1 /*
2 * linux/mm/nommu.c
3 *
4 * Replacement code for mm functions to support CPU's that don't
5 * have any form of memory management unit (thus no virtual memory).
6 *
7 * See Documentation/nommu-mmap.txt
8 *
9 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
10 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
13 * Copyright (c) 2007-2009 Paul Mundt <lethal@linux-sh.org>
14 */
15
16 #include <linux/module.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/swap.h>
20 #include <linux/file.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/tracehook.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mount.h>
29 #include <linux/personality.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32
33 #include <asm/uaccess.h>
34 #include <asm/tlb.h>
35 #include <asm/tlbflush.h>
36 #include "internal.h"
37
38 static inline __attribute__((format(printf, 1, 2)))
39 void no_printk(const char *fmt, ...)
40 {
41 }
42
43 #if 0
44 #define kenter(FMT, ...) \
45 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
46 #define kleave(FMT, ...) \
47 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
48 #define kdebug(FMT, ...) \
49 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
50 #else
51 #define kenter(FMT, ...) \
52 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
53 #define kleave(FMT, ...) \
54 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
55 #define kdebug(FMT, ...) \
56 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
57 #endif
58
59 #include "internal.h"
60
61 void *high_memory;
62 struct page *mem_map;
63 unsigned long max_mapnr;
64 unsigned long num_physpages;
65 struct percpu_counter vm_committed_as;
66 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
67 int sysctl_overcommit_ratio = 50; /* default is 50% */
68 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
69 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
70 int heap_stack_gap = 0;
71
72 atomic_long_t mmap_pages_allocated;
73
74 EXPORT_SYMBOL(mem_map);
75 EXPORT_SYMBOL(num_physpages);
76
77 /* list of mapped, potentially shareable regions */
78 static struct kmem_cache *vm_region_jar;
79 struct rb_root nommu_region_tree = RB_ROOT;
80 DECLARE_RWSEM(nommu_region_sem);
81
82 struct vm_operations_struct generic_file_vm_ops = {
83 };
84
85 /*
86 * Handle all mappings that got truncated by a "truncate()"
87 * system call.
88 *
89 * NOTE! We have to be ready to update the memory sharing
90 * between the file and the memory map for a potential last
91 * incomplete page. Ugly, but necessary.
92 */
93 int vmtruncate(struct inode *inode, loff_t offset)
94 {
95 struct address_space *mapping = inode->i_mapping;
96 unsigned long limit;
97
98 if (inode->i_size < offset)
99 goto do_expand;
100 i_size_write(inode, offset);
101
102 truncate_inode_pages(mapping, offset);
103 goto out_truncate;
104
105 do_expand:
106 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
107 if (limit != RLIM_INFINITY && offset > limit)
108 goto out_sig;
109 if (offset > inode->i_sb->s_maxbytes)
110 goto out;
111 i_size_write(inode, offset);
112
113 out_truncate:
114 if (inode->i_op->truncate)
115 inode->i_op->truncate(inode);
116 return 0;
117 out_sig:
118 send_sig(SIGXFSZ, current, 0);
119 out:
120 return -EFBIG;
121 }
122
123 EXPORT_SYMBOL(vmtruncate);
124
125 /*
126 * Return the total memory allocated for this pointer, not
127 * just what the caller asked for.
128 *
129 * Doesn't have to be accurate, i.e. may have races.
130 */
131 unsigned int kobjsize(const void *objp)
132 {
133 struct page *page;
134
135 /*
136 * If the object we have should not have ksize performed on it,
137 * return size of 0
138 */
139 if (!objp || !virt_addr_valid(objp))
140 return 0;
141
142 page = virt_to_head_page(objp);
143
144 /*
145 * If the allocator sets PageSlab, we know the pointer came from
146 * kmalloc().
147 */
148 if (PageSlab(page))
149 return ksize(objp);
150
151 /*
152 * If it's not a compound page, see if we have a matching VMA
153 * region. This test is intentionally done in reverse order,
154 * so if there's no VMA, we still fall through and hand back
155 * PAGE_SIZE for 0-order pages.
156 */
157 if (!PageCompound(page)) {
158 struct vm_area_struct *vma;
159
160 vma = find_vma(current->mm, (unsigned long)objp);
161 if (vma)
162 return vma->vm_end - vma->vm_start;
163 }
164
165 /*
166 * The ksize() function is only guaranteed to work for pointers
167 * returned by kmalloc(). So handle arbitrary pointers here.
168 */
169 return PAGE_SIZE << compound_order(page);
170 }
171
172 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
173 unsigned long start, int nr_pages, int flags,
174 struct page **pages, struct vm_area_struct **vmas)
175 {
176 struct vm_area_struct *vma;
177 unsigned long vm_flags;
178 int i;
179 int write = !!(flags & GUP_FLAGS_WRITE);
180 int force = !!(flags & GUP_FLAGS_FORCE);
181 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
182
183 /* calculate required read or write permissions.
184 * - if 'force' is set, we only require the "MAY" flags.
185 */
186 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
187 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
188
189 for (i = 0; i < nr_pages; i++) {
190 vma = find_vma(mm, start);
191 if (!vma)
192 goto finish_or_fault;
193
194 /* protect what we can, including chardevs */
195 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
196 (!ignore && !(vm_flags & vma->vm_flags)))
197 goto finish_or_fault;
198
199 if (pages) {
200 pages[i] = virt_to_page(start);
201 if (pages[i])
202 page_cache_get(pages[i]);
203 }
204 if (vmas)
205 vmas[i] = vma;
206 start += PAGE_SIZE;
207 }
208
209 return i;
210
211 finish_or_fault:
212 return i ? : -EFAULT;
213 }
214
215
216 /*
217 * get a list of pages in an address range belonging to the specified process
218 * and indicate the VMA that covers each page
219 * - this is potentially dodgy as we may end incrementing the page count of a
220 * slab page or a secondary page from a compound page
221 * - don't permit access to VMAs that don't support it, such as I/O mappings
222 */
223 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
224 unsigned long start, int nr_pages, int write, int force,
225 struct page **pages, struct vm_area_struct **vmas)
226 {
227 int flags = 0;
228
229 if (write)
230 flags |= GUP_FLAGS_WRITE;
231 if (force)
232 flags |= GUP_FLAGS_FORCE;
233
234 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
235 }
236 EXPORT_SYMBOL(get_user_pages);
237
238 /**
239 * follow_pfn - look up PFN at a user virtual address
240 * @vma: memory mapping
241 * @address: user virtual address
242 * @pfn: location to store found PFN
243 *
244 * Only IO mappings and raw PFN mappings are allowed.
245 *
246 * Returns zero and the pfn at @pfn on success, -ve otherwise.
247 */
248 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
249 unsigned long *pfn)
250 {
251 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
252 return -EINVAL;
253
254 *pfn = address >> PAGE_SHIFT;
255 return 0;
256 }
257 EXPORT_SYMBOL(follow_pfn);
258
259 DEFINE_RWLOCK(vmlist_lock);
260 struct vm_struct *vmlist;
261
262 void vfree(const void *addr)
263 {
264 kfree(addr);
265 }
266 EXPORT_SYMBOL(vfree);
267
268 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
269 {
270 /*
271 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
272 * returns only a logical address.
273 */
274 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
275 }
276 EXPORT_SYMBOL(__vmalloc);
277
278 void *vmalloc_user(unsigned long size)
279 {
280 void *ret;
281
282 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
283 PAGE_KERNEL);
284 if (ret) {
285 struct vm_area_struct *vma;
286
287 down_write(¤t->mm->mmap_sem);
288 vma = find_vma(current->mm, (unsigned long)ret);
289 if (vma)
290 vma->vm_flags |= VM_USERMAP;
291 up_write(¤t->mm->mmap_sem);
292 }
293
294 return ret;
295 }
296 EXPORT_SYMBOL(vmalloc_user);
297
298 struct page *vmalloc_to_page(const void *addr)
299 {
300 return virt_to_page(addr);
301 }
302 EXPORT_SYMBOL(vmalloc_to_page);
303
304 unsigned long vmalloc_to_pfn(const void *addr)
305 {
306 return page_to_pfn(virt_to_page(addr));
307 }
308 EXPORT_SYMBOL(vmalloc_to_pfn);
309
310 long vread(char *buf, char *addr, unsigned long count)
311 {
312 memcpy(buf, addr, count);
313 return count;
314 }
315
316 long vwrite(char *buf, char *addr, unsigned long count)
317 {
318 /* Don't allow overflow */
319 if ((unsigned long) addr + count < count)
320 count = -(unsigned long) addr;
321
322 memcpy(addr, buf, count);
323 return(count);
324 }
325
326 /*
327 * vmalloc - allocate virtually continguos memory
328 *
329 * @size: allocation size
330 *
331 * Allocate enough pages to cover @size from the page level
332 * allocator and map them into continguos kernel virtual space.
333 *
334 * For tight control over page level allocator and protection flags
335 * use __vmalloc() instead.
336 */
337 void *vmalloc(unsigned long size)
338 {
339 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
340 }
341 EXPORT_SYMBOL(vmalloc);
342
343 void *vmalloc_node(unsigned long size, int node)
344 {
345 return vmalloc(size);
346 }
347 EXPORT_SYMBOL(vmalloc_node);
348
349 #ifndef PAGE_KERNEL_EXEC
350 # define PAGE_KERNEL_EXEC PAGE_KERNEL
351 #endif
352
353 /**
354 * vmalloc_exec - allocate virtually contiguous, executable memory
355 * @size: allocation size
356 *
357 * Kernel-internal function to allocate enough pages to cover @size
358 * the page level allocator and map them into contiguous and
359 * executable kernel virtual space.
360 *
361 * For tight control over page level allocator and protection flags
362 * use __vmalloc() instead.
363 */
364
365 void *vmalloc_exec(unsigned long size)
366 {
367 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
368 }
369
370 /**
371 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
372 * @size: allocation size
373 *
374 * Allocate enough 32bit PA addressable pages to cover @size from the
375 * page level allocator and map them into continguos kernel virtual space.
376 */
377 void *vmalloc_32(unsigned long size)
378 {
379 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
380 }
381 EXPORT_SYMBOL(vmalloc_32);
382
383 /**
384 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
385 * @size: allocation size
386 *
387 * The resulting memory area is 32bit addressable and zeroed so it can be
388 * mapped to userspace without leaking data.
389 *
390 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
391 * remap_vmalloc_range() are permissible.
392 */
393 void *vmalloc_32_user(unsigned long size)
394 {
395 /*
396 * We'll have to sort out the ZONE_DMA bits for 64-bit,
397 * but for now this can simply use vmalloc_user() directly.
398 */
399 return vmalloc_user(size);
400 }
401 EXPORT_SYMBOL(vmalloc_32_user);
402
403 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
404 {
405 BUG();
406 return NULL;
407 }
408 EXPORT_SYMBOL(vmap);
409
410 void vunmap(const void *addr)
411 {
412 BUG();
413 }
414 EXPORT_SYMBOL(vunmap);
415
416 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
417 {
418 BUG();
419 return NULL;
420 }
421 EXPORT_SYMBOL(vm_map_ram);
422
423 void vm_unmap_ram(const void *mem, unsigned int count)
424 {
425 BUG();
426 }
427 EXPORT_SYMBOL(vm_unmap_ram);
428
429 void vm_unmap_aliases(void)
430 {
431 }
432 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
433
434 /*
435 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
436 * have one.
437 */
438 void __attribute__((weak)) vmalloc_sync_all(void)
439 {
440 }
441
442 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
443 struct page *page)
444 {
445 return -EINVAL;
446 }
447 EXPORT_SYMBOL(vm_insert_page);
448
449 /*
450 * sys_brk() for the most part doesn't need the global kernel
451 * lock, except when an application is doing something nasty
452 * like trying to un-brk an area that has already been mapped
453 * to a regular file. in this case, the unmapping will need
454 * to invoke file system routines that need the global lock.
455 */
456 SYSCALL_DEFINE1(brk, unsigned long, brk)
457 {
458 struct mm_struct *mm = current->mm;
459
460 if (brk < mm->start_brk || brk > mm->context.end_brk)
461 return mm->brk;
462
463 if (mm->brk == brk)
464 return mm->brk;
465
466 /*
467 * Always allow shrinking brk
468 */
469 if (brk <= mm->brk) {
470 mm->brk = brk;
471 return brk;
472 }
473
474 /*
475 * Ok, looks good - let it rip.
476 */
477 return mm->brk = brk;
478 }
479
480 /*
481 * initialise the VMA and region record slabs
482 */
483 void __init mmap_init(void)
484 {
485 int ret;
486
487 ret = percpu_counter_init(&vm_committed_as, 0);
488 VM_BUG_ON(ret);
489 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
490 }
491
492 /*
493 * validate the region tree
494 * - the caller must hold the region lock
495 */
496 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
497 static noinline void validate_nommu_regions(void)
498 {
499 struct vm_region *region, *last;
500 struct rb_node *p, *lastp;
501
502 lastp = rb_first(&nommu_region_tree);
503 if (!lastp)
504 return;
505
506 last = rb_entry(lastp, struct vm_region, vm_rb);
507 BUG_ON(unlikely(last->vm_end <= last->vm_start));
508 BUG_ON(unlikely(last->vm_top < last->vm_end));
509
510 while ((p = rb_next(lastp))) {
511 region = rb_entry(p, struct vm_region, vm_rb);
512 last = rb_entry(lastp, struct vm_region, vm_rb);
513
514 BUG_ON(unlikely(region->vm_end <= region->vm_start));
515 BUG_ON(unlikely(region->vm_top < region->vm_end));
516 BUG_ON(unlikely(region->vm_start < last->vm_top));
517
518 lastp = p;
519 }
520 }
521 #else
522 static void validate_nommu_regions(void)
523 {
524 }
525 #endif
526
527 /*
528 * add a region into the global tree
529 */
530 static void add_nommu_region(struct vm_region *region)
531 {
532 struct vm_region *pregion;
533 struct rb_node **p, *parent;
534
535 validate_nommu_regions();
536
537 parent = NULL;
538 p = &nommu_region_tree.rb_node;
539 while (*p) {
540 parent = *p;
541 pregion = rb_entry(parent, struct vm_region, vm_rb);
542 if (region->vm_start < pregion->vm_start)
543 p = &(*p)->rb_left;
544 else if (region->vm_start > pregion->vm_start)
545 p = &(*p)->rb_right;
546 else if (pregion == region)
547 return;
548 else
549 BUG();
550 }
551
552 rb_link_node(®ion->vm_rb, parent, p);
553 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
554
555 validate_nommu_regions();
556 }
557
558 /*
559 * delete a region from the global tree
560 */
561 static void delete_nommu_region(struct vm_region *region)
562 {
563 BUG_ON(!nommu_region_tree.rb_node);
564
565 validate_nommu_regions();
566 rb_erase(®ion->vm_rb, &nommu_region_tree);
567 validate_nommu_regions();
568 }
569
570 /*
571 * free a contiguous series of pages
572 */
573 static void free_page_series(unsigned long from, unsigned long to)
574 {
575 for (; from < to; from += PAGE_SIZE) {
576 struct page *page = virt_to_page(from);
577
578 kdebug("- free %lx", from);
579 atomic_long_dec(&mmap_pages_allocated);
580 if (page_count(page) != 1)
581 kdebug("free page %p: refcount not one: %d",
582 page, page_count(page));
583 put_page(page);
584 }
585 }
586
587 /*
588 * release a reference to a region
589 * - the caller must hold the region semaphore for writing, which this releases
590 * - the region may not have been added to the tree yet, in which case vm_top
591 * will equal vm_start
592 */
593 static void __put_nommu_region(struct vm_region *region)
594 __releases(nommu_region_sem)
595 {
596 kenter("%p{%d}", region, atomic_read(®ion->vm_usage));
597
598 BUG_ON(!nommu_region_tree.rb_node);
599
600 if (atomic_dec_and_test(®ion->vm_usage)) {
601 if (region->vm_top > region->vm_start)
602 delete_nommu_region(region);
603 up_write(&nommu_region_sem);
604
605 if (region->vm_file)
606 fput(region->vm_file);
607
608 /* IO memory and memory shared directly out of the pagecache
609 * from ramfs/tmpfs mustn't be released here */
610 if (region->vm_flags & VM_MAPPED_COPY) {
611 kdebug("free series");
612 free_page_series(region->vm_start, region->vm_top);
613 }
614 kmem_cache_free(vm_region_jar, region);
615 } else {
616 up_write(&nommu_region_sem);
617 }
618 }
619
620 /*
621 * release a reference to a region
622 */
623 static void put_nommu_region(struct vm_region *region)
624 {
625 down_write(&nommu_region_sem);
626 __put_nommu_region(region);
627 }
628
629 /*
630 * add a VMA into a process's mm_struct in the appropriate place in the list
631 * and tree and add to the address space's page tree also if not an anonymous
632 * page
633 * - should be called with mm->mmap_sem held writelocked
634 */
635 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
636 {
637 struct vm_area_struct *pvma, **pp;
638 struct address_space *mapping;
639 struct rb_node **p, *parent;
640
641 kenter(",%p", vma);
642
643 BUG_ON(!vma->vm_region);
644
645 mm->map_count++;
646 vma->vm_mm = mm;
647
648 /* add the VMA to the mapping */
649 if (vma->vm_file) {
650 mapping = vma->vm_file->f_mapping;
651
652 flush_dcache_mmap_lock(mapping);
653 vma_prio_tree_insert(vma, &mapping->i_mmap);
654 flush_dcache_mmap_unlock(mapping);
655 }
656
657 /* add the VMA to the tree */
658 parent = NULL;
659 p = &mm->mm_rb.rb_node;
660 while (*p) {
661 parent = *p;
662 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
663
664 /* sort by: start addr, end addr, VMA struct addr in that order
665 * (the latter is necessary as we may get identical VMAs) */
666 if (vma->vm_start < pvma->vm_start)
667 p = &(*p)->rb_left;
668 else if (vma->vm_start > pvma->vm_start)
669 p = &(*p)->rb_right;
670 else if (vma->vm_end < pvma->vm_end)
671 p = &(*p)->rb_left;
672 else if (vma->vm_end > pvma->vm_end)
673 p = &(*p)->rb_right;
674 else if (vma < pvma)
675 p = &(*p)->rb_left;
676 else if (vma > pvma)
677 p = &(*p)->rb_right;
678 else
679 BUG();
680 }
681
682 rb_link_node(&vma->vm_rb, parent, p);
683 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
684
685 /* add VMA to the VMA list also */
686 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
687 if (pvma->vm_start > vma->vm_start)
688 break;
689 if (pvma->vm_start < vma->vm_start)
690 continue;
691 if (pvma->vm_end < vma->vm_end)
692 break;
693 }
694
695 vma->vm_next = *pp;
696 *pp = vma;
697 }
698
699 /*
700 * delete a VMA from its owning mm_struct and address space
701 */
702 static void delete_vma_from_mm(struct vm_area_struct *vma)
703 {
704 struct vm_area_struct **pp;
705 struct address_space *mapping;
706 struct mm_struct *mm = vma->vm_mm;
707
708 kenter("%p", vma);
709
710 mm->map_count--;
711 if (mm->mmap_cache == vma)
712 mm->mmap_cache = NULL;
713
714 /* remove the VMA from the mapping */
715 if (vma->vm_file) {
716 mapping = vma->vm_file->f_mapping;
717
718 flush_dcache_mmap_lock(mapping);
719 vma_prio_tree_remove(vma, &mapping->i_mmap);
720 flush_dcache_mmap_unlock(mapping);
721 }
722
723 /* remove from the MM's tree and list */
724 rb_erase(&vma->vm_rb, &mm->mm_rb);
725 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
726 if (*pp == vma) {
727 *pp = vma->vm_next;
728 break;
729 }
730 }
731
732 vma->vm_mm = NULL;
733 }
734
735 /*
736 * destroy a VMA record
737 */
738 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
739 {
740 kenter("%p", vma);
741 if (vma->vm_ops && vma->vm_ops->close)
742 vma->vm_ops->close(vma);
743 if (vma->vm_file) {
744 fput(vma->vm_file);
745 if (vma->vm_flags & VM_EXECUTABLE)
746 removed_exe_file_vma(mm);
747 }
748 put_nommu_region(vma->vm_region);
749 kmem_cache_free(vm_area_cachep, vma);
750 }
751
752 /*
753 * look up the first VMA in which addr resides, NULL if none
754 * - should be called with mm->mmap_sem at least held readlocked
755 */
756 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
757 {
758 struct vm_area_struct *vma;
759 struct rb_node *n = mm->mm_rb.rb_node;
760
761 /* check the cache first */
762 vma = mm->mmap_cache;
763 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
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 return NULL;
772 if (vma->vm_end > addr) {
773 mm->mmap_cache = vma;
774 return vma;
775 }
776 }
777
778 return NULL;
779 }
780 EXPORT_SYMBOL(find_vma);
781
782 /*
783 * find a VMA
784 * - we don't extend stack VMAs under NOMMU conditions
785 */
786 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
787 {
788 return find_vma(mm, addr);
789 }
790
791 /*
792 * expand a stack to a given address
793 * - not supported under NOMMU conditions
794 */
795 int expand_stack(struct vm_area_struct *vma, unsigned long address)
796 {
797 return -ENOMEM;
798 }
799
800 /*
801 * look up the first VMA exactly that exactly matches addr
802 * - should be called with mm->mmap_sem at least held readlocked
803 */
804 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
805 unsigned long addr,
806 unsigned long len)
807 {
808 struct vm_area_struct *vma;
809 struct rb_node *n = mm->mm_rb.rb_node;
810 unsigned long end = addr + len;
811
812 /* check the cache first */
813 vma = mm->mmap_cache;
814 if (vma && vma->vm_start == addr && vma->vm_end == end)
815 return vma;
816
817 /* trawl the tree (there may be multiple mappings in which addr
818 * resides) */
819 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
820 vma = rb_entry(n, struct vm_area_struct, vm_rb);
821 if (vma->vm_start < addr)
822 continue;
823 if (vma->vm_start > addr)
824 return NULL;
825 if (vma->vm_end == end) {
826 mm->mmap_cache = vma;
827 return vma;
828 }
829 }
830
831 return NULL;
832 }
833
834 /*
835 * determine whether a mapping should be permitted and, if so, what sort of
836 * mapping we're capable of supporting
837 */
838 static int validate_mmap_request(struct file *file,
839 unsigned long addr,
840 unsigned long len,
841 unsigned long prot,
842 unsigned long flags,
843 unsigned long pgoff,
844 unsigned long *_capabilities)
845 {
846 unsigned long capabilities, rlen;
847 unsigned long reqprot = prot;
848 int ret;
849
850 /* do the simple checks first */
851 if (flags & MAP_FIXED || addr) {
852 printk(KERN_DEBUG
853 "%d: Can't do fixed-address/overlay mmap of RAM\n",
854 current->pid);
855 return -EINVAL;
856 }
857
858 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
859 (flags & MAP_TYPE) != MAP_SHARED)
860 return -EINVAL;
861
862 if (!len)
863 return -EINVAL;
864
865 /* Careful about overflows.. */
866 rlen = PAGE_ALIGN(len);
867 if (!rlen || rlen > TASK_SIZE)
868 return -ENOMEM;
869
870 /* offset overflow? */
871 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
872 return -EOVERFLOW;
873
874 if (file) {
875 /* validate file mapping requests */
876 struct address_space *mapping;
877
878 /* files must support mmap */
879 if (!file->f_op || !file->f_op->mmap)
880 return -ENODEV;
881
882 /* work out if what we've got could possibly be shared
883 * - we support chardevs that provide their own "memory"
884 * - we support files/blockdevs that are memory backed
885 */
886 mapping = file->f_mapping;
887 if (!mapping)
888 mapping = file->f_path.dentry->d_inode->i_mapping;
889
890 capabilities = 0;
891 if (mapping && mapping->backing_dev_info)
892 capabilities = mapping->backing_dev_info->capabilities;
893
894 if (!capabilities) {
895 /* no explicit capabilities set, so assume some
896 * defaults */
897 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
898 case S_IFREG:
899 case S_IFBLK:
900 capabilities = BDI_CAP_MAP_COPY;
901 break;
902
903 case S_IFCHR:
904 capabilities =
905 BDI_CAP_MAP_DIRECT |
906 BDI_CAP_READ_MAP |
907 BDI_CAP_WRITE_MAP;
908 break;
909
910 default:
911 return -EINVAL;
912 }
913 }
914
915 /* eliminate any capabilities that we can't support on this
916 * device */
917 if (!file->f_op->get_unmapped_area)
918 capabilities &= ~BDI_CAP_MAP_DIRECT;
919 if (!file->f_op->read)
920 capabilities &= ~BDI_CAP_MAP_COPY;
921
922 /* The file shall have been opened with read permission. */
923 if (!(file->f_mode & FMODE_READ))
924 return -EACCES;
925
926 if (flags & MAP_SHARED) {
927 /* do checks for writing, appending and locking */
928 if ((prot & PROT_WRITE) &&
929 !(file->f_mode & FMODE_WRITE))
930 return -EACCES;
931
932 if (IS_APPEND(file->f_path.dentry->d_inode) &&
933 (file->f_mode & FMODE_WRITE))
934 return -EACCES;
935
936 if (locks_verify_locked(file->f_path.dentry->d_inode))
937 return -EAGAIN;
938
939 if (!(capabilities & BDI_CAP_MAP_DIRECT))
940 return -ENODEV;
941
942 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
943 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
944 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
945 ) {
946 printk("MAP_SHARED not completely supported on !MMU\n");
947 return -EINVAL;
948 }
949
950 /* we mustn't privatise shared mappings */
951 capabilities &= ~BDI_CAP_MAP_COPY;
952 }
953 else {
954 /* we're going to read the file into private memory we
955 * allocate */
956 if (!(capabilities & BDI_CAP_MAP_COPY))
957 return -ENODEV;
958
959 /* we don't permit a private writable mapping to be
960 * shared with the backing device */
961 if (prot & PROT_WRITE)
962 capabilities &= ~BDI_CAP_MAP_DIRECT;
963 }
964
965 /* handle executable mappings and implied executable
966 * mappings */
967 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
968 if (prot & PROT_EXEC)
969 return -EPERM;
970 }
971 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
972 /* handle implication of PROT_EXEC by PROT_READ */
973 if (current->personality & READ_IMPLIES_EXEC) {
974 if (capabilities & BDI_CAP_EXEC_MAP)
975 prot |= PROT_EXEC;
976 }
977 }
978 else if ((prot & PROT_READ) &&
979 (prot & PROT_EXEC) &&
980 !(capabilities & BDI_CAP_EXEC_MAP)
981 ) {
982 /* backing file is not executable, try to copy */
983 capabilities &= ~BDI_CAP_MAP_DIRECT;
984 }
985 }
986 else {
987 /* anonymous mappings are always memory backed and can be
988 * privately mapped
989 */
990 capabilities = BDI_CAP_MAP_COPY;
991
992 /* handle PROT_EXEC implication by PROT_READ */
993 if ((prot & PROT_READ) &&
994 (current->personality & READ_IMPLIES_EXEC))
995 prot |= PROT_EXEC;
996 }
997
998 /* allow the security API to have its say */
999 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1000 if (ret < 0)
1001 return ret;
1002
1003 /* looks okay */
1004 *_capabilities = capabilities;
1005 return 0;
1006 }
1007
1008 /*
1009 * we've determined that we can make the mapping, now translate what we
1010 * now know into VMA flags
1011 */
1012 static unsigned long determine_vm_flags(struct file *file,
1013 unsigned long prot,
1014 unsigned long flags,
1015 unsigned long capabilities)
1016 {
1017 unsigned long vm_flags;
1018
1019 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1020 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1021 /* vm_flags |= mm->def_flags; */
1022
1023 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1024 /* attempt to share read-only copies of mapped file chunks */
1025 if (file && !(prot & PROT_WRITE))
1026 vm_flags |= VM_MAYSHARE;
1027 }
1028 else {
1029 /* overlay a shareable mapping on the backing device or inode
1030 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1031 * romfs/cramfs */
1032 if (flags & MAP_SHARED)
1033 vm_flags |= VM_MAYSHARE | VM_SHARED;
1034 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1035 vm_flags |= VM_MAYSHARE;
1036 }
1037
1038 /* refuse to let anyone share private mappings with this process if
1039 * it's being traced - otherwise breakpoints set in it may interfere
1040 * with another untraced process
1041 */
1042 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1043 vm_flags &= ~VM_MAYSHARE;
1044
1045 return vm_flags;
1046 }
1047
1048 /*
1049 * set up a shared mapping on a file (the driver or filesystem provides and
1050 * pins the storage)
1051 */
1052 static int do_mmap_shared_file(struct vm_area_struct *vma)
1053 {
1054 int ret;
1055
1056 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1057 if (ret == 0) {
1058 vma->vm_region->vm_top = vma->vm_region->vm_end;
1059 return 0;
1060 }
1061 if (ret != -ENOSYS)
1062 return ret;
1063
1064 /* getting an ENOSYS error indicates that direct mmap isn't
1065 * possible (as opposed to tried but failed) so we'll fall
1066 * through to making a private copy of the data and mapping
1067 * that if we can */
1068 return -ENODEV;
1069 }
1070
1071 /*
1072 * set up a private mapping or an anonymous shared mapping
1073 */
1074 static int do_mmap_private(struct vm_area_struct *vma,
1075 struct vm_region *region,
1076 unsigned long len,
1077 unsigned long capabilities)
1078 {
1079 struct page *pages;
1080 unsigned long total, point, n, rlen;
1081 void *base;
1082 int ret, order;
1083
1084 /* invoke the file's mapping function so that it can keep track of
1085 * shared mappings on devices or memory
1086 * - VM_MAYSHARE will be set if it may attempt to share
1087 */
1088 if (capabilities & BDI_CAP_MAP_DIRECT) {
1089 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1090 if (ret == 0) {
1091 /* shouldn't return success if we're not sharing */
1092 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1093 vma->vm_region->vm_top = vma->vm_region->vm_end;
1094 return 0;
1095 }
1096 if (ret != -ENOSYS)
1097 return ret;
1098
1099 /* getting an ENOSYS error indicates that direct mmap isn't
1100 * possible (as opposed to tried but failed) so we'll try to
1101 * make a private copy of the data and map that instead */
1102 }
1103
1104 rlen = PAGE_ALIGN(len);
1105
1106 /* allocate some memory to hold the mapping
1107 * - note that this may not return a page-aligned address if the object
1108 * we're allocating is smaller than a page
1109 */
1110 order = get_order(rlen);
1111 kdebug("alloc order %d for %lx", order, len);
1112
1113 pages = alloc_pages(GFP_KERNEL, order);
1114 if (!pages)
1115 goto enomem;
1116
1117 total = 1 << order;
1118 atomic_long_add(total, &mmap_pages_allocated);
1119
1120 point = rlen >> PAGE_SHIFT;
1121
1122 /* we allocated a power-of-2 sized page set, so we may want to trim off
1123 * the excess */
1124 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1125 while (total > point) {
1126 order = ilog2(total - point);
1127 n = 1 << order;
1128 kdebug("shave %lu/%lu @%lu", n, total - point, total);
1129 atomic_long_sub(n, &mmap_pages_allocated);
1130 total -= n;
1131 set_page_refcounted(pages + total);
1132 __free_pages(pages + total, order);
1133 }
1134 }
1135
1136 for (point = 1; point < total; point++)
1137 set_page_refcounted(&pages[point]);
1138
1139 base = page_address(pages);
1140 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1141 region->vm_start = (unsigned long) base;
1142 region->vm_end = region->vm_start + rlen;
1143 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1144
1145 vma->vm_start = region->vm_start;
1146 vma->vm_end = region->vm_start + len;
1147
1148 if (vma->vm_file) {
1149 /* read the contents of a file into the copy */
1150 mm_segment_t old_fs;
1151 loff_t fpos;
1152
1153 fpos = vma->vm_pgoff;
1154 fpos <<= PAGE_SHIFT;
1155
1156 old_fs = get_fs();
1157 set_fs(KERNEL_DS);
1158 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1159 set_fs(old_fs);
1160
1161 if (ret < 0)
1162 goto error_free;
1163
1164 /* clear the last little bit */
1165 if (ret < rlen)
1166 memset(base + ret, 0, rlen - ret);
1167
1168 } else {
1169 /* if it's an anonymous mapping, then just clear it */
1170 memset(base, 0, rlen);
1171 }
1172
1173 return 0;
1174
1175 error_free:
1176 free_page_series(region->vm_start, region->vm_end);
1177 region->vm_start = vma->vm_start = 0;
1178 region->vm_end = vma->vm_end = 0;
1179 region->vm_top = 0;
1180 return ret;
1181
1182 enomem:
1183 printk("Allocation of length %lu from process %d (%s) failed\n",
1184 len, current->pid, current->comm);
1185 show_free_areas();
1186 return -ENOMEM;
1187 }
1188
1189 /*
1190 * handle mapping creation for uClinux
1191 */
1192 unsigned long do_mmap_pgoff(struct file *file,
1193 unsigned long addr,
1194 unsigned long len,
1195 unsigned long prot,
1196 unsigned long flags,
1197 unsigned long pgoff)
1198 {
1199 struct vm_area_struct *vma;
1200 struct vm_region *region;
1201 struct rb_node *rb;
1202 unsigned long capabilities, vm_flags, result;
1203 int ret;
1204
1205 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1206
1207 if (!(flags & MAP_FIXED))
1208 addr = round_hint_to_min(addr);
1209
1210 /* decide whether we should attempt the mapping, and if so what sort of
1211 * mapping */
1212 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1213 &capabilities);
1214 if (ret < 0) {
1215 kleave(" = %d [val]", ret);
1216 return ret;
1217 }
1218
1219 /* we've determined that we can make the mapping, now translate what we
1220 * now know into VMA flags */
1221 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1222
1223 /* we're going to need to record the mapping */
1224 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1225 if (!region)
1226 goto error_getting_region;
1227
1228 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1229 if (!vma)
1230 goto error_getting_vma;
1231
1232 atomic_set(®ion->vm_usage, 1);
1233 region->vm_flags = vm_flags;
1234 region->vm_pgoff = pgoff;
1235
1236 INIT_LIST_HEAD(&vma->anon_vma_node);
1237 vma->vm_flags = vm_flags;
1238 vma->vm_pgoff = pgoff;
1239
1240 if (file) {
1241 region->vm_file = file;
1242 get_file(file);
1243 vma->vm_file = file;
1244 get_file(file);
1245 if (vm_flags & VM_EXECUTABLE) {
1246 added_exe_file_vma(current->mm);
1247 vma->vm_mm = current->mm;
1248 }
1249 }
1250
1251 down_write(&nommu_region_sem);
1252
1253 /* if we want to share, we need to check for regions created by other
1254 * mmap() calls that overlap with our proposed mapping
1255 * - we can only share with a superset match on most regular files
1256 * - shared mappings on character devices and memory backed files are
1257 * permitted to overlap inexactly as far as we are concerned for in
1258 * these cases, sharing is handled in the driver or filesystem rather
1259 * than here
1260 */
1261 if (vm_flags & VM_MAYSHARE) {
1262 struct vm_region *pregion;
1263 unsigned long pglen, rpglen, pgend, rpgend, start;
1264
1265 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1266 pgend = pgoff + pglen;
1267
1268 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1269 pregion = rb_entry(rb, struct vm_region, vm_rb);
1270
1271 if (!(pregion->vm_flags & VM_MAYSHARE))
1272 continue;
1273
1274 /* search for overlapping mappings on the same file */
1275 if (pregion->vm_file->f_path.dentry->d_inode !=
1276 file->f_path.dentry->d_inode)
1277 continue;
1278
1279 if (pregion->vm_pgoff >= pgend)
1280 continue;
1281
1282 rpglen = pregion->vm_end - pregion->vm_start;
1283 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1284 rpgend = pregion->vm_pgoff + rpglen;
1285 if (pgoff >= rpgend)
1286 continue;
1287
1288 /* handle inexactly overlapping matches between
1289 * mappings */
1290 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1291 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1292 /* new mapping is not a subset of the region */
1293 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1294 goto sharing_violation;
1295 continue;
1296 }
1297
1298 /* we've found a region we can share */
1299 atomic_inc(&pregion->vm_usage);
1300 vma->vm_region = pregion;
1301 start = pregion->vm_start;
1302 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1303 vma->vm_start = start;
1304 vma->vm_end = start + len;
1305
1306 if (pregion->vm_flags & VM_MAPPED_COPY) {
1307 kdebug("share copy");
1308 vma->vm_flags |= VM_MAPPED_COPY;
1309 } else {
1310 kdebug("share mmap");
1311 ret = do_mmap_shared_file(vma);
1312 if (ret < 0) {
1313 vma->vm_region = NULL;
1314 vma->vm_start = 0;
1315 vma->vm_end = 0;
1316 atomic_dec(&pregion->vm_usage);
1317 pregion = NULL;
1318 goto error_just_free;
1319 }
1320 }
1321 fput(region->vm_file);
1322 kmem_cache_free(vm_region_jar, region);
1323 region = pregion;
1324 result = start;
1325 goto share;
1326 }
1327
1328 /* obtain the address at which to make a shared mapping
1329 * - this is the hook for quasi-memory character devices to
1330 * tell us the location of a shared mapping
1331 */
1332 if (capabilities & BDI_CAP_MAP_DIRECT) {
1333 addr = file->f_op->get_unmapped_area(file, addr, len,
1334 pgoff, flags);
1335 if (IS_ERR((void *) addr)) {
1336 ret = addr;
1337 if (ret != (unsigned long) -ENOSYS)
1338 goto error_just_free;
1339
1340 /* the driver refused to tell us where to site
1341 * the mapping so we'll have to attempt to copy
1342 * it */
1343 ret = (unsigned long) -ENODEV;
1344 if (!(capabilities & BDI_CAP_MAP_COPY))
1345 goto error_just_free;
1346
1347 capabilities &= ~BDI_CAP_MAP_DIRECT;
1348 } else {
1349 vma->vm_start = region->vm_start = addr;
1350 vma->vm_end = region->vm_end = addr + len;
1351 }
1352 }
1353 }
1354
1355 vma->vm_region = region;
1356
1357 /* set up the mapping
1358 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1359 */
1360 if (file && vma->vm_flags & VM_SHARED)
1361 ret = do_mmap_shared_file(vma);
1362 else
1363 ret = do_mmap_private(vma, region, len, capabilities);
1364 if (ret < 0)
1365 goto error_just_free;
1366 add_nommu_region(region);
1367
1368 /* okay... we have a mapping; now we have to register it */
1369 result = vma->vm_start;
1370
1371 current->mm->total_vm += len >> PAGE_SHIFT;
1372
1373 share:
1374 add_vma_to_mm(current->mm, vma);
1375
1376 up_write(&nommu_region_sem);
1377
1378 if (prot & PROT_EXEC)
1379 flush_icache_range(result, result + len);
1380
1381 kleave(" = %lx", result);
1382 return result;
1383
1384 error_just_free:
1385 up_write(&nommu_region_sem);
1386 error:
1387 if (region->vm_file)
1388 fput(region->vm_file);
1389 kmem_cache_free(vm_region_jar, region);
1390 if (vma->vm_file)
1391 fput(vma->vm_file);
1392 if (vma->vm_flags & VM_EXECUTABLE)
1393 removed_exe_file_vma(vma->vm_mm);
1394 kmem_cache_free(vm_area_cachep, vma);
1395 kleave(" = %d", ret);
1396 return ret;
1397
1398 sharing_violation:
1399 up_write(&nommu_region_sem);
1400 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1401 ret = -EINVAL;
1402 goto error;
1403
1404 error_getting_vma:
1405 kmem_cache_free(vm_region_jar, region);
1406 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1407 " from process %d failed\n",
1408 len, current->pid);
1409 show_free_areas();
1410 return -ENOMEM;
1411
1412 error_getting_region:
1413 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1414 " from process %d failed\n",
1415 len, current->pid);
1416 show_free_areas();
1417 return -ENOMEM;
1418 }
1419 EXPORT_SYMBOL(do_mmap_pgoff);
1420
1421 /*
1422 * split a vma into two pieces at address 'addr', a new vma is allocated either
1423 * for the first part or the tail.
1424 */
1425 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1426 unsigned long addr, int new_below)
1427 {
1428 struct vm_area_struct *new;
1429 struct vm_region *region;
1430 unsigned long npages;
1431
1432 kenter("");
1433
1434 /* we're only permitted to split anonymous regions that have a single
1435 * owner */
1436 if (vma->vm_file ||
1437 atomic_read(&vma->vm_region->vm_usage) != 1)
1438 return -ENOMEM;
1439
1440 if (mm->map_count >= sysctl_max_map_count)
1441 return -ENOMEM;
1442
1443 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1444 if (!region)
1445 return -ENOMEM;
1446
1447 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1448 if (!new) {
1449 kmem_cache_free(vm_region_jar, region);
1450 return -ENOMEM;
1451 }
1452
1453 /* most fields are the same, copy all, and then fixup */
1454 *new = *vma;
1455 *region = *vma->vm_region;
1456 new->vm_region = region;
1457
1458 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1459
1460 if (new_below) {
1461 region->vm_top = region->vm_end = new->vm_end = addr;
1462 } else {
1463 region->vm_start = new->vm_start = addr;
1464 region->vm_pgoff = new->vm_pgoff += npages;
1465 }
1466
1467 if (new->vm_ops && new->vm_ops->open)
1468 new->vm_ops->open(new);
1469
1470 delete_vma_from_mm(vma);
1471 down_write(&nommu_region_sem);
1472 delete_nommu_region(vma->vm_region);
1473 if (new_below) {
1474 vma->vm_region->vm_start = vma->vm_start = addr;
1475 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1476 } else {
1477 vma->vm_region->vm_end = vma->vm_end = addr;
1478 vma->vm_region->vm_top = addr;
1479 }
1480 add_nommu_region(vma->vm_region);
1481 add_nommu_region(new->vm_region);
1482 up_write(&nommu_region_sem);
1483 add_vma_to_mm(mm, vma);
1484 add_vma_to_mm(mm, new);
1485 return 0;
1486 }
1487
1488 /*
1489 * shrink a VMA by removing the specified chunk from either the beginning or
1490 * the end
1491 */
1492 static int shrink_vma(struct mm_struct *mm,
1493 struct vm_area_struct *vma,
1494 unsigned long from, unsigned long to)
1495 {
1496 struct vm_region *region;
1497
1498 kenter("");
1499
1500 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1501 * and list */
1502 delete_vma_from_mm(vma);
1503 if (from > vma->vm_start)
1504 vma->vm_end = from;
1505 else
1506 vma->vm_start = to;
1507 add_vma_to_mm(mm, vma);
1508
1509 /* cut the backing region down to size */
1510 region = vma->vm_region;
1511 BUG_ON(atomic_read(®ion->vm_usage) != 1);
1512
1513 down_write(&nommu_region_sem);
1514 delete_nommu_region(region);
1515 if (from > region->vm_start) {
1516 to = region->vm_top;
1517 region->vm_top = region->vm_end = from;
1518 } else {
1519 region->vm_start = to;
1520 }
1521 add_nommu_region(region);
1522 up_write(&nommu_region_sem);
1523
1524 free_page_series(from, to);
1525 return 0;
1526 }
1527
1528 /*
1529 * release a mapping
1530 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1531 * VMA, though it need not cover the whole VMA
1532 */
1533 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1534 {
1535 struct vm_area_struct *vma;
1536 struct rb_node *rb;
1537 unsigned long end = start + len;
1538 int ret;
1539
1540 kenter(",%lx,%zx", start, len);
1541
1542 if (len == 0)
1543 return -EINVAL;
1544
1545 /* find the first potentially overlapping VMA */
1546 vma = find_vma(mm, start);
1547 if (!vma) {
1548 static int limit = 0;
1549 if (limit < 5) {
1550 printk(KERN_WARNING
1551 "munmap of memory not mmapped by process %d"
1552 " (%s): 0x%lx-0x%lx\n",
1553 current->pid, current->comm,
1554 start, start + len - 1);
1555 limit++;
1556 }
1557 return -EINVAL;
1558 }
1559
1560 /* we're allowed to split an anonymous VMA but not a file-backed one */
1561 if (vma->vm_file) {
1562 do {
1563 if (start > vma->vm_start) {
1564 kleave(" = -EINVAL [miss]");
1565 return -EINVAL;
1566 }
1567 if (end == vma->vm_end)
1568 goto erase_whole_vma;
1569 rb = rb_next(&vma->vm_rb);
1570 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1571 } while (rb);
1572 kleave(" = -EINVAL [split file]");
1573 return -EINVAL;
1574 } else {
1575 /* the chunk must be a subset of the VMA found */
1576 if (start == vma->vm_start && end == vma->vm_end)
1577 goto erase_whole_vma;
1578 if (start < vma->vm_start || end > vma->vm_end) {
1579 kleave(" = -EINVAL [superset]");
1580 return -EINVAL;
1581 }
1582 if (start & ~PAGE_MASK) {
1583 kleave(" = -EINVAL [unaligned start]");
1584 return -EINVAL;
1585 }
1586 if (end != vma->vm_end && end & ~PAGE_MASK) {
1587 kleave(" = -EINVAL [unaligned split]");
1588 return -EINVAL;
1589 }
1590 if (start != vma->vm_start && end != vma->vm_end) {
1591 ret = split_vma(mm, vma, start, 1);
1592 if (ret < 0) {
1593 kleave(" = %d [split]", ret);
1594 return ret;
1595 }
1596 }
1597 return shrink_vma(mm, vma, start, end);
1598 }
1599
1600 erase_whole_vma:
1601 delete_vma_from_mm(vma);
1602 delete_vma(mm, vma);
1603 kleave(" = 0");
1604 return 0;
1605 }
1606 EXPORT_SYMBOL(do_munmap);
1607
1608 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1609 {
1610 int ret;
1611 struct mm_struct *mm = current->mm;
1612
1613 down_write(&mm->mmap_sem);
1614 ret = do_munmap(mm, addr, len);
1615 up_write(&mm->mmap_sem);
1616 return ret;
1617 }
1618
1619 /*
1620 * release all the mappings made in a process's VM space
1621 */
1622 void exit_mmap(struct mm_struct *mm)
1623 {
1624 struct vm_area_struct *vma;
1625
1626 if (!mm)
1627 return;
1628
1629 kenter("");
1630
1631 mm->total_vm = 0;
1632
1633 while ((vma = mm->mmap)) {
1634 mm->mmap = vma->vm_next;
1635 delete_vma_from_mm(vma);
1636 delete_vma(mm, vma);
1637 }
1638
1639 kleave("");
1640 }
1641
1642 unsigned long do_brk(unsigned long addr, unsigned long len)
1643 {
1644 return -ENOMEM;
1645 }
1646
1647 /*
1648 * expand (or shrink) an existing mapping, potentially moving it at the same
1649 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1650 *
1651 * under NOMMU conditions, we only permit changing a mapping's size, and only
1652 * as long as it stays within the region allocated by do_mmap_private() and the
1653 * block is not shareable
1654 *
1655 * MREMAP_FIXED is not supported under NOMMU conditions
1656 */
1657 unsigned long do_mremap(unsigned long addr,
1658 unsigned long old_len, unsigned long new_len,
1659 unsigned long flags, unsigned long new_addr)
1660 {
1661 struct vm_area_struct *vma;
1662
1663 /* insanity checks first */
1664 if (old_len == 0 || new_len == 0)
1665 return (unsigned long) -EINVAL;
1666
1667 if (addr & ~PAGE_MASK)
1668 return -EINVAL;
1669
1670 if (flags & MREMAP_FIXED && new_addr != addr)
1671 return (unsigned long) -EINVAL;
1672
1673 vma = find_vma_exact(current->mm, addr, old_len);
1674 if (!vma)
1675 return (unsigned long) -EINVAL;
1676
1677 if (vma->vm_end != vma->vm_start + old_len)
1678 return (unsigned long) -EFAULT;
1679
1680 if (vma->vm_flags & VM_MAYSHARE)
1681 return (unsigned long) -EPERM;
1682
1683 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1684 return (unsigned long) -ENOMEM;
1685
1686 /* all checks complete - do it */
1687 vma->vm_end = vma->vm_start + new_len;
1688 return vma->vm_start;
1689 }
1690 EXPORT_SYMBOL(do_mremap);
1691
1692 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1693 unsigned long, new_len, unsigned long, flags,
1694 unsigned long, new_addr)
1695 {
1696 unsigned long ret;
1697
1698 down_write(¤t->mm->mmap_sem);
1699 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1700 up_write(¤t->mm->mmap_sem);
1701 return ret;
1702 }
1703
1704 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1705 unsigned int foll_flags)
1706 {
1707 return NULL;
1708 }
1709
1710 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1711 unsigned long to, unsigned long size, pgprot_t prot)
1712 {
1713 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1714 return 0;
1715 }
1716 EXPORT_SYMBOL(remap_pfn_range);
1717
1718 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1719 unsigned long pgoff)
1720 {
1721 unsigned int size = vma->vm_end - vma->vm_start;
1722
1723 if (!(vma->vm_flags & VM_USERMAP))
1724 return -EINVAL;
1725
1726 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1727 vma->vm_end = vma->vm_start + size;
1728
1729 return 0;
1730 }
1731 EXPORT_SYMBOL(remap_vmalloc_range);
1732
1733 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1734 {
1735 }
1736
1737 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1738 unsigned long len, unsigned long pgoff, unsigned long flags)
1739 {
1740 return -ENOMEM;
1741 }
1742
1743 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1744 {
1745 }
1746
1747 void unmap_mapping_range(struct address_space *mapping,
1748 loff_t const holebegin, loff_t const holelen,
1749 int even_cows)
1750 {
1751 }
1752 EXPORT_SYMBOL(unmap_mapping_range);
1753
1754 /*
1755 * ask for an unmapped area at which to create a mapping on a file
1756 */
1757 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1758 unsigned long len, unsigned long pgoff,
1759 unsigned long flags)
1760 {
1761 unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1762 unsigned long, unsigned long);
1763
1764 get_area = current->mm->get_unmapped_area;
1765 if (file && file->f_op && file->f_op->get_unmapped_area)
1766 get_area = file->f_op->get_unmapped_area;
1767
1768 if (!get_area)
1769 return -ENOSYS;
1770
1771 return get_area(file, addr, len, pgoff, flags);
1772 }
1773 EXPORT_SYMBOL(get_unmapped_area);
1774
1775 /*
1776 * Check that a process has enough memory to allocate a new virtual
1777 * mapping. 0 means there is enough memory for the allocation to
1778 * succeed and -ENOMEM implies there is not.
1779 *
1780 * We currently support three overcommit policies, which are set via the
1781 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1782 *
1783 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1784 * Additional code 2002 Jul 20 by Robert Love.
1785 *
1786 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1787 *
1788 * Note this is a helper function intended to be used by LSMs which
1789 * wish to use this logic.
1790 */
1791 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1792 {
1793 unsigned long free, allowed;
1794
1795 vm_acct_memory(pages);
1796
1797 /*
1798 * Sometimes we want to use more memory than we have
1799 */
1800 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1801 return 0;
1802
1803 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1804 unsigned long n;
1805
1806 free = global_page_state(NR_FILE_PAGES);
1807 free += nr_swap_pages;
1808
1809 /*
1810 * Any slabs which are created with the
1811 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1812 * which are reclaimable, under pressure. The dentry
1813 * cache and most inode caches should fall into this
1814 */
1815 free += global_page_state(NR_SLAB_RECLAIMABLE);
1816
1817 /*
1818 * Leave the last 3% for root
1819 */
1820 if (!cap_sys_admin)
1821 free -= free / 32;
1822
1823 if (free > pages)
1824 return 0;
1825
1826 /*
1827 * nr_free_pages() is very expensive on large systems,
1828 * only call if we're about to fail.
1829 */
1830 n = nr_free_pages();
1831
1832 /*
1833 * Leave reserved pages. The pages are not for anonymous pages.
1834 */
1835 if (n <= totalreserve_pages)
1836 goto error;
1837 else
1838 n -= totalreserve_pages;
1839
1840 /*
1841 * Leave the last 3% for root
1842 */
1843 if (!cap_sys_admin)
1844 n -= n / 32;
1845 free += n;
1846
1847 if (free > pages)
1848 return 0;
1849
1850 goto error;
1851 }
1852
1853 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1854 /*
1855 * Leave the last 3% for root
1856 */
1857 if (!cap_sys_admin)
1858 allowed -= allowed / 32;
1859 allowed += total_swap_pages;
1860
1861 /* Don't let a single process grow too big:
1862 leave 3% of the size of this process for other processes */
1863 if (mm)
1864 allowed -= mm->total_vm / 32;
1865
1866 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1867 return 0;
1868
1869 error:
1870 vm_unacct_memory(pages);
1871
1872 return -ENOMEM;
1873 }
1874
1875 int in_gate_area_no_task(unsigned long addr)
1876 {
1877 return 0;
1878 }
1879
1880 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1881 {
1882 BUG();
1883 return 0;
1884 }
1885 EXPORT_SYMBOL(filemap_fault);
1886
1887 /*
1888 * Access another process' address space.
1889 * - source/target buffer must be kernel space
1890 */
1891 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1892 {
1893 struct vm_area_struct *vma;
1894 struct mm_struct *mm;
1895
1896 if (addr + len < addr)
1897 return 0;
1898
1899 mm = get_task_mm(tsk);
1900 if (!mm)
1901 return 0;
1902
1903 down_read(&mm->mmap_sem);
1904
1905 /* the access must start within one of the target process's mappings */
1906 vma = find_vma(mm, addr);
1907 if (vma) {
1908 /* don't overrun this mapping */
1909 if (addr + len >= vma->vm_end)
1910 len = vma->vm_end - addr;
1911
1912 /* only read or write mappings where it is permitted */
1913 if (write && vma->vm_flags & VM_MAYWRITE)
1914 len -= copy_to_user((void *) addr, buf, len);
1915 else if (!write && vma->vm_flags & VM_MAYREAD)
1916 len -= copy_from_user(buf, (void *) addr, len);
1917 else
1918 len = 0;
1919 } else {
1920 len = 0;
1921 }
1922
1923 up_read(&mm->mmap_sem);
1924 mmput(mm);
1925 return len;
1926 }
1927
|
This page was automatically generated by the
LXR engine.
|