1 /* SPDX-License-Identifier: GPL-2.0 */
3 * include/linux/userfaultfd_k.h
5 * Copyright (C) 2015 Red Hat, Inc.
9 #ifndef _LINUX_USERFAULTFD_K_H
10 #define _LINUX_USERFAULTFD_K_H
12 #ifdef CONFIG_USERFAULTFD
14 #include <linux/userfaultfd.h> /* linux/include/uapi/linux/userfaultfd.h */
16 #include <linux/fcntl.h>
18 #include <linux/swap.h>
19 #include <linux/swapops.h>
20 #include <asm-generic/pgtable_uffd.h>
21 #include <linux/hugetlb_inline.h>
23 /* The set of all possible UFFD-related VM flags. */
24 #define __VM_UFFD_FLAGS (VM_UFFD_MISSING | VM_UFFD_WP | VM_UFFD_MINOR)
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
33 #define UFFD_CLOEXEC O_CLOEXEC
34 #define UFFD_NONBLOCK O_NONBLOCK
36 #define UFFD_SHARED_FCNTL_FLAGS (O_CLOEXEC | O_NONBLOCK)
37 #define UFFD_FLAGS_SET (EFD_SHARED_FCNTL_FLAGS)
39 extern vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason);
41 /* A combined operation mode + behavior flags. */
42 typedef unsigned int __bitwise uffd_flags_t;
44 /* Mutually exclusive modes of operation. */
45 enum mfill_atomic_mode {
47 MFILL_ATOMIC_ZEROPAGE,
48 MFILL_ATOMIC_CONTINUE,
50 NR_MFILL_ATOMIC_MODES,
53 #define MFILL_ATOMIC_MODE_BITS (const_ilog2(NR_MFILL_ATOMIC_MODES - 1) + 1)
54 #define MFILL_ATOMIC_BIT(nr) BIT(MFILL_ATOMIC_MODE_BITS + (nr))
55 #define MFILL_ATOMIC_FLAG(nr) ((__force uffd_flags_t) MFILL_ATOMIC_BIT(nr))
56 #define MFILL_ATOMIC_MODE_MASK ((__force uffd_flags_t) (MFILL_ATOMIC_BIT(0) - 1))
58 static inline bool uffd_flags_mode_is(uffd_flags_t flags, enum mfill_atomic_mode expected)
60 return (flags & MFILL_ATOMIC_MODE_MASK) == ((__force uffd_flags_t) expected);
63 static inline uffd_flags_t uffd_flags_set_mode(uffd_flags_t flags, enum mfill_atomic_mode mode)
65 flags &= ~MFILL_ATOMIC_MODE_MASK;
66 return flags | ((__force uffd_flags_t) mode);
69 /* Flags controlling behavior. These behavior changes are mode-independent. */
70 #define MFILL_ATOMIC_WP MFILL_ATOMIC_FLAG(0)
72 extern int mfill_atomic_install_pte(pmd_t *dst_pmd,
73 struct vm_area_struct *dst_vma,
74 unsigned long dst_addr, struct page *page,
75 bool newly_allocated, uffd_flags_t flags);
77 extern ssize_t mfill_atomic_copy(struct mm_struct *dst_mm, unsigned long dst_start,
78 unsigned long src_start, unsigned long len,
79 atomic_t *mmap_changing, uffd_flags_t flags);
80 extern ssize_t mfill_atomic_zeropage(struct mm_struct *dst_mm,
81 unsigned long dst_start,
83 atomic_t *mmap_changing);
84 extern ssize_t mfill_atomic_continue(struct mm_struct *dst_mm, unsigned long dst_start,
85 unsigned long len, atomic_t *mmap_changing,
87 extern ssize_t mfill_atomic_poison(struct mm_struct *dst_mm, unsigned long start,
88 unsigned long len, atomic_t *mmap_changing,
90 extern int mwriteprotect_range(struct mm_struct *dst_mm,
91 unsigned long start, unsigned long len,
92 bool enable_wp, atomic_t *mmap_changing);
93 extern long uffd_wp_range(struct vm_area_struct *vma,
94 unsigned long start, unsigned long len, bool enable_wp);
97 void double_pt_lock(spinlock_t *ptl1, spinlock_t *ptl2);
98 void double_pt_unlock(spinlock_t *ptl1, spinlock_t *ptl2);
99 ssize_t move_pages(struct userfaultfd_ctx *ctx, struct mm_struct *mm,
100 unsigned long dst_start, unsigned long src_start,
101 unsigned long len, __u64 flags);
102 int move_pages_huge_pmd(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd, pmd_t dst_pmdval,
103 struct vm_area_struct *dst_vma,
104 struct vm_area_struct *src_vma,
105 unsigned long dst_addr, unsigned long src_addr);
108 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
109 struct vm_userfaultfd_ctx vm_ctx)
111 return vma->vm_userfaultfd_ctx.ctx == vm_ctx.ctx;
115 * Never enable huge pmd sharing on some uffd registered vmas:
117 * - VM_UFFD_WP VMAs, because write protect information is per pgtable entry.
119 * - VM_UFFD_MINOR VMAs, because otherwise we would never get minor faults for
120 * VMAs which share huge pmds. (If you have two mappings to the same
121 * underlying pages, and fault in the non-UFFD-registered one with a write,
122 * with huge pmd sharing this would *also* setup the second UFFD-registered
123 * mapping, and we'd not get minor faults.)
125 static inline bool uffd_disable_huge_pmd_share(struct vm_area_struct *vma)
127 return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
131 * Don't do fault around for either WP or MINOR registered uffd range. For
132 * MINOR registered range, fault around will be a total disaster and ptes can
133 * be installed without notifications; for WP it should mostly be fine as long
134 * as the fault around checks for pte_none() before the installation, however
135 * to be super safe we just forbid it.
137 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
139 return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
142 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
144 return vma->vm_flags & VM_UFFD_MISSING;
147 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
149 return vma->vm_flags & VM_UFFD_WP;
152 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
154 return vma->vm_flags & VM_UFFD_MINOR;
157 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
160 return userfaultfd_wp(vma) && pte_uffd_wp(pte);
163 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
166 return userfaultfd_wp(vma) && pmd_uffd_wp(pmd);
169 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
171 return vma->vm_flags & __VM_UFFD_FLAGS;
174 static inline bool vma_can_userfault(struct vm_area_struct *vma,
175 unsigned long vm_flags,
178 vm_flags &= __VM_UFFD_FLAGS;
180 if ((vm_flags & VM_UFFD_MINOR) &&
181 (!is_vm_hugetlb_page(vma) && !vma_is_shmem(vma)))
185 * If wp async enabled, and WP is the only mode enabled, allow any
188 if (wp_async && (vm_flags == VM_UFFD_WP))
191 #ifndef CONFIG_PTE_MARKER_UFFD_WP
193 * If user requested uffd-wp but not enabled pte markers for
194 * uffd-wp, then shmem & hugetlbfs are not supported but only
197 if ((vm_flags & VM_UFFD_WP) && !vma_is_anonymous(vma))
201 /* By default, allow any of anon|shmem|hugetlb */
202 return vma_is_anonymous(vma) || is_vm_hugetlb_page(vma) ||
206 extern int dup_userfaultfd(struct vm_area_struct *, struct list_head *);
207 extern void dup_userfaultfd_complete(struct list_head *);
209 extern void mremap_userfaultfd_prep(struct vm_area_struct *,
210 struct vm_userfaultfd_ctx *);
211 extern void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *,
212 unsigned long from, unsigned long to,
215 extern bool userfaultfd_remove(struct vm_area_struct *vma,
219 extern int userfaultfd_unmap_prep(struct vm_area_struct *vma,
220 unsigned long start, unsigned long end, struct list_head *uf);
221 extern void userfaultfd_unmap_complete(struct mm_struct *mm,
222 struct list_head *uf);
223 extern bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma);
224 extern bool userfaultfd_wp_async(struct vm_area_struct *vma);
226 #else /* CONFIG_USERFAULTFD */
229 static inline vm_fault_t handle_userfault(struct vm_fault *vmf,
230 unsigned long reason)
232 return VM_FAULT_SIGBUS;
235 static inline long uffd_wp_range(struct vm_area_struct *vma,
236 unsigned long start, unsigned long len,
242 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
243 struct vm_userfaultfd_ctx vm_ctx)
248 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
253 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
258 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
263 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
269 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
276 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
281 static inline int dup_userfaultfd(struct vm_area_struct *vma,
287 static inline void dup_userfaultfd_complete(struct list_head *l)
291 static inline void mremap_userfaultfd_prep(struct vm_area_struct *vma,
292 struct vm_userfaultfd_ctx *ctx)
296 static inline void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *ctx,
303 static inline bool userfaultfd_remove(struct vm_area_struct *vma,
310 static inline int userfaultfd_unmap_prep(struct vm_area_struct *vma,
311 unsigned long start, unsigned long end,
312 struct list_head *uf)
317 static inline void userfaultfd_unmap_complete(struct mm_struct *mm,
318 struct list_head *uf)
322 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
327 static inline bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma)
332 static inline bool userfaultfd_wp_async(struct vm_area_struct *vma)
337 #endif /* CONFIG_USERFAULTFD */
339 static inline bool userfaultfd_wp_use_markers(struct vm_area_struct *vma)
341 /* Only wr-protect mode uses pte markers */
342 if (!userfaultfd_wp(vma))
345 /* File-based uffd-wp always need markers */
346 if (!vma_is_anonymous(vma))
350 * Anonymous uffd-wp only needs the markers if WP_UNPOPULATED
351 * enabled (to apply markers on zero pages).
353 return userfaultfd_wp_unpopulated(vma);
356 static inline bool pte_marker_entry_uffd_wp(swp_entry_t entry)
358 #ifdef CONFIG_PTE_MARKER_UFFD_WP
359 return is_pte_marker_entry(entry) &&
360 (pte_marker_get(entry) & PTE_MARKER_UFFD_WP);
366 static inline bool pte_marker_uffd_wp(pte_t pte)
368 #ifdef CONFIG_PTE_MARKER_UFFD_WP
371 if (!is_swap_pte(pte))
374 entry = pte_to_swp_entry(pte);
376 return pte_marker_entry_uffd_wp(entry);
383 * Returns true if this is a swap pte and was uffd-wp wr-protected in either
384 * forms (pte marker or a normal swap pte), false otherwise.
386 static inline bool pte_swp_uffd_wp_any(pte_t pte)
388 #ifdef CONFIG_PTE_MARKER_UFFD_WP
389 if (!is_swap_pte(pte))
392 if (pte_swp_uffd_wp(pte))
395 if (pte_marker_uffd_wp(pte))
401 #endif /* _LINUX_USERFAULTFD_K_H */