mm/hmm: add new helper to hotplug CDM memory region
[linux-2.6-microblaze.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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17
18 #include <linux/export.h>
19 #include <linux/mm.h>
20 #include <linux/sched/mm.h>
21 #include <linux/vmacache.h>
22 #include <linux/mman.h>
23 #include <linux/swap.h>
24 #include <linux/file.h>
25 #include <linux/highmem.h>
26 #include <linux/pagemap.h>
27 #include <linux/slab.h>
28 #include <linux/vmalloc.h>
29 #include <linux/blkdev.h>
30 #include <linux/backing-dev.h>
31 #include <linux/compiler.h>
32 #include <linux/mount.h>
33 #include <linux/personality.h>
34 #include <linux/security.h>
35 #include <linux/syscalls.h>
36 #include <linux/audit.h>
37 #include <linux/printk.h>
38
39 #include <linux/uaccess.h>
40 #include <asm/tlb.h>
41 #include <asm/tlbflush.h>
42 #include <asm/mmu_context.h>
43 #include "internal.h"
44
45 void *high_memory;
46 EXPORT_SYMBOL(high_memory);
47 struct page *mem_map;
48 unsigned long max_mapnr;
49 EXPORT_SYMBOL(max_mapnr);
50 unsigned long highest_memmap_pfn;
51 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
52 int heap_stack_gap = 0;
53
54 atomic_long_t mmap_pages_allocated;
55
56 EXPORT_SYMBOL(mem_map);
57
58 /* list of mapped, potentially shareable regions */
59 static struct kmem_cache *vm_region_jar;
60 struct rb_root nommu_region_tree = RB_ROOT;
61 DECLARE_RWSEM(nommu_region_sem);
62
63 const struct vm_operations_struct generic_file_vm_ops = {
64 };
65
66 /*
67  * Return the total memory allocated for this pointer, not
68  * just what the caller asked for.
69  *
70  * Doesn't have to be accurate, i.e. may have races.
71  */
72 unsigned int kobjsize(const void *objp)
73 {
74         struct page *page;
75
76         /*
77          * If the object we have should not have ksize performed on it,
78          * return size of 0
79          */
80         if (!objp || !virt_addr_valid(objp))
81                 return 0;
82
83         page = virt_to_head_page(objp);
84
85         /*
86          * If the allocator sets PageSlab, we know the pointer came from
87          * kmalloc().
88          */
89         if (PageSlab(page))
90                 return ksize(objp);
91
92         /*
93          * If it's not a compound page, see if we have a matching VMA
94          * region. This test is intentionally done in reverse order,
95          * so if there's no VMA, we still fall through and hand back
96          * PAGE_SIZE for 0-order pages.
97          */
98         if (!PageCompound(page)) {
99                 struct vm_area_struct *vma;
100
101                 vma = find_vma(current->mm, (unsigned long)objp);
102                 if (vma)
103                         return vma->vm_end - vma->vm_start;
104         }
105
106         /*
107          * The ksize() function is only guaranteed to work for pointers
108          * returned by kmalloc(). So handle arbitrary pointers here.
109          */
110         return PAGE_SIZE << compound_order(page);
111 }
112
113 static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
114                       unsigned long start, unsigned long nr_pages,
115                       unsigned int foll_flags, struct page **pages,
116                       struct vm_area_struct **vmas, int *nonblocking)
117 {
118         struct vm_area_struct *vma;
119         unsigned long vm_flags;
120         int i;
121
122         /* calculate required read or write permissions.
123          * If FOLL_FORCE is set, we only require the "MAY" flags.
124          */
125         vm_flags  = (foll_flags & FOLL_WRITE) ?
126                         (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
127         vm_flags &= (foll_flags & FOLL_FORCE) ?
128                         (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
129
130         for (i = 0; i < nr_pages; i++) {
131                 vma = find_vma(mm, start);
132                 if (!vma)
133                         goto finish_or_fault;
134
135                 /* protect what we can, including chardevs */
136                 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
137                     !(vm_flags & vma->vm_flags))
138                         goto finish_or_fault;
139
140                 if (pages) {
141                         pages[i] = virt_to_page(start);
142                         if (pages[i])
143                                 get_page(pages[i]);
144                 }
145                 if (vmas)
146                         vmas[i] = vma;
147                 start = (start + PAGE_SIZE) & PAGE_MASK;
148         }
149
150         return i;
151
152 finish_or_fault:
153         return i ? : -EFAULT;
154 }
155
156 /*
157  * get a list of pages in an address range belonging to the specified process
158  * and indicate the VMA that covers each page
159  * - this is potentially dodgy as we may end incrementing the page count of a
160  *   slab page or a secondary page from a compound page
161  * - don't permit access to VMAs that don't support it, such as I/O mappings
162  */
163 long get_user_pages(unsigned long start, unsigned long nr_pages,
164                     unsigned int gup_flags, struct page **pages,
165                     struct vm_area_struct **vmas)
166 {
167         return __get_user_pages(current, current->mm, start, nr_pages,
168                                 gup_flags, pages, vmas, NULL);
169 }
170 EXPORT_SYMBOL(get_user_pages);
171
172 long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
173                             unsigned int gup_flags, struct page **pages,
174                             int *locked)
175 {
176         return get_user_pages(start, nr_pages, gup_flags, pages, NULL);
177 }
178 EXPORT_SYMBOL(get_user_pages_locked);
179
180 static long __get_user_pages_unlocked(struct task_struct *tsk,
181                         struct mm_struct *mm, unsigned long start,
182                         unsigned long nr_pages, struct page **pages,
183                         unsigned int gup_flags)
184 {
185         long ret;
186         down_read(&mm->mmap_sem);
187         ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
188                                 NULL, NULL);
189         up_read(&mm->mmap_sem);
190         return ret;
191 }
192
193 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
194                              struct page **pages, unsigned int gup_flags)
195 {
196         return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
197                                          pages, gup_flags);
198 }
199 EXPORT_SYMBOL(get_user_pages_unlocked);
200
201 /**
202  * follow_pfn - look up PFN at a user virtual address
203  * @vma: memory mapping
204  * @address: user virtual address
205  * @pfn: location to store found PFN
206  *
207  * Only IO mappings and raw PFN mappings are allowed.
208  *
209  * Returns zero and the pfn at @pfn on success, -ve otherwise.
210  */
211 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
212         unsigned long *pfn)
213 {
214         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
215                 return -EINVAL;
216
217         *pfn = address >> PAGE_SHIFT;
218         return 0;
219 }
220 EXPORT_SYMBOL(follow_pfn);
221
222 LIST_HEAD(vmap_area_list);
223
224 void vfree(const void *addr)
225 {
226         kfree(addr);
227 }
228 EXPORT_SYMBOL(vfree);
229
230 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
231 {
232         /*
233          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
234          * returns only a logical address.
235          */
236         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
237 }
238 EXPORT_SYMBOL(__vmalloc);
239
240 void *__vmalloc_node_flags(unsigned long size, int node, gfp_t flags)
241 {
242         return __vmalloc(size, flags, PAGE_KERNEL);
243 }
244
245 void *vmalloc_user(unsigned long size)
246 {
247         void *ret;
248
249         ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
250         if (ret) {
251                 struct vm_area_struct *vma;
252
253                 down_write(&current->mm->mmap_sem);
254                 vma = find_vma(current->mm, (unsigned long)ret);
255                 if (vma)
256                         vma->vm_flags |= VM_USERMAP;
257                 up_write(&current->mm->mmap_sem);
258         }
259
260         return ret;
261 }
262 EXPORT_SYMBOL(vmalloc_user);
263
264 struct page *vmalloc_to_page(const void *addr)
265 {
266         return virt_to_page(addr);
267 }
268 EXPORT_SYMBOL(vmalloc_to_page);
269
270 unsigned long vmalloc_to_pfn(const void *addr)
271 {
272         return page_to_pfn(virt_to_page(addr));
273 }
274 EXPORT_SYMBOL(vmalloc_to_pfn);
275
276 long vread(char *buf, char *addr, unsigned long count)
277 {
278         /* Don't allow overflow */
279         if ((unsigned long) buf + count < count)
280                 count = -(unsigned long) buf;
281
282         memcpy(buf, addr, count);
283         return count;
284 }
285
286 long vwrite(char *buf, char *addr, unsigned long count)
287 {
288         /* Don't allow overflow */
289         if ((unsigned long) addr + count < count)
290                 count = -(unsigned long) addr;
291
292         memcpy(addr, buf, count);
293         return count;
294 }
295
296 /*
297  *      vmalloc  -  allocate virtually contiguous memory
298  *
299  *      @size:          allocation size
300  *
301  *      Allocate enough pages to cover @size from the page level
302  *      allocator and map them into contiguous kernel virtual space.
303  *
304  *      For tight control over page level allocator and protection flags
305  *      use __vmalloc() instead.
306  */
307 void *vmalloc(unsigned long size)
308 {
309        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
310 }
311 EXPORT_SYMBOL(vmalloc);
312
313 /*
314  *      vzalloc - allocate virtually contiguous memory with zero fill
315  *
316  *      @size:          allocation size
317  *
318  *      Allocate enough pages to cover @size from the page level
319  *      allocator and map them into contiguous kernel virtual space.
320  *      The memory allocated is set to zero.
321  *
322  *      For tight control over page level allocator and protection flags
323  *      use __vmalloc() instead.
324  */
325 void *vzalloc(unsigned long size)
326 {
327         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
328                         PAGE_KERNEL);
329 }
330 EXPORT_SYMBOL(vzalloc);
331
332 /**
333  * vmalloc_node - allocate memory on a specific node
334  * @size:       allocation size
335  * @node:       numa node
336  *
337  * Allocate enough pages to cover @size from the page level
338  * allocator and map them into contiguous kernel virtual space.
339  *
340  * For tight control over page level allocator and protection flags
341  * use __vmalloc() instead.
342  */
343 void *vmalloc_node(unsigned long size, int node)
344 {
345         return vmalloc(size);
346 }
347 EXPORT_SYMBOL(vmalloc_node);
348
349 /**
350  * vzalloc_node - allocate memory on a specific node with zero fill
351  * @size:       allocation size
352  * @node:       numa node
353  *
354  * Allocate enough pages to cover @size from the page level
355  * allocator and map them into contiguous kernel virtual space.
356  * The memory allocated is set to zero.
357  *
358  * For tight control over page level allocator and protection flags
359  * use __vmalloc() instead.
360  */
361 void *vzalloc_node(unsigned long size, int node)
362 {
363         return vzalloc(size);
364 }
365 EXPORT_SYMBOL(vzalloc_node);
366
367 #ifndef PAGE_KERNEL_EXEC
368 # define PAGE_KERNEL_EXEC PAGE_KERNEL
369 #endif
370
371 /**
372  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
373  *      @size:          allocation size
374  *
375  *      Kernel-internal function to allocate enough pages to cover @size
376  *      the page level allocator and map them into contiguous and
377  *      executable kernel virtual space.
378  *
379  *      For tight control over page level allocator and protection flags
380  *      use __vmalloc() instead.
381  */
382
383 void *vmalloc_exec(unsigned long size)
384 {
385         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
386 }
387
388 /**
389  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
390  *      @size:          allocation size
391  *
392  *      Allocate enough 32bit PA addressable pages to cover @size from the
393  *      page level allocator and map them into contiguous kernel virtual space.
394  */
395 void *vmalloc_32(unsigned long size)
396 {
397         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
398 }
399 EXPORT_SYMBOL(vmalloc_32);
400
401 /**
402  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
403  *      @size:          allocation size
404  *
405  * The resulting memory area is 32bit addressable and zeroed so it can be
406  * mapped to userspace without leaking data.
407  *
408  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
409  * remap_vmalloc_range() are permissible.
410  */
411 void *vmalloc_32_user(unsigned long size)
412 {
413         /*
414          * We'll have to sort out the ZONE_DMA bits for 64-bit,
415          * but for now this can simply use vmalloc_user() directly.
416          */
417         return vmalloc_user(size);
418 }
419 EXPORT_SYMBOL(vmalloc_32_user);
420
421 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
422 {
423         BUG();
424         return NULL;
425 }
426 EXPORT_SYMBOL(vmap);
427
428 void vunmap(const void *addr)
429 {
430         BUG();
431 }
432 EXPORT_SYMBOL(vunmap);
433
434 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
435 {
436         BUG();
437         return NULL;
438 }
439 EXPORT_SYMBOL(vm_map_ram);
440
441 void vm_unmap_ram(const void *mem, unsigned int count)
442 {
443         BUG();
444 }
445 EXPORT_SYMBOL(vm_unmap_ram);
446
447 void vm_unmap_aliases(void)
448 {
449 }
450 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
451
452 /*
453  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
454  * have one.
455  */
456 void __weak vmalloc_sync_all(void)
457 {
458 }
459
460 /**
461  *      alloc_vm_area - allocate a range of kernel address space
462  *      @size:          size of the area
463  *
464  *      Returns:        NULL on failure, vm_struct on success
465  *
466  *      This function reserves a range of kernel address space, and
467  *      allocates pagetables to map that range.  No actual mappings
468  *      are created.  If the kernel address space is not shared
469  *      between processes, it syncs the pagetable across all
470  *      processes.
471  */
472 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
473 {
474         BUG();
475         return NULL;
476 }
477 EXPORT_SYMBOL_GPL(alloc_vm_area);
478
479 void free_vm_area(struct vm_struct *area)
480 {
481         BUG();
482 }
483 EXPORT_SYMBOL_GPL(free_vm_area);
484
485 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
486                    struct page *page)
487 {
488         return -EINVAL;
489 }
490 EXPORT_SYMBOL(vm_insert_page);
491
492 /*
493  *  sys_brk() for the most part doesn't need the global kernel
494  *  lock, except when an application is doing something nasty
495  *  like trying to un-brk an area that has already been mapped
496  *  to a regular file.  in this case, the unmapping will need
497  *  to invoke file system routines that need the global lock.
498  */
499 SYSCALL_DEFINE1(brk, unsigned long, brk)
500 {
501         struct mm_struct *mm = current->mm;
502
503         if (brk < mm->start_brk || brk > mm->context.end_brk)
504                 return mm->brk;
505
506         if (mm->brk == brk)
507                 return mm->brk;
508
509         /*
510          * Always allow shrinking brk
511          */
512         if (brk <= mm->brk) {
513                 mm->brk = brk;
514                 return brk;
515         }
516
517         /*
518          * Ok, looks good - let it rip.
519          */
520         flush_icache_range(mm->brk, brk);
521         return mm->brk = brk;
522 }
523
524 /*
525  * initialise the percpu counter for VM and region record slabs
526  */
527 void __init mmap_init(void)
528 {
529         int ret;
530
531         ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
532         VM_BUG_ON(ret);
533         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
534 }
535
536 /*
537  * validate the region tree
538  * - the caller must hold the region lock
539  */
540 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
541 static noinline void validate_nommu_regions(void)
542 {
543         struct vm_region *region, *last;
544         struct rb_node *p, *lastp;
545
546         lastp = rb_first(&nommu_region_tree);
547         if (!lastp)
548                 return;
549
550         last = rb_entry(lastp, struct vm_region, vm_rb);
551         BUG_ON(last->vm_end <= last->vm_start);
552         BUG_ON(last->vm_top < last->vm_end);
553
554         while ((p = rb_next(lastp))) {
555                 region = rb_entry(p, struct vm_region, vm_rb);
556                 last = rb_entry(lastp, struct vm_region, vm_rb);
557
558                 BUG_ON(region->vm_end <= region->vm_start);
559                 BUG_ON(region->vm_top < region->vm_end);
560                 BUG_ON(region->vm_start < last->vm_top);
561
562                 lastp = p;
563         }
564 }
565 #else
566 static void validate_nommu_regions(void)
567 {
568 }
569 #endif
570
571 /*
572  * add a region into the global tree
573  */
574 static void add_nommu_region(struct vm_region *region)
575 {
576         struct vm_region *pregion;
577         struct rb_node **p, *parent;
578
579         validate_nommu_regions();
580
581         parent = NULL;
582         p = &nommu_region_tree.rb_node;
583         while (*p) {
584                 parent = *p;
585                 pregion = rb_entry(parent, struct vm_region, vm_rb);
586                 if (region->vm_start < pregion->vm_start)
587                         p = &(*p)->rb_left;
588                 else if (region->vm_start > pregion->vm_start)
589                         p = &(*p)->rb_right;
590                 else if (pregion == region)
591                         return;
592                 else
593                         BUG();
594         }
595
596         rb_link_node(&region->vm_rb, parent, p);
597         rb_insert_color(&region->vm_rb, &nommu_region_tree);
598
599         validate_nommu_regions();
600 }
601
602 /*
603  * delete a region from the global tree
604  */
605 static void delete_nommu_region(struct vm_region *region)
606 {
607         BUG_ON(!nommu_region_tree.rb_node);
608
609         validate_nommu_regions();
610         rb_erase(&region->vm_rb, &nommu_region_tree);
611         validate_nommu_regions();
612 }
613
614 /*
615  * free a contiguous series of pages
616  */
617 static void free_page_series(unsigned long from, unsigned long to)
618 {
619         for (; from < to; from += PAGE_SIZE) {
620                 struct page *page = virt_to_page(from);
621
622                 atomic_long_dec(&mmap_pages_allocated);
623                 put_page(page);
624         }
625 }
626
627 /*
628  * release a reference to a region
629  * - the caller must hold the region semaphore for writing, which this releases
630  * - the region may not have been added to the tree yet, in which case vm_top
631  *   will equal vm_start
632  */
633 static void __put_nommu_region(struct vm_region *region)
634         __releases(nommu_region_sem)
635 {
636         BUG_ON(!nommu_region_tree.rb_node);
637
638         if (--region->vm_usage == 0) {
639                 if (region->vm_top > region->vm_start)
640                         delete_nommu_region(region);
641                 up_write(&nommu_region_sem);
642
643                 if (region->vm_file)
644                         fput(region->vm_file);
645
646                 /* IO memory and memory shared directly out of the pagecache
647                  * from ramfs/tmpfs mustn't be released here */
648                 if (region->vm_flags & VM_MAPPED_COPY)
649                         free_page_series(region->vm_start, region->vm_top);
650                 kmem_cache_free(vm_region_jar, region);
651         } else {
652                 up_write(&nommu_region_sem);
653         }
654 }
655
656 /*
657  * release a reference to a region
658  */
659 static void put_nommu_region(struct vm_region *region)
660 {
661         down_write(&nommu_region_sem);
662         __put_nommu_region(region);
663 }
664
665 /*
666  * update protection on a vma
667  */
668 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
669 {
670 #ifdef CONFIG_MPU
671         struct mm_struct *mm = vma->vm_mm;
672         long start = vma->vm_start & PAGE_MASK;
673         while (start < vma->vm_end) {
674                 protect_page(mm, start, flags);
675                 start += PAGE_SIZE;
676         }
677         update_protections(mm);
678 #endif
679 }
680
681 /*
682  * add a VMA into a process's mm_struct in the appropriate place in the list
683  * and tree and add to the address space's page tree also if not an anonymous
684  * page
685  * - should be called with mm->mmap_sem held writelocked
686  */
687 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
688 {
689         struct vm_area_struct *pvma, *prev;
690         struct address_space *mapping;
691         struct rb_node **p, *parent, *rb_prev;
692
693         BUG_ON(!vma->vm_region);
694
695         mm->map_count++;
696         vma->vm_mm = mm;
697
698         protect_vma(vma, vma->vm_flags);
699
700         /* add the VMA to the mapping */
701         if (vma->vm_file) {
702                 mapping = vma->vm_file->f_mapping;
703
704                 i_mmap_lock_write(mapping);
705                 flush_dcache_mmap_lock(mapping);
706                 vma_interval_tree_insert(vma, &mapping->i_mmap);
707                 flush_dcache_mmap_unlock(mapping);
708                 i_mmap_unlock_write(mapping);
709         }
710
711         /* add the VMA to the tree */
712         parent = rb_prev = NULL;
713         p = &mm->mm_rb.rb_node;
714         while (*p) {
715                 parent = *p;
716                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
717
718                 /* sort by: start addr, end addr, VMA struct addr in that order
719                  * (the latter is necessary as we may get identical VMAs) */
720                 if (vma->vm_start < pvma->vm_start)
721                         p = &(*p)->rb_left;
722                 else if (vma->vm_start > pvma->vm_start) {
723                         rb_prev = parent;
724                         p = &(*p)->rb_right;
725                 } else if (vma->vm_end < pvma->vm_end)
726                         p = &(*p)->rb_left;
727                 else if (vma->vm_end > pvma->vm_end) {
728                         rb_prev = parent;
729                         p = &(*p)->rb_right;
730                 } else if (vma < pvma)
731                         p = &(*p)->rb_left;
732                 else if (vma > pvma) {
733                         rb_prev = parent;
734                         p = &(*p)->rb_right;
735                 } else
736                         BUG();
737         }
738
739         rb_link_node(&vma->vm_rb, parent, p);
740         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
741
742         /* add VMA to the VMA list also */
743         prev = NULL;
744         if (rb_prev)
745                 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
746
747         __vma_link_list(mm, vma, prev, parent);
748 }
749
750 /*
751  * delete a VMA from its owning mm_struct and address space
752  */
753 static void delete_vma_from_mm(struct vm_area_struct *vma)
754 {
755         int i;
756         struct address_space *mapping;
757         struct mm_struct *mm = vma->vm_mm;
758         struct task_struct *curr = current;
759
760         protect_vma(vma, 0);
761
762         mm->map_count--;
763         for (i = 0; i < VMACACHE_SIZE; i++) {
764                 /* if the vma is cached, invalidate the entire cache */
765                 if (curr->vmacache.vmas[i] == vma) {
766                         vmacache_invalidate(mm);
767                         break;
768                 }
769         }
770
771         /* remove the VMA from the mapping */
772         if (vma->vm_file) {
773                 mapping = vma->vm_file->f_mapping;
774
775                 i_mmap_lock_write(mapping);
776                 flush_dcache_mmap_lock(mapping);
777                 vma_interval_tree_remove(vma, &mapping->i_mmap);
778                 flush_dcache_mmap_unlock(mapping);
779                 i_mmap_unlock_write(mapping);
780         }
781
782         /* remove from the MM's tree and list */
783         rb_erase(&vma->vm_rb, &mm->mm_rb);
784
785         if (vma->vm_prev)
786                 vma->vm_prev->vm_next = vma->vm_next;
787         else
788                 mm->mmap = vma->vm_next;
789
790         if (vma->vm_next)
791                 vma->vm_next->vm_prev = vma->vm_prev;
792 }
793
794 /*
795  * destroy a VMA record
796  */
797 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
798 {
799         if (vma->vm_ops && vma->vm_ops->close)
800                 vma->vm_ops->close(vma);
801         if (vma->vm_file)
802                 fput(vma->vm_file);
803         put_nommu_region(vma->vm_region);
804         kmem_cache_free(vm_area_cachep, vma);
805 }
806
807 /*
808  * look up the first VMA in which addr resides, NULL if none
809  * - should be called with mm->mmap_sem at least held readlocked
810  */
811 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
812 {
813         struct vm_area_struct *vma;
814
815         /* check the cache first */
816         vma = vmacache_find(mm, addr);
817         if (likely(vma))
818                 return vma;
819
820         /* trawl the list (there may be multiple mappings in which addr
821          * resides) */
822         for (vma = mm->mmap; vma; vma = vma->vm_next) {
823                 if (vma->vm_start > addr)
824                         return NULL;
825                 if (vma->vm_end > addr) {
826                         vmacache_update(addr, vma);
827                         return vma;
828                 }
829         }
830
831         return NULL;
832 }
833 EXPORT_SYMBOL(find_vma);
834
835 /*
836  * find a VMA
837  * - we don't extend stack VMAs under NOMMU conditions
838  */
839 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
840 {
841         return find_vma(mm, addr);
842 }
843
844 /*
845  * expand a stack to a given address
846  * - not supported under NOMMU conditions
847  */
848 int expand_stack(struct vm_area_struct *vma, unsigned long address)
849 {
850         return -ENOMEM;
851 }
852
853 /*
854  * look up the first VMA exactly that exactly matches addr
855  * - should be called with mm->mmap_sem at least held readlocked
856  */
857 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
858                                              unsigned long addr,
859                                              unsigned long len)
860 {
861         struct vm_area_struct *vma;
862         unsigned long end = addr + len;
863
864         /* check the cache first */
865         vma = vmacache_find_exact(mm, addr, end);
866         if (vma)
867                 return vma;
868
869         /* trawl the list (there may be multiple mappings in which addr
870          * resides) */
871         for (vma = mm->mmap; vma; vma = vma->vm_next) {
872                 if (vma->vm_start < addr)
873                         continue;
874                 if (vma->vm_start > addr)
875                         return NULL;
876                 if (vma->vm_end == end) {
877                         vmacache_update(addr, vma);
878                         return vma;
879                 }
880         }
881
882         return NULL;
883 }
884
885 /*
886  * determine whether a mapping should be permitted and, if so, what sort of
887  * mapping we're capable of supporting
888  */
889 static int validate_mmap_request(struct file *file,
890                                  unsigned long addr,
891                                  unsigned long len,
892                                  unsigned long prot,
893                                  unsigned long flags,
894                                  unsigned long pgoff,
895                                  unsigned long *_capabilities)
896 {
897         unsigned long capabilities, rlen;
898         int ret;
899
900         /* do the simple checks first */
901         if (flags & MAP_FIXED)
902                 return -EINVAL;
903
904         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
905             (flags & MAP_TYPE) != MAP_SHARED)
906                 return -EINVAL;
907
908         if (!len)
909                 return -EINVAL;
910
911         /* Careful about overflows.. */
912         rlen = PAGE_ALIGN(len);
913         if (!rlen || rlen > TASK_SIZE)
914                 return -ENOMEM;
915
916         /* offset overflow? */
917         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
918                 return -EOVERFLOW;
919
920         if (file) {
921                 /* files must support mmap */
922                 if (!file->f_op->mmap)
923                         return -ENODEV;
924
925                 /* work out if what we've got could possibly be shared
926                  * - we support chardevs that provide their own "memory"
927                  * - we support files/blockdevs that are memory backed
928                  */
929                 if (file->f_op->mmap_capabilities) {
930                         capabilities = file->f_op->mmap_capabilities(file);
931                 } else {
932                         /* no explicit capabilities set, so assume some
933                          * defaults */
934                         switch (file_inode(file)->i_mode & S_IFMT) {
935                         case S_IFREG:
936                         case S_IFBLK:
937                                 capabilities = NOMMU_MAP_COPY;
938                                 break;
939
940                         case S_IFCHR:
941                                 capabilities =
942                                         NOMMU_MAP_DIRECT |
943                                         NOMMU_MAP_READ |
944                                         NOMMU_MAP_WRITE;
945                                 break;
946
947                         default:
948                                 return -EINVAL;
949                         }
950                 }
951
952                 /* eliminate any capabilities that we can't support on this
953                  * device */
954                 if (!file->f_op->get_unmapped_area)
955                         capabilities &= ~NOMMU_MAP_DIRECT;
956                 if (!(file->f_mode & FMODE_CAN_READ))
957                         capabilities &= ~NOMMU_MAP_COPY;
958
959                 /* The file shall have been opened with read permission. */
960                 if (!(file->f_mode & FMODE_READ))
961                         return -EACCES;
962
963                 if (flags & MAP_SHARED) {
964                         /* do checks for writing, appending and locking */
965                         if ((prot & PROT_WRITE) &&
966                             !(file->f_mode & FMODE_WRITE))
967                                 return -EACCES;
968
969                         if (IS_APPEND(file_inode(file)) &&
970                             (file->f_mode & FMODE_WRITE))
971                                 return -EACCES;
972
973                         if (locks_verify_locked(file))
974                                 return -EAGAIN;
975
976                         if (!(capabilities & NOMMU_MAP_DIRECT))
977                                 return -ENODEV;
978
979                         /* we mustn't privatise shared mappings */
980                         capabilities &= ~NOMMU_MAP_COPY;
981                 } else {
982                         /* we're going to read the file into private memory we
983                          * allocate */
984                         if (!(capabilities & NOMMU_MAP_COPY))
985                                 return -ENODEV;
986
987                         /* we don't permit a private writable mapping to be
988                          * shared with the backing device */
989                         if (prot & PROT_WRITE)
990                                 capabilities &= ~NOMMU_MAP_DIRECT;
991                 }
992
993                 if (capabilities & NOMMU_MAP_DIRECT) {
994                         if (((prot & PROT_READ)  && !(capabilities & NOMMU_MAP_READ))  ||
995                             ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
996                             ((prot & PROT_EXEC)  && !(capabilities & NOMMU_MAP_EXEC))
997                             ) {
998                                 capabilities &= ~NOMMU_MAP_DIRECT;
999                                 if (flags & MAP_SHARED) {
1000                                         pr_warn("MAP_SHARED not completely supported on !MMU\n");
1001                                         return -EINVAL;
1002                                 }
1003                         }
1004                 }
1005
1006                 /* handle executable mappings and implied executable
1007                  * mappings */
1008                 if (path_noexec(&file->f_path)) {
1009                         if (prot & PROT_EXEC)
1010                                 return -EPERM;
1011                 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1012                         /* handle implication of PROT_EXEC by PROT_READ */
1013                         if (current->personality & READ_IMPLIES_EXEC) {
1014                                 if (capabilities & NOMMU_MAP_EXEC)
1015                                         prot |= PROT_EXEC;
1016                         }
1017                 } else if ((prot & PROT_READ) &&
1018                          (prot & PROT_EXEC) &&
1019                          !(capabilities & NOMMU_MAP_EXEC)
1020                          ) {
1021                         /* backing file is not executable, try to copy */
1022                         capabilities &= ~NOMMU_MAP_DIRECT;
1023                 }
1024         } else {
1025                 /* anonymous mappings are always memory backed and can be
1026                  * privately mapped
1027                  */
1028                 capabilities = NOMMU_MAP_COPY;
1029
1030                 /* handle PROT_EXEC implication by PROT_READ */
1031                 if ((prot & PROT_READ) &&
1032                     (current->personality & READ_IMPLIES_EXEC))
1033                         prot |= PROT_EXEC;
1034         }
1035
1036         /* allow the security API to have its say */
1037         ret = security_mmap_addr(addr);
1038         if (ret < 0)
1039                 return ret;
1040
1041         /* looks okay */
1042         *_capabilities = capabilities;
1043         return 0;
1044 }
1045
1046 /*
1047  * we've determined that we can make the mapping, now translate what we
1048  * now know into VMA flags
1049  */
1050 static unsigned long determine_vm_flags(struct file *file,
1051                                         unsigned long prot,
1052                                         unsigned long flags,
1053                                         unsigned long capabilities)
1054 {
1055         unsigned long vm_flags;
1056
1057         vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1058         /* vm_flags |= mm->def_flags; */
1059
1060         if (!(capabilities & NOMMU_MAP_DIRECT)) {
1061                 /* attempt to share read-only copies of mapped file chunks */
1062                 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1063                 if (file && !(prot & PROT_WRITE))
1064                         vm_flags |= VM_MAYSHARE;
1065         } else {
1066                 /* overlay a shareable mapping on the backing device or inode
1067                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1068                  * romfs/cramfs */
1069                 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1070                 if (flags & MAP_SHARED)
1071                         vm_flags |= VM_SHARED;
1072         }
1073
1074         /* refuse to let anyone share private mappings with this process if
1075          * it's being traced - otherwise breakpoints set in it may interfere
1076          * with another untraced process
1077          */
1078         if ((flags & MAP_PRIVATE) && current->ptrace)
1079                 vm_flags &= ~VM_MAYSHARE;
1080
1081         return vm_flags;
1082 }
1083
1084 /*
1085  * set up a shared mapping on a file (the driver or filesystem provides and
1086  * pins the storage)
1087  */
1088 static int do_mmap_shared_file(struct vm_area_struct *vma)
1089 {
1090         int ret;
1091
1092         ret = call_mmap(vma->vm_file, vma);
1093         if (ret == 0) {
1094                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1095                 return 0;
1096         }
1097         if (ret != -ENOSYS)
1098                 return ret;
1099
1100         /* getting -ENOSYS indicates that direct mmap isn't possible (as
1101          * opposed to tried but failed) so we can only give a suitable error as
1102          * it's not possible to make a private copy if MAP_SHARED was given */
1103         return -ENODEV;
1104 }
1105
1106 /*
1107  * set up a private mapping or an anonymous shared mapping
1108  */
1109 static int do_mmap_private(struct vm_area_struct *vma,
1110                            struct vm_region *region,
1111                            unsigned long len,
1112                            unsigned long capabilities)
1113 {
1114         unsigned long total, point;
1115         void *base;
1116         int ret, order;
1117
1118         /* invoke the file's mapping function so that it can keep track of
1119          * shared mappings on devices or memory
1120          * - VM_MAYSHARE will be set if it may attempt to share
1121          */
1122         if (capabilities & NOMMU_MAP_DIRECT) {
1123                 ret = call_mmap(vma->vm_file, vma);
1124                 if (ret == 0) {
1125                         /* shouldn't return success if we're not sharing */
1126                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1127                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1128                         return 0;
1129                 }
1130                 if (ret != -ENOSYS)
1131                         return ret;
1132
1133                 /* getting an ENOSYS error indicates that direct mmap isn't
1134                  * possible (as opposed to tried but failed) so we'll try to
1135                  * make a private copy of the data and map that instead */
1136         }
1137
1138
1139         /* allocate some memory to hold the mapping
1140          * - note that this may not return a page-aligned address if the object
1141          *   we're allocating is smaller than a page
1142          */
1143         order = get_order(len);
1144         total = 1 << order;
1145         point = len >> PAGE_SHIFT;
1146
1147         /* we don't want to allocate a power-of-2 sized page set */
1148         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1149                 total = point;
1150
1151         base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1152         if (!base)
1153                 goto enomem;
1154
1155         atomic_long_add(total, &mmap_pages_allocated);
1156
1157         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1158         region->vm_start = (unsigned long) base;
1159         region->vm_end   = region->vm_start + len;
1160         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1161
1162         vma->vm_start = region->vm_start;
1163         vma->vm_end   = region->vm_start + len;
1164
1165         if (vma->vm_file) {
1166                 /* read the contents of a file into the copy */
1167                 mm_segment_t old_fs;
1168                 loff_t fpos;
1169
1170                 fpos = vma->vm_pgoff;
1171                 fpos <<= PAGE_SHIFT;
1172
1173                 old_fs = get_fs();
1174                 set_fs(KERNEL_DS);
1175                 ret = __vfs_read(vma->vm_file, base, len, &fpos);
1176                 set_fs(old_fs);
1177
1178                 if (ret < 0)
1179                         goto error_free;
1180
1181                 /* clear the last little bit */
1182                 if (ret < len)
1183                         memset(base + ret, 0, len - ret);
1184
1185         }
1186
1187         return 0;
1188
1189 error_free:
1190         free_page_series(region->vm_start, region->vm_top);
1191         region->vm_start = vma->vm_start = 0;
1192         region->vm_end   = vma->vm_end = 0;
1193         region->vm_top   = 0;
1194         return ret;
1195
1196 enomem:
1197         pr_err("Allocation of length %lu from process %d (%s) failed\n",
1198                len, current->pid, current->comm);
1199         show_free_areas(0, NULL);
1200         return -ENOMEM;
1201 }
1202
1203 /*
1204  * handle mapping creation for uClinux
1205  */
1206 unsigned long do_mmap(struct file *file,
1207                         unsigned long addr,
1208                         unsigned long len,
1209                         unsigned long prot,
1210                         unsigned long flags,
1211                         vm_flags_t vm_flags,
1212                         unsigned long pgoff,
1213                         unsigned long *populate,
1214                         struct list_head *uf)
1215 {
1216         struct vm_area_struct *vma;
1217         struct vm_region *region;
1218         struct rb_node *rb;
1219         unsigned long capabilities, result;
1220         int ret;
1221
1222         *populate = 0;
1223
1224         /* decide whether we should attempt the mapping, and if so what sort of
1225          * mapping */
1226         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1227                                     &capabilities);
1228         if (ret < 0)
1229                 return ret;
1230
1231         /* we ignore the address hint */
1232         addr = 0;
1233         len = PAGE_ALIGN(len);
1234
1235         /* we've determined that we can make the mapping, now translate what we
1236          * now know into VMA flags */
1237         vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1238
1239         /* we're going to need to record the mapping */
1240         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1241         if (!region)
1242                 goto error_getting_region;
1243
1244         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1245         if (!vma)
1246                 goto error_getting_vma;
1247
1248         region->vm_usage = 1;
1249         region->vm_flags = vm_flags;
1250         region->vm_pgoff = pgoff;
1251
1252         INIT_LIST_HEAD(&vma->anon_vma_chain);
1253         vma->vm_flags = vm_flags;
1254         vma->vm_pgoff = pgoff;
1255
1256         if (file) {
1257                 region->vm_file = get_file(file);
1258                 vma->vm_file = get_file(file);
1259         }
1260
1261         down_write(&nommu_region_sem);
1262
1263         /* if we want to share, we need to check for regions created by other
1264          * mmap() calls that overlap with our proposed mapping
1265          * - we can only share with a superset match on most regular files
1266          * - shared mappings on character devices and memory backed files are
1267          *   permitted to overlap inexactly as far as we are concerned for in
1268          *   these cases, sharing is handled in the driver or filesystem rather
1269          *   than here
1270          */
1271         if (vm_flags & VM_MAYSHARE) {
1272                 struct vm_region *pregion;
1273                 unsigned long pglen, rpglen, pgend, rpgend, start;
1274
1275                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1276                 pgend = pgoff + pglen;
1277
1278                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1279                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1280
1281                         if (!(pregion->vm_flags & VM_MAYSHARE))
1282                                 continue;
1283
1284                         /* search for overlapping mappings on the same file */
1285                         if (file_inode(pregion->vm_file) !=
1286                             file_inode(file))
1287                                 continue;
1288
1289                         if (pregion->vm_pgoff >= pgend)
1290                                 continue;
1291
1292                         rpglen = pregion->vm_end - pregion->vm_start;
1293                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1294                         rpgend = pregion->vm_pgoff + rpglen;
1295                         if (pgoff >= rpgend)
1296                                 continue;
1297
1298                         /* handle inexactly overlapping matches between
1299                          * mappings */
1300                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1301                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1302                                 /* new mapping is not a subset of the region */
1303                                 if (!(capabilities & NOMMU_MAP_DIRECT))
1304                                         goto sharing_violation;
1305                                 continue;
1306                         }
1307
1308                         /* we've found a region we can share */
1309                         pregion->vm_usage++;
1310                         vma->vm_region = pregion;
1311                         start = pregion->vm_start;
1312                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1313                         vma->vm_start = start;
1314                         vma->vm_end = start + len;
1315
1316                         if (pregion->vm_flags & VM_MAPPED_COPY)
1317                                 vma->vm_flags |= VM_MAPPED_COPY;
1318                         else {
1319                                 ret = do_mmap_shared_file(vma);
1320                                 if (ret < 0) {
1321                                         vma->vm_region = NULL;
1322                                         vma->vm_start = 0;
1323                                         vma->vm_end = 0;
1324                                         pregion->vm_usage--;
1325                                         pregion = NULL;
1326                                         goto error_just_free;
1327                                 }
1328                         }
1329                         fput(region->vm_file);
1330                         kmem_cache_free(vm_region_jar, region);
1331                         region = pregion;
1332                         result = start;
1333                         goto share;
1334                 }
1335
1336                 /* obtain the address at which to make a shared mapping
1337                  * - this is the hook for quasi-memory character devices to
1338                  *   tell us the location of a shared mapping
1339                  */
1340                 if (capabilities & NOMMU_MAP_DIRECT) {
1341                         addr = file->f_op->get_unmapped_area(file, addr, len,
1342                                                              pgoff, flags);
1343                         if (IS_ERR_VALUE(addr)) {
1344                                 ret = addr;
1345                                 if (ret != -ENOSYS)
1346                                         goto error_just_free;
1347
1348                                 /* the driver refused to tell us where to site
1349                                  * the mapping so we'll have to attempt to copy
1350                                  * it */
1351                                 ret = -ENODEV;
1352                                 if (!(capabilities & NOMMU_MAP_COPY))
1353                                         goto error_just_free;
1354
1355                                 capabilities &= ~NOMMU_MAP_DIRECT;
1356                         } else {
1357                                 vma->vm_start = region->vm_start = addr;
1358                                 vma->vm_end = region->vm_end = addr + len;
1359                         }
1360                 }
1361         }
1362
1363         vma->vm_region = region;
1364
1365         /* set up the mapping
1366          * - the region is filled in if NOMMU_MAP_DIRECT is still set
1367          */
1368         if (file && vma->vm_flags & VM_SHARED)
1369                 ret = do_mmap_shared_file(vma);
1370         else
1371                 ret = do_mmap_private(vma, region, len, capabilities);
1372         if (ret < 0)
1373                 goto error_just_free;
1374         add_nommu_region(region);
1375
1376         /* clear anonymous mappings that don't ask for uninitialized data */
1377         if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1378                 memset((void *)region->vm_start, 0,
1379                        region->vm_end - region->vm_start);
1380
1381         /* okay... we have a mapping; now we have to register it */
1382         result = vma->vm_start;
1383
1384         current->mm->total_vm += len >> PAGE_SHIFT;
1385
1386 share:
1387         add_vma_to_mm(current->mm, vma);
1388
1389         /* we flush the region from the icache only when the first executable
1390          * mapping of it is made  */
1391         if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1392                 flush_icache_range(region->vm_start, region->vm_end);
1393                 region->vm_icache_flushed = true;
1394         }
1395
1396         up_write(&nommu_region_sem);
1397
1398         return result;
1399
1400 error_just_free:
1401         up_write(&nommu_region_sem);
1402 error:
1403         if (region->vm_file)
1404                 fput(region->vm_file);
1405         kmem_cache_free(vm_region_jar, region);
1406         if (vma->vm_file)
1407                 fput(vma->vm_file);
1408         kmem_cache_free(vm_area_cachep, vma);
1409         return ret;
1410
1411 sharing_violation:
1412         up_write(&nommu_region_sem);
1413         pr_warn("Attempt to share mismatched mappings\n");
1414         ret = -EINVAL;
1415         goto error;
1416
1417 error_getting_vma:
1418         kmem_cache_free(vm_region_jar, region);
1419         pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1420                         len, current->pid);
1421         show_free_areas(0, NULL);
1422         return -ENOMEM;
1423
1424 error_getting_region:
1425         pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1426                         len, current->pid);
1427         show_free_areas(0, NULL);
1428         return -ENOMEM;
1429 }
1430
1431 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1432                 unsigned long, prot, unsigned long, flags,
1433                 unsigned long, fd, unsigned long, pgoff)
1434 {
1435         struct file *file = NULL;
1436         unsigned long retval = -EBADF;
1437
1438         audit_mmap_fd(fd, flags);
1439         if (!(flags & MAP_ANONYMOUS)) {
1440                 file = fget(fd);
1441                 if (!file)
1442                         goto out;
1443         }
1444
1445         flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1446
1447         retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1448
1449         if (file)
1450                 fput(file);
1451 out:
1452         return retval;
1453 }
1454
1455 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1456 struct mmap_arg_struct {
1457         unsigned long addr;
1458         unsigned long len;
1459         unsigned long prot;
1460         unsigned long flags;
1461         unsigned long fd;
1462         unsigned long offset;
1463 };
1464
1465 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1466 {
1467         struct mmap_arg_struct a;
1468
1469         if (copy_from_user(&a, arg, sizeof(a)))
1470                 return -EFAULT;
1471         if (offset_in_page(a.offset))
1472                 return -EINVAL;
1473
1474         return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1475                               a.offset >> PAGE_SHIFT);
1476 }
1477 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1478
1479 /*
1480  * split a vma into two pieces at address 'addr', a new vma is allocated either
1481  * for the first part or the tail.
1482  */
1483 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1484               unsigned long addr, int new_below)
1485 {
1486         struct vm_area_struct *new;
1487         struct vm_region *region;
1488         unsigned long npages;
1489
1490         /* we're only permitted to split anonymous regions (these should have
1491          * only a single usage on the region) */
1492         if (vma->vm_file)
1493                 return -ENOMEM;
1494
1495         if (mm->map_count >= sysctl_max_map_count)
1496                 return -ENOMEM;
1497
1498         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1499         if (!region)
1500                 return -ENOMEM;
1501
1502         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1503         if (!new) {
1504                 kmem_cache_free(vm_region_jar, region);
1505                 return -ENOMEM;
1506         }
1507
1508         /* most fields are the same, copy all, and then fixup */
1509         *new = *vma;
1510         *region = *vma->vm_region;
1511         new->vm_region = region;
1512
1513         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1514
1515         if (new_below) {
1516                 region->vm_top = region->vm_end = new->vm_end = addr;
1517         } else {
1518                 region->vm_start = new->vm_start = addr;
1519                 region->vm_pgoff = new->vm_pgoff += npages;
1520         }
1521
1522         if (new->vm_ops && new->vm_ops->open)
1523                 new->vm_ops->open(new);
1524
1525         delete_vma_from_mm(vma);
1526         down_write(&nommu_region_sem);
1527         delete_nommu_region(vma->vm_region);
1528         if (new_below) {
1529                 vma->vm_region->vm_start = vma->vm_start = addr;
1530                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1531         } else {
1532                 vma->vm_region->vm_end = vma->vm_end = addr;
1533                 vma->vm_region->vm_top = addr;
1534         }
1535         add_nommu_region(vma->vm_region);
1536         add_nommu_region(new->vm_region);
1537         up_write(&nommu_region_sem);
1538         add_vma_to_mm(mm, vma);
1539         add_vma_to_mm(mm, new);
1540         return 0;
1541 }
1542
1543 /*
1544  * shrink a VMA by removing the specified chunk from either the beginning or
1545  * the end
1546  */
1547 static int shrink_vma(struct mm_struct *mm,
1548                       struct vm_area_struct *vma,
1549                       unsigned long from, unsigned long to)
1550 {
1551         struct vm_region *region;
1552
1553         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1554          * and list */
1555         delete_vma_from_mm(vma);
1556         if (from > vma->vm_start)
1557                 vma->vm_end = from;
1558         else
1559                 vma->vm_start = to;
1560         add_vma_to_mm(mm, vma);
1561
1562         /* cut the backing region down to size */
1563         region = vma->vm_region;
1564         BUG_ON(region->vm_usage != 1);
1565
1566         down_write(&nommu_region_sem);
1567         delete_nommu_region(region);
1568         if (from > region->vm_start) {
1569                 to = region->vm_top;
1570                 region->vm_top = region->vm_end = from;
1571         } else {
1572                 region->vm_start = to;
1573         }
1574         add_nommu_region(region);
1575         up_write(&nommu_region_sem);
1576
1577         free_page_series(from, to);
1578         return 0;
1579 }
1580
1581 /*
1582  * release a mapping
1583  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1584  *   VMA, though it need not cover the whole VMA
1585  */
1586 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
1587 {
1588         struct vm_area_struct *vma;
1589         unsigned long end;
1590         int ret;
1591
1592         len = PAGE_ALIGN(len);
1593         if (len == 0)
1594                 return -EINVAL;
1595
1596         end = start + len;
1597
1598         /* find the first potentially overlapping VMA */
1599         vma = find_vma(mm, start);
1600         if (!vma) {
1601                 static int limit;
1602                 if (limit < 5) {
1603                         pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1604                                         current->pid, current->comm,
1605                                         start, start + len - 1);
1606                         limit++;
1607                 }
1608                 return -EINVAL;
1609         }
1610
1611         /* we're allowed to split an anonymous VMA but not a file-backed one */
1612         if (vma->vm_file) {
1613                 do {
1614                         if (start > vma->vm_start)
1615                                 return -EINVAL;
1616                         if (end == vma->vm_end)
1617                                 goto erase_whole_vma;
1618                         vma = vma->vm_next;
1619                 } while (vma);
1620                 return -EINVAL;
1621         } else {
1622                 /* the chunk must be a subset of the VMA found */
1623                 if (start == vma->vm_start && end == vma->vm_end)
1624                         goto erase_whole_vma;
1625                 if (start < vma->vm_start || end > vma->vm_end)
1626                         return -EINVAL;
1627                 if (offset_in_page(start))
1628                         return -EINVAL;
1629                 if (end != vma->vm_end && offset_in_page(end))
1630                         return -EINVAL;
1631                 if (start != vma->vm_start && end != vma->vm_end) {
1632                         ret = split_vma(mm, vma, start, 1);
1633                         if (ret < 0)
1634                                 return ret;
1635                 }
1636                 return shrink_vma(mm, vma, start, end);
1637         }
1638
1639 erase_whole_vma:
1640         delete_vma_from_mm(vma);
1641         delete_vma(mm, vma);
1642         return 0;
1643 }
1644 EXPORT_SYMBOL(do_munmap);
1645
1646 int vm_munmap(unsigned long addr, size_t len)
1647 {
1648         struct mm_struct *mm = current->mm;
1649         int ret;
1650
1651         down_write(&mm->mmap_sem);
1652         ret = do_munmap(mm, addr, len, NULL);
1653         up_write(&mm->mmap_sem);
1654         return ret;
1655 }
1656 EXPORT_SYMBOL(vm_munmap);
1657
1658 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1659 {
1660         return vm_munmap(addr, len);
1661 }
1662
1663 /*
1664  * release all the mappings made in a process's VM space
1665  */
1666 void exit_mmap(struct mm_struct *mm)
1667 {
1668         struct vm_area_struct *vma;
1669
1670         if (!mm)
1671                 return;
1672
1673         mm->total_vm = 0;
1674
1675         while ((vma = mm->mmap)) {
1676                 mm->mmap = vma->vm_next;
1677                 delete_vma_from_mm(vma);
1678                 delete_vma(mm, vma);
1679                 cond_resched();
1680         }
1681 }
1682
1683 int vm_brk(unsigned long addr, unsigned long len)
1684 {
1685         return -ENOMEM;
1686 }
1687
1688 /*
1689  * expand (or shrink) an existing mapping, potentially moving it at the same
1690  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1691  *
1692  * under NOMMU conditions, we only permit changing a mapping's size, and only
1693  * as long as it stays within the region allocated by do_mmap_private() and the
1694  * block is not shareable
1695  *
1696  * MREMAP_FIXED is not supported under NOMMU conditions
1697  */
1698 static unsigned long do_mremap(unsigned long addr,
1699                         unsigned long old_len, unsigned long new_len,
1700                         unsigned long flags, unsigned long new_addr)
1701 {
1702         struct vm_area_struct *vma;
1703
1704         /* insanity checks first */
1705         old_len = PAGE_ALIGN(old_len);
1706         new_len = PAGE_ALIGN(new_len);
1707         if (old_len == 0 || new_len == 0)
1708                 return (unsigned long) -EINVAL;
1709
1710         if (offset_in_page(addr))
1711                 return -EINVAL;
1712
1713         if (flags & MREMAP_FIXED && new_addr != addr)
1714                 return (unsigned long) -EINVAL;
1715
1716         vma = find_vma_exact(current->mm, addr, old_len);
1717         if (!vma)
1718                 return (unsigned long) -EINVAL;
1719
1720         if (vma->vm_end != vma->vm_start + old_len)
1721                 return (unsigned long) -EFAULT;
1722
1723         if (vma->vm_flags & VM_MAYSHARE)
1724                 return (unsigned long) -EPERM;
1725
1726         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1727                 return (unsigned long) -ENOMEM;
1728
1729         /* all checks complete - do it */
1730         vma->vm_end = vma->vm_start + new_len;
1731         return vma->vm_start;
1732 }
1733
1734 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1735                 unsigned long, new_len, unsigned long, flags,
1736                 unsigned long, new_addr)
1737 {
1738         unsigned long ret;
1739
1740         down_write(&current->mm->mmap_sem);
1741         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1742         up_write(&current->mm->mmap_sem);
1743         return ret;
1744 }
1745
1746 struct page *follow_page_mask(struct vm_area_struct *vma,
1747                               unsigned long address, unsigned int flags,
1748                               unsigned int *page_mask)
1749 {
1750         *page_mask = 0;
1751         return NULL;
1752 }
1753
1754 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1755                 unsigned long pfn, unsigned long size, pgprot_t prot)
1756 {
1757         if (addr != (pfn << PAGE_SHIFT))
1758                 return -EINVAL;
1759
1760         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1761         return 0;
1762 }
1763 EXPORT_SYMBOL(remap_pfn_range);
1764
1765 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1766 {
1767         unsigned long pfn = start >> PAGE_SHIFT;
1768         unsigned long vm_len = vma->vm_end - vma->vm_start;
1769
1770         pfn += vma->vm_pgoff;
1771         return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1772 }
1773 EXPORT_SYMBOL(vm_iomap_memory);
1774
1775 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1776                         unsigned long pgoff)
1777 {
1778         unsigned int size = vma->vm_end - vma->vm_start;
1779
1780         if (!(vma->vm_flags & VM_USERMAP))
1781                 return -EINVAL;
1782
1783         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1784         vma->vm_end = vma->vm_start + size;
1785
1786         return 0;
1787 }
1788 EXPORT_SYMBOL(remap_vmalloc_range);
1789
1790 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1791         unsigned long len, unsigned long pgoff, unsigned long flags)
1792 {
1793         return -ENOMEM;
1794 }
1795
1796 void unmap_mapping_range(struct address_space *mapping,
1797                          loff_t const holebegin, loff_t const holelen,
1798                          int even_cows)
1799 {
1800 }
1801 EXPORT_SYMBOL(unmap_mapping_range);
1802
1803 int filemap_fault(struct vm_fault *vmf)
1804 {
1805         BUG();
1806         return 0;
1807 }
1808 EXPORT_SYMBOL(filemap_fault);
1809
1810 void filemap_map_pages(struct vm_fault *vmf,
1811                 pgoff_t start_pgoff, pgoff_t end_pgoff)
1812 {
1813         BUG();
1814 }
1815 EXPORT_SYMBOL(filemap_map_pages);
1816
1817 int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1818                 unsigned long addr, void *buf, int len, unsigned int gup_flags)
1819 {
1820         struct vm_area_struct *vma;
1821         int write = gup_flags & FOLL_WRITE;
1822
1823         down_read(&mm->mmap_sem);
1824
1825         /* the access must start within one of the target process's mappings */
1826         vma = find_vma(mm, addr);
1827         if (vma) {
1828                 /* don't overrun this mapping */
1829                 if (addr + len >= vma->vm_end)
1830                         len = vma->vm_end - addr;
1831
1832                 /* only read or write mappings where it is permitted */
1833                 if (write && vma->vm_flags & VM_MAYWRITE)
1834                         copy_to_user_page(vma, NULL, addr,
1835                                          (void *) addr, buf, len);
1836                 else if (!write && vma->vm_flags & VM_MAYREAD)
1837                         copy_from_user_page(vma, NULL, addr,
1838                                             buf, (void *) addr, len);
1839                 else
1840                         len = 0;
1841         } else {
1842                 len = 0;
1843         }
1844
1845         up_read(&mm->mmap_sem);
1846
1847         return len;
1848 }
1849
1850 /**
1851  * @access_remote_vm - access another process' address space
1852  * @mm:         the mm_struct of the target address space
1853  * @addr:       start address to access
1854  * @buf:        source or destination buffer
1855  * @len:        number of bytes to transfer
1856  * @gup_flags:  flags modifying lookup behaviour
1857  *
1858  * The caller must hold a reference on @mm.
1859  */
1860 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1861                 void *buf, int len, unsigned int gup_flags)
1862 {
1863         return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
1864 }
1865
1866 /*
1867  * Access another process' address space.
1868  * - source/target buffer must be kernel space
1869  */
1870 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1871                 unsigned int gup_flags)
1872 {
1873         struct mm_struct *mm;
1874
1875         if (addr + len < addr)
1876                 return 0;
1877
1878         mm = get_task_mm(tsk);
1879         if (!mm)
1880                 return 0;
1881
1882         len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
1883
1884         mmput(mm);
1885         return len;
1886 }
1887 EXPORT_SYMBOL_GPL(access_process_vm);
1888
1889 /**
1890  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1891  * @inode: The inode to check
1892  * @size: The current filesize of the inode
1893  * @newsize: The proposed filesize of the inode
1894  *
1895  * Check the shared mappings on an inode on behalf of a shrinking truncate to
1896  * make sure that that any outstanding VMAs aren't broken and then shrink the
1897  * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1898  * automatically grant mappings that are too large.
1899  */
1900 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1901                                 size_t newsize)
1902 {
1903         struct vm_area_struct *vma;
1904         struct vm_region *region;
1905         pgoff_t low, high;
1906         size_t r_size, r_top;
1907
1908         low = newsize >> PAGE_SHIFT;
1909         high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1910
1911         down_write(&nommu_region_sem);
1912         i_mmap_lock_read(inode->i_mapping);
1913
1914         /* search for VMAs that fall within the dead zone */
1915         vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1916                 /* found one - only interested if it's shared out of the page
1917                  * cache */
1918                 if (vma->vm_flags & VM_SHARED) {
1919                         i_mmap_unlock_read(inode->i_mapping);
1920                         up_write(&nommu_region_sem);
1921                         return -ETXTBSY; /* not quite true, but near enough */
1922                 }
1923         }
1924
1925         /* reduce any regions that overlap the dead zone - if in existence,
1926          * these will be pointed to by VMAs that don't overlap the dead zone
1927          *
1928          * we don't check for any regions that start beyond the EOF as there
1929          * shouldn't be any
1930          */
1931         vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1932                 if (!(vma->vm_flags & VM_SHARED))
1933                         continue;
1934
1935                 region = vma->vm_region;
1936                 r_size = region->vm_top - region->vm_start;
1937                 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1938
1939                 if (r_top > newsize) {
1940                         region->vm_top -= r_top - newsize;
1941                         if (region->vm_end > region->vm_top)
1942                                 region->vm_end = region->vm_top;
1943                 }
1944         }
1945
1946         i_mmap_unlock_read(inode->i_mapping);
1947         up_write(&nommu_region_sem);
1948         return 0;
1949 }
1950
1951 /*
1952  * Initialise sysctl_user_reserve_kbytes.
1953  *
1954  * This is intended to prevent a user from starting a single memory hogging
1955  * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1956  * mode.
1957  *
1958  * The default value is min(3% of free memory, 128MB)
1959  * 128MB is enough to recover with sshd/login, bash, and top/kill.
1960  */
1961 static int __meminit init_user_reserve(void)
1962 {
1963         unsigned long free_kbytes;
1964
1965         free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1966
1967         sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1968         return 0;
1969 }
1970 subsys_initcall(init_user_reserve);
1971
1972 /*
1973  * Initialise sysctl_admin_reserve_kbytes.
1974  *
1975  * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1976  * to log in and kill a memory hogging process.
1977  *
1978  * Systems with more than 256MB will reserve 8MB, enough to recover
1979  * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1980  * only reserve 3% of free pages by default.
1981  */
1982 static int __meminit init_admin_reserve(void)
1983 {
1984         unsigned long free_kbytes;
1985
1986         free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1987
1988         sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1989         return 0;
1990 }
1991 subsys_initcall(init_admin_reserve);