xref: /linux-6.15/include/linux/userfaultfd_k.h (revision 1db9dbc2)
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 
22 /* The set of all possible UFFD-related VM flags. */
23 #define __VM_UFFD_FLAGS (VM_UFFD_MISSING | VM_UFFD_WP | VM_UFFD_MINOR)
24 
25 /*
26  * CAREFUL: Check include/uapi/asm-generic/fcntl.h when defining
27  * new flags, since they might collide with O_* ones. We want
28  * to re-use O_* flags that couldn't possibly have a meaning
29  * from userfaultfd, in order to leave a free define-space for
30  * shared O_* flags.
31  */
32 #define UFFD_CLOEXEC O_CLOEXEC
33 #define UFFD_NONBLOCK O_NONBLOCK
34 
35 #define UFFD_SHARED_FCNTL_FLAGS (O_CLOEXEC | O_NONBLOCK)
36 #define UFFD_FLAGS_SET (EFD_SHARED_FCNTL_FLAGS)
37 
38 extern int sysctl_unprivileged_userfaultfd;
39 
40 extern vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason);
41 
42 /*
43  * The mode of operation for __mcopy_atomic and its helpers.
44  *
45  * This is almost an implementation detail (mcopy_atomic below doesn't take this
46  * as a parameter), but it's exposed here because memory-kind-specific
47  * implementations (e.g. hugetlbfs) need to know the mode of operation.
48  */
49 enum mcopy_atomic_mode {
50 	/* A normal copy_from_user into the destination range. */
51 	MCOPY_ATOMIC_NORMAL,
52 	/* Don't copy; map the destination range to the zero page. */
53 	MCOPY_ATOMIC_ZEROPAGE,
54 	/* Just install pte(s) with the existing page(s) in the page cache. */
55 	MCOPY_ATOMIC_CONTINUE,
56 };
57 
58 extern int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
59 				    struct vm_area_struct *dst_vma,
60 				    unsigned long dst_addr, struct page *page,
61 				    bool newly_allocated, bool wp_copy);
62 
63 extern ssize_t mcopy_atomic(struct mm_struct *dst_mm, unsigned long dst_start,
64 			    unsigned long src_start, unsigned long len,
65 			    atomic_t *mmap_changing, __u64 mode);
66 extern ssize_t mfill_zeropage(struct mm_struct *dst_mm,
67 			      unsigned long dst_start,
68 			      unsigned long len,
69 			      atomic_t *mmap_changing);
70 extern ssize_t mcopy_continue(struct mm_struct *dst_mm, unsigned long dst_start,
71 			      unsigned long len, atomic_t *mmap_changing);
72 extern int mwriteprotect_range(struct mm_struct *dst_mm,
73 			       unsigned long start, unsigned long len,
74 			       bool enable_wp, atomic_t *mmap_changing);
75 
76 /* mm helpers */
77 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
78 					struct vm_userfaultfd_ctx vm_ctx)
79 {
80 	return vma->vm_userfaultfd_ctx.ctx == vm_ctx.ctx;
81 }
82 
83 /*
84  * Never enable huge pmd sharing on some uffd registered vmas:
85  *
86  * - VM_UFFD_WP VMAs, because write protect information is per pgtable entry.
87  *
88  * - VM_UFFD_MINOR VMAs, because otherwise we would never get minor faults for
89  *   VMAs which share huge pmds. (If you have two mappings to the same
90  *   underlying pages, and fault in the non-UFFD-registered one with a write,
91  *   with huge pmd sharing this would *also* setup the second UFFD-registered
92  *   mapping, and we'd not get minor faults.)
93  */
94 static inline bool uffd_disable_huge_pmd_share(struct vm_area_struct *vma)
95 {
96 	return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
97 }
98 
99 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
100 {
101 	return vma->vm_flags & VM_UFFD_MISSING;
102 }
103 
104 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
105 {
106 	return vma->vm_flags & VM_UFFD_WP;
107 }
108 
109 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
110 {
111 	return vma->vm_flags & VM_UFFD_MINOR;
112 }
113 
114 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
115 				      pte_t pte)
116 {
117 	return userfaultfd_wp(vma) && pte_uffd_wp(pte);
118 }
119 
120 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
121 					   pmd_t pmd)
122 {
123 	return userfaultfd_wp(vma) && pmd_uffd_wp(pmd);
124 }
125 
126 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
127 {
128 	return vma->vm_flags & __VM_UFFD_FLAGS;
129 }
130 
131 extern int dup_userfaultfd(struct vm_area_struct *, struct list_head *);
132 extern void dup_userfaultfd_complete(struct list_head *);
133 
134 extern void mremap_userfaultfd_prep(struct vm_area_struct *,
135 				    struct vm_userfaultfd_ctx *);
136 extern void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *,
137 					unsigned long from, unsigned long to,
138 					unsigned long len);
139 
140 extern bool userfaultfd_remove(struct vm_area_struct *vma,
141 			       unsigned long start,
142 			       unsigned long end);
143 
144 extern int userfaultfd_unmap_prep(struct vm_area_struct *vma,
145 				  unsigned long start, unsigned long end,
146 				  struct list_head *uf);
147 extern void userfaultfd_unmap_complete(struct mm_struct *mm,
148 				       struct list_head *uf);
149 
150 #else /* CONFIG_USERFAULTFD */
151 
152 /* mm helpers */
153 static inline vm_fault_t handle_userfault(struct vm_fault *vmf,
154 				unsigned long reason)
155 {
156 	return VM_FAULT_SIGBUS;
157 }
158 
159 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
160 					struct vm_userfaultfd_ctx vm_ctx)
161 {
162 	return true;
163 }
164 
165 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
166 {
167 	return false;
168 }
169 
170 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
171 {
172 	return false;
173 }
174 
175 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
176 {
177 	return false;
178 }
179 
180 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
181 				      pte_t pte)
182 {
183 	return false;
184 }
185 
186 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
187 					   pmd_t pmd)
188 {
189 	return false;
190 }
191 
192 
193 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
194 {
195 	return false;
196 }
197 
198 static inline int dup_userfaultfd(struct vm_area_struct *vma,
199 				  struct list_head *l)
200 {
201 	return 0;
202 }
203 
204 static inline void dup_userfaultfd_complete(struct list_head *l)
205 {
206 }
207 
208 static inline void mremap_userfaultfd_prep(struct vm_area_struct *vma,
209 					   struct vm_userfaultfd_ctx *ctx)
210 {
211 }
212 
213 static inline void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *ctx,
214 					       unsigned long from,
215 					       unsigned long to,
216 					       unsigned long len)
217 {
218 }
219 
220 static inline bool userfaultfd_remove(struct vm_area_struct *vma,
221 				      unsigned long start,
222 				      unsigned long end)
223 {
224 	return true;
225 }
226 
227 static inline int userfaultfd_unmap_prep(struct vm_area_struct *vma,
228 					 unsigned long start, unsigned long end,
229 					 struct list_head *uf)
230 {
231 	return 0;
232 }
233 
234 static inline void userfaultfd_unmap_complete(struct mm_struct *mm,
235 					      struct list_head *uf)
236 {
237 }
238 
239 #endif /* CONFIG_USERFAULTFD */
240 
241 static inline bool pte_marker_entry_uffd_wp(swp_entry_t entry)
242 {
243 #ifdef CONFIG_PTE_MARKER_UFFD_WP
244 	return is_pte_marker_entry(entry) &&
245 	    (pte_marker_get(entry) & PTE_MARKER_UFFD_WP);
246 #else
247 	return false;
248 #endif
249 }
250 
251 static inline bool pte_marker_uffd_wp(pte_t pte)
252 {
253 #ifdef CONFIG_PTE_MARKER_UFFD_WP
254 	swp_entry_t entry;
255 
256 	if (!is_swap_pte(pte))
257 		return false;
258 
259 	entry = pte_to_swp_entry(pte);
260 
261 	return pte_marker_entry_uffd_wp(entry);
262 #else
263 	return false;
264 #endif
265 }
266 
267 /*
268  * Returns true if this is a swap pte and was uffd-wp wr-protected in either
269  * forms (pte marker or a normal swap pte), false otherwise.
270  */
271 static inline bool pte_swp_uffd_wp_any(pte_t pte)
272 {
273 #ifdef CONFIG_PTE_MARKER_UFFD_WP
274 	if (!is_swap_pte(pte))
275 		return false;
276 
277 	if (pte_swp_uffd_wp(pte))
278 		return true;
279 
280 	if (pte_marker_uffd_wp(pte))
281 		return true;
282 #endif
283 	return false;
284 }
285 
286 #endif /* _LINUX_USERFAULTFD_K_H */
287