xref: /qemu/linux-headers/linux/userfaultfd.h (revision dd873966)
1*dd873966SEric 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)
223a5eb5b4SPaolo Bonzini #define UFFD_API_FEATURES (UFFD_FEATURE_EVENT_FORK |		\
233a5eb5b4SPaolo Bonzini 			   UFFD_FEATURE_EVENT_REMAP |		\
2474c98e20SCornelia Huck 			   UFFD_FEATURE_EVENT_REMOVE |	\
2574c98e20SCornelia Huck 			   UFFD_FEATURE_EVENT_UNMAP |		\
263a5eb5b4SPaolo Bonzini 			   UFFD_FEATURE_MISSING_HUGETLBFS |	\
27d4083f50SAlexey Perevalov 			   UFFD_FEATURE_MISSING_SHMEM |		\
28d4083f50SAlexey Perevalov 			   UFFD_FEATURE_SIGBUS |		\
29d4083f50SAlexey Perevalov 			   UFFD_FEATURE_THREAD_ID)
30c1fcf220SDr. David Alan Gilbert #define UFFD_API_IOCTLS				\
31c1fcf220SDr. David Alan Gilbert 	((__u64)1 << _UFFDIO_REGISTER |		\
32c1fcf220SDr. David Alan Gilbert 	 (__u64)1 << _UFFDIO_UNREGISTER |	\
33c1fcf220SDr. David Alan Gilbert 	 (__u64)1 << _UFFDIO_API)
34c1fcf220SDr. David Alan Gilbert #define UFFD_API_RANGE_IOCTLS			\
35c1fcf220SDr. David Alan Gilbert 	((__u64)1 << _UFFDIO_WAKE |		\
36c1fcf220SDr. David Alan Gilbert 	 (__u64)1 << _UFFDIO_COPY |		\
37c1fcf220SDr. David Alan Gilbert 	 (__u64)1 << _UFFDIO_ZEROPAGE)
383a5eb5b4SPaolo Bonzini #define UFFD_API_RANGE_IOCTLS_BASIC		\
393a5eb5b4SPaolo Bonzini 	((__u64)1 << _UFFDIO_WAKE |		\
403a5eb5b4SPaolo Bonzini 	 (__u64)1 << _UFFDIO_COPY)
41c1fcf220SDr. David Alan Gilbert 
42c1fcf220SDr. David Alan Gilbert /*
43c1fcf220SDr. David Alan Gilbert  * Valid ioctl command number range with this API is from 0x00 to
44c1fcf220SDr. David Alan Gilbert  * 0x3F.  UFFDIO_API is the fixed number, everything else can be
45c1fcf220SDr. David Alan Gilbert  * changed by implementing a different UFFD_API. If sticking to the
46c1fcf220SDr. David Alan Gilbert  * same UFFD_API more ioctl can be added and userland will be aware of
47c1fcf220SDr. David Alan Gilbert  * which ioctl the running kernel implements through the ioctl command
48c1fcf220SDr. David Alan Gilbert  * bitmask written by the UFFDIO_API.
49c1fcf220SDr. David Alan Gilbert  */
50c1fcf220SDr. David Alan Gilbert #define _UFFDIO_REGISTER		(0x00)
51c1fcf220SDr. David Alan Gilbert #define _UFFDIO_UNREGISTER		(0x01)
52c1fcf220SDr. David Alan Gilbert #define _UFFDIO_WAKE			(0x02)
53c1fcf220SDr. David Alan Gilbert #define _UFFDIO_COPY			(0x03)
54c1fcf220SDr. David Alan Gilbert #define _UFFDIO_ZEROPAGE		(0x04)
55c1fcf220SDr. David Alan Gilbert #define _UFFDIO_API			(0x3F)
56c1fcf220SDr. David Alan Gilbert 
57c1fcf220SDr. David Alan Gilbert /* userfaultfd ioctl ids */
58c1fcf220SDr. David Alan Gilbert #define UFFDIO 0xAA
59c1fcf220SDr. David Alan Gilbert #define UFFDIO_API		_IOWR(UFFDIO, _UFFDIO_API,	\
60c1fcf220SDr. David Alan Gilbert 				      struct uffdio_api)
61c1fcf220SDr. David Alan Gilbert #define UFFDIO_REGISTER		_IOWR(UFFDIO, _UFFDIO_REGISTER, \
62c1fcf220SDr. David Alan Gilbert 				      struct uffdio_register)
63c1fcf220SDr. David Alan Gilbert #define UFFDIO_UNREGISTER	_IOR(UFFDIO, _UFFDIO_UNREGISTER,	\
64c1fcf220SDr. David Alan Gilbert 				     struct uffdio_range)
65c1fcf220SDr. David Alan Gilbert #define UFFDIO_WAKE		_IOR(UFFDIO, _UFFDIO_WAKE,	\
66c1fcf220SDr. David Alan Gilbert 				     struct uffdio_range)
67c1fcf220SDr. David Alan Gilbert #define UFFDIO_COPY		_IOWR(UFFDIO, _UFFDIO_COPY,	\
68c1fcf220SDr. David Alan Gilbert 				      struct uffdio_copy)
69c1fcf220SDr. David Alan Gilbert #define UFFDIO_ZEROPAGE		_IOWR(UFFDIO, _UFFDIO_ZEROPAGE,	\
70c1fcf220SDr. David Alan Gilbert 				      struct uffdio_zeropage)
71c1fcf220SDr. David Alan Gilbert 
72c1fcf220SDr. David Alan Gilbert /* read() structure */
73c1fcf220SDr. David Alan Gilbert struct uffd_msg {
74c1fcf220SDr. David Alan Gilbert 	__u8	event;
75c1fcf220SDr. David Alan Gilbert 
76c1fcf220SDr. David Alan Gilbert 	__u8	reserved1;
77c1fcf220SDr. David Alan Gilbert 	__u16	reserved2;
78c1fcf220SDr. David Alan Gilbert 	__u32	reserved3;
79c1fcf220SDr. David Alan Gilbert 
80c1fcf220SDr. David Alan Gilbert 	union {
81c1fcf220SDr. David Alan Gilbert 		struct {
82c1fcf220SDr. David Alan Gilbert 			__u64	flags;
83c1fcf220SDr. David Alan Gilbert 			__u64	address;
84d4083f50SAlexey Perevalov 			union {
85d4083f50SAlexey Perevalov 				__u32 ptid;
86d4083f50SAlexey Perevalov 			} feat;
87c1fcf220SDr. David Alan Gilbert 		} pagefault;
88c1fcf220SDr. David Alan Gilbert 
89c1fcf220SDr. David Alan Gilbert 		struct {
903a5eb5b4SPaolo Bonzini 			__u32	ufd;
913a5eb5b4SPaolo Bonzini 		} fork;
923a5eb5b4SPaolo Bonzini 
933a5eb5b4SPaolo Bonzini 		struct {
943a5eb5b4SPaolo Bonzini 			__u64	from;
953a5eb5b4SPaolo Bonzini 			__u64	to;
963a5eb5b4SPaolo Bonzini 			__u64	len;
973a5eb5b4SPaolo Bonzini 		} remap;
983a5eb5b4SPaolo Bonzini 
993a5eb5b4SPaolo Bonzini 		struct {
1003a5eb5b4SPaolo Bonzini 			__u64	start;
1013a5eb5b4SPaolo Bonzini 			__u64	end;
10274c98e20SCornelia Huck 		} remove;
1033a5eb5b4SPaolo Bonzini 
1043a5eb5b4SPaolo Bonzini 		struct {
105c1fcf220SDr. David Alan Gilbert 			/* unused reserved fields */
106c1fcf220SDr. David Alan Gilbert 			__u64	reserved1;
107c1fcf220SDr. David Alan Gilbert 			__u64	reserved2;
108c1fcf220SDr. David Alan Gilbert 			__u64	reserved3;
109c1fcf220SDr. David Alan Gilbert 		} reserved;
110c1fcf220SDr. David Alan Gilbert 	} arg;
111b89485a5SPaolo Bonzini } __attribute__((packed));
112c1fcf220SDr. David Alan Gilbert 
113c1fcf220SDr. David Alan Gilbert /*
114c1fcf220SDr. David Alan Gilbert  * Start at 0x12 and not at 0 to be more strict against bugs.
115c1fcf220SDr. David Alan Gilbert  */
116c1fcf220SDr. David Alan Gilbert #define UFFD_EVENT_PAGEFAULT	0x12
117c1fcf220SDr. David Alan Gilbert #define UFFD_EVENT_FORK		0x13
1183a5eb5b4SPaolo Bonzini #define UFFD_EVENT_REMAP	0x14
11974c98e20SCornelia Huck #define UFFD_EVENT_REMOVE	0x15
12074c98e20SCornelia Huck #define UFFD_EVENT_UNMAP	0x16
121c1fcf220SDr. David Alan Gilbert 
122c1fcf220SDr. David Alan Gilbert /* flags for UFFD_EVENT_PAGEFAULT */
123c1fcf220SDr. David Alan Gilbert #define UFFD_PAGEFAULT_FLAG_WRITE	(1<<0)	/* If this was a write fault */
124c1fcf220SDr. David Alan Gilbert #define UFFD_PAGEFAULT_FLAG_WP		(1<<1)	/* If reason is VM_UFFD_WP */
125c1fcf220SDr. David Alan Gilbert 
126c1fcf220SDr. David Alan Gilbert struct uffdio_api {
127c1fcf220SDr. David Alan Gilbert 	/* userland asks for an API number and the features to enable */
128c1fcf220SDr. David Alan Gilbert 	__u64 api;
129c1fcf220SDr. David Alan Gilbert 	/*
130c1fcf220SDr. David Alan Gilbert 	 * Kernel answers below with the all available features for
131c1fcf220SDr. David Alan Gilbert 	 * the API, this notifies userland of which events and/or
132c1fcf220SDr. David Alan Gilbert 	 * which flags for each event are enabled in the current
133c1fcf220SDr. David Alan Gilbert 	 * kernel.
134c1fcf220SDr. David Alan Gilbert 	 *
135c1fcf220SDr. David Alan Gilbert 	 * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE
136c1fcf220SDr. David Alan Gilbert 	 * are to be considered implicitly always enabled in all kernels as
137c1fcf220SDr. David Alan Gilbert 	 * long as the uffdio_api.api requested matches UFFD_API.
1383a5eb5b4SPaolo Bonzini 	 *
1393a5eb5b4SPaolo Bonzini 	 * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER
1403a5eb5b4SPaolo Bonzini 	 * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on
1413a5eb5b4SPaolo Bonzini 	 * hugetlbfs virtual memory ranges. Adding or not adding
1423a5eb5b4SPaolo Bonzini 	 * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has
1433a5eb5b4SPaolo Bonzini 	 * no real functional effect after UFFDIO_API returns, but
1443a5eb5b4SPaolo Bonzini 	 * it's only useful for an initial feature set probe at
1453a5eb5b4SPaolo Bonzini 	 * UFFDIO_API time. There are two ways to use it:
1463a5eb5b4SPaolo Bonzini 	 *
1473a5eb5b4SPaolo Bonzini 	 * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the
1483a5eb5b4SPaolo Bonzini 	 *    uffdio_api.features before calling UFFDIO_API, an error
1493a5eb5b4SPaolo Bonzini 	 *    will be returned by UFFDIO_API on a kernel without
1503a5eb5b4SPaolo Bonzini 	 *    hugetlbfs missing support
1513a5eb5b4SPaolo Bonzini 	 *
1523a5eb5b4SPaolo Bonzini 	 * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in
1533a5eb5b4SPaolo Bonzini 	 *    uffdio_api.features and instead it will be set by the
1543a5eb5b4SPaolo Bonzini 	 *    kernel in the uffdio_api.features if the kernel supports
1553a5eb5b4SPaolo Bonzini 	 *    it, so userland can later check if the feature flag is
1563a5eb5b4SPaolo Bonzini 	 *    present in uffdio_api.features after UFFDIO_API
1573a5eb5b4SPaolo Bonzini 	 *    succeeded.
1583a5eb5b4SPaolo Bonzini 	 *
1593a5eb5b4SPaolo Bonzini 	 * UFFD_FEATURE_MISSING_SHMEM works the same as
1603a5eb5b4SPaolo Bonzini 	 * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem
1613a5eb5b4SPaolo Bonzini 	 * (i.e. tmpfs and other shmem based APIs).
162d4083f50SAlexey Perevalov 	 *
163d4083f50SAlexey Perevalov 	 * UFFD_FEATURE_SIGBUS feature means no page-fault
164d4083f50SAlexey Perevalov 	 * (UFFD_EVENT_PAGEFAULT) event will be delivered, instead
165d4083f50SAlexey Perevalov 	 * a SIGBUS signal will be sent to the faulting process.
166d4083f50SAlexey Perevalov 	 *
167d4083f50SAlexey Perevalov 	 * UFFD_FEATURE_THREAD_ID pid of the page faulted task_struct will
168d4083f50SAlexey Perevalov 	 * be returned, if feature is not requested 0 will be returned.
169c1fcf220SDr. David Alan Gilbert 	 */
170c1fcf220SDr. David Alan Gilbert #define UFFD_FEATURE_PAGEFAULT_FLAG_WP		(1<<0)
171c1fcf220SDr. David Alan Gilbert #define UFFD_FEATURE_EVENT_FORK			(1<<1)
1723a5eb5b4SPaolo Bonzini #define UFFD_FEATURE_EVENT_REMAP		(1<<2)
17374c98e20SCornelia Huck #define UFFD_FEATURE_EVENT_REMOVE		(1<<3)
1743a5eb5b4SPaolo Bonzini #define UFFD_FEATURE_MISSING_HUGETLBFS		(1<<4)
1753a5eb5b4SPaolo Bonzini #define UFFD_FEATURE_MISSING_SHMEM		(1<<5)
17674c98e20SCornelia Huck #define UFFD_FEATURE_EVENT_UNMAP		(1<<6)
177d4083f50SAlexey Perevalov #define UFFD_FEATURE_SIGBUS			(1<<7)
178d4083f50SAlexey Perevalov #define UFFD_FEATURE_THREAD_ID			(1<<8)
179c1fcf220SDr. David Alan Gilbert 	__u64 features;
180c1fcf220SDr. David Alan Gilbert 
181c1fcf220SDr. David Alan Gilbert 	__u64 ioctls;
182c1fcf220SDr. David Alan Gilbert };
183c1fcf220SDr. David Alan Gilbert 
184c1fcf220SDr. David Alan Gilbert struct uffdio_range {
185c1fcf220SDr. David Alan Gilbert 	__u64 start;
186c1fcf220SDr. David Alan Gilbert 	__u64 len;
187c1fcf220SDr. David Alan Gilbert };
188c1fcf220SDr. David Alan Gilbert 
189c1fcf220SDr. David Alan Gilbert struct uffdio_register {
190c1fcf220SDr. David Alan Gilbert 	struct uffdio_range range;
191c1fcf220SDr. David Alan Gilbert #define UFFDIO_REGISTER_MODE_MISSING	((__u64)1<<0)
192c1fcf220SDr. David Alan Gilbert #define UFFDIO_REGISTER_MODE_WP		((__u64)1<<1)
193c1fcf220SDr. David Alan Gilbert 	__u64 mode;
194c1fcf220SDr. David Alan Gilbert 
195c1fcf220SDr. David Alan Gilbert 	/*
196c1fcf220SDr. David Alan Gilbert 	 * kernel answers which ioctl commands are available for the
197c1fcf220SDr. David Alan Gilbert 	 * range, keep at the end as the last 8 bytes aren't read.
198c1fcf220SDr. David Alan Gilbert 	 */
199c1fcf220SDr. David Alan Gilbert 	__u64 ioctls;
200c1fcf220SDr. David Alan Gilbert };
201c1fcf220SDr. David Alan Gilbert 
202c1fcf220SDr. David Alan Gilbert struct uffdio_copy {
203c1fcf220SDr. David Alan Gilbert 	__u64 dst;
204c1fcf220SDr. David Alan Gilbert 	__u64 src;
205c1fcf220SDr. David Alan Gilbert 	__u64 len;
206c1fcf220SDr. David Alan Gilbert 	/*
207c1fcf220SDr. David Alan Gilbert 	 * There will be a wrprotection flag later that allows to map
208c1fcf220SDr. David Alan Gilbert 	 * pages wrprotected on the fly. And such a flag will be
209c1fcf220SDr. David Alan Gilbert 	 * available if the wrprotection ioctl are implemented for the
210c1fcf220SDr. David Alan Gilbert 	 * range according to the uffdio_register.ioctls.
211c1fcf220SDr. David Alan Gilbert 	 */
212c1fcf220SDr. David Alan Gilbert #define UFFDIO_COPY_MODE_DONTWAKE		((__u64)1<<0)
213c1fcf220SDr. David Alan Gilbert 	__u64 mode;
214c1fcf220SDr. David Alan Gilbert 
215c1fcf220SDr. David Alan Gilbert 	/*
216c1fcf220SDr. David Alan Gilbert 	 * "copy" is written by the ioctl and must be at the end: the
217c1fcf220SDr. David Alan Gilbert 	 * copy_from_user will not read the last 8 bytes.
218c1fcf220SDr. David Alan Gilbert 	 */
219c1fcf220SDr. David Alan Gilbert 	__s64 copy;
220c1fcf220SDr. David Alan Gilbert };
221c1fcf220SDr. David Alan Gilbert 
222c1fcf220SDr. David Alan Gilbert struct uffdio_zeropage {
223c1fcf220SDr. David Alan Gilbert 	struct uffdio_range range;
224c1fcf220SDr. David Alan Gilbert #define UFFDIO_ZEROPAGE_MODE_DONTWAKE		((__u64)1<<0)
225c1fcf220SDr. David Alan Gilbert 	__u64 mode;
226c1fcf220SDr. David Alan Gilbert 
227c1fcf220SDr. David Alan Gilbert 	/*
228c1fcf220SDr. David Alan Gilbert 	 * "zeropage" is written by the ioctl and must be at the end:
229c1fcf220SDr. David Alan Gilbert 	 * the copy_from_user will not read the last 8 bytes.
230c1fcf220SDr. David Alan Gilbert 	 */
231c1fcf220SDr. David Alan Gilbert 	__s64 zeropage;
232c1fcf220SDr. David Alan Gilbert };
233c1fcf220SDr. David Alan Gilbert 
234c1fcf220SDr. David Alan Gilbert #endif /* _LINUX_USERFAULTFD_H */
235