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