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/vmacache.h>
21#include <linux/mman.h>
22#include <linux/swap.h>
23#include <linux/file.h>
24#include <linux/highmem.h>
25#include <linux/pagemap.h>
26#include <linux/slab.h>
27#include <linux/vmalloc.h>
28#include <linux/blkdev.h>
29#include <linux/backing-dev.h>
30#include <linux/compiler.h>
31#include <linux/mount.h>
32#include <linux/personality.h>
33#include <linux/security.h>
34#include <linux/syscalls.h>
35#include <linux/audit.h>
36#include <linux/printk.h>
37
38#include <asm/uaccess.h>
39#include <asm/tlb.h>
40#include <asm/tlbflush.h>
41#include <asm/mmu_context.h>
42#include "internal.h"
43
44void *high_memory;
45EXPORT_SYMBOL(high_memory);
46struct page *mem_map;
47unsigned long max_mapnr;
48EXPORT_SYMBOL(max_mapnr);
49unsigned long highest_memmap_pfn;
50int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
51int heap_stack_gap = 0;
52
53atomic_long_t mmap_pages_allocated;
54
55EXPORT_SYMBOL(mem_map);
56
57/* list of mapped, potentially shareable regions */
58static struct kmem_cache *vm_region_jar;
59struct rb_root nommu_region_tree = RB_ROOT;
60DECLARE_RWSEM(nommu_region_sem);
61
62const struct vm_operations_struct generic_file_vm_ops = {
63};
64
65/*
66 * Return the total memory allocated for this pointer, not
67 * just what the caller asked for.
68 *
69 * Doesn't have to be accurate, i.e. may have races.
70 */
71unsigned int kobjsize(const void *objp)
72{
73 struct page *page;
74
75 /*
76 * If the object we have should not have ksize performed on it,
77 * return size of 0
78 */
79 if (!objp || !virt_addr_valid(objp))
80 return 0;
81
82 page = virt_to_head_page(objp);
83
84 /*
85 * If the allocator sets PageSlab, we know the pointer came from
86 * kmalloc().
87 */
88 if (PageSlab(page))
89 return ksize(objp);
90
91 /*
92 * If it's not a compound page, see if we have a matching VMA
93 * region. This test is intentionally done in reverse order,
94 * so if there's no VMA, we still fall through and hand back
95 * PAGE_SIZE for 0-order pages.
96 */
97 if (!PageCompound(page)) {
98 struct vm_area_struct *vma;
99
100 vma = find_vma(current->mm, (unsigned long)objp);
101 if (vma)
102 return vma->vm_end - vma->vm_start;
103 }
104
105 /*
106 * The ksize() function is only guaranteed to work for pointers
107 * returned by kmalloc(). So handle arbitrary pointers here.
108 */
109 return PAGE_SIZE << compound_order(page);
110}
111
112long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
113 unsigned long start, unsigned long nr_pages,
114 unsigned int foll_flags, struct page **pages,
115 struct vm_area_struct **vmas, int *nonblocking)
116{
117 struct vm_area_struct *vma;
118 unsigned long vm_flags;
119 int i;
120
121 /* calculate required read or write permissions.
122 * If FOLL_FORCE is set, we only require the "MAY" flags.
123 */
124 vm_flags = (foll_flags & FOLL_WRITE) ?
125 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
126 vm_flags &= (foll_flags & FOLL_FORCE) ?
127 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
128
129 for (i = 0; i < nr_pages; i++) {
130 vma = find_vma(mm, start);
131 if (!vma)
132 goto finish_or_fault;
133
134 /* protect what we can, including chardevs */
135 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
136 !(vm_flags & vma->vm_flags))
137 goto finish_or_fault;
138
139 if (pages) {
140 pages[i] = virt_to_page(start);
141 if (pages[i])
142 get_page(pages[i]);
143 }
144 if (vmas)
145 vmas[i] = vma;
146 start = (start + PAGE_SIZE) & PAGE_MASK;
147 }
148
149 return i;
150
151finish_or_fault:
152 return i ? : -EFAULT;
153}
154
155/*
156 * get a list of pages in an address range belonging to the specified process
157 * and indicate the VMA that covers each page
158 * - this is potentially dodgy as we may end incrementing the page count of a
159 * slab page or a secondary page from a compound page
160 * - don't permit access to VMAs that don't support it, such as I/O mappings
161 */
162long get_user_pages(unsigned long start, unsigned long nr_pages,
163 int write, int force, struct page **pages,
164 struct vm_area_struct **vmas)
165{
166 int flags = 0;
167
168 if (write)
169 flags |= FOLL_WRITE;
170 if (force)
171 flags |= FOLL_FORCE;
172
173 return __get_user_pages(current, current->mm, start, nr_pages, flags,
174 pages, vmas, NULL);
175}
176EXPORT_SYMBOL(get_user_pages);
177
178long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
179 int write, int force, struct page **pages,
180 int *locked)
181{
182 return get_user_pages(start, nr_pages, write, force, pages, NULL);
183}
184EXPORT_SYMBOL(get_user_pages_locked);
185
186long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
187 unsigned long start, unsigned long nr_pages,
188 int write, int force, struct page **pages,
189 unsigned int gup_flags)
190{
191 long ret;
192 down_read(&mm->mmap_sem);
193 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
194 NULL, NULL);
195 up_read(&mm->mmap_sem);
196 return ret;
197}
198EXPORT_SYMBOL(__get_user_pages_unlocked);
199
200long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
201 int write, int force, struct page **pages)
202{
203 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
204 write, force, pages, 0);
205}
206EXPORT_SYMBOL(get_user_pages_unlocked);
207
208/**
209 * follow_pfn - look up PFN at a user virtual address
210 * @vma: memory mapping
211 * @address: user virtual address
212 * @pfn: location to store found PFN
213 *
214 * Only IO mappings and raw PFN mappings are allowed.
215 *
216 * Returns zero and the pfn at @pfn on success, -ve otherwise.
217 */
218int follow_pfn(struct vm_area_struct *vma, unsigned long address,
219 unsigned long *pfn)
220{
221 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
222 return -EINVAL;
223
224 *pfn = address >> PAGE_SHIFT;
225 return 0;
226}
227EXPORT_SYMBOL(follow_pfn);
228
229LIST_HEAD(vmap_area_list);
230
231void vfree(const void *addr)
232{
233 kfree(addr);
234}
235EXPORT_SYMBOL(vfree);
236
237void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
238{
239 /*
240 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
241 * returns only a logical address.
242 */
243 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
244}
245EXPORT_SYMBOL(__vmalloc);
246
247void *vmalloc_user(unsigned long size)
248{
249 void *ret;
250
251 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
252 PAGE_KERNEL);
253 if (ret) {
254 struct vm_area_struct *vma;
255
256 down_write(¤t->mm->mmap_sem);
257 vma = find_vma(current->mm, (unsigned long)ret);
258 if (vma)
259 vma->vm_flags |= VM_USERMAP;
260 up_write(¤t->mm->mmap_sem);
261 }
262
263 return ret;
264}
265EXPORT_SYMBOL(vmalloc_user);
266
267struct page *vmalloc_to_page(const void *addr)
268{
269 return virt_to_page(addr);
270}
271EXPORT_SYMBOL(vmalloc_to_page);
272
273unsigned long vmalloc_to_pfn(const void *addr)
274{
275 return page_to_pfn(virt_to_page(addr));
276}
277EXPORT_SYMBOL(vmalloc_to_pfn);
278
279long vread(char *buf, char *addr, unsigned long count)
280{
281 /* Don't allow overflow */
282 if ((unsigned long) buf + count < count)
283 count = -(unsigned long) buf;
284
285 memcpy(buf, addr, count);
286 return count;
287}
288
289long vwrite(char *buf, char *addr, unsigned long count)
290{
291 /* Don't allow overflow */
292 if ((unsigned long) addr + count < count)
293 count = -(unsigned long) addr;
294
295 memcpy(addr, buf, count);
296 return count;
297}
298
299/*
300 * vmalloc - allocate virtually contiguous memory
301 *
302 * @size: allocation size
303 *
304 * Allocate enough pages to cover @size from the page level
305 * allocator and map them into contiguous kernel virtual space.
306 *
307 * For tight control over page level allocator and protection flags
308 * use __vmalloc() instead.
309 */
310void *vmalloc(unsigned long size)
311{
312 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
313}
314EXPORT_SYMBOL(vmalloc);
315
316/*
317 * vzalloc - allocate virtually contiguous memory with zero fill
318 *
319 * @size: allocation size
320 *
321 * Allocate enough pages to cover @size from the page level
322 * allocator and map them into contiguous kernel virtual space.
323 * The memory allocated is set to zero.
324 *
325 * For tight control over page level allocator and protection flags
326 * use __vmalloc() instead.
327 */
328void *vzalloc(unsigned long size)
329{
330 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
331 PAGE_KERNEL);
332}
333EXPORT_SYMBOL(vzalloc);
334
335/**
336 * vmalloc_node - allocate memory on a specific node
337 * @size: allocation size
338 * @node: numa node
339 *
340 * Allocate enough pages to cover @size from the page level
341 * allocator and map them into contiguous kernel virtual space.
342 *
343 * For tight control over page level allocator and protection flags
344 * use __vmalloc() instead.
345 */
346void *vmalloc_node(unsigned long size, int node)
347{
348 return vmalloc(size);
349}
350EXPORT_SYMBOL(vmalloc_node);
351
352/**
353 * vzalloc_node - allocate memory on a specific node with zero fill
354 * @size: allocation size
355 * @node: numa node
356 *
357 * Allocate enough pages to cover @size from the page level
358 * allocator and map them into contiguous kernel virtual space.
359 * The memory allocated is set to zero.
360 *
361 * For tight control over page level allocator and protection flags
362 * use __vmalloc() instead.
363 */
364void *vzalloc_node(unsigned long size, int node)
365{
366 return vzalloc(size);
367}
368EXPORT_SYMBOL(vzalloc_node);
369
370#ifndef PAGE_KERNEL_EXEC
371# define PAGE_KERNEL_EXEC PAGE_KERNEL
372#endif
373
374/**
375 * vmalloc_exec - allocate virtually contiguous, executable memory
376 * @size: allocation size
377 *
378 * Kernel-internal function to allocate enough pages to cover @size
379 * the page level allocator and map them into contiguous and
380 * executable kernel virtual space.
381 *
382 * For tight control over page level allocator and protection flags
383 * use __vmalloc() instead.
384 */
385
386void *vmalloc_exec(unsigned long size)
387{
388 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
389}
390
391/**
392 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
393 * @size: allocation size
394 *
395 * Allocate enough 32bit PA addressable pages to cover @size from the
396 * page level allocator and map them into contiguous kernel virtual space.
397 */
398void *vmalloc_32(unsigned long size)
399{
400 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
401}
402EXPORT_SYMBOL(vmalloc_32);
403
404/**
405 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
406 * @size: allocation size
407 *
408 * The resulting memory area is 32bit addressable and zeroed so it can be
409 * mapped to userspace without leaking data.
410 *
411 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
412 * remap_vmalloc_range() are permissible.
413 */
414void *vmalloc_32_user(unsigned long size)
415{
416 /*
417 * We'll have to sort out the ZONE_DMA bits for 64-bit,
418 * but for now this can simply use vmalloc_user() directly.
419 */
420 return vmalloc_user(size);
421}
422EXPORT_SYMBOL(vmalloc_32_user);
423
424void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
425{
426 BUG();
427 return NULL;
428}
429EXPORT_SYMBOL(vmap);
430
431void vunmap(const void *addr)
432{
433 BUG();
434}
435EXPORT_SYMBOL(vunmap);
436
437void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
438{
439 BUG();
440 return NULL;
441}
442EXPORT_SYMBOL(vm_map_ram);
443
444void vm_unmap_ram(const void *mem, unsigned int count)
445{
446 BUG();
447}
448EXPORT_SYMBOL(vm_unmap_ram);
449
450void vm_unmap_aliases(void)
451{
452}
453EXPORT_SYMBOL_GPL(vm_unmap_aliases);
454
455/*
456 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
457 * have one.
458 */
459void __weak vmalloc_sync_all(void)
460{
461}
462
463/**
464 * alloc_vm_area - allocate a range of kernel address space
465 * @size: size of the area
466 *
467 * Returns: NULL on failure, vm_struct on success
468 *
469 * This function reserves a range of kernel address space, and
470 * allocates pagetables to map that range. No actual mappings
471 * are created. If the kernel address space is not shared
472 * between processes, it syncs the pagetable across all
473 * processes.
474 */
475struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
476{
477 BUG();
478 return NULL;
479}
480EXPORT_SYMBOL_GPL(alloc_vm_area);
481
482void free_vm_area(struct vm_struct *area)
483{
484 BUG();
485}
486EXPORT_SYMBOL_GPL(free_vm_area);
487
488int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
489 struct page *page)
490{
491 return -EINVAL;
492}
493EXPORT_SYMBOL(vm_insert_page);
494
495/*
496 * sys_brk() for the most part doesn't need the global kernel
497 * lock, except when an application is doing something nasty
498 * like trying to un-brk an area that has already been mapped
499 * to a regular file. in this case, the unmapping will need
500 * to invoke file system routines that need the global lock.
501 */
502SYSCALL_DEFINE1(brk, unsigned long, brk)
503{
504 struct mm_struct *mm = current->mm;
505
506 if (brk < mm->start_brk || brk > mm->context.end_brk)
507 return mm->brk;
508
509 if (mm->brk == brk)
510 return mm->brk;
511
512 /*
513 * Always allow shrinking brk
514 */
515 if (brk <= mm->brk) {
516 mm->brk = brk;
517 return brk;
518 }
519
520 /*
521 * Ok, looks good - let it rip.
522 */
523 flush_icache_range(mm->brk, brk);
524 return mm->brk = brk;
525}
526
527/*
528 * initialise the VMA and region record slabs
529 */
530void __init mmap_init(void)
531{
532 int ret;
533
534 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
535 VM_BUG_ON(ret);
536 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
537}
538
539/*
540 * validate the region tree
541 * - the caller must hold the region lock
542 */
543#ifdef CONFIG_DEBUG_NOMMU_REGIONS
544static noinline void validate_nommu_regions(void)
545{
546 struct vm_region *region, *last;
547 struct rb_node *p, *lastp;
548
549 lastp = rb_first(&nommu_region_tree);
550 if (!lastp)
551 return;
552
553 last = rb_entry(lastp, struct vm_region, vm_rb);
554 BUG_ON(last->vm_end <= last->vm_start);
555 BUG_ON(last->vm_top < last->vm_end);
556
557 while ((p = rb_next(lastp))) {
558 region = rb_entry(p, struct vm_region, vm_rb);
559 last = rb_entry(lastp, struct vm_region, vm_rb);
560
561 BUG_ON(region->vm_end <= region->vm_start);
562 BUG_ON(region->vm_top < region->vm_end);
563 BUG_ON(region->vm_start < last->vm_top);
564
565 lastp = p;
566 }
567}
568#else
569static void validate_nommu_regions(void)
570{
571}
572#endif
573
574/*
575 * add a region into the global tree
576 */
577static void add_nommu_region(struct vm_region *region)
578{
579 struct vm_region *pregion;
580 struct rb_node **p, *parent;
581
582 validate_nommu_regions();
583
584 parent = NULL;
585 p = &nommu_region_tree.rb_node;
586 while (*p) {
587 parent = *p;
588 pregion = rb_entry(parent, struct vm_region, vm_rb);
589 if (region->vm_start < pregion->vm_start)
590 p = &(*p)->rb_left;
591 else if (region->vm_start > pregion->vm_start)
592 p = &(*p)->rb_right;
593 else if (pregion == region)
594 return;
595 else
596 BUG();
597 }
598
599 rb_link_node(®ion->vm_rb, parent, p);
600 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
601
602 validate_nommu_regions();
603}
604
605/*
606 * delete a region from the global tree
607 */
608static void delete_nommu_region(struct vm_region *region)
609{
610 BUG_ON(!nommu_region_tree.rb_node);
611
612 validate_nommu_regions();
613 rb_erase(®ion->vm_rb, &nommu_region_tree);
614 validate_nommu_regions();
615}
616
617/*
618 * free a contiguous series of pages
619 */
620static void free_page_series(unsigned long from, unsigned long to)
621{
622 for (; from < to; from += PAGE_SIZE) {
623 struct page *page = virt_to_page(from);
624
625 atomic_long_dec(&mmap_pages_allocated);
626 put_page(page);
627 }
628}
629
630/*
631 * release a reference to a region
632 * - the caller must hold the region semaphore for writing, which this releases
633 * - the region may not have been added to the tree yet, in which case vm_top
634 * will equal vm_start
635 */
636static void __put_nommu_region(struct vm_region *region)
637 __releases(nommu_region_sem)
638{
639 BUG_ON(!nommu_region_tree.rb_node);
640
641 if (--region->vm_usage == 0) {
642 if (region->vm_top > region->vm_start)
643 delete_nommu_region(region);
644 up_write(&nommu_region_sem);
645
646 if (region->vm_file)
647 fput(region->vm_file);
648
649 /* IO memory and memory shared directly out of the pagecache
650 * from ramfs/tmpfs mustn't be released here */
651 if (region->vm_flags & VM_MAPPED_COPY)
652 free_page_series(region->vm_start, region->vm_top);
653 kmem_cache_free(vm_region_jar, region);
654 } else {
655 up_write(&nommu_region_sem);
656 }
657}
658
659/*
660 * release a reference to a region
661 */
662static void put_nommu_region(struct vm_region *region)
663{
664 down_write(&nommu_region_sem);
665 __put_nommu_region(region);
666}
667
668/*
669 * update protection on a vma
670 */
671static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
672{
673#ifdef CONFIG_MPU
674 struct mm_struct *mm = vma->vm_mm;
675 long start = vma->vm_start & PAGE_MASK;
676 while (start < vma->vm_end) {
677 protect_page(mm, start, flags);
678 start += PAGE_SIZE;
679 }
680 update_protections(mm);
681#endif
682}
683
684/*
685 * add a VMA into a process's mm_struct in the appropriate place in the list
686 * and tree and add to the address space's page tree also if not an anonymous
687 * page
688 * - should be called with mm->mmap_sem held writelocked
689 */
690static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
691{
692 struct vm_area_struct *pvma, *prev;
693 struct address_space *mapping;
694 struct rb_node **p, *parent, *rb_prev;
695
696 BUG_ON(!vma->vm_region);
697
698 mm->map_count++;
699 vma->vm_mm = mm;
700
701 protect_vma(vma, vma->vm_flags);
702
703 /* add the VMA to the mapping */
704 if (vma->vm_file) {
705 mapping = vma->vm_file->f_mapping;
706
707 i_mmap_lock_write(mapping);
708 flush_dcache_mmap_lock(mapping);
709 vma_interval_tree_insert(vma, &mapping->i_mmap);
710 flush_dcache_mmap_unlock(mapping);
711 i_mmap_unlock_write(mapping);
712 }
713
714 /* add the VMA to the tree */
715 parent = rb_prev = NULL;
716 p = &mm->mm_rb.rb_node;
717 while (*p) {
718 parent = *p;
719 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
720
721 /* sort by: start addr, end addr, VMA struct addr in that order
722 * (the latter is necessary as we may get identical VMAs) */
723 if (vma->vm_start < pvma->vm_start)
724 p = &(*p)->rb_left;
725 else if (vma->vm_start > pvma->vm_start) {
726 rb_prev = parent;
727 p = &(*p)->rb_right;
728 } else if (vma->vm_end < pvma->vm_end)
729 p = &(*p)->rb_left;
730 else if (vma->vm_end > pvma->vm_end) {
731 rb_prev = parent;
732 p = &(*p)->rb_right;
733 } else if (vma < pvma)
734 p = &(*p)->rb_left;
735 else if (vma > pvma) {
736 rb_prev = parent;
737 p = &(*p)->rb_right;
738 } else
739 BUG();
740 }
741
742 rb_link_node(&vma->vm_rb, parent, p);
743 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
744
745 /* add VMA to the VMA list also */
746 prev = NULL;
747 if (rb_prev)
748 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
749
750 __vma_link_list(mm, vma, prev, parent);
751}
752
753/*
754 * delete a VMA from its owning mm_struct and address space
755 */
756static void delete_vma_from_mm(struct vm_area_struct *vma)
757{
758 int i;
759 struct address_space *mapping;
760 struct mm_struct *mm = vma->vm_mm;
761 struct task_struct *curr = current;
762
763 protect_vma(vma, 0);
764
765 mm->map_count--;
766 for (i = 0; i < VMACACHE_SIZE; i++) {
767 /* if the vma is cached, invalidate the entire cache */
768 if (curr->vmacache[i] == vma) {
769 vmacache_invalidate(mm);
770 break;
771 }
772 }
773
774 /* remove the VMA from the mapping */
775 if (vma->vm_file) {
776 mapping = vma->vm_file->f_mapping;
777
778 i_mmap_lock_write(mapping);
779 flush_dcache_mmap_lock(mapping);
780 vma_interval_tree_remove(vma, &mapping->i_mmap);
781 flush_dcache_mmap_unlock(mapping);
782 i_mmap_unlock_write(mapping);
783 }
784
785 /* remove from the MM's tree and list */
786 rb_erase(&vma->vm_rb, &mm->mm_rb);
787
788 if (vma->vm_prev)
789 vma->vm_prev->vm_next = vma->vm_next;
790 else
791 mm->mmap = vma->vm_next;
792
793 if (vma->vm_next)
794 vma->vm_next->vm_prev = vma->vm_prev;
795}
796
797/*
798 * destroy a VMA record
799 */
800static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
801{
802 if (vma->vm_ops && vma->vm_ops->close)
803 vma->vm_ops->close(vma);
804 if (vma->vm_file)
805 fput(vma->vm_file);
806 put_nommu_region(vma->vm_region);
807 kmem_cache_free(vm_area_cachep, vma);
808}
809
810/*
811 * look up the first VMA in which addr resides, NULL if none
812 * - should be called with mm->mmap_sem at least held readlocked
813 */
814struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
815{
816 struct vm_area_struct *vma;
817
818 /* check the cache first */
819 vma = vmacache_find(mm, addr);
820 if (likely(vma))
821 return vma;
822
823 /* trawl the list (there may be multiple mappings in which addr
824 * resides) */
825 for (vma = mm->mmap; vma; vma = vma->vm_next) {
826 if (vma->vm_start > addr)
827 return NULL;
828 if (vma->vm_end > addr) {
829 vmacache_update(addr, vma);
830 return vma;
831 }
832 }
833
834 return NULL;
835}
836EXPORT_SYMBOL(find_vma);
837
838/*
839 * find a VMA
840 * - we don't extend stack VMAs under NOMMU conditions
841 */
842struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
843{
844 return find_vma(mm, addr);
845}
846
847/*
848 * expand a stack to a given address
849 * - not supported under NOMMU conditions
850 */
851int expand_stack(struct vm_area_struct *vma, unsigned long address)
852{
853 return -ENOMEM;
854}
855
856/*
857 * look up the first VMA exactly that exactly matches addr
858 * - should be called with mm->mmap_sem at least held readlocked
859 */
860static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
861 unsigned long addr,
862 unsigned long len)
863{
864 struct vm_area_struct *vma;
865 unsigned long end = addr + len;
866
867 /* check the cache first */
868 vma = vmacache_find_exact(mm, addr, end);
869 if (vma)
870 return vma;
871
872 /* trawl the list (there may be multiple mappings in which addr
873 * resides) */
874 for (vma = mm->mmap; vma; vma = vma->vm_next) {
875 if (vma->vm_start < addr)
876 continue;
877 if (vma->vm_start > addr)
878 return NULL;
879 if (vma->vm_end == end) {
880 vmacache_update(addr, vma);
881 return vma;
882 }
883 }
884
885 return NULL;
886}
887
888/*
889 * determine whether a mapping should be permitted and, if so, what sort of
890 * mapping we're capable of supporting
891 */
892static int validate_mmap_request(struct file *file,
893 unsigned long addr,
894 unsigned long len,
895 unsigned long prot,
896 unsigned long flags,
897 unsigned long pgoff,
898 unsigned long *_capabilities)
899{
900 unsigned long capabilities, rlen;
901 int ret;
902
903 /* do the simple checks first */
904 if (flags & MAP_FIXED)
905 return -EINVAL;
906
907 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
908 (flags & MAP_TYPE) != MAP_SHARED)
909 return -EINVAL;
910
911 if (!len)
912 return -EINVAL;
913
914 /* Careful about overflows.. */
915 rlen = PAGE_ALIGN(len);
916 if (!rlen || rlen > TASK_SIZE)
917 return -ENOMEM;
918
919 /* offset overflow? */
920 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
921 return -EOVERFLOW;
922
923 if (file) {
924 /* files must support mmap */
925 if (!file->f_op->mmap)
926 return -ENODEV;
927
928 /* work out if what we've got could possibly be shared
929 * - we support chardevs that provide their own "memory"
930 * - we support files/blockdevs that are memory backed
931 */
932 if (file->f_op->mmap_capabilities) {
933 capabilities = file->f_op->mmap_capabilities(file);
934 } else {
935 /* no explicit capabilities set, so assume some
936 * defaults */
937 switch (file_inode(file)->i_mode & S_IFMT) {
938 case S_IFREG:
939 case S_IFBLK:
940 capabilities = NOMMU_MAP_COPY;
941 break;
942
943 case S_IFCHR:
944 capabilities =
945 NOMMU_MAP_DIRECT |
946 NOMMU_MAP_READ |
947 NOMMU_MAP_WRITE;
948 break;
949
950 default:
951 return -EINVAL;
952 }
953 }
954
955 /* eliminate any capabilities that we can't support on this
956 * device */
957 if (!file->f_op->get_unmapped_area)
958 capabilities &= ~NOMMU_MAP_DIRECT;
959 if (!(file->f_mode & FMODE_CAN_READ))
960 capabilities &= ~NOMMU_MAP_COPY;
961
962 /* The file shall have been opened with read permission. */
963 if (!(file->f_mode & FMODE_READ))
964 return -EACCES;
965
966 if (flags & MAP_SHARED) {
967 /* do checks for writing, appending and locking */
968 if ((prot & PROT_WRITE) &&
969 !(file->f_mode & FMODE_WRITE))
970 return -EACCES;
971
972 if (IS_APPEND(file_inode(file)) &&
973 (file->f_mode & FMODE_WRITE))
974 return -EACCES;
975
976 if (locks_verify_locked(file))
977 return -EAGAIN;
978
979 if (!(capabilities & NOMMU_MAP_DIRECT))
980 return -ENODEV;
981
982 /* we mustn't privatise shared mappings */
983 capabilities &= ~NOMMU_MAP_COPY;
984 } else {
985 /* we're going to read the file into private memory we
986 * allocate */
987 if (!(capabilities & NOMMU_MAP_COPY))
988 return -ENODEV;
989
990 /* we don't permit a private writable mapping to be
991 * shared with the backing device */
992 if (prot & PROT_WRITE)
993 capabilities &= ~NOMMU_MAP_DIRECT;
994 }
995
996 if (capabilities & NOMMU_MAP_DIRECT) {
997 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
998 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
999 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
1000 ) {
1001 capabilities &= ~NOMMU_MAP_DIRECT;
1002 if (flags & MAP_SHARED) {
1003 pr_warn("MAP_SHARED not completely supported on !MMU\n");
1004 return -EINVAL;
1005 }
1006 }
1007 }
1008
1009 /* handle executable mappings and implied executable
1010 * mappings */
1011 if (path_noexec(&file->f_path)) {
1012 if (prot & PROT_EXEC)
1013 return -EPERM;
1014 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1015 /* handle implication of PROT_EXEC by PROT_READ */
1016 if (current->personality & READ_IMPLIES_EXEC) {
1017 if (capabilities & NOMMU_MAP_EXEC)
1018 prot |= PROT_EXEC;
1019 }
1020 } else if ((prot & PROT_READ) &&
1021 (prot & PROT_EXEC) &&
1022 !(capabilities & NOMMU_MAP_EXEC)
1023 ) {
1024 /* backing file is not executable, try to copy */
1025 capabilities &= ~NOMMU_MAP_DIRECT;
1026 }
1027 } else {
1028 /* anonymous mappings are always memory backed and can be
1029 * privately mapped
1030 */
1031 capabilities = NOMMU_MAP_COPY;
1032
1033 /* handle PROT_EXEC implication by PROT_READ */
1034 if ((prot & PROT_READ) &&
1035 (current->personality & READ_IMPLIES_EXEC))
1036 prot |= PROT_EXEC;
1037 }
1038
1039 /* allow the security API to have its say */
1040 ret = security_mmap_addr(addr);
1041 if (ret < 0)
1042 return ret;
1043
1044 /* looks okay */
1045 *_capabilities = capabilities;
1046 return 0;
1047}
1048
1049/*
1050 * we've determined that we can make the mapping, now translate what we
1051 * now know into VMA flags
1052 */
1053static unsigned long determine_vm_flags(struct file *file,
1054 unsigned long prot,
1055 unsigned long flags,
1056 unsigned long capabilities)
1057{
1058 unsigned long vm_flags;
1059
1060 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1061 /* vm_flags |= mm->def_flags; */
1062
1063 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1064 /* attempt to share read-only copies of mapped file chunks */
1065 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1066 if (file && !(prot & PROT_WRITE))
1067 vm_flags |= VM_MAYSHARE;
1068 } else {
1069 /* overlay a shareable mapping on the backing device or inode
1070 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1071 * romfs/cramfs */
1072 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1073 if (flags & MAP_SHARED)
1074 vm_flags |= VM_SHARED;
1075 }
1076
1077 /* refuse to let anyone share private mappings with this process if
1078 * it's being traced - otherwise breakpoints set in it may interfere
1079 * with another untraced process
1080 */
1081 if ((flags & MAP_PRIVATE) && current->ptrace)
1082 vm_flags &= ~VM_MAYSHARE;
1083
1084 return vm_flags;
1085}
1086
1087/*
1088 * set up a shared mapping on a file (the driver or filesystem provides and
1089 * pins the storage)
1090 */
1091static int do_mmap_shared_file(struct vm_area_struct *vma)
1092{
1093 int ret;
1094
1095 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1096 if (ret == 0) {
1097 vma->vm_region->vm_top = vma->vm_region->vm_end;
1098 return 0;
1099 }
1100 if (ret != -ENOSYS)
1101 return ret;
1102
1103 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1104 * opposed to tried but failed) so we can only give a suitable error as
1105 * it's not possible to make a private copy if MAP_SHARED was given */
1106 return -ENODEV;
1107}
1108
1109/*
1110 * set up a private mapping or an anonymous shared mapping
1111 */
1112static int do_mmap_private(struct vm_area_struct *vma,
1113 struct vm_region *region,
1114 unsigned long len,
1115 unsigned long capabilities)
1116{
1117 unsigned long total, point;
1118 void *base;
1119 int ret, order;
1120
1121 /* invoke the file's mapping function so that it can keep track of
1122 * shared mappings on devices or memory
1123 * - VM_MAYSHARE will be set if it may attempt to share
1124 */
1125 if (capabilities & NOMMU_MAP_DIRECT) {
1126 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1127 if (ret == 0) {
1128 /* shouldn't return success if we're not sharing */
1129 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1130 vma->vm_region->vm_top = vma->vm_region->vm_end;
1131 return 0;
1132 }
1133 if (ret != -ENOSYS)
1134 return ret;
1135
1136 /* getting an ENOSYS error indicates that direct mmap isn't
1137 * possible (as opposed to tried but failed) so we'll try to
1138 * make a private copy of the data and map that instead */
1139 }
1140
1141
1142 /* allocate some memory to hold the mapping
1143 * - note that this may not return a page-aligned address if the object
1144 * we're allocating is smaller than a page
1145 */
1146 order = get_order(len);
1147 total = 1 << order;
1148 point = len >> PAGE_SHIFT;
1149
1150 /* we don't want to allocate a power-of-2 sized page set */
1151 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1152 total = point;
1153
1154 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1155 if (!base)
1156 goto enomem;
1157
1158 atomic_long_add(total, &mmap_pages_allocated);
1159
1160 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1161 region->vm_start = (unsigned long) base;
1162 region->vm_end = region->vm_start + len;
1163 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1164
1165 vma->vm_start = region->vm_start;
1166 vma->vm_end = region->vm_start + len;
1167
1168 if (vma->vm_file) {
1169 /* read the contents of a file into the copy */
1170 mm_segment_t old_fs;
1171 loff_t fpos;
1172
1173 fpos = vma->vm_pgoff;
1174 fpos <<= PAGE_SHIFT;
1175
1176 old_fs = get_fs();
1177 set_fs(KERNEL_DS);
1178 ret = __vfs_read(vma->vm_file, base, len, &fpos);
1179 set_fs(old_fs);
1180
1181 if (ret < 0)
1182 goto error_free;
1183
1184 /* clear the last little bit */
1185 if (ret < len)
1186 memset(base + ret, 0, len - ret);
1187
1188 }
1189
1190 return 0;
1191
1192error_free:
1193 free_page_series(region->vm_start, region->vm_top);
1194 region->vm_start = vma->vm_start = 0;
1195 region->vm_end = vma->vm_end = 0;
1196 region->vm_top = 0;
1197 return ret;
1198
1199enomem:
1200 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1201 len, current->pid, current->comm);
1202 show_free_areas(0);
1203 return -ENOMEM;
1204}
1205
1206/*
1207 * handle mapping creation for uClinux
1208 */
1209unsigned long do_mmap(struct file *file,
1210 unsigned long addr,
1211 unsigned long len,
1212 unsigned long prot,
1213 unsigned long flags,
1214 vm_flags_t vm_flags,
1215 unsigned long pgoff,
1216 unsigned long *populate)
1217{
1218 struct vm_area_struct *vma;
1219 struct vm_region *region;
1220 struct rb_node *rb;
1221 unsigned long capabilities, result;
1222 int ret;
1223
1224 *populate = 0;
1225
1226 /* decide whether we should attempt the mapping, and if so what sort of
1227 * mapping */
1228 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1229 &capabilities);
1230 if (ret < 0)
1231 return ret;
1232
1233 /* we ignore the address hint */
1234 addr = 0;
1235 len = PAGE_ALIGN(len);
1236
1237 /* we've determined that we can make the mapping, now translate what we
1238 * now know into VMA flags */
1239 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1240
1241 /* we're going to need to record the mapping */
1242 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1243 if (!region)
1244 goto error_getting_region;
1245
1246 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1247 if (!vma)
1248 goto error_getting_vma;
1249
1250 region->vm_usage = 1;
1251 region->vm_flags = vm_flags;
1252 region->vm_pgoff = pgoff;
1253
1254 INIT_LIST_HEAD(&vma->anon_vma_chain);
1255 vma->vm_flags = vm_flags;
1256 vma->vm_pgoff = pgoff;
1257
1258 if (file) {
1259 region->vm_file = get_file(file);
1260 vma->vm_file = get_file(file);
1261 }
1262
1263 down_write(&nommu_region_sem);
1264
1265 /* if we want to share, we need to check for regions created by other
1266 * mmap() calls that overlap with our proposed mapping
1267 * - we can only share with a superset match on most regular files
1268 * - shared mappings on character devices and memory backed files are
1269 * permitted to overlap inexactly as far as we are concerned for in
1270 * these cases, sharing is handled in the driver or filesystem rather
1271 * than here
1272 */
1273 if (vm_flags & VM_MAYSHARE) {
1274 struct vm_region *pregion;
1275 unsigned long pglen, rpglen, pgend, rpgend, start;
1276
1277 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1278 pgend = pgoff + pglen;
1279
1280 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1281 pregion = rb_entry(rb, struct vm_region, vm_rb);
1282
1283 if (!(pregion->vm_flags & VM_MAYSHARE))
1284 continue;
1285
1286 /* search for overlapping mappings on the same file */
1287 if (file_inode(pregion->vm_file) !=
1288 file_inode(file))
1289 continue;
1290
1291 if (pregion->vm_pgoff >= pgend)
1292 continue;
1293
1294 rpglen = pregion->vm_end - pregion->vm_start;
1295 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1296 rpgend = pregion->vm_pgoff + rpglen;
1297 if (pgoff >= rpgend)
1298 continue;
1299
1300 /* handle inexactly overlapping matches between
1301 * mappings */
1302 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1303 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1304 /* new mapping is not a subset of the region */
1305 if (!(capabilities & NOMMU_MAP_DIRECT))
1306 goto sharing_violation;
1307 continue;
1308 }
1309
1310 /* we've found a region we can share */
1311 pregion->vm_usage++;
1312 vma->vm_region = pregion;
1313 start = pregion->vm_start;
1314 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1315 vma->vm_start = start;
1316 vma->vm_end = start + len;
1317
1318 if (pregion->vm_flags & VM_MAPPED_COPY)
1319 vma->vm_flags |= VM_MAPPED_COPY;
1320 else {
1321 ret = do_mmap_shared_file(vma);
1322 if (ret < 0) {
1323 vma->vm_region = NULL;
1324 vma->vm_start = 0;
1325 vma->vm_end = 0;
1326 pregion->vm_usage--;
1327 pregion = NULL;
1328 goto error_just_free;
1329 }
1330 }
1331 fput(region->vm_file);
1332 kmem_cache_free(vm_region_jar, region);
1333 region = pregion;
1334 result = start;
1335 goto share;
1336 }
1337
1338 /* obtain the address at which to make a shared mapping
1339 * - this is the hook for quasi-memory character devices to
1340 * tell us the location of a shared mapping
1341 */
1342 if (capabilities & NOMMU_MAP_DIRECT) {
1343 addr = file->f_op->get_unmapped_area(file, addr, len,
1344 pgoff, flags);
1345 if (IS_ERR_VALUE(addr)) {
1346 ret = addr;
1347 if (ret != -ENOSYS)
1348 goto error_just_free;
1349
1350 /* the driver refused to tell us where to site
1351 * the mapping so we'll have to attempt to copy
1352 * it */
1353 ret = -ENODEV;
1354 if (!(capabilities & NOMMU_MAP_COPY))
1355 goto error_just_free;
1356
1357 capabilities &= ~NOMMU_MAP_DIRECT;
1358 } else {
1359 vma->vm_start = region->vm_start = addr;
1360 vma->vm_end = region->vm_end = addr + len;
1361 }
1362 }
1363 }
1364
1365 vma->vm_region = region;
1366
1367 /* set up the mapping
1368 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1369 */
1370 if (file && vma->vm_flags & VM_SHARED)
1371 ret = do_mmap_shared_file(vma);
1372 else
1373 ret = do_mmap_private(vma, region, len, capabilities);
1374 if (ret < 0)
1375 goto error_just_free;
1376 add_nommu_region(region);
1377
1378 /* clear anonymous mappings that don't ask for uninitialized data */
1379 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1380 memset((void *)region->vm_start, 0,
1381 region->vm_end - region->vm_start);
1382
1383 /* okay... we have a mapping; now we have to register it */
1384 result = vma->vm_start;
1385
1386 current->mm->total_vm += len >> PAGE_SHIFT;
1387
1388share:
1389 add_vma_to_mm(current->mm, vma);
1390
1391 /* we flush the region from the icache only when the first executable
1392 * mapping of it is made */
1393 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1394 flush_icache_range(region->vm_start, region->vm_end);
1395 region->vm_icache_flushed = true;
1396 }
1397
1398 up_write(&nommu_region_sem);
1399
1400 return result;
1401
1402error_just_free:
1403 up_write(&nommu_region_sem);
1404error:
1405 if (region->vm_file)
1406 fput(region->vm_file);
1407 kmem_cache_free(vm_region_jar, region);
1408 if (vma->vm_file)
1409 fput(vma->vm_file);
1410 kmem_cache_free(vm_area_cachep, vma);
1411 return ret;
1412
1413sharing_violation:
1414 up_write(&nommu_region_sem);
1415 pr_warn("Attempt to share mismatched mappings\n");
1416 ret = -EINVAL;
1417 goto error;
1418
1419error_getting_vma:
1420 kmem_cache_free(vm_region_jar, region);
1421 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1422 len, current->pid);
1423 show_free_areas(0);
1424 return -ENOMEM;
1425
1426error_getting_region:
1427 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1428 len, current->pid);
1429 show_free_areas(0);
1430 return -ENOMEM;
1431}
1432
1433SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1434 unsigned long, prot, unsigned long, flags,
1435 unsigned long, fd, unsigned long, pgoff)
1436{
1437 struct file *file = NULL;
1438 unsigned long retval = -EBADF;
1439
1440 audit_mmap_fd(fd, flags);
1441 if (!(flags & MAP_ANONYMOUS)) {
1442 file = fget(fd);
1443 if (!file)
1444 goto out;
1445 }
1446
1447 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1448
1449 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1450
1451 if (file)
1452 fput(file);
1453out:
1454 return retval;
1455}
1456
1457#ifdef __ARCH_WANT_SYS_OLD_MMAP
1458struct mmap_arg_struct {
1459 unsigned long addr;
1460 unsigned long len;
1461 unsigned long prot;
1462 unsigned long flags;
1463 unsigned long fd;
1464 unsigned long offset;
1465};
1466
1467SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1468{
1469 struct mmap_arg_struct a;
1470
1471 if (copy_from_user(&a, arg, sizeof(a)))
1472 return -EFAULT;
1473 if (offset_in_page(a.offset))
1474 return -EINVAL;
1475
1476 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1477 a.offset >> PAGE_SHIFT);
1478}
1479#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1480
1481/*
1482 * split a vma into two pieces at address 'addr', a new vma is allocated either
1483 * for the first part or the tail.
1484 */
1485int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1486 unsigned long addr, int new_below)
1487{
1488 struct vm_area_struct *new;
1489 struct vm_region *region;
1490 unsigned long npages;
1491
1492 /* we're only permitted to split anonymous regions (these should have
1493 * only a single usage on the region) */
1494 if (vma->vm_file)
1495 return -ENOMEM;
1496
1497 if (mm->map_count >= sysctl_max_map_count)
1498 return -ENOMEM;
1499
1500 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1501 if (!region)
1502 return -ENOMEM;
1503
1504 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1505 if (!new) {
1506 kmem_cache_free(vm_region_jar, region);
1507 return -ENOMEM;
1508 }
1509
1510 /* most fields are the same, copy all, and then fixup */
1511 *new = *vma;
1512 *region = *vma->vm_region;
1513 new->vm_region = region;
1514
1515 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1516
1517 if (new_below) {
1518 region->vm_top = region->vm_end = new->vm_end = addr;
1519 } else {
1520 region->vm_start = new->vm_start = addr;
1521 region->vm_pgoff = new->vm_pgoff += npages;
1522 }
1523
1524 if (new->vm_ops && new->vm_ops->open)
1525 new->vm_ops->open(new);
1526
1527 delete_vma_from_mm(vma);
1528 down_write(&nommu_region_sem);
1529 delete_nommu_region(vma->vm_region);
1530 if (new_below) {
1531 vma->vm_region->vm_start = vma->vm_start = addr;
1532 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1533 } else {
1534 vma->vm_region->vm_end = vma->vm_end = addr;
1535 vma->vm_region->vm_top = addr;
1536 }
1537 add_nommu_region(vma->vm_region);
1538 add_nommu_region(new->vm_region);
1539 up_write(&nommu_region_sem);
1540 add_vma_to_mm(mm, vma);
1541 add_vma_to_mm(mm, new);
1542 return 0;
1543}
1544
1545/*
1546 * shrink a VMA by removing the specified chunk from either the beginning or
1547 * the end
1548 */
1549static int shrink_vma(struct mm_struct *mm,
1550 struct vm_area_struct *vma,
1551 unsigned long from, unsigned long to)
1552{
1553 struct vm_region *region;
1554
1555 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1556 * and list */
1557 delete_vma_from_mm(vma);
1558 if (from > vma->vm_start)
1559 vma->vm_end = from;
1560 else
1561 vma->vm_start = to;
1562 add_vma_to_mm(mm, vma);
1563
1564 /* cut the backing region down to size */
1565 region = vma->vm_region;
1566 BUG_ON(region->vm_usage != 1);
1567
1568 down_write(&nommu_region_sem);
1569 delete_nommu_region(region);
1570 if (from > region->vm_start) {
1571 to = region->vm_top;
1572 region->vm_top = region->vm_end = from;
1573 } else {
1574 region->vm_start = to;
1575 }
1576 add_nommu_region(region);
1577 up_write(&nommu_region_sem);
1578
1579 free_page_series(from, to);
1580 return 0;
1581}
1582
1583/*
1584 * release a mapping
1585 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1586 * VMA, though it need not cover the whole VMA
1587 */
1588int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1589{
1590 struct vm_area_struct *vma;
1591 unsigned long end;
1592 int ret;
1593
1594 len = PAGE_ALIGN(len);
1595 if (len == 0)
1596 return -EINVAL;
1597
1598 end = start + len;
1599
1600 /* find the first potentially overlapping VMA */
1601 vma = find_vma(mm, start);
1602 if (!vma) {
1603 static int limit;
1604 if (limit < 5) {
1605 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1606 current->pid, current->comm,
1607 start, start + len - 1);
1608 limit++;
1609 }
1610 return -EINVAL;
1611 }
1612
1613 /* we're allowed to split an anonymous VMA but not a file-backed one */
1614 if (vma->vm_file) {
1615 do {
1616 if (start > vma->vm_start)
1617 return -EINVAL;
1618 if (end == vma->vm_end)
1619 goto erase_whole_vma;
1620 vma = vma->vm_next;
1621 } while (vma);
1622 return -EINVAL;
1623 } else {
1624 /* the chunk must be a subset of the VMA found */
1625 if (start == vma->vm_start && end == vma->vm_end)
1626 goto erase_whole_vma;
1627 if (start < vma->vm_start || end > vma->vm_end)
1628 return -EINVAL;
1629 if (offset_in_page(start))
1630 return -EINVAL;
1631 if (end != vma->vm_end && offset_in_page(end))
1632 return -EINVAL;
1633 if (start != vma->vm_start && end != vma->vm_end) {
1634 ret = split_vma(mm, vma, start, 1);
1635 if (ret < 0)
1636 return ret;
1637 }
1638 return shrink_vma(mm, vma, start, end);
1639 }
1640
1641erase_whole_vma:
1642 delete_vma_from_mm(vma);
1643 delete_vma(mm, vma);
1644 return 0;
1645}
1646EXPORT_SYMBOL(do_munmap);
1647
1648int vm_munmap(unsigned long addr, size_t len)
1649{
1650 struct mm_struct *mm = current->mm;
1651 int ret;
1652
1653 down_write(&mm->mmap_sem);
1654 ret = do_munmap(mm, addr, len);
1655 up_write(&mm->mmap_sem);
1656 return ret;
1657}
1658EXPORT_SYMBOL(vm_munmap);
1659
1660SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1661{
1662 return vm_munmap(addr, len);
1663}
1664
1665/*
1666 * release all the mappings made in a process's VM space
1667 */
1668void exit_mmap(struct mm_struct *mm)
1669{
1670 struct vm_area_struct *vma;
1671
1672 if (!mm)
1673 return;
1674
1675 mm->total_vm = 0;
1676
1677 while ((vma = mm->mmap)) {
1678 mm->mmap = vma->vm_next;
1679 delete_vma_from_mm(vma);
1680 delete_vma(mm, vma);
1681 cond_resched();
1682 }
1683}
1684
1685unsigned long vm_brk(unsigned long addr, unsigned long len)
1686{
1687 return -ENOMEM;
1688}
1689
1690/*
1691 * expand (or shrink) an existing mapping, potentially moving it at the same
1692 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1693 *
1694 * under NOMMU conditions, we only permit changing a mapping's size, and only
1695 * as long as it stays within the region allocated by do_mmap_private() and the
1696 * block is not shareable
1697 *
1698 * MREMAP_FIXED is not supported under NOMMU conditions
1699 */
1700static unsigned long do_mremap(unsigned long addr,
1701 unsigned long old_len, unsigned long new_len,
1702 unsigned long flags, unsigned long new_addr)
1703{
1704 struct vm_area_struct *vma;
1705
1706 /* insanity checks first */
1707 old_len = PAGE_ALIGN(old_len);
1708 new_len = PAGE_ALIGN(new_len);
1709 if (old_len == 0 || new_len == 0)
1710 return (unsigned long) -EINVAL;
1711
1712 if (offset_in_page(addr))
1713 return -EINVAL;
1714
1715 if (flags & MREMAP_FIXED && new_addr != addr)
1716 return (unsigned long) -EINVAL;
1717
1718 vma = find_vma_exact(current->mm, addr, old_len);
1719 if (!vma)
1720 return (unsigned long) -EINVAL;
1721
1722 if (vma->vm_end != vma->vm_start + old_len)
1723 return (unsigned long) -EFAULT;
1724
1725 if (vma->vm_flags & VM_MAYSHARE)
1726 return (unsigned long) -EPERM;
1727
1728 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1729 return (unsigned long) -ENOMEM;
1730
1731 /* all checks complete - do it */
1732 vma->vm_end = vma->vm_start + new_len;
1733 return vma->vm_start;
1734}
1735
1736SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1737 unsigned long, new_len, unsigned long, flags,
1738 unsigned long, new_addr)
1739{
1740 unsigned long ret;
1741
1742 down_write(¤t->mm->mmap_sem);
1743 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1744 up_write(¤t->mm->mmap_sem);
1745 return ret;
1746}
1747
1748struct page *follow_page_mask(struct vm_area_struct *vma,
1749 unsigned long address, unsigned int flags,
1750 unsigned int *page_mask)
1751{
1752 *page_mask = 0;
1753 return NULL;
1754}
1755
1756int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1757 unsigned long pfn, unsigned long size, pgprot_t prot)
1758{
1759 if (addr != (pfn << PAGE_SHIFT))
1760 return -EINVAL;
1761
1762 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1763 return 0;
1764}
1765EXPORT_SYMBOL(remap_pfn_range);
1766
1767int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1768{
1769 unsigned long pfn = start >> PAGE_SHIFT;
1770 unsigned long vm_len = vma->vm_end - vma->vm_start;
1771
1772 pfn += vma->vm_pgoff;
1773 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1774}
1775EXPORT_SYMBOL(vm_iomap_memory);
1776
1777int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1778 unsigned long pgoff)
1779{
1780 unsigned int size = vma->vm_end - vma->vm_start;
1781
1782 if (!(vma->vm_flags & VM_USERMAP))
1783 return -EINVAL;
1784
1785 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1786 vma->vm_end = vma->vm_start + size;
1787
1788 return 0;
1789}
1790EXPORT_SYMBOL(remap_vmalloc_range);
1791
1792unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1793 unsigned long len, unsigned long pgoff, unsigned long flags)
1794{
1795 return -ENOMEM;
1796}
1797
1798void unmap_mapping_range(struct address_space *mapping,
1799 loff_t const holebegin, loff_t const holelen,
1800 int even_cows)
1801{
1802}
1803EXPORT_SYMBOL(unmap_mapping_range);
1804
1805int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1806{
1807 BUG();
1808 return 0;
1809}
1810EXPORT_SYMBOL(filemap_fault);
1811
1812void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1813{
1814 BUG();
1815}
1816EXPORT_SYMBOL(filemap_map_pages);
1817
1818static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1819 unsigned long addr, void *buf, int len, int write)
1820{
1821 struct vm_area_struct *vma;
1822
1823 down_read(&mm->mmap_sem);
1824
1825 /* the access must start within one of the target process's mappings */
1826 vma = find_vma(mm, addr);
1827 if (vma) {
1828 /* don't overrun this mapping */
1829 if (addr + len >= vma->vm_end)
1830 len = vma->vm_end - addr;
1831
1832 /* only read or write mappings where it is permitted */
1833 if (write && vma->vm_flags & VM_MAYWRITE)
1834 copy_to_user_page(vma, NULL, addr,
1835 (void *) addr, buf, len);
1836 else if (!write && vma->vm_flags & VM_MAYREAD)
1837 copy_from_user_page(vma, NULL, addr,
1838 buf, (void *) addr, len);
1839 else
1840 len = 0;
1841 } else {
1842 len = 0;
1843 }
1844
1845 up_read(&mm->mmap_sem);
1846
1847 return len;
1848}
1849
1850/**
1851 * @access_remote_vm - access another process' address space
1852 * @mm: the mm_struct of the target address space
1853 * @addr: start address to access
1854 * @buf: source or destination buffer
1855 * @len: number of bytes to transfer
1856 * @write: whether the access is a write
1857 *
1858 * The caller must hold a reference on @mm.
1859 */
1860int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1861 void *buf, int len, int write)
1862{
1863 return __access_remote_vm(NULL, mm, addr, buf, len, write);
1864}
1865
1866/*
1867 * Access another process' address space.
1868 * - source/target buffer must be kernel space
1869 */
1870int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1871{
1872 struct mm_struct *mm;
1873
1874 if (addr + len < addr)
1875 return 0;
1876
1877 mm = get_task_mm(tsk);
1878 if (!mm)
1879 return 0;
1880
1881 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
1882
1883 mmput(mm);
1884 return len;
1885}
1886
1887/**
1888 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1889 * @inode: The inode to check
1890 * @size: The current filesize of the inode
1891 * @newsize: The proposed filesize of the inode
1892 *
1893 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1894 * make sure that that any outstanding VMAs aren't broken and then shrink the
1895 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1896 * automatically grant mappings that are too large.
1897 */
1898int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1899 size_t newsize)
1900{
1901 struct vm_area_struct *vma;
1902 struct vm_region *region;
1903 pgoff_t low, high;
1904 size_t r_size, r_top;
1905
1906 low = newsize >> PAGE_SHIFT;
1907 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1908
1909 down_write(&nommu_region_sem);
1910 i_mmap_lock_read(inode->i_mapping);
1911
1912 /* search for VMAs that fall within the dead zone */
1913 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1914 /* found one - only interested if it's shared out of the page
1915 * cache */
1916 if (vma->vm_flags & VM_SHARED) {
1917 i_mmap_unlock_read(inode->i_mapping);
1918 up_write(&nommu_region_sem);
1919 return -ETXTBSY; /* not quite true, but near enough */
1920 }
1921 }
1922
1923 /* reduce any regions that overlap the dead zone - if in existence,
1924 * these will be pointed to by VMAs that don't overlap the dead zone
1925 *
1926 * we don't check for any regions that start beyond the EOF as there
1927 * shouldn't be any
1928 */
1929 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1930 if (!(vma->vm_flags & VM_SHARED))
1931 continue;
1932
1933 region = vma->vm_region;
1934 r_size = region->vm_top - region->vm_start;
1935 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1936
1937 if (r_top > newsize) {
1938 region->vm_top -= r_top - newsize;
1939 if (region->vm_end > region->vm_top)
1940 region->vm_end = region->vm_top;
1941 }
1942 }
1943
1944 i_mmap_unlock_read(inode->i_mapping);
1945 up_write(&nommu_region_sem);
1946 return 0;
1947}
1948
1949/*
1950 * Initialise sysctl_user_reserve_kbytes.
1951 *
1952 * This is intended to prevent a user from starting a single memory hogging
1953 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1954 * mode.
1955 *
1956 * The default value is min(3% of free memory, 128MB)
1957 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1958 */
1959static int __meminit init_user_reserve(void)
1960{
1961 unsigned long free_kbytes;
1962
1963 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1964
1965 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1966 return 0;
1967}
1968subsys_initcall(init_user_reserve);
1969
1970/*
1971 * Initialise sysctl_admin_reserve_kbytes.
1972 *
1973 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1974 * to log in and kill a memory hogging process.
1975 *
1976 * Systems with more than 256MB will reserve 8MB, enough to recover
1977 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1978 * only reserve 3% of free pages by default.
1979 */
1980static int __meminit init_admin_reserve(void)
1981{
1982 unsigned long free_kbytes;
1983
1984 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1985
1986 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1987 return 0;
1988}
1989subsys_initcall(init_admin_reserve);
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#include <linux/export.h>
17#include <linux/mm.h>
18#include <linux/vmacache.h>
19#include <linux/mman.h>
20#include <linux/swap.h>
21#include <linux/file.h>
22#include <linux/highmem.h>
23#include <linux/pagemap.h>
24#include <linux/slab.h>
25#include <linux/vmalloc.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/compiler.h>
29#include <linux/mount.h>
30#include <linux/personality.h>
31#include <linux/security.h>
32#include <linux/syscalls.h>
33#include <linux/audit.h>
34#include <linux/sched/sysctl.h>
35
36#include <asm/uaccess.h>
37#include <asm/tlb.h>
38#include <asm/tlbflush.h>
39#include <asm/mmu_context.h>
40#include "internal.h"
41
42#if 0
43#define kenter(FMT, ...) \
44 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
45#define kleave(FMT, ...) \
46 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
47#define kdebug(FMT, ...) \
48 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
49#else
50#define kenter(FMT, ...) \
51 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
52#define kleave(FMT, ...) \
53 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
54#define kdebug(FMT, ...) \
55 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
56#endif
57
58void *high_memory;
59struct page *mem_map;
60unsigned long max_mapnr;
61unsigned long highest_memmap_pfn;
62struct percpu_counter vm_committed_as;
63int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
64int sysctl_overcommit_ratio = 50; /* default is 50% */
65unsigned long sysctl_overcommit_kbytes __read_mostly;
66int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
67int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
68unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
69unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
70int heap_stack_gap = 0;
71
72atomic_long_t mmap_pages_allocated;
73
74/*
75 * The global memory commitment made in the system can be a metric
76 * that can be used to drive ballooning decisions when Linux is hosted
77 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
78 * balancing memory across competing virtual machines that are hosted.
79 * Several metrics drive this policy engine including the guest reported
80 * memory commitment.
81 */
82unsigned long vm_memory_committed(void)
83{
84 return percpu_counter_read_positive(&vm_committed_as);
85}
86
87EXPORT_SYMBOL_GPL(vm_memory_committed);
88
89EXPORT_SYMBOL(mem_map);
90
91/* list of mapped, potentially shareable regions */
92static struct kmem_cache *vm_region_jar;
93struct rb_root nommu_region_tree = RB_ROOT;
94DECLARE_RWSEM(nommu_region_sem);
95
96const struct vm_operations_struct generic_file_vm_ops = {
97};
98
99/*
100 * Return the total memory allocated for this pointer, not
101 * just what the caller asked for.
102 *
103 * Doesn't have to be accurate, i.e. may have races.
104 */
105unsigned int kobjsize(const void *objp)
106{
107 struct page *page;
108
109 /*
110 * If the object we have should not have ksize performed on it,
111 * return size of 0
112 */
113 if (!objp || !virt_addr_valid(objp))
114 return 0;
115
116 page = virt_to_head_page(objp);
117
118 /*
119 * If the allocator sets PageSlab, we know the pointer came from
120 * kmalloc().
121 */
122 if (PageSlab(page))
123 return ksize(objp);
124
125 /*
126 * If it's not a compound page, see if we have a matching VMA
127 * region. This test is intentionally done in reverse order,
128 * so if there's no VMA, we still fall through and hand back
129 * PAGE_SIZE for 0-order pages.
130 */
131 if (!PageCompound(page)) {
132 struct vm_area_struct *vma;
133
134 vma = find_vma(current->mm, (unsigned long)objp);
135 if (vma)
136 return vma->vm_end - vma->vm_start;
137 }
138
139 /*
140 * The ksize() function is only guaranteed to work for pointers
141 * returned by kmalloc(). So handle arbitrary pointers here.
142 */
143 return PAGE_SIZE << compound_order(page);
144}
145
146long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
147 unsigned long start, unsigned long nr_pages,
148 unsigned int foll_flags, struct page **pages,
149 struct vm_area_struct **vmas, int *nonblocking)
150{
151 struct vm_area_struct *vma;
152 unsigned long vm_flags;
153 int i;
154
155 /* calculate required read or write permissions.
156 * If FOLL_FORCE is set, we only require the "MAY" flags.
157 */
158 vm_flags = (foll_flags & FOLL_WRITE) ?
159 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
160 vm_flags &= (foll_flags & FOLL_FORCE) ?
161 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
162
163 for (i = 0; i < nr_pages; i++) {
164 vma = find_vma(mm, start);
165 if (!vma)
166 goto finish_or_fault;
167
168 /* protect what we can, including chardevs */
169 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
170 !(vm_flags & vma->vm_flags))
171 goto finish_or_fault;
172
173 if (pages) {
174 pages[i] = virt_to_page(start);
175 if (pages[i])
176 page_cache_get(pages[i]);
177 }
178 if (vmas)
179 vmas[i] = vma;
180 start = (start + PAGE_SIZE) & PAGE_MASK;
181 }
182
183 return i;
184
185finish_or_fault:
186 return i ? : -EFAULT;
187}
188
189/*
190 * get a list of pages in an address range belonging to the specified process
191 * and indicate the VMA that covers each page
192 * - this is potentially dodgy as we may end incrementing the page count of a
193 * slab page or a secondary page from a compound page
194 * - don't permit access to VMAs that don't support it, such as I/O mappings
195 */
196long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
197 unsigned long start, unsigned long nr_pages,
198 int write, int force, struct page **pages,
199 struct vm_area_struct **vmas)
200{
201 int flags = 0;
202
203 if (write)
204 flags |= FOLL_WRITE;
205 if (force)
206 flags |= FOLL_FORCE;
207
208 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
209 NULL);
210}
211EXPORT_SYMBOL(get_user_pages);
212
213/**
214 * follow_pfn - look up PFN at a user virtual address
215 * @vma: memory mapping
216 * @address: user virtual address
217 * @pfn: location to store found PFN
218 *
219 * Only IO mappings and raw PFN mappings are allowed.
220 *
221 * Returns zero and the pfn at @pfn on success, -ve otherwise.
222 */
223int follow_pfn(struct vm_area_struct *vma, unsigned long address,
224 unsigned long *pfn)
225{
226 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
227 return -EINVAL;
228
229 *pfn = address >> PAGE_SHIFT;
230 return 0;
231}
232EXPORT_SYMBOL(follow_pfn);
233
234LIST_HEAD(vmap_area_list);
235
236void vfree(const void *addr)
237{
238 kfree(addr);
239}
240EXPORT_SYMBOL(vfree);
241
242void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
243{
244 /*
245 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
246 * returns only a logical address.
247 */
248 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
249}
250EXPORT_SYMBOL(__vmalloc);
251
252void *vmalloc_user(unsigned long size)
253{
254 void *ret;
255
256 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
257 PAGE_KERNEL);
258 if (ret) {
259 struct vm_area_struct *vma;
260
261 down_write(¤t->mm->mmap_sem);
262 vma = find_vma(current->mm, (unsigned long)ret);
263 if (vma)
264 vma->vm_flags |= VM_USERMAP;
265 up_write(¤t->mm->mmap_sem);
266 }
267
268 return ret;
269}
270EXPORT_SYMBOL(vmalloc_user);
271
272struct page *vmalloc_to_page(const void *addr)
273{
274 return virt_to_page(addr);
275}
276EXPORT_SYMBOL(vmalloc_to_page);
277
278unsigned long vmalloc_to_pfn(const void *addr)
279{
280 return page_to_pfn(virt_to_page(addr));
281}
282EXPORT_SYMBOL(vmalloc_to_pfn);
283
284long vread(char *buf, char *addr, unsigned long count)
285{
286 /* Don't allow overflow */
287 if ((unsigned long) buf + count < count)
288 count = -(unsigned long) buf;
289
290 memcpy(buf, addr, count);
291 return count;
292}
293
294long vwrite(char *buf, char *addr, unsigned long count)
295{
296 /* Don't allow overflow */
297 if ((unsigned long) addr + count < count)
298 count = -(unsigned long) addr;
299
300 memcpy(addr, buf, count);
301 return count;
302}
303
304/*
305 * vmalloc - allocate virtually continguos memory
306 *
307 * @size: allocation size
308 *
309 * Allocate enough pages to cover @size from the page level
310 * allocator and map them into continguos kernel virtual space.
311 *
312 * For tight control over page level allocator and protection flags
313 * use __vmalloc() instead.
314 */
315void *vmalloc(unsigned long size)
316{
317 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
318}
319EXPORT_SYMBOL(vmalloc);
320
321/*
322 * vzalloc - allocate virtually continguos memory with zero fill
323 *
324 * @size: allocation size
325 *
326 * Allocate enough pages to cover @size from the page level
327 * allocator and map them into continguos kernel virtual space.
328 * The memory allocated is set to zero.
329 *
330 * For tight control over page level allocator and protection flags
331 * use __vmalloc() instead.
332 */
333void *vzalloc(unsigned long size)
334{
335 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
336 PAGE_KERNEL);
337}
338EXPORT_SYMBOL(vzalloc);
339
340/**
341 * vmalloc_node - allocate memory on a specific node
342 * @size: allocation size
343 * @node: numa node
344 *
345 * Allocate enough pages to cover @size from the page level
346 * allocator and map them into contiguous kernel virtual space.
347 *
348 * For tight control over page level allocator and protection flags
349 * use __vmalloc() instead.
350 */
351void *vmalloc_node(unsigned long size, int node)
352{
353 return vmalloc(size);
354}
355EXPORT_SYMBOL(vmalloc_node);
356
357/**
358 * vzalloc_node - allocate memory on a specific node with zero fill
359 * @size: allocation size
360 * @node: numa node
361 *
362 * Allocate enough pages to cover @size from the page level
363 * allocator and map them into contiguous kernel virtual space.
364 * The memory allocated is set to zero.
365 *
366 * For tight control over page level allocator and protection flags
367 * use __vmalloc() instead.
368 */
369void *vzalloc_node(unsigned long size, int node)
370{
371 return vzalloc(size);
372}
373EXPORT_SYMBOL(vzalloc_node);
374
375#ifndef PAGE_KERNEL_EXEC
376# define PAGE_KERNEL_EXEC PAGE_KERNEL
377#endif
378
379/**
380 * vmalloc_exec - allocate virtually contiguous, executable memory
381 * @size: allocation size
382 *
383 * Kernel-internal function to allocate enough pages to cover @size
384 * the page level allocator and map them into contiguous and
385 * executable kernel virtual space.
386 *
387 * For tight control over page level allocator and protection flags
388 * use __vmalloc() instead.
389 */
390
391void *vmalloc_exec(unsigned long size)
392{
393 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
394}
395
396/**
397 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
398 * @size: allocation size
399 *
400 * Allocate enough 32bit PA addressable pages to cover @size from the
401 * page level allocator and map them into continguos kernel virtual space.
402 */
403void *vmalloc_32(unsigned long size)
404{
405 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
406}
407EXPORT_SYMBOL(vmalloc_32);
408
409/**
410 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
411 * @size: allocation size
412 *
413 * The resulting memory area is 32bit addressable and zeroed so it can be
414 * mapped to userspace without leaking data.
415 *
416 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
417 * remap_vmalloc_range() are permissible.
418 */
419void *vmalloc_32_user(unsigned long size)
420{
421 /*
422 * We'll have to sort out the ZONE_DMA bits for 64-bit,
423 * but for now this can simply use vmalloc_user() directly.
424 */
425 return vmalloc_user(size);
426}
427EXPORT_SYMBOL(vmalloc_32_user);
428
429void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
430{
431 BUG();
432 return NULL;
433}
434EXPORT_SYMBOL(vmap);
435
436void vunmap(const void *addr)
437{
438 BUG();
439}
440EXPORT_SYMBOL(vunmap);
441
442void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
443{
444 BUG();
445 return NULL;
446}
447EXPORT_SYMBOL(vm_map_ram);
448
449void vm_unmap_ram(const void *mem, unsigned int count)
450{
451 BUG();
452}
453EXPORT_SYMBOL(vm_unmap_ram);
454
455void vm_unmap_aliases(void)
456{
457}
458EXPORT_SYMBOL_GPL(vm_unmap_aliases);
459
460/*
461 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
462 * have one.
463 */
464void __weak vmalloc_sync_all(void)
465{
466}
467
468/**
469 * alloc_vm_area - allocate a range of kernel address space
470 * @size: size of the area
471 *
472 * Returns: NULL on failure, vm_struct on success
473 *
474 * This function reserves a range of kernel address space, and
475 * allocates pagetables to map that range. No actual mappings
476 * are created. If the kernel address space is not shared
477 * between processes, it syncs the pagetable across all
478 * processes.
479 */
480struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
481{
482 BUG();
483 return NULL;
484}
485EXPORT_SYMBOL_GPL(alloc_vm_area);
486
487void free_vm_area(struct vm_struct *area)
488{
489 BUG();
490}
491EXPORT_SYMBOL_GPL(free_vm_area);
492
493int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
494 struct page *page)
495{
496 return -EINVAL;
497}
498EXPORT_SYMBOL(vm_insert_page);
499
500/*
501 * sys_brk() for the most part doesn't need the global kernel
502 * lock, except when an application is doing something nasty
503 * like trying to un-brk an area that has already been mapped
504 * to a regular file. in this case, the unmapping will need
505 * to invoke file system routines that need the global lock.
506 */
507SYSCALL_DEFINE1(brk, unsigned long, brk)
508{
509 struct mm_struct *mm = current->mm;
510
511 if (brk < mm->start_brk || brk > mm->context.end_brk)
512 return mm->brk;
513
514 if (mm->brk == brk)
515 return mm->brk;
516
517 /*
518 * Always allow shrinking brk
519 */
520 if (brk <= mm->brk) {
521 mm->brk = brk;
522 return brk;
523 }
524
525 /*
526 * Ok, looks good - let it rip.
527 */
528 flush_icache_range(mm->brk, brk);
529 return mm->brk = brk;
530}
531
532/*
533 * initialise the VMA and region record slabs
534 */
535void __init mmap_init(void)
536{
537 int ret;
538
539 ret = percpu_counter_init(&vm_committed_as, 0);
540 VM_BUG_ON(ret);
541 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
542}
543
544/*
545 * validate the region tree
546 * - the caller must hold the region lock
547 */
548#ifdef CONFIG_DEBUG_NOMMU_REGIONS
549static noinline void validate_nommu_regions(void)
550{
551 struct vm_region *region, *last;
552 struct rb_node *p, *lastp;
553
554 lastp = rb_first(&nommu_region_tree);
555 if (!lastp)
556 return;
557
558 last = rb_entry(lastp, struct vm_region, vm_rb);
559 BUG_ON(unlikely(last->vm_end <= last->vm_start));
560 BUG_ON(unlikely(last->vm_top < last->vm_end));
561
562 while ((p = rb_next(lastp))) {
563 region = rb_entry(p, struct vm_region, vm_rb);
564 last = rb_entry(lastp, struct vm_region, vm_rb);
565
566 BUG_ON(unlikely(region->vm_end <= region->vm_start));
567 BUG_ON(unlikely(region->vm_top < region->vm_end));
568 BUG_ON(unlikely(region->vm_start < last->vm_top));
569
570 lastp = p;
571 }
572}
573#else
574static void validate_nommu_regions(void)
575{
576}
577#endif
578
579/*
580 * add a region into the global tree
581 */
582static void add_nommu_region(struct vm_region *region)
583{
584 struct vm_region *pregion;
585 struct rb_node **p, *parent;
586
587 validate_nommu_regions();
588
589 parent = NULL;
590 p = &nommu_region_tree.rb_node;
591 while (*p) {
592 parent = *p;
593 pregion = rb_entry(parent, struct vm_region, vm_rb);
594 if (region->vm_start < pregion->vm_start)
595 p = &(*p)->rb_left;
596 else if (region->vm_start > pregion->vm_start)
597 p = &(*p)->rb_right;
598 else if (pregion == region)
599 return;
600 else
601 BUG();
602 }
603
604 rb_link_node(®ion->vm_rb, parent, p);
605 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
606
607 validate_nommu_regions();
608}
609
610/*
611 * delete a region from the global tree
612 */
613static void delete_nommu_region(struct vm_region *region)
614{
615 BUG_ON(!nommu_region_tree.rb_node);
616
617 validate_nommu_regions();
618 rb_erase(®ion->vm_rb, &nommu_region_tree);
619 validate_nommu_regions();
620}
621
622/*
623 * free a contiguous series of pages
624 */
625static void free_page_series(unsigned long from, unsigned long to)
626{
627 for (; from < to; from += PAGE_SIZE) {
628 struct page *page = virt_to_page(from);
629
630 kdebug("- free %lx", from);
631 atomic_long_dec(&mmap_pages_allocated);
632 if (page_count(page) != 1)
633 kdebug("free page %p: refcount not one: %d",
634 page, page_count(page));
635 put_page(page);
636 }
637}
638
639/*
640 * release a reference to a region
641 * - the caller must hold the region semaphore for writing, which this releases
642 * - the region may not have been added to the tree yet, in which case vm_top
643 * will equal vm_start
644 */
645static void __put_nommu_region(struct vm_region *region)
646 __releases(nommu_region_sem)
647{
648 kenter("%p{%d}", region, region->vm_usage);
649
650 BUG_ON(!nommu_region_tree.rb_node);
651
652 if (--region->vm_usage == 0) {
653 if (region->vm_top > region->vm_start)
654 delete_nommu_region(region);
655 up_write(&nommu_region_sem);
656
657 if (region->vm_file)
658 fput(region->vm_file);
659
660 /* IO memory and memory shared directly out of the pagecache
661 * from ramfs/tmpfs mustn't be released here */
662 if (region->vm_flags & VM_MAPPED_COPY) {
663 kdebug("free series");
664 free_page_series(region->vm_start, region->vm_top);
665 }
666 kmem_cache_free(vm_region_jar, region);
667 } else {
668 up_write(&nommu_region_sem);
669 }
670}
671
672/*
673 * release a reference to a region
674 */
675static void put_nommu_region(struct vm_region *region)
676{
677 down_write(&nommu_region_sem);
678 __put_nommu_region(region);
679}
680
681/*
682 * update protection on a vma
683 */
684static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
685{
686#ifdef CONFIG_MPU
687 struct mm_struct *mm = vma->vm_mm;
688 long start = vma->vm_start & PAGE_MASK;
689 while (start < vma->vm_end) {
690 protect_page(mm, start, flags);
691 start += PAGE_SIZE;
692 }
693 update_protections(mm);
694#endif
695}
696
697/*
698 * add a VMA into a process's mm_struct in the appropriate place in the list
699 * and tree and add to the address space's page tree also if not an anonymous
700 * page
701 * - should be called with mm->mmap_sem held writelocked
702 */
703static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
704{
705 struct vm_area_struct *pvma, *prev;
706 struct address_space *mapping;
707 struct rb_node **p, *parent, *rb_prev;
708
709 kenter(",%p", vma);
710
711 BUG_ON(!vma->vm_region);
712
713 mm->map_count++;
714 vma->vm_mm = mm;
715
716 protect_vma(vma, vma->vm_flags);
717
718 /* add the VMA to the mapping */
719 if (vma->vm_file) {
720 mapping = vma->vm_file->f_mapping;
721
722 mutex_lock(&mapping->i_mmap_mutex);
723 flush_dcache_mmap_lock(mapping);
724 vma_interval_tree_insert(vma, &mapping->i_mmap);
725 flush_dcache_mmap_unlock(mapping);
726 mutex_unlock(&mapping->i_mmap_mutex);
727 }
728
729 /* add the VMA to the tree */
730 parent = rb_prev = NULL;
731 p = &mm->mm_rb.rb_node;
732 while (*p) {
733 parent = *p;
734 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
735
736 /* sort by: start addr, end addr, VMA struct addr in that order
737 * (the latter is necessary as we may get identical VMAs) */
738 if (vma->vm_start < pvma->vm_start)
739 p = &(*p)->rb_left;
740 else if (vma->vm_start > pvma->vm_start) {
741 rb_prev = parent;
742 p = &(*p)->rb_right;
743 } else if (vma->vm_end < pvma->vm_end)
744 p = &(*p)->rb_left;
745 else if (vma->vm_end > pvma->vm_end) {
746 rb_prev = parent;
747 p = &(*p)->rb_right;
748 } else if (vma < pvma)
749 p = &(*p)->rb_left;
750 else if (vma > pvma) {
751 rb_prev = parent;
752 p = &(*p)->rb_right;
753 } else
754 BUG();
755 }
756
757 rb_link_node(&vma->vm_rb, parent, p);
758 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
759
760 /* add VMA to the VMA list also */
761 prev = NULL;
762 if (rb_prev)
763 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
764
765 __vma_link_list(mm, vma, prev, parent);
766}
767
768/*
769 * delete a VMA from its owning mm_struct and address space
770 */
771static void delete_vma_from_mm(struct vm_area_struct *vma)
772{
773 int i;
774 struct address_space *mapping;
775 struct mm_struct *mm = vma->vm_mm;
776 struct task_struct *curr = current;
777
778 kenter("%p", vma);
779
780 protect_vma(vma, 0);
781
782 mm->map_count--;
783 for (i = 0; i < VMACACHE_SIZE; i++) {
784 /* if the vma is cached, invalidate the entire cache */
785 if (curr->vmacache[i] == vma) {
786 vmacache_invalidate(curr->mm);
787 break;
788 }
789 }
790
791 /* remove the VMA from the mapping */
792 if (vma->vm_file) {
793 mapping = vma->vm_file->f_mapping;
794
795 mutex_lock(&mapping->i_mmap_mutex);
796 flush_dcache_mmap_lock(mapping);
797 vma_interval_tree_remove(vma, &mapping->i_mmap);
798 flush_dcache_mmap_unlock(mapping);
799 mutex_unlock(&mapping->i_mmap_mutex);
800 }
801
802 /* remove from the MM's tree and list */
803 rb_erase(&vma->vm_rb, &mm->mm_rb);
804
805 if (vma->vm_prev)
806 vma->vm_prev->vm_next = vma->vm_next;
807 else
808 mm->mmap = vma->vm_next;
809
810 if (vma->vm_next)
811 vma->vm_next->vm_prev = vma->vm_prev;
812}
813
814/*
815 * destroy a VMA record
816 */
817static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
818{
819 kenter("%p", vma);
820 if (vma->vm_ops && vma->vm_ops->close)
821 vma->vm_ops->close(vma);
822 if (vma->vm_file)
823 fput(vma->vm_file);
824 put_nommu_region(vma->vm_region);
825 kmem_cache_free(vm_area_cachep, vma);
826}
827
828/*
829 * look up the first VMA in which addr resides, NULL if none
830 * - should be called with mm->mmap_sem at least held readlocked
831 */
832struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
833{
834 struct vm_area_struct *vma;
835
836 /* check the cache first */
837 vma = vmacache_find(mm, addr);
838 if (likely(vma))
839 return vma;
840
841 /* trawl the list (there may be multiple mappings in which addr
842 * resides) */
843 for (vma = mm->mmap; vma; vma = vma->vm_next) {
844 if (vma->vm_start > addr)
845 return NULL;
846 if (vma->vm_end > addr) {
847 vmacache_update(addr, vma);
848 return vma;
849 }
850 }
851
852 return NULL;
853}
854EXPORT_SYMBOL(find_vma);
855
856/*
857 * find a VMA
858 * - we don't extend stack VMAs under NOMMU conditions
859 */
860struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
861{
862 return find_vma(mm, addr);
863}
864
865/*
866 * expand a stack to a given address
867 * - not supported under NOMMU conditions
868 */
869int expand_stack(struct vm_area_struct *vma, unsigned long address)
870{
871 return -ENOMEM;
872}
873
874/*
875 * look up the first VMA exactly that exactly matches addr
876 * - should be called with mm->mmap_sem at least held readlocked
877 */
878static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
879 unsigned long addr,
880 unsigned long len)
881{
882 struct vm_area_struct *vma;
883 unsigned long end = addr + len;
884
885 /* check the cache first */
886 vma = vmacache_find_exact(mm, addr, end);
887 if (vma)
888 return vma;
889
890 /* trawl the list (there may be multiple mappings in which addr
891 * resides) */
892 for (vma = mm->mmap; vma; vma = vma->vm_next) {
893 if (vma->vm_start < addr)
894 continue;
895 if (vma->vm_start > addr)
896 return NULL;
897 if (vma->vm_end == end) {
898 vmacache_update(addr, vma);
899 return vma;
900 }
901 }
902
903 return NULL;
904}
905
906/*
907 * determine whether a mapping should be permitted and, if so, what sort of
908 * mapping we're capable of supporting
909 */
910static int validate_mmap_request(struct file *file,
911 unsigned long addr,
912 unsigned long len,
913 unsigned long prot,
914 unsigned long flags,
915 unsigned long pgoff,
916 unsigned long *_capabilities)
917{
918 unsigned long capabilities, rlen;
919 int ret;
920
921 /* do the simple checks first */
922 if (flags & MAP_FIXED) {
923 printk(KERN_DEBUG
924 "%d: Can't do fixed-address/overlay mmap of RAM\n",
925 current->pid);
926 return -EINVAL;
927 }
928
929 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
930 (flags & MAP_TYPE) != MAP_SHARED)
931 return -EINVAL;
932
933 if (!len)
934 return -EINVAL;
935
936 /* Careful about overflows.. */
937 rlen = PAGE_ALIGN(len);
938 if (!rlen || rlen > TASK_SIZE)
939 return -ENOMEM;
940
941 /* offset overflow? */
942 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
943 return -EOVERFLOW;
944
945 if (file) {
946 /* validate file mapping requests */
947 struct address_space *mapping;
948
949 /* files must support mmap */
950 if (!file->f_op->mmap)
951 return -ENODEV;
952
953 /* work out if what we've got could possibly be shared
954 * - we support chardevs that provide their own "memory"
955 * - we support files/blockdevs that are memory backed
956 */
957 mapping = file->f_mapping;
958 if (!mapping)
959 mapping = file_inode(file)->i_mapping;
960
961 capabilities = 0;
962 if (mapping && mapping->backing_dev_info)
963 capabilities = mapping->backing_dev_info->capabilities;
964
965 if (!capabilities) {
966 /* no explicit capabilities set, so assume some
967 * defaults */
968 switch (file_inode(file)->i_mode & S_IFMT) {
969 case S_IFREG:
970 case S_IFBLK:
971 capabilities = BDI_CAP_MAP_COPY;
972 break;
973
974 case S_IFCHR:
975 capabilities =
976 BDI_CAP_MAP_DIRECT |
977 BDI_CAP_READ_MAP |
978 BDI_CAP_WRITE_MAP;
979 break;
980
981 default:
982 return -EINVAL;
983 }
984 }
985
986 /* eliminate any capabilities that we can't support on this
987 * device */
988 if (!file->f_op->get_unmapped_area)
989 capabilities &= ~BDI_CAP_MAP_DIRECT;
990 if (!file->f_op->read)
991 capabilities &= ~BDI_CAP_MAP_COPY;
992
993 /* The file shall have been opened with read permission. */
994 if (!(file->f_mode & FMODE_READ))
995 return -EACCES;
996
997 if (flags & MAP_SHARED) {
998 /* do checks for writing, appending and locking */
999 if ((prot & PROT_WRITE) &&
1000 !(file->f_mode & FMODE_WRITE))
1001 return -EACCES;
1002
1003 if (IS_APPEND(file_inode(file)) &&
1004 (file->f_mode & FMODE_WRITE))
1005 return -EACCES;
1006
1007 if (locks_verify_locked(file))
1008 return -EAGAIN;
1009
1010 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1011 return -ENODEV;
1012
1013 /* we mustn't privatise shared mappings */
1014 capabilities &= ~BDI_CAP_MAP_COPY;
1015 } else {
1016 /* we're going to read the file into private memory we
1017 * allocate */
1018 if (!(capabilities & BDI_CAP_MAP_COPY))
1019 return -ENODEV;
1020
1021 /* we don't permit a private writable mapping to be
1022 * shared with the backing device */
1023 if (prot & PROT_WRITE)
1024 capabilities &= ~BDI_CAP_MAP_DIRECT;
1025 }
1026
1027 if (capabilities & BDI_CAP_MAP_DIRECT) {
1028 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
1029 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1030 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
1031 ) {
1032 capabilities &= ~BDI_CAP_MAP_DIRECT;
1033 if (flags & MAP_SHARED) {
1034 printk(KERN_WARNING
1035 "MAP_SHARED not completely supported on !MMU\n");
1036 return -EINVAL;
1037 }
1038 }
1039 }
1040
1041 /* handle executable mappings and implied executable
1042 * mappings */
1043 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1044 if (prot & PROT_EXEC)
1045 return -EPERM;
1046 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1047 /* handle implication of PROT_EXEC by PROT_READ */
1048 if (current->personality & READ_IMPLIES_EXEC) {
1049 if (capabilities & BDI_CAP_EXEC_MAP)
1050 prot |= PROT_EXEC;
1051 }
1052 } else if ((prot & PROT_READ) &&
1053 (prot & PROT_EXEC) &&
1054 !(capabilities & BDI_CAP_EXEC_MAP)
1055 ) {
1056 /* backing file is not executable, try to copy */
1057 capabilities &= ~BDI_CAP_MAP_DIRECT;
1058 }
1059 } else {
1060 /* anonymous mappings are always memory backed and can be
1061 * privately mapped
1062 */
1063 capabilities = BDI_CAP_MAP_COPY;
1064
1065 /* handle PROT_EXEC implication by PROT_READ */
1066 if ((prot & PROT_READ) &&
1067 (current->personality & READ_IMPLIES_EXEC))
1068 prot |= PROT_EXEC;
1069 }
1070
1071 /* allow the security API to have its say */
1072 ret = security_mmap_addr(addr);
1073 if (ret < 0)
1074 return ret;
1075
1076 /* looks okay */
1077 *_capabilities = capabilities;
1078 return 0;
1079}
1080
1081/*
1082 * we've determined that we can make the mapping, now translate what we
1083 * now know into VMA flags
1084 */
1085static unsigned long determine_vm_flags(struct file *file,
1086 unsigned long prot,
1087 unsigned long flags,
1088 unsigned long capabilities)
1089{
1090 unsigned long vm_flags;
1091
1092 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1093 /* vm_flags |= mm->def_flags; */
1094
1095 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1096 /* attempt to share read-only copies of mapped file chunks */
1097 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1098 if (file && !(prot & PROT_WRITE))
1099 vm_flags |= VM_MAYSHARE;
1100 } else {
1101 /* overlay a shareable mapping on the backing device or inode
1102 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1103 * romfs/cramfs */
1104 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1105 if (flags & MAP_SHARED)
1106 vm_flags |= VM_SHARED;
1107 }
1108
1109 /* refuse to let anyone share private mappings with this process if
1110 * it's being traced - otherwise breakpoints set in it may interfere
1111 * with another untraced process
1112 */
1113 if ((flags & MAP_PRIVATE) && current->ptrace)
1114 vm_flags &= ~VM_MAYSHARE;
1115
1116 return vm_flags;
1117}
1118
1119/*
1120 * set up a shared mapping on a file (the driver or filesystem provides and
1121 * pins the storage)
1122 */
1123static int do_mmap_shared_file(struct vm_area_struct *vma)
1124{
1125 int ret;
1126
1127 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1128 if (ret == 0) {
1129 vma->vm_region->vm_top = vma->vm_region->vm_end;
1130 return 0;
1131 }
1132 if (ret != -ENOSYS)
1133 return ret;
1134
1135 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1136 * opposed to tried but failed) so we can only give a suitable error as
1137 * it's not possible to make a private copy if MAP_SHARED was given */
1138 return -ENODEV;
1139}
1140
1141/*
1142 * set up a private mapping or an anonymous shared mapping
1143 */
1144static int do_mmap_private(struct vm_area_struct *vma,
1145 struct vm_region *region,
1146 unsigned long len,
1147 unsigned long capabilities)
1148{
1149 struct page *pages;
1150 unsigned long total, point, n;
1151 void *base;
1152 int ret, order;
1153
1154 /* invoke the file's mapping function so that it can keep track of
1155 * shared mappings on devices or memory
1156 * - VM_MAYSHARE will be set if it may attempt to share
1157 */
1158 if (capabilities & BDI_CAP_MAP_DIRECT) {
1159 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1160 if (ret == 0) {
1161 /* shouldn't return success if we're not sharing */
1162 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1163 vma->vm_region->vm_top = vma->vm_region->vm_end;
1164 return 0;
1165 }
1166 if (ret != -ENOSYS)
1167 return ret;
1168
1169 /* getting an ENOSYS error indicates that direct mmap isn't
1170 * possible (as opposed to tried but failed) so we'll try to
1171 * make a private copy of the data and map that instead */
1172 }
1173
1174
1175 /* allocate some memory to hold the mapping
1176 * - note that this may not return a page-aligned address if the object
1177 * we're allocating is smaller than a page
1178 */
1179 order = get_order(len);
1180 kdebug("alloc order %d for %lx", order, len);
1181
1182 pages = alloc_pages(GFP_KERNEL, order);
1183 if (!pages)
1184 goto enomem;
1185
1186 total = 1 << order;
1187 atomic_long_add(total, &mmap_pages_allocated);
1188
1189 point = len >> PAGE_SHIFT;
1190
1191 /* we allocated a power-of-2 sized page set, so we may want to trim off
1192 * the excess */
1193 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1194 while (total > point) {
1195 order = ilog2(total - point);
1196 n = 1 << order;
1197 kdebug("shave %lu/%lu @%lu", n, total - point, total);
1198 atomic_long_sub(n, &mmap_pages_allocated);
1199 total -= n;
1200 set_page_refcounted(pages + total);
1201 __free_pages(pages + total, order);
1202 }
1203 }
1204
1205 for (point = 1; point < total; point++)
1206 set_page_refcounted(&pages[point]);
1207
1208 base = page_address(pages);
1209 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1210 region->vm_start = (unsigned long) base;
1211 region->vm_end = region->vm_start + len;
1212 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1213
1214 vma->vm_start = region->vm_start;
1215 vma->vm_end = region->vm_start + len;
1216
1217 if (vma->vm_file) {
1218 /* read the contents of a file into the copy */
1219 mm_segment_t old_fs;
1220 loff_t fpos;
1221
1222 fpos = vma->vm_pgoff;
1223 fpos <<= PAGE_SHIFT;
1224
1225 old_fs = get_fs();
1226 set_fs(KERNEL_DS);
1227 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
1228 set_fs(old_fs);
1229
1230 if (ret < 0)
1231 goto error_free;
1232
1233 /* clear the last little bit */
1234 if (ret < len)
1235 memset(base + ret, 0, len - ret);
1236
1237 }
1238
1239 return 0;
1240
1241error_free:
1242 free_page_series(region->vm_start, region->vm_top);
1243 region->vm_start = vma->vm_start = 0;
1244 region->vm_end = vma->vm_end = 0;
1245 region->vm_top = 0;
1246 return ret;
1247
1248enomem:
1249 printk("Allocation of length %lu from process %d (%s) failed\n",
1250 len, current->pid, current->comm);
1251 show_free_areas(0);
1252 return -ENOMEM;
1253}
1254
1255/*
1256 * handle mapping creation for uClinux
1257 */
1258unsigned long do_mmap_pgoff(struct file *file,
1259 unsigned long addr,
1260 unsigned long len,
1261 unsigned long prot,
1262 unsigned long flags,
1263 unsigned long pgoff,
1264 unsigned long *populate)
1265{
1266 struct vm_area_struct *vma;
1267 struct vm_region *region;
1268 struct rb_node *rb;
1269 unsigned long capabilities, vm_flags, result;
1270 int ret;
1271
1272 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1273
1274 *populate = 0;
1275
1276 /* decide whether we should attempt the mapping, and if so what sort of
1277 * mapping */
1278 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1279 &capabilities);
1280 if (ret < 0) {
1281 kleave(" = %d [val]", ret);
1282 return ret;
1283 }
1284
1285 /* we ignore the address hint */
1286 addr = 0;
1287 len = PAGE_ALIGN(len);
1288
1289 /* we've determined that we can make the mapping, now translate what we
1290 * now know into VMA flags */
1291 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1292
1293 /* we're going to need to record the mapping */
1294 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1295 if (!region)
1296 goto error_getting_region;
1297
1298 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1299 if (!vma)
1300 goto error_getting_vma;
1301
1302 region->vm_usage = 1;
1303 region->vm_flags = vm_flags;
1304 region->vm_pgoff = pgoff;
1305
1306 INIT_LIST_HEAD(&vma->anon_vma_chain);
1307 vma->vm_flags = vm_flags;
1308 vma->vm_pgoff = pgoff;
1309
1310 if (file) {
1311 region->vm_file = get_file(file);
1312 vma->vm_file = get_file(file);
1313 }
1314
1315 down_write(&nommu_region_sem);
1316
1317 /* if we want to share, we need to check for regions created by other
1318 * mmap() calls that overlap with our proposed mapping
1319 * - we can only share with a superset match on most regular files
1320 * - shared mappings on character devices and memory backed files are
1321 * permitted to overlap inexactly as far as we are concerned for in
1322 * these cases, sharing is handled in the driver or filesystem rather
1323 * than here
1324 */
1325 if (vm_flags & VM_MAYSHARE) {
1326 struct vm_region *pregion;
1327 unsigned long pglen, rpglen, pgend, rpgend, start;
1328
1329 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1330 pgend = pgoff + pglen;
1331
1332 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1333 pregion = rb_entry(rb, struct vm_region, vm_rb);
1334
1335 if (!(pregion->vm_flags & VM_MAYSHARE))
1336 continue;
1337
1338 /* search for overlapping mappings on the same file */
1339 if (file_inode(pregion->vm_file) !=
1340 file_inode(file))
1341 continue;
1342
1343 if (pregion->vm_pgoff >= pgend)
1344 continue;
1345
1346 rpglen = pregion->vm_end - pregion->vm_start;
1347 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1348 rpgend = pregion->vm_pgoff + rpglen;
1349 if (pgoff >= rpgend)
1350 continue;
1351
1352 /* handle inexactly overlapping matches between
1353 * mappings */
1354 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1355 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1356 /* new mapping is not a subset of the region */
1357 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1358 goto sharing_violation;
1359 continue;
1360 }
1361
1362 /* we've found a region we can share */
1363 pregion->vm_usage++;
1364 vma->vm_region = pregion;
1365 start = pregion->vm_start;
1366 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1367 vma->vm_start = start;
1368 vma->vm_end = start + len;
1369
1370 if (pregion->vm_flags & VM_MAPPED_COPY) {
1371 kdebug("share copy");
1372 vma->vm_flags |= VM_MAPPED_COPY;
1373 } else {
1374 kdebug("share mmap");
1375 ret = do_mmap_shared_file(vma);
1376 if (ret < 0) {
1377 vma->vm_region = NULL;
1378 vma->vm_start = 0;
1379 vma->vm_end = 0;
1380 pregion->vm_usage--;
1381 pregion = NULL;
1382 goto error_just_free;
1383 }
1384 }
1385 fput(region->vm_file);
1386 kmem_cache_free(vm_region_jar, region);
1387 region = pregion;
1388 result = start;
1389 goto share;
1390 }
1391
1392 /* obtain the address at which to make a shared mapping
1393 * - this is the hook for quasi-memory character devices to
1394 * tell us the location of a shared mapping
1395 */
1396 if (capabilities & BDI_CAP_MAP_DIRECT) {
1397 addr = file->f_op->get_unmapped_area(file, addr, len,
1398 pgoff, flags);
1399 if (IS_ERR_VALUE(addr)) {
1400 ret = addr;
1401 if (ret != -ENOSYS)
1402 goto error_just_free;
1403
1404 /* the driver refused to tell us where to site
1405 * the mapping so we'll have to attempt to copy
1406 * it */
1407 ret = -ENODEV;
1408 if (!(capabilities & BDI_CAP_MAP_COPY))
1409 goto error_just_free;
1410
1411 capabilities &= ~BDI_CAP_MAP_DIRECT;
1412 } else {
1413 vma->vm_start = region->vm_start = addr;
1414 vma->vm_end = region->vm_end = addr + len;
1415 }
1416 }
1417 }
1418
1419 vma->vm_region = region;
1420
1421 /* set up the mapping
1422 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1423 */
1424 if (file && vma->vm_flags & VM_SHARED)
1425 ret = do_mmap_shared_file(vma);
1426 else
1427 ret = do_mmap_private(vma, region, len, capabilities);
1428 if (ret < 0)
1429 goto error_just_free;
1430 add_nommu_region(region);
1431
1432 /* clear anonymous mappings that don't ask for uninitialized data */
1433 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1434 memset((void *)region->vm_start, 0,
1435 region->vm_end - region->vm_start);
1436
1437 /* okay... we have a mapping; now we have to register it */
1438 result = vma->vm_start;
1439
1440 current->mm->total_vm += len >> PAGE_SHIFT;
1441
1442share:
1443 add_vma_to_mm(current->mm, vma);
1444
1445 /* we flush the region from the icache only when the first executable
1446 * mapping of it is made */
1447 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1448 flush_icache_range(region->vm_start, region->vm_end);
1449 region->vm_icache_flushed = true;
1450 }
1451
1452 up_write(&nommu_region_sem);
1453
1454 kleave(" = %lx", result);
1455 return result;
1456
1457error_just_free:
1458 up_write(&nommu_region_sem);
1459error:
1460 if (region->vm_file)
1461 fput(region->vm_file);
1462 kmem_cache_free(vm_region_jar, region);
1463 if (vma->vm_file)
1464 fput(vma->vm_file);
1465 kmem_cache_free(vm_area_cachep, vma);
1466 kleave(" = %d", ret);
1467 return ret;
1468
1469sharing_violation:
1470 up_write(&nommu_region_sem);
1471 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1472 ret = -EINVAL;
1473 goto error;
1474
1475error_getting_vma:
1476 kmem_cache_free(vm_region_jar, region);
1477 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1478 " from process %d failed\n",
1479 len, current->pid);
1480 show_free_areas(0);
1481 return -ENOMEM;
1482
1483error_getting_region:
1484 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1485 " from process %d failed\n",
1486 len, current->pid);
1487 show_free_areas(0);
1488 return -ENOMEM;
1489}
1490
1491SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1492 unsigned long, prot, unsigned long, flags,
1493 unsigned long, fd, unsigned long, pgoff)
1494{
1495 struct file *file = NULL;
1496 unsigned long retval = -EBADF;
1497
1498 audit_mmap_fd(fd, flags);
1499 if (!(flags & MAP_ANONYMOUS)) {
1500 file = fget(fd);
1501 if (!file)
1502 goto out;
1503 }
1504
1505 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1506
1507 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1508
1509 if (file)
1510 fput(file);
1511out:
1512 return retval;
1513}
1514
1515#ifdef __ARCH_WANT_SYS_OLD_MMAP
1516struct mmap_arg_struct {
1517 unsigned long addr;
1518 unsigned long len;
1519 unsigned long prot;
1520 unsigned long flags;
1521 unsigned long fd;
1522 unsigned long offset;
1523};
1524
1525SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1526{
1527 struct mmap_arg_struct a;
1528
1529 if (copy_from_user(&a, arg, sizeof(a)))
1530 return -EFAULT;
1531 if (a.offset & ~PAGE_MASK)
1532 return -EINVAL;
1533
1534 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1535 a.offset >> PAGE_SHIFT);
1536}
1537#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1538
1539/*
1540 * split a vma into two pieces at address 'addr', a new vma is allocated either
1541 * for the first part or the tail.
1542 */
1543int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1544 unsigned long addr, int new_below)
1545{
1546 struct vm_area_struct *new;
1547 struct vm_region *region;
1548 unsigned long npages;
1549
1550 kenter("");
1551
1552 /* we're only permitted to split anonymous regions (these should have
1553 * only a single usage on the region) */
1554 if (vma->vm_file)
1555 return -ENOMEM;
1556
1557 if (mm->map_count >= sysctl_max_map_count)
1558 return -ENOMEM;
1559
1560 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1561 if (!region)
1562 return -ENOMEM;
1563
1564 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1565 if (!new) {
1566 kmem_cache_free(vm_region_jar, region);
1567 return -ENOMEM;
1568 }
1569
1570 /* most fields are the same, copy all, and then fixup */
1571 *new = *vma;
1572 *region = *vma->vm_region;
1573 new->vm_region = region;
1574
1575 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1576
1577 if (new_below) {
1578 region->vm_top = region->vm_end = new->vm_end = addr;
1579 } else {
1580 region->vm_start = new->vm_start = addr;
1581 region->vm_pgoff = new->vm_pgoff += npages;
1582 }
1583
1584 if (new->vm_ops && new->vm_ops->open)
1585 new->vm_ops->open(new);
1586
1587 delete_vma_from_mm(vma);
1588 down_write(&nommu_region_sem);
1589 delete_nommu_region(vma->vm_region);
1590 if (new_below) {
1591 vma->vm_region->vm_start = vma->vm_start = addr;
1592 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1593 } else {
1594 vma->vm_region->vm_end = vma->vm_end = addr;
1595 vma->vm_region->vm_top = addr;
1596 }
1597 add_nommu_region(vma->vm_region);
1598 add_nommu_region(new->vm_region);
1599 up_write(&nommu_region_sem);
1600 add_vma_to_mm(mm, vma);
1601 add_vma_to_mm(mm, new);
1602 return 0;
1603}
1604
1605/*
1606 * shrink a VMA by removing the specified chunk from either the beginning or
1607 * the end
1608 */
1609static int shrink_vma(struct mm_struct *mm,
1610 struct vm_area_struct *vma,
1611 unsigned long from, unsigned long to)
1612{
1613 struct vm_region *region;
1614
1615 kenter("");
1616
1617 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1618 * and list */
1619 delete_vma_from_mm(vma);
1620 if (from > vma->vm_start)
1621 vma->vm_end = from;
1622 else
1623 vma->vm_start = to;
1624 add_vma_to_mm(mm, vma);
1625
1626 /* cut the backing region down to size */
1627 region = vma->vm_region;
1628 BUG_ON(region->vm_usage != 1);
1629
1630 down_write(&nommu_region_sem);
1631 delete_nommu_region(region);
1632 if (from > region->vm_start) {
1633 to = region->vm_top;
1634 region->vm_top = region->vm_end = from;
1635 } else {
1636 region->vm_start = to;
1637 }
1638 add_nommu_region(region);
1639 up_write(&nommu_region_sem);
1640
1641 free_page_series(from, to);
1642 return 0;
1643}
1644
1645/*
1646 * release a mapping
1647 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1648 * VMA, though it need not cover the whole VMA
1649 */
1650int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1651{
1652 struct vm_area_struct *vma;
1653 unsigned long end;
1654 int ret;
1655
1656 kenter(",%lx,%zx", start, len);
1657
1658 len = PAGE_ALIGN(len);
1659 if (len == 0)
1660 return -EINVAL;
1661
1662 end = start + len;
1663
1664 /* find the first potentially overlapping VMA */
1665 vma = find_vma(mm, start);
1666 if (!vma) {
1667 static int limit;
1668 if (limit < 5) {
1669 printk(KERN_WARNING
1670 "munmap of memory not mmapped by process %d"
1671 " (%s): 0x%lx-0x%lx\n",
1672 current->pid, current->comm,
1673 start, start + len - 1);
1674 limit++;
1675 }
1676 return -EINVAL;
1677 }
1678
1679 /* we're allowed to split an anonymous VMA but not a file-backed one */
1680 if (vma->vm_file) {
1681 do {
1682 if (start > vma->vm_start) {
1683 kleave(" = -EINVAL [miss]");
1684 return -EINVAL;
1685 }
1686 if (end == vma->vm_end)
1687 goto erase_whole_vma;
1688 vma = vma->vm_next;
1689 } while (vma);
1690 kleave(" = -EINVAL [split file]");
1691 return -EINVAL;
1692 } else {
1693 /* the chunk must be a subset of the VMA found */
1694 if (start == vma->vm_start && end == vma->vm_end)
1695 goto erase_whole_vma;
1696 if (start < vma->vm_start || end > vma->vm_end) {
1697 kleave(" = -EINVAL [superset]");
1698 return -EINVAL;
1699 }
1700 if (start & ~PAGE_MASK) {
1701 kleave(" = -EINVAL [unaligned start]");
1702 return -EINVAL;
1703 }
1704 if (end != vma->vm_end && end & ~PAGE_MASK) {
1705 kleave(" = -EINVAL [unaligned split]");
1706 return -EINVAL;
1707 }
1708 if (start != vma->vm_start && end != vma->vm_end) {
1709 ret = split_vma(mm, vma, start, 1);
1710 if (ret < 0) {
1711 kleave(" = %d [split]", ret);
1712 return ret;
1713 }
1714 }
1715 return shrink_vma(mm, vma, start, end);
1716 }
1717
1718erase_whole_vma:
1719 delete_vma_from_mm(vma);
1720 delete_vma(mm, vma);
1721 kleave(" = 0");
1722 return 0;
1723}
1724EXPORT_SYMBOL(do_munmap);
1725
1726int vm_munmap(unsigned long addr, size_t len)
1727{
1728 struct mm_struct *mm = current->mm;
1729 int ret;
1730
1731 down_write(&mm->mmap_sem);
1732 ret = do_munmap(mm, addr, len);
1733 up_write(&mm->mmap_sem);
1734 return ret;
1735}
1736EXPORT_SYMBOL(vm_munmap);
1737
1738SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1739{
1740 return vm_munmap(addr, len);
1741}
1742
1743/*
1744 * release all the mappings made in a process's VM space
1745 */
1746void exit_mmap(struct mm_struct *mm)
1747{
1748 struct vm_area_struct *vma;
1749
1750 if (!mm)
1751 return;
1752
1753 kenter("");
1754
1755 mm->total_vm = 0;
1756
1757 while ((vma = mm->mmap)) {
1758 mm->mmap = vma->vm_next;
1759 delete_vma_from_mm(vma);
1760 delete_vma(mm, vma);
1761 cond_resched();
1762 }
1763
1764 kleave("");
1765}
1766
1767unsigned long vm_brk(unsigned long addr, unsigned long len)
1768{
1769 return -ENOMEM;
1770}
1771
1772/*
1773 * expand (or shrink) an existing mapping, potentially moving it at the same
1774 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1775 *
1776 * under NOMMU conditions, we only permit changing a mapping's size, and only
1777 * as long as it stays within the region allocated by do_mmap_private() and the
1778 * block is not shareable
1779 *
1780 * MREMAP_FIXED is not supported under NOMMU conditions
1781 */
1782static unsigned long do_mremap(unsigned long addr,
1783 unsigned long old_len, unsigned long new_len,
1784 unsigned long flags, unsigned long new_addr)
1785{
1786 struct vm_area_struct *vma;
1787
1788 /* insanity checks first */
1789 old_len = PAGE_ALIGN(old_len);
1790 new_len = PAGE_ALIGN(new_len);
1791 if (old_len == 0 || new_len == 0)
1792 return (unsigned long) -EINVAL;
1793
1794 if (addr & ~PAGE_MASK)
1795 return -EINVAL;
1796
1797 if (flags & MREMAP_FIXED && new_addr != addr)
1798 return (unsigned long) -EINVAL;
1799
1800 vma = find_vma_exact(current->mm, addr, old_len);
1801 if (!vma)
1802 return (unsigned long) -EINVAL;
1803
1804 if (vma->vm_end != vma->vm_start + old_len)
1805 return (unsigned long) -EFAULT;
1806
1807 if (vma->vm_flags & VM_MAYSHARE)
1808 return (unsigned long) -EPERM;
1809
1810 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1811 return (unsigned long) -ENOMEM;
1812
1813 /* all checks complete - do it */
1814 vma->vm_end = vma->vm_start + new_len;
1815 return vma->vm_start;
1816}
1817
1818SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1819 unsigned long, new_len, unsigned long, flags,
1820 unsigned long, new_addr)
1821{
1822 unsigned long ret;
1823
1824 down_write(¤t->mm->mmap_sem);
1825 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1826 up_write(¤t->mm->mmap_sem);
1827 return ret;
1828}
1829
1830struct page *follow_page_mask(struct vm_area_struct *vma,
1831 unsigned long address, unsigned int flags,
1832 unsigned int *page_mask)
1833{
1834 *page_mask = 0;
1835 return NULL;
1836}
1837
1838int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1839 unsigned long pfn, unsigned long size, pgprot_t prot)
1840{
1841 if (addr != (pfn << PAGE_SHIFT))
1842 return -EINVAL;
1843
1844 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1845 return 0;
1846}
1847EXPORT_SYMBOL(remap_pfn_range);
1848
1849int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1850{
1851 unsigned long pfn = start >> PAGE_SHIFT;
1852 unsigned long vm_len = vma->vm_end - vma->vm_start;
1853
1854 pfn += vma->vm_pgoff;
1855 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1856}
1857EXPORT_SYMBOL(vm_iomap_memory);
1858
1859int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1860 unsigned long pgoff)
1861{
1862 unsigned int size = vma->vm_end - vma->vm_start;
1863
1864 if (!(vma->vm_flags & VM_USERMAP))
1865 return -EINVAL;
1866
1867 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1868 vma->vm_end = vma->vm_start + size;
1869
1870 return 0;
1871}
1872EXPORT_SYMBOL(remap_vmalloc_range);
1873
1874unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1875 unsigned long len, unsigned long pgoff, unsigned long flags)
1876{
1877 return -ENOMEM;
1878}
1879
1880void unmap_mapping_range(struct address_space *mapping,
1881 loff_t const holebegin, loff_t const holelen,
1882 int even_cows)
1883{
1884}
1885EXPORT_SYMBOL(unmap_mapping_range);
1886
1887/*
1888 * Check that a process has enough memory to allocate a new virtual
1889 * mapping. 0 means there is enough memory for the allocation to
1890 * succeed and -ENOMEM implies there is not.
1891 *
1892 * We currently support three overcommit policies, which are set via the
1893 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1894 *
1895 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1896 * Additional code 2002 Jul 20 by Robert Love.
1897 *
1898 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1899 *
1900 * Note this is a helper function intended to be used by LSMs which
1901 * wish to use this logic.
1902 */
1903int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1904{
1905 unsigned long free, allowed, reserve;
1906
1907 vm_acct_memory(pages);
1908
1909 /*
1910 * Sometimes we want to use more memory than we have
1911 */
1912 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1913 return 0;
1914
1915 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1916 free = global_page_state(NR_FREE_PAGES);
1917 free += global_page_state(NR_FILE_PAGES);
1918
1919 /*
1920 * shmem pages shouldn't be counted as free in this
1921 * case, they can't be purged, only swapped out, and
1922 * that won't affect the overall amount of available
1923 * memory in the system.
1924 */
1925 free -= global_page_state(NR_SHMEM);
1926
1927 free += get_nr_swap_pages();
1928
1929 /*
1930 * Any slabs which are created with the
1931 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1932 * which are reclaimable, under pressure. The dentry
1933 * cache and most inode caches should fall into this
1934 */
1935 free += global_page_state(NR_SLAB_RECLAIMABLE);
1936
1937 /*
1938 * Leave reserved pages. The pages are not for anonymous pages.
1939 */
1940 if (free <= totalreserve_pages)
1941 goto error;
1942 else
1943 free -= totalreserve_pages;
1944
1945 /*
1946 * Reserve some for root
1947 */
1948 if (!cap_sys_admin)
1949 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1950
1951 if (free > pages)
1952 return 0;
1953
1954 goto error;
1955 }
1956
1957 allowed = vm_commit_limit();
1958 /*
1959 * Reserve some 3% for root
1960 */
1961 if (!cap_sys_admin)
1962 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1963
1964 /*
1965 * Don't let a single process grow so big a user can't recover
1966 */
1967 if (mm) {
1968 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1969 allowed -= min(mm->total_vm / 32, reserve);
1970 }
1971
1972 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1973 return 0;
1974
1975error:
1976 vm_unacct_memory(pages);
1977
1978 return -ENOMEM;
1979}
1980
1981int in_gate_area_no_mm(unsigned long addr)
1982{
1983 return 0;
1984}
1985
1986int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1987{
1988 BUG();
1989 return 0;
1990}
1991EXPORT_SYMBOL(filemap_fault);
1992
1993void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1994{
1995 BUG();
1996}
1997EXPORT_SYMBOL(filemap_map_pages);
1998
1999int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr,
2000 unsigned long size, pgoff_t pgoff)
2001{
2002 BUG();
2003 return 0;
2004}
2005EXPORT_SYMBOL(generic_file_remap_pages);
2006
2007static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
2008 unsigned long addr, void *buf, int len, int write)
2009{
2010 struct vm_area_struct *vma;
2011
2012 down_read(&mm->mmap_sem);
2013
2014 /* the access must start within one of the target process's mappings */
2015 vma = find_vma(mm, addr);
2016 if (vma) {
2017 /* don't overrun this mapping */
2018 if (addr + len >= vma->vm_end)
2019 len = vma->vm_end - addr;
2020
2021 /* only read or write mappings where it is permitted */
2022 if (write && vma->vm_flags & VM_MAYWRITE)
2023 copy_to_user_page(vma, NULL, addr,
2024 (void *) addr, buf, len);
2025 else if (!write && vma->vm_flags & VM_MAYREAD)
2026 copy_from_user_page(vma, NULL, addr,
2027 buf, (void *) addr, len);
2028 else
2029 len = 0;
2030 } else {
2031 len = 0;
2032 }
2033
2034 up_read(&mm->mmap_sem);
2035
2036 return len;
2037}
2038
2039/**
2040 * @access_remote_vm - access another process' address space
2041 * @mm: the mm_struct of the target address space
2042 * @addr: start address to access
2043 * @buf: source or destination buffer
2044 * @len: number of bytes to transfer
2045 * @write: whether the access is a write
2046 *
2047 * The caller must hold a reference on @mm.
2048 */
2049int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2050 void *buf, int len, int write)
2051{
2052 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2053}
2054
2055/*
2056 * Access another process' address space.
2057 * - source/target buffer must be kernel space
2058 */
2059int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2060{
2061 struct mm_struct *mm;
2062
2063 if (addr + len < addr)
2064 return 0;
2065
2066 mm = get_task_mm(tsk);
2067 if (!mm)
2068 return 0;
2069
2070 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2071
2072 mmput(mm);
2073 return len;
2074}
2075
2076/**
2077 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2078 * @inode: The inode to check
2079 * @size: The current filesize of the inode
2080 * @newsize: The proposed filesize of the inode
2081 *
2082 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2083 * make sure that that any outstanding VMAs aren't broken and then shrink the
2084 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2085 * automatically grant mappings that are too large.
2086 */
2087int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2088 size_t newsize)
2089{
2090 struct vm_area_struct *vma;
2091 struct vm_region *region;
2092 pgoff_t low, high;
2093 size_t r_size, r_top;
2094
2095 low = newsize >> PAGE_SHIFT;
2096 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2097
2098 down_write(&nommu_region_sem);
2099 mutex_lock(&inode->i_mapping->i_mmap_mutex);
2100
2101 /* search for VMAs that fall within the dead zone */
2102 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
2103 /* found one - only interested if it's shared out of the page
2104 * cache */
2105 if (vma->vm_flags & VM_SHARED) {
2106 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2107 up_write(&nommu_region_sem);
2108 return -ETXTBSY; /* not quite true, but near enough */
2109 }
2110 }
2111
2112 /* reduce any regions that overlap the dead zone - if in existence,
2113 * these will be pointed to by VMAs that don't overlap the dead zone
2114 *
2115 * we don't check for any regions that start beyond the EOF as there
2116 * shouldn't be any
2117 */
2118 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap,
2119 0, ULONG_MAX) {
2120 if (!(vma->vm_flags & VM_SHARED))
2121 continue;
2122
2123 region = vma->vm_region;
2124 r_size = region->vm_top - region->vm_start;
2125 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2126
2127 if (r_top > newsize) {
2128 region->vm_top -= r_top - newsize;
2129 if (region->vm_end > region->vm_top)
2130 region->vm_end = region->vm_top;
2131 }
2132 }
2133
2134 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2135 up_write(&nommu_region_sem);
2136 return 0;
2137}
2138
2139/*
2140 * Initialise sysctl_user_reserve_kbytes.
2141 *
2142 * This is intended to prevent a user from starting a single memory hogging
2143 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2144 * mode.
2145 *
2146 * The default value is min(3% of free memory, 128MB)
2147 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2148 */
2149static int __meminit init_user_reserve(void)
2150{
2151 unsigned long free_kbytes;
2152
2153 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2154
2155 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2156 return 0;
2157}
2158module_init(init_user_reserve)
2159
2160/*
2161 * Initialise sysctl_admin_reserve_kbytes.
2162 *
2163 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2164 * to log in and kill a memory hogging process.
2165 *
2166 * Systems with more than 256MB will reserve 8MB, enough to recover
2167 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2168 * only reserve 3% of free pages by default.
2169 */
2170static int __meminit init_admin_reserve(void)
2171{
2172 unsigned long free_kbytes;
2173
2174 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2175
2176 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2177 return 0;
2178}
2179module_init(init_admin_reserve)