1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */ 2 /* 3 * include/linux/userfaultfd.h 4 * 5 * Copyright (C) 2007 Davide Libenzi <[email protected]> 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_POISON (0x08) 75 #define _UFFDIO_API (0x3F) 76 77 /* userfaultfd ioctl ids */ 78 #define UFFDIO 0xAA 79 #define UFFDIO_API _IOWR(UFFDIO, _UFFDIO_API, \ 80 struct uffdio_api) 81 #define UFFDIO_REGISTER _IOWR(UFFDIO, _UFFDIO_REGISTER, \ 82 struct uffdio_register) 83 #define UFFDIO_UNREGISTER _IOR(UFFDIO, _UFFDIO_UNREGISTER, \ 84 struct uffdio_range) 85 #define UFFDIO_WAKE _IOR(UFFDIO, _UFFDIO_WAKE, \ 86 struct uffdio_range) 87 #define UFFDIO_COPY _IOWR(UFFDIO, _UFFDIO_COPY, \ 88 struct uffdio_copy) 89 #define UFFDIO_ZEROPAGE _IOWR(UFFDIO, _UFFDIO_ZEROPAGE, \ 90 struct uffdio_zeropage) 91 #define UFFDIO_WRITEPROTECT _IOWR(UFFDIO, _UFFDIO_WRITEPROTECT, \ 92 struct uffdio_writeprotect) 93 #define UFFDIO_CONTINUE _IOWR(UFFDIO, _UFFDIO_CONTINUE, \ 94 struct uffdio_continue) 95 #define UFFDIO_POISON _IOWR(UFFDIO, _UFFDIO_POISON, \ 96 struct uffdio_poison) 97 98 /* read() structure */ 99 struct uffd_msg { 100 __u8 event; 101 102 __u8 reserved1; 103 __u16 reserved2; 104 __u32 reserved3; 105 106 union { 107 struct { 108 __u64 flags; 109 __u64 address; 110 union { 111 __u32 ptid; 112 } feat; 113 } pagefault; 114 115 struct { 116 __u32 ufd; 117 } fork; 118 119 struct { 120 __u64 from; 121 __u64 to; 122 __u64 len; 123 } remap; 124 125 struct { 126 __u64 start; 127 __u64 end; 128 } remove; 129 130 struct { 131 /* unused reserved fields */ 132 __u64 reserved1; 133 __u64 reserved2; 134 __u64 reserved3; 135 } reserved; 136 } arg; 137 } __packed; 138 139 /* 140 * Start at 0x12 and not at 0 to be more strict against bugs. 141 */ 142 #define UFFD_EVENT_PAGEFAULT 0x12 143 #define UFFD_EVENT_FORK 0x13 144 #define UFFD_EVENT_REMAP 0x14 145 #define UFFD_EVENT_REMOVE 0x15 146 #define UFFD_EVENT_UNMAP 0x16 147 148 /* flags for UFFD_EVENT_PAGEFAULT */ 149 #define UFFD_PAGEFAULT_FLAG_WRITE (1<<0) /* If this was a write fault */ 150 #define UFFD_PAGEFAULT_FLAG_WP (1<<1) /* If reason is VM_UFFD_WP */ 151 #define UFFD_PAGEFAULT_FLAG_MINOR (1<<2) /* If reason is VM_UFFD_MINOR */ 152 153 struct uffdio_api { 154 /* userland asks for an API number and the features to enable */ 155 __u64 api; 156 /* 157 * Kernel answers below with the all available features for 158 * the API, this notifies userland of which events and/or 159 * which flags for each event are enabled in the current 160 * kernel. 161 * 162 * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE 163 * are to be considered implicitly always enabled in all kernels as 164 * long as the uffdio_api.api requested matches UFFD_API. 165 * 166 * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER 167 * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on 168 * hugetlbfs virtual memory ranges. Adding or not adding 169 * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has 170 * no real functional effect after UFFDIO_API returns, but 171 * it's only useful for an initial feature set probe at 172 * UFFDIO_API time. There are two ways to use it: 173 * 174 * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the 175 * uffdio_api.features before calling UFFDIO_API, an error 176 * will be returned by UFFDIO_API on a kernel without 177 * hugetlbfs missing support 178 * 179 * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in 180 * uffdio_api.features and instead it will be set by the 181 * kernel in the uffdio_api.features if the kernel supports 182 * it, so userland can later check if the feature flag is 183 * present in uffdio_api.features after UFFDIO_API 184 * succeeded. 185 * 186 * UFFD_FEATURE_MISSING_SHMEM works the same as 187 * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem 188 * (i.e. tmpfs and other shmem based APIs). 189 * 190 * UFFD_FEATURE_SIGBUS feature means no page-fault 191 * (UFFD_EVENT_PAGEFAULT) event will be delivered, instead 192 * a SIGBUS signal will be sent to the faulting process. 193 * 194 * UFFD_FEATURE_THREAD_ID pid of the page faulted task_struct will 195 * be returned, if feature is not requested 0 will be returned. 196 * 197 * UFFD_FEATURE_MINOR_HUGETLBFS indicates that minor faults 198 * can be intercepted (via REGISTER_MODE_MINOR) for 199 * hugetlbfs-backed pages. 200 * 201 * UFFD_FEATURE_MINOR_SHMEM indicates the same support as 202 * UFFD_FEATURE_MINOR_HUGETLBFS, but for shmem-backed pages instead. 203 * 204 * UFFD_FEATURE_EXACT_ADDRESS indicates that the exact address of page 205 * faults would be provided and the offset within the page would not be 206 * masked. 207 * 208 * UFFD_FEATURE_WP_HUGETLBFS_SHMEM indicates that userfaultfd 209 * write-protection mode is supported on both shmem and hugetlbfs. 210 * 211 * UFFD_FEATURE_WP_UNPOPULATED indicates that userfaultfd 212 * write-protection mode will always apply to unpopulated pages 213 * (i.e. empty ptes). This will be the default behavior for shmem 214 * & hugetlbfs, so this flag only affects anonymous memory behavior 215 * when userfault write-protection mode is registered. 216 */ 217 #define UFFD_FEATURE_PAGEFAULT_FLAG_WP (1<<0) 218 #define UFFD_FEATURE_EVENT_FORK (1<<1) 219 #define UFFD_FEATURE_EVENT_REMAP (1<<2) 220 #define UFFD_FEATURE_EVENT_REMOVE (1<<3) 221 #define UFFD_FEATURE_MISSING_HUGETLBFS (1<<4) 222 #define UFFD_FEATURE_MISSING_SHMEM (1<<5) 223 #define UFFD_FEATURE_EVENT_UNMAP (1<<6) 224 #define UFFD_FEATURE_SIGBUS (1<<7) 225 #define UFFD_FEATURE_THREAD_ID (1<<8) 226 #define UFFD_FEATURE_MINOR_HUGETLBFS (1<<9) 227 #define UFFD_FEATURE_MINOR_SHMEM (1<<10) 228 #define UFFD_FEATURE_EXACT_ADDRESS (1<<11) 229 #define UFFD_FEATURE_WP_HUGETLBFS_SHMEM (1<<12) 230 #define UFFD_FEATURE_WP_UNPOPULATED (1<<13) 231 #define UFFD_FEATURE_POISON (1<<14) 232 __u64 features; 233 234 __u64 ioctls; 235 }; 236 237 struct uffdio_range { 238 __u64 start; 239 __u64 len; 240 }; 241 242 struct uffdio_register { 243 struct uffdio_range range; 244 #define UFFDIO_REGISTER_MODE_MISSING ((__u64)1<<0) 245 #define UFFDIO_REGISTER_MODE_WP ((__u64)1<<1) 246 #define UFFDIO_REGISTER_MODE_MINOR ((__u64)1<<2) 247 __u64 mode; 248 249 /* 250 * kernel answers which ioctl commands are available for the 251 * range, keep at the end as the last 8 bytes aren't read. 252 */ 253 __u64 ioctls; 254 }; 255 256 struct uffdio_copy { 257 __u64 dst; 258 __u64 src; 259 __u64 len; 260 #define UFFDIO_COPY_MODE_DONTWAKE ((__u64)1<<0) 261 /* 262 * UFFDIO_COPY_MODE_WP will map the page write protected on 263 * the fly. UFFDIO_COPY_MODE_WP is available only if the 264 * write protected ioctl is implemented for the range 265 * according to the uffdio_register.ioctls. 266 */ 267 #define UFFDIO_COPY_MODE_WP ((__u64)1<<1) 268 __u64 mode; 269 270 /* 271 * "copy" is written by the ioctl and must be at the end: the 272 * copy_from_user will not read the last 8 bytes. 273 */ 274 __s64 copy; 275 }; 276 277 struct uffdio_zeropage { 278 struct uffdio_range range; 279 #define UFFDIO_ZEROPAGE_MODE_DONTWAKE ((__u64)1<<0) 280 __u64 mode; 281 282 /* 283 * "zeropage" is written by the ioctl and must be at the end: 284 * the copy_from_user will not read the last 8 bytes. 285 */ 286 __s64 zeropage; 287 }; 288 289 struct uffdio_writeprotect { 290 struct uffdio_range range; 291 /* 292 * UFFDIO_WRITEPROTECT_MODE_WP: set the flag to write protect a range, 293 * unset the flag to undo protection of a range which was previously 294 * write protected. 295 * 296 * UFFDIO_WRITEPROTECT_MODE_DONTWAKE: set the flag to avoid waking up 297 * any wait thread after the operation succeeds. 298 * 299 * NOTE: Write protecting a region (WP=1) is unrelated to page faults, 300 * therefore DONTWAKE flag is meaningless with WP=1. Removing write 301 * protection (WP=0) in response to a page fault wakes the faulting 302 * task unless DONTWAKE is set. 303 */ 304 #define UFFDIO_WRITEPROTECT_MODE_WP ((__u64)1<<0) 305 #define UFFDIO_WRITEPROTECT_MODE_DONTWAKE ((__u64)1<<1) 306 __u64 mode; 307 }; 308 309 struct uffdio_continue { 310 struct uffdio_range range; 311 #define UFFDIO_CONTINUE_MODE_DONTWAKE ((__u64)1<<0) 312 /* 313 * UFFDIO_CONTINUE_MODE_WP will map the page write protected on 314 * the fly. UFFDIO_CONTINUE_MODE_WP is available only if the 315 * write protected ioctl is implemented for the range 316 * according to the uffdio_register.ioctls. 317 */ 318 #define UFFDIO_CONTINUE_MODE_WP ((__u64)1<<1) 319 __u64 mode; 320 321 /* 322 * Fields below here are written by the ioctl and must be at the end: 323 * the copy_from_user will not read past here. 324 */ 325 __s64 mapped; 326 }; 327 328 struct uffdio_poison { 329 struct uffdio_range range; 330 #define UFFDIO_POISON_MODE_DONTWAKE ((__u64)1<<0) 331 __u64 mode; 332 333 /* 334 * Fields below here are written by the ioctl and must be at the end: 335 * the copy_from_user will not read past here. 336 */ 337 __s64 updated; 338 }; 339 340 /* 341 * Flags for the userfaultfd(2) system call itself. 342 */ 343 344 /* 345 * Create a userfaultfd that can handle page faults only in user mode. 346 */ 347 #define UFFD_USER_MODE_ONLY 1 348 349 #endif /* _LINUX_USERFAULTFD_H */ 350