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 /*
40 * Start with fault_pending_wqh and fault_wqh so they're more likely
41 * to be in the same cacheline.
42 *
43 * Locking order:
44 * fd_wqh.lock
45 * fault_pending_wqh.lock
46 * fault_wqh.lock
47 * event_wqh.lock
48 *
49 * To avoid deadlocks, IRQs must be disabled when taking any of the above locks,
50 * since fd_wqh.lock is taken by aio_poll() while it's holding a lock that's
51 * also taken in IRQ context.
52 */
53 struct userfaultfd_ctx {
54 /* waitqueue head for the pending (i.e. not read) userfaults */
55 wait_queue_head_t fault_pending_wqh;
56 /* waitqueue head for the userfaults */
57 wait_queue_head_t fault_wqh;
58 /* waitqueue head for the pseudo fd to wakeup poll/read */
59 wait_queue_head_t fd_wqh;
60 /* waitqueue head for events */
61 wait_queue_head_t event_wqh;
62 /* a refile sequence protected by fault_pending_wqh lock */
63 seqcount_spinlock_t refile_seq;
64 /* pseudo fd refcounting */
65 refcount_t refcount;
66 /* userfaultfd syscall flags */
67 unsigned int flags;
68 /* features requested from the userspace */
69 unsigned int features;
70 /* released */
71 bool released;
72 /*
73 * Prevents userfaultfd operations (fill/move/wp) from happening while
74 * some non-cooperative event(s) is taking place. Increments are done
75 * in write-mode. Whereas, userfaultfd operations, which includes
76 * reading mmap_changing, is done under read-mode.
77 */
78 struct rw_semaphore map_changing_lock;
79 /* memory mappings are changing because of non-cooperative event */
80 atomic_t mmap_changing;
81 /* mm with one ore more vmas attached to this userfaultfd_ctx */
82 struct mm_struct *mm;
83 };
84
85 extern vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason);
86
87 /* A combined operation mode + behavior flags. */
88 typedef unsigned int __bitwise uffd_flags_t;
89
90 /* Mutually exclusive modes of operation. */
91 enum mfill_atomic_mode {
92 MFILL_ATOMIC_COPY,
93 MFILL_ATOMIC_ZEROPAGE,
94 MFILL_ATOMIC_CONTINUE,
95 MFILL_ATOMIC_POISON,
96 NR_MFILL_ATOMIC_MODES,
97 };
98
99 #define MFILL_ATOMIC_MODE_BITS (const_ilog2(NR_MFILL_ATOMIC_MODES - 1) + 1)
100 #define MFILL_ATOMIC_BIT(nr) BIT(MFILL_ATOMIC_MODE_BITS + (nr))
101 #define MFILL_ATOMIC_FLAG(nr) ((__force uffd_flags_t) MFILL_ATOMIC_BIT(nr))
102 #define MFILL_ATOMIC_MODE_MASK ((__force uffd_flags_t) (MFILL_ATOMIC_BIT(0) - 1))
103
uffd_flags_mode_is(uffd_flags_t flags,enum mfill_atomic_mode expected)104 static inline bool uffd_flags_mode_is(uffd_flags_t flags, enum mfill_atomic_mode expected)
105 {
106 return (flags & MFILL_ATOMIC_MODE_MASK) == ((__force uffd_flags_t) expected);
107 }
108
uffd_flags_set_mode(uffd_flags_t flags,enum mfill_atomic_mode mode)109 static inline uffd_flags_t uffd_flags_set_mode(uffd_flags_t flags, enum mfill_atomic_mode mode)
110 {
111 flags &= ~MFILL_ATOMIC_MODE_MASK;
112 return flags | ((__force uffd_flags_t) mode);
113 }
114
115 /* Flags controlling behavior. These behavior changes are mode-independent. */
116 #define MFILL_ATOMIC_WP MFILL_ATOMIC_FLAG(0)
117
118 extern int mfill_atomic_install_pte(pmd_t *dst_pmd,
119 struct vm_area_struct *dst_vma,
120 unsigned long dst_addr, struct page *page,
121 bool newly_allocated, uffd_flags_t flags);
122
123 extern ssize_t mfill_atomic_copy(struct userfaultfd_ctx *ctx, unsigned long dst_start,
124 unsigned long src_start, unsigned long len,
125 uffd_flags_t flags);
126 extern ssize_t mfill_atomic_zeropage(struct userfaultfd_ctx *ctx,
127 unsigned long dst_start,
128 unsigned long len);
129 extern ssize_t mfill_atomic_continue(struct userfaultfd_ctx *ctx, unsigned long dst_start,
130 unsigned long len, uffd_flags_t flags);
131 extern ssize_t mfill_atomic_poison(struct userfaultfd_ctx *ctx, unsigned long start,
132 unsigned long len, uffd_flags_t flags);
133 extern int mwriteprotect_range(struct userfaultfd_ctx *ctx, unsigned long start,
134 unsigned long len, bool enable_wp);
135 extern long uffd_wp_range(struct vm_area_struct *vma,
136 unsigned long start, unsigned long len, bool enable_wp);
137
138 /* move_pages */
139 void double_pt_lock(spinlock_t *ptl1, spinlock_t *ptl2);
140 void double_pt_unlock(spinlock_t *ptl1, spinlock_t *ptl2);
141 ssize_t move_pages(struct userfaultfd_ctx *ctx, unsigned long dst_start,
142 unsigned long src_start, unsigned long len, __u64 flags);
143 int move_pages_huge_pmd(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd, pmd_t dst_pmdval,
144 struct vm_area_struct *dst_vma,
145 struct vm_area_struct *src_vma,
146 unsigned long dst_addr, unsigned long src_addr);
147
148 /* mm helpers */
is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct * vma,struct vm_userfaultfd_ctx vm_ctx)149 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
150 struct vm_userfaultfd_ctx vm_ctx)
151 {
152 return vma->vm_userfaultfd_ctx.ctx == vm_ctx.ctx;
153 }
154
155 /*
156 * Never enable huge pmd sharing on some uffd registered vmas:
157 *
158 * - VM_UFFD_WP VMAs, because write protect information is per pgtable entry.
159 *
160 * - VM_UFFD_MINOR VMAs, because otherwise we would never get minor faults for
161 * VMAs which share huge pmds. (If you have two mappings to the same
162 * underlying pages, and fault in the non-UFFD-registered one with a write,
163 * with huge pmd sharing this would *also* setup the second UFFD-registered
164 * mapping, and we'd not get minor faults.)
165 */
uffd_disable_huge_pmd_share(struct vm_area_struct * vma)166 static inline bool uffd_disable_huge_pmd_share(struct vm_area_struct *vma)
167 {
168 return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
169 }
170
171 /*
172 * Don't do fault around for either WP or MINOR registered uffd range. For
173 * MINOR registered range, fault around will be a total disaster and ptes can
174 * be installed without notifications; for WP it should mostly be fine as long
175 * as the fault around checks for pte_none() before the installation, however
176 * to be super safe we just forbid it.
177 */
uffd_disable_fault_around(struct vm_area_struct * vma)178 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
179 {
180 return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
181 }
182
userfaultfd_missing(struct vm_area_struct * vma)183 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
184 {
185 return vma->vm_flags & VM_UFFD_MISSING;
186 }
187
userfaultfd_wp(struct vm_area_struct * vma)188 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
189 {
190 return vma->vm_flags & VM_UFFD_WP;
191 }
192
userfaultfd_minor(struct vm_area_struct * vma)193 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
194 {
195 return vma->vm_flags & VM_UFFD_MINOR;
196 }
197
userfaultfd_pte_wp(struct vm_area_struct * vma,pte_t pte)198 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
199 pte_t pte)
200 {
201 return userfaultfd_wp(vma) && pte_uffd_wp(pte);
202 }
203
userfaultfd_huge_pmd_wp(struct vm_area_struct * vma,pmd_t pmd)204 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
205 pmd_t pmd)
206 {
207 return userfaultfd_wp(vma) && pmd_uffd_wp(pmd);
208 }
209
userfaultfd_armed(struct vm_area_struct * vma)210 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
211 {
212 return vma->vm_flags & __VM_UFFD_FLAGS;
213 }
214
vma_can_userfault(struct vm_area_struct * vma,unsigned long vm_flags,bool wp_async)215 static inline bool vma_can_userfault(struct vm_area_struct *vma,
216 unsigned long vm_flags,
217 bool wp_async)
218 {
219 vm_flags &= __VM_UFFD_FLAGS;
220
221 if ((vm_flags & VM_UFFD_MINOR) &&
222 (!is_vm_hugetlb_page(vma) && !vma_is_shmem(vma)))
223 return false;
224
225 /*
226 * If wp async enabled, and WP is the only mode enabled, allow any
227 * memory type.
228 */
229 if (wp_async && (vm_flags == VM_UFFD_WP))
230 return true;
231
232 #ifndef CONFIG_PTE_MARKER_UFFD_WP
233 /*
234 * If user requested uffd-wp but not enabled pte markers for
235 * uffd-wp, then shmem & hugetlbfs are not supported but only
236 * anonymous.
237 */
238 if ((vm_flags & VM_UFFD_WP) && !vma_is_anonymous(vma))
239 return false;
240 #endif
241
242 /* By default, allow any of anon|shmem|hugetlb */
243 return vma_is_anonymous(vma) || is_vm_hugetlb_page(vma) ||
244 vma_is_shmem(vma);
245 }
246
247 extern int dup_userfaultfd(struct vm_area_struct *, struct list_head *);
248 extern void dup_userfaultfd_complete(struct list_head *);
249
250 extern void mremap_userfaultfd_prep(struct vm_area_struct *,
251 struct vm_userfaultfd_ctx *);
252 extern void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *,
253 unsigned long from, unsigned long to,
254 unsigned long len);
255
256 extern bool userfaultfd_remove(struct vm_area_struct *vma,
257 unsigned long start,
258 unsigned long end);
259
260 extern int userfaultfd_unmap_prep(struct vm_area_struct *vma,
261 unsigned long start, unsigned long end, struct list_head *uf);
262 extern void userfaultfd_unmap_complete(struct mm_struct *mm,
263 struct list_head *uf);
264 extern bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma);
265 extern bool userfaultfd_wp_async(struct vm_area_struct *vma);
266
267 #else /* CONFIG_USERFAULTFD */
268
269 /* mm helpers */
handle_userfault(struct vm_fault * vmf,unsigned long reason)270 static inline vm_fault_t handle_userfault(struct vm_fault *vmf,
271 unsigned long reason)
272 {
273 return VM_FAULT_SIGBUS;
274 }
275
uffd_wp_range(struct vm_area_struct * vma,unsigned long start,unsigned long len,bool enable_wp)276 static inline long uffd_wp_range(struct vm_area_struct *vma,
277 unsigned long start, unsigned long len,
278 bool enable_wp)
279 {
280 return false;
281 }
282
is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct * vma,struct vm_userfaultfd_ctx vm_ctx)283 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
284 struct vm_userfaultfd_ctx vm_ctx)
285 {
286 return true;
287 }
288
userfaultfd_missing(struct vm_area_struct * vma)289 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
290 {
291 return false;
292 }
293
userfaultfd_wp(struct vm_area_struct * vma)294 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
295 {
296 return false;
297 }
298
userfaultfd_minor(struct vm_area_struct * vma)299 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
300 {
301 return false;
302 }
303
userfaultfd_pte_wp(struct vm_area_struct * vma,pte_t pte)304 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
305 pte_t pte)
306 {
307 return false;
308 }
309
userfaultfd_huge_pmd_wp(struct vm_area_struct * vma,pmd_t pmd)310 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
311 pmd_t pmd)
312 {
313 return false;
314 }
315
316
userfaultfd_armed(struct vm_area_struct * vma)317 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
318 {
319 return false;
320 }
321
dup_userfaultfd(struct vm_area_struct * vma,struct list_head * l)322 static inline int dup_userfaultfd(struct vm_area_struct *vma,
323 struct list_head *l)
324 {
325 return 0;
326 }
327
dup_userfaultfd_complete(struct list_head * l)328 static inline void dup_userfaultfd_complete(struct list_head *l)
329 {
330 }
331
mremap_userfaultfd_prep(struct vm_area_struct * vma,struct vm_userfaultfd_ctx * ctx)332 static inline void mremap_userfaultfd_prep(struct vm_area_struct *vma,
333 struct vm_userfaultfd_ctx *ctx)
334 {
335 }
336
mremap_userfaultfd_complete(struct vm_userfaultfd_ctx * ctx,unsigned long from,unsigned long to,unsigned long len)337 static inline void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *ctx,
338 unsigned long from,
339 unsigned long to,
340 unsigned long len)
341 {
342 }
343
userfaultfd_remove(struct vm_area_struct * vma,unsigned long start,unsigned long end)344 static inline bool userfaultfd_remove(struct vm_area_struct *vma,
345 unsigned long start,
346 unsigned long end)
347 {
348 return true;
349 }
350
userfaultfd_unmap_prep(struct vm_area_struct * vma,unsigned long start,unsigned long end,struct list_head * uf)351 static inline int userfaultfd_unmap_prep(struct vm_area_struct *vma,
352 unsigned long start, unsigned long end,
353 struct list_head *uf)
354 {
355 return 0;
356 }
357
userfaultfd_unmap_complete(struct mm_struct * mm,struct list_head * uf)358 static inline void userfaultfd_unmap_complete(struct mm_struct *mm,
359 struct list_head *uf)
360 {
361 }
362
uffd_disable_fault_around(struct vm_area_struct * vma)363 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
364 {
365 return false;
366 }
367
userfaultfd_wp_unpopulated(struct vm_area_struct * vma)368 static inline bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma)
369 {
370 return false;
371 }
372
userfaultfd_wp_async(struct vm_area_struct * vma)373 static inline bool userfaultfd_wp_async(struct vm_area_struct *vma)
374 {
375 return false;
376 }
377
378 #endif /* CONFIG_USERFAULTFD */
379
userfaultfd_wp_use_markers(struct vm_area_struct * vma)380 static inline bool userfaultfd_wp_use_markers(struct vm_area_struct *vma)
381 {
382 /* Only wr-protect mode uses pte markers */
383 if (!userfaultfd_wp(vma))
384 return false;
385
386 /* File-based uffd-wp always need markers */
387 if (!vma_is_anonymous(vma))
388 return true;
389
390 /*
391 * Anonymous uffd-wp only needs the markers if WP_UNPOPULATED
392 * enabled (to apply markers on zero pages).
393 */
394 return userfaultfd_wp_unpopulated(vma);
395 }
396
pte_marker_entry_uffd_wp(swp_entry_t entry)397 static inline bool pte_marker_entry_uffd_wp(swp_entry_t entry)
398 {
399 #ifdef CONFIG_PTE_MARKER_UFFD_WP
400 return is_pte_marker_entry(entry) &&
401 (pte_marker_get(entry) & PTE_MARKER_UFFD_WP);
402 #else
403 return false;
404 #endif
405 }
406
pte_marker_uffd_wp(pte_t pte)407 static inline bool pte_marker_uffd_wp(pte_t pte)
408 {
409 #ifdef CONFIG_PTE_MARKER_UFFD_WP
410 swp_entry_t entry;
411
412 if (!is_swap_pte(pte))
413 return false;
414
415 entry = pte_to_swp_entry(pte);
416
417 return pte_marker_entry_uffd_wp(entry);
418 #else
419 return false;
420 #endif
421 }
422
423 /*
424 * Returns true if this is a swap pte and was uffd-wp wr-protected in either
425 * forms (pte marker or a normal swap pte), false otherwise.
426 */
pte_swp_uffd_wp_any(pte_t pte)427 static inline bool pte_swp_uffd_wp_any(pte_t pte)
428 {
429 #ifdef CONFIG_PTE_MARKER_UFFD_WP
430 if (!is_swap_pte(pte))
431 return false;
432
433 if (pte_swp_uffd_wp(pte))
434 return true;
435
436 if (pte_marker_uffd_wp(pte))
437 return true;
438 #endif
439 return false;
440 }
441
442 #endif /* _LINUX_USERFAULTFD_K_H */
443