386443e9d2c6cbcc929ee36991259d8f9dd66077
[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 void *high_memory;
60 struct page *mem_map;
61 unsigned long max_mapnr;
62 unsigned long num_physpages;
63 struct percpu_counter vm_committed_as;
64 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
65 int sysctl_overcommit_ratio = 50; /* default is 50% */
66 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
67 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
68 int heap_stack_gap = 0;
69
70 atomic_long_t mmap_pages_allocated;
71
72 EXPORT_SYMBOL(mem_map);
73 EXPORT_SYMBOL(num_physpages);
74
75 /* list of mapped, potentially shareable regions */
76 static struct kmem_cache *vm_region_jar;
77 struct rb_root nommu_region_tree = RB_ROOT;
78 DECLARE_RWSEM(nommu_region_sem);
79
80 struct vm_operations_struct generic_file_vm_ops = {
81 };
82
83 /*
84  * Handle all mappings that got truncated by a "truncate()"
85  * system call.
86  *
87  * NOTE! We have to be ready to update the memory sharing
88  * between the file and the memory map for a potential last
89  * incomplete page.  Ugly, but necessary.
90  */
91 int vmtruncate(struct inode *inode, loff_t offset)
92 {
93         struct address_space *mapping = inode->i_mapping;
94         unsigned long limit;
95
96         if (inode->i_size < offset)
97                 goto do_expand;
98         i_size_write(inode, offset);
99
100         truncate_inode_pages(mapping, offset);
101         goto out_truncate;
102
103 do_expand:
104         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
105         if (limit != RLIM_INFINITY && offset > limit)
106                 goto out_sig;
107         if (offset > inode->i_sb->s_maxbytes)
108                 goto out;
109         i_size_write(inode, offset);
110
111 out_truncate:
112         if (inode->i_op->truncate)
113                 inode->i_op->truncate(inode);
114         return 0;
115 out_sig:
116         send_sig(SIGXFSZ, current, 0);
117 out:
118         return -EFBIG;
119 }
120
121 EXPORT_SYMBOL(vmtruncate);
122
123 /*
124  * Return the total memory allocated for this pointer, not
125  * just what the caller asked for.
126  *
127  * Doesn't have to be accurate, i.e. may have races.
128  */
129 unsigned int kobjsize(const void *objp)
130 {
131         struct page *page;
132
133         /*
134          * If the object we have should not have ksize performed on it,
135          * return size of 0
136          */
137         if (!objp || !virt_addr_valid(objp))
138                 return 0;
139
140         page = virt_to_head_page(objp);
141
142         /*
143          * If the allocator sets PageSlab, we know the pointer came from
144          * kmalloc().
145          */
146         if (PageSlab(page))
147                 return ksize(objp);
148
149         /*
150          * If it's not a compound page, see if we have a matching VMA
151          * region. This test is intentionally done in reverse order,
152          * so if there's no VMA, we still fall through and hand back
153          * PAGE_SIZE for 0-order pages.
154          */
155         if (!PageCompound(page)) {
156                 struct vm_area_struct *vma;
157
158                 vma = find_vma(current->mm, (unsigned long)objp);
159                 if (vma)
160                         return vma->vm_end - vma->vm_start;
161         }
162
163         /*
164          * The ksize() function is only guaranteed to work for pointers
165          * returned by kmalloc(). So handle arbitrary pointers here.
166          */
167         return PAGE_SIZE << compound_order(page);
168 }
169
170 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
171                      unsigned long start, int nr_pages, int flags,
172                      struct page **pages, struct vm_area_struct **vmas)
173 {
174         struct vm_area_struct *vma;
175         unsigned long vm_flags;
176         int i;
177         int write = !!(flags & GUP_FLAGS_WRITE);
178         int force = !!(flags & GUP_FLAGS_FORCE);
179
180         /* calculate required read or write permissions.
181          * - if 'force' is set, we only require the "MAY" flags.
182          */
183         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
184         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
185
186         for (i = 0; i < nr_pages; i++) {
187                 vma = find_vma(mm, start);
188                 if (!vma)
189                         goto finish_or_fault;
190
191                 /* protect what we can, including chardevs */
192                 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
193                     !(vm_flags & vma->vm_flags))
194                         goto finish_or_fault;
195
196                 if (pages) {
197                         pages[i] = virt_to_page(start);
198                         if (pages[i])
199                                 page_cache_get(pages[i]);
200                 }
201                 if (vmas)
202                         vmas[i] = vma;
203                 start += PAGE_SIZE;
204         }
205
206         return i;
207
208 finish_or_fault:
209         return i ? : -EFAULT;
210 }
211
212 /*
213  * get a list of pages in an address range belonging to the specified process
214  * and indicate the VMA that covers each page
215  * - this is potentially dodgy as we may end incrementing the page count of a
216  *   slab page or a secondary page from a compound page
217  * - don't permit access to VMAs that don't support it, such as I/O mappings
218  */
219 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
220         unsigned long start, int nr_pages, int write, int force,
221         struct page **pages, struct vm_area_struct **vmas)
222 {
223         int flags = 0;
224
225         if (write)
226                 flags |= GUP_FLAGS_WRITE;
227         if (force)
228                 flags |= GUP_FLAGS_FORCE;
229
230         return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
231 }
232 EXPORT_SYMBOL(get_user_pages);
233
234 /**
235  * follow_pfn - look up PFN at a user virtual address
236  * @vma: memory mapping
237  * @address: user virtual address
238  * @pfn: location to store found PFN
239  *
240  * Only IO mappings and raw PFN mappings are allowed.
241  *
242  * Returns zero and the pfn at @pfn on success, -ve otherwise.
243  */
244 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
245         unsigned long *pfn)
246 {
247         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
248                 return -EINVAL;
249
250         *pfn = address >> PAGE_SHIFT;
251         return 0;
252 }
253 EXPORT_SYMBOL(follow_pfn);
254
255 DEFINE_RWLOCK(vmlist_lock);
256 struct vm_struct *vmlist;
257
258 void vfree(const void *addr)
259 {
260         kfree(addr);
261 }
262 EXPORT_SYMBOL(vfree);
263
264 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
265 {
266         /*
267          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
268          * returns only a logical address.
269          */
270         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
271 }
272 EXPORT_SYMBOL(__vmalloc);
273
274 void *vmalloc_user(unsigned long size)
275 {
276         void *ret;
277
278         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
279                         PAGE_KERNEL);
280         if (ret) {
281                 struct vm_area_struct *vma;
282
283                 down_write(&current->mm->mmap_sem);
284                 vma = find_vma(current->mm, (unsigned long)ret);
285                 if (vma)
286                         vma->vm_flags |= VM_USERMAP;
287                 up_write(&current->mm->mmap_sem);
288         }
289
290         return ret;
291 }
292 EXPORT_SYMBOL(vmalloc_user);
293
294 struct page *vmalloc_to_page(const void *addr)
295 {
296         return virt_to_page(addr);
297 }
298 EXPORT_SYMBOL(vmalloc_to_page);
299
300 unsigned long vmalloc_to_pfn(const void *addr)
301 {
302         return page_to_pfn(virt_to_page(addr));
303 }
304 EXPORT_SYMBOL(vmalloc_to_pfn);
305
306 long vread(char *buf, char *addr, unsigned long count)
307 {
308         memcpy(buf, addr, count);
309         return count;
310 }
311
312 long vwrite(char *buf, char *addr, unsigned long count)
313 {
314         /* Don't allow overflow */
315         if ((unsigned long) addr + count < count)
316                 count = -(unsigned long) addr;
317
318         memcpy(addr, buf, count);
319         return(count);
320 }
321
322 /*
323  *      vmalloc  -  allocate virtually continguos memory
324  *
325  *      @size:          allocation size
326  *
327  *      Allocate enough pages to cover @size from the page level
328  *      allocator and map them into continguos kernel virtual space.
329  *
330  *      For tight control over page level allocator and protection flags
331  *      use __vmalloc() instead.
332  */
333 void *vmalloc(unsigned long size)
334 {
335        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
336 }
337 EXPORT_SYMBOL(vmalloc);
338
339 void *vmalloc_node(unsigned long size, int node)
340 {
341         return vmalloc(size);
342 }
343 EXPORT_SYMBOL(vmalloc_node);
344
345 #ifndef PAGE_KERNEL_EXEC
346 # define PAGE_KERNEL_EXEC PAGE_KERNEL
347 #endif
348
349 /**
350  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
351  *      @size:          allocation size
352  *
353  *      Kernel-internal function to allocate enough pages to cover @size
354  *      the page level allocator and map them into contiguous and
355  *      executable kernel virtual space.
356  *
357  *      For tight control over page level allocator and protection flags
358  *      use __vmalloc() instead.
359  */
360
361 void *vmalloc_exec(unsigned long size)
362 {
363         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
364 }
365
366 /**
367  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
368  *      @size:          allocation size
369  *
370  *      Allocate enough 32bit PA addressable pages to cover @size from the
371  *      page level allocator and map them into continguos kernel virtual space.
372  */
373 void *vmalloc_32(unsigned long size)
374 {
375         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
376 }
377 EXPORT_SYMBOL(vmalloc_32);
378
379 /**
380  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
381  *      @size:          allocation size
382  *
383  * The resulting memory area is 32bit addressable and zeroed so it can be
384  * mapped to userspace without leaking data.
385  *
386  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
387  * remap_vmalloc_range() are permissible.
388  */
389 void *vmalloc_32_user(unsigned long size)
390 {
391         /*
392          * We'll have to sort out the ZONE_DMA bits for 64-bit,
393          * but for now this can simply use vmalloc_user() directly.
394          */
395         return vmalloc_user(size);
396 }
397 EXPORT_SYMBOL(vmalloc_32_user);
398
399 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
400 {
401         BUG();
402         return NULL;
403 }
404 EXPORT_SYMBOL(vmap);
405
406 void vunmap(const void *addr)
407 {
408         BUG();
409 }
410 EXPORT_SYMBOL(vunmap);
411
412 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
413 {
414         BUG();
415         return NULL;
416 }
417 EXPORT_SYMBOL(vm_map_ram);
418
419 void vm_unmap_ram(const void *mem, unsigned int count)
420 {
421         BUG();
422 }
423 EXPORT_SYMBOL(vm_unmap_ram);
424
425 void vm_unmap_aliases(void)
426 {
427 }
428 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
429
430 /*
431  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
432  * have one.
433  */
434 void  __attribute__((weak)) vmalloc_sync_all(void)
435 {
436 }
437
438 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
439                    struct page *page)
440 {
441         return -EINVAL;
442 }
443 EXPORT_SYMBOL(vm_insert_page);
444
445 /*
446  *  sys_brk() for the most part doesn't need the global kernel
447  *  lock, except when an application is doing something nasty
448  *  like trying to un-brk an area that has already been mapped
449  *  to a regular file.  in this case, the unmapping will need
450  *  to invoke file system routines that need the global lock.
451  */
452 SYSCALL_DEFINE1(brk, unsigned long, brk)
453 {
454         struct mm_struct *mm = current->mm;
455
456         if (brk < mm->start_brk || brk > mm->context.end_brk)
457                 return mm->brk;
458
459         if (mm->brk == brk)
460                 return mm->brk;
461
462         /*
463          * Always allow shrinking brk
464          */
465         if (brk <= mm->brk) {
466                 mm->brk = brk;
467                 return brk;
468         }
469
470         /*
471          * Ok, looks good - let it rip.
472          */
473         return mm->brk = brk;
474 }
475
476 /*
477  * initialise the VMA and region record slabs
478  */
479 void __init mmap_init(void)
480 {
481         int ret;
482
483         ret = percpu_counter_init(&vm_committed_as, 0);
484         VM_BUG_ON(ret);
485         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
486 }
487
488 /*
489  * validate the region tree
490  * - the caller must hold the region lock
491  */
492 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
493 static noinline void validate_nommu_regions(void)
494 {
495         struct vm_region *region, *last;
496         struct rb_node *p, *lastp;
497
498         lastp = rb_first(&nommu_region_tree);
499         if (!lastp)
500                 return;
501
502         last = rb_entry(lastp, struct vm_region, vm_rb);
503         BUG_ON(unlikely(last->vm_end <= last->vm_start));
504         BUG_ON(unlikely(last->vm_top < last->vm_end));
505
506         while ((p = rb_next(lastp))) {
507                 region = rb_entry(p, struct vm_region, vm_rb);
508                 last = rb_entry(lastp, struct vm_region, vm_rb);
509
510                 BUG_ON(unlikely(region->vm_end <= region->vm_start));
511                 BUG_ON(unlikely(region->vm_top < region->vm_end));
512                 BUG_ON(unlikely(region->vm_start < last->vm_top));
513
514                 lastp = p;
515         }
516 }
517 #else
518 static void validate_nommu_regions(void)
519 {
520 }
521 #endif
522
523 /*
524  * add a region into the global tree
525  */
526 static void add_nommu_region(struct vm_region *region)
527 {
528         struct vm_region *pregion;
529         struct rb_node **p, *parent;
530
531         validate_nommu_regions();
532
533         parent = NULL;
534         p = &nommu_region_tree.rb_node;
535         while (*p) {
536                 parent = *p;
537                 pregion = rb_entry(parent, struct vm_region, vm_rb);
538                 if (region->vm_start < pregion->vm_start)
539                         p = &(*p)->rb_left;
540                 else if (region->vm_start > pregion->vm_start)
541                         p = &(*p)->rb_right;
542                 else if (pregion == region)
543                         return;
544                 else
545                         BUG();
546         }
547
548         rb_link_node(&region->vm_rb, parent, p);
549         rb_insert_color(&region->vm_rb, &nommu_region_tree);
550
551         validate_nommu_regions();
552 }
553
554 /*
555  * delete a region from the global tree
556  */
557 static void delete_nommu_region(struct vm_region *region)
558 {
559         BUG_ON(!nommu_region_tree.rb_node);
560
561         validate_nommu_regions();
562         rb_erase(&region->vm_rb, &nommu_region_tree);
563         validate_nommu_regions();
564 }
565
566 /*
567  * free a contiguous series of pages
568  */
569 static void free_page_series(unsigned long from, unsigned long to)
570 {
571         for (; from < to; from += PAGE_SIZE) {
572                 struct page *page = virt_to_page(from);
573
574                 kdebug("- free %lx", from);
575                 atomic_long_dec(&mmap_pages_allocated);
576                 if (page_count(page) != 1)
577                         kdebug("free page %p: refcount not one: %d",
578                                page, page_count(page));
579                 put_page(page);
580         }
581 }
582
583 /*
584  * release a reference to a region
585  * - the caller must hold the region semaphore for writing, which this releases
586  * - the region may not have been added to the tree yet, in which case vm_top
587  *   will equal vm_start
588  */
589 static void __put_nommu_region(struct vm_region *region)
590         __releases(nommu_region_sem)
591 {
592         kenter("%p{%d}", region, atomic_read(&region->vm_usage));
593
594         BUG_ON(!nommu_region_tree.rb_node);
595
596         if (atomic_dec_and_test(&region->vm_usage)) {
597                 if (region->vm_top > region->vm_start)
598                         delete_nommu_region(region);
599                 up_write(&nommu_region_sem);
600
601                 if (region->vm_file)
602                         fput(region->vm_file);
603
604                 /* IO memory and memory shared directly out of the pagecache
605                  * from ramfs/tmpfs mustn't be released here */
606                 if (region->vm_flags & VM_MAPPED_COPY) {
607                         kdebug("free series");
608                         free_page_series(region->vm_start, region->vm_top);
609                 }
610                 kmem_cache_free(vm_region_jar, region);
611         } else {
612                 up_write(&nommu_region_sem);
613         }
614 }
615
616 /*
617  * release a reference to a region
618  */
619 static void put_nommu_region(struct vm_region *region)
620 {
621         down_write(&nommu_region_sem);
622         __put_nommu_region(region);
623 }
624
625 /*
626  * add a VMA into a process's mm_struct in the appropriate place in the list
627  * and tree and add to the address space's page tree also if not an anonymous
628  * page
629  * - should be called with mm->mmap_sem held writelocked
630  */
631 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
632 {
633         struct vm_area_struct *pvma, **pp;
634         struct address_space *mapping;
635         struct rb_node **p, *parent;
636
637         kenter(",%p", vma);
638
639         BUG_ON(!vma->vm_region);
640
641         mm->map_count++;
642         vma->vm_mm = mm;
643
644         /* add the VMA to the mapping */
645         if (vma->vm_file) {
646                 mapping = vma->vm_file->f_mapping;
647
648                 flush_dcache_mmap_lock(mapping);
649                 vma_prio_tree_insert(vma, &mapping->i_mmap);
650                 flush_dcache_mmap_unlock(mapping);
651         }
652
653         /* add the VMA to the tree */
654         parent = NULL;
655         p = &mm->mm_rb.rb_node;
656         while (*p) {
657                 parent = *p;
658                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
659
660                 /* sort by: start addr, end addr, VMA struct addr in that order
661                  * (the latter is necessary as we may get identical VMAs) */
662                 if (vma->vm_start < pvma->vm_start)
663                         p = &(*p)->rb_left;
664                 else if (vma->vm_start > pvma->vm_start)
665                         p = &(*p)->rb_right;
666                 else if (vma->vm_end < pvma->vm_end)
667                         p = &(*p)->rb_left;
668                 else if (vma->vm_end > pvma->vm_end)
669                         p = &(*p)->rb_right;
670                 else if (vma < pvma)
671                         p = &(*p)->rb_left;
672                 else if (vma > pvma)
673                         p = &(*p)->rb_right;
674                 else
675                         BUG();
676         }
677
678         rb_link_node(&vma->vm_rb, parent, p);
679         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
680
681         /* add VMA to the VMA list also */
682         for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
683                 if (pvma->vm_start > vma->vm_start)
684                         break;
685                 if (pvma->vm_start < vma->vm_start)
686                         continue;
687                 if (pvma->vm_end < vma->vm_end)
688                         break;
689         }
690
691         vma->vm_next = *pp;
692         *pp = vma;
693 }
694
695 /*
696  * delete a VMA from its owning mm_struct and address space
697  */
698 static void delete_vma_from_mm(struct vm_area_struct *vma)
699 {
700         struct vm_area_struct **pp;
701         struct address_space *mapping;
702         struct mm_struct *mm = vma->vm_mm;
703
704         kenter("%p", vma);
705
706         mm->map_count--;
707         if (mm->mmap_cache == vma)
708                 mm->mmap_cache = NULL;
709
710         /* remove the VMA from the mapping */
711         if (vma->vm_file) {
712                 mapping = vma->vm_file->f_mapping;
713
714                 flush_dcache_mmap_lock(mapping);
715                 vma_prio_tree_remove(vma, &mapping->i_mmap);
716                 flush_dcache_mmap_unlock(mapping);
717         }
718
719         /* remove from the MM's tree and list */
720         rb_erase(&vma->vm_rb, &mm->mm_rb);
721         for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
722                 if (*pp == vma) {
723                         *pp = vma->vm_next;
724                         break;
725                 }
726         }
727
728         vma->vm_mm = NULL;
729 }
730
731 /*
732  * destroy a VMA record
733  */
734 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
735 {
736         kenter("%p", vma);
737         if (vma->vm_ops && vma->vm_ops->close)
738                 vma->vm_ops->close(vma);
739         if (vma->vm_file) {
740                 fput(vma->vm_file);
741                 if (vma->vm_flags & VM_EXECUTABLE)
742                         removed_exe_file_vma(mm);
743         }
744         put_nommu_region(vma->vm_region);
745         kmem_cache_free(vm_area_cachep, vma);
746 }
747
748 /*
749  * look up the first VMA in which addr resides, NULL if none
750  * - should be called with mm->mmap_sem at least held readlocked
751  */
752 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
753 {
754         struct vm_area_struct *vma;
755         struct rb_node *n = mm->mm_rb.rb_node;
756
757         /* check the cache first */
758         vma = mm->mmap_cache;
759         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
760                 return vma;
761
762         /* trawl the tree (there may be multiple mappings in which addr
763          * resides) */
764         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
765                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
766                 if (vma->vm_start > addr)
767                         return NULL;
768                 if (vma->vm_end > addr) {
769                         mm->mmap_cache = vma;
770                         return vma;
771                 }
772         }
773
774         return NULL;
775 }
776 EXPORT_SYMBOL(find_vma);
777
778 /*
779  * find a VMA
780  * - we don't extend stack VMAs under NOMMU conditions
781  */
782 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
783 {
784         return find_vma(mm, addr);
785 }
786
787 /*
788  * expand a stack to a given address
789  * - not supported under NOMMU conditions
790  */
791 int expand_stack(struct vm_area_struct *vma, unsigned long address)
792 {
793         return -ENOMEM;
794 }
795
796 /*
797  * look up the first VMA exactly that exactly matches addr
798  * - should be called with mm->mmap_sem at least held readlocked
799  */
800 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
801                                              unsigned long addr,
802                                              unsigned long len)
803 {
804         struct vm_area_struct *vma;
805         struct rb_node *n = mm->mm_rb.rb_node;
806         unsigned long end = addr + len;
807
808         /* check the cache first */
809         vma = mm->mmap_cache;
810         if (vma && vma->vm_start == addr && vma->vm_end == end)
811                 return vma;
812
813         /* trawl the tree (there may be multiple mappings in which addr
814          * resides) */
815         for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
816                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
817                 if (vma->vm_start < addr)
818                         continue;
819                 if (vma->vm_start > addr)
820                         return NULL;
821                 if (vma->vm_end == end) {
822                         mm->mmap_cache = vma;
823                         return vma;
824                 }
825         }
826
827         return NULL;
828 }
829
830 /*
831  * determine whether a mapping should be permitted and, if so, what sort of
832  * mapping we're capable of supporting
833  */
834 static int validate_mmap_request(struct file *file,
835                                  unsigned long addr,
836                                  unsigned long len,
837                                  unsigned long prot,
838                                  unsigned long flags,
839                                  unsigned long pgoff,
840                                  unsigned long *_capabilities)
841 {
842         unsigned long capabilities, rlen;
843         unsigned long reqprot = prot;
844         int ret;
845
846         /* do the simple checks first */
847         if (flags & MAP_FIXED || addr) {
848                 printk(KERN_DEBUG
849                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
850                        current->pid);
851                 return -EINVAL;
852         }
853
854         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
855             (flags & MAP_TYPE) != MAP_SHARED)
856                 return -EINVAL;
857
858         if (!len)
859                 return -EINVAL;
860
861         /* Careful about overflows.. */
862         rlen = PAGE_ALIGN(len);
863         if (!rlen || rlen > TASK_SIZE)
864                 return -ENOMEM;
865
866         /* offset overflow? */
867         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
868                 return -EOVERFLOW;
869
870         if (file) {
871                 /* validate file mapping requests */
872                 struct address_space *mapping;
873
874                 /* files must support mmap */
875                 if (!file->f_op || !file->f_op->mmap)
876                         return -ENODEV;
877
878                 /* work out if what we've got could possibly be shared
879                  * - we support chardevs that provide their own "memory"
880                  * - we support files/blockdevs that are memory backed
881                  */
882                 mapping = file->f_mapping;
883                 if (!mapping)
884                         mapping = file->f_path.dentry->d_inode->i_mapping;
885
886                 capabilities = 0;
887                 if (mapping && mapping->backing_dev_info)
888                         capabilities = mapping->backing_dev_info->capabilities;
889
890                 if (!capabilities) {
891                         /* no explicit capabilities set, so assume some
892                          * defaults */
893                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
894                         case S_IFREG:
895                         case S_IFBLK:
896                                 capabilities = BDI_CAP_MAP_COPY;
897                                 break;
898
899                         case S_IFCHR:
900                                 capabilities =
901                                         BDI_CAP_MAP_DIRECT |
902                                         BDI_CAP_READ_MAP |
903                                         BDI_CAP_WRITE_MAP;
904                                 break;
905
906                         default:
907                                 return -EINVAL;
908                         }
909                 }
910
911                 /* eliminate any capabilities that we can't support on this
912                  * device */
913                 if (!file->f_op->get_unmapped_area)
914                         capabilities &= ~BDI_CAP_MAP_DIRECT;
915                 if (!file->f_op->read)
916                         capabilities &= ~BDI_CAP_MAP_COPY;
917
918                 /* The file shall have been opened with read permission. */
919                 if (!(file->f_mode & FMODE_READ))
920                         return -EACCES;
921
922                 if (flags & MAP_SHARED) {
923                         /* do checks for writing, appending and locking */
924                         if ((prot & PROT_WRITE) &&
925                             !(file->f_mode & FMODE_WRITE))
926                                 return -EACCES;
927
928                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
929                             (file->f_mode & FMODE_WRITE))
930                                 return -EACCES;
931
932                         if (locks_verify_locked(file->f_path.dentry->d_inode))
933                                 return -EAGAIN;
934
935                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
936                                 return -ENODEV;
937
938                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
939                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
940                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
941                             ) {
942                                 printk("MAP_SHARED not completely supported on !MMU\n");
943                                 return -EINVAL;
944                         }
945
946                         /* we mustn't privatise shared mappings */
947                         capabilities &= ~BDI_CAP_MAP_COPY;
948                 }
949                 else {
950                         /* we're going to read the file into private memory we
951                          * allocate */
952                         if (!(capabilities & BDI_CAP_MAP_COPY))
953                                 return -ENODEV;
954
955                         /* we don't permit a private writable mapping to be
956                          * shared with the backing device */
957                         if (prot & PROT_WRITE)
958                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
959                 }
960
961                 /* handle executable mappings and implied executable
962                  * mappings */
963                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
964                         if (prot & PROT_EXEC)
965                                 return -EPERM;
966                 }
967                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
968                         /* handle implication of PROT_EXEC by PROT_READ */
969                         if (current->personality & READ_IMPLIES_EXEC) {
970                                 if (capabilities & BDI_CAP_EXEC_MAP)
971                                         prot |= PROT_EXEC;
972                         }
973                 }
974                 else if ((prot & PROT_READ) &&
975                          (prot & PROT_EXEC) &&
976                          !(capabilities & BDI_CAP_EXEC_MAP)
977                          ) {
978                         /* backing file is not executable, try to copy */
979                         capabilities &= ~BDI_CAP_MAP_DIRECT;
980                 }
981         }
982         else {
983                 /* anonymous mappings are always memory backed and can be
984                  * privately mapped
985                  */
986                 capabilities = BDI_CAP_MAP_COPY;
987
988                 /* handle PROT_EXEC implication by PROT_READ */
989                 if ((prot & PROT_READ) &&
990                     (current->personality & READ_IMPLIES_EXEC))
991                         prot |= PROT_EXEC;
992         }
993
994         /* allow the security API to have its say */
995         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
996         if (ret < 0)
997                 return ret;
998
999         /* looks okay */
1000         *_capabilities = capabilities;
1001         return 0;
1002 }
1003
1004 /*
1005  * we've determined that we can make the mapping, now translate what we
1006  * now know into VMA flags
1007  */
1008 static unsigned long determine_vm_flags(struct file *file,
1009                                         unsigned long prot,
1010                                         unsigned long flags,
1011                                         unsigned long capabilities)
1012 {
1013         unsigned long vm_flags;
1014
1015         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1016         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1017         /* vm_flags |= mm->def_flags; */
1018
1019         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1020                 /* attempt to share read-only copies of mapped file chunks */
1021                 if (file && !(prot & PROT_WRITE))
1022                         vm_flags |= VM_MAYSHARE;
1023         }
1024         else {
1025                 /* overlay a shareable mapping on the backing device or inode
1026                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1027                  * romfs/cramfs */
1028                 if (flags & MAP_SHARED)
1029                         vm_flags |= VM_MAYSHARE | VM_SHARED;
1030                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1031                         vm_flags |= VM_MAYSHARE;
1032         }
1033
1034         /* refuse to let anyone share private mappings with this process if
1035          * it's being traced - otherwise breakpoints set in it may interfere
1036          * with another untraced process
1037          */
1038         if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1039                 vm_flags &= ~VM_MAYSHARE;
1040
1041         return vm_flags;
1042 }
1043
1044 /*
1045  * set up a shared mapping on a file (the driver or filesystem provides and
1046  * pins the storage)
1047  */
1048 static int do_mmap_shared_file(struct vm_area_struct *vma)
1049 {
1050         int ret;
1051
1052         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1053         if (ret == 0) {
1054                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1055                 return ret;
1056         }
1057         if (ret != -ENOSYS)
1058                 return ret;
1059
1060         /* getting an ENOSYS error indicates that direct mmap isn't
1061          * possible (as opposed to tried but failed) so we'll fall
1062          * through to making a private copy of the data and mapping
1063          * that if we can */
1064         return -ENODEV;
1065 }
1066
1067 /*
1068  * set up a private mapping or an anonymous shared mapping
1069  */
1070 static int do_mmap_private(struct vm_area_struct *vma,
1071                            struct vm_region *region,
1072                            unsigned long len)
1073 {
1074         struct page *pages;
1075         unsigned long total, point, n, rlen;
1076         void *base;
1077         int ret, order;
1078
1079         /* invoke the file's mapping function so that it can keep track of
1080          * shared mappings on devices or memory
1081          * - VM_MAYSHARE will be set if it may attempt to share
1082          */
1083         if (vma->vm_file) {
1084                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1085                 if (ret == 0) {
1086                         /* shouldn't return success if we're not sharing */
1087                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1088                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1089                         return ret;
1090                 }
1091                 if (ret != -ENOSYS)
1092                         return ret;
1093
1094                 /* getting an ENOSYS error indicates that direct mmap isn't
1095                  * possible (as opposed to tried but failed) so we'll try to
1096                  * make a private copy of the data and map that instead */
1097         }
1098
1099         rlen = PAGE_ALIGN(len);
1100
1101         /* allocate some memory to hold the mapping
1102          * - note that this may not return a page-aligned address if the object
1103          *   we're allocating is smaller than a page
1104          */
1105         order = get_order(rlen);
1106         kdebug("alloc order %d for %lx", order, len);
1107
1108         pages = alloc_pages(GFP_KERNEL, order);
1109         if (!pages)
1110                 goto enomem;
1111
1112         total = 1 << order;
1113         atomic_long_add(total, &mmap_pages_allocated);
1114
1115         point = rlen >> PAGE_SHIFT;
1116
1117         /* we allocated a power-of-2 sized page set, so we may want to trim off
1118          * the excess */
1119         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1120                 while (total > point) {
1121                         order = ilog2(total - point);
1122                         n = 1 << order;
1123                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1124                         atomic_long_sub(n, &mmap_pages_allocated);
1125                         total -= n;
1126                         set_page_refcounted(pages + total);
1127                         __free_pages(pages + total, order);
1128                 }
1129         }
1130
1131         for (point = 1; point < total; point++)
1132                 set_page_refcounted(&pages[point]);
1133
1134         base = page_address(pages);
1135         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1136         region->vm_start = (unsigned long) base;
1137         region->vm_end   = region->vm_start + rlen;
1138         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1139
1140         vma->vm_start = region->vm_start;
1141         vma->vm_end   = region->vm_start + len;
1142
1143         if (vma->vm_file) {
1144                 /* read the contents of a file into the copy */
1145                 mm_segment_t old_fs;
1146                 loff_t fpos;
1147
1148                 fpos = vma->vm_pgoff;
1149                 fpos <<= PAGE_SHIFT;
1150
1151                 old_fs = get_fs();
1152                 set_fs(KERNEL_DS);
1153                 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1154                 set_fs(old_fs);
1155
1156                 if (ret < 0)
1157                         goto error_free;
1158
1159                 /* clear the last little bit */
1160                 if (ret < rlen)
1161                         memset(base + ret, 0, rlen - ret);
1162
1163         } else {
1164                 /* if it's an anonymous mapping, then just clear it */
1165                 memset(base, 0, rlen);
1166         }
1167
1168         return 0;
1169
1170 error_free:
1171         free_page_series(region->vm_start, region->vm_end);
1172         region->vm_start = vma->vm_start = 0;
1173         region->vm_end   = vma->vm_end = 0;
1174         region->vm_top   = 0;
1175         return ret;
1176
1177 enomem:
1178         printk("Allocation of length %lu from process %d (%s) failed\n",
1179                len, current->pid, current->comm);
1180         show_free_areas();
1181         return -ENOMEM;
1182 }
1183
1184 /*
1185  * handle mapping creation for uClinux
1186  */
1187 unsigned long do_mmap_pgoff(struct file *file,
1188                             unsigned long addr,
1189                             unsigned long len,
1190                             unsigned long prot,
1191                             unsigned long flags,
1192                             unsigned long pgoff)
1193 {
1194         struct vm_area_struct *vma;
1195         struct vm_region *region;
1196         struct rb_node *rb;
1197         unsigned long capabilities, vm_flags, result;
1198         int ret;
1199
1200         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1201
1202         if (!(flags & MAP_FIXED))
1203                 addr = round_hint_to_min(addr);
1204
1205         /* decide whether we should attempt the mapping, and if so what sort of
1206          * mapping */
1207         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1208                                     &capabilities);
1209         if (ret < 0) {
1210                 kleave(" = %d [val]", ret);
1211                 return ret;
1212         }
1213
1214         /* we've determined that we can make the mapping, now translate what we
1215          * now know into VMA flags */
1216         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1217
1218         /* we're going to need to record the mapping */
1219         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1220         if (!region)
1221                 goto error_getting_region;
1222
1223         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1224         if (!vma)
1225                 goto error_getting_vma;
1226
1227         atomic_set(&region->vm_usage, 1);
1228         region->vm_flags = vm_flags;
1229         region->vm_pgoff = pgoff;
1230
1231         INIT_LIST_HEAD(&vma->anon_vma_node);
1232         vma->vm_flags = vm_flags;
1233         vma->vm_pgoff = pgoff;
1234
1235         if (file) {
1236                 region->vm_file = file;
1237                 get_file(file);
1238                 vma->vm_file = file;
1239                 get_file(file);
1240                 if (vm_flags & VM_EXECUTABLE) {
1241                         added_exe_file_vma(current->mm);
1242                         vma->vm_mm = current->mm;
1243                 }
1244         }
1245
1246         down_write(&nommu_region_sem);
1247
1248         /* if we want to share, we need to check for regions created by other
1249          * mmap() calls that overlap with our proposed mapping
1250          * - we can only share with a superset match on most regular files
1251          * - shared mappings on character devices and memory backed files are
1252          *   permitted to overlap inexactly as far as we are concerned for in
1253          *   these cases, sharing is handled in the driver or filesystem rather
1254          *   than here
1255          */
1256         if (vm_flags & VM_MAYSHARE) {
1257                 struct vm_region *pregion;
1258                 unsigned long pglen, rpglen, pgend, rpgend, start;
1259
1260                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1261                 pgend = pgoff + pglen;
1262
1263                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1264                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1265
1266                         if (!(pregion->vm_flags & VM_MAYSHARE))
1267                                 continue;
1268
1269                         /* search for overlapping mappings on the same file */
1270                         if (pregion->vm_file->f_path.dentry->d_inode !=
1271                             file->f_path.dentry->d_inode)
1272                                 continue;
1273
1274                         if (pregion->vm_pgoff >= pgend)
1275                                 continue;
1276
1277                         rpglen = pregion->vm_end - pregion->vm_start;
1278                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1279                         rpgend = pregion->vm_pgoff + rpglen;
1280                         if (pgoff >= rpgend)
1281                                 continue;
1282
1283                         /* handle inexactly overlapping matches between
1284                          * mappings */
1285                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1286                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1287                                 /* new mapping is not a subset of the region */
1288                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1289                                         goto sharing_violation;
1290                                 continue;
1291                         }
1292
1293                         /* we've found a region we can share */
1294                         atomic_inc(&pregion->vm_usage);
1295                         vma->vm_region = pregion;
1296                         start = pregion->vm_start;
1297                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1298                         vma->vm_start = start;
1299                         vma->vm_end = start + len;
1300
1301                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1302                                 kdebug("share copy");
1303                                 vma->vm_flags |= VM_MAPPED_COPY;
1304                         } else {
1305                                 kdebug("share mmap");
1306                                 ret = do_mmap_shared_file(vma);
1307                                 if (ret < 0) {
1308                                         vma->vm_region = NULL;
1309                                         vma->vm_start = 0;
1310                                         vma->vm_end = 0;
1311                                         atomic_dec(&pregion->vm_usage);
1312                                         pregion = NULL;
1313                                         goto error_just_free;
1314                                 }
1315                         }
1316                         fput(region->vm_file);
1317                         kmem_cache_free(vm_region_jar, region);
1318                         region = pregion;
1319                         result = start;
1320                         goto share;
1321                 }
1322
1323                 /* obtain the address at which to make a shared mapping
1324                  * - this is the hook for quasi-memory character devices to
1325                  *   tell us the location of a shared mapping
1326                  */
1327                 if (file && file->f_op->get_unmapped_area) {
1328                         addr = file->f_op->get_unmapped_area(file, addr, len,
1329                                                              pgoff, flags);
1330                         if (IS_ERR((void *) addr)) {
1331                                 ret = addr;
1332                                 if (ret != (unsigned long) -ENOSYS)
1333                                         goto error_just_free;
1334
1335                                 /* the driver refused to tell us where to site
1336                                  * the mapping so we'll have to attempt to copy
1337                                  * it */
1338                                 ret = (unsigned long) -ENODEV;
1339                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1340                                         goto error_just_free;
1341
1342                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1343                         } else {
1344                                 vma->vm_start = region->vm_start = addr;
1345                                 vma->vm_end = region->vm_end = addr + len;
1346                         }
1347                 }
1348         }
1349
1350         vma->vm_region = region;
1351         add_nommu_region(region);
1352
1353         /* set up the mapping */
1354         if (file && vma->vm_flags & VM_SHARED)
1355                 ret = do_mmap_shared_file(vma);
1356         else
1357                 ret = do_mmap_private(vma, region, len);
1358         if (ret < 0)
1359                 goto error_put_region;
1360
1361         /* okay... we have a mapping; now we have to register it */
1362         result = vma->vm_start;
1363
1364         current->mm->total_vm += len >> PAGE_SHIFT;
1365
1366 share:
1367         add_vma_to_mm(current->mm, vma);
1368
1369         up_write(&nommu_region_sem);
1370
1371         if (prot & PROT_EXEC)
1372                 flush_icache_range(result, result + len);
1373
1374         kleave(" = %lx", result);
1375         return result;
1376
1377 error_put_region:
1378         __put_nommu_region(region);
1379         if (vma) {
1380                 if (vma->vm_file) {
1381                         fput(vma->vm_file);
1382                         if (vma->vm_flags & VM_EXECUTABLE)
1383                                 removed_exe_file_vma(vma->vm_mm);
1384                 }
1385                 kmem_cache_free(vm_area_cachep, vma);
1386         }
1387         kleave(" = %d [pr]", ret);
1388         return ret;
1389
1390 error_just_free:
1391         up_write(&nommu_region_sem);
1392 error:
1393         fput(region->vm_file);
1394         kmem_cache_free(vm_region_jar, region);
1395         fput(vma->vm_file);
1396         if (vma->vm_flags & VM_EXECUTABLE)
1397                 removed_exe_file_vma(vma->vm_mm);
1398         kmem_cache_free(vm_area_cachep, vma);
1399         kleave(" = %d", ret);
1400         return ret;
1401
1402 sharing_violation:
1403         up_write(&nommu_region_sem);
1404         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1405         ret = -EINVAL;
1406         goto error;
1407
1408 error_getting_vma:
1409         kmem_cache_free(vm_region_jar, region);
1410         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1411                " from process %d failed\n",
1412                len, current->pid);
1413         show_free_areas();
1414         return -ENOMEM;
1415
1416 error_getting_region:
1417         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1418                " from process %d failed\n",
1419                len, current->pid);
1420         show_free_areas();
1421         return -ENOMEM;
1422 }
1423 EXPORT_SYMBOL(do_mmap_pgoff);
1424
1425 /*
1426  * split a vma into two pieces at address 'addr', a new vma is allocated either
1427  * for the first part or the tail.
1428  */
1429 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1430               unsigned long addr, int new_below)
1431 {
1432         struct vm_area_struct *new;
1433         struct vm_region *region;
1434         unsigned long npages;
1435
1436         kenter("");
1437
1438         /* we're only permitted to split anonymous regions that have a single
1439          * owner */
1440         if (vma->vm_file ||
1441             atomic_read(&vma->vm_region->vm_usage) != 1)
1442                 return -ENOMEM;
1443
1444         if (mm->map_count >= sysctl_max_map_count)
1445                 return -ENOMEM;
1446
1447         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1448         if (!region)
1449                 return -ENOMEM;
1450
1451         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1452         if (!new) {
1453                 kmem_cache_free(vm_region_jar, region);
1454                 return -ENOMEM;
1455         }
1456
1457         /* most fields are the same, copy all, and then fixup */
1458         *new = *vma;
1459         *region = *vma->vm_region;
1460         new->vm_region = region;
1461
1462         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1463
1464         if (new_below) {
1465                 region->vm_top = region->vm_end = new->vm_end = addr;
1466         } else {
1467                 region->vm_start = new->vm_start = addr;
1468                 region->vm_pgoff = new->vm_pgoff += npages;
1469         }
1470
1471         if (new->vm_ops && new->vm_ops->open)
1472                 new->vm_ops->open(new);
1473
1474         delete_vma_from_mm(vma);
1475         down_write(&nommu_region_sem);
1476         delete_nommu_region(vma->vm_region);
1477         if (new_below) {
1478                 vma->vm_region->vm_start = vma->vm_start = addr;
1479                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1480         } else {
1481                 vma->vm_region->vm_end = vma->vm_end = addr;
1482                 vma->vm_region->vm_top = addr;
1483         }
1484         add_nommu_region(vma->vm_region);
1485         add_nommu_region(new->vm_region);
1486         up_write(&nommu_region_sem);
1487         add_vma_to_mm(mm, vma);
1488         add_vma_to_mm(mm, new);
1489         return 0;
1490 }
1491
1492 /*
1493  * shrink a VMA by removing the specified chunk from either the beginning or
1494  * the end
1495  */
1496 static int shrink_vma(struct mm_struct *mm,
1497                       struct vm_area_struct *vma,
1498                       unsigned long from, unsigned long to)
1499 {
1500         struct vm_region *region;
1501
1502         kenter("");
1503
1504         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1505          * and list */
1506         delete_vma_from_mm(vma);
1507         if (from > vma->vm_start)
1508                 vma->vm_end = from;
1509         else
1510                 vma->vm_start = to;
1511         add_vma_to_mm(mm, vma);
1512
1513         /* cut the backing region down to size */
1514         region = vma->vm_region;
1515         BUG_ON(atomic_read(&region->vm_usage) != 1);
1516
1517         down_write(&nommu_region_sem);
1518         delete_nommu_region(region);
1519         if (from > region->vm_start) {
1520                 to = region->vm_top;
1521                 region->vm_top = region->vm_end = from;
1522         } else {
1523                 region->vm_start = to;
1524         }
1525         add_nommu_region(region);
1526         up_write(&nommu_region_sem);
1527
1528         free_page_series(from, to);
1529         return 0;
1530 }
1531
1532 /*
1533  * release a mapping
1534  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1535  *   VMA, though it need not cover the whole VMA
1536  */
1537 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1538 {
1539         struct vm_area_struct *vma;
1540         struct rb_node *rb;
1541         unsigned long end = start + len;
1542         int ret;
1543
1544         kenter(",%lx,%zx", start, len);
1545
1546         if (len == 0)
1547                 return -EINVAL;
1548
1549         /* find the first potentially overlapping VMA */
1550         vma = find_vma(mm, start);
1551         if (!vma) {
1552                 static int limit = 0;
1553                 if (limit < 5) {
1554                         printk(KERN_WARNING
1555                                "munmap of memory not mmapped by process %d"
1556                                " (%s): 0x%lx-0x%lx\n",
1557                                current->pid, current->comm,
1558                                start, start + len - 1);
1559                         limit++;
1560                 }
1561                 return -EINVAL;
1562         }
1563
1564         /* we're allowed to split an anonymous VMA but not a file-backed one */
1565         if (vma->vm_file) {
1566                 do {
1567                         if (start > vma->vm_start) {
1568                                 kleave(" = -EINVAL [miss]");
1569                                 return -EINVAL;
1570                         }
1571                         if (end == vma->vm_end)
1572                                 goto erase_whole_vma;
1573                         rb = rb_next(&vma->vm_rb);
1574                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1575                 } while (rb);
1576                 kleave(" = -EINVAL [split file]");
1577                 return -EINVAL;
1578         } else {
1579                 /* the chunk must be a subset of the VMA found */
1580                 if (start == vma->vm_start && end == vma->vm_end)
1581                         goto erase_whole_vma;
1582                 if (start < vma->vm_start || end > vma->vm_end) {
1583                         kleave(" = -EINVAL [superset]");
1584                         return -EINVAL;
1585                 }
1586                 if (start & ~PAGE_MASK) {
1587                         kleave(" = -EINVAL [unaligned start]");
1588                         return -EINVAL;
1589                 }
1590                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1591                         kleave(" = -EINVAL [unaligned split]");
1592                         return -EINVAL;
1593                 }
1594                 if (start != vma->vm_start && end != vma->vm_end) {
1595                         ret = split_vma(mm, vma, start, 1);
1596                         if (ret < 0) {
1597                                 kleave(" = %d [split]", ret);
1598                                 return ret;
1599                         }
1600                 }
1601                 return shrink_vma(mm, vma, start, end);
1602         }
1603
1604 erase_whole_vma:
1605         delete_vma_from_mm(vma);
1606         delete_vma(mm, vma);
1607         kleave(" = 0");
1608         return 0;
1609 }
1610 EXPORT_SYMBOL(do_munmap);
1611
1612 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1613 {
1614         int ret;
1615         struct mm_struct *mm = current->mm;
1616
1617         down_write(&mm->mmap_sem);
1618         ret = do_munmap(mm, addr, len);
1619         up_write(&mm->mmap_sem);
1620         return ret;
1621 }
1622
1623 /*
1624  * release all the mappings made in a process's VM space
1625  */
1626 void exit_mmap(struct mm_struct *mm)
1627 {
1628         struct vm_area_struct *vma;
1629
1630         if (!mm)
1631                 return;
1632
1633         kenter("");
1634
1635         mm->total_vm = 0;
1636
1637         while ((vma = mm->mmap)) {
1638                 mm->mmap = vma->vm_next;
1639                 delete_vma_from_mm(vma);
1640                 delete_vma(mm, vma);
1641         }
1642
1643         kleave("");
1644 }
1645
1646 unsigned long do_brk(unsigned long addr, unsigned long len)
1647 {
1648         return -ENOMEM;
1649 }
1650
1651 /*
1652  * expand (or shrink) an existing mapping, potentially moving it at the same
1653  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1654  *
1655  * under NOMMU conditions, we only permit changing a mapping's size, and only
1656  * as long as it stays within the region allocated by do_mmap_private() and the
1657  * block is not shareable
1658  *
1659  * MREMAP_FIXED is not supported under NOMMU conditions
1660  */
1661 unsigned long do_mremap(unsigned long addr,
1662                         unsigned long old_len, unsigned long new_len,
1663                         unsigned long flags, unsigned long new_addr)
1664 {
1665         struct vm_area_struct *vma;
1666
1667         /* insanity checks first */
1668         if (old_len == 0 || new_len == 0)
1669                 return (unsigned long) -EINVAL;
1670
1671         if (addr & ~PAGE_MASK)
1672                 return -EINVAL;
1673
1674         if (flags & MREMAP_FIXED && new_addr != addr)
1675                 return (unsigned long) -EINVAL;
1676
1677         vma = find_vma_exact(current->mm, addr, old_len);
1678         if (!vma)
1679                 return (unsigned long) -EINVAL;
1680
1681         if (vma->vm_end != vma->vm_start + old_len)
1682                 return (unsigned long) -EFAULT;
1683
1684         if (vma->vm_flags & VM_MAYSHARE)
1685                 return (unsigned long) -EPERM;
1686
1687         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1688                 return (unsigned long) -ENOMEM;
1689
1690         /* all checks complete - do it */
1691         vma->vm_end = vma->vm_start + new_len;
1692         return vma->vm_start;
1693 }
1694 EXPORT_SYMBOL(do_mremap);
1695
1696 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1697                 unsigned long, new_len, unsigned long, flags,
1698                 unsigned long, new_addr)
1699 {
1700         unsigned long ret;
1701
1702         down_write(&current->mm->mmap_sem);
1703         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1704         up_write(&current->mm->mmap_sem);
1705         return ret;
1706 }
1707
1708 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1709                         unsigned int foll_flags)
1710 {
1711         return NULL;
1712 }
1713
1714 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1715                 unsigned long to, unsigned long size, pgprot_t prot)
1716 {
1717         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1718         return 0;
1719 }
1720 EXPORT_SYMBOL(remap_pfn_range);
1721
1722 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1723                         unsigned long pgoff)
1724 {
1725         unsigned int size = vma->vm_end - vma->vm_start;
1726
1727         if (!(vma->vm_flags & VM_USERMAP))
1728                 return -EINVAL;
1729
1730         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1731         vma->vm_end = vma->vm_start + size;
1732
1733         return 0;
1734 }
1735 EXPORT_SYMBOL(remap_vmalloc_range);
1736
1737 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1738 {
1739 }
1740
1741 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1742         unsigned long len, unsigned long pgoff, unsigned long flags)
1743 {
1744         return -ENOMEM;
1745 }
1746
1747 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1748 {
1749 }
1750
1751 void unmap_mapping_range(struct address_space *mapping,
1752                          loff_t const holebegin, loff_t const holelen,
1753                          int even_cows)
1754 {
1755 }
1756 EXPORT_SYMBOL(unmap_mapping_range);
1757
1758 /*
1759  * ask for an unmapped area at which to create a mapping on a file
1760  */
1761 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1762                                 unsigned long len, unsigned long pgoff,
1763                                 unsigned long flags)
1764 {
1765         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1766                                   unsigned long, unsigned long);
1767
1768         get_area = current->mm->get_unmapped_area;
1769         if (file && file->f_op && file->f_op->get_unmapped_area)
1770                 get_area = file->f_op->get_unmapped_area;
1771
1772         if (!get_area)
1773                 return -ENOSYS;
1774
1775         return get_area(file, addr, len, pgoff, flags);
1776 }
1777 EXPORT_SYMBOL(get_unmapped_area);
1778
1779 /*
1780  * Check that a process has enough memory to allocate a new virtual
1781  * mapping. 0 means there is enough memory for the allocation to
1782  * succeed and -ENOMEM implies there is not.
1783  *
1784  * We currently support three overcommit policies, which are set via the
1785  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1786  *
1787  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1788  * Additional code 2002 Jul 20 by Robert Love.
1789  *
1790  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1791  *
1792  * Note this is a helper function intended to be used by LSMs which
1793  * wish to use this logic.
1794  */
1795 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1796 {
1797         unsigned long free, allowed;
1798
1799         vm_acct_memory(pages);
1800
1801         /*
1802          * Sometimes we want to use more memory than we have
1803          */
1804         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1805                 return 0;
1806
1807         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1808                 unsigned long n;
1809
1810                 free = global_page_state(NR_FILE_PAGES);
1811                 free += nr_swap_pages;
1812
1813                 /*
1814                  * Any slabs which are created with the
1815                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1816                  * which are reclaimable, under pressure.  The dentry
1817                  * cache and most inode caches should fall into this
1818                  */
1819                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1820
1821                 /*
1822                  * Leave the last 3% for root
1823                  */
1824                 if (!cap_sys_admin)
1825                         free -= free / 32;
1826
1827                 if (free > pages)
1828                         return 0;
1829
1830                 /*
1831                  * nr_free_pages() is very expensive on large systems,
1832                  * only call if we're about to fail.
1833                  */
1834                 n = nr_free_pages();
1835
1836                 /*
1837                  * Leave reserved pages. The pages are not for anonymous pages.
1838                  */
1839                 if (n <= totalreserve_pages)
1840                         goto error;
1841                 else
1842                         n -= totalreserve_pages;
1843
1844                 /*
1845                  * Leave the last 3% for root
1846                  */
1847                 if (!cap_sys_admin)
1848                         n -= n / 32;
1849                 free += n;
1850
1851                 if (free > pages)
1852                         return 0;
1853
1854                 goto error;
1855         }
1856
1857         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1858         /*
1859          * Leave the last 3% for root
1860          */
1861         if (!cap_sys_admin)
1862                 allowed -= allowed / 32;
1863         allowed += total_swap_pages;
1864
1865         /* Don't let a single process grow too big:
1866            leave 3% of the size of this process for other processes */
1867         if (mm)
1868                 allowed -= mm->total_vm / 32;
1869
1870         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1871                 return 0;
1872
1873 error:
1874         vm_unacct_memory(pages);
1875
1876         return -ENOMEM;
1877 }
1878
1879 int in_gate_area_no_task(unsigned long addr)
1880 {
1881         return 0;
1882 }
1883
1884 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1885 {
1886         BUG();
1887         return 0;
1888 }
1889 EXPORT_SYMBOL(filemap_fault);
1890
1891 /*
1892  * Access another process' address space.
1893  * - source/target buffer must be kernel space
1894  */
1895 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1896 {
1897         struct vm_area_struct *vma;
1898         struct mm_struct *mm;
1899
1900         if (addr + len < addr)
1901                 return 0;
1902
1903         mm = get_task_mm(tsk);
1904         if (!mm)
1905                 return 0;
1906
1907         down_read(&mm->mmap_sem);
1908
1909         /* the access must start within one of the target process's mappings */
1910         vma = find_vma(mm, addr);
1911         if (vma) {
1912                 /* don't overrun this mapping */
1913                 if (addr + len >= vma->vm_end)
1914                         len = vma->vm_end - addr;
1915
1916                 /* only read or write mappings where it is permitted */
1917                 if (write && vma->vm_flags & VM_MAYWRITE)
1918                         len -= copy_to_user((void *) addr, buf, len);
1919                 else if (!write && vma->vm_flags & VM_MAYREAD)
1920                         len -= copy_from_user(buf, (void *) addr, len);
1921                 else
1922                         len = 0;
1923         } else {
1924                 len = 0;
1925         }
1926
1927         up_read(&mm->mmap_sem);
1928         mmput(mm);
1929         return len;
1930 }