userfaultfd.h (9614B)
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 UFFD_FEATURE_EXACT_ADDRESS | \ 37 UFFD_FEATURE_WP_HUGETLBFS_SHMEM) 38 #define UFFD_API_IOCTLS \ 39 ((__u64)1 << _UFFDIO_REGISTER | \ 40 (__u64)1 << _UFFDIO_UNREGISTER | \ 41 (__u64)1 << _UFFDIO_API) 42 #define UFFD_API_RANGE_IOCTLS \ 43 ((__u64)1 << _UFFDIO_WAKE | \ 44 (__u64)1 << _UFFDIO_COPY | \ 45 (__u64)1 << _UFFDIO_ZEROPAGE | \ 46 (__u64)1 << _UFFDIO_WRITEPROTECT | \ 47 (__u64)1 << _UFFDIO_CONTINUE) 48 #define UFFD_API_RANGE_IOCTLS_BASIC \ 49 ((__u64)1 << _UFFDIO_WAKE | \ 50 (__u64)1 << _UFFDIO_COPY | \ 51 (__u64)1 << _UFFDIO_CONTINUE | \ 52 (__u64)1 << _UFFDIO_WRITEPROTECT) 53 54 /* 55 * Valid ioctl command number range with this API is from 0x00 to 56 * 0x3F. UFFDIO_API is the fixed number, everything else can be 57 * changed by implementing a different UFFD_API. If sticking to the 58 * same UFFD_API more ioctl can be added and userland will be aware of 59 * which ioctl the running kernel implements through the ioctl command 60 * bitmask written by the UFFDIO_API. 61 */ 62 #define _UFFDIO_REGISTER (0x00) 63 #define _UFFDIO_UNREGISTER (0x01) 64 #define _UFFDIO_WAKE (0x02) 65 #define _UFFDIO_COPY (0x03) 66 #define _UFFDIO_ZEROPAGE (0x04) 67 #define _UFFDIO_WRITEPROTECT (0x06) 68 #define _UFFDIO_CONTINUE (0x07) 69 #define _UFFDIO_API (0x3F) 70 71 /* userfaultfd ioctl ids */ 72 #define UFFDIO 0xAA 73 #define UFFDIO_API _IOWR(UFFDIO, _UFFDIO_API, \ 74 struct uffdio_api) 75 #define UFFDIO_REGISTER _IOWR(UFFDIO, _UFFDIO_REGISTER, \ 76 struct uffdio_register) 77 #define UFFDIO_UNREGISTER _IOR(UFFDIO, _UFFDIO_UNREGISTER, \ 78 struct uffdio_range) 79 #define UFFDIO_WAKE _IOR(UFFDIO, _UFFDIO_WAKE, \ 80 struct uffdio_range) 81 #define UFFDIO_COPY _IOWR(UFFDIO, _UFFDIO_COPY, \ 82 struct uffdio_copy) 83 #define UFFDIO_ZEROPAGE _IOWR(UFFDIO, _UFFDIO_ZEROPAGE, \ 84 struct uffdio_zeropage) 85 #define UFFDIO_WRITEPROTECT _IOWR(UFFDIO, _UFFDIO_WRITEPROTECT, \ 86 struct uffdio_writeprotect) 87 #define UFFDIO_CONTINUE _IOWR(UFFDIO, _UFFDIO_CONTINUE, \ 88 struct uffdio_continue) 89 90 /* read() structure */ 91 struct uffd_msg { 92 __u8 event; 93 94 __u8 reserved1; 95 __u16 reserved2; 96 __u32 reserved3; 97 98 union { 99 struct { 100 __u64 flags; 101 __u64 address; 102 union { 103 __u32 ptid; 104 } feat; 105 } pagefault; 106 107 struct { 108 __u32 ufd; 109 } fork; 110 111 struct { 112 __u64 from; 113 __u64 to; 114 __u64 len; 115 } remap; 116 117 struct { 118 __u64 start; 119 __u64 end; 120 } remove; 121 122 struct { 123 /* unused reserved fields */ 124 __u64 reserved1; 125 __u64 reserved2; 126 __u64 reserved3; 127 } reserved; 128 } arg; 129 } __attribute__((packed)); 130 131 /* 132 * Start at 0x12 and not at 0 to be more strict against bugs. 133 */ 134 #define UFFD_EVENT_PAGEFAULT 0x12 135 #define UFFD_EVENT_FORK 0x13 136 #define UFFD_EVENT_REMAP 0x14 137 #define UFFD_EVENT_REMOVE 0x15 138 #define UFFD_EVENT_UNMAP 0x16 139 140 /* flags for UFFD_EVENT_PAGEFAULT */ 141 #define UFFD_PAGEFAULT_FLAG_WRITE (1<<0) /* If this was a write fault */ 142 #define UFFD_PAGEFAULT_FLAG_WP (1<<1) /* If reason is VM_UFFD_WP */ 143 #define UFFD_PAGEFAULT_FLAG_MINOR (1<<2) /* If reason is VM_UFFD_MINOR */ 144 145 struct uffdio_api { 146 /* userland asks for an API number and the features to enable */ 147 __u64 api; 148 /* 149 * Kernel answers below with the all available features for 150 * the API, this notifies userland of which events and/or 151 * which flags for each event are enabled in the current 152 * kernel. 153 * 154 * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE 155 * are to be considered implicitly always enabled in all kernels as 156 * long as the uffdio_api.api requested matches UFFD_API. 157 * 158 * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER 159 * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on 160 * hugetlbfs virtual memory ranges. Adding or not adding 161 * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has 162 * no real functional effect after UFFDIO_API returns, but 163 * it's only useful for an initial feature set probe at 164 * UFFDIO_API time. There are two ways to use it: 165 * 166 * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the 167 * uffdio_api.features before calling UFFDIO_API, an error 168 * will be returned by UFFDIO_API on a kernel without 169 * hugetlbfs missing support 170 * 171 * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in 172 * uffdio_api.features and instead it will be set by the 173 * kernel in the uffdio_api.features if the kernel supports 174 * it, so userland can later check if the feature flag is 175 * present in uffdio_api.features after UFFDIO_API 176 * succeeded. 177 * 178 * UFFD_FEATURE_MISSING_SHMEM works the same as 179 * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem 180 * (i.e. tmpfs and other shmem based APIs). 181 * 182 * UFFD_FEATURE_SIGBUS feature means no page-fault 183 * (UFFD_EVENT_PAGEFAULT) event will be delivered, instead 184 * a SIGBUS signal will be sent to the faulting process. 185 * 186 * UFFD_FEATURE_THREAD_ID pid of the page faulted task_struct will 187 * be returned, if feature is not requested 0 will be returned. 188 * 189 * UFFD_FEATURE_MINOR_HUGETLBFS indicates that minor faults 190 * can be intercepted (via REGISTER_MODE_MINOR) for 191 * hugetlbfs-backed pages. 192 * 193 * UFFD_FEATURE_MINOR_SHMEM indicates the same support as 194 * UFFD_FEATURE_MINOR_HUGETLBFS, but for shmem-backed pages instead. 195 * 196 * UFFD_FEATURE_EXACT_ADDRESS indicates that the exact address of page 197 * faults would be provided and the offset within the page would not be 198 * masked. 199 * 200 * UFFD_FEATURE_WP_HUGETLBFS_SHMEM indicates that userfaultfd 201 * write-protection mode is supported on both shmem and hugetlbfs. 202 */ 203 #define UFFD_FEATURE_PAGEFAULT_FLAG_WP (1<<0) 204 #define UFFD_FEATURE_EVENT_FORK (1<<1) 205 #define UFFD_FEATURE_EVENT_REMAP (1<<2) 206 #define UFFD_FEATURE_EVENT_REMOVE (1<<3) 207 #define UFFD_FEATURE_MISSING_HUGETLBFS (1<<4) 208 #define UFFD_FEATURE_MISSING_SHMEM (1<<5) 209 #define UFFD_FEATURE_EVENT_UNMAP (1<<6) 210 #define UFFD_FEATURE_SIGBUS (1<<7) 211 #define UFFD_FEATURE_THREAD_ID (1<<8) 212 #define UFFD_FEATURE_MINOR_HUGETLBFS (1<<9) 213 #define UFFD_FEATURE_MINOR_SHMEM (1<<10) 214 #define UFFD_FEATURE_EXACT_ADDRESS (1<<11) 215 #define UFFD_FEATURE_WP_HUGETLBFS_SHMEM (1<<12) 216 __u64 features; 217 218 __u64 ioctls; 219 }; 220 221 struct uffdio_range { 222 __u64 start; 223 __u64 len; 224 }; 225 226 struct uffdio_register { 227 struct uffdio_range range; 228 #define UFFDIO_REGISTER_MODE_MISSING ((__u64)1<<0) 229 #define UFFDIO_REGISTER_MODE_WP ((__u64)1<<1) 230 #define UFFDIO_REGISTER_MODE_MINOR ((__u64)1<<2) 231 __u64 mode; 232 233 /* 234 * kernel answers which ioctl commands are available for the 235 * range, keep at the end as the last 8 bytes aren't read. 236 */ 237 __u64 ioctls; 238 }; 239 240 struct uffdio_copy { 241 __u64 dst; 242 __u64 src; 243 __u64 len; 244 #define UFFDIO_COPY_MODE_DONTWAKE ((__u64)1<<0) 245 /* 246 * UFFDIO_COPY_MODE_WP will map the page write protected on 247 * the fly. UFFDIO_COPY_MODE_WP is available only if the 248 * write protected ioctl is implemented for the range 249 * according to the uffdio_register.ioctls. 250 */ 251 #define UFFDIO_COPY_MODE_WP ((__u64)1<<1) 252 __u64 mode; 253 254 /* 255 * "copy" is written by the ioctl and must be at the end: the 256 * copy_from_user will not read the last 8 bytes. 257 */ 258 __s64 copy; 259 }; 260 261 struct uffdio_zeropage { 262 struct uffdio_range range; 263 #define UFFDIO_ZEROPAGE_MODE_DONTWAKE ((__u64)1<<0) 264 __u64 mode; 265 266 /* 267 * "zeropage" is written by the ioctl and must be at the end: 268 * the copy_from_user will not read the last 8 bytes. 269 */ 270 __s64 zeropage; 271 }; 272 273 struct uffdio_writeprotect { 274 struct uffdio_range range; 275 /* 276 * UFFDIO_WRITEPROTECT_MODE_WP: set the flag to write protect a range, 277 * unset the flag to undo protection of a range which was previously 278 * write protected. 279 * 280 * UFFDIO_WRITEPROTECT_MODE_DONTWAKE: set the flag to avoid waking up 281 * any wait thread after the operation succeeds. 282 * 283 * NOTE: Write protecting a region (WP=1) is unrelated to page faults, 284 * therefore DONTWAKE flag is meaningless with WP=1. Removing write 285 * protection (WP=0) in response to a page fault wakes the faulting 286 * task unless DONTWAKE is set. 287 */ 288 #define UFFDIO_WRITEPROTECT_MODE_WP ((__u64)1<<0) 289 #define UFFDIO_WRITEPROTECT_MODE_DONTWAKE ((__u64)1<<1) 290 __u64 mode; 291 }; 292 293 struct uffdio_continue { 294 struct uffdio_range range; 295 #define UFFDIO_CONTINUE_MODE_DONTWAKE ((__u64)1<<0) 296 __u64 mode; 297 298 /* 299 * Fields below here are written by the ioctl and must be at the end: 300 * the copy_from_user will not read past here. 301 */ 302 __s64 mapped; 303 }; 304 305 /* 306 * Flags for the userfaultfd(2) system call itself. 307 */ 308 309 /* 310 * Create a userfaultfd that can handle page faults only in user mode. 311 */ 312 #define UFFD_USER_MODE_ONLY 1 313 314 #endif /* _LINUX_USERFAULTFD_H */