1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * include/linux/userfaultfd_k.h 4 * 5 * Copyright (C) 2015 Red Hat, Inc. 6 * 7 */ 8 9 #ifndef _LINUX_USERFAULTFD_K_H 10 #define _LINUX_USERFAULTFD_K_H 11 12 #ifdef CONFIG_USERFAULTFD 13 14 #include <linux/userfaultfd.h> /* linux/include/uapi/linux/userfaultfd.h */ 15 16 #include <linux/fcntl.h> 17 #include <linux/mm.h> 18 #include <linux/swap.h> 19 #include <linux/swapops.h> 20 #include <asm-generic/pgtable_uffd.h> 21 #include <linux/hugetlb_inline.h> 22 23 /* The set of all possible UFFD-related VM flags. */ 24 #define __VM_UFFD_FLAGS (VM_UFFD_MISSING | VM_UFFD_WP | VM_UFFD_MINOR) 25 26 /* 27 * CAREFUL: Check include/uapi/asm-generic/fcntl.h when defining 28 * new flags, since they might collide with O_* ones. We want 29 * to re-use O_* flags that couldn't possibly have a meaning 30 * from userfaultfd, in order to leave a free define-space for 31 * shared O_* flags. 32 */ 33 #define UFFD_CLOEXEC O_CLOEXEC 34 #define UFFD_NONBLOCK O_NONBLOCK 35 36 #define UFFD_SHARED_FCNTL_FLAGS (O_CLOEXEC | O_NONBLOCK) 37 #define UFFD_FLAGS_SET (EFD_SHARED_FCNTL_FLAGS) 38 39 /* 40 * Start with fault_pending_wqh and fault_wqh so they're more likely 41 * to be in the same cacheline. 42 * 43 * Locking order: 44 * fd_wqh.lock 45 * fault_pending_wqh.lock 46 * fault_wqh.lock 47 * event_wqh.lock 48 * 49 * To avoid deadlocks, IRQs must be disabled when taking any of the above locks, 50 * since fd_wqh.lock is taken by aio_poll() while it's holding a lock that's 51 * also taken in IRQ context. 52 */ 53 struct userfaultfd_ctx { 54 /* waitqueue head for the pending (i.e. not read) userfaults */ 55 wait_queue_head_t fault_pending_wqh; 56 /* waitqueue head for the userfaults */ 57 wait_queue_head_t fault_wqh; 58 /* waitqueue head for the pseudo fd to wakeup poll/read */ 59 wait_queue_head_t fd_wqh; 60 /* waitqueue head for events */ 61 wait_queue_head_t event_wqh; 62 /* a refile sequence protected by fault_pending_wqh lock */ 63 seqcount_spinlock_t refile_seq; 64 /* pseudo fd refcounting */ 65 refcount_t refcount; 66 /* userfaultfd syscall flags */ 67 unsigned int flags; 68 /* features requested from the userspace */ 69 unsigned int features; 70 /* released */ 71 bool released; 72 /* 73 * Prevents userfaultfd operations (fill/move/wp) from happening while 74 * some non-cooperative event(s) is taking place. Increments are done 75 * in write-mode. Whereas, userfaultfd operations, which includes 76 * reading mmap_changing, is done under read-mode. 77 */ 78 struct rw_semaphore map_changing_lock; 79 /* memory mappings are changing because of non-cooperative event */ 80 atomic_t mmap_changing; 81 /* mm with one ore more vmas attached to this userfaultfd_ctx */ 82 struct mm_struct *mm; 83 }; 84 85 extern vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason); 86 87 /* A combined operation mode + behavior flags. */ 88 typedef unsigned int __bitwise uffd_flags_t; 89 90 /* Mutually exclusive modes of operation. */ 91 enum mfill_atomic_mode { 92 MFILL_ATOMIC_COPY, 93 MFILL_ATOMIC_ZEROPAGE, 94 MFILL_ATOMIC_CONTINUE, 95 MFILL_ATOMIC_POISON, 96 NR_MFILL_ATOMIC_MODES, 97 }; 98 99 #define MFILL_ATOMIC_MODE_BITS (const_ilog2(NR_MFILL_ATOMIC_MODES - 1) + 1) 100 #define MFILL_ATOMIC_BIT(nr) BIT(MFILL_ATOMIC_MODE_BITS + (nr)) 101 #define MFILL_ATOMIC_FLAG(nr) ((__force uffd_flags_t) MFILL_ATOMIC_BIT(nr)) 102 #define MFILL_ATOMIC_MODE_MASK ((__force uffd_flags_t) (MFILL_ATOMIC_BIT(0) - 1)) 103 104 static inline bool uffd_flags_mode_is(uffd_flags_t flags, enum mfill_atomic_mode expected) 105 { 106 return (flags & MFILL_ATOMIC_MODE_MASK) == ((__force uffd_flags_t) expected); 107 } 108 109 static inline uffd_flags_t uffd_flags_set_mode(uffd_flags_t flags, enum mfill_atomic_mode mode) 110 { 111 flags &= ~MFILL_ATOMIC_MODE_MASK; 112 return flags | ((__force uffd_flags_t) mode); 113 } 114 115 /* Flags controlling behavior. These behavior changes are mode-independent. */ 116 #define MFILL_ATOMIC_WP MFILL_ATOMIC_FLAG(0) 117 118 extern int mfill_atomic_install_pte(pmd_t *dst_pmd, 119 struct vm_area_struct *dst_vma, 120 unsigned long dst_addr, struct page *page, 121 bool newly_allocated, uffd_flags_t flags); 122 123 extern ssize_t mfill_atomic_copy(struct userfaultfd_ctx *ctx, unsigned long dst_start, 124 unsigned long src_start, unsigned long len, 125 uffd_flags_t flags); 126 extern ssize_t mfill_atomic_zeropage(struct userfaultfd_ctx *ctx, 127 unsigned long dst_start, 128 unsigned long len); 129 extern ssize_t mfill_atomic_continue(struct userfaultfd_ctx *ctx, unsigned long dst_start, 130 unsigned long len, uffd_flags_t flags); 131 extern ssize_t mfill_atomic_poison(struct userfaultfd_ctx *ctx, unsigned long start, 132 unsigned long len, uffd_flags_t flags); 133 extern int mwriteprotect_range(struct userfaultfd_ctx *ctx, unsigned long start, 134 unsigned long len, bool enable_wp); 135 extern long uffd_wp_range(struct vm_area_struct *vma, 136 unsigned long start, unsigned long len, bool enable_wp); 137 138 /* move_pages */ 139 void double_pt_lock(spinlock_t *ptl1, spinlock_t *ptl2); 140 void double_pt_unlock(spinlock_t *ptl1, spinlock_t *ptl2); 141 ssize_t move_pages(struct userfaultfd_ctx *ctx, struct mm_struct *mm, 142 unsigned long dst_start, unsigned long src_start, 143 unsigned long len, __u64 flags); 144 int move_pages_huge_pmd(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd, pmd_t dst_pmdval, 145 struct vm_area_struct *dst_vma, 146 struct vm_area_struct *src_vma, 147 unsigned long dst_addr, unsigned long src_addr); 148 149 /* mm helpers */ 150 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma, 151 struct vm_userfaultfd_ctx vm_ctx) 152 { 153 return vma->vm_userfaultfd_ctx.ctx == vm_ctx.ctx; 154 } 155 156 /* 157 * Never enable huge pmd sharing on some uffd registered vmas: 158 * 159 * - VM_UFFD_WP VMAs, because write protect information is per pgtable entry. 160 * 161 * - VM_UFFD_MINOR VMAs, because otherwise we would never get minor faults for 162 * VMAs which share huge pmds. (If you have two mappings to the same 163 * underlying pages, and fault in the non-UFFD-registered one with a write, 164 * with huge pmd sharing this would *also* setup the second UFFD-registered 165 * mapping, and we'd not get minor faults.) 166 */ 167 static inline bool uffd_disable_huge_pmd_share(struct vm_area_struct *vma) 168 { 169 return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR); 170 } 171 172 /* 173 * Don't do fault around for either WP or MINOR registered uffd range. For 174 * MINOR registered range, fault around will be a total disaster and ptes can 175 * be installed without notifications; for WP it should mostly be fine as long 176 * as the fault around checks for pte_none() before the installation, however 177 * to be super safe we just forbid it. 178 */ 179 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma) 180 { 181 return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR); 182 } 183 184 static inline bool userfaultfd_missing(struct vm_area_struct *vma) 185 { 186 return vma->vm_flags & VM_UFFD_MISSING; 187 } 188 189 static inline bool userfaultfd_wp(struct vm_area_struct *vma) 190 { 191 return vma->vm_flags & VM_UFFD_WP; 192 } 193 194 static inline bool userfaultfd_minor(struct vm_area_struct *vma) 195 { 196 return vma->vm_flags & VM_UFFD_MINOR; 197 } 198 199 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma, 200 pte_t pte) 201 { 202 return userfaultfd_wp(vma) && pte_uffd_wp(pte); 203 } 204 205 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma, 206 pmd_t pmd) 207 { 208 return userfaultfd_wp(vma) && pmd_uffd_wp(pmd); 209 } 210 211 static inline bool userfaultfd_armed(struct vm_area_struct *vma) 212 { 213 return vma->vm_flags & __VM_UFFD_FLAGS; 214 } 215 216 static inline bool vma_can_userfault(struct vm_area_struct *vma, 217 unsigned long vm_flags, 218 bool wp_async) 219 { 220 vm_flags &= __VM_UFFD_FLAGS; 221 222 if ((vm_flags & VM_UFFD_MINOR) && 223 (!is_vm_hugetlb_page(vma) && !vma_is_shmem(vma))) 224 return false; 225 226 /* 227 * If wp async enabled, and WP is the only mode enabled, allow any 228 * memory type. 229 */ 230 if (wp_async && (vm_flags == VM_UFFD_WP)) 231 return true; 232 233 #ifndef CONFIG_PTE_MARKER_UFFD_WP 234 /* 235 * If user requested uffd-wp but not enabled pte markers for 236 * uffd-wp, then shmem & hugetlbfs are not supported but only 237 * anonymous. 238 */ 239 if ((vm_flags & VM_UFFD_WP) && !vma_is_anonymous(vma)) 240 return false; 241 #endif 242 243 /* By default, allow any of anon|shmem|hugetlb */ 244 return vma_is_anonymous(vma) || is_vm_hugetlb_page(vma) || 245 vma_is_shmem(vma); 246 } 247 248 extern int dup_userfaultfd(struct vm_area_struct *, struct list_head *); 249 extern void dup_userfaultfd_complete(struct list_head *); 250 251 extern void mremap_userfaultfd_prep(struct vm_area_struct *, 252 struct vm_userfaultfd_ctx *); 253 extern void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *, 254 unsigned long from, unsigned long to, 255 unsigned long len); 256 257 extern bool userfaultfd_remove(struct vm_area_struct *vma, 258 unsigned long start, 259 unsigned long end); 260 261 extern int userfaultfd_unmap_prep(struct vm_area_struct *vma, 262 unsigned long start, unsigned long end, struct list_head *uf); 263 extern void userfaultfd_unmap_complete(struct mm_struct *mm, 264 struct list_head *uf); 265 extern bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma); 266 extern bool userfaultfd_wp_async(struct vm_area_struct *vma); 267 268 #else /* CONFIG_USERFAULTFD */ 269 270 /* mm helpers */ 271 static inline vm_fault_t handle_userfault(struct vm_fault *vmf, 272 unsigned long reason) 273 { 274 return VM_FAULT_SIGBUS; 275 } 276 277 static inline long uffd_wp_range(struct vm_area_struct *vma, 278 unsigned long start, unsigned long len, 279 bool enable_wp) 280 { 281 return false; 282 } 283 284 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma, 285 struct vm_userfaultfd_ctx vm_ctx) 286 { 287 return true; 288 } 289 290 static inline bool userfaultfd_missing(struct vm_area_struct *vma) 291 { 292 return false; 293 } 294 295 static inline bool userfaultfd_wp(struct vm_area_struct *vma) 296 { 297 return false; 298 } 299 300 static inline bool userfaultfd_minor(struct vm_area_struct *vma) 301 { 302 return false; 303 } 304 305 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma, 306 pte_t pte) 307 { 308 return false; 309 } 310 311 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma, 312 pmd_t pmd) 313 { 314 return false; 315 } 316 317 318 static inline bool userfaultfd_armed(struct vm_area_struct *vma) 319 { 320 return false; 321 } 322 323 static inline int dup_userfaultfd(struct vm_area_struct *vma, 324 struct list_head *l) 325 { 326 return 0; 327 } 328 329 static inline void dup_userfaultfd_complete(struct list_head *l) 330 { 331 } 332 333 static inline void mremap_userfaultfd_prep(struct vm_area_struct *vma, 334 struct vm_userfaultfd_ctx *ctx) 335 { 336 } 337 338 static inline void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *ctx, 339 unsigned long from, 340 unsigned long to, 341 unsigned long len) 342 { 343 } 344 345 static inline bool userfaultfd_remove(struct vm_area_struct *vma, 346 unsigned long start, 347 unsigned long end) 348 { 349 return true; 350 } 351 352 static inline int userfaultfd_unmap_prep(struct vm_area_struct *vma, 353 unsigned long start, unsigned long end, 354 struct list_head *uf) 355 { 356 return 0; 357 } 358 359 static inline void userfaultfd_unmap_complete(struct mm_struct *mm, 360 struct list_head *uf) 361 { 362 } 363 364 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma) 365 { 366 return false; 367 } 368 369 static inline bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma) 370 { 371 return false; 372 } 373 374 static inline bool userfaultfd_wp_async(struct vm_area_struct *vma) 375 { 376 return false; 377 } 378 379 #endif /* CONFIG_USERFAULTFD */ 380 381 static inline bool userfaultfd_wp_use_markers(struct vm_area_struct *vma) 382 { 383 /* Only wr-protect mode uses pte markers */ 384 if (!userfaultfd_wp(vma)) 385 return false; 386 387 /* File-based uffd-wp always need markers */ 388 if (!vma_is_anonymous(vma)) 389 return true; 390 391 /* 392 * Anonymous uffd-wp only needs the markers if WP_UNPOPULATED 393 * enabled (to apply markers on zero pages). 394 */ 395 return userfaultfd_wp_unpopulated(vma); 396 } 397 398 static inline bool pte_marker_entry_uffd_wp(swp_entry_t entry) 399 { 400 #ifdef CONFIG_PTE_MARKER_UFFD_WP 401 return is_pte_marker_entry(entry) && 402 (pte_marker_get(entry) & PTE_MARKER_UFFD_WP); 403 #else 404 return false; 405 #endif 406 } 407 408 static inline bool pte_marker_uffd_wp(pte_t pte) 409 { 410 #ifdef CONFIG_PTE_MARKER_UFFD_WP 411 swp_entry_t entry; 412 413 if (!is_swap_pte(pte)) 414 return false; 415 416 entry = pte_to_swp_entry(pte); 417 418 return pte_marker_entry_uffd_wp(entry); 419 #else 420 return false; 421 #endif 422 } 423 424 /* 425 * Returns true if this is a swap pte and was uffd-wp wr-protected in either 426 * forms (pte marker or a normal swap pte), false otherwise. 427 */ 428 static inline bool pte_swp_uffd_wp_any(pte_t pte) 429 { 430 #ifdef CONFIG_PTE_MARKER_UFFD_WP 431 if (!is_swap_pte(pte)) 432 return false; 433 434 if (pte_swp_uffd_wp(pte)) 435 return true; 436 437 if (pte_marker_uffd_wp(pte)) 438 return true; 439 #endif 440 return false; 441 } 442 443 #endif /* _LINUX_USERFAULTFD_K_H */ 444