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