
62#endif
63
64#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
65# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
66# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
67#else
68# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
69# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
70#endif
71
72#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
73# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
74#else
75# define CHECKSUM_PARTIAL CHECKSUM_HW
76# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
77# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
78# else
79# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
80# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
81# else
82# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
83# endif
84/* Versions prior 2.6.10 use stats for both bstats and qstats */
85# define bstats stats
86# define qstats stats
87# endif
88#endif
89
90#ifndef NET_IP_ALIGN
91# define NET_IP_ALIGN 2
92#endif
93
94#if 0
95/** Create scatter / gather segments for fragments. When not used, we will
96 * linearize the socket buffer before creating the internal networking SG. */
97# define VBOXNETFLT_SG_SUPPORT 1
98#endif
99
100#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
101/** Indicates that the linux kernel may send us GSO frames. */
102# define VBOXNETFLT_WITH_GSO 1
103
104/** This enables or disables the transmitting of GSO frame from the internal
105 * network and to the host. */
106# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
107
108# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
109/** This enables or disables the transmitting of GSO frame from the internal
110 * network and to the wire. */
111# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
112# endif
113
114/** This enables or disables the forwarding/flooding of GSO frame from the host
115 * to the internal network. */
116# define VBOXNETFLT_WITH_GSO_RECV 1
117
118#endif
119
120#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
121/** This enables or disables handling of GSO frames coming from the wire (GRO). */
122# define VBOXNETFLT_WITH_GRO 1
123#endif
124/*
125 * GRO support was backported to RHEL 5.4
126 */
127#ifdef RHEL_RELEASE_CODE
128# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
129# define VBOXNETFLT_WITH_GRO 1
130# endif
131#endif
132
133/*******************************************************************************
134* Internal Functions *
135*******************************************************************************/
136static int VBoxNetFltLinuxInit(void);
137static void VBoxNetFltLinuxUnload(void);
138static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
139
140
141/*******************************************************************************
142* Global Variables *
143*******************************************************************************/
144/**
145 * The (common) global data.
146 */
147static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
148
149module_init(VBoxNetFltLinuxInit);
150module_exit(VBoxNetFltLinuxUnload);
151
152MODULE_AUTHOR(VBOX_VENDOR);
153MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
154MODULE_LICENSE("GPL");
155#ifdef MODULE_VERSION
156MODULE_VERSION(VBOX_VERSION_STRING " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
157#endif
158
159
160#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
161unsigned dev_get_flags(const struct net_device *dev)
162{
163 unsigned flags;
164
165 flags = (dev->flags & ~(IFF_PROMISC |
166 IFF_ALLMULTI |
167 IFF_RUNNING)) |
168 (dev->gflags & (IFF_PROMISC |
169 IFF_ALLMULTI));
170
171 if (netif_running(dev) && netif_carrier_ok(dev))
172 flags |= IFF_RUNNING;
173
174 return flags;
175}
176#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
177
178
179/**
180 * Initialize module.
181 *
182 * @returns appropriate status code.
183 */
184static int __init VBoxNetFltLinuxInit(void)
185{
186 int rc;
187 /*
188 * Initialize IPRT.
189 */
190 rc = RTR0Init(0);
191 if (RT_SUCCESS(rc))
192 {
193 Log(("VBoxNetFltLinuxInit\n"));
194
195 /*
196 * Initialize the globals and connect to the support driver.
197 *
198 * This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
199 * for establishing the connect to the support driver.
200 */
201 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
202 rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
203 if (RT_SUCCESS(rc))
204 {
205 LogRel(("VBoxNetFlt: Successfully started.\n"));
206 return 0;
207 }
208
209 LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
210 RTR0Term();
211 }
212 else
213 LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
214
215 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
216 return -RTErrConvertToErrno(rc);
217}
218
219
220/**
221 * Unload the module.
222 *
223 * @todo We have to prevent this if we're busy!
224 */
225static void __exit VBoxNetFltLinuxUnload(void)
226{
227 int rc;
228 Log(("VBoxNetFltLinuxUnload\n"));
229 Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
230
231 /*
232 * Undo the work done during start (in reverse order).
233 */
234 rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
235 AssertRC(rc); NOREF(rc);
236
237 RTR0Term();
238
239 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
240
241 Log(("VBoxNetFltLinuxUnload - done\n"));
242}
243
244
245/**
246 * We filter traffic from the host to the internal network
247 * before it reaches the NIC driver.
248 *
249 * The current code uses a very ugly hack overriding hard_start_xmit
250 * callback in the device structure, but it has been shown to give us a
251 * performance boost of 60-100% though. Eventually we have to find some
252 * less hacky way of getting this job done.
253 */
254#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
255
256#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
257
258# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
259
260# include <linux/ethtool.h>
261
262typedef struct ethtool_ops OVR_OPSTYPE;
263# define OVR_OPS ethtool_ops
264# define OVR_XMIT pfnStartXmit
265
266# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
267
268typedef struct net_device_ops OVR_OPSTYPE;
269# define OVR_OPS netdev_ops
270# define OVR_XMIT pOrgOps->ndo_start_xmit
271
272# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
273
274/**
275 * The overridden net_device_ops of the device we're attached to.
276 *
277 * As there is no net_device_ops structure in pre-2.6.29 kernels we override
278 * ethtool_ops instead along with hard_start_xmit callback in net_device
279 * structure.
280 *
281 * This is a very dirty hack that was created to explore how much we can improve
282 * the host to guest transfers by not CC'ing the NIC. It turns out to be
283 * the only way to filter outgoing packets for devices without TX queue.
284 */
285typedef struct VBoxNetDeviceOpsOverride
286{
287 /** Our overridden ops. */
288 OVR_OPSTYPE Ops;
289 /** Magic word. */
290 uint32_t u32Magic;
291 /** Pointer to the original ops. */
292 OVR_OPSTYPE const *pOrgOps;
293# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
294 /** Pointer to the original hard_start_xmit function. */
295 int (*pfnStartXmit)(struct sk_buff *pSkb, struct net_device *pDev);
296# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
297 /** Pointer to the net filter instance. */
298 PVBOXNETFLTINS pVBoxNetFlt;
299 /** The number of filtered packages. */
300 uint64_t cFiltered;
301 /** The total number of packets */
302 uint64_t cTotal;
303} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
304/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
305#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
306
307/**
308 * ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
309 * because they belong on the internal network.
310 *
311 * @returns NETDEV_TX_XXX.
312 * @param pSkb The socket buffer to transmit.
313 * @param pDev The net device.
314 */
315static int vboxNetFltLinuxStartXmitFilter(struct sk_buff *pSkb, struct net_device *pDev)
316{
317 PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
318 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
319 PCRTNETETHERHDR pEtherHdr;
320 PINTNETTRUNKSWPORT pSwitchPort;
321 uint32_t cbHdrs;
322
323
324 /*
325 * Validate the override structure.
326 *
327 * Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
328 * to be production quality code, we would have to be much more
329 * careful here and avoid the race.
330 */
331 if ( !VALID_PTR(pOverride)
332 || pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
333# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
334 || !VALID_PTR(pOverride->pOrgOps)
335# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
336 )
337 {
338 printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
339 dev_kfree_skb(pSkb);
340 return NETDEV_TX_OK;
341 }
342 pOverride->cTotal++;
343
344 /*
345 * Do the filtering base on the default OUI of our virtual NICs
346 *
347 * Note! In a real solution, we would ask the switch whether the
348 * destination MAC is 100% to be on the internal network and then
349 * drop it.
350 */
351 cbHdrs = skb_headlen(pSkb);
352 cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
353 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
354 if ( pEtherHdr
355 && VALID_PTR(pOverride->pVBoxNetFlt)
356 && (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
357 && VALID_PTR(pSwitchPort)
358 && cbHdrs >= 6)
359 {
360 INTNETSWDECISION enmDecision;
361
362 /** @todo consider reference counting, etc. */
363 enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
364 if (enmDecision == INTNETSWDECISION_INTNET)
365 {
366 dev_kfree_skb(pSkb);
367 pOverride->cFiltered++;
368 return NETDEV_TX_OK;
369 }
370 }
371
372 return pOverride->OVR_XMIT(pSkb, pDev);
373}
374
375/**
376 * Hooks the device ndo_start_xmit operation of the device.
377 *
378 * @param pThis The net filter instance.
379 * @param pDev The net device.
380 */
381static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
382{
383 PVBOXNETDEVICEOPSOVERRIDE pOverride;
384
385 /* Cancel override if ethtool_ops is missing (host-only case, @bugref{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 RTSpinlockAcquire(pThis->hSpinlock); /* (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);
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
422 RTSpinlockAcquire(pThis->hSpinlock);
423 if (!pDev)
424 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
425 if (VALID_PTR(pDev))
426 {
427 pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
428 if ( VALID_PTR(pOverride)
429 && pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
430 && VALID_PTR(pOverride->pOrgOps)
431 )
432 {
433# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
434 ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
435# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
436 ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
437 ASMAtomicWriteU32(&pOverride->u32Magic, 0);
438 }
439 else
440 pOverride = NULL;
441 }
442 else
443 pOverride = NULL;
444 RTSpinlockReleaseNoInts(pThis->hSpinlock);
445
446 if (pOverride)
447 {
448 printk("vboxnetflt: %llu out of %llu packets were not sent (directed to host)\n", pOverride->cFiltered, pOverride->cTotal);
449 RTMemFree(pOverride);
450 }
451}
452
453#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
454
455
456/**
457 * Reads and retains the host interface handle.
458 *
459 * @returns The handle, NULL if detached.
460 * @param pThis
461 */
462DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
463{
464#if 0
465 struct net_device *pDev = NULL;
466
467 Log(("vboxNetFltLinuxRetainNetDev\n"));
468 /*
469 * Be careful here to avoid problems racing the detached callback.
470 */
471 RTSpinlockAcquire(pThis->hSpinlock);
472 if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
473 {
474 pDev = (struct net_device *)ASMAtomicUoReadPtr((void * volatile *)&pThis->u.s.pDev);
475 if (pDev)
476 {
477 dev_hold(pDev);
478 Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
479 pDev, pDev->name,
480#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
481 netdev_refcnt_read(pDev)
482#else
483 atomic_read(&pDev->refcnt)
484#endif
485 ));
486 }
487 }
488 RTSpinlockRelease(pThis->hSpinlock);
489
490 Log(("vboxNetFltLinuxRetainNetDev - done\n"));
491 return pDev;
492#else
493 return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
494#endif
495}
496
497
498/**
499 * Release the host interface handle previously retained
500 * by vboxNetFltLinuxRetainNetDev.
501 *
502 * @param pThis The instance.
503 * @param pDev The vboxNetFltLinuxRetainNetDev
504 * return value, NULL is fine.
505 */
506DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
507{
508#if 0
509 Log(("vboxNetFltLinuxReleaseNetDev\n"));
510 NOREF(pThis);
511 if (pDev)
512 {
513 dev_put(pDev);
514 Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
515 pDev, pDev->name,
516#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
517 netdev_refcnt_read(pDev)
518#else
519 atomic_read(&pDev->refcnt)
520#endif
521 ));
522 }
523 Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
524#endif
525}
526
527#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 | (skb->dev->ifindex & 0xFFFF))
528#define VBOXNETFLT_SKB_TAG(skb) (*(uint32_t*)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
529
530/**
531 * Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
532 * i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
533 *
534 * @returns true / false accordingly.
535 * @param pBuf The sk_buff.
536 */
537DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
538{
539 return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
540}
541
542
543/**
544 * Internal worker that create a linux sk_buff for a
545 * (scatter/)gather list.
546 *
547 * @returns Pointer to the sk_buff.
548 * @param pThis The instance.
549 * @param pSG The (scatter/)gather list.
550 * @param fDstWire Set if the destination is the wire.
551 */
552static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
553{
554 struct sk_buff *pPkt;
555 struct net_device *pDev;
556 unsigned fGsoType = 0;
557
558 if (pSG->cbTotal == 0)
559 {
560 LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
561 return NULL;
562 }
563
564 /** @todo We should use fragments mapping the SG buffers with large packets.
565 * 256 bytes seems to be the a threshold used a lot for this. It
566 * requires some nasty work on the intnet side though... */
567 /*
568 * Allocate a packet and copy over the data.
569 */
570 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
571 pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
572 if (RT_UNLIKELY(!pPkt))
573 {
574 Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
575 pSG->pvUserData = NULL;
576 return NULL;
577 }
578 pPkt->dev = pDev;
579 pPkt->ip_summed = CHECKSUM_NONE;
580
581 /* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
582 skb_reserve(pPkt, NET_IP_ALIGN);
583
584 /* Copy the segments. */
585 skb_put(pPkt, pSG->cbTotal);
586 IntNetSgRead(pSG, pPkt->data);
587
588#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
589 /*
590 * Setup GSO if used by this packet.
591 */
592 switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
593 {
594 default:
595 AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
596 /* fall thru */
597 case PDMNETWORKGSOTYPE_INVALID:
598 fGsoType = 0;
599 break;
600 case PDMNETWORKGSOTYPE_IPV4_TCP:
601 fGsoType = SKB_GSO_TCPV4;
602 break;
603 case PDMNETWORKGSOTYPE_IPV4_UDP:
604 fGsoType = SKB_GSO_UDP;
605 break;
606 case PDMNETWORKGSOTYPE_IPV6_TCP:
607 fGsoType = SKB_GSO_TCPV6;
608 break;
609 }
610 if (fGsoType)
611 {
612 struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
613
614 pShInfo->gso_type = fGsoType | SKB_GSO_DODGY;
615 pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
616 pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
617
618 /*
619 * We need to set checksum fields even if the packet goes to the host
620 * directly as it may be immediately forwarded by IP layer @bugref{5020}.
621 */
622 Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
623 pPkt->ip_summed = CHECKSUM_PARTIAL;
624# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
625 pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
626 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
627 pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
628 else
629 pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
630# else
631 pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
632 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
633 pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
634 else
635 pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
636# endif
637 if (!fDstWire)
638 PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
639 }
640#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE || VBOXNETFLT_WITH_GSO_XMIT_HOST */
641
642 /*
643 * Finish up the socket buffer.
644 */
645 pPkt->protocol = eth_type_trans(pPkt, pDev);
646 if (fDstWire)
647 {
648 VBOX_SKB_RESET_NETWORK_HDR(pPkt);
649
650 /* Restore ethernet header back. */
651 skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
652 VBOX_SKB_RESET_MAC_HDR(pPkt);
653 }
654 VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
655
656 return pPkt;
657}
658
659
660/**
661 * Initializes a SG list from an sk_buff.
662 *
663 * @returns Number of segments.
664 * @param pThis The instance.
665 * @param pBuf The sk_buff.
666 * @param pSG The SG.
667 * @param pvFrame The frame pointer, optional.
668 * @param cSegs The number of segments allocated for the SG.
669 * This should match the number in the mbuf exactly!
670 * @param fSrc The source of the frame.
671 * @param pGso Pointer to the GSO context if it's a GSO
672 * internal network frame. NULL if regular frame.
673 */
674DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
675 unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
676{
677 int i;
678 NOREF(pThis);
679
680 Assert(!skb_shinfo(pBuf)->frag_list);
681
682 if (!pGsoCtx)
683 IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/);
684 else
685 IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/, pGsoCtx);
686
687#ifdef VBOXNETFLT_SG_SUPPORT
688 pSG->aSegs[0].cb = skb_headlen(pBuf);
689 pSG->aSegs[0].pv = pBuf->data;
690 pSG->aSegs[0].Phys = NIL_RTHCPHYS;
691
692 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
693 {
694 skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
695 pSG->aSegs[i+1].cb = pFrag->size;
696 pSG->aSegs[i+1].pv = kmap(pFrag->page);
697 printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
698 pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
699 }
700 ++i;
701
702#else
703 pSG->aSegs[0].cb = pBuf->len;
704 pSG->aSegs[0].pv = pBuf->data;
705 pSG->aSegs[0].Phys = NIL_RTHCPHYS;
706 i = 1;
707#endif
708
709 pSG->cSegsUsed = i;
710
711#ifdef PADD_RUNT_FRAMES_FROM_HOST
712 /*
713 * Add a trailer if the frame is too small.
714 *
715 * Since we're getting to the packet before it is framed, it has not
716 * yet been padded. The current solution is to add a segment pointing
717 * to a buffer containing all zeros and pray that works for all frames...
718 */
719 if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
720 {
721 static uint8_t const s_abZero[128] = {0};
722
723 AssertReturnVoid(i < cSegs);
724
725 pSG->aSegs[i].Phys = NIL_RTHCPHYS;
726 pSG->aSegs[i].pv = (void *)&s_abZero[0];
727 pSG->aSegs[i].cb = 60 - pSG->cbTotal;
728 pSG->cbTotal = 60;
729 pSG->cSegsUsed++;
730 Assert(i + 1 <= pSG->cSegsAlloc)
731 }
732#endif
733
734 Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
735 pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
736 for (i = 0; i < pSG->cSegsUsed; i++)
737 Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
738 i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
739}
740
741/**
742 * Packet handler,
743 *
744 * @returns 0 or EJUSTRETURN.
745 * @param pThis The instance.
746 * @param pMBuf The mbuf.
747 * @param pvFrame The start of the frame, optional.
748 * @param fSrc Where the packet (allegedly) comes from, one INTNETTRUNKDIR_* value.
749 * @param eProtocol The protocol.
750 */
751#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
752static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
753 struct net_device *pSkbDev,
754 struct packet_type *pPacketType,
755 struct net_device *pOrigDev)
756#else
757static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
758 struct net_device *pSkbDev,
759 struct packet_type *pPacketType)
760#endif
761{
762 PVBOXNETFLTINS pThis;
763 struct net_device *pDev;
764 LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
765 pBuf, pSkbDev, pPacketType));
766#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
767 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",
768 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));
769# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
770 Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
771# endif
772#else
773 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",
774 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));
775#endif
776 /*
777 * Drop it immediately?
778 */
779 if (!pBuf)
780 return 0;
781
782 if (pBuf->pkt_type == PACKET_LOOPBACK)
783 {
784 /*
785 * We are not interested in loopbacked packets as they will always have
786 * another copy going to the wire.
787 */
788 Log2(("vboxNetFltLinuxPacketHandler: dropped loopback packet (cb=%u)\n", pBuf->len));
789 dev_kfree_skb(pBuf); /* We must 'consume' all packets we get (@bugref{6539})! */
790 return 0;
791 }
792
793 pThis = VBOX_FLT_PT_TO_INST(pPacketType);
794 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
795 if (pDev != pSkbDev)
796 {
797 Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
798 kfree_skb(pBuf); /* This is a failure, so we use kfree_skb instead of dev_kfree_skb. */
799 return 0;
800 }
801
802 Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
803 if (vboxNetFltLinuxSkBufIsOur(pBuf))
804 {
805 Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
806 dev_kfree_skb(pBuf);
807 return 0;
808 }
809
810#ifndef VBOXNETFLT_SG_SUPPORT
811 {
812 /*
813 * Get rid of fragmented packets, they cause too much trouble.
814 */
815 unsigned int uMacLen = pBuf->mac_len;
816 struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
817 dev_kfree_skb(pBuf);
818 if (!pCopy)
819 {
820 LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
821 return 0;
822 }
823 pBuf = pCopy;
824 /* Somehow skb_copy ignores mac_len */
825 pBuf->mac_len = uMacLen;
826# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
827 /* Restore VLAN tag stripped by host hardware */
828 if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
829 {
830 uint8_t *pMac = (uint8_t*)skb_mac_header(pBuf);
831 struct vlan_ethhdr *pVHdr = (struct vlan_ethhdr *)(pMac - VLAN_HLEN);
832# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
833 memmove(pVHdr, pMac, ETH_ALEN * 2);
834# else
835 memmove(pVHdr, pMac, VLAN_ETH_ALEN * 2);
836# endif
837 pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
838 pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
839 pBuf->mac_header -= VLAN_HLEN;
840 pBuf->mac_len += VLAN_HLEN;
841 }
842# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
843
844# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
845 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",
846 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));
847# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
848 Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
849# endif
850# else
851 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",
852 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));
853# endif
854 }
855#endif
856
857#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
858 /* Forward it to the internal network. */
859 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
860#else
861 /* Add the packet to transmit queue and schedule the bottom half. */
862 skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
863 schedule_work(&pThis->u.s.XmitTask);
864 Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
865 &pThis->u.s.XmitTask, pBuf));
866#endif
867
868 /* It does not really matter what we return, it is ignored by the kernel. */
869 return 0;
870}
871
872/**
873 * Calculate the number of INTNETSEG segments the socket buffer will need.
874 *
875 * @returns Segment count.
876 * @param pBuf The socket buffer.
877 */
878DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
879{
880#ifdef VBOXNETFLT_SG_SUPPORT
881 unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
882#else
883 unsigned cSegs = 1;
884#endif
885#ifdef PADD_RUNT_FRAMES_FROM_HOST
886 /* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
887 if (pBuf->len < 60)
888 cSegs++;
889#endif
890 return cSegs;
891}
892
893/**
894 * Destroy the intnet scatter / gather buffer created by
895 * vboxNetFltLinuxSkBufToSG.
896 */
897static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
898{
899#ifdef VBOXNETFLT_SG_SUPPORT
900 int i;
901
902 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
903 {
904 printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
905 kunmap(pSG->aSegs[i+1].pv);
906 }
907#endif
908 NOREF(pSG);
909}
910
911#ifdef LOG_ENABLED
912/**
913 * Logging helper.
914 */
915static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
916{
917 int i, offSeg;
918 uint8_t *pInt, *pExt;
919 static int iPacketNo = 1;
920 iPacketNo += iIncrement;
921 if (fEgress)
922 {
923 pExt = pSG->aSegs[0].pv;
924 pInt = pExt + 6;
925 }
926 else
927 {
928 pInt = pSG->aSegs[0].pv;
929 pExt = pInt + 6;
930 }
931 Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
932 " %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
933 pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
934 fEgress ? "-->" : "<--", pszWhere,
935 pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
936 pSG->cbTotal, iPacketNo));
937 if (pSG->cSegsUsed == 1)
938 {
939 Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
940 }
941 else
942 {
943 for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
944 {
945 Log3(("-- segment %d at 0x%x (%d bytes) --\n%.*Rhxd\n",
946 i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
947 offSeg += pSG->aSegs[i].cb;
948 }
949 }
950
951}
952#else
953# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
954#endif
955
956#ifdef VBOXNETFLT_WITH_GSO_RECV
957
958/**
959 * Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
960 * GSO socket buffer without having to segment it.
961 *
962 * @returns true on success, false if needs segmenting.
963 * @param pThis The net filter instance.
964 * @param pSkb The GSO socket buffer.
965 * @param fSrc The source.
966 * @param pGsoCtx Where to return the GSO context on success.
967 */
968static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
969 PPDMNETWORKGSO pGsoCtx)
970{
971 PDMNETWORKGSOTYPE enmGsoType;
972 uint16_t uEtherType;
973 unsigned int cbTransport;
974 unsigned int offTransport;
975 unsigned int cbTransportHdr;
976 unsigned uProtocol;
977 union
978 {
979 RTNETIPV4 IPv4;
980 RTNETIPV6 IPv6;
981 RTNETTCP Tcp;
982 uint8_t ab[40];
983 uint16_t au16[40/2];
984 uint32_t au32[40/4];
985 } Buf;
986
987 /*
988 * Check the GSO properties of the socket buffer and make sure it fits.
989 */
990 /** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
991 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP | SKB_GSO_DODGY | SKB_GSO_TCPV6 | SKB_GSO_TCPV4) ))
992 {
993 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
994 return false;
995 }
996 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
997 || pSkb->len > VBOX_MAX_GSO_SIZE ))
998 {
999 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1000 return false;
1001 }
1002 /*
1003 * It is possible to receive GSO packets from wire if GRO is enabled.
1004 */
1005 if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1006 {
1007 Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1008#ifdef VBOXNETFLT_WITH_GRO
1009 /*
1010 * The packet came from the wire and the driver has already consumed
1011 * mac header. We need to restore it back.
1012 */
1013 pSkb->mac_len = skb_network_header(pSkb) - skb_mac_header(pSkb);
1014 skb_push(pSkb, pSkb->mac_len);
1015 Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1016 pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
1017#else /* !VBOXNETFLT_WITH_GRO */
1018 /* Older kernels didn't have GRO. */
1019 return false;
1020#endif /* !VBOXNETFLT_WITH_GRO */
1021 }
1022 else
1023 {
1024 /*
1025 * skb_gso_segment does the following. Do we need to do it as well?
1026 */
1027#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1028 skb_reset_mac_header(pSkb);
1029 pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1030#else
1031 pSkb->mac.raw = pSkb->data;
1032 pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1033#endif
1034 }
1035
1036 /*
1037 * Switch on the ethertype.
1038 */
1039 uEtherType = pSkb->protocol;
1040 if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1041 && pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1042 {
1043 uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1044 if (puEtherType)
1045 uEtherType = *puEtherType;
1046 }
1047 switch (uEtherType)
1048 {
1049 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1050 {
1051 unsigned int cbHdr;
1052 PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1053 if (RT_UNLIKELY(!pIPv4))
1054 {
1055 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1056 return false;
1057 }
1058
1059 cbHdr = pIPv4->ip_hl * 4;
1060 cbTransport = RT_N2H_U16(pIPv4->ip_len);
1061 if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1062 || cbHdr > cbTransport ))
1063 {
1064 Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1065 return false;
1066 }
1067 cbTransport -= cbHdr;
1068 offTransport = pSkb->mac_len + cbHdr;
1069 uProtocol = pIPv4->ip_p;
1070 if (uProtocol == RTNETIPV4_PROT_TCP)
1071 enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1072 else if (uProtocol == RTNETIPV4_PROT_UDP)
1073 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1074 else /** @todo IPv6: 4to6 tunneling */
1075 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1076 break;
1077 }
1078
1079 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1080 {
1081 PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1082 if (RT_UNLIKELY(!pIPv6))
1083 {
1084 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1085 return false;
1086 }
1087
1088 cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1089 offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1090 uProtocol = pIPv6->ip6_nxt;
1091 /** @todo IPv6: Dig our way out of the other headers. */
1092 if (uProtocol == RTNETIPV4_PROT_TCP)
1093 enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1094 else if (uProtocol == RTNETIPV4_PROT_UDP)
1095 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1096 else
1097 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1098 break;
1099 }
1100
1101 default:
1102 Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1103 return false;
1104 }
1105
1106 if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1107 {
1108 Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1109 return false;
1110 }
1111
1112 if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1113 || offTransport + cbTransport > pSkb->len
1114 || cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1115 {
1116 Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1117 offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1118 return false;
1119 }
1120
1121 /*
1122 * Check the TCP/UDP bits.
1123 */
1124 if (uProtocol == RTNETIPV4_PROT_TCP)
1125 {
1126 PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1127 if (RT_UNLIKELY(!pTcp))
1128 {
1129 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1130 return false;
1131 }
1132
1133 cbTransportHdr = pTcp->th_off * 4;
1134 pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1135 if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1136 || cbTransportHdr > cbTransport
1137 || offTransport + cbTransportHdr >= UINT8_MAX
1138 || offTransport + cbTransportHdr >= pSkb->len ))
1139 {
1140 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1141 return false;
1142 }
1143
1144 }
1145 else
1146 {
1147 Assert(uProtocol == RTNETIPV4_PROT_UDP);
1148 cbTransportHdr = sizeof(RTNETUDP);
1149 pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1150 if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1151 || offTransport + cbTransportHdr >= pSkb->len ))
1152 {
1153 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1154 return false;
1155 }
1156 }
1157
1158 /*
1159 * We're good, init the GSO context.
1160 */
1161 pGsoCtx->u8Type = enmGsoType;
1162 pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1163 pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1164 pGsoCtx->offHdr1 = pSkb->mac_len;
1165 pGsoCtx->offHdr2 = offTransport;
1166 pGsoCtx->u8Unused = 0;
1167
1168 return true;
1169}
1170
1171/**
1172 * Forward the socket buffer as a GSO internal network frame.
1173 *
1174 * @returns IPRT status code.
1175 * @param pThis The net filter instance.
1176 * @param pSkb The GSO socket buffer.
1177 * @param fSrc The source.
1178 * @param pGsoCtx Where to return the GSO context on success.
1179 */
1180static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1181{
1182 int rc;
1183 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1184 if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1185 {
1186 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1187 if (RT_LIKELY(pSG))
1188 {
1189 vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1190
1191 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1192 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1193
1194 vboxNetFltLinuxDestroySG(pSG);
1195 rc = VINF_SUCCESS;
1196 }
1197 else
1198 {
1199 Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1200 rc = VERR_NO_MEMORY;
1201 }
1202 }
1203 else
1204 {
1205 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1206 rc = VERR_INTERNAL_ERROR_3;
1207 }
1208
1209 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1210 dev_kfree_skb(pSkb);
1211 return rc;
1212}
1213
1214#endif /* VBOXNETFLT_WITH_GSO_RECV */
1215
1216/**
1217 * Worker for vboxNetFltLinuxForwardToIntNet.
1218 *
1219 * @returns VINF_SUCCESS or VERR_NO_MEMORY.
1220 * @param pThis The net filter instance.
1221 * @param pBuf The socket buffer.
1222 * @param fSrc The source.
1223 */
1224static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1225{
1226 int rc;
1227 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1228 if (cSegs <= MAX_SKB_FRAGS + 1)
1229 {
1230 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1231 if (RT_LIKELY(pSG))
1232 {
1233 if (fSrc & INTNETTRUNKDIR_WIRE)
1234 {
1235 /*
1236 * The packet came from wire, ethernet header was removed by device driver.
1237 * Restore it using mac_len field. This takes into account VLAN headers too.
1238 */
1239 skb_push(pBuf, pBuf->mac_len);
1240 }
1241
1242 vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /*pGsoCtx*/);
1243
1244 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1245 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1246
1247 vboxNetFltLinuxDestroySG(pSG);
1248 rc = VINF_SUCCESS;
1249 }
1250 else
1251 {
1252 Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1253 rc = VERR_NO_MEMORY;
1254 }
1255 }
1256 else
1257 {
1258 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1259 rc = VERR_INTERNAL_ERROR_3;
1260 }
1261
1262 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1263 dev_kfree_skb(pBuf);
1264 return rc;
1265}
1266
1267/**
1268 *
1269 * @param pBuf The socket buffer. This is consumed by this function.
1270 */
1271static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1272{
1273 uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1274
1275#ifdef VBOXNETFLT_WITH_GSO
1276 if (skb_is_gso(pBuf))
1277 {
1278 PDMNETWORKGSO GsoCtx;
1279 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",
1280 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));
1281# ifdef VBOXNETFLT_WITH_GSO_RECV
1282 if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP | SKB_GSO_TCPV6 | SKB_GSO_TCPV4))
1283 && vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1284 vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1285 else
1286# endif
1287 {
1288 /* Need to segment the packet */
1289 struct sk_buff *pNext;
1290 struct sk_buff *pSegment = skb_gso_segment(pBuf, 0 /*supported features*/);
1291 if (IS_ERR(pSegment))
1292 {
1293 dev_kfree_skb(pBuf);
1294 LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1295 return;
1296 }
1297
1298 for (; pSegment; pSegment = pNext)
1299 {
1300 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",
1301 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));
1302 pNext = pSegment->next;
1303 pSegment->next = 0;
1304 vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1305 }
1306 dev_kfree_skb(pBuf);
1307 }
1308 }
1309 else
1310#endif /* VBOXNETFLT_WITH_GSO */
1311 {
1312 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1313 {
1314#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1315 /*
1316 * Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1317 * and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1318 * header length from the header itself and reconstruct 'h' pointer
1319 * to TCP (or whatever) header.
1320 */
1321 unsigned char *tmp = pBuf->h.raw;
1322 if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1323 pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1324#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1325 if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1326 {
1327 LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1328 dev_kfree_skb(pBuf);
1329 return;
1330 }
1331#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1332 /* Restore the original (wrong) pointer. */
1333 pBuf->h.raw = tmp;
1334#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1335 }
1336 vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1337 }
1338}
1339
1340#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1341/**
1342 * Work queue handler that forwards the socket buffers queued by
1343 * vboxNetFltLinuxPacketHandler to the internal network.
1344 *
1345 * @param pWork The work queue.
1346 */
1347# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1348static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1349# else
1350static void vboxNetFltLinuxXmitTask(void *pWork)
1351# endif
1352{
1353 PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1354 struct sk_buff *pBuf;
1355
1356 Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1357
1358 /*
1359 * Active? Retain the instance and increment the busy counter.
1360 */
1361 if (vboxNetFltTryRetainBusyActive(pThis))
1362 {
1363 while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1364 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1365
1366 vboxNetFltRelease(pThis, true /* fBusy */);
1367 }
1368 else
1369 {
1370 /** @todo Shouldn't we just drop the packets here? There is little point in
1371 * making them accumulate when the VM is paused and it'll only waste
1372 * kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1373 * before start draining the packets (goes for the intnet ring buf
1374 * too)? */
1375 }
1376}
1377#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1378
1379/**
1380 * Reports the GSO capabilities of the hardware NIC.
1381 *
1382 * @param pThis The net filter instance. The caller hold a
1383 * reference to this.
1384 */
1385static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1386{
1387#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1388 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1389 {
1390 struct net_device *pDev;
1391 PINTNETTRUNKSWPORT pSwitchPort;
1392 unsigned int fFeatures;
1393
1394 RTSpinlockAcquire(pThis->hSpinlock);
1395
1396 pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1397 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1398 if (pDev)
1399 fFeatures = pDev->features;
1400 else
1401 fFeatures = 0;
1402
1403 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1404
1405 if (pThis->pSwitchPort)
1406 {
1407 /* Set/update the GSO capabilities of the NIC. */
1408 uint32_t fGsoCapabilites = 0;
1409 if (fFeatures & NETIF_F_TSO)
1410 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1411 if (fFeatures & NETIF_F_TSO6)
1412 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1413# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1414 if (fFeatures & NETIF_F_UFO)
1415 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1416 if (fFeatures & NETIF_F_UFO)
1417 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1418# endif
1419 Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1420 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1421 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : "",
1422 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)) ? "ufo " : "",
1423 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)) ? "ufo6 " : ""));
1424 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1425 }
1426
1427 vboxNetFltRelease(pThis, true /*fBusy*/);
1428 }
1429#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1430}
1431
1432/**
1433 * Helper that determines whether the host (ignoreing us) is operating the
1434 * interface in promiscuous mode or not.
1435 */
1436static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1437{
1438 bool fRc = false;
1439 struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1440 if (pDev)
1441 {
1442 fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1443 LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1444 fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1445 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1446 }
1447 return fRc;
1448}
1449
1450#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36)
1451/**
1452 * Helper for detecting TAP devices.
1453 */
1454static bool vboxNetFltIsTapDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1455{
1456 if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1457 {
1458 struct ethtool_drvinfo Info;
1459
1460 memset(&Info, 0, sizeof(Info));
1461 Info.cmd = ETHTOOL_GDRVINFO;
1462 pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1463 Log3(("vboxNetFltIsTapDevice: driver=%s version=%s bus_info=%s\n",
1464 Info.driver, Info.version, Info.bus_info));
1465
1466 return !strncmp(Info.driver, "tun", 4)
1467 && !strncmp(Info.bus_info, "tap", 4);
1468 }
1469
1470 return false;
1471}
1472
1473/**
1474 * Helper for updating the link state of TAP devices.
1475 * Only TAP devices are affected.
1476 */
1477static void vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1478{
1479 if (vboxNetFltIsTapDevice(pThis, pDev))
1480 {
1481 Log3(("vboxNetFltSetTapLinkState: bringing %s tap device link state\n",
1482 fLinkUp ? "up" : "down"));
1483 netif_tx_lock_bh(pDev);
1484 if (fLinkUp)
1485 netif_carrier_on(pDev);
1486 else
1487 netif_carrier_off(pDev);
1488 netif_tx_unlock_bh(pDev);
1489 }
1490}
1491#else /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1492DECLINLINE(void) vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1493{
1494 /* Nothing to do for pre-2.6.36 kernels. */
1495}
1496#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1497
1498/**
1499 * Internal worker for vboxNetFltLinuxNotifierCallback.
1500 *
1501 * @returns VBox status code.
1502 * @param pThis The instance.
1503 * @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1504 * flood the release log.
1505 */
1506static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1507{
1508 LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1509
1510 /*
1511 * Retain and store the device.
1512 */
1513 dev_hold(pDev);
1514
1515 RTSpinlockAcquire(pThis->hSpinlock);
1516 ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1517 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1518
1519 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1520 pDev, pDev->name,
1521#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1522 netdev_refcnt_read(pDev)
1523#else
1524 atomic_read(&pDev->refcnt)
1525#endif
1526 ));
1527 Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1528 pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1529
1530 /* Get the mac address while we still have a valid net_device reference. */
1531 memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1532
1533 /*
1534 * Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1535 */
1536 pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1537 pThis->u.s.PacketType.dev = pDev;
1538 pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1539 dev_add_pack(&pThis->u.s.PacketType);
1540 ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1541 Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1542
1543#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1544 vboxNetFltLinuxHookDev(pThis, pDev);
1545#endif
1546
1547 /*
1548 * If attaching to TAP interface we need to bring the link state up
1549 * starting from 2.6.36 kernel.
1550 */
1551 vboxNetFltSetTapLinkState(pThis, pDev, true);
1552
1553 /*
1554 * Set indicators that require the spinlock. Be abit paranoid about racing
1555 * the device notification handle.
1556 */
1557 RTSpinlockAcquire(pThis->hSpinlock);
1558 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1559 if (pDev)
1560 {
1561 ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1562 ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1563 pDev = NULL; /* don't dereference it */
1564 }
1565 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1566
1567 /*
1568 * If the above succeeded report GSO capabilities, if not undo and
1569 * release the device.
1570 */
1571 if (!pDev)
1572 {
1573 Assert(pThis->pSwitchPort);
1574 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1575 {
1576 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1577 pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1578 pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1579 pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE | INTNETTRUNKDIR_HOST);
1580 vboxNetFltRelease(pThis, true /*fBusy*/);
1581 }
1582 }
1583 else
1584 {
1585#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1586 vboxNetFltLinuxUnhookDev(pThis, pDev);
1587#endif
1588 RTSpinlockAcquire(pThis->hSpinlock);
1589 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1590 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1591 dev_put(pDev);
1592 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n",
1593 pDev, pDev->name,
1594#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1595 netdev_refcnt_read(pDev)
1596#else
1597 atomic_read(&pDev->refcnt)
1598#endif
1599 ));
1600 }
1601
1602 LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1603 return VINF_SUCCESS;
1604}
1605
1606
1607static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1608{
1609 bool fRegistered;
1610 Assert(!pThis->fDisconnectedFromHost);
1611
1612#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1613 vboxNetFltLinuxUnhookDev(pThis, pDev);
1614#endif
1615
1616 if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1617 {
1618 dev_remove_pack(&pThis->u.s.PacketType);
1619 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1620 }
1621
1622 RTSpinlockAcquire(pThis->hSpinlock);
1623 fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1624 if (fRegistered)
1625 {
1626 ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1627 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1628 }
1629 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1630
1631 if (fRegistered)
1632 {
1633#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1634 skb_queue_purge(&pThis->u.s.XmitQueue);
1635#endif
1636 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1637 Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1638 pDev, pDev->name,
1639#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1640 netdev_refcnt_read(pDev)
1641#else
1642 atomic_read(&pDev->refcnt)
1643#endif
1644 ));
1645 dev_put(pDev);
1646 }
1647
1648 return NOTIFY_OK;
1649}
1650
1651static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1652{
1653 /* Check if we are not suspended and promiscuous mode has not been set. */
1654 if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1655 && !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1656 {
1657 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1658 dev_set_promiscuity(pDev, 1);
1659 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1660 Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1661 }
1662 else
1663 Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1664 return NOTIFY_OK;
1665}
1666
1667static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1668{
1669 /* Undo promiscuous mode if we has set it. */
1670 if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1671 {
1672 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1673 dev_set_promiscuity(pDev, -1);
1674 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1675 Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1676 }
1677 else
1678 Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1679 return NOTIFY_OK;
1680}
1681
1682#ifdef LOG_ENABLED
1683/** Stringify the NETDEV_XXX constants. */
1684static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
1685{
1686 const char *pszEvent = "NETDRV_<unknown>";
1687 switch (ulEventType)
1688 {
1689 case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
1690 case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
1691 case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
1692 case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
1693 case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
1694 case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
1695 case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
1696 case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
1697 case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
1698 case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
1699# ifdef NETDEV_FEAT_CHANGE
1700 case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
1701# endif
1702 }
1703 return pszEvent;
1704}
1705#endif /* LOG_ENABLED */
1706
1707/**
1708 * Callback for listening to netdevice events.
1709 *
1710 * This works the rediscovery, clean up on unregistration, promiscuity on
1711 * up/down, and GSO feature changes from ethtool.
1712 *
1713 * @returns NOTIFY_OK
1714 * @param self Pointer to our notifier registration block.
1715 * @param ulEventType The event.
1716 * @param ptr Event specific, but it is usually the device it
1717 * relates to.
1718 */
1719static int vboxNetFltLinuxNotifierCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
1720
1721{
1722 PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
1723 struct net_device *pDev = (struct net_device *)ptr;
1724 int rc = NOTIFY_OK;
1725
1726 Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1727 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1728 if ( ulEventType == NETDEV_REGISTER
1729 && !strcmp(pDev->name, pThis->szName))
1730 {
1731 vboxNetFltLinuxAttachToInterface(pThis, pDev);
1732 }
1733 else
1734 {
1735 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1736 if (pDev == ptr)
1737 {
1738 switch (ulEventType)
1739 {
1740 case NETDEV_UNREGISTER:
1741 rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
1742 break;
1743 case NETDEV_UP:
1744 rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
1745 break;
1746 case NETDEV_GOING_DOWN:
1747 rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
1748 break;
1749 case NETDEV_CHANGENAME:
1750 break;
1751#ifdef NETDEV_FEAT_CHANGE
1752 case NETDEV_FEAT_CHANGE:
1753 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1754 break;
1755#endif
1756 }
1757 }
1758 }
1759
1760 return rc;
1761}
1762
1763bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
1764{
1765 return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
1766}
1767
1768int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
1769{
1770 struct net_device * pDev;
1771 int err;
1772 int rc = VINF_SUCCESS;
1773 NOREF(pvIfData);
1774
1775 LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
1776
1777 pDev = vboxNetFltLinuxRetainNetDev(pThis);
1778 if (pDev)
1779 {
1780 /*
1781 * Create a sk_buff for the gather list and push it onto the wire.
1782 */
1783 if (fDst & INTNETTRUNKDIR_WIRE)
1784 {
1785 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
1786 if (pBuf)
1787 {
1788 vboxNetFltDumpPacket(pSG, true, "wire", 1);
1789 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1790 Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
1791 err = dev_queue_xmit(pBuf);
1792 if (err)
1793 rc = RTErrConvertFromErrno(err);
1794 }
1795 else
1796 rc = VERR_NO_MEMORY;
1797 }
1798
1799 /*
1800 * Create a sk_buff for the gather list and push it onto the host stack.
1801 */
1802 if (fDst & INTNETTRUNKDIR_HOST)
1803 {
1804 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
1805 if (pBuf)
1806 {
1807 vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
1808 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1809 Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
1810 err = netif_rx_ni(pBuf);
1811 if (err)
1812 rc = RTErrConvertFromErrno(err);
1813 }
1814 else
1815 rc = VERR_NO_MEMORY;
1816 }
1817
1818 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1819 }
1820
1821 return rc;
1822}
1823
1824
1825void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
1826{
1827 struct net_device * pDev;
1828
1829 LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
1830 pThis, pThis->szName, fActive?"true":"false",
1831 pThis->fDisablePromiscuous?"true":"false"));
1832
1833 if (pThis->fDisablePromiscuous)
1834 return;
1835
1836 pDev = vboxNetFltLinuxRetainNetDev(pThis);
1837 if (pDev)
1838 {
1839 /*
1840 * This api is a bit weird, the best reference is the code.
1841 *
1842 * Also, we have a bit or race conditions wrt the maintenance of
1843 * host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
1844 */
1845#ifdef LOG_ENABLED
1846 u_int16_t fIf;
1847 unsigned const cPromiscBefore = pDev->promiscuity;
1848#endif
1849 if (fActive)
1850 {
1851 Assert(!pThis->u.s.fPromiscuousSet);
1852
1853 rtnl_lock();
1854 dev_set_promiscuity(pDev, 1);
1855 rtnl_unlock();
1856 pThis->u.s.fPromiscuousSet = true;
1857 Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1858 }
1859 else
1860 {
1861 if (pThis->u.s.fPromiscuousSet)
1862 {
1863 rtnl_lock();
1864 dev_set_promiscuity(pDev, -1);
1865 rtnl_unlock();
1866 Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1867 }
1868 pThis->u.s.fPromiscuousSet = false;
1869
1870#ifdef LOG_ENABLED
1871 fIf = dev_get_flags(pDev);
1872 Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
1873#endif
1874 }
1875
1876 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1877 }
1878}
1879
1880
1881int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
1882{
1883 /*
1884 * Remove packet handler when we get disconnected from internal switch as
1885 * we don't want the handler to forward packets to disconnected switch.
1886 */
1887 if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1888 {
1889 dev_remove_pack(&pThis->u.s.PacketType);
1890 Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
1891 }
1892 return VINF_SUCCESS;
1893}
1894
1895
1896int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
1897{
1898 /*
1899 * Report the GSO capabilities of the host and device (if connected).
1900 * Note! No need to mark ourselves busy here.
1901 */
1902 /** @todo duplicate work here now? Attach */
1903#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1904 Log3(("vboxNetFltOsConnectIt: reporting host tso tso6 ufo\n"));
1905 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
1906 0
1907 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
1908 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
1909 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
1910# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1911 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
1912# endif
1913 , INTNETTRUNKDIR_HOST);
1914
1915#endif
1916 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1917
1918 return VINF_SUCCESS;
1919}
1920
1921
1922void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
1923{
1924 struct net_device *pDev;
1925 bool fRegistered;
1926
1927#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1928 vboxNetFltLinuxUnhookDev(pThis, NULL);
1929#endif
1930
1931 /** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
1932 * unlikely, but none the less). Since it doesn't actually update the
1933 * state (just reads it), it is likely to panic in some interesting
1934 * ways. */
1935
1936 RTSpinlockAcquire(pThis->hSpinlock);
1937 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1938 fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1939 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1940
1941 if (fRegistered)
1942 {
1943 vboxNetFltSetTapLinkState(pThis, pDev, false);
1944
1945#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1946 skb_queue_purge(&pThis->u.s.XmitQueue);
1947#endif
1948 Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
1949 Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
1950 pDev, pDev->name,
1951#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1952 netdev_refcnt_read(pDev)
1953#else
1954 atomic_read(&pDev->refcnt)
1955#endif
1956 ));
1957 dev_put(pDev);
1958 }
1959 Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
1960 unregister_netdevice_notifier(&pThis->u.s.Notifier);
1961 module_put(THIS_MODULE);
1962}
1963
1964
1965int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
1966{
1967 int err;
1968 NOREF(pvContext);
1969
1970 pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
1971 err = register_netdevice_notifier(&pThis->u.s.Notifier);
1972 if (err)
1973 return VERR_INTNET_FLT_IF_FAILED;
1974 if (!pThis->u.s.fRegistered)
1975 {
1976 unregister_netdevice_notifier(&pThis->u.s.Notifier);
1977 LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
1978 return VERR_INTNET_FLT_IF_NOT_FOUND;
1979 }
1980
1981 Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
1982 if ( pThis->fDisconnectedFromHost
1983 || !try_module_get(THIS_MODULE))
1984 return VERR_INTNET_FLT_IF_FAILED;
1985
1986 return VINF_SUCCESS;
1987}
1988
1989int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
1990{
1991 /*
1992 * Init the linux specific members.
1993 */
1994 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1995 pThis->u.s.fRegistered = false;
1996 pThis->u.s.fPromiscuousSet = false;
1997 pThis->u.s.fPacketHandler = false;
1998 memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
1999#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2000 skb_queue_head_init(&pThis->u.s.XmitQueue);
2001# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2002 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2003# else
2004 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2005# endif
2006#endif
2007
2008 return VINF_SUCCESS;
2009}
2010
2011
2012void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2013{
2014 NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2015}
2016
2017
2018int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void *pvIf, void **pvIfData)
2019{
2020 /* Nothing to do */
2021 NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2022 return VINF_SUCCESS;
2023}
2024
2025
2026int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2027{
2028 /* Nothing to do */
2029 NOREF(pThis); NOREF(pvIfData);
2030 return VINF_SUCCESS;
2031}
2032
Note: See TracBrowser for help on using the repository browser.

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