NOMMU: Lock i_mmap_mutex for access to the VMA prio list
[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) 2007-2010 Paul Mundt <lethal@linux-sh.org>
14  */
15
16 #include <linux/export.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/blkdev.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mount.h>
28 #include <linux/personality.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
31 #include <linux/audit.h>
32
33 #include <asm/uaccess.h>
34 #include <asm/tlb.h>
35 #include <asm/tlbflush.h>
36 #include <asm/mmu_context.h>
37 #include "internal.h"
38
39 #if 0
40 #define kenter(FMT, ...) \
41         printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
42 #define kleave(FMT, ...) \
43         printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
44 #define kdebug(FMT, ...) \
45         printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
46 #else
47 #define kenter(FMT, ...) \
48         no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
49 #define kleave(FMT, ...) \
50         no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
51 #define kdebug(FMT, ...) \
52         no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
53 #endif
54
55 void *high_memory;
56 struct page *mem_map;
57 unsigned long max_mapnr;
58 unsigned long num_physpages;
59 unsigned long highest_memmap_pfn;
60 struct percpu_counter vm_committed_as;
61 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
62 int sysctl_overcommit_ratio = 50; /* default is 50% */
63 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
64 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
65 int heap_stack_gap = 0;
66
67 atomic_long_t mmap_pages_allocated;
68
69 EXPORT_SYMBOL(mem_map);
70 EXPORT_SYMBOL(num_physpages);
71
72 /* list of mapped, potentially shareable regions */
73 static struct kmem_cache *vm_region_jar;
74 struct rb_root nommu_region_tree = RB_ROOT;
75 DECLARE_RWSEM(nommu_region_sem);
76
77 const struct vm_operations_struct generic_file_vm_ops = {
78 };
79
80 /*
81  * Return the total memory allocated for this pointer, not
82  * just what the caller asked for.
83  *
84  * Doesn't have to be accurate, i.e. may have races.
85  */
86 unsigned int kobjsize(const void *objp)
87 {
88         struct page *page;
89
90         /*
91          * If the object we have should not have ksize performed on it,
92          * return size of 0
93          */
94         if (!objp || !virt_addr_valid(objp))
95                 return 0;
96
97         page = virt_to_head_page(objp);
98
99         /*
100          * If the allocator sets PageSlab, we know the pointer came from
101          * kmalloc().
102          */
103         if (PageSlab(page))
104                 return ksize(objp);
105
106         /*
107          * If it's not a compound page, see if we have a matching VMA
108          * region. This test is intentionally done in reverse order,
109          * so if there's no VMA, we still fall through and hand back
110          * PAGE_SIZE for 0-order pages.
111          */
112         if (!PageCompound(page)) {
113                 struct vm_area_struct *vma;
114
115                 vma = find_vma(current->mm, (unsigned long)objp);
116                 if (vma)
117                         return vma->vm_end - vma->vm_start;
118         }
119
120         /*
121          * The ksize() function is only guaranteed to work for pointers
122          * returned by kmalloc(). So handle arbitrary pointers here.
123          */
124         return PAGE_SIZE << compound_order(page);
125 }
126
127 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
128                      unsigned long start, int nr_pages, unsigned int foll_flags,
129                      struct page **pages, struct vm_area_struct **vmas,
130                      int *retry)
131 {
132         struct vm_area_struct *vma;
133         unsigned long vm_flags;
134         int i;
135
136         /* calculate required read or write permissions.
137          * If FOLL_FORCE is set, we only require the "MAY" flags.
138          */
139         vm_flags  = (foll_flags & FOLL_WRITE) ?
140                         (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
141         vm_flags &= (foll_flags & FOLL_FORCE) ?
142                         (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
143
144         for (i = 0; i < nr_pages; i++) {
145                 vma = find_vma(mm, start);
146                 if (!vma)
147                         goto finish_or_fault;
148
149                 /* protect what we can, including chardevs */
150                 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
151                     !(vm_flags & vma->vm_flags))
152                         goto finish_or_fault;
153
154                 if (pages) {
155                         pages[i] = virt_to_page(start);
156                         if (pages[i])
157                                 page_cache_get(pages[i]);
158                 }
159                 if (vmas)
160                         vmas[i] = vma;
161                 start = (start + PAGE_SIZE) & PAGE_MASK;
162         }
163
164         return i;
165
166 finish_or_fault:
167         return i ? : -EFAULT;
168 }
169
170 /*
171  * get a list of pages in an address range belonging to the specified process
172  * and indicate the VMA that covers each page
173  * - this is potentially dodgy as we may end incrementing the page count of a
174  *   slab page or a secondary page from a compound page
175  * - don't permit access to VMAs that don't support it, such as I/O mappings
176  */
177 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
178         unsigned long start, int nr_pages, int write, int force,
179         struct page **pages, struct vm_area_struct **vmas)
180 {
181         int flags = 0;
182
183         if (write)
184                 flags |= FOLL_WRITE;
185         if (force)
186                 flags |= FOLL_FORCE;
187
188         return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
189                                 NULL);
190 }
191 EXPORT_SYMBOL(get_user_pages);
192
193 /**
194  * follow_pfn - look up PFN at a user virtual address
195  * @vma: memory mapping
196  * @address: user virtual address
197  * @pfn: location to store found PFN
198  *
199  * Only IO mappings and raw PFN mappings are allowed.
200  *
201  * Returns zero and the pfn at @pfn on success, -ve otherwise.
202  */
203 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
204         unsigned long *pfn)
205 {
206         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
207                 return -EINVAL;
208
209         *pfn = address >> PAGE_SHIFT;
210         return 0;
211 }
212 EXPORT_SYMBOL(follow_pfn);
213
214 DEFINE_RWLOCK(vmlist_lock);
215 struct vm_struct *vmlist;
216
217 void vfree(const void *addr)
218 {
219         kfree(addr);
220 }
221 EXPORT_SYMBOL(vfree);
222
223 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
224 {
225         /*
226          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
227          * returns only a logical address.
228          */
229         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
230 }
231 EXPORT_SYMBOL(__vmalloc);
232
233 void *vmalloc_user(unsigned long size)
234 {
235         void *ret;
236
237         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
238                         PAGE_KERNEL);
239         if (ret) {
240                 struct vm_area_struct *vma;
241
242                 down_write(&current->mm->mmap_sem);
243                 vma = find_vma(current->mm, (unsigned long)ret);
244                 if (vma)
245                         vma->vm_flags |= VM_USERMAP;
246                 up_write(&current->mm->mmap_sem);
247         }
248
249         return ret;
250 }
251 EXPORT_SYMBOL(vmalloc_user);
252
253 struct page *vmalloc_to_page(const void *addr)
254 {
255         return virt_to_page(addr);
256 }
257 EXPORT_SYMBOL(vmalloc_to_page);
258
259 unsigned long vmalloc_to_pfn(const void *addr)
260 {
261         return page_to_pfn(virt_to_page(addr));
262 }
263 EXPORT_SYMBOL(vmalloc_to_pfn);
264
265 long vread(char *buf, char *addr, unsigned long count)
266 {
267         memcpy(buf, addr, count);
268         return count;
269 }
270
271 long vwrite(char *buf, char *addr, unsigned long count)
272 {
273         /* Don't allow overflow */
274         if ((unsigned long) addr + count < count)
275                 count = -(unsigned long) addr;
276
277         memcpy(addr, buf, count);
278         return(count);
279 }
280
281 /*
282  *      vmalloc  -  allocate virtually continguos memory
283  *
284  *      @size:          allocation size
285  *
286  *      Allocate enough pages to cover @size from the page level
287  *      allocator and map them into continguos kernel virtual space.
288  *
289  *      For tight control over page level allocator and protection flags
290  *      use __vmalloc() instead.
291  */
292 void *vmalloc(unsigned long size)
293 {
294        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
295 }
296 EXPORT_SYMBOL(vmalloc);
297
298 /*
299  *      vzalloc - allocate virtually continguos memory with zero fill
300  *
301  *      @size:          allocation size
302  *
303  *      Allocate enough pages to cover @size from the page level
304  *      allocator and map them into continguos kernel virtual space.
305  *      The memory allocated is set to zero.
306  *
307  *      For tight control over page level allocator and protection flags
308  *      use __vmalloc() instead.
309  */
310 void *vzalloc(unsigned long size)
311 {
312         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
313                         PAGE_KERNEL);
314 }
315 EXPORT_SYMBOL(vzalloc);
316
317 /**
318  * vmalloc_node - allocate memory on a specific node
319  * @size:       allocation size
320  * @node:       numa node
321  *
322  * Allocate enough pages to cover @size from the page level
323  * allocator and map them into contiguous kernel virtual space.
324  *
325  * For tight control over page level allocator and protection flags
326  * use __vmalloc() instead.
327  */
328 void *vmalloc_node(unsigned long size, int node)
329 {
330         return vmalloc(size);
331 }
332 EXPORT_SYMBOL(vmalloc_node);
333
334 /**
335  * vzalloc_node - allocate memory on a specific node with zero fill
336  * @size:       allocation size
337  * @node:       numa node
338  *
339  * Allocate enough pages to cover @size from the page level
340  * allocator and map them into contiguous kernel virtual space.
341  * The memory allocated is set to zero.
342  *
343  * For tight control over page level allocator and protection flags
344  * use __vmalloc() instead.
345  */
346 void *vzalloc_node(unsigned long size, int node)
347 {
348         return vzalloc(size);
349 }
350 EXPORT_SYMBOL(vzalloc_node);
351
352 #ifndef PAGE_KERNEL_EXEC
353 # define PAGE_KERNEL_EXEC PAGE_KERNEL
354 #endif
355
356 /**
357  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
358  *      @size:          allocation size
359  *
360  *      Kernel-internal function to allocate enough pages to cover @size
361  *      the page level allocator and map them into contiguous and
362  *      executable kernel virtual space.
363  *
364  *      For tight control over page level allocator and protection flags
365  *      use __vmalloc() instead.
366  */
367
368 void *vmalloc_exec(unsigned long size)
369 {
370         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
371 }
372
373 /**
374  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
375  *      @size:          allocation size
376  *
377  *      Allocate enough 32bit PA addressable pages to cover @size from the
378  *      page level allocator and map them into continguos kernel virtual space.
379  */
380 void *vmalloc_32(unsigned long size)
381 {
382         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
383 }
384 EXPORT_SYMBOL(vmalloc_32);
385
386 /**
387  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
388  *      @size:          allocation size
389  *
390  * The resulting memory area is 32bit addressable and zeroed so it can be
391  * mapped to userspace without leaking data.
392  *
393  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
394  * remap_vmalloc_range() are permissible.
395  */
396 void *vmalloc_32_user(unsigned long size)
397 {
398         /*
399          * We'll have to sort out the ZONE_DMA bits for 64-bit,
400          * but for now this can simply use vmalloc_user() directly.
401          */
402         return vmalloc_user(size);
403 }
404 EXPORT_SYMBOL(vmalloc_32_user);
405
406 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
407 {
408         BUG();
409         return NULL;
410 }
411 EXPORT_SYMBOL(vmap);
412
413 void vunmap(const void *addr)
414 {
415         BUG();
416 }
417 EXPORT_SYMBOL(vunmap);
418
419 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
420 {
421         BUG();
422         return NULL;
423 }
424 EXPORT_SYMBOL(vm_map_ram);
425
426 void vm_unmap_ram(const void *mem, unsigned int count)
427 {
428         BUG();
429 }
430 EXPORT_SYMBOL(vm_unmap_ram);
431
432 void vm_unmap_aliases(void)
433 {
434 }
435 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
436
437 /*
438  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
439  * have one.
440  */
441 void  __attribute__((weak)) vmalloc_sync_all(void)
442 {
443 }
444
445 /**
446  *      alloc_vm_area - allocate a range of kernel address space
447  *      @size:          size of the area
448  *
449  *      Returns:        NULL on failure, vm_struct on success
450  *
451  *      This function reserves a range of kernel address space, and
452  *      allocates pagetables to map that range.  No actual mappings
453  *      are created.  If the kernel address space is not shared
454  *      between processes, it syncs the pagetable across all
455  *      processes.
456  */
457 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
458 {
459         BUG();
460         return NULL;
461 }
462 EXPORT_SYMBOL_GPL(alloc_vm_area);
463
464 void free_vm_area(struct vm_struct *area)
465 {
466         BUG();
467 }
468 EXPORT_SYMBOL_GPL(free_vm_area);
469
470 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
471                    struct page *page)
472 {
473         return -EINVAL;
474 }
475 EXPORT_SYMBOL(vm_insert_page);
476
477 /*
478  *  sys_brk() for the most part doesn't need the global kernel
479  *  lock, except when an application is doing something nasty
480  *  like trying to un-brk an area that has already been mapped
481  *  to a regular file.  in this case, the unmapping will need
482  *  to invoke file system routines that need the global lock.
483  */
484 SYSCALL_DEFINE1(brk, unsigned long, brk)
485 {
486         struct mm_struct *mm = current->mm;
487
488         if (brk < mm->start_brk || brk > mm->context.end_brk)
489                 return mm->brk;
490
491         if (mm->brk == brk)
492                 return mm->brk;
493
494         /*
495          * Always allow shrinking brk
496          */
497         if (brk <= mm->brk) {
498                 mm->brk = brk;
499                 return brk;
500         }
501
502         /*
503          * Ok, looks good - let it rip.
504          */
505         flush_icache_range(mm->brk, brk);
506         return mm->brk = brk;
507 }
508
509 /*
510  * initialise the VMA and region record slabs
511  */
512 void __init mmap_init(void)
513 {
514         int ret;
515
516         ret = percpu_counter_init(&vm_committed_as, 0);
517         VM_BUG_ON(ret);
518         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
519 }
520
521 /*
522  * validate the region tree
523  * - the caller must hold the region lock
524  */
525 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
526 static noinline void validate_nommu_regions(void)
527 {
528         struct vm_region *region, *last;
529         struct rb_node *p, *lastp;
530
531         lastp = rb_first(&nommu_region_tree);
532         if (!lastp)
533                 return;
534
535         last = rb_entry(lastp, struct vm_region, vm_rb);
536         BUG_ON(unlikely(last->vm_end <= last->vm_start));
537         BUG_ON(unlikely(last->vm_top < last->vm_end));
538
539         while ((p = rb_next(lastp))) {
540                 region = rb_entry(p, struct vm_region, vm_rb);
541                 last = rb_entry(lastp, struct vm_region, vm_rb);
542
543                 BUG_ON(unlikely(region->vm_end <= region->vm_start));
544                 BUG_ON(unlikely(region->vm_top < region->vm_end));
545                 BUG_ON(unlikely(region->vm_start < last->vm_top));
546
547                 lastp = p;
548         }
549 }
550 #else
551 static void validate_nommu_regions(void)
552 {
553 }
554 #endif
555
556 /*
557  * add a region into the global tree
558  */
559 static void add_nommu_region(struct vm_region *region)
560 {
561         struct vm_region *pregion;
562         struct rb_node **p, *parent;
563
564         validate_nommu_regions();
565
566         parent = NULL;
567         p = &nommu_region_tree.rb_node;
568         while (*p) {
569                 parent = *p;
570                 pregion = rb_entry(parent, struct vm_region, vm_rb);
571                 if (region->vm_start < pregion->vm_start)
572                         p = &(*p)->rb_left;
573                 else if (region->vm_start > pregion->vm_start)
574                         p = &(*p)->rb_right;
575                 else if (pregion == region)
576                         return;
577                 else
578                         BUG();
579         }
580
581         rb_link_node(&region->vm_rb, parent, p);
582         rb_insert_color(&region->vm_rb, &nommu_region_tree);
583
584         validate_nommu_regions();
585 }
586
587 /*
588  * delete a region from the global tree
589  */
590 static void delete_nommu_region(struct vm_region *region)
591 {
592         BUG_ON(!nommu_region_tree.rb_node);
593
594         validate_nommu_regions();
595         rb_erase(&region->vm_rb, &nommu_region_tree);
596         validate_nommu_regions();
597 }
598
599 /*
600  * free a contiguous series of pages
601  */
602 static void free_page_series(unsigned long from, unsigned long to)
603 {
604         for (; from < to; from += PAGE_SIZE) {
605                 struct page *page = virt_to_page(from);
606
607                 kdebug("- free %lx", from);
608                 atomic_long_dec(&mmap_pages_allocated);
609                 if (page_count(page) != 1)
610                         kdebug("free page %p: refcount not one: %d",
611                                page, page_count(page));
612                 put_page(page);
613         }
614 }
615
616 /*
617  * release a reference to a region
618  * - the caller must hold the region semaphore for writing, which this releases
619  * - the region may not have been added to the tree yet, in which case vm_top
620  *   will equal vm_start
621  */
622 static void __put_nommu_region(struct vm_region *region)
623         __releases(nommu_region_sem)
624 {
625         kenter("%p{%d}", region, region->vm_usage);
626
627         BUG_ON(!nommu_region_tree.rb_node);
628
629         if (--region->vm_usage == 0) {
630                 if (region->vm_top > region->vm_start)
631                         delete_nommu_region(region);
632                 up_write(&nommu_region_sem);
633
634                 if (region->vm_file)
635                         fput(region->vm_file);
636
637                 /* IO memory and memory shared directly out of the pagecache
638                  * from ramfs/tmpfs mustn't be released here */
639                 if (region->vm_flags & VM_MAPPED_COPY) {
640                         kdebug("free series");
641                         free_page_series(region->vm_start, region->vm_top);
642                 }
643                 kmem_cache_free(vm_region_jar, region);
644         } else {
645                 up_write(&nommu_region_sem);
646         }
647 }
648
649 /*
650  * release a reference to a region
651  */
652 static void put_nommu_region(struct vm_region *region)
653 {
654         down_write(&nommu_region_sem);
655         __put_nommu_region(region);
656 }
657
658 /*
659  * update protection on a vma
660  */
661 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
662 {
663 #ifdef CONFIG_MPU
664         struct mm_struct *mm = vma->vm_mm;
665         long start = vma->vm_start & PAGE_MASK;
666         while (start < vma->vm_end) {
667                 protect_page(mm, start, flags);
668                 start += PAGE_SIZE;
669         }
670         update_protections(mm);
671 #endif
672 }
673
674 /*
675  * add a VMA into a process's mm_struct in the appropriate place in the list
676  * and tree and add to the address space's page tree also if not an anonymous
677  * page
678  * - should be called with mm->mmap_sem held writelocked
679  */
680 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
681 {
682         struct vm_area_struct *pvma, *prev;
683         struct address_space *mapping;
684         struct rb_node **p, *parent, *rb_prev;
685
686         kenter(",%p", vma);
687
688         BUG_ON(!vma->vm_region);
689
690         mm->map_count++;
691         vma->vm_mm = mm;
692
693         protect_vma(vma, vma->vm_flags);
694
695         /* add the VMA to the mapping */
696         if (vma->vm_file) {
697                 mapping = vma->vm_file->f_mapping;
698
699                 mutex_lock(&mapping->i_mmap_mutex);
700                 flush_dcache_mmap_lock(mapping);
701                 vma_prio_tree_insert(vma, &mapping->i_mmap);
702                 flush_dcache_mmap_unlock(mapping);
703                 mutex_unlock(&mapping->i_mmap_mutex);
704         }
705
706         /* add the VMA to the tree */
707         parent = rb_prev = NULL;
708         p = &mm->mm_rb.rb_node;
709         while (*p) {
710                 parent = *p;
711                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
712
713                 /* sort by: start addr, end addr, VMA struct addr in that order
714                  * (the latter is necessary as we may get identical VMAs) */
715                 if (vma->vm_start < pvma->vm_start)
716                         p = &(*p)->rb_left;
717                 else if (vma->vm_start > pvma->vm_start) {
718                         rb_prev = parent;
719                         p = &(*p)->rb_right;
720                 } else if (vma->vm_end < pvma->vm_end)
721                         p = &(*p)->rb_left;
722                 else if (vma->vm_end > pvma->vm_end) {
723                         rb_prev = parent;
724                         p = &(*p)->rb_right;
725                 } else if (vma < pvma)
726                         p = &(*p)->rb_left;
727                 else if (vma > pvma) {
728                         rb_prev = parent;
729                         p = &(*p)->rb_right;
730                 } else
731                         BUG();
732         }
733
734         rb_link_node(&vma->vm_rb, parent, p);
735         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
736
737         /* add VMA to the VMA list also */
738         prev = NULL;
739         if (rb_prev)
740                 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
741
742         __vma_link_list(mm, vma, prev, parent);
743 }
744
745 /*
746  * delete a VMA from its owning mm_struct and address space
747  */
748 static void delete_vma_from_mm(struct vm_area_struct *vma)
749 {
750         struct address_space *mapping;
751         struct mm_struct *mm = vma->vm_mm;
752
753         kenter("%p", vma);
754
755         protect_vma(vma, 0);
756
757         mm->map_count--;
758         if (mm->mmap_cache == vma)
759                 mm->mmap_cache = NULL;
760
761         /* remove the VMA from the mapping */
762         if (vma->vm_file) {
763                 mapping = vma->vm_file->f_mapping;
764
765                 mutex_lock(&mapping->i_mmap_mutex);
766                 flush_dcache_mmap_lock(mapping);
767                 vma_prio_tree_remove(vma, &mapping->i_mmap);
768                 flush_dcache_mmap_unlock(mapping);
769                 mutex_unlock(&mapping->i_mmap_mutex);
770         }
771
772         /* remove from the MM's tree and list */
773         rb_erase(&vma->vm_rb, &mm->mm_rb);
774
775         if (vma->vm_prev)
776                 vma->vm_prev->vm_next = vma->vm_next;
777         else
778                 mm->mmap = vma->vm_next;
779
780         if (vma->vm_next)
781                 vma->vm_next->vm_prev = vma->vm_prev;
782
783         vma->vm_mm = NULL;
784 }
785
786 /*
787  * destroy a VMA record
788  */
789 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
790 {
791         kenter("%p", vma);
792         if (vma->vm_ops && vma->vm_ops->close)
793                 vma->vm_ops->close(vma);
794         if (vma->vm_file) {
795                 fput(vma->vm_file);
796                 if (vma->vm_flags & VM_EXECUTABLE)
797                         removed_exe_file_vma(mm);
798         }
799         put_nommu_region(vma->vm_region);
800         kmem_cache_free(vm_area_cachep, vma);
801 }
802
803 /*
804  * look up the first VMA in which addr resides, NULL if none
805  * - should be called with mm->mmap_sem at least held readlocked
806  */
807 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
808 {
809         struct vm_area_struct *vma;
810
811         /* check the cache first */
812         vma = mm->mmap_cache;
813         if (vma && vma->vm_start <= addr && vma->vm_end > addr)
814                 return vma;
815
816         /* trawl the list (there may be multiple mappings in which addr
817          * resides) */
818         for (vma = mm->mmap; vma; vma = vma->vm_next) {
819                 if (vma->vm_start > addr)
820                         return NULL;
821                 if (vma->vm_end > addr) {
822                         mm->mmap_cache = vma;
823                         return vma;
824                 }
825         }
826
827         return NULL;
828 }
829 EXPORT_SYMBOL(find_vma);
830
831 /*
832  * find a VMA
833  * - we don't extend stack VMAs under NOMMU conditions
834  */
835 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
836 {
837         return find_vma(mm, addr);
838 }
839
840 /*
841  * expand a stack to a given address
842  * - not supported under NOMMU conditions
843  */
844 int expand_stack(struct vm_area_struct *vma, unsigned long address)
845 {
846         return -ENOMEM;
847 }
848
849 /*
850  * look up the first VMA exactly that exactly matches addr
851  * - should be called with mm->mmap_sem at least held readlocked
852  */
853 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
854                                              unsigned long addr,
855                                              unsigned long len)
856 {
857         struct vm_area_struct *vma;
858         unsigned long end = addr + len;
859
860         /* check the cache first */
861         vma = mm->mmap_cache;
862         if (vma && vma->vm_start == addr && vma->vm_end == end)
863                 return vma;
864
865         /* trawl the list (there may be multiple mappings in which addr
866          * resides) */
867         for (vma = mm->mmap; vma; vma = vma->vm_next) {
868                 if (vma->vm_start < addr)
869                         continue;
870                 if (vma->vm_start > addr)
871                         return NULL;
872                 if (vma->vm_end == end) {
873                         mm->mmap_cache = vma;
874                         return vma;
875                 }
876         }
877
878         return NULL;
879 }
880
881 /*
882  * determine whether a mapping should be permitted and, if so, what sort of
883  * mapping we're capable of supporting
884  */
885 static int validate_mmap_request(struct file *file,
886                                  unsigned long addr,
887                                  unsigned long len,
888                                  unsigned long prot,
889                                  unsigned long flags,
890                                  unsigned long pgoff,
891                                  unsigned long *_capabilities)
892 {
893         unsigned long capabilities, rlen;
894         unsigned long reqprot = prot;
895         int ret;
896
897         /* do the simple checks first */
898         if (flags & MAP_FIXED) {
899                 printk(KERN_DEBUG
900                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
901                        current->pid);
902                 return -EINVAL;
903         }
904
905         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
906             (flags & MAP_TYPE) != MAP_SHARED)
907                 return -EINVAL;
908
909         if (!len)
910                 return -EINVAL;
911
912         /* Careful about overflows.. */
913         rlen = PAGE_ALIGN(len);
914         if (!rlen || rlen > TASK_SIZE)
915                 return -ENOMEM;
916
917         /* offset overflow? */
918         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
919                 return -EOVERFLOW;
920
921         if (file) {
922                 /* validate file mapping requests */
923                 struct address_space *mapping;
924
925                 /* files must support mmap */
926                 if (!file->f_op || !file->f_op->mmap)
927                         return -ENODEV;
928
929                 /* work out if what we've got could possibly be shared
930                  * - we support chardevs that provide their own "memory"
931                  * - we support files/blockdevs that are memory backed
932                  */
933                 mapping = file->f_mapping;
934                 if (!mapping)
935                         mapping = file->f_path.dentry->d_inode->i_mapping;
936
937                 capabilities = 0;
938                 if (mapping && mapping->backing_dev_info)
939                         capabilities = mapping->backing_dev_info->capabilities;
940
941                 if (!capabilities) {
942                         /* no explicit capabilities set, so assume some
943                          * defaults */
944                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
945                         case S_IFREG:
946                         case S_IFBLK:
947                                 capabilities = BDI_CAP_MAP_COPY;
948                                 break;
949
950                         case S_IFCHR:
951                                 capabilities =
952                                         BDI_CAP_MAP_DIRECT |
953                                         BDI_CAP_READ_MAP |
954                                         BDI_CAP_WRITE_MAP;
955                                 break;
956
957                         default:
958                                 return -EINVAL;
959                         }
960                 }
961
962                 /* eliminate any capabilities that we can't support on this
963                  * device */
964                 if (!file->f_op->get_unmapped_area)
965                         capabilities &= ~BDI_CAP_MAP_DIRECT;
966                 if (!file->f_op->read)
967                         capabilities &= ~BDI_CAP_MAP_COPY;
968
969                 /* The file shall have been opened with read permission. */
970                 if (!(file->f_mode & FMODE_READ))
971                         return -EACCES;
972
973                 if (flags & MAP_SHARED) {
974                         /* do checks for writing, appending and locking */
975                         if ((prot & PROT_WRITE) &&
976                             !(file->f_mode & FMODE_WRITE))
977                                 return -EACCES;
978
979                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
980                             (file->f_mode & FMODE_WRITE))
981                                 return -EACCES;
982
983                         if (locks_verify_locked(file->f_path.dentry->d_inode))
984                                 return -EAGAIN;
985
986                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
987                                 return -ENODEV;
988
989                         /* we mustn't privatise shared mappings */
990                         capabilities &= ~BDI_CAP_MAP_COPY;
991                 }
992                 else {
993                         /* we're going to read the file into private memory we
994                          * allocate */
995                         if (!(capabilities & BDI_CAP_MAP_COPY))
996                                 return -ENODEV;
997
998                         /* we don't permit a private writable mapping to be
999                          * shared with the backing device */
1000                         if (prot & PROT_WRITE)
1001                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1002                 }
1003
1004                 if (capabilities & BDI_CAP_MAP_DIRECT) {
1005                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
1006                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1007                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
1008                             ) {
1009                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1010                                 if (flags & MAP_SHARED) {
1011                                         printk(KERN_WARNING
1012                                                "MAP_SHARED not completely supported on !MMU\n");
1013                                         return -EINVAL;
1014                                 }
1015                         }
1016                 }
1017
1018                 /* handle executable mappings and implied executable
1019                  * mappings */
1020                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1021                         if (prot & PROT_EXEC)
1022                                 return -EPERM;
1023                 }
1024                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1025                         /* handle implication of PROT_EXEC by PROT_READ */
1026                         if (current->personality & READ_IMPLIES_EXEC) {
1027                                 if (capabilities & BDI_CAP_EXEC_MAP)
1028                                         prot |= PROT_EXEC;
1029                         }
1030                 }
1031                 else if ((prot & PROT_READ) &&
1032                          (prot & PROT_EXEC) &&
1033                          !(capabilities & BDI_CAP_EXEC_MAP)
1034                          ) {
1035                         /* backing file is not executable, try to copy */
1036                         capabilities &= ~BDI_CAP_MAP_DIRECT;
1037                 }
1038         }
1039         else {
1040                 /* anonymous mappings are always memory backed and can be
1041                  * privately mapped
1042                  */
1043                 capabilities = BDI_CAP_MAP_COPY;
1044
1045                 /* handle PROT_EXEC implication by PROT_READ */
1046                 if ((prot & PROT_READ) &&
1047                     (current->personality & READ_IMPLIES_EXEC))
1048                         prot |= PROT_EXEC;
1049         }
1050
1051         /* allow the security API to have its say */
1052         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1053         if (ret < 0)
1054                 return ret;
1055
1056         /* looks okay */
1057         *_capabilities = capabilities;
1058         return 0;
1059 }
1060
1061 /*
1062  * we've determined that we can make the mapping, now translate what we
1063  * now know into VMA flags
1064  */
1065 static unsigned long determine_vm_flags(struct file *file,
1066                                         unsigned long prot,
1067                                         unsigned long flags,
1068                                         unsigned long capabilities)
1069 {
1070         unsigned long vm_flags;
1071
1072         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1073         /* vm_flags |= mm->def_flags; */
1074
1075         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1076                 /* attempt to share read-only copies of mapped file chunks */
1077                 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1078                 if (file && !(prot & PROT_WRITE))
1079                         vm_flags |= VM_MAYSHARE;
1080         } else {
1081                 /* overlay a shareable mapping on the backing device or inode
1082                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1083                  * romfs/cramfs */
1084                 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1085                 if (flags & MAP_SHARED)
1086                         vm_flags |= VM_SHARED;
1087         }
1088
1089         /* refuse to let anyone share private mappings with this process if
1090          * it's being traced - otherwise breakpoints set in it may interfere
1091          * with another untraced process
1092          */
1093         if ((flags & MAP_PRIVATE) && current->ptrace)
1094                 vm_flags &= ~VM_MAYSHARE;
1095
1096         return vm_flags;
1097 }
1098
1099 /*
1100  * set up a shared mapping on a file (the driver or filesystem provides and
1101  * pins the storage)
1102  */
1103 static int do_mmap_shared_file(struct vm_area_struct *vma)
1104 {
1105         int ret;
1106
1107         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1108         if (ret == 0) {
1109                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1110                 return 0;
1111         }
1112         if (ret != -ENOSYS)
1113                 return ret;
1114
1115         /* getting -ENOSYS indicates that direct mmap isn't possible (as
1116          * opposed to tried but failed) so we can only give a suitable error as
1117          * it's not possible to make a private copy if MAP_SHARED was given */
1118         return -ENODEV;
1119 }
1120
1121 /*
1122  * set up a private mapping or an anonymous shared mapping
1123  */
1124 static int do_mmap_private(struct vm_area_struct *vma,
1125                            struct vm_region *region,
1126                            unsigned long len,
1127                            unsigned long capabilities)
1128 {
1129         struct page *pages;
1130         unsigned long total, point, n;
1131         void *base;
1132         int ret, order;
1133
1134         /* invoke the file's mapping function so that it can keep track of
1135          * shared mappings on devices or memory
1136          * - VM_MAYSHARE will be set if it may attempt to share
1137          */
1138         if (capabilities & BDI_CAP_MAP_DIRECT) {
1139                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1140                 if (ret == 0) {
1141                         /* shouldn't return success if we're not sharing */
1142                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1143                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1144                         return 0;
1145                 }
1146                 if (ret != -ENOSYS)
1147                         return ret;
1148
1149                 /* getting an ENOSYS error indicates that direct mmap isn't
1150                  * possible (as opposed to tried but failed) so we'll try to
1151                  * make a private copy of the data and map that instead */
1152         }
1153
1154
1155         /* allocate some memory to hold the mapping
1156          * - note that this may not return a page-aligned address if the object
1157          *   we're allocating is smaller than a page
1158          */
1159         order = get_order(len);
1160         kdebug("alloc order %d for %lx", order, len);
1161
1162         pages = alloc_pages(GFP_KERNEL, order);
1163         if (!pages)
1164                 goto enomem;
1165
1166         total = 1 << order;
1167         atomic_long_add(total, &mmap_pages_allocated);
1168
1169         point = len >> PAGE_SHIFT;
1170
1171         /* we allocated a power-of-2 sized page set, so we may want to trim off
1172          * the excess */
1173         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1174                 while (total > point) {
1175                         order = ilog2(total - point);
1176                         n = 1 << order;
1177                         kdebug("shave %lu/%lu @%lu", n, total - point, total);
1178                         atomic_long_sub(n, &mmap_pages_allocated);
1179                         total -= n;
1180                         set_page_refcounted(pages + total);
1181                         __free_pages(pages + total, order);
1182                 }
1183         }
1184
1185         for (point = 1; point < total; point++)
1186                 set_page_refcounted(&pages[point]);
1187
1188         base = page_address(pages);
1189         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1190         region->vm_start = (unsigned long) base;
1191         region->vm_end   = region->vm_start + len;
1192         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1193
1194         vma->vm_start = region->vm_start;
1195         vma->vm_end   = region->vm_start + len;
1196
1197         if (vma->vm_file) {
1198                 /* read the contents of a file into the copy */
1199                 mm_segment_t old_fs;
1200                 loff_t fpos;
1201
1202                 fpos = vma->vm_pgoff;
1203                 fpos <<= PAGE_SHIFT;
1204
1205                 old_fs = get_fs();
1206                 set_fs(KERNEL_DS);
1207                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
1208                 set_fs(old_fs);
1209
1210                 if (ret < 0)
1211                         goto error_free;
1212
1213                 /* clear the last little bit */
1214                 if (ret < len)
1215                         memset(base + ret, 0, len - ret);
1216
1217         }
1218
1219         return 0;
1220
1221 error_free:
1222         free_page_series(region->vm_start, region->vm_top);
1223         region->vm_start = vma->vm_start = 0;
1224         region->vm_end   = vma->vm_end = 0;
1225         region->vm_top   = 0;
1226         return ret;
1227
1228 enomem:
1229         printk("Allocation of length %lu from process %d (%s) failed\n",
1230                len, current->pid, current->comm);
1231         show_free_areas(0);
1232         return -ENOMEM;
1233 }
1234
1235 /*
1236  * handle mapping creation for uClinux
1237  */
1238 unsigned long do_mmap_pgoff(struct file *file,
1239                             unsigned long addr,
1240                             unsigned long len,
1241                             unsigned long prot,
1242                             unsigned long flags,
1243                             unsigned long pgoff)
1244 {
1245         struct vm_area_struct *vma;
1246         struct vm_region *region;
1247         struct rb_node *rb;
1248         unsigned long capabilities, vm_flags, result;
1249         int ret;
1250
1251         kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1252
1253         /* decide whether we should attempt the mapping, and if so what sort of
1254          * mapping */
1255         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1256                                     &capabilities);
1257         if (ret < 0) {
1258                 kleave(" = %d [val]", ret);
1259                 return ret;
1260         }
1261
1262         /* we ignore the address hint */
1263         addr = 0;
1264         len = PAGE_ALIGN(len);
1265
1266         /* we've determined that we can make the mapping, now translate what we
1267          * now know into VMA flags */
1268         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1269
1270         /* we're going to need to record the mapping */
1271         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1272         if (!region)
1273                 goto error_getting_region;
1274
1275         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1276         if (!vma)
1277                 goto error_getting_vma;
1278
1279         region->vm_usage = 1;
1280         region->vm_flags = vm_flags;
1281         region->vm_pgoff = pgoff;
1282
1283         INIT_LIST_HEAD(&vma->anon_vma_chain);
1284         vma->vm_flags = vm_flags;
1285         vma->vm_pgoff = pgoff;
1286
1287         if (file) {
1288                 region->vm_file = file;
1289                 get_file(file);
1290                 vma->vm_file = file;
1291                 get_file(file);
1292                 if (vm_flags & VM_EXECUTABLE) {
1293                         added_exe_file_vma(current->mm);
1294                         vma->vm_mm = current->mm;
1295                 }
1296         }
1297
1298         down_write(&nommu_region_sem);
1299
1300         /* if we want to share, we need to check for regions created by other
1301          * mmap() calls that overlap with our proposed mapping
1302          * - we can only share with a superset match on most regular files
1303          * - shared mappings on character devices and memory backed files are
1304          *   permitted to overlap inexactly as far as we are concerned for in
1305          *   these cases, sharing is handled in the driver or filesystem rather
1306          *   than here
1307          */
1308         if (vm_flags & VM_MAYSHARE) {
1309                 struct vm_region *pregion;
1310                 unsigned long pglen, rpglen, pgend, rpgend, start;
1311
1312                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1313                 pgend = pgoff + pglen;
1314
1315                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1316                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1317
1318                         if (!(pregion->vm_flags & VM_MAYSHARE))
1319                                 continue;
1320
1321                         /* search for overlapping mappings on the same file */
1322                         if (pregion->vm_file->f_path.dentry->d_inode !=
1323                             file->f_path.dentry->d_inode)
1324                                 continue;
1325
1326                         if (pregion->vm_pgoff >= pgend)
1327                                 continue;
1328
1329                         rpglen = pregion->vm_end - pregion->vm_start;
1330                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1331                         rpgend = pregion->vm_pgoff + rpglen;
1332                         if (pgoff >= rpgend)
1333                                 continue;
1334
1335                         /* handle inexactly overlapping matches between
1336                          * mappings */
1337                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1338                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1339                                 /* new mapping is not a subset of the region */
1340                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1341                                         goto sharing_violation;
1342                                 continue;
1343                         }
1344
1345                         /* we've found a region we can share */
1346                         pregion->vm_usage++;
1347                         vma->vm_region = pregion;
1348                         start = pregion->vm_start;
1349                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1350                         vma->vm_start = start;
1351                         vma->vm_end = start + len;
1352
1353                         if (pregion->vm_flags & VM_MAPPED_COPY) {
1354                                 kdebug("share copy");
1355                                 vma->vm_flags |= VM_MAPPED_COPY;
1356                         } else {
1357                                 kdebug("share mmap");
1358                                 ret = do_mmap_shared_file(vma);
1359                                 if (ret < 0) {
1360                                         vma->vm_region = NULL;
1361                                         vma->vm_start = 0;
1362                                         vma->vm_end = 0;
1363                                         pregion->vm_usage--;
1364                                         pregion = NULL;
1365                                         goto error_just_free;
1366                                 }
1367                         }
1368                         fput(region->vm_file);
1369                         kmem_cache_free(vm_region_jar, region);
1370                         region = pregion;
1371                         result = start;
1372                         goto share;
1373                 }
1374
1375                 /* obtain the address at which to make a shared mapping
1376                  * - this is the hook for quasi-memory character devices to
1377                  *   tell us the location of a shared mapping
1378                  */
1379                 if (capabilities & BDI_CAP_MAP_DIRECT) {
1380                         addr = file->f_op->get_unmapped_area(file, addr, len,
1381                                                              pgoff, flags);
1382                         if (IS_ERR_VALUE(addr)) {
1383                                 ret = addr;
1384                                 if (ret != -ENOSYS)
1385                                         goto error_just_free;
1386
1387                                 /* the driver refused to tell us where to site
1388                                  * the mapping so we'll have to attempt to copy
1389                                  * it */
1390                                 ret = -ENODEV;
1391                                 if (!(capabilities & BDI_CAP_MAP_COPY))
1392                                         goto error_just_free;
1393
1394                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
1395                         } else {
1396                                 vma->vm_start = region->vm_start = addr;
1397                                 vma->vm_end = region->vm_end = addr + len;
1398                         }
1399                 }
1400         }
1401
1402         vma->vm_region = region;
1403
1404         /* set up the mapping
1405          * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1406          */
1407         if (file && vma->vm_flags & VM_SHARED)
1408                 ret = do_mmap_shared_file(vma);
1409         else
1410                 ret = do_mmap_private(vma, region, len, capabilities);
1411         if (ret < 0)
1412                 goto error_just_free;
1413         add_nommu_region(region);
1414
1415         /* clear anonymous mappings that don't ask for uninitialized data */
1416         if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1417                 memset((void *)region->vm_start, 0,
1418                        region->vm_end - region->vm_start);
1419
1420         /* okay... we have a mapping; now we have to register it */
1421         result = vma->vm_start;
1422
1423         current->mm->total_vm += len >> PAGE_SHIFT;
1424
1425 share:
1426         add_vma_to_mm(current->mm, vma);
1427
1428         /* we flush the region from the icache only when the first executable
1429          * mapping of it is made  */
1430         if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1431                 flush_icache_range(region->vm_start, region->vm_end);
1432                 region->vm_icache_flushed = true;
1433         }
1434
1435         up_write(&nommu_region_sem);
1436
1437         kleave(" = %lx", result);
1438         return result;
1439
1440 error_just_free:
1441         up_write(&nommu_region_sem);
1442 error:
1443         if (region->vm_file)
1444                 fput(region->vm_file);
1445         kmem_cache_free(vm_region_jar, region);
1446         if (vma->vm_file)
1447                 fput(vma->vm_file);
1448         if (vma->vm_flags & VM_EXECUTABLE)
1449                 removed_exe_file_vma(vma->vm_mm);
1450         kmem_cache_free(vm_area_cachep, vma);
1451         kleave(" = %d", ret);
1452         return ret;
1453
1454 sharing_violation:
1455         up_write(&nommu_region_sem);
1456         printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1457         ret = -EINVAL;
1458         goto error;
1459
1460 error_getting_vma:
1461         kmem_cache_free(vm_region_jar, region);
1462         printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1463                " from process %d failed\n",
1464                len, current->pid);
1465         show_free_areas(0);
1466         return -ENOMEM;
1467
1468 error_getting_region:
1469         printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1470                " from process %d failed\n",
1471                len, current->pid);
1472         show_free_areas(0);
1473         return -ENOMEM;
1474 }
1475 EXPORT_SYMBOL(do_mmap_pgoff);
1476
1477 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1478                 unsigned long, prot, unsigned long, flags,
1479                 unsigned long, fd, unsigned long, pgoff)
1480 {
1481         struct file *file = NULL;
1482         unsigned long retval = -EBADF;
1483
1484         audit_mmap_fd(fd, flags);
1485         if (!(flags & MAP_ANONYMOUS)) {
1486                 file = fget(fd);
1487                 if (!file)
1488                         goto out;
1489         }
1490
1491         flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1492
1493         down_write(&current->mm->mmap_sem);
1494         retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1495         up_write(&current->mm->mmap_sem);
1496
1497         if (file)
1498                 fput(file);
1499 out:
1500         return retval;
1501 }
1502
1503 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1504 struct mmap_arg_struct {
1505         unsigned long addr;
1506         unsigned long len;
1507         unsigned long prot;
1508         unsigned long flags;
1509         unsigned long fd;
1510         unsigned long offset;
1511 };
1512
1513 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1514 {
1515         struct mmap_arg_struct a;
1516
1517         if (copy_from_user(&a, arg, sizeof(a)))
1518                 return -EFAULT;
1519         if (a.offset & ~PAGE_MASK)
1520                 return -EINVAL;
1521
1522         return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1523                               a.offset >> PAGE_SHIFT);
1524 }
1525 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1526
1527 /*
1528  * split a vma into two pieces at address 'addr', a new vma is allocated either
1529  * for the first part or the tail.
1530  */
1531 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1532               unsigned long addr, int new_below)
1533 {
1534         struct vm_area_struct *new;
1535         struct vm_region *region;
1536         unsigned long npages;
1537
1538         kenter("");
1539
1540         /* we're only permitted to split anonymous regions (these should have
1541          * only a single usage on the region) */
1542         if (vma->vm_file)
1543                 return -ENOMEM;
1544
1545         if (mm->map_count >= sysctl_max_map_count)
1546                 return -ENOMEM;
1547
1548         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1549         if (!region)
1550                 return -ENOMEM;
1551
1552         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1553         if (!new) {
1554                 kmem_cache_free(vm_region_jar, region);
1555                 return -ENOMEM;
1556         }
1557
1558         /* most fields are the same, copy all, and then fixup */
1559         *new = *vma;
1560         *region = *vma->vm_region;
1561         new->vm_region = region;
1562
1563         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1564
1565         if (new_below) {
1566                 region->vm_top = region->vm_end = new->vm_end = addr;
1567         } else {
1568                 region->vm_start = new->vm_start = addr;
1569                 region->vm_pgoff = new->vm_pgoff += npages;
1570         }
1571
1572         if (new->vm_ops && new->vm_ops->open)
1573                 new->vm_ops->open(new);
1574
1575         delete_vma_from_mm(vma);
1576         down_write(&nommu_region_sem);
1577         delete_nommu_region(vma->vm_region);
1578         if (new_below) {
1579                 vma->vm_region->vm_start = vma->vm_start = addr;
1580                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1581         } else {
1582                 vma->vm_region->vm_end = vma->vm_end = addr;
1583                 vma->vm_region->vm_top = addr;
1584         }
1585         add_nommu_region(vma->vm_region);
1586         add_nommu_region(new->vm_region);
1587         up_write(&nommu_region_sem);
1588         add_vma_to_mm(mm, vma);
1589         add_vma_to_mm(mm, new);
1590         return 0;
1591 }
1592
1593 /*
1594  * shrink a VMA by removing the specified chunk from either the beginning or
1595  * the end
1596  */
1597 static int shrink_vma(struct mm_struct *mm,
1598                       struct vm_area_struct *vma,
1599                       unsigned long from, unsigned long to)
1600 {
1601         struct vm_region *region;
1602
1603         kenter("");
1604
1605         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1606          * and list */
1607         delete_vma_from_mm(vma);
1608         if (from > vma->vm_start)
1609                 vma->vm_end = from;
1610         else
1611                 vma->vm_start = to;
1612         add_vma_to_mm(mm, vma);
1613
1614         /* cut the backing region down to size */
1615         region = vma->vm_region;
1616         BUG_ON(region->vm_usage != 1);
1617
1618         down_write(&nommu_region_sem);
1619         delete_nommu_region(region);
1620         if (from > region->vm_start) {
1621                 to = region->vm_top;
1622                 region->vm_top = region->vm_end = from;
1623         } else {
1624                 region->vm_start = to;
1625         }
1626         add_nommu_region(region);
1627         up_write(&nommu_region_sem);
1628
1629         free_page_series(from, to);
1630         return 0;
1631 }
1632
1633 /*
1634  * release a mapping
1635  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1636  *   VMA, though it need not cover the whole VMA
1637  */
1638 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1639 {
1640         struct vm_area_struct *vma;
1641         unsigned long end;
1642         int ret;
1643
1644         kenter(",%lx,%zx", start, len);
1645
1646         len = PAGE_ALIGN(len);
1647         if (len == 0)
1648                 return -EINVAL;
1649
1650         end = start + len;
1651
1652         /* find the first potentially overlapping VMA */
1653         vma = find_vma(mm, start);
1654         if (!vma) {
1655                 static int limit = 0;
1656                 if (limit < 5) {
1657                         printk(KERN_WARNING
1658                                "munmap of memory not mmapped by process %d"
1659                                " (%s): 0x%lx-0x%lx\n",
1660                                current->pid, current->comm,
1661                                start, start + len - 1);
1662                         limit++;
1663                 }
1664                 return -EINVAL;
1665         }
1666
1667         /* we're allowed to split an anonymous VMA but not a file-backed one */
1668         if (vma->vm_file) {
1669                 do {
1670                         if (start > vma->vm_start) {
1671                                 kleave(" = -EINVAL [miss]");
1672                                 return -EINVAL;
1673                         }
1674                         if (end == vma->vm_end)
1675                                 goto erase_whole_vma;
1676                         vma = vma->vm_next;
1677                 } while (vma);
1678                 kleave(" = -EINVAL [split file]");
1679                 return -EINVAL;
1680         } else {
1681                 /* the chunk must be a subset of the VMA found */
1682                 if (start == vma->vm_start && end == vma->vm_end)
1683                         goto erase_whole_vma;
1684                 if (start < vma->vm_start || end > vma->vm_end) {
1685                         kleave(" = -EINVAL [superset]");
1686                         return -EINVAL;
1687                 }
1688                 if (start & ~PAGE_MASK) {
1689                         kleave(" = -EINVAL [unaligned start]");
1690                         return -EINVAL;
1691                 }
1692                 if (end != vma->vm_end && end & ~PAGE_MASK) {
1693                         kleave(" = -EINVAL [unaligned split]");
1694                         return -EINVAL;
1695                 }
1696                 if (start != vma->vm_start && end != vma->vm_end) {
1697                         ret = split_vma(mm, vma, start, 1);
1698                         if (ret < 0) {
1699                                 kleave(" = %d [split]", ret);
1700                                 return ret;
1701                         }
1702                 }
1703                 return shrink_vma(mm, vma, start, end);
1704         }
1705
1706 erase_whole_vma:
1707         delete_vma_from_mm(vma);
1708         delete_vma(mm, vma);
1709         kleave(" = 0");
1710         return 0;
1711 }
1712 EXPORT_SYMBOL(do_munmap);
1713
1714 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1715 {
1716         int ret;
1717         struct mm_struct *mm = current->mm;
1718
1719         down_write(&mm->mmap_sem);
1720         ret = do_munmap(mm, addr, len);
1721         up_write(&mm->mmap_sem);
1722         return ret;
1723 }
1724
1725 /*
1726  * release all the mappings made in a process's VM space
1727  */
1728 void exit_mmap(struct mm_struct *mm)
1729 {
1730         struct vm_area_struct *vma;
1731
1732         if (!mm)
1733                 return;
1734
1735         kenter("");
1736
1737         mm->total_vm = 0;
1738
1739         while ((vma = mm->mmap)) {
1740                 mm->mmap = vma->vm_next;
1741                 delete_vma_from_mm(vma);
1742                 delete_vma(mm, vma);
1743                 cond_resched();
1744         }
1745
1746         kleave("");
1747 }
1748
1749 unsigned long do_brk(unsigned long addr, unsigned long len)
1750 {
1751         return -ENOMEM;
1752 }
1753
1754 /*
1755  * expand (or shrink) an existing mapping, potentially moving it at the same
1756  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1757  *
1758  * under NOMMU conditions, we only permit changing a mapping's size, and only
1759  * as long as it stays within the region allocated by do_mmap_private() and the
1760  * block is not shareable
1761  *
1762  * MREMAP_FIXED is not supported under NOMMU conditions
1763  */
1764 unsigned long do_mremap(unsigned long addr,
1765                         unsigned long old_len, unsigned long new_len,
1766                         unsigned long flags, unsigned long new_addr)
1767 {
1768         struct vm_area_struct *vma;
1769
1770         /* insanity checks first */
1771         old_len = PAGE_ALIGN(old_len);
1772         new_len = PAGE_ALIGN(new_len);
1773         if (old_len == 0 || new_len == 0)
1774                 return (unsigned long) -EINVAL;
1775
1776         if (addr & ~PAGE_MASK)
1777                 return -EINVAL;
1778
1779         if (flags & MREMAP_FIXED && new_addr != addr)
1780                 return (unsigned long) -EINVAL;
1781
1782         vma = find_vma_exact(current->mm, addr, old_len);
1783         if (!vma)
1784                 return (unsigned long) -EINVAL;
1785
1786         if (vma->vm_end != vma->vm_start + old_len)
1787                 return (unsigned long) -EFAULT;
1788
1789         if (vma->vm_flags & VM_MAYSHARE)
1790                 return (unsigned long) -EPERM;
1791
1792         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1793                 return (unsigned long) -ENOMEM;
1794
1795         /* all checks complete - do it */
1796         vma->vm_end = vma->vm_start + new_len;
1797         return vma->vm_start;
1798 }
1799 EXPORT_SYMBOL(do_mremap);
1800
1801 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1802                 unsigned long, new_len, unsigned long, flags,
1803                 unsigned long, new_addr)
1804 {
1805         unsigned long ret;
1806
1807         down_write(&current->mm->mmap_sem);
1808         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1809         up_write(&current->mm->mmap_sem);
1810         return ret;
1811 }
1812
1813 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1814                         unsigned int foll_flags)
1815 {
1816         return NULL;
1817 }
1818
1819 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1820                 unsigned long pfn, unsigned long size, pgprot_t prot)
1821 {
1822         if (addr != (pfn << PAGE_SHIFT))
1823                 return -EINVAL;
1824
1825         vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1826         return 0;
1827 }
1828 EXPORT_SYMBOL(remap_pfn_range);
1829
1830 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1831                         unsigned long pgoff)
1832 {
1833         unsigned int size = vma->vm_end - vma->vm_start;
1834
1835         if (!(vma->vm_flags & VM_USERMAP))
1836                 return -EINVAL;
1837
1838         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1839         vma->vm_end = vma->vm_start + size;
1840
1841         return 0;
1842 }
1843 EXPORT_SYMBOL(remap_vmalloc_range);
1844
1845 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1846         unsigned long len, unsigned long pgoff, unsigned long flags)
1847 {
1848         return -ENOMEM;
1849 }
1850
1851 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1852 {
1853 }
1854
1855 void unmap_mapping_range(struct address_space *mapping,
1856                          loff_t const holebegin, loff_t const holelen,
1857                          int even_cows)
1858 {
1859 }
1860 EXPORT_SYMBOL(unmap_mapping_range);
1861
1862 /*
1863  * Check that a process has enough memory to allocate a new virtual
1864  * mapping. 0 means there is enough memory for the allocation to
1865  * succeed and -ENOMEM implies there is not.
1866  *
1867  * We currently support three overcommit policies, which are set via the
1868  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1869  *
1870  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1871  * Additional code 2002 Jul 20 by Robert Love.
1872  *
1873  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1874  *
1875  * Note this is a helper function intended to be used by LSMs which
1876  * wish to use this logic.
1877  */
1878 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1879 {
1880         unsigned long free, allowed;
1881
1882         vm_acct_memory(pages);
1883
1884         /*
1885          * Sometimes we want to use more memory than we have
1886          */
1887         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1888                 return 0;
1889
1890         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1891                 free = global_page_state(NR_FREE_PAGES);
1892                 free += global_page_state(NR_FILE_PAGES);
1893
1894                 /*
1895                  * shmem pages shouldn't be counted as free in this
1896                  * case, they can't be purged, only swapped out, and
1897                  * that won't affect the overall amount of available
1898                  * memory in the system.
1899                  */
1900                 free -= global_page_state(NR_SHMEM);
1901
1902                 free += nr_swap_pages;
1903
1904                 /*
1905                  * Any slabs which are created with the
1906                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1907                  * which are reclaimable, under pressure.  The dentry
1908                  * cache and most inode caches should fall into this
1909                  */
1910                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1911
1912                 /*
1913                  * Leave reserved pages. The pages are not for anonymous pages.
1914                  */
1915                 if (free <= totalreserve_pages)
1916                         goto error;
1917                 else
1918                         free -= totalreserve_pages;
1919
1920                 /*
1921                  * Leave the last 3% for root
1922                  */
1923                 if (!cap_sys_admin)
1924                         free -= free / 32;
1925
1926                 if (free > pages)
1927                         return 0;
1928
1929                 goto error;
1930         }
1931
1932         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1933         /*
1934          * Leave the last 3% for root
1935          */
1936         if (!cap_sys_admin)
1937                 allowed -= allowed / 32;
1938         allowed += total_swap_pages;
1939
1940         /* Don't let a single process grow too big:
1941            leave 3% of the size of this process for other processes */
1942         if (mm)
1943                 allowed -= mm->total_vm / 32;
1944
1945         if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1946                 return 0;
1947
1948 error:
1949         vm_unacct_memory(pages);
1950
1951         return -ENOMEM;
1952 }
1953
1954 int in_gate_area_no_mm(unsigned long addr)
1955 {
1956         return 0;
1957 }
1958
1959 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1960 {
1961         BUG();
1962         return 0;
1963 }
1964 EXPORT_SYMBOL(filemap_fault);
1965
1966 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1967                 unsigned long addr, void *buf, int len, int write)
1968 {
1969         struct vm_area_struct *vma;
1970
1971         down_read(&mm->mmap_sem);
1972
1973         /* the access must start within one of the target process's mappings */
1974         vma = find_vma(mm, addr);
1975         if (vma) {
1976                 /* don't overrun this mapping */
1977                 if (addr + len >= vma->vm_end)
1978                         len = vma->vm_end - addr;
1979
1980                 /* only read or write mappings where it is permitted */
1981                 if (write && vma->vm_flags & VM_MAYWRITE)
1982                         copy_to_user_page(vma, NULL, addr,
1983                                          (void *) addr, buf, len);
1984                 else if (!write && vma->vm_flags & VM_MAYREAD)
1985                         copy_from_user_page(vma, NULL, addr,
1986                                             buf, (void *) addr, len);
1987                 else
1988                         len = 0;
1989         } else {
1990                 len = 0;
1991         }
1992
1993         up_read(&mm->mmap_sem);
1994
1995         return len;
1996 }
1997
1998 /**
1999  * @access_remote_vm - access another process' address space
2000  * @mm:         the mm_struct of the target address space
2001  * @addr:       start address to access
2002  * @buf:        source or destination buffer
2003  * @len:        number of bytes to transfer
2004  * @write:      whether the access is a write
2005  *
2006  * The caller must hold a reference on @mm.
2007  */
2008 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2009                 void *buf, int len, int write)
2010 {
2011         return __access_remote_vm(NULL, mm, addr, buf, len, write);
2012 }
2013
2014 /*
2015  * Access another process' address space.
2016  * - source/target buffer must be kernel space
2017  */
2018 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2019 {
2020         struct mm_struct *mm;
2021
2022         if (addr + len < addr)
2023                 return 0;
2024
2025         mm = get_task_mm(tsk);
2026         if (!mm)
2027                 return 0;
2028
2029         len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2030
2031         mmput(mm);
2032         return len;
2033 }
2034
2035 /**
2036  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2037  * @inode: The inode to check
2038  * @size: The current filesize of the inode
2039  * @newsize: The proposed filesize of the inode
2040  *
2041  * Check the shared mappings on an inode on behalf of a shrinking truncate to
2042  * make sure that that any outstanding VMAs aren't broken and then shrink the
2043  * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2044  * automatically grant mappings that are too large.
2045  */
2046 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2047                                 size_t newsize)
2048 {
2049         struct vm_area_struct *vma;
2050         struct prio_tree_iter iter;
2051         struct vm_region *region;
2052         pgoff_t low, high;
2053         size_t r_size, r_top;
2054
2055         low = newsize >> PAGE_SHIFT;
2056         high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2057
2058         down_write(&nommu_region_sem);
2059         mutex_lock(&inode->i_mapping->i_mmap_mutex);
2060
2061         /* search for VMAs that fall within the dead zone */
2062         vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2063                               low, high) {
2064                 /* found one - only interested if it's shared out of the page
2065                  * cache */
2066                 if (vma->vm_flags & VM_SHARED) {
2067                         mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2068                         up_write(&nommu_region_sem);
2069                         return -ETXTBSY; /* not quite true, but near enough */
2070                 }
2071         }
2072
2073         /* reduce any regions that overlap the dead zone - if in existence,
2074          * these will be pointed to by VMAs that don't overlap the dead zone
2075          *
2076          * we don't check for any regions that start beyond the EOF as there
2077          * shouldn't be any
2078          */
2079         vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2080                               0, ULONG_MAX) {
2081                 if (!(vma->vm_flags & VM_SHARED))
2082                         continue;
2083
2084                 region = vma->vm_region;
2085                 r_size = region->vm_top - region->vm_start;
2086                 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2087
2088                 if (r_top > newsize) {
2089                         region->vm_top -= r_top - newsize;
2090                         if (region->vm_end > region->vm_top)
2091                                 region->vm_end = region->vm_top;
2092                 }
2093         }
2094
2095         mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2096         up_write(&nommu_region_sem);
2097         return 0;
2098 }