VirtualBox

source: vbox/trunk/src/VBox/HostDrivers/VBoxNetFlt/linux/VBoxNetFlt-linux.c@ 60688

Last change on this file since 60688 was 60590, checked in by vboxsync, 9 years ago

fix for Linux 4.5 if CONFIG_NET_CLS_ACT is enabled, see ticketref:15327
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 83.2 KB
Line 
1/* $Id: VBoxNetFlt-linux.c 60590 2016-04-20 10:22:31Z vboxsync $ */
2/** @file
3 * VBoxNetFlt - Network Filter Driver (Host), Linux Specific Code.
4 */
5
6/*
7 * Copyright (C) 2006-2015 Oracle Corporation
8 *
9 * This file is part of VirtualBox Open Source Edition (OSE), as
10 * available from http://www.215389.xyz. 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(2, 6, 34)
90# define VBOX_NETDEV_NAME(dev) netdev_name(dev)
91#else
92# define VBOX_NETDEV_NAME(dev) ((dev)->reg_state != NETREG_REGISTERED ? "(unregistered net_device)" : (dev)->name)
93#endif
94
95#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25)
96# define VBOX_IPV4_IS_LOOPBACK(addr) ipv4_is_loopback(addr)
97# define VBOX_IPV4_IS_LINKLOCAL_169(addr) ipv4_is_linklocal_169(addr)
98#else
99# define VBOX_IPV4_IS_LOOPBACK(addr) ((addr & htonl(IN_CLASSA_NET)) == htonl(0x7f000000))
100# define VBOX_IPV4_IS_LINKLOCAL_169(addr) ((addr & htonl(IN_CLASSB_NET)) == htonl(0xa9fe0000))
101#endif
102
103#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
104# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb_reset_network_header(skb)
105# define VBOX_SKB_RESET_MAC_HDR(skb) skb_reset_mac_header(skb)
106#else
107# define VBOX_SKB_RESET_NETWORK_HDR(skb) skb->nh.raw = skb->data
108# define VBOX_SKB_RESET_MAC_HDR(skb) skb->mac.raw = skb->data
109#endif
110
111#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
112# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb)
113#else
114# define CHECKSUM_PARTIAL CHECKSUM_HW
115# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 10)
116# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(skb, 0)
117# else
118# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 7)
119# define VBOX_SKB_CHECKSUM_HELP(skb) skb_checksum_help(&skb, 0)
120# else
121# define VBOX_SKB_CHECKSUM_HELP(skb) (!skb_checksum_help(skb))
122# endif
123/* Versions prior 2.6.10 use stats for both bstats and qstats */
124# define bstats stats
125# define qstats stats
126# endif
127#endif
128
129#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 20, 0)
130# define VBOX_HAVE_SKB_VLAN
131#else
132# ifdef RHEL_RELEASE_CODE
133# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(7, 2)
134# define VBOX_HAVE_SKB_VLAN
135# endif
136# endif
137#endif
138
139#ifdef VBOX_HAVE_SKB_VLAN
140# define vlan_tx_tag_get(skb) skb_vlan_tag_get(skb)
141# define vlan_tx_tag_present(skb) skb_vlan_tag_present(skb)
142#endif
143
144#ifndef NET_IP_ALIGN
145# define NET_IP_ALIGN 2
146#endif
147
148#if 0
149/** Create scatter / gather segments for fragments. When not used, we will
150 * linearize the socket buffer before creating the internal networking SG. */
151# define VBOXNETFLT_SG_SUPPORT 1
152#endif
153
154#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
155
156/** Indicates that the linux kernel may send us GSO frames. */
157# define VBOXNETFLT_WITH_GSO 1
158
159/** This enables or disables the transmitting of GSO frame from the internal
160 * network and to the host. */
161# define VBOXNETFLT_WITH_GSO_XMIT_HOST 1
162
163# if 0 /** @todo This is currently disable because it causes performance loss of 5-10%. */
164/** This enables or disables the transmitting of GSO frame from the internal
165 * network and to the wire. */
166# define VBOXNETFLT_WITH_GSO_XMIT_WIRE 1
167# endif
168
169/** This enables or disables the forwarding/flooding of GSO frame from the host
170 * to the internal network. */
171# define VBOXNETFLT_WITH_GSO_RECV 1
172
173#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18) */
174
175#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
176/** This enables or disables handling of GSO frames coming from the wire (GRO). */
177# define VBOXNETFLT_WITH_GRO 1
178#endif
179
180/*
181 * GRO support was backported to RHEL 5.4
182 */
183#ifdef RHEL_RELEASE_CODE
184# if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(5, 4)
185# define VBOXNETFLT_WITH_GRO 1
186# endif
187#endif
188
189
190/*********************************************************************************************************************************
191* Internal Functions *
192*********************************************************************************************************************************/
193static int VBoxNetFltLinuxInit(void);
194static void VBoxNetFltLinuxUnload(void);
195static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf);
196
197
198/*********************************************************************************************************************************
199* Global Variables *
200*********************************************************************************************************************************/
201/**
202 * The (common) global data.
203 */
204static VBOXNETFLTGLOBALS g_VBoxNetFltGlobals;
205
206module_init(VBoxNetFltLinuxInit);
207module_exit(VBoxNetFltLinuxUnload);
208
209MODULE_AUTHOR(VBOX_VENDOR);
210MODULE_DESCRIPTION(VBOX_PRODUCT " Network Filter Driver");
211MODULE_LICENSE("GPL");
212#ifdef MODULE_VERSION
213MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV) " (" RT_XSTR(INTNETTRUNKIFPORT_VERSION) ")");
214#endif
215
216
217#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) && defined(LOG_ENABLED)
218unsigned dev_get_flags(const struct net_device *dev)
219{
220 unsigned flags;
221
222 flags = (dev->flags & ~(IFF_PROMISC |
223 IFF_ALLMULTI |
224 IFF_RUNNING)) |
225 (dev->gflags & (IFF_PROMISC |
226 IFF_ALLMULTI));
227
228 if (netif_running(dev) && netif_carrier_ok(dev))
229 flags |= IFF_RUNNING;
230
231 return flags;
232}
233#endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 12) */
234
235
236/**
237 * Initialize module.
238 *
239 * @returns appropriate status code.
240 */
241static int __init VBoxNetFltLinuxInit(void)
242{
243 int rc;
244 /*
245 * Initialize IPRT.
246 */
247 rc = RTR0Init(0);
248 if (RT_SUCCESS(rc))
249 {
250 Log(("VBoxNetFltLinuxInit\n"));
251
252 /*
253 * Initialize the globals and connect to the support driver.
254 *
255 * This will call back vboxNetFltOsOpenSupDrv (and maybe vboxNetFltOsCloseSupDrv)
256 * for establishing the connect to the support driver.
257 */
258 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
259 rc = vboxNetFltInitGlobalsAndIdc(&g_VBoxNetFltGlobals);
260 if (RT_SUCCESS(rc))
261 {
262 LogRel(("VBoxNetFlt: Successfully started.\n"));
263 return 0;
264 }
265
266 LogRel(("VBoxNetFlt: failed to initialize device extension (rc=%d)\n", rc));
267 RTR0Term();
268 }
269 else
270 LogRel(("VBoxNetFlt: failed to initialize IPRT (rc=%d)\n", rc));
271
272 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
273 return -RTErrConvertToErrno(rc);
274}
275
276
277/**
278 * Unload the module.
279 *
280 * @todo We have to prevent this if we're busy!
281 */
282static void __exit VBoxNetFltLinuxUnload(void)
283{
284 int rc;
285 Log(("VBoxNetFltLinuxUnload\n"));
286 Assert(vboxNetFltCanUnload(&g_VBoxNetFltGlobals));
287
288 /*
289 * Undo the work done during start (in reverse order).
290 */
291 rc = vboxNetFltTryDeleteIdcAndGlobals(&g_VBoxNetFltGlobals);
292 AssertRC(rc); NOREF(rc);
293
294 RTR0Term();
295
296 memset(&g_VBoxNetFltGlobals, 0, sizeof(g_VBoxNetFltGlobals));
297
298 Log(("VBoxNetFltLinuxUnload - done\n"));
299}
300
301
302/**
303 * We filter traffic from the host to the internal network
304 * before it reaches the NIC driver.
305 *
306 * The current code uses a very ugly hack overriding hard_start_xmit
307 * callback in the device structure, but it has been shown to give us a
308 * performance boost of 60-100% though. Eventually we have to find some
309 * less hacky way of getting this job done.
310 */
311#define VBOXNETFLT_WITH_HOST2WIRE_FILTER
312
313#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
314
315# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
316
317# include <linux/ethtool.h>
318
319typedef struct ethtool_ops OVR_OPSTYPE;
320# define OVR_OPS ethtool_ops
321# define OVR_XMIT pfnStartXmit
322
323# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
324
325typedef struct net_device_ops OVR_OPSTYPE;
326# define OVR_OPS netdev_ops
327# define OVR_XMIT pOrgOps->ndo_start_xmit
328
329# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
330
331/**
332 * The overridden net_device_ops of the device we're attached to.
333 *
334 * As there is no net_device_ops structure in pre-2.6.29 kernels we override
335 * ethtool_ops instead along with hard_start_xmit callback in net_device
336 * structure.
337 *
338 * This is a very dirty hack that was created to explore how much we can improve
339 * the host to guest transfers by not CC'ing the NIC. It turns out to be
340 * the only way to filter outgoing packets for devices without TX queue.
341 */
342typedef struct VBoxNetDeviceOpsOverride
343{
344 /** Our overridden ops. */
345 OVR_OPSTYPE Ops;
346 /** Magic word. */
347 uint32_t u32Magic;
348 /** Pointer to the original ops. */
349 OVR_OPSTYPE const *pOrgOps;
350# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
351 /** Pointer to the original hard_start_xmit function. */
352 int (*pfnStartXmit)(struct sk_buff *pSkb, struct net_device *pDev);
353# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
354 /** Pointer to the net filter instance. */
355 PVBOXNETFLTINS pVBoxNetFlt;
356 /** The number of filtered packages. */
357 uint64_t cFiltered;
358 /** The total number of packets */
359 uint64_t cTotal;
360} VBOXNETDEVICEOPSOVERRIDE, *PVBOXNETDEVICEOPSOVERRIDE;
361/** VBOXNETDEVICEOPSOVERRIDE::u32Magic value. */
362#define VBOXNETDEVICEOPSOVERRIDE_MAGIC UINT32_C(0x00c0ffee)
363
364/**
365 * ndo_start_xmit wrapper that drops packets that shouldn't go to the wire
366 * because they belong on the internal network.
367 *
368 * @returns NETDEV_TX_XXX.
369 * @param pSkb The socket buffer to transmit.
370 * @param pDev The net device.
371 */
372static int vboxNetFltLinuxStartXmitFilter(struct sk_buff *pSkb, struct net_device *pDev)
373{
374 PVBOXNETDEVICEOPSOVERRIDE pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
375 uint8_t abHdrBuf[sizeof(RTNETETHERHDR) + sizeof(uint32_t) + RTNETIPV4_MIN_LEN];
376 PCRTNETETHERHDR pEtherHdr;
377 PINTNETTRUNKSWPORT pSwitchPort;
378 uint32_t cbHdrs;
379
380
381 /*
382 * Validate the override structure.
383 *
384 * Note! We're racing vboxNetFltLinuxUnhookDev here. If this was supposed
385 * to be production quality code, we would have to be much more
386 * careful here and avoid the race.
387 */
388 if ( !VALID_PTR(pOverride)
389 || pOverride->u32Magic != VBOXNETDEVICEOPSOVERRIDE_MAGIC
390# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
391 || !VALID_PTR(pOverride->pOrgOps)
392# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
393 )
394 {
395 printk("vboxNetFltLinuxStartXmitFilter: bad override %p\n", pOverride);
396 dev_kfree_skb(pSkb);
397 return NETDEV_TX_OK;
398 }
399 pOverride->cTotal++;
400
401 /*
402 * Do the filtering base on the default OUI of our virtual NICs
403 *
404 * Note! In a real solution, we would ask the switch whether the
405 * destination MAC is 100% to be on the internal network and then
406 * drop it.
407 */
408 cbHdrs = skb_headlen(pSkb);
409 cbHdrs = RT_MIN(cbHdrs, sizeof(abHdrBuf));
410 pEtherHdr = (PCRTNETETHERHDR)skb_header_pointer(pSkb, 0, cbHdrs, &abHdrBuf[0]);
411 if ( pEtherHdr
412 && VALID_PTR(pOverride->pVBoxNetFlt)
413 && (pSwitchPort = pOverride->pVBoxNetFlt->pSwitchPort) != NULL
414 && VALID_PTR(pSwitchPort)
415 && cbHdrs >= 6)
416 {
417 INTNETSWDECISION enmDecision;
418
419 /** @todo consider reference counting, etc. */
420 enmDecision = pSwitchPort->pfnPreRecv(pSwitchPort, pEtherHdr, cbHdrs, INTNETTRUNKDIR_HOST);
421 if (enmDecision == INTNETSWDECISION_INTNET)
422 {
423 dev_kfree_skb(pSkb);
424 pOverride->cFiltered++;
425 return NETDEV_TX_OK;
426 }
427 }
428
429 return pOverride->OVR_XMIT(pSkb, pDev);
430}
431
432/**
433 * Hooks the device ndo_start_xmit operation of the device.
434 *
435 * @param pThis The net filter instance.
436 * @param pDev The net device.
437 */
438static void vboxNetFltLinuxHookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
439{
440 PVBOXNETDEVICEOPSOVERRIDE pOverride;
441
442 /* Cancel override if ethtool_ops is missing (host-only case, @bugref{5712}) */
443 if (!VALID_PTR(pDev->OVR_OPS))
444 return;
445 pOverride = RTMemAlloc(sizeof(*pOverride));
446 if (!pOverride)
447 return;
448 pOverride->pOrgOps = pDev->OVR_OPS;
449 pOverride->Ops = *pDev->OVR_OPS;
450# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
451 pOverride->pfnStartXmit = pDev->hard_start_xmit;
452# else /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
453 pOverride->Ops.ndo_start_xmit = vboxNetFltLinuxStartXmitFilter;
454# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29) */
455 pOverride->u32Magic = VBOXNETDEVICEOPSOVERRIDE_MAGIC;
456 pOverride->cTotal = 0;
457 pOverride->cFiltered = 0;
458 pOverride->pVBoxNetFlt = pThis;
459
460 RTSpinlockAcquire(pThis->hSpinlock); /* (this isn't necessary, but so what) */
461 ASMAtomicWritePtr((void * volatile *)&pDev->OVR_OPS, pOverride);
462# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
463 ASMAtomicXchgPtr((void * volatile *)&pDev->hard_start_xmit, vboxNetFltLinuxStartXmitFilter);
464# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
465 RTSpinlockRelease(pThis->hSpinlock);
466}
467
468/**
469 * Undos what vboxNetFltLinuxHookDev did.
470 *
471 * @param pThis The net filter instance.
472 * @param pDev The net device. Can be NULL, in which case
473 * we'll try retrieve it from @a pThis.
474 */
475static void vboxNetFltLinuxUnhookDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
476{
477 PVBOXNETDEVICEOPSOVERRIDE pOverride;
478
479 RTSpinlockAcquire(pThis->hSpinlock);
480 if (!pDev)
481 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
482 if (VALID_PTR(pDev))
483 {
484 pOverride = (PVBOXNETDEVICEOPSOVERRIDE)pDev->OVR_OPS;
485 if ( VALID_PTR(pOverride)
486 && pOverride->u32Magic == VBOXNETDEVICEOPSOVERRIDE_MAGIC
487 && VALID_PTR(pOverride->pOrgOps)
488 )
489 {
490# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29)
491 ASMAtomicWritePtr((void * volatile *)&pDev->hard_start_xmit, pOverride->pfnStartXmit);
492# endif /* LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 29) */
493 ASMAtomicWritePtr((void const * volatile *)&pDev->OVR_OPS, pOverride->pOrgOps);
494 ASMAtomicWriteU32(&pOverride->u32Magic, 0);
495 }
496 else
497 pOverride = NULL;
498 }
499 else
500 pOverride = NULL;
501 RTSpinlockRelease(pThis->hSpinlock);
502
503 if (pOverride)
504 {
505 printk("vboxnetflt: %llu out of %llu packets were not sent (directed to host)\n", pOverride->cFiltered, pOverride->cTotal);
506 RTMemFree(pOverride);
507 }
508}
509
510#endif /* VBOXNETFLT_WITH_HOST2WIRE_FILTER */
511
512
513/**
514 * Reads and retains the host interface handle.
515 *
516 * @returns The handle, NULL if detached.
517 * @param pThis
518 */
519DECLINLINE(struct net_device *) vboxNetFltLinuxRetainNetDev(PVBOXNETFLTINS pThis)
520{
521#if 0
522 struct net_device *pDev = NULL;
523
524 Log(("vboxNetFltLinuxRetainNetDev\n"));
525 /*
526 * Be careful here to avoid problems racing the detached callback.
527 */
528 RTSpinlockAcquire(pThis->hSpinlock);
529 if (!ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost))
530 {
531 pDev = (struct net_device *)ASMAtomicUoReadPtr((void * volatile *)&pThis->u.s.pDev);
532 if (pDev)
533 {
534 dev_hold(pDev);
535 Log(("vboxNetFltLinuxRetainNetDev: Device %p(%s) retained. ref=%d\n",
536 pDev, pDev->name,
537#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
538 netdev_refcnt_read(pDev)
539#else
540 atomic_read(&pDev->refcnt)
541#endif
542 ));
543 }
544 }
545 RTSpinlockRelease(pThis->hSpinlock);
546
547 Log(("vboxNetFltLinuxRetainNetDev - done\n"));
548 return pDev;
549#else
550 return ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
551#endif
552}
553
554
555/**
556 * Release the host interface handle previously retained
557 * by vboxNetFltLinuxRetainNetDev.
558 *
559 * @param pThis The instance.
560 * @param pDev The vboxNetFltLinuxRetainNetDev
561 * return value, NULL is fine.
562 */
563DECLINLINE(void) vboxNetFltLinuxReleaseNetDev(PVBOXNETFLTINS pThis, struct net_device *pDev)
564{
565#if 0
566 Log(("vboxNetFltLinuxReleaseNetDev\n"));
567 NOREF(pThis);
568 if (pDev)
569 {
570 dev_put(pDev);
571 Log(("vboxNetFltLinuxReleaseNetDev: Device %p(%s) released. ref=%d\n",
572 pDev, pDev->name,
573#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
574 netdev_refcnt_read(pDev)
575#else
576 atomic_read(&pDev->refcnt)
577#endif
578 ));
579 }
580 Log(("vboxNetFltLinuxReleaseNetDev - done\n"));
581#endif
582}
583
584#define VBOXNETFLT_CB_TAG(skb) (0xA1C90000 | (skb->dev->ifindex & 0xFFFF))
585#define VBOXNETFLT_SKB_TAG(skb) (*(uint32_t*)&((skb)->cb[sizeof((skb)->cb)-sizeof(uint32_t)]))
586
587/**
588 * Checks whether this is an mbuf created by vboxNetFltLinuxMBufFromSG,
589 * i.e. a buffer which we're pushing and should be ignored by the filter callbacks.
590 *
591 * @returns true / false accordingly.
592 * @param pBuf The sk_buff.
593 */
594DECLINLINE(bool) vboxNetFltLinuxSkBufIsOur(struct sk_buff *pBuf)
595{
596 return VBOXNETFLT_SKB_TAG(pBuf) == VBOXNETFLT_CB_TAG(pBuf);
597}
598
599
600/**
601 * Checks whether this SG list contains a GSO packet.
602 *
603 * @returns true / false accordingly.
604 * @param pSG The (scatter/)gather list.
605 */
606DECLINLINE(bool) vboxNetFltLinuxIsGso(PINTNETSG pSG)
607{
608#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
609 return !((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type == PDMNETWORKGSOTYPE_INVALID);
610#else /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
611 return false;
612#endif /* !VBOXNETFLT_WITH_GSO_XMIT_WIRE && !VBOXNETFLT_WITH_GSO_XMIT_HOST */
613}
614
615
616/**
617 * Find out the frame size (of a single segment in case of GSO frames).
618 *
619 * @returns the frame size.
620 * @param pSG The (scatter/)gather list.
621 */
622DECLINLINE(uint32_t) vboxNetFltLinuxFrameSize(PINTNETSG pSG)
623{
624 uint16_t u16Type = 0;
625 uint32_t cbVlanTag = 0;
626 if (pSG->aSegs[0].cb >= sizeof(RTNETETHERHDR))
627 u16Type = RT_BE2H_U16(((PCRTNETETHERHDR)pSG->aSegs[0].pv)->EtherType);
628 else if (pSG->cbTotal >= sizeof(RTNETETHERHDR))
629 {
630 uint32_t off = RT_OFFSETOF(RTNETETHERHDR, EtherType);
631 uint32_t i;
632 for (i = 0; i < pSG->cSegsUsed; ++i)
633 {
634 if (off <= pSG->aSegs[i].cb)
635 {
636 if (off + sizeof(uint16_t) <= pSG->aSegs[i].cb)
637 u16Type = RT_BE2H_U16(*(uint16_t *)((uintptr_t)pSG->aSegs[i].pv + off));
638 else if (i + 1 < pSG->cSegsUsed)
639 u16Type = RT_BE2H_U16( ((uint16_t)( ((uint8_t *)pSG->aSegs[i].pv)[off] ) << 8)
640 + *(uint8_t *)pSG->aSegs[i + 1].pv); /* ASSUMES no empty segments! */
641 /* else: frame is too short. */
642 break;
643 }
644 off -= pSG->aSegs[i].cb;
645 }
646 }
647 if (u16Type == RTNET_ETHERTYPE_VLAN)
648 cbVlanTag = 4;
649 return (vboxNetFltLinuxIsGso(pSG) ? (uint32_t)pSG->GsoCtx.cbMaxSeg + pSG->GsoCtx.cbHdrsTotal : pSG->cbTotal) - cbVlanTag;
650}
651
652
653/**
654 * Internal worker that create a linux sk_buff for a
655 * (scatter/)gather list.
656 *
657 * @returns Pointer to the sk_buff.
658 * @param pThis The instance.
659 * @param pSG The (scatter/)gather list.
660 * @param fDstWire Set if the destination is the wire.
661 */
662static struct sk_buff *vboxNetFltLinuxSkBufFromSG(PVBOXNETFLTINS pThis, PINTNETSG pSG, bool fDstWire)
663{
664 struct sk_buff *pPkt;
665 struct net_device *pDev;
666 unsigned fGsoType = 0;
667
668 if (pSG->cbTotal == 0)
669 {
670 LogRel(("VBoxNetFlt: Dropped empty packet coming from internal network.\n"));
671 return NULL;
672 }
673 Log5(("VBoxNetFlt: Packet to %s of %d bytes (frame=%d).\n", fDstWire?"wire":"host", pSG->cbTotal, vboxNetFltLinuxFrameSize(pSG)));
674 if (fDstWire && (vboxNetFltLinuxFrameSize(pSG) > ASMAtomicReadU32(&pThis->u.s.cbMtu) + 14))
675 {
676 static bool s_fOnce = true;
677 if (s_fOnce)
678 {
679 s_fOnce = false;
680 printk("VBoxNetFlt: Dropped over-sized packet (%d bytes) coming from internal network.\n", vboxNetFltLinuxFrameSize(pSG));
681 }
682 return NULL;
683 }
684
685 /** @todo We should use fragments mapping the SG buffers with large packets.
686 * 256 bytes seems to be the a threshold used a lot for this. It
687 * requires some nasty work on the intnet side though... */
688 /*
689 * Allocate a packet and copy over the data.
690 */
691 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
692 pPkt = dev_alloc_skb(pSG->cbTotal + NET_IP_ALIGN);
693 if (RT_UNLIKELY(!pPkt))
694 {
695 Log(("vboxNetFltLinuxSkBufFromSG: Failed to allocate sk_buff(%u).\n", pSG->cbTotal));
696 pSG->pvUserData = NULL;
697 return NULL;
698 }
699 pPkt->dev = pDev;
700 pPkt->ip_summed = CHECKSUM_NONE;
701
702 /* Align IP header on 16-byte boundary: 2 + 14 (ethernet hdr size). */
703 skb_reserve(pPkt, NET_IP_ALIGN);
704
705 /* Copy the segments. */
706 skb_put(pPkt, pSG->cbTotal);
707 IntNetSgRead(pSG, pPkt->data);
708
709#if defined(VBOXNETFLT_WITH_GSO_XMIT_WIRE) || defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
710 /*
711 * Setup GSO if used by this packet.
712 */
713 switch ((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type)
714 {
715 default:
716 AssertMsgFailed(("%u (%s)\n", pSG->GsoCtx.u8Type, PDMNetGsoTypeName((PDMNETWORKGSOTYPE)pSG->GsoCtx.u8Type) ));
717 /* fall thru */
718 case PDMNETWORKGSOTYPE_INVALID:
719 fGsoType = 0;
720 break;
721 case PDMNETWORKGSOTYPE_IPV4_TCP:
722 fGsoType = SKB_GSO_TCPV4;
723 break;
724 case PDMNETWORKGSOTYPE_IPV4_UDP:
725 fGsoType = SKB_GSO_UDP;
726 break;
727 case PDMNETWORKGSOTYPE_IPV6_TCP:
728 fGsoType = SKB_GSO_TCPV6;
729 break;
730 }
731 if (fGsoType)
732 {
733 struct skb_shared_info *pShInfo = skb_shinfo(pPkt);
734
735 pShInfo->gso_type = fGsoType | SKB_GSO_DODGY;
736 pShInfo->gso_size = pSG->GsoCtx.cbMaxSeg;
737 pShInfo->gso_segs = PDMNetGsoCalcSegmentCount(&pSG->GsoCtx, pSG->cbTotal);
738
739 /*
740 * We need to set checksum fields even if the packet goes to the host
741 * directly as it may be immediately forwarded by IP layer @bugref{5020}.
742 */
743 Assert(skb_headlen(pPkt) >= pSG->GsoCtx.cbHdrsTotal);
744 pPkt->ip_summed = CHECKSUM_PARTIAL;
745# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
746 pPkt->csum_start = skb_headroom(pPkt) + pSG->GsoCtx.offHdr2;
747 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
748 pPkt->csum_offset = RT_OFFSETOF(RTNETTCP, th_sum);
749 else
750 pPkt->csum_offset = RT_OFFSETOF(RTNETUDP, uh_sum);
751# else
752 pPkt->h.raw = pPkt->data + pSG->GsoCtx.offHdr2;
753 if (fGsoType & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))
754 pPkt->csum = RT_OFFSETOF(RTNETTCP, th_sum);
755 else
756 pPkt->csum = RT_OFFSETOF(RTNETUDP, uh_sum);
757# endif
758 if (!fDstWire)
759 PDMNetGsoPrepForDirectUse(&pSG->GsoCtx, pPkt->data, pSG->cbTotal, PDMNETCSUMTYPE_PSEUDO);
760 }
761#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE || VBOXNETFLT_WITH_GSO_XMIT_HOST */
762
763 /*
764 * Finish up the socket buffer.
765 */
766 pPkt->protocol = eth_type_trans(pPkt, pDev);
767 if (fDstWire)
768 {
769 VBOX_SKB_RESET_NETWORK_HDR(pPkt);
770
771 /* Restore ethernet header back. */
772 skb_push(pPkt, ETH_HLEN); /** @todo VLAN: +4 if VLAN? */
773 VBOX_SKB_RESET_MAC_HDR(pPkt);
774 }
775 VBOXNETFLT_SKB_TAG(pPkt) = VBOXNETFLT_CB_TAG(pPkt);
776
777 return pPkt;
778}
779
780
781/**
782 * Initializes a SG list from an sk_buff.
783 *
784 * @returns Number of segments.
785 * @param pThis The instance.
786 * @param pBuf The sk_buff.
787 * @param pSG The SG.
788 * @param cSegs The number of segments allocated for the SG.
789 * This should match the number in the mbuf exactly!
790 * @param fSrc The source of the frame.
791 * @param pGsoCtx Pointer to the GSO context if it's a GSO
792 * internal network frame. NULL if regular frame.
793 */
794DECLINLINE(void) vboxNetFltLinuxSkBufToSG(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, PINTNETSG pSG,
795 unsigned cSegs, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
796{
797 int i;
798 NOREF(pThis);
799
800 Assert(!skb_shinfo(pBuf)->frag_list);
801
802 if (!pGsoCtx)
803 IntNetSgInitTempSegs(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/);
804 else
805 IntNetSgInitTempSegsGso(pSG, pBuf->len, cSegs, 0 /*cSegsUsed*/, pGsoCtx);
806
807#ifdef VBOXNETFLT_SG_SUPPORT
808 pSG->aSegs[0].cb = skb_headlen(pBuf);
809 pSG->aSegs[0].pv = pBuf->data;
810 pSG->aSegs[0].Phys = NIL_RTHCPHYS;
811
812 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
813 {
814 skb_frag_t *pFrag = &skb_shinfo(pBuf)->frags[i];
815 pSG->aSegs[i+1].cb = pFrag->size;
816 pSG->aSegs[i+1].pv = kmap(pFrag->page);
817 printk("%p = kmap()\n", pSG->aSegs[i+1].pv);
818 pSG->aSegs[i+1].Phys = NIL_RTHCPHYS;
819 }
820 ++i;
821
822#else
823 pSG->aSegs[0].cb = pBuf->len;
824 pSG->aSegs[0].pv = pBuf->data;
825 pSG->aSegs[0].Phys = NIL_RTHCPHYS;
826 i = 1;
827#endif
828
829 pSG->cSegsUsed = i;
830
831#ifdef PADD_RUNT_FRAMES_FROM_HOST
832 /*
833 * Add a trailer if the frame is too small.
834 *
835 * Since we're getting to the packet before it is framed, it has not
836 * yet been padded. The current solution is to add a segment pointing
837 * to a buffer containing all zeros and pray that works for all frames...
838 */
839 if (pSG->cbTotal < 60 && (fSrc & INTNETTRUNKDIR_HOST))
840 {
841 static uint8_t const s_abZero[128] = {0};
842
843 AssertReturnVoid(i < cSegs);
844
845 pSG->aSegs[i].Phys = NIL_RTHCPHYS;
846 pSG->aSegs[i].pv = (void *)&s_abZero[0];
847 pSG->aSegs[i].cb = 60 - pSG->cbTotal;
848 pSG->cbTotal = 60;
849 pSG->cSegsUsed++;
850 Assert(i + 1 <= pSG->cSegsAlloc)
851 }
852#endif
853
854 Log4(("vboxNetFltLinuxSkBufToSG: allocated=%d, segments=%d frags=%d next=%p frag_list=%p pkt_type=%x fSrc=%x\n",
855 pSG->cSegsAlloc, pSG->cSegsUsed, skb_shinfo(pBuf)->nr_frags, pBuf->next, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, fSrc));
856 for (i = 0; i < pSG->cSegsUsed; i++)
857 Log4(("vboxNetFltLinuxSkBufToSG: #%d: cb=%d pv=%p\n",
858 i, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
859}
860
861/**
862 * Packet handler; not really documented - figure it out yourself.
863 *
864 * @returns 0 or EJUSTRETURN - this is probably copy & pastry and thus wrong.
865 */
866#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
867static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
868 struct net_device *pSkbDev,
869 struct packet_type *pPacketType,
870 struct net_device *pOrigDev)
871#else
872static int vboxNetFltLinuxPacketHandler(struct sk_buff *pBuf,
873 struct net_device *pSkbDev,
874 struct packet_type *pPacketType)
875#endif
876{
877 PVBOXNETFLTINS pThis;
878 struct net_device *pDev;
879 LogFlow(("vboxNetFltLinuxPacketHandler: pBuf=%p pSkbDev=%p pPacketType=%p\n",
880 pBuf, pSkbDev, pPacketType));
881#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
882 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",
883 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));
884# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
885 Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
886# endif
887#else
888 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",
889 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));
890#endif
891 /*
892 * Drop it immediately?
893 */
894 if (!pBuf)
895 return 0;
896
897 if (pBuf->pkt_type == PACKET_LOOPBACK)
898 {
899 /*
900 * We are not interested in loopbacked packets as they will always have
901 * another copy going to the wire.
902 */
903 Log2(("vboxNetFltLinuxPacketHandler: dropped loopback packet (cb=%u)\n", pBuf->len));
904 dev_kfree_skb(pBuf); /* We must 'consume' all packets we get (@bugref{6539})! */
905 return 0;
906 }
907
908 pThis = VBOX_FLT_PT_TO_INST(pPacketType);
909 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
910 if (pDev != pSkbDev)
911 {
912 Log(("vboxNetFltLinuxPacketHandler: Devices do not match, pThis may be wrong! pThis=%p\n", pThis));
913 kfree_skb(pBuf); /* This is a failure, so we use kfree_skb instead of dev_kfree_skb. */
914 return 0;
915 }
916
917 Log4(("vboxNetFltLinuxPacketHandler: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
918 if (vboxNetFltLinuxSkBufIsOur(pBuf))
919 {
920 Log2(("vboxNetFltLinuxPacketHandler: got our own sk_buff, drop it.\n"));
921 dev_kfree_skb(pBuf);
922 return 0;
923 }
924
925#ifndef VBOXNETFLT_SG_SUPPORT
926 {
927 /*
928 * Get rid of fragmented packets, they cause too much trouble.
929 */
930 unsigned int uMacLen = pBuf->mac_len;
931 struct sk_buff *pCopy = skb_copy(pBuf, GFP_ATOMIC);
932 dev_kfree_skb(pBuf);
933 if (!pCopy)
934 {
935 LogRel(("VBoxNetFlt: Failed to allocate packet buffer, dropping the packet.\n"));
936 return 0;
937 }
938 pBuf = pCopy;
939 /* Somehow skb_copy ignores mac_len */
940 pBuf->mac_len = uMacLen;
941# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27)
942 /* Restore VLAN tag stripped by host hardware */
943 if (vlan_tx_tag_present(pBuf) && skb_headroom(pBuf) >= VLAN_ETH_HLEN)
944 {
945 uint8_t *pMac = (uint8_t*)skb_mac_header(pBuf);
946 struct vlan_ethhdr *pVHdr = (struct vlan_ethhdr *)(pMac - VLAN_HLEN);
947 memmove(pVHdr, pMac, ETH_ALEN * 2);
948 pVHdr->h_vlan_proto = RT_H2N_U16(ETH_P_8021Q);
949 pVHdr->h_vlan_TCI = RT_H2N_U16(vlan_tx_tag_get(pBuf));
950 pBuf->mac_header -= VLAN_HLEN;
951 pBuf->mac_len += VLAN_HLEN;
952 }
953# endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 27) */
954
955# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
956 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",
957 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));
958# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
959 Log4(("vboxNetFltLinuxPacketHandler: packet dump follows:\n%.*Rhxd\n", pBuf->len-pBuf->data_len, skb_mac_header(pBuf)));
960# endif
961# else
962 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",
963 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));
964# endif
965 }
966#endif
967
968#ifdef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
969 /* Forward it to the internal network. */
970 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
971#else
972 /* Add the packet to transmit queue and schedule the bottom half. */
973 skb_queue_tail(&pThis->u.s.XmitQueue, pBuf);
974 schedule_work(&pThis->u.s.XmitTask);
975 Log4(("vboxNetFltLinuxPacketHandler: scheduled work %p for sk_buff %p\n",
976 &pThis->u.s.XmitTask, pBuf));
977#endif
978
979 /* It does not really matter what we return, it is ignored by the kernel. */
980 return 0;
981}
982
983/**
984 * Calculate the number of INTNETSEG segments the socket buffer will need.
985 *
986 * @returns Segment count.
987 * @param pBuf The socket buffer.
988 */
989DECLINLINE(unsigned) vboxNetFltLinuxCalcSGSegments(struct sk_buff *pBuf)
990{
991#ifdef VBOXNETFLT_SG_SUPPORT
992 unsigned cSegs = 1 + skb_shinfo(pBuf)->nr_frags;
993#else
994 unsigned cSegs = 1;
995#endif
996#ifdef PADD_RUNT_FRAMES_FROM_HOST
997 /* vboxNetFltLinuxSkBufToSG adds a padding segment if it's a runt. */
998 if (pBuf->len < 60)
999 cSegs++;
1000#endif
1001 return cSegs;
1002}
1003
1004/**
1005 * Destroy the intnet scatter / gather buffer created by
1006 * vboxNetFltLinuxSkBufToSG.
1007 */
1008static void vboxNetFltLinuxDestroySG(PINTNETSG pSG)
1009{
1010#ifdef VBOXNETFLT_SG_SUPPORT
1011 int i;
1012
1013 for (i = 0; i < skb_shinfo(pBuf)->nr_frags; i++)
1014 {
1015 printk("kunmap(%p)\n", pSG->aSegs[i+1].pv);
1016 kunmap(pSG->aSegs[i+1].pv);
1017 }
1018#endif
1019 NOREF(pSG);
1020}
1021
1022#ifdef LOG_ENABLED
1023/**
1024 * Logging helper.
1025 */
1026static void vboxNetFltDumpPacket(PINTNETSG pSG, bool fEgress, const char *pszWhere, int iIncrement)
1027{
1028 int i, offSeg;
1029 uint8_t *pInt, *pExt;
1030 static int iPacketNo = 1;
1031 iPacketNo += iIncrement;
1032 if (fEgress)
1033 {
1034 pExt = pSG->aSegs[0].pv;
1035 pInt = pExt + 6;
1036 }
1037 else
1038 {
1039 pInt = pSG->aSegs[0].pv;
1040 pExt = pInt + 6;
1041 }
1042 Log(("VBoxNetFlt: (int)%02x:%02x:%02x:%02x:%02x:%02x"
1043 " %s (%s)%02x:%02x:%02x:%02x:%02x:%02x (%u bytes) packet #%u\n",
1044 pInt[0], pInt[1], pInt[2], pInt[3], pInt[4], pInt[5],
1045 fEgress ? "-->" : "<--", pszWhere,
1046 pExt[0], pExt[1], pExt[2], pExt[3], pExt[4], pExt[5],
1047 pSG->cbTotal, iPacketNo));
1048 if (pSG->cSegsUsed == 1)
1049 {
1050 Log3(("%.*Rhxd\n", pSG->aSegs[0].cb, pSG->aSegs[0].pv));
1051 }
1052 else
1053 {
1054 for (i = 0, offSeg = 0; i < pSG->cSegsUsed; i++)
1055 {
1056 Log3(("-- segment %d at 0x%x (%d bytes) --\n%.*Rhxd\n",
1057 i, offSeg, pSG->aSegs[i].cb, pSG->aSegs[i].cb, pSG->aSegs[i].pv));
1058 offSeg += pSG->aSegs[i].cb;
1059 }
1060 }
1061
1062}
1063#else
1064# define vboxNetFltDumpPacket(a, b, c, d) do {} while (0)
1065#endif
1066
1067#ifdef VBOXNETFLT_WITH_GSO_RECV
1068
1069/**
1070 * Worker for vboxNetFltLinuxForwardToIntNet that checks if we can forwards a
1071 * GSO socket buffer without having to segment it.
1072 *
1073 * @returns true on success, false if needs segmenting.
1074 * @param pThis The net filter instance.
1075 * @param pSkb The GSO socket buffer.
1076 * @param fSrc The source.
1077 * @param pGsoCtx Where to return the GSO context on success.
1078 */
1079static bool vboxNetFltLinuxCanForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc,
1080 PPDMNETWORKGSO pGsoCtx)
1081{
1082 PDMNETWORKGSOTYPE enmGsoType;
1083 uint16_t uEtherType;
1084 unsigned int cbTransport;
1085 unsigned int offTransport;
1086 unsigned int cbTransportHdr;
1087 unsigned uProtocol;
1088 union
1089 {
1090 RTNETIPV4 IPv4;
1091 RTNETIPV6 IPv6;
1092 RTNETTCP Tcp;
1093 uint8_t ab[40];
1094 uint16_t au16[40/2];
1095 uint32_t au32[40/4];
1096 } Buf;
1097
1098 /*
1099 * Check the GSO properties of the socket buffer and make sure it fits.
1100 */
1101 /** @todo Figure out how to handle SKB_GSO_TCP_ECN! */
1102 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_type & ~(SKB_GSO_UDP | SKB_GSO_DODGY | SKB_GSO_TCPV6 | SKB_GSO_TCPV4) ))
1103 {
1104 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_type=%#x\n", skb_shinfo(pSkb)->gso_type));
1105 return false;
1106 }
1107 if (RT_UNLIKELY( skb_shinfo(pSkb)->gso_size < 1
1108 || pSkb->len > VBOX_MAX_GSO_SIZE ))
1109 {
1110 Log5(("vboxNetFltLinuxCanForwardAsGso: gso_size=%#x skb_len=%#x (max=%#x)\n", skb_shinfo(pSkb)->gso_size, pSkb->len, VBOX_MAX_GSO_SIZE));
1111 return false;
1112 }
1113 /*
1114 * It is possible to receive GSO packets from wire if GRO is enabled.
1115 */
1116 if (RT_UNLIKELY(fSrc & INTNETTRUNKDIR_WIRE))
1117 {
1118 Log5(("vboxNetFltLinuxCanForwardAsGso: fSrc=wire\n"));
1119#ifdef VBOXNETFLT_WITH_GRO
1120 /*
1121 * The packet came from the wire and the driver has already consumed
1122 * mac header. We need to restore it back.
1123 */
1124 pSkb->mac_len = skb_network_header(pSkb) - skb_mac_header(pSkb);
1125 skb_push(pSkb, pSkb->mac_len);
1126 Log5(("vboxNetFltLinuxCanForwardAsGso: mac_len=%d data=%p mac_header=%p network_header=%p\n",
1127 pSkb->mac_len, pSkb->data, skb_mac_header(pSkb), skb_network_header(pSkb)));
1128#else /* !VBOXNETFLT_WITH_GRO */
1129 /* Older kernels didn't have GRO. */
1130 return false;
1131#endif /* !VBOXNETFLT_WITH_GRO */
1132 }
1133 else
1134 {
1135 /*
1136 * skb_gso_segment does the following. Do we need to do it as well?
1137 */
1138#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1139 skb_reset_mac_header(pSkb);
1140 pSkb->mac_len = pSkb->network_header - pSkb->mac_header;
1141#else
1142 pSkb->mac.raw = pSkb->data;
1143 pSkb->mac_len = pSkb->nh.raw - pSkb->data;
1144#endif
1145 }
1146
1147 /*
1148 * Switch on the ethertype.
1149 */
1150 uEtherType = pSkb->protocol;
1151 if ( uEtherType == RT_H2N_U16_C(RTNET_ETHERTYPE_VLAN)
1152 && pSkb->mac_len == sizeof(RTNETETHERHDR) + sizeof(uint32_t))
1153 {
1154 uint16_t const *puEtherType = skb_header_pointer(pSkb, sizeof(RTNETETHERHDR) + sizeof(uint16_t), sizeof(uint16_t), &Buf);
1155 if (puEtherType)
1156 uEtherType = *puEtherType;
1157 }
1158 switch (uEtherType)
1159 {
1160 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV4):
1161 {
1162 unsigned int cbHdr;
1163 PCRTNETIPV4 pIPv4 = (PCRTNETIPV4)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv4), &Buf);
1164 if (RT_UNLIKELY(!pIPv4))
1165 {
1166 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv4 hdr\n"));
1167 return false;
1168 }
1169
1170 cbHdr = pIPv4->ip_hl * 4;
1171 cbTransport = RT_N2H_U16(pIPv4->ip_len);
1172 if (RT_UNLIKELY( cbHdr < RTNETIPV4_MIN_LEN
1173 || cbHdr > cbTransport ))
1174 {
1175 Log5(("vboxNetFltLinuxCanForwardAsGso: invalid IPv4 lengths: ip_hl=%u ip_len=%u\n", pIPv4->ip_hl, RT_N2H_U16(pIPv4->ip_len)));
1176 return false;
1177 }
1178 cbTransport -= cbHdr;
1179 offTransport = pSkb->mac_len + cbHdr;
1180 uProtocol = pIPv4->ip_p;
1181 if (uProtocol == RTNETIPV4_PROT_TCP)
1182 enmGsoType = PDMNETWORKGSOTYPE_IPV4_TCP;
1183 else if (uProtocol == RTNETIPV4_PROT_UDP)
1184 enmGsoType = PDMNETWORKGSOTYPE_IPV4_UDP;
1185 else /** @todo IPv6: 4to6 tunneling */
1186 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1187 break;
1188 }
1189
1190 case RT_H2N_U16_C(RTNET_ETHERTYPE_IPV6):
1191 {
1192 PCRTNETIPV6 pIPv6 = (PCRTNETIPV6)skb_header_pointer(pSkb, pSkb->mac_len, sizeof(Buf.IPv6), &Buf);
1193 if (RT_UNLIKELY(!pIPv6))
1194 {
1195 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access IPv6 hdr\n"));
1196 return false;
1197 }
1198
1199 cbTransport = RT_N2H_U16(pIPv6->ip6_plen);
1200 offTransport = pSkb->mac_len + sizeof(RTNETIPV6);
1201 uProtocol = pIPv6->ip6_nxt;
1202 /** @todo IPv6: Dig our way out of the other headers. */
1203 if (uProtocol == RTNETIPV4_PROT_TCP)
1204 enmGsoType = PDMNETWORKGSOTYPE_IPV6_TCP;
1205 else if (uProtocol == RTNETIPV4_PROT_UDP)
1206 enmGsoType = PDMNETWORKGSOTYPE_IPV6_UDP;
1207 else
1208 enmGsoType = PDMNETWORKGSOTYPE_INVALID;
1209 break;
1210 }
1211
1212 default:
1213 Log5(("vboxNetFltLinuxCanForwardAsGso: uEtherType=%#x\n", RT_H2N_U16(uEtherType)));
1214 return false;
1215 }
1216
1217 if (enmGsoType == PDMNETWORKGSOTYPE_INVALID)
1218 {
1219 Log5(("vboxNetFltLinuxCanForwardAsGso: Unsupported protocol %d\n", uProtocol));
1220 return false;
1221 }
1222
1223 if (RT_UNLIKELY( offTransport + cbTransport <= offTransport
1224 || offTransport + cbTransport > pSkb->len
1225 || cbTransport < (uProtocol == RTNETIPV4_PROT_TCP ? RTNETTCP_MIN_LEN : RTNETUDP_MIN_LEN)) )
1226 {
1227 Log5(("vboxNetFltLinuxCanForwardAsGso: Bad transport length; off=%#x + cb=%#x => %#x; skb_len=%#x (%s)\n",
1228 offTransport, cbTransport, offTransport + cbTransport, pSkb->len, PDMNetGsoTypeName(enmGsoType) ));
1229 return false;
1230 }
1231
1232 /*
1233 * Check the TCP/UDP bits.
1234 */
1235 if (uProtocol == RTNETIPV4_PROT_TCP)
1236 {
1237 PCRTNETTCP pTcp = (PCRTNETTCP)skb_header_pointer(pSkb, offTransport, sizeof(Buf.Tcp), &Buf);
1238 if (RT_UNLIKELY(!pTcp))
1239 {
1240 Log5(("vboxNetFltLinuxCanForwardAsGso: failed to access TCP hdr\n"));
1241 return false;
1242 }
1243
1244 cbTransportHdr = pTcp->th_off * 4;
1245 pGsoCtx->cbHdrsSeg = offTransport + cbTransportHdr;
1246 if (RT_UNLIKELY( cbTransportHdr < RTNETTCP_MIN_LEN
1247 || cbTransportHdr > cbTransport
1248 || offTransport + cbTransportHdr >= UINT8_MAX
1249 || offTransport + cbTransportHdr >= pSkb->len ))
1250 {
1251 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for TCP header; off=%#x cb=%#x skb_len=%#x\n", offTransport, cbTransportHdr, pSkb->len));
1252 return false;
1253 }
1254
1255 }
1256 else
1257 {
1258 Assert(uProtocol == RTNETIPV4_PROT_UDP);
1259 cbTransportHdr = sizeof(RTNETUDP);
1260 pGsoCtx->cbHdrsSeg = offTransport; /* Exclude UDP header */
1261 if (RT_UNLIKELY( offTransport + cbTransportHdr >= UINT8_MAX
1262 || offTransport + cbTransportHdr >= pSkb->len ))
1263 {
1264 Log5(("vboxNetFltLinuxCanForwardAsGso: No space for UDP header; off=%#x skb_len=%#x\n", offTransport, pSkb->len));
1265 return false;
1266 }
1267 }
1268
1269 /*
1270 * We're good, init the GSO context.
1271 */
1272 pGsoCtx->u8Type = enmGsoType;
1273 pGsoCtx->cbHdrsTotal = offTransport + cbTransportHdr;
1274 pGsoCtx->cbMaxSeg = skb_shinfo(pSkb)->gso_size;
1275 pGsoCtx->offHdr1 = pSkb->mac_len;
1276 pGsoCtx->offHdr2 = offTransport;
1277 pGsoCtx->u8Unused = 0;
1278
1279 return true;
1280}
1281
1282/**
1283 * Forward the socket buffer as a GSO internal network frame.
1284 *
1285 * @returns IPRT status code.
1286 * @param pThis The net filter instance.
1287 * @param pSkb The GSO socket buffer.
1288 * @param fSrc The source.
1289 * @param pGsoCtx Where to return the GSO context on success.
1290 */
1291static int vboxNetFltLinuxForwardAsGso(PVBOXNETFLTINS pThis, struct sk_buff *pSkb, uint32_t fSrc, PCPDMNETWORKGSO pGsoCtx)
1292{
1293 int rc;
1294 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pSkb);
1295 if (RT_LIKELY(cSegs <= MAX_SKB_FRAGS + 1))
1296 {
1297 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1298 if (RT_LIKELY(pSG))
1299 {
1300 vboxNetFltLinuxSkBufToSG(pThis, pSkb, pSG, cSegs, fSrc, pGsoCtx);
1301
1302 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1303 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1304
1305 vboxNetFltLinuxDestroySG(pSG);
1306 rc = VINF_SUCCESS;
1307 }
1308 else
1309 {
1310 Log(("VBoxNetFlt: Dropping the sk_buff (failure case).\n"));
1311 rc = VERR_NO_MEMORY;
1312 }
1313 }
1314 else
1315 {
1316 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1317 rc = VERR_INTERNAL_ERROR_3;
1318 }
1319
1320 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1321 dev_kfree_skb(pSkb);
1322 return rc;
1323}
1324
1325#endif /* VBOXNETFLT_WITH_GSO_RECV */
1326
1327/**
1328 * Worker for vboxNetFltLinuxForwardToIntNet.
1329 *
1330 * @returns VINF_SUCCESS or VERR_NO_MEMORY.
1331 * @param pThis The net filter instance.
1332 * @param pBuf The socket buffer.
1333 * @param fSrc The source.
1334 */
1335static int vboxNetFltLinuxForwardSegment(PVBOXNETFLTINS pThis, struct sk_buff *pBuf, uint32_t fSrc)
1336{
1337 int rc;
1338 unsigned cSegs = vboxNetFltLinuxCalcSGSegments(pBuf);
1339 if (cSegs <= MAX_SKB_FRAGS + 1)
1340 {
1341 PINTNETSG pSG = (PINTNETSG)alloca(RT_OFFSETOF(INTNETSG, aSegs[cSegs]));
1342 if (RT_LIKELY(pSG))
1343 {
1344 if (fSrc & INTNETTRUNKDIR_WIRE)
1345 {
1346 /*
1347 * The packet came from wire, ethernet header was removed by device driver.
1348 * Restore it using mac_len field. This takes into account VLAN headers too.
1349 */
1350 skb_push(pBuf, pBuf->mac_len);
1351 }
1352
1353 vboxNetFltLinuxSkBufToSG(pThis, pBuf, pSG, cSegs, fSrc, NULL /*pGsoCtx*/);
1354
1355 vboxNetFltDumpPacket(pSG, false, (fSrc & INTNETTRUNKDIR_HOST) ? "host" : "wire", 1);
1356 pThis->pSwitchPort->pfnRecv(pThis->pSwitchPort, NULL /* pvIf */, pSG, fSrc);
1357
1358 vboxNetFltLinuxDestroySG(pSG);
1359 rc = VINF_SUCCESS;
1360 }
1361 else
1362 {
1363 Log(("VBoxNetFlt: Failed to allocate SG buffer.\n"));
1364 rc = VERR_NO_MEMORY;
1365 }
1366 }
1367 else
1368 {
1369 Log(("VBoxNetFlt: Bad sk_buff? cSegs=%#x.\n", cSegs));
1370 rc = VERR_INTERNAL_ERROR_3;
1371 }
1372
1373 Log4(("VBoxNetFlt: Dropping the sk_buff.\n"));
1374 dev_kfree_skb(pBuf);
1375 return rc;
1376}
1377
1378/**
1379 * I won't disclose what I do, figure it out yourself, including pThis referencing.
1380 *
1381 * @param pThis The net filter instance.
1382 * @param pBuf The socket buffer. This is consumed by this function.
1383 */
1384static void vboxNetFltLinuxForwardToIntNet(PVBOXNETFLTINS pThis, struct sk_buff *pBuf)
1385{
1386 uint32_t fSrc = pBuf->pkt_type == PACKET_OUTGOING ? INTNETTRUNKDIR_HOST : INTNETTRUNKDIR_WIRE;
1387
1388#ifdef VBOXNETFLT_WITH_GSO
1389 if (skb_is_gso(pBuf))
1390 {
1391 PDMNETWORKGSO GsoCtx;
1392 Log3(("vboxNetFltLinuxForwardToIntNet: skb len=%u data_len=%u truesize=%u next=%p nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x ip_summed=%d\n",
1393 pBuf->len, pBuf->data_len, pBuf->truesize, pBuf->next, skb_shinfo(pBuf)->nr_frags, skb_shinfo(pBuf)->gso_size, skb_shinfo(pBuf)->gso_segs, skb_shinfo(pBuf)->gso_type, skb_shinfo(pBuf)->frag_list, pBuf->pkt_type, pBuf->ip_summed));
1394# ifdef VBOXNETFLT_WITH_GSO_RECV
1395 if ( (skb_shinfo(pBuf)->gso_type & (SKB_GSO_UDP | SKB_GSO_TCPV6 | SKB_GSO_TCPV4))
1396 && vboxNetFltLinuxCanForwardAsGso(pThis, pBuf, fSrc, &GsoCtx) )
1397 vboxNetFltLinuxForwardAsGso(pThis, pBuf, fSrc, &GsoCtx);
1398 else
1399# endif
1400 {
1401 /* Need to segment the packet */
1402 struct sk_buff *pNext;
1403 struct sk_buff *pSegment = skb_gso_segment(pBuf, 0 /*supported features*/);
1404 if (IS_ERR(pSegment))
1405 {
1406 dev_kfree_skb(pBuf);
1407 LogRel(("VBoxNetFlt: Failed to segment a packet (%d).\n", PTR_ERR(pSegment)));
1408 return;
1409 }
1410
1411 for (; pSegment; pSegment = pNext)
1412 {
1413 Log3(("vboxNetFltLinuxForwardToIntNet: segment len=%u data_len=%u truesize=%u next=%p nr_frags=%u gso_size=%u gso_seqs=%u gso_type=%x frag_list=%p pkt_type=%x\n",
1414 pSegment->len, pSegment->data_len, pSegment->truesize, pSegment->next, skb_shinfo(pSegment)->nr_frags, skb_shinfo(pSegment)->gso_size, skb_shinfo(pSegment)->gso_segs, skb_shinfo(pSegment)->gso_type, skb_shinfo(pSegment)->frag_list, pSegment->pkt_type));
1415 pNext = pSegment->next;
1416 pSegment->next = 0;
1417 vboxNetFltLinuxForwardSegment(pThis, pSegment, fSrc);
1418 }
1419 dev_kfree_skb(pBuf);
1420 }
1421 }
1422 else
1423#endif /* VBOXNETFLT_WITH_GSO */
1424 {
1425 if (pBuf->ip_summed == CHECKSUM_PARTIAL && pBuf->pkt_type == PACKET_OUTGOING)
1426 {
1427#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1428 /*
1429 * Try to work around the problem with CentOS 4.7 and 5.2 (2.6.9
1430 * and 2.6.18 kernels), they pass wrong 'h' pointer down. We take IP
1431 * header length from the header itself and reconstruct 'h' pointer
1432 * to TCP (or whatever) header.
1433 */
1434 unsigned char *tmp = pBuf->h.raw;
1435 if (pBuf->h.raw == pBuf->nh.raw && pBuf->protocol == htons(ETH_P_IP))
1436 pBuf->h.raw = pBuf->nh.raw + pBuf->nh.iph->ihl * 4;
1437#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1438 if (VBOX_SKB_CHECKSUM_HELP(pBuf))
1439 {
1440 LogRel(("VBoxNetFlt: Failed to compute checksum, dropping the packet.\n"));
1441 dev_kfree_skb(pBuf);
1442 return;
1443 }
1444#if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18)
1445 /* Restore the original (wrong) pointer. */
1446 pBuf->h.raw = tmp;
1447#endif /* LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 18) */
1448 }
1449 vboxNetFltLinuxForwardSegment(pThis, pBuf, fSrc);
1450 }
1451}
1452
1453#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1454/**
1455 * Work queue handler that forwards the socket buffers queued by
1456 * vboxNetFltLinuxPacketHandler to the internal network.
1457 *
1458 * @param pWork The work queue.
1459 */
1460# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
1461static void vboxNetFltLinuxXmitTask(struct work_struct *pWork)
1462# else
1463static void vboxNetFltLinuxXmitTask(void *pWork)
1464# endif
1465{
1466 PVBOXNETFLTINS pThis = VBOX_FLT_XT_TO_INST(pWork);
1467 struct sk_buff *pBuf;
1468
1469 Log4(("vboxNetFltLinuxXmitTask: Got work %p.\n", pWork));
1470
1471 /*
1472 * Active? Retain the instance and increment the busy counter.
1473 */
1474 if (vboxNetFltTryRetainBusyActive(pThis))
1475 {
1476 while ((pBuf = skb_dequeue(&pThis->u.s.XmitQueue)) != NULL)
1477 vboxNetFltLinuxForwardToIntNet(pThis, pBuf);
1478
1479 vboxNetFltRelease(pThis, true /* fBusy */);
1480 }
1481 else
1482 {
1483 /** @todo Shouldn't we just drop the packets here? There is little point in
1484 * making them accumulate when the VM is paused and it'll only waste
1485 * kernel memory anyway... Hmm. maybe wait a short while (2-5 secs)
1486 * before start draining the packets (goes for the intnet ring buf
1487 * too)? */
1488 }
1489}
1490#endif /* !VBOXNETFLT_LINUX_NO_XMIT_QUEUE */
1491
1492/**
1493 * Reports the GSO capabilities of the hardware NIC.
1494 *
1495 * @param pThis The net filter instance. The caller hold a
1496 * reference to this.
1497 */
1498static void vboxNetFltLinuxReportNicGsoCapabilities(PVBOXNETFLTINS pThis)
1499{
1500#ifdef VBOXNETFLT_WITH_GSO_XMIT_WIRE
1501 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1502 {
1503 struct net_device *pDev;
1504 PINTNETTRUNKSWPORT pSwitchPort;
1505 unsigned int fFeatures;
1506
1507 RTSpinlockAcquire(pThis->hSpinlock);
1508
1509 pSwitchPort = pThis->pSwitchPort; /* this doesn't need to be here, but it doesn't harm. */
1510 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1511 if (pDev)
1512 fFeatures = pDev->features;
1513 else
1514 fFeatures = 0;
1515
1516 RTSpinlockRelease(pThis->hSpinlock);
1517
1518 if (pThis->pSwitchPort)
1519 {
1520 /* Set/update the GSO capabilities of the NIC. */
1521 uint32_t fGsoCapabilites = 0;
1522 if (fFeatures & NETIF_F_TSO)
1523 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP);
1524 if (fFeatures & NETIF_F_TSO6)
1525 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP);
1526# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
1527 if (fFeatures & NETIF_F_UFO)
1528 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP);
1529 if (fFeatures & NETIF_F_UFO)
1530 fGsoCapabilites |= RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP);
1531# endif
1532 Log3(("vboxNetFltLinuxReportNicGsoCapabilities: reporting wire %s%s%s%s\n",
1533 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)) ? "tso " : "",
1534 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)) ? "tso6 " : "",
1535 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)) ? "ufo " : "",
1536 (fGsoCapabilites & RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)) ? "ufo6 " : ""));
1537 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort, fGsoCapabilites, INTNETTRUNKDIR_WIRE);
1538 }
1539
1540 vboxNetFltRelease(pThis, true /*fBusy*/);
1541 }
1542#endif /* VBOXNETFLT_WITH_GSO_XMIT_WIRE */
1543}
1544
1545/**
1546 * Helper that determines whether the host (ignoreing us) is operating the
1547 * interface in promiscuous mode or not.
1548 */
1549static bool vboxNetFltLinuxPromiscuous(PVBOXNETFLTINS pThis)
1550{
1551 bool fRc = false;
1552 struct net_device * pDev = vboxNetFltLinuxRetainNetDev(pThis);
1553 if (pDev)
1554 {
1555 fRc = !!(pDev->promiscuity - (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet) & 1));
1556 LogFlow(("vboxNetFltPortOsIsPromiscuous: returns %d, pDev->promiscuity=%d, fPromiscuousSet=%d\n",
1557 fRc, pDev->promiscuity, pThis->u.s.fPromiscuousSet));
1558 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
1559 }
1560 return fRc;
1561}
1562
1563/**
1564 * Does this device needs link state change signaled?
1565 * Currently we need it for our own VBoxNetAdp and TAP.
1566 */
1567static bool vboxNetFltNeedsLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev)
1568{
1569 if (pDev->ethtool_ops && pDev->ethtool_ops->get_drvinfo)
1570 {
1571 struct ethtool_drvinfo Info;
1572
1573 memset(&Info, 0, sizeof(Info));
1574 Info.cmd = ETHTOOL_GDRVINFO;
1575 pDev->ethtool_ops->get_drvinfo(pDev, &Info);
1576 Log3(("%s: driver=%.*s version=%.*s bus_info=%.*s\n",
1577 __FUNCTION__,
1578 sizeof(Info.driver), Info.driver,
1579 sizeof(Info.version), Info.version,
1580 sizeof(Info.bus_info), Info.bus_info));
1581
1582 if (!strncmp(Info.driver, "vboxnet", sizeof(Info.driver)))
1583 return true;
1584
1585#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 36) /* TAP started doing carrier */
1586 return !strncmp(Info.driver, "tun", 4)
1587 && !strncmp(Info.bus_info, "tap", 4);
1588#endif
1589 }
1590
1591 return false;
1592}
1593
1594/**
1595 * Some devices need link state change when filter attaches/detaches
1596 * since the filter is their link in a sense.
1597 */
1598static void vboxNetFltSetLinkState(PVBOXNETFLTINS pThis, struct net_device *pDev, bool fLinkUp)
1599{
1600 if (vboxNetFltNeedsLinkState(pThis, pDev))
1601 {
1602 Log3(("%s: bringing device link %s\n",
1603 __FUNCTION__, fLinkUp ? "up" : "down"));
1604 netif_tx_lock_bh(pDev);
1605 if (fLinkUp)
1606 netif_carrier_on(pDev);
1607 else
1608 netif_carrier_off(pDev);
1609 netif_tx_unlock_bh(pDev);
1610 }
1611}
1612
1613/**
1614 * Internal worker for vboxNetFltLinuxNotifierCallback.
1615 *
1616 * @returns VBox status code.
1617 * @param pThis The instance.
1618 * @param pDev The device to attach to.
1619 */
1620static int vboxNetFltLinuxAttachToInterface(PVBOXNETFLTINS pThis, struct net_device *pDev)
1621{
1622 LogFlow(("vboxNetFltLinuxAttachToInterface: pThis=%p (%s)\n", pThis, pThis->szName));
1623
1624 /*
1625 * Retain and store the device.
1626 */
1627 dev_hold(pDev);
1628
1629 RTSpinlockAcquire(pThis->hSpinlock);
1630 ASMAtomicUoWritePtr(&pThis->u.s.pDev, pDev);
1631 RTSpinlockRelease(pThis->hSpinlock);
1632
1633 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) retained. ref=%d\n",
1634 pDev, pDev->name,
1635#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1636 netdev_refcnt_read(pDev)
1637#else
1638 atomic_read(&pDev->refcnt)
1639#endif
1640 ));
1641 Log(("vboxNetFltLinuxAttachToInterface: Got pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1642 pDev, pThis, ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *)));
1643
1644 /* Get the mac address while we still have a valid net_device reference. */
1645 memcpy(&pThis->u.s.MacAddr, pDev->dev_addr, sizeof(pThis->u.s.MacAddr));
1646 /* Initialize MTU */
1647 pThis->u.s.cbMtu = pDev->mtu;
1648
1649 /*
1650 * Install a packet filter for this device with a protocol wildcard (ETH_P_ALL).
1651 */
1652 pThis->u.s.PacketType.type = __constant_htons(ETH_P_ALL);
1653 pThis->u.s.PacketType.dev = pDev;
1654 pThis->u.s.PacketType.func = vboxNetFltLinuxPacketHandler;
1655 dev_add_pack(&pThis->u.s.PacketType);
1656 ASMAtomicUoWriteBool(&pThis->u.s.fPacketHandler, true);
1657 Log(("vboxNetFltLinuxAttachToInterface: this=%p: Packet handler installed.\n", pThis));
1658
1659#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1660 vboxNetFltLinuxHookDev(pThis, pDev);
1661#endif
1662
1663 /*
1664 * Are we the "carrier" for this device (e.g. vboxnet or tap)?
1665 */
1666 vboxNetFltSetLinkState(pThis, pDev, true);
1667
1668 /*
1669 * Set indicators that require the spinlock. Be abit paranoid about racing
1670 * the device notification handle.
1671 */
1672 RTSpinlockAcquire(pThis->hSpinlock);
1673 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1674 if (pDev)
1675 {
1676 ASMAtomicUoWriteBool(&pThis->fDisconnectedFromHost, false);
1677 ASMAtomicUoWriteBool(&pThis->u.s.fRegistered, true);
1678 pDev = NULL; /* don't dereference it */
1679 }
1680 RTSpinlockRelease(pThis->hSpinlock);
1681
1682 /*
1683 * If the above succeeded report GSO capabilities, if not undo and
1684 * release the device.
1685 */
1686 if (!pDev)
1687 {
1688 Assert(pThis->pSwitchPort);
1689 if (vboxNetFltTryRetainBusyNotDisconnected(pThis))
1690 {
1691 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1692 pThis->pSwitchPort->pfnReportMacAddress(pThis->pSwitchPort, &pThis->u.s.MacAddr);
1693 pThis->pSwitchPort->pfnReportPromiscuousMode(pThis->pSwitchPort, vboxNetFltLinuxPromiscuous(pThis));
1694 pThis->pSwitchPort->pfnReportNoPreemptDsts(pThis->pSwitchPort, INTNETTRUNKDIR_WIRE | INTNETTRUNKDIR_HOST);
1695 vboxNetFltRelease(pThis, true /*fBusy*/);
1696 }
1697 }
1698 else
1699 {
1700#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1701 vboxNetFltLinuxUnhookDev(pThis, pDev);
1702#endif
1703 RTSpinlockAcquire(pThis->hSpinlock);
1704 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1705 RTSpinlockRelease(pThis->hSpinlock);
1706 dev_put(pDev);
1707 Log(("vboxNetFltLinuxAttachToInterface: Device %p(%s) released. ref=%d\n",
1708 pDev, pDev->name,
1709#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1710 netdev_refcnt_read(pDev)
1711#else
1712 atomic_read(&pDev->refcnt)
1713#endif
1714 ));
1715 }
1716
1717 LogRel(("VBoxNetFlt: attached to '%s' / %RTmac\n", pThis->szName, &pThis->u.s.MacAddr));
1718 return VINF_SUCCESS;
1719}
1720
1721
1722static int vboxNetFltLinuxUnregisterDevice(PVBOXNETFLTINS pThis, struct net_device *pDev)
1723{
1724 bool fRegistered;
1725 Assert(!pThis->fDisconnectedFromHost);
1726
1727#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
1728 vboxNetFltLinuxUnhookDev(pThis, pDev);
1729#endif
1730
1731 if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
1732 {
1733 dev_remove_pack(&pThis->u.s.PacketType);
1734 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: packet handler removed.\n", pThis));
1735 }
1736
1737 RTSpinlockAcquire(pThis->hSpinlock);
1738 fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
1739 if (fRegistered)
1740 {
1741 ASMAtomicWriteBool(&pThis->fDisconnectedFromHost, true);
1742 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
1743 }
1744 RTSpinlockRelease(pThis->hSpinlock);
1745
1746 if (fRegistered)
1747 {
1748#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
1749 skb_queue_purge(&pThis->u.s.XmitQueue);
1750#endif
1751 Log(("vboxNetFltLinuxUnregisterDevice: this=%p: xmit queue purged.\n", pThis));
1752 Log(("vboxNetFltLinuxUnregisterDevice: Device %p(%s) released. ref=%d\n",
1753 pDev, pDev->name,
1754#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
1755 netdev_refcnt_read(pDev)
1756#else
1757 atomic_read(&pDev->refcnt)
1758#endif
1759 ));
1760 dev_put(pDev);
1761 }
1762
1763 return NOTIFY_OK;
1764}
1765
1766static int vboxNetFltLinuxDeviceIsUp(PVBOXNETFLTINS pThis, struct net_device *pDev)
1767{
1768 /* Check if we are not suspended and promiscuous mode has not been set. */
1769 if ( pThis->enmTrunkState == INTNETTRUNKIFSTATE_ACTIVE
1770 && !ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1771 {
1772 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1773 dev_set_promiscuity(pDev, 1);
1774 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, true);
1775 Log(("vboxNetFltLinuxDeviceIsUp: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1776 }
1777 else
1778 Log(("vboxNetFltLinuxDeviceIsUp: no need to enable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1779 return NOTIFY_OK;
1780}
1781
1782static int vboxNetFltLinuxDeviceGoingDown(PVBOXNETFLTINS pThis, struct net_device *pDev)
1783{
1784 /* Undo promiscuous mode if we has set it. */
1785 if (ASMAtomicUoReadBool(&pThis->u.s.fPromiscuousSet))
1786 {
1787 /* Note that there is no need for locking as the kernel got hold of the lock already. */
1788 dev_set_promiscuity(pDev, -1);
1789 ASMAtomicWriteBool(&pThis->u.s.fPromiscuousSet, false);
1790 Log(("vboxNetFltLinuxDeviceGoingDown: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1791 }
1792 else
1793 Log(("vboxNetFltLinuxDeviceGoingDown: no need to disable promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
1794 return NOTIFY_OK;
1795}
1796
1797/**
1798 * Callback for listening to MTU change event.
1799 *
1800 * We need to track changes of host's inteface MTU to discard over-sized frames
1801 * coming from the internal network as they may hang the TX queue of host's
1802 * adapter.
1803 *
1804 * @returns NOTIFY_OK
1805 * @param pThis The netfilter instance.
1806 * @param pDev Pointer to device structure of host's interface.
1807 */
1808static int vboxNetFltLinuxDeviceMtuChange(PVBOXNETFLTINS pThis, struct net_device *pDev)
1809{
1810 ASMAtomicWriteU32(&pThis->u.s.cbMtu, pDev->mtu);
1811 Log(("vboxNetFltLinuxDeviceMtuChange: set MTU for %s to %d\n", pThis->szName, pDev->mtu));
1812 return NOTIFY_OK;
1813}
1814
1815#ifdef LOG_ENABLED
1816/** Stringify the NETDEV_XXX constants. */
1817static const char *vboxNetFltLinuxGetNetDevEventName(unsigned long ulEventType)
1818{
1819 const char *pszEvent = "NETDEV_<unknown>";
1820 switch (ulEventType)
1821 {
1822 case NETDEV_REGISTER: pszEvent = "NETDEV_REGISTER"; break;
1823 case NETDEV_UNREGISTER: pszEvent = "NETDEV_UNREGISTER"; break;
1824 case NETDEV_UP: pszEvent = "NETDEV_UP"; break;
1825 case NETDEV_DOWN: pszEvent = "NETDEV_DOWN"; break;
1826 case NETDEV_REBOOT: pszEvent = "NETDEV_REBOOT"; break;
1827 case NETDEV_CHANGENAME: pszEvent = "NETDEV_CHANGENAME"; break;
1828 case NETDEV_CHANGE: pszEvent = "NETDEV_CHANGE"; break;
1829 case NETDEV_CHANGEMTU: pszEvent = "NETDEV_CHANGEMTU"; break;
1830 case NETDEV_CHANGEADDR: pszEvent = "NETDEV_CHANGEADDR"; break;
1831 case NETDEV_GOING_DOWN: pszEvent = "NETDEV_GOING_DOWN"; break;
1832# ifdef NETDEV_FEAT_CHANGE
1833 case NETDEV_FEAT_CHANGE: pszEvent = "NETDEV_FEAT_CHANGE"; break;
1834# endif
1835 }
1836 return pszEvent;
1837}
1838#endif /* LOG_ENABLED */
1839
1840/**
1841 * Callback for listening to netdevice events.
1842 *
1843 * This works the rediscovery, clean up on unregistration, promiscuity on
1844 * up/down, and GSO feature changes from ethtool.
1845 *
1846 * @returns NOTIFY_OK
1847 * @param self Pointer to our notifier registration block.
1848 * @param ulEventType The event.
1849 * @param ptr Event specific, but it is usually the device it
1850 * relates to.
1851 */
1852static int vboxNetFltLinuxNotifierCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
1853
1854{
1855 PVBOXNETFLTINS pThis = VBOX_FLT_NB_TO_INST(self);
1856 struct net_device *pMyDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
1857 struct net_device *pDev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
1858 int rc = NOTIFY_OK;
1859
1860 Log(("VBoxNetFlt: got event %s(0x%lx) on %s, pDev=%p pThis=%p pThis->u.s.pDev=%p\n",
1861 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType, pDev->name, pDev, pThis, pMyDev));
1862
1863 if (ulEventType == NETDEV_REGISTER)
1864 {
1865#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 24) /* cgroups/namespaces introduced */
1866# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 26)
1867# define VBOX_DEV_NET(dev) dev_net(dev)
1868# define VBOX_NET_EQ(n1, n2) net_eq((n1), (n2))
1869# else
1870# define VBOX_DEV_NET(dev) ((dev)->nd_net)
1871# define VBOX_NET_EQ(n1, n2) ((n1) == (n2))
1872# endif
1873 struct net *pMyNet = current->nsproxy->net_ns;
1874 struct net *pDevNet = VBOX_DEV_NET(pDev);
1875
1876 if (VBOX_NET_EQ(pDevNet, pMyNet))
1877#endif /* namespaces */
1878 {
1879 if (strcmp(pDev->name, pThis->szName) == 0)
1880 {
1881 vboxNetFltLinuxAttachToInterface(pThis, pDev);
1882 }
1883 }
1884 }
1885 else
1886 {
1887 if (pDev == pMyDev)
1888 {
1889 switch (ulEventType)
1890 {
1891 case NETDEV_UNREGISTER:
1892 rc = vboxNetFltLinuxUnregisterDevice(pThis, pDev);
1893 break;
1894 case NETDEV_UP:
1895 rc = vboxNetFltLinuxDeviceIsUp(pThis, pDev);
1896 break;
1897 case NETDEV_GOING_DOWN:
1898 rc = vboxNetFltLinuxDeviceGoingDown(pThis, pDev);
1899 break;
1900 case NETDEV_CHANGEMTU:
1901 rc = vboxNetFltLinuxDeviceMtuChange(pThis, pDev);
1902 break;
1903 case NETDEV_CHANGENAME:
1904 break;
1905#ifdef NETDEV_FEAT_CHANGE
1906 case NETDEV_FEAT_CHANGE:
1907 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
1908 break;
1909#endif
1910 }
1911 }
1912 }
1913
1914 return rc;
1915}
1916
1917/*
1918 * Initial enumeration of netdevs. Called with NETDEV_REGISTER by
1919 * register_netdevice_notifier() under rtnl lock.
1920 */
1921static int vboxNetFltLinuxEnumeratorCallback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
1922{
1923 PVBOXNETFLTINS pThis = ((PVBOXNETFLTNOTIFIER)self)->pThis;
1924 struct net_device *dev = VBOX_NETDEV_NOTIFIER_INFO_TO_DEV(ptr);
1925 struct in_device *in_dev;
1926 struct inet6_dev *in6_dev;
1927
1928 if (ulEventType != NETDEV_REGISTER)
1929 return NOTIFY_OK;
1930
1931 if (RT_UNLIKELY(pThis->pSwitchPort->pfnNotifyHostAddress == NULL))
1932 return NOTIFY_OK;
1933
1934 /*
1935 * IPv4
1936 */
1937#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 14)
1938 in_dev = __in_dev_get_rtnl(dev);
1939#else
1940 in_dev = __in_dev_get(dev);
1941#endif
1942 if (in_dev != NULL)
1943 {
1944 for_ifa(in_dev) {
1945 if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
1946 return NOTIFY_OK;
1947
1948 if ( dev != pThis->u.s.pDev
1949 && VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
1950 continue;
1951
1952 Log(("%s: %s: IPv4 addr %RTnaipv4 mask %RTnaipv4\n",
1953 __FUNCTION__, VBOX_NETDEV_NAME(dev),
1954 ifa->ifa_address, ifa->ifa_mask));
1955
1956 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1957 /* :fAdded */ true, kIntNetAddrType_IPv4, &ifa->ifa_address);
1958 } endfor_ifa(in_dev);
1959 }
1960
1961 /*
1962 * IPv6
1963 */
1964 in6_dev = __in6_dev_get(dev);
1965 if (in6_dev != NULL)
1966 {
1967 struct inet6_ifaddr *ifa;
1968
1969 read_lock_bh(&in6_dev->lock);
1970#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 35)
1971 list_for_each_entry(ifa, &in6_dev->addr_list, if_list)
1972#else
1973 for (ifa = in6_dev->addr_list; ifa != NULL; ifa = ifa->if_next)
1974#endif
1975 {
1976 if ( dev != pThis->u.s.pDev
1977 && ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_LOOPBACK))
1978 continue;
1979
1980 Log(("%s: %s: IPv6 addr %RTnaipv6/%u\n",
1981 __FUNCTION__, VBOX_NETDEV_NAME(dev),
1982 &ifa->addr, (unsigned)ifa->prefix_len));
1983
1984 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort,
1985 /* :fAdded */ true, kIntNetAddrType_IPv6, &ifa->addr);
1986 }
1987 read_unlock_bh(&in6_dev->lock);
1988 }
1989
1990 return NOTIFY_OK;
1991}
1992
1993
1994static int vboxNetFltLinuxNotifierIPv4Callback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
1995{
1996 PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv4);
1997 struct net_device *pDev, *pEventDev;
1998 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1999 bool fMyDev;
2000 int rc = NOTIFY_OK;
2001
2002 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2003 pEventDev = ifa->ifa_dev->dev;
2004 fMyDev = (pDev == pEventDev);
2005 Log(("VBoxNetFlt: %s: IPv4 event %s(0x%lx) %s: addr %RTnaipv4 mask %RTnaipv4\n",
2006 pDev ? VBOX_NETDEV_NAME(pDev) : "<???>",
2007 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2008 pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<???>",
2009 ifa->ifa_address, ifa->ifa_mask));
2010
2011 if (pDev != NULL)
2012 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2013
2014 if (VBOX_IPV4_IS_LOOPBACK(ifa->ifa_address))
2015 return NOTIFY_OK;
2016
2017 if ( !fMyDev
2018 && VBOX_IPV4_IS_LINKLOCAL_169(ifa->ifa_address))
2019 return NOTIFY_OK;
2020
2021 if (pThis->pSwitchPort->pfnNotifyHostAddress)
2022 {
2023 bool fAdded;
2024 if (ulEventType == NETDEV_UP)
2025 fAdded = true;
2026 else if (ulEventType == NETDEV_DOWN)
2027 fAdded = false;
2028 else
2029 return NOTIFY_OK;
2030
2031 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2032 kIntNetAddrType_IPv4, &ifa->ifa_local);
2033 }
2034
2035 return rc;
2036}
2037
2038
2039static int vboxNetFltLinuxNotifierIPv6Callback(struct notifier_block *self, unsigned long ulEventType, void *ptr)
2040{
2041 PVBOXNETFLTINS pThis = RT_FROM_MEMBER(self, VBOXNETFLTINS, u.s.NotifierIPv6);
2042 struct net_device *pDev, *pEventDev;
2043 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
2044 bool fMyDev;
2045 int rc = NOTIFY_OK;
2046
2047 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2048 pEventDev = ifa->idev->dev;
2049 fMyDev = (pDev == pEventDev);
2050 Log(("VBoxNetFlt: %s: IPv6 event %s(0x%lx) %s: %RTnaipv6\n",
2051 pDev ? VBOX_NETDEV_NAME(pDev) : "<???>",
2052 vboxNetFltLinuxGetNetDevEventName(ulEventType), ulEventType,
2053 pEventDev ? VBOX_NETDEV_NAME(pEventDev) : "<???>",
2054 &ifa->addr));
2055
2056 if (pDev != NULL)
2057 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2058
2059 if ( !fMyDev
2060 && ipv6_addr_type(&ifa->addr) & (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_LOOPBACK))
2061 return NOTIFY_OK;
2062
2063 if (pThis->pSwitchPort->pfnNotifyHostAddress)
2064 {
2065 bool fAdded;
2066 if (ulEventType == NETDEV_UP)
2067 fAdded = true;
2068 else if (ulEventType == NETDEV_DOWN)
2069 fAdded = false;
2070 else
2071 return NOTIFY_OK;
2072
2073 pThis->pSwitchPort->pfnNotifyHostAddress(pThis->pSwitchPort, fAdded,
2074 kIntNetAddrType_IPv6, &ifa->addr);
2075 }
2076
2077 return rc;
2078}
2079
2080
2081bool vboxNetFltOsMaybeRediscovered(PVBOXNETFLTINS pThis)
2082{
2083 return !ASMAtomicUoReadBool(&pThis->fDisconnectedFromHost);
2084}
2085
2086int vboxNetFltPortOsXmit(PVBOXNETFLTINS pThis, void *pvIfData, PINTNETSG pSG, uint32_t fDst)
2087{
2088 struct net_device * pDev;
2089 int err;
2090 int rc = VINF_SUCCESS;
2091 IPRT_LINUX_SAVE_EFL_AC();
2092 NOREF(pvIfData);
2093
2094 LogFlow(("vboxNetFltPortOsXmit: pThis=%p (%s)\n", pThis, pThis->szName));
2095
2096 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2097 if (pDev)
2098 {
2099 /*
2100 * Create a sk_buff for the gather list and push it onto the wire.
2101 */
2102 if (fDst & INTNETTRUNKDIR_WIRE)
2103 {
2104 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, true);
2105 if (pBuf)
2106 {
2107 vboxNetFltDumpPacket(pSG, true, "wire", 1);
2108 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2109 Log4(("vboxNetFltPortOsXmit: dev_queue_xmit(%p)\n", pBuf));
2110 err = dev_queue_xmit(pBuf);
2111 if (err)
2112 rc = RTErrConvertFromErrno(err);
2113 }
2114 else
2115 rc = VERR_NO_MEMORY;
2116 }
2117
2118 /*
2119 * Create a sk_buff for the gather list and push it onto the host stack.
2120 */
2121 if (fDst & INTNETTRUNKDIR_HOST)
2122 {
2123 struct sk_buff *pBuf = vboxNetFltLinuxSkBufFromSG(pThis, pSG, false);
2124 if (pBuf)
2125 {
2126 vboxNetFltDumpPacket(pSG, true, "host", (fDst & INTNETTRUNKDIR_WIRE) ? 0 : 1);
2127 Log4(("vboxNetFltPortOsXmit: pBuf->cb dump:\n%.*Rhxd\n", sizeof(pBuf->cb), pBuf->cb));
2128 Log4(("vboxNetFltPortOsXmit: netif_rx_ni(%p)\n", pBuf));
2129 err = netif_rx_ni(pBuf);
2130 if (err)
2131 rc = RTErrConvertFromErrno(err);
2132 }
2133 else
2134 rc = VERR_NO_MEMORY;
2135 }
2136
2137 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2138 }
2139
2140 IPRT_LINUX_RESTORE_EFL_AC();
2141 return rc;
2142}
2143
2144
2145void vboxNetFltPortOsSetActive(PVBOXNETFLTINS pThis, bool fActive)
2146{
2147 struct net_device *pDev;
2148 IPRT_LINUX_SAVE_EFL_AC();
2149
2150 LogFlow(("vboxNetFltPortOsSetActive: pThis=%p (%s), fActive=%RTbool, fDisablePromiscuous=%RTbool\n",
2151 pThis, pThis->szName, fActive, pThis->fDisablePromiscuous));
2152
2153 if (pThis->fDisablePromiscuous)
2154 return;
2155
2156 pDev = vboxNetFltLinuxRetainNetDev(pThis);
2157 if (pDev)
2158 {
2159 /*
2160 * This api is a bit weird, the best reference is the code.
2161 *
2162 * Also, we have a bit or race conditions wrt the maintenance of
2163 * host the interface promiscuity for vboxNetFltPortOsIsPromiscuous.
2164 */
2165#ifdef LOG_ENABLED
2166 u_int16_t fIf;
2167 unsigned const cPromiscBefore = pDev->promiscuity;
2168#endif
2169 if (fActive)
2170 {
2171 Assert(!pThis->u.s.fPromiscuousSet);
2172
2173 rtnl_lock();
2174 dev_set_promiscuity(pDev, 1);
2175 rtnl_unlock();
2176 pThis->u.s.fPromiscuousSet = true;
2177 Log(("vboxNetFltPortOsSetActive: enabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2178 }
2179 else
2180 {
2181 if (pThis->u.s.fPromiscuousSet)
2182 {
2183 rtnl_lock();
2184 dev_set_promiscuity(pDev, -1);
2185 rtnl_unlock();
2186 Log(("vboxNetFltPortOsSetActive: disabled promiscuous mode on %s (%d)\n", pThis->szName, pDev->promiscuity));
2187 }
2188 pThis->u.s.fPromiscuousSet = false;
2189
2190#ifdef LOG_ENABLED
2191 fIf = dev_get_flags(pDev);
2192 Log(("VBoxNetFlt: fIf=%#x; %d->%d\n", fIf, cPromiscBefore, pDev->promiscuity));
2193#endif
2194 }
2195
2196 vboxNetFltLinuxReleaseNetDev(pThis, pDev);
2197 }
2198 IPRT_LINUX_RESTORE_EFL_AC();
2199}
2200
2201
2202int vboxNetFltOsDisconnectIt(PVBOXNETFLTINS pThis)
2203{
2204 /*
2205 * Remove packet handler when we get disconnected from internal switch as
2206 * we don't want the handler to forward packets to disconnected switch.
2207 */
2208 if (ASMAtomicCmpXchgBool(&pThis->u.s.fPacketHandler, false, true))
2209 {
2210 IPRT_LINUX_SAVE_EFL_AC();
2211 dev_remove_pack(&pThis->u.s.PacketType);
2212 Log(("vboxNetFltOsDisconnectIt: this=%p: Packet handler removed.\n", pThis));
2213 IPRT_LINUX_RESTORE_EFL_AC();
2214 }
2215 return VINF_SUCCESS;
2216}
2217
2218
2219int vboxNetFltOsConnectIt(PVBOXNETFLTINS pThis)
2220{
2221 IPRT_LINUX_SAVE_EFL_AC();
2222
2223 /*
2224 * Report the GSO capabilities of the host and device (if connected).
2225 * Note! No need to mark ourselves busy here.
2226 */
2227 /** @todo duplicate work here now? Attach */
2228#if defined(VBOXNETFLT_WITH_GSO_XMIT_HOST)
2229 Log3(("vboxNetFltOsConnectIt: reporting host tso tso6 ufo\n"));
2230 pThis->pSwitchPort->pfnReportGsoCapabilities(pThis->pSwitchPort,
2231 0
2232 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_TCP)
2233 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_TCP)
2234 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV4_UDP)
2235# if 0 /** @todo GSO: Test UDP offloading (UFO) on linux. */
2236 | RT_BIT_32(PDMNETWORKGSOTYPE_IPV6_UDP)
2237# endif
2238 , INTNETTRUNKDIR_HOST);
2239
2240#endif
2241 vboxNetFltLinuxReportNicGsoCapabilities(pThis);
2242
2243 IPRT_LINUX_RESTORE_EFL_AC();
2244 return VINF_SUCCESS;
2245}
2246
2247
2248void vboxNetFltOsDeleteInstance(PVBOXNETFLTINS pThis)
2249{
2250 struct net_device *pDev;
2251 bool fRegistered;
2252 IPRT_LINUX_SAVE_EFL_AC();
2253
2254#ifdef VBOXNETFLT_WITH_HOST2WIRE_FILTER
2255 vboxNetFltLinuxUnhookDev(pThis, NULL);
2256#endif
2257
2258 /** @todo This code may race vboxNetFltLinuxUnregisterDevice (very very
2259 * unlikely, but none the less). Since it doesn't actually update the
2260 * state (just reads it), it is likely to panic in some interesting
2261 * ways. */
2262
2263 RTSpinlockAcquire(pThis->hSpinlock);
2264 pDev = ASMAtomicUoReadPtrT(&pThis->u.s.pDev, struct net_device *);
2265 fRegistered = ASMAtomicXchgBool(&pThis->u.s.fRegistered, false);
2266 RTSpinlockRelease(pThis->hSpinlock);
2267
2268 if (fRegistered)
2269 {
2270 vboxNetFltSetLinkState(pThis, pDev, false);
2271
2272#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2273 skb_queue_purge(&pThis->u.s.XmitQueue);
2274#endif
2275 Log(("vboxNetFltOsDeleteInstance: this=%p: xmit queue purged.\n", pThis));
2276 Log(("vboxNetFltOsDeleteInstance: Device %p(%s) released. ref=%d\n",
2277 pDev, pDev->name,
2278#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
2279 netdev_refcnt_read(pDev)
2280#else
2281 atomic_read(&pDev->refcnt)
2282#endif
2283 ));
2284 dev_put(pDev);
2285 }
2286
2287 unregister_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2288 unregister_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2289
2290 Log(("vboxNetFltOsDeleteInstance: this=%p: Notifier removed.\n", pThis));
2291 unregister_netdevice_notifier(&pThis->u.s.Notifier);
2292 module_put(THIS_MODULE);
2293
2294 IPRT_LINUX_RESTORE_EFL_AC();
2295}
2296
2297
2298int vboxNetFltOsInitInstance(PVBOXNETFLTINS pThis, void *pvContext)
2299{
2300 int err;
2301 IPRT_LINUX_SAVE_EFL_AC();
2302 NOREF(pvContext);
2303
2304 pThis->u.s.Notifier.notifier_call = vboxNetFltLinuxNotifierCallback;
2305 err = register_netdevice_notifier(&pThis->u.s.Notifier);
2306 if (err)
2307 {
2308 IPRT_LINUX_RESTORE_EFL_AC();
2309 return VERR_INTNET_FLT_IF_FAILED;
2310 }
2311 if (!pThis->u.s.fRegistered)
2312 {
2313 unregister_netdevice_notifier(&pThis->u.s.Notifier);
2314 LogRel(("VBoxNetFlt: failed to find %s.\n", pThis->szName));
2315 IPRT_LINUX_RESTORE_EFL_AC();
2316 return VERR_INTNET_FLT_IF_NOT_FOUND;
2317 }
2318
2319 Log(("vboxNetFltOsInitInstance: this=%p: Notifier installed.\n", pThis));
2320 if ( pThis->fDisconnectedFromHost
2321 || !try_module_get(THIS_MODULE))
2322 {
2323 IPRT_LINUX_RESTORE_EFL_AC();
2324 return VERR_INTNET_FLT_IF_FAILED;
2325 }
2326
2327 if (pThis->pSwitchPort->pfnNotifyHostAddress)
2328 {
2329 VBOXNETFLTNOTIFIER Enumerator;
2330
2331 /*
2332 * register_inetaddr_notifier() and register_inet6addr_notifier()
2333 * do not call the callback for existing devices. Enumerating
2334 * all network devices explicitly is a bit of an ifdef mess,
2335 * so co-opt register_netdevice_notifier() to do that for us.
2336 */
2337 RT_ZERO(Enumerator);
2338 Enumerator.Notifier.notifier_call = vboxNetFltLinuxEnumeratorCallback;
2339 Enumerator.pThis = pThis;
2340
2341 err = register_netdevice_notifier(&Enumerator.Notifier);
2342 if (err)
2343 {
2344 LogRel(("%s: failed to enumerate network devices: error %d\n", __FUNCTION__, err));
2345 IPRT_LINUX_RESTORE_EFL_AC();
2346 return VINF_SUCCESS;
2347 }
2348
2349 unregister_netdevice_notifier(&Enumerator.Notifier);
2350
2351 pThis->u.s.NotifierIPv4.notifier_call = vboxNetFltLinuxNotifierIPv4Callback;
2352 err = register_inetaddr_notifier(&pThis->u.s.NotifierIPv4);
2353 if (err)
2354 LogRel(("%s: failed to register IPv4 notifier: error %d\n", __FUNCTION__, err));
2355
2356 pThis->u.s.NotifierIPv6.notifier_call = vboxNetFltLinuxNotifierIPv6Callback;
2357 err = register_inet6addr_notifier(&pThis->u.s.NotifierIPv6);
2358 if (err)
2359 LogRel(("%s: failed to register IPv6 notifier: error %d\n", __FUNCTION__, err));
2360 }
2361
2362 IPRT_LINUX_RESTORE_EFL_AC();
2363 return VINF_SUCCESS;
2364}
2365
2366int vboxNetFltOsPreInitInstance(PVBOXNETFLTINS pThis)
2367{
2368 IPRT_LINUX_SAVE_EFL_AC();
2369
2370 /*
2371 * Init the linux specific members.
2372 */
2373 ASMAtomicUoWriteNullPtr(&pThis->u.s.pDev);
2374 pThis->u.s.fRegistered = false;
2375 pThis->u.s.fPromiscuousSet = false;
2376 pThis->u.s.fPacketHandler = false;
2377 memset(&pThis->u.s.PacketType, 0, sizeof(pThis->u.s.PacketType));
2378#ifndef VBOXNETFLT_LINUX_NO_XMIT_QUEUE
2379 skb_queue_head_init(&pThis->u.s.XmitQueue);
2380# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
2381 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask);
2382# else
2383 INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);
2384# endif
2385#endif
2386
2387 IPRT_LINUX_RESTORE_EFL_AC();
2388 return VINF_SUCCESS;
2389}
2390
2391
2392void vboxNetFltPortOsNotifyMacAddress(PVBOXNETFLTINS pThis, void *pvIfData, PCRTMAC pMac)
2393{
2394 NOREF(pThis); NOREF(pvIfData); NOREF(pMac);
2395}
2396
2397
2398int vboxNetFltPortOsConnectInterface(PVBOXNETFLTINS pThis, void *pvIf, void **pvIfData)
2399{
2400 /* Nothing to do */
2401 NOREF(pThis); NOREF(pvIf); NOREF(pvIfData);
2402 return VINF_SUCCESS;
2403}
2404
2405
2406int vboxNetFltPortOsDisconnectInterface(PVBOXNETFLTINS pThis, void *pvIfData)
2407{
2408 /* Nothing to do */
2409 NOREF(pThis); NOREF(pvIfData);
2410 return VINF_SUCCESS;
2411}
2412
Note: See TracBrowser for help on using the repository browser.

© 2025 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette