SUPDrv-linux.c@ 82957

Last change on this file since 82957 was 82590, checked in by vboxsync, 5 years ago
SUPDrv,IPRT: Adding SUPR0HCPhysToVirt to linux & solaris; introducing IPRT_WITHOUT_EFLAGS_AC_PRESERVING and VBOX_WITHOUT_EFLAGS_AC_SET_IN_VBOXDRV config macros for disabling EFLAGS.AC hacks (not active). bugref:9627
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 45.8 KB

Line
1	/* $Id: SUPDrv-linux.c 82590 2019-12-16 17:48:40Z vboxsync $ */
2	/** @file
3	* VBoxDrv - The VirtualBox Support Driver - Linux specifics.
4	*/
5
6	/*
7	* Copyright (C) 2006-2019 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	* 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_SUP_DRV
32	#include "../SUPDrvInternal.h"
33	#include "the-linux-kernel.h"
34	#include "version-generated.h"
35	#include "product-generated.h"
36	#include "revision-generated.h"
37
38	#include <iprt/assert.h>
39	#include <iprt/spinlock.h>
40	#include <iprt/semaphore.h>
41	#include <iprt/initterm.h>
42	#include <iprt/process.h>
43	#include <VBox/err.h>
44	#include <iprt/mem.h>
45	#include <VBox/log.h>
46	#include <iprt/mp.h>
47
48	/** @todo figure out the exact version number */
49	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 16)
50	# include <iprt/power.h>
51	# define VBOX_WITH_SUSPEND_NOTIFICATION
52	#endif
53
54	#include <linux/sched.h>
55	#include <linux/miscdevice.h>
56	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
57	# include <linux/platform_device.h>
58	#endif
59	#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 28)) && defined(SUPDRV_WITH_MSR_PROBER)
60	# define SUPDRV_LINUX_HAS_SAFE_MSR_API
61	# include <asm/msr.h>
62	#endif
63
64	#include <asm/desc.h>
65
66	#include <iprt/asm-amd64-x86.h>
67
68
69	/*********************************************************************************************************************************
70	* Defined Constants And Macros *
71	*********************************************************************************************************************************/
72	/* check kernel version */
73	# ifndef SUPDRV_AGNOSTIC
74	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
75	# error Unsupported kernel version!
76	# endif
77	# endif
78
79	#ifdef CONFIG_X86_HIGH_ENTRY
80	# error "CONFIG_X86_HIGH_ENTRY is not supported by VBoxDrv at this time."
81	#endif
82
83	/* We cannot include x86.h, so we copy the defines we need here: */
84	#define X86_EFL_IF RT_BIT(9)
85	#define X86_EFL_AC RT_BIT(18)
86	#define X86_EFL_DF RT_BIT(10)
87	#define X86_EFL_IOPL (RT_BIT(12) \| RT_BIT(13))
88
89	/* To include the version number of VirtualBox into kernel backtraces: */
90	#define VBoxDrvLinuxVersion RT_CONCAT3(RT_CONCAT(VBOX_VERSION_MAJOR, _), \
91	RT_CONCAT(VBOX_VERSION_MINOR, _), \
92	VBOX_VERSION_BUILD)
93	#define VBoxDrvLinuxIOCtl RT_CONCAT(VBoxDrvLinuxIOCtl_,VBoxDrvLinuxVersion)
94
95
96
97	/*********************************************************************************************************************************
98	* Internal Functions *
99	*********************************************************************************************************************************/
100	static int __init VBoxDrvLinuxInit(void);
101	static void __exit VBoxDrvLinuxUnload(void);
102	static int VBoxDrvLinuxCreateSys(struct inode pInode, struct file pFilp);
103	static int VBoxDrvLinuxCreateUsr(struct inode pInode, struct file pFilp);
104	static int VBoxDrvLinuxClose(struct inode pInode, struct file pFilp);
105	#ifdef HAVE_UNLOCKED_IOCTL
106	static long VBoxDrvLinuxIOCtl(struct file *pFilp, unsigned int uCmd, unsigned long ulArg);
107	#else
108	static int VBoxDrvLinuxIOCtl(struct inode pInode, struct file pFilp, unsigned int uCmd, unsigned long ulArg);
109	#endif
110	static int VBoxDrvLinuxIOCtlSlow(struct file *pFilp, unsigned int uCmd, unsigned long ulArg, PSUPDRVSESSION pSession);
111	static int VBoxDrvLinuxErr2LinuxErr(int);
112	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
113	static int VBoxDrvProbe(struct platform_device *pDev);
114	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 30)
115	static int VBoxDrvSuspend(struct device *pDev);
116	static int VBoxDrvResume(struct device *pDev);
117	# else
118	static int VBoxDrvSuspend(struct platform_device *pDev, pm_message_t State);
119	static int VBoxDrvResume(struct platform_device *pDev);
120	# endif
121	static void VBoxDevRelease(struct device *pDev);
122	#endif
123
124
125	/*********************************************************************************************************************************
126	* Global Variables *
127	*********************************************************************************************************************************/
128	/**
129	* Device extention & session data association structure.
130	*/
131	static SUPDRVDEVEXT g_DevExt;
132
133	/** Module parameter.
134	* Not prefixed because the name is used by macros and the end of this file. */
135	static int force_async_tsc = 0;
136
137	/** The system device name. */
138	#define DEVICE_NAME_SYS "vboxdrv"
139	/** The user device name. */
140	#define DEVICE_NAME_USR "vboxdrvu"
141
142	#if (defined(RT_ARCH_AMD64) && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 23)) \|\| defined(VBOX_WITH_TEXT_MODMEM_HACK)
143	/**
144	* Memory for the executable memory heap (in IPRT).
145	*/
146	# ifdef DEBUG
147	# define EXEC_MEMORY_SIZE 8388608 /* 8 MB */
148	# else
149	# define EXEC_MEMORY_SIZE 2097152 /* 2 MB */
150	# endif
151	extern uint8_t g_abExecMemory[EXEC_MEMORY_SIZE];
152	# ifndef VBOX_WITH_TEXT_MODMEM_HACK
153	__asm__(".section execmemory, \"awx\", @progbits\n\t"
154	".align 32\n\t"
155	".globl g_abExecMemory\n"
156	"g_abExecMemory:\n\t"
157	".zero " RT_XSTR(EXEC_MEMORY_SIZE) "\n\t"
158	".type g_abExecMemory, @object\n\t"
159	".size g_abExecMemory, " RT_XSTR(EXEC_MEMORY_SIZE) "\n\t"
160	".text\n\t");
161	# else
162	__asm__(".text\n\t"
163	".align 4096\n\t"
164	".globl g_abExecMemory\n"
165	"g_abExecMemory:\n\t"
166	".zero " RT_XSTR(EXEC_MEMORY_SIZE) "\n\t"
167	".type g_abExecMemory, @object\n\t"
168	".size g_abExecMemory, " RT_XSTR(EXEC_MEMORY_SIZE) "\n\t"
169	".text\n\t");
170	# endif
171	#endif
172
173	/** The file_operations structure. */
174	static struct file_operations gFileOpsVBoxDrvSys =
175	{
176	owner: THIS_MODULE,
177	open: VBoxDrvLinuxCreateSys,
178	release: VBoxDrvLinuxClose,
179	#ifdef HAVE_UNLOCKED_IOCTL
180	unlocked_ioctl: VBoxDrvLinuxIOCtl,
181	#else
182	ioctl: VBoxDrvLinuxIOCtl,
183	#endif
184	};
185
186	/** The file_operations structure. */
187	static struct file_operations gFileOpsVBoxDrvUsr =
188	{
189	owner: THIS_MODULE,
190	open: VBoxDrvLinuxCreateUsr,
191	release: VBoxDrvLinuxClose,
192	#ifdef HAVE_UNLOCKED_IOCTL
193	unlocked_ioctl: VBoxDrvLinuxIOCtl,
194	#else
195	ioctl: VBoxDrvLinuxIOCtl,
196	#endif
197	};
198
199	/** The miscdevice structure for vboxdrv. */
200	static struct miscdevice gMiscDeviceSys =
201	{
202	minor: MISC_DYNAMIC_MINOR,
203	name: DEVICE_NAME_SYS,
204	fops: &gFileOpsVBoxDrvSys,
205	# if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 17)
206	devfs_name: DEVICE_NAME_SYS,
207	# endif
208	};
209	/** The miscdevice structure for vboxdrvu. */
210	static struct miscdevice gMiscDeviceUsr =
211	{
212	minor: MISC_DYNAMIC_MINOR,
213	name: DEVICE_NAME_USR,
214	fops: &gFileOpsVBoxDrvUsr,
215	# if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 17)
216	devfs_name: DEVICE_NAME_USR,
217	# endif
218	};
219
220
221	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
222	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 30)
223	static struct dev_pm_ops gPlatformPMOps =
224	{
225	.suspend = VBoxDrvSuspend, /* before entering deep sleep */
226	.resume = VBoxDrvResume, /* after wakeup from deep sleep */
227	.freeze = VBoxDrvSuspend, /* before creating hibernation image */
228	.restore = VBoxDrvResume, /* after waking up from hibernation */
229	};
230	# endif
231
232	static struct platform_driver gPlatformDriver =
233	{
234	.probe = VBoxDrvProbe,
235	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 30)
236	.suspend = VBoxDrvSuspend,
237	.resume = VBoxDrvResume,
238	# endif
239	/** @todo .shutdown? */
240	.driver =
241	{
242	.name = "vboxdrv",
243	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 30)
244	.pm = &gPlatformPMOps,
245	# endif
246	}
247	};
248
249	static struct platform_device gPlatformDevice =
250	{
251	.name = "vboxdrv",
252	.dev =
253	{
254	.release = VBoxDevRelease
255	}
256	};
257	#endif /* VBOX_WITH_SUSPEND_NOTIFICATION */
258
259
260	DECLINLINE(RTUID) vboxdrvLinuxUid(void)
261	{
262	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
263	# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
264	return from_kuid(current_user_ns(), current->cred->uid);
265	# else
266	return current->cred->uid;
267	# endif
268	#else
269	return current->uid;
270	#endif
271	}
272
273	DECLINLINE(RTGID) vboxdrvLinuxGid(void)
274	{
275	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
276	# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
277	return from_kgid(current_user_ns(), current->cred->gid);
278	# else
279	return current->cred->gid;
280	# endif
281	#else
282	return current->gid;
283	#endif
284	}
285
286	DECLINLINE(RTUID) vboxdrvLinuxEuid(void)
287	{
288	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
289	# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
290	return from_kuid(current_user_ns(), current->cred->euid);
291	# else
292	return current->cred->euid;
293	# endif
294	#else
295	return current->euid;
296	#endif
297	}
298
299	/**
300	* Initialize module.
301	*
302	* @returns appropriate status code.
303	*/
304	static int __init VBoxDrvLinuxInit(void)
305	{
306	int rc;
307
308	/*
309	* Check for synchronous/asynchronous TSC mode.
310	*/
311	printk(KERN_DEBUG "vboxdrv: Found %u processor cores\n", (unsigned)RTMpGetOnlineCount());
312	rc = misc_register(&gMiscDeviceSys);
313	if (rc)
314	{
315	printk(KERN_ERR "vboxdrv: Can't register system misc device! rc=%d\n", rc);
316	return rc;
317	}
318	rc = misc_register(&gMiscDeviceUsr);
319	if (rc)
320	{
321	printk(KERN_ERR "vboxdrv: Can't register user misc device! rc=%d\n", rc);
322	misc_deregister(&gMiscDeviceSys);
323	return rc;
324	}
325	if (!rc)
326	{
327	/*
328	* Initialize the runtime.
329	* On AMD64 we'll have to donate the high rwx memory block to the exec allocator.
330	*/
331	rc = RTR0Init(0);
332	if (RT_SUCCESS(rc))
333	{
334	#if (defined(RT_ARCH_AMD64) && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 23)) \|\| defined(VBOX_WITH_TEXT_MODMEM_HACK)
335	# ifdef VBOX_WITH_TEXT_MODMEM_HACK
336	set_memory_x(&g_abExecMemory[0], sizeof(g_abExecMemory) / PAGE_SIZE);
337	set_memory_rw(&g_abExecMemory[0], sizeof(g_abExecMemory) / PAGE_SIZE);
338	# endif
339	rc = RTR0MemExecDonate(&g_abExecMemory[0], sizeof(g_abExecMemory));
340	printk(KERN_DEBUG "VBoxDrv: dbg - g_abExecMemory=%p\n", (void *)&g_abExecMemory[0]);
341	#endif
342	Log(("VBoxDrv::ModuleInit\n"));
343
344	/*
345	* Initialize the device extension.
346	*/
347	if (RT_SUCCESS(rc))
348	rc = supdrvInitDevExt(&g_DevExt, sizeof(SUPDRVSESSION));
349	if (RT_SUCCESS(rc))
350	{
351	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
352	rc = platform_driver_register(&gPlatformDriver);
353	if (rc == 0)
354	{
355	rc = platform_device_register(&gPlatformDevice);
356	if (rc == 0)
357	#endif
358	{
359	printk(KERN_INFO "vboxdrv: TSC mode is %s, tentative frequency %llu Hz\n",
360	SUPGetGIPModeName(g_DevExt.pGip), g_DevExt.pGip->u64CpuHz);
361	LogFlow(("VBoxDrv::ModuleInit returning %#x\n", rc));
362	printk(KERN_DEBUG "vboxdrv: Successfully loaded version "
363	VBOX_VERSION_STRING " (interface " RT_XSTR(SUPDRV_IOC_VERSION) ")\n");
364	return rc;
365	}
366	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
367	else
368	platform_driver_unregister(&gPlatformDriver);
369	}
370	#endif
371	}
372
373	rc = -EINVAL;
374	RTR0TermForced();
375	}
376	else
377	rc = -EINVAL;
378
379	/*
380	* Failed, cleanup and return the error code.
381	*/
382	}
383	misc_deregister(&gMiscDeviceSys);
384	misc_deregister(&gMiscDeviceUsr);
385	Log(("VBoxDrv::ModuleInit returning %#x (minor:%d & %d)\n", rc, gMiscDeviceSys.minor, gMiscDeviceUsr.minor));
386	return rc;
387	}
388
389
390	/**
391	* Unload the module.
392	*/
393	static void __exit VBoxDrvLinuxUnload(void)
394	{
395	Log(("VBoxDrvLinuxUnload\n"));
396
397	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
398	platform_device_unregister(&gPlatformDevice);
399	platform_driver_unregister(&gPlatformDriver);
400	#endif
401
402	/*
403	* I Don't think it's possible to unload a driver which processes have
404	* opened, at least we'll blindly assume that here.
405	*/
406	misc_deregister(&gMiscDeviceUsr);
407	misc_deregister(&gMiscDeviceSys);
408
409	/*
410	* Destroy GIP, delete the device extension and terminate IPRT.
411	*/
412	supdrvDeleteDevExt(&g_DevExt);
413	RTR0TermForced();
414	}
415
416
417	/**
418	* Common open code.
419	*
420	* @param pInode Pointer to inode info structure.
421	* @param pFilp Associated file pointer.
422	* @param fUnrestricted Indicates which device node which was opened.
423	*/
424	static int vboxdrvLinuxCreateCommon(struct inode pInode, struct file pFilp, bool fUnrestricted)
425	{
426	int rc;
427	PSUPDRVSESSION pSession;
428	Log(("VBoxDrvLinuxCreate: pFilp=%p pid=%d/%d %s\n", pFilp, RTProcSelf(), current->pid, current->comm));
429
430	#ifdef VBOX_WITH_HARDENING
431	/*
432	* Only root is allowed to access the unrestricted device, enforce it!
433	*/
434	if ( fUnrestricted
435	&& vboxdrvLinuxEuid() != 0 /* root */ )
436	{
437	Log(("VBoxDrvLinuxCreate: euid=%d, expected 0 (root)\n", vboxdrvLinuxEuid()));
438	return -EPERM;
439	}
440	#endif /* VBOX_WITH_HARDENING */
441
442	/*
443	* Call common code for the rest.
444	*/
445	rc = supdrvCreateSession(&g_DevExt, true /* fUser */, fUnrestricted, &pSession);
446	if (!rc)
447	{
448	pSession->Uid = vboxdrvLinuxUid();
449	pSession->Gid = vboxdrvLinuxGid();
450	}
451
452	pFilp->private_data = pSession;
453
454	Log(("VBoxDrvLinuxCreate: g_DevExt=%p pSession=%p rc=%d/%d (pid=%d/%d %s)\n",
455	&g_DevExt, pSession, rc, VBoxDrvLinuxErr2LinuxErr(rc),
456	RTProcSelf(), current->pid, current->comm));
457	return VBoxDrvLinuxErr2LinuxErr(rc);
458	}
459
460
461	/** /dev/vboxdrv. */
462	static int VBoxDrvLinuxCreateSys(struct inode pInode, struct file pFilp)
463	{
464	return vboxdrvLinuxCreateCommon(pInode, pFilp, true);
465	}
466
467
468	/** /dev/vboxdrvu. */
469	static int VBoxDrvLinuxCreateUsr(struct inode pInode, struct file pFilp)
470	{
471	return vboxdrvLinuxCreateCommon(pInode, pFilp, false);
472	}
473
474
475	/**
476	* Close device.
477	*
478	* @param pInode Pointer to inode info structure.
479	* @param pFilp Associated file pointer.
480	*/
481	static int VBoxDrvLinuxClose(struct inode pInode, struct file pFilp)
482	{
483	Log(("VBoxDrvLinuxClose: pFilp=%p pSession=%p pid=%d/%d %s\n",
484	pFilp, pFilp->private_data, RTProcSelf(), current->pid, current->comm));
485	supdrvSessionRelease((PSUPDRVSESSION)pFilp->private_data);
486	pFilp->private_data = NULL;
487	return 0;
488	}
489
490
491	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
492	/**
493	* Dummy device release function. We have to provide this function,
494	* otherwise the kernel will complain.
495	*
496	* @param pDev Pointer to the platform device.
497	*/
498	static void VBoxDevRelease(struct device *pDev)
499	{
500	}
501
502	/**
503	* Dummy probe function.
504	*
505	* @param pDev Pointer to the platform device.
506	*/
507	static int VBoxDrvProbe(struct platform_device *pDev)
508	{
509	return 0;
510	}
511
512	/**
513	* Suspend callback.
514	* @param pDev Pointer to the platform device.
515	* @param State Message type, see Documentation/power/devices.txt.
516	* Ignored.
517	*/
518	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 30) && !defined(DOXYGEN_RUNNING)
519	static int VBoxDrvSuspend(struct device *pDev)
520	# else
521	static int VBoxDrvSuspend(struct platform_device *pDev, pm_message_t State)
522	# endif
523	{
524	RTPowerSignalEvent(RTPOWEREVENT_SUSPEND);
525	return 0;
526	}
527
528	/**
529	* Resume callback.
530	*
531	* @param pDev Pointer to the platform device.
532	*/
533	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 30)
534	static int VBoxDrvResume(struct device *pDev)
535	# else
536	static int VBoxDrvResume(struct platform_device *pDev)
537	# endif
538	{
539	RTPowerSignalEvent(RTPOWEREVENT_RESUME);
540	return 0;
541	}
542	#endif /* VBOX_WITH_SUSPEND_NOTIFICATION */
543
544
545	/**
546	* Device I/O Control entry point.
547	*
548	* @param pFilp Associated file pointer.
549	* @param uCmd The function specified to ioctl().
550	* @param ulArg The argument specified to ioctl().
551	*/
552	#if defined(HAVE_UNLOCKED_IOCTL) \|\| defined(DOXYGEN_RUNNING)
553	static long VBoxDrvLinuxIOCtl(struct file *pFilp, unsigned int uCmd, unsigned long ulArg)
554	#else
555	static int VBoxDrvLinuxIOCtl(struct inode pInode, struct file pFilp, unsigned int uCmd, unsigned long ulArg)
556	#endif
557	{
558	PSUPDRVSESSION pSession = (PSUPDRVSESSION)pFilp->private_data;
559	int rc;
560	#ifndef VBOX_WITHOUT_EFLAGS_AC_SET_IN_VBOXDRV
561	# if defined(VBOX_STRICT) \|\| defined(VBOX_WITH_EFLAGS_AC_SET_IN_VBOXDRV)
562	RTCCUINTREG fSavedEfl;
563
564	/*
565	* Refuse all I/O control calls if we've ever detected EFLAGS.AC being cleared.
566	*
567	* This isn't a problem, as there is absolutely nothing in the kernel context that
568	* depend on user context triggering cleanups. That would be pretty wild, right?
569	*/
570	if (RT_UNLIKELY(g_DevExt.cBadContextCalls > 0))
571	{
572	SUPR0Printf("VBoxDrvLinuxIOCtl: EFLAGS.AC=0 detected %u times, refusing all I/O controls!\n", g_DevExt.cBadContextCalls);
573	return ESPIPE;
574	}
575
576	fSavedEfl = ASMAddFlags(X86_EFL_AC);
577	# else
578	stac();
579	# endif
580	#endif
581
582	/*
583	* Deal with the two high-speed IOCtl that takes it's arguments from
584	* the session and iCmd, and only returns a VBox status code.
585	*/
586	AssertCompile(_IOC_NRSHIFT == 0 && _IOC_NRBITS == 8);
587	#ifdef HAVE_UNLOCKED_IOCTL
588	if (RT_LIKELY( (unsigned int)(uCmd - SUP_IOCTL_FAST_DO_FIRST) < (unsigned int)32
589	&& pSession->fUnrestricted))
590	rc = supdrvIOCtlFast(uCmd - SUP_IOCTL_FAST_DO_FIRST, ulArg, &g_DevExt, pSession);
591	else
592	rc = VBoxDrvLinuxIOCtlSlow(pFilp, uCmd, ulArg, pSession);
593	#else /* !HAVE_UNLOCKED_IOCTL */
594	unlock_kernel();
595	if (RT_LIKELY( (unsigned int)(uCmd - SUP_IOCTL_FAST_DO_FIRST) < (unsigned int)32
596	&& pSession->fUnrestricted))
597	rc = supdrvIOCtlFast(uCmd - SUP_IOCTL_FAST_DO_FIRST, ulArg, &g_DevExt, pSession);
598	else
599	rc = VBoxDrvLinuxIOCtlSlow(pFilp, uCmd, ulArg, pSession);
600	lock_kernel();
601	#endif /* !HAVE_UNLOCKED_IOCTL */
602
603	#ifndef VBOX_WITHOUT_EFLAGS_AC_SET_IN_VBOXDRV
604	# if defined(VBOX_STRICT) \|\| defined(VBOX_WITH_EFLAGS_AC_SET_IN_VBOXDRV)
605	/*
606	* Before we restore AC and the rest of EFLAGS, check if the IOCtl handler code
607	* accidentially modified it or some other important flag.
608	*/
609	if (RT_UNLIKELY( (ASMGetFlags() & (X86_EFL_AC \| X86_EFL_IF \| X86_EFL_DF))
610	!= ((fSavedEfl & (X86_EFL_AC \| X86_EFL_IF \| X86_EFL_DF)) \| X86_EFL_AC) ))
611	{
612	char szTmp[48];
613	RTStrPrintf(szTmp, sizeof(szTmp), "uCmd=%#x: %#x->%#x!", _IOC_NR(uCmd), (uint32_t)fSavedEfl, (uint32_t)ASMGetFlags());
614	supdrvBadContext(&g_DevExt, "SUPDrv-linux.c", __LINE__, szTmp);
615	}
616	ASMSetFlags(fSavedEfl);
617	# else
618	clac();
619	# endif
620	#endif
621	return rc;
622	}
623
624
625	/**
626	* Device I/O Control entry point.
627	*
628	* @param pFilp Associated file pointer.
629	* @param uCmd The function specified to ioctl().
630	* @param ulArg The argument specified to ioctl().
631	* @param pSession The session instance.
632	*/
633	static int VBoxDrvLinuxIOCtlSlow(struct file *pFilp, unsigned int uCmd, unsigned long ulArg, PSUPDRVSESSION pSession)
634	{
635	int rc;
636	SUPREQHDR Hdr;
637	PSUPREQHDR pHdr;
638	uint32_t cbBuf;
639
640	Log6(("VBoxDrvLinuxIOCtl: pFilp=%p uCmd=%#x ulArg=%p pid=%d/%d\n", pFilp, uCmd, (void *)ulArg, RTProcSelf(), current->pid));
641
642	/*
643	* Read the header.
644	*/
645	if (RT_FAILURE(RTR0MemUserCopyFrom(&Hdr, ulArg, sizeof(Hdr))))
646	{
647	Log(("VBoxDrvLinuxIOCtl: copy_from_user(,%#lx,) failed; uCmd=%#x\n", ulArg, uCmd));
648	return -EFAULT;
649	}
650	if (RT_UNLIKELY((Hdr.fFlags & SUPREQHDR_FLAGS_MAGIC_MASK) != SUPREQHDR_FLAGS_MAGIC))
651	{
652	Log(("VBoxDrvLinuxIOCtl: bad header magic %#x; uCmd=%#x\n", Hdr.fFlags & SUPREQHDR_FLAGS_MAGIC_MASK, uCmd));
653	return -EINVAL;
654	}
655
656	/*
657	* Buffer the request.
658	*/
659	cbBuf = RT_MAX(Hdr.cbIn, Hdr.cbOut);
660	if (RT_UNLIKELY(cbBuf > _1M*16))
661	{
662	Log(("VBoxDrvLinuxIOCtl: too big cbBuf=%#x; uCmd=%#x\n", cbBuf, uCmd));
663	return -E2BIG;
664	}
665	if (RT_UNLIKELY(_IOC_SIZE(uCmd) ? cbBuf != _IOC_SIZE(uCmd) : Hdr.cbIn < sizeof(Hdr)))
666	{
667	Log(("VBoxDrvLinuxIOCtl: bad ioctl cbBuf=%#x _IOC_SIZE=%#x; uCmd=%#x\n", cbBuf, _IOC_SIZE(uCmd), uCmd));
668	return -EINVAL;
669	}
670	pHdr = RTMemAlloc(cbBuf);
671	if (RT_UNLIKELY(!pHdr))
672	{
673	OSDBGPRINT(("VBoxDrvLinuxIOCtl: failed to allocate buffer of %d bytes for uCmd=%#x\n", cbBuf, uCmd));
674	return -ENOMEM;
675	}
676	if (RT_FAILURE(RTR0MemUserCopyFrom(pHdr, ulArg, Hdr.cbIn)))
677	{
678	Log(("VBoxDrvLinuxIOCtl: copy_from_user(,%#lx, %#x) failed; uCmd=%#x\n", ulArg, Hdr.cbIn, uCmd));
679	RTMemFree(pHdr);
680	return -EFAULT;
681	}
682	if (Hdr.cbIn < cbBuf)
683	RT_BZERO((uint8_t *)pHdr + Hdr.cbIn, cbBuf - Hdr.cbIn);
684
685	/*
686	* Process the IOCtl.
687	*/
688	rc = supdrvIOCtl(uCmd, &g_DevExt, pSession, pHdr, cbBuf);
689
690	/*
691	* Copy ioctl data and output buffer back to user space.
692	*/
693	if (RT_LIKELY(!rc))
694	{
695	uint32_t cbOut = pHdr->cbOut;
696	if (RT_UNLIKELY(cbOut > cbBuf))
697	{
698	OSDBGPRINT(("VBoxDrvLinuxIOCtl: too much output! %#x > %#x; uCmd=%#x!\n", cbOut, cbBuf, uCmd));
699	cbOut = cbBuf;
700	}
701	if (RT_FAILURE(RTR0MemUserCopyTo(ulArg, pHdr, cbOut)))
702	{
703	/* this is really bad! */
704	OSDBGPRINT(("VBoxDrvLinuxIOCtl: copy_to_user(%#lx,,%#x); uCmd=%#x!\n", ulArg, cbOut, uCmd));
705	rc = -EFAULT;
706	}
707	}
708	else
709	{
710	Log(("VBoxDrvLinuxIOCtl: pFilp=%p uCmd=%#x ulArg=%p failed, rc=%d\n", pFilp, uCmd, (void *)ulArg, rc));
711	rc = -EINVAL;
712	}
713	RTMemFree(pHdr);
714
715	Log6(("VBoxDrvLinuxIOCtl: returns %d (pid=%d/%d)\n", rc, RTProcSelf(), current->pid));
716	return rc;
717	}
718
719
720	/**
721	* The SUPDRV IDC entry point.
722	*
723	* @returns VBox status code, see supdrvIDC.
724	* @param uReq The request code.
725	* @param pReq The request.
726	*/
727	int VBOXCALL SUPDrvLinuxIDC(uint32_t uReq, PSUPDRVIDCREQHDR pReq)
728	{
729	PSUPDRVSESSION pSession;
730
731	/*
732	* Some quick validations.
733	*/
734	if (RT_UNLIKELY(!VALID_PTR(pReq)))
735	return VERR_INVALID_POINTER;
736
737	pSession = pReq->pSession;
738	if (pSession)
739	{
740	if (RT_UNLIKELY(!VALID_PTR(pSession)))
741	return VERR_INVALID_PARAMETER;
742	if (RT_UNLIKELY(pSession->pDevExt != &g_DevExt))
743	return VERR_INVALID_PARAMETER;
744	}
745	else if (RT_UNLIKELY(uReq != SUPDRV_IDC_REQ_CONNECT))
746	return VERR_INVALID_PARAMETER;
747
748	/*
749	* Do the job.
750	*/
751	return supdrvIDC(uReq, &g_DevExt, pSession, pReq);
752	}
753
754	EXPORT_SYMBOL(SUPDrvLinuxIDC);
755
756
757	RTCCUINTREG VBOXCALL supdrvOSChangeCR4(RTCCUINTREG fOrMask, RTCCUINTREG fAndMask)
758	{
759	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 20, 0)
760	RTCCUINTREG uOld = this_cpu_read(cpu_tlbstate.cr4);
761	RTCCUINTREG uNew = (uOld & fAndMask) \| fOrMask;
762	if (uNew != uOld)
763	{
764	this_cpu_write(cpu_tlbstate.cr4, uNew);
765	__write_cr4(uNew);
766	}
767	#else
768	RTCCUINTREG uOld = ASMGetCR4();
769	RTCCUINTREG uNew = (uOld & fAndMask) \| fOrMask;
770	if (uNew != uOld)
771	ASMSetCR4(uNew);
772	#endif
773	return uOld;
774	}
775
776
777	void VBOXCALL supdrvOSCleanupSession(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession)
778	{
779	NOREF(pDevExt);
780	NOREF(pSession);
781	}
782
783
784	void VBOXCALL supdrvOSSessionHashTabInserted(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, void *pvUser)
785	{
786	NOREF(pDevExt); NOREF(pSession); NOREF(pvUser);
787	}
788
789
790	void VBOXCALL supdrvOSSessionHashTabRemoved(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, void *pvUser)
791	{
792	NOREF(pDevExt); NOREF(pSession); NOREF(pvUser);
793	}
794
795
796	/**
797	* Initializes any OS specific object creator fields.
798	*/
799	void VBOXCALL supdrvOSObjInitCreator(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession)
800	{
801	NOREF(pObj);
802	NOREF(pSession);
803	}
804
805
806	/**
807	* Checks if the session can access the object.
808	*
809	* @returns true if a decision has been made.
810	* @returns false if the default access policy should be applied.
811	*
812	* @param pObj The object in question.
813	* @param pSession The session wanting to access the object.
814	* @param pszObjName The object name, can be NULL.
815	* @param prc Where to store the result when returning true.
816	*/
817	bool VBOXCALL supdrvOSObjCanAccess(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession, const char pszObjName, int prc)
818	{
819	NOREF(pObj);
820	NOREF(pSession);
821	NOREF(pszObjName);
822	NOREF(prc);
823	return false;
824	}
825
826
827	bool VBOXCALL supdrvOSGetForcedAsyncTscMode(PSUPDRVDEVEXT pDevExt)
828	{
829	return force_async_tsc != 0;
830	}
831
832
833	bool VBOXCALL supdrvOSAreCpusOfflinedOnSuspend(void)
834	{
835	return true;
836	}
837
838
839	bool VBOXCALL supdrvOSAreTscDeltasInSync(void)
840	{
841	return false;
842	}
843
844
845	int VBOXCALL supdrvOSLdrOpen(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, const char *pszFilename)
846	{
847	NOREF(pDevExt); NOREF(pImage); NOREF(pszFilename);
848	return VERR_NOT_SUPPORTED;
849	}
850
851
852	int VBOXCALL supdrvOSLdrValidatePointer(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, void *pv,
853	const uint8_t pbImageBits, const char pszSymbol)
854	{
855	NOREF(pDevExt); NOREF(pImage); NOREF(pv); NOREF(pbImageBits); NOREF(pszSymbol);
856	return VERR_NOT_SUPPORTED;
857	}
858
859
860	int VBOXCALL supdrvOSLdrLoad(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, const uint8_t *pbImageBits, PSUPLDRLOAD pReq)
861	{
862	NOREF(pDevExt); NOREF(pImage); NOREF(pbImageBits); NOREF(pReq);
863	return VERR_NOT_SUPPORTED;
864	}
865
866
867	void VBOXCALL supdrvOSLdrUnload(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
868	{
869	NOREF(pDevExt); NOREF(pImage);
870	}
871
872
873	/** @def VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS
874	* A very crude hack for debugging using perf and dtrace.
875	*
876	* DO ABSOLUTELY NOT ENABLE IN PRODUCTION BUILDS! DEVELOPMENT ONLY!!
877	* DO ABSOLUTELY NOT ENABLE IN PRODUCTION BUILDS! DEVELOPMENT ONLY!!
878	* DO ABSOLUTELY NOT ENABLE IN PRODUCTION BUILDS! DEVELOPMENT ONLY!!
879	*
880	*/
881	#if 0 \|\| defined(DOXYGEN_RUNNING)
882	# define VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS
883	#endif
884
885	#if defined(VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS) && defined(CONFIG_MODULES_TREE_LOOKUP)
886	/** Whether g_pfnModTreeInsert and g_pfnModTreeRemove have been initialized.
887	* @remarks can still be NULL after init. */
888	static volatile bool g_fLookedForModTreeFunctions = false;
889	static void (g_pfnModTreeInsert)(struct mod_tree_node ) = NULL; /*< __mod_tree_insert /
890	static void (g_pfnModTreeRemove)(struct mod_tree_node ) = NULL; /*< __mod_tree_remove /
891	#endif
892
893
894	void VBOXCALL supdrvOSLdrNotifyOpened(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, const char *pszFilename)
895	{
896	#ifdef VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS /* Not for production use!! Debugging only! */
897	/*
898	* This trick stops working with 4.2 when CONFIG_MODULES_TREE_LOOKUP is
899	* defined. The module lookups are done via a tree structure and we
900	* cannot get at the root of it. :-(
901	*/
902	# ifdef CONFIG_KALLSYMS
903	size_t const cchName = strlen(pImage->szName);
904	# endif
905	struct module pMyMod, pSelfMod, pTestMod, pTestModByName;
906	IPRT_LINUX_SAVE_EFL_AC();
907
908	pImage->pLnxModHack = NULL;
909
910	# ifdef CONFIG_MODULES_TREE_LOOKUP
911	/*
912	* This is pretty naive, but works for 4.2 on arch linux. I don't think we
913	* can count on finding __mod_tree_remove in all kernel builds as it's not
914	* marked noinline like __mod_tree_insert.
915	*/
916	if (!g_fLookedForModTreeFunctions)
917	{
918	unsigned long ulInsert = kallsyms_lookup_name("__mod_tree_insert");
919	unsigned long ulRemove = kallsyms_lookup_name("__mod_tree_remove");
920	if (!ulInsert \|\| !ulRemove)
921	{
922	g_fLookedForModTreeFunctions = true;
923	printk(KERN_ERR "vboxdrv: failed to locate __mod_tree_insert and __mod_tree_remove.\n");
924	IPRT_LINUX_RESTORE_EFL_AC();
925	return;
926	}
927	(unsigned long )&g_pfnModTreeInsert = ulInsert;
928	(unsigned long )&g_pfnModTreeRemove = ulRemove;
929	ASMCompilerBarrier();
930	g_fLookedForModTreeFunctions = true;
931	}
932	else if (!g_pfnModTreeInsert \|\| !g_pfnModTreeRemove)
933	return;
934	#endif
935
936	/*
937	* Make sure we've found our own module, otherwise we cannot access the linked list.
938	*/
939	mutex_lock(&module_mutex);
940	pSelfMod = find_module("vboxdrv");
941	mutex_unlock(&module_mutex);
942	if (!pSelfMod)
943	{
944	IPRT_LINUX_RESTORE_EFL_AC();
945	return;
946	}
947
948	/*
949	* Cook up a module structure for the image.
950	* We allocate symbol and string tables in the allocation and the module to keep things simple.
951	*/
952	# ifdef CONFIG_KALLSYMS
953	pMyMod = (struct module )RTMemAllocZ(sizeof(pMyMod)
954	+ sizeof(Elf_Sym) * 3
955	+ 1 + cchName * 2 + sizeof("_start") + sizeof("_end") + 4 );
956	# else
957	pMyMod = (struct module )RTMemAllocZ(sizeof(pMyMod));
958	# endif
959	if (pMyMod)
960	{
961	int rc = VINF_SUCCESS;
962	# ifdef CONFIG_KALLSYMS
963	Elf_Sym paSymbols = (Elf_Sym )(pMyMod + 1);
964	char pchStrTab = (char )(paSymbols + 3);
965	# endif
966
967	pMyMod->state = MODULE_STATE_LIVE;
968	INIT_LIST_HEAD(&pMyMod->list); /* just in case */
969
970	/* Perf only matches up files with a .ko extension (maybe .ko.gz),
971	so in order for this crap to work smoothly, we append .ko to the
972	module name and require the user to create symbolic links in
973	/lib/modules/`uname -r`:
974	for i in VMMR0.r0 VBoxDDR0.r0 VBoxDD2R0.r0; do
975	sudo ln -s /mnt/scratch/vbox/svn/trunk/out/linux.amd64/debug/bin/$i /lib/modules/`uname -r`/$i.ko;
976	done */
977	RTStrPrintf(pMyMod->name, sizeof(pMyMod->name), "%s", pImage->szName);
978
979	/* sysfs bits. */
980	INIT_LIST_HEAD(&pMyMod->mkobj.kobj.entry); /* rest of kobj is already zeroed, hopefully never accessed... */
981	pMyMod->mkobj.mod = pMyMod;
982	pMyMod->mkobj.drivers_dir = NULL;
983	pMyMod->mkobj.mp = NULL;
984	pMyMod->mkobj.kobj_completion = NULL;
985
986	pMyMod->modinfo_attrs = NULL; /* hopefully not accessed after setup. */
987	pMyMod->holders_dir = NULL; /* hopefully not accessed. */
988	pMyMod->version = "N/A";
989	pMyMod->srcversion = "N/A";
990
991	/* We export no symbols. */
992	pMyMod->num_syms = 0;
993	pMyMod->syms = NULL;
994	pMyMod->crcs = NULL;
995
996	pMyMod->num_gpl_syms = 0;
997	pMyMod->gpl_syms = NULL;
998	pMyMod->gpl_crcs = NULL;
999
1000	pMyMod->num_gpl_future_syms = 0;
1001	pMyMod->gpl_future_syms = NULL;
1002	pMyMod->gpl_future_crcs = NULL;
1003
1004	# if CONFIG_UNUSED_SYMBOLS
1005	pMyMod->num_unused_syms = 0;
1006	pMyMod->unused_syms = NULL;
1007	pMyMod->unused_crcs = NULL;
1008
1009	pMyMod->num_unused_gpl_syms = 0;
1010	pMyMod->unused_gpl_syms = NULL;
1011	pMyMod->unused_gpl_crcs = NULL;
1012	# endif
1013	/* No kernel parameters either. */
1014	pMyMod->kp = NULL;
1015	pMyMod->num_kp = 0;
1016
1017	# ifdef CONFIG_MODULE_SIG
1018	/* Pretend ok signature. */
1019	pMyMod->sig_ok = true;
1020	# endif
1021	/* No exception table. */
1022	pMyMod->num_exentries = 0;
1023	pMyMod->extable = NULL;
1024
1025	/* No init function */
1026	pMyMod->init = NULL;
1027	pMyMod->module_init = NULL;
1028	pMyMod->init_size = 0;
1029	pMyMod->init_ro_size = 0;
1030	pMyMod->init_text_size = 0;
1031
1032	/* The module address and size. It's all text. */
1033	pMyMod->module_core = pImage->pvImage;
1034	pMyMod->core_size = pImage->cbImageBits;
1035	pMyMod->core_text_size = pImage->cbImageBits;
1036	pMyMod->core_ro_size = pImage->cbImageBits;
1037
1038	#ifdef CONFIG_MODULES_TREE_LOOKUP
1039	/* Fill in the self pointers for the tree nodes. */
1040	pMyMod->mtn_core.mod = pMyMod;
1041	pMyMod->mtn_init.mod = pMyMod;
1042	#endif
1043	/* They invented the tained bit for us, didn't they? */
1044	pMyMod->taints = 1;
1045
1046	# ifdef CONFIG_GENERIC_BUGS
1047	/* No BUGs in our modules. */
1048	pMyMod->num_bugs = 0;
1049	INIT_LIST_HEAD(&pMyMod->bug_list);
1050	pMyMod->bug_table = NULL;
1051	# endif
1052
1053	# ifdef CONFIG_KALLSYMS
1054	/* The core stuff is documented as only used when loading. So just zero them. */
1055	pMyMod->core_num_syms = 0;
1056	pMyMod->core_symtab = NULL;
1057	pMyMod->core_strtab = NULL;
1058
1059	/* Construct a symbol table with start and end symbols.
1060	Note! We don't have our own symbol table at this point, image bit
1061	are not uploaded yet! */
1062	pMyMod->num_symtab = 3;
1063	pMyMod->symtab = paSymbols;
1064	pMyMod->strtab = pchStrTab;
1065	RT_ZERO(paSymbols[0]);
1066	pchStrTab[0] = '\0';
1067	paSymbols[1].st_name = 1;
1068	paSymbols[2].st_name = 2 + RTStrPrintf(&pchStrTab[paSymbols[1].st_name], cchName + sizeof("_start"),
1069	"%s_start", pImage->szName);
1070	RTStrPrintf(&pchStrTab[paSymbols[2].st_name], cchName + sizeof("_end"), "%s_end", pImage->szName);
1071	paSymbols[1].st_info = 't';
1072	paSymbols[2].st_info = 'b';
1073	paSymbols[1].st_other = 0;
1074	paSymbols[2].st_other = 0;
1075	paSymbols[1].st_shndx = 0;
1076	paSymbols[2].st_shndx = 0;
1077	paSymbols[1].st_value = (uintptr_t)pImage->pvImage;
1078	paSymbols[2].st_value = (uintptr_t)pImage->pvImage + pImage->cbImageBits - 1;
1079	paSymbols[1].st_size = pImage->cbImageBits - 1;
1080	paSymbols[2].st_size = 1;
1081	# endif
1082	/* No arguments, but seems its always non-NULL so put empty string there. */
1083	pMyMod->args = "";
1084
1085	# ifdef CONFIG_SMP
1086	/* No per CPU data. */
1087	pMyMod->percpu = NULL;
1088	pMyMod->percpu_size = 0;
1089	# endif
1090	# ifdef CONFIG_TRACEPOINTS
1091	/* No tracepoints we like to share. */
1092	pMyMod->num_tracepoints = 0;
1093	pMyMod->tracepoints_ptrs = NULL;
1094	#endif
1095	# ifdef HAVE_JUMP_LABEL
1096	/* No jump lable stuff either. */
1097	pMyMod->jump_entries = NULL;
1098	pMyMod->num_jump_entries = 0;
1099	# endif
1100	# ifdef CONFIG_TRACING
1101	pMyMod->num_trace_bprintk_fmt = 0;
1102	pMyMod->trace_bprintk_fmt_start = NULL;
1103	# endif
1104	# ifdef CONFIG_EVENT_TRACING
1105	pMyMod->trace_events = NULL;
1106	pMyMod->num_trace_events = 0;
1107	# endif
1108	# ifdef CONFIG_FTRACE_MCOUNT_RECORD
1109	pMyMod->num_ftrace_callsites = 0;
1110	pMyMod->ftrace_callsites = NULL;
1111	# endif
1112	# ifdef CONFIG_MODULE_UNLOAD
1113	/* Dependency lists, not worth sharing */
1114	INIT_LIST_HEAD(&pMyMod->source_list);
1115	INIT_LIST_HEAD(&pMyMod->target_list);
1116
1117	/* Nobody waiting and no exit function. */
1118	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 13, 0)
1119	pMyMod->waiter = NULL;
1120	# endif
1121	pMyMod->exit = NULL;
1122
1123	/* References, very important as we must not allow the module
1124	to be unloaded using rmmod. */
1125	# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 19, 0)
1126	atomic_set(&pMyMod->refcnt, 42);
1127	# else
1128	pMyMod->refptr = alloc_percpu(struct module_ref);
1129	if (pMyMod->refptr)
1130	{
1131	int iCpu;
1132	for_each_possible_cpu(iCpu)
1133	{
1134	per_cpu_ptr(pMyMod->refptr, iCpu)->decs = 0;
1135	per_cpu_ptr(pMyMod->refptr, iCpu)->incs = 1;
1136	}
1137	}
1138	else
1139	rc = VERR_NO_MEMORY;
1140	# endif
1141	# endif
1142	# ifdef CONFIG_CONSTRUCTORS
1143	/* No constructors. */
1144	pMyMod->ctors = NULL;
1145	pMyMod->num_ctors = 0;
1146	# endif
1147	if (RT_SUCCESS(rc))
1148	{
1149	bool fIsModText;
1150
1151	/*
1152	* Add the module to the list.
1153	*/
1154	mutex_lock(&module_mutex);
1155	list_add_rcu(&pMyMod->list, &pSelfMod->list);
1156	pImage->pLnxModHack = pMyMod;
1157	# ifdef CONFIG_MODULES_TREE_LOOKUP
1158	g_pfnModTreeInsert(&pMyMod->mtn_core); /* __mod_tree_insert */
1159	# endif
1160	mutex_unlock(&module_mutex);
1161
1162	/*
1163	* Test it.
1164	*/
1165	mutex_lock(&module_mutex);
1166	pTestModByName = find_module(pMyMod->name);
1167	pTestMod = __module_address((uintptr_t)pImage->pvImage + pImage->cbImageBits / 4);
1168	fIsModText = __module_text_address((uintptr_t)pImage->pvImage + pImage->cbImageBits / 2);
1169	mutex_unlock(&module_mutex);
1170	if ( pTestMod == pMyMod
1171	&& pTestModByName == pMyMod
1172	&& fIsModText)
1173	printk(KERN_ERR "vboxdrv: fake module works for '%s' (%#lx to %#lx)\n",
1174	pMyMod->name, (unsigned long)paSymbols[1].st_value, (unsigned long)paSymbols[2].st_value);
1175	else
1176	printk(KERN_ERR "vboxdrv: failed to find fake module (pTestMod=%p, pTestModByName=%p, pMyMod=%p, fIsModText=%d)\n",
1177	pTestMod, pTestModByName, pMyMod, fIsModText);
1178	}
1179	else
1180	RTMemFree(pMyMod);
1181	}
1182
1183	IPRT_LINUX_RESTORE_EFL_AC();
1184	#else
1185	pImage->pLnxModHack = NULL;
1186	#endif
1187	NOREF(pDevExt); NOREF(pImage);
1188	}
1189
1190
1191	void VBOXCALL supdrvOSLdrNotifyUnloaded(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
1192	{
1193	#ifdef VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS /* Not for production use!! Debugging only! */
1194	struct module *pMyMod = pImage->pLnxModHack;
1195	pImage->pLnxModHack = NULL;
1196	if (pMyMod)
1197	{
1198	/*
1199	* Remove the fake module list entry and free it.
1200	*/
1201	IPRT_LINUX_SAVE_EFL_AC();
1202	mutex_lock(&module_mutex);
1203	list_del_rcu(&pMyMod->list);
1204	# ifdef CONFIG_MODULES_TREE_LOOKUP
1205	g_pfnModTreeRemove(&pMyMod->mtn_core);
1206	# endif
1207	synchronize_sched();
1208	mutex_unlock(&module_mutex);
1209
1210	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 19, 0)
1211	free_percpu(pMyMod->refptr);
1212	# endif
1213	RTMemFree(pMyMod);
1214	IPRT_LINUX_RESTORE_EFL_AC();
1215	}
1216
1217	#else
1218	Assert(pImage->pLnxModHack == NULL);
1219	#endif
1220	NOREF(pDevExt); NOREF(pImage);
1221	}
1222
1223
1224	int VBOXCALL supdrvOSLdrQuerySymbol(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage,
1225	const char pszSymbol, size_t cchSymbol, void *ppvSymbol)
1226	{
1227	#ifdef VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS
1228	# error "implement me!"
1229	#endif
1230	RT_NOREF(pDevExt, pImage, pszSymbol, cchSymbol, ppvSymbol);
1231	return VERR_WRONG_ORDER;
1232	}
1233
1234
1235	#ifdef SUPDRV_WITH_MSR_PROBER
1236
1237	int VBOXCALL supdrvOSMsrProberRead(uint32_t uMsr, RTCPUID idCpu, uint64_t *puValue)
1238	{
1239	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1240	uint32_t u32Low, u32High;
1241	int rc;
1242
1243	IPRT_LINUX_SAVE_EFL_AC();
1244	if (idCpu == NIL_RTCPUID)
1245	rc = rdmsr_safe(uMsr, &u32Low, &u32High);
1246	else if (RTMpIsCpuOnline(idCpu))
1247	rc = rdmsr_safe_on_cpu(idCpu, uMsr, &u32Low, &u32High);
1248	else
1249	return VERR_CPU_OFFLINE;
1250	IPRT_LINUX_RESTORE_EFL_AC();
1251	if (rc == 0)
1252	{
1253	*puValue = RT_MAKE_U64(u32Low, u32High);
1254	return VINF_SUCCESS;
1255	}
1256	return VERR_ACCESS_DENIED;
1257	# else
1258	return VERR_NOT_SUPPORTED;
1259	# endif
1260	}
1261
1262
1263	int VBOXCALL supdrvOSMsrProberWrite(uint32_t uMsr, RTCPUID idCpu, uint64_t uValue)
1264	{
1265	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1266	int rc;
1267
1268	IPRT_LINUX_SAVE_EFL_AC();
1269	if (idCpu == NIL_RTCPUID)
1270	rc = wrmsr_safe(uMsr, RT_LODWORD(uValue), RT_HIDWORD(uValue));
1271	else if (RTMpIsCpuOnline(idCpu))
1272	rc = wrmsr_safe_on_cpu(idCpu, uMsr, RT_LODWORD(uValue), RT_HIDWORD(uValue));
1273	else
1274	return VERR_CPU_OFFLINE;
1275	IPRT_LINUX_RESTORE_EFL_AC();
1276
1277	if (rc == 0)
1278	return VINF_SUCCESS;
1279	return VERR_ACCESS_DENIED;
1280	# else
1281	return VERR_NOT_SUPPORTED;
1282	# endif
1283	}
1284
1285	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1286	/**
1287	* Worker for supdrvOSMsrProberModify.
1288	*/
1289	static DECLCALLBACK(void) supdrvLnxMsrProberModifyOnCpu(RTCPUID idCpu, void pvUser1, void pvUser2)
1290	{
1291	PSUPMSRPROBER pReq = (PSUPMSRPROBER)pvUser1;
1292	register uint32_t uMsr = pReq->u.In.uMsr;
1293	bool const fFaster = pReq->u.In.enmOp == SUPMSRPROBEROP_MODIFY_FASTER;
1294	uint64_t uBefore;
1295	uint64_t uWritten;
1296	uint64_t uAfter;
1297	int rcBefore, rcWrite, rcAfter, rcRestore;
1298	RTCCUINTREG fOldFlags;
1299
1300	/* Initialize result variables. */
1301	uBefore = uWritten = uAfter = 0;
1302	rcWrite = rcAfter = rcRestore = -EIO;
1303
1304	/*
1305	* Do the job.
1306	*/
1307	fOldFlags = ASMIntDisableFlags();
1308	ASMCompilerBarrier(); /* paranoia */
1309	if (!fFaster)
1310	ASMWriteBackAndInvalidateCaches();
1311
1312	rcBefore = rdmsrl_safe(uMsr, &uBefore);
1313	if (rcBefore >= 0)
1314	{
1315	register uint64_t uRestore = uBefore;
1316	uWritten = uRestore;
1317	uWritten &= pReq->u.In.uArgs.Modify.fAndMask;
1318	uWritten \|= pReq->u.In.uArgs.Modify.fOrMask;
1319
1320	rcWrite = wrmsr_safe(uMsr, RT_LODWORD(uWritten), RT_HIDWORD(uWritten));
1321	rcAfter = rdmsrl_safe(uMsr, &uAfter);
1322	rcRestore = wrmsr_safe(uMsr, RT_LODWORD(uRestore), RT_HIDWORD(uRestore));
1323
1324	if (!fFaster)
1325	{
1326	ASMWriteBackAndInvalidateCaches();
1327	ASMReloadCR3();
1328	ASMNopPause();
1329	}
1330	}
1331
1332	ASMCompilerBarrier(); /* paranoia */
1333	ASMSetFlags(fOldFlags);
1334
1335	/*
1336	* Write out the results.
1337	*/
1338	pReq->u.Out.uResults.Modify.uBefore = uBefore;
1339	pReq->u.Out.uResults.Modify.uWritten = uWritten;
1340	pReq->u.Out.uResults.Modify.uAfter = uAfter;
1341	pReq->u.Out.uResults.Modify.fBeforeGp = rcBefore != 0;
1342	pReq->u.Out.uResults.Modify.fModifyGp = rcWrite != 0;
1343	pReq->u.Out.uResults.Modify.fAfterGp = rcAfter != 0;
1344	pReq->u.Out.uResults.Modify.fRestoreGp = rcRestore != 0;
1345	RT_ZERO(pReq->u.Out.uResults.Modify.afReserved);
1346	}
1347	# endif
1348
1349
1350	int VBOXCALL supdrvOSMsrProberModify(RTCPUID idCpu, PSUPMSRPROBER pReq)
1351	{
1352	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1353	if (idCpu == NIL_RTCPUID)
1354	{
1355	supdrvLnxMsrProberModifyOnCpu(idCpu, pReq, NULL);
1356	return VINF_SUCCESS;
1357	}
1358	return RTMpOnSpecific(idCpu, supdrvLnxMsrProberModifyOnCpu, pReq, NULL);
1359	# else
1360	return VERR_NOT_SUPPORTED;
1361	# endif
1362	}
1363
1364	#endif /* SUPDRV_WITH_MSR_PROBER */
1365
1366
1367	/**
1368	* Converts a supdrv error code to an linux error code.
1369	*
1370	* @returns corresponding linux error code.
1371	* @param rc IPRT status code.
1372	*/
1373	static int VBoxDrvLinuxErr2LinuxErr(int rc)
1374	{
1375	switch (rc)
1376	{
1377	case VINF_SUCCESS: return 0;
1378	case VERR_GENERAL_FAILURE: return -EACCES;
1379	case VERR_INVALID_PARAMETER: return -EINVAL;
1380	case VERR_INVALID_MAGIC: return -EILSEQ;
1381	case VERR_INVALID_HANDLE: return -ENXIO;
1382	case VERR_INVALID_POINTER: return -EFAULT;
1383	case VERR_LOCK_FAILED: return -ENOLCK;
1384	case VERR_ALREADY_LOADED: return -EEXIST;
1385	case VERR_PERMISSION_DENIED: return -EPERM;
1386	case VERR_VERSION_MISMATCH: return -ENOSYS;
1387	case VERR_IDT_FAILED: return -1000;
1388	}
1389
1390	return -EPERM;
1391	}
1392
1393
1394	SUPR0DECL(int) SUPR0HCPhysToVirt(RTHCPHYS HCPhys, void **ppv)
1395	{
1396	AssertReturn(!(HCPhys & PAGE_OFFSET_MASK), VERR_INVALID_POINTER);
1397	AssertReturn(HCPhys != NIL_RTHCPHYS, VERR_INVALID_POINTER);
1398	/* Would've like to use valid_phys_addr_range for this test, but it isn't exported. */
1399	AssertReturn((HCPhys \| PAGE_OFFSET_MASK) < __pa(high_memory), VERR_INVALID_POINTER);
1400	*ppv = phys_to_virt(HCPhys);
1401	return VINF_SUCCESS;
1402	}
1403
1404
1405	RTDECL(int) SUPR0Printf(const char *pszFormat, ...)
1406	{
1407	va_list va;
1408	char szMsg[512];
1409	IPRT_LINUX_SAVE_EFL_AC();
1410
1411	va_start(va, pszFormat);
1412	RTStrPrintfV(szMsg, sizeof(szMsg) - 1, pszFormat, va);
1413	va_end(va);
1414	szMsg[sizeof(szMsg) - 1] = '\0';
1415
1416	printk("%s", szMsg);
1417
1418	IPRT_LINUX_RESTORE_EFL_AC();
1419	return 0;
1420	}
1421
1422
1423	SUPR0DECL(uint32_t) SUPR0GetKernelFeatures(void)
1424	{
1425	uint32_t fFlags = 0;
1426	#ifdef CONFIG_PAX_KERNEXEC
1427	fFlags \|= SUPKERNELFEATURES_GDT_READ_ONLY;
1428	#endif
1429	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 12, 0)
1430	fFlags \|= SUPKERNELFEATURES_GDT_NEED_WRITABLE;
1431	#endif
1432	#if defined(VBOX_STRICT) \|\| defined(VBOX_WITH_EFLAGS_AC_SET_IN_VBOXDRV)
1433	fFlags \|= SUPKERNELFEATURES_SMAP;
1434	#elif defined(CONFIG_X86_SMAP)
1435	if (ASMGetCR4() & X86_CR4_SMAP)
1436	fFlags \|= SUPKERNELFEATURES_SMAP;
1437	#endif
1438	return fFlags;
1439	}
1440
1441
1442	int VBOXCALL supdrvOSGetCurrentGdtRw(RTHCUINTPTR *pGdtRw)
1443	{
1444	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 12, 0)
1445	*pGdtRw = (RTHCUINTPTR)get_current_gdt_rw();
1446	return VINF_SUCCESS;
1447	#else
1448	return VERR_NOT_IMPLEMENTED;
1449	#endif
1450	}
1451
1452
1453	module_init(VBoxDrvLinuxInit);
1454	module_exit(VBoxDrvLinuxUnload);
1455
1456	MODULE_AUTHOR(VBOX_VENDOR);
1457	MODULE_DESCRIPTION(VBOX_PRODUCT " Support Driver");
1458	MODULE_LICENSE("GPL");
1459	#ifdef MODULE_VERSION
1460	MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV) " (" RT_XSTR(SUPDRV_IOC_VERSION) ")");
1461	#endif
1462
1463	module_param(force_async_tsc, int, 0444);
1464	MODULE_PARM_DESC(force_async_tsc, "force the asynchronous TSC mode");
1465

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/HostDrivers/Support/linux/SUPDrv-linux.c@ 82957

Download in other formats: