SUPDrv-linux.c@ 90780

Last change on this file since 90780 was 90780, checked in by vboxsync, 3 years ago
SUP: VALID_PTR -> RT_VALID_PTR.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 49.0 KB

Line
1	/* $Id: SUPDrv-linux.c 90780 2021-08-23 09:25:17Z 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	/* Once externally provided, this string will be printed into kernel log on
96	* module start together with the rest of versioning information. */
97	#ifndef VBOX_EXTRA_VERSION_STRING
98	# define VBOX_EXTRA_VERSION_STRING ""
99	#endif
100
101
102	/*********************************************************************************************************************************
103	* Internal Functions *
104	*********************************************************************************************************************************/
105	static int __init VBoxDrvLinuxInit(void);
106	static void __exit VBoxDrvLinuxUnload(void);
107	static int VBoxDrvLinuxCreateSys(struct inode pInode, struct file pFilp);
108	static int VBoxDrvLinuxCreateUsr(struct inode pInode, struct file pFilp);
109	static int VBoxDrvLinuxClose(struct inode pInode, struct file pFilp);
110	#ifdef HAVE_UNLOCKED_IOCTL
111	static long VBoxDrvLinuxIOCtl(struct file *pFilp, unsigned int uCmd, unsigned long ulArg);
112	#else
113	static int VBoxDrvLinuxIOCtl(struct inode pInode, struct file pFilp, unsigned int uCmd, unsigned long ulArg);
114	#endif
115	static int VBoxDrvLinuxIOCtlSlow(struct file *pFilp, unsigned int uCmd, unsigned long ulArg, PSUPDRVSESSION pSession);
116	static int VBoxDrvLinuxErr2LinuxErr(int);
117	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
118	static int VBoxDrvProbe(struct platform_device *pDev);
119	# if RTLNX_VER_MIN(2,6,30)
120	static int VBoxDrvSuspend(struct device *pDev);
121	static int VBoxDrvResume(struct device *pDev);
122	# else
123	static int VBoxDrvSuspend(struct platform_device *pDev, pm_message_t State);
124	static int VBoxDrvResume(struct platform_device *pDev);
125	# endif
126	static void VBoxDevRelease(struct device *pDev);
127	#endif
128
129
130	/*********************************************************************************************************************************
131	* Global Variables *
132	*********************************************************************************************************************************/
133	/**
134	* Device extention & session data association structure.
135	*/
136	static SUPDRVDEVEXT g_DevExt;
137
138	/** Module parameter.
139	* Not prefixed because the name is used by macros and the end of this file. */
140	static int force_async_tsc = 0;
141
142	/** The system device name. */
143	#define DEVICE_NAME_SYS "vboxdrv"
144	/** The user device name. */
145	#define DEVICE_NAME_USR "vboxdrvu"
146
147	#if (defined(RT_ARCH_AMD64) && RTLNX_VER_MAX(2,6,23)) \|\| defined(VBOX_WITH_TEXT_MODMEM_HACK)
148	/**
149	* Memory for the executable memory heap (in IPRT).
150	*/
151	# ifdef DEBUG
152	# define EXEC_MEMORY_SIZE 10485760 /* 10 MB */
153	# else
154	# define EXEC_MEMORY_SIZE 8388608 /* 8 MB */
155	# endif
156	extern uint8_t g_abExecMemory[EXEC_MEMORY_SIZE];
157	# ifndef VBOX_WITH_TEXT_MODMEM_HACK
158	__asm__(".section execmemory, \"awx\", @progbits\n\t"
159	".align 32\n\t"
160	".globl g_abExecMemory\n"
161	"g_abExecMemory:\n\t"
162	".zero " RT_XSTR(EXEC_MEMORY_SIZE) "\n\t"
163	".type g_abExecMemory, @object\n\t"
164	".size g_abExecMemory, " RT_XSTR(EXEC_MEMORY_SIZE) "\n\t"
165	".text\n\t");
166	# else
167	__asm__(".text\n\t"
168	".align 4096\n\t"
169	".globl g_abExecMemory\n"
170	"g_abExecMemory:\n\t"
171	".zero " RT_XSTR(EXEC_MEMORY_SIZE) "\n\t"
172	".type g_abExecMemory, @object\n\t"
173	".size g_abExecMemory, " RT_XSTR(EXEC_MEMORY_SIZE) "\n\t"
174	".text\n\t");
175	# endif
176	#endif
177
178	/** The file_operations structure. */
179	static struct file_operations gFileOpsVBoxDrvSys =
180	{
181	owner: THIS_MODULE,
182	open: VBoxDrvLinuxCreateSys,
183	release: VBoxDrvLinuxClose,
184	#ifdef HAVE_UNLOCKED_IOCTL
185	unlocked_ioctl: VBoxDrvLinuxIOCtl,
186	#else
187	ioctl: VBoxDrvLinuxIOCtl,
188	#endif
189	};
190
191	/** The file_operations structure. */
192	static struct file_operations gFileOpsVBoxDrvUsr =
193	{
194	owner: THIS_MODULE,
195	open: VBoxDrvLinuxCreateUsr,
196	release: VBoxDrvLinuxClose,
197	#ifdef HAVE_UNLOCKED_IOCTL
198	unlocked_ioctl: VBoxDrvLinuxIOCtl,
199	#else
200	ioctl: VBoxDrvLinuxIOCtl,
201	#endif
202	};
203
204	/** The miscdevice structure for vboxdrv. */
205	static struct miscdevice gMiscDeviceSys =
206	{
207	minor: MISC_DYNAMIC_MINOR,
208	name: DEVICE_NAME_SYS,
209	fops: &gFileOpsVBoxDrvSys,
210	# if RTLNX_VER_MAX(2,6,18)
211	devfs_name: DEVICE_NAME_SYS,
212	# endif
213	};
214	/** The miscdevice structure for vboxdrvu. */
215	static struct miscdevice gMiscDeviceUsr =
216	{
217	minor: MISC_DYNAMIC_MINOR,
218	name: DEVICE_NAME_USR,
219	fops: &gFileOpsVBoxDrvUsr,
220	# if RTLNX_VER_MAX(2,6,18)
221	devfs_name: DEVICE_NAME_USR,
222	# endif
223	};
224
225
226	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
227	# if RTLNX_VER_MIN(2,6,30)
228	static struct dev_pm_ops gPlatformPMOps =
229	{
230	.suspend = VBoxDrvSuspend, /* before entering deep sleep */
231	.resume = VBoxDrvResume, /* after wakeup from deep sleep */
232	.freeze = VBoxDrvSuspend, /* before creating hibernation image */
233	.restore = VBoxDrvResume, /* after waking up from hibernation */
234	};
235	# endif
236
237	static struct platform_driver gPlatformDriver =
238	{
239	.probe = VBoxDrvProbe,
240	# if RTLNX_VER_MAX(2,6,30)
241	.suspend = VBoxDrvSuspend,
242	.resume = VBoxDrvResume,
243	# endif
244	/** @todo .shutdown? */
245	.driver =
246	{
247	.name = "vboxdrv",
248	# if RTLNX_VER_MIN(2,6,30)
249	.pm = &gPlatformPMOps,
250	# endif
251	}
252	};
253
254	static struct platform_device gPlatformDevice =
255	{
256	.name = "vboxdrv",
257	.dev =
258	{
259	.release = VBoxDevRelease
260	}
261	};
262	#endif /* VBOX_WITH_SUSPEND_NOTIFICATION */
263
264
265	DECLINLINE(RTUID) vboxdrvLinuxUid(void)
266	{
267	#if RTLNX_VER_MIN(2,6,29)
268	# if RTLNX_VER_MIN(3,5,0)
269	return from_kuid(current_user_ns(), current->cred->uid);
270	# else
271	return current->cred->uid;
272	# endif
273	#else
274	return current->uid;
275	#endif
276	}
277
278	DECLINLINE(RTGID) vboxdrvLinuxGid(void)
279	{
280	#if RTLNX_VER_MIN(2,6,29)
281	# if RTLNX_VER_MIN(3,5,0)
282	return from_kgid(current_user_ns(), current->cred->gid);
283	# else
284	return current->cred->gid;
285	# endif
286	#else
287	return current->gid;
288	#endif
289	}
290
291	DECLINLINE(RTUID) vboxdrvLinuxEuid(void)
292	{
293	#if RTLNX_VER_MIN(2,6,29)
294	# if RTLNX_VER_MIN(3,5,0)
295	return from_kuid(current_user_ns(), current->cred->euid);
296	# else
297	return current->cred->euid;
298	# endif
299	#else
300	return current->euid;
301	#endif
302	}
303
304	/**
305	* Initialize module.
306	*
307	* @returns appropriate status code.
308	*/
309	static int __init VBoxDrvLinuxInit(void)
310	{
311	int rc;
312
313	/*
314	* Check for synchronous/asynchronous TSC mode.
315	*/
316	printk(KERN_DEBUG "vboxdrv: Found %u processor cores/threads\n", (unsigned)RTMpGetOnlineCount());
317	rc = misc_register(&gMiscDeviceSys);
318	if (rc)
319	{
320	printk(KERN_ERR "vboxdrv: Can't register system misc device! rc=%d\n", rc);
321	return rc;
322	}
323	rc = misc_register(&gMiscDeviceUsr);
324	if (rc)
325	{
326	printk(KERN_ERR "vboxdrv: Can't register user misc device! rc=%d\n", rc);
327	misc_deregister(&gMiscDeviceSys);
328	return rc;
329	}
330	if (!rc)
331	{
332	/*
333	* Initialize the runtime.
334	* On AMD64 we'll have to donate the high rwx memory block to the exec allocator.
335	*/
336	rc = RTR0Init(0);
337	if (RT_SUCCESS(rc))
338	{
339	#if (defined(RT_ARCH_AMD64) && RTLNX_VER_MAX(2,6,23)) \|\| defined(VBOX_WITH_TEXT_MODMEM_HACK)
340	# ifdef VBOX_WITH_TEXT_MODMEM_HACK
341	set_memory_x(&g_abExecMemory[0], sizeof(g_abExecMemory) / PAGE_SIZE);
342	set_memory_rw(&g_abExecMemory[0], sizeof(g_abExecMemory) / PAGE_SIZE);
343	# endif
344	rc = RTR0MemExecDonate(&g_abExecMemory[0], sizeof(g_abExecMemory));
345	printk(KERN_DEBUG "VBoxDrv: dbg - g_abExecMemory=%p\n", (void *)&g_abExecMemory[0]);
346	#endif
347	Log(("VBoxDrv::ModuleInit\n"));
348
349	/*
350	* Initialize the device extension.
351	*/
352	if (RT_SUCCESS(rc))
353	rc = supdrvInitDevExt(&g_DevExt, sizeof(SUPDRVSESSION));
354	if (RT_SUCCESS(rc))
355	{
356	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
357	rc = platform_driver_register(&gPlatformDriver);
358	if (rc == 0)
359	{
360	rc = platform_device_register(&gPlatformDevice);
361	if (rc == 0)
362	#endif
363	{
364	printk(KERN_INFO "vboxdrv: TSC mode is %s, tentative frequency %llu Hz\n",
365	SUPGetGIPModeName(g_DevExt.pGip), g_DevExt.pGip->u64CpuHz);
366	LogFlow(("VBoxDrv::ModuleInit returning %#x\n", rc));
367	printk(KERN_DEBUG "vboxdrv: Successfully loaded version "
368	VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV)
369	VBOX_EXTRA_VERSION_STRING
370	" (interface " RT_XSTR(SUPDRV_IOC_VERSION) ")\n");
371	return rc;
372	}
373	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
374	else
375	platform_driver_unregister(&gPlatformDriver);
376	}
377	#endif
378	}
379
380	rc = -EINVAL;
381	RTR0TermForced();
382	}
383	else
384	rc = -EINVAL;
385
386	/*
387	* Failed, cleanup and return the error code.
388	*/
389	}
390	misc_deregister(&gMiscDeviceSys);
391	misc_deregister(&gMiscDeviceUsr);
392	Log(("VBoxDrv::ModuleInit returning %#x (minor:%d & %d)\n", rc, gMiscDeviceSys.minor, gMiscDeviceUsr.minor));
393	return rc;
394	}
395
396
397	/**
398	* Unload the module.
399	*/
400	static void __exit VBoxDrvLinuxUnload(void)
401	{
402	Log(("VBoxDrvLinuxUnload\n"));
403
404	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
405	platform_device_unregister(&gPlatformDevice);
406	platform_driver_unregister(&gPlatformDriver);
407	#endif
408
409	/*
410	* I Don't think it's possible to unload a driver which processes have
411	* opened, at least we'll blindly assume that here.
412	*/
413	misc_deregister(&gMiscDeviceUsr);
414	misc_deregister(&gMiscDeviceSys);
415
416	/*
417	* Destroy GIP, delete the device extension and terminate IPRT.
418	*/
419	supdrvDeleteDevExt(&g_DevExt);
420	RTR0TermForced();
421	}
422
423
424	/**
425	* Common open code.
426	*
427	* @param pInode Pointer to inode info structure.
428	* @param pFilp Associated file pointer.
429	* @param fUnrestricted Indicates which device node which was opened.
430	*/
431	static int vboxdrvLinuxCreateCommon(struct inode pInode, struct file pFilp, bool fUnrestricted)
432	{
433	int rc;
434	PSUPDRVSESSION pSession;
435	Log(("VBoxDrvLinuxCreate: pFilp=%p pid=%d/%d %s\n", pFilp, RTProcSelf(), current->pid, current->comm));
436
437	#ifdef VBOX_WITH_HARDENING
438	/*
439	* Only root is allowed to access the unrestricted device, enforce it!
440	*/
441	if ( fUnrestricted
442	&& vboxdrvLinuxEuid() != 0 /* root */ )
443	{
444	Log(("VBoxDrvLinuxCreate: euid=%d, expected 0 (root)\n", vboxdrvLinuxEuid()));
445	return -EPERM;
446	}
447	#endif /* VBOX_WITH_HARDENING */
448
449	/*
450	* Call common code for the rest.
451	*/
452	rc = supdrvCreateSession(&g_DevExt, true /* fUser */, fUnrestricted, &pSession);
453	if (!rc)
454	{
455	pSession->Uid = vboxdrvLinuxUid();
456	pSession->Gid = vboxdrvLinuxGid();
457	}
458
459	pFilp->private_data = pSession;
460
461	Log(("VBoxDrvLinuxCreate: g_DevExt=%p pSession=%p rc=%d/%d (pid=%d/%d %s)\n",
462	&g_DevExt, pSession, rc, VBoxDrvLinuxErr2LinuxErr(rc),
463	RTProcSelf(), current->pid, current->comm));
464	return VBoxDrvLinuxErr2LinuxErr(rc);
465	}
466
467
468	/** /dev/vboxdrv. */
469	static int VBoxDrvLinuxCreateSys(struct inode pInode, struct file pFilp)
470	{
471	return vboxdrvLinuxCreateCommon(pInode, pFilp, true);
472	}
473
474
475	/** /dev/vboxdrvu. */
476	static int VBoxDrvLinuxCreateUsr(struct inode pInode, struct file pFilp)
477	{
478	return vboxdrvLinuxCreateCommon(pInode, pFilp, false);
479	}
480
481
482	/**
483	* Close device.
484	*
485	* @param pInode Pointer to inode info structure.
486	* @param pFilp Associated file pointer.
487	*/
488	static int VBoxDrvLinuxClose(struct inode pInode, struct file pFilp)
489	{
490	Log(("VBoxDrvLinuxClose: pFilp=%p pSession=%p pid=%d/%d %s\n",
491	pFilp, pFilp->private_data, RTProcSelf(), current->pid, current->comm));
492	supdrvSessionRelease((PSUPDRVSESSION)pFilp->private_data);
493	pFilp->private_data = NULL;
494	return 0;
495	}
496
497
498	#ifdef VBOX_WITH_SUSPEND_NOTIFICATION
499	/**
500	* Dummy device release function. We have to provide this function,
501	* otherwise the kernel will complain.
502	*
503	* @param pDev Pointer to the platform device.
504	*/
505	static void VBoxDevRelease(struct device *pDev)
506	{
507	}
508
509	/**
510	* Dummy probe function.
511	*
512	* @param pDev Pointer to the platform device.
513	*/
514	static int VBoxDrvProbe(struct platform_device *pDev)
515	{
516	return 0;
517	}
518
519	/**
520	* Suspend callback.
521	* @param pDev Pointer to the platform device.
522	* @param State Message type, see Documentation/power/devices.txt.
523	* Ignored.
524	*/
525	# if RTLNX_VER_MIN(2,6,30) && !defined(DOXYGEN_RUNNING)
526	static int VBoxDrvSuspend(struct device *pDev)
527	# else
528	static int VBoxDrvSuspend(struct platform_device *pDev, pm_message_t State)
529	# endif
530	{
531	RTPowerSignalEvent(RTPOWEREVENT_SUSPEND);
532	return 0;
533	}
534
535	/**
536	* Resume callback.
537	*
538	* @param pDev Pointer to the platform device.
539	*/
540	# if RTLNX_VER_MIN(2,6,30)
541	static int VBoxDrvResume(struct device *pDev)
542	# else
543	static int VBoxDrvResume(struct platform_device *pDev)
544	# endif
545	{
546	RTPowerSignalEvent(RTPOWEREVENT_RESUME);
547	return 0;
548	}
549	#endif /* VBOX_WITH_SUSPEND_NOTIFICATION */
550
551
552	/**
553	* Device I/O Control entry point.
554	*
555	* @param pFilp Associated file pointer.
556	* @param uCmd The function specified to ioctl().
557	* @param ulArg The argument specified to ioctl().
558	*/
559	#if defined(HAVE_UNLOCKED_IOCTL) \|\| defined(DOXYGEN_RUNNING)
560	static long VBoxDrvLinuxIOCtl(struct file *pFilp, unsigned int uCmd, unsigned long ulArg)
561	#else
562	static int VBoxDrvLinuxIOCtl(struct inode pInode, struct file pFilp, unsigned int uCmd, unsigned long ulArg)
563	#endif
564	{
565	PSUPDRVSESSION pSession = (PSUPDRVSESSION)pFilp->private_data;
566	int rc;
567	#ifndef VBOX_WITHOUT_EFLAGS_AC_SET_IN_VBOXDRV
568	# if defined(VBOX_STRICT) \|\| defined(VBOX_WITH_EFLAGS_AC_SET_IN_VBOXDRV)
569	RTCCUINTREG fSavedEfl;
570
571	/*
572	* Refuse all I/O control calls if we've ever detected EFLAGS.AC being cleared.
573	*
574	* This isn't a problem, as there is absolutely nothing in the kernel context that
575	* depend on user context triggering cleanups. That would be pretty wild, right?
576	*/
577	if (RT_UNLIKELY(g_DevExt.cBadContextCalls > 0))
578	{
579	SUPR0Printf("VBoxDrvLinuxIOCtl: EFLAGS.AC=0 detected %u times, refusing all I/O controls!\n", g_DevExt.cBadContextCalls);
580	return ESPIPE;
581	}
582
583	fSavedEfl = ASMAddFlags(X86_EFL_AC);
584	# else
585	stac();
586	# endif
587	#endif
588
589	/*
590	* Deal with the two high-speed IOCtl that takes it's arguments from
591	* the session and iCmd, and only returns a VBox status code.
592	*/
593	AssertCompile(_IOC_NRSHIFT == 0 && _IOC_NRBITS == 8);
594	#ifdef HAVE_UNLOCKED_IOCTL
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	#else /* !HAVE_UNLOCKED_IOCTL */
601	unlock_kernel();
602	if (RT_LIKELY( (unsigned int)(uCmd - SUP_IOCTL_FAST_DO_FIRST) < (unsigned int)32
603	&& pSession->fUnrestricted))
604	rc = supdrvIOCtlFast(uCmd - SUP_IOCTL_FAST_DO_FIRST, ulArg, &g_DevExt, pSession);
605	else
606	rc = VBoxDrvLinuxIOCtlSlow(pFilp, uCmd, ulArg, pSession);
607	lock_kernel();
608	#endif /* !HAVE_UNLOCKED_IOCTL */
609
610	#ifndef VBOX_WITHOUT_EFLAGS_AC_SET_IN_VBOXDRV
611	# if defined(VBOX_STRICT) \|\| defined(VBOX_WITH_EFLAGS_AC_SET_IN_VBOXDRV)
612	/*
613	* Before we restore AC and the rest of EFLAGS, check if the IOCtl handler code
614	* accidentially modified it or some other important flag.
615	*/
616	if (RT_UNLIKELY( (ASMGetFlags() & (X86_EFL_AC \| X86_EFL_IF \| X86_EFL_DF))
617	!= ((fSavedEfl & (X86_EFL_AC \| X86_EFL_IF \| X86_EFL_DF)) \| X86_EFL_AC) ))
618	{
619	char szTmp[48];
620	RTStrPrintf(szTmp, sizeof(szTmp), "uCmd=%#x: %#x->%#x!", _IOC_NR(uCmd), (uint32_t)fSavedEfl, (uint32_t)ASMGetFlags());
621	supdrvBadContext(&g_DevExt, "SUPDrv-linux.c", __LINE__, szTmp);
622	}
623	ASMSetFlags(fSavedEfl);
624	# else
625	clac();
626	# endif
627	#endif
628	return rc;
629	}
630
631
632	/**
633	* Device I/O Control entry point.
634	*
635	* @param pFilp Associated file pointer.
636	* @param uCmd The function specified to ioctl().
637	* @param ulArg The argument specified to ioctl().
638	* @param pSession The session instance.
639	*/
640	static int VBoxDrvLinuxIOCtlSlow(struct file *pFilp, unsigned int uCmd, unsigned long ulArg, PSUPDRVSESSION pSession)
641	{
642	int rc;
643	SUPREQHDR Hdr;
644	PSUPREQHDR pHdr;
645	uint32_t cbBuf;
646
647	Log6(("VBoxDrvLinuxIOCtl: pFilp=%p uCmd=%#x ulArg=%p pid=%d/%d\n", pFilp, uCmd, (void *)ulArg, RTProcSelf(), current->pid));
648
649	/*
650	* Read the header.
651	*/
652	if (RT_FAILURE(RTR0MemUserCopyFrom(&Hdr, ulArg, sizeof(Hdr))))
653	{
654	Log(("VBoxDrvLinuxIOCtl: copy_from_user(,%#lx,) failed; uCmd=%#x\n", ulArg, uCmd));
655	return -EFAULT;
656	}
657	if (RT_UNLIKELY((Hdr.fFlags & SUPREQHDR_FLAGS_MAGIC_MASK) != SUPREQHDR_FLAGS_MAGIC))
658	{
659	Log(("VBoxDrvLinuxIOCtl: bad header magic %#x; uCmd=%#x\n", Hdr.fFlags & SUPREQHDR_FLAGS_MAGIC_MASK, uCmd));
660	return -EINVAL;
661	}
662
663	/*
664	* Buffer the request.
665	*/
666	cbBuf = RT_MAX(Hdr.cbIn, Hdr.cbOut);
667	if (RT_UNLIKELY(cbBuf > _1M*16))
668	{
669	Log(("VBoxDrvLinuxIOCtl: too big cbBuf=%#x; uCmd=%#x\n", cbBuf, uCmd));
670	return -E2BIG;
671	}
672	if (RT_UNLIKELY(_IOC_SIZE(uCmd) ? cbBuf != _IOC_SIZE(uCmd) : Hdr.cbIn < sizeof(Hdr)))
673	{
674	Log(("VBoxDrvLinuxIOCtl: bad ioctl cbBuf=%#x _IOC_SIZE=%#x; uCmd=%#x\n", cbBuf, _IOC_SIZE(uCmd), uCmd));
675	return -EINVAL;
676	}
677	pHdr = RTMemAlloc(cbBuf);
678	if (RT_UNLIKELY(!pHdr))
679	{
680	OSDBGPRINT(("VBoxDrvLinuxIOCtl: failed to allocate buffer of %d bytes for uCmd=%#x\n", cbBuf, uCmd));
681	return -ENOMEM;
682	}
683	if (RT_FAILURE(RTR0MemUserCopyFrom(pHdr, ulArg, Hdr.cbIn)))
684	{
685	Log(("VBoxDrvLinuxIOCtl: copy_from_user(,%#lx, %#x) failed; uCmd=%#x\n", ulArg, Hdr.cbIn, uCmd));
686	RTMemFree(pHdr);
687	return -EFAULT;
688	}
689	if (Hdr.cbIn < cbBuf)
690	RT_BZERO((uint8_t *)pHdr + Hdr.cbIn, cbBuf - Hdr.cbIn);
691
692	/*
693	* Process the IOCtl.
694	*/
695	rc = supdrvIOCtl(uCmd, &g_DevExt, pSession, pHdr, cbBuf);
696
697	/*
698	* Copy ioctl data and output buffer back to user space.
699	*/
700	if (RT_LIKELY(!rc))
701	{
702	uint32_t cbOut = pHdr->cbOut;
703	if (RT_UNLIKELY(cbOut > cbBuf))
704	{
705	OSDBGPRINT(("VBoxDrvLinuxIOCtl: too much output! %#x > %#x; uCmd=%#x!\n", cbOut, cbBuf, uCmd));
706	cbOut = cbBuf;
707	}
708	if (RT_FAILURE(RTR0MemUserCopyTo(ulArg, pHdr, cbOut)))
709	{
710	/* this is really bad! */
711	OSDBGPRINT(("VBoxDrvLinuxIOCtl: copy_to_user(%#lx,,%#x); uCmd=%#x!\n", ulArg, cbOut, uCmd));
712	rc = -EFAULT;
713	}
714	}
715	else
716	{
717	Log(("VBoxDrvLinuxIOCtl: pFilp=%p uCmd=%#x ulArg=%p failed, rc=%d\n", pFilp, uCmd, (void *)ulArg, rc));
718	rc = -EINVAL;
719	}
720	RTMemFree(pHdr);
721
722	Log6(("VBoxDrvLinuxIOCtl: returns %d (pid=%d/%d)\n", rc, RTProcSelf(), current->pid));
723	return rc;
724	}
725
726
727	/**
728	* The SUPDRV IDC entry point.
729	*
730	* @returns VBox status code, see supdrvIDC.
731	* @param uReq The request code.
732	* @param pReq The request.
733	*/
734	int VBOXCALL SUPDrvLinuxIDC(uint32_t uReq, PSUPDRVIDCREQHDR pReq)
735	{
736	PSUPDRVSESSION pSession;
737
738	/*
739	* Some quick validations.
740	*/
741	if (RT_UNLIKELY(!RT_VALID_PTR(pReq)))
742	return VERR_INVALID_POINTER;
743
744	pSession = pReq->pSession;
745	if (pSession)
746	{
747	if (RT_UNLIKELY(!RT_VALID_PTR(pSession)))
748	return VERR_INVALID_PARAMETER;
749	if (RT_UNLIKELY(pSession->pDevExt != &g_DevExt))
750	return VERR_INVALID_PARAMETER;
751	}
752	else if (RT_UNLIKELY(uReq != SUPDRV_IDC_REQ_CONNECT))
753	return VERR_INVALID_PARAMETER;
754
755	/*
756	* Do the job.
757	*/
758	return supdrvIDC(uReq, &g_DevExt, pSession, pReq);
759	}
760	EXPORT_SYMBOL(SUPDrvLinuxIDC);
761
762
763	# if RTLNX_VER_MAX(5,12,0)
764	/**
765	* Used by native wrapper modules, forwarding to supdrvLdrRegisterWrappedModule
766	* with device extension prepended to the argument list.
767	*/
768	SUPR0DECL(int) SUPDrvLinuxLdrRegisterWrappedModule(PCSUPLDRWRAPPEDMODULE pWrappedModInfo,
769	const char pszLnxModName, void *phMod)
770	{
771	# if RTLNX_VER_MIN(2,6,30)
772	/* Locate the module structure for the caller so can later reference
773	and dereference it to prevent unloading while it is being used.
774
775	Before Linux v5.9 this could be done by address (__module_address()
776	or __module_text_address()), but someone (guess who) apparently on
777	a mission to make life miserable for out-of-tree modules or something,
778	decided it was only used by build-in code and unexported both of them.
779
780	I could find no init callouts getting a struct module pointer either,
781	nor any module name hint anywhere I could see. So, we're left with
782	hardcoding the module name via the compiler and pass it along to
783	SUPDrv so we can call find_module() here.
784
785	Sigh^2. */
786	AssertPtrReturn(pszLnxModName, VERR_INVALID_POINTER);
787	AssertReturn(*pszLnxModName, VERR_INVALID_NAME);
788	if (mutex_lock_interruptible(&module_mutex) == 0)
789	{
790	struct module *pLnxModule = find_module(pszLnxModName);
791	mutex_unlock(&module_mutex);
792	if (pLnxModule)
793	return supdrvLdrRegisterWrappedModule(&g_DevExt, pWrappedModInfo, pLnxModule, phMod);
794	printk("vboxdrv: find_module(%s) failed in SUPDrvLinuxLdrRegisterWrappedModule!\n", pszLnxModName);
795	return VERR_MODULE_NOT_FOUND;
796	}
797	return VERR_INTERRUPTED;
798	# else
799	printk("vboxdrv: wrapper modules are not supported on 2.6.29 and earlier. sorry.\n");
800	return VERR_NOT_SUPPORTED;
801	# endif
802	}
803	EXPORT_SYMBOL(SUPDrvLinuxLdrRegisterWrappedModule);
804	#endif /* < 5.12.0 */
805
806
807	/**
808	* Used by native wrapper modules, forwarding to supdrvLdrDeregisterWrappedModule
809	* with device extension prepended to the argument list.
810	*/
811	SUPR0DECL(int) SUPDrvLinuxLdrDeregisterWrappedModule(PCSUPLDRWRAPPEDMODULE pWrappedModInfo, void **phMod)
812	{
813	return supdrvLdrDeregisterWrappedModule(&g_DevExt, pWrappedModInfo, phMod);
814	}
815	EXPORT_SYMBOL(SUPDrvLinuxLdrDeregisterWrappedModule);
816
817
818	RTCCUINTREG VBOXCALL supdrvOSChangeCR4(RTCCUINTREG fOrMask, RTCCUINTREG fAndMask)
819	{
820	#if RTLNX_VER_MIN(5,8,0)
821	unsigned long fSavedFlags;
822	local_irq_save(fSavedFlags);
823	RTCCUINTREG const uOld = cr4_read_shadow();
824	cr4_update_irqsoff(fOrMask, ~fAndMask); /* Same as this function, only it is not returning the old value. */
825	AssertMsg(cr4_read_shadow() == ((uOld & fAndMask) \| fOrMask),
826	("fOrMask=%#RTreg fAndMask=%#RTreg uOld=%#RTreg; new cr4=%#llx\n", fOrMask, fAndMask, uOld, cr4_read_shadow()));
827	local_irq_restore(fSavedFlags);
828	#else
829	# if RTLNX_VER_MIN(3,20,0)
830	RTCCUINTREG const uOld = this_cpu_read(cpu_tlbstate.cr4);
831	# else
832	RTCCUINTREG const uOld = ASMGetCR4();
833	# endif
834	RTCCUINTREG const uNew = (uOld & fAndMask) \| fOrMask;
835	if (uNew != uOld)
836	{
837	# if RTLNX_VER_MIN(3,20,0)
838	this_cpu_write(cpu_tlbstate.cr4, uNew);
839	__write_cr4(uNew);
840	# else
841	ASMSetCR4(uNew);
842	# endif
843	}
844	#endif
845	return uOld;
846	}
847
848
849	void VBOXCALL supdrvOSCleanupSession(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession)
850	{
851	NOREF(pDevExt);
852	NOREF(pSession);
853	}
854
855
856	void VBOXCALL supdrvOSSessionHashTabInserted(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, void *pvUser)
857	{
858	NOREF(pDevExt); NOREF(pSession); NOREF(pvUser);
859	}
860
861
862	void VBOXCALL supdrvOSSessionHashTabRemoved(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, void *pvUser)
863	{
864	NOREF(pDevExt); NOREF(pSession); NOREF(pvUser);
865	}
866
867
868	/**
869	* Initializes any OS specific object creator fields.
870	*/
871	void VBOXCALL supdrvOSObjInitCreator(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession)
872	{
873	NOREF(pObj);
874	NOREF(pSession);
875	}
876
877
878	/**
879	* Checks if the session can access the object.
880	*
881	* @returns true if a decision has been made.
882	* @returns false if the default access policy should be applied.
883	*
884	* @param pObj The object in question.
885	* @param pSession The session wanting to access the object.
886	* @param pszObjName The object name, can be NULL.
887	* @param prc Where to store the result when returning true.
888	*/
889	bool VBOXCALL supdrvOSObjCanAccess(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession, const char pszObjName, int prc)
890	{
891	NOREF(pObj);
892	NOREF(pSession);
893	NOREF(pszObjName);
894	NOREF(prc);
895	return false;
896	}
897
898
899	bool VBOXCALL supdrvOSGetForcedAsyncTscMode(PSUPDRVDEVEXT pDevExt)
900	{
901	return force_async_tsc != 0;
902	}
903
904
905	bool VBOXCALL supdrvOSAreCpusOfflinedOnSuspend(void)
906	{
907	return true;
908	}
909
910
911	bool VBOXCALL supdrvOSAreTscDeltasInSync(void)
912	{
913	return false;
914	}
915
916
917	int VBOXCALL supdrvOSLdrOpen(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, const char *pszFilename)
918	{
919	NOREF(pDevExt); NOREF(pImage); NOREF(pszFilename);
920	return VERR_NOT_SUPPORTED;
921	}
922
923
924	int VBOXCALL supdrvOSLdrValidatePointer(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, void *pv,
925	const uint8_t pbImageBits, const char pszSymbol)
926	{
927	NOREF(pDevExt); NOREF(pImage); NOREF(pv); NOREF(pbImageBits); NOREF(pszSymbol);
928	return VERR_NOT_SUPPORTED;
929	}
930
931
932	int VBOXCALL supdrvOSLdrLoad(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, const uint8_t *pbImageBits, PSUPLDRLOAD pReq)
933	{
934	NOREF(pDevExt); NOREF(pImage); NOREF(pbImageBits); NOREF(pReq);
935	return VERR_NOT_SUPPORTED;
936	}
937
938
939	void VBOXCALL supdrvOSLdrUnload(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
940	{
941	NOREF(pDevExt); NOREF(pImage);
942	}
943
944
945	/** @def VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS
946	* A very crude hack for debugging using perf and dtrace.
947	*
948	* DO ABSOLUTELY NOT ENABLE IN PRODUCTION BUILDS! DEVELOPMENT ONLY!!
949	* DO ABSOLUTELY NOT ENABLE IN PRODUCTION BUILDS! DEVELOPMENT ONLY!!
950	* DO ABSOLUTELY NOT ENABLE IN PRODUCTION BUILDS! DEVELOPMENT ONLY!!
951	*
952	*/
953	#if 0 \|\| defined(DOXYGEN_RUNNING)
954	# define VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS
955	#endif
956
957	#if defined(VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS) && defined(CONFIG_MODULES_TREE_LOOKUP)
958	/** Whether g_pfnModTreeInsert and g_pfnModTreeRemove have been initialized.
959	* @remarks can still be NULL after init. */
960	static volatile bool g_fLookedForModTreeFunctions = false;
961	static void (g_pfnModTreeInsert)(struct mod_tree_node ) = NULL; /*< __mod_tree_insert /
962	static void (g_pfnModTreeRemove)(struct mod_tree_node ) = NULL; /*< __mod_tree_remove /
963	#endif
964
965
966	void VBOXCALL supdrvOSLdrNotifyOpened(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage, const char *pszFilename)
967	{
968	#ifdef VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS /* Not for production use!! Debugging only! */
969	/*
970	* This trick stops working with 4.2 when CONFIG_MODULES_TREE_LOOKUP is
971	* defined. The module lookups are done via a tree structure and we
972	* cannot get at the root of it. :-(
973	*/
974	# ifdef CONFIG_KALLSYMS
975	size_t const cchName = strlen(pImage->szName);
976	# endif
977	struct module pMyMod, pSelfMod, pTestMod, pTestModByName;
978	IPRT_LINUX_SAVE_EFL_AC();
979
980	pImage->pLnxModHack = NULL;
981
982	# ifdef CONFIG_MODULES_TREE_LOOKUP
983	/*
984	* This is pretty naive, but works for 4.2 on arch linux. I don't think we
985	* can count on finding __mod_tree_remove in all kernel builds as it's not
986	* marked noinline like __mod_tree_insert.
987	*/
988	if (!g_fLookedForModTreeFunctions)
989	{
990	unsigned long ulInsert = kallsyms_lookup_name("__mod_tree_insert");
991	unsigned long ulRemove = kallsyms_lookup_name("__mod_tree_remove");
992	if (!ulInsert \|\| !ulRemove)
993	{
994	g_fLookedForModTreeFunctions = true;
995	printk(KERN_ERR "vboxdrv: failed to locate __mod_tree_insert and __mod_tree_remove.\n");
996	IPRT_LINUX_RESTORE_EFL_AC();
997	return;
998	}
999	(unsigned long )&g_pfnModTreeInsert = ulInsert;
1000	(unsigned long )&g_pfnModTreeRemove = ulRemove;
1001	ASMCompilerBarrier();
1002	g_fLookedForModTreeFunctions = true;
1003	}
1004	else if (!g_pfnModTreeInsert \|\| !g_pfnModTreeRemove)
1005	return;
1006	#endif
1007
1008	/*
1009	* Make sure we've found our own module, otherwise we cannot access the linked list.
1010	*/
1011	mutex_lock(&module_mutex);
1012	pSelfMod = find_module("vboxdrv");
1013	mutex_unlock(&module_mutex);
1014	if (!pSelfMod)
1015	{
1016	IPRT_LINUX_RESTORE_EFL_AC();
1017	return;
1018	}
1019
1020	/*
1021	* Cook up a module structure for the image.
1022	* We allocate symbol and string tables in the allocation and the module to keep things simple.
1023	*/
1024	# ifdef CONFIG_KALLSYMS
1025	pMyMod = (struct module )RTMemAllocZ(sizeof(pMyMod)
1026	+ sizeof(Elf_Sym) * 3
1027	+ 1 + cchName * 2 + sizeof("_start") + sizeof("_end") + 4 );
1028	# else
1029	pMyMod = (struct module )RTMemAllocZ(sizeof(pMyMod));
1030	# endif
1031	if (pMyMod)
1032	{
1033	int rc = VINF_SUCCESS;
1034	# ifdef CONFIG_KALLSYMS
1035	Elf_Sym paSymbols = (Elf_Sym )(pMyMod + 1);
1036	char pchStrTab = (char )(paSymbols + 3);
1037	# endif
1038
1039	pMyMod->state = MODULE_STATE_LIVE;
1040	INIT_LIST_HEAD(&pMyMod->list); /* just in case */
1041
1042	/* Perf only matches up files with a .ko extension (maybe .ko.gz),
1043	so in order for this crap to work smoothly, we append .ko to the
1044	module name and require the user to create symbolic links in
1045	/lib/modules/`uname -r`:
1046	for i in VMMR0.r0 VBoxDDR0.r0 VBoxDD2R0.r0; do
1047	sudo ln -s /mnt/scratch/vbox/svn/trunk/out/linux.amd64/debug/bin/$i /lib/modules/`uname -r`/$i.ko;
1048	done */
1049	RTStrPrintf(pMyMod->name, sizeof(pMyMod->name), "%s", pImage->szName);
1050
1051	/* sysfs bits. */
1052	INIT_LIST_HEAD(&pMyMod->mkobj.kobj.entry); /* rest of kobj is already zeroed, hopefully never accessed... */
1053	pMyMod->mkobj.mod = pMyMod;
1054	pMyMod->mkobj.drivers_dir = NULL;
1055	pMyMod->mkobj.mp = NULL;
1056	pMyMod->mkobj.kobj_completion = NULL;
1057
1058	pMyMod->modinfo_attrs = NULL; /* hopefully not accessed after setup. */
1059	pMyMod->holders_dir = NULL; /* hopefully not accessed. */
1060	pMyMod->version = "N/A";
1061	pMyMod->srcversion = "N/A";
1062
1063	/* We export no symbols. */
1064	pMyMod->num_syms = 0;
1065	pMyMod->syms = NULL;
1066	pMyMod->crcs = NULL;
1067
1068	pMyMod->num_gpl_syms = 0;
1069	pMyMod->gpl_syms = NULL;
1070	pMyMod->gpl_crcs = NULL;
1071
1072	pMyMod->num_gpl_future_syms = 0;
1073	pMyMod->gpl_future_syms = NULL;
1074	pMyMod->gpl_future_crcs = NULL;
1075
1076	# if CONFIG_UNUSED_SYMBOLS
1077	pMyMod->num_unused_syms = 0;
1078	pMyMod->unused_syms = NULL;
1079	pMyMod->unused_crcs = NULL;
1080
1081	pMyMod->num_unused_gpl_syms = 0;
1082	pMyMod->unused_gpl_syms = NULL;
1083	pMyMod->unused_gpl_crcs = NULL;
1084	# endif
1085	/* No kernel parameters either. */
1086	pMyMod->kp = NULL;
1087	pMyMod->num_kp = 0;
1088
1089	# ifdef CONFIG_MODULE_SIG
1090	/* Pretend ok signature. */
1091	pMyMod->sig_ok = true;
1092	# endif
1093	/* No exception table. */
1094	pMyMod->num_exentries = 0;
1095	pMyMod->extable = NULL;
1096
1097	/* No init function */
1098	pMyMod->init = NULL;
1099	pMyMod->module_init = NULL;
1100	pMyMod->init_size = 0;
1101	pMyMod->init_ro_size = 0;
1102	pMyMod->init_text_size = 0;
1103
1104	/* The module address and size. It's all text. */
1105	pMyMod->module_core = pImage->pvImage;
1106	pMyMod->core_size = pImage->cbImageBits;
1107	pMyMod->core_text_size = pImage->cbImageBits;
1108	pMyMod->core_ro_size = pImage->cbImageBits;
1109
1110	#ifdef CONFIG_MODULES_TREE_LOOKUP
1111	/* Fill in the self pointers for the tree nodes. */
1112	pMyMod->mtn_core.mod = pMyMod;
1113	pMyMod->mtn_init.mod = pMyMod;
1114	#endif
1115	/* They invented the tained bit for us, didn't they? */
1116	pMyMod->taints = 1;
1117
1118	# ifdef CONFIG_GENERIC_BUGS
1119	/* No BUGs in our modules. */
1120	pMyMod->num_bugs = 0;
1121	INIT_LIST_HEAD(&pMyMod->bug_list);
1122	pMyMod->bug_table = NULL;
1123	# endif
1124
1125	# ifdef CONFIG_KALLSYMS
1126	/* The core stuff is documented as only used when loading. So just zero them. */
1127	pMyMod->core_num_syms = 0;
1128	pMyMod->core_symtab = NULL;
1129	pMyMod->core_strtab = NULL;
1130
1131	/* Construct a symbol table with start and end symbols.
1132	Note! We don't have our own symbol table at this point, image bit
1133	are not uploaded yet! */
1134	pMyMod->num_symtab = 3;
1135	pMyMod->symtab = paSymbols;
1136	pMyMod->strtab = pchStrTab;
1137	RT_ZERO(paSymbols[0]);
1138	pchStrTab[0] = '\0';
1139	paSymbols[1].st_name = 1;
1140	paSymbols[2].st_name = 2 + RTStrPrintf(&pchStrTab[paSymbols[1].st_name], cchName + sizeof("_start"),
1141	"%s_start", pImage->szName);
1142	RTStrPrintf(&pchStrTab[paSymbols[2].st_name], cchName + sizeof("_end"), "%s_end", pImage->szName);
1143	paSymbols[1].st_info = 't';
1144	paSymbols[2].st_info = 'b';
1145	paSymbols[1].st_other = 0;
1146	paSymbols[2].st_other = 0;
1147	paSymbols[1].st_shndx = 0;
1148	paSymbols[2].st_shndx = 0;
1149	paSymbols[1].st_value = (uintptr_t)pImage->pvImage;
1150	paSymbols[2].st_value = (uintptr_t)pImage->pvImage + pImage->cbImageBits - 1;
1151	paSymbols[1].st_size = pImage->cbImageBits - 1;
1152	paSymbols[2].st_size = 1;
1153	# endif
1154	/* No arguments, but seems its always non-NULL so put empty string there. */
1155	pMyMod->args = "";
1156
1157	# ifdef CONFIG_SMP
1158	/* No per CPU data. */
1159	pMyMod->percpu = NULL;
1160	pMyMod->percpu_size = 0;
1161	# endif
1162	# ifdef CONFIG_TRACEPOINTS
1163	/* No tracepoints we like to share. */
1164	pMyMod->num_tracepoints = 0;
1165	pMyMod->tracepoints_ptrs = NULL;
1166	#endif
1167	# ifdef HAVE_JUMP_LABEL
1168	/* No jump lable stuff either. */
1169	pMyMod->jump_entries = NULL;
1170	pMyMod->num_jump_entries = 0;
1171	# endif
1172	# ifdef CONFIG_TRACING
1173	pMyMod->num_trace_bprintk_fmt = 0;
1174	pMyMod->trace_bprintk_fmt_start = NULL;
1175	# endif
1176	# ifdef CONFIG_EVENT_TRACING
1177	pMyMod->trace_events = NULL;
1178	pMyMod->num_trace_events = 0;
1179	# endif
1180	# ifdef CONFIG_FTRACE_MCOUNT_RECORD
1181	pMyMod->num_ftrace_callsites = 0;
1182	pMyMod->ftrace_callsites = NULL;
1183	# endif
1184	# ifdef CONFIG_MODULE_UNLOAD
1185	/* Dependency lists, not worth sharing */
1186	INIT_LIST_HEAD(&pMyMod->source_list);
1187	INIT_LIST_HEAD(&pMyMod->target_list);
1188
1189	/* Nobody waiting and no exit function. */
1190	# if RTLNX_VER_MAX(3,13,0)
1191	pMyMod->waiter = NULL;
1192	# endif
1193	pMyMod->exit = NULL;
1194
1195	/* References, very important as we must not allow the module
1196	to be unloaded using rmmod. */
1197	# if RTLNX_VER_MIN(3,19,0)
1198	atomic_set(&pMyMod->refcnt, 42);
1199	# else
1200	pMyMod->refptr = alloc_percpu(struct module_ref);
1201	if (pMyMod->refptr)
1202	{
1203	int iCpu;
1204	for_each_possible_cpu(iCpu)
1205	{
1206	per_cpu_ptr(pMyMod->refptr, iCpu)->decs = 0;
1207	per_cpu_ptr(pMyMod->refptr, iCpu)->incs = 1;
1208	}
1209	}
1210	else
1211	rc = VERR_NO_MEMORY;
1212	# endif
1213	# endif
1214	# ifdef CONFIG_CONSTRUCTORS
1215	/* No constructors. */
1216	pMyMod->ctors = NULL;
1217	pMyMod->num_ctors = 0;
1218	# endif
1219	if (RT_SUCCESS(rc))
1220	{
1221	bool fIsModText;
1222
1223	/*
1224	* Add the module to the list.
1225	*/
1226	mutex_lock(&module_mutex);
1227	list_add_rcu(&pMyMod->list, &pSelfMod->list);
1228	pImage->pLnxModHack = pMyMod;
1229	# ifdef CONFIG_MODULES_TREE_LOOKUP
1230	g_pfnModTreeInsert(&pMyMod->mtn_core); /* __mod_tree_insert */
1231	# endif
1232	mutex_unlock(&module_mutex);
1233
1234	/*
1235	* Test it.
1236	*/
1237	mutex_lock(&module_mutex);
1238	pTestModByName = find_module(pMyMod->name);
1239	pTestMod = __module_address((uintptr_t)pImage->pvImage + pImage->cbImageBits / 4);
1240	fIsModText = __module_text_address((uintptr_t)pImage->pvImage + pImage->cbImageBits / 2);
1241	mutex_unlock(&module_mutex);
1242	if ( pTestMod == pMyMod
1243	&& pTestModByName == pMyMod
1244	&& fIsModText)
1245	printk(KERN_ERR "vboxdrv: fake module works for '%s' (%#lx to %#lx)\n",
1246	pMyMod->name, (unsigned long)paSymbols[1].st_value, (unsigned long)paSymbols[2].st_value);
1247	else
1248	printk(KERN_ERR "vboxdrv: failed to find fake module (pTestMod=%p, pTestModByName=%p, pMyMod=%p, fIsModText=%d)\n",
1249	pTestMod, pTestModByName, pMyMod, fIsModText);
1250	}
1251	else
1252	RTMemFree(pMyMod);
1253	}
1254
1255	IPRT_LINUX_RESTORE_EFL_AC();
1256	#else
1257	pImage->pLnxModHack = NULL;
1258	#endif
1259	NOREF(pDevExt); NOREF(pImage);
1260	}
1261
1262
1263	void VBOXCALL supdrvOSLdrNotifyUnloaded(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
1264	{
1265	#ifdef VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS /* Not for production use!! Debugging only! */
1266	struct module *pMyMod = pImage->pLnxModHack;
1267	pImage->pLnxModHack = NULL;
1268	if (pMyMod)
1269	{
1270	/*
1271	* Remove the fake module list entry and free it.
1272	*/
1273	IPRT_LINUX_SAVE_EFL_AC();
1274	mutex_lock(&module_mutex);
1275	list_del_rcu(&pMyMod->list);
1276	# ifdef CONFIG_MODULES_TREE_LOOKUP
1277	g_pfnModTreeRemove(&pMyMod->mtn_core);
1278	# endif
1279	synchronize_sched();
1280	mutex_unlock(&module_mutex);
1281
1282	# if RTLNX_VER_MAX(3,19,0)
1283	free_percpu(pMyMod->refptr);
1284	# endif
1285	RTMemFree(pMyMod);
1286	IPRT_LINUX_RESTORE_EFL_AC();
1287	}
1288
1289	#else
1290	Assert(pImage->pLnxModHack == NULL);
1291	#endif
1292	NOREF(pDevExt); NOREF(pImage);
1293	}
1294
1295
1296	int VBOXCALL supdrvOSLdrQuerySymbol(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage,
1297	const char pszSymbol, size_t cchSymbol, void *ppvSymbol)
1298	{
1299	#ifdef VBOX_WITH_NON_PROD_HACK_FOR_PERF_STACKS
1300	# error "implement me!"
1301	#endif
1302	RT_NOREF(pDevExt, pImage, pszSymbol, cchSymbol, ppvSymbol);
1303	return VERR_WRONG_ORDER;
1304	}
1305
1306
1307	void VBOXCALL supdrvOSLdrRetainWrapperModule(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
1308	{
1309	struct module pLnxMod = (struct module )pImage->pvWrappedNative;
1310	Assert(!pImage->fLnxWrapperRef);
1311	AssertReturnVoid(pLnxMod);
1312	pImage->fLnxWrapperRef = try_module_get(pLnxMod);
1313	RT_NOREF(pDevExt);
1314	}
1315
1316
1317	void VBOXCALL supdrvOSLdrReleaseWrapperModule(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
1318	{
1319	if (pImage->fLnxWrapperRef)
1320	{
1321	struct module pLnxMod = (struct module )pImage->pvWrappedNative;
1322	pImage->fLnxWrapperRef = false;
1323	module_put(pLnxMod);
1324	}
1325	RT_NOREF(pDevExt);
1326	}
1327
1328
1329	#ifdef SUPDRV_WITH_MSR_PROBER
1330
1331	int VBOXCALL supdrvOSMsrProberRead(uint32_t uMsr, RTCPUID idCpu, uint64_t *puValue)
1332	{
1333	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1334	uint32_t u32Low, u32High;
1335	int rc;
1336
1337	IPRT_LINUX_SAVE_EFL_AC();
1338	if (idCpu == NIL_RTCPUID)
1339	rc = rdmsr_safe(uMsr, &u32Low, &u32High);
1340	else if (RTMpIsCpuOnline(idCpu))
1341	rc = rdmsr_safe_on_cpu(idCpu, uMsr, &u32Low, &u32High);
1342	else
1343	return VERR_CPU_OFFLINE;
1344	IPRT_LINUX_RESTORE_EFL_AC();
1345	if (rc == 0)
1346	{
1347	*puValue = RT_MAKE_U64(u32Low, u32High);
1348	return VINF_SUCCESS;
1349	}
1350	return VERR_ACCESS_DENIED;
1351	# else
1352	return VERR_NOT_SUPPORTED;
1353	# endif
1354	}
1355
1356
1357	int VBOXCALL supdrvOSMsrProberWrite(uint32_t uMsr, RTCPUID idCpu, uint64_t uValue)
1358	{
1359	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1360	int rc;
1361
1362	IPRT_LINUX_SAVE_EFL_AC();
1363	if (idCpu == NIL_RTCPUID)
1364	rc = wrmsr_safe(uMsr, RT_LODWORD(uValue), RT_HIDWORD(uValue));
1365	else if (RTMpIsCpuOnline(idCpu))
1366	rc = wrmsr_safe_on_cpu(idCpu, uMsr, RT_LODWORD(uValue), RT_HIDWORD(uValue));
1367	else
1368	return VERR_CPU_OFFLINE;
1369	IPRT_LINUX_RESTORE_EFL_AC();
1370
1371	if (rc == 0)
1372	return VINF_SUCCESS;
1373	return VERR_ACCESS_DENIED;
1374	# else
1375	return VERR_NOT_SUPPORTED;
1376	# endif
1377	}
1378
1379	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1380	/**
1381	* Worker for supdrvOSMsrProberModify.
1382	*/
1383	static DECLCALLBACK(void) supdrvLnxMsrProberModifyOnCpu(RTCPUID idCpu, void pvUser1, void pvUser2)
1384	{
1385	PSUPMSRPROBER pReq = (PSUPMSRPROBER)pvUser1;
1386	register uint32_t uMsr = pReq->u.In.uMsr;
1387	bool const fFaster = pReq->u.In.enmOp == SUPMSRPROBEROP_MODIFY_FASTER;
1388	uint64_t uBefore;
1389	uint64_t uWritten;
1390	uint64_t uAfter;
1391	int rcBefore, rcWrite, rcAfter, rcRestore;
1392	RTCCUINTREG fOldFlags;
1393
1394	/* Initialize result variables. */
1395	uBefore = uWritten = uAfter = 0;
1396	rcWrite = rcAfter = rcRestore = -EIO;
1397
1398	/*
1399	* Do the job.
1400	*/
1401	fOldFlags = ASMIntDisableFlags();
1402	ASMCompilerBarrier(); /* paranoia */
1403	if (!fFaster)
1404	ASMWriteBackAndInvalidateCaches();
1405
1406	rcBefore = rdmsrl_safe(uMsr, &uBefore);
1407	if (rcBefore >= 0)
1408	{
1409	register uint64_t uRestore = uBefore;
1410	uWritten = uRestore;
1411	uWritten &= pReq->u.In.uArgs.Modify.fAndMask;
1412	uWritten \|= pReq->u.In.uArgs.Modify.fOrMask;
1413
1414	rcWrite = wrmsr_safe(uMsr, RT_LODWORD(uWritten), RT_HIDWORD(uWritten));
1415	rcAfter = rdmsrl_safe(uMsr, &uAfter);
1416	rcRestore = wrmsr_safe(uMsr, RT_LODWORD(uRestore), RT_HIDWORD(uRestore));
1417
1418	if (!fFaster)
1419	{
1420	ASMWriteBackAndInvalidateCaches();
1421	ASMReloadCR3();
1422	ASMNopPause();
1423	}
1424	}
1425
1426	ASMCompilerBarrier(); /* paranoia */
1427	ASMSetFlags(fOldFlags);
1428
1429	/*
1430	* Write out the results.
1431	*/
1432	pReq->u.Out.uResults.Modify.uBefore = uBefore;
1433	pReq->u.Out.uResults.Modify.uWritten = uWritten;
1434	pReq->u.Out.uResults.Modify.uAfter = uAfter;
1435	pReq->u.Out.uResults.Modify.fBeforeGp = rcBefore != 0;
1436	pReq->u.Out.uResults.Modify.fModifyGp = rcWrite != 0;
1437	pReq->u.Out.uResults.Modify.fAfterGp = rcAfter != 0;
1438	pReq->u.Out.uResults.Modify.fRestoreGp = rcRestore != 0;
1439	RT_ZERO(pReq->u.Out.uResults.Modify.afReserved);
1440	}
1441	# endif
1442
1443
1444	int VBOXCALL supdrvOSMsrProberModify(RTCPUID idCpu, PSUPMSRPROBER pReq)
1445	{
1446	# ifdef SUPDRV_LINUX_HAS_SAFE_MSR_API
1447	if (idCpu == NIL_RTCPUID)
1448	{
1449	supdrvLnxMsrProberModifyOnCpu(idCpu, pReq, NULL);
1450	return VINF_SUCCESS;
1451	}
1452	return RTMpOnSpecific(idCpu, supdrvLnxMsrProberModifyOnCpu, pReq, NULL);
1453	# else
1454	return VERR_NOT_SUPPORTED;
1455	# endif
1456	}
1457
1458	#endif /* SUPDRV_WITH_MSR_PROBER */
1459
1460
1461	/**
1462	* Converts a supdrv error code to an linux error code.
1463	*
1464	* @returns corresponding linux error code.
1465	* @param rc IPRT status code.
1466	*/
1467	static int VBoxDrvLinuxErr2LinuxErr(int rc)
1468	{
1469	switch (rc)
1470	{
1471	case VINF_SUCCESS: return 0;
1472	case VERR_GENERAL_FAILURE: return -EACCES;
1473	case VERR_INVALID_PARAMETER: return -EINVAL;
1474	case VERR_INVALID_MAGIC: return -EILSEQ;
1475	case VERR_INVALID_HANDLE: return -ENXIO;
1476	case VERR_INVALID_POINTER: return -EFAULT;
1477	case VERR_LOCK_FAILED: return -ENOLCK;
1478	case VERR_ALREADY_LOADED: return -EEXIST;
1479	case VERR_PERMISSION_DENIED: return -EPERM;
1480	case VERR_VERSION_MISMATCH: return -ENOSYS;
1481	case VERR_IDT_FAILED: return -1000;
1482	}
1483
1484	return -EPERM;
1485	}
1486
1487
1488	SUPR0DECL(int) SUPR0HCPhysToVirt(RTHCPHYS HCPhys, void **ppv)
1489	{
1490	AssertReturn(!(HCPhys & PAGE_OFFSET_MASK), VERR_INVALID_POINTER);
1491	AssertReturn(HCPhys != NIL_RTHCPHYS, VERR_INVALID_POINTER);
1492	/* Would've like to use valid_phys_addr_range for this test, but it isn't exported. */
1493	AssertReturn((HCPhys \| PAGE_OFFSET_MASK) < __pa(high_memory), VERR_INVALID_POINTER);
1494	*ppv = phys_to_virt(HCPhys);
1495	return VINF_SUCCESS;
1496	}
1497	SUPR0_EXPORT_SYMBOL(SUPR0HCPhysToVirt);
1498
1499
1500	RTDECL(int) SUPR0Printf(const char *pszFormat, ...)
1501	{
1502	va_list va;
1503	char szMsg[512];
1504	IPRT_LINUX_SAVE_EFL_AC();
1505
1506	va_start(va, pszFormat);
1507	RTStrPrintfV(szMsg, sizeof(szMsg) - 1, pszFormat, va);
1508	va_end(va);
1509	szMsg[sizeof(szMsg) - 1] = '\0';
1510
1511	printk("%s", szMsg);
1512
1513	IPRT_LINUX_RESTORE_EFL_AC();
1514	return 0;
1515	}
1516	SUPR0_EXPORT_SYMBOL(SUPR0Printf);
1517
1518
1519	SUPR0DECL(uint32_t) SUPR0GetKernelFeatures(void)
1520	{
1521	uint32_t fFlags = 0;
1522	#ifdef CONFIG_PAX_KERNEXEC
1523	fFlags \|= SUPKERNELFEATURES_GDT_READ_ONLY;
1524	#endif
1525	#if RTLNX_VER_MIN(4,12,0)
1526	fFlags \|= SUPKERNELFEATURES_GDT_NEED_WRITABLE;
1527	#endif
1528	#if defined(VBOX_STRICT) \|\| defined(VBOX_WITH_EFLAGS_AC_SET_IN_VBOXDRV)
1529	fFlags \|= SUPKERNELFEATURES_SMAP;
1530	#elif defined(CONFIG_X86_SMAP)
1531	if (ASMGetCR4() & X86_CR4_SMAP)
1532	fFlags \|= SUPKERNELFEATURES_SMAP;
1533	#endif
1534	return fFlags;
1535	}
1536	SUPR0_EXPORT_SYMBOL(SUPR0GetKernelFeatures);
1537
1538
1539	int VBOXCALL supdrvOSGetCurrentGdtRw(RTHCUINTPTR *pGdtRw)
1540	{
1541	#if RTLNX_VER_MIN(4,12,0)
1542	*pGdtRw = (RTHCUINTPTR)get_current_gdt_rw();
1543	return VINF_SUCCESS;
1544	#else
1545	return VERR_NOT_IMPLEMENTED;
1546	#endif
1547	}
1548
1549
1550	module_init(VBoxDrvLinuxInit);
1551	module_exit(VBoxDrvLinuxUnload);
1552
1553	MODULE_AUTHOR(VBOX_VENDOR);
1554	MODULE_DESCRIPTION(VBOX_PRODUCT " Support Driver");
1555	MODULE_LICENSE("GPL");
1556	#ifdef MODULE_VERSION
1557	MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV) " (" RT_XSTR(SUPDRV_IOC_VERSION) ")");
1558	#endif
1559
1560	module_param(force_async_tsc, int, 0444);
1561	MODULE_PARM_DESC(force_async_tsc, "force the asynchronous TSC mode");
1562

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

source: vbox/trunk/src/VBox/HostDrivers/Support/linux/SUPDrv-linux.c@ 90780

Download in other formats: