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 /* ioctls for /dev/userfaultfd */
16 #define USERFAULTFD_IOC 0xAA
17 #define USERFAULTFD_IOC_NEW _IO(USERFAULTFD_IOC, 0x00)
18 
19 /*
20  * If the UFFDIO_API is upgraded someday, the UFFDIO_UNREGISTER and
21  * UFFDIO_WAKE ioctls should be defined as _IOW and not as _IOR.  In
22  * userfaultfd.h we assumed the kernel was reading (instead _IOC_READ
23  * means the userland is reading).
24  */
25 #define UFFD_API ((__u64)0xAA)
26 #define UFFD_API_REGISTER_MODES (UFFDIO_REGISTER_MODE_MISSING |	\
27 				 UFFDIO_REGISTER_MODE_WP |	\
28 				 UFFDIO_REGISTER_MODE_MINOR)
29 #define UFFD_API_FEATURES (UFFD_FEATURE_PAGEFAULT_FLAG_WP |	\
30 			   UFFD_FEATURE_EVENT_FORK |		\
31 			   UFFD_FEATURE_EVENT_REMAP |		\
32 			   UFFD_FEATURE_EVENT_REMOVE |		\
33 			   UFFD_FEATURE_EVENT_UNMAP |		\
34 			   UFFD_FEATURE_MISSING_HUGETLBFS |	\
35 			   UFFD_FEATURE_MISSING_SHMEM |		\
36 			   UFFD_FEATURE_SIGBUS |		\
37 			   UFFD_FEATURE_THREAD_ID |		\
38 			   UFFD_FEATURE_MINOR_HUGETLBFS |	\
39 			   UFFD_FEATURE_MINOR_SHMEM |		\
40 			   UFFD_FEATURE_EXACT_ADDRESS |		\
41 			   UFFD_FEATURE_WP_HUGETLBFS_SHMEM)
42 #define UFFD_API_IOCTLS				\
43 	((__u64)1 << _UFFDIO_REGISTER |		\
44 	 (__u64)1 << _UFFDIO_UNREGISTER |	\
45 	 (__u64)1 << _UFFDIO_API)
46 #define UFFD_API_RANGE_IOCTLS			\
47 	((__u64)1 << _UFFDIO_WAKE |		\
48 	 (__u64)1 << _UFFDIO_COPY |		\
49 	 (__u64)1 << _UFFDIO_ZEROPAGE |		\
50 	 (__u64)1 << _UFFDIO_WRITEPROTECT |	\
51 	 (__u64)1 << _UFFDIO_CONTINUE)
52 #define UFFD_API_RANGE_IOCTLS_BASIC		\
53 	((__u64)1 << _UFFDIO_WAKE |		\
54 	 (__u64)1 << _UFFDIO_COPY |		\
55 	 (__u64)1 << _UFFDIO_CONTINUE |		\
56 	 (__u64)1 << _UFFDIO_WRITEPROTECT)
57 
58 /*
59  * Valid ioctl command number range with this API is from 0x00 to
60  * 0x3F.  UFFDIO_API is the fixed number, everything else can be
61  * changed by implementing a different UFFD_API. If sticking to the
62  * same UFFD_API more ioctl can be added and userland will be aware of
63  * which ioctl the running kernel implements through the ioctl command
64  * bitmask written by the UFFDIO_API.
65  */
66 #define _UFFDIO_REGISTER		(0x00)
67 #define _UFFDIO_UNREGISTER		(0x01)
68 #define _UFFDIO_WAKE			(0x02)
69 #define _UFFDIO_COPY			(0x03)
70 #define _UFFDIO_ZEROPAGE		(0x04)
71 #define _UFFDIO_WRITEPROTECT		(0x06)
72 #define _UFFDIO_CONTINUE		(0x07)
73 #define _UFFDIO_API			(0x3F)
74 
75 /* userfaultfd ioctl ids */
76 #define UFFDIO 0xAA
77 #define UFFDIO_API		_IOWR(UFFDIO, _UFFDIO_API,	\
78 				      struct uffdio_api)
79 #define UFFDIO_REGISTER		_IOWR(UFFDIO, _UFFDIO_REGISTER, \
80 				      struct uffdio_register)
81 #define UFFDIO_UNREGISTER	_IOR(UFFDIO, _UFFDIO_UNREGISTER,	\
82 				     struct uffdio_range)
83 #define UFFDIO_WAKE		_IOR(UFFDIO, _UFFDIO_WAKE,	\
84 				     struct uffdio_range)
85 #define UFFDIO_COPY		_IOWR(UFFDIO, _UFFDIO_COPY,	\
86 				      struct uffdio_copy)
87 #define UFFDIO_ZEROPAGE		_IOWR(UFFDIO, _UFFDIO_ZEROPAGE,	\
88 				      struct uffdio_zeropage)
89 #define UFFDIO_WRITEPROTECT	_IOWR(UFFDIO, _UFFDIO_WRITEPROTECT, \
90 				      struct uffdio_writeprotect)
91 #define UFFDIO_CONTINUE		_IOWR(UFFDIO, _UFFDIO_CONTINUE,	\
92 				      struct uffdio_continue)
93 
94 /* read() structure */
95 struct uffd_msg {
96 	__u8	event;
97 
98 	__u8	reserved1;
99 	__u16	reserved2;
100 	__u32	reserved3;
101 
102 	union {
103 		struct {
104 			__u64	flags;
105 			__u64	address;
106 			union {
107 				__u32 ptid;
108 			} feat;
109 		} pagefault;
110 
111 		struct {
112 			__u32	ufd;
113 		} fork;
114 
115 		struct {
116 			__u64	from;
117 			__u64	to;
118 			__u64	len;
119 		} remap;
120 
121 		struct {
122 			__u64	start;
123 			__u64	end;
124 		} remove;
125 
126 		struct {
127 			/* unused reserved fields */
128 			__u64	reserved1;
129 			__u64	reserved2;
130 			__u64	reserved3;
131 		} reserved;
132 	} arg;
133 } __attribute__((packed));
134 
135 /*
136  * Start at 0x12 and not at 0 to be more strict against bugs.
137  */
138 #define UFFD_EVENT_PAGEFAULT	0x12
139 #define UFFD_EVENT_FORK		0x13
140 #define UFFD_EVENT_REMAP	0x14
141 #define UFFD_EVENT_REMOVE	0x15
142 #define UFFD_EVENT_UNMAP	0x16
143 
144 /* flags for UFFD_EVENT_PAGEFAULT */
145 #define UFFD_PAGEFAULT_FLAG_WRITE	(1<<0)	/* If this was a write fault */
146 #define UFFD_PAGEFAULT_FLAG_WP		(1<<1)	/* If reason is VM_UFFD_WP */
147 #define UFFD_PAGEFAULT_FLAG_MINOR	(1<<2)	/* If reason is VM_UFFD_MINOR */
148 
149 struct uffdio_api {
150 	/* userland asks for an API number and the features to enable */
151 	__u64 api;
152 	/*
153 	 * Kernel answers below with the all available features for
154 	 * the API, this notifies userland of which events and/or
155 	 * which flags for each event are enabled in the current
156 	 * kernel.
157 	 *
158 	 * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE
159 	 * are to be considered implicitly always enabled in all kernels as
160 	 * long as the uffdio_api.api requested matches UFFD_API.
161 	 *
162 	 * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER
163 	 * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on
164 	 * hugetlbfs virtual memory ranges. Adding or not adding
165 	 * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has
166 	 * no real functional effect after UFFDIO_API returns, but
167 	 * it's only useful for an initial feature set probe at
168 	 * UFFDIO_API time. There are two ways to use it:
169 	 *
170 	 * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the
171 	 *    uffdio_api.features before calling UFFDIO_API, an error
172 	 *    will be returned by UFFDIO_API on a kernel without
173 	 *    hugetlbfs missing support
174 	 *
175 	 * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in
176 	 *    uffdio_api.features and instead it will be set by the
177 	 *    kernel in the uffdio_api.features if the kernel supports
178 	 *    it, so userland can later check if the feature flag is
179 	 *    present in uffdio_api.features after UFFDIO_API
180 	 *    succeeded.
181 	 *
182 	 * UFFD_FEATURE_MISSING_SHMEM works the same as
183 	 * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem
184 	 * (i.e. tmpfs and other shmem based APIs).
185 	 *
186 	 * UFFD_FEATURE_SIGBUS feature means no page-fault
187 	 * (UFFD_EVENT_PAGEFAULT) event will be delivered, instead
188 	 * a SIGBUS signal will be sent to the faulting process.
189 	 *
190 	 * UFFD_FEATURE_THREAD_ID pid of the page faulted task_struct will
191 	 * be returned, if feature is not requested 0 will be returned.
192 	 *
193 	 * UFFD_FEATURE_MINOR_HUGETLBFS indicates that minor faults
194 	 * can be intercepted (via REGISTER_MODE_MINOR) for
195 	 * hugetlbfs-backed pages.
196 	 *
197 	 * UFFD_FEATURE_MINOR_SHMEM indicates the same support as
198 	 * UFFD_FEATURE_MINOR_HUGETLBFS, but for shmem-backed pages instead.
199 	 *
200 	 * UFFD_FEATURE_EXACT_ADDRESS indicates that the exact address of page
201 	 * faults would be provided and the offset within the page would not be
202 	 * masked.
203 	 *
204 	 * UFFD_FEATURE_WP_HUGETLBFS_SHMEM indicates that userfaultfd
205 	 * write-protection mode is supported on both shmem and hugetlbfs.
206 	 */
207 #define UFFD_FEATURE_PAGEFAULT_FLAG_WP		(1<<0)
208 #define UFFD_FEATURE_EVENT_FORK			(1<<1)
209 #define UFFD_FEATURE_EVENT_REMAP		(1<<2)
210 #define UFFD_FEATURE_EVENT_REMOVE		(1<<3)
211 #define UFFD_FEATURE_MISSING_HUGETLBFS		(1<<4)
212 #define UFFD_FEATURE_MISSING_SHMEM		(1<<5)
213 #define UFFD_FEATURE_EVENT_UNMAP		(1<<6)
214 #define UFFD_FEATURE_SIGBUS			(1<<7)
215 #define UFFD_FEATURE_THREAD_ID			(1<<8)
216 #define UFFD_FEATURE_MINOR_HUGETLBFS		(1<<9)
217 #define UFFD_FEATURE_MINOR_SHMEM		(1<<10)
218 #define UFFD_FEATURE_EXACT_ADDRESS		(1<<11)
219 #define UFFD_FEATURE_WP_HUGETLBFS_SHMEM		(1<<12)
220 	__u64 features;
221 
222 	__u64 ioctls;
223 };
224 
225 struct uffdio_range {
226 	__u64 start;
227 	__u64 len;
228 };
229 
230 struct uffdio_register {
231 	struct uffdio_range range;
232 #define UFFDIO_REGISTER_MODE_MISSING	((__u64)1<<0)
233 #define UFFDIO_REGISTER_MODE_WP		((__u64)1<<1)
234 #define UFFDIO_REGISTER_MODE_MINOR	((__u64)1<<2)
235 	__u64 mode;
236 
237 	/*
238 	 * kernel answers which ioctl commands are available for the
239 	 * range, keep at the end as the last 8 bytes aren't read.
240 	 */
241 	__u64 ioctls;
242 };
243 
244 struct uffdio_copy {
245 	__u64 dst;
246 	__u64 src;
247 	__u64 len;
248 #define UFFDIO_COPY_MODE_DONTWAKE		((__u64)1<<0)
249 	/*
250 	 * UFFDIO_COPY_MODE_WP will map the page write protected on
251 	 * the fly.  UFFDIO_COPY_MODE_WP is available only if the
252 	 * write protected ioctl is implemented for the range
253 	 * according to the uffdio_register.ioctls.
254 	 */
255 #define UFFDIO_COPY_MODE_WP			((__u64)1<<1)
256 	__u64 mode;
257 
258 	/*
259 	 * "copy" is written by the ioctl and must be at the end: the
260 	 * copy_from_user will not read the last 8 bytes.
261 	 */
262 	__s64 copy;
263 };
264 
265 struct uffdio_zeropage {
266 	struct uffdio_range range;
267 #define UFFDIO_ZEROPAGE_MODE_DONTWAKE		((__u64)1<<0)
268 	__u64 mode;
269 
270 	/*
271 	 * "zeropage" is written by the ioctl and must be at the end:
272 	 * the copy_from_user will not read the last 8 bytes.
273 	 */
274 	__s64 zeropage;
275 };
276 
277 struct uffdio_writeprotect {
278 	struct uffdio_range range;
279 /*
280  * UFFDIO_WRITEPROTECT_MODE_WP: set the flag to write protect a range,
281  * unset the flag to undo protection of a range which was previously
282  * write protected.
283  *
284  * UFFDIO_WRITEPROTECT_MODE_DONTWAKE: set the flag to avoid waking up
285  * any wait thread after the operation succeeds.
286  *
287  * NOTE: Write protecting a region (WP=1) is unrelated to page faults,
288  * therefore DONTWAKE flag is meaningless with WP=1.  Removing write
289  * protection (WP=0) in response to a page fault wakes the faulting
290  * task unless DONTWAKE is set.
291  */
292 #define UFFDIO_WRITEPROTECT_MODE_WP		((__u64)1<<0)
293 #define UFFDIO_WRITEPROTECT_MODE_DONTWAKE	((__u64)1<<1)
294 	__u64 mode;
295 };
296 
297 struct uffdio_continue {
298 	struct uffdio_range range;
299 #define UFFDIO_CONTINUE_MODE_DONTWAKE		((__u64)1<<0)
300 	__u64 mode;
301 
302 	/*
303 	 * Fields below here are written by the ioctl and must be at the end:
304 	 * the copy_from_user will not read past here.
305 	 */
306 	__s64 mapped;
307 };
308 
309 /*
310  * Flags for the userfaultfd(2) system call itself.
311  */
312 
313 /*
314  * Create a userfaultfd that can handle page faults only in user mode.
315  */
316 #define UFFD_USER_MODE_ONLY 1
317 
318 #endif /* _LINUX_USERFAULTFD_H */
319