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 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
98 struct vm_userfaultfd_ctx vm_ctx)
100 return vma->vm_userfaultfd_ctx.ctx == vm_ctx.ctx;
104 * Never enable huge pmd sharing on some uffd registered vmas:
106 * - VM_UFFD_WP VMAs, because write protect information is per pgtable entry.
108 * - VM_UFFD_MINOR VMAs, because otherwise we would never get minor faults for
109 * VMAs which share huge pmds. (If you have two mappings to the same
110 * underlying pages, and fault in the non-UFFD-registered one with a write,
111 * with huge pmd sharing this would *also* setup the second UFFD-registered
112 * mapping, and we'd not get minor faults.)
114 static inline bool uffd_disable_huge_pmd_share(struct vm_area_struct *vma)
116 return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
120 * Don't do fault around for either WP or MINOR registered uffd range. For
121 * MINOR registered range, fault around will be a total disaster and ptes can
122 * be installed without notifications; for WP it should mostly be fine as long
123 * as the fault around checks for pte_none() before the installation, however
124 * to be super safe we just forbid it.
126 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
128 return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
131 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
133 return vma->vm_flags & VM_UFFD_MISSING;
136 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
138 return vma->vm_flags & VM_UFFD_WP;
141 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
143 return vma->vm_flags & VM_UFFD_MINOR;
146 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
149 return userfaultfd_wp(vma) && pte_uffd_wp(pte);
152 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
155 return userfaultfd_wp(vma) && pmd_uffd_wp(pmd);
158 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
160 return vma->vm_flags & __VM_UFFD_FLAGS;
163 static inline bool vma_can_userfault(struct vm_area_struct *vma,
164 unsigned long vm_flags,
167 vm_flags &= __VM_UFFD_FLAGS;
169 if ((vm_flags & VM_UFFD_MINOR) &&
170 (!is_vm_hugetlb_page(vma) && !vma_is_shmem(vma)))
174 * If wp async enabled, and WP is the only mode enabled, allow any
177 if (wp_async && (vm_flags == VM_UFFD_WP))
180 #ifndef CONFIG_PTE_MARKER_UFFD_WP
182 * If user requested uffd-wp but not enabled pte markers for
183 * uffd-wp, then shmem & hugetlbfs are not supported but only
186 if ((vm_flags & VM_UFFD_WP) && !vma_is_anonymous(vma))
190 /* By default, allow any of anon|shmem|hugetlb */
191 return vma_is_anonymous(vma) || is_vm_hugetlb_page(vma) ||
195 extern int dup_userfaultfd(struct vm_area_struct *, struct list_head *);
196 extern void dup_userfaultfd_complete(struct list_head *);
198 extern void mremap_userfaultfd_prep(struct vm_area_struct *,
199 struct vm_userfaultfd_ctx *);
200 extern void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *,
201 unsigned long from, unsigned long to,
204 extern bool userfaultfd_remove(struct vm_area_struct *vma,
208 extern int userfaultfd_unmap_prep(struct vm_area_struct *vma,
209 unsigned long start, unsigned long end, struct list_head *uf);
210 extern void userfaultfd_unmap_complete(struct mm_struct *mm,
211 struct list_head *uf);
212 extern bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma);
213 extern bool userfaultfd_wp_async(struct vm_area_struct *vma);
215 #else /* CONFIG_USERFAULTFD */
218 static inline vm_fault_t handle_userfault(struct vm_fault *vmf,
219 unsigned long reason)
221 return VM_FAULT_SIGBUS;
224 static inline long uffd_wp_range(struct vm_area_struct *vma,
225 unsigned long start, unsigned long len,
231 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
232 struct vm_userfaultfd_ctx vm_ctx)
237 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
242 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
247 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
252 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
258 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
265 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
270 static inline int dup_userfaultfd(struct vm_area_struct *vma,
276 static inline void dup_userfaultfd_complete(struct list_head *l)
280 static inline void mremap_userfaultfd_prep(struct vm_area_struct *vma,
281 struct vm_userfaultfd_ctx *ctx)
285 static inline void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *ctx,
292 static inline bool userfaultfd_remove(struct vm_area_struct *vma,
299 static inline int userfaultfd_unmap_prep(struct vm_area_struct *vma,
300 unsigned long start, unsigned long end,
301 struct list_head *uf)
306 static inline void userfaultfd_unmap_complete(struct mm_struct *mm,
307 struct list_head *uf)
311 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
316 static inline bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma)
321 static inline bool userfaultfd_wp_async(struct vm_area_struct *vma)
326 #endif /* CONFIG_USERFAULTFD */
328 static inline bool userfaultfd_wp_use_markers(struct vm_area_struct *vma)
330 /* Only wr-protect mode uses pte markers */
331 if (!userfaultfd_wp(vma))
334 /* File-based uffd-wp always need markers */
335 if (!vma_is_anonymous(vma))
339 * Anonymous uffd-wp only needs the markers if WP_UNPOPULATED
340 * enabled (to apply markers on zero pages).
342 return userfaultfd_wp_unpopulated(vma);
345 static inline bool pte_marker_entry_uffd_wp(swp_entry_t entry)
347 #ifdef CONFIG_PTE_MARKER_UFFD_WP
348 return is_pte_marker_entry(entry) &&
349 (pte_marker_get(entry) & PTE_MARKER_UFFD_WP);
355 static inline bool pte_marker_uffd_wp(pte_t pte)
357 #ifdef CONFIG_PTE_MARKER_UFFD_WP
360 if (!is_swap_pte(pte))
363 entry = pte_to_swp_entry(pte);
365 return pte_marker_entry_uffd_wp(entry);
372 * Returns true if this is a swap pte and was uffd-wp wr-protected in either
373 * forms (pte marker or a normal swap pte), false otherwise.
375 static inline bool pte_swp_uffd_wp_any(pte_t pte)
377 #ifdef CONFIG_PTE_MARKER_UFFD_WP
378 if (!is_swap_pte(pte))
381 if (pte_swp_uffd_wp(pte))
384 if (pte_marker_uffd_wp(pte))
390 #endif /* _LINUX_USERFAULTFD_K_H */