x86: Fix checking of SRAT when node 0 ram is not from 0
[linux-2.6.git] / mm / nommu.c
1 /*
2  *  linux/mm/nommu.c
3  *
4  *  Replacement code for mm functions to support CPU's that don't
5  *  have any form of memory management unit (thus no virtual memory).
6  *
7  *  See Documentation/nommu-mmap.txt
8  *
9  *  Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
10  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
13  *  Copyright (c) 2007-2009 Paul Mundt <lethal@linux-sh.org>
14  */
15
16 #include <linux/module.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/swap.h>
20 #include <linux/file.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/tracehook.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mount.h>
29 #include <linux/personality.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32
33 #include <asm/uaccess.h>
34 #include <asm/tlb.h>
35 #include <asm/tlbflush.h>
36 #include <asm/mmu_context.h>
37 #include "internal.h"
38
39 static inline __attribute__((format(printf, 1, 2)))
40 void no_printk(const char *fmt, ...)
41 {
42 }
43
44 #if 0
45 #define kenter(FMT, ...) \
46         printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
47 #define kleave(FMT, ...) \
48         printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
49 #define kdebug(FMT, ...) \
50         printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
51 #else
52 #define kenter(FMT, ...) \
53         no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
54 #define kleave(FMT, ...) \
55         no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
56 #define kdebug(FMT, ...) \
57         no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
58 #endif
59
60 void *high_memory;
61 struct page *mem_map;
62 unsigned long max_mapnr;
63 unsigned long num_physpages;
64 unsigned long highest_memmap_pfn;
65 struct percpu_counter vm_committed_as;
66 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
67 int sysctl_overcommit_ratio = 50; /* default is 50% */
68 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
69 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
70 int heap_stack_gap = 0;
71
72 atomic_long_t mmap_pages_allocated;
73
74 EXPORT_SYMBOL(mem_map);
75 EXPORT_SYMBOL(num_physpages);
76
77 /* list of mapped, potentially shareable regions */
78 static struct kmem_cache *vm_region_jar;
79 struct rb_root nommu_region_tree = RB_ROOT;
80 DECLARE_RWSEM(nommu_region_sem);
81
82 const struct vm_operations_struct generic_file_vm_ops = {
83 };
84
85 /*
86  * Return the total memory allocated for this pointer, not
87  * just what the caller asked for.
88  *
89  * Doesn't have to be accurate, i.e. may have races.
90  */
91 unsigned int kobjsize(const void *objp)
92 {
93         struct page *page;
94
95         /*
96          * If the object we have should not have ksize performed on it,
97          * return size of 0
98          */
99         if (!objp || !virt_addr_valid(objp))
100                 return 0;
101
102         page = virt_to_head_page(objp);
103
104         /*
105          * If the allocator sets PageSlab, we know the pointer came from
106          * kmalloc().
107          */
108         if (PageSlab(page))
109                 return ksize(objp);
110
111         /*
112          * If it's not a compound page, see if we have a matching VMA
113          * region. This test is intentionally done in reverse order,
114          * so if there's no VMA, we still fall through and hand back
115          * PAGE_SIZE for 0-order pages.
116          */
117         if (!PageCompound(page)) {
118                 struct vm_area_struct *vma;
119
120                 vma = find_vma(current->mm, (unsigned long)objp);
121                 if (vma)
122                         return vma->vm_end - vma->vm_start;
123         }
124
125         /*
126          * The ksize() function is only guaranteed to work for pointers
127          * returned by kmalloc(). So handle arbitrary pointers here.
128          */
129         return PAGE_SIZE << compound_order(page);
130 }
131
132 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
133                      unsigned long start, int nr_pages, unsigned int foll_flags,
134                      struct page **pages, struct vm_area_struct **vmas)
135 {
136         struct vm_area_struct *vma;
137         unsigned long vm_flags;
138         int i;
139
140         /* calculate required read or write permissions.
141          * If FOLL_FORCE is set, we only require the "MAY" flags.
142          */
143         vm_flags  = (foll_flags & FOLL_WRITE) ?
144                         (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
145         vm_flags &= (foll_flags & FOLL_FORCE) ?
146                         (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
147
148         for (i = 0; i < nr_pages; i++) {
149                 vma = find_vma(mm, start);
150                 if (!vma)
151                         goto finish_or_fault;
152
153                 /* protect what we can, including chardevs */
154                 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
155                     !(vm_flags & vma->vm_flags))
156                         goto finish_or_fault;
157
158                 if (pages) {
159                         pages[i] = virt_to_page(start);
160                         if (pages[i])
161                                 page_cache_get(pages[i]);
162                 }
163                 if (vmas)
164                         vmas[i] = vma;
165                 start += PAGE_SIZE;
166         }
167
168         return i;
169
170 finish_or_fault:
171         return i ? : -EFAULT;
172 }
173
174 /*
175  * get a list of pages in an address range belonging to the specified process
176  * and indicate the VMA that covers each page
177  * - this is potentially dodgy as we may end incrementing the page count of a
178  *   slab page or a secondary page from a compound page
179  * - don't permit access to VMAs that don't support it, such as I/O mappings
180  */
181 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
182         unsigned long start, int nr_pages, int write, int force,
183         struct page **pages, struct vm_area_struct **vmas)
184 {
185         int flags = 0;
186
187         if (write)
188                 flags |= FOLL_WRITE;
189         if (force)
190                 flags |= FOLL_FORCE;
191
192         return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
193 }
194 EXPORT_SYMBOL(get_user_pages);
195
196 /**
197  * follow_pfn - look up PFN at a user virtual address
198  * @vma: memory mapping
199  * @address: user virtual address
200  * @pfn: location to store found PFN
201  *
202  * Only IO mappings and raw PFN mappings are allowed.
203  *
204  * Returns zero and the pfn at @pfn on success, -ve otherwise.
205  */
206 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
207         unsigned long *pfn)
208 {
209         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
210                 return -EINVAL;
211
212         *pfn = address >> PAGE_SHIFT;
213         return 0;
214 }
215 EXPORT_SYMBOL(follow_pfn);
216
217 DEFINE_RWLOCK(vmlist_lock);
218 struct vm_struct *vmlist;
219
220 void vfree(const void *addr)
221 {
222         kfree(addr);
223 }
224 EXPORT_SYMBOL(vfree);
225
226 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
227 {
228         /*
229          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
230          * returns only a logical address.
231          */
232         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
233 }
234 EXPORT_SYMBOL(__vmalloc);
235
236 void *vmalloc_user(unsigned long size)
237 {
238         void *ret;
239
240         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
241                         PAGE_KERNEL);
242         if (ret) {
243                 struct vm_area_struct *vma;
244
245                 down_write(&current->mm->mmap_sem);
246                 vma = find_vma(current->mm, (unsigned long)ret);
247                 if (vma)
248                         vma->vm_flags |= VM_USERMAP;
249                 up_write(&current->mm->mmap_sem);
250         }
251
252         return ret;
253 }
254 EXPORT_SYMBOL(vmalloc_user);
255
256 struct page *vmalloc_to_page(const void *addr)
257 {
258         return virt_to_page(addr);
259 }
260 EXPORT_SYMBOL(vmalloc_to_page);
261
262 unsigned long vmalloc_to_pfn(const void *addr)
263 {
264         return page_to_pfn(virt_to_page(addr));
265 }
266 EXPORT_SYMBOL(vmalloc_to_pfn);
267
268 long vread(char *buf, char *addr, unsigned long count)
269 {
270         memcpy(buf, addr, count);
271         return count;
272 }
273
274 long vwrite(char *buf, char *addr, unsigned long count)
275 {
276         /* Don't allow overflow */
277         if ((unsigned long) addr + count < count)
278                 count = -(unsigned long) addr;
279
280         memcpy(addr, buf, count);
281         return(count);
282 }
283
284 /*
285  *      vmalloc  -  allocate virtually continguos memory
286  *
287  *      @size:          allocation size
288  *
289  *      Allocate enough pages to cover @size from the page level
290  *      allocator and map them into continguos kernel virtual space.
291  *
292  *      For tight control over page level allocator and protection flags
293  *      use __vmalloc() instead.
294  */
295 void *vmalloc(unsigned long size)
296 {
297        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
298 }
299 EXPORT_SYMBOL(vmalloc);
300
301 void *vmalloc_node(unsigned long size, int node)
302 {
303         return vmalloc(size);
304 }
305 EXPORT_SYMBOL(vmalloc_node);
306
307 #ifndef PAGE_KERNEL_EXEC
308 # define PAGE_KERNEL_EXEC PAGE_KERNEL
309 #endif
310
311 /**
312  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
313  *      @size:          allocation size
314  *
315  *      Kernel-internal function to allocate enough pages to cover @size
316  *      the page level allocator and map them into contiguous and
317  *      executable kernel virtual space.
318  *
319  *      For tight control over page level allocator and protection flags
320  *      use __vmalloc() instead.
321  */
322
323 void *vmalloc_exec(unsigned long size)
324 {
325         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
326 }
327
328 /**
329  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
330  *      @size:          allocation size
331  *
332  *      Allocate enough 32bit PA addressable pages to cover @size from the
333  *      page level allocator and map them into continguos kernel virtual space.
334  */
335 void *vmalloc_32(unsigned long size)
336 {
337         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
338 }
339 EXPORT_SYMBOL(vmalloc_32);
340
341 /**
342  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
343  *      @size:          allocation size
344  *
345  * The resulting memory area is 32bit addressable and zeroed so it can be
346  * mapped to userspace without leaking data.
347  *
348  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
349  * remap_vmalloc_range() are permissible.
350  */
351 void *vmalloc_32_user(unsigned long size)
352 {
353         /*
354          * We'll have to sort out the ZONE_DMA bits for 64-bit,
355          * but for now this can simply use vmalloc_user() directly.
356          */
357         return vmalloc_user(size);
358 }
359 EXPORT_SYMBOL(vmalloc_32_user);
360
361 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
362 {
363         BUG();
364         return NULL;
365 }
366 EXPORT_SYMBOL(vmap);
367
368 void vunmap(const void *addr)
369 {
370         BUG();
371 }
372 EXPORT_SYMBOL(vunmap);
373
374 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
375 {
376         BUG();
377         return NULL;
378 }
379 EXPORT_SYMBOL(vm_map_ram);
380
381 void vm_unmap_ram(const void *mem, unsigned int count)
382 {
383         BUG();
384 }
385 EXPORT_SYMBOL(vm_unmap_ram);
386
387 void vm_unmap_aliases(void)
388 {
389 }
390 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
391
392 /*
393  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
394  * have one.
395  */
396 void  __attribute__((weak)) vmalloc_sync_all(void)
397 {
398 }
399
400 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
401                    struct page *page)
402 {
403         return -EINVAL;
404 }
405 EXPORT_SYMBOL(vm_insert_page);
406
407 /*
408  *  sys_brk() for the most part doesn't need the global kernel
409  *  lock, except when an application is doing something nasty
410  *  like trying to un-brk an area that has already been mapped
411  *  to a regular file.  in this case, the unmapping will need
412  *  to invoke file system routines that need the global lock.
413  */
414 SYSCALL_DEFINE1(brk, unsigned long, brk)
415 {
416         struct mm_struct *mm = current->mm;
417
418         if (brk < mm->start_brk || brk > mm->context.end_brk)
419                 return mm->brk;
420
421         if (mm->brk == brk)
422                 return mm->brk;
423
424         /*
425          * Always allow shrinking brk
426          */
427         if (brk <= mm->brk) {
428                 mm->brk = brk;
429                 return brk;
430         }
431
432         /*
433          * Ok, looks good - let it rip.
434          */
435         return mm->brk = brk;
436 }
437
438 /*
439  * initialise the VMA and region record slabs
440  */
441 void __init mmap_init(void)
442 {
443         int ret;
444
445         ret = percpu_counter_init(&vm_committed_as, 0);
446         VM_BUG_ON(ret);
447         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
448 }
449
450 /*
451  * validate the region tree
452  * - the caller must hold the region lock
453  */
454 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
455 static noinline void validate_nommu_regions(void)
456 {
457         struct vm_region *region, *last;
458         struct rb_node *p, *lastp;
459
460         lastp = rb_first(&nommu_region_tree);
461         if (!lastp)
462                 return;
463
464         last = rb_entry(lastp, struct vm_region, vm_rb);
465         BUG_ON(unlikely(last->vm_end <= last->vm_start));
466         BUG_ON(unlikely(last->vm_top < last->vm_end));
467
468         while ((p = rb_next(lastp))) {
469                 region = rb_entry(p, struct vm_region, vm_rb);
470                 last = rb_entry(lastp, struct vm_region, vm_rb);
471
472                 BUG_ON(unlikely(region->vm_end <= region->vm_start));
473                 BUG_ON(unlikely(region->vm_top < region->vm_end));
474                 BUG_ON(unlikely(region->vm_start < last->vm_top));
475
476                 lastp = p;
477         }
478 }
479 #else
480 static void validate_nommu_regions(void)
481 {
482 }
483 #endif
484
485 /*
486  * add a region into the global tree
487  */
488 static void add_nommu_region(struct vm_region *region)
489 {
490         struct vm_region *pregion;
491         struct rb_node **p, *parent;
492
493         validate_nommu_regions();
494
495         parent = NULL;
496         p = &nommu_region_tree.rb_node;
497         while (*p) {
498                 parent = *p;
499                 pregion = rb_entry(parent, struct vm_region, vm_rb);
500                 if (region->vm_start < pregion->vm_start)
501                         p = &(*p)->rb_left;
502                 else if (region->vm_start > pregion->vm_start)
503                         p = &(*p)->rb_right;
504                 else if (pregion == region)
505                         return;
506                 else
507                         BUG();
508         }
509
510         rb_link_node(&region->vm_rb, parent, p);
511         rb_insert_color(&region->vm_rb, &nommu_region_tree);
512
513         validate_nommu_regions();
514 }
515
516 /*
517  * delete a region from the global tree
518  */
519 static void delete_nommu_region(struct vm_region *region)
520 {
521         BUG_ON(!nommu_region_tree.rb_node);
522
523         validate_nommu_regions();
524         rb_erase(&region->vm_rb, &nommu_region_tree);
525         validate_nommu_regions();
526 }
527
528 /*
529  * free a contiguous series of pages
530  */
531 static void free_page_series(unsigned long from, unsigned long to)
532 {
533         for (; from < to; from += PAGE_SIZE) {
534                 struct page *page = virt_to_page(from);
535
536                 kdebug("- free %lx", from);
537                 atomic_long_dec(&mmap_pages_allocated);
538                 if (page_count(page) != 1)
539                         kdebug("free page %p: refcount not one: %d",
540                                page, page_count(page));
541                 put_page(page);
542         }
543 }
544
545 /*
546  * release a reference to a region
547  * - the caller must hold the region semaphore for writing, which this releases
548  * - the region may not have been added to the tree yet, in which case vm_top
549  *   will equal vm_start
550  */
551 static void __put_nommu_region(struct vm_region *region)
552         __releases(nommu_region_sem)
553 {
554         kenter("%p{%d}", region, atomic_read(&region->vm_usage));
555
556         BUG_ON(!nommu_region_tree.rb_node);
557
558         if (atomic_dec_and_test(&region->vm_usage)) {
559                 if (region->vm_top > region->vm_start)
560                         delete_nommu_region(region);
561                 up_write(&nommu_region_sem);
562
563                 if (region->vm_file)
564                         fput(region->vm_file);
565
566                 /* IO memory and memory shared directly out of the pagecache
567                  * from ramfs/tmpfs mustn't be released here */
568                 if (region->vm_flags & VM_MAPPED_COPY) {
569                         kdebug("free series");
570                         free_page_series(region->vm_start, region->vm_top);
571                 }
572                 kmem_cache_free(vm_region_jar, region);
573         } else {
574                 up_write(&nommu_region_sem);
575         }
576 }
577
578 /*
579  * release a reference to a region
580  */
581 static void put_nommu_region(struct vm_region *region)
582 {
583         down_write(&nommu_region_sem);
584         __put_nommu_region(region);
585 }
586
587 /*
588  * update protection on a vma
589  */
590 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
591 {
592 #ifdef CONFIG_MPU
593         struct mm_struct *mm = vma->vm_mm;
594         long start = vma->vm_start & PAGE_MASK;
595         while (start < vma->vm_end) {
596                 protect_page(mm, start, flags);
597                 start += PAGE_SIZE;
598         }
599         update_protections(mm);
600 #endif
601 }
602
603 /*
604  * add a VMA into a process's mm_struct in the appropriate place in the list
605  * and tree and add to the address space's page tree also if not an anonymous
606  * page
607  * - should be called with mm->mmap_sem held writelocked
608  */
609 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
610 {
611         struct vm_area_struct *pvma, **pp;
612         struct address_space *mapping;
613         struct rb_node **p, *parent;
614
615         kenter(",%p", vma);
616
617         BUG_ON(!vma->vm_region);
618
619         mm->map_count++;
620         vma->vm_mm = mm;
621
622         protect_vma(vma, vma->vm_flags);
623
624         /* add the VMA to the mapping */
625         if (vma->vm_file) {
626                 mapping = vma->vm_file->f_mapping;
627
628                 flush_dcache_mmap_lock(mapping);
629                 vma_prio_tree_insert(vma, &mapping->i_mmap);
630                 flush_dcache_mmap_unlock(mapping);
631         }
632
633         /* add the VMA to the tree */
634         parent = NULL;
635         p = &mm->mm_rb.rb_node;
636         while (*p) {
637                 parent = *p;
638                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
639
640                 /* sort by: start addr, end addr, VMA struct addr in that order
641                  * (the latter is necessary as we may get identical VMAs) */
642                 if (vma->vm_start < pvma->vm_start)
643                         p = &(*p)->rb_left;
644                 else if (vma->vm_start > pvma->vm_start)
645                         p = &(*p)->rb_right;
646                 else if (vma->vm_end < pvma->vm_end)
647                         p = &(*p)->rb_left;
648                 else if (vma->vm_end > pvma->vm_end)
649                         p = &(*p)->rb_right;
650                 else if (vma < pvma)
651                         p = &(*p)->rb_left;
652                 else if (vma > pvma)
653                         p = &(*p)->rb_right;
654                 else
655                         BUG();
656         }
657
658         rb_link_node(&vma->vm_rb, parent, p);
659         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
660
661         /* add VMA to the VMA list also */
662         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
663                 if (pvma->vm_start > vma->vm_start)
664                         break;
665                 if (pvma->vm_start < vma->vm_start)
666                         continue;
667                 if (pvma->vm_end < vma->vm_end)
668                         break;
669         }
670
671         vma->vm_next = *pp;
672         *pp = vma;
673 }
674
675 /*
676  * delete a VMA from its owning mm_struct and address space
677  */
678 static void delete_vma_from_mm(struct vm_area_struct *vma)
679 {
680         struct vm_area_struct **pp;
681         struct address_space *mapping;
682         struct mm_struct *mm = vma->vm_mm;
683
684         kenter("%p", vma);
685
686         protect_vma(vma, 0);
687
688         mm->map_count--;
689         if (mm->mmap_cache == vma)
690                 mm->mmap_cache = NULL;
691
692         /* remove the VMA from the mapping */
693         if (vma->vm_file) {
694                 mapping = vma->vm_file->f_mapping;
695
696                 flush_dcache_mmap_lock(mapping);
697                 vma_prio_tree_remove(vma, &mapping->i_mmap);
698                 flush_dcache_mmap_unlock(mapping);
699         }
700
701         /* remove from the MM's tree and list */
702         rb_erase(&vma->vm_rb, &mm->mm_rb);
703         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
704                 if (*pp == vma) {
705                         *pp = vma->vm_next;
706                         break;
707                 }
708         }
709
710         vma->vm_mm = NULL;
711 }
712
713 /*
714  * destroy a VMA record
715  */
716 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
717 {
718         kenter("%p", vma);
719         if (vma->vm_ops && vma->vm_ops->close)
720                 vma->vm_ops->close(vma);
721         if (vma->vm_file) {
722                 fput(vma->vm_file);
723                 if (vma->vm_flags & VM_EXECUTABLE)
724                         removed_exe_file_vma(mm);
725         }
726         put_nommu_region(vma->vm_region);
727         kmem_cache_free(vm_area_cachep, vma);
728 }
729
730 /*
731  * look up the first VMA in which addr resides, NULL if none
732  * - should be called with mm->mmap_sem at least held readlocked
733  */
734 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
735 {
736         struct vm_area_struct *vma;
737         struct rb_node *n = mm->mm_rb.rb_node;
738
739         /* check the cache first */
740         vma = mm->mmap_cache;
741         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
742                 return vma;
743
744         /* trawl the tree (there may be multiple mappings in which addr
745          * resides) */
746         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
747                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
748                 if (vma->vm_start > addr)
749                         return NULL;
750                 if (vma->vm_end > addr) {
751                         mm->mmap_cache = vma;
752                         return vma;
753                 }
754         }
755
756         return NULL;
757 }
758 EXPORT_SYMBOL(find_vma);
759
760 /*
761  * find a VMA
762  * - we don't extend stack VMAs under NOMMU conditions
763  */
764 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
765 {
766         return find_vma(mm, addr);
767 }
768
769 /*
770  * expand a stack to a given address
771  * - not supported under NOMMU conditions
772  */
773 int expand_stack(struct vm_area_struct *vma, unsigned long address)
774 {
775         return -ENOMEM;
776 }
777
778 /*
779  * look up the first VMA exactly that exactly matches addr
780  * - should be called with mm->mmap_sem at least held readlocked
781  */
782 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
783                                              unsigned long addr,
784                                              unsigned long len)
785 {
786         struct vm_area_struct *vma;
787         struct rb_node *n = mm->mm_rb.rb_node;
788         unsigned long end = addr + len;
789
790         /* check the cache first */
791         vma = mm->mmap_cache;
792         if (vma && vma->vm_start == addr && vma->vm_end == end)
793                 return vma;
794
795         /* trawl the tree (there may be multiple mappings in which addr
796          * resides) */
797         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
798                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
799                 if (vma->vm_start < addr)
800                         continue;
801                 if (vma->vm_start > addr)
802                         return NULL;
803                 if (vma->vm_end == end) {
804                         mm->mmap_cache = vma;
805                         return vma;
806                 }
807         }
808
809         return NULL;
810 }
811
812 /*
813  * determine whether a mapping should be permitted and, if so, what sort of
814  * mapping we're capable of supporting
815  */
816 static int validate_mmap_request(struct file *file,
817                                  unsigned long addr,
818                                  unsigned long len,
819                                  unsigned long prot,
820                                  unsigned long flags,
821                                  unsigned long pgoff,
822                                  unsigned long *_capabilities)
823 {
824         unsigned long capabilities, rlen;
825         unsigned long reqprot = prot;
826         int ret;
827
828         /* do the simple checks first */
829         if (flags & MAP_FIXED) {
830                 printk(KERN_DEBUG
831                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
832                        current->pid);
833                 return -EINVAL;
834         }
835
836         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
837             (flags & MAP_TYPE) != MAP_SHARED)
838                 return -EINVAL;
839
840         if (!len)
841                 return -EINVAL;
842
843         /* Careful about overflows.. */
844         rlen = PAGE_ALIGN(len);
845         if (!rlen || rlen > TASK_SIZE)
846                 return -ENOMEM;
847
848         /* offset overflow? */
849         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
850                 return -EOVERFLOW;
851
852         if (file) {
853                 /* validate file mapping requests */
854                 struct address_space *mapping;
855
856                 /* files must support mmap */
857                 if (!file->f_op || !file->f_op->mmap)
858                         return -ENODEV;
859
860                 /* work out if what we've got could possibly be shared
861                  * - we support chardevs that provide their own "memory"
862                  * - we support files/blockdevs that are memory backed
863                  */
864                 mapping = file->f_mapping;
865                 if (!mapping)
866                         mapping = file->f_path.dentry->d_inode->i_mapping;
867
868                 capabilities = 0;
869                 if (mapping && mapping->backing_dev_info)
870                         capabilities = mapping->backing_dev_info->capabilities;
871
872                 if (!capabilities) {
873                         /* no explicit capabilities set, so assume some
874                          * defaults */
875                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
876                         case S_IFREG:
877                         case S_IFBLK:
878                                 capabilities = BDI_CAP_MAP_COPY;
879                                 break;
880
881                         case S_IFCHR:
882                                 capabilities =
883                                         BDI_CAP_MAP_DIRECT |
884                                         BDI_CAP_READ_MAP |
885                                         BDI_CAP_WRITE_MAP;
886                                 break;
887
888                         default:
889                                 return -EINVAL;
890                         }
891                 }
892
893                 /* eliminate any capabilities that we can't support on this
894                  * device */
895                 if (!file->f_op->get_unmapped_area)
896                         capabilities &= ~BDI_CAP_MAP_DIRECT;
897                 if (!file->f_op->read)
898                         capabilities &= ~BDI_CAP_MAP_COPY;
899
900                 /* The file shall have been opened with read permission. */
901                 if (!(file->f_mode & FMODE_READ))
902                         return -EACCES;
903
904                 if (flags & MAP_SHARED) {
905                         /* do checks for writing, appending and locking */
906                         if ((prot & PROT_WRITE) &&
907                             !(file->f_mode & FMODE_WRITE))
908                                 return -EACCES;
909
910                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
911                             (file->f_mode & FMODE_WRITE))
912                                 return -EACCES;
913
914                         if (locks_verify_locked(file->f_path.dentry->d_inode))
915                                 return -EAGAIN;
916
917                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
918                                 return -ENODEV;
919
920                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
921                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
922                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
923                             ) {
924                                 printk("MAP_SHARED not completely supported on !MMU\n");
925                                 return -EINVAL;
926                         }
927
928                         /* we mustn't privatise shared mappings */
929                         capabilities &= ~BDI_CAP_MAP_COPY;
930                 }
931                 else {
932                         /* we're going to read the file into private memory we
933                          * allocate */
934                         if (!(capabilities & BDI_CAP_MAP_COPY))
935                                 return -ENODEV;
936
937                         /* we don't permit a private writable mapping to be
938                          * shared with the backing device */
939                         if (prot & PROT_WRITE)
940                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
941                 }
942
943                 /* handle executable mappings and implied executable
944                  * mappings */
945                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
946                         if (prot & PROT_EXEC)
947                                 return -EPERM;
948                 }
949                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
950                         /* handle implication of PROT_EXEC by PROT_READ */
951                         if (current->personality & READ_IMPLIES_EXEC) {
952                                 if (capabilities & BDI_CAP_EXEC_MAP)
953                                         prot |= PROT_EXEC;
954                         }
955                 }
956                 else if ((prot & PROT_READ) &&
957                          (prot & PROT_EXEC) &&
958                          !(capabilities & BDI_CAP_EXEC_MAP)
959                          ) {
960                         /* backing file is not executable, try to copy */
961                         capabilities &= ~BDI_CAP_MAP_DIRECT;
962                 }
963         }
964         else {
965                 /* anonymous mappings are always memory backed and can be
966                  * privately mapped
967                  */
968                 capabilities = BDI_CAP_MAP_COPY;
969
970                 /* handle PROT_EXEC implication by PROT_READ */
971                 if ((prot & PROT_READ) &&
972                     (current->personality & READ_IMPLIES_EXEC))
973                         prot |= PROT_EXEC;
974         }
975
976         /* allow the security API to have its say */
977         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
978         if (ret < 0)
979                 return ret;
980
981         /* looks okay */
982         *_capabilities = capabilities;
983         return 0;
984 }
985
986 /*
987  * we've determined that we can make the mapping, now translate what we
988  * now know into VMA flags
989  */
990 static unsigned long determine_vm_flags(struct file *file,
991                                         unsigned long prot,
992                                         unsigned long flags,
993                                         unsigned long capabilities)
994 {
995         unsigned long vm_flags;
996
997         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
998         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
999         /* vm_flags |= mm->def_flags; */
1000
1001         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1002                 /* attempt to share read-only copies of mapped file chunks */
1003                 if (file && !(prot & PROT_WRITE))
1004                         vm_flags |= VM_MAYSHARE;
1005         }
1006         else {
1007                 /* overlay a shareable mapping on the backing device or inode
1008                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1009                  * romfs/cramfs */
1010                 if (flags & MAP_SHARED)
1011                         vm_flags |= VM_MAYSHARE | VM_SHARED;
1012                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1013                         vm_flags |= VM_MAYSHARE;
1014         }
1015
1016         /* refuse to let anyone share private mappings with this process if
1017          * it's being traced - otherwise breakpoints set in it may interfere
1018          * with another untraced process
1019          */
1020         if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1021                 vm_flags &= ~VM_MAYSHARE;
1022
1023         return vm_flags;
1024 }
1025
1026 /*
1027  * set up a shared mapping on a file (the driver or filesystem provides and
1028  * pins the storage)
1029  */
1030 static int do_mmap_shared_file(struct vm_area_struct *vma)
1031 {
1032         int ret;
1033
1034         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1035         if (ret == 0) {
1036                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1037                 return 0;
1038         }
1039         if (ret != -ENOSYS)
1040                 return ret;
1041
1042         /* getting an ENOSYS error indicates that direct mmap isn't
1043          * possible (as opposed to tried but failed) so we'll fall
1044          * through to making a private copy of the data and mapping
1045          * that if we can */
1046         return -ENODEV;
1047 }
1048
1049 /*
1050  * set up a private mapping or an anonymous shared mapping
1051  */
1052 static int do_mmap_private(struct vm_area_struct *vma,
1053                            struct vm_region *region,
1054                            unsigned long len,
1055                            unsigned long capabilities)
1056 {
1057         struct page *pages;
1058         unsigned long total, point, n, rlen;
1059         void *base;
1060         int ret, order;
1061
1062         /* invoke the file's mapping function so that it can keep track of
1063          * shared mappings on devices or memory
1064          * - VM_MAYSHARE will be set if it may attempt to share
1065          */
1066         if (capabilities & BDI_CAP_MAP_DIRECT) {
1067                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1068                 if (ret == 0) {
1069                         /* shouldn't return success if we're not sharing */
1070                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1071                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1072                         return 0;
1073                 }
1074                 if (ret != -ENOSYS)
1075                         return ret;
1076
1077                 /* getting an ENOSYS error indicates that direct mmap isn't
1078                  * possible (as opposed to tried but failed) so we'll try to
1079                  * make a private copy of the data and map that instead */
1080         }
1081
1082         rlen = PAGE_ALIGN(len);
1083
1084         /* allocate some memory to hold the mapping
1085          * - note that this may not return a page-aligned address if the object
1086          *   we're allocating is smaller than a page
1087          */
1088         order = get_order(rlen);
1089         kdebug("alloc order %d for %lx", order, len);
1090
1091         pages = alloc_pages(GFP_KERNEL, order);
1092         if (!pages)
1093                 goto enomem;
1094
1095         total = 1 << order;
1096         atomic_long_add(total, &mmap_pages_allocated);
1097
1098         point = rlen >> PAGE_SHIFT;
1099
1100         /* we allocated a power-of-2 sized page set, so we may want to trim off
1101          * the excess */
1102         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1103                 while (total > point) {
1104                         order = ilog2(total - point);
1105                         n = 1 << order;
1106                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1107                         atomic_long_sub(n, &mmap_pages_allocated);
1108                         total -= n;
1109                         set_page_refcounted(pages + total);
1110                         __free_pages(pages + total, order);
1111                 }
1112         }
1113
1114         for (point = 1; point < total; point++)
1115                 set_page_refcounted(&pages[point]);
1116
1117         base = page_address(pages);
1118         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1119         region->vm_start = (unsigned long) base;
1120         region->vm_end   = region->vm_start + rlen;
1121         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1122
1123         vma->vm_start = region->vm_start;
1124         vma->vm_end   = region->vm_start + len;
1125
1126         if (vma->vm_file) {
1127                 /* read the contents of a file into the copy */
1128                 mm_segment_t old_fs;
1129                 loff_t fpos;
1130
1131                 fpos = vma->vm_pgoff;
1132                 fpos <<= PAGE_SHIFT;
1133
1134                 old_fs = get_fs();
1135                 set_fs(KERNEL_DS);
1136                 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1137                 set_fs(old_fs);
1138
1139                 if (ret < 0)
1140                         goto error_free;
1141
1142                 /* clear the last little bit */
1143                 if (ret < rlen)
1144                         memset(base + ret, 0, rlen - ret);
1145
1146         } else {
1147                 /* if it's an anonymous mapping, then just clear it */
1148                 memset(base, 0, rlen);
1149         }
1150
1151         return 0;
1152
1153 error_free:
1154         free_page_series(region->vm_start, region->vm_end);
1155         region->vm_start = vma->vm_start = 0;
1156         region->vm_end   = vma->vm_end = 0;
1157         region->vm_top   = 0;
1158         return ret;
1159
1160 enomem:
1161         printk("Allocation of length %lu from process %d (%s) failed\n",
1162                len, current->pid, current->comm);
1163         show_free_areas();
1164         return -ENOMEM;
1165 }
1166
1167 /*
1168  * handle mapping creation for uClinux
1169  */
1170 unsigned long do_mmap_pgoff(struct file *file,
1171                             unsigned long addr,
1172                             unsigned long len,
1173                             unsigned long prot,
1174                             unsigned long flags,
1175                             unsigned long pgoff)
1176 {
1177         struct vm_area_struct *vma;
1178         struct vm_region *region;
1179         struct rb_node *rb;
1180         unsigned long capabilities, vm_flags, result;
1181         int ret;
1182
1183         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1184
1185         /* decide whether we should attempt the mapping, and if so what sort of
1186          * mapping */
1187         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1188                                     &capabilities);
1189         if (ret < 0) {
1190                 kleave(" = %d [val]", ret);
1191                 return ret;
1192         }
1193
1194         /* we ignore the address hint */
1195         addr = 0;
1196
1197         /* we've determined that we can make the mapping, now translate what we
1198          * now know into VMA flags */
1199         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1200
1201         /* we're going to need to record the mapping */
1202         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1203         if (!region)
1204                 goto error_getting_region;
1205
1206         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1207         if (!vma)
1208                 goto error_getting_vma;
1209
1210         atomic_set(&region->vm_usage, 1);
1211         region->vm_flags = vm_flags;
1212         region->vm_pgoff = pgoff;
1213
1214         INIT_LIST_HEAD(&vma->anon_vma_node);
1215         vma->vm_flags = vm_flags;
1216         vma->vm_pgoff = pgoff;
1217
1218         if (file) {
1219                 region->vm_file = file;
1220                 get_file(file);
1221                 vma->vm_file = file;
1222                 get_file(file);
1223                 if (vm_flags & VM_EXECUTABLE) {
1224                         added_exe_file_vma(current->mm);
1225                         vma->vm_mm = current->mm;
1226                 }
1227         }
1228
1229         down_write(&nommu_region_sem);
1230
1231         /* if we want to share, we need to check for regions created by other
1232          * mmap() calls that overlap with our proposed mapping
1233          * - we can only share with a superset match on most regular files
1234          * - shared mappings on character devices and memory backed files are
1235          *   permitted to overlap inexactly as far as we are concerned for in
1236          *   these cases, sharing is handled in the driver or filesystem rather
1237          *   than here
1238          */
1239         if (vm_flags & VM_MAYSHARE) {
1240                 struct vm_region *pregion;
1241                 unsigned long pglen, rpglen, pgend, rpgend, start;
1242
1243                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1244                 pgend = pgoff + pglen;
1245
1246                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1247                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1248
1249                         if (!(pregion->vm_flags & VM_MAYSHARE))
1250                                 continue;
1251
1252                         /* search for overlapping mappings on the same file */
1253                         if (pregion->vm_file->f_path.dentry->d_inode !=
1254                             file->f_path.dentry->d_inode)
1255                                 continue;
1256
1257                         if (pregion->vm_pgoff >= pgend)
1258                                 continue;
1259
1260                         rpglen = pregion->vm_end - pregion->vm_start;
1261                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1262                         rpgend = pregion->vm_pgoff + rpglen;
1263                         if (pgoff >= rpgend)
1264                                 continue;
1265
1266                         /* handle inexactly overlapping matches between
1267                          * mappings */
1268                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1269                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1270                                 /* new mapping is not a subset of the region */
1271                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1272                                         goto sharing_violation;
1273                                 continue;
1274                         }
1275
1276                         /* we've found a region we can share */
1277                         atomic_inc(&pregion->vm_usage);
1278                         vma->vm_region = pregion;
1279                         start = pregion->vm_start;
1280                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1281                         vma->vm_start = start;
1282                         vma->vm_end = start + len;
1283
1284                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1285                                 kdebug("share copy");
1286                                 vma->vm_flags |= VM_MAPPED_COPY;
1287                         } else {
1288                                 kdebug("share mmap");
1289                                 ret = do_mmap_shared_file(vma);
1290                                 if (ret < 0) {
1291                                         vma->vm_region = NULL;
1292                                         vma->vm_start = 0;
1293                                         vma->vm_end = 0;
1294                                         atomic_dec(&pregion->vm_usage);
1295                                         pregion = NULL;
1296                                         goto error_just_free;
1297                                 }
1298                         }
1299                         fput(region->vm_file);
1300                         kmem_cache_free(vm_region_jar, region);
1301                         region = pregion;
1302                         result = start;
1303                         goto share;
1304                 }
1305
1306                 /* obtain the address at which to make a shared mapping
1307                  * - this is the hook for quasi-memory character devices to
1308                  *   tell us the location of a shared mapping
1309                  */
1310                 if (capabilities & BDI_CAP_MAP_DIRECT) {
1311                         addr = file->f_op->get_unmapped_area(file, addr, len,
1312                                                              pgoff, flags);
1313                         if (IS_ERR((void *) addr)) {
1314                                 ret = addr;
1315                                 if (ret != (unsigned long) -ENOSYS)
1316                                         goto error_just_free;
1317
1318                                 /* the driver refused to tell us where to site
1319                                  * the mapping so we'll have to attempt to copy
1320                                  * it */
1321                                 ret = (unsigned long) -ENODEV;
1322                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1323                                         goto error_just_free;
1324
1325                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1326                         } else {
1327                                 vma->vm_start = region->vm_start = addr;
1328                                 vma->vm_end = region->vm_end = addr + len;
1329                         }
1330                 }
1331         }
1332
1333         vma->vm_region = region;
1334
1335         /* set up the mapping
1336          * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1337          */
1338         if (file && vma->vm_flags & VM_SHARED)
1339                 ret = do_mmap_shared_file(vma);
1340         else
1341                 ret = do_mmap_private(vma, region, len, capabilities);
1342         if (ret < 0)
1343                 goto error_just_free;
1344         add_nommu_region(region);
1345
1346         /* okay... we have a mapping; now we have to register it */
1347         result = vma->vm_start;
1348
1349         current->mm->total_vm += len >> PAGE_SHIFT;
1350
1351 share:
1352         add_vma_to_mm(current->mm, vma);
1353
1354         up_write(&nommu_region_sem);
1355
1356         if (prot & PROT_EXEC)
1357                 flush_icache_range(result, result + len);
1358
1359         kleave(" = %lx", result);
1360         return result;
1361
1362 error_just_free:
1363         up_write(&nommu_region_sem);
1364 error:
1365         if (region->vm_file)
1366                 fput(region->vm_file);
1367         kmem_cache_free(vm_region_jar, region);
1368         if (vma->vm_file)
1369                 fput(vma->vm_file);
1370         if (vma->vm_flags & VM_EXECUTABLE)
1371                 removed_exe_file_vma(vma->vm_mm);
1372         kmem_cache_free(vm_area_cachep, vma);
1373         kleave(" = %d", ret);
1374         return ret;
1375
1376 sharing_violation:
1377         up_write(&nommu_region_sem);
1378         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1379         ret = -EINVAL;
1380         goto error;
1381
1382 error_getting_vma:
1383         kmem_cache_free(vm_region_jar, region);
1384         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1385                " from process %d failed\n",
1386                len, current->pid);
1387         show_free_areas();
1388         return -ENOMEM;
1389
1390 error_getting_region:
1391         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1392                " from process %d failed\n",
1393                len, current->pid);
1394         show_free_areas();
1395         return -ENOMEM;
1396 }
1397 EXPORT_SYMBOL(do_mmap_pgoff);
1398
1399 /*
1400  * split a vma into two pieces at address 'addr', a new vma is allocated either
1401  * for the first part or the tail.
1402  */
1403 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1404               unsigned long addr, int new_below)
1405 {
1406         struct vm_area_struct *new;
1407         struct vm_region *region;
1408         unsigned long npages;
1409
1410         kenter("");
1411
1412         /* we're only permitted to split anonymous regions that have a single
1413          * owner */
1414         if (vma->vm_file ||
1415             atomic_read(&vma->vm_region->vm_usage) != 1)
1416                 return -ENOMEM;
1417
1418         if (mm->map_count >= sysctl_max_map_count)
1419                 return -ENOMEM;
1420
1421         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1422         if (!region)
1423                 return -ENOMEM;
1424
1425         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1426         if (!new) {
1427                 kmem_cache_free(vm_region_jar, region);
1428                 return -ENOMEM;
1429         }
1430
1431         /* most fields are the same, copy all, and then fixup */
1432         *new = *vma;
1433         *region = *vma->vm_region;
1434         new->vm_region = region;
1435
1436         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1437
1438         if (new_below) {
1439                 region->vm_top = region->vm_end = new->vm_end = addr;
1440         } else {
1441                 region->vm_start = new->vm_start = addr;
1442                 region->vm_pgoff = new->vm_pgoff += npages;
1443         }
1444
1445         if (new->vm_ops && new->vm_ops->open)
1446                 new->vm_ops->open(new);
1447
1448         delete_vma_from_mm(vma);
1449         down_write(&nommu_region_sem);
1450         delete_nommu_region(vma->vm_region);
1451         if (new_below) {
1452                 vma->vm_region->vm_start = vma->vm_start = addr;
1453                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1454         } else {
1455                 vma->vm_region->vm_end = vma->vm_end = addr;
1456                 vma->vm_region->vm_top = addr;
1457         }
1458         add_nommu_region(vma->vm_region);
1459         add_nommu_region(new->vm_region);
1460         up_write(&nommu_region_sem);
1461         add_vma_to_mm(mm, vma);
1462         add_vma_to_mm(mm, new);
1463         return 0;
1464 }
1465
1466 /*
1467  * shrink a VMA by removing the specified chunk from either the beginning or
1468  * the end
1469  */
1470 static int shrink_vma(struct mm_struct *mm,
1471                       struct vm_area_struct *vma,
1472                       unsigned long from, unsigned long to)
1473 {
1474         struct vm_region *region;
1475
1476         kenter("");
1477
1478         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1479          * and list */
1480         delete_vma_from_mm(vma);
1481         if (from > vma->vm_start)
1482                 vma->vm_end = from;
1483         else
1484                 vma->vm_start = to;
1485         add_vma_to_mm(mm, vma);
1486
1487         /* cut the backing region down to size */
1488         region = vma->vm_region;
1489         BUG_ON(atomic_read(&region->vm_usage) != 1);
1490
1491         down_write(&nommu_region_sem);
1492         delete_nommu_region(region);
1493         if (from > region->vm_start) {
1494                 to = region->vm_top;
1495                 region->vm_top = region->vm_end = from;
1496         } else {
1497                 region->vm_start = to;
1498         }
1499         add_nommu_region(region);
1500         up_write(&nommu_region_sem);
1501
1502         free_page_series(from, to);
1503         return 0;
1504 }
1505
1506 /*
1507  * release a mapping
1508  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1509  *   VMA, though it need not cover the whole VMA
1510  */
1511 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1512 {
1513         struct vm_area_struct *vma;
1514         struct rb_node *rb;
1515         unsigned long end = start + len;
1516         int ret;
1517
1518         kenter(",%lx,%zx", start, len);
1519
1520         if (len == 0)
1521                 return -EINVAL;
1522
1523         /* find the first potentially overlapping VMA */
1524         vma = find_vma(mm, start);
1525         if (!vma) {
1526                 static int limit = 0;
1527                 if (limit < 5) {
1528                         printk(KERN_WARNING
1529                                "munmap of memory not mmapped by process %d"
1530                                " (%s): 0x%lx-0x%lx\n",
1531                                current->pid, current->comm,
1532                                start, start + len - 1);
1533                         limit++;
1534                 }
1535                 return -EINVAL;
1536         }
1537
1538         /* we're allowed to split an anonymous VMA but not a file-backed one */
1539         if (vma->vm_file) {
1540                 do {
1541                         if (start > vma->vm_start) {
1542                                 kleave(" = -EINVAL [miss]");
1543                                 return -EINVAL;
1544                         }
1545                         if (end == vma->vm_end)
1546                                 goto erase_whole_vma;
1547                         rb = rb_next(&vma->vm_rb);
1548                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1549                 } while (rb);
1550                 kleave(" = -EINVAL [split file]");
1551                 return -EINVAL;
1552         } else {
1553                 /* the chunk must be a subset of the VMA found */
1554                 if (start == vma->vm_start && end == vma->vm_end)
1555                         goto erase_whole_vma;
1556                 if (start < vma->vm_start || end > vma->vm_end) {
1557                         kleave(" = -EINVAL [superset]");
1558                         return -EINVAL;
1559                 }
1560                 if (start & ~PAGE_MASK) {
1561                         kleave(" = -EINVAL [unaligned start]");
1562                         return -EINVAL;
1563                 }
1564                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1565                         kleave(" = -EINVAL [unaligned split]");
1566                         return -EINVAL;
1567                 }
1568                 if (start != vma->vm_start && end != vma->vm_end) {
1569                         ret = split_vma(mm, vma, start, 1);
1570                         if (ret < 0) {
1571                                 kleave(" = %d [split]", ret);
1572                                 return ret;
1573                         }
1574                 }
1575                 return shrink_vma(mm, vma, start, end);
1576         }
1577
1578 erase_whole_vma:
1579         delete_vma_from_mm(vma);
1580         delete_vma(mm, vma);
1581         kleave(" = 0");
1582         return 0;
1583 }
1584 EXPORT_SYMBOL(do_munmap);
1585
1586 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1587 {
1588         int ret;
1589         struct mm_struct *mm = current->mm;
1590
1591         down_write(&mm->mmap_sem);
1592         ret = do_munmap(mm, addr, len);
1593         up_write(&mm->mmap_sem);
1594         return ret;
1595 }
1596
1597 /*
1598  * release all the mappings made in a process's VM space
1599  */
1600 void exit_mmap(struct mm_struct *mm)
1601 {
1602         struct vm_area_struct *vma;
1603
1604         if (!mm)
1605                 return;
1606
1607         kenter("");
1608
1609         mm->total_vm = 0;
1610
1611         while ((vma = mm->mmap)) {
1612                 mm->mmap = vma->vm_next;
1613                 delete_vma_from_mm(vma);
1614                 delete_vma(mm, vma);
1615         }
1616
1617         kleave("");
1618 }
1619
1620 unsigned long do_brk(unsigned long addr, unsigned long len)
1621 {
1622         return -ENOMEM;
1623 }
1624
1625 /*
1626  * expand (or shrink) an existing mapping, potentially moving it at the same
1627  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1628  *
1629  * under NOMMU conditions, we only permit changing a mapping's size, and only
1630  * as long as it stays within the region allocated by do_mmap_private() and the
1631  * block is not shareable
1632  *
1633  * MREMAP_FIXED is not supported under NOMMU conditions
1634  */
1635 unsigned long do_mremap(unsigned long addr,
1636                         unsigned long old_len, unsigned long new_len,
1637                         unsigned long flags, unsigned long new_addr)
1638 {
1639         struct vm_area_struct *vma;
1640
1641         /* insanity checks first */
1642         if (old_len == 0 || new_len == 0)
1643                 return (unsigned long) -EINVAL;
1644
1645         if (addr & ~PAGE_MASK)
1646                 return -EINVAL;
1647
1648         if (flags & MREMAP_FIXED && new_addr != addr)
1649                 return (unsigned long) -EINVAL;
1650
1651         vma = find_vma_exact(current->mm, addr, old_len);
1652         if (!vma)
1653                 return (unsigned long) -EINVAL;
1654
1655         if (vma->vm_end != vma->vm_start + old_len)
1656                 return (unsigned long) -EFAULT;
1657
1658         if (vma->vm_flags & VM_MAYSHARE)
1659                 return (unsigned long) -EPERM;
1660
1661         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1662                 return (unsigned long) -ENOMEM;
1663
1664         /* all checks complete - do it */
1665         vma->vm_end = vma->vm_start + new_len;
1666         return vma->vm_start;
1667 }
1668 EXPORT_SYMBOL(do_mremap);
1669
1670 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1671                 unsigned long, new_len, unsigned long, flags,
1672                 unsigned long, new_addr)
1673 {
1674         unsigned long ret;
1675
1676         down_write(&current->mm->mmap_sem);
1677         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1678         up_write(&current->mm->mmap_sem);
1679         return ret;
1680 }
1681
1682 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1683                         unsigned int foll_flags)
1684 {
1685         return NULL;
1686 }
1687
1688 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1689                 unsigned long to, unsigned long size, pgprot_t prot)
1690 {
1691         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1692         return 0;
1693 }
1694 EXPORT_SYMBOL(remap_pfn_range);
1695
1696 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1697                         unsigned long pgoff)
1698 {
1699         unsigned int size = vma->vm_end - vma->vm_start;
1700
1701         if (!(vma->vm_flags & VM_USERMAP))
1702                 return -EINVAL;
1703
1704         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1705         vma->vm_end = vma->vm_start + size;
1706
1707         return 0;
1708 }
1709 EXPORT_SYMBOL(remap_vmalloc_range);
1710
1711 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1712 {
1713 }
1714
1715 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1716         unsigned long len, unsigned long pgoff, unsigned long flags)
1717 {
1718         return -ENOMEM;
1719 }
1720
1721 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1722 {
1723 }
1724
1725 void unmap_mapping_range(struct address_space *mapping,
1726                          loff_t const holebegin, loff_t const holelen,
1727                          int even_cows)
1728 {
1729 }
1730 EXPORT_SYMBOL(unmap_mapping_range);
1731
1732 /*
1733  * ask for an unmapped area at which to create a mapping on a file
1734  */
1735 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1736                                 unsigned long len, unsigned long pgoff,
1737                                 unsigned long flags)
1738 {
1739         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1740                                   unsigned long, unsigned long);
1741
1742         get_area = current->mm->get_unmapped_area;
1743         if (file && file->f_op && file->f_op->get_unmapped_area)
1744                 get_area = file->f_op->get_unmapped_area;
1745
1746         if (!get_area)
1747                 return -ENOSYS;
1748
1749         return get_area(file, addr, len, pgoff, flags);
1750 }
1751 EXPORT_SYMBOL(get_unmapped_area);
1752
1753 /*
1754  * Check that a process has enough memory to allocate a new virtual
1755  * mapping. 0 means there is enough memory for the allocation to
1756  * succeed and -ENOMEM implies there is not.
1757  *
1758  * We currently support three overcommit policies, which are set via the
1759  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1760  *
1761  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1762  * Additional code 2002 Jul 20 by Robert Love.
1763  *
1764  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1765  *
1766  * Note this is a helper function intended to be used by LSMs which
1767  * wish to use this logic.
1768  */
1769 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1770 {
1771         unsigned long free, allowed;
1772
1773         vm_acct_memory(pages);
1774
1775         /*
1776          * Sometimes we want to use more memory than we have
1777          */
1778         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1779                 return 0;
1780
1781         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1782                 unsigned long n;
1783
1784                 free = global_page_state(NR_FILE_PAGES);
1785                 free += nr_swap_pages;
1786
1787                 /*
1788                  * Any slabs which are created with the
1789                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1790                  * which are reclaimable, under pressure.  The dentry
1791                  * cache and most inode caches should fall into this
1792                  */
1793                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1794
1795                 /*
1796                  * Leave the last 3% for root
1797                  */
1798                 if (!cap_sys_admin)
1799                         free -= free / 32;
1800
1801                 if (free > pages)
1802                         return 0;
1803
1804                 /*
1805                  * nr_free_pages() is very expensive on large systems,
1806                  * only call if we're about to fail.
1807                  */
1808                 n = nr_free_pages();
1809
1810                 /*
1811                  * Leave reserved pages. The pages are not for anonymous pages.
1812                  */
1813                 if (n <= totalreserve_pages)
1814                         goto error;
1815                 else
1816                         n -= totalreserve_pages;
1817
1818                 /*
1819                  * Leave the last 3% for root
1820                  */
1821                 if (!cap_sys_admin)
1822                         n -= n / 32;
1823                 free += n;
1824
1825                 if (free > pages)
1826                         return 0;
1827
1828                 goto error;
1829         }
1830
1831         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1832         /*
1833          * Leave the last 3% for root
1834          */
1835         if (!cap_sys_admin)
1836                 allowed -= allowed / 32;
1837         allowed += total_swap_pages;
1838
1839         /* Don't let a single process grow too big:
1840            leave 3% of the size of this process for other processes */
1841         if (mm)
1842                 allowed -= mm->total_vm / 32;
1843
1844         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1845                 return 0;
1846
1847 error:
1848         vm_unacct_memory(pages);
1849
1850         return -ENOMEM;
1851 }
1852
1853 int in_gate_area_no_task(unsigned long addr)
1854 {
1855         return 0;
1856 }
1857
1858 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1859 {
1860         BUG();
1861         return 0;
1862 }
1863 EXPORT_SYMBOL(filemap_fault);
1864
1865 /*
1866  * Access another process' address space.
1867  * - source/target buffer must be kernel space
1868  */
1869 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1870 {
1871         struct vm_area_struct *vma;
1872         struct mm_struct *mm;
1873
1874         if (addr + len < addr)
1875                 return 0;
1876
1877         mm = get_task_mm(tsk);
1878         if (!mm)
1879                 return 0;
1880
1881         down_read(&mm->mmap_sem);
1882
1883         /* the access must start within one of the target process's mappings */
1884         vma = find_vma(mm, addr);
1885         if (vma) {
1886                 /* don't overrun this mapping */
1887                 if (addr + len >= vma->vm_end)
1888                         len = vma->vm_end - addr;
1889
1890                 /* only read or write mappings where it is permitted */
1891                 if (write && vma->vm_flags & VM_MAYWRITE)
1892                         len -= copy_to_user((void *) addr, buf, len);
1893                 else if (!write && vma->vm_flags & VM_MAYREAD)
1894                         len -= copy_from_user(buf, (void *) addr, len);
1895                 else
1896                         len = 0;
1897         } else {
1898                 len = 0;
1899         }
1900
1901         up_read(&mm->mmap_sem);
1902         mmput(mm);
1903         return len;
1904 }