audit mmap
[linux-2.6.git] / mm / nommu.c
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) 2007-2009 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/tracehook.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mount.h>
29 #include <linux/personality.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32 #include <linux/audit.h>
33
34 #include <asm/uaccess.h>
35 #include <asm/tlb.h>
36 #include <asm/tlbflush.h>
37 #include <asm/mmu_context.h>
38 #include "internal.h"
39
40 #if 0
41 #define kenter(FMT, ...) \
42         printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
43 #define kleave(FMT, ...) \
44         printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
45 #define kdebug(FMT, ...) \
46         printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
47 #else
48 #define kenter(FMT, ...) \
49         no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
50 #define kleave(FMT, ...) \
51         no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
52 #define kdebug(FMT, ...) \
53         no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
54 #endif
55
56 void *high_memory;
57 struct page *mem_map;
58 unsigned long max_mapnr;
59 unsigned long num_physpages;
60 unsigned long highest_memmap_pfn;
61 struct percpu_counter vm_committed_as;
62 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
63 int sysctl_overcommit_ratio = 50; /* default is 50% */
64 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
65 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
66 int heap_stack_gap = 0;
67
68 atomic_long_t mmap_pages_allocated;
69
70 EXPORT_SYMBOL(mem_map);
71 EXPORT_SYMBOL(num_physpages);
72
73 /* list of mapped, potentially shareable regions */
74 static struct kmem_cache *vm_region_jar;
75 struct rb_root nommu_region_tree = RB_ROOT;
76 DECLARE_RWSEM(nommu_region_sem);
77
78 const struct vm_operations_struct generic_file_vm_ops = {
79 };
80
81 /*
82  * Return the total memory allocated for this pointer, not
83  * just what the caller asked for.
84  *
85  * Doesn't have to be accurate, i.e. may have races.
86  */
87 unsigned int kobjsize(const void *objp)
88 {
89         struct page *page;
90
91         /*
92          * If the object we have should not have ksize performed on it,
93          * return size of 0
94          */
95         if (!objp || !virt_addr_valid(objp))
96                 return 0;
97
98         page = virt_to_head_page(objp);
99
100         /*
101          * If the allocator sets PageSlab, we know the pointer came from
102          * kmalloc().
103          */
104         if (PageSlab(page))
105                 return ksize(objp);
106
107         /*
108          * If it's not a compound page, see if we have a matching VMA
109          * region. This test is intentionally done in reverse order,
110          * so if there's no VMA, we still fall through and hand back
111          * PAGE_SIZE for 0-order pages.
112          */
113         if (!PageCompound(page)) {
114                 struct vm_area_struct *vma;
115
116                 vma = find_vma(current->mm, (unsigned long)objp);
117                 if (vma)
118                         return vma->vm_end - vma->vm_start;
119         }
120
121         /*
122          * The ksize() function is only guaranteed to work for pointers
123          * returned by kmalloc(). So handle arbitrary pointers here.
124          */
125         return PAGE_SIZE << compound_order(page);
126 }
127
128 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
129                      unsigned long start, int nr_pages, unsigned int foll_flags,
130                      struct page **pages, struct vm_area_struct **vmas)
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
166 finish_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  */
177 int 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 }
190 EXPORT_SYMBOL(get_user_pages);
191
192 /**
193  * follow_pfn - look up PFN at a user virtual address
194  * @vma: memory mapping
195  * @address: user virtual address
196  * @pfn: location to store found PFN
197  *
198  * Only IO mappings and raw PFN mappings are allowed.
199  *
200  * Returns zero and the pfn at @pfn on success, -ve otherwise.
201  */
202 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
203         unsigned long *pfn)
204 {
205         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
206                 return -EINVAL;
207
208         *pfn = address >> PAGE_SHIFT;
209         return 0;
210 }
211 EXPORT_SYMBOL(follow_pfn);
212
213 DEFINE_RWLOCK(vmlist_lock);
214 struct vm_struct *vmlist;
215
216 void vfree(const void *addr)
217 {
218         kfree(addr);
219 }
220 EXPORT_SYMBOL(vfree);
221
222 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
223 {
224         /*
225          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
226          * returns only a logical address.
227          */
228         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
229 }
230 EXPORT_SYMBOL(__vmalloc);
231
232 void *vmalloc_user(unsigned long size)
233 {
234         void *ret;
235
236         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
237                         PAGE_KERNEL);
238         if (ret) {
239                 struct vm_area_struct *vma;
240
241                 down_write(&current->mm->mmap_sem);
242                 vma = find_vma(current->mm, (unsigned long)ret);
243                 if (vma)
244                         vma->vm_flags |= VM_USERMAP;
245                 up_write(&current->mm->mmap_sem);
246         }
247
248         return ret;
249 }
250 EXPORT_SYMBOL(vmalloc_user);
251
252 struct page *vmalloc_to_page(const void *addr)
253 {
254         return virt_to_page(addr);
255 }
256 EXPORT_SYMBOL(vmalloc_to_page);
257
258 unsigned long vmalloc_to_pfn(const void *addr)
259 {
260         return page_to_pfn(virt_to_page(addr));
261 }
262 EXPORT_SYMBOL(vmalloc_to_pfn);
263
264 long vread(char *buf, char *addr, unsigned long count)
265 {
266         memcpy(buf, addr, count);
267         return count;
268 }
269
270 long vwrite(char *buf, char *addr, unsigned long count)
271 {
272         /* Don't allow overflow */
273         if ((unsigned long) addr + count < count)
274                 count = -(unsigned long) addr;
275
276         memcpy(addr, buf, count);
277         return(count);
278 }
279
280 /*
281  *      vmalloc  -  allocate virtually continguos memory
282  *
283  *      @size:          allocation size
284  *
285  *      Allocate enough pages to cover @size from the page level
286  *      allocator and map them into continguos kernel virtual space.
287  *
288  *      For tight control over page level allocator and protection flags
289  *      use __vmalloc() instead.
290  */
291 void *vmalloc(unsigned long size)
292 {
293        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
294 }
295 EXPORT_SYMBOL(vmalloc);
296
297 /*
298  *      vzalloc - allocate virtually continguos memory with zero fill
299  *
300  *      @size:          allocation size
301  *
302  *      Allocate enough pages to cover @size from the page level
303  *      allocator and map them into continguos kernel virtual space.
304  *      The memory allocated is set to zero.
305  *
306  *      For tight control over page level allocator and protection flags
307  *      use __vmalloc() instead.
308  */
309 void *vzalloc(unsigned long size)
310 {
311         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
312                         PAGE_KERNEL);
313 }
314 EXPORT_SYMBOL(vzalloc);
315
316 /**
317  * vmalloc_node - allocate memory on a specific node
318  * @size:       allocation size
319  * @node:       numa node
320  *
321  * Allocate enough pages to cover @size from the page level
322  * allocator and map them into contiguous kernel virtual space.
323  *
324  * For tight control over page level allocator and protection flags
325  * use __vmalloc() instead.
326  */
327 void *vmalloc_node(unsigned long size, int node)
328 {
329         return vmalloc(size);
330 }
331
332 /**
333  * vzalloc_node - allocate memory on a specific node with zero fill
334  * @size:       allocation size
335  * @node:       numa node
336  *
337  * Allocate enough pages to cover @size from the page level
338  * allocator and map them into contiguous kernel virtual space.
339  * The memory allocated is set to zero.
340  *
341  * For tight control over page level allocator and protection flags
342  * use __vmalloc() instead.
343  */
344 void *vzalloc_node(unsigned long size, int node)
345 {
346         return vzalloc(size);
347 }
348 EXPORT_SYMBOL(vzalloc_node);
349
350 #ifndef PAGE_KERNEL_EXEC
351 # define PAGE_KERNEL_EXEC PAGE_KERNEL
352 #endif
353
354 /**
355  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
356  *      @size:          allocation size
357  *
358  *      Kernel-internal function to allocate enough pages to cover @size
359  *      the page level allocator and map them into contiguous and
360  *      executable kernel virtual space.
361  *
362  *      For tight control over page level allocator and protection flags
363  *      use __vmalloc() instead.
364  */
365
366 void *vmalloc_exec(unsigned long size)
367 {
368         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
369 }
370
371 /**
372  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
373  *      @size:          allocation size
374  *
375  *      Allocate enough 32bit PA addressable pages to cover @size from the
376  *      page level allocator and map them into continguos kernel virtual space.
377  */
378 void *vmalloc_32(unsigned long size)
379 {
380         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
381 }
382 EXPORT_SYMBOL(vmalloc_32);
383
384 /**
385  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
386  *      @size:          allocation size
387  *
388  * The resulting memory area is 32bit addressable and zeroed so it can be
389  * mapped to userspace without leaking data.
390  *
391  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
392  * remap_vmalloc_range() are permissible.
393  */
394 void *vmalloc_32_user(unsigned long size)
395 {
396         /*
397          * We'll have to sort out the ZONE_DMA bits for 64-bit,
398          * but for now this can simply use vmalloc_user() directly.
399          */
400         return vmalloc_user(size);
401 }
402 EXPORT_SYMBOL(vmalloc_32_user);
403
404 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
405 {
406         BUG();
407         return NULL;
408 }
409 EXPORT_SYMBOL(vmap);
410
411 void vunmap(const void *addr)
412 {
413         BUG();
414 }
415 EXPORT_SYMBOL(vunmap);
416
417 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
418 {
419         BUG();
420         return NULL;
421 }
422 EXPORT_SYMBOL(vm_map_ram);
423
424 void vm_unmap_ram(const void *mem, unsigned int count)
425 {
426         BUG();
427 }
428 EXPORT_SYMBOL(vm_unmap_ram);
429
430 void vm_unmap_aliases(void)
431 {
432 }
433 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
434
435 /*
436  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
437  * have one.
438  */
439 void  __attribute__((weak)) vmalloc_sync_all(void)
440 {
441 }
442
443 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
444                    struct page *page)
445 {
446         return -EINVAL;
447 }
448 EXPORT_SYMBOL(vm_insert_page);
449
450 /*
451  *  sys_brk() for the most part doesn't need the global kernel
452  *  lock, except when an application is doing something nasty
453  *  like trying to un-brk an area that has already been mapped
454  *  to a regular file.  in this case, the unmapping will need
455  *  to invoke file system routines that need the global lock.
456  */
457 SYSCALL_DEFINE1(brk, unsigned long, brk)
458 {
459         struct mm_struct *mm = current->mm;
460
461         if (brk < mm->start_brk || brk > mm->context.end_brk)
462                 return mm->brk;
463
464         if (mm->brk == brk)
465                 return mm->brk;
466
467         /*
468          * Always allow shrinking brk
469          */
470         if (brk <= mm->brk) {
471                 mm->brk = brk;
472                 return brk;
473         }
474
475         /*
476          * Ok, looks good - let it rip.
477          */
478         flush_icache_range(mm->brk, brk);
479         return mm->brk = brk;
480 }
481
482 /*
483  * initialise the VMA and region record slabs
484  */
485 void __init mmap_init(void)
486 {
487         int ret;
488
489         ret = percpu_counter_init(&vm_committed_as, 0);
490         VM_BUG_ON(ret);
491         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
492 }
493
494 /*
495  * validate the region tree
496  * - the caller must hold the region lock
497  */
498 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
499 static noinline void validate_nommu_regions(void)
500 {
501         struct vm_region *region, *last;
502         struct rb_node *p, *lastp;
503
504         lastp = rb_first(&nommu_region_tree);
505         if (!lastp)
506                 return;
507
508         last = rb_entry(lastp, struct vm_region, vm_rb);
509         BUG_ON(unlikely(last->vm_end <= last->vm_start));
510         BUG_ON(unlikely(last->vm_top < last->vm_end));
511
512         while ((p = rb_next(lastp))) {
513                 region = rb_entry(p, struct vm_region, vm_rb);
514                 last = rb_entry(lastp, struct vm_region, vm_rb);
515
516                 BUG_ON(unlikely(region->vm_end <= region->vm_start));
517                 BUG_ON(unlikely(region->vm_top < region->vm_end));
518                 BUG_ON(unlikely(region->vm_start < last->vm_top));
519
520                 lastp = p;
521         }
522 }
523 #else
524 static void validate_nommu_regions(void)
525 {
526 }
527 #endif
528
529 /*
530  * add a region into the global tree
531  */
532 static void add_nommu_region(struct vm_region *region)
533 {
534         struct vm_region *pregion;
535         struct rb_node **p, *parent;
536
537         validate_nommu_regions();
538
539         parent = NULL;
540         p = &nommu_region_tree.rb_node;
541         while (*p) {
542                 parent = *p;
543                 pregion = rb_entry(parent, struct vm_region, vm_rb);
544                 if (region->vm_start < pregion->vm_start)
545                         p = &(*p)->rb_left;
546                 else if (region->vm_start > pregion->vm_start)
547                         p = &(*p)->rb_right;
548                 else if (pregion == region)
549                         return;
550                 else
551                         BUG();
552         }
553
554         rb_link_node(&region->vm_rb, parent, p);
555         rb_insert_color(&region->vm_rb, &nommu_region_tree);
556
557         validate_nommu_regions();
558 }
559
560 /*
561  * delete a region from the global tree
562  */
563 static void delete_nommu_region(struct vm_region *region)
564 {
565         BUG_ON(!nommu_region_tree.rb_node);
566
567         validate_nommu_regions();
568         rb_erase(&region->vm_rb, &nommu_region_tree);
569         validate_nommu_regions();
570 }
571
572 /*
573  * free a contiguous series of pages
574  */
575 static void free_page_series(unsigned long from, unsigned long to)
576 {
577         for (; from < to; from += PAGE_SIZE) {
578                 struct page *page = virt_to_page(from);
579
580                 kdebug("- free %lx", from);
581                 atomic_long_dec(&mmap_pages_allocated);
582                 if (page_count(page) != 1)
583                         kdebug("free page %p: refcount not one: %d",
584                                page, page_count(page));
585                 put_page(page);
586         }
587 }
588
589 /*
590  * release a reference to a region
591  * - the caller must hold the region semaphore for writing, which this releases
592  * - the region may not have been added to the tree yet, in which case vm_top
593  *   will equal vm_start
594  */
595 static void __put_nommu_region(struct vm_region *region)
596         __releases(nommu_region_sem)
597 {
598         kenter("%p{%d}", region, region->vm_usage);
599
600         BUG_ON(!nommu_region_tree.rb_node);
601
602         if (--region->vm_usage == 0) {
603                 if (region->vm_top > region->vm_start)
604                         delete_nommu_region(region);
605                 up_write(&nommu_region_sem);
606
607                 if (region->vm_file)
608                         fput(region->vm_file);
609
610                 /* IO memory and memory shared directly out of the pagecache
611                  * from ramfs/tmpfs mustn't be released here */
612                 if (region->vm_flags & VM_MAPPED_COPY) {
613                         kdebug("free series");
614                         free_page_series(region->vm_start, region->vm_top);
615                 }
616                 kmem_cache_free(vm_region_jar, region);
617         } else {
618                 up_write(&nommu_region_sem);
619         }
620 }
621
622 /*
623  * release a reference to a region
624  */
625 static void put_nommu_region(struct vm_region *region)
626 {
627         down_write(&nommu_region_sem);
628         __put_nommu_region(region);
629 }
630
631 /*
632  * update protection on a vma
633  */
634 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
635 {
636 #ifdef CONFIG_MPU
637         struct mm_struct *mm = vma->vm_mm;
638         long start = vma->vm_start & PAGE_MASK;
639         while (start < vma->vm_end) {
640                 protect_page(mm, start, flags);
641                 start += PAGE_SIZE;
642         }
643         update_protections(mm);
644 #endif
645 }
646
647 /*
648  * add a VMA into a process's mm_struct in the appropriate place in the list
649  * and tree and add to the address space's page tree also if not an anonymous
650  * page
651  * - should be called with mm->mmap_sem held writelocked
652  */
653 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
654 {
655         struct vm_area_struct *pvma, **pp, *next;
656         struct address_space *mapping;
657         struct rb_node **p, *parent;
658
659         kenter(",%p", vma);
660
661         BUG_ON(!vma->vm_region);
662
663         mm->map_count++;
664         vma->vm_mm = mm;
665
666         protect_vma(vma, vma->vm_flags);
667
668         /* add the VMA to the mapping */
669         if (vma->vm_file) {
670                 mapping = vma->vm_file->f_mapping;
671
672                 flush_dcache_mmap_lock(mapping);
673                 vma_prio_tree_insert(vma, &mapping->i_mmap);
674                 flush_dcache_mmap_unlock(mapping);
675         }
676
677         /* add the VMA to the tree */
678         parent = NULL;
679         p = &mm->mm_rb.rb_node;
680         while (*p) {
681                 parent = *p;
682                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
683
684                 /* sort by: start addr, end addr, VMA struct addr in that order
685                  * (the latter is necessary as we may get identical VMAs) */
686                 if (vma->vm_start < pvma->vm_start)
687                         p = &(*p)->rb_left;
688                 else if (vma->vm_start > pvma->vm_start)
689                         p = &(*p)->rb_right;
690                 else if (vma->vm_end < pvma->vm_end)
691                         p = &(*p)->rb_left;
692                 else if (vma->vm_end > pvma->vm_end)
693                         p = &(*p)->rb_right;
694                 else if (vma < pvma)
695                         p = &(*p)->rb_left;
696                 else if (vma > pvma)
697                         p = &(*p)->rb_right;
698                 else
699                         BUG();
700         }
701
702         rb_link_node(&vma->vm_rb, parent, p);
703         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
704
705         /* add VMA to the VMA list also */
706         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
707                 if (pvma->vm_start > vma->vm_start)
708                         break;
709                 if (pvma->vm_start < vma->vm_start)
710                         continue;
711                 if (pvma->vm_end < vma->vm_end)
712                         break;
713         }
714
715         next = *pp;
716         *pp = vma;
717         vma->vm_next = next;
718         if (next)
719                 next->vm_prev = vma;
720 }
721
722 /*
723  * delete a VMA from its owning mm_struct and address space
724  */
725 static void delete_vma_from_mm(struct vm_area_struct *vma)
726 {
727         struct vm_area_struct **pp;
728         struct address_space *mapping;
729         struct mm_struct *mm = vma->vm_mm;
730
731         kenter("%p", vma);
732
733         protect_vma(vma, 0);
734
735         mm->map_count--;
736         if (mm->mmap_cache == vma)
737                 mm->mmap_cache = NULL;
738
739         /* remove the VMA from the mapping */
740         if (vma->vm_file) {
741                 mapping = vma->vm_file->f_mapping;
742
743                 flush_dcache_mmap_lock(mapping);
744                 vma_prio_tree_remove(vma, &mapping->i_mmap);
745                 flush_dcache_mmap_unlock(mapping);
746         }
747
748         /* remove from the MM's tree and list */
749         rb_erase(&vma->vm_rb, &mm->mm_rb);
750         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
751                 if (*pp == vma) {
752                         *pp = vma->vm_next;
753                         break;
754                 }
755         }
756
757         vma->vm_mm = NULL;
758 }
759
760 /*
761  * destroy a VMA record
762  */
763 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
764 {
765         kenter("%p", vma);
766         if (vma->vm_ops && vma->vm_ops->close)
767                 vma->vm_ops->close(vma);
768         if (vma->vm_file) {
769                 fput(vma->vm_file);
770                 if (vma->vm_flags & VM_EXECUTABLE)
771                         removed_exe_file_vma(mm);
772         }
773         put_nommu_region(vma->vm_region);
774         kmem_cache_free(vm_area_cachep, vma);
775 }
776
777 /*
778  * look up the first VMA in which addr resides, NULL if none
779  * - should be called with mm->mmap_sem at least held readlocked
780  */
781 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
782 {
783         struct vm_area_struct *vma;
784         struct rb_node *n = mm->mm_rb.rb_node;
785
786         /* check the cache first */
787         vma = mm->mmap_cache;
788         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
789                 return vma;
790
791         /* trawl the tree (there may be multiple mappings in which addr
792          * resides) */
793         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
794                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
795                 if (vma->vm_start > addr)
796                         return NULL;
797                 if (vma->vm_end > addr) {
798                         mm->mmap_cache = vma;
799                         return vma;
800                 }
801         }
802
803         return NULL;
804 }
805 EXPORT_SYMBOL(find_vma);
806
807 /*
808  * find a VMA
809  * - we don't extend stack VMAs under NOMMU conditions
810  */
811 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
812 {
813         return find_vma(mm, addr);
814 }
815
816 /*
817  * expand a stack to a given address
818  * - not supported under NOMMU conditions
819  */
820 int expand_stack(struct vm_area_struct *vma, unsigned long address)
821 {
822         return -ENOMEM;
823 }
824
825 /*
826  * look up the first VMA exactly that exactly matches addr
827  * - should be called with mm->mmap_sem at least held readlocked
828  */
829 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
830                                              unsigned long addr,
831                                              unsigned long len)
832 {
833         struct vm_area_struct *vma;
834         struct rb_node *n = mm->mm_rb.rb_node;
835         unsigned long end = addr + len;
836
837         /* check the cache first */
838         vma = mm->mmap_cache;
839         if (vma && vma->vm_start == addr && vma->vm_end == end)
840                 return vma;
841
842         /* trawl the tree (there may be multiple mappings in which addr
843          * resides) */
844         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
845                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
846                 if (vma->vm_start < addr)
847                         continue;
848                 if (vma->vm_start > addr)
849                         return NULL;
850                 if (vma->vm_end == end) {
851                         mm->mmap_cache = vma;
852                         return vma;
853                 }
854         }
855
856         return NULL;
857 }
858
859 /*
860  * determine whether a mapping should be permitted and, if so, what sort of
861  * mapping we're capable of supporting
862  */
863 static int validate_mmap_request(struct file *file,
864                                  unsigned long addr,
865                                  unsigned long len,
866                                  unsigned long prot,
867                                  unsigned long flags,
868                                  unsigned long pgoff,
869                                  unsigned long *_capabilities)
870 {
871         unsigned long capabilities, rlen;
872         unsigned long reqprot = prot;
873         int ret;
874
875         /* do the simple checks first */
876         if (flags & MAP_FIXED) {
877                 printk(KERN_DEBUG
878                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
879                        current->pid);
880                 return -EINVAL;
881         }
882
883         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
884             (flags & MAP_TYPE) != MAP_SHARED)
885                 return -EINVAL;
886
887         if (!len)
888                 return -EINVAL;
889
890         /* Careful about overflows.. */
891         rlen = PAGE_ALIGN(len);
892         if (!rlen || rlen > TASK_SIZE)
893                 return -ENOMEM;
894
895         /* offset overflow? */
896         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
897                 return -EOVERFLOW;
898
899         if (file) {
900                 /* validate file mapping requests */
901                 struct address_space *mapping;
902
903                 /* files must support mmap */
904                 if (!file->f_op || !file->f_op->mmap)
905                         return -ENODEV;
906
907                 /* work out if what we've got could possibly be shared
908                  * - we support chardevs that provide their own "memory"
909                  * - we support files/blockdevs that are memory backed
910                  */
911                 mapping = file->f_mapping;
912                 if (!mapping)
913                         mapping = file->f_path.dentry->d_inode->i_mapping;
914
915                 capabilities = 0;
916                 if (mapping && mapping->backing_dev_info)
917                         capabilities = mapping->backing_dev_info->capabilities;
918
919                 if (!capabilities) {
920                         /* no explicit capabilities set, so assume some
921                          * defaults */
922                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
923                         case S_IFREG:
924                         case S_IFBLK:
925                                 capabilities = BDI_CAP_MAP_COPY;
926                                 break;
927
928                         case S_IFCHR:
929                                 capabilities =
930                                         BDI_CAP_MAP_DIRECT |
931                                         BDI_CAP_READ_MAP |
932                                         BDI_CAP_WRITE_MAP;
933                                 break;
934
935                         default:
936                                 return -EINVAL;
937                         }
938                 }
939
940                 /* eliminate any capabilities that we can't support on this
941                  * device */
942                 if (!file->f_op->get_unmapped_area)
943                         capabilities &= ~BDI_CAP_MAP_DIRECT;
944                 if (!file->f_op->read)
945                         capabilities &= ~BDI_CAP_MAP_COPY;
946
947                 /* The file shall have been opened with read permission. */
948                 if (!(file->f_mode & FMODE_READ))
949                         return -EACCES;
950
951                 if (flags & MAP_SHARED) {
952                         /* do checks for writing, appending and locking */
953                         if ((prot & PROT_WRITE) &&
954                             !(file->f_mode & FMODE_WRITE))
955                                 return -EACCES;
956
957                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
958                             (file->f_mode & FMODE_WRITE))
959                                 return -EACCES;
960
961                         if (locks_verify_locked(file->f_path.dentry->d_inode))
962                                 return -EAGAIN;
963
964                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
965                                 return -ENODEV;
966
967                         /* we mustn't privatise shared mappings */
968                         capabilities &= ~BDI_CAP_MAP_COPY;
969                 }
970                 else {
971                         /* we're going to read the file into private memory we
972                          * allocate */
973                         if (!(capabilities & BDI_CAP_MAP_COPY))
974                                 return -ENODEV;
975
976                         /* we don't permit a private writable mapping to be
977                          * shared with the backing device */
978                         if (prot & PROT_WRITE)
979                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
980                 }
981
982                 if (capabilities & BDI_CAP_MAP_DIRECT) {
983                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
984                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
985                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
986                             ) {
987                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
988                                 if (flags & MAP_SHARED) {
989                                         printk(KERN_WARNING
990                                                "MAP_SHARED not completely supported on !MMU\n");
991                                         return -EINVAL;
992                                 }
993                         }
994                 }
995
996                 /* handle executable mappings and implied executable
997                  * mappings */
998                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
999                         if (prot & PROT_EXEC)
1000                                 return -EPERM;
1001                 }
1002                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1003                         /* handle implication of PROT_EXEC by PROT_READ */
1004                         if (current->personality & READ_IMPLIES_EXEC) {
1005                                 if (capabilities & BDI_CAP_EXEC_MAP)
1006                                         prot |= PROT_EXEC;
1007                         }
1008                 }
1009                 else if ((prot & PROT_READ) &&
1010                          (prot & PROT_EXEC) &&
1011                          !(capabilities & BDI_CAP_EXEC_MAP)
1012                          ) {
1013                         /* backing file is not executable, try to copy */
1014                         capabilities &= ~BDI_CAP_MAP_DIRECT;
1015                 }
1016         }
1017         else {
1018                 /* anonymous mappings are always memory backed and can be
1019                  * privately mapped
1020                  */
1021                 capabilities = BDI_CAP_MAP_COPY;
1022
1023                 /* handle PROT_EXEC implication by PROT_READ */
1024                 if ((prot & PROT_READ) &&
1025                     (current->personality & READ_IMPLIES_EXEC))
1026                         prot |= PROT_EXEC;
1027         }
1028
1029         /* allow the security API to have its say */
1030         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1031         if (ret < 0)
1032                 return ret;
1033
1034         /* looks okay */
1035         *_capabilities = capabilities;
1036         return 0;
1037 }
1038
1039 /*
1040  * we've determined that we can make the mapping, now translate what we
1041  * now know into VMA flags
1042  */
1043 static unsigned long determine_vm_flags(struct file *file,
1044                                         unsigned long prot,
1045                                         unsigned long flags,
1046                                         unsigned long capabilities)
1047 {
1048         unsigned long vm_flags;
1049
1050         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1051         /* vm_flags |= mm->def_flags; */
1052
1053         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1054                 /* attempt to share read-only copies of mapped file chunks */
1055                 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1056                 if (file && !(prot & PROT_WRITE))
1057                         vm_flags |= VM_MAYSHARE;
1058         } else {
1059                 /* overlay a shareable mapping on the backing device or inode
1060                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1061                  * romfs/cramfs */
1062                 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1063                 if (flags & MAP_SHARED)
1064                         vm_flags |= VM_SHARED;
1065         }
1066
1067         /* refuse to let anyone share private mappings with this process if
1068          * it's being traced - otherwise breakpoints set in it may interfere
1069          * with another untraced process
1070          */
1071         if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1072                 vm_flags &= ~VM_MAYSHARE;
1073
1074         return vm_flags;
1075 }
1076
1077 /*
1078  * set up a shared mapping on a file (the driver or filesystem provides and
1079  * pins the storage)
1080  */
1081 static int do_mmap_shared_file(struct vm_area_struct *vma)
1082 {
1083         int ret;
1084
1085         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1086         if (ret == 0) {
1087                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1088                 return 0;
1089         }
1090         if (ret != -ENOSYS)
1091                 return ret;
1092
1093         /* getting -ENOSYS indicates that direct mmap isn't possible (as
1094          * opposed to tried but failed) so we can only give a suitable error as
1095          * it's not possible to make a private copy if MAP_SHARED was given */
1096         return -ENODEV;
1097 }
1098
1099 /*
1100  * set up a private mapping or an anonymous shared mapping
1101  */
1102 static int do_mmap_private(struct vm_area_struct *vma,
1103                            struct vm_region *region,
1104                            unsigned long len,
1105                            unsigned long capabilities)
1106 {
1107         struct page *pages;
1108         unsigned long total, point, n, rlen;
1109         void *base;
1110         int ret, order;
1111
1112         /* invoke the file's mapping function so that it can keep track of
1113          * shared mappings on devices or memory
1114          * - VM_MAYSHARE will be set if it may attempt to share
1115          */
1116         if (capabilities & BDI_CAP_MAP_DIRECT) {
1117                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1118                 if (ret == 0) {
1119                         /* shouldn't return success if we're not sharing */
1120                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1121                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1122                         return 0;
1123                 }
1124                 if (ret != -ENOSYS)
1125                         return ret;
1126
1127                 /* getting an ENOSYS error indicates that direct mmap isn't
1128                  * possible (as opposed to tried but failed) so we'll try to
1129                  * make a private copy of the data and map that instead */
1130         }
1131
1132         rlen = PAGE_ALIGN(len);
1133
1134         /* allocate some memory to hold the mapping
1135          * - note that this may not return a page-aligned address if the object
1136          *   we're allocating is smaller than a page
1137          */
1138         order = get_order(rlen);
1139         kdebug("alloc order %d for %lx", order, len);
1140
1141         pages = alloc_pages(GFP_KERNEL, order);
1142         if (!pages)
1143                 goto enomem;
1144
1145         total = 1 << order;
1146         atomic_long_add(total, &mmap_pages_allocated);
1147
1148         point = rlen >> PAGE_SHIFT;
1149
1150         /* we allocated a power-of-2 sized page set, so we may want to trim off
1151          * the excess */
1152         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1153                 while (total > point) {
1154                         order = ilog2(total - point);
1155                         n = 1 << order;
1156                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1157                         atomic_long_sub(n, &mmap_pages_allocated);
1158                         total -= n;
1159                         set_page_refcounted(pages + total);
1160                         __free_pages(pages + total, order);
1161                 }
1162         }
1163
1164         for (point = 1; point < total; point++)
1165                 set_page_refcounted(&pages[point]);
1166
1167         base = page_address(pages);
1168         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1169         region->vm_start = (unsigned long) base;
1170         region->vm_end   = region->vm_start + rlen;
1171         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1172
1173         vma->vm_start = region->vm_start;
1174         vma->vm_end   = region->vm_start + len;
1175
1176         if (vma->vm_file) {
1177                 /* read the contents of a file into the copy */
1178                 mm_segment_t old_fs;
1179                 loff_t fpos;
1180
1181                 fpos = vma->vm_pgoff;
1182                 fpos <<= PAGE_SHIFT;
1183
1184                 old_fs = get_fs();
1185                 set_fs(KERNEL_DS);
1186                 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1187                 set_fs(old_fs);
1188
1189                 if (ret < 0)
1190                         goto error_free;
1191
1192                 /* clear the last little bit */
1193                 if (ret < rlen)
1194                         memset(base + ret, 0, rlen - ret);
1195
1196         }
1197
1198         return 0;
1199
1200 error_free:
1201         free_page_series(region->vm_start, region->vm_end);
1202         region->vm_start = vma->vm_start = 0;
1203         region->vm_end   = vma->vm_end = 0;
1204         region->vm_top   = 0;
1205         return ret;
1206
1207 enomem:
1208         printk("Allocation of length %lu from process %d (%s) failed\n",
1209                len, current->pid, current->comm);
1210         show_free_areas();
1211         return -ENOMEM;
1212 }
1213
1214 /*
1215  * handle mapping creation for uClinux
1216  */
1217 unsigned long do_mmap_pgoff(struct file *file,
1218                             unsigned long addr,
1219                             unsigned long len,
1220                             unsigned long prot,
1221                             unsigned long flags,
1222                             unsigned long pgoff)
1223 {
1224         struct vm_area_struct *vma;
1225         struct vm_region *region;
1226         struct rb_node *rb;
1227         unsigned long capabilities, vm_flags, result;
1228         int ret;
1229
1230         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1231
1232         /* decide whether we should attempt the mapping, and if so what sort of
1233          * mapping */
1234         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1235                                     &capabilities);
1236         if (ret < 0) {
1237                 kleave(" = %d [val]", ret);
1238                 return ret;
1239         }
1240
1241         /* we ignore the address hint */
1242         addr = 0;
1243
1244         /* we've determined that we can make the mapping, now translate what we
1245          * now know into VMA flags */
1246         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1247
1248         /* we're going to need to record the mapping */
1249         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1250         if (!region)
1251                 goto error_getting_region;
1252
1253         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1254         if (!vma)
1255                 goto error_getting_vma;
1256
1257         region->vm_usage = 1;
1258         region->vm_flags = vm_flags;
1259         region->vm_pgoff = pgoff;
1260
1261         INIT_LIST_HEAD(&vma->anon_vma_chain);
1262         vma->vm_flags = vm_flags;
1263         vma->vm_pgoff = pgoff;
1264
1265         if (file) {
1266                 region->vm_file = file;
1267                 get_file(file);
1268                 vma->vm_file = file;
1269                 get_file(file);
1270                 if (vm_flags & VM_EXECUTABLE) {
1271                         added_exe_file_vma(current->mm);
1272                         vma->vm_mm = current->mm;
1273                 }
1274         }
1275
1276         down_write(&nommu_region_sem);
1277
1278         /* if we want to share, we need to check for regions created by other
1279          * mmap() calls that overlap with our proposed mapping
1280          * - we can only share with a superset match on most regular files
1281          * - shared mappings on character devices and memory backed files are
1282          *   permitted to overlap inexactly as far as we are concerned for in
1283          *   these cases, sharing is handled in the driver or filesystem rather
1284          *   than here
1285          */
1286         if (vm_flags & VM_MAYSHARE) {
1287                 struct vm_region *pregion;
1288                 unsigned long pglen, rpglen, pgend, rpgend, start;
1289
1290                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1291                 pgend = pgoff + pglen;
1292
1293                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1294                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1295
1296                         if (!(pregion->vm_flags & VM_MAYSHARE))
1297                                 continue;
1298
1299                         /* search for overlapping mappings on the same file */
1300                         if (pregion->vm_file->f_path.dentry->d_inode !=
1301                             file->f_path.dentry->d_inode)
1302                                 continue;
1303
1304                         if (pregion->vm_pgoff >= pgend)
1305                                 continue;
1306
1307                         rpglen = pregion->vm_end - pregion->vm_start;
1308                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1309                         rpgend = pregion->vm_pgoff + rpglen;
1310                         if (pgoff >= rpgend)
1311                                 continue;
1312
1313                         /* handle inexactly overlapping matches between
1314                          * mappings */
1315                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1316                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1317                                 /* new mapping is not a subset of the region */
1318                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1319                                         goto sharing_violation;
1320                                 continue;
1321                         }
1322
1323                         /* we've found a region we can share */
1324                         pregion->vm_usage++;
1325                         vma->vm_region = pregion;
1326                         start = pregion->vm_start;
1327                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1328                         vma->vm_start = start;
1329                         vma->vm_end = start + len;
1330
1331                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1332                                 kdebug("share copy");
1333                                 vma->vm_flags |= VM_MAPPED_COPY;
1334                         } else {
1335                                 kdebug("share mmap");
1336                                 ret = do_mmap_shared_file(vma);
1337                                 if (ret < 0) {
1338                                         vma->vm_region = NULL;
1339                                         vma->vm_start = 0;
1340                                         vma->vm_end = 0;
1341                                         pregion->vm_usage--;
1342                                         pregion = NULL;
1343                                         goto error_just_free;
1344                                 }
1345                         }
1346                         fput(region->vm_file);
1347                         kmem_cache_free(vm_region_jar, region);
1348                         region = pregion;
1349                         result = start;
1350                         goto share;
1351                 }
1352
1353                 /* obtain the address at which to make a shared mapping
1354                  * - this is the hook for quasi-memory character devices to
1355                  *   tell us the location of a shared mapping
1356                  */
1357                 if (capabilities & BDI_CAP_MAP_DIRECT) {
1358                         addr = file->f_op->get_unmapped_area(file, addr, len,
1359                                                              pgoff, flags);
1360                         if (IS_ERR((void *) addr)) {
1361                                 ret = addr;
1362                                 if (ret != (unsigned long) -ENOSYS)
1363                                         goto error_just_free;
1364
1365                                 /* the driver refused to tell us where to site
1366                                  * the mapping so we'll have to attempt to copy
1367                                  * it */
1368                                 ret = (unsigned long) -ENODEV;
1369                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1370                                         goto error_just_free;
1371
1372                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1373                         } else {
1374                                 vma->vm_start = region->vm_start = addr;
1375                                 vma->vm_end = region->vm_end = addr + len;
1376                         }
1377                 }
1378         }
1379
1380         vma->vm_region = region;
1381
1382         /* set up the mapping
1383          * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1384          */
1385         if (file && vma->vm_flags & VM_SHARED)
1386                 ret = do_mmap_shared_file(vma);
1387         else
1388                 ret = do_mmap_private(vma, region, len, capabilities);
1389         if (ret < 0)
1390                 goto error_just_free;
1391         add_nommu_region(region);
1392
1393         /* clear anonymous mappings that don't ask for uninitialized data */
1394         if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1395                 memset((void *)region->vm_start, 0,
1396                        region->vm_end - region->vm_start);
1397
1398         /* okay... we have a mapping; now we have to register it */
1399         result = vma->vm_start;
1400
1401         current->mm->total_vm += len >> PAGE_SHIFT;
1402
1403 share:
1404         add_vma_to_mm(current->mm, vma);
1405
1406         /* we flush the region from the icache only when the first executable
1407          * mapping of it is made  */
1408         if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1409                 flush_icache_range(region->vm_start, region->vm_end);
1410                 region->vm_icache_flushed = true;
1411         }
1412
1413         up_write(&nommu_region_sem);
1414
1415         kleave(" = %lx", result);
1416         return result;
1417
1418 error_just_free:
1419         up_write(&nommu_region_sem);
1420 error:
1421         if (region->vm_file)
1422                 fput(region->vm_file);
1423         kmem_cache_free(vm_region_jar, region);
1424         if (vma->vm_file)
1425                 fput(vma->vm_file);
1426         if (vma->vm_flags & VM_EXECUTABLE)
1427                 removed_exe_file_vma(vma->vm_mm);
1428         kmem_cache_free(vm_area_cachep, vma);
1429         kleave(" = %d", ret);
1430         return ret;
1431
1432 sharing_violation:
1433         up_write(&nommu_region_sem);
1434         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1435         ret = -EINVAL;
1436         goto error;
1437
1438 error_getting_vma:
1439         kmem_cache_free(vm_region_jar, region);
1440         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1441                " from process %d failed\n",
1442                len, current->pid);
1443         show_free_areas();
1444         return -ENOMEM;
1445
1446 error_getting_region:
1447         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1448                " from process %d failed\n",
1449                len, current->pid);
1450         show_free_areas();
1451         return -ENOMEM;
1452 }
1453 EXPORT_SYMBOL(do_mmap_pgoff);
1454
1455 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1456                 unsigned long, prot, unsigned long, flags,
1457                 unsigned long, fd, unsigned long, pgoff)
1458 {
1459         struct file *file = NULL;
1460         unsigned long retval = -EBADF;
1461
1462         audit_mmap_fd(fd, flags);
1463         if (!(flags & MAP_ANONYMOUS)) {
1464                 file = fget(fd);
1465                 if (!file)
1466                         goto out;
1467         }
1468
1469         flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1470
1471         down_write(&current->mm->mmap_sem);
1472         retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1473         up_write(&current->mm->mmap_sem);
1474
1475         if (file)
1476                 fput(file);
1477 out:
1478         return retval;
1479 }
1480
1481 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1482 struct mmap_arg_struct {
1483         unsigned long addr;
1484         unsigned long len;
1485         unsigned long prot;
1486         unsigned long flags;
1487         unsigned long fd;
1488         unsigned long offset;
1489 };
1490
1491 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1492 {
1493         struct mmap_arg_struct a;
1494
1495         if (copy_from_user(&a, arg, sizeof(a)))
1496                 return -EFAULT;
1497         if (a.offset & ~PAGE_MASK)
1498                 return -EINVAL;
1499
1500         return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1501                               a.offset >> PAGE_SHIFT);
1502 }
1503 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1504
1505 /*
1506  * split a vma into two pieces at address 'addr', a new vma is allocated either
1507  * for the first part or the tail.
1508  */
1509 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1510               unsigned long addr, int new_below)
1511 {
1512         struct vm_area_struct *new;
1513         struct vm_region *region;
1514         unsigned long npages;
1515
1516         kenter("");
1517
1518         /* we're only permitted to split anonymous regions (these should have
1519          * only a single usage on the region) */
1520         if (vma->vm_file)
1521                 return -ENOMEM;
1522
1523         if (mm->map_count >= sysctl_max_map_count)
1524                 return -ENOMEM;
1525
1526         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1527         if (!region)
1528                 return -ENOMEM;
1529
1530         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1531         if (!new) {
1532                 kmem_cache_free(vm_region_jar, region);
1533                 return -ENOMEM;
1534         }
1535
1536         /* most fields are the same, copy all, and then fixup */
1537         *new = *vma;
1538         *region = *vma->vm_region;
1539         new->vm_region = region;
1540
1541         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1542
1543         if (new_below) {
1544                 region->vm_top = region->vm_end = new->vm_end = addr;
1545         } else {
1546                 region->vm_start = new->vm_start = addr;
1547                 region->vm_pgoff = new->vm_pgoff += npages;
1548         }
1549
1550         if (new->vm_ops && new->vm_ops->open)
1551                 new->vm_ops->open(new);
1552
1553         delete_vma_from_mm(vma);
1554         down_write(&nommu_region_sem);
1555         delete_nommu_region(vma->vm_region);
1556         if (new_below) {
1557                 vma->vm_region->vm_start = vma->vm_start = addr;
1558                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1559         } else {
1560                 vma->vm_region->vm_end = vma->vm_end = addr;
1561                 vma->vm_region->vm_top = addr;
1562         }
1563         add_nommu_region(vma->vm_region);
1564         add_nommu_region(new->vm_region);
1565         up_write(&nommu_region_sem);
1566         add_vma_to_mm(mm, vma);
1567         add_vma_to_mm(mm, new);
1568         return 0;
1569 }
1570
1571 /*
1572  * shrink a VMA by removing the specified chunk from either the beginning or
1573  * the end
1574  */
1575 static int shrink_vma(struct mm_struct *mm,
1576                       struct vm_area_struct *vma,
1577                       unsigned long from, unsigned long to)
1578 {
1579         struct vm_region *region;
1580
1581         kenter("");
1582
1583         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1584          * and list */
1585         delete_vma_from_mm(vma);
1586         if (from > vma->vm_start)
1587                 vma->vm_end = from;
1588         else
1589                 vma->vm_start = to;
1590         add_vma_to_mm(mm, vma);
1591
1592         /* cut the backing region down to size */
1593         region = vma->vm_region;
1594         BUG_ON(region->vm_usage != 1);
1595
1596         down_write(&nommu_region_sem);
1597         delete_nommu_region(region);
1598         if (from > region->vm_start) {
1599                 to = region->vm_top;
1600                 region->vm_top = region->vm_end = from;
1601         } else {
1602                 region->vm_start = to;
1603         }
1604         add_nommu_region(region);
1605         up_write(&nommu_region_sem);
1606
1607         free_page_series(from, to);
1608         return 0;
1609 }
1610
1611 /*
1612  * release a mapping
1613  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1614  *   VMA, though it need not cover the whole VMA
1615  */
1616 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1617 {
1618         struct vm_area_struct *vma;
1619         struct rb_node *rb;
1620         unsigned long end = start + len;
1621         int ret;
1622
1623         kenter(",%lx,%zx", start, len);
1624
1625         if (len == 0)
1626                 return -EINVAL;
1627
1628         /* find the first potentially overlapping VMA */
1629         vma = find_vma(mm, start);
1630         if (!vma) {
1631                 static int limit = 0;
1632                 if (limit < 5) {
1633                         printk(KERN_WARNING
1634                                "munmap of memory not mmapped by process %d"
1635                                " (%s): 0x%lx-0x%lx\n",
1636                                current->pid, current->comm,
1637                                start, start + len - 1);
1638                         limit++;
1639                 }
1640                 return -EINVAL;
1641         }
1642
1643         /* we're allowed to split an anonymous VMA but not a file-backed one */
1644         if (vma->vm_file) {
1645                 do {
1646                         if (start > vma->vm_start) {
1647                                 kleave(" = -EINVAL [miss]");
1648                                 return -EINVAL;
1649                         }
1650                         if (end == vma->vm_end)
1651                                 goto erase_whole_vma;
1652                         rb = rb_next(&vma->vm_rb);
1653                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1654                 } while (rb);
1655                 kleave(" = -EINVAL [split file]");
1656                 return -EINVAL;
1657         } else {
1658                 /* the chunk must be a subset of the VMA found */
1659                 if (start == vma->vm_start && end == vma->vm_end)
1660                         goto erase_whole_vma;
1661                 if (start < vma->vm_start || end > vma->vm_end) {
1662                         kleave(" = -EINVAL [superset]");
1663                         return -EINVAL;
1664                 }
1665                 if (start & ~PAGE_MASK) {
1666                         kleave(" = -EINVAL [unaligned start]");
1667                         return -EINVAL;
1668                 }
1669                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1670                         kleave(" = -EINVAL [unaligned split]");
1671                         return -EINVAL;
1672                 }
1673                 if (start != vma->vm_start && end != vma->vm_end) {
1674                         ret = split_vma(mm, vma, start, 1);
1675                         if (ret < 0) {
1676                                 kleave(" = %d [split]", ret);
1677                                 return ret;
1678                         }
1679                 }
1680                 return shrink_vma(mm, vma, start, end);
1681         }
1682
1683 erase_whole_vma:
1684         delete_vma_from_mm(vma);
1685         delete_vma(mm, vma);
1686         kleave(" = 0");
1687         return 0;
1688 }
1689 EXPORT_SYMBOL(do_munmap);
1690
1691 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1692 {
1693         int ret;
1694         struct mm_struct *mm = current->mm;
1695
1696         down_write(&mm->mmap_sem);
1697         ret = do_munmap(mm, addr, len);
1698         up_write(&mm->mmap_sem);
1699         return ret;
1700 }
1701
1702 /*
1703  * release all the mappings made in a process's VM space
1704  */
1705 void exit_mmap(struct mm_struct *mm)
1706 {
1707         struct vm_area_struct *vma;
1708
1709         if (!mm)
1710                 return;
1711
1712         kenter("");
1713
1714         mm->total_vm = 0;
1715
1716         while ((vma = mm->mmap)) {
1717                 mm->mmap = vma->vm_next;
1718                 delete_vma_from_mm(vma);
1719                 delete_vma(mm, vma);
1720         }
1721
1722         kleave("");
1723 }
1724
1725 unsigned long do_brk(unsigned long addr, unsigned long len)
1726 {
1727         return -ENOMEM;
1728 }
1729
1730 /*
1731  * expand (or shrink) an existing mapping, potentially moving it at the same
1732  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1733  *
1734  * under NOMMU conditions, we only permit changing a mapping's size, and only
1735  * as long as it stays within the region allocated by do_mmap_private() and the
1736  * block is not shareable
1737  *
1738  * MREMAP_FIXED is not supported under NOMMU conditions
1739  */
1740 unsigned long do_mremap(unsigned long addr,
1741                         unsigned long old_len, unsigned long new_len,
1742                         unsigned long flags, unsigned long new_addr)
1743 {
1744         struct vm_area_struct *vma;
1745
1746         /* insanity checks first */
1747         if (old_len == 0 || new_len == 0)
1748                 return (unsigned long) -EINVAL;
1749
1750         if (addr & ~PAGE_MASK)
1751                 return -EINVAL;
1752
1753         if (flags & MREMAP_FIXED && new_addr != addr)
1754                 return (unsigned long) -EINVAL;
1755
1756         vma = find_vma_exact(current->mm, addr, old_len);
1757         if (!vma)
1758                 return (unsigned long) -EINVAL;
1759
1760         if (vma->vm_end != vma->vm_start + old_len)
1761                 return (unsigned long) -EFAULT;
1762
1763         if (vma->vm_flags & VM_MAYSHARE)
1764                 return (unsigned long) -EPERM;
1765
1766         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1767                 return (unsigned long) -ENOMEM;
1768
1769         /* all checks complete - do it */
1770         vma->vm_end = vma->vm_start + new_len;
1771         return vma->vm_start;
1772 }
1773 EXPORT_SYMBOL(do_mremap);
1774
1775 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1776                 unsigned long, new_len, unsigned long, flags,
1777                 unsigned long, new_addr)
1778 {
1779         unsigned long ret;
1780
1781         down_write(&current->mm->mmap_sem);
1782         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1783         up_write(&current->mm->mmap_sem);
1784         return ret;
1785 }
1786
1787 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1788                         unsigned int foll_flags)
1789 {
1790         return NULL;
1791 }
1792
1793 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1794                 unsigned long to, unsigned long size, pgprot_t prot)
1795 {
1796         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1797         return 0;
1798 }
1799 EXPORT_SYMBOL(remap_pfn_range);
1800
1801 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1802                         unsigned long pgoff)
1803 {
1804         unsigned int size = vma->vm_end - vma->vm_start;
1805
1806         if (!(vma->vm_flags & VM_USERMAP))
1807                 return -EINVAL;
1808
1809         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1810         vma->vm_end = vma->vm_start + size;
1811
1812         return 0;
1813 }
1814 EXPORT_SYMBOL(remap_vmalloc_range);
1815
1816 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1817 {
1818 }
1819
1820 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1821         unsigned long len, unsigned long pgoff, unsigned long flags)
1822 {
1823         return -ENOMEM;
1824 }
1825
1826 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1827 {
1828 }
1829
1830 void unmap_mapping_range(struct address_space *mapping,
1831                          loff_t const holebegin, loff_t const holelen,
1832                          int even_cows)
1833 {
1834 }
1835 EXPORT_SYMBOL(unmap_mapping_range);
1836
1837 /*
1838  * Check that a process has enough memory to allocate a new virtual
1839  * mapping. 0 means there is enough memory for the allocation to
1840  * succeed and -ENOMEM implies there is not.
1841  *
1842  * We currently support three overcommit policies, which are set via the
1843  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1844  *
1845  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1846  * Additional code 2002 Jul 20 by Robert Love.
1847  *
1848  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1849  *
1850  * Note this is a helper function intended to be used by LSMs which
1851  * wish to use this logic.
1852  */
1853 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1854 {
1855         unsigned long free, allowed;
1856
1857         vm_acct_memory(pages);
1858
1859         /*
1860          * Sometimes we want to use more memory than we have
1861          */
1862         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1863                 return 0;
1864
1865         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1866                 unsigned long n;
1867
1868                 free = global_page_state(NR_FILE_PAGES);
1869                 free += nr_swap_pages;
1870
1871                 /*
1872                  * Any slabs which are created with the
1873                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1874                  * which are reclaimable, under pressure.  The dentry
1875                  * cache and most inode caches should fall into this
1876                  */
1877                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1878
1879                 /*
1880                  * Leave the last 3% for root
1881                  */
1882                 if (!cap_sys_admin)
1883                         free -= free / 32;
1884
1885                 if (free > pages)
1886                         return 0;
1887
1888                 /*
1889                  * nr_free_pages() is very expensive on large systems,
1890                  * only call if we're about to fail.
1891                  */
1892                 n = nr_free_pages();
1893
1894                 /*
1895                  * Leave reserved pages. The pages are not for anonymous pages.
1896                  */
1897                 if (n <= totalreserve_pages)
1898                         goto error;
1899                 else
1900                         n -= totalreserve_pages;
1901
1902                 /*
1903                  * Leave the last 3% for root
1904                  */
1905                 if (!cap_sys_admin)
1906                         n -= n / 32;
1907                 free += n;
1908
1909                 if (free > pages)
1910                         return 0;
1911
1912                 goto error;
1913         }
1914
1915         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1916         /*
1917          * Leave the last 3% for root
1918          */
1919         if (!cap_sys_admin)
1920                 allowed -= allowed / 32;
1921         allowed += total_swap_pages;
1922
1923         /* Don't let a single process grow too big:
1924            leave 3% of the size of this process for other processes */
1925         if (mm)
1926                 allowed -= mm->total_vm / 32;
1927
1928         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1929                 return 0;
1930
1931 error:
1932         vm_unacct_memory(pages);
1933
1934         return -ENOMEM;
1935 }
1936
1937 int in_gate_area_no_task(unsigned long addr)
1938 {
1939         return 0;
1940 }
1941
1942 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1943 {
1944         BUG();
1945         return 0;
1946 }
1947 EXPORT_SYMBOL(filemap_fault);
1948
1949 /*
1950  * Access another process' address space.
1951  * - source/target buffer must be kernel space
1952  */
1953 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1954 {
1955         struct vm_area_struct *vma;
1956         struct mm_struct *mm;
1957
1958         if (addr + len < addr)
1959                 return 0;
1960
1961         mm = get_task_mm(tsk);
1962         if (!mm)
1963                 return 0;
1964
1965         down_read(&mm->mmap_sem);
1966
1967         /* the access must start within one of the target process's mappings */
1968         vma = find_vma(mm, addr);
1969         if (vma) {
1970                 /* don't overrun this mapping */
1971                 if (addr + len >= vma->vm_end)
1972                         len = vma->vm_end - addr;
1973
1974                 /* only read or write mappings where it is permitted */
1975                 if (write && vma->vm_flags & VM_MAYWRITE)
1976                         copy_to_user_page(vma, NULL, addr,
1977                                          (void *) addr, buf, len);
1978                 else if (!write && vma->vm_flags & VM_MAYREAD)
1979                         copy_from_user_page(vma, NULL, addr,
1980                                             buf, (void *) addr, len);
1981                 else
1982                         len = 0;
1983         } else {
1984                 len = 0;
1985         }
1986
1987         up_read(&mm->mmap_sem);
1988         mmput(mm);
1989         return len;
1990 }
1991
1992 /**
1993  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1994  * @inode: The inode to check
1995  * @size: The current filesize of the inode
1996  * @newsize: The proposed filesize of the inode
1997  *
1998  * Check the shared mappings on an inode on behalf of a shrinking truncate to
1999  * make sure that that any outstanding VMAs aren't broken and then shrink the
2000  * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2001  * automatically grant mappings that are too large.
2002  */
2003 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2004                                 size_t newsize)
2005 {
2006         struct vm_area_struct *vma;
2007         struct prio_tree_iter iter;
2008         struct vm_region *region;
2009         pgoff_t low, high;
2010         size_t r_size, r_top;
2011
2012         low = newsize >> PAGE_SHIFT;
2013         high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2014
2015         down_write(&nommu_region_sem);
2016
2017         /* search for VMAs that fall within the dead zone */
2018         vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2019                               low, high) {
2020                 /* found one - only interested if it's shared out of the page
2021                  * cache */
2022                 if (vma->vm_flags & VM_SHARED) {
2023                         up_write(&nommu_region_sem);
2024                         return -ETXTBSY; /* not quite true, but near enough */
2025                 }
2026         }
2027
2028         /* reduce any regions that overlap the dead zone - if in existence,
2029          * these will be pointed to by VMAs that don't overlap the dead zone
2030          *
2031          * we don't check for any regions that start beyond the EOF as there
2032          * shouldn't be any
2033          */
2034         vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2035                               0, ULONG_MAX) {
2036                 if (!(vma->vm_flags & VM_SHARED))
2037                         continue;
2038
2039                 region = vma->vm_region;
2040                 r_size = region->vm_top - region->vm_start;
2041                 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2042
2043                 if (r_top > newsize) {
2044                         region->vm_top -= r_top - newsize;
2045                         if (region->vm_end > region->vm_top)
2046                                 region->vm_end = region->vm_top;
2047                 }
2048         }
2049
2050         up_write(&nommu_region_sem);
2051         return 0;
2052 }