VirtualBox

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 82968

Last change on this file since 82968 was 82968, checked in by vboxsync, 5 years ago

Copyright year updates by scm.
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 92.1 KB
Line 
1/* $Id: VBoxNetFlt-linux.c 82968 2020-02-04 10:35:17Z vboxsync $ */
2/** @file
3 * VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4 */
5
6/*
7 * Copyright (C) 2006-2020 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 * The contents of this file may alternatively be used under the terms
18 * of the Common Development and Distribution License Version 1.0
19 * (CDDL) only, as it comes in the "COPYING.CDDL" file of the
20 * VirtualBox OSE distribution, in which case the provisions of the
21 * CDDL are applicable instead of those of the GPL.
22 *
23 * You may elect to license modified versions of this file under the
24 * terms and conditions of either the GPL or the CDDL or both.
25 */
26
27
28/*********************************************************************************************************************************
29* Header Files *
30*********************************************************************************************************************************/
31#define LOG_GROUP LOG_GROUP_NET_FLT_DRV
32#define VBOXNETFLT_LINUX_NO_XMIT_QUEUE
33#include "the-linux-kernel.h"
34#include "version-generated.h"
35#include "revision-generated.h"
36#include "product-generated.h"
37#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 24)
38#include <linux/nsproxy.h>
39#endif
40#include <linux/netdevice.h>
41#include <linux/etherdevice.h>
42#include <linux/rtnetlink.h>
43#include <linux/miscdevice.h>
44#include <linux/inetdevice.h>
45#include <linux/in.h>
46#include <linux/ip.h>
47#include <linux/if_vlan.h>
48#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 5, 0)
49#include <uapi/linux/pkt_cls.h>
50#endif
51#include <net/ipv6.h>
52#include <net/if_inet6.h>
53#include <net/addrconf.h>
54
55#include <VBox/log.h>
56#include <VBox/err.h>
57#include <VBox/intnetinline.h>
58#include <VBox/vmm/pdmnetinline.h>
59#include <VBox/param.h>
60#include <iprt/alloca.h>
61#include <iprt/assert.h>
62#include <iprt/spinlock.h>
63#include <iprt/semaphore.h>
64#include <iprt/initterm.h>
65#include <iprt/process.h>
66#include <iprt/mem.h>
67#include <iprt/net.h>
68#include <iprt/log.h>
69#include <iprt/mp.h>
70#include <iprt/mem.h>
71#include <iprt/time.h>
72
73#define VBOXNETFLT_OS_SPECFIC 1
74#include "../VBoxNetFltInternal.h"
75
76typedef struct VBOXNETFLTNOTIFIER {
77 struct notifier_block Notifier;
78 PVBOXNETFLTINS pThis;
79} VBOXNETFLTNOTIFIER;
80typedef struct VBOXNETFLTNOTIFIER *PVBOXNETFLTNOTIFIER;
81
82
83/*********************************************************************************************************************************
84* Defined Constants And Macros *
85*********************************************************************************************************************************/
86#define VBOX_FLT_NB_TO_INST(pNB) RT_FROM_MEMBER(pNB, VBOXNETFLTINS, u.s.Notifier)
87#define VBOX_FLT_PT_TO_INST(pPT) RT_FROM_MEMBER(pPT, VBOXNETFLTINS, u.s.PacketType)
88#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
89# define VBOX_FLT_XT_TO_INST(pXT) RT_FROM_MEMBER(pXT, VBOXNETFLTINS, u.s.XmitTask)
90#endif
91
92#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 11, 0)
93# define VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr) netdev_notifier_info_to_dev(ptr)
94#else
95# define VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr) ((struct net_device *)ptr)
96#endif
97
98#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
99# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(skb_frag_page(frag))
100# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr)
101#else
102# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 2, 0)
103# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(skb_frag_page(frag), KM_SKB_DATA_SOFTIRQ)
104# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ)
105# else
106# define VBOX_SKB_KMAP_FRAG(frag) kmap_atomic(frag->page, KM_SKB_DATA_SOFTIRQ)
107# define VBOX_SKB_KUNMAP_FRAG(vaddr) kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ)
108# endif
109#endif
110
111#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 34)
112# define VBOX_NETDEV_NAME(dev) netdev_name(dev)
113#else
114# define VBOX_NETDEV_NAME(dev) ((dev)->reg_state != NETREG_REGISTERED ? "(unregistered net_device)" : (dev)->name)
115#endif
116
117#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25)
118# define VBOX_IPV4_IS_LOOPBACK(addr) ipv4_is_loopback(addr)
119# define VBOX_IPV4_IS_LINKLOCAL_169(addr) ipv4_is_linklocal_169(addr)
120#else
121# define VBOX_IPV4_IS_LOOPBACK(addr) ((addr & htonl(IN_CLASSA_NET)) == htonl(0x7f000000))
122# define VBOX_IPV4_IS_LINKLOCAL_169(addr) ((addr & htonl(IN_CLASSB_NET)) == htonl(0xa9fe0000))
123#endif
124
125#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
126# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
127# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
128# define VBOX_SKB_CSUM_OFFSET(skb) skb->csum_offset
129#else
130# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
131# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
132# define VBOX_SKB_CSUM_OFFSET(skb) skb->csum
133#endif
134
135#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
136# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
137#else
138# define CHECKSUM_PARTIAL CHECKSUM_HW
139# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
140# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
141# else
142# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
143# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
144# else
145# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
146# endif
147/* Versions prior 2.6.10 use stats for both bstats and qstats */
148# define bstats stats
149# define qstats stats
150# endif
151#endif
152
153#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 20, 0)
154# define VBOX_HAVE_SKB_VLAN
155#else
156# ifdef RHEL_RELEASE_CODE
157# if (RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(7, 2) && RHEL_RELEASE_CODE < RHEL_RELEASE_VERSION(8, 0)) || \
158 (RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(6, 8) && RHEL_RELEASE_CODE < RHEL_RELEASE_VERSION(7, 0))
159# define VBOX_HAVE_SKB_VLAN
160# endif
161# endif
162#endif
163
164#ifdef VBOX_HAVE_SKB_VLAN
165# define vlan_tx_tag_get(skb) skb_vlan_tag_get(skb)
166# define vlan_tx_tag_present(skb) skb_vlan_tag_present(skb)
167#endif
168
169#ifndef NET_IP_ALIGN
170# define NET_IP_ALIGN 2
171#endif
172
173#if 1
174/** Create scatter / gather segments for fragments. When not used, we will
175 * linearize the socket buffer before creating the internal networking SG. */
176# define VBOXNETFLT_SG_SUPPORT 1
177#endif
178
179#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
180
181/** Indicates that the linux kernel may send us GSO frames. */
182# define VBOXNETFLT_WITH_GSO 1
183
184/** This enables or disables the transmitting of GSO frame from the internal
185 * network and to the host. */
186# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
187
188# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
189/** This enables or disables the transmitting of GSO frame from the internal
190 * network and to the wire. */
191# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
192# endif
193
194/** This enables or disables the forwarding/flooding of GSO frame from the host
195 * to the internal network. */
196# define VBOXNETFLT_WITH_GSO_RECV 1
197
198#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18) */
199
200#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
201/** This enables or disables handling of GSO frames coming from the wire (GRO). */
202# define VBOXNETFLT_WITH_GRO 1
203#endif
204
205/*
206 * GRO support was backported to RHEL 5.4
207 */
208#ifdef RHEL_RELEASE_CODE
209# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
210# define VBOXNETFLT_WITH_GRO 1
211# endif
212#endif
213
214
215/*********************************************************************************************************************************
216* Internal Functions *
217*********************************************************************************************************************************/
218static int __init VBoxNetFltLinuxInit(void);
219static void __exit VBoxNetFltLinuxUnload(void);
220static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
221
222
223/*********************************************************************************************************************************
224* Global Variables *
225*********************************************************************************************************************************/
226/**
227 * The (common) global data.
228 */
229static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
230
231module_init(VBoxNetFltLinuxInit);
232module_exit(VBoxNetFltLinuxUnload);
233
234MODULE_AUTHOR(VBOX_VENDOR);
235MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
236MODULE_LICENSE("GPL");
237#ifdef MODULE_VERSION
238MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV) " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
239#endif
240
241
242#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
243unsigned dev_get_flags(const struct net_device *dev)
244{
245 unsigned flags;
246
247 flags = (dev->flags & ~(IFF_PROMISC |
248 IFF_ALLMULTI |
249 IFF_RUNNING)) |
250 (dev->gflags & (IFF_PROMISC |
251 IFF_ALLMULTI));
252
253 if (netif_running(dev) && netif_carrier_ok(dev))
254 flags |= IFF_RUNNING;
255
256 return flags;
257}
258#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
259
260
261/**
262 * Initialize module.
263 *
264 * @returns appropriate status code.
265 */
266static int __init VBoxNetFltLinuxInit(void)
267{
268 int rc;
269 /*
270 * Initialize IPRT.
271 */
272 rc = RTR0Init(0);
273 if (RT_SUCCESS(rc))
274 {
275 Log(("VBoxNetFltLinuxInit\n"));
276
277 /*
278 * Initialize the globals and connect to the support driver.
279 *
280 * This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
281 * for establishing the connect to the support driver.
282 */
283 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
284 rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
285 if (RT_SUCCESS(rc))
286 {
287 LogRel(("VBoxNetFlt: Successfully started.\n"));
288 return 0;
289 }
290
291 LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
292 RTR0Term();
293 }
294 else
295 LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
296
297 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
298 return -RTErrConvertToErrno(rc);
299}
300
301
302/**
303 * Unload the module.
304 *
305 * @todo We have to prevent this if we're busy!
306 */
307static void __exit VBoxNetFltLinuxUnload(void)
308{
309 int rc;
310 Log(("VBoxNetFltLinuxUnload\n"));
311 Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
312
313 /*
314 * Undo the work done during start (in reverse order).
315 */
316 rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
317 AssertRC(rc); NOREF(rc);
318
319 RTR0Term();
320
321 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
322
323 Log(("VBoxNetFltLinuxUnload - done\n"));
324}
325
326
327/**
328 * We filter traffic from the host to the internal network
329 * before it reaches the NIC driver.
330 *
331 * The current code uses a very ugly hack overriding hard_start_xmit
332 * callback in the device structure, but it has been shown to give us a
333 * performance boost of 60-100% though. Eventually we have to find some
334 * less hacky way of getting this job done.
335 */
336#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
337
338#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
339
340# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
341
342# include <linux/ethtool.h>
343
344typedef struct ethtool_ops OVR_OPSTYPE;
345# define OVR_OPS ethtool_ops
346# define OVR_XMIT pfnStartXmit
347
348# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
349
350typedef struct net_device_ops OVR_OPSTYPE;
351# define OVR_OPS netdev_ops
352# define OVR_XMIT pOrgOps->ndo_start_xmit
353
354# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
355
356/**
357 * The overridden net_device_ops of the device we're attached to.
358 *
359 * As there is no net_device_ops structure in pre-2.6.29 kernels we override
360 * ethtool_ops instead along with hard_start_xmit callback in net_device
361 * structure.
362 *
363 * This is a very dirty hack that was created to explore how much we can improve
364 * the host to guest transfers by not CC'ing the NIC. It turns out to be
365 * the only way to filter outgoing packets for devices without TX queue.
366 */
367typedef struct VBoxNetDeviceOpsOverride
368{
369 /** Our overridden ops. */
370 OVR_OPSTYPE Ops;
371 /** Magic word. */
372 uint32_t u32Magic;
373 /** Pointer to the original ops. */
374 OVR_OPSTYPE const *pOrgOps;
375# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
376 /** Pointer to the original hard_start_xmit function. */
377 int (*pfnStartXmit)(struct sk_buff *pSkb, struct net_device *pDev);
378# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
379 /** Pointer to the net filter instance. */
380 PVBOXNETFLTINS pVBoxNetFlt;
381 /** The number of filtered packages. */
382 uint64_t cFiltered;
383 /** The total number of packets */
384 uint64_t cTotal;
385} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
386/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
387#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
388
389/**
390 * ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
391 * because they belong on the internal network.
392 *
393 * @returns NETDEV_TX_XXX.
394 * @param pSkb The socket buffer to transmit.
395 * @param pDev The net device.
396 */
397static int vboxNetFltLinuxStartXmitFilter(struct sk_buff *pSkb, struct net_device *pDev)
398{
399 PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
400 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
401 PCRTNETETHERHDR pEtherHdr;
402 PINTNETTRUNKSWPORT pSwitchPort;
403 uint32_t cbHdrs;
404
405
406 /*
407 * Validate the override structure.
408 *
409 * Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
410 * to be production quality code, we would have to be much more
411 * careful here and avoid the race.
412 */
413 if ( !VALID_PTR(pOverride)
414 || pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
415# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
416 || !VALID_PTR(pOverride->pOrgOps)
417# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
418 )
419 {
420 printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
421 dev_kfree_skb(pSkb);
422 return NETDEV_TX_OK;
423 }
424 pOverride->cTotal++;
425
426 /*
427 * Do the filtering base on the default OUI of our virtual NICs
428 *
429 * Note! In a real solution, we would ask the switch whether the
430 * destination MAC is 100% to be on the internal network and then
431 * drop it.
432 */
433 cbHdrs = skb_headlen(pSkb);
434 cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
435 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
436 if ( pEtherHdr
437 && VALID_PTR(pOverride->pVBoxNetFlt)
438 && (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
439 && VALID_PTR(pSwitchPort)
440 && cbHdrs >= 6)
441 {
442 INTNETSWDECISION enmDecision;
443
444 /** @todo consider reference counting, etc. */
445 enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
446 if (enmDecision == INTNETSWDECISION_INTNET)
447 {
448 dev_kfree_skb(pSkb);
449 pOverride->cFiltered++;
450 return NETDEV_TX_OK;
451 }
452 }
453
454 return pOverride->OVR_XMIT(pSkb, pDev);
455}
456
457/**
458 * Hooks the device ndo_start_xmit operation of the device.
459 *
460 * @param pThis The net filter instance.
461 * @param pDev The net device.
462 */
463static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
464{
465 PVBOXNETDEVICEOPSOVERRIDE pOverride;
466
467 /* Cancel override if ethtool_ops is missing (host-only case, @bugref{5712}) */
468 if (!VALID_PTR(pDev->OVR_OPS))
469 return;
470 pOverride = RTMemAlloc(sizeof(*pOverride));
471 if (!pOverride)
472 return;
473 pOverride->pOrgOps = pDev->OVR_OPS;
474 pOverride->Ops = *pDev->OVR_OPS;
475# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
476 pOverride->pfnStartXmit = pDev->hard_start_xmit;
477# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
478 pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
479# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
480 pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
481 pOverride->cTotal = 0;
482 pOverride->cFiltered = 0;
483 pOverride->pVBoxNetFlt = pThis;
484
485 RTSpinlockAcquire(pThis->hSpinlock); /* (this isn't necessary, but so what) */
486 ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
487# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
488 ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
489# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
490 RTSpinlockRelease(pThis->hSpinlock);
491}
492
493/**
494 * Undos what vboxNetFltLinuxHookDev did.
495 *
496 * @param pThis The net filter instance.
497 * @param pDev The net device. Can be NULL, in which case
498 * we'll try retrieve it from @a pThis.
499 */
500static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
501{
502 PVBOXNETDEVICEOPSOVERRIDE pOverride;
503
504 RTSpinlockAcquire(pThis->hSpinlock);
505 if (!pDev)
506 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
507 if (VALID_PTR(pDev))
508 {
509 pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
510 if ( VALID_PTR(pOverride)
511 && pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
512 && VALID_PTR(pOverride->pOrgOps)
513 )
514 {
515# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
516 ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
517# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
518 ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
519 ASMAtomicWriteU32(&pOverride->u32Magic, 0);
520 }
521 else
522 pOverride = NULL;
523 }
524 else
525 pOverride = NULL;
526 RTSpinlockRelease(pThis->hSpinlock);
527
528 if (pOverride)
529 {
530 printk("vboxnetflt: %llu out of %llu packets were not sent (directed to host)\n", pOverride->cFiltered, pOverride->cTotal);
531 RTMemFree(pOverride);
532 }
533}
534
535#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
536
537
538/**
539 * Reads and retains the host interface handle.
540 *
541 * @returns The handle, NULL if detached.
542 * @param pThis
543 */
544DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
545{
546#if 0
547 struct net_device *pDev = NULL;
548
549 Log(("vboxNetFltLinuxRetainNetDev\n"));
550 /*
551 * Be careful here to avoid problems racing the detached callback.
552 */
553 RTSpinlockAcquire(pThis->hSpinlock);
554 if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
555 {
556 pDev = (struct net_device *)ASMAtomicUoReadPtr((void * volatile *)&pThis->u.s.pDev);
557 if (pDev)
558 {
559 dev_hold(pDev);
560 Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
561 pDev, pDev->name,
562#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
563 netdev_refcnt_read(pDev)
564#else
565 atomic_read(&pDev->refcnt)
566#endif
567 ));
568 }
569 }
570 RTSpinlockRelease(pThis->hSpinlock);
571
572 Log(("vboxNetFltLinuxRetainNetDev - done\n"));
573 return pDev;
574#else
575 return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
576#endif
577}
578
579
580/**
581 * Release the host interface handle previously retained
582 * by vboxNetFltLinuxRetainNetDev.
583 *
584 * @param pThis The instance.
585 * @param pDev The vboxNetFltLinuxRetainNetDev
586 * return value, NULL is fine.
587 */
588DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
589{
590#if 0
591 Log(("vboxNetFltLinuxReleaseNetDev\n"));
592 NOREF(pThis);
593 if (pDev)
594 {
595 dev_put(pDev);
596 Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
597 pDev, pDev->name,
598#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
599 netdev_refcnt_read(pDev)
600#else
601 atomic_read(&pDev->refcnt)
602#endif
603 ));
604 }
605 Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
606#endif
607}
608
609#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 | (skb->dev->ifindex & 0xFFFF))
610#define VBOXNETFLT_SKB_TAG(skb) (*(uint32_t*)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
611
612/**
613 * Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
614 * i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
615 *
616 * @returns true / false accordingly.
617 * @param pBuf The sk_buff.
618 */
619DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
620{
621 return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
622}
623
624
625/**
626 * Checks whether this SG list contains a GSO packet.
627 *
628 * @returns true / false accordingly.
629 * @param pSG The (scatter/)gather list.
630 */
631DECLINLINE(bool) vboxNetFltLinuxIsGso(PINTNETSG pSG)
632{
633#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
634 return !((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type == PDMNETWORKGSOTYPE_INVALID);
635#else /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
636 return false;
637#endif /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
638}
639
640
641/**
642 * Find out the frame size (of a single segment in case of GSO frames).
643 *
644 * @returns the frame size.
645 * @param pSG The (scatter/)gather list.
646 */
647DECLINLINE(uint32_t) vboxNetFltLinuxFrameSize(PINTNETSG pSG)
648{
649 uint16_t u16Type = 0;
650 uint32_t cbVlanTag = 0;
651 if (pSG->aSegs[0].cb >= sizeof(RTNETETHERHDR))
652 u16Type = RT_BE2H_U16(((PCRTNETETHERHDR)pSG->aSegs[0].pv)->EtherType);
653 else if (pSG->cbTotal >= sizeof(RTNETETHERHDR))
654 {
655 uint32_t off = RT_UOFFSETOF(RTNETETHERHDR, EtherType);
656 uint32_t i;
657 for (i = 0; i < pSG->cSegsUsed; ++i)
658 {
659 if (off <= pSG->aSegs[i].cb)
660 {
661 if (off + sizeof(uint16_t) <= pSG->aSegs[i].cb)
662 u16Type = RT_BE2H_U16(*(uint16_t *)((uintptr_t)pSG->aSegs[i].pv + off));
663 else if (i + 1 < pSG->cSegsUsed)
664 u16Type = RT_BE2H_U16( ((uint16_t)( ((uint8_t *)pSG->aSegs[i].pv)[off] ) << 8)
665 + *(uint8_t *)pSG->aSegs[i + 1].pv); /* ASSUMES no empty segments! */
666 /* else: frame is too short. */
667 break;
668 }
669 off -= pSG->aSegs[i].cb;
670 }
671 }
672 if (u16Type == RTNET_ETHERTYPE_VLAN)
673 cbVlanTag = 4;
674 return (vboxNetFltLinuxIsGso(pSG) ? (uint32_t)pSG->GsoCtx.cbMaxSeg + pSG->GsoCtx.cbHdrsTotal : pSG->cbTotal) - cbVlanTag;
675}
676
677
678/**
679 * Internal worker that create a linux sk_buff for a
680 * (scatter/)gather list.
681 *
682 * @returns Pointer to the sk_buff.
683 * @param pThis The instance.
684 * @param pSG The (scatter/)gather list.
685 * @param fDstWire Set if the destination is the wire.
686 */
687static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
688{
689 struct sk_buff *pPkt;
690 struct net_device *pDev;
691#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
692 unsigned fGsoType = 0;
693#endif
694
695 if (pSG->cbTotal == 0)
696 {
697 LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
698 return NULL;
699 }
700 Log5(("VBoxNetFlt: Packet to %s of %d bytes (frame=%d).\n", fDstWire?"wire":"host", pSG->cbTotal, vboxNetFltLinuxFrameSize(pSG)));
701 if (fDstWire && (vboxNetFltLinuxFrameSize(pSG) > ASMAtomicReadU32(&pThis->u.s.cbMtu) + 14))
702 {
703 static bool s_fOnce = true;
704 if (s_fOnce)
705 {
706 s_fOnce = false;
707 printk("VBoxNetFlt: Dropped over-sized packet (%d bytes) coming from internal network.\n", vboxNetFltLinuxFrameSize(pSG));
708 }
709 return NULL;
710 }
711
712 /** @todo We should use fragments mapping the SG buffers with large packets.
713 * 256 bytes seems to be the a threshold used a lot for this. It
714 * requires some nasty work on the intnet side though... */
715 /*
716 * Allocate a packet and copy over the data.
717 */
718 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
719 pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
720 if (RT_UNLIKELY(!pPkt))
721 {
722 Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
723 pSG->pvUserData = NULL;
724 return NULL;
725 }
726 pPkt->dev = pDev;
727 pPkt->ip_summed = CHECKSUM_NONE;
728
729 /* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
730 skb_reserve(pPkt, NET_IP_ALIGN);
731
732 /* Copy the segments. */
733 skb_put(pPkt, pSG->cbTotal);
734 IntNetSgRead(pSG, pPkt->data);
735
736#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
737 /*
738 * Setup GSO if used by this packet.
739 */
740 switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
741 {
742 default:
743 AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
744 /* fall thru */
745 case PDMNETWORKGSOTYPE_INVALID:
746 fGsoType = 0;
747 break;
748 case PDMNETWORKGSOTYPE_IPV4_TCP:
749 fGsoType = SKB_GSO_TCPV4;
750 break;
751 case PDMNETWORKGSOTYPE_IPV6_TCP:
752 fGsoType = SKB_GSO_TCPV6;
753 break;
754 }
755 if (fGsoType)
756 {
757 struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
758
759 pShInfo->gso_type = fGsoType | SKB_GSO_DODGY;
760 pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
761 pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
762
763 /*
764 * We need to set checksum fields even if the packet goes to the host
765 * directly as it may be immediately forwarded by IP layer @bugref{5020}.
766 */
767 Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
768 pPkt->ip_summed = CHECKSUM_PARTIAL;
769# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
770 pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
771 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
772 pPkt->csum_offset = RT_UOFFSETOF(RTNETTCP, th_sum);
773 else
774 pPkt->csum_offset = RT_UOFFSETOF(RTNETUDP, uh_sum);
775# else
776 pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
777 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
778 pPkt->csum = RT_UOFFSETOF(RTNETTCP, th_sum);
779 else
780 pPkt->csum = RT_UOFFSETOF(RTNETUDP, uh_sum);
781# endif
782 if (!fDstWire)
783 PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
784 }
785#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE || VBOXNETFLT_WITH_GSO_XMIT_HOST */
786
787 /*
788 * Finish up the socket buffer.
789 */
790 pPkt->protocol = eth_type_trans(pPkt, pDev);
791 if (fDstWire)
792 {
793 VBOX_SKB_RESET_NETWORK_HDR(pPkt);
794
795 /* Restore ethernet header back. */
796 skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
797 VBOX_SKB_RESET_MAC_HDR(pPkt);
798 }
799 VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
800
801 return pPkt;
802}
803
804
805/**
806 * Return the offset where to start checksum computation from.
807 *
808 * @returns the offset relative to pBuf->data.
809 * @param pBuf The socket buffer.
810 */
811DECLINLINE(unsigned) vboxNetFltLinuxGetChecksumStartOffset(struct sk_buff *pBuf)
812{
813# if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 21)
814 unsigned char *pTransportHdr = pBuf->h.raw;
815# if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
816 /*
817 * Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
818 * and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
819 * header length from the header itself and reconstruct 'h' pointer
820 * to TCP (or whatever) header.
821 */
822 if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
823 pTransportHdr = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
824# endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
825 return pTransportHdr - pBuf->data;
826# else /* LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 21) */
827# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 38)
828 return pBuf->csum_start - skb_headroom(pBuf);
829# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 38) */
830 return skb_checksum_start_offset(pBuf);
831# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 38) */
832# endif /* LINUX_VERSION_CODE > KERNEL_VERSION(2, 6, 21) */
833}
834
835
836/**
837 * Initializes a SG list from an sk_buff.
838 *
839 * @returns Number of segments.
840 * @param pThis The instance.
841 * @param pBuf The sk_buff.
842 * @param pSG The SG.
843 * @param cbExtra The number of bytes of extra space allocated immediately after the SG.
844 * @param cSegs The number of segments allocated for the SG.
845 * This should match the number in the mbuf exactly!
846 * @param fSrc The source of the frame.
847 * @param pGsoCtx Pointer to the GSO context if it's a GSO
848 * internal network frame. NULL if regular frame.
849 */
850static void vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
851 unsigned cbExtra, unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
852{
853 int i;
854 NOREF(pThis);
855
856#ifndef VBOXNETFLT_SG_SUPPORT
857 Assert(!skb_shinfo(pBuf)->frag_list);
858#else /* VBOXNETFLT_SG_SUPPORT */
859 uint8_t *pExtra = (uint8_t *)&pSG->aSegs[cSegs];
860 unsigned cbConsumed = 0;
861 unsigned cbProduced = 0;
862
863# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
864 /* Restore VLAN tag stripped by host hardware */
865 if (vlan_tx_tag_present(pBuf))
866 {
867 uint8_t *pMac = pBuf->data;
868 struct vlan_ethhdr *pVHdr = (struct vlan_ethhdr *)pExtra;
869 Assert(ETH_ALEN * 2 + VLAN_HLEN <= cbExtra);
870 memmove(pVHdr, pMac, ETH_ALEN * 2);
871 /* Consume whole Ethernet header: 2 addresses + EtherType (see @bugref{8599}) */
872 cbConsumed += ETH_ALEN * 2 + sizeof(uint16_t);
873 pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
874 pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
875 pVHdr->h_vlan_encapsulated_proto = *(uint16_t*)(pMac + ETH_ALEN * 2);
876 cbProduced += VLAN_ETH_HLEN;
877 }
878# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
879
880 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
881 {
882 unsigned uCsumStartOffset = vboxNetFltLinuxGetChecksumStartOffset(pBuf);
883 unsigned uCsumStoreOffset = uCsumStartOffset + VBOX_SKB_CSUM_OFFSET(pBuf) - cbConsumed;
884 Log3(("cbConsumed=%u cbProduced=%u uCsumStartOffset=%u uCsumStoreOffset=%u\n",
885 cbConsumed, cbProduced, uCsumStartOffset, uCsumStoreOffset));
886 Assert(cbProduced + uCsumStoreOffset + sizeof(uint16_t) <= cbExtra);
887 /*
888 * We assume that the checksum is stored at the very end of the transport header
889 * so we will have all headers in a single fragment. If our assumption is wrong
890 * we may see suboptimal performance.
891 */
892 memmove(pExtra + cbProduced,
893 pBuf->data + cbConsumed,
894 uCsumStoreOffset);
895 unsigned uChecksum = skb_checksum(pBuf, uCsumStartOffset, pBuf->len - uCsumStartOffset, 0);
896 *(uint16_t*)(pExtra + cbProduced + uCsumStoreOffset) = csum_fold(uChecksum);
897 cbProduced += uCsumStoreOffset + sizeof(uint16_t);
898 cbConsumed += uCsumStoreOffset + sizeof(uint16_t);
899 }
900#endif /* VBOXNETFLT_SG_SUPPORT */
901
902 if (!pGsoCtx)
903 IntNetSgInitTempSegs(pSG, pBuf->len + cbProduced - cbConsumed, cSegs, 0 /*cSegsUsed*/);
904 else
905 IntNetSgInitTempSegsGso(pSG, pBuf->len + cbProduced - cbConsumed, cSegs, 0 /*cSegsUsed*/, pGsoCtx);
906
907 int iSeg = 0;
908#ifdef VBOXNETFLT_SG_SUPPORT
909 if (cbProduced)
910 {
911 pSG->aSegs[iSeg].cb = cbProduced;
912 pSG->aSegs[iSeg].pv = pExtra;
913 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
914 }
915 pSG->aSegs[iSeg].cb = skb_headlen(pBuf) - cbConsumed;
916 pSG->aSegs[iSeg].pv = pBuf->data + cbConsumed;
917 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
918 Assert(iSeg <= pSG->cSegsAlloc);
919
920# ifdef LOG_ENABLED
921 if (pBuf->data_len)
922 Log6((" kmap_atomic:"));
923# endif /* LOG_ENABLED */
924 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
925 {
926 skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
927# if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 4, 0)
928 pSG->aSegs[iSeg].cb = pFrag->bv_len;
929 pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->bv_offset;
930# else /* < KERNEL_VERSION(5, 4, 0) */
931 pSG->aSegs[iSeg].cb = pFrag->size;
932 pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
933# endif /* >= KERNEL_VERSION(5, 4, 0) */
934 Log6((" %p", pSG->aSegs[iSeg].pv));
935 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
936 Assert(iSeg <= pSG->cSegsAlloc);
937 }
938 struct sk_buff *pFragBuf;
939 for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
940 {
941 pSG->aSegs[iSeg].cb = skb_headlen(pFragBuf);
942 pSG->aSegs[iSeg].pv = pFragBuf->data;
943 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
944 Assert(iSeg <= pSG->cSegsAlloc);
945 for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
946 {
947 skb_frag_t *pFrag = &skb_shinfo(pFragBuf)->frags[i];
948# if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 4, 0)
949 pSG->aSegs[iSeg].cb = pFrag->bv_len;
950 pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->bv_offset;
951# else /* < KERNEL_VERSION(5, 4, 0) */
952 pSG->aSegs[iSeg].cb = pFrag->size;
953 pSG->aSegs[iSeg].pv = VBOX_SKB_KMAP_FRAG(pFrag) + pFrag->page_offset;
954# endif /* >= KERNEL_VERSION(5, 4, 0) */
955 Log6((" %p", pSG->aSegs[iSeg].pv));
956 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
957 Assert(iSeg <= pSG->cSegsAlloc);
958 }
959 }
960# ifdef LOG_ENABLED
961 if (pBuf->data_len)
962 Log6(("\n"));
963# endif /* LOG_ENABLED */
964#else
965 pSG->aSegs[iSeg].cb = pBuf->len;
966 pSG->aSegs[iSeg].pv = pBuf->data;
967 pSG->aSegs[iSeg++].Phys = NIL_RTHCPHYS;
968#endif
969
970 pSG->cSegsUsed = iSeg;
971
972#if 0
973 if (cbProduced)
974 {
975 LogRel(("vboxNetFltLinuxSkBufToSG: original packet dump:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
976 LogRel(("vboxNetFltLinuxSkBufToSG: cbConsumed=%u cbProduced=%u cbExtra=%u\n", cbConsumed, cbProduced, cbExtra));
977 uint32_t offset = 0;
978 for (i = 0; i < pSG->cSegsUsed; ++i)
979 {
980 LogRel(("vboxNetFltLinuxSkBufToSG: seg#%d (%d bytes, starting at 0x%x):\n%.*Rhxd\n",
981 i, pSG->aSegs[i].cb, offset, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
982 offset += pSG->aSegs[i].cb;
983 }
984 }
985#endif
986
987#ifdef PADD_RUNT_FRAMES_FROM_HOST
988 /*
989 * Add a trailer if the frame is too small.
990 *
991 * Since we're getting to the packet before it is framed, it has not
992 * yet been padded. The current solution is to add a segment pointing
993 * to a buffer containing all zeros and pray that works for all frames...
994 */
995 if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
996 {
997 Assert(pBuf->data_len == 0); /* Packets with fragments are never small! */
998 static uint8_t const s_abZero[128] = {0};
999
1000 AssertReturnVoid(iSeg < cSegs);
1001
1002 pSG->aSegs[iSeg].Phys = NIL_RTHCPHYS;
1003 pSG->aSegs[iSeg].pv = (void *)&s_abZero[0];
1004 pSG->aSegs[iSeg++].cb = 60 - pSG->cbTotal;
1005 pSG->cbTotal = 60;
1006 pSG->cSegsUsed++;
1007 Assert(iSeg <= pSG->cSegsAlloc)
1008 }
1009#endif
1010
1011 Log6(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
1012 pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
1013 for (i = 0; i < pSG->cSegsUsed; i++)
1014 Log6(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
1015 i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1016}
1017
1018/**
1019 * Packet handler; not really documented - figure it out yourself.
1020 *
1021 * @returns 0 or EJUSTRETURN - this is probably copy & pastry and thus wrong.
1022 */
1023#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
1024static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1025 struct net_device *pSkbDev,
1026 struct packet_type *pPacketType,
1027 struct net_device *pOrigDev)
1028#else
1029static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
1030 struct net_device *pSkbDev,
1031 struct packet_type *pPacketType)
1032#endif
1033{
1034 PVBOXNETFLTINS pThis;
1035 struct net_device *pDev;
1036 LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
1037 pBuf, pSkbDev, pPacketType));
1038#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1039 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",
1040 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));
1041# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1042 Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1043# endif
1044#else
1045 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",
1046 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));
1047#endif
1048 /*
1049 * Drop it immediately?
1050 */
1051 if (!pBuf)
1052 return 0;
1053
1054 if (pBuf->pkt_type == PACKET_LOOPBACK)
1055 {
1056 /*
1057 * We are not interested in loopbacked packets as they will always have
1058 * another copy going to the wire.
1059 */
1060 Log2(("vboxNetFltLinuxPacketHandler: dropped loopback packet (cb=%u)\n", pBuf->len));
1061 dev_kfree_skb(pBuf); /* We must 'consume' all packets we get (@bugref{6539})! */
1062 return 0;
1063 }
1064
1065 pThis = VBOX_FLT_PT_TO_INST(pPacketType);
1066 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1067 if (pDev != pSkbDev)
1068 {
1069 Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
1070 kfree_skb(pBuf); /* This is a failure, so we use kfree_skb instead of dev_kfree_skb. */
1071 return 0;
1072 }
1073
1074 Log6(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
1075 if (vboxNetFltLinuxSkBufIsOur(pBuf))
1076 {
1077 Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
1078 dev_kfree_skb(pBuf);
1079 return 0;
1080 }
1081
1082#ifndef VBOXNETFLT_SG_SUPPORT
1083 {
1084 /*
1085 * Get rid of fragmented packets, they cause too much trouble.
1086 */
1087 unsigned int uMacLen = pBuf->mac_len;
1088 struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
1089 dev_kfree_skb(pBuf);
1090 if (!pCopy)
1091 {
1092 LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
1093 return 0;
1094 }
1095 pBuf = pCopy;
1096 /* Somehow skb_copy ignores mac_len */
1097 pBuf->mac_len = uMacLen;
1098# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
1099 /* Restore VLAN tag stripped by host hardware */
1100 if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
1101 {
1102 uint8_t *pMac = (uint8_t*)skb_mac_header(pBuf);
1103 struct vlan_ethhdr *pVHdr = (struct vlan_ethhdr *)(pMac - VLAN_HLEN);
1104 memmove(pVHdr, pMac, ETH_ALEN * 2);
1105 pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
1106 pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
1107 pBuf->mac_header -= VLAN_HLEN;
1108 pBuf->mac_len += VLAN_HLEN;
1109 }
1110# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
1111
1112# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
1113 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",
1114 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));
1115# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1116 Log6(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
1117# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22) */
1118# else /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 18) */
1119 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",
1120 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));
1121# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 18) */
1122 }
1123#endif /* !VBOXNETFLT_SG_SUPPORT */
1124
1125#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1126 /* Forward it to the internal network. */
1127 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1128#else /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1129 /* Add the packet to transmit queue and schedule the bottom half. */
1130 skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
1131 schedule_work(&pThis->u.s.XmitTask);
1132 Log6(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
1133 &pThis->u.s.XmitTask, pBuf));
1134#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1135
1136 /* It does not really matter what we return, it is ignored by the kernel. */
1137 return 0;
1138}
1139
1140/**
1141 * Calculate the number of INTNETSEG segments the socket buffer will need.
1142 *
1143 * @returns Segment count.
1144 * @param pBuf The socket buffer.
1145 * @param pcbTemp Where to store the number of bytes of the part
1146 * of the socket buffer that will be copied to
1147 * a temporary storage.
1148 */
1149DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf, unsigned *pcbTemp)
1150{
1151 *pcbTemp = 0;
1152#ifdef VBOXNETFLT_SG_SUPPORT
1153 unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
1154 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1155 {
1156 *pcbTemp = vboxNetFltLinuxGetChecksumStartOffset(pBuf) + VBOX_SKB_CSUM_OFFSET(pBuf) + sizeof(uint16_t);
1157 }
1158# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
1159 if (vlan_tx_tag_present(pBuf))
1160 {
1161 if (*pcbTemp)
1162 *pcbTemp += VLAN_HLEN;
1163 else
1164 *pcbTemp = VLAN_ETH_HLEN;
1165 }
1166# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
1167 if (*pcbTemp)
1168 ++cSegs;
1169 struct sk_buff *pFrag;
1170 for (pFrag = skb_shinfo(pBuf)->frag_list; pFrag; pFrag = pFrag->next)
1171 {
1172 Log6(("vboxNetFltLinuxCalcSGSegments: frag=%p len=%d data_len=%d frags=%d frag_list=%p next=%p\n",
1173 pFrag, pFrag->len, pFrag->data_len, skb_shinfo(pFrag)->nr_frags, skb_shinfo(pFrag)->frag_list, pFrag->next));
1174 cSegs += 1 + skb_shinfo(pFrag)->nr_frags;
1175 }
1176#else
1177 unsigned cSegs = 1;
1178#endif
1179#ifdef PADD_RUNT_FRAMES_FROM_HOST
1180 /* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
1181 if (pBuf->len < 60)
1182 cSegs++;
1183#endif
1184 return cSegs;
1185}
1186
1187
1188/**
1189 * Destroy the intnet scatter / gather buffer created by
1190 * vboxNetFltLinuxSkBufToSG.
1191 *
1192 * @param pSG The (scatter/)gather list.
1193 * @param pBuf The original socket buffer that was used to create
1194 * the scatter/gather list.
1195 */
1196static void vboxNetFltLinuxDestroySG(PINTNETSG pSG, struct sk_buff *pBuf)
1197{
1198#ifdef VBOXNETFLT_SG_SUPPORT
1199 int i, iSeg = 1; /* Skip non-paged part of SKB */
1200 /* Check if the extra buffer behind SG structure was used for modified packet header */
1201 if (pBuf->data != pSG->aSegs[0].pv)
1202 ++iSeg; /* Skip it as well */
1203# ifdef LOG_ENABLED
1204 if (pBuf->data_len)
1205 Log6(("kunmap_atomic:"));
1206# endif /* LOG_ENABLED */
1207 /* iSeg now points to the first mapped fragment if there are any */
1208 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1209 {
1210 Log6((" %p", pSG->aSegs[iSeg].pv));
1211 VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1212 }
1213 struct sk_buff *pFragBuf;
1214 for (pFragBuf = skb_shinfo(pBuf)->frag_list; pFragBuf; pFragBuf = pFragBuf->next)
1215 {
1216 ++iSeg; /* Non-fragment (unmapped) portion of chained SKB */
1217 for (i = 0; i < skb_shinfo(pFragBuf)->nr_frags; i++)
1218 {
1219 Log6((" %p", pSG->aSegs[iSeg].pv));
1220 VBOX_SKB_KUNMAP_FRAG(pSG->aSegs[iSeg++].pv);
1221 }
1222 }
1223# ifdef LOG_ENABLED
1224 if (pBuf->data_len)
1225 Log6(("\n"));
1226# endif /* LOG_ENABLED */
1227#endif
1228 NOREF(pSG);
1229}
1230
1231#ifdef LOG_ENABLED
1232/**
1233 * Logging helper.
1234 */
1235static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1236{
1237 int i, offSeg;
1238 uint8_t *pInt, *pExt;
1239 static int iPacketNo = 1;
1240 iPacketNo += iIncrement;
1241 if (fEgress)
1242 {
1243 pExt = pSG->aSegs[0].pv;
1244 pInt = pExt + 6;
1245 }
1246 else
1247 {
1248 pInt = pSG->aSegs[0].pv;
1249 pExt = pInt + 6;
1250 }
1251 Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1252 " %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1253 pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1254 fEgress ? "-->" : "<--", pszWhere,
1255 pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1256 pSG->cbTotal, iPacketNo));
1257 if (pSG->cSegsUsed == 1)
1258 {
1259 Log4(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1260 }
1261 else
1262 {
1263 for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
1264 {
1265 Log4(("-- segment %d at 0x%x (%d bytes)\n --\n%.*Rhxd\n",
1266 i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1267 offSeg += pSG->aSegs[i].cb;
1268 }
1269 }
1270}
1271#else
1272# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1273#endif
1274
1275#ifdef VBOXNETFLT_WITH_GSO_RECV
1276
1277/**
1278 * Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1279 * GSO socket buffer without having to segment it.
1280 *
1281 * @returns true on success, false if needs segmenting.
1282 * @param pThis The net filter instance.
1283 * @param pSkb The GSO socket buffer.
1284 * @param fSrc The source.
1285 * @param pGsoCtx Where to return the GSO context on success.
1286 */
1287static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1288 PPDMNETWORKGSO pGsoCtx)
1289{
1290 PDMNETWORKGSOTYPE enmGsoType;
1291 uint16_t uEtherType;
1292 unsigned int cbTransport;
1293 unsigned int offTransport;
1294 unsigned int cbTransportHdr;
1295 unsigned uProtocol;
1296 union
1297 {
1298 RTNETIPV4 IPv4;
1299 RTNETIPV6 IPv6;
1300 RTNETTCP Tcp;
1301 uint8_t ab[40];
1302 uint16_t au16[40/2];
1303 uint32_t au32[40/4];
1304 } Buf;
1305
1306 /*
1307 * Check the GSO properties of the socket buffer and make sure it fits.
1308 */
1309 /** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1310 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_DODGY | SKB_GSO_TCPV6 | SKB_GSO_TCPV4) ))
1311 {
1312 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1313 return false;
1314 }
1315 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1316 || pSkb->len > VBOX_MAX_GSO_SIZE ))
1317 {
1318 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1319 return false;
1320 }
1321
1322 /*
1323 * Switch on the ethertype.
1324 */
1325 uEtherType = pSkb->protocol;
1326 if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1327 && pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1328 {
1329 uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1330 if (puEtherType)
1331 uEtherType = *puEtherType;
1332 }
1333 switch (uEtherType)
1334 {
1335 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1336 {
1337 unsigned int cbHdr;
1338 PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1339 if (RT_UNLIKELY(!pIPv4))
1340 {
1341 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1342 return false;
1343 }
1344
1345 cbHdr = pIPv4->ip_hl * 4;
1346 cbTransport = RT_N2H_U16(pIPv4->ip_len);
1347 if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1348 || cbHdr > cbTransport ))
1349 {
1350 Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1351 return false;
1352 }
1353 cbTransport -= cbHdr;
1354 offTransport = pSkb->mac_len + cbHdr;
1355 uProtocol = pIPv4->ip_p;
1356 if (uProtocol == RTNETIPV4_PROT_TCP)
1357 enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1358 else if (uProtocol == RTNETIPV4_PROT_UDP)
1359 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1360 else /** @todo IPv6: 4to6 tunneling */
1361 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1362 break;
1363 }
1364
1365 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1366 {
1367 PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1368 if (RT_UNLIKELY(!pIPv6))
1369 {
1370 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1371 return false;
1372 }
1373
1374 cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1375 offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1376 uProtocol = pIPv6->ip6_nxt;
1377 /** @todo IPv6: Dig our way out of the other headers. */
1378 if (uProtocol == RTNETIPV4_PROT_TCP)
1379 enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1380 else if (uProtocol == RTNETIPV4_PROT_UDP)
1381 enmGsoType = PDMNETWORKGSOTYPE_IPV6_UDP;
1382 else
1383 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1384 break;
1385 }
1386
1387 default:
1388 Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1389 return false;
1390 }
1391
1392 if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1393 {
1394 Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1395 return false;
1396 }
1397
1398 if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1399 || offTransport + cbTransport > pSkb->len
1400 || cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1401 {
1402 Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1403 offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1404 return false;
1405 }
1406
1407 /*
1408 * Check the TCP/UDP bits.
1409 */
1410 if (uProtocol == RTNETIPV4_PROT_TCP)
1411 {
1412 PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1413 if (RT_UNLIKELY(!pTcp))
1414 {
1415 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1416 return false;
1417 }
1418
1419 cbTransportHdr = pTcp->th_off * 4;
1420 pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1421 if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1422 || cbTransportHdr > cbTransport
1423 || offTransport + cbTransportHdr >= UINT8_MAX
1424 || offTransport + cbTransportHdr >= pSkb->len ))
1425 {
1426 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1427 return false;
1428 }
1429
1430 }
1431 else
1432 {
1433 Assert(uProtocol == RTNETIPV4_PROT_UDP);
1434 cbTransportHdr = sizeof(RTNETUDP);
1435 pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1436 if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1437 || offTransport + cbTransportHdr >= pSkb->len ))
1438 {
1439 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1440 return false;
1441 }
1442 }
1443
1444 /*
1445 * We're good, init the GSO context.
1446 */
1447 pGsoCtx->u8Type = enmGsoType;
1448 pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1449 pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1450 pGsoCtx->offHdr1 = pSkb->mac_len;
1451 pGsoCtx->offHdr2 = offTransport;
1452 pGsoCtx->u8Unused = 0;
1453
1454 return true;
1455}
1456
1457/**
1458 * Forward the socket buffer as a GSO internal network frame.
1459 *
1460 * @returns IPRT status code.
1461 * @param pThis The net filter instance.
1462 * @param pSkb The GSO socket buffer.
1463 * @param fSrc The source.
1464 * @param pGsoCtx Where to return the GSO context on success.
1465 */
1466static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1467{
1468 int rc;
1469 unsigned cbExtra;
1470 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb, &cbExtra);
1471 PINTNETSG pSG = (PINTNETSG)alloca(RT_UOFFSETOF_DYN(INTNETSG, aSegs[cSegs]) + cbExtra);
1472 if (RT_LIKELY(pSG))
1473 {
1474 vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cbExtra, cSegs, fSrc, pGsoCtx);
1475
1476 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1477 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1478
1479 vboxNetFltLinuxDestroySG(pSG, pSkb);
1480 rc = VINF_SUCCESS;
1481 }
1482 else
1483 {
1484 Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1485 rc = VERR_NO_MEMORY;
1486 }
1487 return rc;
1488}
1489
1490#endif /* VBOXNETFLT_WITH_GSO_RECV */
1491
1492/**
1493 * Worker for vboxNetFltLinuxForwardToIntNet.
1494 *
1495 * @returns VINF_SUCCESS or VERR_NO_MEMORY.
1496 * @param pThis The net filter instance.
1497 * @param pBuf The socket buffer.
1498 * @param fSrc The source.
1499 */
1500static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1501{
1502 int rc;
1503 unsigned cbExtra;
1504 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf, &cbExtra);
1505 PINTNETSG pSG = (PINTNETSG)alloca(RT_UOFFSETOF_DYN(INTNETSG, aSegs[cSegs]) + cbExtra);
1506 if (RT_LIKELY(pSG))
1507 {
1508 vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cbExtra, cSegs, fSrc, NULL /*pGsoCtx*/);
1509
1510 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1511 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1512
1513 vboxNetFltLinuxDestroySG(pSG, pBuf);
1514 rc = VINF_SUCCESS;
1515 }
1516 else
1517 {
1518 Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1519 rc = VERR_NO_MEMORY;
1520 }
1521 return rc;
1522}
1523
1524
1525/**
1526 * I won't disclose what I do, figure it out yourself, including pThis referencing.
1527 *
1528 * @param pThis The net filter instance.
1529 * @param pBuf The socket buffer.
1530 * @param fSrc Where the packet comes from.
1531 */
1532static void vboxNetFltLinuxForwardToIntNetInner(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1533{
1534#ifdef VBOXNETFLT_WITH_GSO
1535 if (skb_is_gso(pBuf))
1536 {
1537 PDMNETWORKGSO GsoCtx;
1538 Log6(("vboxNetFltLinuxForwardToIntNetInner: skb len=%u data_len=%u truesize=%u next=%p"
1539 " nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x ip_summed=%d\n",
1540 pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next,
1541 skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->gso_size,
1542 skb_shinfo(pBuf)->gso_segs, skb_shinfo(pBuf)->gso_type,
1543 skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, pBuf->ip_summed));
1544#ifndef VBOXNETFLT_SG_SUPPORT
1545 if (RT_LIKELY(fSrc & INTNETTRUNKDIR_HOST))
1546 {
1547 /*
1548 * skb_gso_segment does the following. Do we need to do it as well?
1549 */
1550# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1551 skb_reset_mac_header(pBuf);
1552 pBuf->mac_len = pBuf->network_header - pBuf->mac_header;
1553# else
1554 pBuf->mac.raw = pBuf->data;
1555 pBuf->mac_len = pBuf->nh.raw - pBuf->data;
1556# endif
1557 }
1558#endif /* !VBOXNETFLT_SG_SUPPORT */
1559# ifdef VBOXNETFLT_WITH_GSO_RECV
1560 if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_TCPV6 | SKB_GSO_TCPV4))
1561 && vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1562 vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1563 else
1564# endif /* VBOXNETFLT_WITH_GSO_RECV */
1565 {
1566 /* Need to segment the packet */
1567 struct sk_buff *pNext;
1568 struct sk_buff *pSegment = skb_gso_segment(pBuf, 0 /*supported features*/);
1569 if (IS_ERR(pSegment))
1570 {
1571 LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1572 return;
1573 }
1574
1575 for (; pSegment; pSegment = pNext)
1576 {
1577 Log6(("vboxNetFltLinuxForwardToIntNetInner: segment len=%u data_len=%u truesize=%u next=%p"
1578 " nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x\n",
1579 pSegment->len, pSegment->data_len, pSegment->truesize, pSegment->next,
1580 skb_shinfo(pSegment)->nr_frags, skb_shinfo(pSegment)->gso_size,
1581 skb_shinfo(pSegment)->gso_segs, skb_shinfo(pSegment)->gso_type,
1582 skb_shinfo(pSegment)->frag_list, pSegment->pkt_type));
1583 pNext = pSegment->next;
1584 pSegment->next = 0;
1585 vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1586 dev_kfree_skb(pSegment);
1587 }
1588 }
1589 }
1590 else
1591#endif /* VBOXNETFLT_WITH_GSO */
1592 {
1593 Log6(("vboxNetFltLinuxForwardToIntNetInner: ptk_type=%d ip_summed=%d len=%d"
1594 " data_len=%d headroom=%d hdr_len=%d csum_offset=%d\n",
1595 pBuf->pkt_type, pBuf->ip_summed, pBuf->len, pBuf->data_len, skb_headroom(pBuf),
1596 skb_headlen(pBuf), vboxNetFltLinuxGetChecksumStartOffset(pBuf)));
1597#ifndef VBOXNETFLT_SG_SUPPORT
1598 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1599 {
1600#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1601 /*
1602 * Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1603 * and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1604 * header length from the header itself and reconstruct 'h' pointer
1605 * to TCP (or whatever) header.
1606 */
1607 unsigned char *tmp = pBuf->h.raw;
1608 if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1609 pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1610#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1611 int rc = VBOX_SKB_CHECKSUM_HELP(pBuf);
1612#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1613 /* Restore the original (wrong) pointer. */
1614 pBuf->h.raw = tmp;
1615#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1616 if (rc) {
1617 LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1618 return;
1619 }
1620 }
1621#endif /* !VBOXNETFLT_SG_SUPPORT */
1622 vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1623 }
1624}
1625
1626
1627/**
1628 * Temporarily adjust pBuf->data so it always points to the Ethernet header,
1629 * then forward it to the internal network.
1630 *
1631 * @param pThis The net filter instance.
1632 * @param pBuf The socket buffer. This is consumed by this function.
1633 */
1634static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1635{
1636 uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1637
1638 if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1639 {
1640 /*
1641 * The packet came from the wire and the driver has already consumed
1642 * mac header. We need to restore it back. Moreover, after we are
1643 * through with this skb we need to restore its original state!
1644 */
1645 skb_push(pBuf, pBuf->mac_len);
1646 Log5(("vboxNetFltLinuxForwardToIntNet: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1647 pBuf->mac_len, pBuf->data, skb_mac_header(pBuf), skb_network_header(pBuf)));
1648 }
1649
1650 vboxNetFltLinuxForwardToIntNetInner(pThis, pBuf, fSrc);
1651
1652 /*
1653 * Restore the original state of skb as there are other handlers this skb
1654 * will be provided to.
1655 */
1656 if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1657 skb_pull(pBuf, pBuf->mac_len);
1658
1659 dev_kfree_skb(pBuf);
1660}
1661
1662
1663#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1664/**
1665 * Work queue handler that forwards the socket buffers queued by
1666 * vboxNetFltLinuxPacketHandler to the internal network.
1667 *
1668 * @param pWork The work queue.
1669 */
1670# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1671static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1672# else
1673static void vboxNetFltLinuxXmitTask(void *pWork)
1674# endif
1675{
1676 PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1677 struct sk_buff *pBuf;
1678
1679 Log6(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1680
1681 /*
1682 * Active? Retain the instance and increment the busy counter.
1683 */
1684 if (vboxNetFltTryRetainBusyActive(pThis))
1685 {
1686 while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1687 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1688
1689 vboxNetFltRelease(pThis, true /* fBusy */);
1690 }
1691 else
1692 {
1693 /** @todo Shouldn't we just drop the packets here? There is little point in
1694 * making them accumulate when the VM is paused and it'll only waste
1695 * kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1696 * before start draining the packets (goes for the intnet ring buf
1697 * too)? */
1698 }
1699}
1700#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1701
1702/**
1703 * Reports the GSO capabilities of the hardware NIC.
1704 *
1705 * @param pThis The net filter instance. The caller hold a
1706 * reference to this.
1707 */
1708static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1709{
1710#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1711 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1712 {
1713 struct net_device *pDev;
1714 PINTNETTRUNKSWPORT pSwitchPort;
1715 unsigned int fFeatures;
1716
1717 RTSpinlockAcquire(pThis->hSpinlock);
1718
1719 pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1720 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1721 if (pDev)
1722 fFeatures = pDev->features;
1723 else
1724 fFeatures = 0;
1725
1726 RTSpinlockRelease(pThis->hSpinlock);
1727
1728 if (pThis->pSwitchPort)
1729 {
1730 /* Set/update the GSO capabilities of the NIC. */
1731 uint32_t fGsoCapabilites = 0;
1732 if (fFeatures & NETIF_F_TSO)
1733 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1734 if (fFeatures & NETIF_F_TSO6)
1735 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1736 Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1737 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1738 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : ""));
1739 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1740 }
1741
1742 vboxNetFltRelease(pThis, true /*fBusy*/);
1743 }
1744#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1745}
1746
1747/**
1748 * Helper that determines whether the host (ignoreing us) is operating the
1749 * interface in promiscuous mode or not.
1750 */
1751static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1752{
1753 bool fRc = false;
1754 struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1755 if (pDev)
1756 {
1757 fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1758 LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1759 fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1760 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1761 }
1762 return fRc;
1763}
1764
1765/**
1766 * Does this device needs link state change signaled?
1767 * Currently we need it for our own VBoxNetAdp and TAP.
1768 */
1769static bool vboxNetFltNeedsLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev)
1770{
1771 if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1772 {
1773 struct ethtool_drvinfo Info;
1774
1775 memset(&Info, 0, sizeof(Info));
1776 Info.cmd = ETHTOOL_GDRVINFO;
1777 pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1778 Log3(("%s: driver=%.*s version=%.*s bus_info=%.*s\n",
1779 __FUNCTION__,
1780 sizeof(Info.driver), Info.driver,
1781 sizeof(Info.version), Info.version,
1782 sizeof(Info.bus_info), Info.bus_info));
1783
1784 if (!strncmp(Info.driver, "vboxnet", sizeof(Info.driver)))
1785 return true;
1786
1787#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36) /* TAP started doing carrier */
1788 return !strncmp(Info.driver, "tun", 4)
1789 && !strncmp(Info.bus_info, "tap", 4);
1790#endif
1791 }
1792
1793 return false;
1794}
1795
1796#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 18)
1797DECLINLINE(void) netif_tx_lock_bh(struct net_device *pDev)
1798{
1799 spin_lock_bh(&pDev->xmit_lock);
1800}
1801
1802DECLINLINE(void) netif_tx_unlock_bh(struct net_device *pDev)
1803{
1804 spin_unlock_bh(&pDev->xmit_lock);
1805}
1806#endif
1807
1808/**
1809 * Some devices need link state change when filter attaches/detaches
1810 * since the filter is their link in a sense.
1811 */
1812static void vboxNetFltSetLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1813{
1814 if (vboxNetFltNeedsLinkState(pThis, pDev))
1815 {
1816 Log3(("%s: bringing device link %s\n",
1817 __FUNCTION__, fLinkUp ? "up" : "down"));
1818 netif_tx_lock_bh(pDev);
1819 if (fLinkUp)
1820 netif_carrier_on(pDev);
1821 else
1822 netif_carrier_off(pDev);
1823 netif_tx_unlock_bh(pDev);
1824 }
1825}
1826
1827/**
1828 * Internal worker for vboxNetFltLinuxNotifierCallback.
1829 *
1830 * @returns VBox status code.
1831 * @param pThis The instance.
1832 * @param pDev The device to attach to.
1833 */
1834static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1835{
1836 LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1837
1838 /*
1839 * Retain and store the device.
1840 */
1841 dev_hold(pDev);
1842
1843 RTSpinlockAcquire(pThis->hSpinlock);
1844 ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1845 RTSpinlockRelease(pThis->hSpinlock);
1846
1847 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1848 pDev, pDev->name,
1849#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1850 netdev_refcnt_read(pDev)
1851#else
1852 atomic_read(&pDev->refcnt)
1853#endif
1854 ));
1855 Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1856 pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1857
1858 /* Get the mac address while we still have a valid net_device reference. */
1859 memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1860 /* Initialize MTU */
1861 pThis->u.s.cbMtu = pDev->mtu;
1862
1863 /*
1864 * Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1865 */
1866 pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1867 pThis->u.s.PacketType.dev = pDev;
1868 pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1869 dev_add_pack(&pThis->u.s.PacketType);
1870 ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1871 Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1872
1873#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1874 vboxNetFltLinuxHookDev(pThis, pDev);
1875#endif
1876
1877 /*
1878 * Are we the "carrier" for this device (e.g. vboxnet or tap)?
1879 */
1880 vboxNetFltSetLinkState(pThis, pDev, true);
1881
1882 /*
1883 * Set indicators that require the spinlock. Be abit paranoid about racing
1884 * the device notification handle.
1885 */
1886 RTSpinlockAcquire(pThis->hSpinlock);
1887 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1888 if (pDev)
1889 {
1890 ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1891 ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1892 pDev = NULL; /* don't dereference it */
1893 }
1894 RTSpinlockRelease(pThis->hSpinlock);
1895
1896 /*
1897 * Report GSO capabilities
1898 */
1899 Assert(pThis->pSwitchPort);
1900 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1901 {
1902 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1903 pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1904 pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1905 pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE | INTNETTRUNKDIR_HOST);
1906 vboxNetFltRelease(pThis, true /*fBusy*/);
1907 }
1908
1909 LogRel(("VBoxNetFlt: attached to '%s' / %RTmac\n", pThis->szName, &pThis->u.s.MacAddr));
1910 return VINF_SUCCESS;
1911}
1912
1913
1914static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1915{
1916 bool fRegistered;
1917 Assert(!pThis->fDisconnectedFromHost);
1918
1919#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1920 vboxNetFltLinuxUnhookDev(pThis, pDev);
1921#endif
1922
1923 if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1924 {
1925 dev_remove_pack(&pThis->u.s.PacketType);
1926 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1927 }
1928
1929 RTSpinlockAcquire(pThis->hSpinlock);
1930 fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1931 if (fRegistered)
1932 {
1933 ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1934 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1935 }
1936 RTSpinlockRelease(pThis->hSpinlock);
1937
1938 if (fRegistered)
1939 {
1940#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1941 skb_queue_purge(&pThis->u.s.XmitQueue);
1942#endif
1943 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1944 Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1945 pDev, pDev->name,
1946#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1947 netdev_refcnt_read(pDev)
1948#else
1949 atomic_read(&pDev->refcnt)
1950#endif
1951 ));
1952 dev_put(pDev);
1953 }
1954
1955 return NOTIFY_OK;
1956}
1957
1958static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1959{
1960 /* Check if we are not suspended and promiscuous mode has not been set. */
1961 if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1962 && !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1963 {
1964 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1965 dev_set_promiscuity(pDev, 1);
1966 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1967 Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1968 }
1969 else
1970 Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1971 return NOTIFY_OK;
1972}
1973
1974static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1975{
1976 /* Undo promiscuous mode if we has set it. */
1977 if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1978 {
1979 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1980 dev_set_promiscuity(pDev, -1);
1981 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1982 Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1983 }
1984 else
1985 Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1986 return NOTIFY_OK;
1987}
1988
1989/**
1990 * Callback for listening to MTU change event.
1991 *
1992 * We need to track changes of host's inteface MTU to discard over-sized frames
1993 * coming from the internal network as they may hang the TX queue of host's
1994 * adapter.
1995 *
1996 * @returns NOTIFY_OK
1997 * @param pThis The netfilter instance.
1998 * @param pDev Pointer to device structure of host's interface.
1999 */
2000static int vboxNetFltLinuxDeviceMtuChange(PVBOXNETFLTINS pThis, struct net_device *pDev)
2001{
2002 ASMAtomicWriteU32(&pThis->u.s.cbMtu, pDev->mtu);
2003 Log(("vboxNetFltLinuxDeviceMtuChange: set MTU for %s to %d\n", pThis->szName, pDev->mtu));
2004 return NOTIFY_OK;
2005}
2006
2007#ifdef LOG_ENABLED
2008/** Stringify the NETDEV_XXX constants. */
2009static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
2010{
2011 const char *pszEvent = "NETDEV_<unknown>";
2012 switch (ulEventType)
2013 {
2014 case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
2015 case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
2016 case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
2017 case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
2018 case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
2019 case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
2020 case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
2021 case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
2022 case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
2023 case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
2024# ifdef NETDEV_FEAT_CHANGE
2025 case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
2026# endif
2027 }
2028 return pszEvent;
2029}
2030#endif /* LOG_ENABLED */
2031
2032/**
2033 * Callback for listening to netdevice events.
2034 *
2035 * This works the rediscovery, clean up on unregistration, promiscuity on
2036 * up/down, and GSO feature changes from ethtool.
2037 *
2038 * @returns NOTIFY_OK
2039 * @param self Pointer to our notifier registration block.
2040 * @param ulEventType The event.
2041 * @param ptr Event specific, but it is usually the device it
2042 * relates to.
2043 */
2044static int vboxNetFltLinuxNotifierCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2045
2046{
2047 PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
2048 struct net_device *pMyDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2049 struct net_device *pDev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2050 int rc = NOTIFY_OK;
2051
2052 Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
2053 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, pMyDev));
2054
2055 if (ulEventType == NETDEV_REGISTER)
2056 {
2057#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 24) /* cgroups/namespaces introduced */
2058# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 26)
2059# define VBOX_DEV_NET(dev) dev_net(dev)
2060# define VBOX_NET_EQ(n1, n2) net_eq((n1), (n2))
2061# else
2062# define VBOX_DEV_NET(dev) ((dev)->nd_net)
2063# define VBOX_NET_EQ(n1, n2) ((n1) == (n2))
2064# endif
2065 struct net *pMyNet = current->nsproxy->net_ns;
2066 struct net *pDevNet = VBOX_DEV_NET(pDev);
2067
2068 if (VBOX_NET_EQ(pDevNet, pMyNet))
2069#endif /* namespaces */
2070 {
2071 if (strcmp(pDev->name, pThis->szName) == 0)
2072 {
2073 vboxNetFltLinuxAttachToInterface(pThis, pDev);
2074 }
2075 }
2076 }
2077 else
2078 {
2079 if (pDev == pMyDev)
2080 {
2081 switch (ulEventType)
2082 {
2083 case NETDEV_UNREGISTER:
2084 rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
2085 break;
2086 case NETDEV_UP:
2087 rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
2088 break;
2089 case NETDEV_GOING_DOWN:
2090 rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
2091 break;
2092 case NETDEV_CHANGEMTU:
2093 rc = vboxNetFltLinuxDeviceMtuChange(pThis, pDev);
2094 break;
2095 case NETDEV_CHANGENAME:
2096 break;
2097#ifdef NETDEV_FEAT_CHANGE
2098 case NETDEV_FEAT_CHANGE:
2099 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2100 break;
2101#endif
2102 }
2103 }
2104 }
2105
2106 return rc;
2107}
2108
2109/*
2110 * Initial enumeration of netdevs. Called with NETDEV_REGISTER by
2111 * register_netdevice_notifier() under rtnl lock.
2112 */
2113static int vboxNetFltLinuxEnumeratorCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2114{
2115 PVBOXNETFLTINS pThis = ((PVBOXNETFLTNOTIFIER)self)->pThis;
2116 struct net_device *dev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
2117 struct in_device *in_dev;
2118 struct inet6_dev *in6_dev;
2119
2120 if (ulEventType != NETDEV_REGISTER)
2121 return NOTIFY_OK;
2122
2123 if (RT_UNLIKELY(pThis->pSwitchPort->pfnNotifyHostAddress == NULL))
2124 return NOTIFY_OK;
2125
2126 /*
2127 * IPv4
2128 */
2129#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
2130 in_dev = __in_dev_get_rtnl(dev);
2131#else
2132 in_dev = __in_dev_get(dev);
2133#endif
2134 if (in_dev != NULL)
2135 {
2136 struct in_ifaddr *ifa;
2137
2138 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
2139 if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2140 return NOTIFY_OK;
2141
2142 if ( dev != pThis->u.s.pDev
2143 && VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2144 continue;
2145
2146 Log(("%s: %s: IPv4 addr %RTnaipv4 mask %RTnaipv4\n",
2147 __FUNCTION__, VBOX_NETDEV_NAME(dev),
2148 ifa->ifa_address, ifa->ifa_mask));
2149
2150 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2151 /* :fAdded */ true, kIntNetAddrType_IPv4, &ifa->ifa_address);
2152 }
2153 }
2154
2155 /*
2156 * IPv6
2157 */
2158 in6_dev = __in6_dev_get(dev);
2159 if (in6_dev != NULL)
2160 {
2161 struct inet6_ifaddr *ifa;
2162
2163 read_lock_bh(&in6_dev->lock);
2164#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 35)
2165 list_for_each_entry(ifa, &in6_dev->addr_list, if_list)
2166#else
2167 for (ifa = in6_dev->addr_list; ifa != NULL; ifa = ifa->if_next)
2168#endif
2169 {
2170 if ( dev != pThis->u.s.pDev
2171 && ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_LOOPBACK))
2172 continue;
2173
2174 Log(("%s: %s: IPv6 addr %RTnaipv6/%u\n",
2175 __FUNCTION__, VBOX_NETDEV_NAME(dev),
2176 &ifa->addr, (unsigned)ifa->prefix_len));
2177
2178 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
2179 /* :fAdded */ true, kIntNetAddrType_IPv6, &ifa->addr);
2180 }
2181 read_unlock_bh(&in6_dev->lock);
2182 }
2183
2184 return NOTIFY_OK;
2185}
2186
2187
2188static int vboxNetFltLinuxNotifierIPv4Callback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2189{
2190 PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv4);
2191 struct net_device *pDev, *pEventDev;
2192 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
2193 bool fMyDev;
2194 int rc = NOTIFY_OK;
2195
2196 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2197 pEventDev = ifa->ifa_dev->dev;
2198 fMyDev = (pDev == pEventDev);
2199 Log(("VBoxNetFlt: %s: IPv4 event %s(0x%lx) %s: addr %RTnaipv4 mask %RTnaipv4\n",
2200 pDev ? VBOX_NETDEV_NAME(pDev) : "<unknown>",
2201 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2202 pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<unknown>",
2203 ifa->ifa_address, ifa->ifa_mask));
2204
2205 if (pDev != NULL)
2206 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2207
2208 if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2209 return NOTIFY_OK;
2210
2211 if ( !fMyDev
2212 && VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2213 return NOTIFY_OK;
2214
2215 if (pThis->pSwitchPort->pfnNotifyHostAddress)
2216 {
2217 bool fAdded;
2218 if (ulEventType == NETDEV_UP)
2219 fAdded = true;
2220 else if (ulEventType == NETDEV_DOWN)
2221 fAdded = false;
2222 else
2223 return NOTIFY_OK;
2224
2225 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2226 kIntNetAddrType_IPv4, &ifa->ifa_local);
2227 }
2228
2229 return rc;
2230}
2231
2232
2233static int vboxNetFltLinuxNotifierIPv6Callback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2234{
2235 PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv6);
2236 struct net_device *pDev, *pEventDev;
2237 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
2238 bool fMyDev;
2239 int rc = NOTIFY_OK;
2240
2241 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2242 pEventDev = ifa->idev->dev;
2243 fMyDev = (pDev == pEventDev);
2244 Log(("VBoxNetFlt: %s: IPv6 event %s(0x%lx) %s: %RTnaipv6\n",
2245 pDev ? VBOX_NETDEV_NAME(pDev) : "<unknown>",
2246 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2247 pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<unknown>",
2248 &ifa->addr));
2249
2250 if (pDev != NULL)
2251 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2252
2253 if ( !fMyDev
2254 && ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_LOOPBACK))
2255 return NOTIFY_OK;
2256
2257 if (pThis->pSwitchPort->pfnNotifyHostAddress)
2258 {
2259 bool fAdded;
2260 if (ulEventType == NETDEV_UP)
2261 fAdded = true;
2262 else if (ulEventType == NETDEV_DOWN)
2263 fAdded = false;
2264 else
2265 return NOTIFY_OK;
2266
2267 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2268 kIntNetAddrType_IPv6, &ifa->addr);
2269 }
2270
2271 return rc;
2272}
2273
2274
2275bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2276{
2277 return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2278}
2279
2280int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2281{
2282 struct net_device * pDev;
2283 int err;
2284 int rc = VINF_SUCCESS;
2285 IPRT_LINUX_SAVE_EFL_AC();
2286 NOREF(pvIfData);
2287
2288 LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2289
2290 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2291 if (pDev)
2292 {
2293 /*
2294 * Create a sk_buff for the gather list and push it onto the wire.
2295 */
2296 if (fDst & INTNETTRUNKDIR_WIRE)
2297 {
2298 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2299 if (pBuf)
2300 {
2301 vboxNetFltDumpPacket(pSG, true, "wire", 1);
2302 Log6(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2303 Log6(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2304 err = dev_queue_xmit(pBuf);
2305 if (err)
2306 rc = RTErrConvertFromErrno(err);
2307 }
2308 else
2309 rc = VERR_NO_MEMORY;
2310 }
2311
2312 /*
2313 * Create a sk_buff for the gather list and push it onto the host stack.
2314 */
2315 if (fDst & INTNETTRUNKDIR_HOST)
2316 {
2317 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2318 if (pBuf)
2319 {
2320 vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2321 Log6(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2322 Log6(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2323 err = netif_rx_ni(pBuf);
2324 if (err)
2325 rc = RTErrConvertFromErrno(err);
2326 }
2327 else
2328 rc = VERR_NO_MEMORY;
2329 }
2330
2331 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2332 }
2333
2334 IPRT_LINUX_RESTORE_EFL_AC();
2335 return rc;
2336}
2337
2338
2339void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2340{
2341 struct net_device *pDev;
2342 IPRT_LINUX_SAVE_EFL_AC();
2343
2344 LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%RTbool, fDisablePromiscuous=%RTbool\n",
2345 pThis, pThis->szName, fActive, pThis->fDisablePromiscuous));
2346
2347 if (pThis->fDisablePromiscuous)
2348 return;
2349
2350 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2351 if (pDev)
2352 {
2353 /*
2354 * This api is a bit weird, the best reference is the code.
2355 *
2356 * Also, we have a bit or race conditions wrt the maintenance of
2357 * host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2358 */
2359#ifdef LOG_ENABLED
2360 u_int16_t fIf;
2361 unsigned const cPromiscBefore = pDev->promiscuity;
2362#endif
2363 if (fActive)
2364 {
2365 Assert(!pThis->u.s.fPromiscuousSet);
2366
2367 rtnl_lock();
2368 dev_set_promiscuity(pDev, 1);
2369 rtnl_unlock();
2370 pThis->u.s.fPromiscuousSet = true;
2371 Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2372 }
2373 else
2374 {
2375 if (pThis->u.s.fPromiscuousSet)
2376 {
2377 rtnl_lock();
2378 dev_set_promiscuity(pDev, -1);
2379 rtnl_unlock();
2380 Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2381 }
2382 pThis->u.s.fPromiscuousSet = false;
2383
2384#ifdef LOG_ENABLED
2385 fIf = dev_get_flags(pDev);
2386 Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2387#endif
2388 }
2389
2390 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2391 }
2392 IPRT_LINUX_RESTORE_EFL_AC();
2393}
2394
2395
2396int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2397{
2398 /*
2399 * Remove packet handler when we get disconnected from internal switch as
2400 * we don't want the handler to forward packets to disconnected switch.
2401 */
2402 if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
2403 {
2404 IPRT_LINUX_SAVE_EFL_AC();
2405 dev_remove_pack(&pThis->u.s.PacketType);
2406 Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
2407 IPRT_LINUX_RESTORE_EFL_AC();
2408 }
2409 return VINF_SUCCESS;
2410}
2411
2412
2413int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2414{
2415 IPRT_LINUX_SAVE_EFL_AC();
2416
2417 /*
2418 * Report the GSO capabilities of the host and device (if connected).
2419 * Note! No need to mark ourselves busy here.
2420 */
2421 /** @todo duplicate work here now? Attach */
2422#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2423 Log3(("vboxNetFltOsConnectIt: reporting host tso tso6\n"));
2424 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2425 0
2426 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2427 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2428 , INTNETTRUNKDIR_HOST);
2429
2430#endif
2431 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2432
2433 IPRT_LINUX_RESTORE_EFL_AC();
2434 return VINF_SUCCESS;
2435}
2436
2437
2438void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2439{
2440 struct net_device *pDev;
2441 bool fRegistered;
2442 IPRT_LINUX_SAVE_EFL_AC();
2443
2444#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
2445 vboxNetFltLinuxUnhookDev(pThis, NULL);
2446#endif
2447
2448 /** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2449 * unlikely, but none the less). Since it doesn't actually update the
2450 * state (just reads it), it is likely to panic in some interesting
2451 * ways. */
2452
2453 RTSpinlockAcquire(pThis->hSpinlock);
2454 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2455 fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
2456 RTSpinlockRelease(pThis->hSpinlock);
2457
2458 if (fRegistered)
2459 {
2460 vboxNetFltSetLinkState(pThis, pDev, false);
2461
2462#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2463 skb_queue_purge(&pThis->u.s.XmitQueue);
2464#endif
2465 Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
2466 Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
2467 pDev, pDev->name,
2468#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
2469 netdev_refcnt_read(pDev)
2470#else
2471 atomic_read(&pDev->refcnt)
2472#endif
2473 ));
2474 dev_put(pDev);
2475 }
2476
2477 unregister_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2478 unregister_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2479
2480 Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2481 unregister_netdevice_notifier(&pThis->u.s.Notifier);
2482 module_put(THIS_MODULE);
2483
2484 IPRT_LINUX_RESTORE_EFL_AC();
2485}
2486
2487
2488int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2489{
2490 int err;
2491 IPRT_LINUX_SAVE_EFL_AC();
2492 NOREF(pvContext);
2493
2494 pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2495 err = register_netdevice_notifier(&pThis->u.s.Notifier);
2496 if (err)
2497 {
2498 IPRT_LINUX_RESTORE_EFL_AC();
2499 return VERR_INTNET_FLT_IF_FAILED;
2500 }
2501 if (!pThis->u.s.fRegistered)
2502 {
2503 unregister_netdevice_notifier(&pThis->u.s.Notifier);
2504 LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2505 IPRT_LINUX_RESTORE_EFL_AC();
2506 return VERR_INTNET_FLT_IF_NOT_FOUND;
2507 }
2508
2509 Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2510 if ( pThis->fDisconnectedFromHost
2511 || !try_module_get(THIS_MODULE))
2512 {
2513 IPRT_LINUX_RESTORE_EFL_AC();
2514 return VERR_INTNET_FLT_IF_FAILED;
2515 }
2516
2517 if (pThis->pSwitchPort->pfnNotifyHostAddress)
2518 {
2519 VBOXNETFLTNOTIFIER Enumerator;
2520
2521 /*
2522 * register_inetaddr_notifier() and register_inet6addr_notifier()
2523 * do not call the callback for existing devices. Enumerating
2524 * all network devices explicitly is a bit of an ifdef mess,
2525 * so co-opt register_netdevice_notifier() to do that for us.
2526 */
2527 RT_ZERO(Enumerator);
2528 Enumerator.Notifier.notifier_call = vboxNetFltLinuxEnumeratorCallback;
2529 Enumerator.pThis = pThis;
2530
2531 err = register_netdevice_notifier(&Enumerator.Notifier);
2532 if (err)
2533 {
2534 LogRel(("%s: failed to enumerate network devices: error %d\n", __FUNCTION__, err));
2535 IPRT_LINUX_RESTORE_EFL_AC();
2536 return VINF_SUCCESS;
2537 }
2538
2539 unregister_netdevice_notifier(&Enumerator.Notifier);
2540
2541 pThis->u.s.NotifierIPv4.notifier_call = vboxNetFltLinuxNotifierIPv4Callback;
2542 err = register_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2543 if (err)
2544 LogRel(("%s: failed to register IPv4 notifier: error %d\n", __FUNCTION__, err));
2545
2546 pThis->u.s.NotifierIPv6.notifier_call = vboxNetFltLinuxNotifierIPv6Callback;
2547 err = register_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2548 if (err)
2549 LogRel(("%s: failed to register IPv6 notifier: error %d\n", __FUNCTION__, err));
2550 }
2551
2552 IPRT_LINUX_RESTORE_EFL_AC();
2553 return VINF_SUCCESS;
2554}
2555
2556int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2557{
2558 IPRT_LINUX_SAVE_EFL_AC();
2559
2560 /*
2561 * Init the linux specific members.
2562 */
2563 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2564 pThis->u.s.fRegistered = false;
2565 pThis->u.s.fPromiscuousSet = false;
2566 pThis->u.s.fPacketHandler = false;
2567 memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2568#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2569 skb_queue_head_init(&pThis->u.s.XmitQueue);
2570# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2571 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2572# else
2573 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2574# endif
2575#endif
2576
2577 IPRT_LINUX_RESTORE_EFL_AC();
2578 return VINF_SUCCESS;
2579}
2580
2581
2582void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2583{
2584 NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2585}
2586
2587
2588int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void *pvIf, void **pvIfData)
2589{
2590 /* Nothing to do */
2591 NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2592 return VINF_SUCCESS;
2593}
2594
2595
2596int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2597{
2598 /* Nothing to do */
2599 NOREF(pThis); NOREF(pvIfData);
2600 return VINF_SUCCESS;
2601}
2602
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