include/linux/userfaultfd_k.h

   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, unsigned long dst_start,
 142                    unsigned long src_start, unsigned long len, __u64 flags);
 143 int move_pages_huge_pmd(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd, pmd_t dst_pmdval,
 144                         struct vm_area_struct *dst_vma,
 145                         struct vm_area_struct *src_vma,
 146                         unsigned long dst_addr, unsigned long src_addr);
 147
 148 /* mm helpers */
 149 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
 150                                         struct vm_userfaultfd_ctx vm_ctx)
 151 {
 152         return vma->vm_userfaultfd_ctx.ctx == vm_ctx.ctx;
 153 }
 154
 155 /*
 156  * Never enable huge pmd sharing on some uffd registered vmas:
 157  *
 158  * - VM_UFFD_WP VMAs, because write protect information is per pgtable entry.
 159  *
 160  * - VM_UFFD_MINOR VMAs, because otherwise we would never get minor faults for
 161  *   VMAs which share huge pmds. (If you have two mappings to the same
 162  *   underlying pages, and fault in the non-UFFD-registered one with a write,
 163  *   with huge pmd sharing this would *also* setup the second UFFD-registered
 164  *   mapping, and we'd not get minor faults.)
 165  */
 166 static inline bool uffd_disable_huge_pmd_share(struct vm_area_struct *vma)
 167 {
 168         return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
 169 }
 170
 171 /*
 172  * Don't do fault around for either WP or MINOR registered uffd range.  For
 173  * MINOR registered range, fault around will be a total disaster and ptes can
 174  * be installed without notifications; for WP it should mostly be fine as long
 175  * as the fault around checks for pte_none() before the installation, however
 176  * to be super safe we just forbid it.
 177  */
 178 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
 179 {
 180         return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
 181 }
 182
 183 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
 184 {
 185         return vma->vm_flags & VM_UFFD_MISSING;
 186 }
 187
 188 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
 189 {
 190         return vma->vm_flags & VM_UFFD_WP;
 191 }
 192
 193 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
 194 {
 195         return vma->vm_flags & VM_UFFD_MINOR;
 196 }
 197
 198 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
 199                                       pte_t pte)
 200 {
 201         return userfaultfd_wp(vma) && pte_uffd_wp(pte);
 202 }
 203
 204 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
 205                                            pmd_t pmd)
 206 {
 207         return userfaultfd_wp(vma) && pmd_uffd_wp(pmd);
 208 }
 209
 210 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
 211 {
 212         return vma->vm_flags & __VM_UFFD_FLAGS;
 213 }
 214
 215 static inline bool vma_can_userfault(struct vm_area_struct *vma,
 216                                      unsigned long vm_flags,
 217                                      bool wp_async)
 218 {
 219         vm_flags &= __VM_UFFD_FLAGS;
 220
 221         if (vm_flags & VM_DROPPABLE)
 222                 return false;
 223
 224         if ((vm_flags & VM_UFFD_MINOR) &&
 225             (!is_vm_hugetlb_page(vma) && !vma_is_shmem(vma)))
 226                 return false;
 227
 228         /*
 229          * If wp async enabled, and WP is the only mode enabled, allow any
 230          * memory type.
 231          */
 232         if (wp_async && (vm_flags == VM_UFFD_WP))
 233                 return true;
 234
 235 #ifndef CONFIG_PTE_MARKER_UFFD_WP
 236         /*
 237          * If user requested uffd-wp but not enabled pte markers for
 238          * uffd-wp, then shmem & hugetlbfs are not supported but only
 239          * anonymous.
 240          */
 241         if ((vm_flags & VM_UFFD_WP) && !vma_is_anonymous(vma))
 242                 return false;
 243 #endif
 244
 245         /* By default, allow any of anon|shmem|hugetlb */
 246         return vma_is_anonymous(vma) || is_vm_hugetlb_page(vma) ||
 247             vma_is_shmem(vma);
 248 }
 249
 250 extern int dup_userfaultfd(struct vm_area_struct *, struct list_head *);
 251 extern void dup_userfaultfd_complete(struct list_head *);
 252 void dup_userfaultfd_fail(struct list_head *);
 253
 254 extern void mremap_userfaultfd_prep(struct vm_area_struct *,
 255                                     struct vm_userfaultfd_ctx *);
 256 extern void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *,
 257                                         unsigned long from, unsigned long to,
 258                                         unsigned long len);
 259
 260 extern bool userfaultfd_remove(struct vm_area_struct *vma,
 261                                unsigned long start,
 262                                unsigned long end);
 263
 264 extern int userfaultfd_unmap_prep(struct vm_area_struct *vma,
 265                 unsigned long start, unsigned long end, struct list_head *uf);
 266 extern void userfaultfd_unmap_complete(struct mm_struct *mm,
 267                                        struct list_head *uf);
 268 extern bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma);
 269 extern bool userfaultfd_wp_async(struct vm_area_struct *vma);
 270
 271 void userfaultfd_reset_ctx(struct vm_area_struct *vma);
 272
 273 struct vm_area_struct *userfaultfd_clear_vma(struct vma_iterator *vmi,
 274                                              struct vm_area_struct *prev,
 275                                              struct vm_area_struct *vma,
 276                                              unsigned long start,
 277                                              unsigned long end);
 278
 279 int userfaultfd_register_range(struct userfaultfd_ctx *ctx,
 280                                struct vm_area_struct *vma,
 281                                unsigned long vm_flags,
 282                                unsigned long start, unsigned long end,
 283                                bool wp_async);
 284
 285 void userfaultfd_release_new(struct userfaultfd_ctx *ctx);
 286
 287 void userfaultfd_release_all(struct mm_struct *mm,
 288                              struct userfaultfd_ctx *ctx);
 289
 290 #else /* CONFIG_USERFAULTFD */
 291
 292 /* mm helpers */
 293 static inline vm_fault_t handle_userfault(struct vm_fault *vmf,
 294                                 unsigned long reason)
 295 {
 296         return VM_FAULT_SIGBUS;
 297 }
 298
 299 static inline long uffd_wp_range(struct vm_area_struct *vma,
 300                                  unsigned long start, unsigned long len,
 301                                  bool enable_wp)
 302 {
 303         return false;
 304 }
 305
 306 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
 307                                         struct vm_userfaultfd_ctx vm_ctx)
 308 {
 309         return true;
 310 }
 311
 312 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
 313 {
 314         return false;
 315 }
 316
 317 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
 318 {
 319         return false;
 320 }
 321
 322 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
 323 {
 324         return false;
 325 }
 326
 327 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
 328                                       pte_t pte)
 329 {
 330         return false;
 331 }
 332
 333 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
 334                                            pmd_t pmd)
 335 {
 336         return false;
 337 }
 338
 339
 340 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
 341 {
 342         return false;
 343 }
 344
 345 static inline int dup_userfaultfd(struct vm_area_struct *vma,
 346                                   struct list_head *l)
 347 {
 348         return 0;
 349 }
 350
 351 static inline void dup_userfaultfd_complete(struct list_head *l)
 352 {
 353 }
 354
 355 static inline void dup_userfaultfd_fail(struct list_head *l)
 356 {
 357 }
 358
 359 static inline void mremap_userfaultfd_prep(struct vm_area_struct *vma,
 360                                            struct vm_userfaultfd_ctx *ctx)
 361 {
 362 }
 363
 364 static inline void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *ctx,
 365                                                unsigned long from,
 366                                                unsigned long to,
 367                                                unsigned long len)
 368 {
 369 }
 370
 371 static inline bool userfaultfd_remove(struct vm_area_struct *vma,
 372                                       unsigned long start,
 373                                       unsigned long end)
 374 {
 375         return true;
 376 }
 377
 378 static inline int userfaultfd_unmap_prep(struct vm_area_struct *vma,
 379                                          unsigned long start, unsigned long end,
 380                                          struct list_head *uf)
 381 {
 382         return 0;
 383 }
 384
 385 static inline void userfaultfd_unmap_complete(struct mm_struct *mm,
 386                                               struct list_head *uf)
 387 {
 388 }
 389
 390 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
 391 {
 392         return false;
 393 }
 394
 395 static inline bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma)
 396 {
 397         return false;
 398 }
 399
 400 static inline bool userfaultfd_wp_async(struct vm_area_struct *vma)
 401 {
 402         return false;
 403 }
 404
 405 #endif /* CONFIG_USERFAULTFD */
 406
 407 static inline bool userfaultfd_wp_use_markers(struct vm_area_struct *vma)
 408 {
 409         /* Only wr-protect mode uses pte markers */
 410         if (!userfaultfd_wp(vma))
 411                 return false;
 412
 413         /* File-based uffd-wp always need markers */
 414         if (!vma_is_anonymous(vma))
 415                 return true;
 416
 417         /*
 418          * Anonymous uffd-wp only needs the markers if WP_UNPOPULATED
 419          * enabled (to apply markers on zero pages).
 420          */
 421         return userfaultfd_wp_unpopulated(vma);
 422 }
 423
 424 static inline bool pte_marker_entry_uffd_wp(swp_entry_t entry)
 425 {
 426 #ifdef CONFIG_PTE_MARKER_UFFD_WP
 427         return is_pte_marker_entry(entry) &&
 428             (pte_marker_get(entry) & PTE_MARKER_UFFD_WP);
 429 #else
 430         return false;
 431 #endif
 432 }
 433
 434 static inline bool pte_marker_uffd_wp(pte_t pte)
 435 {
 436 #ifdef CONFIG_PTE_MARKER_UFFD_WP
 437         swp_entry_t entry;
 438
 439         if (!is_swap_pte(pte))
 440                 return false;
 441
 442         entry = pte_to_swp_entry(pte);
 443
 444         return pte_marker_entry_uffd_wp(entry);
 445 #else
 446         return false;
 447 #endif
 448 }
 449
 450 /*
 451  * Returns true if this is a swap pte and was uffd-wp wr-protected in either
 452  * forms (pte marker or a normal swap pte), false otherwise.
 453  */
 454 static inline bool pte_swp_uffd_wp_any(pte_t pte)
 455 {
 456 #ifdef CONFIG_PTE_MARKER_UFFD_WP
 457         if (!is_swap_pte(pte))
 458                 return false;
 459
 460         if (pte_swp_uffd_wp(pte))
 461                 return true;
 462
 463         if (pte_marker_uffd_wp(pte))
 464                 return true;
 465 #endif
 466         return false;
 467 }
 468
 469 #endif /* _LINUX_USERFAULTFD_K_H */