xref: /linux-6.15/include/uapi/linux/userfaultfd.h (revision ca49ca71)
1 /*
2  *  include/linux/userfaultfd.h
3  *
4  *  Copyright (C) 2007  Davide Libenzi <[email protected]>
5  *  Copyright (C) 2015  Red Hat, Inc.
6  *
7  */
8 
9 #ifndef _LINUX_USERFAULTFD_H
10 #define _LINUX_USERFAULTFD_H
11 
12 #include <linux/types.h>
13 
14 /*
15  * If the UFFDIO_API is upgraded someday, the UFFDIO_UNREGISTER and
16  * UFFDIO_WAKE ioctls should be defined as _IOW and not as _IOR.  In
17  * userfaultfd.h we assumed the kernel was reading (instead _IOC_READ
18  * means the userland is reading).
19  */
20 #define UFFD_API ((__u64)0xAA)
21 #define UFFD_API_FEATURES (UFFD_FEATURE_EVENT_EXIT |		\
22 			   UFFD_FEATURE_EVENT_FORK |		\
23 			   UFFD_FEATURE_EVENT_REMAP |		\
24 			   UFFD_FEATURE_EVENT_REMOVE |	\
25 			   UFFD_FEATURE_EVENT_UNMAP |		\
26 			   UFFD_FEATURE_MISSING_HUGETLBFS |	\
27 			   UFFD_FEATURE_MISSING_SHMEM)
28 #define UFFD_API_IOCTLS				\
29 	((__u64)1 << _UFFDIO_REGISTER |		\
30 	 (__u64)1 << _UFFDIO_UNREGISTER |	\
31 	 (__u64)1 << _UFFDIO_API)
32 #define UFFD_API_RANGE_IOCTLS			\
33 	((__u64)1 << _UFFDIO_WAKE |		\
34 	 (__u64)1 << _UFFDIO_COPY |		\
35 	 (__u64)1 << _UFFDIO_ZEROPAGE)
36 #define UFFD_API_RANGE_IOCTLS_BASIC		\
37 	((__u64)1 << _UFFDIO_WAKE |		\
38 	 (__u64)1 << _UFFDIO_COPY)
39 
40 /*
41  * Valid ioctl command number range with this API is from 0x00 to
42  * 0x3F.  UFFDIO_API is the fixed number, everything else can be
43  * changed by implementing a different UFFD_API. If sticking to the
44  * same UFFD_API more ioctl can be added and userland will be aware of
45  * which ioctl the running kernel implements through the ioctl command
46  * bitmask written by the UFFDIO_API.
47  */
48 #define _UFFDIO_REGISTER		(0x00)
49 #define _UFFDIO_UNREGISTER		(0x01)
50 #define _UFFDIO_WAKE			(0x02)
51 #define _UFFDIO_COPY			(0x03)
52 #define _UFFDIO_ZEROPAGE		(0x04)
53 #define _UFFDIO_API			(0x3F)
54 
55 /* userfaultfd ioctl ids */
56 #define UFFDIO 0xAA
57 #define UFFDIO_API		_IOWR(UFFDIO, _UFFDIO_API,	\
58 				      struct uffdio_api)
59 #define UFFDIO_REGISTER		_IOWR(UFFDIO, _UFFDIO_REGISTER, \
60 				      struct uffdio_register)
61 #define UFFDIO_UNREGISTER	_IOR(UFFDIO, _UFFDIO_UNREGISTER,	\
62 				     struct uffdio_range)
63 #define UFFDIO_WAKE		_IOR(UFFDIO, _UFFDIO_WAKE,	\
64 				     struct uffdio_range)
65 #define UFFDIO_COPY		_IOWR(UFFDIO, _UFFDIO_COPY,	\
66 				      struct uffdio_copy)
67 #define UFFDIO_ZEROPAGE		_IOWR(UFFDIO, _UFFDIO_ZEROPAGE,	\
68 				      struct uffdio_zeropage)
69 
70 /* read() structure */
71 struct uffd_msg {
72 	__u8	event;
73 
74 	__u8	reserved1;
75 	__u16	reserved2;
76 	__u32	reserved3;
77 
78 	union {
79 		struct {
80 			__u64	flags;
81 			__u64	address;
82 		} pagefault;
83 
84 		struct {
85 			__u32	ufd;
86 		} fork;
87 
88 		struct {
89 			__u64	from;
90 			__u64	to;
91 			__u64	len;
92 		} remap;
93 
94 		struct {
95 			__u64	start;
96 			__u64	end;
97 		} remove;
98 
99 		struct {
100 			/* unused reserved fields */
101 			__u64	reserved1;
102 			__u64	reserved2;
103 			__u64	reserved3;
104 		} reserved;
105 	} arg;
106 } __packed;
107 
108 /*
109  * Start at 0x12 and not at 0 to be more strict against bugs.
110  */
111 #define UFFD_EVENT_PAGEFAULT	0x12
112 #define UFFD_EVENT_FORK		0x13
113 #define UFFD_EVENT_REMAP	0x14
114 #define UFFD_EVENT_REMOVE	0x15
115 #define UFFD_EVENT_UNMAP	0x16
116 #define UFFD_EVENT_EXIT		0x17
117 
118 /* flags for UFFD_EVENT_PAGEFAULT */
119 #define UFFD_PAGEFAULT_FLAG_WRITE	(1<<0)	/* If this was a write fault */
120 #define UFFD_PAGEFAULT_FLAG_WP		(1<<1)	/* If reason is VM_UFFD_WP */
121 
122 struct uffdio_api {
123 	/* userland asks for an API number and the features to enable */
124 	__u64 api;
125 	/*
126 	 * Kernel answers below with the all available features for
127 	 * the API, this notifies userland of which events and/or
128 	 * which flags for each event are enabled in the current
129 	 * kernel.
130 	 *
131 	 * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE
132 	 * are to be considered implicitly always enabled in all kernels as
133 	 * long as the uffdio_api.api requested matches UFFD_API.
134 	 *
135 	 * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER
136 	 * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on
137 	 * hugetlbfs virtual memory ranges. Adding or not adding
138 	 * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has
139 	 * no real functional effect after UFFDIO_API returns, but
140 	 * it's only useful for an initial feature set probe at
141 	 * UFFDIO_API time. There are two ways to use it:
142 	 *
143 	 * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the
144 	 *    uffdio_api.features before calling UFFDIO_API, an error
145 	 *    will be returned by UFFDIO_API on a kernel without
146 	 *    hugetlbfs missing support
147 	 *
148 	 * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in
149 	 *    uffdio_api.features and instead it will be set by the
150 	 *    kernel in the uffdio_api.features if the kernel supports
151 	 *    it, so userland can later check if the feature flag is
152 	 *    present in uffdio_api.features after UFFDIO_API
153 	 *    succeeded.
154 	 *
155 	 * UFFD_FEATURE_MISSING_SHMEM works the same as
156 	 * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem
157 	 * (i.e. tmpfs and other shmem based APIs).
158 	 */
159 #define UFFD_FEATURE_PAGEFAULT_FLAG_WP		(1<<0)
160 #define UFFD_FEATURE_EVENT_FORK			(1<<1)
161 #define UFFD_FEATURE_EVENT_REMAP		(1<<2)
162 #define UFFD_FEATURE_EVENT_REMOVE		(1<<3)
163 #define UFFD_FEATURE_MISSING_HUGETLBFS		(1<<4)
164 #define UFFD_FEATURE_MISSING_SHMEM		(1<<5)
165 #define UFFD_FEATURE_EVENT_UNMAP		(1<<6)
166 #define UFFD_FEATURE_EVENT_EXIT			(1<<7)
167 	__u64 features;
168 
169 	__u64 ioctls;
170 };
171 
172 struct uffdio_range {
173 	__u64 start;
174 	__u64 len;
175 };
176 
177 struct uffdio_register {
178 	struct uffdio_range range;
179 #define UFFDIO_REGISTER_MODE_MISSING	((__u64)1<<0)
180 #define UFFDIO_REGISTER_MODE_WP		((__u64)1<<1)
181 	__u64 mode;
182 
183 	/*
184 	 * kernel answers which ioctl commands are available for the
185 	 * range, keep at the end as the last 8 bytes aren't read.
186 	 */
187 	__u64 ioctls;
188 };
189 
190 struct uffdio_copy {
191 	__u64 dst;
192 	__u64 src;
193 	__u64 len;
194 	/*
195 	 * There will be a wrprotection flag later that allows to map
196 	 * pages wrprotected on the fly. And such a flag will be
197 	 * available if the wrprotection ioctl are implemented for the
198 	 * range according to the uffdio_register.ioctls.
199 	 */
200 #define UFFDIO_COPY_MODE_DONTWAKE		((__u64)1<<0)
201 	__u64 mode;
202 
203 	/*
204 	 * "copy" is written by the ioctl and must be at the end: the
205 	 * copy_from_user will not read the last 8 bytes.
206 	 */
207 	__s64 copy;
208 };
209 
210 struct uffdio_zeropage {
211 	struct uffdio_range range;
212 #define UFFDIO_ZEROPAGE_MODE_DONTWAKE		((__u64)1<<0)
213 	__u64 mode;
214 
215 	/*
216 	 * "zeropage" is written by the ioctl and must be at the end:
217 	 * the copy_from_user will not read the last 8 bytes.
218 	 */
219 	__s64 zeropage;
220 };
221 
222 #endif /* _LINUX_USERFAULTFD_H */
223