xref: /qemu/linux-headers/linux/userfaultfd.h (revision dc6f8d45)
1dd873966SEric Auger /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
2c1fcf220SDr. David Alan Gilbert /*
3c1fcf220SDr. David Alan Gilbert  *  include/linux/userfaultfd.h
4c1fcf220SDr. David Alan Gilbert  *
5c1fcf220SDr. David Alan Gilbert  *  Copyright (C) 2007  Davide Libenzi <davidel@xmailserver.org>
6c1fcf220SDr. David Alan Gilbert  *  Copyright (C) 2015  Red Hat, Inc.
7c1fcf220SDr. David Alan Gilbert  *
8c1fcf220SDr. David Alan Gilbert  */
9c1fcf220SDr. David Alan Gilbert 
10c1fcf220SDr. David Alan Gilbert #ifndef _LINUX_USERFAULTFD_H
11c1fcf220SDr. David Alan Gilbert #define _LINUX_USERFAULTFD_H
12c1fcf220SDr. David Alan Gilbert 
13c1fcf220SDr. David Alan Gilbert #include <linux/types.h>
14c1fcf220SDr. David Alan Gilbert 
15c1fcf220SDr. David Alan Gilbert /*
163a5eb5b4SPaolo Bonzini  * If the UFFDIO_API is upgraded someday, the UFFDIO_UNREGISTER and
173a5eb5b4SPaolo Bonzini  * UFFDIO_WAKE ioctls should be defined as _IOW and not as _IOR.  In
183a5eb5b4SPaolo Bonzini  * userfaultfd.h we assumed the kernel was reading (instead _IOC_READ
193a5eb5b4SPaolo Bonzini  * means the userland is reading).
20c1fcf220SDr. David Alan Gilbert  */
213a5eb5b4SPaolo Bonzini #define UFFD_API ((__u64)0xAA)
22*dc6f8d45SCornelia Huck #define UFFD_API_FEATURES (UFFD_FEATURE_PAGEFAULT_FLAG_WP |	\
23*dc6f8d45SCornelia Huck 			   UFFD_FEATURE_EVENT_FORK |		\
243a5eb5b4SPaolo Bonzini 			   UFFD_FEATURE_EVENT_REMAP |		\
2574c98e20SCornelia Huck 			   UFFD_FEATURE_EVENT_REMOVE |	\
2674c98e20SCornelia Huck 			   UFFD_FEATURE_EVENT_UNMAP |		\
273a5eb5b4SPaolo Bonzini 			   UFFD_FEATURE_MISSING_HUGETLBFS |	\
28d4083f50SAlexey Perevalov 			   UFFD_FEATURE_MISSING_SHMEM |		\
29d4083f50SAlexey Perevalov 			   UFFD_FEATURE_SIGBUS |		\
30d4083f50SAlexey Perevalov 			   UFFD_FEATURE_THREAD_ID)
31c1fcf220SDr. David Alan Gilbert #define UFFD_API_IOCTLS				\
32c1fcf220SDr. David Alan Gilbert 	((__u64)1 << _UFFDIO_REGISTER |		\
33c1fcf220SDr. David Alan Gilbert 	 (__u64)1 << _UFFDIO_UNREGISTER |	\
34c1fcf220SDr. David Alan Gilbert 	 (__u64)1 << _UFFDIO_API)
35c1fcf220SDr. David Alan Gilbert #define UFFD_API_RANGE_IOCTLS			\
36c1fcf220SDr. David Alan Gilbert 	((__u64)1 << _UFFDIO_WAKE |		\
37c1fcf220SDr. David Alan Gilbert 	 (__u64)1 << _UFFDIO_COPY |		\
38*dc6f8d45SCornelia Huck 	 (__u64)1 << _UFFDIO_ZEROPAGE |		\
39*dc6f8d45SCornelia Huck 	 (__u64)1 << _UFFDIO_WRITEPROTECT)
403a5eb5b4SPaolo Bonzini #define UFFD_API_RANGE_IOCTLS_BASIC		\
413a5eb5b4SPaolo Bonzini 	((__u64)1 << _UFFDIO_WAKE |		\
423a5eb5b4SPaolo Bonzini 	 (__u64)1 << _UFFDIO_COPY)
43c1fcf220SDr. David Alan Gilbert 
44c1fcf220SDr. David Alan Gilbert /*
45c1fcf220SDr. David Alan Gilbert  * Valid ioctl command number range with this API is from 0x00 to
46c1fcf220SDr. David Alan Gilbert  * 0x3F.  UFFDIO_API is the fixed number, everything else can be
47c1fcf220SDr. David Alan Gilbert  * changed by implementing a different UFFD_API. If sticking to the
48c1fcf220SDr. David Alan Gilbert  * same UFFD_API more ioctl can be added and userland will be aware of
49c1fcf220SDr. David Alan Gilbert  * which ioctl the running kernel implements through the ioctl command
50c1fcf220SDr. David Alan Gilbert  * bitmask written by the UFFDIO_API.
51c1fcf220SDr. David Alan Gilbert  */
52c1fcf220SDr. David Alan Gilbert #define _UFFDIO_REGISTER		(0x00)
53c1fcf220SDr. David Alan Gilbert #define _UFFDIO_UNREGISTER		(0x01)
54c1fcf220SDr. David Alan Gilbert #define _UFFDIO_WAKE			(0x02)
55c1fcf220SDr. David Alan Gilbert #define _UFFDIO_COPY			(0x03)
56c1fcf220SDr. David Alan Gilbert #define _UFFDIO_ZEROPAGE		(0x04)
57*dc6f8d45SCornelia Huck #define _UFFDIO_WRITEPROTECT		(0x06)
58c1fcf220SDr. David Alan Gilbert #define _UFFDIO_API			(0x3F)
59c1fcf220SDr. David Alan Gilbert 
60c1fcf220SDr. David Alan Gilbert /* userfaultfd ioctl ids */
61c1fcf220SDr. David Alan Gilbert #define UFFDIO 0xAA
62c1fcf220SDr. David Alan Gilbert #define UFFDIO_API		_IOWR(UFFDIO, _UFFDIO_API,	\
63c1fcf220SDr. David Alan Gilbert 				      struct uffdio_api)
64c1fcf220SDr. David Alan Gilbert #define UFFDIO_REGISTER		_IOWR(UFFDIO, _UFFDIO_REGISTER, \
65c1fcf220SDr. David Alan Gilbert 				      struct uffdio_register)
66c1fcf220SDr. David Alan Gilbert #define UFFDIO_UNREGISTER	_IOR(UFFDIO, _UFFDIO_UNREGISTER,	\
67c1fcf220SDr. David Alan Gilbert 				     struct uffdio_range)
68c1fcf220SDr. David Alan Gilbert #define UFFDIO_WAKE		_IOR(UFFDIO, _UFFDIO_WAKE,	\
69c1fcf220SDr. David Alan Gilbert 				     struct uffdio_range)
70c1fcf220SDr. David Alan Gilbert #define UFFDIO_COPY		_IOWR(UFFDIO, _UFFDIO_COPY,	\
71c1fcf220SDr. David Alan Gilbert 				      struct uffdio_copy)
72c1fcf220SDr. David Alan Gilbert #define UFFDIO_ZEROPAGE		_IOWR(UFFDIO, _UFFDIO_ZEROPAGE,	\
73c1fcf220SDr. David Alan Gilbert 				      struct uffdio_zeropage)
74*dc6f8d45SCornelia Huck #define UFFDIO_WRITEPROTECT	_IOWR(UFFDIO, _UFFDIO_WRITEPROTECT, \
75*dc6f8d45SCornelia Huck 				      struct uffdio_writeprotect)
76c1fcf220SDr. David Alan Gilbert 
77c1fcf220SDr. David Alan Gilbert /* read() structure */
78c1fcf220SDr. David Alan Gilbert struct uffd_msg {
79c1fcf220SDr. David Alan Gilbert 	__u8	event;
80c1fcf220SDr. David Alan Gilbert 
81c1fcf220SDr. David Alan Gilbert 	__u8	reserved1;
82c1fcf220SDr. David Alan Gilbert 	__u16	reserved2;
83c1fcf220SDr. David Alan Gilbert 	__u32	reserved3;
84c1fcf220SDr. David Alan Gilbert 
85c1fcf220SDr. David Alan Gilbert 	union {
86c1fcf220SDr. David Alan Gilbert 		struct {
87c1fcf220SDr. David Alan Gilbert 			__u64	flags;
88c1fcf220SDr. David Alan Gilbert 			__u64	address;
89d4083f50SAlexey Perevalov 			union {
90d4083f50SAlexey Perevalov 				__u32 ptid;
91d4083f50SAlexey Perevalov 			} feat;
92c1fcf220SDr. David Alan Gilbert 		} pagefault;
93c1fcf220SDr. David Alan Gilbert 
94c1fcf220SDr. David Alan Gilbert 		struct {
953a5eb5b4SPaolo Bonzini 			__u32	ufd;
963a5eb5b4SPaolo Bonzini 		} fork;
973a5eb5b4SPaolo Bonzini 
983a5eb5b4SPaolo Bonzini 		struct {
993a5eb5b4SPaolo Bonzini 			__u64	from;
1003a5eb5b4SPaolo Bonzini 			__u64	to;
1013a5eb5b4SPaolo Bonzini 			__u64	len;
1023a5eb5b4SPaolo Bonzini 		} remap;
1033a5eb5b4SPaolo Bonzini 
1043a5eb5b4SPaolo Bonzini 		struct {
1053a5eb5b4SPaolo Bonzini 			__u64	start;
1063a5eb5b4SPaolo Bonzini 			__u64	end;
10774c98e20SCornelia Huck 		} remove;
1083a5eb5b4SPaolo Bonzini 
1093a5eb5b4SPaolo Bonzini 		struct {
110c1fcf220SDr. David Alan Gilbert 			/* unused reserved fields */
111c1fcf220SDr. David Alan Gilbert 			__u64	reserved1;
112c1fcf220SDr. David Alan Gilbert 			__u64	reserved2;
113c1fcf220SDr. David Alan Gilbert 			__u64	reserved3;
114c1fcf220SDr. David Alan Gilbert 		} reserved;
115c1fcf220SDr. David Alan Gilbert 	} arg;
116b89485a5SPaolo Bonzini } __attribute__((packed));
117c1fcf220SDr. David Alan Gilbert 
118c1fcf220SDr. David Alan Gilbert /*
119c1fcf220SDr. David Alan Gilbert  * Start at 0x12 and not at 0 to be more strict against bugs.
120c1fcf220SDr. David Alan Gilbert  */
121c1fcf220SDr. David Alan Gilbert #define UFFD_EVENT_PAGEFAULT	0x12
122c1fcf220SDr. David Alan Gilbert #define UFFD_EVENT_FORK		0x13
1233a5eb5b4SPaolo Bonzini #define UFFD_EVENT_REMAP	0x14
12474c98e20SCornelia Huck #define UFFD_EVENT_REMOVE	0x15
12574c98e20SCornelia Huck #define UFFD_EVENT_UNMAP	0x16
126c1fcf220SDr. David Alan Gilbert 
127c1fcf220SDr. David Alan Gilbert /* flags for UFFD_EVENT_PAGEFAULT */
128c1fcf220SDr. David Alan Gilbert #define UFFD_PAGEFAULT_FLAG_WRITE	(1<<0)	/* If this was a write fault */
129c1fcf220SDr. David Alan Gilbert #define UFFD_PAGEFAULT_FLAG_WP		(1<<1)	/* If reason is VM_UFFD_WP */
130c1fcf220SDr. David Alan Gilbert 
131c1fcf220SDr. David Alan Gilbert struct uffdio_api {
132c1fcf220SDr. David Alan Gilbert 	/* userland asks for an API number and the features to enable */
133c1fcf220SDr. David Alan Gilbert 	__u64 api;
134c1fcf220SDr. David Alan Gilbert 	/*
135c1fcf220SDr. David Alan Gilbert 	 * Kernel answers below with the all available features for
136c1fcf220SDr. David Alan Gilbert 	 * the API, this notifies userland of which events and/or
137c1fcf220SDr. David Alan Gilbert 	 * which flags for each event are enabled in the current
138c1fcf220SDr. David Alan Gilbert 	 * kernel.
139c1fcf220SDr. David Alan Gilbert 	 *
140c1fcf220SDr. David Alan Gilbert 	 * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE
141c1fcf220SDr. David Alan Gilbert 	 * are to be considered implicitly always enabled in all kernels as
142c1fcf220SDr. David Alan Gilbert 	 * long as the uffdio_api.api requested matches UFFD_API.
1433a5eb5b4SPaolo Bonzini 	 *
1443a5eb5b4SPaolo Bonzini 	 * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER
1453a5eb5b4SPaolo Bonzini 	 * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on
1463a5eb5b4SPaolo Bonzini 	 * hugetlbfs virtual memory ranges. Adding or not adding
1473a5eb5b4SPaolo Bonzini 	 * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has
1483a5eb5b4SPaolo Bonzini 	 * no real functional effect after UFFDIO_API returns, but
1493a5eb5b4SPaolo Bonzini 	 * it's only useful for an initial feature set probe at
1503a5eb5b4SPaolo Bonzini 	 * UFFDIO_API time. There are two ways to use it:
1513a5eb5b4SPaolo Bonzini 	 *
1523a5eb5b4SPaolo Bonzini 	 * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the
1533a5eb5b4SPaolo Bonzini 	 *    uffdio_api.features before calling UFFDIO_API, an error
1543a5eb5b4SPaolo Bonzini 	 *    will be returned by UFFDIO_API on a kernel without
1553a5eb5b4SPaolo Bonzini 	 *    hugetlbfs missing support
1563a5eb5b4SPaolo Bonzini 	 *
1573a5eb5b4SPaolo Bonzini 	 * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in
1583a5eb5b4SPaolo Bonzini 	 *    uffdio_api.features and instead it will be set by the
1593a5eb5b4SPaolo Bonzini 	 *    kernel in the uffdio_api.features if the kernel supports
1603a5eb5b4SPaolo Bonzini 	 *    it, so userland can later check if the feature flag is
1613a5eb5b4SPaolo Bonzini 	 *    present in uffdio_api.features after UFFDIO_API
1623a5eb5b4SPaolo Bonzini 	 *    succeeded.
1633a5eb5b4SPaolo Bonzini 	 *
1643a5eb5b4SPaolo Bonzini 	 * UFFD_FEATURE_MISSING_SHMEM works the same as
1653a5eb5b4SPaolo Bonzini 	 * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem
1663a5eb5b4SPaolo Bonzini 	 * (i.e. tmpfs and other shmem based APIs).
167d4083f50SAlexey Perevalov 	 *
168d4083f50SAlexey Perevalov 	 * UFFD_FEATURE_SIGBUS feature means no page-fault
169d4083f50SAlexey Perevalov 	 * (UFFD_EVENT_PAGEFAULT) event will be delivered, instead
170d4083f50SAlexey Perevalov 	 * a SIGBUS signal will be sent to the faulting process.
171d4083f50SAlexey Perevalov 	 *
172d4083f50SAlexey Perevalov 	 * UFFD_FEATURE_THREAD_ID pid of the page faulted task_struct will
173d4083f50SAlexey Perevalov 	 * be returned, if feature is not requested 0 will be returned.
174c1fcf220SDr. David Alan Gilbert 	 */
175c1fcf220SDr. David Alan Gilbert #define UFFD_FEATURE_PAGEFAULT_FLAG_WP		(1<<0)
176c1fcf220SDr. David Alan Gilbert #define UFFD_FEATURE_EVENT_FORK			(1<<1)
1773a5eb5b4SPaolo Bonzini #define UFFD_FEATURE_EVENT_REMAP		(1<<2)
17874c98e20SCornelia Huck #define UFFD_FEATURE_EVENT_REMOVE		(1<<3)
1793a5eb5b4SPaolo Bonzini #define UFFD_FEATURE_MISSING_HUGETLBFS		(1<<4)
1803a5eb5b4SPaolo Bonzini #define UFFD_FEATURE_MISSING_SHMEM		(1<<5)
18174c98e20SCornelia Huck #define UFFD_FEATURE_EVENT_UNMAP		(1<<6)
182d4083f50SAlexey Perevalov #define UFFD_FEATURE_SIGBUS			(1<<7)
183d4083f50SAlexey Perevalov #define UFFD_FEATURE_THREAD_ID			(1<<8)
184c1fcf220SDr. David Alan Gilbert 	__u64 features;
185c1fcf220SDr. David Alan Gilbert 
186c1fcf220SDr. David Alan Gilbert 	__u64 ioctls;
187c1fcf220SDr. David Alan Gilbert };
188c1fcf220SDr. David Alan Gilbert 
189c1fcf220SDr. David Alan Gilbert struct uffdio_range {
190c1fcf220SDr. David Alan Gilbert 	__u64 start;
191c1fcf220SDr. David Alan Gilbert 	__u64 len;
192c1fcf220SDr. David Alan Gilbert };
193c1fcf220SDr. David Alan Gilbert 
194c1fcf220SDr. David Alan Gilbert struct uffdio_register {
195c1fcf220SDr. David Alan Gilbert 	struct uffdio_range range;
196c1fcf220SDr. David Alan Gilbert #define UFFDIO_REGISTER_MODE_MISSING	((__u64)1<<0)
197c1fcf220SDr. David Alan Gilbert #define UFFDIO_REGISTER_MODE_WP		((__u64)1<<1)
198c1fcf220SDr. David Alan Gilbert 	__u64 mode;
199c1fcf220SDr. David Alan Gilbert 
200c1fcf220SDr. David Alan Gilbert 	/*
201c1fcf220SDr. David Alan Gilbert 	 * kernel answers which ioctl commands are available for the
202c1fcf220SDr. David Alan Gilbert 	 * range, keep at the end as the last 8 bytes aren't read.
203c1fcf220SDr. David Alan Gilbert 	 */
204c1fcf220SDr. David Alan Gilbert 	__u64 ioctls;
205c1fcf220SDr. David Alan Gilbert };
206c1fcf220SDr. David Alan Gilbert 
207c1fcf220SDr. David Alan Gilbert struct uffdio_copy {
208c1fcf220SDr. David Alan Gilbert 	__u64 dst;
209c1fcf220SDr. David Alan Gilbert 	__u64 src;
210c1fcf220SDr. David Alan Gilbert 	__u64 len;
211c1fcf220SDr. David Alan Gilbert #define UFFDIO_COPY_MODE_DONTWAKE		((__u64)1<<0)
212*dc6f8d45SCornelia Huck 	/*
213*dc6f8d45SCornelia Huck 	 * UFFDIO_COPY_MODE_WP will map the page write protected on
214*dc6f8d45SCornelia Huck 	 * the fly.  UFFDIO_COPY_MODE_WP is available only if the
215*dc6f8d45SCornelia Huck 	 * write protected ioctl is implemented for the range
216*dc6f8d45SCornelia Huck 	 * according to the uffdio_register.ioctls.
217*dc6f8d45SCornelia Huck 	 */
218*dc6f8d45SCornelia Huck #define UFFDIO_COPY_MODE_WP			((__u64)1<<1)
219c1fcf220SDr. David Alan Gilbert 	__u64 mode;
220c1fcf220SDr. David Alan Gilbert 
221c1fcf220SDr. David Alan Gilbert 	/*
222c1fcf220SDr. David Alan Gilbert 	 * "copy" is written by the ioctl and must be at the end: the
223c1fcf220SDr. David Alan Gilbert 	 * copy_from_user will not read the last 8 bytes.
224c1fcf220SDr. David Alan Gilbert 	 */
225c1fcf220SDr. David Alan Gilbert 	__s64 copy;
226c1fcf220SDr. David Alan Gilbert };
227c1fcf220SDr. David Alan Gilbert 
228c1fcf220SDr. David Alan Gilbert struct uffdio_zeropage {
229c1fcf220SDr. David Alan Gilbert 	struct uffdio_range range;
230c1fcf220SDr. David Alan Gilbert #define UFFDIO_ZEROPAGE_MODE_DONTWAKE		((__u64)1<<0)
231c1fcf220SDr. David Alan Gilbert 	__u64 mode;
232c1fcf220SDr. David Alan Gilbert 
233c1fcf220SDr. David Alan Gilbert 	/*
234c1fcf220SDr. David Alan Gilbert 	 * "zeropage" is written by the ioctl and must be at the end:
235c1fcf220SDr. David Alan Gilbert 	 * the copy_from_user will not read the last 8 bytes.
236c1fcf220SDr. David Alan Gilbert 	 */
237c1fcf220SDr. David Alan Gilbert 	__s64 zeropage;
238c1fcf220SDr. David Alan Gilbert };
239c1fcf220SDr. David Alan Gilbert 
240*dc6f8d45SCornelia Huck struct uffdio_writeprotect {
241*dc6f8d45SCornelia Huck 	struct uffdio_range range;
242*dc6f8d45SCornelia Huck /*
243*dc6f8d45SCornelia Huck  * UFFDIO_WRITEPROTECT_MODE_WP: set the flag to write protect a range,
244*dc6f8d45SCornelia Huck  * unset the flag to undo protection of a range which was previously
245*dc6f8d45SCornelia Huck  * write protected.
246*dc6f8d45SCornelia Huck  *
247*dc6f8d45SCornelia Huck  * UFFDIO_WRITEPROTECT_MODE_DONTWAKE: set the flag to avoid waking up
248*dc6f8d45SCornelia Huck  * any wait thread after the operation succeeds.
249*dc6f8d45SCornelia Huck  *
250*dc6f8d45SCornelia Huck  * NOTE: Write protecting a region (WP=1) is unrelated to page faults,
251*dc6f8d45SCornelia Huck  * therefore DONTWAKE flag is meaningless with WP=1.  Removing write
252*dc6f8d45SCornelia Huck  * protection (WP=0) in response to a page fault wakes the faulting
253*dc6f8d45SCornelia Huck  * task unless DONTWAKE is set.
254*dc6f8d45SCornelia Huck  */
255*dc6f8d45SCornelia Huck #define UFFDIO_WRITEPROTECT_MODE_WP		((__u64)1<<0)
256*dc6f8d45SCornelia Huck #define UFFDIO_WRITEPROTECT_MODE_DONTWAKE	((__u64)1<<1)
257*dc6f8d45SCornelia Huck 	__u64 mode;
258*dc6f8d45SCornelia Huck };
259*dc6f8d45SCornelia Huck 
260c1fcf220SDr. David Alan Gilbert #endif /* _LINUX_USERFAULTFD_H */
261