xref: /linux-6.15/include/linux/userfaultfd_k.h (revision 2d337b71)
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 extern vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason);
40 
41 /*
42  * The mode of operation for __mcopy_atomic and its helpers.
43  *
44  * This is almost an implementation detail (mcopy_atomic below doesn't take this
45  * as a parameter), but it's exposed here because memory-kind-specific
46  * implementations (e.g. hugetlbfs) need to know the mode of operation.
47  */
48 enum mcopy_atomic_mode {
49 	/* A normal copy_from_user into the destination range. */
50 	MCOPY_ATOMIC_NORMAL,
51 	/* Don't copy; map the destination range to the zero page. */
52 	MCOPY_ATOMIC_ZEROPAGE,
53 	/* Just install pte(s) with the existing page(s) in the page cache. */
54 	MCOPY_ATOMIC_CONTINUE,
55 };
56 
57 extern int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
58 				    struct vm_area_struct *dst_vma,
59 				    unsigned long dst_addr, struct page *page,
60 				    bool newly_allocated, bool wp_copy);
61 
62 extern ssize_t mcopy_atomic(struct mm_struct *dst_mm, unsigned long dst_start,
63 			    unsigned long src_start, unsigned long len,
64 			    atomic_t *mmap_changing, __u64 mode);
65 extern ssize_t mfill_zeropage(struct mm_struct *dst_mm,
66 			      unsigned long dst_start,
67 			      unsigned long len,
68 			      atomic_t *mmap_changing);
69 extern ssize_t mcopy_continue(struct mm_struct *dst_mm, unsigned long dst_start,
70 			      unsigned long len, atomic_t *mmap_changing);
71 extern int mwriteprotect_range(struct mm_struct *dst_mm,
72 			       unsigned long start, unsigned long len,
73 			       bool enable_wp, atomic_t *mmap_changing);
74 extern long uffd_wp_range(struct mm_struct *dst_mm, struct vm_area_struct *vma,
75 			  unsigned long start, unsigned long len, bool enable_wp);
76 
77 /* mm helpers */
78 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
79 					struct vm_userfaultfd_ctx vm_ctx)
80 {
81 	return vma->vm_userfaultfd_ctx.ctx == vm_ctx.ctx;
82 }
83 
84 /*
85  * Never enable huge pmd sharing on some uffd registered vmas:
86  *
87  * - VM_UFFD_WP VMAs, because write protect information is per pgtable entry.
88  *
89  * - VM_UFFD_MINOR VMAs, because otherwise we would never get minor faults for
90  *   VMAs which share huge pmds. (If you have two mappings to the same
91  *   underlying pages, and fault in the non-UFFD-registered one with a write,
92  *   with huge pmd sharing this would *also* setup the second UFFD-registered
93  *   mapping, and we'd not get minor faults.)
94  */
95 static inline bool uffd_disable_huge_pmd_share(struct vm_area_struct *vma)
96 {
97 	return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
98 }
99 
100 /*
101  * Don't do fault around for either WP or MINOR registered uffd range.  For
102  * MINOR registered range, fault around will be a total disaster and ptes can
103  * be installed without notifications; for WP it should mostly be fine as long
104  * as the fault around checks for pte_none() before the installation, however
105  * to be super safe we just forbid it.
106  */
107 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
108 {
109 	return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
110 }
111 
112 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
113 {
114 	return vma->vm_flags & VM_UFFD_MISSING;
115 }
116 
117 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
118 {
119 	return vma->vm_flags & VM_UFFD_WP;
120 }
121 
122 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
123 {
124 	return vma->vm_flags & VM_UFFD_MINOR;
125 }
126 
127 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
128 				      pte_t pte)
129 {
130 	return userfaultfd_wp(vma) && pte_uffd_wp(pte);
131 }
132 
133 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
134 					   pmd_t pmd)
135 {
136 	return userfaultfd_wp(vma) && pmd_uffd_wp(pmd);
137 }
138 
139 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
140 {
141 	return vma->vm_flags & __VM_UFFD_FLAGS;
142 }
143 
144 static inline bool vma_can_userfault(struct vm_area_struct *vma,
145 				     unsigned long vm_flags)
146 {
147 	if ((vm_flags & VM_UFFD_MINOR) &&
148 	    (!is_vm_hugetlb_page(vma) && !vma_is_shmem(vma)))
149 		return false;
150 #ifndef CONFIG_PTE_MARKER_UFFD_WP
151 	/*
152 	 * If user requested uffd-wp but not enabled pte markers for
153 	 * uffd-wp, then shmem & hugetlbfs are not supported but only
154 	 * anonymous.
155 	 */
156 	if ((vm_flags & VM_UFFD_WP) && !vma_is_anonymous(vma))
157 		return false;
158 #endif
159 	return vma_is_anonymous(vma) || is_vm_hugetlb_page(vma) ||
160 	    vma_is_shmem(vma);
161 }
162 
163 extern int dup_userfaultfd(struct vm_area_struct *, struct list_head *);
164 extern void dup_userfaultfd_complete(struct list_head *);
165 
166 extern void mremap_userfaultfd_prep(struct vm_area_struct *,
167 				    struct vm_userfaultfd_ctx *);
168 extern void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *,
169 					unsigned long from, unsigned long to,
170 					unsigned long len);
171 
172 extern bool userfaultfd_remove(struct vm_area_struct *vma,
173 			       unsigned long start,
174 			       unsigned long end);
175 
176 extern int userfaultfd_unmap_prep(struct mm_struct *mm, unsigned long start,
177 				  unsigned long end, struct list_head *uf);
178 extern void userfaultfd_unmap_complete(struct mm_struct *mm,
179 				       struct list_head *uf);
180 
181 #else /* CONFIG_USERFAULTFD */
182 
183 /* mm helpers */
184 static inline vm_fault_t handle_userfault(struct vm_fault *vmf,
185 				unsigned long reason)
186 {
187 	return VM_FAULT_SIGBUS;
188 }
189 
190 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
191 					struct vm_userfaultfd_ctx vm_ctx)
192 {
193 	return true;
194 }
195 
196 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
197 {
198 	return false;
199 }
200 
201 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
202 {
203 	return false;
204 }
205 
206 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
207 {
208 	return false;
209 }
210 
211 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
212 				      pte_t pte)
213 {
214 	return false;
215 }
216 
217 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
218 					   pmd_t pmd)
219 {
220 	return false;
221 }
222 
223 
224 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
225 {
226 	return false;
227 }
228 
229 static inline int dup_userfaultfd(struct vm_area_struct *vma,
230 				  struct list_head *l)
231 {
232 	return 0;
233 }
234 
235 static inline void dup_userfaultfd_complete(struct list_head *l)
236 {
237 }
238 
239 static inline void mremap_userfaultfd_prep(struct vm_area_struct *vma,
240 					   struct vm_userfaultfd_ctx *ctx)
241 {
242 }
243 
244 static inline void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *ctx,
245 					       unsigned long from,
246 					       unsigned long to,
247 					       unsigned long len)
248 {
249 }
250 
251 static inline bool userfaultfd_remove(struct vm_area_struct *vma,
252 				      unsigned long start,
253 				      unsigned long end)
254 {
255 	return true;
256 }
257 
258 static inline int userfaultfd_unmap_prep(struct mm_struct *mm,
259 					 unsigned long start, unsigned long end,
260 					 struct list_head *uf)
261 {
262 	return 0;
263 }
264 
265 static inline void userfaultfd_unmap_complete(struct mm_struct *mm,
266 					      struct list_head *uf)
267 {
268 }
269 
270 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
271 {
272 	return false;
273 }
274 
275 #endif /* CONFIG_USERFAULTFD */
276 
277 static inline bool pte_marker_entry_uffd_wp(swp_entry_t entry)
278 {
279 #ifdef CONFIG_PTE_MARKER_UFFD_WP
280 	return is_pte_marker_entry(entry) &&
281 	    (pte_marker_get(entry) & PTE_MARKER_UFFD_WP);
282 #else
283 	return false;
284 #endif
285 }
286 
287 static inline bool pte_marker_uffd_wp(pte_t pte)
288 {
289 #ifdef CONFIG_PTE_MARKER_UFFD_WP
290 	swp_entry_t entry;
291 
292 	if (!is_swap_pte(pte))
293 		return false;
294 
295 	entry = pte_to_swp_entry(pte);
296 
297 	return pte_marker_entry_uffd_wp(entry);
298 #else
299 	return false;
300 #endif
301 }
302 
303 /*
304  * Returns true if this is a swap pte and was uffd-wp wr-protected in either
305  * forms (pte marker or a normal swap pte), false otherwise.
306  */
307 static inline bool pte_swp_uffd_wp_any(pte_t pte)
308 {
309 #ifdef CONFIG_PTE_MARKER_UFFD_WP
310 	if (!is_swap_pte(pte))
311 		return false;
312 
313 	if (pte_swp_uffd_wp(pte))
314 		return true;
315 
316 	if (pte_marker_uffd_wp(pte))
317 		return true;
318 #endif
319 	return false;
320 }
321 
322 #endif /* _LINUX_USERFAULTFD_K_H */
323