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