xref: /linux/mm/vma.h (revision 334fbe734e687404f346eba7d5d96ed2b44d35ab)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * vma.h
4  *
5  * Core VMA manipulation API implemented in vma.c.
6  */
7 #ifndef __MM_VMA_H
8 #define __MM_VMA_H
9 
10 /*
11  * VMA lock generalization
12  */
13 struct vma_prepare {
14 	struct vm_area_struct *vma;
15 	struct vm_area_struct *adj_next;
16 	struct file *file;
17 	struct address_space *mapping;
18 	struct anon_vma *anon_vma;
19 	struct vm_area_struct *insert;
20 	struct vm_area_struct *remove;
21 	struct vm_area_struct *remove2;
22 
23 	bool skip_vma_uprobe :1;
24 };
25 
26 struct unlink_vma_file_batch {
27 	int count;
28 	struct vm_area_struct *vmas[8];
29 };
30 
31 /*
32  * vma munmap operation
33  */
34 struct vma_munmap_struct {
35 	struct vma_iterator *vmi;
36 	struct vm_area_struct *vma;     /* The first vma to munmap */
37 	struct vm_area_struct *prev;    /* vma before the munmap area */
38 	struct vm_area_struct *next;    /* vma after the munmap area */
39 	struct list_head *uf;           /* Userfaultfd list_head */
40 	unsigned long start;            /* Aligned start addr (inclusive) */
41 	unsigned long end;              /* Aligned end addr (exclusive) */
42 	unsigned long unmap_start;      /* Unmap PTE start */
43 	unsigned long unmap_end;        /* Unmap PTE end */
44 	int vma_count;                  /* Number of vmas that will be removed */
45 	bool unlock;                    /* Unlock after the munmap */
46 	bool clear_ptes;                /* If there are outstanding PTE to be cleared */
47 	/* 2 byte hole */
48 	unsigned long nr_pages;         /* Number of pages being removed */
49 	unsigned long locked_vm;        /* Number of locked pages */
50 	unsigned long nr_accounted;     /* Number of VM_ACCOUNT pages */
51 	unsigned long exec_vm;
52 	unsigned long stack_vm;
53 	unsigned long data_vm;
54 };
55 
56 enum vma_merge_state {
57 	VMA_MERGE_START,
58 	VMA_MERGE_ERROR_NOMEM,
59 	VMA_MERGE_NOMERGE,
60 	VMA_MERGE_SUCCESS,
61 };
62 
63 /*
64  * Describes a VMA merge operation and is threaded throughout it.
65  *
66  * Any of the fields may be mutated by the merge operation, so no guarantees are
67  * made to the contents of this structure after a merge operation has completed.
68  */
69 struct vma_merge_struct {
70 	struct mm_struct *mm;
71 	struct vma_iterator *vmi;
72 	/*
73 	 * Adjacent VMAs, any of which may be NULL if not present:
74 	 *
75 	 * |------|--------|------|
76 	 * | prev | middle | next |
77 	 * |------|--------|------|
78 	 *
79 	 * middle may not yet exist in the case of a proposed new VMA being
80 	 * merged, or it may be an existing VMA.
81 	 *
82 	 * next may be assigned by the caller.
83 	 */
84 	struct vm_area_struct *prev;
85 	struct vm_area_struct *middle;
86 	struct vm_area_struct *next;
87 	/* This is the VMA we ultimately target to become the merged VMA. */
88 	struct vm_area_struct *target;
89 	/*
90 	 * Initially, the start, end, pgoff fields are provided by the caller
91 	 * and describe the proposed new VMA range, whether modifying an
92 	 * existing VMA (which will be 'middle'), or adding a new one.
93 	 *
94 	 * During the merge process these fields are updated to describe the new
95 	 * range _including those VMAs which will be merged_.
96 	 */
97 	unsigned long start;
98 	unsigned long end;
99 	pgoff_t pgoff;
100 
101 	union {
102 		/* Temporary while VMA flags are being converted. */
103 		vm_flags_t vm_flags;
104 		vma_flags_t vma_flags;
105 	};
106 	struct file *file;
107 	struct anon_vma *anon_vma;
108 	struct mempolicy *policy;
109 	struct vm_userfaultfd_ctx uffd_ctx;
110 	struct anon_vma_name *anon_name;
111 	enum vma_merge_state state;
112 
113 	/* If copied from (i.e. mremap()'d) the VMA from which we are copying. */
114 	struct vm_area_struct *copied_from;
115 
116 	/* Flags which callers can use to modify merge behaviour: */
117 
118 	/*
119 	 * If we can expand, simply do so. We know there is nothing to merge to
120 	 * the right. Does not reset state upon failure to merge. The VMA
121 	 * iterator is assumed to be positioned at the previous VMA, rather than
122 	 * at the gap.
123 	 */
124 	bool just_expand :1;
125 
126 	/*
127 	 * If a merge is possible, but an OOM error occurs, give up and don't
128 	 * execute the merge, returning NULL.
129 	 */
130 	bool give_up_on_oom :1;
131 
132 	/*
133 	 * If set, skip uprobe_mmap upon merged vma.
134 	 */
135 	bool skip_vma_uprobe :1;
136 
137 	/* Internal flags set during merge process: */
138 
139 	/*
140 	 * Internal flag indicating the merge increases vmg->middle->vm_start
141 	 * (and thereby, vmg->prev->vm_end).
142 	 */
143 	bool __adjust_middle_start :1;
144 	/*
145 	 * Internal flag indicating the merge decreases vmg->next->vm_start
146 	 * (and thereby, vmg->middle->vm_end).
147 	 */
148 	bool __adjust_next_start :1;
149 	/*
150 	 * Internal flag used during the merge operation to indicate we will
151 	 * remove vmg->middle.
152 	 */
153 	bool __remove_middle :1;
154 	/*
155 	 * Internal flag used during the merge operation to indicate we will
156 	 * remove vmg->next.
157 	 */
158 	bool __remove_next :1;
159 
160 };
161 
162 struct unmap_desc {
163 	struct  ma_state *mas;        /* the maple state point to the first vma */
164 	struct vm_area_struct *first; /* The first vma */
165 	unsigned long pg_start;       /* The first pagetable address to free (floor) */
166 	unsigned long pg_end;         /* The last pagetable address to free (ceiling) */
167 	unsigned long vma_start;      /* The min vma address */
168 	unsigned long vma_end;        /* The max vma address */
169 	unsigned long tree_end;       /* Maximum for the vma tree search */
170 	unsigned long tree_reset;     /* Where to reset the vma tree walk */
171 	bool mm_wr_locked;            /* If the mmap write lock is held */
172 };
173 
174 /*
175  * unmap_all_init() - Initialize unmap_desc to remove all vmas, point the
176  * pg_start and pg_end to a safe location.
177  */
unmap_all_init(struct unmap_desc * unmap,struct vma_iterator * vmi,struct vm_area_struct * vma)178 static inline void unmap_all_init(struct unmap_desc *unmap,
179 		struct vma_iterator *vmi, struct vm_area_struct *vma)
180 {
181 	unmap->mas = &vmi->mas;
182 	unmap->first = vma;
183 	unmap->pg_start = FIRST_USER_ADDRESS;
184 	unmap->pg_end = USER_PGTABLES_CEILING;
185 	unmap->vma_start = 0;
186 	unmap->vma_end = ULONG_MAX;
187 	unmap->tree_end = ULONG_MAX;
188 	unmap->tree_reset = vma->vm_end;
189 	unmap->mm_wr_locked = false;
190 }
191 
192 /*
193  * unmap_pgtable_init() - Initialize unmap_desc to remove all page tables within
194  * the user range.
195  *
196  * ARM can have mappings outside of vmas.
197  * See: e2cdef8c847b4 ("[PATCH] freepgt: free_pgtables from FIRST_USER_ADDRESS")
198  *
199  * ARM LPAE uses page table mappings beyond the USER_PGTABLES_CEILING
200  * See: CONFIG_ARM_LPAE in arch/arm/include/asm/pgtable.h
201  */
unmap_pgtable_init(struct unmap_desc * unmap,struct vma_iterator * vmi)202 static inline void unmap_pgtable_init(struct unmap_desc *unmap,
203 				      struct vma_iterator *vmi)
204 {
205 	vma_iter_set(vmi, unmap->tree_reset);
206 	unmap->vma_start = FIRST_USER_ADDRESS;
207 	unmap->vma_end = USER_PGTABLES_CEILING;
208 	unmap->tree_end = USER_PGTABLES_CEILING;
209 }
210 
211 #define UNMAP_STATE(name, _vmi, _vma, _vma_start, _vma_end, _prev, _next)      \
212 	struct unmap_desc name = {                                             \
213 		.mas = &(_vmi)->mas,                                           \
214 		.first = _vma,                                                 \
215 		.pg_start = _prev ? ((struct vm_area_struct *)_prev)->vm_end : \
216 			FIRST_USER_ADDRESS,                                    \
217 		.pg_end = _next ? ((struct vm_area_struct *)_next)->vm_start : \
218 			USER_PGTABLES_CEILING,                                 \
219 		.vma_start = _vma_start,                                       \
220 		.vma_end = _vma_end,                                           \
221 		.tree_end = _next ?                                            \
222 			((struct vm_area_struct *)_next)->vm_start :           \
223 			USER_PGTABLES_CEILING,                                 \
224 		.tree_reset = _vma->vm_end,                                    \
225 		.mm_wr_locked = true,                                          \
226 	}
227 
vmg_nomem(struct vma_merge_struct * vmg)228 static inline bool vmg_nomem(struct vma_merge_struct *vmg)
229 {
230 	return vmg->state == VMA_MERGE_ERROR_NOMEM;
231 }
232 
233 /* Assumes addr >= vma->vm_start. */
vma_pgoff_offset(struct vm_area_struct * vma,unsigned long addr)234 static inline pgoff_t vma_pgoff_offset(struct vm_area_struct *vma,
235 				       unsigned long addr)
236 {
237 	return vma->vm_pgoff + PHYS_PFN(addr - vma->vm_start);
238 }
239 
240 #define VMG_STATE(name, mm_, vmi_, start_, end_, vma_flags_, pgoff_)	\
241 	struct vma_merge_struct name = {				\
242 		.mm = mm_,						\
243 		.vmi = vmi_,						\
244 		.start = start_,					\
245 		.end = end_,						\
246 		.vma_flags = vma_flags_,				\
247 		.pgoff = pgoff_,					\
248 		.state = VMA_MERGE_START,				\
249 	}
250 
251 #define VMG_VMA_STATE(name, vmi_, prev_, vma_, start_, end_)	\
252 	struct vma_merge_struct name = {			\
253 		.mm = vma_->vm_mm,				\
254 		.vmi = vmi_,					\
255 		.prev = prev_,					\
256 		.middle = vma_,					\
257 		.next = NULL,					\
258 		.start = start_,				\
259 		.end = end_,					\
260 		.vm_flags = vma_->vm_flags,			\
261 		.pgoff = vma_pgoff_offset(vma_, start_),	\
262 		.file = vma_->vm_file,				\
263 		.anon_vma = vma_->anon_vma,			\
264 		.policy = vma_policy(vma_),			\
265 		.uffd_ctx = vma_->vm_userfaultfd_ctx,		\
266 		.anon_name = anon_vma_name(vma_),		\
267 		.state = VMA_MERGE_START,			\
268 	}
269 
270 #ifdef CONFIG_DEBUG_VM_MAPLE_TREE
271 void validate_mm(struct mm_struct *mm);
272 #else
273 #define validate_mm(mm) do { } while (0)
274 #endif
275 
276 __must_check int vma_expand(struct vma_merge_struct *vmg);
277 __must_check int vma_shrink(struct vma_iterator *vmi,
278 		struct vm_area_struct *vma,
279 		unsigned long start, unsigned long end, pgoff_t pgoff);
280 
vma_iter_store_gfp(struct vma_iterator * vmi,struct vm_area_struct * vma,gfp_t gfp)281 static inline int vma_iter_store_gfp(struct vma_iterator *vmi,
282 			struct vm_area_struct *vma, gfp_t gfp)
283 
284 {
285 	if (vmi->mas.status != ma_start &&
286 	    ((vmi->mas.index > vma->vm_start) || (vmi->mas.last < vma->vm_start)))
287 		vma_iter_invalidate(vmi);
288 
289 	__mas_set_range(&vmi->mas, vma->vm_start, vma->vm_end - 1);
290 	mas_store_gfp(&vmi->mas, vma, gfp);
291 	if (unlikely(mas_is_err(&vmi->mas)))
292 		return -ENOMEM;
293 
294 	vma_mark_attached(vma);
295 	return 0;
296 }
297 
298 /*
299  * Temporary helper function for stacked mmap handlers which specify
300  * f_op->mmap() but which might have an underlying file system which implements
301  * f_op->mmap_prepare().
302  */
compat_set_vma_from_desc(struct vm_area_struct * vma,struct vm_area_desc * desc)303 static inline void compat_set_vma_from_desc(struct vm_area_struct *vma,
304 		struct vm_area_desc *desc)
305 {
306 	/*
307 	 * Since we're invoking .mmap_prepare() despite having a partially
308 	 * established VMA, we must take care to handle setting fields
309 	 * correctly.
310 	 */
311 
312 	/* Mutable fields. Populated with initial state. */
313 	vma->vm_pgoff = desc->pgoff;
314 	if (desc->vm_file != vma->vm_file)
315 		vma_set_file(vma, desc->vm_file);
316 	vma->flags = desc->vma_flags;
317 	vma->vm_page_prot = desc->page_prot;
318 
319 	/* User-defined fields. */
320 	vma->vm_ops = desc->vm_ops;
321 	vma->vm_private_data = desc->private_data;
322 }
323 
324 int
325 do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma,
326 		    struct mm_struct *mm, unsigned long start,
327 		    unsigned long end, struct list_head *uf, bool unlock);
328 
329 int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm,
330 		  unsigned long start, size_t len, struct list_head *uf,
331 		  bool unlock);
332 
333 void remove_vma(struct vm_area_struct *vma);
334 void unmap_region(struct unmap_desc *unmap);
335 
336 /**
337  * vma_modify_flags() - Perform any necessary split/merge in preparation for
338  * setting VMA flags to *@vm_flags in the range @start to @end contained within
339  * @vma.
340  * @vmi: Valid VMA iterator positioned at @vma.
341  * @prev: The VMA immediately prior to @vma or NULL if @vma is the first.
342  * @vma: The VMA containing the range @start to @end to be updated.
343  * @start: The start of the range to update. May be offset within @vma.
344  * @end: The exclusive end of the range to update, may be offset within @vma.
345  * @vma_flags_ptr: A pointer to the VMA flags that the @start to @end range is
346  * about to be set to. On merge, this will be updated to include sticky flags.
347  *
348  * IMPORTANT: The actual modification being requested here is NOT applied,
349  * rather the VMA is perhaps split, perhaps merged to accommodate the change,
350  * and the caller is expected to perform the actual modification.
351  *
352  * In order to account for sticky VMA flags, the @vma_flags_ptr parameter points
353  * to the requested flags which are then updated so the caller, should they
354  * overwrite any existing flags, correctly retains these.
355  *
356  * Returns: A VMA which contains the range @start to @end ready to have its
357  * flags altered to *@vma_flags.
358  */
359 __must_check struct vm_area_struct *vma_modify_flags(struct vma_iterator *vmi,
360 		struct vm_area_struct *prev, struct vm_area_struct *vma,
361 		unsigned long start, unsigned long end, vma_flags_t *vma_flags_ptr);
362 
363 /**
364  * vma_modify_name() - Perform any necessary split/merge in preparation for
365  * setting anonymous VMA name to @new_name in the range @start to @end contained
366  * within @vma.
367  * @vmi: Valid VMA iterator positioned at @vma.
368  * @prev: The VMA immediately prior to @vma or NULL if @vma is the first.
369  * @vma: The VMA containing the range @start to @end to be updated.
370  * @start: The start of the range to update. May be offset within @vma.
371  * @end: The exclusive end of the range to update, may be offset within @vma.
372  * @new_name: The anonymous VMA name that the @start to @end range is about to
373  * be set to.
374  *
375  * IMPORTANT: The actual modification being requested here is NOT applied,
376  * rather the VMA is perhaps split, perhaps merged to accommodate the change,
377  * and the caller is expected to perform the actual modification.
378  *
379  * Returns: A VMA which contains the range @start to @end ready to have its
380  * anonymous VMA name changed to @new_name.
381  */
382 __must_check struct vm_area_struct *vma_modify_name(struct vma_iterator *vmi,
383 		struct vm_area_struct *prev, struct vm_area_struct *vma,
384 		unsigned long start, unsigned long end,
385 		struct anon_vma_name *new_name);
386 
387 /**
388  * vma_modify_policy() - Perform any necessary split/merge in preparation for
389  * setting NUMA policy to @new_pol in the range @start to @end contained
390  * within @vma.
391  * @vmi: Valid VMA iterator positioned at @vma.
392  * @prev: The VMA immediately prior to @vma or NULL if @vma is the first.
393  * @vma: The VMA containing the range @start to @end to be updated.
394  * @start: The start of the range to update. May be offset within @vma.
395  * @end: The exclusive end of the range to update, may be offset within @vma.
396  * @new_pol: The NUMA policy that the @start to @end range is about to be set
397  * to.
398  *
399  * IMPORTANT: The actual modification being requested here is NOT applied,
400  * rather the VMA is perhaps split, perhaps merged to accommodate the change,
401  * and the caller is expected to perform the actual modification.
402  *
403  * Returns: A VMA which contains the range @start to @end ready to have its
404  * NUMA policy changed to @new_pol.
405  */
406 __must_check struct vm_area_struct *vma_modify_policy(struct vma_iterator *vmi,
407 		   struct vm_area_struct *prev, struct vm_area_struct *vma,
408 		   unsigned long start, unsigned long end,
409 		   struct mempolicy *new_pol);
410 
411 /**
412  * vma_modify_flags_uffd() - Perform any necessary split/merge in preparation for
413  * setting VMA flags to @vm_flags and UFFD context to @new_ctx in the range
414  * @start to @end contained within @vma.
415  * @vmi: Valid VMA iterator positioned at @vma.
416  * @prev: The VMA immediately prior to @vma or NULL if @vma is the first.
417  * @vma: The VMA containing the range @start to @end to be updated.
418  * @start: The start of the range to update. May be offset within @vma.
419  * @end: The exclusive end of the range to update, may be offset within @vma.
420  * @vma_flags: The VMA flags that the @start to @end range is about to be set to.
421  * @new_ctx: The userfaultfd context that the @start to @end range is about to
422  * be set to.
423  * @give_up_on_oom: If an out of memory condition occurs on merge, simply give
424  * up on it and treat the merge as best-effort.
425  *
426  * IMPORTANT: The actual modification being requested here is NOT applied,
427  * rather the VMA is perhaps split, perhaps merged to accommodate the change,
428  * and the caller is expected to perform the actual modification.
429  *
430  * Returns: A VMA which contains the range @start to @end ready to have its VMA
431  * flags changed to @vma_flags and its userfaultfd context changed to @new_ctx.
432  */
433 __must_check struct vm_area_struct *vma_modify_flags_uffd(struct vma_iterator *vmi,
434 		struct vm_area_struct *prev, struct vm_area_struct *vma,
435 		unsigned long start, unsigned long end, const vma_flags_t *vma_flags,
436 		struct vm_userfaultfd_ctx new_ctx, bool give_up_on_oom);
437 
438 __must_check struct vm_area_struct *vma_merge_new_range(struct vma_merge_struct *vmg);
439 
440 __must_check struct vm_area_struct *vma_merge_extend(struct vma_iterator *vmi,
441 		  struct vm_area_struct *vma, unsigned long delta);
442 
443 void unlink_file_vma_batch_init(struct unlink_vma_file_batch *vb);
444 
445 void unlink_file_vma_batch_final(struct unlink_vma_file_batch *vb);
446 
447 void unlink_file_vma_batch_add(struct unlink_vma_file_batch *vb,
448 			       struct vm_area_struct *vma);
449 
450 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
451 	unsigned long addr, unsigned long len, pgoff_t pgoff,
452 	bool *need_rmap_locks);
453 
454 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma);
455 
456 bool vma_needs_dirty_tracking(struct vm_area_struct *vma);
457 bool vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot);
458 
459 int mm_take_all_locks(struct mm_struct *mm);
460 void mm_drop_all_locks(struct mm_struct *mm);
461 
462 unsigned long mmap_region(struct file *file, unsigned long addr,
463 		unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
464 		struct list_head *uf);
465 
466 int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *brkvma,
467 		 unsigned long addr, unsigned long request,
468 		 vma_flags_t vma_flags);
469 
470 unsigned long unmapped_area(struct vm_unmapped_area_info *info);
471 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info);
472 
vma_wants_manual_pte_write_upgrade(struct vm_area_struct * vma)473 static inline bool vma_wants_manual_pte_write_upgrade(struct vm_area_struct *vma)
474 {
475 	/*
476 	 * We want to check manually if we can change individual PTEs writable
477 	 * if we can't do that automatically for all PTEs in a mapping. For
478 	 * private mappings, that's always the case when we have write
479 	 * permissions as we properly have to handle COW.
480 	 */
481 	if (vma->vm_flags & VM_SHARED)
482 		return vma_wants_writenotify(vma, vma->vm_page_prot);
483 	return !!(vma->vm_flags & VM_WRITE);
484 }
485 
486 #ifdef CONFIG_MMU
vm_pgprot_modify(pgprot_t oldprot,vm_flags_t vm_flags)487 static inline pgprot_t vm_pgprot_modify(pgprot_t oldprot, vm_flags_t vm_flags)
488 {
489 	return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
490 }
491 #endif
492 
vma_prev_limit(struct vma_iterator * vmi,unsigned long min)493 static inline struct vm_area_struct *vma_prev_limit(struct vma_iterator *vmi,
494 						    unsigned long min)
495 {
496 	return mas_prev(&vmi->mas, min);
497 }
498 
499 /*
500  * These three helpers classifies VMAs for virtual memory accounting.
501  */
502 
503 /*
504  * Executable code area - executable, not writable, not stack
505  */
is_exec_mapping(vm_flags_t flags)506 static inline bool is_exec_mapping(vm_flags_t flags)
507 {
508 	return (flags & (VM_EXEC | VM_WRITE | VM_STACK)) == VM_EXEC;
509 }
510 
511 /*
512  * Stack area (including shadow stacks)
513  *
514  * VM_GROWSUP / VM_GROWSDOWN VMAs are always private anonymous:
515  * do_mmap() forbids all other combinations.
516  */
is_stack_mapping(vm_flags_t flags)517 static inline bool is_stack_mapping(vm_flags_t flags)
518 {
519 	return ((flags & VM_STACK) == VM_STACK) || (flags & VM_SHADOW_STACK);
520 }
521 
522 /*
523  * Data area - private, writable, not stack
524  */
is_data_mapping(vm_flags_t flags)525 static inline bool is_data_mapping(vm_flags_t flags)
526 {
527 	return (flags & (VM_WRITE | VM_SHARED | VM_STACK)) == VM_WRITE;
528 }
529 
is_data_mapping_vma_flags(const vma_flags_t * vma_flags)530 static inline bool is_data_mapping_vma_flags(const vma_flags_t *vma_flags)
531 {
532 	return vma_flags_test(vma_flags, VMA_WRITE_BIT) &&
533 		!vma_flags_test_any(vma_flags, VMA_SHARED_BIT, VMA_STACK_BIT);
534 }
535 
vma_iter_config(struct vma_iterator * vmi,unsigned long index,unsigned long last)536 static inline void vma_iter_config(struct vma_iterator *vmi,
537 		unsigned long index, unsigned long last)
538 {
539 	__mas_set_range(&vmi->mas, index, last - 1);
540 }
541 
vma_iter_reset(struct vma_iterator * vmi)542 static inline void vma_iter_reset(struct vma_iterator *vmi)
543 {
544 	mas_reset(&vmi->mas);
545 }
546 
547 static inline
vma_iter_prev_range_limit(struct vma_iterator * vmi,unsigned long min)548 struct vm_area_struct *vma_iter_prev_range_limit(struct vma_iterator *vmi, unsigned long min)
549 {
550 	return mas_prev_range(&vmi->mas, min);
551 }
552 
553 static inline
vma_iter_next_range_limit(struct vma_iterator * vmi,unsigned long max)554 struct vm_area_struct *vma_iter_next_range_limit(struct vma_iterator *vmi, unsigned long max)
555 {
556 	return mas_next_range(&vmi->mas, max);
557 }
558 
vma_iter_area_lowest(struct vma_iterator * vmi,unsigned long min,unsigned long max,unsigned long size)559 static inline int vma_iter_area_lowest(struct vma_iterator *vmi, unsigned long min,
560 				       unsigned long max, unsigned long size)
561 {
562 	return mas_empty_area(&vmi->mas, min, max - 1, size);
563 }
564 
vma_iter_area_highest(struct vma_iterator * vmi,unsigned long min,unsigned long max,unsigned long size)565 static inline int vma_iter_area_highest(struct vma_iterator *vmi, unsigned long min,
566 					unsigned long max, unsigned long size)
567 {
568 	return mas_empty_area_rev(&vmi->mas, min, max - 1, size);
569 }
570 
571 /*
572  * VMA Iterator functions shared between nommu and mmap
573  */
vma_iter_prealloc(struct vma_iterator * vmi,struct vm_area_struct * vma)574 static inline int vma_iter_prealloc(struct vma_iterator *vmi,
575 		struct vm_area_struct *vma)
576 {
577 	return mas_preallocate(&vmi->mas, vma, GFP_KERNEL);
578 }
579 
vma_iter_clear(struct vma_iterator * vmi)580 static inline void vma_iter_clear(struct vma_iterator *vmi)
581 {
582 	mas_store_prealloc(&vmi->mas, NULL);
583 }
584 
vma_iter_load(struct vma_iterator * vmi)585 static inline struct vm_area_struct *vma_iter_load(struct vma_iterator *vmi)
586 {
587 	return mas_walk(&vmi->mas);
588 }
589 
590 /* Store a VMA with preallocated memory */
vma_iter_store_overwrite(struct vma_iterator * vmi,struct vm_area_struct * vma)591 static inline void vma_iter_store_overwrite(struct vma_iterator *vmi,
592 					    struct vm_area_struct *vma)
593 {
594 	vma_assert_attached(vma);
595 
596 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
597 	if (MAS_WARN_ON(&vmi->mas, vmi->mas.status != ma_start &&
598 			vmi->mas.index > vma->vm_start)) {
599 		pr_warn("%lx > %lx\n store vma %lx-%lx\n into slot %lx-%lx\n",
600 			vmi->mas.index, vma->vm_start, vma->vm_start,
601 			vma->vm_end, vmi->mas.index, vmi->mas.last);
602 	}
603 	if (MAS_WARN_ON(&vmi->mas, vmi->mas.status != ma_start &&
604 			vmi->mas.last <  vma->vm_start)) {
605 		pr_warn("%lx < %lx\nstore vma %lx-%lx\ninto slot %lx-%lx\n",
606 		       vmi->mas.last, vma->vm_start, vma->vm_start, vma->vm_end,
607 		       vmi->mas.index, vmi->mas.last);
608 	}
609 #endif
610 
611 	if (vmi->mas.status != ma_start &&
612 	    ((vmi->mas.index > vma->vm_start) || (vmi->mas.last < vma->vm_start)))
613 		vma_iter_invalidate(vmi);
614 
615 	__mas_set_range(&vmi->mas, vma->vm_start, vma->vm_end - 1);
616 	mas_store_prealloc(&vmi->mas, vma);
617 }
618 
vma_iter_store_new(struct vma_iterator * vmi,struct vm_area_struct * vma)619 static inline void vma_iter_store_new(struct vma_iterator *vmi,
620 				      struct vm_area_struct *vma)
621 {
622 	vma_mark_attached(vma);
623 	vma_iter_store_overwrite(vmi, vma);
624 }
625 
vma_iter_addr(struct vma_iterator * vmi)626 static inline unsigned long vma_iter_addr(struct vma_iterator *vmi)
627 {
628 	return vmi->mas.index;
629 }
630 
vma_iter_end(struct vma_iterator * vmi)631 static inline unsigned long vma_iter_end(struct vma_iterator *vmi)
632 {
633 	return vmi->mas.last + 1;
634 }
635 
636 static inline
vma_iter_prev_range(struct vma_iterator * vmi)637 struct vm_area_struct *vma_iter_prev_range(struct vma_iterator *vmi)
638 {
639 	return mas_prev_range(&vmi->mas, 0);
640 }
641 
642 /*
643  * Retrieve the next VMA and rewind the iterator to end of the previous VMA, or
644  * if no previous VMA, to index 0.
645  */
646 static inline
vma_iter_next_rewind(struct vma_iterator * vmi,struct vm_area_struct ** pprev)647 struct vm_area_struct *vma_iter_next_rewind(struct vma_iterator *vmi,
648 		struct vm_area_struct **pprev)
649 {
650 	struct vm_area_struct *next = vma_next(vmi);
651 	struct vm_area_struct *prev = vma_prev(vmi);
652 
653 	/*
654 	 * Consider the case where no previous VMA exists. We advance to the
655 	 * next VMA, skipping any gap, then rewind to the start of the range.
656 	 *
657 	 * If we were to unconditionally advance to the next range we'd wind up
658 	 * at the next VMA again, so we check to ensure there is a previous VMA
659 	 * to skip over.
660 	 */
661 	if (prev)
662 		vma_iter_next_range(vmi);
663 
664 	if (pprev)
665 		*pprev = prev;
666 
667 	return next;
668 }
669 
670 #ifdef CONFIG_64BIT
vma_is_sealed(struct vm_area_struct * vma)671 static inline bool vma_is_sealed(struct vm_area_struct *vma)
672 {
673 	return (vma->vm_flags & VM_SEALED);
674 }
675 #else
vma_is_sealed(struct vm_area_struct * vma)676 static inline bool vma_is_sealed(struct vm_area_struct *vma)
677 {
678 	return false;
679 }
680 #endif
681 
682 #if defined(CONFIG_STACK_GROWSUP)
683 int expand_upwards(struct vm_area_struct *vma, unsigned long address);
684 #endif
685 
686 int expand_downwards(struct vm_area_struct *vma, unsigned long address);
687 
688 int __vm_munmap(unsigned long start, size_t len, bool unlock);
689 
690 int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma);
691 
692 /* vma_init.h, shared between CONFIG_MMU and nommu. */
693 void __init vma_state_init(void);
694 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm);
695 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig);
696 void vm_area_free(struct vm_area_struct *vma);
697 
698 /* vma_exec.c */
699 #ifdef CONFIG_MMU
700 int create_init_stack_vma(struct mm_struct *mm, struct vm_area_struct **vmap,
701 			  unsigned long *top_mem_p);
702 int relocate_vma_down(struct vm_area_struct *vma, unsigned long shift);
703 #endif
704 
705 #ifdef CONFIG_MMU
706 /*
707  * Denies creating a writable executable mapping or gaining executable permissions.
708  *
709  * This denies the following:
710  *
711  *	a)	mmap(PROT_WRITE | PROT_EXEC)
712  *
713  *	b)	mmap(PROT_WRITE)
714  *		mprotect(PROT_EXEC)
715  *
716  *	c)	mmap(PROT_WRITE)
717  *		mprotect(PROT_READ)
718  *		mprotect(PROT_EXEC)
719  *
720  * But allows the following:
721  *
722  *	d)	mmap(PROT_READ | PROT_EXEC)
723  *		mmap(PROT_READ | PROT_EXEC | PROT_BTI)
724  *
725  * This is only applicable if the user has set the Memory-Deny-Write-Execute
726  * (MDWE) protection mask for the current process.
727  *
728  * @old specifies the VMA flags the VMA originally possessed, and @new the ones
729  * we propose to set.
730  *
731  * Return: false if proposed change is OK, true if not ok and should be denied.
732  */
map_deny_write_exec(const vma_flags_t * old,const vma_flags_t * new)733 static inline bool map_deny_write_exec(const vma_flags_t *old,
734 				       const vma_flags_t *new)
735 {
736 	/* If MDWE is disabled, we have nothing to deny. */
737 	if (!mm_flags_test(MMF_HAS_MDWE, current->mm))
738 		return false;
739 
740 	/* If the new VMA is not executable, we have nothing to deny. */
741 	if (!vma_flags_test(new, VMA_EXEC_BIT))
742 		return false;
743 
744 	/* Under MDWE we do not accept newly writably executable VMAs... */
745 	if (vma_flags_test(new, VMA_WRITE_BIT))
746 		return true;
747 
748 	/* ...nor previously non-executable VMAs becoming executable. */
749 	if (!vma_flags_test(old, VMA_EXEC_BIT))
750 		return true;
751 
752 	return false;
753 }
754 #endif
755 
756 #endif	/* __MM_VMA_H */
757