SUPDrv-linux.c@ 87702

Last change on this file since 87702 was 87700, checked in by vboxsync, 4 years ago
SUPDrv,++: Experimental support for wrapping .r0 modules in native kernel modules on linux, so that perf and similar tools work better. Minor IOC version increase as SUP_IOCTL_LDR_OPEN now support just opening a module w/o preparing the loading. SUPDrv must export all the symbols in g_aFunctions the linux way now, or linux won't see them, so introduced a SUPR0_EXPORT_SYMBOL macro similar to RT_EXPORT_SYMBOL. bugref:9937
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 47.2 KB

Line
1	/* $Id: SUPDrv-linux.c 87700 2021-02-10 20:21:04Z vboxsync $ */
2	/** @file
3	* VBoxDrv - The VirtualBox Support Driver - Linux specifics.
4	*/
5
6	/*
7	* Copyright (C) 2006-2020 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* The contents of this file may alternatively be used under the terms
18	* of the Common Development and Distribution License Version 1.0
19	* (CDDL) only, as it comes in the "COPYING.CDDL" file of the
20	* VirtualBox OSE distribution, in which case the provisions of the
21	* CDDL are applicable instead of those of the GPL.
22	*
23	* You may elect to license modified versions of this file under the
24	* terms and conditions of either the GPL or the CDDL or both.
25	*/
26
27
28	/*********************************************************************************************************************************
29	* Header Files *
30	*********************************************************************************************************************************/
31	#define LOG_GROUP LOG_GROUP_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 RTLNX_VER_MIN(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 (RTLNX_VER_MIN(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 RTLNX_VER_MAX(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 RTLNX_VER_MIN(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) && RTLNX_VER_MAX(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 10485760 /* 10 MB */
148	# else
149	# define EXEC_MEMORY_SIZE 8388608 /* 8 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 RTLNX_VER_MAX(2,6,18)
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 RTLNX_VER_MAX(2,6,18)
216	devfs_name: DEVICE_NAME_USR,
217	# endif
218	};
219
220
221	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
222	# if RTLNX_VER_MIN(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 RTLNX_VER_MAX(2,6,30)
236	.suspend = VBoxDrvSuspend,
237	.resume = VBoxDrvResume,
238	# endif
239	/** @todo .shutdown? */
240	.driver =
241	{
242	.name = "vboxdrv",
243	# if RTLNX_VER_MIN(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 RTLNX_VER_MIN(2,6,29)
263	# if RTLNX_VER_MIN(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 RTLNX_VER_MIN(2,6,29)
276	# if RTLNX_VER_MIN(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 RTLNX_VER_MIN(2,6,29)
289	# if RTLNX_VER_MIN(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/threads\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) && RTLNX_VER_MAX(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 RTLNX_VER_MIN(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 RTLNX_VER_MIN(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	EXPORT_SYMBOL(SUPDrvLinuxIDC);
754
755
756	/**
757	* Used by native wrapper modules, forwarding to supdrvLdrRegisterWrappedModule
758	* with device extension prepended to the argument list.
759	*/
760	SUPR0DECL(int) SUPDrvLinuxLdrRegisterWrappedModule(PCSUPLDRWRAPPEDMODULE pWrappedModInfo, void pvLnxModule, void *phMod)
761	{
762	AssertPtrReturn(pvLnxModule, VERR_INVALID_POINTER);
763	return supdrvLdrRegisterWrappedModule(&g_DevExt, pWrappedModInfo, pvLnxModule, phMod);
764	}
765	EXPORT_SYMBOL(SUPDrvLinuxLdrRegisterWrappedModule);
766
767
768	/**
769	* Used by native wrapper modules, forwarding to supdrvLdrDeregisterWrappedModule
770	* with device extension prepended to the argument list.
771	*/
772	SUPR0DECL(int) SUPDrvLinuxLdrDeregisterWrappedModule(PCSUPLDRWRAPPEDMODULE pWrappedModInfo, void **phMod)
773	{
774	return supdrvLdrDeregisterWrappedModule(&g_DevExt, pWrappedModInfo, phMod);
775	}
776	EXPORT_SYMBOL(SUPDrvLinuxLdrDeregisterWrappedModule);
777
778
779	RTCCUINTREG VBOXCALL supdrvOSChangeCR4(RTCCUINTREG fOrMask, RTCCUINTREG fAndMask)
780	{
781	#if RTLNX_VER_MIN(5,8,0)
782	unsigned long fSavedFlags;
783	local_irq_save(fSavedFlags);
784	RTCCUINTREG const uOld = cr4_read_shadow();
785	cr4_update_irqsoff(fOrMask, ~fAndMask); /* Same as this function, only it is not returning the old value. */
786	AssertMsg(cr4_read_shadow() == ((uOld & fAndMask) \| fOrMask),
787	("fOrMask=%#RTreg fAndMask=%#RTreg uOld=%#RTreg; new cr4=%#llx\n", fOrMask, fAndMask, uOld, cr4_read_shadow()));
788	local_irq_restore(fSavedFlags);
789	#else
790	# if RTLNX_VER_MIN(3,20,0)
791	RTCCUINTREG const uOld = this_cpu_read(cpu_tlbstate.cr4);
792	# else
793	RTCCUINTREG const uOld = ASMGetCR4();
794	# endif
795	RTCCUINTREG const uNew = (uOld & fAndMask) \| fOrMask;
796	if (uNew != uOld)
797	{
798	# if RTLNX_VER_MIN(3,20,0)
799	this_cpu_write(cpu_tlbstate.cr4, uNew);
800	__write_cr4(uNew);
801	# else
802	ASMSetCR4(uNew);
803	# endif
804	}
805	#endif
806	return uOld;
807	}
808
809
810	void VBOXCALL supdrvOSCleanupSession(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession)
811	{
812	NOREF(pDevExt);
813	NOREF(pSession);
814	}
815
816
817	void VBOXCALL supdrvOSSessionHashTabInserted(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, void *pvUser)
818	{
819	NOREF(pDevExt); NOREF(pSession); NOREF(pvUser);
820	}
821
822
823	void VBOXCALL supdrvOSSessionHashTabRemoved(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, void *pvUser)
824	{
825	NOREF(pDevExt); NOREF(pSession); NOREF(pvUser);
826	}
827
828
829	/**
830	* Initializes any OS specific object creator fields.
831	*/
832	void VBOXCALL supdrvOSObjInitCreator(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession)
833	{
834	NOREF(pObj);
835	NOREF(pSession);
836	}
837
838
839	/**
840	* Checks if the session can access the object.
841	*
842	* @returns true if a decision has been made.
843	* @returns false if the default access policy should be applied.
844	*
845	* @param pObj The object in question.
846	* @param pSession The session wanting to access the object.
847	* @param pszObjName The object name, can be NULL.
848	* @param prc Where to store the result when returning true.
849	*/
850	bool VBOXCALL supdrvOSObjCanAccess(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession, const char pszObjName, int prc)
851	{
852	NOREF(pObj);
853	NOREF(pSession);
854	NOREF(pszObjName);
855	NOREF(prc);
856	return false;
857	}
858
859
860	bool VBOXCALL supdrvOSGetForcedAsyncTscMode(PSUPDRVDEVEXT pDevExt)
861	{
862	return force_async_tsc != 0;
863	}
864
865
866	bool VBOXCALL supdrvOSAreCpusOfflinedOnSuspend(void)
867	{
868	return true;
869	}
870
871
872	bool VBOXCALL supdrvOSAreTscDeltasInSync(void)
873	{
874	return false;
875	}
876
877
878	int VBOXCALL supdrvOSLdrOpen(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, const char *pszFilename)
879	{
880	NOREF(pDevExt); NOREF(pImage); NOREF(pszFilename);
881	return VERR_NOT_SUPPORTED;
882	}
883
884
885	int VBOXCALL supdrvOSLdrValidatePointer(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, void *pv,
886	const uint8_t pbImageBits, const char pszSymbol)
887	{
888	NOREF(pDevExt); NOREF(pImage); NOREF(pv); NOREF(pbImageBits); NOREF(pszSymbol);
889	return VERR_NOT_SUPPORTED;
890	}
891
892
893	int VBOXCALL supdrvOSLdrLoad(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, const uint8_t *pbImageBits, PSUPLDRLOAD pReq)
894	{
895	NOREF(pDevExt); NOREF(pImage); NOREF(pbImageBits); NOREF(pReq);
896	return VERR_NOT_SUPPORTED;
897	}
898
899
900	void VBOXCALL supdrvOSLdrUnload(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
901	{
902	NOREF(pDevExt); NOREF(pImage);
903	}
904
905
906	/** @def VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS
907	* A very crude hack for debugging using perf and dtrace.
908	*
909	* DO ABSOLUTELY NOT ENABLE IN PRODUCTION BUILDS! DEVELOPMENT ONLY!!
910	* DO ABSOLUTELY NOT ENABLE IN PRODUCTION BUILDS! DEVELOPMENT ONLY!!
911	* DO ABSOLUTELY NOT ENABLE IN PRODUCTION BUILDS! DEVELOPMENT ONLY!!
912	*
913	*/
914	#if 0 \|\| defined(DOXYGEN_RUNNING)
915	# define VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS
916	#endif
917
918	#if defined(VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS) && defined(CONFIG_MODULES_TREE_LOOKUP)
919	/** Whether g_pfnModTreeInsert and g_pfnModTreeRemove have been initialized.
920	* @remarks can still be NULL after init. */
921	static volatile bool g_fLookedForModTreeFunctions = false;
922	static void (g_pfnModTreeInsert)(struct mod_tree_node ) = NULL; /*< __mod_tree_insert /
923	static void (g_pfnModTreeRemove)(struct mod_tree_node ) = NULL; /*< __mod_tree_remove /
924	#endif
925
926
927	void VBOXCALL supdrvOSLdrNotifyOpened(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, const char *pszFilename)
928	{
929	#ifdef VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS /* Not for production use!! Debugging only! */
930	/*
931	* This trick stops working with 4.2 when CONFIG_MODULES_TREE_LOOKUP is
932	* defined. The module lookups are done via a tree structure and we
933	* cannot get at the root of it. :-(
934	*/
935	# ifdef CONFIG_KALLSYMS
936	size_t const cchName = strlen(pImage->szName);
937	# endif
938	struct module pMyMod, pSelfMod, pTestMod, pTestModByName;
939	IPRT_LINUX_SAVE_EFL_AC();
940
941	pImage->pLnxModHack = NULL;
942
943	# ifdef CONFIG_MODULES_TREE_LOOKUP
944	/*
945	* This is pretty naive, but works for 4.2 on arch linux. I don't think we
946	* can count on finding __mod_tree_remove in all kernel builds as it's not
947	* marked noinline like __mod_tree_insert.
948	*/
949	if (!g_fLookedForModTreeFunctions)
950	{
951	unsigned long ulInsert = kallsyms_lookup_name("__mod_tree_insert");
952	unsigned long ulRemove = kallsyms_lookup_name("__mod_tree_remove");
953	if (!ulInsert \|\| !ulRemove)
954	{
955	g_fLookedForModTreeFunctions = true;
956	printk(KERN_ERR "vboxdrv: failed to locate __mod_tree_insert and __mod_tree_remove.\n");
957	IPRT_LINUX_RESTORE_EFL_AC();
958	return;
959	}
960	(unsigned long )&g_pfnModTreeInsert = ulInsert;
961	(unsigned long )&g_pfnModTreeRemove = ulRemove;
962	ASMCompilerBarrier();
963	g_fLookedForModTreeFunctions = true;
964	}
965	else if (!g_pfnModTreeInsert \|\| !g_pfnModTreeRemove)
966	return;
967	#endif
968
969	/*
970	* Make sure we've found our own module, otherwise we cannot access the linked list.
971	*/
972	mutex_lock(&module_mutex);
973	pSelfMod = find_module("vboxdrv");
974	mutex_unlock(&module_mutex);
975	if (!pSelfMod)
976	{
977	IPRT_LINUX_RESTORE_EFL_AC();
978	return;
979	}
980
981	/*
982	* Cook up a module structure for the image.
983	* We allocate symbol and string tables in the allocation and the module to keep things simple.
984	*/
985	# ifdef CONFIG_KALLSYMS
986	pMyMod = (struct module )RTMemAllocZ(sizeof(pMyMod)
987	+ sizeof(Elf_Sym) * 3
988	+ 1 + cchName * 2 + sizeof("_start") + sizeof("_end") + 4 );
989	# else
990	pMyMod = (struct module )RTMemAllocZ(sizeof(pMyMod));
991	# endif
992	if (pMyMod)
993	{
994	int rc = VINF_SUCCESS;
995	# ifdef CONFIG_KALLSYMS
996	Elf_Sym paSymbols = (Elf_Sym )(pMyMod + 1);
997	char pchStrTab = (char )(paSymbols + 3);
998	# endif
999
1000	pMyMod->state = MODULE_STATE_LIVE;
1001	INIT_LIST_HEAD(&pMyMod->list); /* just in case */
1002
1003	/* Perf only matches up files with a .ko extension (maybe .ko.gz),
1004	so in order for this crap to work smoothly, we append .ko to the
1005	module name and require the user to create symbolic links in
1006	/lib/modules/`uname -r`:
1007	for i in VMMR0.r0 VBoxDDR0.r0 VBoxDD2R0.r0; do
1008	sudo ln -s /mnt/scratch/vbox/svn/trunk/out/linux.amd64/debug/bin/$i /lib/modules/`uname -r`/$i.ko;
1009	done */
1010	RTStrPrintf(pMyMod->name, sizeof(pMyMod->name), "%s", pImage->szName);
1011
1012	/* sysfs bits. */
1013	INIT_LIST_HEAD(&pMyMod->mkobj.kobj.entry); /* rest of kobj is already zeroed, hopefully never accessed... */
1014	pMyMod->mkobj.mod = pMyMod;
1015	pMyMod->mkobj.drivers_dir = NULL;
1016	pMyMod->mkobj.mp = NULL;
1017	pMyMod->mkobj.kobj_completion = NULL;
1018
1019	pMyMod->modinfo_attrs = NULL; /* hopefully not accessed after setup. */
1020	pMyMod->holders_dir = NULL; /* hopefully not accessed. */
1021	pMyMod->version = "N/A";
1022	pMyMod->srcversion = "N/A";
1023
1024	/* We export no symbols. */
1025	pMyMod->num_syms = 0;
1026	pMyMod->syms = NULL;
1027	pMyMod->crcs = NULL;
1028
1029	pMyMod->num_gpl_syms = 0;
1030	pMyMod->gpl_syms = NULL;
1031	pMyMod->gpl_crcs = NULL;
1032
1033	pMyMod->num_gpl_future_syms = 0;
1034	pMyMod->gpl_future_syms = NULL;
1035	pMyMod->gpl_future_crcs = NULL;
1036
1037	# if CONFIG_UNUSED_SYMBOLS
1038	pMyMod->num_unused_syms = 0;
1039	pMyMod->unused_syms = NULL;
1040	pMyMod->unused_crcs = NULL;
1041
1042	pMyMod->num_unused_gpl_syms = 0;
1043	pMyMod->unused_gpl_syms = NULL;
1044	pMyMod->unused_gpl_crcs = NULL;
1045	# endif
1046	/* No kernel parameters either. */
1047	pMyMod->kp = NULL;
1048	pMyMod->num_kp = 0;
1049
1050	# ifdef CONFIG_MODULE_SIG
1051	/* Pretend ok signature. */
1052	pMyMod->sig_ok = true;
1053	# endif
1054	/* No exception table. */
1055	pMyMod->num_exentries = 0;
1056	pMyMod->extable = NULL;
1057
1058	/* No init function */
1059	pMyMod->init = NULL;
1060	pMyMod->module_init = NULL;
1061	pMyMod->init_size = 0;
1062	pMyMod->init_ro_size = 0;
1063	pMyMod->init_text_size = 0;
1064
1065	/* The module address and size. It's all text. */
1066	pMyMod->module_core = pImage->pvImage;
1067	pMyMod->core_size = pImage->cbImageBits;
1068	pMyMod->core_text_size = pImage->cbImageBits;
1069	pMyMod->core_ro_size = pImage->cbImageBits;
1070
1071	#ifdef CONFIG_MODULES_TREE_LOOKUP
1072	/* Fill in the self pointers for the tree nodes. */
1073	pMyMod->mtn_core.mod = pMyMod;
1074	pMyMod->mtn_init.mod = pMyMod;
1075	#endif
1076	/* They invented the tained bit for us, didn't they? */
1077	pMyMod->taints = 1;
1078
1079	# ifdef CONFIG_GENERIC_BUGS
1080	/* No BUGs in our modules. */
1081	pMyMod->num_bugs = 0;
1082	INIT_LIST_HEAD(&pMyMod->bug_list);
1083	pMyMod->bug_table = NULL;
1084	# endif
1085
1086	# ifdef CONFIG_KALLSYMS
1087	/* The core stuff is documented as only used when loading. So just zero them. */
1088	pMyMod->core_num_syms = 0;
1089	pMyMod->core_symtab = NULL;
1090	pMyMod->core_strtab = NULL;
1091
1092	/* Construct a symbol table with start and end symbols.
1093	Note! We don't have our own symbol table at this point, image bit
1094	are not uploaded yet! */
1095	pMyMod->num_symtab = 3;
1096	pMyMod->symtab = paSymbols;
1097	pMyMod->strtab = pchStrTab;
1098	RT_ZERO(paSymbols[0]);
1099	pchStrTab[0] = '\0';
1100	paSymbols[1].st_name = 1;
1101	paSymbols[2].st_name = 2 + RTStrPrintf(&pchStrTab[paSymbols[1].st_name], cchName + sizeof("_start"),
1102	"%s_start", pImage->szName);
1103	RTStrPrintf(&pchStrTab[paSymbols[2].st_name], cchName + sizeof("_end"), "%s_end", pImage->szName);
1104	paSymbols[1].st_info = 't';
1105	paSymbols[2].st_info = 'b';
1106	paSymbols[1].st_other = 0;
1107	paSymbols[2].st_other = 0;
1108	paSymbols[1].st_shndx = 0;
1109	paSymbols[2].st_shndx = 0;
1110	paSymbols[1].st_value = (uintptr_t)pImage->pvImage;
1111	paSymbols[2].st_value = (uintptr_t)pImage->pvImage + pImage->cbImageBits - 1;
1112	paSymbols[1].st_size = pImage->cbImageBits - 1;
1113	paSymbols[2].st_size = 1;
1114	# endif
1115	/* No arguments, but seems its always non-NULL so put empty string there. */
1116	pMyMod->args = "";
1117
1118	# ifdef CONFIG_SMP
1119	/* No per CPU data. */
1120	pMyMod->percpu = NULL;
1121	pMyMod->percpu_size = 0;
1122	# endif
1123	# ifdef CONFIG_TRACEPOINTS
1124	/* No tracepoints we like to share. */
1125	pMyMod->num_tracepoints = 0;
1126	pMyMod->tracepoints_ptrs = NULL;
1127	#endif
1128	# ifdef HAVE_JUMP_LABEL
1129	/* No jump lable stuff either. */
1130	pMyMod->jump_entries = NULL;
1131	pMyMod->num_jump_entries = 0;
1132	# endif
1133	# ifdef CONFIG_TRACING
1134	pMyMod->num_trace_bprintk_fmt = 0;
1135	pMyMod->trace_bprintk_fmt_start = NULL;
1136	# endif
1137	# ifdef CONFIG_EVENT_TRACING
1138	pMyMod->trace_events = NULL;
1139	pMyMod->num_trace_events = 0;
1140	# endif
1141	# ifdef CONFIG_FTRACE_MCOUNT_RECORD
1142	pMyMod->num_ftrace_callsites = 0;
1143	pMyMod->ftrace_callsites = NULL;
1144	# endif
1145	# ifdef CONFIG_MODULE_UNLOAD
1146	/* Dependency lists, not worth sharing */
1147	INIT_LIST_HEAD(&pMyMod->source_list);
1148	INIT_LIST_HEAD(&pMyMod->target_list);
1149
1150	/* Nobody waiting and no exit function. */
1151	# if RTLNX_VER_MAX(3,13,0)
1152	pMyMod->waiter = NULL;
1153	# endif
1154	pMyMod->exit = NULL;
1155
1156	/* References, very important as we must not allow the module
1157	to be unloaded using rmmod. */
1158	# if RTLNX_VER_MIN(3,19,0)
1159	atomic_set(&pMyMod->refcnt, 42);
1160	# else
1161	pMyMod->refptr = alloc_percpu(struct module_ref);
1162	if (pMyMod->refptr)
1163	{
1164	int iCpu;
1165	for_each_possible_cpu(iCpu)
1166	{
1167	per_cpu_ptr(pMyMod->refptr, iCpu)->decs = 0;
1168	per_cpu_ptr(pMyMod->refptr, iCpu)->incs = 1;
1169	}
1170	}
1171	else
1172	rc = VERR_NO_MEMORY;
1173	# endif
1174	# endif
1175	# ifdef CONFIG_CONSTRUCTORS
1176	/* No constructors. */
1177	pMyMod->ctors = NULL;
1178	pMyMod->num_ctors = 0;
1179	# endif
1180	if (RT_SUCCESS(rc))
1181	{
1182	bool fIsModText;
1183
1184	/*
1185	* Add the module to the list.
1186	*/
1187	mutex_lock(&module_mutex);
1188	list_add_rcu(&pMyMod->list, &pSelfMod->list);
1189	pImage->pLnxModHack = pMyMod;
1190	# ifdef CONFIG_MODULES_TREE_LOOKUP
1191	g_pfnModTreeInsert(&pMyMod->mtn_core); /* __mod_tree_insert */
1192	# endif
1193	mutex_unlock(&module_mutex);
1194
1195	/*
1196	* Test it.
1197	*/
1198	mutex_lock(&module_mutex);
1199	pTestModByName = find_module(pMyMod->name);
1200	pTestMod = __module_address((uintptr_t)pImage->pvImage + pImage->cbImageBits / 4);
1201	fIsModText = __module_text_address((uintptr_t)pImage->pvImage + pImage->cbImageBits / 2);
1202	mutex_unlock(&module_mutex);
1203	if ( pTestMod == pMyMod
1204	&& pTestModByName == pMyMod
1205	&& fIsModText)
1206	printk(KERN_ERR "vboxdrv: fake module works for '%s' (%#lx to %#lx)\n",
1207	pMyMod->name, (unsigned long)paSymbols[1].st_value, (unsigned long)paSymbols[2].st_value);
1208	else
1209	printk(KERN_ERR "vboxdrv: failed to find fake module (pTestMod=%p, pTestModByName=%p, pMyMod=%p, fIsModText=%d)\n",
1210	pTestMod, pTestModByName, pMyMod, fIsModText);
1211	}
1212	else
1213	RTMemFree(pMyMod);
1214	}
1215
1216	IPRT_LINUX_RESTORE_EFL_AC();
1217	#else
1218	pImage->pLnxModHack = NULL;
1219	#endif
1220	NOREF(pDevExt); NOREF(pImage);
1221	}
1222
1223
1224	void VBOXCALL supdrvOSLdrNotifyUnloaded(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
1225	{
1226	#ifdef VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS /* Not for production use!! Debugging only! */
1227	struct module *pMyMod = pImage->pLnxModHack;
1228	pImage->pLnxModHack = NULL;
1229	if (pMyMod)
1230	{
1231	/*
1232	* Remove the fake module list entry and free it.
1233	*/
1234	IPRT_LINUX_SAVE_EFL_AC();
1235	mutex_lock(&module_mutex);
1236	list_del_rcu(&pMyMod->list);
1237	# ifdef CONFIG_MODULES_TREE_LOOKUP
1238	g_pfnModTreeRemove(&pMyMod->mtn_core);
1239	# endif
1240	synchronize_sched();
1241	mutex_unlock(&module_mutex);
1242
1243	# if RTLNX_VER_MAX(3,19,0)
1244	free_percpu(pMyMod->refptr);
1245	# endif
1246	RTMemFree(pMyMod);
1247	IPRT_LINUX_RESTORE_EFL_AC();
1248	}
1249
1250	#else
1251	Assert(pImage->pLnxModHack == NULL);
1252	#endif
1253	NOREF(pDevExt); NOREF(pImage);
1254	}
1255
1256
1257	int VBOXCALL supdrvOSLdrQuerySymbol(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage,
1258	const char pszSymbol, size_t cchSymbol, void *ppvSymbol)
1259	{
1260	#ifdef VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS
1261	# error "implement me!"
1262	#endif
1263	RT_NOREF(pDevExt, pImage, pszSymbol, cchSymbol, ppvSymbol);
1264	return VERR_WRONG_ORDER;
1265	}
1266
1267
1268	void VBOXCALL supdrvOSLdrRetainWrapperModule(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
1269	{
1270	struct module pLnxMod = (struct module )pImage->pvWrappedNative;
1271	Assert(!pImage->fLnxWrapperRef);
1272	AssertReturnVoid(pLnxMod);
1273	pImage->fLnxWrapperRef = try_module_get(pLnxMod);
1274	RT_NOREF(pDevExt);
1275	}
1276
1277
1278	void VBOXCALL supdrvOSLdrReleaseWrapperModule(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
1279	{
1280	if (pImage->fLnxWrapperRef)
1281	{
1282	struct module pLnxMod = (struct module )pImage->pvWrappedNative;
1283	pImage->fLnxWrapperRef = false;
1284	module_put(pLnxMod);
1285	}
1286	RT_NOREF(pDevExt);
1287	}
1288
1289
1290	#ifdef SUPDRV_WITH_MSR_PROBER
1291
1292	int VBOXCALL supdrvOSMsrProberRead(uint32_t uMsr, RTCPUID idCpu, uint64_t *puValue)
1293	{
1294	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1295	uint32_t u32Low, u32High;
1296	int rc;
1297
1298	IPRT_LINUX_SAVE_EFL_AC();
1299	if (idCpu == NIL_RTCPUID)
1300	rc = rdmsr_safe(uMsr, &u32Low, &u32High);
1301	else if (RTMpIsCpuOnline(idCpu))
1302	rc = rdmsr_safe_on_cpu(idCpu, uMsr, &u32Low, &u32High);
1303	else
1304	return VERR_CPU_OFFLINE;
1305	IPRT_LINUX_RESTORE_EFL_AC();
1306	if (rc == 0)
1307	{
1308	*puValue = RT_MAKE_U64(u32Low, u32High);
1309	return VINF_SUCCESS;
1310	}
1311	return VERR_ACCESS_DENIED;
1312	# else
1313	return VERR_NOT_SUPPORTED;
1314	# endif
1315	}
1316
1317
1318	int VBOXCALL supdrvOSMsrProberWrite(uint32_t uMsr, RTCPUID idCpu, uint64_t uValue)
1319	{
1320	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1321	int rc;
1322
1323	IPRT_LINUX_SAVE_EFL_AC();
1324	if (idCpu == NIL_RTCPUID)
1325	rc = wrmsr_safe(uMsr, RT_LODWORD(uValue), RT_HIDWORD(uValue));
1326	else if (RTMpIsCpuOnline(idCpu))
1327	rc = wrmsr_safe_on_cpu(idCpu, uMsr, RT_LODWORD(uValue), RT_HIDWORD(uValue));
1328	else
1329	return VERR_CPU_OFFLINE;
1330	IPRT_LINUX_RESTORE_EFL_AC();
1331
1332	if (rc == 0)
1333	return VINF_SUCCESS;
1334	return VERR_ACCESS_DENIED;
1335	# else
1336	return VERR_NOT_SUPPORTED;
1337	# endif
1338	}
1339
1340	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1341	/**
1342	* Worker for supdrvOSMsrProberModify.
1343	*/
1344	static DECLCALLBACK(void) supdrvLnxMsrProberModifyOnCpu(RTCPUID idCpu, void pvUser1, void pvUser2)
1345	{
1346	PSUPMSRPROBER pReq = (PSUPMSRPROBER)pvUser1;
1347	register uint32_t uMsr = pReq->u.In.uMsr;
1348	bool const fFaster = pReq->u.In.enmOp == SUPMSRPROBEROP_MODIFY_FASTER;
1349	uint64_t uBefore;
1350	uint64_t uWritten;
1351	uint64_t uAfter;
1352	int rcBefore, rcWrite, rcAfter, rcRestore;
1353	RTCCUINTREG fOldFlags;
1354
1355	/* Initialize result variables. */
1356	uBefore = uWritten = uAfter = 0;
1357	rcWrite = rcAfter = rcRestore = -EIO;
1358
1359	/*
1360	* Do the job.
1361	*/
1362	fOldFlags = ASMIntDisableFlags();
1363	ASMCompilerBarrier(); /* paranoia */
1364	if (!fFaster)
1365	ASMWriteBackAndInvalidateCaches();
1366
1367	rcBefore = rdmsrl_safe(uMsr, &uBefore);
1368	if (rcBefore >= 0)
1369	{
1370	register uint64_t uRestore = uBefore;
1371	uWritten = uRestore;
1372	uWritten &= pReq->u.In.uArgs.Modify.fAndMask;
1373	uWritten \|= pReq->u.In.uArgs.Modify.fOrMask;
1374
1375	rcWrite = wrmsr_safe(uMsr, RT_LODWORD(uWritten), RT_HIDWORD(uWritten));
1376	rcAfter = rdmsrl_safe(uMsr, &uAfter);
1377	rcRestore = wrmsr_safe(uMsr, RT_LODWORD(uRestore), RT_HIDWORD(uRestore));
1378
1379	if (!fFaster)
1380	{
1381	ASMWriteBackAndInvalidateCaches();
1382	ASMReloadCR3();
1383	ASMNopPause();
1384	}
1385	}
1386
1387	ASMCompilerBarrier(); /* paranoia */
1388	ASMSetFlags(fOldFlags);
1389
1390	/*
1391	* Write out the results.
1392	*/
1393	pReq->u.Out.uResults.Modify.uBefore = uBefore;
1394	pReq->u.Out.uResults.Modify.uWritten = uWritten;
1395	pReq->u.Out.uResults.Modify.uAfter = uAfter;
1396	pReq->u.Out.uResults.Modify.fBeforeGp = rcBefore != 0;
1397	pReq->u.Out.uResults.Modify.fModifyGp = rcWrite != 0;
1398	pReq->u.Out.uResults.Modify.fAfterGp = rcAfter != 0;
1399	pReq->u.Out.uResults.Modify.fRestoreGp = rcRestore != 0;
1400	RT_ZERO(pReq->u.Out.uResults.Modify.afReserved);
1401	}
1402	# endif
1403
1404
1405	int VBOXCALL supdrvOSMsrProberModify(RTCPUID idCpu, PSUPMSRPROBER pReq)
1406	{
1407	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1408	if (idCpu == NIL_RTCPUID)
1409	{
1410	supdrvLnxMsrProberModifyOnCpu(idCpu, pReq, NULL);
1411	return VINF_SUCCESS;
1412	}
1413	return RTMpOnSpecific(idCpu, supdrvLnxMsrProberModifyOnCpu, pReq, NULL);
1414	# else
1415	return VERR_NOT_SUPPORTED;
1416	# endif
1417	}
1418
1419	#endif /* SUPDRV_WITH_MSR_PROBER */
1420
1421
1422	/**
1423	* Converts a supdrv error code to an linux error code.
1424	*
1425	* @returns corresponding linux error code.
1426	* @param rc IPRT status code.
1427	*/
1428	static int VBoxDrvLinuxErr2LinuxErr(int rc)
1429	{
1430	switch (rc)
1431	{
1432	case VINF_SUCCESS: return 0;
1433	case VERR_GENERAL_FAILURE: return -EACCES;
1434	case VERR_INVALID_PARAMETER: return -EINVAL;
1435	case VERR_INVALID_MAGIC: return -EILSEQ;
1436	case VERR_INVALID_HANDLE: return -ENXIO;
1437	case VERR_INVALID_POINTER: return -EFAULT;
1438	case VERR_LOCK_FAILED: return -ENOLCK;
1439	case VERR_ALREADY_LOADED: return -EEXIST;
1440	case VERR_PERMISSION_DENIED: return -EPERM;
1441	case VERR_VERSION_MISMATCH: return -ENOSYS;
1442	case VERR_IDT_FAILED: return -1000;
1443	}
1444
1445	return -EPERM;
1446	}
1447
1448
1449	SUPR0DECL(int) SUPR0HCPhysToVirt(RTHCPHYS HCPhys, void **ppv)
1450	{
1451	AssertReturn(!(HCPhys & PAGE_OFFSET_MASK), VERR_INVALID_POINTER);
1452	AssertReturn(HCPhys != NIL_RTHCPHYS, VERR_INVALID_POINTER);
1453	/* Would've like to use valid_phys_addr_range for this test, but it isn't exported. */
1454	AssertReturn((HCPhys \| PAGE_OFFSET_MASK) < __pa(high_memory), VERR_INVALID_POINTER);
1455	*ppv = phys_to_virt(HCPhys);
1456	return VINF_SUCCESS;
1457	}
1458	SUPR0_EXPORT_SYMBOL(SUPR0HCPhysToVirt);
1459
1460
1461	RTDECL(int) SUPR0Printf(const char *pszFormat, ...)
1462	{
1463	va_list va;
1464	char szMsg[512];
1465	IPRT_LINUX_SAVE_EFL_AC();
1466
1467	va_start(va, pszFormat);
1468	RTStrPrintfV(szMsg, sizeof(szMsg) - 1, pszFormat, va);
1469	va_end(va);
1470	szMsg[sizeof(szMsg) - 1] = '\0';
1471
1472	printk("%s", szMsg);
1473
1474	IPRT_LINUX_RESTORE_EFL_AC();
1475	return 0;
1476	}
1477	SUPR0_EXPORT_SYMBOL(SUPR0Printf);
1478
1479
1480	SUPR0DECL(uint32_t) SUPR0GetKernelFeatures(void)
1481	{
1482	uint32_t fFlags = 0;
1483	#ifdef CONFIG_PAX_KERNEXEC
1484	fFlags \|= SUPKERNELFEATURES_GDT_READ_ONLY;
1485	#endif
1486	#if RTLNX_VER_MIN(4,12,0)
1487	fFlags \|= SUPKERNELFEATURES_GDT_NEED_WRITABLE;
1488	#endif
1489	#if defined(VBOX_STRICT) \|\| defined(VBOX_WITH_EFLAGS_AC_SET_IN_VBOXDRV)
1490	fFlags \|= SUPKERNELFEATURES_SMAP;
1491	#elif defined(CONFIG_X86_SMAP)
1492	if (ASMGetCR4() & X86_CR4_SMAP)
1493	fFlags \|= SUPKERNELFEATURES_SMAP;
1494	#endif
1495	return fFlags;
1496	}
1497	SUPR0_EXPORT_SYMBOL(SUPR0GetKernelFeatures);
1498
1499
1500	int VBOXCALL supdrvOSGetCurrentGdtRw(RTHCUINTPTR *pGdtRw)
1501	{
1502	#if RTLNX_VER_MIN(4,12,0)
1503	*pGdtRw = (RTHCUINTPTR)get_current_gdt_rw();
1504	return VINF_SUCCESS;
1505	#else
1506	return VERR_NOT_IMPLEMENTED;
1507	#endif
1508	}
1509
1510
1511	module_init(VBoxDrvLinuxInit);
1512	module_exit(VBoxDrvLinuxUnload);
1513
1514	MODULE_AUTHOR(VBOX_VENDOR);
1515	MODULE_DESCRIPTION(VBOX_PRODUCT " Support Driver");
1516	MODULE_LICENSE("GPL");
1517	#ifdef MODULE_VERSION
1518	MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV) " (" RT_XSTR(SUPDRV_IOC_VERSION) ")");
1519	#endif
1520
1521	module_param(force_async_tsc, int, 0444);
1522	MODULE_PARM_DESC(force_async_tsc, "force the asynchronous TSC mode");
1523

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

source: vbox/trunk/src/VBox/HostDrivers/Support/linux/SUPDrv-linux.c@ 87702

Download in other formats: