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