projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[linux-block.git] / include / linux / userfaultfd_k.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
932b18e0
AA		 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>
55adf4de 17#include <linux/mm.h>
1db9dbc2
PX		 18#include <linux/swap.h>
19#include <linux/swapops.h>
55adf4de 20#include <asm-generic/pgtable_uffd.h>
b1f9e876 21#include <linux/hugetlb_inline.h>
932b18e0 22
7677f7fd
AR		 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
932b18e0
AA		 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
2b740303 39extern vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason);
932b18e0 40
d9712937
AR		 41/* A combined operation mode + behavior flags. */
42typedef unsigned int __bitwise uffd_flags_t;
43
44/* Mutually exclusive modes of operation. */
45enum mfill_atomic_mode {
46 MFILL_ATOMIC_COPY,
47 MFILL_ATOMIC_ZEROPAGE,
48 MFILL_ATOMIC_CONTINUE,
49 NR_MFILL_ATOMIC_MODES,
f6191471
AR		 50};
51
d9712937
AR		 52#define MFILL_ATOMIC_MODE_BITS (const_ilog2(NR_MFILL_ATOMIC_MODES - 1) + 1)
53#define MFILL_ATOMIC_BIT(nr) BIT(MFILL_ATOMIC_MODE_BITS + (nr))
54#define MFILL_ATOMIC_FLAG(nr) ((__force uffd_flags_t) MFILL_ATOMIC_BIT(nr))
55#define MFILL_ATOMIC_MODE_MASK ((__force uffd_flags_t) (MFILL_ATOMIC_BIT(0) - 1))
56
57static inline bool uffd_flags_mode_is(uffd_flags_t flags, enum mfill_atomic_mode expected)
58{
59 return (flags & MFILL_ATOMIC_MODE_MASK) == ((__force uffd_flags_t) expected);
60}
61
62static inline uffd_flags_t uffd_flags_set_mode(uffd_flags_t flags, enum mfill_atomic_mode mode)
63{
64 flags &= ~MFILL_ATOMIC_MODE_MASK;
65 return flags | ((__force uffd_flags_t) mode);
66}
67
68/* Flags controlling behavior. These behavior changes are mode-independent. */
69#define MFILL_ATOMIC_WP MFILL_ATOMIC_FLAG(0)
70
61c50040 71extern int mfill_atomic_install_pte(pmd_t *dst_pmd,
7d64ae3a
AR		 72 struct vm_area_struct *dst_vma,
73 unsigned long dst_addr, struct page *page,
d9712937 74 bool newly_allocated, uffd_flags_t flags);
7d64ae3a 75
a734991c
AR		 76extern ssize_t mfill_atomic_copy(struct mm_struct *dst_mm, unsigned long dst_start,
77 unsigned long src_start, unsigned long len,
d9712937 78 atomic_t *mmap_changing, uffd_flags_t flags);
a734991c
AR		 79extern ssize_t mfill_atomic_zeropage(struct mm_struct *dst_mm,
80 unsigned long dst_start,
81 unsigned long len,
82 atomic_t *mmap_changing);
83extern ssize_t mfill_atomic_continue(struct mm_struct *dst_mm, unsigned long dst_start,
02891844
AR		 84 unsigned long len, atomic_t *mmap_changing,
85 uffd_flags_t flags);
ffd05793
SL		 86extern int mwriteprotect_range(struct mm_struct *dst_mm,
87 unsigned long start, unsigned long len,
a759a909 88 bool enable_wp, atomic_t *mmap_changing);
61c50040 89extern long uffd_wp_range(struct vm_area_struct *vma,
f369b07c 90 unsigned long start, unsigned long len, bool enable_wp);
c1a4de99 91
932b18e0
AA		 92/* mm helpers */
93static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
94 struct vm_userfaultfd_ctx vm_ctx)
95{
96 return vma->vm_userfaultfd_ctx.ctx == vm_ctx.ctx;
97}
98
c1991e07 99/*
0d9cadab
AR		 100 * Never enable huge pmd sharing on some uffd registered vmas:
101 *
102 * - VM_UFFD_WP VMAs, because write protect information is per pgtable entry.
103 *
104 * - VM_UFFD_MINOR VMAs, because otherwise we would never get minor faults for
105 * VMAs which share huge pmds. (If you have two mappings to the same
106 * underlying pages, and fault in the non-UFFD-registered one with a write,
107 * with huge pmd sharing this would *also* setup the second UFFD-registered
108 * mapping, and we'd not get minor faults.)
c1991e07
PX		 109 */
110static inline bool uffd_disable_huge_pmd_share(struct vm_area_struct *vma)
111{
0d9cadab 112 return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
c1991e07
PX		 113}
114
9c28a205
PX		 115/*
116 * Don't do fault around for either WP or MINOR registered uffd range. For
117 * MINOR registered range, fault around will be a total disaster and ptes can
118 * be installed without notifications; for WP it should mostly be fine as long
119 * as the fault around checks for pte_none() before the installation, however
120 * to be super safe we just forbid it.
121 */
122static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
123{
124 return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
125}
126
932b18e0
AA		 127static inline bool userfaultfd_missing(struct vm_area_struct *vma)
128{
129 return vma->vm_flags & VM_UFFD_MISSING;
130}
131
1df319e0
SL		 132static inline bool userfaultfd_wp(struct vm_area_struct *vma)
133{
134 return vma->vm_flags & VM_UFFD_WP;
135}
136
7677f7fd
AR		 137static inline bool userfaultfd_minor(struct vm_area_struct *vma)
138{
139 return vma->vm_flags & VM_UFFD_MINOR;
140}
141
55adf4de
AA		 142static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
143 pte_t pte)
144{
145 return userfaultfd_wp(vma) && pte_uffd_wp(pte);
146}
147
148static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
149 pmd_t pmd)
150{
151 return userfaultfd_wp(vma) && pmd_uffd_wp(pmd);
152}
153
932b18e0
AA		 154static inline bool userfaultfd_armed(struct vm_area_struct *vma)
155{
7677f7fd 156 return vma->vm_flags & __VM_UFFD_FLAGS;
932b18e0
AA		 157}
158
b1f9e876
PX		 159static inline bool vma_can_userfault(struct vm_area_struct *vma,
160 unsigned long vm_flags)
161{
67eae54b
PX		 162 if ((vm_flags & VM_UFFD_MINOR) &&
163 (!is_vm_hugetlb_page(vma) && !vma_is_shmem(vma)))
164 return false;
b1f9e876
PX		 165#ifndef CONFIG_PTE_MARKER_UFFD_WP
166 /*
167 * If user requested uffd-wp but not enabled pte markers for
168 * uffd-wp, then shmem & hugetlbfs are not supported but only
169 * anonymous.
170 */
171 if ((vm_flags & VM_UFFD_WP) && !vma_is_anonymous(vma))
172 return false;
173#endif
174 return vma_is_anonymous(vma) || is_vm_hugetlb_page(vma) ||
175 vma_is_shmem(vma);
176}
177
893e26e6
PE		 178extern int dup_userfaultfd(struct vm_area_struct *, struct list_head *);
179extern void dup_userfaultfd_complete(struct list_head *);
180
72f87654
PE		 181extern void mremap_userfaultfd_prep(struct vm_area_struct *,
182 struct vm_userfaultfd_ctx *);
90794bf1 183extern void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *,
72f87654
PE		 184 unsigned long from, unsigned long to,
185 unsigned long len);
186
70ccb92f 187extern bool userfaultfd_remove(struct vm_area_struct *vma,
d811914d
MR		 188 unsigned long start,
189 unsigned long end);
05ce7724 190
65ac1320
LH		 191extern int userfaultfd_unmap_prep(struct vm_area_struct *vma,
192 unsigned long start, unsigned long end, struct list_head *uf);
897ab3e0
MR		 193extern void userfaultfd_unmap_complete(struct mm_struct *mm,
194 struct list_head *uf);
2bad466c 195extern bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma);
897ab3e0 196
932b18e0
AA		 197#else /* CONFIG_USERFAULTFD */
198
199/* mm helpers */
2b740303
SJ		 200static inline vm_fault_t handle_userfault(struct vm_fault *vmf,
201 unsigned long reason)
932b18e0
AA		 202{
203 return VM_FAULT_SIGBUS;
204}
205
206static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
207 struct vm_userfaultfd_ctx vm_ctx)
208{
209 return true;
210}
211
212static inline bool userfaultfd_missing(struct vm_area_struct *vma)
213{
214 return false;
215}
216
1df319e0
SL		 217static inline bool userfaultfd_wp(struct vm_area_struct *vma)
218{
219 return false;
220}
221
7677f7fd
AR		 222static inline bool userfaultfd_minor(struct vm_area_struct *vma)
223{
224 return false;
225}
226
55adf4de
AA		 227static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
228 pte_t pte)
229{
230 return false;
231}
232
233static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
234 pmd_t pmd)
235{
236 return false;
237}
238
239
932b18e0
AA		 240static inline bool userfaultfd_armed(struct vm_area_struct *vma)
241{
242 return false;
243}
244
893e26e6
PE		 245static inline int dup_userfaultfd(struct vm_area_struct *vma,
246 struct list_head *l)
247{
248 return 0;
249}
250
251static inline void dup_userfaultfd_complete(struct list_head *l)
252{
253}
254
72f87654
PE		 255static inline void mremap_userfaultfd_prep(struct vm_area_struct *vma,
256 struct vm_userfaultfd_ctx *ctx)
257{
258}
259
90794bf1 260static inline void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *ctx,
72f87654
PE		 261 unsigned long from,
262 unsigned long to,
263 unsigned long len)
264{
265}
05ce7724 266
70ccb92f 267static inline bool userfaultfd_remove(struct vm_area_struct *vma,
d811914d
MR		 268 unsigned long start,
269 unsigned long end)
05ce7724 270{
70ccb92f 271 return true;
05ce7724 272}
897ab3e0 273
65ac1320 274static inline int userfaultfd_unmap_prep(struct vm_area_struct *vma,
897ab3e0
MR		 275 unsigned long start, unsigned long end,
276 struct list_head *uf)
277{
278 return 0;
279}
280
281static inline void userfaultfd_unmap_complete(struct mm_struct *mm,
282 struct list_head *uf)
283{
284}
ca49ca71 285
9c28a205
PX		 286static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
287{
288 return false;
289}
290
2bad466c
PX		 291static inline bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma)
292{
293 return false;
294}
295
932b18e0
AA		 296#endif /* CONFIG_USERFAULTFD */
297
2bad466c
PX		 298static inline bool userfaultfd_wp_use_markers(struct vm_area_struct *vma)
299{
300 /* Only wr-protect mode uses pte markers */
301 if (!userfaultfd_wp(vma))
302 return false;
303
304 /* File-based uffd-wp always need markers */
305 if (!vma_is_anonymous(vma))
306 return true;
307
308 /*
309 * Anonymous uffd-wp only needs the markers if WP_UNPOPULATED
310 * enabled (to apply markers on zero pages).
311 */
312 return userfaultfd_wp_unpopulated(vma);
313}
314
1db9dbc2
PX		 315static inline bool pte_marker_entry_uffd_wp(swp_entry_t entry)
316{
317#ifdef CONFIG_PTE_MARKER_UFFD_WP
318 return is_pte_marker_entry(entry) &&
319 (pte_marker_get(entry) & PTE_MARKER_UFFD_WP);
320#else
321 return false;
322#endif
323}
324
325static inline bool pte_marker_uffd_wp(pte_t pte)
326{
327#ifdef CONFIG_PTE_MARKER_UFFD_WP
328 swp_entry_t entry;
329
330 if (!is_swap_pte(pte))
331 return false;
332
333 entry = pte_to_swp_entry(pte);
334
335 return pte_marker_entry_uffd_wp(entry);
336#else
337 return false;
338#endif
339}
340
341/*
342 * Returns true if this is a swap pte and was uffd-wp wr-protected in either
343 * forms (pte marker or a normal swap pte), false otherwise.
344 */
345static inline bool pte_swp_uffd_wp_any(pte_t pte)
346{
347#ifdef CONFIG_PTE_MARKER_UFFD_WP
348 if (!is_swap_pte(pte))
349 return false;
350
351 if (pte_swp_uffd_wp(pte))
352 return true;
353
354 if (pte_marker_uffd_wp(pte))
355 return true;
356#endif
357 return false;
358}
359
932b18e0 360#endif /* _LINUX_USERFAULTFD_K_H */
Linux 6.x block layer and io_uring tree(s)