linux-headers/linux/userfaultfd.h

   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                            UFFD_FEATURE_POISON)
  44 #define UFFD_API_IOCTLS                         \
  45         ((__u64)1 << _UFFDIO_REGISTER |         \
  46          (__u64)1 << _UFFDIO_UNREGISTER |       \
  47          (__u64)1 << _UFFDIO_API)
  48 #define UFFD_API_RANGE_IOCTLS                   \
  49         ((__u64)1 << _UFFDIO_WAKE |             \
  50          (__u64)1 << _UFFDIO_COPY |             \
  51          (__u64)1 << _UFFDIO_ZEROPAGE |         \
  52          (__u64)1 << _UFFDIO_WRITEPROTECT |     \
  53          (__u64)1 << _UFFDIO_CONTINUE |         \
  54          (__u64)1 << _UFFDIO_POISON)
  55 #define UFFD_API_RANGE_IOCTLS_BASIC             \
  56         ((__u64)1 << _UFFDIO_WAKE |             \
  57          (__u64)1 << _UFFDIO_COPY |             \
  58          (__u64)1 << _UFFDIO_WRITEPROTECT |     \
  59          (__u64)1 << _UFFDIO_CONTINUE |         \
  60          (__u64)1 << _UFFDIO_POISON)
  61
  62 /*
  63  * Valid ioctl command number range with this API is from 0x00 to
  64  * 0x3F.  UFFDIO_API is the fixed number, everything else can be
  65  * changed by implementing a different UFFD_API. If sticking to the
  66  * same UFFD_API more ioctl can be added and userland will be aware of
  67  * which ioctl the running kernel implements through the ioctl command
  68  * bitmask written by the UFFDIO_API.
  69  */
  70 #define _UFFDIO_REGISTER                (0x00)
  71 #define _UFFDIO_UNREGISTER              (0x01)
  72 #define _UFFDIO_WAKE                    (0x02)
  73 #define _UFFDIO_COPY                    (0x03)
  74 #define _UFFDIO_ZEROPAGE                (0x04)
  75 #define _UFFDIO_WRITEPROTECT            (0x06)
  76 #define _UFFDIO_CONTINUE                (0x07)
  77 #define _UFFDIO_POISON                  (0x08)
  78 #define _UFFDIO_API                     (0x3F)
  79
  80 /* userfaultfd ioctl ids */
  81 #define UFFDIO 0xAA
  82 #define UFFDIO_API              _IOWR(UFFDIO, _UFFDIO_API,      \
  83                                       struct uffdio_api)
  84 #define UFFDIO_REGISTER         _IOWR(UFFDIO, _UFFDIO_REGISTER, \
  85                                       struct uffdio_register)
  86 #define UFFDIO_UNREGISTER       _IOR(UFFDIO, _UFFDIO_UNREGISTER,        \
  87                                      struct uffdio_range)
  88 #define UFFDIO_WAKE             _IOR(UFFDIO, _UFFDIO_WAKE,      \
  89                                      struct uffdio_range)
  90 #define UFFDIO_COPY             _IOWR(UFFDIO, _UFFDIO_COPY,     \
  91                                       struct uffdio_copy)
  92 #define UFFDIO_ZEROPAGE         _IOWR(UFFDIO, _UFFDIO_ZEROPAGE, \
  93                                       struct uffdio_zeropage)
  94 #define UFFDIO_WRITEPROTECT     _IOWR(UFFDIO, _UFFDIO_WRITEPROTECT, \
  95                                       struct uffdio_writeprotect)
  96 #define UFFDIO_CONTINUE         _IOWR(UFFDIO, _UFFDIO_CONTINUE, \
  97                                       struct uffdio_continue)
  98 #define UFFDIO_POISON           _IOWR(UFFDIO, _UFFDIO_POISON, \
  99                                       struct uffdio_poison)
 100
 101 /* read() structure */
 102 struct uffd_msg {
 103         __u8    event;
 104
 105         __u8    reserved1;
 106         __u16   reserved2;
 107         __u32   reserved3;
 108
 109         union {
 110                 struct {
 111                         __u64   flags;
 112                         __u64   address;
 113                         union {
 114                                 __u32 ptid;
 115                         } feat;
 116                 } pagefault;
 117
 118                 struct {
 119                         __u32   ufd;
 120                 } fork;
 121
 122                 struct {
 123                         __u64   from;
 124                         __u64   to;
 125                         __u64   len;
 126                 } remap;
 127
 128                 struct {
 129                         __u64   start;
 130                         __u64   end;
 131                 } remove;
 132
 133                 struct {
 134                         /* unused reserved fields */
 135                         __u64   reserved1;
 136                         __u64   reserved2;
 137                         __u64   reserved3;
 138                 } reserved;
 139         } arg;
 140 } __attribute__((packed));
 141
 142 /*
 143  * Start at 0x12 and not at 0 to be more strict against bugs.
 144  */
 145 #define UFFD_EVENT_PAGEFAULT    0x12
 146 #define UFFD_EVENT_FORK         0x13
 147 #define UFFD_EVENT_REMAP        0x14
 148 #define UFFD_EVENT_REMOVE       0x15
 149 #define UFFD_EVENT_UNMAP        0x16
 150
 151 /* flags for UFFD_EVENT_PAGEFAULT */
 152 #define UFFD_PAGEFAULT_FLAG_WRITE       (1<<0)  /* If this was a write fault */
 153 #define UFFD_PAGEFAULT_FLAG_WP          (1<<1)  /* If reason is VM_UFFD_WP */
 154 #define UFFD_PAGEFAULT_FLAG_MINOR       (1<<2)  /* If reason is VM_UFFD_MINOR */
 155
 156 struct uffdio_api {
 157         /* userland asks for an API number and the features to enable */
 158         __u64 api;
 159         /*
 160          * Kernel answers below with the all available features for
 161          * the API, this notifies userland of which events and/or
 162          * which flags for each event are enabled in the current
 163          * kernel.
 164          *
 165          * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE
 166          * are to be considered implicitly always enabled in all kernels as
 167          * long as the uffdio_api.api requested matches UFFD_API.
 168          *
 169          * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER
 170          * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on
 171          * hugetlbfs virtual memory ranges. Adding or not adding
 172          * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has
 173          * no real functional effect after UFFDIO_API returns, but
 174          * it's only useful for an initial feature set probe at
 175          * UFFDIO_API time. There are two ways to use it:
 176          *
 177          * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the
 178          *    uffdio_api.features before calling UFFDIO_API, an error
 179          *    will be returned by UFFDIO_API on a kernel without
 180          *    hugetlbfs missing support
 181          *
 182          * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in
 183          *    uffdio_api.features and instead it will be set by the
 184          *    kernel in the uffdio_api.features if the kernel supports
 185          *    it, so userland can later check if the feature flag is
 186          *    present in uffdio_api.features after UFFDIO_API
 187          *    succeeded.
 188          *
 189          * UFFD_FEATURE_MISSING_SHMEM works the same as
 190          * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem
 191          * (i.e. tmpfs and other shmem based APIs).
 192          *
 193          * UFFD_FEATURE_SIGBUS feature means no page-fault
 194          * (UFFD_EVENT_PAGEFAULT) event will be delivered, instead
 195          * a SIGBUS signal will be sent to the faulting process.
 196          *
 197          * UFFD_FEATURE_THREAD_ID pid of the page faulted task_struct will
 198          * be returned, if feature is not requested 0 will be returned.
 199          *
 200          * UFFD_FEATURE_MINOR_HUGETLBFS indicates that minor faults
 201          * can be intercepted (via REGISTER_MODE_MINOR) for
 202          * hugetlbfs-backed pages.
 203          *
 204          * UFFD_FEATURE_MINOR_SHMEM indicates the same support as
 205          * UFFD_FEATURE_MINOR_HUGETLBFS, but for shmem-backed pages instead.
 206          *
 207          * UFFD_FEATURE_EXACT_ADDRESS indicates that the exact address of page
 208          * faults would be provided and the offset within the page would not be
 209          * masked.
 210          *
 211          * UFFD_FEATURE_WP_HUGETLBFS_SHMEM indicates that userfaultfd
 212          * write-protection mode is supported on both shmem and hugetlbfs.
 213          *
 214          * UFFD_FEATURE_WP_UNPOPULATED indicates that userfaultfd
 215          * write-protection mode will always apply to unpopulated pages
 216          * (i.e. empty ptes).  This will be the default behavior for shmem
 217          * & hugetlbfs, so this flag only affects anonymous memory behavior
 218          * when userfault write-protection mode is registered.
 219          */
 220 #define UFFD_FEATURE_PAGEFAULT_FLAG_WP          (1<<0)
 221 #define UFFD_FEATURE_EVENT_FORK                 (1<<1)
 222 #define UFFD_FEATURE_EVENT_REMAP                (1<<2)
 223 #define UFFD_FEATURE_EVENT_REMOVE               (1<<3)
 224 #define UFFD_FEATURE_MISSING_HUGETLBFS          (1<<4)
 225 #define UFFD_FEATURE_MISSING_SHMEM              (1<<5)
 226 #define UFFD_FEATURE_EVENT_UNMAP                (1<<6)
 227 #define UFFD_FEATURE_SIGBUS                     (1<<7)
 228 #define UFFD_FEATURE_THREAD_ID                  (1<<8)
 229 #define UFFD_FEATURE_MINOR_HUGETLBFS            (1<<9)
 230 #define UFFD_FEATURE_MINOR_SHMEM                (1<<10)
 231 #define UFFD_FEATURE_EXACT_ADDRESS              (1<<11)
 232 #define UFFD_FEATURE_WP_HUGETLBFS_SHMEM         (1<<12)
 233 #define UFFD_FEATURE_WP_UNPOPULATED             (1<<13)
 234 #define UFFD_FEATURE_POISON                     (1<<14)
 235         __u64 features;
 236
 237         __u64 ioctls;
 238 };
 239
 240 struct uffdio_range {
 241         __u64 start;
 242         __u64 len;
 243 };
 244
 245 struct uffdio_register {
 246         struct uffdio_range range;
 247 #define UFFDIO_REGISTER_MODE_MISSING    ((__u64)1<<0)
 248 #define UFFDIO_REGISTER_MODE_WP         ((__u64)1<<1)
 249 #define UFFDIO_REGISTER_MODE_MINOR      ((__u64)1<<2)
 250         __u64 mode;
 251
 252         /*
 253          * kernel answers which ioctl commands are available for the
 254          * range, keep at the end as the last 8 bytes aren't read.
 255          */
 256         __u64 ioctls;
 257 };
 258
 259 struct uffdio_copy {
 260         __u64 dst;
 261         __u64 src;
 262         __u64 len;
 263 #define UFFDIO_COPY_MODE_DONTWAKE               ((__u64)1<<0)
 264         /*
 265          * UFFDIO_COPY_MODE_WP will map the page write protected on
 266          * the fly.  UFFDIO_COPY_MODE_WP is available only if the
 267          * write protected ioctl is implemented for the range
 268          * according to the uffdio_register.ioctls.
 269          */
 270 #define UFFDIO_COPY_MODE_WP                     ((__u64)1<<1)
 271         __u64 mode;
 272
 273         /*
 274          * "copy" is written by the ioctl and must be at the end: the
 275          * copy_from_user will not read the last 8 bytes.
 276          */
 277         __s64 copy;
 278 };
 279
 280 struct uffdio_zeropage {
 281         struct uffdio_range range;
 282 #define UFFDIO_ZEROPAGE_MODE_DONTWAKE           ((__u64)1<<0)
 283         __u64 mode;
 284
 285         /*
 286          * "zeropage" is written by the ioctl and must be at the end:
 287          * the copy_from_user will not read the last 8 bytes.
 288          */
 289         __s64 zeropage;
 290 };
 291
 292 struct uffdio_writeprotect {
 293         struct uffdio_range range;
 294 /*
 295  * UFFDIO_WRITEPROTECT_MODE_WP: set the flag to write protect a range,
 296  * unset the flag to undo protection of a range which was previously
 297  * write protected.
 298  *
 299  * UFFDIO_WRITEPROTECT_MODE_DONTWAKE: set the flag to avoid waking up
 300  * any wait thread after the operation succeeds.
 301  *
 302  * NOTE: Write protecting a region (WP=1) is unrelated to page faults,
 303  * therefore DONTWAKE flag is meaningless with WP=1.  Removing write
 304  * protection (WP=0) in response to a page fault wakes the faulting
 305  * task unless DONTWAKE is set.
 306  */
 307 #define UFFDIO_WRITEPROTECT_MODE_WP             ((__u64)1<<0)
 308 #define UFFDIO_WRITEPROTECT_MODE_DONTWAKE       ((__u64)1<<1)
 309         __u64 mode;
 310 };
 311
 312 struct uffdio_continue {
 313         struct uffdio_range range;
 314 #define UFFDIO_CONTINUE_MODE_DONTWAKE           ((__u64)1<<0)
 315         /*
 316          * UFFDIO_CONTINUE_MODE_WP will map the page write protected on
 317          * the fly.  UFFDIO_CONTINUE_MODE_WP is available only if the
 318          * write protected ioctl is implemented for the range
 319          * according to the uffdio_register.ioctls.
 320          */
 321 #define UFFDIO_CONTINUE_MODE_WP                 ((__u64)1<<1)
 322         __u64 mode;
 323
 324         /*
 325          * Fields below here are written by the ioctl and must be at the end:
 326          * the copy_from_user will not read past here.
 327          */
 328         __s64 mapped;
 329 };
 330
 331 struct uffdio_poison {
 332         struct uffdio_range range;
 333 #define UFFDIO_POISON_MODE_DONTWAKE             ((__u64)1<<0)
 334         __u64 mode;
 335
 336         /*
 337          * Fields below here are written by the ioctl and must be at the end:
 338          * the copy_from_user will not read past here.
 339          */
 340         __s64 updated;
 341 };
 342
 343 /*
 344  * Flags for the userfaultfd(2) system call itself.
 345  */
 346
 347 /*
 348  * Create a userfaultfd that can handle page faults only in user mode.
 349  */
 350 #define UFFD_USER_MODE_ONLY 1
 351
 352 #endif /* _LINUX_USERFAULTFD_H */