userfaultfd.h 9.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318
  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. #ifndef _LINUX_USERFAULTFD_H
  10. #define _LINUX_USERFAULTFD_H
  11. #include <linux/types.h>
  12. /* ioctls for /dev/userfaultfd */
  13. #define USERFAULTFD_IOC 0xAA
  14. #define USERFAULTFD_IOC_NEW _IO(USERFAULTFD_IOC, 0x00)
  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. * Valid ioctl command number range with this API is from 0x00 to
  55. * 0x3F. UFFDIO_API is the fixed number, everything else can be
  56. * changed by implementing a different UFFD_API. If sticking to the
  57. * same UFFD_API more ioctl can be added and userland will be aware of
  58. * which ioctl the running kernel implements through the ioctl command
  59. * bitmask written by the UFFDIO_API.
  60. */
  61. #define _UFFDIO_REGISTER (0x00)
  62. #define _UFFDIO_UNREGISTER (0x01)
  63. #define _UFFDIO_WAKE (0x02)
  64. #define _UFFDIO_COPY (0x03)
  65. #define _UFFDIO_ZEROPAGE (0x04)
  66. #define _UFFDIO_WRITEPROTECT (0x06)
  67. #define _UFFDIO_CONTINUE (0x07)
  68. #define _UFFDIO_API (0x3F)
  69. /* userfaultfd ioctl ids */
  70. #define UFFDIO 0xAA
  71. #define UFFDIO_API _IOWR(UFFDIO, _UFFDIO_API, \
  72. struct uffdio_api)
  73. #define UFFDIO_REGISTER _IOWR(UFFDIO, _UFFDIO_REGISTER, \
  74. struct uffdio_register)
  75. #define UFFDIO_UNREGISTER _IOR(UFFDIO, _UFFDIO_UNREGISTER, \
  76. struct uffdio_range)
  77. #define UFFDIO_WAKE _IOR(UFFDIO, _UFFDIO_WAKE, \
  78. struct uffdio_range)
  79. #define UFFDIO_COPY _IOWR(UFFDIO, _UFFDIO_COPY, \
  80. struct uffdio_copy)
  81. #define UFFDIO_ZEROPAGE _IOWR(UFFDIO, _UFFDIO_ZEROPAGE, \
  82. struct uffdio_zeropage)
  83. #define UFFDIO_WRITEPROTECT _IOWR(UFFDIO, _UFFDIO_WRITEPROTECT, \
  84. struct uffdio_writeprotect)
  85. #define UFFDIO_CONTINUE _IOWR(UFFDIO, _UFFDIO_CONTINUE, \
  86. struct uffdio_continue)
  87. /* read() structure */
  88. struct uffd_msg {
  89. __u8 event;
  90. __u8 reserved1;
  91. __u16 reserved2;
  92. __u32 reserved3;
  93. union {
  94. struct {
  95. __u64 flags;
  96. __u64 address;
  97. union {
  98. __u32 ptid;
  99. } feat;
  100. } pagefault;
  101. struct {
  102. __u32 ufd;
  103. } fork;
  104. struct {
  105. __u64 from;
  106. __u64 to;
  107. __u64 len;
  108. } remap;
  109. struct {
  110. __u64 start;
  111. __u64 end;
  112. } remove;
  113. struct {
  114. /* unused reserved fields */
  115. __u64 reserved1;
  116. __u64 reserved2;
  117. __u64 reserved3;
  118. } reserved;
  119. } arg;
  120. } __packed;
  121. /*
  122. * Start at 0x12 and not at 0 to be more strict against bugs.
  123. */
  124. #define UFFD_EVENT_PAGEFAULT 0x12
  125. #define UFFD_EVENT_FORK 0x13
  126. #define UFFD_EVENT_REMAP 0x14
  127. #define UFFD_EVENT_REMOVE 0x15
  128. #define UFFD_EVENT_UNMAP 0x16
  129. /* flags for UFFD_EVENT_PAGEFAULT */
  130. #define UFFD_PAGEFAULT_FLAG_WRITE (1<<0) /* If this was a write fault */
  131. #define UFFD_PAGEFAULT_FLAG_WP (1<<1) /* If reason is VM_UFFD_WP */
  132. #define UFFD_PAGEFAULT_FLAG_MINOR (1<<2) /* If reason is VM_UFFD_MINOR */
  133. struct uffdio_api {
  134. /* userland asks for an API number and the features to enable */
  135. __u64 api;
  136. /*
  137. * Kernel answers below with the all available features for
  138. * the API, this notifies userland of which events and/or
  139. * which flags for each event are enabled in the current
  140. * kernel.
  141. *
  142. * Note: UFFD_EVENT_PAGEFAULT and UFFD_PAGEFAULT_FLAG_WRITE
  143. * are to be considered implicitly always enabled in all kernels as
  144. * long as the uffdio_api.api requested matches UFFD_API.
  145. *
  146. * UFFD_FEATURE_MISSING_HUGETLBFS means an UFFDIO_REGISTER
  147. * with UFFDIO_REGISTER_MODE_MISSING mode will succeed on
  148. * hugetlbfs virtual memory ranges. Adding or not adding
  149. * UFFD_FEATURE_MISSING_HUGETLBFS to uffdio_api.features has
  150. * no real functional effect after UFFDIO_API returns, but
  151. * it's only useful for an initial feature set probe at
  152. * UFFDIO_API time. There are two ways to use it:
  153. *
  154. * 1) by adding UFFD_FEATURE_MISSING_HUGETLBFS to the
  155. * uffdio_api.features before calling UFFDIO_API, an error
  156. * will be returned by UFFDIO_API on a kernel without
  157. * hugetlbfs missing support
  158. *
  159. * 2) the UFFD_FEATURE_MISSING_HUGETLBFS can not be added in
  160. * uffdio_api.features and instead it will be set by the
  161. * kernel in the uffdio_api.features if the kernel supports
  162. * it, so userland can later check if the feature flag is
  163. * present in uffdio_api.features after UFFDIO_API
  164. * succeeded.
  165. *
  166. * UFFD_FEATURE_MISSING_SHMEM works the same as
  167. * UFFD_FEATURE_MISSING_HUGETLBFS, but it applies to shmem
  168. * (i.e. tmpfs and other shmem based APIs).
  169. *
  170. * UFFD_FEATURE_SIGBUS feature means no page-fault
  171. * (UFFD_EVENT_PAGEFAULT) event will be delivered, instead
  172. * a SIGBUS signal will be sent to the faulting process.
  173. *
  174. * UFFD_FEATURE_THREAD_ID pid of the page faulted task_struct will
  175. * be returned, if feature is not requested 0 will be returned.
  176. *
  177. * UFFD_FEATURE_MINOR_HUGETLBFS indicates that minor faults
  178. * can be intercepted (via REGISTER_MODE_MINOR) for
  179. * hugetlbfs-backed pages.
  180. *
  181. * UFFD_FEATURE_MINOR_SHMEM indicates the same support as
  182. * UFFD_FEATURE_MINOR_HUGETLBFS, but for shmem-backed pages instead.
  183. *
  184. * UFFD_FEATURE_EXACT_ADDRESS indicates that the exact address of page
  185. * faults would be provided and the offset within the page would not be
  186. * masked.
  187. *
  188. * UFFD_FEATURE_WP_HUGETLBFS_SHMEM indicates that userfaultfd
  189. * write-protection mode is supported on both shmem and hugetlbfs.
  190. */
  191. #define UFFD_FEATURE_PAGEFAULT_FLAG_WP (1<<0)
  192. #define UFFD_FEATURE_EVENT_FORK (1<<1)
  193. #define UFFD_FEATURE_EVENT_REMAP (1<<2)
  194. #define UFFD_FEATURE_EVENT_REMOVE (1<<3)
  195. #define UFFD_FEATURE_MISSING_HUGETLBFS (1<<4)
  196. #define UFFD_FEATURE_MISSING_SHMEM (1<<5)
  197. #define UFFD_FEATURE_EVENT_UNMAP (1<<6)
  198. #define UFFD_FEATURE_SIGBUS (1<<7)
  199. #define UFFD_FEATURE_THREAD_ID (1<<8)
  200. #define UFFD_FEATURE_MINOR_HUGETLBFS (1<<9)
  201. #define UFFD_FEATURE_MINOR_SHMEM (1<<10)
  202. #define UFFD_FEATURE_EXACT_ADDRESS (1<<11)
  203. #define UFFD_FEATURE_WP_HUGETLBFS_SHMEM (1<<12)
  204. __u64 features;
  205. __u64 ioctls;
  206. };
  207. struct uffdio_range {
  208. __u64 start;
  209. __u64 len;
  210. };
  211. struct uffdio_register {
  212. struct uffdio_range range;
  213. #define UFFDIO_REGISTER_MODE_MISSING ((__u64)1<<0)
  214. #define UFFDIO_REGISTER_MODE_WP ((__u64)1<<1)
  215. #define UFFDIO_REGISTER_MODE_MINOR ((__u64)1<<2)
  216. __u64 mode;
  217. /*
  218. * kernel answers which ioctl commands are available for the
  219. * range, keep at the end as the last 8 bytes aren't read.
  220. */
  221. __u64 ioctls;
  222. };
  223. struct uffdio_copy {
  224. __u64 dst;
  225. __u64 src;
  226. __u64 len;
  227. #define UFFDIO_COPY_MODE_DONTWAKE ((__u64)1<<0)
  228. /*
  229. * UFFDIO_COPY_MODE_WP will map the page write protected on
  230. * the fly. UFFDIO_COPY_MODE_WP is available only if the
  231. * write protected ioctl is implemented for the range
  232. * according to the uffdio_register.ioctls.
  233. */
  234. #define UFFDIO_COPY_MODE_WP ((__u64)1<<1)
  235. __u64 mode;
  236. /*
  237. * "copy" is written by the ioctl and must be at the end: the
  238. * copy_from_user will not read the last 8 bytes.
  239. */
  240. __s64 copy;
  241. };
  242. struct uffdio_zeropage {
  243. struct uffdio_range range;
  244. #define UFFDIO_ZEROPAGE_MODE_DONTWAKE ((__u64)1<<0)
  245. __u64 mode;
  246. /*
  247. * "zeropage" is written by the ioctl and must be at the end:
  248. * the copy_from_user will not read the last 8 bytes.
  249. */
  250. __s64 zeropage;
  251. };
  252. struct uffdio_writeprotect {
  253. struct uffdio_range range;
  254. /*
  255. * UFFDIO_WRITEPROTECT_MODE_WP: set the flag to write protect a range,
  256. * unset the flag to undo protection of a range which was previously
  257. * write protected.
  258. *
  259. * UFFDIO_WRITEPROTECT_MODE_DONTWAKE: set the flag to avoid waking up
  260. * any wait thread after the operation succeeds.
  261. *
  262. * NOTE: Write protecting a region (WP=1) is unrelated to page faults,
  263. * therefore DONTWAKE flag is meaningless with WP=1. Removing write
  264. * protection (WP=0) in response to a page fault wakes the faulting
  265. * task unless DONTWAKE is set.
  266. */
  267. #define UFFDIO_WRITEPROTECT_MODE_WP ((__u64)1<<0)
  268. #define UFFDIO_WRITEPROTECT_MODE_DONTWAKE ((__u64)1<<1)
  269. __u64 mode;
  270. };
  271. struct uffdio_continue {
  272. struct uffdio_range range;
  273. #define UFFDIO_CONTINUE_MODE_DONTWAKE ((__u64)1<<0)
  274. __u64 mode;
  275. /*
  276. * Fields below here are written by the ioctl and must be at the end:
  277. * the copy_from_user will not read past here.
  278. */
  279. __s64 mapped;
  280. };
  281. /*
  282. * Flags for the userfaultfd(2) system call itself.
  283. */
  284. /*
  285. * Create a userfaultfd that can handle page faults only in user mode.
  286. */
  287. #define UFFD_USER_MODE_ONLY 1
  288. #endif /* _LINUX_USERFAULTFD_H */