SUPLib-linux.cpp

Last change on this file was 108725, checked in by vboxsync, 6 weeks ago
HostDrivers/Support: Make SUPR3 work on hosts where the page size is not known during compile time, bugref:10391 [build fix]
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 12.4 KB

Line
1	/* $Id: SUPLib-linux.cpp 108725 2025-03-24 18:54:47Z vboxsync $ */
2	/** @file
3	* VirtualBox Support Library - GNU/Linux specific parts.
4	*/
5
6	/*
7	* Copyright (C) 2006-2024 Oracle and/or its affiliates.
8	*
9	* This file is part of VirtualBox base platform packages, as
10	* available from https://www.virtualbox.org.
11	*
12	* This program is free software; you can redistribute it and/or
13	* modify it under the terms of the GNU General Public License
14	* as published by the Free Software Foundation, in version 3 of the
15	* License.
16	*
17	* This program is distributed in the hope that it will be useful, but
18	* WITHOUT ANY WARRANTY; without even the implied warranty of
19	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20	* General Public License for more details.
21	*
22	* You should have received a copy of the GNU General Public License
23	* along with this program; if not, see <https://www.gnu.org/licenses>.
24	*
25	* The contents of this file may alternatively be used under the terms
26	* of the Common Development and Distribution License Version 1.0
27	* (CDDL), a copy of it is provided in the "COPYING.CDDL" file included
28	* in the VirtualBox distribution, in which case the provisions of the
29	* CDDL are applicable instead of those of the GPL.
30	*
31	* You may elect to license modified versions of this file under the
32	* terms and conditions of either the GPL or the CDDL or both.
33	*
34	* SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0
35	*/
36
37
38	/*********************************************************************************************************************************
39	* Header Files *
40	*********************************************************************************************************************************/
41	#define LOG_GROUP LOG_GROUP_SUP
42	#ifdef IN_SUP_HARDENED_R3
43	# undef DEBUG /* Warning: disables RT_STRICT */
44	# undef RT_STRICT
45	# ifndef LOG_DISABLED
46	# define LOG_DISABLED
47	# endif
48	# define RTLOG_REL_DISABLED
49	# include <iprt/log.h>
50	#endif
51
52	#include <sys/fcntl.h>
53	#include <sys/ioctl.h>
54	#include <sys/mman.h>
55	#include <errno.h>
56	#include <unistd.h>
57	#include <stdlib.h>
58	#include <malloc.h>
59
60	#include <VBox/log.h>
61	#include <VBox/sup.h>
62	#include <iprt/path.h>
63	#include <iprt/assert.h>
64	#include <VBox/types.h>
65	#include <iprt/string.h>
66	#include <iprt/system.h>
67	#include <VBox/err.h>
68	#include <VBox/param.h>
69	#include "../SUPLibInternal.h"
70	#include "../SUPDrvIOC.h"
71
72
73	/*********************************************************************************************************************************
74	* Defined Constants And Macros *
75	*********************************************************************************************************************************/
76	/** System device name. */
77	#define DEVICE_NAME_SYS "/dev/vboxdrv"
78	/** User device name. */
79	#define DEVICE_NAME_USR "/dev/vboxdrvu"
80
81	/* define MADV_DONTFORK if it's missing from the system headers. */
82	#ifndef MADV_DONTFORK
83	# define MADV_DONTFORK 10
84	#endif
85
86
87
88	DECLHIDDEN(int) suplibOsInit(PSUPLIBDATA pThis, bool fPreInited, uint32_t fFlags, SUPINITOP *penmWhat, PRTERRINFO pErrInfo)
89	{
90	RT_NOREF2(penmWhat, pErrInfo);
91
92	/*
93	* Nothing to do if pre-inited.
94	*/
95	if (fPreInited)
96	return VINF_SUCCESS;
97	Assert(pThis->hDevice == (intptr_t)NIL_RTFILE);
98
99	/*
100	* Check if madvise works.
101	*/
102	uint32_t const cbPage = SUP_PAGE_SIZE;
103	void *pv = mmap(NULL, cbPage, PROT_READ \| PROT_WRITE, MAP_PRIVATE \| MAP_ANONYMOUS, -1, 0);
104	if (pv == MAP_FAILED)
105	return VERR_NO_MEMORY;
106	pThis->fSysMadviseWorks = (0 == madvise(pv, cbPage, MADV_DONTFORK));
107	munmap(pv, cbPage);
108
109	/*
110	* Driverless?
111	*/
112	if (fFlags & SUPR3INIT_F_DRIVERLESS)
113	{
114	pThis->fDriverless = true;
115	return VINF_SUCCESS;
116	}
117
118	/*
119	* Try open the device.
120	*/
121	const char *pszDeviceNm = fFlags & SUPR3INIT_F_UNRESTRICTED ? DEVICE_NAME_SYS : DEVICE_NAME_USR;
122	int hDevice = open(pszDeviceNm, O_RDWR, 0);
123	if (hDevice < 0)
124	{
125	/*
126	* Try load the device.
127	*/
128	hDevice = open(pszDeviceNm, O_RDWR, 0);
129	if (hDevice < 0)
130	{
131	int rc;
132	switch (errno)
133	{
134	case ENXIO: /* see man 2 open, ENODEV is actually a kernel bug */
135	case ENODEV: rc = VERR_VM_DRIVER_LOAD_ERROR; break;
136	case EPERM:
137	case EACCES: rc = VERR_VM_DRIVER_NOT_ACCESSIBLE; break;
138	case ENOENT: rc = VERR_VM_DRIVER_NOT_INSTALLED; break;
139	default: rc = VERR_VM_DRIVER_OPEN_ERROR; break;
140	}
141	if (fFlags & SUPR3INIT_F_DRIVERLESS_MASK)
142	{
143	LogRel(("Failed to open \"%s\", errno=%d, rc=%Rrc - Switching to driverless mode.\n", pszDeviceNm, errno, rc));
144	pThis->fDriverless = true;
145	return VINF_SUCCESS;
146	}
147	LogRel(("Failed to open \"%s\", errno=%d, rc=%Rrc\n", pszDeviceNm, errno, rc));
148	return rc;
149	}
150	}
151
152	/*
153	* Mark the file handle close on exec.
154	*/
155	if (fcntl(hDevice, F_SETFD, FD_CLOEXEC) == -1)
156	{
157	close(hDevice);
158	#ifdef IN_SUP_HARDENED_R3
159	return VERR_INTERNAL_ERROR;
160	#else
161	return RTErrConvertFromErrno(errno);
162	#endif
163	}
164
165	/*
166	* We're done.
167	*/
168	pThis->hDevice = hDevice;
169	pThis->fUnrestricted = RT_BOOL(fFlags & SUPR3INIT_F_UNRESTRICTED);
170	return VINF_SUCCESS;
171	}
172
173
174	DECLHIDDEN(int) suplibOsTerm(PSUPLIBDATA pThis)
175	{
176	/*
177	* Close the device if it's actually open.
178	*/
179	if (pThis->hDevice != (intptr_t)NIL_RTFILE)
180	{
181	if (close(pThis->hDevice))
182	AssertFailed();
183	pThis->hDevice = (intptr_t)NIL_RTFILE;
184	}
185
186	return 0;
187	}
188
189
190	#ifndef IN_SUP_HARDENED_R3
191
192	DECLHIDDEN(int) suplibOsInstall(void)
193	{
194	// nothing to do on Linux
195	return VERR_NOT_IMPLEMENTED;
196	}
197
198
199	DECLHIDDEN(int) suplibOsUninstall(void)
200	{
201	// nothing to do on Linux
202	return VERR_NOT_IMPLEMENTED;
203	}
204
205
206	DECLHIDDEN(int) suplibOsIOCtl(PSUPLIBDATA pThis, uintptr_t uFunction, void *pvReq, size_t cbReq)
207	{
208	AssertMsg(pThis->hDevice != (intptr_t)NIL_RTFILE, ("SUPLIB not initiated successfully!\n"));
209	NOREF(cbReq);
210
211	/*
212	* Issue device iocontrol.
213	*/
214	if (RT_LIKELY(ioctl(pThis->hDevice, uFunction, pvReq) >= 0))
215	return VINF_SUCCESS;
216
217	/* This is the reverse operation of the one found in SUPDrv-linux.c */
218	switch (errno)
219	{
220	case EACCES: return VERR_GENERAL_FAILURE;
221	case EINVAL: return VERR_INVALID_PARAMETER;
222	case EILSEQ: return VERR_INVALID_MAGIC;
223	case ENXIO: return VERR_INVALID_HANDLE;
224	case EFAULT: return VERR_INVALID_POINTER;
225	case ENOLCK: return VERR_LOCK_FAILED;
226	case EEXIST: return VERR_ALREADY_LOADED;
227	case EPERM: return VERR_PERMISSION_DENIED;
228	case ENOSYS: return VERR_VERSION_MISMATCH;
229	case 1000: return VERR_IDT_FAILED;
230	}
231
232	return RTErrConvertFromErrno(errno);
233	}
234
235
236	DECLHIDDEN(int) suplibOsIOCtlFast(PSUPLIBDATA pThis, uintptr_t uFunction, uintptr_t idCpu)
237	{
238	int rc = ioctl(pThis->hDevice, uFunction, idCpu);
239	if (rc == -1)
240	rc = -errno;
241	return rc;
242	}
243
244
245	DECLHIDDEN(int) suplibOsPageAlloc(PSUPLIBDATA pThis, size_t cPages, uint32_t fFlags, void **ppvPages)
246	{
247	/*
248	* If large pages are requested, try use the MAP_HUGETBL flags. This takes
249	* pages from the reserved huge page pool (see sysctl vm.nr_hugepages) and
250	* is typically not configured. Also, when the pool is exhausted we get
251	* ENOMEM back at us. So, when it fails try again w/o MAP_HUGETLB.
252	*/
253	int fMmap = MAP_PRIVATE \| MAP_ANONYMOUS;
254	#ifdef MAP_HUGETLB
255	if ((fFlags & SUP_PAGE_ALLOC_F_LARGE_PAGES) && !(cPages & 511))
256	fMmap \|= MAP_HUGETLB;
257	#endif
258
259	uint32_t const cbPage = SUP_PAGE_SIZE;
260	uint32_t const cPageShift = SUP_PAGE_SHIFT;
261
262	size_t cbMmap = cPages << cPageShift;
263	if ( !pThis->fSysMadviseWorks
264	&& (fFlags & (SUP_PAGE_ALLOC_F_FOR_LOCKING \| SUP_PAGE_ALLOC_F_LARGE_PAGES)) == SUP_PAGE_ALLOC_F_FOR_LOCKING)
265	cbMmap += cbPage * 2;
266
267	uint8_t pbPages = (uint8_t )mmap(NULL, cbMmap, PROT_READ \| PROT_WRITE, fMmap, -1, 0);
268	#ifdef MAP_HUGETLB
269	if (pbPages == MAP_FAILED && (fMmap & MAP_HUGETLB))
270	{
271	/* Try again without MAP_HUGETLB if mmap fails: */
272	fMmap &= ~MAP_HUGETLB;
273	if (!pThis->fSysMadviseWorks && (fFlags & SUP_PAGE_ALLOC_F_FOR_LOCKING))
274	cbMmap = (cPages + 2) << cPageShift;
275	pbPages = (uint8_t *)mmap(NULL, cbMmap, PROT_READ \| PROT_WRITE, fMmap, -1, 0);
276	}
277	#endif
278	if (pbPages != MAP_FAILED)
279	{
280	if ( !(fFlags & SUP_PAGE_ALLOC_F_FOR_LOCKING)
281	\|\| pThis->fSysMadviseWorks
282	#ifdef MAP_HUGETLB
283	\|\| (fMmap & MAP_HUGETLB)
284	#endif
285	)
286	{
287	/*
288	* It is not fatal if we fail here but a forked child (e.g. the ALSA sound server)
289	* could crash. Linux < 2.6.16 does not implement madvise(MADV_DONTFORK) but the
290	* kernel seems to split bigger VMAs and that is all that we want -- later we set the
291	* VM_DONTCOPY attribute in supdrvOSLockMemOne().
292	*/
293	if ( madvise(pbPages, cbMmap, MADV_DONTFORK)
294	#ifdef MAP_HUGETLB
295	&& !(fMmap & MAP_HUGETLB)
296	#endif
297	)
298	LogRel(("SUPLib: madvise %p-%p failed\n", pbPages, cbMmap));
299
300	#ifdef MADV_HUGEPAGE
301	/*
302	* Try enable transparent huge pages for the allocation if desired
303	* and we weren't able to use MAP_HUGETBL above.
304	* Note! KVM doesn't seem to benefit much from this.
305	*/
306	if ( !(fMmap & MAP_HUGETLB)
307	&& (fFlags & SUP_PAGE_ALLOC_F_LARGE_PAGES)
308	&& !(cPages & 511)) /** @todo PORTME: x86 assumption */
309	madvise(pbPages, cbMmap, MADV_HUGEPAGE);
310	#endif
311	}
312	else
313	{
314	/*
315	* madvise(MADV_DONTFORK) is not available (most probably Linux 2.4). Enclose any
316	* mmapped region by two unmapped pages to guarantee that there is exactly one VM
317	* area struct of the very same size as the mmap area.
318	*/
319	mprotect(pbPages, cbPage, PROT_NONE);
320	mprotect(pbPages + cbMmap - cbPage, cbPage, PROT_NONE);
321	pbPages += cPageShift;
322	}
323
324	/** @todo Dunno why we do this, really. It's a waste of time. Maybe it was
325	* to try make sure the pages were allocated or something before we locked them,
326	* so I qualified it with SUP_PAGE_ALLOC_F_FOR_LOCKING (unused) for now... */
327	if (fFlags & SUP_PAGE_ALLOC_F_FOR_LOCKING)
328	memset(pbPages, 0, cPages << cPageShift);
329
330	*ppvPages = pbPages;
331	return VINF_SUCCESS;
332	}
333	return VERR_NO_MEMORY;
334	}
335
336
337	DECLHIDDEN(int) suplibOsPageFree(PSUPLIBDATA pThis, void *pvPages, size_t cPages)
338	{
339	NOREF(pThis);
340	munmap(pvPages, cPages << SUP_PAGE_SHIFT);
341	return VINF_SUCCESS;
342	}
343
344
345	/**
346	* Check if the host kernel supports VT-x or not.
347	*
348	* Older Linux kernels clear the VMXE bit in the CR4 register (function
349	* tlb_flush_all()) leading to a host kernel panic.
350	*
351	* @returns VBox status code (no info).
352	* @param ppszWhy Where to return explanatory message.
353	*/
354	DECLHIDDEN(int) suplibOsQueryVTxSupported(const char **ppszWhy)
355	{
356	char szBuf[256];
357	int rc = RTSystemQueryOSInfo(RTSYSOSINFO_RELEASE, szBuf, sizeof(szBuf));
358	if (RT_SUCCESS(rc))
359	{
360	char *pszNext;
361	uint32_t uA, uB, uC;
362
363	rc = RTStrToUInt32Ex(szBuf, &pszNext, 10, &uA);
364	if ( RT_SUCCESS(rc)
365	&& *pszNext == '.')
366	{
367	/*
368	* new version number scheme starting with Linux 3.0
369	*/
370	if (uA >= 3)
371	return VINF_SUCCESS;
372	rc = RTStrToUInt32Ex(pszNext+1, &pszNext, 10, &uB);
373	if ( RT_SUCCESS(rc)
374	&& *pszNext == '.')
375	{
376	rc = RTStrToUInt32Ex(pszNext+1, &pszNext, 10, &uC);
377	if (RT_SUCCESS(rc))
378	{
379	uint32_t uLinuxVersion = (uA << 16) + (uB << 8) + uC;
380	if (uLinuxVersion >= (2 << 16) + (6 << 8) + 13)
381	return VINF_SUCCESS;
382	}
383	}
384	}
385	}
386
387	*ppszWhy = "Linux 2.6.13 or newer required!";
388	return VERR_SUPDRV_KERNEL_TOO_OLD_FOR_VTX;
389	}
390
391	#endif /* !IN_SUP_HARDENED_R3 */
392

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

source: vbox/trunk/src/VBox/HostDrivers/Support/linux/SUPLib-linux.cpp

Download in other formats: