VirtualBox

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 64198

Last change on this file since 64198 was 64198, checked in by vboxsync, 8 years ago

NetFlt/Linux (bugref:8599) Yet even more build fixes for r111185
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 91.2 KB
Line 
1/* $Id: VBoxNetFlt-linux.c 64198 2016-10-11 09:04:54Z vboxsync $ */
2/** @file
3 * VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4 */
5
6/*
7 * Copyright (C) 2006-2016 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/*********************************************************************************************************************************
20* Header Files *
21*********************************************************************************************************************************/
22#define LOG_GROUP LOG_GROUP_NET_FLT_DRV
23#define VBOXNETFLT_LINUX_NO_XMIT_QUEUE
24#include "the-linux-kernel.h"
25#include "version-generated.h"
26#include "revision-generated.h"
27#include "product-generated.h"
28#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 24)
29#include <linux/nsproxy.h>
30#endif
31#include <linux/netdevice.h>
32#include <linux/etherdevice.h>
33#include <linux/rtnetlink.h>
34#include <linux/miscdevice.h>
35#include <linux/inetdevice.h>
36#include <linux/in.h>
37#include <linux/ip.h>
38#include <linux/if_vlan.h>
39#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 5, 0)
40#include <uapi/linux/pkt_cls.h>
41#endif
42#include <net/ipv6.h>
43#include <net/if_inet6.h>
44#include <net/addrconf.h>
45
46#include <VBox/log.h>
47#include <VBox/err.h>
48#include <VBox/intnetinline.h>
49#include <VBox/vmm/pdmnetinline.h>
50#include <VBox/param.h>
51#include <iprt/alloca.h>
52#include <iprt/assert.h>
53#include <iprt/spinlock.h>
54#include <iprt/semaphore.h>
55#include <iprt/initterm.h>
56#include <iprt/process.h>
57#include <iprt/mem.h>
58#include <iprt/net.h>
59#include <iprt/log.h>
60#include <iprt/mp.h>
61#include <iprt/mem.h>
62#include <iprt/time.h>
63
64#define VBOXNETFLT_OS_SPECFIC 1
65#include "../VBoxNetFltInternal.h"
66
67typedef struct VBOXNETFLTNOTIFIER {
68 struct notifier_block Notifier;
69 PVBOXNETFLTINS pThis;
70} VBOXNETFLTNOTIFIER;
71typedef struct VBOXNETFLTNOTIFIER *PVBOXNETFLTNOTIFIER;
72
73
74/*********************************************************************************************************************************
75* Defined Constants And Macros *
76*********************************************************************************************************************************/
77#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
78#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
79#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
80# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
81#endif
82
83#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 11, 0)
84# define VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr) netdev_notifier_info_to_dev(ptr)
85#else
86# define VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr) ((struct net_device *)ptr)
87#endif
88
89#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
90# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(skb_frag_page(frag))
91# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr)
92#else
93# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 2, 0)
94# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(skb_frag_page(frag), KM_SKB_DATA_SOFTIRQ)
95# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ)
96# else
97# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(frag->page, KM_SKB_DATA_SOFTIRQ)
98# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ)
99# endif
100#endif
101
102#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 34)
103# define VBOX_NETDEV_NAME(dev) netdev_name(dev)
104#else
105# define VBOX_NETDEV_NAME(dev) ((dev)->reg_state != NETREG_REGISTERED ? "(unregistered net_device)" : (dev)->name)
106#endif
107
108#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25)
109# define VBOX_IPV4_IS_LOOPBACK(addr) ipv4_is_loopback(addr)
110# define VBOX_IPV4_IS_LINKLOCAL_169(addr) ipv4_is_linklocal_169(addr)
111#else
112# define VBOX_IPV4_IS_LOOPBACK(addr) ((addr & htonl(IN_CLASSA_NET)) == htonl(0x7f000000))
113# define VBOX_IPV4_IS_LINKLOCAL_169(addr) ((addr & htonl(IN_CLASSB_NET)) == htonl(0xa9fe0000))
114#endif
115
116#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
117# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
118# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
119# define VBOX_SKB_CSUM_OFFSET(skb) skb->csum_offset
120#else
121# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
122# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
123# define VBOX_SKB_CSUM_OFFSET(skb) skb->csum
124#endif
125
126#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
127# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
128#else
129# define CHECKSUM_PARTIAL CHECKSUM_HW
130# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
131# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
132# else
133# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
134# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
135# else
136# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
137# endif
138/* Versions prior 2.6.10 use stats for both bstats and qstats */
139# define bstats stats
140# define qstats stats
141# endif
142#endif
143
144#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 20, 0)
145# define VBOX_HAVE_SKB_VLAN
146#else
147# ifdef RHEL_RELEASE_CODE
148# if (RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(7, 2) && RHEL_RELEASE_CODE < RHEL_RELEASE_VERSION(8, 0)) || \
149 (RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(6, 8) && RHEL_RELEASE_CODE < RHEL_RELEASE_VERSION(7, 0))
150# define VBOX_HAVE_SKB_VLAN
151# endif
152# endif
153#endif
154
155#ifdef VBOX_HAVE_SKB_VLAN
156# define vlan_tx_tag_get(skb) skb_vlan_tag_get(skb)
157# define vlan_tx_tag_present(skb) skb_vlan_tag_present(skb)
158#endif
159
160#ifndef NET_IP_ALIGN
161# define NET_IP_ALIGN 2
162#endif
163
164#if 1
165/** Create scatter / gather segments for fragments. When not used, we will
166 * linearize the socket buffer before creating the internal networking SG. */
167# define VBOXNETFLT_SG_SUPPORT 1
168#endif
169
170#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
171
172/** Indicates that the linux kernel may send us GSO frames. */
173# define VBOXNETFLT_WITH_GSO 1
174
175/** This enables or disables the transmitting of GSO frame from the internal
176 * network and to the host. */
177# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
178
179# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
180/** This enables or disables the transmitting of GSO frame from the internal
181 * network and to the wire. */
182# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
183# endif
184
185/** This enables or disables the forwarding/flooding of GSO frame from the host
186 * to the internal network. */
187# define VBOXNETFLT_WITH_GSO_RECV 1
188
189#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18) */
190
191#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
192/** This enables or disables handling of GSO frames coming from the wire (GRO). */
193# define VBOXNETFLT_WITH_GRO 1
194#endif
195
196/*
197 * GRO support was backported to RHEL 5.4
198 */
199#ifdef RHEL_RELEASE_CODE
200# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
201# define VBOXNETFLT_WITH_GRO 1
202# endif
203#endif
204
205
206/*********************************************************************************************************************************
207* Internal Functions *
208*********************************************************************************************************************************/
209static int VBoxNetFltLinuxInit(void);
210static void VBoxNetFltLinuxUnload(void);
211static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
212
213
214/*********************************************************************************************************************************
215* Global Variables *
216*********************************************************************************************************************************/
217/**
218 * The (common) global data.
219 */
220static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
221
222module_init(VBoxNetFltLinuxInit);
223module_exit(VBoxNetFltLinuxUnload);
224
225MODULE_AUTHOR(VBOX_VENDOR);
226MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
227MODULE_LICENSE("GPL");
228#ifdef MODULE_VERSION
229MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV) " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
230#endif
231
232
233#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
234unsigned dev_get_flags(const struct net_device *dev)
235{
236 unsigned flags;
237
238 flags = (dev->flags & ~(IFF_PROMISC |
239 IFF_ALLMULTI |
240 IFF_RUNNING)) |
241 (dev->gflags & (IFF_PROMISC |
242 IFF_ALLMULTI));
243
244 if (netif_running(dev) && netif_carrier_ok(dev))
245 flags |= IFF_RUNNING;
246
247 return flags;
248}
249#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
250
251
252/**
253 * Initialize module.
254 *
255 * @returns appropriate status code.
256 */
257static int __init VBoxNetFltLinuxInit(void)
258{
259 int rc;
260 /*
261 * Initialize IPRT.
262 */
263 rc = RTR0Init(0);
264 if (RT_SUCCESS(rc))
265 {
266 Log(("VBoxNetFltLinuxInit\n"));
267
268 /*
269 * Initialize the globals and connect to the support driver.
270 *
271 * This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
272 * for establishing the connect to the support driver.
273 */
274 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
275 rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
276 if (RT_SUCCESS(rc))
277 {
278 LogRel(("VBoxNetFlt: Successfully started.\n"));
279 return 0;
280 }
281
282 LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
283 RTR0Term();
284 }
285 else
286 LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
287
288 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
289 return -RTErrConvertToErrno(rc);
290}
291
292
293/**
294 * Unload the module.
295 *
296 * @todo We have to prevent this if we're busy!
297 */
298static void __exit VBoxNetFltLinuxUnload(void)
299{
300 int rc;
301 Log(("VBoxNetFltLinuxUnload\n"));
302 Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
303
304 /*
305 * Undo the work done during start (in reverse order).
306 */
307 rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
308 AssertRC(rc); NOREF(rc);
309
310 RTR0Term();
311
312 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
313
314 Log(("VBoxNetFltLinuxUnload - done\n"));
315}
316
317
318/**
319 * We filter traffic from the host to the internal network
320 * before it reaches the NIC driver.
321 *
322 * The current code uses a very ugly hack overriding hard_start_xmit
323 * callback in the device structure, but it has been shown to give us a
324 * performance boost of 60-100% though. Eventually we have to find some
325 * less hacky way of getting this job done.
326 */
327#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
328
329#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
330
331# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
332
333# include <linux/ethtool.h>
334
335typedef struct ethtool_ops OVR_OPSTYPE;
336# define OVR_OPS ethtool_ops
337# define OVR_XMIT pfnStartXmit
338
339# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
340
341typedef struct net_device_ops OVR_OPSTYPE;
342# define OVR_OPS netdev_ops
343# define OVR_XMIT pOrgOps->ndo_start_xmit
344
345# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
346
347/**
348 * The overridden net_device_ops of the device we're attached to.
349 *
350 * As there is no net_device_ops structure in pre-2.6.29 kernels we override
351 * ethtool_ops instead along with hard_start_xmit callback in net_device
352 * structure.
353 *
354 * This is a very dirty hack that was created to explore how much we can improve
355 * the host to guest transfers by not CC'ing the NIC. It turns out to be
356 * the only way to filter outgoing packets for devices without TX queue.
357 */
358typedef struct VBoxNetDeviceOpsOverride
359{
360 /** Our overridden ops. */
361 OVR_OPSTYPE Ops;
362 /** Magic word. */
363 uint32_t u32Magic;
364 /** Pointer to the original ops. */
365 OVR_OPSTYPE const *pOrgOps;
366# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
367 /** Pointer to the original hard_start_xmit function. */
368 int (*pfnStartXmit)(struct sk_buff *pSkb, struct net_device *pDev);
369# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
370 /** Pointer to the net filter instance. */
371 PVBOXNETFLTINS pVBoxNetFlt;
372 /** The number of filtered packages. */
373 uint64_t cFiltered;
374 /** The total number of packets */
375 uint64_t cTotal;
376} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
377/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
378#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
379
380/**
381 * ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
382 * because they belong on the internal network.
383 *
384 * @returns NETDEV_TX_XXX.
385 * @param pSkb The socket buffer to transmit.
386 * @param pDev The net device.
387 */
388static int vboxNetFltLinuxStartXmitFilter(struct sk_buff *pSkb, struct net_device *pDev)
389{
390 PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
391 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
392 PCRTNETETHERHDR pEtherHdr;
393 PINTNETTRUNKSWPORT pSwitchPort;
394 uint32_t cbHdrs;
395
396
397 /*
398 * Validate the override structure.
399 *
400 * Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
401 * to be production quality code, we would have to be much more
402 * careful here and avoid the race.
403 */
404 if ( !VALID_PTR(pOverride)
405 || pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
406# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
407 || !VALID_PTR(pOverride->pOrgOps)
408# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
409 )
410 {
411 printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
412 dev_kfree_skb(pSkb);
413 return NETDEV_TX_OK;
414 }
415 pOverride->cTotal++;
416
417 /*
418 * Do the filtering base on the default OUI of our virtual NICs
419 *
420 * Note! In a real solution, we would ask the switch whether the
421 * destination MAC is 100% to be on the internal network and then
422 * drop it.
423 */
424 cbHdrs = skb_headlen(pSkb);
425 cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
426 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
427 if ( pEtherHdr
428 && VALID_PTR(pOverride->pVBoxNetFlt)
429 && (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
430 && VALID_PTR(pSwitchPort)
431 && cbHdrs >= 6)
432 {
433 INTNETSWDECISION enmDecision;
434
435 /** @todo consider reference counting, etc. */
436 enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
437 if (enmDecision == INTNETSWDECISION_INTNET)
438 {
439 dev_kfree_skb(pSkb);
440 pOverride->cFiltered++;
441 return NETDEV_TX_OK;
442 }
443 }
444
445 return pOverride->OVR_XMIT(pSkb, pDev);
446}
447
448/**
449 * Hooks the device ndo_start_xmit operation of the device.
450 *
451 * @param pThis The net filter instance.
452 * @param pDev The net device.
453 */
454static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
455{
456 PVBOXNETDEVICEOPSOVERRIDE pOverride;
457
458 /* Cancel override if ethtool_ops is missing (host-only case, @bugref{5712}) */
459 if (!VALID_PTR(pDev->OVR_OPS))
460 return;
461 pOverride = RTMemAlloc(sizeof(*pOverride));
462 if (!pOverride)
463 return;
464 pOverride->pOrgOps = pDev->OVR_OPS;
465 pOverride->Ops = *pDev->OVR_OPS;
466# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
467 pOverride->pfnStartXmit = pDev->hard_start_xmit;
468# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
469 pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
470# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
471 pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
472 pOverride->cTotal = 0;
473 pOverride->cFiltered = 0;
474 pOverride->pVBoxNetFlt = pThis;
475
476 RTSpinlockAcquire(pThis->hSpinlock); /* (this isn't necessary, but so what) */
477 ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
478# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
479 ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
480# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
481 RTSpinlockRelease(pThis->hSpinlock);
482}
483
484/**
485 * Undos what vboxNetFltLinuxHookDev did.
486 *
487 * @param pThis The net filter instance.
488 * @param pDev The net device. Can be NULL, in which case
489 * we'll try retrieve it from @a pThis.
490 */
491static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
492{
493 PVBOXNETDEVICEOPSOVERRIDE pOverride;
494
495 RTSpinlockAcquire(pThis->hSpinlock);
496 if (!pDev)
497 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
498 if (VALID_PTR(pDev))
499 {
500 pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
501 if ( VALID_PTR(pOverride)
502 && pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
503 && VALID_PTR(pOverride->pOrgOps)
504 )
505 {
506# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
507 ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
508# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
509 ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
510 ASMAtomicWriteU32(&pOverride->u32Magic, 0);
511 }
512 else
513 pOverride = NULL;
514 }
515 else
516 pOverride = NULL;
517 RTSpinlockRelease(pThis->hSpinlock);
518
519 if (pOverride)
520 {
521 printk("vboxnetflt: %llu out of %llu packets were not sent (directed to host)\n", pOverride->cFiltered, pOverride->cTotal);
522 RTMemFree(pOverride);
523 }
524}
525
526#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
527
528
529/**
530 * Reads and retains the host interface handle.
531 *
532 * @returns The handle, NULL if detached.
533 * @param pThis
534 */
535DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
536{
537#if 0
538 struct net_device *pDev = NULL;
539
540 Log(("vboxNetFltLinuxRetainNetDev\n"));
541 /*
542 * Be careful here to avoid problems racing the detached callback.
543 */
544 RTSpinlockAcquire(pThis->hSpinlock);
545 if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
546 {
547 pDev = (struct net_device *)ASMAtomicUoReadPtr((void * volatile *)&pThis->u.s.pDev);
548 if (pDev)
549 {
550 dev_hold(pDev);
551 Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
552 pDev, pDev->name,
553#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
554 netdev_refcnt_read(pDev)
555#else
556 atomic_read(&pDev->refcnt)
557#endif
558 ));
559 }
560 }
561 RTSpinlockRelease(pThis->hSpinlock);
562
563 Log(("vboxNetFltLinuxRetainNetDev - done\n"));
564 return pDev;
565#else
566 return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
567#endif
568}
569
570
571/**
572 * Release the host interface handle previously retained
573 * by vboxNetFltLinuxRetainNetDev.
574 *
575 * @param pThis The instance.
576 * @param pDev The vboxNetFltLinuxRetainNetDev
577 * return value, NULL is fine.
578 */
579DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
580{
581#if 0
582 Log(("vboxNetFltLinuxReleaseNetDev\n"));
583 NOREF(pThis);
584 if (pDev)
585 {
586 dev_put(pDev);
587 Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
588 pDev, pDev->name,
589#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
590 netdev_refcnt_read(pDev)
591#else
592 atomic_read(&pDev->refcnt)
593#endif
594 ));
595 }
596 Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
597#endif
598}
599
600#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 | (skb->dev->ifindex & 0xFFFF))
601#define VBOXNETFLT_SKB_TAG(skb) (*(uint32_t*)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
602
603/**
604 * Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
605 * i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
606 *
607 * @returns true / false accordingly.
608 * @param pBuf The sk_buff.
609 */
610DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
611{
612 return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
613}
614
615
616/**
617 * Checks whether this SG list contains a GSO packet.
618 *
619 * @returns true / false accordingly.
620 * @param pSG The (scatter/)gather list.
621 */
622DECLINLINE(bool) vboxNetFltLinuxIsGso(PINTNETSG pSG)
623{
624#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
625 return !((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type == PDMNETWORKGSOTYPE_INVALID);
626#else /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
627 return false;
628#endif /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
629}
630
631
632/**
633 * Find out the frame size (of a single segment in case of GSO frames).
634 *
635 * @returns the frame size.
636 * @param pSG The (scatter/)gather list.
637 */
638DECLINLINE(uint32_t) vboxNetFltLinuxFrameSize(PINTNETSG pSG)
639{
640 uint16_t u16Type = 0;
641 uint32_t cbVlanTag = 0;
642 if (pSG->aSegs[0].cb >= sizeof(RTNETETHERHDR))
643 u16Type = RT_BE2H_U16(((PCRTNETETHERHDR)pSG->aSegs[0].pv)->EtherType);
644 else if (pSG->cbTotal >= sizeof(RTNETETHERHDR))
645 {
646 uint32_t off = RT_OFFSETOF(RTNETETHERHDR, EtherType);
647 uint32_t i;
648 for (i = 0; i < pSG->cSegsUsed; ++i)
649 {
650 if (off <= pSG->aSegs[i].cb)
651 {
652 if (off + sizeof(uint16_t) <= pSG->aSegs[i].cb)
653 u16Type = RT_BE2H_U16(*(uint16_t *)((uintptr_t)pSG->aSegs[i].pv + off));
654 else if (i + 1 < pSG->cSegsUsed)
655 u16Type = RT_BE2H_U16( ((uint16_t)( ((uint8_t *)pSG->aSegs[i].pv)[off] ) << 8)
656 + *(uint8_t *)pSG->aSegs[i + 1].pv); /* ASSUMES no empty segments! */
657 /* else: frame is too short. */
658 break;
659 }
660 off -= pSG->aSegs[i].cb;
661 }
662 }
663 if (u16Type == RTNET_ETHERTYPE_VLAN)
664 cbVlanTag = 4;
665 return (vboxNetFltLinuxIsGso(pSG) ? (uint32_t)pSG->GsoCtx.cbMaxSeg + pSG->GsoCtx.cbHdrsTotal : pSG->cbTotal) - cbVlanTag;
666}
667
668
669/**
670 * Internal worker that create a linux sk_buff for a
671 * (scatter/)gather list.
672 *
673 * @returns Pointer to the sk_buff.
674 * @param pThis The instance.
675 * @param pSG The (scatter/)gather list.
676 * @param fDstWire Set if the destination is the wire.
677 */
678static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
679{
680 struct sk_buff *pPkt;
681 struct net_device *pDev;
682 unsigned fGsoType = 0;
683
684 if (pSG->cbTotal == 0)
685 {
686 LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
687 return NULL;
688 }
689 Log5(("VBoxNetFlt: Packet to %s of %d bytes (frame=%d).\n", fDstWire?"wire":"host", pSG->cbTotal, vboxNetFltLinuxFrameSize(pSG)));
690 if (fDstWire && (vboxNetFltLinuxFrameSize(pSG) > ASMAtomicReadU32(&pThis->u.s.cbMtu) + 14))
691 {
692 static bool s_fOnce = true;
693 if (s_fOnce)
694 {
695 s_fOnce = false;
696 printk("VBoxNetFlt: Dropped over-sized packet (%d bytes) coming from internal network.\n", vboxNetFltLinuxFrameSize(pSG));
697 }
698 return NULL;
699 }
700
701 /** @todo We should use fragments mapping the SG buffers with large packets.
702 * 256 bytes seems to be the a threshold used a lot for this. It
703 * requires some nasty work on the intnet side though... */
704 /*
705 * Allocate a packet and copy over the data.
706 */
707 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
708 pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
709 if (RT_UNLIKELY(!pPkt))
710 {
711 Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
712 pSG->pvUserData = NULL;
713 return NULL;
714 }
715 pPkt->dev = pDev;
716 pPkt->ip_summed = CHECKSUM_NONE;
717
718 /* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
719 skb_reserve(pPkt, NET_IP_ALIGN);
720
721 /* Copy the segments. */
722 skb_put(pPkt, pSG->cbTotal);
723 IntNetSgRead(pSG, pPkt->data);
724
725#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
726 /*
727 * Setup GSO if used by this packet.
728 */
729 switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
730 {
731 default:
732 AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
733 /* fall thru */
734 case PDMNETWORKGSOTYPE_INVALID:
735 fGsoType = 0;
736 break;
737 case PDMNETWORKGSOTYPE_IPV4_TCP:
738 fGsoType = SKB_GSO_TCPV4;
739 break;
740 case PDMNETWORKGSOTYPE_IPV4_UDP:
741 fGsoType = SKB_GSO_UDP;
742 break;
743 case PDMNETWORKGSOTYPE_IPV6_TCP:
744 fGsoType = SKB_GSO_TCPV6;
745 break;
746 }
747 if (fGsoType)
748 {
749 struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
750
751 pShInfo->gso_type = fGsoType | SKB_GSO_DODGY;
752 pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
753 pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
754
755 /*
756 * We need to set checksum fields even if the packet goes to the host
757 * directly as it may be immediately forwarded by IP layer @bugref{5020}.
758 */
759 Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
760 pPkt->ip_summed = CHECKSUM_PARTIAL;
761# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
762 pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
763 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
764 pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
765 else
766 pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
767# else
768 pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
769 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
770 pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
771 else
772 pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
773# endif
774 if (!fDstWire)
775 PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
776 }
777#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE || VBOXNETFLT_WITH_GSO_XMIT_HOST */
778
779 /*
780 * Finish up the socket buffer.
781 */
782 pPkt->protocol = eth_type_trans(pPkt, pDev);
783 if (fDstWire)
784 {
785 VBOX_SKB_RESET_NETWORK_HDR(pPkt);
786
787 /* Restore ethernet header back. */
788 skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
789 VBOX_SKB_RESET_MAC_HDR(pPkt);
790 }
791 VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
792
793 return pPkt;
794}
795
796
797/**
798 * Return the offset where to start checksum computation from.
799 *
800 * @returns the offset relative to pBuf->data.
801 * @param pBuf The socket buffer.
802 */
803DECLINLINE(unsigned) vboxNetFltLinuxGetChecksumStartOffset(struct sk_buff *pBuf)
804{
805# if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 21)
806 unsigned char *pTransportHdr = pBuf->h.raw;
807# if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
808 /*
809 * Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
810 * and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
811 * header length from the header itself and reconstruct 'h' pointer
812 * to TCP (or whatever) header.
813 */
814 if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
815 pTransportHdr = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
816# endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
817 return pTransportHdr - pBuf->data;
818# else /* LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 21) */
819 return skb_checksum_start_offset(pBuf);
820# endif /* LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 21) */
821}
822
823
824/**
825 * Initializes a SG list from an sk_buff.
826 *
827 * @returns Number of segments.
828 * @param pThis The instance.
829 * @param pBuf The sk_buff.
830 * @param pSG The SG.
831 * @param cExtra The number of bytes of extra space allocated immediately after the SG.
832 * @param cSegs The number of segments allocated for the SG.
833 * This should match the number in the mbuf exactly!
834 * @param fSrc The source of the frame.
835 * @param pGsoCtx Pointer to the GSO context if it's a GSO
836 * internal network frame. NULL if regular frame.
837 */
838static void vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
839 unsigned cbExtra, unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
840{
841 int i;
842 NOREF(pThis);
843
844#ifndef VBOXNETFLT_SG_SUPPORT
845 Assert(!skb_shinfo(pBuf)->frag_list);
846#else /* VBOXNETFLT_SG_SUPPORT */
847 uint8_t *pExtra = (uint8_t *)&pSG->aSegs[cSegs];
848 unsigned cbConsumed = 0;
849 unsigned cbProduced = 0;
850
851# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
852 /* Restore VLAN tag stripped by host hardware */
853 if (vlan_tx_tag_present(pBuf))
854 {
855 uint8_t *pMac = pBuf->data;
856 struct vlan_ethhdr *pVHdr = (struct vlan_ethhdr *)pExtra;
857 Assert(ETH_ALEN * 2 + VLAN_HLEN <= cbExtra);
858 memmove(pVHdr, pMac, ETH_ALEN * 2);
859 cbConsumed += ETH_ALEN * 2;
860 pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
861 pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
862 pVHdr->h_vlan_encapsulated_proto = *(uint16_t*)(pMac + ETH_ALEN * 2);
863 cbProduced += VLAN_ETH_HLEN;
864 }
865# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
866
867 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
868 {
869 unsigned uCsumStartOffset = vboxNetFltLinuxGetChecksumStartOffset(pBuf);
870 unsigned uCsumStoreOffset = uCsumStartOffset + VBOX_SKB_CSUM_OFFSET(pBuf) - cbConsumed;
871 Log3(("cbConsumed=%u cbProduced=%u uCsumStartOffset=%u uCsumStoreOffset=%u\n",
872 cbConsumed, cbProduced, uCsumStartOffset, uCsumStoreOffset));
873 Assert(cbProduced + uCsumStoreOffset + sizeof(uint16_t) <= cbExtra);
874 /*
875 * We assume that the checksum is stored at the very end of the transport header
876 * so we will have all headers in a single fragment. If our assumption is wrong
877 * we may see suboptimal performance.
878 */
879 memmove(pExtra + cbProduced,
880 pBuf->data + cbConsumed,
881 uCsumStoreOffset);
882 unsigned uChecksum = skb_checksum(pBuf, uCsumStartOffset, pBuf->len - uCsumStartOffset, 0);
883 *(uint16_t*)(pExtra + cbProduced + uCsumStoreOffset) = csum_fold(uChecksum);
884 cbProduced += uCsumStoreOffset + sizeof(uint16_t);
885 cbConsumed += uCsumStoreOffset + sizeof(uint16_t);
886 }
887#endif /* VBOXNETFLT_SG_SUPPORT */
888
889 if (!pGsoCtx)
890 IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/);
891 else
892 IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/, pGsoCtx);
893
894 int iSeg = 0;
895#ifdef VBOXNETFLT_SG_SUPPORT
896 if (cbProduced)
897 {
898 pSG->aSegs[iSeg].cb = cbProduced;
899 pSG->aSegs[iSeg].pv = pExtra;
900 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
901 }
902 pSG->aSegs[iSeg].cb = skb_headlen(pBuf) - cbConsumed;
903 pSG->aSegs[iSeg].pv = pBuf->data + cbConsumed;
904 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
905 Assert(iSeg <= pSG->cSegsAlloc);
906
907# ifdef LOG_ENABLED
908 if (pBuf->data_len)
909 Log6((" kmap_atomic:"));
910# endif /* LOG_ENABLED */
911 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
912 {
913 skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
914 pSG->aSegs[iSeg].cb = pFrag->size;
915 pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
916 Log6((" %p", pSG->aSegs[iSeg].pv));
917 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
918 Assert(iSeg <= pSG->cSegsAlloc);
919 }
920 struct sk_buff *pFragBuf;
921 for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
922 {
923 pSG->aSegs[iSeg].cb = skb_headlen(pFragBuf);
924 pSG->aSegs[iSeg].pv = pFragBuf->data;
925 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
926 Assert(iSeg <= pSG->cSegsAlloc);
927 for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
928 {
929 skb_frag_t *pFrag = &skb_shinfo(pFragBuf)->frags[i];
930 pSG->aSegs[iSeg].cb = pFrag->size;
931 pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
932 Log6((" %p", pSG->aSegs[iSeg].pv));
933 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
934 Assert(iSeg <= pSG->cSegsAlloc);
935 }
936 }
937# ifdef LOG_ENABLED
938 if (pBuf->data_len)
939 Log6(("\n"));
940# endif /* LOG_ENABLED */
941#else
942 pSG->aSegs[iSeg].cb = pBuf->len;
943 pSG->aSegs[iSeg].pv = pBuf->data;
944 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
945#endif
946
947 pSG->cSegsUsed = iSeg;
948
949#ifdef PADD_RUNT_FRAMES_FROM_HOST
950 /*
951 * Add a trailer if the frame is too small.
952 *
953 * Since we're getting to the packet before it is framed, it has not
954 * yet been padded. The current solution is to add a segment pointing
955 * to a buffer containing all zeros and pray that works for all frames...
956 */
957 if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
958 {
959 Assert(pBuf->data_len == 0); /* Packets with fragments are never small! */
960 static uint8_t const s_abZero[128] = {0};
961
962 AssertReturnVoid(iSeg < cSegs);
963
964 pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
965 pSG->aSegs[iSeg].pv = (void *)&s_abZero[0];
966 pSG->aSegs[iSeg++].cb = 60 - pSG->cbTotal;
967 pSG->cbTotal = 60;
968 pSG->cSegsUsed++;
969 Assert(iSeg <= pSG->cSegsAlloc)
970 }
971#endif
972
973 Log6(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
974 pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
975 for (i = 0; i < pSG->cSegsUsed; i++)
976 Log6(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
977 i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
978}
979
980/**
981 * Packet handler; not really documented - figure it out yourself.
982 *
983 * @returns 0 or EJUSTRETURN - this is probably copy & pastry and thus wrong.
984 */
985#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
986static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
987 struct net_device *pSkbDev,
988 struct packet_type *pPacketType,
989 struct net_device *pOrigDev)
990#else
991static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
992 struct net_device *pSkbDev,
993 struct packet_type *pPacketType)
994#endif
995{
996 PVBOXNETFLTINS pThis;
997 struct net_device *pDev;
998 LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
999 pBuf, pSkbDev, pPacketType));
1000#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1001 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",
1002 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));
1003# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1004 Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1005# endif
1006#else
1007 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",
1008 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));
1009#endif
1010 /*
1011 * Drop it immediately?
1012 */
1013 if (!pBuf)
1014 return 0;
1015
1016 if (pBuf->pkt_type == PACKET_LOOPBACK)
1017 {
1018 /*
1019 * We are not interested in loopbacked packets as they will always have
1020 * another copy going to the wire.
1021 */
1022 Log2(("vboxNetFltLinuxPacketHandler: dropped loopback packet (cb=%u)\n", pBuf->len));
1023 dev_kfree_skb(pBuf); /* We must 'consume' all packets we get (@bugref{6539})! */
1024 return 0;
1025 }
1026
1027 pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1028 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1029 if (pDev != pSkbDev)
1030 {
1031 Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1032 kfree_skb(pBuf); /* This is a failure, so we use kfree_skb instead of dev_kfree_skb. */
1033 return 0;
1034 }
1035
1036 Log6(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1037 if (vboxNetFltLinuxSkBufIsOur(pBuf))
1038 {
1039 Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1040 dev_kfree_skb(pBuf);
1041 return 0;
1042 }
1043
1044#ifndef VBOXNETFLT_SG_SUPPORT
1045 {
1046 /*
1047 * Get rid of fragmented packets, they cause too much trouble.
1048 */
1049 unsigned int uMacLen = pBuf->mac_len;
1050 struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1051 dev_kfree_skb(pBuf);
1052 if (!pCopy)
1053 {
1054 LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1055 return 0;
1056 }
1057 pBuf = pCopy;
1058 /* Somehow skb_copy ignores mac_len */
1059 pBuf->mac_len = uMacLen;
1060# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
1061 /* Restore VLAN tag stripped by host hardware */
1062 if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
1063 {
1064 uint8_t *pMac = (uint8_t*)skb_mac_header(pBuf);
1065 struct vlan_ethhdr *pVHdr = (struct vlan_ethhdr *)(pMac - VLAN_HLEN);
1066 memmove(pVHdr, pMac, ETH_ALEN * 2);
1067 pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
1068 pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
1069 pBuf->mac_header -= VLAN_HLEN;
1070 pBuf->mac_len += VLAN_HLEN;
1071 }
1072# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
1073
1074# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1075 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",
1076 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));
1077# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1078 Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1079# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22) */
1080# else /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 18) */
1081 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",
1082 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));
1083# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 18) */
1084 }
1085#endif /* !VBOXNETFLT_SG_SUPPORT */
1086
1087#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1088 /* Forward it to the internal network. */
1089 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1090#else /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1091 /* Add the packet to transmit queue and schedule the bottom half. */
1092 skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1093 schedule_work(&pThis->u.s.XmitTask);
1094 Log6(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1095 &pThis->u.s.XmitTask, pBuf));
1096#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1097
1098 /* It does not really matter what we return, it is ignored by the kernel. */
1099 return 0;
1100}
1101
1102/**
1103 * Calculate the number of INTNETSEG segments the socket buffer will need.
1104 *
1105 * @returns Segment count.
1106 * @param pBuf The socket buffer.
1107 * @param pcbTemp Where to store the number of bytes of the part
1108 * of the socket buffer that will be copied to
1109 * a temporary storage.
1110 */
1111DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf, unsigned *pcbTemp)
1112{
1113 *pcbTemp = 0;
1114#ifdef VBOXNETFLT_SG_SUPPORT
1115 unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1116 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1117 {
1118 *pcbTemp = vboxNetFltLinuxGetChecksumStartOffset(pBuf) + VBOX_SKB_CSUM_OFFSET(pBuf) + sizeof(uint16_t);
1119 }
1120# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
1121 if (vlan_tx_tag_present(pBuf))
1122 {
1123 if (*pcbTemp)
1124 *pcbTemp += VLAN_HLEN;
1125 else
1126 *pcbTemp = VLAN_ETH_HLEN;
1127 }
1128# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
1129 if (*pcbTemp)
1130 ++cSegs;
1131 struct sk_buff *pFrag;
1132 for (pFrag = skb_shinfo(pBuf)->frag_list; pFrag; pFrag = pFrag->next)
1133 {
1134 Log6(("vboxNetFltLinuxCalcSGSegments: frag=%p len=%d data_len=%d frags=%d frag_list=%p next=%p\n",
1135 pFrag, pFrag->len, pFrag->data_len, skb_shinfo(pFrag)->nr_frags, skb_shinfo(pFrag)->frag_list, pFrag->next));
1136 cSegs += 1 + skb_shinfo(pFrag)->nr_frags;
1137 }
1138#else
1139 unsigned cSegs = 1;
1140#endif
1141#ifdef PADD_RUNT_FRAMES_FROM_HOST
1142 /* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1143 if (pBuf->len < 60)
1144 cSegs++;
1145#endif
1146 return cSegs;
1147}
1148
1149
1150/**
1151 * Destroy the intnet scatter / gather buffer created by
1152 * vboxNetFltLinuxSkBufToSG.
1153 *
1154 * @param pSG The (scatter/)gather list.
1155 * @param pBuf The original socket buffer that was used to create
1156 * the scatter/gather list.
1157 */
1158static void vboxNetFltLinuxDestroySG(PINTNETSG pSG, struct sk_buff *pBuf)
1159{
1160#ifdef VBOXNETFLT_SG_SUPPORT
1161 int i, iSeg = 1; /* Skip non-paged part of SKB */
1162 /* Check if the extra buffer behind SG structure was used for modified packet header */
1163 if (pBuf->data != pSG->aSegs[0].pv)
1164 ++iSeg; /* Skip it as well */
1165# ifdef LOG_ENABLED
1166 if (pBuf->data_len)
1167 Log6(("kunmap_atomic:"));
1168# endif /* LOG_ENABLED */
1169 /* iSeg now points to the first mapped fragment if there are any */
1170 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1171 {
1172 Log6((" %p", pSG->aSegs[iSeg].pv));
1173 VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1174 }
1175 struct sk_buff *pFragBuf;
1176 for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
1177 {
1178 ++iSeg; /* Non-fragment (unmapped) portion of chained SKB */
1179 for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
1180 {
1181 Log6((" %p", pSG->aSegs[iSeg].pv));
1182 VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1183 }
1184 }
1185# ifdef LOG_ENABLED
1186 if (pBuf->data_len)
1187 Log6(("\n"));
1188# endif /* LOG_ENABLED */
1189#endif
1190 NOREF(pSG);
1191}
1192
1193#ifdef LOG_ENABLED
1194/**
1195 * Logging helper.
1196 */
1197static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1198{
1199 int i, offSeg;
1200 uint8_t *pInt, *pExt;
1201 static int iPacketNo = 1;
1202 iPacketNo += iIncrement;
1203 if (fEgress)
1204 {
1205 pExt = pSG->aSegs[0].pv;
1206 pInt = pExt + 6;
1207 }
1208 else
1209 {
1210 pInt = pSG->aSegs[0].pv;
1211 pExt = pInt + 6;
1212 }
1213 Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1214 " %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1215 pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1216 fEgress ? "-->" : "<--", pszWhere,
1217 pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1218 pSG->cbTotal, iPacketNo));
1219 if (pSG->cSegsUsed == 1)
1220 {
1221 Log4(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1222 }
1223 else
1224 {
1225 for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
1226 {
1227 Log4(("-- segment %d at 0x%x (%d bytes)\n --\n%.*Rhxd\n",
1228 i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1229 offSeg += pSG->aSegs[i].cb;
1230 }
1231 }
1232}
1233#else
1234# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1235#endif
1236
1237#ifdef VBOXNETFLT_WITH_GSO_RECV
1238
1239/**
1240 * Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1241 * GSO socket buffer without having to segment it.
1242 *
1243 * @returns true on success, false if needs segmenting.
1244 * @param pThis The net filter instance.
1245 * @param pSkb The GSO socket buffer.
1246 * @param fSrc The source.
1247 * @param pGsoCtx Where to return the GSO context on success.
1248 */
1249static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1250 PPDMNETWORKGSO pGsoCtx)
1251{
1252 PDMNETWORKGSOTYPE enmGsoType;
1253 uint16_t uEtherType;
1254 unsigned int cbTransport;
1255 unsigned int offTransport;
1256 unsigned int cbTransportHdr;
1257 unsigned uProtocol;
1258 union
1259 {
1260 RTNETIPV4 IPv4;
1261 RTNETIPV6 IPv6;
1262 RTNETTCP Tcp;
1263 uint8_t ab[40];
1264 uint16_t au16[40/2];
1265 uint32_t au32[40/4];
1266 } Buf;
1267
1268 /*
1269 * Check the GSO properties of the socket buffer and make sure it fits.
1270 */
1271 /** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1272 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP | SKB_GSO_DODGY | SKB_GSO_TCPV6 | SKB_GSO_TCPV4) ))
1273 {
1274 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1275 return false;
1276 }
1277 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1278 || pSkb->len > VBOX_MAX_GSO_SIZE ))
1279 {
1280 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1281 return false;
1282 }
1283
1284 /*
1285 * Switch on the ethertype.
1286 */
1287 uEtherType = pSkb->protocol;
1288 if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1289 && pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1290 {
1291 uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1292 if (puEtherType)
1293 uEtherType = *puEtherType;
1294 }
1295 switch (uEtherType)
1296 {
1297 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1298 {
1299 unsigned int cbHdr;
1300 PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1301 if (RT_UNLIKELY(!pIPv4))
1302 {
1303 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1304 return false;
1305 }
1306
1307 cbHdr = pIPv4->ip_hl * 4;
1308 cbTransport = RT_N2H_U16(pIPv4->ip_len);
1309 if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1310 || cbHdr > cbTransport ))
1311 {
1312 Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1313 return false;
1314 }
1315 cbTransport -= cbHdr;
1316 offTransport = pSkb->mac_len + cbHdr;
1317 uProtocol = pIPv4->ip_p;
1318 if (uProtocol == RTNETIPV4_PROT_TCP)
1319 enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1320 else if (uProtocol == RTNETIPV4_PROT_UDP)
1321 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1322 else /** @todo IPv6: 4to6 tunneling */
1323 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1324 break;
1325 }
1326
1327 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1328 {
1329 PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1330 if (RT_UNLIKELY(!pIPv6))
1331 {
1332 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1333 return false;
1334 }
1335
1336 cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1337 offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1338 uProtocol = pIPv6->ip6_nxt;
1339 /** @todo IPv6: Dig our way out of the other headers. */
1340 if (uProtocol == RTNETIPV4_PROT_TCP)
1341 enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1342 else if (uProtocol == RTNETIPV4_PROT_UDP)
1343 enmGsoType = PDMNETWORKGSOTYPE_IPV6_UDP;
1344 else
1345 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1346 break;
1347 }
1348
1349 default:
1350 Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1351 return false;
1352 }
1353
1354 if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1355 {
1356 Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1357 return false;
1358 }
1359
1360 if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1361 || offTransport + cbTransport > pSkb->len
1362 || cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1363 {
1364 Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1365 offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1366 return false;
1367 }
1368
1369 /*
1370 * Check the TCP/UDP bits.
1371 */
1372 if (uProtocol == RTNETIPV4_PROT_TCP)
1373 {
1374 PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1375 if (RT_UNLIKELY(!pTcp))
1376 {
1377 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1378 return false;
1379 }
1380
1381 cbTransportHdr = pTcp->th_off * 4;
1382 pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1383 if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1384 || cbTransportHdr > cbTransport
1385 || offTransport + cbTransportHdr >= UINT8_MAX
1386 || offTransport + cbTransportHdr >= pSkb->len ))
1387 {
1388 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1389 return false;
1390 }
1391
1392 }
1393 else
1394 {
1395 Assert(uProtocol == RTNETIPV4_PROT_UDP);
1396 cbTransportHdr = sizeof(RTNETUDP);
1397 pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1398 if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1399 || offTransport + cbTransportHdr >= pSkb->len ))
1400 {
1401 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1402 return false;
1403 }
1404 }
1405
1406 /*
1407 * We're good, init the GSO context.
1408 */
1409 pGsoCtx->u8Type = enmGsoType;
1410 pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1411 pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1412 pGsoCtx->offHdr1 = pSkb->mac_len;
1413 pGsoCtx->offHdr2 = offTransport;
1414 pGsoCtx->u8Unused = 0;
1415
1416 return true;
1417}
1418
1419/**
1420 * Forward the socket buffer as a GSO internal network frame.
1421 *
1422 * @returns IPRT status code.
1423 * @param pThis The net filter instance.
1424 * @param pSkb The GSO socket buffer.
1425 * @param fSrc The source.
1426 * @param pGsoCtx Where to return the GSO context on success.
1427 */
1428static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1429{
1430 int rc;
1431 unsigned cbExtra;
1432 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb, &cbExtra);
1433 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]) + cbExtra);
1434 if (RT_LIKELY(pSG))
1435 {
1436 vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cbExtra, cSegs, fSrc, pGsoCtx);
1437
1438 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1439 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1440
1441 vboxNetFltLinuxDestroySG(pSG, pSkb);
1442 rc = VINF_SUCCESS;
1443 }
1444 else
1445 {
1446 Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1447 rc = VERR_NO_MEMORY;
1448 }
1449 return rc;
1450}
1451
1452#endif /* VBOXNETFLT_WITH_GSO_RECV */
1453
1454/**
1455 * Worker for vboxNetFltLinuxForwardToIntNet.
1456 *
1457 * @returns VINF_SUCCESS or VERR_NO_MEMORY.
1458 * @param pThis The net filter instance.
1459 * @param pBuf The socket buffer.
1460 * @param fSrc The source.
1461 */
1462static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1463{
1464 int rc;
1465 unsigned cbExtra;
1466 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf, &cbExtra);
1467 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]) + cbExtra);
1468 if (RT_LIKELY(pSG))
1469 {
1470 vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cbExtra, cSegs, fSrc, NULL /*pGsoCtx*/);
1471
1472 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1473 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1474
1475 vboxNetFltLinuxDestroySG(pSG, pBuf);
1476 rc = VINF_SUCCESS;
1477 }
1478 else
1479 {
1480 Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1481 rc = VERR_NO_MEMORY;
1482 }
1483 return rc;
1484}
1485
1486
1487/**
1488 * I won't disclose what I do, figure it out yourself, including pThis referencing.
1489 *
1490 * @param pThis The net filter instance.
1491 * @param pBuf The socket buffer.
1492 * @param fSrc Where the packet comes from.
1493 */
1494static void vboxNetFltLinuxForwardToIntNetInner(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1495{
1496#ifdef VBOXNETFLT_WITH_GSO
1497 if (skb_is_gso(pBuf))
1498 {
1499 PDMNETWORKGSO GsoCtx;
1500 Log6(("vboxNetFltLinuxForwardToIntNetInner: skb len=%u data_len=%u truesize=%u next=%p"
1501 " nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x ip_summed=%d\n",
1502 pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next,
1503 skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->gso_size,
1504 skb_shinfo(pBuf)->gso_segs, skb_shinfo(pBuf)->gso_type,
1505 skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, pBuf->ip_summed));
1506#ifndef VBOXNETFLT_SG_SUPPORT
1507 if (RT_LIKELY(fSrc & INTNETTRUNKDIR_HOST))
1508 {
1509 /*
1510 * skb_gso_segment does the following. Do we need to do it as well?
1511 */
1512# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1513 skb_reset_mac_header(pBuf);
1514 pBuf->mac_len = pBuf->network_header - pBuf->mac_header;
1515# else
1516 pBuf->mac.raw = pBuf->data;
1517 pBuf->mac_len = pBuf->nh.raw - pBuf->data;
1518# endif
1519 }
1520#endif /* !VBOXNETFLT_SG_SUPPORT */
1521# ifdef VBOXNETFLT_WITH_GSO_RECV
1522 if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP | SKB_GSO_TCPV6 | SKB_GSO_TCPV4))
1523 && vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1524 vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1525 else
1526# endif /* VBOXNETFLT_WITH_GSO_RECV */
1527 {
1528 /* Need to segment the packet */
1529 struct sk_buff *pNext;
1530 struct sk_buff *pSegment = skb_gso_segment(pBuf, 0 /*supported features*/);
1531 if (IS_ERR(pSegment))
1532 {
1533 LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1534 return;
1535 }
1536
1537 for (; pSegment; pSegment = pNext)
1538 {
1539 Log6(("vboxNetFltLinuxForwardToIntNetInner: segment len=%u data_len=%u truesize=%u next=%p"
1540 " nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x\n",
1541 pSegment->len, pSegment->data_len, pSegment->truesize, pSegment->next,
1542 skb_shinfo(pSegment)->nr_frags, skb_shinfo(pSegment)->gso_size,
1543 skb_shinfo(pSegment)->gso_segs, skb_shinfo(pSegment)->gso_type,
1544 skb_shinfo(pSegment)->frag_list, pSegment->pkt_type));
1545 pNext = pSegment->next;
1546 pSegment->next = 0;
1547 vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1548 dev_kfree_skb(pSegment);
1549 }
1550 }
1551 }
1552 else
1553#endif /* VBOXNETFLT_WITH_GSO */
1554 {
1555 Log6(("vboxNetFltLinuxForwardToIntNetInner: ptk_type=%d ip_summed=%d len=%d"
1556 " data_len=%d headroom=%d hdr_len=%d csum_offset=%d\n",
1557 pBuf->pkt_type, pBuf->ip_summed, pBuf->len, pBuf->data_len, skb_headroom(pBuf),
1558 skb_headlen(pBuf), skb_checksum_start_offset(pBuf)));
1559#ifndef VBOXNETFLT_SG_SUPPORT
1560 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1561 {
1562#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1563 /*
1564 * Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1565 * and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1566 * header length from the header itself and reconstruct 'h' pointer
1567 * to TCP (or whatever) header.
1568 */
1569 unsigned char *tmp = pBuf->h.raw;
1570 if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1571 pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1572#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1573 int rc = VBOX_SKB_CHECKSUM_HELP(pBuf);
1574#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1575 /* Restore the original (wrong) pointer. */
1576 pBuf->h.raw = tmp;
1577#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1578 if (rc) {
1579 LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1580 return;
1581 }
1582 }
1583#endif /* !VBOXNETFLT_SG_SUPPORT */
1584 vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1585 }
1586}
1587
1588
1589/**
1590 * Temporarily adjust pBuf->data so it always points to the Ethernet header,
1591 * then forward it to the internal network.
1592 *
1593 * @param pThis The net filter instance.
1594 * @param pBuf The socket buffer. This is consumed by this function.
1595 */
1596static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1597{
1598 uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1599
1600 if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1601 {
1602 /*
1603 * The packet came from the wire and the driver has already consumed
1604 * mac header. We need to restore it back. Moreover, after we are
1605 * through with this skb we need to restore its original state!
1606 */
1607 skb_push(pBuf, pBuf->mac_len);
1608 Log5(("vboxNetFltLinuxForwardToIntNet: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1609 pBuf->mac_len, pBuf->data, skb_mac_header(pBuf), skb_network_header(pBuf)));
1610 }
1611
1612 vboxNetFltLinuxForwardToIntNetInner(pThis, pBuf, fSrc);
1613
1614 /*
1615 * Restore the original state of skb as there are other handlers this skb
1616 * will be provided to.
1617 */
1618 if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1619 skb_pull(pBuf, pBuf->mac_len);
1620
1621 dev_kfree_skb(pBuf);
1622}
1623
1624
1625#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1626/**
1627 * Work queue handler that forwards the socket buffers queued by
1628 * vboxNetFltLinuxPacketHandler to the internal network.
1629 *
1630 * @param pWork The work queue.
1631 */
1632# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1633static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1634# else
1635static void vboxNetFltLinuxXmitTask(void *pWork)
1636# endif
1637{
1638 PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1639 struct sk_buff *pBuf;
1640
1641 Log6(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1642
1643 /*
1644 * Active? Retain the instance and increment the busy counter.
1645 */
1646 if (vboxNetFltTryRetainBusyActive(pThis))
1647 {
1648 while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1649 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1650
1651 vboxNetFltRelease(pThis, true /* fBusy */);
1652 }
1653 else
1654 {
1655 /** @todo Shouldn't we just drop the packets here? There is little point in
1656 * making them accumulate when the VM is paused and it'll only waste
1657 * kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1658 * before start draining the packets (goes for the intnet ring buf
1659 * too)? */
1660 }
1661}
1662#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1663
1664/**
1665 * Reports the GSO capabilities of the hardware NIC.
1666 *
1667 * @param pThis The net filter instance. The caller hold a
1668 * reference to this.
1669 */
1670static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1671{
1672#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1673 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1674 {
1675 struct net_device *pDev;
1676 PINTNETTRUNKSWPORT pSwitchPort;
1677 unsigned int fFeatures;
1678
1679 RTSpinlockAcquire(pThis->hSpinlock);
1680
1681 pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1682 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1683 if (pDev)
1684 fFeatures = pDev->features;
1685 else
1686 fFeatures = 0;
1687
1688 RTSpinlockRelease(pThis->hSpinlock);
1689
1690 if (pThis->pSwitchPort)
1691 {
1692 /* Set/update the GSO capabilities of the NIC. */
1693 uint32_t fGsoCapabilites = 0;
1694 if (fFeatures & NETIF_F_TSO)
1695 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1696 if (fFeatures & NETIF_F_TSO6)
1697 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1698# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1699 if (fFeatures & NETIF_F_UFO)
1700 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1701 if (fFeatures & NETIF_F_UFO)
1702 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1703# endif
1704 Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1705 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1706 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : "",
1707 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)) ? "ufo " : "",
1708 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)) ? "ufo6 " : ""));
1709 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1710 }
1711
1712 vboxNetFltRelease(pThis, true /*fBusy*/);
1713 }
1714#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1715}
1716
1717/**
1718 * Helper that determines whether the host (ignoreing us) is operating the
1719 * interface in promiscuous mode or not.
1720 */
1721static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1722{
1723 bool fRc = false;
1724 struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1725 if (pDev)
1726 {
1727 fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1728 LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1729 fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1730 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1731 }
1732 return fRc;
1733}
1734
1735/**
1736 * Does this device needs link state change signaled?
1737 * Currently we need it for our own VBoxNetAdp and TAP.
1738 */
1739static bool vboxNetFltNeedsLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev)
1740{
1741 if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1742 {
1743 struct ethtool_drvinfo Info;
1744
1745 memset(&Info, 0, sizeof(Info));
1746 Info.cmd = ETHTOOL_GDRVINFO;
1747 pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1748 Log3(("%s: driver=%.*s version=%.*s bus_info=%.*s\n",
1749 __FUNCTION__,
1750 sizeof(Info.driver), Info.driver,
1751 sizeof(Info.version), Info.version,
1752 sizeof(Info.bus_info), Info.bus_info));
1753
1754 if (!strncmp(Info.driver, "vboxnet", sizeof(Info.driver)))
1755 return true;
1756
1757#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36) /* TAP started doing carrier */
1758 return !strncmp(Info.driver, "tun", 4)
1759 && !strncmp(Info.bus_info, "tap", 4);
1760#endif
1761 }
1762
1763 return false;
1764}
1765
1766/**
1767 * Some devices need link state change when filter attaches/detaches
1768 * since the filter is their link in a sense.
1769 */
1770static void vboxNetFltSetLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1771{
1772 if (vboxNetFltNeedsLinkState(pThis, pDev))
1773 {
1774 Log3(("%s: bringing device link %s\n",
1775 __FUNCTION__, fLinkUp ? "up" : "down"));
1776 netif_tx_lock_bh(pDev);
1777 if (fLinkUp)
1778 netif_carrier_on(pDev);
1779 else
1780 netif_carrier_off(pDev);
1781 netif_tx_unlock_bh(pDev);
1782 }
1783}
1784
1785/**
1786 * Internal worker for vboxNetFltLinuxNotifierCallback.
1787 *
1788 * @returns VBox status code.
1789 * @param pThis The instance.
1790 * @param pDev The device to attach to.
1791 */
1792static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1793{
1794 LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1795
1796 /*
1797 * Retain and store the device.
1798 */
1799 dev_hold(pDev);
1800
1801 RTSpinlockAcquire(pThis->hSpinlock);
1802 ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1803 RTSpinlockRelease(pThis->hSpinlock);
1804
1805 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1806 pDev, pDev->name,
1807#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1808 netdev_refcnt_read(pDev)
1809#else
1810 atomic_read(&pDev->refcnt)
1811#endif
1812 ));
1813 Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1814 pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1815
1816 /* Get the mac address while we still have a valid net_device reference. */
1817 memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1818 /* Initialize MTU */
1819 pThis->u.s.cbMtu = pDev->mtu;
1820
1821 /*
1822 * Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1823 */
1824 pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1825 pThis->u.s.PacketType.dev = pDev;
1826 pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1827 dev_add_pack(&pThis->u.s.PacketType);
1828 ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1829 Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1830
1831#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1832 vboxNetFltLinuxHookDev(pThis, pDev);
1833#endif
1834
1835 /*
1836 * Are we the "carrier" for this device (e.g. vboxnet or tap)?
1837 */
1838 vboxNetFltSetLinkState(pThis, pDev, true);
1839
1840 /*
1841 * Set indicators that require the spinlock. Be abit paranoid about racing
1842 * the device notification handle.
1843 */
1844 RTSpinlockAcquire(pThis->hSpinlock);
1845 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1846 if (pDev)
1847 {
1848 ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1849 ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1850 pDev = NULL; /* don't dereference it */
1851 }
1852 RTSpinlockRelease(pThis->hSpinlock);
1853
1854 /*
1855 * If the above succeeded report GSO capabilities, if not undo and
1856 * release the device.
1857 */
1858 if (!pDev)
1859 {
1860 Assert(pThis->pSwitchPort);
1861 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1862 {
1863 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1864 pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1865 pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1866 pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE | INTNETTRUNKDIR_HOST);
1867 vboxNetFltRelease(pThis, true /*fBusy*/);
1868 }
1869 }
1870 else
1871 {
1872#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1873 vboxNetFltLinuxUnhookDev(pThis, pDev);
1874#endif
1875 RTSpinlockAcquire(pThis->hSpinlock);
1876 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1877 RTSpinlockRelease(pThis->hSpinlock);
1878 dev_put(pDev);
1879 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n",
1880 pDev, pDev->name,
1881#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1882 netdev_refcnt_read(pDev)
1883#else
1884 atomic_read(&pDev->refcnt)
1885#endif
1886 ));
1887 }
1888
1889 LogRel(("VBoxNetFlt: attached to '%s' / %RTmac\n", pThis->szName, &pThis->u.s.MacAddr));
1890 return VINF_SUCCESS;
1891}
1892
1893
1894static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1895{
1896 bool fRegistered;
1897 Assert(!pThis->fDisconnectedFromHost);
1898
1899#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1900 vboxNetFltLinuxUnhookDev(pThis, pDev);
1901#endif
1902
1903 if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1904 {
1905 dev_remove_pack(&pThis->u.s.PacketType);
1906 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1907 }
1908
1909 RTSpinlockAcquire(pThis->hSpinlock);
1910 fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1911 if (fRegistered)
1912 {
1913 ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1914 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1915 }
1916 RTSpinlockRelease(pThis->hSpinlock);
1917
1918 if (fRegistered)
1919 {
1920#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1921 skb_queue_purge(&pThis->u.s.XmitQueue);
1922#endif
1923 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1924 Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1925 pDev, pDev->name,
1926#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1927 netdev_refcnt_read(pDev)
1928#else
1929 atomic_read(&pDev->refcnt)
1930#endif
1931 ));
1932 dev_put(pDev);
1933 }
1934
1935 return NOTIFY_OK;
1936}
1937
1938static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1939{
1940 /* Check if we are not suspended and promiscuous mode has not been set. */
1941 if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1942 && !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1943 {
1944 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1945 dev_set_promiscuity(pDev, 1);
1946 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1947 Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1948 }
1949 else
1950 Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1951 return NOTIFY_OK;
1952}
1953
1954static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1955{
1956 /* Undo promiscuous mode if we has set it. */
1957 if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1958 {
1959 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1960 dev_set_promiscuity(pDev, -1);
1961 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1962 Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1963 }
1964 else
1965 Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1966 return NOTIFY_OK;
1967}
1968
1969/**
1970 * Callback for listening to MTU change event.
1971 *
1972 * We need to track changes of host's inteface MTU to discard over-sized frames
1973 * coming from the internal network as they may hang the TX queue of host's
1974 * adapter.
1975 *
1976 * @returns NOTIFY_OK
1977 * @param pThis The netfilter instance.
1978 * @param pDev Pointer to device structure of host's interface.
1979 */
1980static int vboxNetFltLinuxDeviceMtuChange(PVBOXNETFLTINS pThis, struct net_device *pDev)
1981{
1982 ASMAtomicWriteU32(&pThis->u.s.cbMtu, pDev->mtu);
1983 Log(("vboxNetFltLinuxDeviceMtuChange: set MTU for %s to %d\n", pThis->szName, pDev->mtu));
1984 return NOTIFY_OK;
1985}
1986
1987#ifdef LOG_ENABLED
1988/** Stringify the NETDEV_XXX constants. */
1989static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
1990{
1991 const char *pszEvent = "NETDEV_<unknown>";
1992 switch (ulEventType)
1993 {
1994 case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
1995 case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
1996 case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
1997 case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
1998 case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
1999 case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2000 case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2001 case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2002 case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2003 case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2004# ifdef NETDEV_FEAT_CHANGE
2005 case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2006# endif
2007 }
2008 return pszEvent;
2009}
2010#endif /* LOG_ENABLED */
2011
2012/**
2013 * Callback for listening to netdevice events.
2014 *
2015 * This works the rediscovery, clean up on unregistration, promiscuity on
2016 * up/down, and GSO feature changes from ethtool.
2017 *
2018 * @returns NOTIFY_OK
2019 * @param self Pointer to our notifier registration block.
2020 * @param ulEventType The event.
2021 * @param ptr Event specific, but it is usually the device it
2022 * relates to.
2023 */
2024static int vboxNetFltLinuxNotifierCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2025
2026{
2027 PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2028 struct net_device *pMyDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2029 struct net_device *pDev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2030 int rc = NOTIFY_OK;
2031
2032 Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2033 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, pMyDev));
2034
2035 if (ulEventType == NETDEV_REGISTER)
2036 {
2037#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 24) /* cgroups/namespaces introduced */
2038# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 26)
2039# define VBOX_DEV_NET(dev) dev_net(dev)
2040# define VBOX_NET_EQ(n1, n2) net_eq((n1), (n2))
2041# else
2042# define VBOX_DEV_NET(dev) ((dev)->nd_net)
2043# define VBOX_NET_EQ(n1, n2) ((n1) == (n2))
2044# endif
2045 struct net *pMyNet = current->nsproxy->net_ns;
2046 struct net *pDevNet = VBOX_DEV_NET(pDev);
2047
2048 if (VBOX_NET_EQ(pDevNet, pMyNet))
2049#endif /* namespaces */
2050 {
2051 if (strcmp(pDev->name, pThis->szName) == 0)
2052 {
2053 vboxNetFltLinuxAttachToInterface(pThis, pDev);
2054 }
2055 }
2056 }
2057 else
2058 {
2059 if (pDev == pMyDev)
2060 {
2061 switch (ulEventType)
2062 {
2063 case NETDEV_UNREGISTER:
2064 rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2065 break;
2066 case NETDEV_UP:
2067 rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2068 break;
2069 case NETDEV_GOING_DOWN:
2070 rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2071 break;
2072 case NETDEV_CHANGEMTU:
2073 rc = vboxNetFltLinuxDeviceMtuChange(pThis, pDev);
2074 break;
2075 case NETDEV_CHANGENAME:
2076 break;
2077#ifdef NETDEV_FEAT_CHANGE
2078 case NETDEV_FEAT_CHANGE:
2079 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2080 break;
2081#endif
2082 }
2083 }
2084 }
2085
2086 return rc;
2087}
2088
2089/*
2090 * Initial enumeration of netdevs. Called with NETDEV_REGISTER by
2091 * register_netdevice_notifier() under rtnl lock.
2092 */
2093static int vboxNetFltLinuxEnumeratorCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2094{
2095 PVBOXNETFLTINS pThis = ((PVBOXNETFLTNOTIFIER)self)->pThis;
2096 struct net_device *dev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2097 struct in_device *in_dev;
2098 struct inet6_dev *in6_dev;
2099
2100 if (ulEventType != NETDEV_REGISTER)
2101 return NOTIFY_OK;
2102
2103 if (RT_UNLIKELY(pThis->pSwitchPort->pfnNotifyHostAddress == NULL))
2104 return NOTIFY_OK;
2105
2106 /*
2107 * IPv4
2108 */
2109#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
2110 in_dev = __in_dev_get_rtnl(dev);
2111#else
2112 in_dev = __in_dev_get(dev);
2113#endif
2114 if (in_dev != NULL)
2115 {
2116 for_ifa(in_dev) {
2117 if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2118 return NOTIFY_OK;
2119
2120 if ( dev != pThis->u.s.pDev
2121 && VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2122 continue;
2123
2124 Log(("%s: %s: IPv4 addr %RTnaipv4 mask %RTnaipv4\n",
2125 __FUNCTION__, VBOX_NETDEV_NAME(dev),
2126 ifa->ifa_address, ifa->ifa_mask));
2127
2128 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2129 /* :fAdded */ true, kIntNetAddrType_IPv4, &ifa->ifa_address);
2130 } endfor_ifa(in_dev);
2131 }
2132
2133 /*
2134 * IPv6
2135 */
2136 in6_dev = __in6_dev_get(dev);
2137 if (in6_dev != NULL)
2138 {
2139 struct inet6_ifaddr *ifa;
2140
2141 read_lock_bh(&in6_dev->lock);
2142#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 35)
2143 list_for_each_entry(ifa, &in6_dev->addr_list, if_list)
2144#else
2145 for (ifa = in6_dev->addr_list; ifa != NULL; ifa = ifa->if_next)
2146#endif
2147 {
2148 if ( dev != pThis->u.s.pDev
2149 && ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_LOOPBACK))
2150 continue;
2151
2152 Log(("%s: %s: IPv6 addr %RTnaipv6/%u\n",
2153 __FUNCTION__, VBOX_NETDEV_NAME(dev),
2154 &ifa->addr, (unsigned)ifa->prefix_len));
2155
2156 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2157 /* :fAdded */ true, kIntNetAddrType_IPv6, &ifa->addr);
2158 }
2159 read_unlock_bh(&in6_dev->lock);
2160 }
2161
2162 return NOTIFY_OK;
2163}
2164
2165
2166static int vboxNetFltLinuxNotifierIPv4Callback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2167{
2168 PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv4);
2169 struct net_device *pDev, *pEventDev;
2170 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
2171 bool fMyDev;
2172 int rc = NOTIFY_OK;
2173
2174 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2175 pEventDev = ifa->ifa_dev->dev;
2176 fMyDev = (pDev == pEventDev);
2177 Log(("VBoxNetFlt: %s: IPv4 event %s(0x%lx) %s: addr %RTnaipv4 mask %RTnaipv4\n",
2178 pDev ? VBOX_NETDEV_NAME(pDev) : "<???>",
2179 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2180 pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<???>",
2181 ifa->ifa_address, ifa->ifa_mask));
2182
2183 if (pDev != NULL)
2184 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2185
2186 if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2187 return NOTIFY_OK;
2188
2189 if ( !fMyDev
2190 && VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2191 return NOTIFY_OK;
2192
2193 if (pThis->pSwitchPort->pfnNotifyHostAddress)
2194 {
2195 bool fAdded;
2196 if (ulEventType == NETDEV_UP)
2197 fAdded = true;
2198 else if (ulEventType == NETDEV_DOWN)
2199 fAdded = false;
2200 else
2201 return NOTIFY_OK;
2202
2203 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2204 kIntNetAddrType_IPv4, &ifa->ifa_local);
2205 }
2206
2207 return rc;
2208}
2209
2210
2211static int vboxNetFltLinuxNotifierIPv6Callback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2212{
2213 PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv6);
2214 struct net_device *pDev, *pEventDev;
2215 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
2216 bool fMyDev;
2217 int rc = NOTIFY_OK;
2218
2219 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2220 pEventDev = ifa->idev->dev;
2221 fMyDev = (pDev == pEventDev);
2222 Log(("VBoxNetFlt: %s: IPv6 event %s(0x%lx) %s: %RTnaipv6\n",
2223 pDev ? VBOX_NETDEV_NAME(pDev) : "<???>",
2224 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2225 pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<???>",
2226 &ifa->addr));
2227
2228 if (pDev != NULL)
2229 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2230
2231 if ( !fMyDev
2232 && ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_LOOPBACK))
2233 return NOTIFY_OK;
2234
2235 if (pThis->pSwitchPort->pfnNotifyHostAddress)
2236 {
2237 bool fAdded;
2238 if (ulEventType == NETDEV_UP)
2239 fAdded = true;
2240 else if (ulEventType == NETDEV_DOWN)
2241 fAdded = false;
2242 else
2243 return NOTIFY_OK;
2244
2245 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2246 kIntNetAddrType_IPv6, &ifa->addr);
2247 }
2248
2249 return rc;
2250}
2251
2252
2253bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2254{
2255 return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2256}
2257
2258int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2259{
2260 struct net_device * pDev;
2261 int err;
2262 int rc = VINF_SUCCESS;
2263 IPRT_LINUX_SAVE_EFL_AC();
2264 NOREF(pvIfData);
2265
2266 LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2267
2268 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2269 if (pDev)
2270 {
2271 /*
2272 * Create a sk_buff for the gather list and push it onto the wire.
2273 */
2274 if (fDst & INTNETTRUNKDIR_WIRE)
2275 {
2276 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2277 if (pBuf)
2278 {
2279 vboxNetFltDumpPacket(pSG, true, "wire", 1);
2280 Log6(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2281 Log6(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2282 err = dev_queue_xmit(pBuf);
2283 if (err)
2284 rc = RTErrConvertFromErrno(err);
2285 }
2286 else
2287 rc = VERR_NO_MEMORY;
2288 }
2289
2290 /*
2291 * Create a sk_buff for the gather list and push it onto the host stack.
2292 */
2293 if (fDst & INTNETTRUNKDIR_HOST)
2294 {
2295 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2296 if (pBuf)
2297 {
2298 vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2299 Log6(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2300 Log6(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2301 err = netif_rx_ni(pBuf);
2302 if (err)
2303 rc = RTErrConvertFromErrno(err);
2304 }
2305 else
2306 rc = VERR_NO_MEMORY;
2307 }
2308
2309 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2310 }
2311
2312 IPRT_LINUX_RESTORE_EFL_AC();
2313 return rc;
2314}
2315
2316
2317void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2318{
2319 struct net_device *pDev;
2320 IPRT_LINUX_SAVE_EFL_AC();
2321
2322 LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%RTbool, fDisablePromiscuous=%RTbool\n",
2323 pThis, pThis->szName, fActive, pThis->fDisablePromiscuous));
2324
2325 if (pThis->fDisablePromiscuous)
2326 return;
2327
2328 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2329 if (pDev)
2330 {
2331 /*
2332 * This api is a bit weird, the best reference is the code.
2333 *
2334 * Also, we have a bit or race conditions wrt the maintenance of
2335 * host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2336 */
2337#ifdef LOG_ENABLED
2338 u_int16_t fIf;
2339 unsigned const cPromiscBefore = pDev->promiscuity;
2340#endif
2341 if (fActive)
2342 {
2343 Assert(!pThis->u.s.fPromiscuousSet);
2344
2345 rtnl_lock();
2346 dev_set_promiscuity(pDev, 1);
2347 rtnl_unlock();
2348 pThis->u.s.fPromiscuousSet = true;
2349 Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2350 }
2351 else
2352 {
2353 if (pThis->u.s.fPromiscuousSet)
2354 {
2355 rtnl_lock();
2356 dev_set_promiscuity(pDev, -1);
2357 rtnl_unlock();
2358 Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2359 }
2360 pThis->u.s.fPromiscuousSet = false;
2361
2362#ifdef LOG_ENABLED
2363 fIf = dev_get_flags(pDev);
2364 Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2365#endif
2366 }
2367
2368 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2369 }
2370 IPRT_LINUX_RESTORE_EFL_AC();
2371}
2372
2373
2374int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2375{
2376 /*
2377 * Remove packet handler when we get disconnected from internal switch as
2378 * we don't want the handler to forward packets to disconnected switch.
2379 */
2380 if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
2381 {
2382 IPRT_LINUX_SAVE_EFL_AC();
2383 dev_remove_pack(&pThis->u.s.PacketType);
2384 Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
2385 IPRT_LINUX_RESTORE_EFL_AC();
2386 }
2387 return VINF_SUCCESS;
2388}
2389
2390
2391int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2392{
2393 IPRT_LINUX_SAVE_EFL_AC();
2394
2395 /*
2396 * Report the GSO capabilities of the host and device (if connected).
2397 * Note! No need to mark ourselves busy here.
2398 */
2399 /** @todo duplicate work here now? Attach */
2400#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2401 Log3(("vboxNetFltOsConnectIt: reporting host tso tso6 ufo\n"));
2402 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2403 0
2404 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2405 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2406 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2407# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2408 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2409# endif
2410 , INTNETTRUNKDIR_HOST);
2411
2412#endif
2413 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2414
2415 IPRT_LINUX_RESTORE_EFL_AC();
2416 return VINF_SUCCESS;
2417}
2418
2419
2420void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2421{
2422 struct net_device *pDev;
2423 bool fRegistered;
2424 IPRT_LINUX_SAVE_EFL_AC();
2425
2426#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
2427 vboxNetFltLinuxUnhookDev(pThis, NULL);
2428#endif
2429
2430 /** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2431 * unlikely, but none the less). Since it doesn't actually update the
2432 * state (just reads it), it is likely to panic in some interesting
2433 * ways. */
2434
2435 RTSpinlockAcquire(pThis->hSpinlock);
2436 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2437 fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
2438 RTSpinlockRelease(pThis->hSpinlock);
2439
2440 if (fRegistered)
2441 {
2442 vboxNetFltSetLinkState(pThis, pDev, false);
2443
2444#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2445 skb_queue_purge(&pThis->u.s.XmitQueue);
2446#endif
2447 Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
2448 Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
2449 pDev, pDev->name,
2450#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
2451 netdev_refcnt_read(pDev)
2452#else
2453 atomic_read(&pDev->refcnt)
2454#endif
2455 ));
2456 dev_put(pDev);
2457 }
2458
2459 unregister_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2460 unregister_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2461
2462 Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2463 unregister_netdevice_notifier(&pThis->u.s.Notifier);
2464 module_put(THIS_MODULE);
2465
2466 IPRT_LINUX_RESTORE_EFL_AC();
2467}
2468
2469
2470int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2471{
2472 int err;
2473 IPRT_LINUX_SAVE_EFL_AC();
2474 NOREF(pvContext);
2475
2476 pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2477 err = register_netdevice_notifier(&pThis->u.s.Notifier);
2478 if (err)
2479 {
2480 IPRT_LINUX_RESTORE_EFL_AC();
2481 return VERR_INTNET_FLT_IF_FAILED;
2482 }
2483 if (!pThis->u.s.fRegistered)
2484 {
2485 unregister_netdevice_notifier(&pThis->u.s.Notifier);
2486 LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2487 IPRT_LINUX_RESTORE_EFL_AC();
2488 return VERR_INTNET_FLT_IF_NOT_FOUND;
2489 }
2490
2491 Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2492 if ( pThis->fDisconnectedFromHost
2493 || !try_module_get(THIS_MODULE))
2494 {
2495 IPRT_LINUX_RESTORE_EFL_AC();
2496 return VERR_INTNET_FLT_IF_FAILED;
2497 }
2498
2499 if (pThis->pSwitchPort->pfnNotifyHostAddress)
2500 {
2501 VBOXNETFLTNOTIFIER Enumerator;
2502
2503 /*
2504 * register_inetaddr_notifier() and register_inet6addr_notifier()
2505 * do not call the callback for existing devices. Enumerating
2506 * all network devices explicitly is a bit of an ifdef mess,
2507 * so co-opt register_netdevice_notifier() to do that for us.
2508 */
2509 RT_ZERO(Enumerator);
2510 Enumerator.Notifier.notifier_call = vboxNetFltLinuxEnumeratorCallback;
2511 Enumerator.pThis = pThis;
2512
2513 err = register_netdevice_notifier(&Enumerator.Notifier);
2514 if (err)
2515 {
2516 LogRel(("%s: failed to enumerate network devices: error %d\n", __FUNCTION__, err));
2517 IPRT_LINUX_RESTORE_EFL_AC();
2518 return VINF_SUCCESS;
2519 }
2520
2521 unregister_netdevice_notifier(&Enumerator.Notifier);
2522
2523 pThis->u.s.NotifierIPv4.notifier_call = vboxNetFltLinuxNotifierIPv4Callback;
2524 err = register_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2525 if (err)
2526 LogRel(("%s: failed to register IPv4 notifier: error %d\n", __FUNCTION__, err));
2527
2528 pThis->u.s.NotifierIPv6.notifier_call = vboxNetFltLinuxNotifierIPv6Callback;
2529 err = register_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2530 if (err)
2531 LogRel(("%s: failed to register IPv6 notifier: error %d\n", __FUNCTION__, err));
2532 }
2533
2534 IPRT_LINUX_RESTORE_EFL_AC();
2535 return VINF_SUCCESS;
2536}
2537
2538int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2539{
2540 IPRT_LINUX_SAVE_EFL_AC();
2541
2542 /*
2543 * Init the linux specific members.
2544 */
2545 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2546 pThis->u.s.fRegistered = false;
2547 pThis->u.s.fPromiscuousSet = false;
2548 pThis->u.s.fPacketHandler = false;
2549 memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2550#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2551 skb_queue_head_init(&pThis->u.s.XmitQueue);
2552# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2553 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2554# else
2555 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2556# endif
2557#endif
2558
2559 IPRT_LINUX_RESTORE_EFL_AC();
2560 return VINF_SUCCESS;
2561}
2562
2563
2564void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2565{
2566 NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2567}
2568
2569
2570int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void *pvIf, void **pvIfData)
2571{
2572 /* Nothing to do */
2573 NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2574 return VINF_SUCCESS;
2575}
2576
2577
2578int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2579{
2580 /* Nothing to do */
2581 NOREF(pThis); NOREF(pvIfData);
2582 return VINF_SUCCESS;
2583}
2584
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