Merge tag 'v4.4.69' into linux-linaro-lsk-v4.4
[firefly-linux-kernel-4.4.55.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) 2007-2010 Paul Mundt <lethal@linux-sh.org>
14  */
15
16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17
18 #include <linux/export.h>
19 #include <linux/mm.h>
20 #include <linux/vmacache.h>
21 #include <linux/mman.h>
22 #include <linux/swap.h>
23 #include <linux/file.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <linux/slab.h>
27 #include <linux/vmalloc.h>
28 #include <linux/blkdev.h>
29 #include <linux/backing-dev.h>
30 #include <linux/compiler.h>
31 #include <linux/mount.h>
32 #include <linux/personality.h>
33 #include <linux/security.h>
34 #include <linux/syscalls.h>
35 #include <linux/audit.h>
36 #include <linux/sched/sysctl.h>
37 #include <linux/printk.h>
38
39 #include <asm/uaccess.h>
40 #include <asm/tlb.h>
41 #include <asm/tlbflush.h>
42 #include <asm/mmu_context.h>
43 #include "internal.h"
44
45 void *high_memory;
46 EXPORT_SYMBOL(high_memory);
47 struct page *mem_map;
48 unsigned long max_mapnr;
49 EXPORT_SYMBOL(max_mapnr);
50 unsigned long highest_memmap_pfn;
51 struct percpu_counter vm_committed_as;
52 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
53 int sysctl_overcommit_ratio = 50; /* default is 50% */
54 unsigned long sysctl_overcommit_kbytes __read_mostly;
55 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
56 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
57 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
58 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
59 int heap_stack_gap = 0;
60
61 atomic_long_t mmap_pages_allocated;
62
63 /*
64  * The global memory commitment made in the system can be a metric
65  * that can be used to drive ballooning decisions when Linux is hosted
66  * as a guest. On Hyper-V, the host implements a policy engine for dynamically
67  * balancing memory across competing virtual machines that are hosted.
68  * Several metrics drive this policy engine including the guest reported
69  * memory commitment.
70  */
71 unsigned long vm_memory_committed(void)
72 {
73         return percpu_counter_read_positive(&vm_committed_as);
74 }
75
76 EXPORT_SYMBOL_GPL(vm_memory_committed);
77
78 EXPORT_SYMBOL(mem_map);
79
80 /* list of mapped, potentially shareable regions */
81 static struct kmem_cache *vm_region_jar;
82 struct rb_root nommu_region_tree = RB_ROOT;
83 DECLARE_RWSEM(nommu_region_sem);
84
85 const struct vm_operations_struct generic_file_vm_ops = {
86 };
87
88 /*
89  * Return the total memory allocated for this pointer, not
90  * just what the caller asked for.
91  *
92  * Doesn't have to be accurate, i.e. may have races.
93  */
94 unsigned int kobjsize(const void *objp)
95 {
96         struct page *page;
97
98         /*
99          * If the object we have should not have ksize performed on it,
100          * return size of 0
101          */
102         if (!objp || !virt_addr_valid(objp))
103                 return 0;
104
105         page = virt_to_head_page(objp);
106
107         /*
108          * If the allocator sets PageSlab, we know the pointer came from
109          * kmalloc().
110          */
111         if (PageSlab(page))
112                 return ksize(objp);
113
114         /*
115          * If it's not a compound page, see if we have a matching VMA
116          * region. This test is intentionally done in reverse order,
117          * so if there's no VMA, we still fall through and hand back
118          * PAGE_SIZE for 0-order pages.
119          */
120         if (!PageCompound(page)) {
121                 struct vm_area_struct *vma;
122
123                 vma = find_vma(current->mm, (unsigned long)objp);
124                 if (vma)
125                         return vma->vm_end - vma->vm_start;
126         }
127
128         /*
129          * The ksize() function is only guaranteed to work for pointers
130          * returned by kmalloc(). So handle arbitrary pointers here.
131          */
132         return PAGE_SIZE << compound_order(page);
133 }
134
135 long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
136                       unsigned long start, unsigned long nr_pages,
137                       unsigned int foll_flags, struct page **pages,
138                       struct vm_area_struct **vmas, int *nonblocking)
139 {
140         struct vm_area_struct *vma;
141         unsigned long vm_flags;
142         int i;
143
144         /* calculate required read or write permissions.
145          * If FOLL_FORCE is set, we only require the "MAY" flags.
146          */
147         vm_flags  = (foll_flags & FOLL_WRITE) ?
148                         (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
149         vm_flags &= (foll_flags & FOLL_FORCE) ?
150                         (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
151
152         for (i = 0; i < nr_pages; i++) {
153                 vma = find_vma(mm, start);
154                 if (!vma)
155                         goto finish_or_fault;
156
157                 /* protect what we can, including chardevs */
158                 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
159                     !(vm_flags & vma->vm_flags))
160                         goto finish_or_fault;
161
162                 if (pages) {
163                         pages[i] = virt_to_page(start);
164                         if (pages[i])
165                                 page_cache_get(pages[i]);
166                 }
167                 if (vmas)
168                         vmas[i] = vma;
169                 start = (start + PAGE_SIZE) & PAGE_MASK;
170         }
171
172         return i;
173
174 finish_or_fault:
175         return i ? : -EFAULT;
176 }
177
178 /*
179  * get a list of pages in an address range belonging to the specified process
180  * and indicate the VMA that covers each page
181  * - this is potentially dodgy as we may end incrementing the page count of a
182  *   slab page or a secondary page from a compound page
183  * - don't permit access to VMAs that don't support it, such as I/O mappings
184  */
185 long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
186                     unsigned long start, unsigned long nr_pages,
187                     int write, int force, struct page **pages,
188                     struct vm_area_struct **vmas)
189 {
190         int flags = 0;
191
192         if (write)
193                 flags |= FOLL_WRITE;
194         if (force)
195                 flags |= FOLL_FORCE;
196
197         return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
198                                 NULL);
199 }
200 EXPORT_SYMBOL(get_user_pages);
201
202 long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm,
203                            unsigned long start, unsigned long nr_pages,
204                            int write, int force, struct page **pages,
205                            int *locked)
206 {
207         return get_user_pages(tsk, mm, start, nr_pages, write, force,
208                               pages, NULL);
209 }
210 EXPORT_SYMBOL(get_user_pages_locked);
211
212 long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
213                                unsigned long start, unsigned long nr_pages,
214                                int write, int force, struct page **pages,
215                                unsigned int gup_flags)
216 {
217         long ret;
218         down_read(&mm->mmap_sem);
219         ret = get_user_pages(tsk, mm, start, nr_pages, write, force,
220                              pages, NULL);
221         up_read(&mm->mmap_sem);
222         return ret;
223 }
224 EXPORT_SYMBOL(__get_user_pages_unlocked);
225
226 long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
227                              unsigned long start, unsigned long nr_pages,
228                              int write, int force, struct page **pages)
229 {
230         return __get_user_pages_unlocked(tsk, mm, start, nr_pages, write,
231                                          force, pages, 0);
232 }
233 EXPORT_SYMBOL(get_user_pages_unlocked);
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 LIST_HEAD(vmap_area_list);
257
258 void vfree(const void *addr)
259 {
260         kfree(addr);
261 }
262 EXPORT_SYMBOL(vfree);
263
264 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
265 {
266         /*
267          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
268          * returns only a logical address.
269          */
270         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
271 }
272 EXPORT_SYMBOL(__vmalloc);
273
274 void *vmalloc_user(unsigned long size)
275 {
276         void *ret;
277
278         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
279                         PAGE_KERNEL);
280         if (ret) {
281                 struct vm_area_struct *vma;
282
283                 down_write(&current->mm->mmap_sem);
284                 vma = find_vma(current->mm, (unsigned long)ret);
285                 if (vma)
286                         vma->vm_flags |= VM_USERMAP;
287                 up_write(&current->mm->mmap_sem);
288         }
289
290         return ret;
291 }
292 EXPORT_SYMBOL(vmalloc_user);
293
294 struct page *vmalloc_to_page(const void *addr)
295 {
296         return virt_to_page(addr);
297 }
298 EXPORT_SYMBOL(vmalloc_to_page);
299
300 unsigned long vmalloc_to_pfn(const void *addr)
301 {
302         return page_to_pfn(virt_to_page(addr));
303 }
304 EXPORT_SYMBOL(vmalloc_to_pfn);
305
306 long vread(char *buf, char *addr, unsigned long count)
307 {
308         /* Don't allow overflow */
309         if ((unsigned long) buf + count < count)
310                 count = -(unsigned long) buf;
311
312         memcpy(buf, addr, count);
313         return count;
314 }
315
316 long vwrite(char *buf, char *addr, unsigned long count)
317 {
318         /* Don't allow overflow */
319         if ((unsigned long) addr + count < count)
320                 count = -(unsigned long) addr;
321
322         memcpy(addr, buf, count);
323         return count;
324 }
325
326 /*
327  *      vmalloc  -  allocate virtually contiguous memory
328  *
329  *      @size:          allocation size
330  *
331  *      Allocate enough pages to cover @size from the page level
332  *      allocator and map them into contiguous kernel virtual space.
333  *
334  *      For tight control over page level allocator and protection flags
335  *      use __vmalloc() instead.
336  */
337 void *vmalloc(unsigned long size)
338 {
339        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
340 }
341 EXPORT_SYMBOL(vmalloc);
342
343 /*
344  *      vzalloc - allocate virtually contiguous memory with zero fill
345  *
346  *      @size:          allocation size
347  *
348  *      Allocate enough pages to cover @size from the page level
349  *      allocator and map them into contiguous kernel virtual space.
350  *      The memory allocated is set to zero.
351  *
352  *      For tight control over page level allocator and protection flags
353  *      use __vmalloc() instead.
354  */
355 void *vzalloc(unsigned long size)
356 {
357         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
358                         PAGE_KERNEL);
359 }
360 EXPORT_SYMBOL(vzalloc);
361
362 /**
363  * vmalloc_node - allocate memory on a specific node
364  * @size:       allocation size
365  * @node:       numa node
366  *
367  * Allocate enough pages to cover @size from the page level
368  * allocator and map them into contiguous kernel virtual space.
369  *
370  * For tight control over page level allocator and protection flags
371  * use __vmalloc() instead.
372  */
373 void *vmalloc_node(unsigned long size, int node)
374 {
375         return vmalloc(size);
376 }
377 EXPORT_SYMBOL(vmalloc_node);
378
379 /**
380  * vzalloc_node - allocate memory on a specific node with zero fill
381  * @size:       allocation size
382  * @node:       numa node
383  *
384  * Allocate enough pages to cover @size from the page level
385  * allocator and map them into contiguous kernel virtual space.
386  * The memory allocated is set to zero.
387  *
388  * For tight control over page level allocator and protection flags
389  * use __vmalloc() instead.
390  */
391 void *vzalloc_node(unsigned long size, int node)
392 {
393         return vzalloc(size);
394 }
395 EXPORT_SYMBOL(vzalloc_node);
396
397 #ifndef PAGE_KERNEL_EXEC
398 # define PAGE_KERNEL_EXEC PAGE_KERNEL
399 #endif
400
401 /**
402  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
403  *      @size:          allocation size
404  *
405  *      Kernel-internal function to allocate enough pages to cover @size
406  *      the page level allocator and map them into contiguous and
407  *      executable kernel virtual space.
408  *
409  *      For tight control over page level allocator and protection flags
410  *      use __vmalloc() instead.
411  */
412
413 void *vmalloc_exec(unsigned long size)
414 {
415         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
416 }
417
418 /**
419  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
420  *      @size:          allocation size
421  *
422  *      Allocate enough 32bit PA addressable pages to cover @size from the
423  *      page level allocator and map them into contiguous kernel virtual space.
424  */
425 void *vmalloc_32(unsigned long size)
426 {
427         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
428 }
429 EXPORT_SYMBOL(vmalloc_32);
430
431 /**
432  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
433  *      @size:          allocation size
434  *
435  * The resulting memory area is 32bit addressable and zeroed so it can be
436  * mapped to userspace without leaking data.
437  *
438  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
439  * remap_vmalloc_range() are permissible.
440  */
441 void *vmalloc_32_user(unsigned long size)
442 {
443         /*
444          * We'll have to sort out the ZONE_DMA bits for 64-bit,
445          * but for now this can simply use vmalloc_user() directly.
446          */
447         return vmalloc_user(size);
448 }
449 EXPORT_SYMBOL(vmalloc_32_user);
450
451 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
452 {
453         BUG();
454         return NULL;
455 }
456 EXPORT_SYMBOL(vmap);
457
458 void vunmap(const void *addr)
459 {
460         BUG();
461 }
462 EXPORT_SYMBOL(vunmap);
463
464 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
465 {
466         BUG();
467         return NULL;
468 }
469 EXPORT_SYMBOL(vm_map_ram);
470
471 void vm_unmap_ram(const void *mem, unsigned int count)
472 {
473         BUG();
474 }
475 EXPORT_SYMBOL(vm_unmap_ram);
476
477 void vm_unmap_aliases(void)
478 {
479 }
480 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
481
482 /*
483  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
484  * have one.
485  */
486 void __weak vmalloc_sync_all(void)
487 {
488 }
489
490 /**
491  *      alloc_vm_area - allocate a range of kernel address space
492  *      @size:          size of the area
493  *
494  *      Returns:        NULL on failure, vm_struct on success
495  *
496  *      This function reserves a range of kernel address space, and
497  *      allocates pagetables to map that range.  No actual mappings
498  *      are created.  If the kernel address space is not shared
499  *      between processes, it syncs the pagetable across all
500  *      processes.
501  */
502 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
503 {
504         BUG();
505         return NULL;
506 }
507 EXPORT_SYMBOL_GPL(alloc_vm_area);
508
509 void free_vm_area(struct vm_struct *area)
510 {
511         BUG();
512 }
513 EXPORT_SYMBOL_GPL(free_vm_area);
514
515 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
516                    struct page *page)
517 {
518         return -EINVAL;
519 }
520 EXPORT_SYMBOL(vm_insert_page);
521
522 /*
523  *  sys_brk() for the most part doesn't need the global kernel
524  *  lock, except when an application is doing something nasty
525  *  like trying to un-brk an area that has already been mapped
526  *  to a regular file.  in this case, the unmapping will need
527  *  to invoke file system routines that need the global lock.
528  */
529 SYSCALL_DEFINE1(brk, unsigned long, brk)
530 {
531         struct mm_struct *mm = current->mm;
532
533         if (brk < mm->start_brk || brk > mm->context.end_brk)
534                 return mm->brk;
535
536         if (mm->brk == brk)
537                 return mm->brk;
538
539         /*
540          * Always allow shrinking brk
541          */
542         if (brk <= mm->brk) {
543                 mm->brk = brk;
544                 return brk;
545         }
546
547         /*
548          * Ok, looks good - let it rip.
549          */
550         flush_icache_range(mm->brk, brk);
551         return mm->brk = brk;
552 }
553
554 /*
555  * initialise the VMA and region record slabs
556  */
557 void __init mmap_init(void)
558 {
559         int ret;
560
561         ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
562         VM_BUG_ON(ret);
563         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
564 }
565
566 /*
567  * validate the region tree
568  * - the caller must hold the region lock
569  */
570 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
571 static noinline void validate_nommu_regions(void)
572 {
573         struct vm_region *region, *last;
574         struct rb_node *p, *lastp;
575
576         lastp = rb_first(&nommu_region_tree);
577         if (!lastp)
578                 return;
579
580         last = rb_entry(lastp, struct vm_region, vm_rb);
581         BUG_ON(last->vm_end <= last->vm_start);
582         BUG_ON(last->vm_top < last->vm_end);
583
584         while ((p = rb_next(lastp))) {
585                 region = rb_entry(p, struct vm_region, vm_rb);
586                 last = rb_entry(lastp, struct vm_region, vm_rb);
587
588                 BUG_ON(region->vm_end <= region->vm_start);
589                 BUG_ON(region->vm_top < region->vm_end);
590                 BUG_ON(region->vm_start < last->vm_top);
591
592                 lastp = p;
593         }
594 }
595 #else
596 static void validate_nommu_regions(void)
597 {
598 }
599 #endif
600
601 /*
602  * add a region into the global tree
603  */
604 static void add_nommu_region(struct vm_region *region)
605 {
606         struct vm_region *pregion;
607         struct rb_node **p, *parent;
608
609         validate_nommu_regions();
610
611         parent = NULL;
612         p = &nommu_region_tree.rb_node;
613         while (*p) {
614                 parent = *p;
615                 pregion = rb_entry(parent, struct vm_region, vm_rb);
616                 if (region->vm_start < pregion->vm_start)
617                         p = &(*p)->rb_left;
618                 else if (region->vm_start > pregion->vm_start)
619                         p = &(*p)->rb_right;
620                 else if (pregion == region)
621                         return;
622                 else
623                         BUG();
624         }
625
626         rb_link_node(&region->vm_rb, parent, p);
627         rb_insert_color(&region->vm_rb, &nommu_region_tree);
628
629         validate_nommu_regions();
630 }
631
632 /*
633  * delete a region from the global tree
634  */
635 static void delete_nommu_region(struct vm_region *region)
636 {
637         BUG_ON(!nommu_region_tree.rb_node);
638
639         validate_nommu_regions();
640         rb_erase(&region->vm_rb, &nommu_region_tree);
641         validate_nommu_regions();
642 }
643
644 /*
645  * free a contiguous series of pages
646  */
647 static void free_page_series(unsigned long from, unsigned long to)
648 {
649         for (; from < to; from += PAGE_SIZE) {
650                 struct page *page = virt_to_page(from);
651
652                 atomic_long_dec(&mmap_pages_allocated);
653                 put_page(page);
654         }
655 }
656
657 /*
658  * release a reference to a region
659  * - the caller must hold the region semaphore for writing, which this releases
660  * - the region may not have been added to the tree yet, in which case vm_top
661  *   will equal vm_start
662  */
663 static void __put_nommu_region(struct vm_region *region)
664         __releases(nommu_region_sem)
665 {
666         BUG_ON(!nommu_region_tree.rb_node);
667
668         if (--region->vm_usage == 0) {
669                 if (region->vm_top > region->vm_start)
670                         delete_nommu_region(region);
671                 up_write(&nommu_region_sem);
672
673                 if (region->vm_file)
674                         fput(region->vm_file);
675
676                 /* IO memory and memory shared directly out of the pagecache
677                  * from ramfs/tmpfs mustn't be released here */
678                 if (region->vm_flags & VM_MAPPED_COPY)
679                         free_page_series(region->vm_start, region->vm_top);
680                 kmem_cache_free(vm_region_jar, region);
681         } else {
682                 up_write(&nommu_region_sem);
683         }
684 }
685
686 /*
687  * release a reference to a region
688  */
689 static void put_nommu_region(struct vm_region *region)
690 {
691         down_write(&nommu_region_sem);
692         __put_nommu_region(region);
693 }
694
695 /*
696  * update protection on a vma
697  */
698 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
699 {
700 #ifdef CONFIG_MPU
701         struct mm_struct *mm = vma->vm_mm;
702         long start = vma->vm_start & PAGE_MASK;
703         while (start < vma->vm_end) {
704                 protect_page(mm, start, flags);
705                 start += PAGE_SIZE;
706         }
707         update_protections(mm);
708 #endif
709 }
710
711 /*
712  * add a VMA into a process's mm_struct in the appropriate place in the list
713  * and tree and add to the address space's page tree also if not an anonymous
714  * page
715  * - should be called with mm->mmap_sem held writelocked
716  */
717 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
718 {
719         struct vm_area_struct *pvma, *prev;
720         struct address_space *mapping;
721         struct rb_node **p, *parent, *rb_prev;
722
723         BUG_ON(!vma->vm_region);
724
725         mm->map_count++;
726         vma->vm_mm = mm;
727
728         protect_vma(vma, vma->vm_flags);
729
730         /* add the VMA to the mapping */
731         if (vma->vm_file) {
732                 mapping = vma->vm_file->f_mapping;
733
734                 i_mmap_lock_write(mapping);
735                 flush_dcache_mmap_lock(mapping);
736                 vma_interval_tree_insert(vma, &mapping->i_mmap);
737                 flush_dcache_mmap_unlock(mapping);
738                 i_mmap_unlock_write(mapping);
739         }
740
741         /* add the VMA to the tree */
742         parent = rb_prev = NULL;
743         p = &mm->mm_rb.rb_node;
744         while (*p) {
745                 parent = *p;
746                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
747
748                 /* sort by: start addr, end addr, VMA struct addr in that order
749                  * (the latter is necessary as we may get identical VMAs) */
750                 if (vma->vm_start < pvma->vm_start)
751                         p = &(*p)->rb_left;
752                 else if (vma->vm_start > pvma->vm_start) {
753                         rb_prev = parent;
754                         p = &(*p)->rb_right;
755                 } else if (vma->vm_end < pvma->vm_end)
756                         p = &(*p)->rb_left;
757                 else if (vma->vm_end > pvma->vm_end) {
758                         rb_prev = parent;
759                         p = &(*p)->rb_right;
760                 } else if (vma < pvma)
761                         p = &(*p)->rb_left;
762                 else if (vma > pvma) {
763                         rb_prev = parent;
764                         p = &(*p)->rb_right;
765                 } else
766                         BUG();
767         }
768
769         rb_link_node(&vma->vm_rb, parent, p);
770         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
771
772         /* add VMA to the VMA list also */
773         prev = NULL;
774         if (rb_prev)
775                 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
776
777         __vma_link_list(mm, vma, prev, parent);
778 }
779
780 /*
781  * delete a VMA from its owning mm_struct and address space
782  */
783 static void delete_vma_from_mm(struct vm_area_struct *vma)
784 {
785         int i;
786         struct address_space *mapping;
787         struct mm_struct *mm = vma->vm_mm;
788         struct task_struct *curr = current;
789
790         protect_vma(vma, 0);
791
792         mm->map_count--;
793         for (i = 0; i < VMACACHE_SIZE; i++) {
794                 /* if the vma is cached, invalidate the entire cache */
795                 if (curr->vmacache[i] == vma) {
796                         vmacache_invalidate(mm);
797                         break;
798                 }
799         }
800
801         /* remove the VMA from the mapping */
802         if (vma->vm_file) {
803                 mapping = vma->vm_file->f_mapping;
804
805                 i_mmap_lock_write(mapping);
806                 flush_dcache_mmap_lock(mapping);
807                 vma_interval_tree_remove(vma, &mapping->i_mmap);
808                 flush_dcache_mmap_unlock(mapping);
809                 i_mmap_unlock_write(mapping);
810         }
811
812         /* remove from the MM's tree and list */
813         rb_erase(&vma->vm_rb, &mm->mm_rb);
814
815         if (vma->vm_prev)
816                 vma->vm_prev->vm_next = vma->vm_next;
817         else
818                 mm->mmap = vma->vm_next;
819
820         if (vma->vm_next)
821                 vma->vm_next->vm_prev = vma->vm_prev;
822 }
823
824 /*
825  * destroy a VMA record
826  */
827 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
828 {
829         if (vma->vm_ops && vma->vm_ops->close)
830                 vma->vm_ops->close(vma);
831         if (vma->vm_file)
832                 fput(vma->vm_file);
833         put_nommu_region(vma->vm_region);
834         kmem_cache_free(vm_area_cachep, vma);
835 }
836
837 /*
838  * look up the first VMA in which addr resides, NULL if none
839  * - should be called with mm->mmap_sem at least held readlocked
840  */
841 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
842 {
843         struct vm_area_struct *vma;
844
845         /* check the cache first */
846         vma = vmacache_find(mm, addr);
847         if (likely(vma))
848                 return vma;
849
850         /* trawl the list (there may be multiple mappings in which addr
851          * resides) */
852         for (vma = mm->mmap; vma; vma = vma->vm_next) {
853                 if (vma->vm_start > addr)
854                         return NULL;
855                 if (vma->vm_end > addr) {
856                         vmacache_update(addr, vma);
857                         return vma;
858                 }
859         }
860
861         return NULL;
862 }
863 EXPORT_SYMBOL(find_vma);
864
865 /*
866  * find a VMA
867  * - we don't extend stack VMAs under NOMMU conditions
868  */
869 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
870 {
871         return find_vma(mm, addr);
872 }
873
874 /*
875  * expand a stack to a given address
876  * - not supported under NOMMU conditions
877  */
878 int expand_stack(struct vm_area_struct *vma, unsigned long address)
879 {
880         return -ENOMEM;
881 }
882
883 /*
884  * look up the first VMA exactly that exactly matches addr
885  * - should be called with mm->mmap_sem at least held readlocked
886  */
887 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
888                                              unsigned long addr,
889                                              unsigned long len)
890 {
891         struct vm_area_struct *vma;
892         unsigned long end = addr + len;
893
894         /* check the cache first */
895         vma = vmacache_find_exact(mm, addr, end);
896         if (vma)
897                 return vma;
898
899         /* trawl the list (there may be multiple mappings in which addr
900          * resides) */
901         for (vma = mm->mmap; vma; vma = vma->vm_next) {
902                 if (vma->vm_start < addr)
903                         continue;
904                 if (vma->vm_start > addr)
905                         return NULL;
906                 if (vma->vm_end == end) {
907                         vmacache_update(addr, vma);
908                         return vma;
909                 }
910         }
911
912         return NULL;
913 }
914
915 /*
916  * determine whether a mapping should be permitted and, if so, what sort of
917  * mapping we're capable of supporting
918  */
919 static int validate_mmap_request(struct file *file,
920                                  unsigned long addr,
921                                  unsigned long len,
922                                  unsigned long prot,
923                                  unsigned long flags,
924                                  unsigned long pgoff,
925                                  unsigned long *_capabilities)
926 {
927         unsigned long capabilities, rlen;
928         int ret;
929
930         /* do the simple checks first */
931         if (flags & MAP_FIXED)
932                 return -EINVAL;
933
934         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
935             (flags & MAP_TYPE) != MAP_SHARED)
936                 return -EINVAL;
937
938         if (!len)
939                 return -EINVAL;
940
941         /* Careful about overflows.. */
942         rlen = PAGE_ALIGN(len);
943         if (!rlen || rlen > TASK_SIZE)
944                 return -ENOMEM;
945
946         /* offset overflow? */
947         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
948                 return -EOVERFLOW;
949
950         if (file) {
951                 /* files must support mmap */
952                 if (!file->f_op->mmap)
953                         return -ENODEV;
954
955                 /* work out if what we've got could possibly be shared
956                  * - we support chardevs that provide their own "memory"
957                  * - we support files/blockdevs that are memory backed
958                  */
959                 if (file->f_op->mmap_capabilities) {
960                         capabilities = file->f_op->mmap_capabilities(file);
961                 } else {
962                         /* no explicit capabilities set, so assume some
963                          * defaults */
964                         switch (file_inode(file)->i_mode & S_IFMT) {
965                         case S_IFREG:
966                         case S_IFBLK:
967                                 capabilities = NOMMU_MAP_COPY;
968                                 break;
969
970                         case S_IFCHR:
971                                 capabilities =
972                                         NOMMU_MAP_DIRECT |
973                                         NOMMU_MAP_READ |
974                                         NOMMU_MAP_WRITE;
975                                 break;
976
977                         default:
978                                 return -EINVAL;
979                         }
980                 }
981
982                 /* eliminate any capabilities that we can't support on this
983                  * device */
984                 if (!file->f_op->get_unmapped_area)
985                         capabilities &= ~NOMMU_MAP_DIRECT;
986                 if (!(file->f_mode & FMODE_CAN_READ))
987                         capabilities &= ~NOMMU_MAP_COPY;
988
989                 /* The file shall have been opened with read permission. */
990                 if (!(file->f_mode & FMODE_READ))
991                         return -EACCES;
992
993                 if (flags & MAP_SHARED) {
994                         /* do checks for writing, appending and locking */
995                         if ((prot & PROT_WRITE) &&
996                             !(file->f_mode & FMODE_WRITE))
997                                 return -EACCES;
998
999                         if (IS_APPEND(file_inode(file)) &&
1000                             (file->f_mode & FMODE_WRITE))
1001                                 return -EACCES;
1002
1003                         if (locks_verify_locked(file))
1004                                 return -EAGAIN;
1005
1006                         if (!(capabilities & NOMMU_MAP_DIRECT))
1007                                 return -ENODEV;
1008
1009                         /* we mustn't privatise shared mappings */
1010                         capabilities &= ~NOMMU_MAP_COPY;
1011                 } else {
1012                         /* we're going to read the file into private memory we
1013                          * allocate */
1014                         if (!(capabilities & NOMMU_MAP_COPY))
1015                                 return -ENODEV;
1016
1017                         /* we don't permit a private writable mapping to be
1018                          * shared with the backing device */
1019                         if (prot & PROT_WRITE)
1020                                 capabilities &= ~NOMMU_MAP_DIRECT;
1021                 }
1022
1023                 if (capabilities & NOMMU_MAP_DIRECT) {
1024                         if (((prot & PROT_READ)  && !(capabilities & NOMMU_MAP_READ))  ||
1025                             ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
1026                             ((prot & PROT_EXEC)  && !(capabilities & NOMMU_MAP_EXEC))
1027                             ) {
1028                                 capabilities &= ~NOMMU_MAP_DIRECT;
1029                                 if (flags & MAP_SHARED) {
1030                                         pr_warn("MAP_SHARED not completely supported on !MMU\n");
1031                                         return -EINVAL;
1032                                 }
1033                         }
1034                 }
1035
1036                 /* handle executable mappings and implied executable
1037                  * mappings */
1038                 if (path_noexec(&file->f_path)) {
1039                         if (prot & PROT_EXEC)
1040                                 return -EPERM;
1041                 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1042                         /* handle implication of PROT_EXEC by PROT_READ */
1043                         if (current->personality & READ_IMPLIES_EXEC) {
1044                                 if (capabilities & NOMMU_MAP_EXEC)
1045                                         prot |= PROT_EXEC;
1046                         }
1047                 } else if ((prot & PROT_READ) &&
1048                          (prot & PROT_EXEC) &&
1049                          !(capabilities & NOMMU_MAP_EXEC)
1050                          ) {
1051                         /* backing file is not executable, try to copy */
1052                         capabilities &= ~NOMMU_MAP_DIRECT;
1053                 }
1054         } else {
1055                 /* anonymous mappings are always memory backed and can be
1056                  * privately mapped
1057                  */
1058                 capabilities = NOMMU_MAP_COPY;
1059
1060                 /* handle PROT_EXEC implication by PROT_READ */
1061                 if ((prot & PROT_READ) &&
1062                     (current->personality & READ_IMPLIES_EXEC))
1063                         prot |= PROT_EXEC;
1064         }
1065
1066         /* allow the security API to have its say */
1067         ret = security_mmap_addr(addr);
1068         if (ret < 0)
1069                 return ret;
1070
1071         /* looks okay */
1072         *_capabilities = capabilities;
1073         return 0;
1074 }
1075
1076 /*
1077  * we've determined that we can make the mapping, now translate what we
1078  * now know into VMA flags
1079  */
1080 static unsigned long determine_vm_flags(struct file *file,
1081                                         unsigned long prot,
1082                                         unsigned long flags,
1083                                         unsigned long capabilities)
1084 {
1085         unsigned long vm_flags;
1086
1087         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1088         /* vm_flags |= mm->def_flags; */
1089
1090         if (!(capabilities & NOMMU_MAP_DIRECT)) {
1091                 /* attempt to share read-only copies of mapped file chunks */
1092                 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1093                 if (file && !(prot & PROT_WRITE))
1094                         vm_flags |= VM_MAYSHARE;
1095         } else {
1096                 /* overlay a shareable mapping on the backing device or inode
1097                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1098                  * romfs/cramfs */
1099                 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1100                 if (flags & MAP_SHARED)
1101                         vm_flags |= VM_SHARED;
1102         }
1103
1104         /* refuse to let anyone share private mappings with this process if
1105          * it's being traced - otherwise breakpoints set in it may interfere
1106          * with another untraced process
1107          */
1108         if ((flags & MAP_PRIVATE) && current->ptrace)
1109                 vm_flags &= ~VM_MAYSHARE;
1110
1111         return vm_flags;
1112 }
1113
1114 /*
1115  * set up a shared mapping on a file (the driver or filesystem provides and
1116  * pins the storage)
1117  */
1118 static int do_mmap_shared_file(struct vm_area_struct *vma)
1119 {
1120         int ret;
1121
1122         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1123         if (ret == 0) {
1124                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1125                 return 0;
1126         }
1127         if (ret != -ENOSYS)
1128                 return ret;
1129
1130         /* getting -ENOSYS indicates that direct mmap isn't possible (as
1131          * opposed to tried but failed) so we can only give a suitable error as
1132          * it's not possible to make a private copy if MAP_SHARED was given */
1133         return -ENODEV;
1134 }
1135
1136 /*
1137  * set up a private mapping or an anonymous shared mapping
1138  */
1139 static int do_mmap_private(struct vm_area_struct *vma,
1140                            struct vm_region *region,
1141                            unsigned long len,
1142                            unsigned long capabilities)
1143 {
1144         unsigned long total, point;
1145         void *base;
1146         int ret, order;
1147
1148         /* invoke the file's mapping function so that it can keep track of
1149          * shared mappings on devices or memory
1150          * - VM_MAYSHARE will be set if it may attempt to share
1151          */
1152         if (capabilities & NOMMU_MAP_DIRECT) {
1153                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1154                 if (ret == 0) {
1155                         /* shouldn't return success if we're not sharing */
1156                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1157                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1158                         return 0;
1159                 }
1160                 if (ret != -ENOSYS)
1161                         return ret;
1162
1163                 /* getting an ENOSYS error indicates that direct mmap isn't
1164                  * possible (as opposed to tried but failed) so we'll try to
1165                  * make a private copy of the data and map that instead */
1166         }
1167
1168
1169         /* allocate some memory to hold the mapping
1170          * - note that this may not return a page-aligned address if the object
1171          *   we're allocating is smaller than a page
1172          */
1173         order = get_order(len);
1174         total = 1 << order;
1175         point = len >> PAGE_SHIFT;
1176
1177         /* we don't want to allocate a power-of-2 sized page set */
1178         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1179                 total = point;
1180
1181         base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1182         if (!base)
1183                 goto enomem;
1184
1185         atomic_long_add(total, &mmap_pages_allocated);
1186
1187         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1188         region->vm_start = (unsigned long) base;
1189         region->vm_end   = region->vm_start + len;
1190         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1191
1192         vma->vm_start = region->vm_start;
1193         vma->vm_end   = region->vm_start + len;
1194
1195         if (vma->vm_file) {
1196                 /* read the contents of a file into the copy */
1197                 mm_segment_t old_fs;
1198                 loff_t fpos;
1199
1200                 fpos = vma->vm_pgoff;
1201                 fpos <<= PAGE_SHIFT;
1202
1203                 old_fs = get_fs();
1204                 set_fs(KERNEL_DS);
1205                 ret = __vfs_read(vma->vm_file, base, len, &fpos);
1206                 set_fs(old_fs);
1207
1208                 if (ret < 0)
1209                         goto error_free;
1210
1211                 /* clear the last little bit */
1212                 if (ret < len)
1213                         memset(base + ret, 0, len - ret);
1214
1215         }
1216
1217         return 0;
1218
1219 error_free:
1220         free_page_series(region->vm_start, region->vm_top);
1221         region->vm_start = vma->vm_start = 0;
1222         region->vm_end   = vma->vm_end = 0;
1223         region->vm_top   = 0;
1224         return ret;
1225
1226 enomem:
1227         pr_err("Allocation of length %lu from process %d (%s) failed\n",
1228                len, current->pid, current->comm);
1229         show_free_areas(0);
1230         return -ENOMEM;
1231 }
1232
1233 /*
1234  * handle mapping creation for uClinux
1235  */
1236 unsigned long do_mmap(struct file *file,
1237                         unsigned long addr,
1238                         unsigned long len,
1239                         unsigned long prot,
1240                         unsigned long flags,
1241                         vm_flags_t vm_flags,
1242                         unsigned long pgoff,
1243                         unsigned long *populate)
1244 {
1245         struct vm_area_struct *vma;
1246         struct vm_region *region;
1247         struct rb_node *rb;
1248         unsigned long capabilities, result;
1249         int ret;
1250
1251         *populate = 0;
1252
1253         /* decide whether we should attempt the mapping, and if so what sort of
1254          * mapping */
1255         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1256                                     &capabilities);
1257         if (ret < 0)
1258                 return ret;
1259
1260         /* we ignore the address hint */
1261         addr = 0;
1262         len = PAGE_ALIGN(len);
1263
1264         /* we've determined that we can make the mapping, now translate what we
1265          * now know into VMA flags */
1266         vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1267
1268         /* we're going to need to record the mapping */
1269         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1270         if (!region)
1271                 goto error_getting_region;
1272
1273         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1274         if (!vma)
1275                 goto error_getting_vma;
1276
1277         region->vm_usage = 1;
1278         region->vm_flags = vm_flags;
1279         region->vm_pgoff = pgoff;
1280
1281         INIT_LIST_HEAD(&vma->anon_vma_chain);
1282         vma->vm_flags = vm_flags;
1283         vma->vm_pgoff = pgoff;
1284
1285         if (file) {
1286                 region->vm_file = get_file(file);
1287                 vma->vm_file = get_file(file);
1288         }
1289
1290         down_write(&nommu_region_sem);
1291
1292         /* if we want to share, we need to check for regions created by other
1293          * mmap() calls that overlap with our proposed mapping
1294          * - we can only share with a superset match on most regular files
1295          * - shared mappings on character devices and memory backed files are
1296          *   permitted to overlap inexactly as far as we are concerned for in
1297          *   these cases, sharing is handled in the driver or filesystem rather
1298          *   than here
1299          */
1300         if (vm_flags & VM_MAYSHARE) {
1301                 struct vm_region *pregion;
1302                 unsigned long pglen, rpglen, pgend, rpgend, start;
1303
1304                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1305                 pgend = pgoff + pglen;
1306
1307                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1308                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1309
1310                         if (!(pregion->vm_flags & VM_MAYSHARE))
1311                                 continue;
1312
1313                         /* search for overlapping mappings on the same file */
1314                         if (file_inode(pregion->vm_file) !=
1315                             file_inode(file))
1316                                 continue;
1317
1318                         if (pregion->vm_pgoff >= pgend)
1319                                 continue;
1320
1321                         rpglen = pregion->vm_end - pregion->vm_start;
1322                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1323                         rpgend = pregion->vm_pgoff + rpglen;
1324                         if (pgoff >= rpgend)
1325                                 continue;
1326
1327                         /* handle inexactly overlapping matches between
1328                          * mappings */
1329                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1330                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1331                                 /* new mapping is not a subset of the region */
1332                                 if (!(capabilities & NOMMU_MAP_DIRECT))
1333                                         goto sharing_violation;
1334                                 continue;
1335                         }
1336
1337                         /* we've found a region we can share */
1338                         pregion->vm_usage++;
1339                         vma->vm_region = pregion;
1340                         start = pregion->vm_start;
1341                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1342                         vma->vm_start = start;
1343                         vma->vm_end = start + len;
1344
1345                         if (pregion->vm_flags & VM_MAPPED_COPY)
1346                                 vma->vm_flags |= VM_MAPPED_COPY;
1347                         else {
1348                                 ret = do_mmap_shared_file(vma);
1349                                 if (ret < 0) {
1350                                         vma->vm_region = NULL;
1351                                         vma->vm_start = 0;
1352                                         vma->vm_end = 0;
1353                                         pregion->vm_usage--;
1354                                         pregion = NULL;
1355                                         goto error_just_free;
1356                                 }
1357                         }
1358                         fput(region->vm_file);
1359                         kmem_cache_free(vm_region_jar, region);
1360                         region = pregion;
1361                         result = start;
1362                         goto share;
1363                 }
1364
1365                 /* obtain the address at which to make a shared mapping
1366                  * - this is the hook for quasi-memory character devices to
1367                  *   tell us the location of a shared mapping
1368                  */
1369                 if (capabilities & NOMMU_MAP_DIRECT) {
1370                         addr = file->f_op->get_unmapped_area(file, addr, len,
1371                                                              pgoff, flags);
1372                         if (IS_ERR_VALUE(addr)) {
1373                                 ret = addr;
1374                                 if (ret != -ENOSYS)
1375                                         goto error_just_free;
1376
1377                                 /* the driver refused to tell us where to site
1378                                  * the mapping so we'll have to attempt to copy
1379                                  * it */
1380                                 ret = -ENODEV;
1381                                 if (!(capabilities & NOMMU_MAP_COPY))
1382                                         goto error_just_free;
1383
1384                                 capabilities &= ~NOMMU_MAP_DIRECT;
1385                         } else {
1386                                 vma->vm_start = region->vm_start = addr;
1387                                 vma->vm_end = region->vm_end = addr + len;
1388                         }
1389                 }
1390         }
1391
1392         vma->vm_region = region;
1393
1394         /* set up the mapping
1395          * - the region is filled in if NOMMU_MAP_DIRECT is still set
1396          */
1397         if (file && vma->vm_flags & VM_SHARED)
1398                 ret = do_mmap_shared_file(vma);
1399         else
1400                 ret = do_mmap_private(vma, region, len, capabilities);
1401         if (ret < 0)
1402                 goto error_just_free;
1403         add_nommu_region(region);
1404
1405         /* clear anonymous mappings that don't ask for uninitialized data */
1406         if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1407                 memset((void *)region->vm_start, 0,
1408                        region->vm_end - region->vm_start);
1409
1410         /* okay... we have a mapping; now we have to register it */
1411         result = vma->vm_start;
1412
1413         current->mm->total_vm += len >> PAGE_SHIFT;
1414
1415 share:
1416         add_vma_to_mm(current->mm, vma);
1417
1418         /* we flush the region from the icache only when the first executable
1419          * mapping of it is made  */
1420         if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1421                 flush_icache_range(region->vm_start, region->vm_end);
1422                 region->vm_icache_flushed = true;
1423         }
1424
1425         up_write(&nommu_region_sem);
1426
1427         return result;
1428
1429 error_just_free:
1430         up_write(&nommu_region_sem);
1431 error:
1432         if (region->vm_file)
1433                 fput(region->vm_file);
1434         kmem_cache_free(vm_region_jar, region);
1435         if (vma->vm_file)
1436                 fput(vma->vm_file);
1437         kmem_cache_free(vm_area_cachep, vma);
1438         return ret;
1439
1440 sharing_violation:
1441         up_write(&nommu_region_sem);
1442         pr_warn("Attempt to share mismatched mappings\n");
1443         ret = -EINVAL;
1444         goto error;
1445
1446 error_getting_vma:
1447         kmem_cache_free(vm_region_jar, region);
1448         pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1449                         len, current->pid);
1450         show_free_areas(0);
1451         return -ENOMEM;
1452
1453 error_getting_region:
1454         pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1455                         len, current->pid);
1456         show_free_areas(0);
1457         return -ENOMEM;
1458 }
1459
1460 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1461                 unsigned long, prot, unsigned long, flags,
1462                 unsigned long, fd, unsigned long, pgoff)
1463 {
1464         struct file *file = NULL;
1465         unsigned long retval = -EBADF;
1466
1467         audit_mmap_fd(fd, flags);
1468         if (!(flags & MAP_ANONYMOUS)) {
1469                 file = fget(fd);
1470                 if (!file)
1471                         goto out;
1472         }
1473
1474         flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1475
1476         retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1477
1478         if (file)
1479                 fput(file);
1480 out:
1481         return retval;
1482 }
1483
1484 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1485 struct mmap_arg_struct {
1486         unsigned long addr;
1487         unsigned long len;
1488         unsigned long prot;
1489         unsigned long flags;
1490         unsigned long fd;
1491         unsigned long offset;
1492 };
1493
1494 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1495 {
1496         struct mmap_arg_struct a;
1497
1498         if (copy_from_user(&a, arg, sizeof(a)))
1499                 return -EFAULT;
1500         if (offset_in_page(a.offset))
1501                 return -EINVAL;
1502
1503         return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1504                               a.offset >> PAGE_SHIFT);
1505 }
1506 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1507
1508 /*
1509  * split a vma into two pieces at address 'addr', a new vma is allocated either
1510  * for the first part or the tail.
1511  */
1512 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1513               unsigned long addr, int new_below)
1514 {
1515         struct vm_area_struct *new;
1516         struct vm_region *region;
1517         unsigned long npages;
1518
1519         /* we're only permitted to split anonymous regions (these should have
1520          * only a single usage on the region) */
1521         if (vma->vm_file)
1522                 return -ENOMEM;
1523
1524         if (mm->map_count >= sysctl_max_map_count)
1525                 return -ENOMEM;
1526
1527         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1528         if (!region)
1529                 return -ENOMEM;
1530
1531         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1532         if (!new) {
1533                 kmem_cache_free(vm_region_jar, region);
1534                 return -ENOMEM;
1535         }
1536
1537         /* most fields are the same, copy all, and then fixup */
1538         *new = *vma;
1539         *region = *vma->vm_region;
1540         new->vm_region = region;
1541
1542         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1543
1544         if (new_below) {
1545                 region->vm_top = region->vm_end = new->vm_end = addr;
1546         } else {
1547                 region->vm_start = new->vm_start = addr;
1548                 region->vm_pgoff = new->vm_pgoff += npages;
1549         }
1550
1551         if (new->vm_ops && new->vm_ops->open)
1552                 new->vm_ops->open(new);
1553
1554         delete_vma_from_mm(vma);
1555         down_write(&nommu_region_sem);
1556         delete_nommu_region(vma->vm_region);
1557         if (new_below) {
1558                 vma->vm_region->vm_start = vma->vm_start = addr;
1559                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1560         } else {
1561                 vma->vm_region->vm_end = vma->vm_end = addr;
1562                 vma->vm_region->vm_top = addr;
1563         }
1564         add_nommu_region(vma->vm_region);
1565         add_nommu_region(new->vm_region);
1566         up_write(&nommu_region_sem);
1567         add_vma_to_mm(mm, vma);
1568         add_vma_to_mm(mm, new);
1569         return 0;
1570 }
1571
1572 /*
1573  * shrink a VMA by removing the specified chunk from either the beginning or
1574  * the end
1575  */
1576 static int shrink_vma(struct mm_struct *mm,
1577                       struct vm_area_struct *vma,
1578                       unsigned long from, unsigned long to)
1579 {
1580         struct vm_region *region;
1581
1582         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1583          * and list */
1584         delete_vma_from_mm(vma);
1585         if (from > vma->vm_start)
1586                 vma->vm_end = from;
1587         else
1588                 vma->vm_start = to;
1589         add_vma_to_mm(mm, vma);
1590
1591         /* cut the backing region down to size */
1592         region = vma->vm_region;
1593         BUG_ON(region->vm_usage != 1);
1594
1595         down_write(&nommu_region_sem);
1596         delete_nommu_region(region);
1597         if (from > region->vm_start) {
1598                 to = region->vm_top;
1599                 region->vm_top = region->vm_end = from;
1600         } else {
1601                 region->vm_start = to;
1602         }
1603         add_nommu_region(region);
1604         up_write(&nommu_region_sem);
1605
1606         free_page_series(from, to);
1607         return 0;
1608 }
1609
1610 /*
1611  * release a mapping
1612  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1613  *   VMA, though it need not cover the whole VMA
1614  */
1615 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1616 {
1617         struct vm_area_struct *vma;
1618         unsigned long end;
1619         int ret;
1620
1621         len = PAGE_ALIGN(len);
1622         if (len == 0)
1623                 return -EINVAL;
1624
1625         end = start + len;
1626
1627         /* find the first potentially overlapping VMA */
1628         vma = find_vma(mm, start);
1629         if (!vma) {
1630                 static int limit;
1631                 if (limit < 5) {
1632                         pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1633                                         current->pid, current->comm,
1634                                         start, start + len - 1);
1635                         limit++;
1636                 }
1637                 return -EINVAL;
1638         }
1639
1640         /* we're allowed to split an anonymous VMA but not a file-backed one */
1641         if (vma->vm_file) {
1642                 do {
1643                         if (start > vma->vm_start)
1644                                 return -EINVAL;
1645                         if (end == vma->vm_end)
1646                                 goto erase_whole_vma;
1647                         vma = vma->vm_next;
1648                 } while (vma);
1649                 return -EINVAL;
1650         } else {
1651                 /* the chunk must be a subset of the VMA found */
1652                 if (start == vma->vm_start && end == vma->vm_end)
1653                         goto erase_whole_vma;
1654                 if (start < vma->vm_start || end > vma->vm_end)
1655                         return -EINVAL;
1656                 if (offset_in_page(start))
1657                         return -EINVAL;
1658                 if (end != vma->vm_end && offset_in_page(end))
1659                         return -EINVAL;
1660                 if (start != vma->vm_start && end != vma->vm_end) {
1661                         ret = split_vma(mm, vma, start, 1);
1662                         if (ret < 0)
1663                                 return ret;
1664                 }
1665                 return shrink_vma(mm, vma, start, end);
1666         }
1667
1668 erase_whole_vma:
1669         delete_vma_from_mm(vma);
1670         delete_vma(mm, vma);
1671         return 0;
1672 }
1673 EXPORT_SYMBOL(do_munmap);
1674
1675 int vm_munmap(unsigned long addr, size_t len)
1676 {
1677         struct mm_struct *mm = current->mm;
1678         int ret;
1679
1680         down_write(&mm->mmap_sem);
1681         ret = do_munmap(mm, addr, len);
1682         up_write(&mm->mmap_sem);
1683         return ret;
1684 }
1685 EXPORT_SYMBOL(vm_munmap);
1686
1687 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1688 {
1689         return vm_munmap(addr, len);
1690 }
1691
1692 /*
1693  * release all the mappings made in a process's VM space
1694  */
1695 void exit_mmap(struct mm_struct *mm)
1696 {
1697         struct vm_area_struct *vma;
1698
1699         if (!mm)
1700                 return;
1701
1702         mm->total_vm = 0;
1703
1704         while ((vma = mm->mmap)) {
1705                 mm->mmap = vma->vm_next;
1706                 delete_vma_from_mm(vma);
1707                 delete_vma(mm, vma);
1708                 cond_resched();
1709         }
1710 }
1711
1712 unsigned long vm_brk(unsigned long addr, unsigned long len)
1713 {
1714         return -ENOMEM;
1715 }
1716
1717 /*
1718  * expand (or shrink) an existing mapping, potentially moving it at the same
1719  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1720  *
1721  * under NOMMU conditions, we only permit changing a mapping's size, and only
1722  * as long as it stays within the region allocated by do_mmap_private() and the
1723  * block is not shareable
1724  *
1725  * MREMAP_FIXED is not supported under NOMMU conditions
1726  */
1727 static unsigned long do_mremap(unsigned long addr,
1728                         unsigned long old_len, unsigned long new_len,
1729                         unsigned long flags, unsigned long new_addr)
1730 {
1731         struct vm_area_struct *vma;
1732
1733         /* insanity checks first */
1734         old_len = PAGE_ALIGN(old_len);
1735         new_len = PAGE_ALIGN(new_len);
1736         if (old_len == 0 || new_len == 0)
1737                 return (unsigned long) -EINVAL;
1738
1739         if (offset_in_page(addr))
1740                 return -EINVAL;
1741
1742         if (flags & MREMAP_FIXED && new_addr != addr)
1743                 return (unsigned long) -EINVAL;
1744
1745         vma = find_vma_exact(current->mm, addr, old_len);
1746         if (!vma)
1747                 return (unsigned long) -EINVAL;
1748
1749         if (vma->vm_end != vma->vm_start + old_len)
1750                 return (unsigned long) -EFAULT;
1751
1752         if (vma->vm_flags & VM_MAYSHARE)
1753                 return (unsigned long) -EPERM;
1754
1755         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1756                 return (unsigned long) -ENOMEM;
1757
1758         /* all checks complete - do it */
1759         vma->vm_end = vma->vm_start + new_len;
1760         return vma->vm_start;
1761 }
1762
1763 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1764                 unsigned long, new_len, unsigned long, flags,
1765                 unsigned long, new_addr)
1766 {
1767         unsigned long ret;
1768
1769         down_write(&current->mm->mmap_sem);
1770         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1771         up_write(&current->mm->mmap_sem);
1772         return ret;
1773 }
1774
1775 struct page *follow_page_mask(struct vm_area_struct *vma,
1776                               unsigned long address, unsigned int flags,
1777                               unsigned int *page_mask)
1778 {
1779         *page_mask = 0;
1780         return NULL;
1781 }
1782
1783 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1784                 unsigned long pfn, unsigned long size, pgprot_t prot)
1785 {
1786         if (addr != (pfn << PAGE_SHIFT))
1787                 return -EINVAL;
1788
1789         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1790         return 0;
1791 }
1792 EXPORT_SYMBOL(remap_pfn_range);
1793
1794 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1795 {
1796         unsigned long pfn = start >> PAGE_SHIFT;
1797         unsigned long vm_len = vma->vm_end - vma->vm_start;
1798
1799         pfn += vma->vm_pgoff;
1800         return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1801 }
1802 EXPORT_SYMBOL(vm_iomap_memory);
1803
1804 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1805                         unsigned long pgoff)
1806 {
1807         unsigned int size = vma->vm_end - vma->vm_start;
1808
1809         if (!(vma->vm_flags & VM_USERMAP))
1810                 return -EINVAL;
1811
1812         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1813         vma->vm_end = vma->vm_start + size;
1814
1815         return 0;
1816 }
1817 EXPORT_SYMBOL(remap_vmalloc_range);
1818
1819 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1820         unsigned long len, unsigned long pgoff, unsigned long flags)
1821 {
1822         return -ENOMEM;
1823 }
1824
1825 void unmap_mapping_range(struct address_space *mapping,
1826                          loff_t const holebegin, loff_t const holelen,
1827                          int even_cows)
1828 {
1829 }
1830 EXPORT_SYMBOL(unmap_mapping_range);
1831
1832 /*
1833  * Check that a process has enough memory to allocate a new virtual
1834  * mapping. 0 means there is enough memory for the allocation to
1835  * succeed and -ENOMEM implies there is not.
1836  *
1837  * We currently support three overcommit policies, which are set via the
1838  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1839  *
1840  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1841  * Additional code 2002 Jul 20 by Robert Love.
1842  *
1843  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1844  *
1845  * Note this is a helper function intended to be used by LSMs which
1846  * wish to use this logic.
1847  */
1848 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1849 {
1850         long free, allowed, reserve;
1851
1852         vm_acct_memory(pages);
1853
1854         /*
1855          * Sometimes we want to use more memory than we have
1856          */
1857         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1858                 return 0;
1859
1860         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1861                 free = global_page_state(NR_FREE_PAGES);
1862                 free += global_page_state(NR_FILE_PAGES);
1863
1864                 /*
1865                  * shmem pages shouldn't be counted as free in this
1866                  * case, they can't be purged, only swapped out, and
1867                  * that won't affect the overall amount of available
1868                  * memory in the system.
1869                  */
1870                 free -= global_page_state(NR_SHMEM);
1871
1872                 free += get_nr_swap_pages();
1873
1874                 /*
1875                  * Any slabs which are created with the
1876                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1877                  * which are reclaimable, under pressure.  The dentry
1878                  * cache and most inode caches should fall into this
1879                  */
1880                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1881
1882                 /*
1883                  * Leave reserved pages. The pages are not for anonymous pages.
1884                  */
1885                 if (free <= totalreserve_pages)
1886                         goto error;
1887                 else
1888                         free -= totalreserve_pages;
1889
1890                 /*
1891                  * Reserve some for root
1892                  */
1893                 if (!cap_sys_admin)
1894                         free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1895
1896                 if (free > pages)
1897                         return 0;
1898
1899                 goto error;
1900         }
1901
1902         allowed = vm_commit_limit();
1903         /*
1904          * Reserve some 3% for root
1905          */
1906         if (!cap_sys_admin)
1907                 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1908
1909         /*
1910          * Don't let a single process grow so big a user can't recover
1911          */
1912         if (mm) {
1913                 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1914                 allowed -= min_t(long, mm->total_vm / 32, reserve);
1915         }
1916
1917         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1918                 return 0;
1919
1920 error:
1921         vm_unacct_memory(pages);
1922
1923         return -ENOMEM;
1924 }
1925
1926 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1927 {
1928         BUG();
1929         return 0;
1930 }
1931 EXPORT_SYMBOL(filemap_fault);
1932
1933 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1934 {
1935         BUG();
1936 }
1937 EXPORT_SYMBOL(filemap_map_pages);
1938
1939 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1940                 unsigned long addr, void *buf, int len, int write)
1941 {
1942         struct vm_area_struct *vma;
1943
1944         down_read(&mm->mmap_sem);
1945
1946         /* the access must start within one of the target process's mappings */
1947         vma = find_vma(mm, addr);
1948         if (vma) {
1949                 /* don't overrun this mapping */
1950                 if (addr + len >= vma->vm_end)
1951                         len = vma->vm_end - addr;
1952
1953                 /* only read or write mappings where it is permitted */
1954                 if (write && vma->vm_flags & VM_MAYWRITE)
1955                         copy_to_user_page(vma, NULL, addr,
1956                                          (void *) addr, buf, len);
1957                 else if (!write && vma->vm_flags & VM_MAYREAD)
1958                         copy_from_user_page(vma, NULL, addr,
1959                                             buf, (void *) addr, len);
1960                 else
1961                         len = 0;
1962         } else {
1963                 len = 0;
1964         }
1965
1966         up_read(&mm->mmap_sem);
1967
1968         return len;
1969 }
1970
1971 /**
1972  * @access_remote_vm - access another process' address space
1973  * @mm:         the mm_struct of the target address space
1974  * @addr:       start address to access
1975  * @buf:        source or destination buffer
1976  * @len:        number of bytes to transfer
1977  * @write:      whether the access is a write
1978  *
1979  * The caller must hold a reference on @mm.
1980  */
1981 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1982                 void *buf, int len, int write)
1983 {
1984         return __access_remote_vm(NULL, mm, addr, buf, len, write);
1985 }
1986
1987 /*
1988  * Access another process' address space.
1989  * - source/target buffer must be kernel space
1990  */
1991 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1992 {
1993         struct mm_struct *mm;
1994
1995         if (addr + len < addr)
1996                 return 0;
1997
1998         mm = get_task_mm(tsk);
1999         if (!mm)
2000                 return 0;
2001
2002         len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2003
2004         mmput(mm);
2005         return len;
2006 }
2007
2008 /**
2009  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2010  * @inode: The inode to check
2011  * @size: The current filesize of the inode
2012  * @newsize: The proposed filesize of the inode
2013  *
2014  * Check the shared mappings on an inode on behalf of a shrinking truncate to
2015  * make sure that that any outstanding VMAs aren't broken and then shrink the
2016  * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2017  * automatically grant mappings that are too large.
2018  */
2019 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2020                                 size_t newsize)
2021 {
2022         struct vm_area_struct *vma;
2023         struct vm_region *region;
2024         pgoff_t low, high;
2025         size_t r_size, r_top;
2026
2027         low = newsize >> PAGE_SHIFT;
2028         high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2029
2030         down_write(&nommu_region_sem);
2031         i_mmap_lock_read(inode->i_mapping);
2032
2033         /* search for VMAs that fall within the dead zone */
2034         vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
2035                 /* found one - only interested if it's shared out of the page
2036                  * cache */
2037                 if (vma->vm_flags & VM_SHARED) {
2038                         i_mmap_unlock_read(inode->i_mapping);
2039                         up_write(&nommu_region_sem);
2040                         return -ETXTBSY; /* not quite true, but near enough */
2041                 }
2042         }
2043
2044         /* reduce any regions that overlap the dead zone - if in existence,
2045          * these will be pointed to by VMAs that don't overlap the dead zone
2046          *
2047          * we don't check for any regions that start beyond the EOF as there
2048          * shouldn't be any
2049          */
2050         vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
2051                 if (!(vma->vm_flags & VM_SHARED))
2052                         continue;
2053
2054                 region = vma->vm_region;
2055                 r_size = region->vm_top - region->vm_start;
2056                 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2057
2058                 if (r_top > newsize) {
2059                         region->vm_top -= r_top - newsize;
2060                         if (region->vm_end > region->vm_top)
2061                                 region->vm_end = region->vm_top;
2062                 }
2063         }
2064
2065         i_mmap_unlock_read(inode->i_mapping);
2066         up_write(&nommu_region_sem);
2067         return 0;
2068 }
2069
2070 /*
2071  * Initialise sysctl_user_reserve_kbytes.
2072  *
2073  * This is intended to prevent a user from starting a single memory hogging
2074  * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2075  * mode.
2076  *
2077  * The default value is min(3% of free memory, 128MB)
2078  * 128MB is enough to recover with sshd/login, bash, and top/kill.
2079  */
2080 static int __meminit init_user_reserve(void)
2081 {
2082         unsigned long free_kbytes;
2083
2084         free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2085
2086         sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2087         return 0;
2088 }
2089 subsys_initcall(init_user_reserve);
2090
2091 /*
2092  * Initialise sysctl_admin_reserve_kbytes.
2093  *
2094  * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2095  * to log in and kill a memory hogging process.
2096  *
2097  * Systems with more than 256MB will reserve 8MB, enough to recover
2098  * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2099  * only reserve 3% of free pages by default.
2100  */
2101 static int __meminit init_admin_reserve(void)
2102 {
2103         unsigned long free_kbytes;
2104
2105         free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2106
2107         sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2108         return 0;
2109 }
2110 subsys_initcall(init_admin_reserve);