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_READ
19  * means the userland is reading).
20  */
21 #define UFFD_API ((__u64)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 			   UFFD_FEATURE_MINOR_SHMEM)
36 #define UFFD_API_IOCTLS				\
37 	((__u64)1 << _UFFDIO_REGISTER |		\
38 	 (__u64)1 << _UFFDIO_UNREGISTER |	\
39 	 (__u64)1 << _UFFDIO_API)
40 #define UFFD_API_RANGE_IOCTLS			\
41 	((__u64)1 << _UFFDIO_WAKE |		\
42 	 (__u64)1 << _UFFDIO_COPY |		\
43 	 (__u64)1 << _UFFDIO_ZEROPAGE |		\
44 	 (__u64)1 << _UFFDIO_WRITEPROTECT |	\
45 	 (__u64)1 << _UFFDIO_CONTINUE)
46 #define UFFD_API_RANGE_IOCTLS_BASIC		\
47 	((__u64)1 << _UFFDIO_WAKE |		\
48 	 (__u64)1 << _UFFDIO_COPY |		\
49 	 (__u64)1 << _UFFDIO_CONTINUE)
50 
51 /*
52  * Valid ioctl command number range with this API is from 0x00 to
53  * 0x3F.  UFFDIO_API is the fixed number, everything else can be
54  * changed by implementing a different UFFD_API. If sticking to the
55  * same UFFD_API more ioctl can be added and userland will be aware of
56  * which ioctl the running kernel implements through the ioctl command
57  * bitmask written by the UFFDIO_API.
58  */
59 #define _UFFDIO_REGISTER		(0x00)
60 #define _UFFDIO_UNREGISTER		(0x01)
61 #define _UFFDIO_WAKE			(0x02)
62 #define _UFFDIO_COPY			(0x03)
63 #define _UFFDIO_ZEROPAGE		(0x04)
64 #define _UFFDIO_WRITEPROTECT		(0x06)
65 #define _UFFDIO_CONTINUE		(0x07)
66 #define _UFFDIO_API			(0x3F)
67 
68 /* userfaultfd ioctl ids */
69 #define UFFDIO 0xAA
70 #define UFFDIO_API		_IOWR(UFFDIO, _UFFDIO_API,	\
71 				      struct uffdio_api)
72 #define UFFDIO_REGISTER		_IOWR(UFFDIO, _UFFDIO_REGISTER, \
73 				      struct uffdio_register)
74 #define UFFDIO_UNREGISTER	_IOR(UFFDIO, _UFFDIO_UNREGISTER,	\
75 				     struct uffdio_range)
76 #define UFFDIO_WAKE		_IOR(UFFDIO, _UFFDIO_WAKE,	\
77 				     struct uffdio_range)
78 #define UFFDIO_COPY		_IOWR(UFFDIO, _UFFDIO_COPY,	\
79 				      struct uffdio_copy)
80 #define UFFDIO_ZEROPAGE		_IOWR(UFFDIO, _UFFDIO_ZEROPAGE,	\
81 				      struct uffdio_zeropage)
82 #define UFFDIO_WRITEPROTECT	_IOWR(UFFDIO, _UFFDIO_WRITEPROTECT, \
83 				      struct uffdio_writeprotect)
84 #define UFFDIO_CONTINUE		_IOWR(UFFDIO, _UFFDIO_CONTINUE,	\
85 				      struct uffdio_continue)
86 
87 /* read() structure */
88 struct uffd_msg {
89 	__u8	event;
90 
91 	__u8	reserved1;
92 	__u16	reserved2;
93 	__u32	reserved3;
94 
95 	union {
96 		struct {
97 			__u64	flags;
98 			__u64	address;
99 			union {
100 				__u32 ptid;
101 			} feat;
102 		} pagefault;
103 
104 		struct {
105 			__u32	ufd;
106 		} fork;
107 
108 		struct {
109 			__u64	from;
110 			__u64	to;
111 			__u64	len;
112 		} remap;
113 
114 		struct {
115 			__u64	start;
116 			__u64	end;
117 		} remove;
118 
119 		struct {
120 			/* unused reserved fields */
121 			__u64	reserved1;
122 			__u64	reserved2;
123 			__u64	reserved3;
124 		} reserved;
125 	} arg;
126 } __packed;
127 
128 /*
129  * Start at 0x12 and not at 0 to be more strict against bugs.
130  */
131 #define UFFD_EVENT_PAGEFAULT	0x12
132 #define UFFD_EVENT_FORK		0x13
133 #define UFFD_EVENT_REMAP	0x14
134 #define UFFD_EVENT_REMOVE	0x15
135 #define UFFD_EVENT_UNMAP	0x16
136 
137 /* flags for UFFD_EVENT_PAGEFAULT */
138 #define UFFD_PAGEFAULT_FLAG_WRITE	(1<<0)	/* If this was a write fault */
139 #define UFFD_PAGEFAULT_FLAG_WP		(1<<1)	/* If reason is VM_UFFD_WP */
140 #define UFFD_PAGEFAULT_FLAG_MINOR	(1<<2)	/* If reason is VM_UFFD_MINOR */
141 
142 struct uffdio_api {
143 	/* userland asks for an API number and the features to enable */
144 	__u64 api;
145 	/*
146 	 * Kernel answers below with the all available features for
147 	 * the API, this notifies userland of which events and/or
148 	 * which flags for each event are enabled in the current
149 	 * kernel.
150 	 *
151 	 * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE
152 	 * are to be considered implicitly always enabled in all kernels as
153 	 * long as the uffdio_api.api requested matches UFFD_API.
154 	 *
155 	 * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER
156 	 * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on
157 	 * hugetlbfs virtual memory ranges. Adding or not adding
158 	 * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has
159 	 * no real functional effect after UFFDIO_API returns, but
160 	 * it's only useful for an initial feature set probe at
161 	 * UFFDIO_API time. There are two ways to use it:
162 	 *
163 	 * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the
164 	 *    uffdio_api.features before calling UFFDIO_API, an error
165 	 *    will be returned by UFFDIO_API on a kernel without
166 	 *    hugetlbfs missing support
167 	 *
168 	 * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in
169 	 *    uffdio_api.features and instead it will be set by the
170 	 *    kernel in the uffdio_api.features if the kernel supports
171 	 *    it, so userland can later check if the feature flag is
172 	 *    present in uffdio_api.features after UFFDIO_API
173 	 *    succeeded.
174 	 *
175 	 * UFFD_FEATURE_MISSING_SHMEM works the same as
176 	 * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem
177 	 * (i.e. tmpfs and other shmem based APIs).
178 	 *
179 	 * UFFD_FEATURE_SIGBUS feature means no page-fault
180 	 * (UFFD_EVENT_PAGEFAULT) event will be delivered, instead
181 	 * a SIGBUS signal will be sent to the faulting process.
182 	 *
183 	 * UFFD_FEATURE_THREAD_ID pid of the page faulted task_struct will
184 	 * be returned, if feature is not requested 0 will be returned.
185 	 *
186 	 * UFFD_FEATURE_MINOR_HUGETLBFS indicates that minor faults
187 	 * can be intercepted (via REGISTER_MODE_MINOR) for
188 	 * hugetlbfs-backed pages.
189 	 *
190 	 * UFFD_FEATURE_MINOR_SHMEM indicates the same support as
191 	 * UFFD_FEATURE_MINOR_HUGETLBFS, but for shmem-backed pages instead.
192 	 */
193 #define UFFD_FEATURE_PAGEFAULT_FLAG_WP		(1<<0)
194 #define UFFD_FEATURE_EVENT_FORK			(1<<1)
195 #define UFFD_FEATURE_EVENT_REMAP		(1<<2)
196 #define UFFD_FEATURE_EVENT_REMOVE		(1<<3)
197 #define UFFD_FEATURE_MISSING_HUGETLBFS		(1<<4)
198 #define UFFD_FEATURE_MISSING_SHMEM		(1<<5)
199 #define UFFD_FEATURE_EVENT_UNMAP		(1<<6)
200 #define UFFD_FEATURE_SIGBUS			(1<<7)
201 #define UFFD_FEATURE_THREAD_ID			(1<<8)
202 #define UFFD_FEATURE_MINOR_HUGETLBFS		(1<<9)
203 #define UFFD_FEATURE_MINOR_SHMEM		(1<<10)
204 	__u64 features;
205 
206 	__u64 ioctls;
207 };
208 
209 struct uffdio_range {
210 	__u64 start;
211 	__u64 len;
212 };
213 
214 struct uffdio_register {
215 	struct uffdio_range range;
216 #define UFFDIO_REGISTER_MODE_MISSING	((__u64)1<<0)
217 #define UFFDIO_REGISTER_MODE_WP		((__u64)1<<1)
218 #define UFFDIO_REGISTER_MODE_MINOR	((__u64)1<<2)
219 	__u64 mode;
220 
221 	/*
222 	 * kernel answers which ioctl commands are available for the
223 	 * range, keep at the end as the last 8 bytes aren't read.
224 	 */
225 	__u64 ioctls;
226 };
227 
228 struct uffdio_copy {
229 	__u64 dst;
230 	__u64 src;
231 	__u64 len;
232 #define UFFDIO_COPY_MODE_DONTWAKE		((__u64)1<<0)
233 	/*
234 	 * UFFDIO_COPY_MODE_WP will map the page write protected on
235 	 * the fly.  UFFDIO_COPY_MODE_WP is available only if the
236 	 * write protected ioctl is implemented for the range
237 	 * according to the uffdio_register.ioctls.
238 	 */
239 #define UFFDIO_COPY_MODE_WP			((__u64)1<<1)
240 	__u64 mode;
241 
242 	/*
243 	 * "copy" is written by the ioctl and must be at the end: the
244 	 * copy_from_user will not read the last 8 bytes.
245 	 */
246 	__s64 copy;
247 };
248 
249 struct uffdio_zeropage {
250 	struct uffdio_range range;
251 #define UFFDIO_ZEROPAGE_MODE_DONTWAKE		((__u64)1<<0)
252 	__u64 mode;
253 
254 	/*
255 	 * "zeropage" is written by the ioctl and must be at the end:
256 	 * the copy_from_user will not read the last 8 bytes.
257 	 */
258 	__s64 zeropage;
259 };
260 
261 struct uffdio_writeprotect {
262 	struct uffdio_range range;
263 /*
264  * UFFDIO_WRITEPROTECT_MODE_WP: set the flag to write protect a range,
265  * unset the flag to undo protection of a range which was previously
266  * write protected.
267  *
268  * UFFDIO_WRITEPROTECT_MODE_DONTWAKE: set the flag to avoid waking up
269  * any wait thread after the operation succeeds.
270  *
271  * NOTE: Write protecting a region (WP=1) is unrelated to page faults,
272  * therefore DONTWAKE flag is meaningless with WP=1.  Removing write
273  * protection (WP=0) in response to a page fault wakes the faulting
274  * task unless DONTWAKE is set.
275  */
276 #define UFFDIO_WRITEPROTECT_MODE_WP		((__u64)1<<0)
277 #define UFFDIO_WRITEPROTECT_MODE_DONTWAKE	((__u64)1<<1)
278 	__u64 mode;
279 };
280 
281 struct uffdio_continue {
282 	struct uffdio_range range;
283 #define UFFDIO_CONTINUE_MODE_DONTWAKE		((__u64)1<<0)
284 	__u64 mode;
285 
286 	/*
287 	 * Fields below here are written by the ioctl and must be at the end:
288 	 * the copy_from_user will not read past here.
289 	 */
290 	__s64 mapped;
291 };
292 
293 /*
294  * Flags for the userfaultfd(2) system call itself.
295  */
296 
297 /*
298  * Create a userfaultfd that can handle page faults only in user mode.
299  */
300 #define UFFD_USER_MODE_ONLY 1
301 
302 #endif /* _LINUX_USERFAULTFD_H */
303