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