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 			   UFFD_FEATURE_WP_UNPOPULATED)
43 #define UFFD_API_IOCTLS				\
44 	((__u64)1 << _UFFDIO_REGISTER |		\
45 	 (__u64)1 << _UFFDIO_UNREGISTER |	\
46 	 (__u64)1 << _UFFDIO_API)
47 #define UFFD_API_RANGE_IOCTLS			\
48 	((__u64)1 << _UFFDIO_WAKE |		\
49 	 (__u64)1 << _UFFDIO_COPY |		\
50 	 (__u64)1 << _UFFDIO_ZEROPAGE |		\
51 	 (__u64)1 << _UFFDIO_WRITEPROTECT |	\
52 	 (__u64)1 << _UFFDIO_CONTINUE)
53 #define UFFD_API_RANGE_IOCTLS_BASIC		\
54 	((__u64)1 << _UFFDIO_WAKE |		\
55 	 (__u64)1 << _UFFDIO_COPY |		\
56 	 (__u64)1 << _UFFDIO_CONTINUE |		\
57 	 (__u64)1 << _UFFDIO_WRITEPROTECT)
58 
59 /*
60  * Valid ioctl command number range with this API is from 0x00 to
61  * 0x3F.  UFFDIO_API is the fixed number, everything else can be
62  * changed by implementing a different UFFD_API. If sticking to the
63  * same UFFD_API more ioctl can be added and userland will be aware of
64  * which ioctl the running kernel implements through the ioctl command
65  * bitmask written by the UFFDIO_API.
66  */
67 #define _UFFDIO_REGISTER		(0x00)
68 #define _UFFDIO_UNREGISTER		(0x01)
69 #define _UFFDIO_WAKE			(0x02)
70 #define _UFFDIO_COPY			(0x03)
71 #define _UFFDIO_ZEROPAGE		(0x04)
72 #define _UFFDIO_WRITEPROTECT		(0x06)
73 #define _UFFDIO_CONTINUE		(0x07)
74 #define _UFFDIO_API			(0x3F)
75 
76 /* userfaultfd ioctl ids */
77 #define UFFDIO 0xAA
78 #define UFFDIO_API		_IOWR(UFFDIO, _UFFDIO_API,	\
79 				      struct uffdio_api)
80 #define UFFDIO_REGISTER		_IOWR(UFFDIO, _UFFDIO_REGISTER, \
81 				      struct uffdio_register)
82 #define UFFDIO_UNREGISTER	_IOR(UFFDIO, _UFFDIO_UNREGISTER,	\
83 				     struct uffdio_range)
84 #define UFFDIO_WAKE		_IOR(UFFDIO, _UFFDIO_WAKE,	\
85 				     struct uffdio_range)
86 #define UFFDIO_COPY		_IOWR(UFFDIO, _UFFDIO_COPY,	\
87 				      struct uffdio_copy)
88 #define UFFDIO_ZEROPAGE		_IOWR(UFFDIO, _UFFDIO_ZEROPAGE,	\
89 				      struct uffdio_zeropage)
90 #define UFFDIO_WRITEPROTECT	_IOWR(UFFDIO, _UFFDIO_WRITEPROTECT, \
91 				      struct uffdio_writeprotect)
92 #define UFFDIO_CONTINUE		_IOWR(UFFDIO, _UFFDIO_CONTINUE,	\
93 				      struct uffdio_continue)
94 
95 /* read() structure */
96 struct uffd_msg {
97 	__u8	event;
98 
99 	__u8	reserved1;
100 	__u16	reserved2;
101 	__u32	reserved3;
102 
103 	union {
104 		struct {
105 			__u64	flags;
106 			__u64	address;
107 			union {
108 				__u32 ptid;
109 			} feat;
110 		} pagefault;
111 
112 		struct {
113 			__u32	ufd;
114 		} fork;
115 
116 		struct {
117 			__u64	from;
118 			__u64	to;
119 			__u64	len;
120 		} remap;
121 
122 		struct {
123 			__u64	start;
124 			__u64	end;
125 		} remove;
126 
127 		struct {
128 			/* unused reserved fields */
129 			__u64	reserved1;
130 			__u64	reserved2;
131 			__u64	reserved3;
132 		} reserved;
133 	} arg;
134 } __attribute__((packed));
135 
136 /*
137  * Start at 0x12 and not at 0 to be more strict against bugs.
138  */
139 #define UFFD_EVENT_PAGEFAULT	0x12
140 #define UFFD_EVENT_FORK		0x13
141 #define UFFD_EVENT_REMAP	0x14
142 #define UFFD_EVENT_REMOVE	0x15
143 #define UFFD_EVENT_UNMAP	0x16
144 
145 /* flags for UFFD_EVENT_PAGEFAULT */
146 #define UFFD_PAGEFAULT_FLAG_WRITE	(1<<0)	/* If this was a write fault */
147 #define UFFD_PAGEFAULT_FLAG_WP		(1<<1)	/* If reason is VM_UFFD_WP */
148 #define UFFD_PAGEFAULT_FLAG_MINOR	(1<<2)	/* If reason is VM_UFFD_MINOR */
149 
150 struct uffdio_api {
151 	/* userland asks for an API number and the features to enable */
152 	__u64 api;
153 	/*
154 	 * Kernel answers below with the all available features for
155 	 * the API, this notifies userland of which events and/or
156 	 * which flags for each event are enabled in the current
157 	 * kernel.
158 	 *
159 	 * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE
160 	 * are to be considered implicitly always enabled in all kernels as
161 	 * long as the uffdio_api.api requested matches UFFD_API.
162 	 *
163 	 * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER
164 	 * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on
165 	 * hugetlbfs virtual memory ranges. Adding or not adding
166 	 * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has
167 	 * no real functional effect after UFFDIO_API returns, but
168 	 * it's only useful for an initial feature set probe at
169 	 * UFFDIO_API time. There are two ways to use it:
170 	 *
171 	 * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the
172 	 *    uffdio_api.features before calling UFFDIO_API, an error
173 	 *    will be returned by UFFDIO_API on a kernel without
174 	 *    hugetlbfs missing support
175 	 *
176 	 * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in
177 	 *    uffdio_api.features and instead it will be set by the
178 	 *    kernel in the uffdio_api.features if the kernel supports
179 	 *    it, so userland can later check if the feature flag is
180 	 *    present in uffdio_api.features after UFFDIO_API
181 	 *    succeeded.
182 	 *
183 	 * UFFD_FEATURE_MISSING_SHMEM works the same as
184 	 * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem
185 	 * (i.e. tmpfs and other shmem based APIs).
186 	 *
187 	 * UFFD_FEATURE_SIGBUS feature means no page-fault
188 	 * (UFFD_EVENT_PAGEFAULT) event will be delivered, instead
189 	 * a SIGBUS signal will be sent to the faulting process.
190 	 *
191 	 * UFFD_FEATURE_THREAD_ID pid of the page faulted task_struct will
192 	 * be returned, if feature is not requested 0 will be returned.
193 	 *
194 	 * UFFD_FEATURE_MINOR_HUGETLBFS indicates that minor faults
195 	 * can be intercepted (via REGISTER_MODE_MINOR) for
196 	 * hugetlbfs-backed pages.
197 	 *
198 	 * UFFD_FEATURE_MINOR_SHMEM indicates the same support as
199 	 * UFFD_FEATURE_MINOR_HUGETLBFS, but for shmem-backed pages instead.
200 	 *
201 	 * UFFD_FEATURE_EXACT_ADDRESS indicates that the exact address of page
202 	 * faults would be provided and the offset within the page would not be
203 	 * masked.
204 	 *
205 	 * UFFD_FEATURE_WP_HUGETLBFS_SHMEM indicates that userfaultfd
206 	 * write-protection mode is supported on both shmem and hugetlbfs.
207 	 *
208 	 * UFFD_FEATURE_WP_UNPOPULATED indicates that userfaultfd
209 	 * write-protection mode will always apply to unpopulated pages
210 	 * (i.e. empty ptes).  This will be the default behavior for shmem
211 	 * & hugetlbfs, so this flag only affects anonymous memory behavior
212 	 * when userfault write-protection mode is registered.
213 	 */
214 #define UFFD_FEATURE_PAGEFAULT_FLAG_WP		(1<<0)
215 #define UFFD_FEATURE_EVENT_FORK			(1<<1)
216 #define UFFD_FEATURE_EVENT_REMAP		(1<<2)
217 #define UFFD_FEATURE_EVENT_REMOVE		(1<<3)
218 #define UFFD_FEATURE_MISSING_HUGETLBFS		(1<<4)
219 #define UFFD_FEATURE_MISSING_SHMEM		(1<<5)
220 #define UFFD_FEATURE_EVENT_UNMAP		(1<<6)
221 #define UFFD_FEATURE_SIGBUS			(1<<7)
222 #define UFFD_FEATURE_THREAD_ID			(1<<8)
223 #define UFFD_FEATURE_MINOR_HUGETLBFS		(1<<9)
224 #define UFFD_FEATURE_MINOR_SHMEM		(1<<10)
225 #define UFFD_FEATURE_EXACT_ADDRESS		(1<<11)
226 #define UFFD_FEATURE_WP_HUGETLBFS_SHMEM		(1<<12)
227 #define UFFD_FEATURE_WP_UNPOPULATED		(1<<13)
228 	__u64 features;
229 
230 	__u64 ioctls;
231 };
232 
233 struct uffdio_range {
234 	__u64 start;
235 	__u64 len;
236 };
237 
238 struct uffdio_register {
239 	struct uffdio_range range;
240 #define UFFDIO_REGISTER_MODE_MISSING	((__u64)1<<0)
241 #define UFFDIO_REGISTER_MODE_WP		((__u64)1<<1)
242 #define UFFDIO_REGISTER_MODE_MINOR	((__u64)1<<2)
243 	__u64 mode;
244 
245 	/*
246 	 * kernel answers which ioctl commands are available for the
247 	 * range, keep at the end as the last 8 bytes aren't read.
248 	 */
249 	__u64 ioctls;
250 };
251 
252 struct uffdio_copy {
253 	__u64 dst;
254 	__u64 src;
255 	__u64 len;
256 #define UFFDIO_COPY_MODE_DONTWAKE		((__u64)1<<0)
257 	/*
258 	 * UFFDIO_COPY_MODE_WP will map the page write protected on
259 	 * the fly.  UFFDIO_COPY_MODE_WP is available only if the
260 	 * write protected ioctl is implemented for the range
261 	 * according to the uffdio_register.ioctls.
262 	 */
263 #define UFFDIO_COPY_MODE_WP			((__u64)1<<1)
264 	__u64 mode;
265 
266 	/*
267 	 * "copy" is written by the ioctl and must be at the end: the
268 	 * copy_from_user will not read the last 8 bytes.
269 	 */
270 	__s64 copy;
271 };
272 
273 struct uffdio_zeropage {
274 	struct uffdio_range range;
275 #define UFFDIO_ZEROPAGE_MODE_DONTWAKE		((__u64)1<<0)
276 	__u64 mode;
277 
278 	/*
279 	 * "zeropage" is written by the ioctl and must be at the end:
280 	 * the copy_from_user will not read the last 8 bytes.
281 	 */
282 	__s64 zeropage;
283 };
284 
285 struct uffdio_writeprotect {
286 	struct uffdio_range range;
287 /*
288  * UFFDIO_WRITEPROTECT_MODE_WP: set the flag to write protect a range,
289  * unset the flag to undo protection of a range which was previously
290  * write protected.
291  *
292  * UFFDIO_WRITEPROTECT_MODE_DONTWAKE: set the flag to avoid waking up
293  * any wait thread after the operation succeeds.
294  *
295  * NOTE: Write protecting a region (WP=1) is unrelated to page faults,
296  * therefore DONTWAKE flag is meaningless with WP=1.  Removing write
297  * protection (WP=0) in response to a page fault wakes the faulting
298  * task unless DONTWAKE is set.
299  */
300 #define UFFDIO_WRITEPROTECT_MODE_WP		((__u64)1<<0)
301 #define UFFDIO_WRITEPROTECT_MODE_DONTWAKE	((__u64)1<<1)
302 	__u64 mode;
303 };
304 
305 struct uffdio_continue {
306 	struct uffdio_range range;
307 #define UFFDIO_CONTINUE_MODE_DONTWAKE		((__u64)1<<0)
308 	/*
309 	 * UFFDIO_CONTINUE_MODE_WP will map the page write protected on
310 	 * the fly.  UFFDIO_CONTINUE_MODE_WP is available only if the
311 	 * write protected ioctl is implemented for the range
312 	 * according to the uffdio_register.ioctls.
313 	 */
314 #define UFFDIO_CONTINUE_MODE_WP			((__u64)1<<1)
315 	__u64 mode;
316 
317 	/*
318 	 * Fields below here are written by the ioctl and must be at the end:
319 	 * the copy_from_user will not read past here.
320 	 */
321 	__s64 mapped;
322 };
323 
324 /*
325  * Flags for the userfaultfd(2) system call itself.
326  */
327 
328 /*
329  * Create a userfaultfd that can handle page faults only in user mode.
330  */
331 #define UFFD_USER_MODE_ONLY 1
332 
333 #endif /* _LINUX_USERFAULTFD_H */
334