xref: /openbmc/linux/mm/nommu.c (revision 54618888)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/mm/nommu.c
4  *
5  *  Replacement code for mm functions to support CPU's that don't
6  *  have any form of memory management unit (thus no virtual memory).
7  *
8  *  See Documentation/admin-guide/mm/nommu-mmap.rst
9  *
10  *  Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
11  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
12  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
13  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
14  *  Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
15  */
16 
17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18 
19 #include <linux/export.h>
20 #include <linux/mm.h>
21 #include <linux/sched/mm.h>
22 #include <linux/mman.h>
23 #include <linux/swap.h>
24 #include <linux/file.h>
25 #include <linux/highmem.h>
26 #include <linux/pagemap.h>
27 #include <linux/slab.h>
28 #include <linux/vmalloc.h>
29 #include <linux/backing-dev.h>
30 #include <linux/compiler.h>
31 #include <linux/mount.h>
32 #include <linux/personality.h>
33 #include <linux/security.h>
34 #include <linux/syscalls.h>
35 #include <linux/audit.h>
36 #include <linux/printk.h>
37 
38 #include <linux/uaccess.h>
39 #include <asm/tlb.h>
40 #include <asm/tlbflush.h>
41 #include <asm/mmu_context.h>
42 #include "internal.h"
43 
44 void *high_memory;
45 EXPORT_SYMBOL(high_memory);
46 struct page *mem_map;
47 unsigned long max_mapnr;
48 EXPORT_SYMBOL(max_mapnr);
49 unsigned long highest_memmap_pfn;
50 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
51 int heap_stack_gap = 0;
52 
53 atomic_long_t mmap_pages_allocated;
54 
55 EXPORT_SYMBOL(mem_map);
56 
57 /* list of mapped, potentially shareable regions */
58 static struct kmem_cache *vm_region_jar;
59 struct rb_root nommu_region_tree = RB_ROOT;
60 DECLARE_RWSEM(nommu_region_sem);
61 
62 const struct vm_operations_struct generic_file_vm_ops = {
63 };
64 
65 /*
66  * Return the total memory allocated for this pointer, not
67  * just what the caller asked for.
68  *
69  * Doesn't have to be accurate, i.e. may have races.
70  */
71 unsigned int kobjsize(const void *objp)
72 {
73 	struct page *page;
74 
75 	/*
76 	 * If the object we have should not have ksize performed on it,
77 	 * return size of 0
78 	 */
79 	if (!objp || !virt_addr_valid(objp))
80 		return 0;
81 
82 	page = virt_to_head_page(objp);
83 
84 	/*
85 	 * If the allocator sets PageSlab, we know the pointer came from
86 	 * kmalloc().
87 	 */
88 	if (PageSlab(page))
89 		return ksize(objp);
90 
91 	/*
92 	 * If it's not a compound page, see if we have a matching VMA
93 	 * region. This test is intentionally done in reverse order,
94 	 * so if there's no VMA, we still fall through and hand back
95 	 * PAGE_SIZE for 0-order pages.
96 	 */
97 	if (!PageCompound(page)) {
98 		struct vm_area_struct *vma;
99 
100 		vma = find_vma(current->mm, (unsigned long)objp);
101 		if (vma)
102 			return vma->vm_end - vma->vm_start;
103 	}
104 
105 	/*
106 	 * The ksize() function is only guaranteed to work for pointers
107 	 * returned by kmalloc(). So handle arbitrary pointers here.
108 	 */
109 	return page_size(page);
110 }
111 
112 /**
113  * follow_pfn - look up PFN at a user virtual address
114  * @vma: memory mapping
115  * @address: user virtual address
116  * @pfn: location to store found PFN
117  *
118  * Only IO mappings and raw PFN mappings are allowed.
119  *
120  * Returns zero and the pfn at @pfn on success, -ve otherwise.
121  */
122 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
123 	unsigned long *pfn)
124 {
125 	if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
126 		return -EINVAL;
127 
128 	*pfn = address >> PAGE_SHIFT;
129 	return 0;
130 }
131 EXPORT_SYMBOL(follow_pfn);
132 
133 LIST_HEAD(vmap_area_list);
134 
135 void vfree(const void *addr)
136 {
137 	kfree(addr);
138 }
139 EXPORT_SYMBOL(vfree);
140 
141 void *__vmalloc(unsigned long size, gfp_t gfp_mask)
142 {
143 	/*
144 	 *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
145 	 * returns only a logical address.
146 	 */
147 	return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
148 }
149 EXPORT_SYMBOL(__vmalloc);
150 
151 void *__vmalloc_node_range(unsigned long size, unsigned long align,
152 		unsigned long start, unsigned long end, gfp_t gfp_mask,
153 		pgprot_t prot, unsigned long vm_flags, int node,
154 		const void *caller)
155 {
156 	return __vmalloc(size, gfp_mask);
157 }
158 
159 void *__vmalloc_node(unsigned long size, unsigned long align, gfp_t gfp_mask,
160 		int node, const void *caller)
161 {
162 	return __vmalloc(size, gfp_mask);
163 }
164 
165 static void *__vmalloc_user_flags(unsigned long size, gfp_t flags)
166 {
167 	void *ret;
168 
169 	ret = __vmalloc(size, flags);
170 	if (ret) {
171 		struct vm_area_struct *vma;
172 
173 		mmap_write_lock(current->mm);
174 		vma = find_vma(current->mm, (unsigned long)ret);
175 		if (vma)
176 			vma->vm_flags |= VM_USERMAP;
177 		mmap_write_unlock(current->mm);
178 	}
179 
180 	return ret;
181 }
182 
183 void *vmalloc_user(unsigned long size)
184 {
185 	return __vmalloc_user_flags(size, GFP_KERNEL | __GFP_ZERO);
186 }
187 EXPORT_SYMBOL(vmalloc_user);
188 
189 struct page *vmalloc_to_page(const void *addr)
190 {
191 	return virt_to_page(addr);
192 }
193 EXPORT_SYMBOL(vmalloc_to_page);
194 
195 unsigned long vmalloc_to_pfn(const void *addr)
196 {
197 	return page_to_pfn(virt_to_page(addr));
198 }
199 EXPORT_SYMBOL(vmalloc_to_pfn);
200 
201 long vread(char *buf, char *addr, unsigned long count)
202 {
203 	/* Don't allow overflow */
204 	if ((unsigned long) buf + count < count)
205 		count = -(unsigned long) buf;
206 
207 	memcpy(buf, addr, count);
208 	return count;
209 }
210 
211 /*
212  *	vmalloc  -  allocate virtually contiguous memory
213  *
214  *	@size:		allocation size
215  *
216  *	Allocate enough pages to cover @size from the page level
217  *	allocator and map them into contiguous kernel virtual space.
218  *
219  *	For tight control over page level allocator and protection flags
220  *	use __vmalloc() instead.
221  */
222 void *vmalloc(unsigned long size)
223 {
224 	return __vmalloc(size, GFP_KERNEL);
225 }
226 EXPORT_SYMBOL(vmalloc);
227 
228 void *vmalloc_huge(unsigned long size, gfp_t gfp_mask) __weak __alias(__vmalloc);
229 
230 /*
231  *	vzalloc - allocate virtually contiguous memory with zero fill
232  *
233  *	@size:		allocation size
234  *
235  *	Allocate enough pages to cover @size from the page level
236  *	allocator and map them into contiguous kernel virtual space.
237  *	The memory allocated is set to zero.
238  *
239  *	For tight control over page level allocator and protection flags
240  *	use __vmalloc() instead.
241  */
242 void *vzalloc(unsigned long size)
243 {
244 	return __vmalloc(size, GFP_KERNEL | __GFP_ZERO);
245 }
246 EXPORT_SYMBOL(vzalloc);
247 
248 /**
249  * vmalloc_node - allocate memory on a specific node
250  * @size:	allocation size
251  * @node:	numa node
252  *
253  * Allocate enough pages to cover @size from the page level
254  * allocator and map them into contiguous kernel virtual space.
255  *
256  * For tight control over page level allocator and protection flags
257  * use __vmalloc() instead.
258  */
259 void *vmalloc_node(unsigned long size, int node)
260 {
261 	return vmalloc(size);
262 }
263 EXPORT_SYMBOL(vmalloc_node);
264 
265 /**
266  * vzalloc_node - allocate memory on a specific node with zero fill
267  * @size:	allocation size
268  * @node:	numa node
269  *
270  * Allocate enough pages to cover @size from the page level
271  * allocator and map them into contiguous kernel virtual space.
272  * The memory allocated is set to zero.
273  *
274  * For tight control over page level allocator and protection flags
275  * use __vmalloc() instead.
276  */
277 void *vzalloc_node(unsigned long size, int node)
278 {
279 	return vzalloc(size);
280 }
281 EXPORT_SYMBOL(vzalloc_node);
282 
283 /**
284  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
285  *	@size:		allocation size
286  *
287  *	Allocate enough 32bit PA addressable pages to cover @size from the
288  *	page level allocator and map them into contiguous kernel virtual space.
289  */
290 void *vmalloc_32(unsigned long size)
291 {
292 	return __vmalloc(size, GFP_KERNEL);
293 }
294 EXPORT_SYMBOL(vmalloc_32);
295 
296 /**
297  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
298  *	@size:		allocation size
299  *
300  * The resulting memory area is 32bit addressable and zeroed so it can be
301  * mapped to userspace without leaking data.
302  *
303  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
304  * remap_vmalloc_range() are permissible.
305  */
306 void *vmalloc_32_user(unsigned long size)
307 {
308 	/*
309 	 * We'll have to sort out the ZONE_DMA bits for 64-bit,
310 	 * but for now this can simply use vmalloc_user() directly.
311 	 */
312 	return vmalloc_user(size);
313 }
314 EXPORT_SYMBOL(vmalloc_32_user);
315 
316 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
317 {
318 	BUG();
319 	return NULL;
320 }
321 EXPORT_SYMBOL(vmap);
322 
323 void vunmap(const void *addr)
324 {
325 	BUG();
326 }
327 EXPORT_SYMBOL(vunmap);
328 
329 void *vm_map_ram(struct page **pages, unsigned int count, int node)
330 {
331 	BUG();
332 	return NULL;
333 }
334 EXPORT_SYMBOL(vm_map_ram);
335 
336 void vm_unmap_ram(const void *mem, unsigned int count)
337 {
338 	BUG();
339 }
340 EXPORT_SYMBOL(vm_unmap_ram);
341 
342 void vm_unmap_aliases(void)
343 {
344 }
345 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
346 
347 void free_vm_area(struct vm_struct *area)
348 {
349 	BUG();
350 }
351 EXPORT_SYMBOL_GPL(free_vm_area);
352 
353 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
354 		   struct page *page)
355 {
356 	return -EINVAL;
357 }
358 EXPORT_SYMBOL(vm_insert_page);
359 
360 int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
361 			unsigned long num)
362 {
363 	return -EINVAL;
364 }
365 EXPORT_SYMBOL(vm_map_pages);
366 
367 int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
368 				unsigned long num)
369 {
370 	return -EINVAL;
371 }
372 EXPORT_SYMBOL(vm_map_pages_zero);
373 
374 /*
375  *  sys_brk() for the most part doesn't need the global kernel
376  *  lock, except when an application is doing something nasty
377  *  like trying to un-brk an area that has already been mapped
378  *  to a regular file.  in this case, the unmapping will need
379  *  to invoke file system routines that need the global lock.
380  */
381 SYSCALL_DEFINE1(brk, unsigned long, brk)
382 {
383 	struct mm_struct *mm = current->mm;
384 
385 	if (brk < mm->start_brk || brk > mm->context.end_brk)
386 		return mm->brk;
387 
388 	if (mm->brk == brk)
389 		return mm->brk;
390 
391 	/*
392 	 * Always allow shrinking brk
393 	 */
394 	if (brk <= mm->brk) {
395 		mm->brk = brk;
396 		return brk;
397 	}
398 
399 	/*
400 	 * Ok, looks good - let it rip.
401 	 */
402 	flush_icache_user_range(mm->brk, brk);
403 	return mm->brk = brk;
404 }
405 
406 /*
407  * initialise the percpu counter for VM and region record slabs
408  */
409 void __init mmap_init(void)
410 {
411 	int ret;
412 
413 	ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
414 	VM_BUG_ON(ret);
415 	vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
416 }
417 
418 /*
419  * validate the region tree
420  * - the caller must hold the region lock
421  */
422 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
423 static noinline void validate_nommu_regions(void)
424 {
425 	struct vm_region *region, *last;
426 	struct rb_node *p, *lastp;
427 
428 	lastp = rb_first(&nommu_region_tree);
429 	if (!lastp)
430 		return;
431 
432 	last = rb_entry(lastp, struct vm_region, vm_rb);
433 	BUG_ON(last->vm_end <= last->vm_start);
434 	BUG_ON(last->vm_top < last->vm_end);
435 
436 	while ((p = rb_next(lastp))) {
437 		region = rb_entry(p, struct vm_region, vm_rb);
438 		last = rb_entry(lastp, struct vm_region, vm_rb);
439 
440 		BUG_ON(region->vm_end <= region->vm_start);
441 		BUG_ON(region->vm_top < region->vm_end);
442 		BUG_ON(region->vm_start < last->vm_top);
443 
444 		lastp = p;
445 	}
446 }
447 #else
448 static void validate_nommu_regions(void)
449 {
450 }
451 #endif
452 
453 /*
454  * add a region into the global tree
455  */
456 static void add_nommu_region(struct vm_region *region)
457 {
458 	struct vm_region *pregion;
459 	struct rb_node **p, *parent;
460 
461 	validate_nommu_regions();
462 
463 	parent = NULL;
464 	p = &nommu_region_tree.rb_node;
465 	while (*p) {
466 		parent = *p;
467 		pregion = rb_entry(parent, struct vm_region, vm_rb);
468 		if (region->vm_start < pregion->vm_start)
469 			p = &(*p)->rb_left;
470 		else if (region->vm_start > pregion->vm_start)
471 			p = &(*p)->rb_right;
472 		else if (pregion == region)
473 			return;
474 		else
475 			BUG();
476 	}
477 
478 	rb_link_node(&region->vm_rb, parent, p);
479 	rb_insert_color(&region->vm_rb, &nommu_region_tree);
480 
481 	validate_nommu_regions();
482 }
483 
484 /*
485  * delete a region from the global tree
486  */
487 static void delete_nommu_region(struct vm_region *region)
488 {
489 	BUG_ON(!nommu_region_tree.rb_node);
490 
491 	validate_nommu_regions();
492 	rb_erase(&region->vm_rb, &nommu_region_tree);
493 	validate_nommu_regions();
494 }
495 
496 /*
497  * free a contiguous series of pages
498  */
499 static void free_page_series(unsigned long from, unsigned long to)
500 {
501 	for (; from < to; from += PAGE_SIZE) {
502 		struct page *page = virt_to_page((void *)from);
503 
504 		atomic_long_dec(&mmap_pages_allocated);
505 		put_page(page);
506 	}
507 }
508 
509 /*
510  * release a reference to a region
511  * - the caller must hold the region semaphore for writing, which this releases
512  * - the region may not have been added to the tree yet, in which case vm_top
513  *   will equal vm_start
514  */
515 static void __put_nommu_region(struct vm_region *region)
516 	__releases(nommu_region_sem)
517 {
518 	BUG_ON(!nommu_region_tree.rb_node);
519 
520 	if (--region->vm_usage == 0) {
521 		if (region->vm_top > region->vm_start)
522 			delete_nommu_region(region);
523 		up_write(&nommu_region_sem);
524 
525 		if (region->vm_file)
526 			fput(region->vm_file);
527 
528 		/* IO memory and memory shared directly out of the pagecache
529 		 * from ramfs/tmpfs mustn't be released here */
530 		if (region->vm_flags & VM_MAPPED_COPY)
531 			free_page_series(region->vm_start, region->vm_top);
532 		kmem_cache_free(vm_region_jar, region);
533 	} else {
534 		up_write(&nommu_region_sem);
535 	}
536 }
537 
538 /*
539  * release a reference to a region
540  */
541 static void put_nommu_region(struct vm_region *region)
542 {
543 	down_write(&nommu_region_sem);
544 	__put_nommu_region(region);
545 }
546 
547 void vma_mas_store(struct vm_area_struct *vma, struct ma_state *mas)
548 {
549 	mas_set_range(mas, vma->vm_start, vma->vm_end - 1);
550 	mas_store_prealloc(mas, vma);
551 }
552 
553 void vma_mas_remove(struct vm_area_struct *vma, struct ma_state *mas)
554 {
555 	mas->index = vma->vm_start;
556 	mas->last = vma->vm_end - 1;
557 	mas_store_prealloc(mas, NULL);
558 }
559 
560 static void setup_vma_to_mm(struct vm_area_struct *vma, struct mm_struct *mm)
561 {
562 	vma->vm_mm = mm;
563 
564 	/* add the VMA to the mapping */
565 	if (vma->vm_file) {
566 		struct address_space *mapping = vma->vm_file->f_mapping;
567 
568 		i_mmap_lock_write(mapping);
569 		flush_dcache_mmap_lock(mapping);
570 		vma_interval_tree_insert(vma, &mapping->i_mmap);
571 		flush_dcache_mmap_unlock(mapping);
572 		i_mmap_unlock_write(mapping);
573 	}
574 }
575 
576 /*
577  * mas_add_vma_to_mm() - Maple state variant of add_mas_to_mm().
578  * @mas: The maple state with preallocations.
579  * @mm: The mm_struct
580  * @vma: The vma to add
581  *
582  */
583 static void mas_add_vma_to_mm(struct ma_state *mas, struct mm_struct *mm,
584 			      struct vm_area_struct *vma)
585 {
586 	BUG_ON(!vma->vm_region);
587 
588 	setup_vma_to_mm(vma, mm);
589 	mm->map_count++;
590 
591 	/* add the VMA to the tree */
592 	vma_mas_store(vma, mas);
593 }
594 
595 /*
596  * add a VMA into a process's mm_struct in the appropriate place in the list
597  * and tree and add to the address space's page tree also if not an anonymous
598  * page
599  * - should be called with mm->mmap_lock held writelocked
600  */
601 static int add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
602 {
603 	MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_end);
604 
605 	if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
606 		pr_warn("Allocation of vma tree for process %d failed\n",
607 		       current->pid);
608 		return -ENOMEM;
609 	}
610 	mas_add_vma_to_mm(&mas, mm, vma);
611 	return 0;
612 }
613 
614 static void cleanup_vma_from_mm(struct vm_area_struct *vma)
615 {
616 	vma->vm_mm->map_count--;
617 	/* remove the VMA from the mapping */
618 	if (vma->vm_file) {
619 		struct address_space *mapping;
620 		mapping = vma->vm_file->f_mapping;
621 
622 		i_mmap_lock_write(mapping);
623 		flush_dcache_mmap_lock(mapping);
624 		vma_interval_tree_remove(vma, &mapping->i_mmap);
625 		flush_dcache_mmap_unlock(mapping);
626 		i_mmap_unlock_write(mapping);
627 	}
628 }
629 /*
630  * delete a VMA from its owning mm_struct and address space
631  */
632 static int delete_vma_from_mm(struct vm_area_struct *vma)
633 {
634 	MA_STATE(mas, &vma->vm_mm->mm_mt, 0, 0);
635 
636 	if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
637 		pr_warn("Allocation of vma tree for process %d failed\n",
638 		       current->pid);
639 		return -ENOMEM;
640 	}
641 	cleanup_vma_from_mm(vma);
642 
643 	/* remove from the MM's tree and list */
644 	vma_mas_remove(vma, &mas);
645 	return 0;
646 }
647 
648 /*
649  * destroy a VMA record
650  */
651 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
652 {
653 	if (vma->vm_ops && vma->vm_ops->close)
654 		vma->vm_ops->close(vma);
655 	if (vma->vm_file)
656 		fput(vma->vm_file);
657 	put_nommu_region(vma->vm_region);
658 	vm_area_free(vma);
659 }
660 
661 struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
662 					     unsigned long start_addr,
663 					     unsigned long end_addr)
664 {
665 	unsigned long index = start_addr;
666 
667 	mmap_assert_locked(mm);
668 	return mt_find(&mm->mm_mt, &index, end_addr - 1);
669 }
670 EXPORT_SYMBOL(find_vma_intersection);
671 
672 /*
673  * look up the first VMA in which addr resides, NULL if none
674  * - should be called with mm->mmap_lock at least held readlocked
675  */
676 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
677 {
678 	MA_STATE(mas, &mm->mm_mt, addr, addr);
679 
680 	return mas_walk(&mas);
681 }
682 EXPORT_SYMBOL(find_vma);
683 
684 /*
685  * find a VMA
686  * - we don't extend stack VMAs under NOMMU conditions
687  */
688 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
689 {
690 	return find_vma(mm, addr);
691 }
692 
693 /*
694  * expand a stack to a given address
695  * - not supported under NOMMU conditions
696  */
697 int expand_stack(struct vm_area_struct *vma, unsigned long address)
698 {
699 	return -ENOMEM;
700 }
701 
702 /*
703  * look up the first VMA exactly that exactly matches addr
704  * - should be called with mm->mmap_lock at least held readlocked
705  */
706 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
707 					     unsigned long addr,
708 					     unsigned long len)
709 {
710 	struct vm_area_struct *vma;
711 	unsigned long end = addr + len;
712 	MA_STATE(mas, &mm->mm_mt, addr, addr);
713 
714 	vma = mas_walk(&mas);
715 	if (!vma)
716 		return NULL;
717 	if (vma->vm_start != addr)
718 		return NULL;
719 	if (vma->vm_end != end)
720 		return NULL;
721 
722 	return vma;
723 }
724 
725 /*
726  * determine whether a mapping should be permitted and, if so, what sort of
727  * mapping we're capable of supporting
728  */
729 static int validate_mmap_request(struct file *file,
730 				 unsigned long addr,
731 				 unsigned long len,
732 				 unsigned long prot,
733 				 unsigned long flags,
734 				 unsigned long pgoff,
735 				 unsigned long *_capabilities)
736 {
737 	unsigned long capabilities, rlen;
738 	int ret;
739 
740 	/* do the simple checks first */
741 	if (flags & MAP_FIXED)
742 		return -EINVAL;
743 
744 	if ((flags & MAP_TYPE) != MAP_PRIVATE &&
745 	    (flags & MAP_TYPE) != MAP_SHARED)
746 		return -EINVAL;
747 
748 	if (!len)
749 		return -EINVAL;
750 
751 	/* Careful about overflows.. */
752 	rlen = PAGE_ALIGN(len);
753 	if (!rlen || rlen > TASK_SIZE)
754 		return -ENOMEM;
755 
756 	/* offset overflow? */
757 	if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
758 		return -EOVERFLOW;
759 
760 	if (file) {
761 		/* files must support mmap */
762 		if (!file->f_op->mmap)
763 			return -ENODEV;
764 
765 		/* work out if what we've got could possibly be shared
766 		 * - we support chardevs that provide their own "memory"
767 		 * - we support files/blockdevs that are memory backed
768 		 */
769 		if (file->f_op->mmap_capabilities) {
770 			capabilities = file->f_op->mmap_capabilities(file);
771 		} else {
772 			/* no explicit capabilities set, so assume some
773 			 * defaults */
774 			switch (file_inode(file)->i_mode & S_IFMT) {
775 			case S_IFREG:
776 			case S_IFBLK:
777 				capabilities = NOMMU_MAP_COPY;
778 				break;
779 
780 			case S_IFCHR:
781 				capabilities =
782 					NOMMU_MAP_DIRECT |
783 					NOMMU_MAP_READ |
784 					NOMMU_MAP_WRITE;
785 				break;
786 
787 			default:
788 				return -EINVAL;
789 			}
790 		}
791 
792 		/* eliminate any capabilities that we can't support on this
793 		 * device */
794 		if (!file->f_op->get_unmapped_area)
795 			capabilities &= ~NOMMU_MAP_DIRECT;
796 		if (!(file->f_mode & FMODE_CAN_READ))
797 			capabilities &= ~NOMMU_MAP_COPY;
798 
799 		/* The file shall have been opened with read permission. */
800 		if (!(file->f_mode & FMODE_READ))
801 			return -EACCES;
802 
803 		if (flags & MAP_SHARED) {
804 			/* do checks for writing, appending and locking */
805 			if ((prot & PROT_WRITE) &&
806 			    !(file->f_mode & FMODE_WRITE))
807 				return -EACCES;
808 
809 			if (IS_APPEND(file_inode(file)) &&
810 			    (file->f_mode & FMODE_WRITE))
811 				return -EACCES;
812 
813 			if (!(capabilities & NOMMU_MAP_DIRECT))
814 				return -ENODEV;
815 
816 			/* we mustn't privatise shared mappings */
817 			capabilities &= ~NOMMU_MAP_COPY;
818 		} else {
819 			/* we're going to read the file into private memory we
820 			 * allocate */
821 			if (!(capabilities & NOMMU_MAP_COPY))
822 				return -ENODEV;
823 
824 			/* we don't permit a private writable mapping to be
825 			 * shared with the backing device */
826 			if (prot & PROT_WRITE)
827 				capabilities &= ~NOMMU_MAP_DIRECT;
828 		}
829 
830 		if (capabilities & NOMMU_MAP_DIRECT) {
831 			if (((prot & PROT_READ)  && !(capabilities & NOMMU_MAP_READ))  ||
832 			    ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
833 			    ((prot & PROT_EXEC)  && !(capabilities & NOMMU_MAP_EXEC))
834 			    ) {
835 				capabilities &= ~NOMMU_MAP_DIRECT;
836 				if (flags & MAP_SHARED) {
837 					pr_warn("MAP_SHARED not completely supported on !MMU\n");
838 					return -EINVAL;
839 				}
840 			}
841 		}
842 
843 		/* handle executable mappings and implied executable
844 		 * mappings */
845 		if (path_noexec(&file->f_path)) {
846 			if (prot & PROT_EXEC)
847 				return -EPERM;
848 		} else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
849 			/* handle implication of PROT_EXEC by PROT_READ */
850 			if (current->personality & READ_IMPLIES_EXEC) {
851 				if (capabilities & NOMMU_MAP_EXEC)
852 					prot |= PROT_EXEC;
853 			}
854 		} else if ((prot & PROT_READ) &&
855 			 (prot & PROT_EXEC) &&
856 			 !(capabilities & NOMMU_MAP_EXEC)
857 			 ) {
858 			/* backing file is not executable, try to copy */
859 			capabilities &= ~NOMMU_MAP_DIRECT;
860 		}
861 	} else {
862 		/* anonymous mappings are always memory backed and can be
863 		 * privately mapped
864 		 */
865 		capabilities = NOMMU_MAP_COPY;
866 
867 		/* handle PROT_EXEC implication by PROT_READ */
868 		if ((prot & PROT_READ) &&
869 		    (current->personality & READ_IMPLIES_EXEC))
870 			prot |= PROT_EXEC;
871 	}
872 
873 	/* allow the security API to have its say */
874 	ret = security_mmap_addr(addr);
875 	if (ret < 0)
876 		return ret;
877 
878 	/* looks okay */
879 	*_capabilities = capabilities;
880 	return 0;
881 }
882 
883 /*
884  * we've determined that we can make the mapping, now translate what we
885  * now know into VMA flags
886  */
887 static unsigned long determine_vm_flags(struct file *file,
888 					unsigned long prot,
889 					unsigned long flags,
890 					unsigned long capabilities)
891 {
892 	unsigned long vm_flags;
893 
894 	vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
895 	/* vm_flags |= mm->def_flags; */
896 
897 	if (!(capabilities & NOMMU_MAP_DIRECT)) {
898 		/* attempt to share read-only copies of mapped file chunks */
899 		vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
900 		if (file && !(prot & PROT_WRITE))
901 			vm_flags |= VM_MAYSHARE;
902 	} else {
903 		/* overlay a shareable mapping on the backing device or inode
904 		 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
905 		 * romfs/cramfs */
906 		vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
907 		if (flags & MAP_SHARED)
908 			vm_flags |= VM_SHARED;
909 	}
910 
911 	/* refuse to let anyone share private mappings with this process if
912 	 * it's being traced - otherwise breakpoints set in it may interfere
913 	 * with another untraced process
914 	 */
915 	if ((flags & MAP_PRIVATE) && current->ptrace)
916 		vm_flags &= ~VM_MAYSHARE;
917 
918 	return vm_flags;
919 }
920 
921 /*
922  * set up a shared mapping on a file (the driver or filesystem provides and
923  * pins the storage)
924  */
925 static int do_mmap_shared_file(struct vm_area_struct *vma)
926 {
927 	int ret;
928 
929 	ret = call_mmap(vma->vm_file, vma);
930 	if (ret == 0) {
931 		vma->vm_region->vm_top = vma->vm_region->vm_end;
932 		return 0;
933 	}
934 	if (ret != -ENOSYS)
935 		return ret;
936 
937 	/* getting -ENOSYS indicates that direct mmap isn't possible (as
938 	 * opposed to tried but failed) so we can only give a suitable error as
939 	 * it's not possible to make a private copy if MAP_SHARED was given */
940 	return -ENODEV;
941 }
942 
943 /*
944  * set up a private mapping or an anonymous shared mapping
945  */
946 static int do_mmap_private(struct vm_area_struct *vma,
947 			   struct vm_region *region,
948 			   unsigned long len,
949 			   unsigned long capabilities)
950 {
951 	unsigned long total, point;
952 	void *base;
953 	int ret, order;
954 
955 	/* invoke the file's mapping function so that it can keep track of
956 	 * shared mappings on devices or memory
957 	 * - VM_MAYSHARE will be set if it may attempt to share
958 	 */
959 	if (capabilities & NOMMU_MAP_DIRECT) {
960 		ret = call_mmap(vma->vm_file, vma);
961 		if (ret == 0) {
962 			/* shouldn't return success if we're not sharing */
963 			BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
964 			vma->vm_region->vm_top = vma->vm_region->vm_end;
965 			return 0;
966 		}
967 		if (ret != -ENOSYS)
968 			return ret;
969 
970 		/* getting an ENOSYS error indicates that direct mmap isn't
971 		 * possible (as opposed to tried but failed) so we'll try to
972 		 * make a private copy of the data and map that instead */
973 	}
974 
975 
976 	/* allocate some memory to hold the mapping
977 	 * - note that this may not return a page-aligned address if the object
978 	 *   we're allocating is smaller than a page
979 	 */
980 	order = get_order(len);
981 	total = 1 << order;
982 	point = len >> PAGE_SHIFT;
983 
984 	/* we don't want to allocate a power-of-2 sized page set */
985 	if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
986 		total = point;
987 
988 	base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
989 	if (!base)
990 		goto enomem;
991 
992 	atomic_long_add(total, &mmap_pages_allocated);
993 
994 	region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
995 	region->vm_start = (unsigned long) base;
996 	region->vm_end   = region->vm_start + len;
997 	region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
998 
999 	vma->vm_start = region->vm_start;
1000 	vma->vm_end   = region->vm_start + len;
1001 
1002 	if (vma->vm_file) {
1003 		/* read the contents of a file into the copy */
1004 		loff_t fpos;
1005 
1006 		fpos = vma->vm_pgoff;
1007 		fpos <<= PAGE_SHIFT;
1008 
1009 		ret = kernel_read(vma->vm_file, base, len, &fpos);
1010 		if (ret < 0)
1011 			goto error_free;
1012 
1013 		/* clear the last little bit */
1014 		if (ret < len)
1015 			memset(base + ret, 0, len - ret);
1016 
1017 	} else {
1018 		vma_set_anonymous(vma);
1019 	}
1020 
1021 	return 0;
1022 
1023 error_free:
1024 	free_page_series(region->vm_start, region->vm_top);
1025 	region->vm_start = vma->vm_start = 0;
1026 	region->vm_end   = vma->vm_end = 0;
1027 	region->vm_top   = 0;
1028 	return ret;
1029 
1030 enomem:
1031 	pr_err("Allocation of length %lu from process %d (%s) failed\n",
1032 	       len, current->pid, current->comm);
1033 	show_free_areas(0, NULL);
1034 	return -ENOMEM;
1035 }
1036 
1037 /*
1038  * handle mapping creation for uClinux
1039  */
1040 unsigned long do_mmap(struct file *file,
1041 			unsigned long addr,
1042 			unsigned long len,
1043 			unsigned long prot,
1044 			unsigned long flags,
1045 			unsigned long pgoff,
1046 			unsigned long *populate,
1047 			struct list_head *uf)
1048 {
1049 	struct vm_area_struct *vma;
1050 	struct vm_region *region;
1051 	struct rb_node *rb;
1052 	vm_flags_t vm_flags;
1053 	unsigned long capabilities, result;
1054 	int ret;
1055 	MA_STATE(mas, &current->mm->mm_mt, 0, 0);
1056 
1057 	*populate = 0;
1058 
1059 	/* decide whether we should attempt the mapping, and if so what sort of
1060 	 * mapping */
1061 	ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1062 				    &capabilities);
1063 	if (ret < 0)
1064 		return ret;
1065 
1066 	/* we ignore the address hint */
1067 	addr = 0;
1068 	len = PAGE_ALIGN(len);
1069 
1070 	/* we've determined that we can make the mapping, now translate what we
1071 	 * now know into VMA flags */
1072 	vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1073 
1074 
1075 	/* we're going to need to record the mapping */
1076 	region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1077 	if (!region)
1078 		goto error_getting_region;
1079 
1080 	vma = vm_area_alloc(current->mm);
1081 	if (!vma)
1082 		goto error_getting_vma;
1083 
1084 	if (mas_preallocate(&mas, vma, GFP_KERNEL))
1085 		goto error_maple_preallocate;
1086 
1087 	region->vm_usage = 1;
1088 	region->vm_flags = vm_flags;
1089 	region->vm_pgoff = pgoff;
1090 
1091 	vma->vm_flags = vm_flags;
1092 	vma->vm_pgoff = pgoff;
1093 
1094 	if (file) {
1095 		region->vm_file = get_file(file);
1096 		vma->vm_file = get_file(file);
1097 	}
1098 
1099 	down_write(&nommu_region_sem);
1100 
1101 	/* if we want to share, we need to check for regions created by other
1102 	 * mmap() calls that overlap with our proposed mapping
1103 	 * - we can only share with a superset match on most regular files
1104 	 * - shared mappings on character devices and memory backed files are
1105 	 *   permitted to overlap inexactly as far as we are concerned for in
1106 	 *   these cases, sharing is handled in the driver or filesystem rather
1107 	 *   than here
1108 	 */
1109 	if (vm_flags & VM_MAYSHARE) {
1110 		struct vm_region *pregion;
1111 		unsigned long pglen, rpglen, pgend, rpgend, start;
1112 
1113 		pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1114 		pgend = pgoff + pglen;
1115 
1116 		for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1117 			pregion = rb_entry(rb, struct vm_region, vm_rb);
1118 
1119 			if (!(pregion->vm_flags & VM_MAYSHARE))
1120 				continue;
1121 
1122 			/* search for overlapping mappings on the same file */
1123 			if (file_inode(pregion->vm_file) !=
1124 			    file_inode(file))
1125 				continue;
1126 
1127 			if (pregion->vm_pgoff >= pgend)
1128 				continue;
1129 
1130 			rpglen = pregion->vm_end - pregion->vm_start;
1131 			rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1132 			rpgend = pregion->vm_pgoff + rpglen;
1133 			if (pgoff >= rpgend)
1134 				continue;
1135 
1136 			/* handle inexactly overlapping matches between
1137 			 * mappings */
1138 			if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1139 			    !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1140 				/* new mapping is not a subset of the region */
1141 				if (!(capabilities & NOMMU_MAP_DIRECT))
1142 					goto sharing_violation;
1143 				continue;
1144 			}
1145 
1146 			/* we've found a region we can share */
1147 			pregion->vm_usage++;
1148 			vma->vm_region = pregion;
1149 			start = pregion->vm_start;
1150 			start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1151 			vma->vm_start = start;
1152 			vma->vm_end = start + len;
1153 
1154 			if (pregion->vm_flags & VM_MAPPED_COPY)
1155 				vma->vm_flags |= VM_MAPPED_COPY;
1156 			else {
1157 				ret = do_mmap_shared_file(vma);
1158 				if (ret < 0) {
1159 					vma->vm_region = NULL;
1160 					vma->vm_start = 0;
1161 					vma->vm_end = 0;
1162 					pregion->vm_usage--;
1163 					pregion = NULL;
1164 					goto error_just_free;
1165 				}
1166 			}
1167 			fput(region->vm_file);
1168 			kmem_cache_free(vm_region_jar, region);
1169 			region = pregion;
1170 			result = start;
1171 			goto share;
1172 		}
1173 
1174 		/* obtain the address at which to make a shared mapping
1175 		 * - this is the hook for quasi-memory character devices to
1176 		 *   tell us the location of a shared mapping
1177 		 */
1178 		if (capabilities & NOMMU_MAP_DIRECT) {
1179 			addr = file->f_op->get_unmapped_area(file, addr, len,
1180 							     pgoff, flags);
1181 			if (IS_ERR_VALUE(addr)) {
1182 				ret = addr;
1183 				if (ret != -ENOSYS)
1184 					goto error_just_free;
1185 
1186 				/* the driver refused to tell us where to site
1187 				 * the mapping so we'll have to attempt to copy
1188 				 * it */
1189 				ret = -ENODEV;
1190 				if (!(capabilities & NOMMU_MAP_COPY))
1191 					goto error_just_free;
1192 
1193 				capabilities &= ~NOMMU_MAP_DIRECT;
1194 			} else {
1195 				vma->vm_start = region->vm_start = addr;
1196 				vma->vm_end = region->vm_end = addr + len;
1197 			}
1198 		}
1199 	}
1200 
1201 	vma->vm_region = region;
1202 
1203 	/* set up the mapping
1204 	 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1205 	 */
1206 	if (file && vma->vm_flags & VM_SHARED)
1207 		ret = do_mmap_shared_file(vma);
1208 	else
1209 		ret = do_mmap_private(vma, region, len, capabilities);
1210 	if (ret < 0)
1211 		goto error_just_free;
1212 	add_nommu_region(region);
1213 
1214 	/* clear anonymous mappings that don't ask for uninitialized data */
1215 	if (!vma->vm_file &&
1216 	    (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) ||
1217 	     !(flags & MAP_UNINITIALIZED)))
1218 		memset((void *)region->vm_start, 0,
1219 		       region->vm_end - region->vm_start);
1220 
1221 	/* okay... we have a mapping; now we have to register it */
1222 	result = vma->vm_start;
1223 
1224 	current->mm->total_vm += len >> PAGE_SHIFT;
1225 
1226 share:
1227 	mas_add_vma_to_mm(&mas, current->mm, vma);
1228 
1229 	/* we flush the region from the icache only when the first executable
1230 	 * mapping of it is made  */
1231 	if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1232 		flush_icache_user_range(region->vm_start, region->vm_end);
1233 		region->vm_icache_flushed = true;
1234 	}
1235 
1236 	up_write(&nommu_region_sem);
1237 
1238 	return result;
1239 
1240 error_just_free:
1241 	up_write(&nommu_region_sem);
1242 error:
1243 	mas_destroy(&mas);
1244 	if (region->vm_file)
1245 		fput(region->vm_file);
1246 	kmem_cache_free(vm_region_jar, region);
1247 	if (vma->vm_file)
1248 		fput(vma->vm_file);
1249 	vm_area_free(vma);
1250 	return ret;
1251 
1252 sharing_violation:
1253 	up_write(&nommu_region_sem);
1254 	pr_warn("Attempt to share mismatched mappings\n");
1255 	ret = -EINVAL;
1256 	goto error;
1257 
1258 error_getting_vma:
1259 	kmem_cache_free(vm_region_jar, region);
1260 	pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1261 			len, current->pid);
1262 	show_free_areas(0, NULL);
1263 	return -ENOMEM;
1264 
1265 error_getting_region:
1266 	pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1267 			len, current->pid);
1268 	show_free_areas(0, NULL);
1269 	return -ENOMEM;
1270 
1271 error_maple_preallocate:
1272 	kmem_cache_free(vm_region_jar, region);
1273 	vm_area_free(vma);
1274 	pr_warn("Allocation of vma tree for process %d failed\n", current->pid);
1275 	show_free_areas(0, NULL);
1276 	return -ENOMEM;
1277 
1278 }
1279 
1280 unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1281 			      unsigned long prot, unsigned long flags,
1282 			      unsigned long fd, unsigned long pgoff)
1283 {
1284 	struct file *file = NULL;
1285 	unsigned long retval = -EBADF;
1286 
1287 	audit_mmap_fd(fd, flags);
1288 	if (!(flags & MAP_ANONYMOUS)) {
1289 		file = fget(fd);
1290 		if (!file)
1291 			goto out;
1292 	}
1293 
1294 	retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1295 
1296 	if (file)
1297 		fput(file);
1298 out:
1299 	return retval;
1300 }
1301 
1302 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1303 		unsigned long, prot, unsigned long, flags,
1304 		unsigned long, fd, unsigned long, pgoff)
1305 {
1306 	return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1307 }
1308 
1309 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1310 struct mmap_arg_struct {
1311 	unsigned long addr;
1312 	unsigned long len;
1313 	unsigned long prot;
1314 	unsigned long flags;
1315 	unsigned long fd;
1316 	unsigned long offset;
1317 };
1318 
1319 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1320 {
1321 	struct mmap_arg_struct a;
1322 
1323 	if (copy_from_user(&a, arg, sizeof(a)))
1324 		return -EFAULT;
1325 	if (offset_in_page(a.offset))
1326 		return -EINVAL;
1327 
1328 	return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1329 			       a.offset >> PAGE_SHIFT);
1330 }
1331 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1332 
1333 /*
1334  * split a vma into two pieces at address 'addr', a new vma is allocated either
1335  * for the first part or the tail.
1336  */
1337 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1338 	      unsigned long addr, int new_below)
1339 {
1340 	struct vm_area_struct *new;
1341 	struct vm_region *region;
1342 	unsigned long npages;
1343 	MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_end);
1344 
1345 	/* we're only permitted to split anonymous regions (these should have
1346 	 * only a single usage on the region) */
1347 	if (vma->vm_file)
1348 		return -ENOMEM;
1349 
1350 	mm = vma->vm_mm;
1351 	if (mm->map_count >= sysctl_max_map_count)
1352 		return -ENOMEM;
1353 
1354 	region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1355 	if (!region)
1356 		return -ENOMEM;
1357 
1358 	new = vm_area_dup(vma);
1359 	if (!new)
1360 		goto err_vma_dup;
1361 
1362 	if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
1363 		pr_warn("Allocation of vma tree for process %d failed\n",
1364 			current->pid);
1365 		goto err_mas_preallocate;
1366 	}
1367 
1368 	/* most fields are the same, copy all, and then fixup */
1369 	*region = *vma->vm_region;
1370 	new->vm_region = region;
1371 
1372 	npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1373 
1374 	if (new_below) {
1375 		region->vm_top = region->vm_end = new->vm_end = addr;
1376 	} else {
1377 		region->vm_start = new->vm_start = addr;
1378 		region->vm_pgoff = new->vm_pgoff += npages;
1379 	}
1380 
1381 	if (new->vm_ops && new->vm_ops->open)
1382 		new->vm_ops->open(new);
1383 
1384 	down_write(&nommu_region_sem);
1385 	delete_nommu_region(vma->vm_region);
1386 	if (new_below) {
1387 		vma->vm_region->vm_start = vma->vm_start = addr;
1388 		vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1389 	} else {
1390 		vma->vm_region->vm_end = vma->vm_end = addr;
1391 		vma->vm_region->vm_top = addr;
1392 	}
1393 	add_nommu_region(vma->vm_region);
1394 	add_nommu_region(new->vm_region);
1395 	up_write(&nommu_region_sem);
1396 
1397 	setup_vma_to_mm(vma, mm);
1398 	setup_vma_to_mm(new, mm);
1399 	mas_set_range(&mas, vma->vm_start, vma->vm_end - 1);
1400 	mas_store(&mas, vma);
1401 	vma_mas_store(new, &mas);
1402 	mm->map_count++;
1403 	return 0;
1404 
1405 err_mas_preallocate:
1406 	vm_area_free(new);
1407 err_vma_dup:
1408 	kmem_cache_free(vm_region_jar, region);
1409 	return -ENOMEM;
1410 }
1411 
1412 /*
1413  * shrink a VMA by removing the specified chunk from either the beginning or
1414  * the end
1415  */
1416 static int shrink_vma(struct mm_struct *mm,
1417 		      struct vm_area_struct *vma,
1418 		      unsigned long from, unsigned long to)
1419 {
1420 	struct vm_region *region;
1421 
1422 	/* adjust the VMA's pointers, which may reposition it in the MM's tree
1423 	 * and list */
1424 	if (delete_vma_from_mm(vma))
1425 		return -ENOMEM;
1426 	if (from > vma->vm_start)
1427 		vma->vm_end = from;
1428 	else
1429 		vma->vm_start = to;
1430 	if (add_vma_to_mm(mm, vma))
1431 		return -ENOMEM;
1432 
1433 	/* cut the backing region down to size */
1434 	region = vma->vm_region;
1435 	BUG_ON(region->vm_usage != 1);
1436 
1437 	down_write(&nommu_region_sem);
1438 	delete_nommu_region(region);
1439 	if (from > region->vm_start) {
1440 		to = region->vm_top;
1441 		region->vm_top = region->vm_end = from;
1442 	} else {
1443 		region->vm_start = to;
1444 	}
1445 	add_nommu_region(region);
1446 	up_write(&nommu_region_sem);
1447 
1448 	free_page_series(from, to);
1449 	return 0;
1450 }
1451 
1452 /*
1453  * release a mapping
1454  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1455  *   VMA, though it need not cover the whole VMA
1456  */
1457 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
1458 {
1459 	MA_STATE(mas, &mm->mm_mt, start, start);
1460 	struct vm_area_struct *vma;
1461 	unsigned long end;
1462 	int ret = 0;
1463 
1464 	len = PAGE_ALIGN(len);
1465 	if (len == 0)
1466 		return -EINVAL;
1467 
1468 	end = start + len;
1469 
1470 	/* find the first potentially overlapping VMA */
1471 	vma = mas_find(&mas, end - 1);
1472 	if (!vma) {
1473 		static int limit;
1474 		if (limit < 5) {
1475 			pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1476 					current->pid, current->comm,
1477 					start, start + len - 1);
1478 			limit++;
1479 		}
1480 		return -EINVAL;
1481 	}
1482 
1483 	/* we're allowed to split an anonymous VMA but not a file-backed one */
1484 	if (vma->vm_file) {
1485 		do {
1486 			if (start > vma->vm_start)
1487 				return -EINVAL;
1488 			if (end == vma->vm_end)
1489 				goto erase_whole_vma;
1490 			vma = mas_next(&mas, end - 1);
1491 		} while (vma);
1492 		return -EINVAL;
1493 	} else {
1494 		/* the chunk must be a subset of the VMA found */
1495 		if (start == vma->vm_start && end == vma->vm_end)
1496 			goto erase_whole_vma;
1497 		if (start < vma->vm_start || end > vma->vm_end)
1498 			return -EINVAL;
1499 		if (offset_in_page(start))
1500 			return -EINVAL;
1501 		if (end != vma->vm_end && offset_in_page(end))
1502 			return -EINVAL;
1503 		if (start != vma->vm_start && end != vma->vm_end) {
1504 			ret = split_vma(mm, vma, start, 1);
1505 			if (ret < 0)
1506 				return ret;
1507 		}
1508 		return shrink_vma(mm, vma, start, end);
1509 	}
1510 
1511 erase_whole_vma:
1512 	if (delete_vma_from_mm(vma))
1513 		ret = -ENOMEM;
1514 	else
1515 		delete_vma(mm, vma);
1516 	return ret;
1517 }
1518 
1519 int vm_munmap(unsigned long addr, size_t len)
1520 {
1521 	struct mm_struct *mm = current->mm;
1522 	int ret;
1523 
1524 	mmap_write_lock(mm);
1525 	ret = do_munmap(mm, addr, len, NULL);
1526 	mmap_write_unlock(mm);
1527 	return ret;
1528 }
1529 EXPORT_SYMBOL(vm_munmap);
1530 
1531 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1532 {
1533 	return vm_munmap(addr, len);
1534 }
1535 
1536 /*
1537  * release all the mappings made in a process's VM space
1538  */
1539 void exit_mmap(struct mm_struct *mm)
1540 {
1541 	VMA_ITERATOR(vmi, mm, 0);
1542 	struct vm_area_struct *vma;
1543 
1544 	if (!mm)
1545 		return;
1546 
1547 	mm->total_vm = 0;
1548 
1549 	/*
1550 	 * Lock the mm to avoid assert complaining even though this is the only
1551 	 * user of the mm
1552 	 */
1553 	mmap_write_lock(mm);
1554 	for_each_vma(vmi, vma) {
1555 		cleanup_vma_from_mm(vma);
1556 		delete_vma(mm, vma);
1557 		cond_resched();
1558 	}
1559 	__mt_destroy(&mm->mm_mt);
1560 	mmap_write_unlock(mm);
1561 }
1562 
1563 int vm_brk(unsigned long addr, unsigned long len)
1564 {
1565 	return -ENOMEM;
1566 }
1567 
1568 /*
1569  * expand (or shrink) an existing mapping, potentially moving it at the same
1570  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1571  *
1572  * under NOMMU conditions, we only permit changing a mapping's size, and only
1573  * as long as it stays within the region allocated by do_mmap_private() and the
1574  * block is not shareable
1575  *
1576  * MREMAP_FIXED is not supported under NOMMU conditions
1577  */
1578 static unsigned long do_mremap(unsigned long addr,
1579 			unsigned long old_len, unsigned long new_len,
1580 			unsigned long flags, unsigned long new_addr)
1581 {
1582 	struct vm_area_struct *vma;
1583 
1584 	/* insanity checks first */
1585 	old_len = PAGE_ALIGN(old_len);
1586 	new_len = PAGE_ALIGN(new_len);
1587 	if (old_len == 0 || new_len == 0)
1588 		return (unsigned long) -EINVAL;
1589 
1590 	if (offset_in_page(addr))
1591 		return -EINVAL;
1592 
1593 	if (flags & MREMAP_FIXED && new_addr != addr)
1594 		return (unsigned long) -EINVAL;
1595 
1596 	vma = find_vma_exact(current->mm, addr, old_len);
1597 	if (!vma)
1598 		return (unsigned long) -EINVAL;
1599 
1600 	if (vma->vm_end != vma->vm_start + old_len)
1601 		return (unsigned long) -EFAULT;
1602 
1603 	if (vma->vm_flags & VM_MAYSHARE)
1604 		return (unsigned long) -EPERM;
1605 
1606 	if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1607 		return (unsigned long) -ENOMEM;
1608 
1609 	/* all checks complete - do it */
1610 	vma->vm_end = vma->vm_start + new_len;
1611 	return vma->vm_start;
1612 }
1613 
1614 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1615 		unsigned long, new_len, unsigned long, flags,
1616 		unsigned long, new_addr)
1617 {
1618 	unsigned long ret;
1619 
1620 	mmap_write_lock(current->mm);
1621 	ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1622 	mmap_write_unlock(current->mm);
1623 	return ret;
1624 }
1625 
1626 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1627 			 unsigned int foll_flags)
1628 {
1629 	return NULL;
1630 }
1631 
1632 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1633 		unsigned long pfn, unsigned long size, pgprot_t prot)
1634 {
1635 	if (addr != (pfn << PAGE_SHIFT))
1636 		return -EINVAL;
1637 
1638 	vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1639 	return 0;
1640 }
1641 EXPORT_SYMBOL(remap_pfn_range);
1642 
1643 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1644 {
1645 	unsigned long pfn = start >> PAGE_SHIFT;
1646 	unsigned long vm_len = vma->vm_end - vma->vm_start;
1647 
1648 	pfn += vma->vm_pgoff;
1649 	return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1650 }
1651 EXPORT_SYMBOL(vm_iomap_memory);
1652 
1653 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1654 			unsigned long pgoff)
1655 {
1656 	unsigned int size = vma->vm_end - vma->vm_start;
1657 
1658 	if (!(vma->vm_flags & VM_USERMAP))
1659 		return -EINVAL;
1660 
1661 	vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1662 	vma->vm_end = vma->vm_start + size;
1663 
1664 	return 0;
1665 }
1666 EXPORT_SYMBOL(remap_vmalloc_range);
1667 
1668 vm_fault_t filemap_fault(struct vm_fault *vmf)
1669 {
1670 	BUG();
1671 	return 0;
1672 }
1673 EXPORT_SYMBOL(filemap_fault);
1674 
1675 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
1676 		pgoff_t start_pgoff, pgoff_t end_pgoff)
1677 {
1678 	BUG();
1679 	return 0;
1680 }
1681 EXPORT_SYMBOL(filemap_map_pages);
1682 
1683 int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf,
1684 		       int len, unsigned int gup_flags)
1685 {
1686 	struct vm_area_struct *vma;
1687 	int write = gup_flags & FOLL_WRITE;
1688 
1689 	if (mmap_read_lock_killable(mm))
1690 		return 0;
1691 
1692 	/* the access must start within one of the target process's mappings */
1693 	vma = find_vma(mm, addr);
1694 	if (vma) {
1695 		/* don't overrun this mapping */
1696 		if (addr + len >= vma->vm_end)
1697 			len = vma->vm_end - addr;
1698 
1699 		/* only read or write mappings where it is permitted */
1700 		if (write && vma->vm_flags & VM_MAYWRITE)
1701 			copy_to_user_page(vma, NULL, addr,
1702 					 (void *) addr, buf, len);
1703 		else if (!write && vma->vm_flags & VM_MAYREAD)
1704 			copy_from_user_page(vma, NULL, addr,
1705 					    buf, (void *) addr, len);
1706 		else
1707 			len = 0;
1708 	} else {
1709 		len = 0;
1710 	}
1711 
1712 	mmap_read_unlock(mm);
1713 
1714 	return len;
1715 }
1716 
1717 /**
1718  * access_remote_vm - access another process' address space
1719  * @mm:		the mm_struct of the target address space
1720  * @addr:	start address to access
1721  * @buf:	source or destination buffer
1722  * @len:	number of bytes to transfer
1723  * @gup_flags:	flags modifying lookup behaviour
1724  *
1725  * The caller must hold a reference on @mm.
1726  */
1727 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1728 		void *buf, int len, unsigned int gup_flags)
1729 {
1730 	return __access_remote_vm(mm, addr, buf, len, gup_flags);
1731 }
1732 
1733 /*
1734  * Access another process' address space.
1735  * - source/target buffer must be kernel space
1736  */
1737 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1738 		unsigned int gup_flags)
1739 {
1740 	struct mm_struct *mm;
1741 
1742 	if (addr + len < addr)
1743 		return 0;
1744 
1745 	mm = get_task_mm(tsk);
1746 	if (!mm)
1747 		return 0;
1748 
1749 	len = __access_remote_vm(mm, addr, buf, len, gup_flags);
1750 
1751 	mmput(mm);
1752 	return len;
1753 }
1754 EXPORT_SYMBOL_GPL(access_process_vm);
1755 
1756 /**
1757  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1758  * @inode: The inode to check
1759  * @size: The current filesize of the inode
1760  * @newsize: The proposed filesize of the inode
1761  *
1762  * Check the shared mappings on an inode on behalf of a shrinking truncate to
1763  * make sure that any outstanding VMAs aren't broken and then shrink the
1764  * vm_regions that extend beyond so that do_mmap() doesn't
1765  * automatically grant mappings that are too large.
1766  */
1767 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1768 				size_t newsize)
1769 {
1770 	struct vm_area_struct *vma;
1771 	struct vm_region *region;
1772 	pgoff_t low, high;
1773 	size_t r_size, r_top;
1774 
1775 	low = newsize >> PAGE_SHIFT;
1776 	high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1777 
1778 	down_write(&nommu_region_sem);
1779 	i_mmap_lock_read(inode->i_mapping);
1780 
1781 	/* search for VMAs that fall within the dead zone */
1782 	vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1783 		/* found one - only interested if it's shared out of the page
1784 		 * cache */
1785 		if (vma->vm_flags & VM_SHARED) {
1786 			i_mmap_unlock_read(inode->i_mapping);
1787 			up_write(&nommu_region_sem);
1788 			return -ETXTBSY; /* not quite true, but near enough */
1789 		}
1790 	}
1791 
1792 	/* reduce any regions that overlap the dead zone - if in existence,
1793 	 * these will be pointed to by VMAs that don't overlap the dead zone
1794 	 *
1795 	 * we don't check for any regions that start beyond the EOF as there
1796 	 * shouldn't be any
1797 	 */
1798 	vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1799 		if (!(vma->vm_flags & VM_SHARED))
1800 			continue;
1801 
1802 		region = vma->vm_region;
1803 		r_size = region->vm_top - region->vm_start;
1804 		r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1805 
1806 		if (r_top > newsize) {
1807 			region->vm_top -= r_top - newsize;
1808 			if (region->vm_end > region->vm_top)
1809 				region->vm_end = region->vm_top;
1810 		}
1811 	}
1812 
1813 	i_mmap_unlock_read(inode->i_mapping);
1814 	up_write(&nommu_region_sem);
1815 	return 0;
1816 }
1817 
1818 /*
1819  * Initialise sysctl_user_reserve_kbytes.
1820  *
1821  * This is intended to prevent a user from starting a single memory hogging
1822  * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1823  * mode.
1824  *
1825  * The default value is min(3% of free memory, 128MB)
1826  * 128MB is enough to recover with sshd/login, bash, and top/kill.
1827  */
1828 static int __meminit init_user_reserve(void)
1829 {
1830 	unsigned long free_kbytes;
1831 
1832 	free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1833 
1834 	sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1835 	return 0;
1836 }
1837 subsys_initcall(init_user_reserve);
1838 
1839 /*
1840  * Initialise sysctl_admin_reserve_kbytes.
1841  *
1842  * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1843  * to log in and kill a memory hogging process.
1844  *
1845  * Systems with more than 256MB will reserve 8MB, enough to recover
1846  * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1847  * only reserve 3% of free pages by default.
1848  */
1849 static int __meminit init_admin_reserve(void)
1850 {
1851 	unsigned long free_kbytes;
1852 
1853 	free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1854 
1855 	sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1856 	return 0;
1857 }
1858 subsys_initcall(init_admin_reserve);
1859