f59e1424d3db650fc23617f60d060d6b68ce271b
[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) 2007-2010 Paul Mundt <lethal@linux-sh.org>
14  */
15
16 #include <linux/module.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/swap.h>
20 #include <linux/file.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/tracehook.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mount.h>
29 #include <linux/personality.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32 #include <linux/audit.h>
33
34 #include <asm/uaccess.h>
35 #include <asm/tlb.h>
36 #include <asm/tlbflush.h>
37 #include <asm/mmu_context.h>
38 #include "internal.h"
39
40 #if 0
41 #define kenter(FMT, ...) \
42         printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
43 #define kleave(FMT, ...) \
44         printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
45 #define kdebug(FMT, ...) \
46         printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
47 #else
48 #define kenter(FMT, ...) \
49         no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
50 #define kleave(FMT, ...) \
51         no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
52 #define kdebug(FMT, ...) \
53         no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
54 #endif
55
56 void *high_memory;
57 struct page *mem_map;
58 unsigned long max_mapnr;
59 unsigned long num_physpages;
60 unsigned long highest_memmap_pfn;
61 struct percpu_counter vm_committed_as;
62 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
63 int sysctl_overcommit_ratio = 50; /* default is 50% */
64 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
65 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
66 int heap_stack_gap = 0;
67
68 atomic_long_t mmap_pages_allocated;
69
70 EXPORT_SYMBOL(mem_map);
71 EXPORT_SYMBOL(num_physpages);
72
73 /* list of mapped, potentially shareable regions */
74 static struct kmem_cache *vm_region_jar;
75 struct rb_root nommu_region_tree = RB_ROOT;
76 DECLARE_RWSEM(nommu_region_sem);
77
78 const struct vm_operations_struct generic_file_vm_ops = {
79 };
80
81 /*
82  * Return the total memory allocated for this pointer, not
83  * just what the caller asked for.
84  *
85  * Doesn't have to be accurate, i.e. may have races.
86  */
87 unsigned int kobjsize(const void *objp)
88 {
89         struct page *page;
90
91         /*
92          * If the object we have should not have ksize performed on it,
93          * return size of 0
94          */
95         if (!objp || !virt_addr_valid(objp))
96                 return 0;
97
98         page = virt_to_head_page(objp);
99
100         /*
101          * If the allocator sets PageSlab, we know the pointer came from
102          * kmalloc().
103          */
104         if (PageSlab(page))
105                 return ksize(objp);
106
107         /*
108          * If it's not a compound page, see if we have a matching VMA
109          * region. This test is intentionally done in reverse order,
110          * so if there's no VMA, we still fall through and hand back
111          * PAGE_SIZE for 0-order pages.
112          */
113         if (!PageCompound(page)) {
114                 struct vm_area_struct *vma;
115
116                 vma = find_vma(current->mm, (unsigned long)objp);
117                 if (vma)
118                         return vma->vm_end - vma->vm_start;
119         }
120
121         /*
122          * The ksize() function is only guaranteed to work for pointers
123          * returned by kmalloc(). So handle arbitrary pointers here.
124          */
125         return PAGE_SIZE << compound_order(page);
126 }
127
128 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
129                      unsigned long start, int nr_pages, unsigned int foll_flags,
130                      struct page **pages, struct vm_area_struct **vmas,
131                      int *retry)
132 {
133         struct vm_area_struct *vma;
134         unsigned long vm_flags;
135         int i;
136
137         /* calculate required read or write permissions.
138          * If FOLL_FORCE is set, we only require the "MAY" flags.
139          */
140         vm_flags  = (foll_flags & FOLL_WRITE) ?
141                         (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
142         vm_flags &= (foll_flags & FOLL_FORCE) ?
143                         (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
144
145         for (i = 0; i < nr_pages; i++) {
146                 vma = find_vma(mm, start);
147                 if (!vma)
148                         goto finish_or_fault;
149
150                 /* protect what we can, including chardevs */
151                 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
152                     !(vm_flags & vma->vm_flags))
153                         goto finish_or_fault;
154
155                 if (pages) {
156                         pages[i] = virt_to_page(start);
157                         if (pages[i])
158                                 page_cache_get(pages[i]);
159                 }
160                 if (vmas)
161                         vmas[i] = vma;
162                 start = (start + PAGE_SIZE) & PAGE_MASK;
163         }
164
165         return i;
166
167 finish_or_fault:
168         return i ? : -EFAULT;
169 }
170
171 /*
172  * get a list of pages in an address range belonging to the specified process
173  * and indicate the VMA that covers each page
174  * - this is potentially dodgy as we may end incrementing the page count of a
175  *   slab page or a secondary page from a compound page
176  * - don't permit access to VMAs that don't support it, such as I/O mappings
177  */
178 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
179         unsigned long start, int nr_pages, int write, int force,
180         struct page **pages, struct vm_area_struct **vmas)
181 {
182         int flags = 0;
183
184         if (write)
185                 flags |= FOLL_WRITE;
186         if (force)
187                 flags |= FOLL_FORCE;
188
189         return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
190                                 NULL);
191 }
192 EXPORT_SYMBOL(get_user_pages);
193
194 /**
195  * follow_pfn - look up PFN at a user virtual address
196  * @vma: memory mapping
197  * @address: user virtual address
198  * @pfn: location to store found PFN
199  *
200  * Only IO mappings and raw PFN mappings are allowed.
201  *
202  * Returns zero and the pfn at @pfn on success, -ve otherwise.
203  */
204 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
205         unsigned long *pfn)
206 {
207         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
208                 return -EINVAL;
209
210         *pfn = address >> PAGE_SHIFT;
211         return 0;
212 }
213 EXPORT_SYMBOL(follow_pfn);
214
215 DEFINE_RWLOCK(vmlist_lock);
216 struct vm_struct *vmlist;
217
218 void vfree(const void *addr)
219 {
220         kfree(addr);
221 }
222 EXPORT_SYMBOL(vfree);
223
224 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
225 {
226         /*
227          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
228          * returns only a logical address.
229          */
230         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
231 }
232 EXPORT_SYMBOL(__vmalloc);
233
234 void *vmalloc_user(unsigned long size)
235 {
236         void *ret;
237
238         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
239                         PAGE_KERNEL);
240         if (ret) {
241                 struct vm_area_struct *vma;
242
243                 down_write(&current->mm->mmap_sem);
244                 vma = find_vma(current->mm, (unsigned long)ret);
245                 if (vma)
246                         vma->vm_flags |= VM_USERMAP;
247                 up_write(&current->mm->mmap_sem);
248         }
249
250         return ret;
251 }
252 EXPORT_SYMBOL(vmalloc_user);
253
254 struct page *vmalloc_to_page(const void *addr)
255 {
256         return virt_to_page(addr);
257 }
258 EXPORT_SYMBOL(vmalloc_to_page);
259
260 unsigned long vmalloc_to_pfn(const void *addr)
261 {
262         return page_to_pfn(virt_to_page(addr));
263 }
264 EXPORT_SYMBOL(vmalloc_to_pfn);
265
266 long vread(char *buf, char *addr, unsigned long count)
267 {
268         memcpy(buf, addr, count);
269         return count;
270 }
271
272 long vwrite(char *buf, char *addr, unsigned long count)
273 {
274         /* Don't allow overflow */
275         if ((unsigned long) addr + count < count)
276                 count = -(unsigned long) addr;
277
278         memcpy(addr, buf, count);
279         return(count);
280 }
281
282 /*
283  *      vmalloc  -  allocate virtually continguos memory
284  *
285  *      @size:          allocation size
286  *
287  *      Allocate enough pages to cover @size from the page level
288  *      allocator and map them into continguos kernel virtual space.
289  *
290  *      For tight control over page level allocator and protection flags
291  *      use __vmalloc() instead.
292  */
293 void *vmalloc(unsigned long size)
294 {
295        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
296 }
297 EXPORT_SYMBOL(vmalloc);
298
299 /*
300  *      vzalloc - allocate virtually continguos memory with zero fill
301  *
302  *      @size:          allocation size
303  *
304  *      Allocate enough pages to cover @size from the page level
305  *      allocator and map them into continguos kernel virtual space.
306  *      The memory allocated is set to zero.
307  *
308  *      For tight control over page level allocator and protection flags
309  *      use __vmalloc() instead.
310  */
311 void *vzalloc(unsigned long size)
312 {
313         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
314                         PAGE_KERNEL);
315 }
316 EXPORT_SYMBOL(vzalloc);
317
318 /**
319  * vmalloc_node - allocate memory on a specific node
320  * @size:       allocation size
321  * @node:       numa node
322  *
323  * Allocate enough pages to cover @size from the page level
324  * allocator and map them into contiguous kernel virtual space.
325  *
326  * For tight control over page level allocator and protection flags
327  * use __vmalloc() instead.
328  */
329 void *vmalloc_node(unsigned long size, int node)
330 {
331         return vmalloc(size);
332 }
333 EXPORT_SYMBOL(vmalloc_node);
334
335 /**
336  * vzalloc_node - allocate memory on a specific node with zero fill
337  * @size:       allocation size
338  * @node:       numa node
339  *
340  * Allocate enough pages to cover @size from the page level
341  * allocator and map them into contiguous kernel virtual space.
342  * The memory allocated is set to zero.
343  *
344  * For tight control over page level allocator and protection flags
345  * use __vmalloc() instead.
346  */
347 void *vzalloc_node(unsigned long size, int node)
348 {
349         return vzalloc(size);
350 }
351 EXPORT_SYMBOL(vzalloc_node);
352
353 #ifndef PAGE_KERNEL_EXEC
354 # define PAGE_KERNEL_EXEC PAGE_KERNEL
355 #endif
356
357 /**
358  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
359  *      @size:          allocation size
360  *
361  *      Kernel-internal function to allocate enough pages to cover @size
362  *      the page level allocator and map them into contiguous and
363  *      executable kernel virtual space.
364  *
365  *      For tight control over page level allocator and protection flags
366  *      use __vmalloc() instead.
367  */
368
369 void *vmalloc_exec(unsigned long size)
370 {
371         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
372 }
373
374 /**
375  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
376  *      @size:          allocation size
377  *
378  *      Allocate enough 32bit PA addressable pages to cover @size from the
379  *      page level allocator and map them into continguos kernel virtual space.
380  */
381 void *vmalloc_32(unsigned long size)
382 {
383         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
384 }
385 EXPORT_SYMBOL(vmalloc_32);
386
387 /**
388  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
389  *      @size:          allocation size
390  *
391  * The resulting memory area is 32bit addressable and zeroed so it can be
392  * mapped to userspace without leaking data.
393  *
394  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
395  * remap_vmalloc_range() are permissible.
396  */
397 void *vmalloc_32_user(unsigned long size)
398 {
399         /*
400          * We'll have to sort out the ZONE_DMA bits for 64-bit,
401          * but for now this can simply use vmalloc_user() directly.
402          */
403         return vmalloc_user(size);
404 }
405 EXPORT_SYMBOL(vmalloc_32_user);
406
407 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
408 {
409         BUG();
410         return NULL;
411 }
412 EXPORT_SYMBOL(vmap);
413
414 void vunmap(const void *addr)
415 {
416         BUG();
417 }
418 EXPORT_SYMBOL(vunmap);
419
420 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
421 {
422         BUG();
423         return NULL;
424 }
425 EXPORT_SYMBOL(vm_map_ram);
426
427 void vm_unmap_ram(const void *mem, unsigned int count)
428 {
429         BUG();
430 }
431 EXPORT_SYMBOL(vm_unmap_ram);
432
433 void vm_unmap_aliases(void)
434 {
435 }
436 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
437
438 /*
439  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
440  * have one.
441  */
442 void  __attribute__((weak)) vmalloc_sync_all(void)
443 {
444 }
445
446 /**
447  *      alloc_vm_area - allocate a range of kernel address space
448  *      @size:          size of the area
449  *
450  *      Returns:        NULL on failure, vm_struct on success
451  *
452  *      This function reserves a range of kernel address space, and
453  *      allocates pagetables to map that range.  No actual mappings
454  *      are created.  If the kernel address space is not shared
455  *      between processes, it syncs the pagetable across all
456  *      processes.
457  */
458 struct vm_struct *alloc_vm_area(size_t size)
459 {
460         BUG();
461         return NULL;
462 }
463 EXPORT_SYMBOL_GPL(alloc_vm_area);
464
465 void free_vm_area(struct vm_struct *area)
466 {
467         BUG();
468 }
469 EXPORT_SYMBOL_GPL(free_vm_area);
470
471 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
472                    struct page *page)
473 {
474         return -EINVAL;
475 }
476 EXPORT_SYMBOL(vm_insert_page);
477
478 /*
479  *  sys_brk() for the most part doesn't need the global kernel
480  *  lock, except when an application is doing something nasty
481  *  like trying to un-brk an area that has already been mapped
482  *  to a regular file.  in this case, the unmapping will need
483  *  to invoke file system routines that need the global lock.
484  */
485 SYSCALL_DEFINE1(brk, unsigned long, brk)
486 {
487         struct mm_struct *mm = current->mm;
488
489         if (brk < mm->start_brk || brk > mm->context.end_brk)
490                 return mm->brk;
491
492         if (mm->brk == brk)
493                 return mm->brk;
494
495         /*
496          * Always allow shrinking brk
497          */
498         if (brk <= mm->brk) {
499                 mm->brk = brk;
500                 return brk;
501         }
502
503         /*
504          * Ok, looks good - let it rip.
505          */
506         flush_icache_range(mm->brk, brk);
507         return mm->brk = brk;
508 }
509
510 /*
511  * initialise the VMA and region record slabs
512  */
513 void __init mmap_init(void)
514 {
515         int ret;
516
517         ret = percpu_counter_init(&vm_committed_as, 0);
518         VM_BUG_ON(ret);
519         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
520 }
521
522 /*
523  * validate the region tree
524  * - the caller must hold the region lock
525  */
526 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
527 static noinline void validate_nommu_regions(void)
528 {
529         struct vm_region *region, *last;
530         struct rb_node *p, *lastp;
531
532         lastp = rb_first(&nommu_region_tree);
533         if (!lastp)
534                 return;
535
536         last = rb_entry(lastp, struct vm_region, vm_rb);
537         BUG_ON(unlikely(last->vm_end <= last->vm_start));
538         BUG_ON(unlikely(last->vm_top < last->vm_end));
539
540         while ((p = rb_next(lastp))) {
541                 region = rb_entry(p, struct vm_region, vm_rb);
542                 last = rb_entry(lastp, struct vm_region, vm_rb);
543
544                 BUG_ON(unlikely(region->vm_end <= region->vm_start));
545                 BUG_ON(unlikely(region->vm_top < region->vm_end));
546                 BUG_ON(unlikely(region->vm_start < last->vm_top));
547
548                 lastp = p;
549         }
550 }
551 #else
552 static void validate_nommu_regions(void)
553 {
554 }
555 #endif
556
557 /*
558  * add a region into the global tree
559  */
560 static void add_nommu_region(struct vm_region *region)
561 {
562         struct vm_region *pregion;
563         struct rb_node **p, *parent;
564
565         validate_nommu_regions();
566
567         parent = NULL;
568         p = &nommu_region_tree.rb_node;
569         while (*p) {
570                 parent = *p;
571                 pregion = rb_entry(parent, struct vm_region, vm_rb);
572                 if (region->vm_start < pregion->vm_start)
573                         p = &(*p)->rb_left;
574                 else if (region->vm_start > pregion->vm_start)
575                         p = &(*p)->rb_right;
576                 else if (pregion == region)
577                         return;
578                 else
579                         BUG();
580         }
581
582         rb_link_node(&region->vm_rb, parent, p);
583         rb_insert_color(&region->vm_rb, &nommu_region_tree);
584
585         validate_nommu_regions();
586 }
587
588 /*
589  * delete a region from the global tree
590  */
591 static void delete_nommu_region(struct vm_region *region)
592 {
593         BUG_ON(!nommu_region_tree.rb_node);
594
595         validate_nommu_regions();
596         rb_erase(&region->vm_rb, &nommu_region_tree);
597         validate_nommu_regions();
598 }
599
600 /*
601  * free a contiguous series of pages
602  */
603 static void free_page_series(unsigned long from, unsigned long to)
604 {
605         for (; from < to; from += PAGE_SIZE) {
606                 struct page *page = virt_to_page(from);
607
608                 kdebug("- free %lx", from);
609                 atomic_long_dec(&mmap_pages_allocated);
610                 if (page_count(page) != 1)
611                         kdebug("free page %p: refcount not one: %d",
612                                page, page_count(page));
613                 put_page(page);
614         }
615 }
616
617 /*
618  * release a reference to a region
619  * - the caller must hold the region semaphore for writing, which this releases
620  * - the region may not have been added to the tree yet, in which case vm_top
621  *   will equal vm_start
622  */
623 static void __put_nommu_region(struct vm_region *region)
624         __releases(nommu_region_sem)
625 {
626         kenter("%p{%d}", region, region->vm_usage);
627
628         BUG_ON(!nommu_region_tree.rb_node);
629
630         if (--region->vm_usage == 0) {
631                 if (region->vm_top > region->vm_start)
632                         delete_nommu_region(region);
633                 up_write(&nommu_region_sem);
634
635                 if (region->vm_file)
636                         fput(region->vm_file);
637
638                 /* IO memory and memory shared directly out of the pagecache
639                  * from ramfs/tmpfs mustn't be released here */
640                 if (region->vm_flags & VM_MAPPED_COPY) {
641                         kdebug("free series");
642                         free_page_series(region->vm_start, region->vm_top);
643                 }
644                 kmem_cache_free(vm_region_jar, region);
645         } else {
646                 up_write(&nommu_region_sem);
647         }
648 }
649
650 /*
651  * release a reference to a region
652  */
653 static void put_nommu_region(struct vm_region *region)
654 {
655         down_write(&nommu_region_sem);
656         __put_nommu_region(region);
657 }
658
659 /*
660  * update protection on a vma
661  */
662 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
663 {
664 #ifdef CONFIG_MPU
665         struct mm_struct *mm = vma->vm_mm;
666         long start = vma->vm_start & PAGE_MASK;
667         while (start < vma->vm_end) {
668                 protect_page(mm, start, flags);
669                 start += PAGE_SIZE;
670         }
671         update_protections(mm);
672 #endif
673 }
674
675 /*
676  * add a VMA into a process's mm_struct in the appropriate place in the list
677  * and tree and add to the address space's page tree also if not an anonymous
678  * page
679  * - should be called with mm->mmap_sem held writelocked
680  */
681 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
682 {
683         struct vm_area_struct *pvma, **pp, *next;
684         struct address_space *mapping;
685         struct rb_node **p, *parent;
686
687         kenter(",%p", vma);
688
689         BUG_ON(!vma->vm_region);
690
691         mm->map_count++;
692         vma->vm_mm = mm;
693
694         protect_vma(vma, vma->vm_flags);
695
696         /* add the VMA to the mapping */
697         if (vma->vm_file) {
698                 mapping = vma->vm_file->f_mapping;
699
700                 flush_dcache_mmap_lock(mapping);
701                 vma_prio_tree_insert(vma, &mapping->i_mmap);
702                 flush_dcache_mmap_unlock(mapping);
703         }
704
705         /* add the VMA to the tree */
706         parent = NULL;
707         p = &mm->mm_rb.rb_node;
708         while (*p) {
709                 parent = *p;
710                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
711
712                 /* sort by: start addr, end addr, VMA struct addr in that order
713                  * (the latter is necessary as we may get identical VMAs) */
714                 if (vma->vm_start < pvma->vm_start)
715                         p = &(*p)->rb_left;
716                 else if (vma->vm_start > pvma->vm_start)
717                         p = &(*p)->rb_right;
718                 else if (vma->vm_end < pvma->vm_end)
719                         p = &(*p)->rb_left;
720                 else if (vma->vm_end > pvma->vm_end)
721                         p = &(*p)->rb_right;
722                 else if (vma < pvma)
723                         p = &(*p)->rb_left;
724                 else if (vma > pvma)
725                         p = &(*p)->rb_right;
726                 else
727                         BUG();
728         }
729
730         rb_link_node(&vma->vm_rb, parent, p);
731         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
732
733         /* add VMA to the VMA list also */
734         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
735                 if (pvma->vm_start > vma->vm_start)
736                         break;
737                 if (pvma->vm_start < vma->vm_start)
738                         continue;
739                 if (pvma->vm_end < vma->vm_end)
740                         break;
741         }
742
743         next = *pp;
744         *pp = vma;
745         vma->vm_next = next;
746         if (next)
747                 next->vm_prev = vma;
748 }
749
750 /*
751  * delete a VMA from its owning mm_struct and address space
752  */
753 static void delete_vma_from_mm(struct vm_area_struct *vma)
754 {
755         struct vm_area_struct **pp;
756         struct address_space *mapping;
757         struct mm_struct *mm = vma->vm_mm;
758
759         kenter("%p", vma);
760
761         protect_vma(vma, 0);
762
763         mm->map_count--;
764         if (mm->mmap_cache == vma)
765                 mm->mmap_cache = NULL;
766
767         /* remove the VMA from the mapping */
768         if (vma->vm_file) {
769                 mapping = vma->vm_file->f_mapping;
770
771                 flush_dcache_mmap_lock(mapping);
772                 vma_prio_tree_remove(vma, &mapping->i_mmap);
773                 flush_dcache_mmap_unlock(mapping);
774         }
775
776         /* remove from the MM's tree and list */
777         rb_erase(&vma->vm_rb, &mm->mm_rb);
778         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
779                 if (*pp == vma) {
780                         *pp = vma->vm_next;
781                         break;
782                 }
783         }
784
785         vma->vm_mm = NULL;
786 }
787
788 /*
789  * destroy a VMA record
790  */
791 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
792 {
793         kenter("%p", vma);
794         if (vma->vm_ops && vma->vm_ops->close)
795                 vma->vm_ops->close(vma);
796         if (vma->vm_file) {
797                 fput(vma->vm_file);
798                 if (vma->vm_flags & VM_EXECUTABLE)
799                         removed_exe_file_vma(mm);
800         }
801         put_nommu_region(vma->vm_region);
802         kmem_cache_free(vm_area_cachep, vma);
803 }
804
805 /*
806  * look up the first VMA in which addr resides, NULL if none
807  * - should be called with mm->mmap_sem at least held readlocked
808  */
809 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
810 {
811         struct vm_area_struct *vma;
812         struct rb_node *n = mm->mm_rb.rb_node;
813
814         /* check the cache first */
815         vma = mm->mmap_cache;
816         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
817                 return vma;
818
819         /* trawl the tree (there may be multiple mappings in which addr
820          * resides) */
821         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
822                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
823                 if (vma->vm_start > addr)
824                         return NULL;
825                 if (vma->vm_end > addr) {
826                         mm->mmap_cache = vma;
827                         return vma;
828                 }
829         }
830
831         return NULL;
832 }
833 EXPORT_SYMBOL(find_vma);
834
835 /*
836  * find a VMA
837  * - we don't extend stack VMAs under NOMMU conditions
838  */
839 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
840 {
841         return find_vma(mm, addr);
842 }
843
844 /*
845  * expand a stack to a given address
846  * - not supported under NOMMU conditions
847  */
848 int expand_stack(struct vm_area_struct *vma, unsigned long address)
849 {
850         return -ENOMEM;
851 }
852
853 /*
854  * look up the first VMA exactly that exactly matches addr
855  * - should be called with mm->mmap_sem at least held readlocked
856  */
857 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
858                                              unsigned long addr,
859                                              unsigned long len)
860 {
861         struct vm_area_struct *vma;
862         struct rb_node *n = mm->mm_rb.rb_node;
863         unsigned long end = addr + len;
864
865         /* check the cache first */
866         vma = mm->mmap_cache;
867         if (vma && vma->vm_start == addr && vma->vm_end == end)
868                 return vma;
869
870         /* trawl the tree (there may be multiple mappings in which addr
871          * resides) */
872         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
873                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
874                 if (vma->vm_start < addr)
875                         continue;
876                 if (vma->vm_start > addr)
877                         return NULL;
878                 if (vma->vm_end == end) {
879                         mm->mmap_cache = vma;
880                         return vma;
881                 }
882         }
883
884         return NULL;
885 }
886
887 /*
888  * determine whether a mapping should be permitted and, if so, what sort of
889  * mapping we're capable of supporting
890  */
891 static int validate_mmap_request(struct file *file,
892                                  unsigned long addr,
893                                  unsigned long len,
894                                  unsigned long prot,
895                                  unsigned long flags,
896                                  unsigned long pgoff,
897                                  unsigned long *_capabilities)
898 {
899         unsigned long capabilities, rlen;
900         unsigned long reqprot = prot;
901         int ret;
902
903         /* do the simple checks first */
904         if (flags & MAP_FIXED) {
905                 printk(KERN_DEBUG
906                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
907                        current->pid);
908                 return -EINVAL;
909         }
910
911         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
912             (flags & MAP_TYPE) != MAP_SHARED)
913                 return -EINVAL;
914
915         if (!len)
916                 return -EINVAL;
917
918         /* Careful about overflows.. */
919         rlen = PAGE_ALIGN(len);
920         if (!rlen || rlen > TASK_SIZE)
921                 return -ENOMEM;
922
923         /* offset overflow? */
924         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
925                 return -EOVERFLOW;
926
927         if (file) {
928                 /* validate file mapping requests */
929                 struct address_space *mapping;
930
931                 /* files must support mmap */
932                 if (!file->f_op || !file->f_op->mmap)
933                         return -ENODEV;
934
935                 /* work out if what we've got could possibly be shared
936                  * - we support chardevs that provide their own "memory"
937                  * - we support files/blockdevs that are memory backed
938                  */
939                 mapping = file->f_mapping;
940                 if (!mapping)
941                         mapping = file->f_path.dentry->d_inode->i_mapping;
942
943                 capabilities = 0;
944                 if (mapping && mapping->backing_dev_info)
945                         capabilities = mapping->backing_dev_info->capabilities;
946
947                 if (!capabilities) {
948                         /* no explicit capabilities set, so assume some
949                          * defaults */
950                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
951                         case S_IFREG:
952                         case S_IFBLK:
953                                 capabilities = BDI_CAP_MAP_COPY;
954                                 break;
955
956                         case S_IFCHR:
957                                 capabilities =
958                                         BDI_CAP_MAP_DIRECT |
959                                         BDI_CAP_READ_MAP |
960                                         BDI_CAP_WRITE_MAP;
961                                 break;
962
963                         default:
964                                 return -EINVAL;
965                         }
966                 }
967
968                 /* eliminate any capabilities that we can't support on this
969                  * device */
970                 if (!file->f_op->get_unmapped_area)
971                         capabilities &= ~BDI_CAP_MAP_DIRECT;
972                 if (!file->f_op->read)
973                         capabilities &= ~BDI_CAP_MAP_COPY;
974
975                 /* The file shall have been opened with read permission. */
976                 if (!(file->f_mode & FMODE_READ))
977                         return -EACCES;
978
979                 if (flags & MAP_SHARED) {
980                         /* do checks for writing, appending and locking */
981                         if ((prot & PROT_WRITE) &&
982                             !(file->f_mode & FMODE_WRITE))
983                                 return -EACCES;
984
985                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
986                             (file->f_mode & FMODE_WRITE))
987                                 return -EACCES;
988
989                         if (locks_verify_locked(file->f_path.dentry->d_inode))
990                                 return -EAGAIN;
991
992                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
993                                 return -ENODEV;
994
995                         /* we mustn't privatise shared mappings */
996                         capabilities &= ~BDI_CAP_MAP_COPY;
997                 }
998                 else {
999                         /* we're going to read the file into private memory we
1000                          * allocate */
1001                         if (!(capabilities & BDI_CAP_MAP_COPY))
1002                                 return -ENODEV;
1003
1004                         /* we don't permit a private writable mapping to be
1005                          * shared with the backing device */
1006                         if (prot & PROT_WRITE)
1007                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1008                 }
1009
1010                 if (capabilities & BDI_CAP_MAP_DIRECT) {
1011                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
1012                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1013                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
1014                             ) {
1015                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1016                                 if (flags & MAP_SHARED) {
1017                                         printk(KERN_WARNING
1018                                                "MAP_SHARED not completely supported on !MMU\n");
1019                                         return -EINVAL;
1020                                 }
1021                         }
1022                 }
1023
1024                 /* handle executable mappings and implied executable
1025                  * mappings */
1026                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1027                         if (prot & PROT_EXEC)
1028                                 return -EPERM;
1029                 }
1030                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1031                         /* handle implication of PROT_EXEC by PROT_READ */
1032                         if (current->personality & READ_IMPLIES_EXEC) {
1033                                 if (capabilities & BDI_CAP_EXEC_MAP)
1034                                         prot |= PROT_EXEC;
1035                         }
1036                 }
1037                 else if ((prot & PROT_READ) &&
1038                          (prot & PROT_EXEC) &&
1039                          !(capabilities & BDI_CAP_EXEC_MAP)
1040                          ) {
1041                         /* backing file is not executable, try to copy */
1042                         capabilities &= ~BDI_CAP_MAP_DIRECT;
1043                 }
1044         }
1045         else {
1046                 /* anonymous mappings are always memory backed and can be
1047                  * privately mapped
1048                  */
1049                 capabilities = BDI_CAP_MAP_COPY;
1050
1051                 /* handle PROT_EXEC implication by PROT_READ */
1052                 if ((prot & PROT_READ) &&
1053                     (current->personality & READ_IMPLIES_EXEC))
1054                         prot |= PROT_EXEC;
1055         }
1056
1057         /* allow the security API to have its say */
1058         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1059         if (ret < 0)
1060                 return ret;
1061
1062         /* looks okay */
1063         *_capabilities = capabilities;
1064         return 0;
1065 }
1066
1067 /*
1068  * we've determined that we can make the mapping, now translate what we
1069  * now know into VMA flags
1070  */
1071 static unsigned long determine_vm_flags(struct file *file,
1072                                         unsigned long prot,
1073                                         unsigned long flags,
1074                                         unsigned long capabilities)
1075 {
1076         unsigned long vm_flags;
1077
1078         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1079         /* vm_flags |= mm->def_flags; */
1080
1081         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1082                 /* attempt to share read-only copies of mapped file chunks */
1083                 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1084                 if (file && !(prot & PROT_WRITE))
1085                         vm_flags |= VM_MAYSHARE;
1086         } else {
1087                 /* overlay a shareable mapping on the backing device or inode
1088                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1089                  * romfs/cramfs */
1090                 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1091                 if (flags & MAP_SHARED)
1092                         vm_flags |= VM_SHARED;
1093         }
1094
1095         /* refuse to let anyone share private mappings with this process if
1096          * it's being traced - otherwise breakpoints set in it may interfere
1097          * with another untraced process
1098          */
1099         if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1100                 vm_flags &= ~VM_MAYSHARE;
1101
1102         return vm_flags;
1103 }
1104
1105 /*
1106  * set up a shared mapping on a file (the driver or filesystem provides and
1107  * pins the storage)
1108  */
1109 static int do_mmap_shared_file(struct vm_area_struct *vma)
1110 {
1111         int ret;
1112
1113         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1114         if (ret == 0) {
1115                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1116                 return 0;
1117         }
1118         if (ret != -ENOSYS)
1119                 return ret;
1120
1121         /* getting -ENOSYS indicates that direct mmap isn't possible (as
1122          * opposed to tried but failed) so we can only give a suitable error as
1123          * it's not possible to make a private copy if MAP_SHARED was given */
1124         return -ENODEV;
1125 }
1126
1127 /*
1128  * set up a private mapping or an anonymous shared mapping
1129  */
1130 static int do_mmap_private(struct vm_area_struct *vma,
1131                            struct vm_region *region,
1132                            unsigned long len,
1133                            unsigned long capabilities)
1134 {
1135         struct page *pages;
1136         unsigned long total, point, n, rlen;
1137         void *base;
1138         int ret, order;
1139
1140         /* invoke the file's mapping function so that it can keep track of
1141          * shared mappings on devices or memory
1142          * - VM_MAYSHARE will be set if it may attempt to share
1143          */
1144         if (capabilities & BDI_CAP_MAP_DIRECT) {
1145                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1146                 if (ret == 0) {
1147                         /* shouldn't return success if we're not sharing */
1148                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1149                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1150                         return 0;
1151                 }
1152                 if (ret != -ENOSYS)
1153                         return ret;
1154
1155                 /* getting an ENOSYS error indicates that direct mmap isn't
1156                  * possible (as opposed to tried but failed) so we'll try to
1157                  * make a private copy of the data and map that instead */
1158         }
1159
1160         rlen = PAGE_ALIGN(len);
1161
1162         /* allocate some memory to hold the mapping
1163          * - note that this may not return a page-aligned address if the object
1164          *   we're allocating is smaller than a page
1165          */
1166         order = get_order(rlen);
1167         kdebug("alloc order %d for %lx", order, len);
1168
1169         pages = alloc_pages(GFP_KERNEL, order);
1170         if (!pages)
1171                 goto enomem;
1172
1173         total = 1 << order;
1174         atomic_long_add(total, &mmap_pages_allocated);
1175
1176         point = rlen >> PAGE_SHIFT;
1177
1178         /* we allocated a power-of-2 sized page set, so we may want to trim off
1179          * the excess */
1180         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1181                 while (total > point) {
1182                         order = ilog2(total - point);
1183                         n = 1 << order;
1184                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1185                         atomic_long_sub(n, &mmap_pages_allocated);
1186                         total -= n;
1187                         set_page_refcounted(pages + total);
1188                         __free_pages(pages + total, order);
1189                 }
1190         }
1191
1192         for (point = 1; point < total; point++)
1193                 set_page_refcounted(&pages[point]);
1194
1195         base = page_address(pages);
1196         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1197         region->vm_start = (unsigned long) base;
1198         region->vm_end   = region->vm_start + rlen;
1199         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1200
1201         vma->vm_start = region->vm_start;
1202         vma->vm_end   = region->vm_start + len;
1203
1204         if (vma->vm_file) {
1205                 /* read the contents of a file into the copy */
1206                 mm_segment_t old_fs;
1207                 loff_t fpos;
1208
1209                 fpos = vma->vm_pgoff;
1210                 fpos <<= PAGE_SHIFT;
1211
1212                 old_fs = get_fs();
1213                 set_fs(KERNEL_DS);
1214                 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1215                 set_fs(old_fs);
1216
1217                 if (ret < 0)
1218                         goto error_free;
1219
1220                 /* clear the last little bit */
1221                 if (ret < rlen)
1222                         memset(base + ret, 0, rlen - ret);
1223
1224         }
1225
1226         return 0;
1227
1228 error_free:
1229         free_page_series(region->vm_start, region->vm_end);
1230         region->vm_start = vma->vm_start = 0;
1231         region->vm_end   = vma->vm_end = 0;
1232         region->vm_top   = 0;
1233         return ret;
1234
1235 enomem:
1236         printk("Allocation of length %lu from process %d (%s) failed\n",
1237                len, current->pid, current->comm);
1238         show_free_areas();
1239         return -ENOMEM;
1240 }
1241
1242 /*
1243  * handle mapping creation for uClinux
1244  */
1245 unsigned long do_mmap_pgoff(struct file *file,
1246                             unsigned long addr,
1247                             unsigned long len,
1248                             unsigned long prot,
1249                             unsigned long flags,
1250                             unsigned long pgoff)
1251 {
1252         struct vm_area_struct *vma;
1253         struct vm_region *region;
1254         struct rb_node *rb;
1255         unsigned long capabilities, vm_flags, result;
1256         int ret;
1257
1258         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1259
1260         /* decide whether we should attempt the mapping, and if so what sort of
1261          * mapping */
1262         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1263                                     &capabilities);
1264         if (ret < 0) {
1265                 kleave(" = %d [val]", ret);
1266                 return ret;
1267         }
1268
1269         /* we ignore the address hint */
1270         addr = 0;
1271
1272         /* we've determined that we can make the mapping, now translate what we
1273          * now know into VMA flags */
1274         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1275
1276         /* we're going to need to record the mapping */
1277         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1278         if (!region)
1279                 goto error_getting_region;
1280
1281         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1282         if (!vma)
1283                 goto error_getting_vma;
1284
1285         region->vm_usage = 1;
1286         region->vm_flags = vm_flags;
1287         region->vm_pgoff = pgoff;
1288
1289         INIT_LIST_HEAD(&vma->anon_vma_chain);
1290         vma->vm_flags = vm_flags;
1291         vma->vm_pgoff = pgoff;
1292
1293         if (file) {
1294                 region->vm_file = file;
1295                 get_file(file);
1296                 vma->vm_file = file;
1297                 get_file(file);
1298                 if (vm_flags & VM_EXECUTABLE) {
1299                         added_exe_file_vma(current->mm);
1300                         vma->vm_mm = current->mm;
1301                 }
1302         }
1303
1304         down_write(&nommu_region_sem);
1305
1306         /* if we want to share, we need to check for regions created by other
1307          * mmap() calls that overlap with our proposed mapping
1308          * - we can only share with a superset match on most regular files
1309          * - shared mappings on character devices and memory backed files are
1310          *   permitted to overlap inexactly as far as we are concerned for in
1311          *   these cases, sharing is handled in the driver or filesystem rather
1312          *   than here
1313          */
1314         if (vm_flags & VM_MAYSHARE) {
1315                 struct vm_region *pregion;
1316                 unsigned long pglen, rpglen, pgend, rpgend, start;
1317
1318                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1319                 pgend = pgoff + pglen;
1320
1321                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1322                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1323
1324                         if (!(pregion->vm_flags & VM_MAYSHARE))
1325                                 continue;
1326
1327                         /* search for overlapping mappings on the same file */
1328                         if (pregion->vm_file->f_path.dentry->d_inode !=
1329                             file->f_path.dentry->d_inode)
1330                                 continue;
1331
1332                         if (pregion->vm_pgoff >= pgend)
1333                                 continue;
1334
1335                         rpglen = pregion->vm_end - pregion->vm_start;
1336                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1337                         rpgend = pregion->vm_pgoff + rpglen;
1338                         if (pgoff >= rpgend)
1339                                 continue;
1340
1341                         /* handle inexactly overlapping matches between
1342                          * mappings */
1343                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1344                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1345                                 /* new mapping is not a subset of the region */
1346                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1347                                         goto sharing_violation;
1348                                 continue;
1349                         }
1350
1351                         /* we've found a region we can share */
1352                         pregion->vm_usage++;
1353                         vma->vm_region = pregion;
1354                         start = pregion->vm_start;
1355                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1356                         vma->vm_start = start;
1357                         vma->vm_end = start + len;
1358
1359                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1360                                 kdebug("share copy");
1361                                 vma->vm_flags |= VM_MAPPED_COPY;
1362                         } else {
1363                                 kdebug("share mmap");
1364                                 ret = do_mmap_shared_file(vma);
1365                                 if (ret < 0) {
1366                                         vma->vm_region = NULL;
1367                                         vma->vm_start = 0;
1368                                         vma->vm_end = 0;
1369                                         pregion->vm_usage--;
1370                                         pregion = NULL;
1371                                         goto error_just_free;
1372                                 }
1373                         }
1374                         fput(region->vm_file);
1375                         kmem_cache_free(vm_region_jar, region);
1376                         region = pregion;
1377                         result = start;
1378                         goto share;
1379                 }
1380
1381                 /* obtain the address at which to make a shared mapping
1382                  * - this is the hook for quasi-memory character devices to
1383                  *   tell us the location of a shared mapping
1384                  */
1385                 if (capabilities & BDI_CAP_MAP_DIRECT) {
1386                         addr = file->f_op->get_unmapped_area(file, addr, len,
1387                                                              pgoff, flags);
1388                         if (IS_ERR((void *) addr)) {
1389                                 ret = addr;
1390                                 if (ret != (unsigned long) -ENOSYS)
1391                                         goto error_just_free;
1392
1393                                 /* the driver refused to tell us where to site
1394                                  * the mapping so we'll have to attempt to copy
1395                                  * it */
1396                                 ret = (unsigned long) -ENODEV;
1397                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1398                                         goto error_just_free;
1399
1400                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1401                         } else {
1402                                 vma->vm_start = region->vm_start = addr;
1403                                 vma->vm_end = region->vm_end = addr + len;
1404                         }
1405                 }
1406         }
1407
1408         vma->vm_region = region;
1409
1410         /* set up the mapping
1411          * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1412          */
1413         if (file && vma->vm_flags & VM_SHARED)
1414                 ret = do_mmap_shared_file(vma);
1415         else
1416                 ret = do_mmap_private(vma, region, len, capabilities);
1417         if (ret < 0)
1418                 goto error_just_free;
1419         add_nommu_region(region);
1420
1421         /* clear anonymous mappings that don't ask for uninitialized data */
1422         if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1423                 memset((void *)region->vm_start, 0,
1424                        region->vm_end - region->vm_start);
1425
1426         /* okay... we have a mapping; now we have to register it */
1427         result = vma->vm_start;
1428
1429         current->mm->total_vm += len >> PAGE_SHIFT;
1430
1431 share:
1432         add_vma_to_mm(current->mm, vma);
1433
1434         /* we flush the region from the icache only when the first executable
1435          * mapping of it is made  */
1436         if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1437                 flush_icache_range(region->vm_start, region->vm_end);
1438                 region->vm_icache_flushed = true;
1439         }
1440
1441         up_write(&nommu_region_sem);
1442
1443         kleave(" = %lx", result);
1444         return result;
1445
1446 error_just_free:
1447         up_write(&nommu_region_sem);
1448 error:
1449         if (region->vm_file)
1450                 fput(region->vm_file);
1451         kmem_cache_free(vm_region_jar, region);
1452         if (vma->vm_file)
1453                 fput(vma->vm_file);
1454         if (vma->vm_flags & VM_EXECUTABLE)
1455                 removed_exe_file_vma(vma->vm_mm);
1456         kmem_cache_free(vm_area_cachep, vma);
1457         kleave(" = %d", ret);
1458         return ret;
1459
1460 sharing_violation:
1461         up_write(&nommu_region_sem);
1462         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1463         ret = -EINVAL;
1464         goto error;
1465
1466 error_getting_vma:
1467         kmem_cache_free(vm_region_jar, region);
1468         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1469                " from process %d failed\n",
1470                len, current->pid);
1471         show_free_areas();
1472         return -ENOMEM;
1473
1474 error_getting_region:
1475         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1476                " from process %d failed\n",
1477                len, current->pid);
1478         show_free_areas();
1479         return -ENOMEM;
1480 }
1481 EXPORT_SYMBOL(do_mmap_pgoff);
1482
1483 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1484                 unsigned long, prot, unsigned long, flags,
1485                 unsigned long, fd, unsigned long, pgoff)
1486 {
1487         struct file *file = NULL;
1488         unsigned long retval = -EBADF;
1489
1490         audit_mmap_fd(fd, flags);
1491         if (!(flags & MAP_ANONYMOUS)) {
1492                 file = fget(fd);
1493                 if (!file)
1494                         goto out;
1495         }
1496
1497         flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1498
1499         down_write(&current->mm->mmap_sem);
1500         retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1501         up_write(&current->mm->mmap_sem);
1502
1503         if (file)
1504                 fput(file);
1505 out:
1506         return retval;
1507 }
1508
1509 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1510 struct mmap_arg_struct {
1511         unsigned long addr;
1512         unsigned long len;
1513         unsigned long prot;
1514         unsigned long flags;
1515         unsigned long fd;
1516         unsigned long offset;
1517 };
1518
1519 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1520 {
1521         struct mmap_arg_struct a;
1522
1523         if (copy_from_user(&a, arg, sizeof(a)))
1524                 return -EFAULT;
1525         if (a.offset & ~PAGE_MASK)
1526                 return -EINVAL;
1527
1528         return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1529                               a.offset >> PAGE_SHIFT);
1530 }
1531 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1532
1533 /*
1534  * split a vma into two pieces at address 'addr', a new vma is allocated either
1535  * for the first part or the tail.
1536  */
1537 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1538               unsigned long addr, int new_below)
1539 {
1540         struct vm_area_struct *new;
1541         struct vm_region *region;
1542         unsigned long npages;
1543
1544         kenter("");
1545
1546         /* we're only permitted to split anonymous regions (these should have
1547          * only a single usage on the region) */
1548         if (vma->vm_file)
1549                 return -ENOMEM;
1550
1551         if (mm->map_count >= sysctl_max_map_count)
1552                 return -ENOMEM;
1553
1554         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1555         if (!region)
1556                 return -ENOMEM;
1557
1558         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1559         if (!new) {
1560                 kmem_cache_free(vm_region_jar, region);
1561                 return -ENOMEM;
1562         }
1563
1564         /* most fields are the same, copy all, and then fixup */
1565         *new = *vma;
1566         *region = *vma->vm_region;
1567         new->vm_region = region;
1568
1569         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1570
1571         if (new_below) {
1572                 region->vm_top = region->vm_end = new->vm_end = addr;
1573         } else {
1574                 region->vm_start = new->vm_start = addr;
1575                 region->vm_pgoff = new->vm_pgoff += npages;
1576         }
1577
1578         if (new->vm_ops && new->vm_ops->open)
1579                 new->vm_ops->open(new);
1580
1581         delete_vma_from_mm(vma);
1582         down_write(&nommu_region_sem);
1583         delete_nommu_region(vma->vm_region);
1584         if (new_below) {
1585                 vma->vm_region->vm_start = vma->vm_start = addr;
1586                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1587         } else {
1588                 vma->vm_region->vm_end = vma->vm_end = addr;
1589                 vma->vm_region->vm_top = addr;
1590         }
1591         add_nommu_region(vma->vm_region);
1592         add_nommu_region(new->vm_region);
1593         up_write(&nommu_region_sem);
1594         add_vma_to_mm(mm, vma);
1595         add_vma_to_mm(mm, new);
1596         return 0;
1597 }
1598
1599 /*
1600  * shrink a VMA by removing the specified chunk from either the beginning or
1601  * the end
1602  */
1603 static int shrink_vma(struct mm_struct *mm,
1604                       struct vm_area_struct *vma,
1605                       unsigned long from, unsigned long to)
1606 {
1607         struct vm_region *region;
1608
1609         kenter("");
1610
1611         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1612          * and list */
1613         delete_vma_from_mm(vma);
1614         if (from > vma->vm_start)
1615                 vma->vm_end = from;
1616         else
1617                 vma->vm_start = to;
1618         add_vma_to_mm(mm, vma);
1619
1620         /* cut the backing region down to size */
1621         region = vma->vm_region;
1622         BUG_ON(region->vm_usage != 1);
1623
1624         down_write(&nommu_region_sem);
1625         delete_nommu_region(region);
1626         if (from > region->vm_start) {
1627                 to = region->vm_top;
1628                 region->vm_top = region->vm_end = from;
1629         } else {
1630                 region->vm_start = to;
1631         }
1632         add_nommu_region(region);
1633         up_write(&nommu_region_sem);
1634
1635         free_page_series(from, to);
1636         return 0;
1637 }
1638
1639 /*
1640  * release a mapping
1641  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1642  *   VMA, though it need not cover the whole VMA
1643  */
1644 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1645 {
1646         struct vm_area_struct *vma;
1647         struct rb_node *rb;
1648         unsigned long end = start + len;
1649         int ret;
1650
1651         kenter(",%lx,%zx", start, len);
1652
1653         if (len == 0)
1654                 return -EINVAL;
1655
1656         /* find the first potentially overlapping VMA */
1657         vma = find_vma(mm, start);
1658         if (!vma) {
1659                 static int limit = 0;
1660                 if (limit < 5) {
1661                         printk(KERN_WARNING
1662                                "munmap of memory not mmapped by process %d"
1663                                " (%s): 0x%lx-0x%lx\n",
1664                                current->pid, current->comm,
1665                                start, start + len - 1);
1666                         limit++;
1667                 }
1668                 return -EINVAL;
1669         }
1670
1671         /* we're allowed to split an anonymous VMA but not a file-backed one */
1672         if (vma->vm_file) {
1673                 do {
1674                         if (start > vma->vm_start) {
1675                                 kleave(" = -EINVAL [miss]");
1676                                 return -EINVAL;
1677                         }
1678                         if (end == vma->vm_end)
1679                                 goto erase_whole_vma;
1680                         rb = rb_next(&vma->vm_rb);
1681                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1682                 } while (rb);
1683                 kleave(" = -EINVAL [split file]");
1684                 return -EINVAL;
1685         } else {
1686                 /* the chunk must be a subset of the VMA found */
1687                 if (start == vma->vm_start && end == vma->vm_end)
1688                         goto erase_whole_vma;
1689                 if (start < vma->vm_start || end > vma->vm_end) {
1690                         kleave(" = -EINVAL [superset]");
1691                         return -EINVAL;
1692                 }
1693                 if (start & ~PAGE_MASK) {
1694                         kleave(" = -EINVAL [unaligned start]");
1695                         return -EINVAL;
1696                 }
1697                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1698                         kleave(" = -EINVAL [unaligned split]");
1699                         return -EINVAL;
1700                 }
1701                 if (start != vma->vm_start && end != vma->vm_end) {
1702                         ret = split_vma(mm, vma, start, 1);
1703                         if (ret < 0) {
1704                                 kleave(" = %d [split]", ret);
1705                                 return ret;
1706                         }
1707                 }
1708                 return shrink_vma(mm, vma, start, end);
1709         }
1710
1711 erase_whole_vma:
1712         delete_vma_from_mm(vma);
1713         delete_vma(mm, vma);
1714         kleave(" = 0");
1715         return 0;
1716 }
1717 EXPORT_SYMBOL(do_munmap);
1718
1719 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1720 {
1721         int ret;
1722         struct mm_struct *mm = current->mm;
1723
1724         down_write(&mm->mmap_sem);
1725         ret = do_munmap(mm, addr, len);
1726         up_write(&mm->mmap_sem);
1727         return ret;
1728 }
1729
1730 /*
1731  * release all the mappings made in a process's VM space
1732  */
1733 void exit_mmap(struct mm_struct *mm)
1734 {
1735         struct vm_area_struct *vma;
1736
1737         if (!mm)
1738                 return;
1739
1740         kenter("");
1741
1742         mm->total_vm = 0;
1743
1744         while ((vma = mm->mmap)) {
1745                 mm->mmap = vma->vm_next;
1746                 delete_vma_from_mm(vma);
1747                 delete_vma(mm, vma);
1748                 cond_resched();
1749         }
1750
1751         kleave("");
1752 }
1753
1754 unsigned long do_brk(unsigned long addr, unsigned long len)
1755 {
1756         return -ENOMEM;
1757 }
1758
1759 /*
1760  * expand (or shrink) an existing mapping, potentially moving it at the same
1761  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1762  *
1763  * under NOMMU conditions, we only permit changing a mapping's size, and only
1764  * as long as it stays within the region allocated by do_mmap_private() and the
1765  * block is not shareable
1766  *
1767  * MREMAP_FIXED is not supported under NOMMU conditions
1768  */
1769 unsigned long do_mremap(unsigned long addr,
1770                         unsigned long old_len, unsigned long new_len,
1771                         unsigned long flags, unsigned long new_addr)
1772 {
1773         struct vm_area_struct *vma;
1774
1775         /* insanity checks first */
1776         if (old_len == 0 || new_len == 0)
1777                 return (unsigned long) -EINVAL;
1778
1779         if (addr & ~PAGE_MASK)
1780                 return -EINVAL;
1781
1782         if (flags & MREMAP_FIXED && new_addr != addr)
1783                 return (unsigned long) -EINVAL;
1784
1785         vma = find_vma_exact(current->mm, addr, old_len);
1786         if (!vma)
1787                 return (unsigned long) -EINVAL;
1788
1789         if (vma->vm_end != vma->vm_start + old_len)
1790                 return (unsigned long) -EFAULT;
1791
1792         if (vma->vm_flags & VM_MAYSHARE)
1793                 return (unsigned long) -EPERM;
1794
1795         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1796                 return (unsigned long) -ENOMEM;
1797
1798         /* all checks complete - do it */
1799         vma->vm_end = vma->vm_start + new_len;
1800         return vma->vm_start;
1801 }
1802 EXPORT_SYMBOL(do_mremap);
1803
1804 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1805                 unsigned long, new_len, unsigned long, flags,
1806                 unsigned long, new_addr)
1807 {
1808         unsigned long ret;
1809
1810         down_write(&current->mm->mmap_sem);
1811         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1812         up_write(&current->mm->mmap_sem);
1813         return ret;
1814 }
1815
1816 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1817                         unsigned int foll_flags)
1818 {
1819         return NULL;
1820 }
1821
1822 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1823                 unsigned long to, unsigned long size, pgprot_t prot)
1824 {
1825         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1826         return 0;
1827 }
1828 EXPORT_SYMBOL(remap_pfn_range);
1829
1830 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1831                         unsigned long pgoff)
1832 {
1833         unsigned int size = vma->vm_end - vma->vm_start;
1834
1835         if (!(vma->vm_flags & VM_USERMAP))
1836                 return -EINVAL;
1837
1838         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1839         vma->vm_end = vma->vm_start + size;
1840
1841         return 0;
1842 }
1843 EXPORT_SYMBOL(remap_vmalloc_range);
1844
1845 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1846 {
1847 }
1848
1849 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1850         unsigned long len, unsigned long pgoff, unsigned long flags)
1851 {
1852         return -ENOMEM;
1853 }
1854
1855 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1856 {
1857 }
1858
1859 void unmap_mapping_range(struct address_space *mapping,
1860                          loff_t const holebegin, loff_t const holelen,
1861                          int even_cows)
1862 {
1863 }
1864 EXPORT_SYMBOL(unmap_mapping_range);
1865
1866 /*
1867  * Check that a process has enough memory to allocate a new virtual
1868  * mapping. 0 means there is enough memory for the allocation to
1869  * succeed and -ENOMEM implies there is not.
1870  *
1871  * We currently support three overcommit policies, which are set via the
1872  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1873  *
1874  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1875  * Additional code 2002 Jul 20 by Robert Love.
1876  *
1877  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1878  *
1879  * Note this is a helper function intended to be used by LSMs which
1880  * wish to use this logic.
1881  */
1882 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1883 {
1884         unsigned long free, allowed;
1885
1886         vm_acct_memory(pages);
1887
1888         /*
1889          * Sometimes we want to use more memory than we have
1890          */
1891         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1892                 return 0;
1893
1894         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1895                 unsigned long n;
1896
1897                 free = global_page_state(NR_FILE_PAGES);
1898                 free += nr_swap_pages;
1899
1900                 /*
1901                  * Any slabs which are created with the
1902                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1903                  * which are reclaimable, under pressure.  The dentry
1904                  * cache and most inode caches should fall into this
1905                  */
1906                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1907
1908                 /*
1909                  * Leave the last 3% for root
1910                  */
1911                 if (!cap_sys_admin)
1912                         free -= free / 32;
1913
1914                 if (free > pages)
1915                         return 0;
1916
1917                 /*
1918                  * nr_free_pages() is very expensive on large systems,
1919                  * only call if we're about to fail.
1920                  */
1921                 n = nr_free_pages();
1922
1923                 /*
1924                  * Leave reserved pages. The pages are not for anonymous pages.
1925                  */
1926                 if (n <= totalreserve_pages)
1927                         goto error;
1928                 else
1929                         n -= totalreserve_pages;
1930
1931                 /*
1932                  * Leave the last 3% for root
1933                  */
1934                 if (!cap_sys_admin)
1935                         n -= n / 32;
1936                 free += n;
1937
1938                 if (free > pages)
1939                         return 0;
1940
1941                 goto error;
1942         }
1943
1944         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1945         /*
1946          * Leave the last 3% for root
1947          */
1948         if (!cap_sys_admin)
1949                 allowed -= allowed / 32;
1950         allowed += total_swap_pages;
1951
1952         /* Don't let a single process grow too big:
1953            leave 3% of the size of this process for other processes */
1954         if (mm)
1955                 allowed -= mm->total_vm / 32;
1956
1957         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1958                 return 0;
1959
1960 error:
1961         vm_unacct_memory(pages);
1962
1963         return -ENOMEM;
1964 }
1965
1966 int in_gate_area_no_task(unsigned long addr)
1967 {
1968         return 0;
1969 }
1970
1971 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1972 {
1973         BUG();
1974         return 0;
1975 }
1976 EXPORT_SYMBOL(filemap_fault);
1977
1978 /*
1979  * Access another process' address space.
1980  * - source/target buffer must be kernel space
1981  */
1982 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1983 {
1984         struct vm_area_struct *vma;
1985         struct mm_struct *mm;
1986
1987         if (addr + len < addr)
1988                 return 0;
1989
1990         mm = get_task_mm(tsk);
1991         if (!mm)
1992                 return 0;
1993
1994         down_read(&mm->mmap_sem);
1995
1996         /* the access must start within one of the target process's mappings */
1997         vma = find_vma(mm, addr);
1998         if (vma) {
1999                 /* don't overrun this mapping */
2000                 if (addr + len >= vma->vm_end)
2001                         len = vma->vm_end - addr;
2002
2003                 /* only read or write mappings where it is permitted */
2004                 if (write && vma->vm_flags & VM_MAYWRITE)
2005                         copy_to_user_page(vma, NULL, addr,
2006                                          (void *) addr, buf, len);
2007                 else if (!write && vma->vm_flags & VM_MAYREAD)
2008                         copy_from_user_page(vma, NULL, addr,
2009                                             buf, (void *) addr, len);
2010                 else
2011                         len = 0;
2012         } else {
2013                 len = 0;
2014         }
2015
2016         up_read(&mm->mmap_sem);
2017         mmput(mm);
2018         return len;
2019 }
2020
2021 /**
2022  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2023  * @inode: The inode to check
2024  * @size: The current filesize of the inode
2025  * @newsize: The proposed filesize of the inode
2026  *
2027  * Check the shared mappings on an inode on behalf of a shrinking truncate to
2028  * make sure that that any outstanding VMAs aren't broken and then shrink the
2029  * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2030  * automatically grant mappings that are too large.
2031  */
2032 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2033                                 size_t newsize)
2034 {
2035         struct vm_area_struct *vma;
2036         struct prio_tree_iter iter;
2037         struct vm_region *region;
2038         pgoff_t low, high;
2039         size_t r_size, r_top;
2040
2041         low = newsize >> PAGE_SHIFT;
2042         high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2043
2044         down_write(&nommu_region_sem);
2045
2046         /* search for VMAs that fall within the dead zone */
2047         vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2048                               low, high) {
2049                 /* found one - only interested if it's shared out of the page
2050                  * cache */
2051                 if (vma->vm_flags & VM_SHARED) {
2052                         up_write(&nommu_region_sem);
2053                         return -ETXTBSY; /* not quite true, but near enough */
2054                 }
2055         }
2056
2057         /* reduce any regions that overlap the dead zone - if in existence,
2058          * these will be pointed to by VMAs that don't overlap the dead zone
2059          *
2060          * we don't check for any regions that start beyond the EOF as there
2061          * shouldn't be any
2062          */
2063         vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2064                               0, ULONG_MAX) {
2065                 if (!(vma->vm_flags & VM_SHARED))
2066                         continue;
2067
2068                 region = vma->vm_region;
2069                 r_size = region->vm_top - region->vm_start;
2070                 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2071
2072                 if (r_top > newsize) {
2073                         region->vm_top -= r_top - newsize;
2074                         if (region->vm_end > region->vm_top)
2075                                 region->vm_end = region->vm_top;
2076                 }
2077         }
2078
2079         up_write(&nommu_region_sem);
2080         return 0;
2081 }