1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
2 /*
3  *  include/linux/userfaultfd.h
4  *
5  *  Copyright (C) 2007  Davide Libenzi <davidel@xmailserver.org>
6  *  Copyright (C) 2015  Red Hat, Inc.
7  *
8  */
9 
10 #ifndef _LINUX_USERFAULTFD_H
11 #define _LINUX_USERFAULTFD_H
12 
13 #include <linux/types.h>
14 
15 /*
16  * If the UFFDIO_API is upgraded someday, the UFFDIO_UNREGISTER and
17  * UFFDIO_WAKE ioctls should be defined as _IOW and not as _IOR.  In
18  * userfaultfd.h we assumed the kernel was reading (instead IOC_OUT
19  * means the userland is reading).
20  */
21 #define UFFD_API ((uint64_t)0xAA)
22 #define UFFD_API_REGISTER_MODES (UFFDIO_REGISTER_MODE_MISSING |	\
23 				 UFFDIO_REGISTER_MODE_WP |	\
24 				 UFFDIO_REGISTER_MODE_MINOR)
25 #define UFFD_API_FEATURES (UFFD_FEATURE_PAGEFAULT_FLAG_WP |	\
26 			   UFFD_FEATURE_EVENT_FORK |		\
27 			   UFFD_FEATURE_EVENT_REMAP |		\
28 			   UFFD_FEATURE_EVENT_REMOVE |		\
29 			   UFFD_FEATURE_EVENT_UNMAP |		\
30 			   UFFD_FEATURE_MISSING_HUGETLBFS |	\
31 			   UFFD_FEATURE_MISSING_SHMEM |		\
32 			   UFFD_FEATURE_SIGBUS |		\
33 			   UFFD_FEATURE_THREAD_ID |		\
34 			   UFFD_FEATURE_MINOR_HUGETLBFS)
35 #define UFFD_API_IOCTLS				\
36 	((uint64_t)1 << _UFFDIO_REGISTER |		\
37 	 (uint64_t)1 << _UFFDIO_UNREGISTER |	\
38 	 (uint64_t)1 << _UFFDIO_API)
39 #define UFFD_API_RANGE_IOCTLS			\
40 	((uint64_t)1 << _UFFDIO_WAKE |		\
41 	 (uint64_t)1 << _UFFDIO_COPY |		\
42 	 (uint64_t)1 << _UFFDIO_ZEROPAGE |		\
43 	 (uint64_t)1 << _UFFDIO_WRITEPROTECT |	\
44 	 (uint64_t)1 << _UFFDIO_CONTINUE)
45 #define UFFD_API_RANGE_IOCTLS_BASIC		\
46 	((uint64_t)1 << _UFFDIO_WAKE |		\
47 	 (uint64_t)1 << _UFFDIO_COPY |		\
48 	 (uint64_t)1 << _UFFDIO_CONTINUE)
49 
50 /*
51  * Valid ioctl command number range with this API is from 0x00 to
52  * 0x3F.  UFFDIO_API is the fixed number, everything else can be
53  * changed by implementing a different UFFD_API. If sticking to the
54  * same UFFD_API more ioctl can be added and userland will be aware of
55  * which ioctl the running kernel implements through the ioctl command
56  * bitmask written by the UFFDIO_API.
57  */
58 #define _UFFDIO_REGISTER		(0x00)
59 #define _UFFDIO_UNREGISTER		(0x01)
60 #define _UFFDIO_WAKE			(0x02)
61 #define _UFFDIO_COPY			(0x03)
62 #define _UFFDIO_ZEROPAGE		(0x04)
63 #define _UFFDIO_WRITEPROTECT		(0x06)
64 #define _UFFDIO_CONTINUE		(0x07)
65 #define _UFFDIO_API			(0x3F)
66 
67 /* userfaultfd ioctl ids */
68 #define UFFDIO 0xAA
69 #define UFFDIO_API		_IOWR(UFFDIO, _UFFDIO_API,	\
70 				      struct uffdio_api)
71 #define UFFDIO_REGISTER		_IOWR(UFFDIO, _UFFDIO_REGISTER, \
72 				      struct uffdio_register)
73 #define UFFDIO_UNREGISTER	_IOR(UFFDIO, _UFFDIO_UNREGISTER,	\
74 				     struct uffdio_range)
75 #define UFFDIO_WAKE		_IOR(UFFDIO, _UFFDIO_WAKE,	\
76 				     struct uffdio_range)
77 #define UFFDIO_COPY		_IOWR(UFFDIO, _UFFDIO_COPY,	\
78 				      struct uffdio_copy)
79 #define UFFDIO_ZEROPAGE		_IOWR(UFFDIO, _UFFDIO_ZEROPAGE,	\
80 				      struct uffdio_zeropage)
81 #define UFFDIO_WRITEPROTECT	_IOWR(UFFDIO, _UFFDIO_WRITEPROTECT, \
82 				      struct uffdio_writeprotect)
83 #define UFFDIO_CONTINUE		_IOR(UFFDIO, _UFFDIO_CONTINUE,	\
84 				     struct uffdio_continue)
85 
86 /* read() structure */
87 struct uffd_msg {
88 	uint8_t	event;
89 
90 	uint8_t	reserved1;
91 	uint16_t	reserved2;
92 	uint32_t	reserved3;
93 
94 	union {
95 		struct {
96 			uint64_t	flags;
97 			uint64_t	address;
98 			union {
99 				uint32_t ptid;
100 			} feat;
101 		} pagefault;
102 
103 		struct {
104 			uint32_t	ufd;
105 		} fork;
106 
107 		struct {
108 			uint64_t	from;
109 			uint64_t	to;
110 			uint64_t	len;
111 		} remap;
112 
113 		struct {
114 			uint64_t	start;
115 			uint64_t	end;
116 		} remove;
117 
118 		struct {
119 			/* unused reserved fields */
120 			uint64_t	reserved1;
121 			uint64_t	reserved2;
122 			uint64_t	reserved3;
123 		} reserved;
124 	} arg;
125 } __packed;
126 
127 /*
128  * Start at 0x12 and not at 0 to be more strict against bugs.
129  */
130 #define UFFD_EVENT_PAGEFAULT	0x12
131 #define UFFD_EVENT_FORK		0x13
132 #define UFFD_EVENT_REMAP	0x14
133 #define UFFD_EVENT_REMOVE	0x15
134 #define UFFD_EVENT_UNMAP	0x16
135 
136 /* flags for UFFD_EVENT_PAGEFAULT */
137 #define UFFD_PAGEFAULT_FLAG_WRITE	(1<<0)	/* If this was a write fault */
138 #define UFFD_PAGEFAULT_FLAG_WP		(1<<1)	/* If reason is VM_UFFD_WP */
139 #define UFFD_PAGEFAULT_FLAG_MINOR	(1<<2)	/* If reason is VM_UFFD_MINOR */
140 
141 struct uffdio_api {
142 	/* userland asks for an API number and the features to enable */
143 	uint64_t api;
144 	/*
145 	 * Kernel answers below with the all available features for
146 	 * the API, this notifies userland of which events and/or
147 	 * which flags for each event are enabled in the current
148 	 * kernel.
149 	 *
150 	 * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE
151 	 * are to be considered implicitly always enabled in all kernels as
152 	 * long as the uffdio_api.api requested matches UFFD_API.
153 	 *
154 	 * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER
155 	 * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on
156 	 * hugetlbfs virtual memory ranges. Adding or not adding
157 	 * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has
158 	 * no real functional effect after UFFDIO_API returns, but
159 	 * it's only useful for an initial feature set probe at
160 	 * UFFDIO_API time. There are two ways to use it:
161 	 *
162 	 * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the
163 	 *    uffdio_api.features before calling UFFDIO_API, an error
164 	 *    will be returned by UFFDIO_API on a kernel without
165 	 *    hugetlbfs missing support
166 	 *
167 	 * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in
168 	 *    uffdio_api.features and instead it will be set by the
169 	 *    kernel in the uffdio_api.features if the kernel supports
170 	 *    it, so userland can later check if the feature flag is
171 	 *    present in uffdio_api.features after UFFDIO_API
172 	 *    succeeded.
173 	 *
174 	 * UFFD_FEATURE_MISSING_SHMEM works the same as
175 	 * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem
176 	 * (i.e. tmpfs and other shmem based APIs).
177 	 *
178 	 * UFFD_FEATURE_SIGBUS feature means no page-fault
179 	 * (UFFD_EVENT_PAGEFAULT) event will be delivered, instead
180 	 * a SIGBUS signal will be sent to the faulting process.
181 	 *
182 	 * UFFD_FEATURE_THREAD_ID pid of the page faulted task_struct will
183 	 * be returned, if feature is not requested 0 will be returned.
184 	 *
185 	 * UFFD_FEATURE_MINOR_HUGETLBFS indicates that minor faults
186 	 * can be intercepted (via REGISTER_MODE_MINOR) for
187 	 * hugetlbfs-backed pages.
188 	 */
189 #define UFFD_FEATURE_PAGEFAULT_FLAG_WP		(1<<0)
190 #define UFFD_FEATURE_EVENT_FORK			(1<<1)
191 #define UFFD_FEATURE_EVENT_REMAP		(1<<2)
192 #define UFFD_FEATURE_EVENT_REMOVE		(1<<3)
193 #define UFFD_FEATURE_MISSING_HUGETLBFS		(1<<4)
194 #define UFFD_FEATURE_MISSING_SHMEM		(1<<5)
195 #define UFFD_FEATURE_EVENT_UNMAP		(1<<6)
196 #define UFFD_FEATURE_SIGBUS			(1<<7)
197 #define UFFD_FEATURE_THREAD_ID			(1<<8)
198 #define UFFD_FEATURE_MINOR_HUGETLBFS		(1<<9)
199 	uint64_t features;
200 
201 	uint64_t ioctls;
202 };
203 
204 struct uffdio_range {
205 	uint64_t start;
206 	uint64_t len;
207 };
208 
209 struct uffdio_register {
210 	struct uffdio_range range;
211 #define UFFDIO_REGISTER_MODE_MISSING	((uint64_t)1<<0)
212 #define UFFDIO_REGISTER_MODE_WP		((uint64_t)1<<1)
213 #define UFFDIO_REGISTER_MODE_MINOR	((uint64_t)1<<2)
214 	uint64_t mode;
215 
216 	/*
217 	 * kernel answers which ioctl commands are available for the
218 	 * range, keep at the end as the last 8 bytes aren't read.
219 	 */
220 	uint64_t ioctls;
221 };
222 
223 struct uffdio_copy {
224 	uint64_t dst;
225 	uint64_t src;
226 	uint64_t len;
227 #define UFFDIO_COPY_MODE_DONTWAKE		((uint64_t)1<<0)
228 	/*
229 	 * UFFDIO_COPY_MODE_WP will map the page write protected on
230 	 * the fly.  UFFDIO_COPY_MODE_WP is available only if the
231 	 * write protected ioctl is implemented for the range
232 	 * according to the uffdio_register.ioctls.
233 	 */
234 #define UFFDIO_COPY_MODE_WP			((uint64_t)1<<1)
235 	uint64_t mode;
236 
237 	/*
238 	 * "copy" is written by the ioctl and must be at the end: the
239 	 * copy_from_user will not read the last 8 bytes.
240 	 */
241 	int64_t copy;
242 };
243 
244 struct uffdio_zeropage {
245 	struct uffdio_range range;
246 #define UFFDIO_ZEROPAGE_MODE_DONTWAKE		((uint64_t)1<<0)
247 	uint64_t mode;
248 
249 	/*
250 	 * "zeropage" is written by the ioctl and must be at the end:
251 	 * the copy_from_user will not read the last 8 bytes.
252 	 */
253 	int64_t zeropage;
254 };
255 
256 struct uffdio_writeprotect {
257 	struct uffdio_range range;
258 /*
259  * UFFDIO_WRITEPROTECT_MODE_WP: set the flag to write protect a range,
260  * unset the flag to undo protection of a range which was previously
261  * write protected.
262  *
263  * UFFDIO_WRITEPROTECT_MODE_DONTWAKE: set the flag to avoid waking up
264  * any wait thread after the operation succeeds.
265  *
266  * NOTE: Write protecting a region (WP=1) is unrelated to page faults,
267  * therefore DONTWAKE flag is meaningless with WP=1.  Removing write
268  * protection (WP=0) in response to a page fault wakes the faulting
269  * task unless DONTWAKE is set.
270  */
271 #define UFFDIO_WRITEPROTECT_MODE_WP		((uint64_t)1<<0)
272 #define UFFDIO_WRITEPROTECT_MODE_DONTWAKE	((uint64_t)1<<1)
273 	uint64_t mode;
274 };
275 
276 struct uffdio_continue {
277 	struct uffdio_range range;
278 #define UFFDIO_CONTINUE_MODE_DONTWAKE		((uint64_t)1<<0)
279 	uint64_t mode;
280 
281 	/*
282 	 * Fields below here are written by the ioctl and must be at the end:
283 	 * the copy_from_user will not read past here.
284 	 */
285 	int64_t mapped;
286 };
287 
288 /*
289  * Flags for the userfaultfd(2) system call itself.
290  */
291 
292 /*
293  * Create a userfaultfd that can handle page faults only in user mode.
294  */
295 #define UFFD_USER_MODE_ONLY 1
296 
297 #endif /* _LINUX_USERFAULTFD_H */
298