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