xref: /linux/include/linux/userfaultfd_k.h (revision beace86e61e465dba204a268ab3f3377153a4973)
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_SHARED_FCNTL_FLAGS (O_CLOEXEC | O_NONBLOCK)
34 
35 /*
36  * Start with fault_pending_wqh and fault_wqh so they're more likely
37  * to be in the same cacheline.
38  *
39  * Locking order:
40  *	fd_wqh.lock
41  *		fault_pending_wqh.lock
42  *			fault_wqh.lock
43  *		event_wqh.lock
44  *
45  * To avoid deadlocks, IRQs must be disabled when taking any of the above locks,
46  * since fd_wqh.lock is taken by aio_poll() while it's holding a lock that's
47  * also taken in IRQ context.
48  */
49 struct userfaultfd_ctx {
50 	/* waitqueue head for the pending (i.e. not read) userfaults */
51 	wait_queue_head_t fault_pending_wqh;
52 	/* waitqueue head for the userfaults */
53 	wait_queue_head_t fault_wqh;
54 	/* waitqueue head for the pseudo fd to wakeup poll/read */
55 	wait_queue_head_t fd_wqh;
56 	/* waitqueue head for events */
57 	wait_queue_head_t event_wqh;
58 	/* a refile sequence protected by fault_pending_wqh lock */
59 	seqcount_spinlock_t refile_seq;
60 	/* pseudo fd refcounting */
61 	refcount_t refcount;
62 	/* userfaultfd syscall flags */
63 	unsigned int flags;
64 	/* features requested from the userspace */
65 	unsigned int features;
66 	/* released */
67 	bool released;
68 	/*
69 	 * Prevents userfaultfd operations (fill/move/wp) from happening while
70 	 * some non-cooperative event(s) is taking place. Increments are done
71 	 * in write-mode. Whereas, userfaultfd operations, which includes
72 	 * reading mmap_changing, is done under read-mode.
73 	 */
74 	struct rw_semaphore map_changing_lock;
75 	/* memory mappings are changing because of non-cooperative event */
76 	atomic_t mmap_changing;
77 	/* mm with one ore more vmas attached to this userfaultfd_ctx */
78 	struct mm_struct *mm;
79 };
80 
81 extern vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason);
82 
83 /* A combined operation mode + behavior flags. */
84 typedef unsigned int __bitwise uffd_flags_t;
85 
86 /* Mutually exclusive modes of operation. */
87 enum mfill_atomic_mode {
88 	MFILL_ATOMIC_COPY,
89 	MFILL_ATOMIC_ZEROPAGE,
90 	MFILL_ATOMIC_CONTINUE,
91 	MFILL_ATOMIC_POISON,
92 	NR_MFILL_ATOMIC_MODES,
93 };
94 
95 #define MFILL_ATOMIC_MODE_BITS (const_ilog2(NR_MFILL_ATOMIC_MODES - 1) + 1)
96 #define MFILL_ATOMIC_BIT(nr) BIT(MFILL_ATOMIC_MODE_BITS + (nr))
97 #define MFILL_ATOMIC_FLAG(nr) ((__force uffd_flags_t) MFILL_ATOMIC_BIT(nr))
98 #define MFILL_ATOMIC_MODE_MASK ((__force uffd_flags_t) (MFILL_ATOMIC_BIT(0) - 1))
99 
100 static inline bool uffd_flags_mode_is(uffd_flags_t flags, enum mfill_atomic_mode expected)
101 {
102 	return (flags & MFILL_ATOMIC_MODE_MASK) == ((__force uffd_flags_t) expected);
103 }
104 
105 static inline uffd_flags_t uffd_flags_set_mode(uffd_flags_t flags, enum mfill_atomic_mode mode)
106 {
107 	flags &= ~MFILL_ATOMIC_MODE_MASK;
108 	return flags | ((__force uffd_flags_t) mode);
109 }
110 
111 /* Flags controlling behavior. These behavior changes are mode-independent. */
112 #define MFILL_ATOMIC_WP MFILL_ATOMIC_FLAG(0)
113 
114 extern int mfill_atomic_install_pte(pmd_t *dst_pmd,
115 				    struct vm_area_struct *dst_vma,
116 				    unsigned long dst_addr, struct page *page,
117 				    bool newly_allocated, uffd_flags_t flags);
118 
119 extern ssize_t mfill_atomic_copy(struct userfaultfd_ctx *ctx, unsigned long dst_start,
120 				 unsigned long src_start, unsigned long len,
121 				 uffd_flags_t flags);
122 extern ssize_t mfill_atomic_zeropage(struct userfaultfd_ctx *ctx,
123 				     unsigned long dst_start,
124 				     unsigned long len);
125 extern ssize_t mfill_atomic_continue(struct userfaultfd_ctx *ctx, unsigned long dst_start,
126 				     unsigned long len, uffd_flags_t flags);
127 extern ssize_t mfill_atomic_poison(struct userfaultfd_ctx *ctx, unsigned long start,
128 				   unsigned long len, uffd_flags_t flags);
129 extern int mwriteprotect_range(struct userfaultfd_ctx *ctx, unsigned long start,
130 			       unsigned long len, bool enable_wp);
131 extern long uffd_wp_range(struct vm_area_struct *vma,
132 			  unsigned long start, unsigned long len, bool enable_wp);
133 
134 /* move_pages */
135 void double_pt_lock(spinlock_t *ptl1, spinlock_t *ptl2);
136 void double_pt_unlock(spinlock_t *ptl1, spinlock_t *ptl2);
137 ssize_t move_pages(struct userfaultfd_ctx *ctx, unsigned long dst_start,
138 		   unsigned long src_start, unsigned long len, __u64 flags);
139 int move_pages_huge_pmd(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd, pmd_t dst_pmdval,
140 			struct vm_area_struct *dst_vma,
141 			struct vm_area_struct *src_vma,
142 			unsigned long dst_addr, unsigned long src_addr);
143 
144 /* mm helpers */
145 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
146 					struct vm_userfaultfd_ctx vm_ctx)
147 {
148 	return vma->vm_userfaultfd_ctx.ctx == vm_ctx.ctx;
149 }
150 
151 /*
152  * Never enable huge pmd sharing on some uffd registered vmas:
153  *
154  * - VM_UFFD_WP VMAs, because write protect information is per pgtable entry.
155  *
156  * - VM_UFFD_MINOR VMAs, because otherwise we would never get minor faults for
157  *   VMAs which share huge pmds. (If you have two mappings to the same
158  *   underlying pages, and fault in the non-UFFD-registered one with a write,
159  *   with huge pmd sharing this would *also* setup the second UFFD-registered
160  *   mapping, and we'd not get minor faults.)
161  */
162 static inline bool uffd_disable_huge_pmd_share(struct vm_area_struct *vma)
163 {
164 	return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
165 }
166 
167 /*
168  * Don't do fault around for either WP or MINOR registered uffd range.  For
169  * MINOR registered range, fault around will be a total disaster and ptes can
170  * be installed without notifications; for WP it should mostly be fine as long
171  * as the fault around checks for pte_none() before the installation, however
172  * to be super safe we just forbid it.
173  */
174 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
175 {
176 	return vma->vm_flags & (VM_UFFD_WP | VM_UFFD_MINOR);
177 }
178 
179 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
180 {
181 	return vma->vm_flags & VM_UFFD_MISSING;
182 }
183 
184 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
185 {
186 	return vma->vm_flags & VM_UFFD_WP;
187 }
188 
189 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
190 {
191 	return vma->vm_flags & VM_UFFD_MINOR;
192 }
193 
194 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
195 				      pte_t pte)
196 {
197 	return userfaultfd_wp(vma) && pte_uffd_wp(pte);
198 }
199 
200 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
201 					   pmd_t pmd)
202 {
203 	return userfaultfd_wp(vma) && pmd_uffd_wp(pmd);
204 }
205 
206 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
207 {
208 	return vma->vm_flags & __VM_UFFD_FLAGS;
209 }
210 
211 static inline bool vma_can_userfault(struct vm_area_struct *vma,
212 				     vm_flags_t vm_flags,
213 				     bool wp_async)
214 {
215 	vm_flags &= __VM_UFFD_FLAGS;
216 
217 	if (vma->vm_flags & VM_DROPPABLE)
218 		return false;
219 
220 	if ((vm_flags & VM_UFFD_MINOR) &&
221 	    (!is_vm_hugetlb_page(vma) && !vma_is_shmem(vma)))
222 		return false;
223 
224 	/*
225 	 * If wp async enabled, and WP is the only mode enabled, allow any
226 	 * memory type.
227 	 */
228 	if (wp_async && (vm_flags == VM_UFFD_WP))
229 		return true;
230 
231 #ifndef CONFIG_PTE_MARKER_UFFD_WP
232 	/*
233 	 * If user requested uffd-wp but not enabled pte markers for
234 	 * uffd-wp, then shmem & hugetlbfs are not supported but only
235 	 * anonymous.
236 	 */
237 	if ((vm_flags & VM_UFFD_WP) && !vma_is_anonymous(vma))
238 		return false;
239 #endif
240 
241 	/* By default, allow any of anon|shmem|hugetlb */
242 	return vma_is_anonymous(vma) || is_vm_hugetlb_page(vma) ||
243 	    vma_is_shmem(vma);
244 }
245 
246 static inline bool vma_has_uffd_without_event_remap(struct vm_area_struct *vma)
247 {
248 	struct userfaultfd_ctx *uffd_ctx = vma->vm_userfaultfd_ctx.ctx;
249 
250 	return uffd_ctx && (uffd_ctx->features & UFFD_FEATURE_EVENT_REMAP) == 0;
251 }
252 
253 extern int dup_userfaultfd(struct vm_area_struct *, struct list_head *);
254 extern void dup_userfaultfd_complete(struct list_head *);
255 void dup_userfaultfd_fail(struct list_head *);
256 
257 extern void mremap_userfaultfd_prep(struct vm_area_struct *,
258 				    struct vm_userfaultfd_ctx *);
259 extern void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *,
260 					unsigned long from, unsigned long to,
261 					unsigned long len);
262 void mremap_userfaultfd_fail(struct vm_userfaultfd_ctx *);
263 
264 extern bool userfaultfd_remove(struct vm_area_struct *vma,
265 			       unsigned long start,
266 			       unsigned long end);
267 
268 extern int userfaultfd_unmap_prep(struct vm_area_struct *vma,
269 		unsigned long start, unsigned long end, struct list_head *uf);
270 extern void userfaultfd_unmap_complete(struct mm_struct *mm,
271 				       struct list_head *uf);
272 extern bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma);
273 extern bool userfaultfd_wp_async(struct vm_area_struct *vma);
274 
275 void userfaultfd_reset_ctx(struct vm_area_struct *vma);
276 
277 struct vm_area_struct *userfaultfd_clear_vma(struct vma_iterator *vmi,
278 					     struct vm_area_struct *prev,
279 					     struct vm_area_struct *vma,
280 					     unsigned long start,
281 					     unsigned long end);
282 
283 int userfaultfd_register_range(struct userfaultfd_ctx *ctx,
284 			       struct vm_area_struct *vma,
285 			       vm_flags_t vm_flags,
286 			       unsigned long start, unsigned long end,
287 			       bool wp_async);
288 
289 void userfaultfd_release_new(struct userfaultfd_ctx *ctx);
290 
291 void userfaultfd_release_all(struct mm_struct *mm,
292 			     struct userfaultfd_ctx *ctx);
293 
294 #else /* CONFIG_USERFAULTFD */
295 
296 /* mm helpers */
297 static inline vm_fault_t handle_userfault(struct vm_fault *vmf,
298 				unsigned long reason)
299 {
300 	return VM_FAULT_SIGBUS;
301 }
302 
303 static inline long uffd_wp_range(struct vm_area_struct *vma,
304 				 unsigned long start, unsigned long len,
305 				 bool enable_wp)
306 {
307 	return false;
308 }
309 
310 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
311 					struct vm_userfaultfd_ctx vm_ctx)
312 {
313 	return true;
314 }
315 
316 static inline bool userfaultfd_missing(struct vm_area_struct *vma)
317 {
318 	return false;
319 }
320 
321 static inline bool userfaultfd_wp(struct vm_area_struct *vma)
322 {
323 	return false;
324 }
325 
326 static inline bool userfaultfd_minor(struct vm_area_struct *vma)
327 {
328 	return false;
329 }
330 
331 static inline bool userfaultfd_pte_wp(struct vm_area_struct *vma,
332 				      pte_t pte)
333 {
334 	return false;
335 }
336 
337 static inline bool userfaultfd_huge_pmd_wp(struct vm_area_struct *vma,
338 					   pmd_t pmd)
339 {
340 	return false;
341 }
342 
343 
344 static inline bool userfaultfd_armed(struct vm_area_struct *vma)
345 {
346 	return false;
347 }
348 
349 static inline int dup_userfaultfd(struct vm_area_struct *vma,
350 				  struct list_head *l)
351 {
352 	return 0;
353 }
354 
355 static inline void dup_userfaultfd_complete(struct list_head *l)
356 {
357 }
358 
359 static inline void dup_userfaultfd_fail(struct list_head *l)
360 {
361 }
362 
363 static inline void mremap_userfaultfd_prep(struct vm_area_struct *vma,
364 					   struct vm_userfaultfd_ctx *ctx)
365 {
366 }
367 
368 static inline void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *ctx,
369 					       unsigned long from,
370 					       unsigned long to,
371 					       unsigned long len)
372 {
373 }
374 
375 static inline void mremap_userfaultfd_fail(struct vm_userfaultfd_ctx *ctx)
376 {
377 }
378 
379 static inline bool userfaultfd_remove(struct vm_area_struct *vma,
380 				      unsigned long start,
381 				      unsigned long end)
382 {
383 	return true;
384 }
385 
386 static inline int userfaultfd_unmap_prep(struct vm_area_struct *vma,
387 					 unsigned long start, unsigned long end,
388 					 struct list_head *uf)
389 {
390 	return 0;
391 }
392 
393 static inline void userfaultfd_unmap_complete(struct mm_struct *mm,
394 					      struct list_head *uf)
395 {
396 }
397 
398 static inline bool uffd_disable_fault_around(struct vm_area_struct *vma)
399 {
400 	return false;
401 }
402 
403 static inline bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma)
404 {
405 	return false;
406 }
407 
408 static inline bool userfaultfd_wp_async(struct vm_area_struct *vma)
409 {
410 	return false;
411 }
412 
413 static inline bool vma_has_uffd_without_event_remap(struct vm_area_struct *vma)
414 {
415 	return false;
416 }
417 
418 #endif /* CONFIG_USERFAULTFD */
419 
420 static inline bool userfaultfd_wp_use_markers(struct vm_area_struct *vma)
421 {
422 	/* Only wr-protect mode uses pte markers */
423 	if (!userfaultfd_wp(vma))
424 		return false;
425 
426 	/* File-based uffd-wp always need markers */
427 	if (!vma_is_anonymous(vma))
428 		return true;
429 
430 	/*
431 	 * Anonymous uffd-wp only needs the markers if WP_UNPOPULATED
432 	 * enabled (to apply markers on zero pages).
433 	 */
434 	return userfaultfd_wp_unpopulated(vma);
435 }
436 
437 static inline bool pte_marker_entry_uffd_wp(swp_entry_t entry)
438 {
439 #ifdef CONFIG_PTE_MARKER_UFFD_WP
440 	return is_pte_marker_entry(entry) &&
441 	    (pte_marker_get(entry) & PTE_MARKER_UFFD_WP);
442 #else
443 	return false;
444 #endif
445 }
446 
447 static inline bool pte_marker_uffd_wp(pte_t pte)
448 {
449 #ifdef CONFIG_PTE_MARKER_UFFD_WP
450 	swp_entry_t entry;
451 
452 	if (!is_swap_pte(pte))
453 		return false;
454 
455 	entry = pte_to_swp_entry(pte);
456 
457 	return pte_marker_entry_uffd_wp(entry);
458 #else
459 	return false;
460 #endif
461 }
462 
463 /*
464  * Returns true if this is a swap pte and was uffd-wp wr-protected in either
465  * forms (pte marker or a normal swap pte), false otherwise.
466  */
467 static inline bool pte_swp_uffd_wp_any(pte_t pte)
468 {
469 #ifdef CONFIG_PTE_MARKER_UFFD_WP
470 	if (!is_swap_pte(pte))
471 		return false;
472 
473 	if (pte_swp_uffd_wp(pte))
474 		return true;
475 
476 	if (pte_marker_uffd_wp(pte))
477 		return true;
478 #endif
479 	return false;
480 }
481 
482 #endif /* _LINUX_USERFAULTFD_K_H */
483