VirtualBox

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 41548

Last change on this file since 41548 was 41548, checked in by vboxsync, 13 years ago

netflt: fix dev unregistration issues on kernels 3.2.18 and newer (#6225)
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 70.0 KB
Line 
1/* $Id: VBoxNetFlt-linux.c 41548 2012-06-01 16:46:34Z 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#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, #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: dropped %llu out of %llu packets\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 return 0;
790 }
791
792 pThis = VBOX_FLT_PT_TO_INST(pPacketType);
793 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
794 if (pDev != pSkbDev)
795 {
796 Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
797 return 0;
798 }
799
800 Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
801 if (vboxNetFltLinuxSkBufIsOur(pBuf))
802 {
803 Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
804 dev_kfree_skb(pBuf);
805 return 0;
806 }
807
808#ifndef VBOXNETFLT_SG_SUPPORT
809 {
810 /*
811 * Get rid of fragmented packets, they cause too much trouble.
812 */
813 unsigned int uMacLen = pBuf->mac_len;
814 struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
815 kfree_skb(pBuf);
816 if (!pCopy)
817 {
818 LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
819 return 0;
820 }
821 pBuf = pCopy;
822 /* Somehow skb_copy ignores mac_len */
823 pBuf->mac_len = uMacLen;
824# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
825 /* Restore VLAN tag stripped by host hardware */
826 if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
827 {
828 uint8_t *pMac = (uint8_t*)skb_mac_header(pBuf);
829 struct vlan_ethhdr *pVHdr = (struct vlan_ethhdr *)(pMac - VLAN_HLEN);
830# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
831 memmove(pVHdr, pMac, ETH_ALEN * 2);
832# else
833 memmove(pVHdr, pMac, VLAN_ETH_ALEN * 2);
834# endif
835 pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
836 pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
837 pBuf->mac_header -= VLAN_HLEN;
838 pBuf->mac_len += VLAN_HLEN;
839 }
840# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
841
842# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
843 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",
844 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));
845# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
846 Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
847# endif
848# else
849 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",
850 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));
851# endif
852 }
853#endif
854
855#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
856 /* Forward it to the internal network. */
857 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
858#else
859 /* Add the packet to transmit queue and schedule the bottom half. */
860 skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
861 schedule_work(&pThis->u.s.XmitTask);
862 Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
863 &pThis->u.s.XmitTask, pBuf));
864#endif
865
866 /* It does not really matter what we return, it is ignored by the kernel. */
867 return 0;
868}
869
870/**
871 * Calculate the number of INTNETSEG segments the socket buffer will need.
872 *
873 * @returns Segment count.
874 * @param pBuf The socket buffer.
875 */
876DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
877{
878#ifdef VBOXNETFLT_SG_SUPPORT
879 unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
880#else
881 unsigned cSegs = 1;
882#endif
883#ifdef PADD_RUNT_FRAMES_FROM_HOST
884 /* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
885 if (pBuf->len < 60)
886 cSegs++;
887#endif
888 return cSegs;
889}
890
891/**
892 * Destroy the intnet scatter / gather buffer created by
893 * vboxNetFltLinuxSkBufToSG.
894 */
895static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
896{
897#ifdef VBOXNETFLT_SG_SUPPORT
898 int i;
899
900 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
901 {
902 printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
903 kunmap(pSG->aSegs[i+1].pv);
904 }
905#endif
906 NOREF(pSG);
907}
908
909#ifdef LOG_ENABLED
910/**
911 * Logging helper.
912 */
913static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
914{
915 int i, offSeg;
916 uint8_t *pInt, *pExt;
917 static int iPacketNo = 1;
918 iPacketNo += iIncrement;
919 if (fEgress)
920 {
921 pExt = pSG->aSegs[0].pv;
922 pInt = pExt + 6;
923 }
924 else
925 {
926 pInt = pSG->aSegs[0].pv;
927 pExt = pInt + 6;
928 }
929 Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
930 " %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
931 pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
932 fEgress ? "-->" : "<--", pszWhere,
933 pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
934 pSG->cbTotal, iPacketNo));
935 if (pSG->cSegsUsed == 1)
936 {
937 Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
938 }
939 else
940 {
941 for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
942 {
943 Log3(("-- segment %d at 0x%x (%d bytes) --\n%.*Rhxd\n",
944 i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
945 offSeg += pSG->aSegs[i].cb;
946 }
947 }
948
949}
950#else
951# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
952#endif
953
954#ifdef VBOXNETFLT_WITH_GSO_RECV
955
956/**
957 * Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
958 * GSO socket buffer without having to segment it.
959 *
960 * @returns true on success, false if needs segmenting.
961 * @param pThis The net filter instance.
962 * @param pSkb The GSO socket buffer.
963 * @param fSrc The source.
964 * @param pGsoCtx Where to return the GSO context on success.
965 */
966static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
967 PPDMNETWORKGSO pGsoCtx)
968{
969 PDMNETWORKGSOTYPE enmGsoType;
970 uint16_t uEtherType;
971 unsigned int cbTransport;
972 unsigned int offTransport;
973 unsigned int cbTransportHdr;
974 unsigned uProtocol;
975 union
976 {
977 RTNETIPV4 IPv4;
978 RTNETIPV6 IPv6;
979 RTNETTCP Tcp;
980 uint8_t ab[40];
981 uint16_t au16[40/2];
982 uint32_t au32[40/4];
983 } Buf;
984
985 /*
986 * Check the GSO properties of the socket buffer and make sure it fits.
987 */
988 /** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
989 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP | SKB_GSO_DODGY | SKB_GSO_TCPV6 | SKB_GSO_TCPV4) ))
990 {
991 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
992 return false;
993 }
994 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
995 || pSkb->len > VBOX_MAX_GSO_SIZE ))
996 {
997 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
998 return false;
999 }
1000 /*
1001 * It is possible to receive GSO packets from wire if GRO is enabled.
1002 */
1003 if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1004 {
1005 Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1006#ifdef VBOXNETFLT_WITH_GRO
1007 /*
1008 * The packet came from the wire and the driver has already consumed
1009 * mac header. We need to restore it back.
1010 */
1011 pSkb->mac_len = skb_network_header(pSkb) - skb_mac_header(pSkb);
1012 skb_push(pSkb, pSkb->mac_len);
1013 Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1014 pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
1015#else /* !VBOXNETFLT_WITH_GRO */
1016 /* Older kernels didn't have GRO. */
1017 return false;
1018#endif /* !VBOXNETFLT_WITH_GRO */
1019 }
1020 else
1021 {
1022 /*
1023 * skb_gso_segment does the following. Do we need to do it as well?
1024 */
1025#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1026 skb_reset_mac_header(pSkb);
1027 pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1028#else
1029 pSkb->mac.raw = pSkb->data;
1030 pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1031#endif
1032 }
1033
1034 /*
1035 * Switch on the ethertype.
1036 */
1037 uEtherType = pSkb->protocol;
1038 if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1039 && pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1040 {
1041 uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1042 if (puEtherType)
1043 uEtherType = *puEtherType;
1044 }
1045 switch (uEtherType)
1046 {
1047 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1048 {
1049 unsigned int cbHdr;
1050 PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1051 if (RT_UNLIKELY(!pIPv4))
1052 {
1053 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1054 return false;
1055 }
1056
1057 cbHdr = pIPv4->ip_hl * 4;
1058 cbTransport = RT_N2H_U16(pIPv4->ip_len);
1059 if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1060 || cbHdr > cbTransport ))
1061 {
1062 Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1063 return false;
1064 }
1065 cbTransport -= cbHdr;
1066 offTransport = pSkb->mac_len + cbHdr;
1067 uProtocol = pIPv4->ip_p;
1068 if (uProtocol == RTNETIPV4_PROT_TCP)
1069 enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1070 else if (uProtocol == RTNETIPV4_PROT_UDP)
1071 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1072 else /** @todo IPv6: 4to6 tunneling */
1073 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1074 break;
1075 }
1076
1077 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1078 {
1079 PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1080 if (RT_UNLIKELY(!pIPv6))
1081 {
1082 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1083 return false;
1084 }
1085
1086 cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1087 offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1088 uProtocol = pIPv6->ip6_nxt;
1089 /** @todo IPv6: Dig our way out of the other headers. */
1090 if (uProtocol == RTNETIPV4_PROT_TCP)
1091 enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1092 else if (uProtocol == RTNETIPV4_PROT_UDP)
1093 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1094 else
1095 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1096 break;
1097 }
1098
1099 default:
1100 Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1101 return false;
1102 }
1103
1104 if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1105 {
1106 Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1107 return false;
1108 }
1109
1110 if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1111 || offTransport + cbTransport > pSkb->len
1112 || cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1113 {
1114 Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1115 offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1116 return false;
1117 }
1118
1119 /*
1120 * Check the TCP/UDP bits.
1121 */
1122 if (uProtocol == RTNETIPV4_PROT_TCP)
1123 {
1124 PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1125 if (RT_UNLIKELY(!pTcp))
1126 {
1127 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1128 return false;
1129 }
1130
1131 cbTransportHdr = pTcp->th_off * 4;
1132 pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1133 if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1134 || cbTransportHdr > cbTransport
1135 || offTransport + cbTransportHdr >= UINT8_MAX
1136 || offTransport + cbTransportHdr >= pSkb->len ))
1137 {
1138 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1139 return false;
1140 }
1141
1142 }
1143 else
1144 {
1145 Assert(uProtocol == RTNETIPV4_PROT_UDP);
1146 cbTransportHdr = sizeof(RTNETUDP);
1147 pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1148 if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1149 || offTransport + cbTransportHdr >= pSkb->len ))
1150 {
1151 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1152 return false;
1153 }
1154 }
1155
1156 /*
1157 * We're good, init the GSO context.
1158 */
1159 pGsoCtx->u8Type = enmGsoType;
1160 pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1161 pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1162 pGsoCtx->offHdr1 = pSkb->mac_len;
1163 pGsoCtx->offHdr2 = offTransport;
1164 pGsoCtx->u8Unused = 0;
1165
1166 return true;
1167}
1168
1169/**
1170 * Forward the socket buffer as a GSO internal network frame.
1171 *
1172 * @returns IPRT status code.
1173 * @param pThis The net filter instance.
1174 * @param pSkb The GSO socket buffer.
1175 * @param fSrc The source.
1176 * @param pGsoCtx Where to return the GSO context on success.
1177 */
1178static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1179{
1180 int rc;
1181 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1182 if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1183 {
1184 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1185 if (RT_LIKELY(pSG))
1186 {
1187 vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1188
1189 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1190 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1191
1192 vboxNetFltLinuxDestroySG(pSG);
1193 rc = VINF_SUCCESS;
1194 }
1195 else
1196 {
1197 Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1198 rc = VERR_NO_MEMORY;
1199 }
1200 }
1201 else
1202 {
1203 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1204 rc = VERR_INTERNAL_ERROR_3;
1205 }
1206
1207 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1208 dev_kfree_skb(pSkb);
1209 return rc;
1210}
1211
1212#endif /* VBOXNETFLT_WITH_GSO_RECV */
1213
1214/**
1215 * Worker for vboxNetFltLinuxForwardToIntNet.
1216 *
1217 * @returns VINF_SUCCESS or VERR_NO_MEMORY.
1218 * @param pThis The net filter instance.
1219 * @param pBuf The socket buffer.
1220 * @param fSrc The source.
1221 */
1222static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1223{
1224 int rc;
1225 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1226 if (cSegs <= MAX_SKB_FRAGS + 1)
1227 {
1228 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1229 if (RT_LIKELY(pSG))
1230 {
1231 if (fSrc & INTNETTRUNKDIR_WIRE)
1232 {
1233 /*
1234 * The packet came from wire, ethernet header was removed by device driver.
1235 * Restore it using mac_len field. This takes into account VLAN headers too.
1236 */
1237 skb_push(pBuf, pBuf->mac_len);
1238 }
1239
1240 vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /*pGsoCtx*/);
1241
1242 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1243 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1244
1245 vboxNetFltLinuxDestroySG(pSG);
1246 rc = VINF_SUCCESS;
1247 }
1248 else
1249 {
1250 Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1251 rc = VERR_NO_MEMORY;
1252 }
1253 }
1254 else
1255 {
1256 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1257 rc = VERR_INTERNAL_ERROR_3;
1258 }
1259
1260 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1261 dev_kfree_skb(pBuf);
1262 return rc;
1263}
1264
1265/**
1266 *
1267 * @param pBuf The socket buffer. This is consumed by this function.
1268 */
1269static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1270{
1271 uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1272
1273#ifdef VBOXNETFLT_WITH_GSO
1274 if (skb_is_gso(pBuf))
1275 {
1276 PDMNETWORKGSO GsoCtx;
1277 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",
1278 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));
1279# ifdef VBOXNETFLT_WITH_GSO_RECV
1280 if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP | SKB_GSO_TCPV6 | SKB_GSO_TCPV4))
1281 && vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1282 vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1283 else
1284# endif
1285 {
1286 /* Need to segment the packet */
1287 struct sk_buff *pNext;
1288 struct sk_buff *pSegment = skb_gso_segment(pBuf, 0 /*supported features*/);
1289 if (IS_ERR(pSegment))
1290 {
1291 dev_kfree_skb(pBuf);
1292 LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1293 return;
1294 }
1295
1296 for (; pSegment; pSegment = pNext)
1297 {
1298 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",
1299 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));
1300 pNext = pSegment->next;
1301 pSegment->next = 0;
1302 vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1303 }
1304 dev_kfree_skb(pBuf);
1305 }
1306 }
1307 else
1308#endif /* VBOXNETFLT_WITH_GSO */
1309 {
1310 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1311 {
1312#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1313 /*
1314 * Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1315 * and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1316 * header length from the header itself and reconstruct 'h' pointer
1317 * to TCP (or whatever) header.
1318 */
1319 unsigned char *tmp = pBuf->h.raw;
1320 if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1321 pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1322#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1323 if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1324 {
1325 LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1326 dev_kfree_skb(pBuf);
1327 return;
1328 }
1329#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1330 /* Restore the original (wrong) pointer. */
1331 pBuf->h.raw = tmp;
1332#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1333 }
1334 vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1335 }
1336}
1337
1338#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1339/**
1340 * Work queue handler that forwards the socket buffers queued by
1341 * vboxNetFltLinuxPacketHandler to the internal network.
1342 *
1343 * @param pWork The work queue.
1344 */
1345# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1346static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1347# else
1348static void vboxNetFltLinuxXmitTask(void *pWork)
1349# endif
1350{
1351 PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1352 struct sk_buff *pBuf;
1353
1354 Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1355
1356 /*
1357 * Active? Retain the instance and increment the busy counter.
1358 */
1359 if (vboxNetFltTryRetainBusyActive(pThis))
1360 {
1361 while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1362 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1363
1364 vboxNetFltRelease(pThis, true /* fBusy */);
1365 }
1366 else
1367 {
1368 /** @todo Shouldn't we just drop the packets here? There is little point in
1369 * making them accumulate when the VM is paused and it'll only waste
1370 * kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1371 * before start draining the packets (goes for the intnet ring buf
1372 * too)? */
1373 }
1374}
1375#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1376
1377/**
1378 * Reports the GSO capabilities of the hardware NIC.
1379 *
1380 * @param pThis The net filter instance. The caller hold a
1381 * reference to this.
1382 */
1383static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1384{
1385#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1386 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1387 {
1388 struct net_device *pDev;
1389 PINTNETTRUNKSWPORT pSwitchPort;
1390 unsigned int fFeatures;
1391
1392 RTSpinlockAcquire(pThis->hSpinlock);
1393
1394 pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1395 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1396 if (pDev)
1397 fFeatures = pDev->features;
1398 else
1399 fFeatures = 0;
1400
1401 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1402
1403 if (pThis->pSwitchPort)
1404 {
1405 /* Set/update the GSO capabilities of the NIC. */
1406 uint32_t fGsoCapabilites = 0;
1407 if (fFeatures & NETIF_F_TSO)
1408 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1409 if (fFeatures & NETIF_F_TSO6)
1410 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1411# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1412 if (fFeatures & NETIF_F_UFO)
1413 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1414 if (fFeatures & NETIF_F_UFO)
1415 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1416# endif
1417 Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1418 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1419 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : "",
1420 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)) ? "ufo " : "",
1421 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)) ? "ufo6 " : ""));
1422 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1423 }
1424
1425 vboxNetFltRelease(pThis, true /*fBusy*/);
1426 }
1427#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1428}
1429
1430/**
1431 * Helper that determines whether the host (ignoreing us) is operating the
1432 * interface in promiscuous mode or not.
1433 */
1434static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1435{
1436 bool fRc = false;
1437 struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1438 if (pDev)
1439 {
1440 fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1441 LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1442 fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1443 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1444 }
1445 return fRc;
1446}
1447
1448#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36)
1449/**
1450 * Helper for detecting TAP devices.
1451 */
1452static bool vboxNetFltIsTapDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1453{
1454 if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1455 {
1456 struct ethtool_drvinfo Info;
1457
1458 memset(&Info, 0, sizeof(Info));
1459 Info.cmd = ETHTOOL_GDRVINFO;
1460 pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1461 Log3(("vboxNetFltIsTapDevice: driver=%s version=%s bus_info=%s\n",
1462 Info.driver, Info.version, Info.bus_info));
1463
1464 return !strncmp(Info.driver, "tun", 4)
1465 && !strncmp(Info.bus_info, "tap", 4);
1466 }
1467
1468 return false;
1469}
1470
1471/**
1472 * Helper for updating the link state of TAP devices.
1473 * Only TAP devices are affected.
1474 */
1475static void vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1476{
1477 if (vboxNetFltIsTapDevice(pThis, pDev))
1478 {
1479 Log3(("vboxNetFltSetTapLinkState: bringing %s tap device link state\n",
1480 fLinkUp ? "up" : "down"));
1481 netif_tx_lock_bh(pDev);
1482 if (fLinkUp)
1483 netif_carrier_on(pDev);
1484 else
1485 netif_carrier_off(pDev);
1486 netif_tx_unlock_bh(pDev);
1487 }
1488}
1489#else /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1490DECLINLINE(void) vboxNetFltSetTapLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1491{
1492 /* Nothing to do for pre-2.6.36 kernels. */
1493}
1494#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36) */
1495
1496/**
1497 * Internal worker for vboxNetFltLinuxNotifierCallback.
1498 *
1499 * @returns VBox status code.
1500 * @param pThis The instance.
1501 * @param fRediscovery If set we're doing a rediscovery attempt, so, don't
1502 * flood the release log.
1503 */
1504static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1505{
1506 LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1507
1508 /*
1509 * Retain and store the device.
1510 */
1511 dev_hold(pDev);
1512
1513 RTSpinlockAcquire(pThis->hSpinlock);
1514 ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1515 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1516
1517 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1518 pDev, pDev->name,
1519#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1520 netdev_refcnt_read(pDev)
1521#else
1522 atomic_read(&pDev->refcnt)
1523#endif
1524 ));
1525 Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1526 pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1527
1528 /* Get the mac address while we still have a valid net_device reference. */
1529 memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1530
1531 /*
1532 * Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1533 */
1534 pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1535 pThis->u.s.PacketType.dev = pDev;
1536 pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1537 dev_add_pack(&pThis->u.s.PacketType);
1538 ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1539 Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1540
1541#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1542 vboxNetFltLinuxHookDev(pThis, pDev);
1543#endif
1544
1545 /*
1546 * If attaching to TAP interface we need to bring the link state up
1547 * starting from 2.6.36 kernel.
1548 */
1549 vboxNetFltSetTapLinkState(pThis, pDev, true);
1550
1551 /*
1552 * Set indicators that require the spinlock. Be abit paranoid about racing
1553 * the device notification handle.
1554 */
1555 RTSpinlockAcquire(pThis->hSpinlock);
1556 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1557 if (pDev)
1558 {
1559 ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1560 ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1561 pDev = NULL; /* don't dereference it */
1562 }
1563 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1564
1565 /*
1566 * If the above succeeded report GSO capabilities, if not undo and
1567 * release the device.
1568 */
1569 if (!pDev)
1570 {
1571 Assert(pThis->pSwitchPort);
1572 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1573 {
1574 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1575 pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1576 pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1577 pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE | INTNETTRUNKDIR_HOST);
1578 vboxNetFltRelease(pThis, true /*fBusy*/);
1579 }
1580 }
1581 else
1582 {
1583#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1584 vboxNetFltLinuxUnhookDev(pThis, pDev);
1585#endif
1586 RTSpinlockAcquire(pThis->hSpinlock);
1587 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1588 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1589 dev_put(pDev);
1590 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n",
1591 pDev, pDev->name,
1592#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1593 netdev_refcnt_read(pDev)
1594#else
1595 atomic_read(&pDev->refcnt)
1596#endif
1597 ));
1598 }
1599
1600 LogRel(("VBoxNetFlt: attached to '%s' / %.*Rhxs\n", pThis->szName, sizeof(pThis->u.s.MacAddr), &pThis->u.s.MacAddr));
1601 return VINF_SUCCESS;
1602}
1603
1604
1605static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1606{
1607 bool fRegistered;
1608 Assert(!pThis->fDisconnectedFromHost);
1609
1610#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1611 vboxNetFltLinuxUnhookDev(pThis, pDev);
1612#endif
1613
1614 if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1615 {
1616 dev_remove_pack(&pThis->u.s.PacketType);
1617 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1618 }
1619
1620 RTSpinlockAcquire(pThis->hSpinlock);
1621 fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1622 if (fRegistered)
1623 {
1624 ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1625 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1626 }
1627 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1628
1629 if (fRegistered)
1630 {
1631#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1632 skb_queue_purge(&pThis->u.s.XmitQueue);
1633#endif
1634 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1635 Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1636 pDev, pDev->name,
1637#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1638 netdev_refcnt_read(pDev)
1639#else
1640 atomic_read(&pDev->refcnt)
1641#endif
1642 ));
1643 dev_put(pDev);
1644 }
1645
1646 return NOTIFY_OK;
1647}
1648
1649static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1650{
1651 /* Check if we are not suspended and promiscuous mode has not been set. */
1652 if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1653 && !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1654 {
1655 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1656 dev_set_promiscuity(pDev, 1);
1657 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1658 Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1659 }
1660 else
1661 Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1662 return NOTIFY_OK;
1663}
1664
1665static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1666{
1667 /* Undo promiscuous mode if we has set it. */
1668 if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1669 {
1670 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1671 dev_set_promiscuity(pDev, -1);
1672 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1673 Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1674 }
1675 else
1676 Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1677 return NOTIFY_OK;
1678}
1679
1680#ifdef LOG_ENABLED
1681/** Stringify the NETDEV_XXX constants. */
1682static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
1683{
1684 const char *pszEvent = "NETDRV_<unknown>";
1685 switch (ulEventType)
1686 {
1687 case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
1688 case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
1689 case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
1690 case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
1691 case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
1692 case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
1693 case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
1694 case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
1695 case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
1696 case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
1697# ifdef NETDEV_FEAT_CHANGE
1698 case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
1699# endif
1700 }
1701 return pszEvent;
1702}
1703#endif /* LOG_ENABLED */
1704
1705/**
1706 * Callback for listening to netdevice events.
1707 *
1708 * This works the rediscovery, clean up on unregistration, promiscuity on
1709 * up/down, and GSO feature changes from ethtool.
1710 *
1711 * @returns NOTIFY_OK
1712 * @param self Pointer to our notifier registration block.
1713 * @param ulEventType The event.
1714 * @param ptr Event specific, but it is usually the device it
1715 * relates to.
1716 */
1717static int vboxNetFltLinuxNotifierCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
1718
1719{
1720 PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
1721 struct net_device *pDev = (struct net_device *)ptr;
1722 int rc = NOTIFY_OK;
1723
1724 Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1725 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1726 if ( ulEventType == NETDEV_REGISTER
1727 && !strcmp(pDev->name, pThis->szName))
1728 {
1729 vboxNetFltLinuxAttachToInterface(pThis, pDev);
1730 }
1731 else
1732 {
1733 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1734 if (pDev == ptr)
1735 {
1736 switch (ulEventType)
1737 {
1738 case NETDEV_UNREGISTER:
1739 rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
1740 break;
1741 case NETDEV_UP:
1742 rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
1743 break;
1744 case NETDEV_GOING_DOWN:
1745 rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
1746 break;
1747 case NETDEV_CHANGENAME:
1748 break;
1749#ifdef NETDEV_FEAT_CHANGE
1750 case NETDEV_FEAT_CHANGE:
1751 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1752 break;
1753#endif
1754 }
1755 }
1756 }
1757
1758 return rc;
1759}
1760
1761bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
1762{
1763 return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
1764}
1765
1766int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
1767{
1768 struct net_device * pDev;
1769 int err;
1770 int rc = VINF_SUCCESS;
1771 NOREF(pvIfData);
1772
1773 LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
1774
1775 pDev = vboxNetFltLinuxRetainNetDev(pThis);
1776 if (pDev)
1777 {
1778 /*
1779 * Create a sk_buff for the gather list and push it onto the wire.
1780 */
1781 if (fDst & INTNETTRUNKDIR_WIRE)
1782 {
1783 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
1784 if (pBuf)
1785 {
1786 vboxNetFltDumpPacket(pSG, true, "wire", 1);
1787 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1788 Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
1789 err = dev_queue_xmit(pBuf);
1790 if (err)
1791 rc = RTErrConvertFromErrno(err);
1792 }
1793 else
1794 rc = VERR_NO_MEMORY;
1795 }
1796
1797 /*
1798 * Create a sk_buff for the gather list and push it onto the host stack.
1799 */
1800 if (fDst & INTNETTRUNKDIR_HOST)
1801 {
1802 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
1803 if (pBuf)
1804 {
1805 vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
1806 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1807 Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
1808 err = netif_rx_ni(pBuf);
1809 if (err)
1810 rc = RTErrConvertFromErrno(err);
1811 }
1812 else
1813 rc = VERR_NO_MEMORY;
1814 }
1815
1816 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1817 }
1818
1819 return rc;
1820}
1821
1822
1823void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
1824{
1825 struct net_device * pDev;
1826
1827 LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%s, fDisablePromiscuous=%s\n",
1828 pThis, pThis->szName, fActive?"true":"false",
1829 pThis->fDisablePromiscuous?"true":"false"));
1830
1831 if (pThis->fDisablePromiscuous)
1832 return;
1833
1834 pDev = vboxNetFltLinuxRetainNetDev(pThis);
1835 if (pDev)
1836 {
1837 /*
1838 * This api is a bit weird, the best reference is the code.
1839 *
1840 * Also, we have a bit or race conditions wrt the maintenance of
1841 * host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
1842 */
1843#ifdef LOG_ENABLED
1844 u_int16_t fIf;
1845 unsigned const cPromiscBefore = pDev->promiscuity;
1846#endif
1847 if (fActive)
1848 {
1849 Assert(!pThis->u.s.fPromiscuousSet);
1850
1851 rtnl_lock();
1852 dev_set_promiscuity(pDev, 1);
1853 rtnl_unlock();
1854 pThis->u.s.fPromiscuousSet = true;
1855 Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1856 }
1857 else
1858 {
1859 if (pThis->u.s.fPromiscuousSet)
1860 {
1861 rtnl_lock();
1862 dev_set_promiscuity(pDev, -1);
1863 rtnl_unlock();
1864 Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1865 }
1866 pThis->u.s.fPromiscuousSet = false;
1867
1868#ifdef LOG_ENABLED
1869 fIf = dev_get_flags(pDev);
1870 Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
1871#endif
1872 }
1873
1874 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1875 }
1876}
1877
1878
1879int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
1880{
1881 /*
1882 * Remove packet handler when we get disconnected from internal switch as
1883 * we don't want the handler to forward packets to disconnected switch.
1884 */
1885 if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1886 {
1887 dev_remove_pack(&pThis->u.s.PacketType);
1888 Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
1889 }
1890 return VINF_SUCCESS;
1891}
1892
1893
1894int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
1895{
1896 /*
1897 * Report the GSO capabilities of the host and device (if connected).
1898 * Note! No need to mark ourselves busy here.
1899 */
1900 /** @todo duplicate work here now? Attach */
1901#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
1902 Log3(("vboxNetFltOsConnectIt: reporting host tso tso6 ufo\n"));
1903 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
1904 0
1905 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
1906 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
1907 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
1908# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1909 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
1910# endif
1911 , INTNETTRUNKDIR_HOST);
1912
1913#endif
1914 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1915
1916 return VINF_SUCCESS;
1917}
1918
1919
1920void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
1921{
1922 struct net_device *pDev;
1923 bool fRegistered;
1924
1925#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1926 vboxNetFltLinuxUnhookDev(pThis, NULL);
1927#endif
1928
1929 /** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
1930 * unlikely, but none the less). Since it doesn't actually update the
1931 * state (just reads it), it is likely to panic in some interesting
1932 * ways. */
1933
1934 RTSpinlockAcquire(pThis->hSpinlock);
1935 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1936 fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1937 RTSpinlockReleaseNoInts(pThis->hSpinlock);
1938
1939 if (fRegistered)
1940 {
1941 vboxNetFltSetTapLinkState(pThis, pDev, false);
1942
1943#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1944 skb_queue_purge(&pThis->u.s.XmitQueue);
1945#endif
1946 Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
1947 Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
1948 pDev, pDev->name,
1949#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1950 netdev_refcnt_read(pDev)
1951#else
1952 atomic_read(&pDev->refcnt)
1953#endif
1954 ));
1955 dev_put(pDev);
1956 }
1957 Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
1958 unregister_netdevice_notifier(&pThis->u.s.Notifier);
1959 module_put(THIS_MODULE);
1960}
1961
1962
1963int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
1964{
1965 int err;
1966 NOREF(pvContext);
1967
1968 pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
1969 err = register_netdevice_notifier(&pThis->u.s.Notifier);
1970 if (err)
1971 return VERR_INTNET_FLT_IF_FAILED;
1972 if (!pThis->u.s.fRegistered)
1973 {
1974 unregister_netdevice_notifier(&pThis->u.s.Notifier);
1975 LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
1976 return VERR_INTNET_FLT_IF_NOT_FOUND;
1977 }
1978
1979 Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
1980 if ( pThis->fDisconnectedFromHost
1981 || !try_module_get(THIS_MODULE))
1982 return VERR_INTNET_FLT_IF_FAILED;
1983
1984 return VINF_SUCCESS;
1985}
1986
1987int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
1988{
1989 /*
1990 * Init the linux specific members.
1991 */
1992 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1993 pThis->u.s.fRegistered = false;
1994 pThis->u.s.fPromiscuousSet = false;
1995 pThis->u.s.fPacketHandler = false;
1996 memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
1997#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1998 skb_queue_head_init(&pThis->u.s.XmitQueue);
1999# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2000 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2001# else
2002 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2003# endif
2004#endif
2005
2006 return VINF_SUCCESS;
2007}
2008
2009
2010void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2011{
2012 NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2013}
2014
2015
2016int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void *pvIf, void **pvIfData)
2017{
2018 /* Nothing to do */
2019 NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2020 return VINF_SUCCESS;
2021}
2022
2023
2024int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2025{
2026 /* Nothing to do */
2027 NOREF(pThis); NOREF(pvIfData);
2028 return VINF_SUCCESS;
2029}
2030
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