Merge git://git.kernel.org/pub/scm/linux/kernel/git/lethal/sh-2.6
[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 "internal.h"
37
38 static inline __attribute__((format(printf, 1, 2)))
39 void no_printk(const char *fmt, ...)
40 {
41 }
42
43 #if 0
44 #define kenter(FMT, ...) \
45         printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
46 #define kleave(FMT, ...) \
47         printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
48 #define kdebug(FMT, ...) \
49         printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
50 #else
51 #define kenter(FMT, ...) \
52         no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
53 #define kleave(FMT, ...) \
54         no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
55 #define kdebug(FMT, ...) \
56         no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
57 #endif
58
59 #include "internal.h"
60
61 void *high_memory;
62 struct page *mem_map;
63 unsigned long max_mapnr;
64 unsigned long num_physpages;
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 /* amount of vm to protect from userspace access */
73 unsigned long mmap_min_addr = CONFIG_DEFAULT_MMAP_MIN_ADDR;
74
75 atomic_long_t mmap_pages_allocated;
76
77 EXPORT_SYMBOL(mem_map);
78 EXPORT_SYMBOL(num_physpages);
79
80 /* list of mapped, potentially shareable regions */
81 static struct kmem_cache *vm_region_jar;
82 struct rb_root nommu_region_tree = RB_ROOT;
83 DECLARE_RWSEM(nommu_region_sem);
84
85 struct vm_operations_struct generic_file_vm_ops = {
86 };
87
88 /*
89  * Handle all mappings that got truncated by a "truncate()"
90  * system call.
91  *
92  * NOTE! We have to be ready to update the memory sharing
93  * between the file and the memory map for a potential last
94  * incomplete page.  Ugly, but necessary.
95  */
96 int vmtruncate(struct inode *inode, loff_t offset)
97 {
98         struct address_space *mapping = inode->i_mapping;
99         unsigned long limit;
100
101         if (inode->i_size < offset)
102                 goto do_expand;
103         i_size_write(inode, offset);
104
105         truncate_inode_pages(mapping, offset);
106         goto out_truncate;
107
108 do_expand:
109         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
110         if (limit != RLIM_INFINITY && offset > limit)
111                 goto out_sig;
112         if (offset > inode->i_sb->s_maxbytes)
113                 goto out;
114         i_size_write(inode, offset);
115
116 out_truncate:
117         if (inode->i_op->truncate)
118                 inode->i_op->truncate(inode);
119         return 0;
120 out_sig:
121         send_sig(SIGXFSZ, current, 0);
122 out:
123         return -EFBIG;
124 }
125
126 EXPORT_SYMBOL(vmtruncate);
127
128 /*
129  * Return the total memory allocated for this pointer, not
130  * just what the caller asked for.
131  *
132  * Doesn't have to be accurate, i.e. may have races.
133  */
134 unsigned int kobjsize(const void *objp)
135 {
136         struct page *page;
137
138         /*
139          * If the object we have should not have ksize performed on it,
140          * return size of 0
141          */
142         if (!objp || !virt_addr_valid(objp))
143                 return 0;
144
145         page = virt_to_head_page(objp);
146
147         /*
148          * If the allocator sets PageSlab, we know the pointer came from
149          * kmalloc().
150          */
151         if (PageSlab(page))
152                 return ksize(objp);
153
154         /*
155          * If it's not a compound page, see if we have a matching VMA
156          * region. This test is intentionally done in reverse order,
157          * so if there's no VMA, we still fall through and hand back
158          * PAGE_SIZE for 0-order pages.
159          */
160         if (!PageCompound(page)) {
161                 struct vm_area_struct *vma;
162
163                 vma = find_vma(current->mm, (unsigned long)objp);
164                 if (vma)
165                         return vma->vm_end - vma->vm_start;
166         }
167
168         /*
169          * The ksize() function is only guaranteed to work for pointers
170          * returned by kmalloc(). So handle arbitrary pointers here.
171          */
172         return PAGE_SIZE << compound_order(page);
173 }
174
175 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
176                      unsigned long start, int nr_pages, int flags,
177                      struct page **pages, struct vm_area_struct **vmas)
178 {
179         struct vm_area_struct *vma;
180         unsigned long vm_flags;
181         int i;
182         int write = !!(flags & GUP_FLAGS_WRITE);
183         int force = !!(flags & GUP_FLAGS_FORCE);
184         int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
185
186         /* calculate required read or write permissions.
187          * - if 'force' is set, we only require the "MAY" flags.
188          */
189         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
190         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
191
192         for (i = 0; i < nr_pages; i++) {
193                 vma = find_vma(mm, start);
194                 if (!vma)
195                         goto finish_or_fault;
196
197                 /* protect what we can, including chardevs */
198                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
199                     (!ignore && !(vm_flags & vma->vm_flags)))
200                         goto finish_or_fault;
201
202                 if (pages) {
203                         pages[i] = virt_to_page(start);
204                         if (pages[i])
205                                 page_cache_get(pages[i]);
206                 }
207                 if (vmas)
208                         vmas[i] = vma;
209                 start += PAGE_SIZE;
210         }
211
212         return i;
213
214 finish_or_fault:
215         return i ? : -EFAULT;
216 }
217
218
219 /*
220  * get a list of pages in an address range belonging to the specified process
221  * and indicate the VMA that covers each page
222  * - this is potentially dodgy as we may end incrementing the page count of a
223  *   slab page or a secondary page from a compound page
224  * - don't permit access to VMAs that don't support it, such as I/O mappings
225  */
226 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
227         unsigned long start, int nr_pages, int write, int force,
228         struct page **pages, struct vm_area_struct **vmas)
229 {
230         int flags = 0;
231
232         if (write)
233                 flags |= GUP_FLAGS_WRITE;
234         if (force)
235                 flags |= GUP_FLAGS_FORCE;
236
237         return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
238 }
239 EXPORT_SYMBOL(get_user_pages);
240
241 /**
242  * follow_pfn - look up PFN at a user virtual address
243  * @vma: memory mapping
244  * @address: user virtual address
245  * @pfn: location to store found PFN
246  *
247  * Only IO mappings and raw PFN mappings are allowed.
248  *
249  * Returns zero and the pfn at @pfn on success, -ve otherwise.
250  */
251 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
252         unsigned long *pfn)
253 {
254         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
255                 return -EINVAL;
256
257         *pfn = address >> PAGE_SHIFT;
258         return 0;
259 }
260 EXPORT_SYMBOL(follow_pfn);
261
262 DEFINE_RWLOCK(vmlist_lock);
263 struct vm_struct *vmlist;
264
265 void vfree(const void *addr)
266 {
267         kfree(addr);
268 }
269 EXPORT_SYMBOL(vfree);
270
271 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
272 {
273         /*
274          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
275          * returns only a logical address.
276          */
277         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
278 }
279 EXPORT_SYMBOL(__vmalloc);
280
281 void *vmalloc_user(unsigned long size)
282 {
283         void *ret;
284
285         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
286                         PAGE_KERNEL);
287         if (ret) {
288                 struct vm_area_struct *vma;
289
290                 down_write(&current->mm->mmap_sem);
291                 vma = find_vma(current->mm, (unsigned long)ret);
292                 if (vma)
293                         vma->vm_flags |= VM_USERMAP;
294                 up_write(&current->mm->mmap_sem);
295         }
296
297         return ret;
298 }
299 EXPORT_SYMBOL(vmalloc_user);
300
301 struct page *vmalloc_to_page(const void *addr)
302 {
303         return virt_to_page(addr);
304 }
305 EXPORT_SYMBOL(vmalloc_to_page);
306
307 unsigned long vmalloc_to_pfn(const void *addr)
308 {
309         return page_to_pfn(virt_to_page(addr));
310 }
311 EXPORT_SYMBOL(vmalloc_to_pfn);
312
313 long vread(char *buf, char *addr, unsigned long count)
314 {
315         memcpy(buf, addr, count);
316         return count;
317 }
318
319 long vwrite(char *buf, char *addr, unsigned long count)
320 {
321         /* Don't allow overflow */
322         if ((unsigned long) addr + count < count)
323                 count = -(unsigned long) addr;
324
325         memcpy(addr, buf, count);
326         return(count);
327 }
328
329 /*
330  *      vmalloc  -  allocate virtually continguos memory
331  *
332  *      @size:          allocation size
333  *
334  *      Allocate enough pages to cover @size from the page level
335  *      allocator and map them into continguos kernel virtual space.
336  *
337  *      For tight control over page level allocator and protection flags
338  *      use __vmalloc() instead.
339  */
340 void *vmalloc(unsigned long size)
341 {
342        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
343 }
344 EXPORT_SYMBOL(vmalloc);
345
346 void *vmalloc_node(unsigned long size, int node)
347 {
348         return vmalloc(size);
349 }
350 EXPORT_SYMBOL(vmalloc_node);
351
352 #ifndef PAGE_KERNEL_EXEC
353 # define PAGE_KERNEL_EXEC PAGE_KERNEL
354 #endif
355
356 /**
357  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
358  *      @size:          allocation size
359  *
360  *      Kernel-internal function to allocate enough pages to cover @size
361  *      the page level allocator and map them into contiguous and
362  *      executable kernel virtual space.
363  *
364  *      For tight control over page level allocator and protection flags
365  *      use __vmalloc() instead.
366  */
367
368 void *vmalloc_exec(unsigned long size)
369 {
370         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
371 }
372
373 /**
374  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
375  *      @size:          allocation size
376  *
377  *      Allocate enough 32bit PA addressable pages to cover @size from the
378  *      page level allocator and map them into continguos kernel virtual space.
379  */
380 void *vmalloc_32(unsigned long size)
381 {
382         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
383 }
384 EXPORT_SYMBOL(vmalloc_32);
385
386 /**
387  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
388  *      @size:          allocation size
389  *
390  * The resulting memory area is 32bit addressable and zeroed so it can be
391  * mapped to userspace without leaking data.
392  *
393  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
394  * remap_vmalloc_range() are permissible.
395  */
396 void *vmalloc_32_user(unsigned long size)
397 {
398         /*
399          * We'll have to sort out the ZONE_DMA bits for 64-bit,
400          * but for now this can simply use vmalloc_user() directly.
401          */
402         return vmalloc_user(size);
403 }
404 EXPORT_SYMBOL(vmalloc_32_user);
405
406 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
407 {
408         BUG();
409         return NULL;
410 }
411 EXPORT_SYMBOL(vmap);
412
413 void vunmap(const void *addr)
414 {
415         BUG();
416 }
417 EXPORT_SYMBOL(vunmap);
418
419 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
420 {
421         BUG();
422         return NULL;
423 }
424 EXPORT_SYMBOL(vm_map_ram);
425
426 void vm_unmap_ram(const void *mem, unsigned int count)
427 {
428         BUG();
429 }
430 EXPORT_SYMBOL(vm_unmap_ram);
431
432 void vm_unmap_aliases(void)
433 {
434 }
435 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
436
437 /*
438  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
439  * have one.
440  */
441 void  __attribute__((weak)) vmalloc_sync_all(void)
442 {
443 }
444
445 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
446                    struct page *page)
447 {
448         return -EINVAL;
449 }
450 EXPORT_SYMBOL(vm_insert_page);
451
452 /*
453  *  sys_brk() for the most part doesn't need the global kernel
454  *  lock, except when an application is doing something nasty
455  *  like trying to un-brk an area that has already been mapped
456  *  to a regular file.  in this case, the unmapping will need
457  *  to invoke file system routines that need the global lock.
458  */
459 SYSCALL_DEFINE1(brk, unsigned long, brk)
460 {
461         struct mm_struct *mm = current->mm;
462
463         if (brk < mm->start_brk || brk > mm->context.end_brk)
464                 return mm->brk;
465
466         if (mm->brk == brk)
467                 return mm->brk;
468
469         /*
470          * Always allow shrinking brk
471          */
472         if (brk <= mm->brk) {
473                 mm->brk = brk;
474                 return brk;
475         }
476
477         /*
478          * Ok, looks good - let it rip.
479          */
480         return mm->brk = brk;
481 }
482
483 /*
484  * initialise the VMA and region record slabs
485  */
486 void __init mmap_init(void)
487 {
488         int ret;
489
490         ret = percpu_counter_init(&vm_committed_as, 0);
491         VM_BUG_ON(ret);
492         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
493 }
494
495 /*
496  * validate the region tree
497  * - the caller must hold the region lock
498  */
499 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
500 static noinline void validate_nommu_regions(void)
501 {
502         struct vm_region *region, *last;
503         struct rb_node *p, *lastp;
504
505         lastp = rb_first(&nommu_region_tree);
506         if (!lastp)
507                 return;
508
509         last = rb_entry(lastp, struct vm_region, vm_rb);
510         BUG_ON(unlikely(last->vm_end <= last->vm_start));
511         BUG_ON(unlikely(last->vm_top < last->vm_end));
512
513         while ((p = rb_next(lastp))) {
514                 region = rb_entry(p, struct vm_region, vm_rb);
515                 last = rb_entry(lastp, struct vm_region, vm_rb);
516
517                 BUG_ON(unlikely(region->vm_end <= region->vm_start));
518                 BUG_ON(unlikely(region->vm_top < region->vm_end));
519                 BUG_ON(unlikely(region->vm_start < last->vm_top));
520
521                 lastp = p;
522         }
523 }
524 #else
525 static void validate_nommu_regions(void)
526 {
527 }
528 #endif
529
530 /*
531  * add a region into the global tree
532  */
533 static void add_nommu_region(struct vm_region *region)
534 {
535         struct vm_region *pregion;
536         struct rb_node **p, *parent;
537
538         validate_nommu_regions();
539
540         parent = NULL;
541         p = &nommu_region_tree.rb_node;
542         while (*p) {
543                 parent = *p;
544                 pregion = rb_entry(parent, struct vm_region, vm_rb);
545                 if (region->vm_start < pregion->vm_start)
546                         p = &(*p)->rb_left;
547                 else if (region->vm_start > pregion->vm_start)
548                         p = &(*p)->rb_right;
549                 else if (pregion == region)
550                         return;
551                 else
552                         BUG();
553         }
554
555         rb_link_node(&region->vm_rb, parent, p);
556         rb_insert_color(&region->vm_rb, &nommu_region_tree);
557
558         validate_nommu_regions();
559 }
560
561 /*
562  * delete a region from the global tree
563  */
564 static void delete_nommu_region(struct vm_region *region)
565 {
566         BUG_ON(!nommu_region_tree.rb_node);
567
568         validate_nommu_regions();
569         rb_erase(&region->vm_rb, &nommu_region_tree);
570         validate_nommu_regions();
571 }
572
573 /*
574  * free a contiguous series of pages
575  */
576 static void free_page_series(unsigned long from, unsigned long to)
577 {
578         for (; from < to; from += PAGE_SIZE) {
579                 struct page *page = virt_to_page(from);
580
581                 kdebug("- free %lx", from);
582                 atomic_long_dec(&mmap_pages_allocated);
583                 if (page_count(page) != 1)
584                         kdebug("free page %p: refcount not one: %d",
585                                page, page_count(page));
586                 put_page(page);
587         }
588 }
589
590 /*
591  * release a reference to a region
592  * - the caller must hold the region semaphore for writing, which this releases
593  * - the region may not have been added to the tree yet, in which case vm_top
594  *   will equal vm_start
595  */
596 static void __put_nommu_region(struct vm_region *region)
597         __releases(nommu_region_sem)
598 {
599         kenter("%p{%d}", region, atomic_read(&region->vm_usage));
600
601         BUG_ON(!nommu_region_tree.rb_node);
602
603         if (atomic_dec_and_test(&region->vm_usage)) {
604                 if (region->vm_top > region->vm_start)
605                         delete_nommu_region(region);
606                 up_write(&nommu_region_sem);
607
608                 if (region->vm_file)
609                         fput(region->vm_file);
610
611                 /* IO memory and memory shared directly out of the pagecache
612                  * from ramfs/tmpfs mustn't be released here */
613                 if (region->vm_flags & VM_MAPPED_COPY) {
614                         kdebug("free series");
615                         free_page_series(region->vm_start, region->vm_top);
616                 }
617                 kmem_cache_free(vm_region_jar, region);
618         } else {
619                 up_write(&nommu_region_sem);
620         }
621 }
622
623 /*
624  * release a reference to a region
625  */
626 static void put_nommu_region(struct vm_region *region)
627 {
628         down_write(&nommu_region_sem);
629         __put_nommu_region(region);
630 }
631
632 /*
633  * add a VMA into a process's mm_struct in the appropriate place in the list
634  * and tree and add to the address space's page tree also if not an anonymous
635  * page
636  * - should be called with mm->mmap_sem held writelocked
637  */
638 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
639 {
640         struct vm_area_struct *pvma, **pp;
641         struct address_space *mapping;
642         struct rb_node **p, *parent;
643
644         kenter(",%p", vma);
645
646         BUG_ON(!vma->vm_region);
647
648         mm->map_count++;
649         vma->vm_mm = mm;
650
651         /* add the VMA to the mapping */
652         if (vma->vm_file) {
653                 mapping = vma->vm_file->f_mapping;
654
655                 flush_dcache_mmap_lock(mapping);
656                 vma_prio_tree_insert(vma, &mapping->i_mmap);
657                 flush_dcache_mmap_unlock(mapping);
658         }
659
660         /* add the VMA to the tree */
661         parent = NULL;
662         p = &mm->mm_rb.rb_node;
663         while (*p) {
664                 parent = *p;
665                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
666
667                 /* sort by: start addr, end addr, VMA struct addr in that order
668                  * (the latter is necessary as we may get identical VMAs) */
669                 if (vma->vm_start < pvma->vm_start)
670                         p = &(*p)->rb_left;
671                 else if (vma->vm_start > pvma->vm_start)
672                         p = &(*p)->rb_right;
673                 else if (vma->vm_end < pvma->vm_end)
674                         p = &(*p)->rb_left;
675                 else if (vma->vm_end > pvma->vm_end)
676                         p = &(*p)->rb_right;
677                 else if (vma < pvma)
678                         p = &(*p)->rb_left;
679                 else if (vma > pvma)
680                         p = &(*p)->rb_right;
681                 else
682                         BUG();
683         }
684
685         rb_link_node(&vma->vm_rb, parent, p);
686         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
687
688         /* add VMA to the VMA list also */
689         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
690                 if (pvma->vm_start > vma->vm_start)
691                         break;
692                 if (pvma->vm_start < vma->vm_start)
693                         continue;
694                 if (pvma->vm_end < vma->vm_end)
695                         break;
696         }
697
698         vma->vm_next = *pp;
699         *pp = vma;
700 }
701
702 /*
703  * delete a VMA from its owning mm_struct and address space
704  */
705 static void delete_vma_from_mm(struct vm_area_struct *vma)
706 {
707         struct vm_area_struct **pp;
708         struct address_space *mapping;
709         struct mm_struct *mm = vma->vm_mm;
710
711         kenter("%p", vma);
712
713         mm->map_count--;
714         if (mm->mmap_cache == vma)
715                 mm->mmap_cache = NULL;
716
717         /* remove the VMA from the mapping */
718         if (vma->vm_file) {
719                 mapping = vma->vm_file->f_mapping;
720
721                 flush_dcache_mmap_lock(mapping);
722                 vma_prio_tree_remove(vma, &mapping->i_mmap);
723                 flush_dcache_mmap_unlock(mapping);
724         }
725
726         /* remove from the MM's tree and list */
727         rb_erase(&vma->vm_rb, &mm->mm_rb);
728         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
729                 if (*pp == vma) {
730                         *pp = vma->vm_next;
731                         break;
732                 }
733         }
734
735         vma->vm_mm = NULL;
736 }
737
738 /*
739  * destroy a VMA record
740  */
741 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
742 {
743         kenter("%p", vma);
744         if (vma->vm_ops && vma->vm_ops->close)
745                 vma->vm_ops->close(vma);
746         if (vma->vm_file) {
747                 fput(vma->vm_file);
748                 if (vma->vm_flags & VM_EXECUTABLE)
749                         removed_exe_file_vma(mm);
750         }
751         put_nommu_region(vma->vm_region);
752         kmem_cache_free(vm_area_cachep, vma);
753 }
754
755 /*
756  * look up the first VMA in which addr resides, NULL if none
757  * - should be called with mm->mmap_sem at least held readlocked
758  */
759 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
760 {
761         struct vm_area_struct *vma;
762         struct rb_node *n = mm->mm_rb.rb_node;
763
764         /* check the cache first */
765         vma = mm->mmap_cache;
766         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
767                 return vma;
768
769         /* trawl the tree (there may be multiple mappings in which addr
770          * resides) */
771         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
772                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
773                 if (vma->vm_start > addr)
774                         return NULL;
775                 if (vma->vm_end > addr) {
776                         mm->mmap_cache = vma;
777                         return vma;
778                 }
779         }
780
781         return NULL;
782 }
783 EXPORT_SYMBOL(find_vma);
784
785 /*
786  * find a VMA
787  * - we don't extend stack VMAs under NOMMU conditions
788  */
789 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
790 {
791         return find_vma(mm, addr);
792 }
793
794 /*
795  * expand a stack to a given address
796  * - not supported under NOMMU conditions
797  */
798 int expand_stack(struct vm_area_struct *vma, unsigned long address)
799 {
800         return -ENOMEM;
801 }
802
803 /*
804  * look up the first VMA exactly that exactly matches addr
805  * - should be called with mm->mmap_sem at least held readlocked
806  */
807 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
808                                              unsigned long addr,
809                                              unsigned long len)
810 {
811         struct vm_area_struct *vma;
812         struct rb_node *n = mm->mm_rb.rb_node;
813         unsigned long end = addr + len;
814
815         /* check the cache first */
816         vma = mm->mmap_cache;
817         if (vma && vma->vm_start == addr && vma->vm_end == end)
818                 return vma;
819
820         /* trawl the tree (there may be multiple mappings in which addr
821          * resides) */
822         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
823                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
824                 if (vma->vm_start < addr)
825                         continue;
826                 if (vma->vm_start > addr)
827                         return NULL;
828                 if (vma->vm_end == end) {
829                         mm->mmap_cache = vma;
830                         return vma;
831                 }
832         }
833
834         return NULL;
835 }
836
837 /*
838  * determine whether a mapping should be permitted and, if so, what sort of
839  * mapping we're capable of supporting
840  */
841 static int validate_mmap_request(struct file *file,
842                                  unsigned long addr,
843                                  unsigned long len,
844                                  unsigned long prot,
845                                  unsigned long flags,
846                                  unsigned long pgoff,
847                                  unsigned long *_capabilities)
848 {
849         unsigned long capabilities, rlen;
850         unsigned long reqprot = prot;
851         int ret;
852
853         /* do the simple checks first */
854         if (flags & MAP_FIXED || addr) {
855                 printk(KERN_DEBUG
856                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
857                        current->pid);
858                 return -EINVAL;
859         }
860
861         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
862             (flags & MAP_TYPE) != MAP_SHARED)
863                 return -EINVAL;
864
865         if (!len)
866                 return -EINVAL;
867
868         /* Careful about overflows.. */
869         rlen = PAGE_ALIGN(len);
870         if (!rlen || rlen > TASK_SIZE)
871                 return -ENOMEM;
872
873         /* offset overflow? */
874         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
875                 return -EOVERFLOW;
876
877         if (file) {
878                 /* validate file mapping requests */
879                 struct address_space *mapping;
880
881                 /* files must support mmap */
882                 if (!file->f_op || !file->f_op->mmap)
883                         return -ENODEV;
884
885                 /* work out if what we've got could possibly be shared
886                  * - we support chardevs that provide their own "memory"
887                  * - we support files/blockdevs that are memory backed
888                  */
889                 mapping = file->f_mapping;
890                 if (!mapping)
891                         mapping = file->f_path.dentry->d_inode->i_mapping;
892
893                 capabilities = 0;
894                 if (mapping && mapping->backing_dev_info)
895                         capabilities = mapping->backing_dev_info->capabilities;
896
897                 if (!capabilities) {
898                         /* no explicit capabilities set, so assume some
899                          * defaults */
900                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
901                         case S_IFREG:
902                         case S_IFBLK:
903                                 capabilities = BDI_CAP_MAP_COPY;
904                                 break;
905
906                         case S_IFCHR:
907                                 capabilities =
908                                         BDI_CAP_MAP_DIRECT |
909                                         BDI_CAP_READ_MAP |
910                                         BDI_CAP_WRITE_MAP;
911                                 break;
912
913                         default:
914                                 return -EINVAL;
915                         }
916                 }
917
918                 /* eliminate any capabilities that we can't support on this
919                  * device */
920                 if (!file->f_op->get_unmapped_area)
921                         capabilities &= ~BDI_CAP_MAP_DIRECT;
922                 if (!file->f_op->read)
923                         capabilities &= ~BDI_CAP_MAP_COPY;
924
925                 if (flags & MAP_SHARED) {
926                         /* do checks for writing, appending and locking */
927                         if ((prot & PROT_WRITE) &&
928                             !(file->f_mode & FMODE_WRITE))
929                                 return -EACCES;
930
931                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
932                             (file->f_mode & FMODE_WRITE))
933                                 return -EACCES;
934
935                         if (locks_verify_locked(file->f_path.dentry->d_inode))
936                                 return -EAGAIN;
937
938                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
939                                 return -ENODEV;
940
941                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
942                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
943                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
944                             ) {
945                                 printk("MAP_SHARED not completely supported on !MMU\n");
946                                 return -EINVAL;
947                         }
948
949                         /* we mustn't privatise shared mappings */
950                         capabilities &= ~BDI_CAP_MAP_COPY;
951                 }
952                 else {
953                         /* we're going to read the file into private memory we
954                          * allocate */
955                         if (!(capabilities & BDI_CAP_MAP_COPY))
956                                 return -ENODEV;
957
958                         /* we don't permit a private writable mapping to be
959                          * shared with the backing device */
960                         if (prot & PROT_WRITE)
961                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
962                 }
963
964                 /* handle executable mappings and implied executable
965                  * mappings */
966                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
967                         if (prot & PROT_EXEC)
968                                 return -EPERM;
969                 }
970                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
971                         /* handle implication of PROT_EXEC by PROT_READ */
972                         if (current->personality & READ_IMPLIES_EXEC) {
973                                 if (capabilities & BDI_CAP_EXEC_MAP)
974                                         prot |= PROT_EXEC;
975                         }
976                 }
977                 else if ((prot & PROT_READ) &&
978                          (prot & PROT_EXEC) &&
979                          !(capabilities & BDI_CAP_EXEC_MAP)
980                          ) {
981                         /* backing file is not executable, try to copy */
982                         capabilities &= ~BDI_CAP_MAP_DIRECT;
983                 }
984         }
985         else {
986                 /* anonymous mappings are always memory backed and can be
987                  * privately mapped
988                  */
989                 capabilities = BDI_CAP_MAP_COPY;
990
991                 /* handle PROT_EXEC implication by PROT_READ */
992                 if ((prot & PROT_READ) &&
993                     (current->personality & READ_IMPLIES_EXEC))
994                         prot |= PROT_EXEC;
995         }
996
997         /* allow the security API to have its say */
998         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
999         if (ret < 0)
1000                 return ret;
1001
1002         /* looks okay */
1003         *_capabilities = capabilities;
1004         return 0;
1005 }
1006
1007 /*
1008  * we've determined that we can make the mapping, now translate what we
1009  * now know into VMA flags
1010  */
1011 static unsigned long determine_vm_flags(struct file *file,
1012                                         unsigned long prot,
1013                                         unsigned long flags,
1014                                         unsigned long capabilities)
1015 {
1016         unsigned long vm_flags;
1017
1018         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1019         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1020         /* vm_flags |= mm->def_flags; */
1021
1022         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1023                 /* attempt to share read-only copies of mapped file chunks */
1024                 if (file && !(prot & PROT_WRITE))
1025                         vm_flags |= VM_MAYSHARE;
1026         }
1027         else {
1028                 /* overlay a shareable mapping on the backing device or inode
1029                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1030                  * romfs/cramfs */
1031                 if (flags & MAP_SHARED)
1032                         vm_flags |= VM_MAYSHARE | VM_SHARED;
1033                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1034                         vm_flags |= VM_MAYSHARE;
1035         }
1036
1037         /* refuse to let anyone share private mappings with this process if
1038          * it's being traced - otherwise breakpoints set in it may interfere
1039          * with another untraced process
1040          */
1041         if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1042                 vm_flags &= ~VM_MAYSHARE;
1043
1044         return vm_flags;
1045 }
1046
1047 /*
1048  * set up a shared mapping on a file (the driver or filesystem provides and
1049  * pins the storage)
1050  */
1051 static int do_mmap_shared_file(struct vm_area_struct *vma)
1052 {
1053         int ret;
1054
1055         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1056         if (ret == 0) {
1057                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1058                 return ret;
1059         }
1060         if (ret != -ENOSYS)
1061                 return ret;
1062
1063         /* getting an ENOSYS error indicates that direct mmap isn't
1064          * possible (as opposed to tried but failed) so we'll fall
1065          * through to making a private copy of the data and mapping
1066          * that if we can */
1067         return -ENODEV;
1068 }
1069
1070 /*
1071  * set up a private mapping or an anonymous shared mapping
1072  */
1073 static int do_mmap_private(struct vm_area_struct *vma,
1074                            struct vm_region *region,
1075                            unsigned long len)
1076 {
1077         struct page *pages;
1078         unsigned long total, point, n, rlen;
1079         void *base;
1080         int ret, order;
1081
1082         /* invoke the file's mapping function so that it can keep track of
1083          * shared mappings on devices or memory
1084          * - VM_MAYSHARE will be set if it may attempt to share
1085          */
1086         if (vma->vm_file) {
1087                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1088                 if (ret == 0) {
1089                         /* shouldn't return success if we're not sharing */
1090                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1091                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1092                         return ret;
1093                 }
1094                 if (ret != -ENOSYS)
1095                         return ret;
1096
1097                 /* getting an ENOSYS error indicates that direct mmap isn't
1098                  * possible (as opposed to tried but failed) so we'll try to
1099                  * make a private copy of the data and map that instead */
1100         }
1101
1102         rlen = PAGE_ALIGN(len);
1103
1104         /* allocate some memory to hold the mapping
1105          * - note that this may not return a page-aligned address if the object
1106          *   we're allocating is smaller than a page
1107          */
1108         order = get_order(rlen);
1109         kdebug("alloc order %d for %lx", order, len);
1110
1111         pages = alloc_pages(GFP_KERNEL, order);
1112         if (!pages)
1113                 goto enomem;
1114
1115         total = 1 << order;
1116         atomic_long_add(total, &mmap_pages_allocated);
1117
1118         point = rlen >> PAGE_SHIFT;
1119
1120         /* we allocated a power-of-2 sized page set, so we may want to trim off
1121          * the excess */
1122         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1123                 while (total > point) {
1124                         order = ilog2(total - point);
1125                         n = 1 << order;
1126                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1127                         atomic_long_sub(n, &mmap_pages_allocated);
1128                         total -= n;
1129                         set_page_refcounted(pages + total);
1130                         __free_pages(pages + total, order);
1131                 }
1132         }
1133
1134         for (point = 1; point < total; point++)
1135                 set_page_refcounted(&pages[point]);
1136
1137         base = page_address(pages);
1138         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1139         region->vm_start = (unsigned long) base;
1140         region->vm_end   = region->vm_start + rlen;
1141         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1142
1143         vma->vm_start = region->vm_start;
1144         vma->vm_end   = region->vm_start + len;
1145
1146         if (vma->vm_file) {
1147                 /* read the contents of a file into the copy */
1148                 mm_segment_t old_fs;
1149                 loff_t fpos;
1150
1151                 fpos = vma->vm_pgoff;
1152                 fpos <<= PAGE_SHIFT;
1153
1154                 old_fs = get_fs();
1155                 set_fs(KERNEL_DS);
1156                 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1157                 set_fs(old_fs);
1158
1159                 if (ret < 0)
1160                         goto error_free;
1161
1162                 /* clear the last little bit */
1163                 if (ret < rlen)
1164                         memset(base + ret, 0, rlen - ret);
1165
1166         } else {
1167                 /* if it's an anonymous mapping, then just clear it */
1168                 memset(base, 0, rlen);
1169         }
1170
1171         return 0;
1172
1173 error_free:
1174         free_page_series(region->vm_start, region->vm_end);
1175         region->vm_start = vma->vm_start = 0;
1176         region->vm_end   = vma->vm_end = 0;
1177         region->vm_top   = 0;
1178         return ret;
1179
1180 enomem:
1181         printk("Allocation of length %lu from process %d (%s) failed\n",
1182                len, current->pid, current->comm);
1183         show_free_areas();
1184         return -ENOMEM;
1185 }
1186
1187 /*
1188  * handle mapping creation for uClinux
1189  */
1190 unsigned long do_mmap_pgoff(struct file *file,
1191                             unsigned long addr,
1192                             unsigned long len,
1193                             unsigned long prot,
1194                             unsigned long flags,
1195                             unsigned long pgoff)
1196 {
1197         struct vm_area_struct *vma;
1198         struct vm_region *region;
1199         struct rb_node *rb;
1200         unsigned long capabilities, vm_flags, result;
1201         int ret;
1202
1203         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1204
1205         if (!(flags & MAP_FIXED))
1206                 addr = round_hint_to_min(addr);
1207
1208         /* decide whether we should attempt the mapping, and if so what sort of
1209          * mapping */
1210         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1211                                     &capabilities);
1212         if (ret < 0) {
1213                 kleave(" = %d [val]", ret);
1214                 return ret;
1215         }
1216
1217         /* we've determined that we can make the mapping, now translate what we
1218          * now know into VMA flags */
1219         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1220
1221         /* we're going to need to record the mapping */
1222         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1223         if (!region)
1224                 goto error_getting_region;
1225
1226         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1227         if (!vma)
1228                 goto error_getting_vma;
1229
1230         atomic_set(&region->vm_usage, 1);
1231         region->vm_flags = vm_flags;
1232         region->vm_pgoff = pgoff;
1233
1234         INIT_LIST_HEAD(&vma->anon_vma_node);
1235         vma->vm_flags = vm_flags;
1236         vma->vm_pgoff = pgoff;
1237
1238         if (file) {
1239                 region->vm_file = file;
1240                 get_file(file);
1241                 vma->vm_file = file;
1242                 get_file(file);
1243                 if (vm_flags & VM_EXECUTABLE) {
1244                         added_exe_file_vma(current->mm);
1245                         vma->vm_mm = current->mm;
1246                 }
1247         }
1248
1249         down_write(&nommu_region_sem);
1250
1251         /* if we want to share, we need to check for regions created by other
1252          * mmap() calls that overlap with our proposed mapping
1253          * - we can only share with a superset match on most regular files
1254          * - shared mappings on character devices and memory backed files are
1255          *   permitted to overlap inexactly as far as we are concerned for in
1256          *   these cases, sharing is handled in the driver or filesystem rather
1257          *   than here
1258          */
1259         if (vm_flags & VM_MAYSHARE) {
1260                 struct vm_region *pregion;
1261                 unsigned long pglen, rpglen, pgend, rpgend, start;
1262
1263                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1264                 pgend = pgoff + pglen;
1265
1266                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1267                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1268
1269                         if (!(pregion->vm_flags & VM_MAYSHARE))
1270                                 continue;
1271
1272                         /* search for overlapping mappings on the same file */
1273                         if (pregion->vm_file->f_path.dentry->d_inode !=
1274                             file->f_path.dentry->d_inode)
1275                                 continue;
1276
1277                         if (pregion->vm_pgoff >= pgend)
1278                                 continue;
1279
1280                         rpglen = pregion->vm_end - pregion->vm_start;
1281                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1282                         rpgend = pregion->vm_pgoff + rpglen;
1283                         if (pgoff >= rpgend)
1284                                 continue;
1285
1286                         /* handle inexactly overlapping matches between
1287                          * mappings */
1288                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1289                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1290                                 /* new mapping is not a subset of the region */
1291                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1292                                         goto sharing_violation;
1293                                 continue;
1294                         }
1295
1296                         /* we've found a region we can share */
1297                         atomic_inc(&pregion->vm_usage);
1298                         vma->vm_region = pregion;
1299                         start = pregion->vm_start;
1300                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1301                         vma->vm_start = start;
1302                         vma->vm_end = start + len;
1303
1304                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1305                                 kdebug("share copy");
1306                                 vma->vm_flags |= VM_MAPPED_COPY;
1307                         } else {
1308                                 kdebug("share mmap");
1309                                 ret = do_mmap_shared_file(vma);
1310                                 if (ret < 0) {
1311                                         vma->vm_region = NULL;
1312                                         vma->vm_start = 0;
1313                                         vma->vm_end = 0;
1314                                         atomic_dec(&pregion->vm_usage);
1315                                         pregion = NULL;
1316                                         goto error_just_free;
1317                                 }
1318                         }
1319                         fput(region->vm_file);
1320                         kmem_cache_free(vm_region_jar, region);
1321                         region = pregion;
1322                         result = start;
1323                         goto share;
1324                 }
1325
1326                 /* obtain the address at which to make a shared mapping
1327                  * - this is the hook for quasi-memory character devices to
1328                  *   tell us the location of a shared mapping
1329                  */
1330                 if (file && file->f_op->get_unmapped_area) {
1331                         addr = file->f_op->get_unmapped_area(file, addr, len,
1332                                                              pgoff, flags);
1333                         if (IS_ERR((void *) addr)) {
1334                                 ret = addr;
1335                                 if (ret != (unsigned long) -ENOSYS)
1336                                         goto error_just_free;
1337
1338                                 /* the driver refused to tell us where to site
1339                                  * the mapping so we'll have to attempt to copy
1340                                  * it */
1341                                 ret = (unsigned long) -ENODEV;
1342                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1343                                         goto error_just_free;
1344
1345                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1346                         } else {
1347                                 vma->vm_start = region->vm_start = addr;
1348                                 vma->vm_end = region->vm_end = addr + len;
1349                         }
1350                 }
1351         }
1352
1353         vma->vm_region = region;
1354
1355         /* set up the mapping */
1356         if (file && vma->vm_flags & VM_SHARED)
1357                 ret = do_mmap_shared_file(vma);
1358         else
1359                 ret = do_mmap_private(vma, region, len);
1360         if (ret < 0)
1361                 goto error_put_region;
1362
1363         add_nommu_region(region);
1364
1365         /* okay... we have a mapping; now we have to register it */
1366         result = vma->vm_start;
1367
1368         current->mm->total_vm += len >> PAGE_SHIFT;
1369
1370 share:
1371         add_vma_to_mm(current->mm, vma);
1372
1373         up_write(&nommu_region_sem);
1374
1375         if (prot & PROT_EXEC)
1376                 flush_icache_range(result, result + len);
1377
1378         kleave(" = %lx", result);
1379         return result;
1380
1381 error_put_region:
1382         __put_nommu_region(region);
1383         if (vma) {
1384                 if (vma->vm_file) {
1385                         fput(vma->vm_file);
1386                         if (vma->vm_flags & VM_EXECUTABLE)
1387                                 removed_exe_file_vma(vma->vm_mm);
1388                 }
1389                 kmem_cache_free(vm_area_cachep, vma);
1390         }
1391         kleave(" = %d [pr]", ret);
1392         return ret;
1393
1394 error_just_free:
1395         up_write(&nommu_region_sem);
1396 error:
1397         fput(region->vm_file);
1398         kmem_cache_free(vm_region_jar, region);
1399         fput(vma->vm_file);
1400         if (vma->vm_flags & VM_EXECUTABLE)
1401                 removed_exe_file_vma(vma->vm_mm);
1402         kmem_cache_free(vm_area_cachep, vma);
1403         kleave(" = %d", ret);
1404         return ret;
1405
1406 sharing_violation:
1407         up_write(&nommu_region_sem);
1408         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1409         ret = -EINVAL;
1410         goto error;
1411
1412 error_getting_vma:
1413         kmem_cache_free(vm_region_jar, region);
1414         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1415                " from process %d failed\n",
1416                len, current->pid);
1417         show_free_areas();
1418         return -ENOMEM;
1419
1420 error_getting_region:
1421         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1422                " from process %d failed\n",
1423                len, current->pid);
1424         show_free_areas();
1425         return -ENOMEM;
1426 }
1427 EXPORT_SYMBOL(do_mmap_pgoff);
1428
1429 /*
1430  * split a vma into two pieces at address 'addr', a new vma is allocated either
1431  * for the first part or the tail.
1432  */
1433 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1434               unsigned long addr, int new_below)
1435 {
1436         struct vm_area_struct *new;
1437         struct vm_region *region;
1438         unsigned long npages;
1439
1440         kenter("");
1441
1442         /* we're only permitted to split anonymous regions that have a single
1443          * owner */
1444         if (vma->vm_file ||
1445             atomic_read(&vma->vm_region->vm_usage) != 1)
1446                 return -ENOMEM;
1447
1448         if (mm->map_count >= sysctl_max_map_count)
1449                 return -ENOMEM;
1450
1451         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1452         if (!region)
1453                 return -ENOMEM;
1454
1455         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1456         if (!new) {
1457                 kmem_cache_free(vm_region_jar, region);
1458                 return -ENOMEM;
1459         }
1460
1461         /* most fields are the same, copy all, and then fixup */
1462         *new = *vma;
1463         *region = *vma->vm_region;
1464         new->vm_region = region;
1465
1466         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1467
1468         if (new_below) {
1469                 region->vm_top = region->vm_end = new->vm_end = addr;
1470         } else {
1471                 region->vm_start = new->vm_start = addr;
1472                 region->vm_pgoff = new->vm_pgoff += npages;
1473         }
1474
1475         if (new->vm_ops && new->vm_ops->open)
1476                 new->vm_ops->open(new);
1477
1478         delete_vma_from_mm(vma);
1479         down_write(&nommu_region_sem);
1480         delete_nommu_region(vma->vm_region);
1481         if (new_below) {
1482                 vma->vm_region->vm_start = vma->vm_start = addr;
1483                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1484         } else {
1485                 vma->vm_region->vm_end = vma->vm_end = addr;
1486                 vma->vm_region->vm_top = addr;
1487         }
1488         add_nommu_region(vma->vm_region);
1489         add_nommu_region(new->vm_region);
1490         up_write(&nommu_region_sem);
1491         add_vma_to_mm(mm, vma);
1492         add_vma_to_mm(mm, new);
1493         return 0;
1494 }
1495
1496 /*
1497  * shrink a VMA by removing the specified chunk from either the beginning or
1498  * the end
1499  */
1500 static int shrink_vma(struct mm_struct *mm,
1501                       struct vm_area_struct *vma,
1502                       unsigned long from, unsigned long to)
1503 {
1504         struct vm_region *region;
1505
1506         kenter("");
1507
1508         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1509          * and list */
1510         delete_vma_from_mm(vma);
1511         if (from > vma->vm_start)
1512                 vma->vm_end = from;
1513         else
1514                 vma->vm_start = to;
1515         add_vma_to_mm(mm, vma);
1516
1517         /* cut the backing region down to size */
1518         region = vma->vm_region;
1519         BUG_ON(atomic_read(&region->vm_usage) != 1);
1520
1521         down_write(&nommu_region_sem);
1522         delete_nommu_region(region);
1523         if (from > region->vm_start) {
1524                 to = region->vm_top;
1525                 region->vm_top = region->vm_end = from;
1526         } else {
1527                 region->vm_start = to;
1528         }
1529         add_nommu_region(region);
1530         up_write(&nommu_region_sem);
1531
1532         free_page_series(from, to);
1533         return 0;
1534 }
1535
1536 /*
1537  * release a mapping
1538  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1539  *   VMA, though it need not cover the whole VMA
1540  */
1541 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1542 {
1543         struct vm_area_struct *vma;
1544         struct rb_node *rb;
1545         unsigned long end = start + len;
1546         int ret;
1547
1548         kenter(",%lx,%zx", start, len);
1549
1550         if (len == 0)
1551                 return -EINVAL;
1552
1553         /* find the first potentially overlapping VMA */
1554         vma = find_vma(mm, start);
1555         if (!vma) {
1556                 static int limit = 0;
1557                 if (limit < 5) {
1558                         printk(KERN_WARNING
1559                                "munmap of memory not mmapped by process %d"
1560                                " (%s): 0x%lx-0x%lx\n",
1561                                current->pid, current->comm,
1562                                start, start + len - 1);
1563                         limit++;
1564                 }
1565                 return -EINVAL;
1566         }
1567
1568         /* we're allowed to split an anonymous VMA but not a file-backed one */
1569         if (vma->vm_file) {
1570                 do {
1571                         if (start > vma->vm_start) {
1572                                 kleave(" = -EINVAL [miss]");
1573                                 return -EINVAL;
1574                         }
1575                         if (end == vma->vm_end)
1576                                 goto erase_whole_vma;
1577                         rb = rb_next(&vma->vm_rb);
1578                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1579                 } while (rb);
1580                 kleave(" = -EINVAL [split file]");
1581                 return -EINVAL;
1582         } else {
1583                 /* the chunk must be a subset of the VMA found */
1584                 if (start == vma->vm_start && end == vma->vm_end)
1585                         goto erase_whole_vma;
1586                 if (start < vma->vm_start || end > vma->vm_end) {
1587                         kleave(" = -EINVAL [superset]");
1588                         return -EINVAL;
1589                 }
1590                 if (start & ~PAGE_MASK) {
1591                         kleave(" = -EINVAL [unaligned start]");
1592                         return -EINVAL;
1593                 }
1594                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1595                         kleave(" = -EINVAL [unaligned split]");
1596                         return -EINVAL;
1597                 }
1598                 if (start != vma->vm_start && end != vma->vm_end) {
1599                         ret = split_vma(mm, vma, start, 1);
1600                         if (ret < 0) {
1601                                 kleave(" = %d [split]", ret);
1602                                 return ret;
1603                         }
1604                 }
1605                 return shrink_vma(mm, vma, start, end);
1606         }
1607
1608 erase_whole_vma:
1609         delete_vma_from_mm(vma);
1610         delete_vma(mm, vma);
1611         kleave(" = 0");
1612         return 0;
1613 }
1614 EXPORT_SYMBOL(do_munmap);
1615
1616 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1617 {
1618         int ret;
1619         struct mm_struct *mm = current->mm;
1620
1621         down_write(&mm->mmap_sem);
1622         ret = do_munmap(mm, addr, len);
1623         up_write(&mm->mmap_sem);
1624         return ret;
1625 }
1626
1627 /*
1628  * release all the mappings made in a process's VM space
1629  */
1630 void exit_mmap(struct mm_struct *mm)
1631 {
1632         struct vm_area_struct *vma;
1633
1634         if (!mm)
1635                 return;
1636
1637         kenter("");
1638
1639         mm->total_vm = 0;
1640
1641         while ((vma = mm->mmap)) {
1642                 mm->mmap = vma->vm_next;
1643                 delete_vma_from_mm(vma);
1644                 delete_vma(mm, vma);
1645         }
1646
1647         kleave("");
1648 }
1649
1650 unsigned long do_brk(unsigned long addr, unsigned long len)
1651 {
1652         return -ENOMEM;
1653 }
1654
1655 /*
1656  * expand (or shrink) an existing mapping, potentially moving it at the same
1657  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1658  *
1659  * under NOMMU conditions, we only permit changing a mapping's size, and only
1660  * as long as it stays within the region allocated by do_mmap_private() and the
1661  * block is not shareable
1662  *
1663  * MREMAP_FIXED is not supported under NOMMU conditions
1664  */
1665 unsigned long do_mremap(unsigned long addr,
1666                         unsigned long old_len, unsigned long new_len,
1667                         unsigned long flags, unsigned long new_addr)
1668 {
1669         struct vm_area_struct *vma;
1670
1671         /* insanity checks first */
1672         if (old_len == 0 || new_len == 0)
1673                 return (unsigned long) -EINVAL;
1674
1675         if (addr & ~PAGE_MASK)
1676                 return -EINVAL;
1677
1678         if (flags & MREMAP_FIXED && new_addr != addr)
1679                 return (unsigned long) -EINVAL;
1680
1681         vma = find_vma_exact(current->mm, addr, old_len);
1682         if (!vma)
1683                 return (unsigned long) -EINVAL;
1684
1685         if (vma->vm_end != vma->vm_start + old_len)
1686                 return (unsigned long) -EFAULT;
1687
1688         if (vma->vm_flags & VM_MAYSHARE)
1689                 return (unsigned long) -EPERM;
1690
1691         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1692                 return (unsigned long) -ENOMEM;
1693
1694         /* all checks complete - do it */
1695         vma->vm_end = vma->vm_start + new_len;
1696         return vma->vm_start;
1697 }
1698 EXPORT_SYMBOL(do_mremap);
1699
1700 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1701                 unsigned long, new_len, unsigned long, flags,
1702                 unsigned long, new_addr)
1703 {
1704         unsigned long ret;
1705
1706         down_write(&current->mm->mmap_sem);
1707         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1708         up_write(&current->mm->mmap_sem);
1709         return ret;
1710 }
1711
1712 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1713                         unsigned int foll_flags)
1714 {
1715         return NULL;
1716 }
1717
1718 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1719                 unsigned long to, unsigned long size, pgprot_t prot)
1720 {
1721         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1722         return 0;
1723 }
1724 EXPORT_SYMBOL(remap_pfn_range);
1725
1726 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1727                         unsigned long pgoff)
1728 {
1729         unsigned int size = vma->vm_end - vma->vm_start;
1730
1731         if (!(vma->vm_flags & VM_USERMAP))
1732                 return -EINVAL;
1733
1734         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1735         vma->vm_end = vma->vm_start + size;
1736
1737         return 0;
1738 }
1739 EXPORT_SYMBOL(remap_vmalloc_range);
1740
1741 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1742 {
1743 }
1744
1745 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1746         unsigned long len, unsigned long pgoff, unsigned long flags)
1747 {
1748         return -ENOMEM;
1749 }
1750
1751 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1752 {
1753 }
1754
1755 void unmap_mapping_range(struct address_space *mapping,
1756                          loff_t const holebegin, loff_t const holelen,
1757                          int even_cows)
1758 {
1759 }
1760 EXPORT_SYMBOL(unmap_mapping_range);
1761
1762 /*
1763  * ask for an unmapped area at which to create a mapping on a file
1764  */
1765 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1766                                 unsigned long len, unsigned long pgoff,
1767                                 unsigned long flags)
1768 {
1769         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1770                                   unsigned long, unsigned long);
1771
1772         get_area = current->mm->get_unmapped_area;
1773         if (file && file->f_op && file->f_op->get_unmapped_area)
1774                 get_area = file->f_op->get_unmapped_area;
1775
1776         if (!get_area)
1777                 return -ENOSYS;
1778
1779         return get_area(file, addr, len, pgoff, flags);
1780 }
1781 EXPORT_SYMBOL(get_unmapped_area);
1782
1783 /*
1784  * Check that a process has enough memory to allocate a new virtual
1785  * mapping. 0 means there is enough memory for the allocation to
1786  * succeed and -ENOMEM implies there is not.
1787  *
1788  * We currently support three overcommit policies, which are set via the
1789  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1790  *
1791  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1792  * Additional code 2002 Jul 20 by Robert Love.
1793  *
1794  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1795  *
1796  * Note this is a helper function intended to be used by LSMs which
1797  * wish to use this logic.
1798  */
1799 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1800 {
1801         unsigned long free, allowed;
1802
1803         vm_acct_memory(pages);
1804
1805         /*
1806          * Sometimes we want to use more memory than we have
1807          */
1808         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1809                 return 0;
1810
1811         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1812                 unsigned long n;
1813
1814                 free = global_page_state(NR_FILE_PAGES);
1815                 free += nr_swap_pages;
1816
1817                 /*
1818                  * Any slabs which are created with the
1819                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1820                  * which are reclaimable, under pressure.  The dentry
1821                  * cache and most inode caches should fall into this
1822                  */
1823                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1824
1825                 /*
1826                  * Leave the last 3% for root
1827                  */
1828                 if (!cap_sys_admin)
1829                         free -= free / 32;
1830
1831                 if (free > pages)
1832                         return 0;
1833
1834                 /*
1835                  * nr_free_pages() is very expensive on large systems,
1836                  * only call if we're about to fail.
1837                  */
1838                 n = nr_free_pages();
1839
1840                 /*
1841                  * Leave reserved pages. The pages are not for anonymous pages.
1842                  */
1843                 if (n <= totalreserve_pages)
1844                         goto error;
1845                 else
1846                         n -= totalreserve_pages;
1847
1848                 /*
1849                  * Leave the last 3% for root
1850                  */
1851                 if (!cap_sys_admin)
1852                         n -= n / 32;
1853                 free += n;
1854
1855                 if (free > pages)
1856                         return 0;
1857
1858                 goto error;
1859         }
1860
1861         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1862         /*
1863          * Leave the last 3% for root
1864          */
1865         if (!cap_sys_admin)
1866                 allowed -= allowed / 32;
1867         allowed += total_swap_pages;
1868
1869         /* Don't let a single process grow too big:
1870            leave 3% of the size of this process for other processes */
1871         if (mm)
1872                 allowed -= mm->total_vm / 32;
1873
1874         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1875                 return 0;
1876
1877 error:
1878         vm_unacct_memory(pages);
1879
1880         return -ENOMEM;
1881 }
1882
1883 int in_gate_area_no_task(unsigned long addr)
1884 {
1885         return 0;
1886 }
1887
1888 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1889 {
1890         BUG();
1891         return 0;
1892 }
1893 EXPORT_SYMBOL(filemap_fault);
1894
1895 /*
1896  * Access another process' address space.
1897  * - source/target buffer must be kernel space
1898  */
1899 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1900 {
1901         struct vm_area_struct *vma;
1902         struct mm_struct *mm;
1903
1904         if (addr + len < addr)
1905                 return 0;
1906
1907         mm = get_task_mm(tsk);
1908         if (!mm)
1909                 return 0;
1910
1911         down_read(&mm->mmap_sem);
1912
1913         /* the access must start within one of the target process's mappings */
1914         vma = find_vma(mm, addr);
1915         if (vma) {
1916                 /* don't overrun this mapping */
1917                 if (addr + len >= vma->vm_end)
1918                         len = vma->vm_end - addr;
1919
1920                 /* only read or write mappings where it is permitted */
1921                 if (write && vma->vm_flags & VM_MAYWRITE)
1922                         len -= copy_to_user((void *) addr, buf, len);
1923                 else if (!write && vma->vm_flags & VM_MAYREAD)
1924                         len -= copy_from_user(buf, (void *) addr, len);
1925                 else
1926                         len = 0;
1927         } else {
1928                 len = 0;
1929         }
1930
1931         up_read(&mm->mmap_sem);
1932         mmput(mm);
1933         return len;
1934 }