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