xref: /openbmc/linux/mm/nommu.c (revision 92a2c6b2)
1 /*
2  *  linux/mm/nommu.c
3  *
4  *  Replacement code for mm functions to support CPU's that don't
5  *  have any form of memory management unit (thus no virtual memory).
6  *
7  *  See Documentation/nommu-mmap.txt
8  *
9  *  Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
10  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
13  *  Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
14  */
15 
16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 
18 #include <linux/export.h>
19 #include <linux/mm.h>
20 #include <linux/vmacache.h>
21 #include <linux/mman.h>
22 #include <linux/swap.h>
23 #include <linux/file.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <linux/slab.h>
27 #include <linux/vmalloc.h>
28 #include <linux/blkdev.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/sched/sysctl.h>
37 #include <linux/printk.h>
38 
39 #include <asm/uaccess.h>
40 #include <asm/tlb.h>
41 #include <asm/tlbflush.h>
42 #include <asm/mmu_context.h>
43 #include "internal.h"
44 
45 #if 0
46 #define kenter(FMT, ...) \
47 	printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
48 #define kleave(FMT, ...) \
49 	printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
50 #define kdebug(FMT, ...) \
51 	printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
52 #else
53 #define kenter(FMT, ...) \
54 	no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
55 #define kleave(FMT, ...) \
56 	no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
57 #define kdebug(FMT, ...) \
58 	no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
59 #endif
60 
61 void *high_memory;
62 EXPORT_SYMBOL(high_memory);
63 struct page *mem_map;
64 unsigned long max_mapnr;
65 unsigned long highest_memmap_pfn;
66 struct percpu_counter vm_committed_as;
67 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
68 int sysctl_overcommit_ratio = 50; /* default is 50% */
69 unsigned long sysctl_overcommit_kbytes __read_mostly;
70 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
71 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
72 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
73 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
74 int heap_stack_gap = 0;
75 
76 atomic_long_t mmap_pages_allocated;
77 
78 /*
79  * The global memory commitment made in the system can be a metric
80  * that can be used to drive ballooning decisions when Linux is hosted
81  * as a guest. On Hyper-V, the host implements a policy engine for dynamically
82  * balancing memory across competing virtual machines that are hosted.
83  * Several metrics drive this policy engine including the guest reported
84  * memory commitment.
85  */
86 unsigned long vm_memory_committed(void)
87 {
88 	return percpu_counter_read_positive(&vm_committed_as);
89 }
90 
91 EXPORT_SYMBOL_GPL(vm_memory_committed);
92 
93 EXPORT_SYMBOL(mem_map);
94 
95 /* list of mapped, potentially shareable regions */
96 static struct kmem_cache *vm_region_jar;
97 struct rb_root nommu_region_tree = RB_ROOT;
98 DECLARE_RWSEM(nommu_region_sem);
99 
100 const struct vm_operations_struct generic_file_vm_ops = {
101 };
102 
103 /*
104  * Return the total memory allocated for this pointer, not
105  * just what the caller asked for.
106  *
107  * Doesn't have to be accurate, i.e. may have races.
108  */
109 unsigned int kobjsize(const void *objp)
110 {
111 	struct page *page;
112 
113 	/*
114 	 * If the object we have should not have ksize performed on it,
115 	 * return size of 0
116 	 */
117 	if (!objp || !virt_addr_valid(objp))
118 		return 0;
119 
120 	page = virt_to_head_page(objp);
121 
122 	/*
123 	 * If the allocator sets PageSlab, we know the pointer came from
124 	 * kmalloc().
125 	 */
126 	if (PageSlab(page))
127 		return ksize(objp);
128 
129 	/*
130 	 * If it's not a compound page, see if we have a matching VMA
131 	 * region. This test is intentionally done in reverse order,
132 	 * so if there's no VMA, we still fall through and hand back
133 	 * PAGE_SIZE for 0-order pages.
134 	 */
135 	if (!PageCompound(page)) {
136 		struct vm_area_struct *vma;
137 
138 		vma = find_vma(current->mm, (unsigned long)objp);
139 		if (vma)
140 			return vma->vm_end - vma->vm_start;
141 	}
142 
143 	/*
144 	 * The ksize() function is only guaranteed to work for pointers
145 	 * returned by kmalloc(). So handle arbitrary pointers here.
146 	 */
147 	return PAGE_SIZE << compound_order(page);
148 }
149 
150 long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
151 		      unsigned long start, unsigned long nr_pages,
152 		      unsigned int foll_flags, struct page **pages,
153 		      struct vm_area_struct **vmas, int *nonblocking)
154 {
155 	struct vm_area_struct *vma;
156 	unsigned long vm_flags;
157 	int i;
158 
159 	/* calculate required read or write permissions.
160 	 * If FOLL_FORCE is set, we only require the "MAY" flags.
161 	 */
162 	vm_flags  = (foll_flags & FOLL_WRITE) ?
163 			(VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
164 	vm_flags &= (foll_flags & FOLL_FORCE) ?
165 			(VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
166 
167 	for (i = 0; i < nr_pages; i++) {
168 		vma = find_vma(mm, start);
169 		if (!vma)
170 			goto finish_or_fault;
171 
172 		/* protect what we can, including chardevs */
173 		if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
174 		    !(vm_flags & vma->vm_flags))
175 			goto finish_or_fault;
176 
177 		if (pages) {
178 			pages[i] = virt_to_page(start);
179 			if (pages[i])
180 				page_cache_get(pages[i]);
181 		}
182 		if (vmas)
183 			vmas[i] = vma;
184 		start = (start + PAGE_SIZE) & PAGE_MASK;
185 	}
186 
187 	return i;
188 
189 finish_or_fault:
190 	return i ? : -EFAULT;
191 }
192 
193 /*
194  * get a list of pages in an address range belonging to the specified process
195  * and indicate the VMA that covers each page
196  * - this is potentially dodgy as we may end incrementing the page count of a
197  *   slab page or a secondary page from a compound page
198  * - don't permit access to VMAs that don't support it, such as I/O mappings
199  */
200 long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
201 		    unsigned long start, unsigned long nr_pages,
202 		    int write, int force, struct page **pages,
203 		    struct vm_area_struct **vmas)
204 {
205 	int flags = 0;
206 
207 	if (write)
208 		flags |= FOLL_WRITE;
209 	if (force)
210 		flags |= FOLL_FORCE;
211 
212 	return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
213 				NULL);
214 }
215 EXPORT_SYMBOL(get_user_pages);
216 
217 long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm,
218 			   unsigned long start, unsigned long nr_pages,
219 			   int write, int force, struct page **pages,
220 			   int *locked)
221 {
222 	return get_user_pages(tsk, mm, start, nr_pages, write, force,
223 			      pages, NULL);
224 }
225 EXPORT_SYMBOL(get_user_pages_locked);
226 
227 long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
228 			       unsigned long start, unsigned long nr_pages,
229 			       int write, int force, struct page **pages,
230 			       unsigned int gup_flags)
231 {
232 	long ret;
233 	down_read(&mm->mmap_sem);
234 	ret = get_user_pages(tsk, mm, start, nr_pages, write, force,
235 			     pages, NULL);
236 	up_read(&mm->mmap_sem);
237 	return ret;
238 }
239 EXPORT_SYMBOL(__get_user_pages_unlocked);
240 
241 long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
242 			     unsigned long start, unsigned long nr_pages,
243 			     int write, int force, struct page **pages)
244 {
245 	return __get_user_pages_unlocked(tsk, mm, start, nr_pages, write,
246 					 force, pages, 0);
247 }
248 EXPORT_SYMBOL(get_user_pages_unlocked);
249 
250 /**
251  * follow_pfn - look up PFN at a user virtual address
252  * @vma: memory mapping
253  * @address: user virtual address
254  * @pfn: location to store found PFN
255  *
256  * Only IO mappings and raw PFN mappings are allowed.
257  *
258  * Returns zero and the pfn at @pfn on success, -ve otherwise.
259  */
260 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
261 	unsigned long *pfn)
262 {
263 	if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
264 		return -EINVAL;
265 
266 	*pfn = address >> PAGE_SHIFT;
267 	return 0;
268 }
269 EXPORT_SYMBOL(follow_pfn);
270 
271 LIST_HEAD(vmap_area_list);
272 
273 void vfree(const void *addr)
274 {
275 	kfree(addr);
276 }
277 EXPORT_SYMBOL(vfree);
278 
279 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
280 {
281 	/*
282 	 *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
283 	 * returns only a logical address.
284 	 */
285 	return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
286 }
287 EXPORT_SYMBOL(__vmalloc);
288 
289 void *vmalloc_user(unsigned long size)
290 {
291 	void *ret;
292 
293 	ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
294 			PAGE_KERNEL);
295 	if (ret) {
296 		struct vm_area_struct *vma;
297 
298 		down_write(&current->mm->mmap_sem);
299 		vma = find_vma(current->mm, (unsigned long)ret);
300 		if (vma)
301 			vma->vm_flags |= VM_USERMAP;
302 		up_write(&current->mm->mmap_sem);
303 	}
304 
305 	return ret;
306 }
307 EXPORT_SYMBOL(vmalloc_user);
308 
309 struct page *vmalloc_to_page(const void *addr)
310 {
311 	return virt_to_page(addr);
312 }
313 EXPORT_SYMBOL(vmalloc_to_page);
314 
315 unsigned long vmalloc_to_pfn(const void *addr)
316 {
317 	return page_to_pfn(virt_to_page(addr));
318 }
319 EXPORT_SYMBOL(vmalloc_to_pfn);
320 
321 long vread(char *buf, char *addr, unsigned long count)
322 {
323 	/* Don't allow overflow */
324 	if ((unsigned long) buf + count < count)
325 		count = -(unsigned long) buf;
326 
327 	memcpy(buf, addr, count);
328 	return count;
329 }
330 
331 long vwrite(char *buf, char *addr, unsigned long count)
332 {
333 	/* Don't allow overflow */
334 	if ((unsigned long) addr + count < count)
335 		count = -(unsigned long) addr;
336 
337 	memcpy(addr, buf, count);
338 	return count;
339 }
340 
341 /*
342  *	vmalloc  -  allocate virtually continguos memory
343  *
344  *	@size:		allocation size
345  *
346  *	Allocate enough pages to cover @size from the page level
347  *	allocator and map them into continguos kernel virtual space.
348  *
349  *	For tight control over page level allocator and protection flags
350  *	use __vmalloc() instead.
351  */
352 void *vmalloc(unsigned long size)
353 {
354        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
355 }
356 EXPORT_SYMBOL(vmalloc);
357 
358 /*
359  *	vzalloc - allocate virtually continguos memory with zero fill
360  *
361  *	@size:		allocation size
362  *
363  *	Allocate enough pages to cover @size from the page level
364  *	allocator and map them into continguos kernel virtual space.
365  *	The memory allocated is set to zero.
366  *
367  *	For tight control over page level allocator and protection flags
368  *	use __vmalloc() instead.
369  */
370 void *vzalloc(unsigned long size)
371 {
372 	return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
373 			PAGE_KERNEL);
374 }
375 EXPORT_SYMBOL(vzalloc);
376 
377 /**
378  * vmalloc_node - allocate memory on a specific node
379  * @size:	allocation size
380  * @node:	numa node
381  *
382  * Allocate enough pages to cover @size from the page level
383  * allocator and map them into contiguous kernel virtual space.
384  *
385  * For tight control over page level allocator and protection flags
386  * use __vmalloc() instead.
387  */
388 void *vmalloc_node(unsigned long size, int node)
389 {
390 	return vmalloc(size);
391 }
392 EXPORT_SYMBOL(vmalloc_node);
393 
394 /**
395  * vzalloc_node - allocate memory on a specific node with zero fill
396  * @size:	allocation size
397  * @node:	numa node
398  *
399  * Allocate enough pages to cover @size from the page level
400  * allocator and map them into contiguous kernel virtual space.
401  * The memory allocated is set to zero.
402  *
403  * For tight control over page level allocator and protection flags
404  * use __vmalloc() instead.
405  */
406 void *vzalloc_node(unsigned long size, int node)
407 {
408 	return vzalloc(size);
409 }
410 EXPORT_SYMBOL(vzalloc_node);
411 
412 #ifndef PAGE_KERNEL_EXEC
413 # define PAGE_KERNEL_EXEC PAGE_KERNEL
414 #endif
415 
416 /**
417  *	vmalloc_exec  -  allocate virtually contiguous, executable memory
418  *	@size:		allocation size
419  *
420  *	Kernel-internal function to allocate enough pages to cover @size
421  *	the page level allocator and map them into contiguous and
422  *	executable kernel virtual space.
423  *
424  *	For tight control over page level allocator and protection flags
425  *	use __vmalloc() instead.
426  */
427 
428 void *vmalloc_exec(unsigned long size)
429 {
430 	return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
431 }
432 
433 /**
434  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
435  *	@size:		allocation size
436  *
437  *	Allocate enough 32bit PA addressable pages to cover @size from the
438  *	page level allocator and map them into continguos kernel virtual space.
439  */
440 void *vmalloc_32(unsigned long size)
441 {
442 	return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
443 }
444 EXPORT_SYMBOL(vmalloc_32);
445 
446 /**
447  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
448  *	@size:		allocation size
449  *
450  * The resulting memory area is 32bit addressable and zeroed so it can be
451  * mapped to userspace without leaking data.
452  *
453  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
454  * remap_vmalloc_range() are permissible.
455  */
456 void *vmalloc_32_user(unsigned long size)
457 {
458 	/*
459 	 * We'll have to sort out the ZONE_DMA bits for 64-bit,
460 	 * but for now this can simply use vmalloc_user() directly.
461 	 */
462 	return vmalloc_user(size);
463 }
464 EXPORT_SYMBOL(vmalloc_32_user);
465 
466 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
467 {
468 	BUG();
469 	return NULL;
470 }
471 EXPORT_SYMBOL(vmap);
472 
473 void vunmap(const void *addr)
474 {
475 	BUG();
476 }
477 EXPORT_SYMBOL(vunmap);
478 
479 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
480 {
481 	BUG();
482 	return NULL;
483 }
484 EXPORT_SYMBOL(vm_map_ram);
485 
486 void vm_unmap_ram(const void *mem, unsigned int count)
487 {
488 	BUG();
489 }
490 EXPORT_SYMBOL(vm_unmap_ram);
491 
492 void vm_unmap_aliases(void)
493 {
494 }
495 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
496 
497 /*
498  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
499  * have one.
500  */
501 void __weak vmalloc_sync_all(void)
502 {
503 }
504 
505 /**
506  *	alloc_vm_area - allocate a range of kernel address space
507  *	@size:		size of the area
508  *
509  *	Returns:	NULL on failure, vm_struct on success
510  *
511  *	This function reserves a range of kernel address space, and
512  *	allocates pagetables to map that range.  No actual mappings
513  *	are created.  If the kernel address space is not shared
514  *	between processes, it syncs the pagetable across all
515  *	processes.
516  */
517 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
518 {
519 	BUG();
520 	return NULL;
521 }
522 EXPORT_SYMBOL_GPL(alloc_vm_area);
523 
524 void free_vm_area(struct vm_struct *area)
525 {
526 	BUG();
527 }
528 EXPORT_SYMBOL_GPL(free_vm_area);
529 
530 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
531 		   struct page *page)
532 {
533 	return -EINVAL;
534 }
535 EXPORT_SYMBOL(vm_insert_page);
536 
537 /*
538  *  sys_brk() for the most part doesn't need the global kernel
539  *  lock, except when an application is doing something nasty
540  *  like trying to un-brk an area that has already been mapped
541  *  to a regular file.  in this case, the unmapping will need
542  *  to invoke file system routines that need the global lock.
543  */
544 SYSCALL_DEFINE1(brk, unsigned long, brk)
545 {
546 	struct mm_struct *mm = current->mm;
547 
548 	if (brk < mm->start_brk || brk > mm->context.end_brk)
549 		return mm->brk;
550 
551 	if (mm->brk == brk)
552 		return mm->brk;
553 
554 	/*
555 	 * Always allow shrinking brk
556 	 */
557 	if (brk <= mm->brk) {
558 		mm->brk = brk;
559 		return brk;
560 	}
561 
562 	/*
563 	 * Ok, looks good - let it rip.
564 	 */
565 	flush_icache_range(mm->brk, brk);
566 	return mm->brk = brk;
567 }
568 
569 /*
570  * initialise the VMA and region record slabs
571  */
572 void __init mmap_init(void)
573 {
574 	int ret;
575 
576 	ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
577 	VM_BUG_ON(ret);
578 	vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
579 }
580 
581 /*
582  * validate the region tree
583  * - the caller must hold the region lock
584  */
585 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
586 static noinline void validate_nommu_regions(void)
587 {
588 	struct vm_region *region, *last;
589 	struct rb_node *p, *lastp;
590 
591 	lastp = rb_first(&nommu_region_tree);
592 	if (!lastp)
593 		return;
594 
595 	last = rb_entry(lastp, struct vm_region, vm_rb);
596 	BUG_ON(unlikely(last->vm_end <= last->vm_start));
597 	BUG_ON(unlikely(last->vm_top < last->vm_end));
598 
599 	while ((p = rb_next(lastp))) {
600 		region = rb_entry(p, struct vm_region, vm_rb);
601 		last = rb_entry(lastp, struct vm_region, vm_rb);
602 
603 		BUG_ON(unlikely(region->vm_end <= region->vm_start));
604 		BUG_ON(unlikely(region->vm_top < region->vm_end));
605 		BUG_ON(unlikely(region->vm_start < last->vm_top));
606 
607 		lastp = p;
608 	}
609 }
610 #else
611 static void validate_nommu_regions(void)
612 {
613 }
614 #endif
615 
616 /*
617  * add a region into the global tree
618  */
619 static void add_nommu_region(struct vm_region *region)
620 {
621 	struct vm_region *pregion;
622 	struct rb_node **p, *parent;
623 
624 	validate_nommu_regions();
625 
626 	parent = NULL;
627 	p = &nommu_region_tree.rb_node;
628 	while (*p) {
629 		parent = *p;
630 		pregion = rb_entry(parent, struct vm_region, vm_rb);
631 		if (region->vm_start < pregion->vm_start)
632 			p = &(*p)->rb_left;
633 		else if (region->vm_start > pregion->vm_start)
634 			p = &(*p)->rb_right;
635 		else if (pregion == region)
636 			return;
637 		else
638 			BUG();
639 	}
640 
641 	rb_link_node(&region->vm_rb, parent, p);
642 	rb_insert_color(&region->vm_rb, &nommu_region_tree);
643 
644 	validate_nommu_regions();
645 }
646 
647 /*
648  * delete a region from the global tree
649  */
650 static void delete_nommu_region(struct vm_region *region)
651 {
652 	BUG_ON(!nommu_region_tree.rb_node);
653 
654 	validate_nommu_regions();
655 	rb_erase(&region->vm_rb, &nommu_region_tree);
656 	validate_nommu_regions();
657 }
658 
659 /*
660  * free a contiguous series of pages
661  */
662 static void free_page_series(unsigned long from, unsigned long to)
663 {
664 	for (; from < to; from += PAGE_SIZE) {
665 		struct page *page = virt_to_page(from);
666 
667 		kdebug("- free %lx", from);
668 		atomic_long_dec(&mmap_pages_allocated);
669 		if (page_count(page) != 1)
670 			kdebug("free page %p: refcount not one: %d",
671 			       page, page_count(page));
672 		put_page(page);
673 	}
674 }
675 
676 /*
677  * release a reference to a region
678  * - the caller must hold the region semaphore for writing, which this releases
679  * - the region may not have been added to the tree yet, in which case vm_top
680  *   will equal vm_start
681  */
682 static void __put_nommu_region(struct vm_region *region)
683 	__releases(nommu_region_sem)
684 {
685 	kenter("%p{%d}", region, region->vm_usage);
686 
687 	BUG_ON(!nommu_region_tree.rb_node);
688 
689 	if (--region->vm_usage == 0) {
690 		if (region->vm_top > region->vm_start)
691 			delete_nommu_region(region);
692 		up_write(&nommu_region_sem);
693 
694 		if (region->vm_file)
695 			fput(region->vm_file);
696 
697 		/* IO memory and memory shared directly out of the pagecache
698 		 * from ramfs/tmpfs mustn't be released here */
699 		if (region->vm_flags & VM_MAPPED_COPY) {
700 			kdebug("free series");
701 			free_page_series(region->vm_start, region->vm_top);
702 		}
703 		kmem_cache_free(vm_region_jar, region);
704 	} else {
705 		up_write(&nommu_region_sem);
706 	}
707 }
708 
709 /*
710  * release a reference to a region
711  */
712 static void put_nommu_region(struct vm_region *region)
713 {
714 	down_write(&nommu_region_sem);
715 	__put_nommu_region(region);
716 }
717 
718 /*
719  * update protection on a vma
720  */
721 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
722 {
723 #ifdef CONFIG_MPU
724 	struct mm_struct *mm = vma->vm_mm;
725 	long start = vma->vm_start & PAGE_MASK;
726 	while (start < vma->vm_end) {
727 		protect_page(mm, start, flags);
728 		start += PAGE_SIZE;
729 	}
730 	update_protections(mm);
731 #endif
732 }
733 
734 /*
735  * add a VMA into a process's mm_struct in the appropriate place in the list
736  * and tree and add to the address space's page tree also if not an anonymous
737  * page
738  * - should be called with mm->mmap_sem held writelocked
739  */
740 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
741 {
742 	struct vm_area_struct *pvma, *prev;
743 	struct address_space *mapping;
744 	struct rb_node **p, *parent, *rb_prev;
745 
746 	kenter(",%p", vma);
747 
748 	BUG_ON(!vma->vm_region);
749 
750 	mm->map_count++;
751 	vma->vm_mm = mm;
752 
753 	protect_vma(vma, vma->vm_flags);
754 
755 	/* add the VMA to the mapping */
756 	if (vma->vm_file) {
757 		mapping = vma->vm_file->f_mapping;
758 
759 		i_mmap_lock_write(mapping);
760 		flush_dcache_mmap_lock(mapping);
761 		vma_interval_tree_insert(vma, &mapping->i_mmap);
762 		flush_dcache_mmap_unlock(mapping);
763 		i_mmap_unlock_write(mapping);
764 	}
765 
766 	/* add the VMA to the tree */
767 	parent = rb_prev = NULL;
768 	p = &mm->mm_rb.rb_node;
769 	while (*p) {
770 		parent = *p;
771 		pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
772 
773 		/* sort by: start addr, end addr, VMA struct addr in that order
774 		 * (the latter is necessary as we may get identical VMAs) */
775 		if (vma->vm_start < pvma->vm_start)
776 			p = &(*p)->rb_left;
777 		else if (vma->vm_start > pvma->vm_start) {
778 			rb_prev = parent;
779 			p = &(*p)->rb_right;
780 		} else if (vma->vm_end < pvma->vm_end)
781 			p = &(*p)->rb_left;
782 		else if (vma->vm_end > pvma->vm_end) {
783 			rb_prev = parent;
784 			p = &(*p)->rb_right;
785 		} else if (vma < pvma)
786 			p = &(*p)->rb_left;
787 		else if (vma > pvma) {
788 			rb_prev = parent;
789 			p = &(*p)->rb_right;
790 		} else
791 			BUG();
792 	}
793 
794 	rb_link_node(&vma->vm_rb, parent, p);
795 	rb_insert_color(&vma->vm_rb, &mm->mm_rb);
796 
797 	/* add VMA to the VMA list also */
798 	prev = NULL;
799 	if (rb_prev)
800 		prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
801 
802 	__vma_link_list(mm, vma, prev, parent);
803 }
804 
805 /*
806  * delete a VMA from its owning mm_struct and address space
807  */
808 static void delete_vma_from_mm(struct vm_area_struct *vma)
809 {
810 	int i;
811 	struct address_space *mapping;
812 	struct mm_struct *mm = vma->vm_mm;
813 	struct task_struct *curr = current;
814 
815 	kenter("%p", vma);
816 
817 	protect_vma(vma, 0);
818 
819 	mm->map_count--;
820 	for (i = 0; i < VMACACHE_SIZE; i++) {
821 		/* if the vma is cached, invalidate the entire cache */
822 		if (curr->vmacache[i] == vma) {
823 			vmacache_invalidate(mm);
824 			break;
825 		}
826 	}
827 
828 	/* remove the VMA from the mapping */
829 	if (vma->vm_file) {
830 		mapping = vma->vm_file->f_mapping;
831 
832 		i_mmap_lock_write(mapping);
833 		flush_dcache_mmap_lock(mapping);
834 		vma_interval_tree_remove(vma, &mapping->i_mmap);
835 		flush_dcache_mmap_unlock(mapping);
836 		i_mmap_unlock_write(mapping);
837 	}
838 
839 	/* remove from the MM's tree and list */
840 	rb_erase(&vma->vm_rb, &mm->mm_rb);
841 
842 	if (vma->vm_prev)
843 		vma->vm_prev->vm_next = vma->vm_next;
844 	else
845 		mm->mmap = vma->vm_next;
846 
847 	if (vma->vm_next)
848 		vma->vm_next->vm_prev = vma->vm_prev;
849 }
850 
851 /*
852  * destroy a VMA record
853  */
854 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
855 {
856 	kenter("%p", vma);
857 	if (vma->vm_ops && vma->vm_ops->close)
858 		vma->vm_ops->close(vma);
859 	if (vma->vm_file)
860 		fput(vma->vm_file);
861 	put_nommu_region(vma->vm_region);
862 	kmem_cache_free(vm_area_cachep, vma);
863 }
864 
865 /*
866  * look up the first VMA in which addr resides, NULL if none
867  * - should be called with mm->mmap_sem at least held readlocked
868  */
869 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
870 {
871 	struct vm_area_struct *vma;
872 
873 	/* check the cache first */
874 	vma = vmacache_find(mm, addr);
875 	if (likely(vma))
876 		return vma;
877 
878 	/* trawl the list (there may be multiple mappings in which addr
879 	 * resides) */
880 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
881 		if (vma->vm_start > addr)
882 			return NULL;
883 		if (vma->vm_end > addr) {
884 			vmacache_update(addr, vma);
885 			return vma;
886 		}
887 	}
888 
889 	return NULL;
890 }
891 EXPORT_SYMBOL(find_vma);
892 
893 /*
894  * find a VMA
895  * - we don't extend stack VMAs under NOMMU conditions
896  */
897 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
898 {
899 	return find_vma(mm, addr);
900 }
901 
902 /*
903  * expand a stack to a given address
904  * - not supported under NOMMU conditions
905  */
906 int expand_stack(struct vm_area_struct *vma, unsigned long address)
907 {
908 	return -ENOMEM;
909 }
910 
911 /*
912  * look up the first VMA exactly that exactly matches addr
913  * - should be called with mm->mmap_sem at least held readlocked
914  */
915 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
916 					     unsigned long addr,
917 					     unsigned long len)
918 {
919 	struct vm_area_struct *vma;
920 	unsigned long end = addr + len;
921 
922 	/* check the cache first */
923 	vma = vmacache_find_exact(mm, addr, end);
924 	if (vma)
925 		return vma;
926 
927 	/* trawl the list (there may be multiple mappings in which addr
928 	 * resides) */
929 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
930 		if (vma->vm_start < addr)
931 			continue;
932 		if (vma->vm_start > addr)
933 			return NULL;
934 		if (vma->vm_end == end) {
935 			vmacache_update(addr, vma);
936 			return vma;
937 		}
938 	}
939 
940 	return NULL;
941 }
942 
943 /*
944  * determine whether a mapping should be permitted and, if so, what sort of
945  * mapping we're capable of supporting
946  */
947 static int validate_mmap_request(struct file *file,
948 				 unsigned long addr,
949 				 unsigned long len,
950 				 unsigned long prot,
951 				 unsigned long flags,
952 				 unsigned long pgoff,
953 				 unsigned long *_capabilities)
954 {
955 	unsigned long capabilities, rlen;
956 	int ret;
957 
958 	/* do the simple checks first */
959 	if (flags & MAP_FIXED) {
960 		printk(KERN_DEBUG
961 		       "%d: Can't do fixed-address/overlay mmap of RAM\n",
962 		       current->pid);
963 		return -EINVAL;
964 	}
965 
966 	if ((flags & MAP_TYPE) != MAP_PRIVATE &&
967 	    (flags & MAP_TYPE) != MAP_SHARED)
968 		return -EINVAL;
969 
970 	if (!len)
971 		return -EINVAL;
972 
973 	/* Careful about overflows.. */
974 	rlen = PAGE_ALIGN(len);
975 	if (!rlen || rlen > TASK_SIZE)
976 		return -ENOMEM;
977 
978 	/* offset overflow? */
979 	if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
980 		return -EOVERFLOW;
981 
982 	if (file) {
983 		/* files must support mmap */
984 		if (!file->f_op->mmap)
985 			return -ENODEV;
986 
987 		/* work out if what we've got could possibly be shared
988 		 * - we support chardevs that provide their own "memory"
989 		 * - we support files/blockdevs that are memory backed
990 		 */
991 		if (file->f_op->mmap_capabilities) {
992 			capabilities = file->f_op->mmap_capabilities(file);
993 		} else {
994 			/* no explicit capabilities set, so assume some
995 			 * defaults */
996 			switch (file_inode(file)->i_mode & S_IFMT) {
997 			case S_IFREG:
998 			case S_IFBLK:
999 				capabilities = NOMMU_MAP_COPY;
1000 				break;
1001 
1002 			case S_IFCHR:
1003 				capabilities =
1004 					NOMMU_MAP_DIRECT |
1005 					NOMMU_MAP_READ |
1006 					NOMMU_MAP_WRITE;
1007 				break;
1008 
1009 			default:
1010 				return -EINVAL;
1011 			}
1012 		}
1013 
1014 		/* eliminate any capabilities that we can't support on this
1015 		 * device */
1016 		if (!file->f_op->get_unmapped_area)
1017 			capabilities &= ~NOMMU_MAP_DIRECT;
1018 		if (!file->f_op->read)
1019 			capabilities &= ~NOMMU_MAP_COPY;
1020 
1021 		/* The file shall have been opened with read permission. */
1022 		if (!(file->f_mode & FMODE_READ))
1023 			return -EACCES;
1024 
1025 		if (flags & MAP_SHARED) {
1026 			/* do checks for writing, appending and locking */
1027 			if ((prot & PROT_WRITE) &&
1028 			    !(file->f_mode & FMODE_WRITE))
1029 				return -EACCES;
1030 
1031 			if (IS_APPEND(file_inode(file)) &&
1032 			    (file->f_mode & FMODE_WRITE))
1033 				return -EACCES;
1034 
1035 			if (locks_verify_locked(file))
1036 				return -EAGAIN;
1037 
1038 			if (!(capabilities & NOMMU_MAP_DIRECT))
1039 				return -ENODEV;
1040 
1041 			/* we mustn't privatise shared mappings */
1042 			capabilities &= ~NOMMU_MAP_COPY;
1043 		} else {
1044 			/* we're going to read the file into private memory we
1045 			 * allocate */
1046 			if (!(capabilities & NOMMU_MAP_COPY))
1047 				return -ENODEV;
1048 
1049 			/* we don't permit a private writable mapping to be
1050 			 * shared with the backing device */
1051 			if (prot & PROT_WRITE)
1052 				capabilities &= ~NOMMU_MAP_DIRECT;
1053 		}
1054 
1055 		if (capabilities & NOMMU_MAP_DIRECT) {
1056 			if (((prot & PROT_READ)  && !(capabilities & NOMMU_MAP_READ))  ||
1057 			    ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
1058 			    ((prot & PROT_EXEC)  && !(capabilities & NOMMU_MAP_EXEC))
1059 			    ) {
1060 				capabilities &= ~NOMMU_MAP_DIRECT;
1061 				if (flags & MAP_SHARED) {
1062 					printk(KERN_WARNING
1063 					       "MAP_SHARED not completely supported on !MMU\n");
1064 					return -EINVAL;
1065 				}
1066 			}
1067 		}
1068 
1069 		/* handle executable mappings and implied executable
1070 		 * mappings */
1071 		if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1072 			if (prot & PROT_EXEC)
1073 				return -EPERM;
1074 		} else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1075 			/* handle implication of PROT_EXEC by PROT_READ */
1076 			if (current->personality & READ_IMPLIES_EXEC) {
1077 				if (capabilities & NOMMU_MAP_EXEC)
1078 					prot |= PROT_EXEC;
1079 			}
1080 		} else if ((prot & PROT_READ) &&
1081 			 (prot & PROT_EXEC) &&
1082 			 !(capabilities & NOMMU_MAP_EXEC)
1083 			 ) {
1084 			/* backing file is not executable, try to copy */
1085 			capabilities &= ~NOMMU_MAP_DIRECT;
1086 		}
1087 	} else {
1088 		/* anonymous mappings are always memory backed and can be
1089 		 * privately mapped
1090 		 */
1091 		capabilities = NOMMU_MAP_COPY;
1092 
1093 		/* handle PROT_EXEC implication by PROT_READ */
1094 		if ((prot & PROT_READ) &&
1095 		    (current->personality & READ_IMPLIES_EXEC))
1096 			prot |= PROT_EXEC;
1097 	}
1098 
1099 	/* allow the security API to have its say */
1100 	ret = security_mmap_addr(addr);
1101 	if (ret < 0)
1102 		return ret;
1103 
1104 	/* looks okay */
1105 	*_capabilities = capabilities;
1106 	return 0;
1107 }
1108 
1109 /*
1110  * we've determined that we can make the mapping, now translate what we
1111  * now know into VMA flags
1112  */
1113 static unsigned long determine_vm_flags(struct file *file,
1114 					unsigned long prot,
1115 					unsigned long flags,
1116 					unsigned long capabilities)
1117 {
1118 	unsigned long vm_flags;
1119 
1120 	vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1121 	/* vm_flags |= mm->def_flags; */
1122 
1123 	if (!(capabilities & NOMMU_MAP_DIRECT)) {
1124 		/* attempt to share read-only copies of mapped file chunks */
1125 		vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1126 		if (file && !(prot & PROT_WRITE))
1127 			vm_flags |= VM_MAYSHARE;
1128 	} else {
1129 		/* overlay a shareable mapping on the backing device or inode
1130 		 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1131 		 * romfs/cramfs */
1132 		vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1133 		if (flags & MAP_SHARED)
1134 			vm_flags |= VM_SHARED;
1135 	}
1136 
1137 	/* refuse to let anyone share private mappings with this process if
1138 	 * it's being traced - otherwise breakpoints set in it may interfere
1139 	 * with another untraced process
1140 	 */
1141 	if ((flags & MAP_PRIVATE) && current->ptrace)
1142 		vm_flags &= ~VM_MAYSHARE;
1143 
1144 	return vm_flags;
1145 }
1146 
1147 /*
1148  * set up a shared mapping on a file (the driver or filesystem provides and
1149  * pins the storage)
1150  */
1151 static int do_mmap_shared_file(struct vm_area_struct *vma)
1152 {
1153 	int ret;
1154 
1155 	ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1156 	if (ret == 0) {
1157 		vma->vm_region->vm_top = vma->vm_region->vm_end;
1158 		return 0;
1159 	}
1160 	if (ret != -ENOSYS)
1161 		return ret;
1162 
1163 	/* getting -ENOSYS indicates that direct mmap isn't possible (as
1164 	 * opposed to tried but failed) so we can only give a suitable error as
1165 	 * it's not possible to make a private copy if MAP_SHARED was given */
1166 	return -ENODEV;
1167 }
1168 
1169 /*
1170  * set up a private mapping or an anonymous shared mapping
1171  */
1172 static int do_mmap_private(struct vm_area_struct *vma,
1173 			   struct vm_region *region,
1174 			   unsigned long len,
1175 			   unsigned long capabilities)
1176 {
1177 	unsigned long total, point;
1178 	void *base;
1179 	int ret, order;
1180 
1181 	/* invoke the file's mapping function so that it can keep track of
1182 	 * shared mappings on devices or memory
1183 	 * - VM_MAYSHARE will be set if it may attempt to share
1184 	 */
1185 	if (capabilities & NOMMU_MAP_DIRECT) {
1186 		ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1187 		if (ret == 0) {
1188 			/* shouldn't return success if we're not sharing */
1189 			BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1190 			vma->vm_region->vm_top = vma->vm_region->vm_end;
1191 			return 0;
1192 		}
1193 		if (ret != -ENOSYS)
1194 			return ret;
1195 
1196 		/* getting an ENOSYS error indicates that direct mmap isn't
1197 		 * possible (as opposed to tried but failed) so we'll try to
1198 		 * make a private copy of the data and map that instead */
1199 	}
1200 
1201 
1202 	/* allocate some memory to hold the mapping
1203 	 * - note that this may not return a page-aligned address if the object
1204 	 *   we're allocating is smaller than a page
1205 	 */
1206 	order = get_order(len);
1207 	kdebug("alloc order %d for %lx", order, len);
1208 
1209 	total = 1 << order;
1210 	point = len >> PAGE_SHIFT;
1211 
1212 	/* we don't want to allocate a power-of-2 sized page set */
1213 	if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1214 		total = point;
1215 		kdebug("try to alloc exact %lu pages", total);
1216 		base = alloc_pages_exact(len, GFP_KERNEL);
1217 	} else {
1218 		base = (void *)__get_free_pages(GFP_KERNEL, order);
1219 	}
1220 
1221 	if (!base)
1222 		goto enomem;
1223 
1224 	atomic_long_add(total, &mmap_pages_allocated);
1225 
1226 	region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1227 	region->vm_start = (unsigned long) base;
1228 	region->vm_end   = region->vm_start + len;
1229 	region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1230 
1231 	vma->vm_start = region->vm_start;
1232 	vma->vm_end   = region->vm_start + len;
1233 
1234 	if (vma->vm_file) {
1235 		/* read the contents of a file into the copy */
1236 		mm_segment_t old_fs;
1237 		loff_t fpos;
1238 
1239 		fpos = vma->vm_pgoff;
1240 		fpos <<= PAGE_SHIFT;
1241 
1242 		old_fs = get_fs();
1243 		set_fs(KERNEL_DS);
1244 		ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
1245 		set_fs(old_fs);
1246 
1247 		if (ret < 0)
1248 			goto error_free;
1249 
1250 		/* clear the last little bit */
1251 		if (ret < len)
1252 			memset(base + ret, 0, len - ret);
1253 
1254 	}
1255 
1256 	return 0;
1257 
1258 error_free:
1259 	free_page_series(region->vm_start, region->vm_top);
1260 	region->vm_start = vma->vm_start = 0;
1261 	region->vm_end   = vma->vm_end = 0;
1262 	region->vm_top   = 0;
1263 	return ret;
1264 
1265 enomem:
1266 	pr_err("Allocation of length %lu from process %d (%s) failed\n",
1267 	       len, current->pid, current->comm);
1268 	show_free_areas(0);
1269 	return -ENOMEM;
1270 }
1271 
1272 /*
1273  * handle mapping creation for uClinux
1274  */
1275 unsigned long do_mmap_pgoff(struct file *file,
1276 			    unsigned long addr,
1277 			    unsigned long len,
1278 			    unsigned long prot,
1279 			    unsigned long flags,
1280 			    unsigned long pgoff,
1281 			    unsigned long *populate)
1282 {
1283 	struct vm_area_struct *vma;
1284 	struct vm_region *region;
1285 	struct rb_node *rb;
1286 	unsigned long capabilities, vm_flags, result;
1287 	int ret;
1288 
1289 	kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1290 
1291 	*populate = 0;
1292 
1293 	/* decide whether we should attempt the mapping, and if so what sort of
1294 	 * mapping */
1295 	ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1296 				    &capabilities);
1297 	if (ret < 0) {
1298 		kleave(" = %d [val]", ret);
1299 		return ret;
1300 	}
1301 
1302 	/* we ignore the address hint */
1303 	addr = 0;
1304 	len = PAGE_ALIGN(len);
1305 
1306 	/* we've determined that we can make the mapping, now translate what we
1307 	 * now know into VMA flags */
1308 	vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1309 
1310 	/* we're going to need to record the mapping */
1311 	region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1312 	if (!region)
1313 		goto error_getting_region;
1314 
1315 	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1316 	if (!vma)
1317 		goto error_getting_vma;
1318 
1319 	region->vm_usage = 1;
1320 	region->vm_flags = vm_flags;
1321 	region->vm_pgoff = pgoff;
1322 
1323 	INIT_LIST_HEAD(&vma->anon_vma_chain);
1324 	vma->vm_flags = vm_flags;
1325 	vma->vm_pgoff = pgoff;
1326 
1327 	if (file) {
1328 		region->vm_file = get_file(file);
1329 		vma->vm_file = get_file(file);
1330 	}
1331 
1332 	down_write(&nommu_region_sem);
1333 
1334 	/* if we want to share, we need to check for regions created by other
1335 	 * mmap() calls that overlap with our proposed mapping
1336 	 * - we can only share with a superset match on most regular files
1337 	 * - shared mappings on character devices and memory backed files are
1338 	 *   permitted to overlap inexactly as far as we are concerned for in
1339 	 *   these cases, sharing is handled in the driver or filesystem rather
1340 	 *   than here
1341 	 */
1342 	if (vm_flags & VM_MAYSHARE) {
1343 		struct vm_region *pregion;
1344 		unsigned long pglen, rpglen, pgend, rpgend, start;
1345 
1346 		pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1347 		pgend = pgoff + pglen;
1348 
1349 		for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1350 			pregion = rb_entry(rb, struct vm_region, vm_rb);
1351 
1352 			if (!(pregion->vm_flags & VM_MAYSHARE))
1353 				continue;
1354 
1355 			/* search for overlapping mappings on the same file */
1356 			if (file_inode(pregion->vm_file) !=
1357 			    file_inode(file))
1358 				continue;
1359 
1360 			if (pregion->vm_pgoff >= pgend)
1361 				continue;
1362 
1363 			rpglen = pregion->vm_end - pregion->vm_start;
1364 			rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1365 			rpgend = pregion->vm_pgoff + rpglen;
1366 			if (pgoff >= rpgend)
1367 				continue;
1368 
1369 			/* handle inexactly overlapping matches between
1370 			 * mappings */
1371 			if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1372 			    !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1373 				/* new mapping is not a subset of the region */
1374 				if (!(capabilities & NOMMU_MAP_DIRECT))
1375 					goto sharing_violation;
1376 				continue;
1377 			}
1378 
1379 			/* we've found a region we can share */
1380 			pregion->vm_usage++;
1381 			vma->vm_region = pregion;
1382 			start = pregion->vm_start;
1383 			start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1384 			vma->vm_start = start;
1385 			vma->vm_end = start + len;
1386 
1387 			if (pregion->vm_flags & VM_MAPPED_COPY) {
1388 				kdebug("share copy");
1389 				vma->vm_flags |= VM_MAPPED_COPY;
1390 			} else {
1391 				kdebug("share mmap");
1392 				ret = do_mmap_shared_file(vma);
1393 				if (ret < 0) {
1394 					vma->vm_region = NULL;
1395 					vma->vm_start = 0;
1396 					vma->vm_end = 0;
1397 					pregion->vm_usage--;
1398 					pregion = NULL;
1399 					goto error_just_free;
1400 				}
1401 			}
1402 			fput(region->vm_file);
1403 			kmem_cache_free(vm_region_jar, region);
1404 			region = pregion;
1405 			result = start;
1406 			goto share;
1407 		}
1408 
1409 		/* obtain the address at which to make a shared mapping
1410 		 * - this is the hook for quasi-memory character devices to
1411 		 *   tell us the location of a shared mapping
1412 		 */
1413 		if (capabilities & NOMMU_MAP_DIRECT) {
1414 			addr = file->f_op->get_unmapped_area(file, addr, len,
1415 							     pgoff, flags);
1416 			if (IS_ERR_VALUE(addr)) {
1417 				ret = addr;
1418 				if (ret != -ENOSYS)
1419 					goto error_just_free;
1420 
1421 				/* the driver refused to tell us where to site
1422 				 * the mapping so we'll have to attempt to copy
1423 				 * it */
1424 				ret = -ENODEV;
1425 				if (!(capabilities & NOMMU_MAP_COPY))
1426 					goto error_just_free;
1427 
1428 				capabilities &= ~NOMMU_MAP_DIRECT;
1429 			} else {
1430 				vma->vm_start = region->vm_start = addr;
1431 				vma->vm_end = region->vm_end = addr + len;
1432 			}
1433 		}
1434 	}
1435 
1436 	vma->vm_region = region;
1437 
1438 	/* set up the mapping
1439 	 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1440 	 */
1441 	if (file && vma->vm_flags & VM_SHARED)
1442 		ret = do_mmap_shared_file(vma);
1443 	else
1444 		ret = do_mmap_private(vma, region, len, capabilities);
1445 	if (ret < 0)
1446 		goto error_just_free;
1447 	add_nommu_region(region);
1448 
1449 	/* clear anonymous mappings that don't ask for uninitialized data */
1450 	if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1451 		memset((void *)region->vm_start, 0,
1452 		       region->vm_end - region->vm_start);
1453 
1454 	/* okay... we have a mapping; now we have to register it */
1455 	result = vma->vm_start;
1456 
1457 	current->mm->total_vm += len >> PAGE_SHIFT;
1458 
1459 share:
1460 	add_vma_to_mm(current->mm, vma);
1461 
1462 	/* we flush the region from the icache only when the first executable
1463 	 * mapping of it is made  */
1464 	if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1465 		flush_icache_range(region->vm_start, region->vm_end);
1466 		region->vm_icache_flushed = true;
1467 	}
1468 
1469 	up_write(&nommu_region_sem);
1470 
1471 	kleave(" = %lx", result);
1472 	return result;
1473 
1474 error_just_free:
1475 	up_write(&nommu_region_sem);
1476 error:
1477 	if (region->vm_file)
1478 		fput(region->vm_file);
1479 	kmem_cache_free(vm_region_jar, region);
1480 	if (vma->vm_file)
1481 		fput(vma->vm_file);
1482 	kmem_cache_free(vm_area_cachep, vma);
1483 	kleave(" = %d", ret);
1484 	return ret;
1485 
1486 sharing_violation:
1487 	up_write(&nommu_region_sem);
1488 	printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1489 	ret = -EINVAL;
1490 	goto error;
1491 
1492 error_getting_vma:
1493 	kmem_cache_free(vm_region_jar, region);
1494 	printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1495 	       " from process %d failed\n",
1496 	       len, current->pid);
1497 	show_free_areas(0);
1498 	return -ENOMEM;
1499 
1500 error_getting_region:
1501 	printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1502 	       " from process %d failed\n",
1503 	       len, current->pid);
1504 	show_free_areas(0);
1505 	return -ENOMEM;
1506 }
1507 
1508 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1509 		unsigned long, prot, unsigned long, flags,
1510 		unsigned long, fd, unsigned long, pgoff)
1511 {
1512 	struct file *file = NULL;
1513 	unsigned long retval = -EBADF;
1514 
1515 	audit_mmap_fd(fd, flags);
1516 	if (!(flags & MAP_ANONYMOUS)) {
1517 		file = fget(fd);
1518 		if (!file)
1519 			goto out;
1520 	}
1521 
1522 	flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1523 
1524 	retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1525 
1526 	if (file)
1527 		fput(file);
1528 out:
1529 	return retval;
1530 }
1531 
1532 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1533 struct mmap_arg_struct {
1534 	unsigned long addr;
1535 	unsigned long len;
1536 	unsigned long prot;
1537 	unsigned long flags;
1538 	unsigned long fd;
1539 	unsigned long offset;
1540 };
1541 
1542 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1543 {
1544 	struct mmap_arg_struct a;
1545 
1546 	if (copy_from_user(&a, arg, sizeof(a)))
1547 		return -EFAULT;
1548 	if (a.offset & ~PAGE_MASK)
1549 		return -EINVAL;
1550 
1551 	return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1552 			      a.offset >> PAGE_SHIFT);
1553 }
1554 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1555 
1556 /*
1557  * split a vma into two pieces at address 'addr', a new vma is allocated either
1558  * for the first part or the tail.
1559  */
1560 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1561 	      unsigned long addr, int new_below)
1562 {
1563 	struct vm_area_struct *new;
1564 	struct vm_region *region;
1565 	unsigned long npages;
1566 
1567 	kenter("");
1568 
1569 	/* we're only permitted to split anonymous regions (these should have
1570 	 * only a single usage on the region) */
1571 	if (vma->vm_file)
1572 		return -ENOMEM;
1573 
1574 	if (mm->map_count >= sysctl_max_map_count)
1575 		return -ENOMEM;
1576 
1577 	region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1578 	if (!region)
1579 		return -ENOMEM;
1580 
1581 	new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1582 	if (!new) {
1583 		kmem_cache_free(vm_region_jar, region);
1584 		return -ENOMEM;
1585 	}
1586 
1587 	/* most fields are the same, copy all, and then fixup */
1588 	*new = *vma;
1589 	*region = *vma->vm_region;
1590 	new->vm_region = region;
1591 
1592 	npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1593 
1594 	if (new_below) {
1595 		region->vm_top = region->vm_end = new->vm_end = addr;
1596 	} else {
1597 		region->vm_start = new->vm_start = addr;
1598 		region->vm_pgoff = new->vm_pgoff += npages;
1599 	}
1600 
1601 	if (new->vm_ops && new->vm_ops->open)
1602 		new->vm_ops->open(new);
1603 
1604 	delete_vma_from_mm(vma);
1605 	down_write(&nommu_region_sem);
1606 	delete_nommu_region(vma->vm_region);
1607 	if (new_below) {
1608 		vma->vm_region->vm_start = vma->vm_start = addr;
1609 		vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1610 	} else {
1611 		vma->vm_region->vm_end = vma->vm_end = addr;
1612 		vma->vm_region->vm_top = addr;
1613 	}
1614 	add_nommu_region(vma->vm_region);
1615 	add_nommu_region(new->vm_region);
1616 	up_write(&nommu_region_sem);
1617 	add_vma_to_mm(mm, vma);
1618 	add_vma_to_mm(mm, new);
1619 	return 0;
1620 }
1621 
1622 /*
1623  * shrink a VMA by removing the specified chunk from either the beginning or
1624  * the end
1625  */
1626 static int shrink_vma(struct mm_struct *mm,
1627 		      struct vm_area_struct *vma,
1628 		      unsigned long from, unsigned long to)
1629 {
1630 	struct vm_region *region;
1631 
1632 	kenter("");
1633 
1634 	/* adjust the VMA's pointers, which may reposition it in the MM's tree
1635 	 * and list */
1636 	delete_vma_from_mm(vma);
1637 	if (from > vma->vm_start)
1638 		vma->vm_end = from;
1639 	else
1640 		vma->vm_start = to;
1641 	add_vma_to_mm(mm, vma);
1642 
1643 	/* cut the backing region down to size */
1644 	region = vma->vm_region;
1645 	BUG_ON(region->vm_usage != 1);
1646 
1647 	down_write(&nommu_region_sem);
1648 	delete_nommu_region(region);
1649 	if (from > region->vm_start) {
1650 		to = region->vm_top;
1651 		region->vm_top = region->vm_end = from;
1652 	} else {
1653 		region->vm_start = to;
1654 	}
1655 	add_nommu_region(region);
1656 	up_write(&nommu_region_sem);
1657 
1658 	free_page_series(from, to);
1659 	return 0;
1660 }
1661 
1662 /*
1663  * release a mapping
1664  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1665  *   VMA, though it need not cover the whole VMA
1666  */
1667 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1668 {
1669 	struct vm_area_struct *vma;
1670 	unsigned long end;
1671 	int ret;
1672 
1673 	kenter(",%lx,%zx", start, len);
1674 
1675 	len = PAGE_ALIGN(len);
1676 	if (len == 0)
1677 		return -EINVAL;
1678 
1679 	end = start + len;
1680 
1681 	/* find the first potentially overlapping VMA */
1682 	vma = find_vma(mm, start);
1683 	if (!vma) {
1684 		static int limit;
1685 		if (limit < 5) {
1686 			printk(KERN_WARNING
1687 			       "munmap of memory not mmapped by process %d"
1688 			       " (%s): 0x%lx-0x%lx\n",
1689 			       current->pid, current->comm,
1690 			       start, start + len - 1);
1691 			limit++;
1692 		}
1693 		return -EINVAL;
1694 	}
1695 
1696 	/* we're allowed to split an anonymous VMA but not a file-backed one */
1697 	if (vma->vm_file) {
1698 		do {
1699 			if (start > vma->vm_start) {
1700 				kleave(" = -EINVAL [miss]");
1701 				return -EINVAL;
1702 			}
1703 			if (end == vma->vm_end)
1704 				goto erase_whole_vma;
1705 			vma = vma->vm_next;
1706 		} while (vma);
1707 		kleave(" = -EINVAL [split file]");
1708 		return -EINVAL;
1709 	} else {
1710 		/* the chunk must be a subset of the VMA found */
1711 		if (start == vma->vm_start && end == vma->vm_end)
1712 			goto erase_whole_vma;
1713 		if (start < vma->vm_start || end > vma->vm_end) {
1714 			kleave(" = -EINVAL [superset]");
1715 			return -EINVAL;
1716 		}
1717 		if (start & ~PAGE_MASK) {
1718 			kleave(" = -EINVAL [unaligned start]");
1719 			return -EINVAL;
1720 		}
1721 		if (end != vma->vm_end && end & ~PAGE_MASK) {
1722 			kleave(" = -EINVAL [unaligned split]");
1723 			return -EINVAL;
1724 		}
1725 		if (start != vma->vm_start && end != vma->vm_end) {
1726 			ret = split_vma(mm, vma, start, 1);
1727 			if (ret < 0) {
1728 				kleave(" = %d [split]", ret);
1729 				return ret;
1730 			}
1731 		}
1732 		return shrink_vma(mm, vma, start, end);
1733 	}
1734 
1735 erase_whole_vma:
1736 	delete_vma_from_mm(vma);
1737 	delete_vma(mm, vma);
1738 	kleave(" = 0");
1739 	return 0;
1740 }
1741 EXPORT_SYMBOL(do_munmap);
1742 
1743 int vm_munmap(unsigned long addr, size_t len)
1744 {
1745 	struct mm_struct *mm = current->mm;
1746 	int ret;
1747 
1748 	down_write(&mm->mmap_sem);
1749 	ret = do_munmap(mm, addr, len);
1750 	up_write(&mm->mmap_sem);
1751 	return ret;
1752 }
1753 EXPORT_SYMBOL(vm_munmap);
1754 
1755 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1756 {
1757 	return vm_munmap(addr, len);
1758 }
1759 
1760 /*
1761  * release all the mappings made in a process's VM space
1762  */
1763 void exit_mmap(struct mm_struct *mm)
1764 {
1765 	struct vm_area_struct *vma;
1766 
1767 	if (!mm)
1768 		return;
1769 
1770 	kenter("");
1771 
1772 	mm->total_vm = 0;
1773 
1774 	while ((vma = mm->mmap)) {
1775 		mm->mmap = vma->vm_next;
1776 		delete_vma_from_mm(vma);
1777 		delete_vma(mm, vma);
1778 		cond_resched();
1779 	}
1780 
1781 	kleave("");
1782 }
1783 
1784 unsigned long vm_brk(unsigned long addr, unsigned long len)
1785 {
1786 	return -ENOMEM;
1787 }
1788 
1789 /*
1790  * expand (or shrink) an existing mapping, potentially moving it at the same
1791  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1792  *
1793  * under NOMMU conditions, we only permit changing a mapping's size, and only
1794  * as long as it stays within the region allocated by do_mmap_private() and the
1795  * block is not shareable
1796  *
1797  * MREMAP_FIXED is not supported under NOMMU conditions
1798  */
1799 static unsigned long do_mremap(unsigned long addr,
1800 			unsigned long old_len, unsigned long new_len,
1801 			unsigned long flags, unsigned long new_addr)
1802 {
1803 	struct vm_area_struct *vma;
1804 
1805 	/* insanity checks first */
1806 	old_len = PAGE_ALIGN(old_len);
1807 	new_len = PAGE_ALIGN(new_len);
1808 	if (old_len == 0 || new_len == 0)
1809 		return (unsigned long) -EINVAL;
1810 
1811 	if (addr & ~PAGE_MASK)
1812 		return -EINVAL;
1813 
1814 	if (flags & MREMAP_FIXED && new_addr != addr)
1815 		return (unsigned long) -EINVAL;
1816 
1817 	vma = find_vma_exact(current->mm, addr, old_len);
1818 	if (!vma)
1819 		return (unsigned long) -EINVAL;
1820 
1821 	if (vma->vm_end != vma->vm_start + old_len)
1822 		return (unsigned long) -EFAULT;
1823 
1824 	if (vma->vm_flags & VM_MAYSHARE)
1825 		return (unsigned long) -EPERM;
1826 
1827 	if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1828 		return (unsigned long) -ENOMEM;
1829 
1830 	/* all checks complete - do it */
1831 	vma->vm_end = vma->vm_start + new_len;
1832 	return vma->vm_start;
1833 }
1834 
1835 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1836 		unsigned long, new_len, unsigned long, flags,
1837 		unsigned long, new_addr)
1838 {
1839 	unsigned long ret;
1840 
1841 	down_write(&current->mm->mmap_sem);
1842 	ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1843 	up_write(&current->mm->mmap_sem);
1844 	return ret;
1845 }
1846 
1847 struct page *follow_page_mask(struct vm_area_struct *vma,
1848 			      unsigned long address, unsigned int flags,
1849 			      unsigned int *page_mask)
1850 {
1851 	*page_mask = 0;
1852 	return NULL;
1853 }
1854 
1855 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1856 		unsigned long pfn, unsigned long size, pgprot_t prot)
1857 {
1858 	if (addr != (pfn << PAGE_SHIFT))
1859 		return -EINVAL;
1860 
1861 	vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1862 	return 0;
1863 }
1864 EXPORT_SYMBOL(remap_pfn_range);
1865 
1866 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1867 {
1868 	unsigned long pfn = start >> PAGE_SHIFT;
1869 	unsigned long vm_len = vma->vm_end - vma->vm_start;
1870 
1871 	pfn += vma->vm_pgoff;
1872 	return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1873 }
1874 EXPORT_SYMBOL(vm_iomap_memory);
1875 
1876 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1877 			unsigned long pgoff)
1878 {
1879 	unsigned int size = vma->vm_end - vma->vm_start;
1880 
1881 	if (!(vma->vm_flags & VM_USERMAP))
1882 		return -EINVAL;
1883 
1884 	vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1885 	vma->vm_end = vma->vm_start + size;
1886 
1887 	return 0;
1888 }
1889 EXPORT_SYMBOL(remap_vmalloc_range);
1890 
1891 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1892 	unsigned long len, unsigned long pgoff, unsigned long flags)
1893 {
1894 	return -ENOMEM;
1895 }
1896 
1897 void unmap_mapping_range(struct address_space *mapping,
1898 			 loff_t const holebegin, loff_t const holelen,
1899 			 int even_cows)
1900 {
1901 }
1902 EXPORT_SYMBOL(unmap_mapping_range);
1903 
1904 /*
1905  * Check that a process has enough memory to allocate a new virtual
1906  * mapping. 0 means there is enough memory for the allocation to
1907  * succeed and -ENOMEM implies there is not.
1908  *
1909  * We currently support three overcommit policies, which are set via the
1910  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1911  *
1912  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1913  * Additional code 2002 Jul 20 by Robert Love.
1914  *
1915  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1916  *
1917  * Note this is a helper function intended to be used by LSMs which
1918  * wish to use this logic.
1919  */
1920 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1921 {
1922 	long free, allowed, reserve;
1923 
1924 	vm_acct_memory(pages);
1925 
1926 	/*
1927 	 * Sometimes we want to use more memory than we have
1928 	 */
1929 	if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1930 		return 0;
1931 
1932 	if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1933 		free = global_page_state(NR_FREE_PAGES);
1934 		free += global_page_state(NR_FILE_PAGES);
1935 
1936 		/*
1937 		 * shmem pages shouldn't be counted as free in this
1938 		 * case, they can't be purged, only swapped out, and
1939 		 * that won't affect the overall amount of available
1940 		 * memory in the system.
1941 		 */
1942 		free -= global_page_state(NR_SHMEM);
1943 
1944 		free += get_nr_swap_pages();
1945 
1946 		/*
1947 		 * Any slabs which are created with the
1948 		 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1949 		 * which are reclaimable, under pressure.  The dentry
1950 		 * cache and most inode caches should fall into this
1951 		 */
1952 		free += global_page_state(NR_SLAB_RECLAIMABLE);
1953 
1954 		/*
1955 		 * Leave reserved pages. The pages are not for anonymous pages.
1956 		 */
1957 		if (free <= totalreserve_pages)
1958 			goto error;
1959 		else
1960 			free -= totalreserve_pages;
1961 
1962 		/*
1963 		 * Reserve some for root
1964 		 */
1965 		if (!cap_sys_admin)
1966 			free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1967 
1968 		if (free > pages)
1969 			return 0;
1970 
1971 		goto error;
1972 	}
1973 
1974 	allowed = vm_commit_limit();
1975 	/*
1976 	 * Reserve some 3% for root
1977 	 */
1978 	if (!cap_sys_admin)
1979 		allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1980 
1981 	/*
1982 	 * Don't let a single process grow so big a user can't recover
1983 	 */
1984 	if (mm) {
1985 		reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1986 		allowed -= min_t(long, mm->total_vm / 32, reserve);
1987 	}
1988 
1989 	if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1990 		return 0;
1991 
1992 error:
1993 	vm_unacct_memory(pages);
1994 
1995 	return -ENOMEM;
1996 }
1997 
1998 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1999 {
2000 	BUG();
2001 	return 0;
2002 }
2003 EXPORT_SYMBOL(filemap_fault);
2004 
2005 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2006 {
2007 	BUG();
2008 }
2009 EXPORT_SYMBOL(filemap_map_pages);
2010 
2011 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
2012 		unsigned long addr, void *buf, int len, int write)
2013 {
2014 	struct vm_area_struct *vma;
2015 
2016 	down_read(&mm->mmap_sem);
2017 
2018 	/* the access must start within one of the target process's mappings */
2019 	vma = find_vma(mm, addr);
2020 	if (vma) {
2021 		/* don't overrun this mapping */
2022 		if (addr + len >= vma->vm_end)
2023 			len = vma->vm_end - addr;
2024 
2025 		/* only read or write mappings where it is permitted */
2026 		if (write && vma->vm_flags & VM_MAYWRITE)
2027 			copy_to_user_page(vma, NULL, addr,
2028 					 (void *) addr, buf, len);
2029 		else if (!write && vma->vm_flags & VM_MAYREAD)
2030 			copy_from_user_page(vma, NULL, addr,
2031 					    buf, (void *) addr, len);
2032 		else
2033 			len = 0;
2034 	} else {
2035 		len = 0;
2036 	}
2037 
2038 	up_read(&mm->mmap_sem);
2039 
2040 	return len;
2041 }
2042 
2043 /**
2044  * @access_remote_vm - access another process' address space
2045  * @mm:		the mm_struct of the target address space
2046  * @addr:	start address to access
2047  * @buf:	source or destination buffer
2048  * @len:	number of bytes to transfer
2049  * @write:	whether the access is a write
2050  *
2051  * The caller must hold a reference on @mm.
2052  */
2053 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2054 		void *buf, int len, int write)
2055 {
2056 	return __access_remote_vm(NULL, mm, addr, buf, len, write);
2057 }
2058 
2059 /*
2060  * Access another process' address space.
2061  * - source/target buffer must be kernel space
2062  */
2063 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2064 {
2065 	struct mm_struct *mm;
2066 
2067 	if (addr + len < addr)
2068 		return 0;
2069 
2070 	mm = get_task_mm(tsk);
2071 	if (!mm)
2072 		return 0;
2073 
2074 	len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2075 
2076 	mmput(mm);
2077 	return len;
2078 }
2079 
2080 /**
2081  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2082  * @inode: The inode to check
2083  * @size: The current filesize of the inode
2084  * @newsize: The proposed filesize of the inode
2085  *
2086  * Check the shared mappings on an inode on behalf of a shrinking truncate to
2087  * make sure that that any outstanding VMAs aren't broken and then shrink the
2088  * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2089  * automatically grant mappings that are too large.
2090  */
2091 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2092 				size_t newsize)
2093 {
2094 	struct vm_area_struct *vma;
2095 	struct vm_region *region;
2096 	pgoff_t low, high;
2097 	size_t r_size, r_top;
2098 
2099 	low = newsize >> PAGE_SHIFT;
2100 	high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2101 
2102 	down_write(&nommu_region_sem);
2103 	i_mmap_lock_read(inode->i_mapping);
2104 
2105 	/* search for VMAs that fall within the dead zone */
2106 	vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
2107 		/* found one - only interested if it's shared out of the page
2108 		 * cache */
2109 		if (vma->vm_flags & VM_SHARED) {
2110 			i_mmap_unlock_read(inode->i_mapping);
2111 			up_write(&nommu_region_sem);
2112 			return -ETXTBSY; /* not quite true, but near enough */
2113 		}
2114 	}
2115 
2116 	/* reduce any regions that overlap the dead zone - if in existence,
2117 	 * these will be pointed to by VMAs that don't overlap the dead zone
2118 	 *
2119 	 * we don't check for any regions that start beyond the EOF as there
2120 	 * shouldn't be any
2121 	 */
2122 	vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
2123 		if (!(vma->vm_flags & VM_SHARED))
2124 			continue;
2125 
2126 		region = vma->vm_region;
2127 		r_size = region->vm_top - region->vm_start;
2128 		r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2129 
2130 		if (r_top > newsize) {
2131 			region->vm_top -= r_top - newsize;
2132 			if (region->vm_end > region->vm_top)
2133 				region->vm_end = region->vm_top;
2134 		}
2135 	}
2136 
2137 	i_mmap_unlock_read(inode->i_mapping);
2138 	up_write(&nommu_region_sem);
2139 	return 0;
2140 }
2141 
2142 /*
2143  * Initialise sysctl_user_reserve_kbytes.
2144  *
2145  * This is intended to prevent a user from starting a single memory hogging
2146  * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2147  * mode.
2148  *
2149  * The default value is min(3% of free memory, 128MB)
2150  * 128MB is enough to recover with sshd/login, bash, and top/kill.
2151  */
2152 static int __meminit init_user_reserve(void)
2153 {
2154 	unsigned long free_kbytes;
2155 
2156 	free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2157 
2158 	sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2159 	return 0;
2160 }
2161 module_init(init_user_reserve)
2162 
2163 /*
2164  * Initialise sysctl_admin_reserve_kbytes.
2165  *
2166  * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2167  * to log in and kill a memory hogging process.
2168  *
2169  * Systems with more than 256MB will reserve 8MB, enough to recover
2170  * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2171  * only reserve 3% of free pages by default.
2172  */
2173 static int __meminit init_admin_reserve(void)
2174 {
2175 	unsigned long free_kbytes;
2176 
2177 	free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2178 
2179 	sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2180 	return 0;
2181 }
2182 module_init(init_admin_reserve)
2183