Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/mm/nommu.c
4 *
5 * Replacement code for mm functions to support CPU's that don't
6 * have any form of memory management unit (thus no virtual memory).
7 *
8 * See Documentation/admin-guide/mm/nommu-mmap.rst
9 *
10 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
11 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
12 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
13 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
14 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
15 */
16
17#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18
19#include <linux/export.h>
20#include <linux/mm.h>
21#include <linux/sched/mm.h>
22#include <linux/mman.h>
23#include <linux/swap.h>
24#include <linux/file.h>
25#include <linux/highmem.h>
26#include <linux/pagemap.h>
27#include <linux/slab.h>
28#include <linux/vmalloc.h>
29#include <linux/backing-dev.h>
30#include <linux/compiler.h>
31#include <linux/mount.h>
32#include <linux/personality.h>
33#include <linux/security.h>
34#include <linux/syscalls.h>
35#include <linux/audit.h>
36#include <linux/printk.h>
37
38#include <linux/uaccess.h>
39#include <asm/tlb.h>
40#include <asm/tlbflush.h>
41#include <asm/mmu_context.h>
42#include "internal.h"
43
44void *high_memory;
45EXPORT_SYMBOL(high_memory);
46struct page *mem_map;
47unsigned long max_mapnr;
48EXPORT_SYMBOL(max_mapnr);
49unsigned long highest_memmap_pfn;
50int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
51int heap_stack_gap = 0;
52
53atomic_long_t mmap_pages_allocated;
54
55EXPORT_SYMBOL(mem_map);
56
57/* list of mapped, potentially shareable regions */
58static struct kmem_cache *vm_region_jar;
59struct rb_root nommu_region_tree = RB_ROOT;
60DECLARE_RWSEM(nommu_region_sem);
61
62const struct vm_operations_struct generic_file_vm_ops = {
63};
64
65/*
66 * Return the total memory allocated for this pointer, not
67 * just what the caller asked for.
68 *
69 * Doesn't have to be accurate, i.e. may have races.
70 */
71unsigned int kobjsize(const void *objp)
72{
73 struct page *page;
74
75 /*
76 * If the object we have should not have ksize performed on it,
77 * return size of 0
78 */
79 if (!objp || !virt_addr_valid(objp))
80 return 0;
81
82 page = virt_to_head_page(objp);
83
84 /*
85 * If the allocator sets PageSlab, we know the pointer came from
86 * kmalloc().
87 */
88 if (PageSlab(page))
89 return ksize(objp);
90
91 /*
92 * If it's not a compound page, see if we have a matching VMA
93 * region. This test is intentionally done in reverse order,
94 * so if there's no VMA, we still fall through and hand back
95 * PAGE_SIZE for 0-order pages.
96 */
97 if (!PageCompound(page)) {
98 struct vm_area_struct *vma;
99
100 vma = find_vma(current->mm, (unsigned long)objp);
101 if (vma)
102 return vma->vm_end - vma->vm_start;
103 }
104
105 /*
106 * The ksize() function is only guaranteed to work for pointers
107 * returned by kmalloc(). So handle arbitrary pointers here.
108 */
109 return page_size(page);
110}
111
112/**
113 * follow_pfn - look up PFN at a user virtual address
114 * @vma: memory mapping
115 * @address: user virtual address
116 * @pfn: location to store found PFN
117 *
118 * Only IO mappings and raw PFN mappings are allowed.
119 *
120 * Returns zero and the pfn at @pfn on success, -ve otherwise.
121 */
122int follow_pfn(struct vm_area_struct *vma, unsigned long address,
123 unsigned long *pfn)
124{
125 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
126 return -EINVAL;
127
128 *pfn = address >> PAGE_SHIFT;
129 return 0;
130}
131EXPORT_SYMBOL(follow_pfn);
132
133LIST_HEAD(vmap_area_list);
134
135void vfree(const void *addr)
136{
137 kfree(addr);
138}
139EXPORT_SYMBOL(vfree);
140
141void *__vmalloc(unsigned long size, gfp_t gfp_mask)
142{
143 /*
144 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
145 * returns only a logical address.
146 */
147 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
148}
149EXPORT_SYMBOL(__vmalloc);
150
151void *__vmalloc_node_range(unsigned long size, unsigned long align,
152 unsigned long start, unsigned long end, gfp_t gfp_mask,
153 pgprot_t prot, unsigned long vm_flags, int node,
154 const void *caller)
155{
156 return __vmalloc(size, gfp_mask);
157}
158
159void *__vmalloc_node(unsigned long size, unsigned long align, gfp_t gfp_mask,
160 int node, const void *caller)
161{
162 return __vmalloc(size, gfp_mask);
163}
164
165static void *__vmalloc_user_flags(unsigned long size, gfp_t flags)
166{
167 void *ret;
168
169 ret = __vmalloc(size, flags);
170 if (ret) {
171 struct vm_area_struct *vma;
172
173 mmap_write_lock(current->mm);
174 vma = find_vma(current->mm, (unsigned long)ret);
175 if (vma)
176 vma->vm_flags |= VM_USERMAP;
177 mmap_write_unlock(current->mm);
178 }
179
180 return ret;
181}
182
183void *vmalloc_user(unsigned long size)
184{
185 return __vmalloc_user_flags(size, GFP_KERNEL | __GFP_ZERO);
186}
187EXPORT_SYMBOL(vmalloc_user);
188
189struct page *vmalloc_to_page(const void *addr)
190{
191 return virt_to_page(addr);
192}
193EXPORT_SYMBOL(vmalloc_to_page);
194
195unsigned long vmalloc_to_pfn(const void *addr)
196{
197 return page_to_pfn(virt_to_page(addr));
198}
199EXPORT_SYMBOL(vmalloc_to_pfn);
200
201long vread(char *buf, char *addr, unsigned long count)
202{
203 /* Don't allow overflow */
204 if ((unsigned long) buf + count < count)
205 count = -(unsigned long) buf;
206
207 memcpy(buf, addr, count);
208 return count;
209}
210
211/*
212 * vmalloc - allocate virtually contiguous memory
213 *
214 * @size: allocation size
215 *
216 * Allocate enough pages to cover @size from the page level
217 * allocator and map them into contiguous kernel virtual space.
218 *
219 * For tight control over page level allocator and protection flags
220 * use __vmalloc() instead.
221 */
222void *vmalloc(unsigned long size)
223{
224 return __vmalloc(size, GFP_KERNEL);
225}
226EXPORT_SYMBOL(vmalloc);
227
228void *vmalloc_huge(unsigned long size, gfp_t gfp_mask) __weak __alias(__vmalloc);
229
230/*
231 * vzalloc - allocate virtually contiguous memory with zero fill
232 *
233 * @size: allocation size
234 *
235 * Allocate enough pages to cover @size from the page level
236 * allocator and map them into contiguous kernel virtual space.
237 * The memory allocated is set to zero.
238 *
239 * For tight control over page level allocator and protection flags
240 * use __vmalloc() instead.
241 */
242void *vzalloc(unsigned long size)
243{
244 return __vmalloc(size, GFP_KERNEL | __GFP_ZERO);
245}
246EXPORT_SYMBOL(vzalloc);
247
248/**
249 * vmalloc_node - allocate memory on a specific node
250 * @size: allocation size
251 * @node: numa node
252 *
253 * Allocate enough pages to cover @size from the page level
254 * allocator and map them into contiguous kernel virtual space.
255 *
256 * For tight control over page level allocator and protection flags
257 * use __vmalloc() instead.
258 */
259void *vmalloc_node(unsigned long size, int node)
260{
261 return vmalloc(size);
262}
263EXPORT_SYMBOL(vmalloc_node);
264
265/**
266 * vzalloc_node - allocate memory on a specific node with zero fill
267 * @size: allocation size
268 * @node: numa node
269 *
270 * Allocate enough pages to cover @size from the page level
271 * allocator and map them into contiguous kernel virtual space.
272 * The memory allocated is set to zero.
273 *
274 * For tight control over page level allocator and protection flags
275 * use __vmalloc() instead.
276 */
277void *vzalloc_node(unsigned long size, int node)
278{
279 return vzalloc(size);
280}
281EXPORT_SYMBOL(vzalloc_node);
282
283/**
284 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
285 * @size: allocation size
286 *
287 * Allocate enough 32bit PA addressable pages to cover @size from the
288 * page level allocator and map them into contiguous kernel virtual space.
289 */
290void *vmalloc_32(unsigned long size)
291{
292 return __vmalloc(size, GFP_KERNEL);
293}
294EXPORT_SYMBOL(vmalloc_32);
295
296/**
297 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
298 * @size: allocation size
299 *
300 * The resulting memory area is 32bit addressable and zeroed so it can be
301 * mapped to userspace without leaking data.
302 *
303 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
304 * remap_vmalloc_range() are permissible.
305 */
306void *vmalloc_32_user(unsigned long size)
307{
308 /*
309 * We'll have to sort out the ZONE_DMA bits for 64-bit,
310 * but for now this can simply use vmalloc_user() directly.
311 */
312 return vmalloc_user(size);
313}
314EXPORT_SYMBOL(vmalloc_32_user);
315
316void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
317{
318 BUG();
319 return NULL;
320}
321EXPORT_SYMBOL(vmap);
322
323void vunmap(const void *addr)
324{
325 BUG();
326}
327EXPORT_SYMBOL(vunmap);
328
329void *vm_map_ram(struct page **pages, unsigned int count, int node)
330{
331 BUG();
332 return NULL;
333}
334EXPORT_SYMBOL(vm_map_ram);
335
336void vm_unmap_ram(const void *mem, unsigned int count)
337{
338 BUG();
339}
340EXPORT_SYMBOL(vm_unmap_ram);
341
342void vm_unmap_aliases(void)
343{
344}
345EXPORT_SYMBOL_GPL(vm_unmap_aliases);
346
347void free_vm_area(struct vm_struct *area)
348{
349 BUG();
350}
351EXPORT_SYMBOL_GPL(free_vm_area);
352
353int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
354 struct page *page)
355{
356 return -EINVAL;
357}
358EXPORT_SYMBOL(vm_insert_page);
359
360int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
361 unsigned long num)
362{
363 return -EINVAL;
364}
365EXPORT_SYMBOL(vm_map_pages);
366
367int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
368 unsigned long num)
369{
370 return -EINVAL;
371}
372EXPORT_SYMBOL(vm_map_pages_zero);
373
374/*
375 * sys_brk() for the most part doesn't need the global kernel
376 * lock, except when an application is doing something nasty
377 * like trying to un-brk an area that has already been mapped
378 * to a regular file. in this case, the unmapping will need
379 * to invoke file system routines that need the global lock.
380 */
381SYSCALL_DEFINE1(brk, unsigned long, brk)
382{
383 struct mm_struct *mm = current->mm;
384
385 if (brk < mm->start_brk || brk > mm->context.end_brk)
386 return mm->brk;
387
388 if (mm->brk == brk)
389 return mm->brk;
390
391 /*
392 * Always allow shrinking brk
393 */
394 if (brk <= mm->brk) {
395 mm->brk = brk;
396 return brk;
397 }
398
399 /*
400 * Ok, looks good - let it rip.
401 */
402 flush_icache_user_range(mm->brk, brk);
403 return mm->brk = brk;
404}
405
406/*
407 * initialise the percpu counter for VM and region record slabs
408 */
409void __init mmap_init(void)
410{
411 int ret;
412
413 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
414 VM_BUG_ON(ret);
415 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
416}
417
418/*
419 * validate the region tree
420 * - the caller must hold the region lock
421 */
422#ifdef CONFIG_DEBUG_NOMMU_REGIONS
423static noinline void validate_nommu_regions(void)
424{
425 struct vm_region *region, *last;
426 struct rb_node *p, *lastp;
427
428 lastp = rb_first(&nommu_region_tree);
429 if (!lastp)
430 return;
431
432 last = rb_entry(lastp, struct vm_region, vm_rb);
433 BUG_ON(last->vm_end <= last->vm_start);
434 BUG_ON(last->vm_top < last->vm_end);
435
436 while ((p = rb_next(lastp))) {
437 region = rb_entry(p, struct vm_region, vm_rb);
438 last = rb_entry(lastp, struct vm_region, vm_rb);
439
440 BUG_ON(region->vm_end <= region->vm_start);
441 BUG_ON(region->vm_top < region->vm_end);
442 BUG_ON(region->vm_start < last->vm_top);
443
444 lastp = p;
445 }
446}
447#else
448static void validate_nommu_regions(void)
449{
450}
451#endif
452
453/*
454 * add a region into the global tree
455 */
456static void add_nommu_region(struct vm_region *region)
457{
458 struct vm_region *pregion;
459 struct rb_node **p, *parent;
460
461 validate_nommu_regions();
462
463 parent = NULL;
464 p = &nommu_region_tree.rb_node;
465 while (*p) {
466 parent = *p;
467 pregion = rb_entry(parent, struct vm_region, vm_rb);
468 if (region->vm_start < pregion->vm_start)
469 p = &(*p)->rb_left;
470 else if (region->vm_start > pregion->vm_start)
471 p = &(*p)->rb_right;
472 else if (pregion == region)
473 return;
474 else
475 BUG();
476 }
477
478 rb_link_node(®ion->vm_rb, parent, p);
479 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
480
481 validate_nommu_regions();
482}
483
484/*
485 * delete a region from the global tree
486 */
487static void delete_nommu_region(struct vm_region *region)
488{
489 BUG_ON(!nommu_region_tree.rb_node);
490
491 validate_nommu_regions();
492 rb_erase(®ion->vm_rb, &nommu_region_tree);
493 validate_nommu_regions();
494}
495
496/*
497 * free a contiguous series of pages
498 */
499static void free_page_series(unsigned long from, unsigned long to)
500{
501 for (; from < to; from += PAGE_SIZE) {
502 struct page *page = virt_to_page((void *)from);
503
504 atomic_long_dec(&mmap_pages_allocated);
505 put_page(page);
506 }
507}
508
509/*
510 * release a reference to a region
511 * - the caller must hold the region semaphore for writing, which this releases
512 * - the region may not have been added to the tree yet, in which case vm_top
513 * will equal vm_start
514 */
515static void __put_nommu_region(struct vm_region *region)
516 __releases(nommu_region_sem)
517{
518 BUG_ON(!nommu_region_tree.rb_node);
519
520 if (--region->vm_usage == 0) {
521 if (region->vm_top > region->vm_start)
522 delete_nommu_region(region);
523 up_write(&nommu_region_sem);
524
525 if (region->vm_file)
526 fput(region->vm_file);
527
528 /* IO memory and memory shared directly out of the pagecache
529 * from ramfs/tmpfs mustn't be released here */
530 if (region->vm_flags & VM_MAPPED_COPY)
531 free_page_series(region->vm_start, region->vm_top);
532 kmem_cache_free(vm_region_jar, region);
533 } else {
534 up_write(&nommu_region_sem);
535 }
536}
537
538/*
539 * release a reference to a region
540 */
541static void put_nommu_region(struct vm_region *region)
542{
543 down_write(&nommu_region_sem);
544 __put_nommu_region(region);
545}
546
547void vma_mas_store(struct vm_area_struct *vma, struct ma_state *mas)
548{
549 mas_set_range(mas, vma->vm_start, vma->vm_end - 1);
550 mas_store_prealloc(mas, vma);
551}
552
553void vma_mas_remove(struct vm_area_struct *vma, struct ma_state *mas)
554{
555 mas->index = vma->vm_start;
556 mas->last = vma->vm_end - 1;
557 mas_store_prealloc(mas, NULL);
558}
559
560static void setup_vma_to_mm(struct vm_area_struct *vma, struct mm_struct *mm)
561{
562 vma->vm_mm = mm;
563
564 /* add the VMA to the mapping */
565 if (vma->vm_file) {
566 struct address_space *mapping = vma->vm_file->f_mapping;
567
568 i_mmap_lock_write(mapping);
569 flush_dcache_mmap_lock(mapping);
570 vma_interval_tree_insert(vma, &mapping->i_mmap);
571 flush_dcache_mmap_unlock(mapping);
572 i_mmap_unlock_write(mapping);
573 }
574}
575
576/*
577 * mas_add_vma_to_mm() - Maple state variant of add_mas_to_mm().
578 * @mas: The maple state with preallocations.
579 * @mm: The mm_struct
580 * @vma: The vma to add
581 *
582 */
583static void mas_add_vma_to_mm(struct ma_state *mas, struct mm_struct *mm,
584 struct vm_area_struct *vma)
585{
586 BUG_ON(!vma->vm_region);
587
588 setup_vma_to_mm(vma, mm);
589 mm->map_count++;
590
591 /* add the VMA to the tree */
592 vma_mas_store(vma, mas);
593}
594
595/*
596 * add a VMA into a process's mm_struct in the appropriate place in the list
597 * and tree and add to the address space's page tree also if not an anonymous
598 * page
599 * - should be called with mm->mmap_lock held writelocked
600 */
601static int add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
602{
603 MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_end);
604
605 if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
606 pr_warn("Allocation of vma tree for process %d failed\n",
607 current->pid);
608 return -ENOMEM;
609 }
610 mas_add_vma_to_mm(&mas, mm, vma);
611 return 0;
612}
613
614static void cleanup_vma_from_mm(struct vm_area_struct *vma)
615{
616 vma->vm_mm->map_count--;
617 /* remove the VMA from the mapping */
618 if (vma->vm_file) {
619 struct address_space *mapping;
620 mapping = vma->vm_file->f_mapping;
621
622 i_mmap_lock_write(mapping);
623 flush_dcache_mmap_lock(mapping);
624 vma_interval_tree_remove(vma, &mapping->i_mmap);
625 flush_dcache_mmap_unlock(mapping);
626 i_mmap_unlock_write(mapping);
627 }
628}
629/*
630 * delete a VMA from its owning mm_struct and address space
631 */
632static int delete_vma_from_mm(struct vm_area_struct *vma)
633{
634 MA_STATE(mas, &vma->vm_mm->mm_mt, 0, 0);
635
636 if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
637 pr_warn("Allocation of vma tree for process %d failed\n",
638 current->pid);
639 return -ENOMEM;
640 }
641 cleanup_vma_from_mm(vma);
642
643 /* remove from the MM's tree and list */
644 vma_mas_remove(vma, &mas);
645 return 0;
646}
647
648/*
649 * destroy a VMA record
650 */
651static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
652{
653 if (vma->vm_ops && vma->vm_ops->close)
654 vma->vm_ops->close(vma);
655 if (vma->vm_file)
656 fput(vma->vm_file);
657 put_nommu_region(vma->vm_region);
658 vm_area_free(vma);
659}
660
661struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
662 unsigned long start_addr,
663 unsigned long end_addr)
664{
665 unsigned long index = start_addr;
666
667 mmap_assert_locked(mm);
668 return mt_find(&mm->mm_mt, &index, end_addr - 1);
669}
670EXPORT_SYMBOL(find_vma_intersection);
671
672/*
673 * look up the first VMA in which addr resides, NULL if none
674 * - should be called with mm->mmap_lock at least held readlocked
675 */
676struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
677{
678 MA_STATE(mas, &mm->mm_mt, addr, addr);
679
680 return mas_walk(&mas);
681}
682EXPORT_SYMBOL(find_vma);
683
684/*
685 * find a VMA
686 * - we don't extend stack VMAs under NOMMU conditions
687 */
688struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
689{
690 return find_vma(mm, addr);
691}
692
693/*
694 * expand a stack to a given address
695 * - not supported under NOMMU conditions
696 */
697int expand_stack(struct vm_area_struct *vma, unsigned long address)
698{
699 return -ENOMEM;
700}
701
702/*
703 * look up the first VMA exactly that exactly matches addr
704 * - should be called with mm->mmap_lock at least held readlocked
705 */
706static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
707 unsigned long addr,
708 unsigned long len)
709{
710 struct vm_area_struct *vma;
711 unsigned long end = addr + len;
712 MA_STATE(mas, &mm->mm_mt, addr, addr);
713
714 vma = mas_walk(&mas);
715 if (!vma)
716 return NULL;
717 if (vma->vm_start != addr)
718 return NULL;
719 if (vma->vm_end != end)
720 return NULL;
721
722 return vma;
723}
724
725/*
726 * determine whether a mapping should be permitted and, if so, what sort of
727 * mapping we're capable of supporting
728 */
729static int validate_mmap_request(struct file *file,
730 unsigned long addr,
731 unsigned long len,
732 unsigned long prot,
733 unsigned long flags,
734 unsigned long pgoff,
735 unsigned long *_capabilities)
736{
737 unsigned long capabilities, rlen;
738 int ret;
739
740 /* do the simple checks first */
741 if (flags & MAP_FIXED)
742 return -EINVAL;
743
744 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
745 (flags & MAP_TYPE) != MAP_SHARED)
746 return -EINVAL;
747
748 if (!len)
749 return -EINVAL;
750
751 /* Careful about overflows.. */
752 rlen = PAGE_ALIGN(len);
753 if (!rlen || rlen > TASK_SIZE)
754 return -ENOMEM;
755
756 /* offset overflow? */
757 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
758 return -EOVERFLOW;
759
760 if (file) {
761 /* files must support mmap */
762 if (!file->f_op->mmap)
763 return -ENODEV;
764
765 /* work out if what we've got could possibly be shared
766 * - we support chardevs that provide their own "memory"
767 * - we support files/blockdevs that are memory backed
768 */
769 if (file->f_op->mmap_capabilities) {
770 capabilities = file->f_op->mmap_capabilities(file);
771 } else {
772 /* no explicit capabilities set, so assume some
773 * defaults */
774 switch (file_inode(file)->i_mode & S_IFMT) {
775 case S_IFREG:
776 case S_IFBLK:
777 capabilities = NOMMU_MAP_COPY;
778 break;
779
780 case S_IFCHR:
781 capabilities =
782 NOMMU_MAP_DIRECT |
783 NOMMU_MAP_READ |
784 NOMMU_MAP_WRITE;
785 break;
786
787 default:
788 return -EINVAL;
789 }
790 }
791
792 /* eliminate any capabilities that we can't support on this
793 * device */
794 if (!file->f_op->get_unmapped_area)
795 capabilities &= ~NOMMU_MAP_DIRECT;
796 if (!(file->f_mode & FMODE_CAN_READ))
797 capabilities &= ~NOMMU_MAP_COPY;
798
799 /* The file shall have been opened with read permission. */
800 if (!(file->f_mode & FMODE_READ))
801 return -EACCES;
802
803 if (flags & MAP_SHARED) {
804 /* do checks for writing, appending and locking */
805 if ((prot & PROT_WRITE) &&
806 !(file->f_mode & FMODE_WRITE))
807 return -EACCES;
808
809 if (IS_APPEND(file_inode(file)) &&
810 (file->f_mode & FMODE_WRITE))
811 return -EACCES;
812
813 if (!(capabilities & NOMMU_MAP_DIRECT))
814 return -ENODEV;
815
816 /* we mustn't privatise shared mappings */
817 capabilities &= ~NOMMU_MAP_COPY;
818 } else {
819 /* we're going to read the file into private memory we
820 * allocate */
821 if (!(capabilities & NOMMU_MAP_COPY))
822 return -ENODEV;
823
824 /* we don't permit a private writable mapping to be
825 * shared with the backing device */
826 if (prot & PROT_WRITE)
827 capabilities &= ~NOMMU_MAP_DIRECT;
828 }
829
830 if (capabilities & NOMMU_MAP_DIRECT) {
831 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
832 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
833 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
834 ) {
835 capabilities &= ~NOMMU_MAP_DIRECT;
836 if (flags & MAP_SHARED) {
837 pr_warn("MAP_SHARED not completely supported on !MMU\n");
838 return -EINVAL;
839 }
840 }
841 }
842
843 /* handle executable mappings and implied executable
844 * mappings */
845 if (path_noexec(&file->f_path)) {
846 if (prot & PROT_EXEC)
847 return -EPERM;
848 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
849 /* handle implication of PROT_EXEC by PROT_READ */
850 if (current->personality & READ_IMPLIES_EXEC) {
851 if (capabilities & NOMMU_MAP_EXEC)
852 prot |= PROT_EXEC;
853 }
854 } else if ((prot & PROT_READ) &&
855 (prot & PROT_EXEC) &&
856 !(capabilities & NOMMU_MAP_EXEC)
857 ) {
858 /* backing file is not executable, try to copy */
859 capabilities &= ~NOMMU_MAP_DIRECT;
860 }
861 } else {
862 /* anonymous mappings are always memory backed and can be
863 * privately mapped
864 */
865 capabilities = NOMMU_MAP_COPY;
866
867 /* handle PROT_EXEC implication by PROT_READ */
868 if ((prot & PROT_READ) &&
869 (current->personality & READ_IMPLIES_EXEC))
870 prot |= PROT_EXEC;
871 }
872
873 /* allow the security API to have its say */
874 ret = security_mmap_addr(addr);
875 if (ret < 0)
876 return ret;
877
878 /* looks okay */
879 *_capabilities = capabilities;
880 return 0;
881}
882
883/*
884 * we've determined that we can make the mapping, now translate what we
885 * now know into VMA flags
886 */
887static unsigned long determine_vm_flags(struct file *file,
888 unsigned long prot,
889 unsigned long flags,
890 unsigned long capabilities)
891{
892 unsigned long vm_flags;
893
894 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
895 /* vm_flags |= mm->def_flags; */
896
897 if (!(capabilities & NOMMU_MAP_DIRECT)) {
898 /* attempt to share read-only copies of mapped file chunks */
899 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
900 if (file && !(prot & PROT_WRITE))
901 vm_flags |= VM_MAYSHARE;
902 } else {
903 /* overlay a shareable mapping on the backing device or inode
904 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
905 * romfs/cramfs */
906 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
907 if (flags & MAP_SHARED)
908 vm_flags |= VM_SHARED;
909 }
910
911 /* refuse to let anyone share private mappings with this process if
912 * it's being traced - otherwise breakpoints set in it may interfere
913 * with another untraced process
914 */
915 if ((flags & MAP_PRIVATE) && current->ptrace)
916 vm_flags &= ~VM_MAYSHARE;
917
918 return vm_flags;
919}
920
921/*
922 * set up a shared mapping on a file (the driver or filesystem provides and
923 * pins the storage)
924 */
925static int do_mmap_shared_file(struct vm_area_struct *vma)
926{
927 int ret;
928
929 ret = call_mmap(vma->vm_file, vma);
930 if (ret == 0) {
931 vma->vm_region->vm_top = vma->vm_region->vm_end;
932 return 0;
933 }
934 if (ret != -ENOSYS)
935 return ret;
936
937 /* getting -ENOSYS indicates that direct mmap isn't possible (as
938 * opposed to tried but failed) so we can only give a suitable error as
939 * it's not possible to make a private copy if MAP_SHARED was given */
940 return -ENODEV;
941}
942
943/*
944 * set up a private mapping or an anonymous shared mapping
945 */
946static int do_mmap_private(struct vm_area_struct *vma,
947 struct vm_region *region,
948 unsigned long len,
949 unsigned long capabilities)
950{
951 unsigned long total, point;
952 void *base;
953 int ret, order;
954
955 /* invoke the file's mapping function so that it can keep track of
956 * shared mappings on devices or memory
957 * - VM_MAYSHARE will be set if it may attempt to share
958 */
959 if (capabilities & NOMMU_MAP_DIRECT) {
960 ret = call_mmap(vma->vm_file, vma);
961 if (ret == 0) {
962 /* shouldn't return success if we're not sharing */
963 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
964 vma->vm_region->vm_top = vma->vm_region->vm_end;
965 return 0;
966 }
967 if (ret != -ENOSYS)
968 return ret;
969
970 /* getting an ENOSYS error indicates that direct mmap isn't
971 * possible (as opposed to tried but failed) so we'll try to
972 * make a private copy of the data and map that instead */
973 }
974
975
976 /* allocate some memory to hold the mapping
977 * - note that this may not return a page-aligned address if the object
978 * we're allocating is smaller than a page
979 */
980 order = get_order(len);
981 total = 1 << order;
982 point = len >> PAGE_SHIFT;
983
984 /* we don't want to allocate a power-of-2 sized page set */
985 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
986 total = point;
987
988 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
989 if (!base)
990 goto enomem;
991
992 atomic_long_add(total, &mmap_pages_allocated);
993
994 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
995 region->vm_start = (unsigned long) base;
996 region->vm_end = region->vm_start + len;
997 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
998
999 vma->vm_start = region->vm_start;
1000 vma->vm_end = region->vm_start + len;
1001
1002 if (vma->vm_file) {
1003 /* read the contents of a file into the copy */
1004 loff_t fpos;
1005
1006 fpos = vma->vm_pgoff;
1007 fpos <<= PAGE_SHIFT;
1008
1009 ret = kernel_read(vma->vm_file, base, len, &fpos);
1010 if (ret < 0)
1011 goto error_free;
1012
1013 /* clear the last little bit */
1014 if (ret < len)
1015 memset(base + ret, 0, len - ret);
1016
1017 } else {
1018 vma_set_anonymous(vma);
1019 }
1020
1021 return 0;
1022
1023error_free:
1024 free_page_series(region->vm_start, region->vm_top);
1025 region->vm_start = vma->vm_start = 0;
1026 region->vm_end = vma->vm_end = 0;
1027 region->vm_top = 0;
1028 return ret;
1029
1030enomem:
1031 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1032 len, current->pid, current->comm);
1033 show_free_areas(0, NULL);
1034 return -ENOMEM;
1035}
1036
1037/*
1038 * handle mapping creation for uClinux
1039 */
1040unsigned long do_mmap(struct file *file,
1041 unsigned long addr,
1042 unsigned long len,
1043 unsigned long prot,
1044 unsigned long flags,
1045 unsigned long pgoff,
1046 unsigned long *populate,
1047 struct list_head *uf)
1048{
1049 struct vm_area_struct *vma;
1050 struct vm_region *region;
1051 struct rb_node *rb;
1052 vm_flags_t vm_flags;
1053 unsigned long capabilities, result;
1054 int ret;
1055 MA_STATE(mas, ¤t->mm->mm_mt, 0, 0);
1056
1057 *populate = 0;
1058
1059 /* decide whether we should attempt the mapping, and if so what sort of
1060 * mapping */
1061 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1062 &capabilities);
1063 if (ret < 0)
1064 return ret;
1065
1066 /* we ignore the address hint */
1067 addr = 0;
1068 len = PAGE_ALIGN(len);
1069
1070 /* we've determined that we can make the mapping, now translate what we
1071 * now know into VMA flags */
1072 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1073
1074
1075 /* we're going to need to record the mapping */
1076 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1077 if (!region)
1078 goto error_getting_region;
1079
1080 vma = vm_area_alloc(current->mm);
1081 if (!vma)
1082 goto error_getting_vma;
1083
1084 if (mas_preallocate(&mas, vma, GFP_KERNEL))
1085 goto error_maple_preallocate;
1086
1087 region->vm_usage = 1;
1088 region->vm_flags = vm_flags;
1089 region->vm_pgoff = pgoff;
1090
1091 vma->vm_flags = vm_flags;
1092 vma->vm_pgoff = pgoff;
1093
1094 if (file) {
1095 region->vm_file = get_file(file);
1096 vma->vm_file = get_file(file);
1097 }
1098
1099 down_write(&nommu_region_sem);
1100
1101 /* if we want to share, we need to check for regions created by other
1102 * mmap() calls that overlap with our proposed mapping
1103 * - we can only share with a superset match on most regular files
1104 * - shared mappings on character devices and memory backed files are
1105 * permitted to overlap inexactly as far as we are concerned for in
1106 * these cases, sharing is handled in the driver or filesystem rather
1107 * than here
1108 */
1109 if (vm_flags & VM_MAYSHARE) {
1110 struct vm_region *pregion;
1111 unsigned long pglen, rpglen, pgend, rpgend, start;
1112
1113 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1114 pgend = pgoff + pglen;
1115
1116 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1117 pregion = rb_entry(rb, struct vm_region, vm_rb);
1118
1119 if (!(pregion->vm_flags & VM_MAYSHARE))
1120 continue;
1121
1122 /* search for overlapping mappings on the same file */
1123 if (file_inode(pregion->vm_file) !=
1124 file_inode(file))
1125 continue;
1126
1127 if (pregion->vm_pgoff >= pgend)
1128 continue;
1129
1130 rpglen = pregion->vm_end - pregion->vm_start;
1131 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1132 rpgend = pregion->vm_pgoff + rpglen;
1133 if (pgoff >= rpgend)
1134 continue;
1135
1136 /* handle inexactly overlapping matches between
1137 * mappings */
1138 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1139 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1140 /* new mapping is not a subset of the region */
1141 if (!(capabilities & NOMMU_MAP_DIRECT))
1142 goto sharing_violation;
1143 continue;
1144 }
1145
1146 /* we've found a region we can share */
1147 pregion->vm_usage++;
1148 vma->vm_region = pregion;
1149 start = pregion->vm_start;
1150 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1151 vma->vm_start = start;
1152 vma->vm_end = start + len;
1153
1154 if (pregion->vm_flags & VM_MAPPED_COPY)
1155 vma->vm_flags |= VM_MAPPED_COPY;
1156 else {
1157 ret = do_mmap_shared_file(vma);
1158 if (ret < 0) {
1159 vma->vm_region = NULL;
1160 vma->vm_start = 0;
1161 vma->vm_end = 0;
1162 pregion->vm_usage--;
1163 pregion = NULL;
1164 goto error_just_free;
1165 }
1166 }
1167 fput(region->vm_file);
1168 kmem_cache_free(vm_region_jar, region);
1169 region = pregion;
1170 result = start;
1171 goto share;
1172 }
1173
1174 /* obtain the address at which to make a shared mapping
1175 * - this is the hook for quasi-memory character devices to
1176 * tell us the location of a shared mapping
1177 */
1178 if (capabilities & NOMMU_MAP_DIRECT) {
1179 addr = file->f_op->get_unmapped_area(file, addr, len,
1180 pgoff, flags);
1181 if (IS_ERR_VALUE(addr)) {
1182 ret = addr;
1183 if (ret != -ENOSYS)
1184 goto error_just_free;
1185
1186 /* the driver refused to tell us where to site
1187 * the mapping so we'll have to attempt to copy
1188 * it */
1189 ret = -ENODEV;
1190 if (!(capabilities & NOMMU_MAP_COPY))
1191 goto error_just_free;
1192
1193 capabilities &= ~NOMMU_MAP_DIRECT;
1194 } else {
1195 vma->vm_start = region->vm_start = addr;
1196 vma->vm_end = region->vm_end = addr + len;
1197 }
1198 }
1199 }
1200
1201 vma->vm_region = region;
1202
1203 /* set up the mapping
1204 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1205 */
1206 if (file && vma->vm_flags & VM_SHARED)
1207 ret = do_mmap_shared_file(vma);
1208 else
1209 ret = do_mmap_private(vma, region, len, capabilities);
1210 if (ret < 0)
1211 goto error_just_free;
1212 add_nommu_region(region);
1213
1214 /* clear anonymous mappings that don't ask for uninitialized data */
1215 if (!vma->vm_file &&
1216 (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) ||
1217 !(flags & MAP_UNINITIALIZED)))
1218 memset((void *)region->vm_start, 0,
1219 region->vm_end - region->vm_start);
1220
1221 /* okay... we have a mapping; now we have to register it */
1222 result = vma->vm_start;
1223
1224 current->mm->total_vm += len >> PAGE_SHIFT;
1225
1226share:
1227 mas_add_vma_to_mm(&mas, current->mm, vma);
1228
1229 /* we flush the region from the icache only when the first executable
1230 * mapping of it is made */
1231 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1232 flush_icache_user_range(region->vm_start, region->vm_end);
1233 region->vm_icache_flushed = true;
1234 }
1235
1236 up_write(&nommu_region_sem);
1237
1238 return result;
1239
1240error_just_free:
1241 up_write(&nommu_region_sem);
1242error:
1243 mas_destroy(&mas);
1244 if (region->vm_file)
1245 fput(region->vm_file);
1246 kmem_cache_free(vm_region_jar, region);
1247 if (vma->vm_file)
1248 fput(vma->vm_file);
1249 vm_area_free(vma);
1250 return ret;
1251
1252sharing_violation:
1253 up_write(&nommu_region_sem);
1254 pr_warn("Attempt to share mismatched mappings\n");
1255 ret = -EINVAL;
1256 goto error;
1257
1258error_getting_vma:
1259 kmem_cache_free(vm_region_jar, region);
1260 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1261 len, current->pid);
1262 show_free_areas(0, NULL);
1263 return -ENOMEM;
1264
1265error_getting_region:
1266 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1267 len, current->pid);
1268 show_free_areas(0, NULL);
1269 return -ENOMEM;
1270
1271error_maple_preallocate:
1272 kmem_cache_free(vm_region_jar, region);
1273 vm_area_free(vma);
1274 pr_warn("Allocation of vma tree for process %d failed\n", current->pid);
1275 show_free_areas(0, NULL);
1276 return -ENOMEM;
1277
1278}
1279
1280unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1281 unsigned long prot, unsigned long flags,
1282 unsigned long fd, unsigned long pgoff)
1283{
1284 struct file *file = NULL;
1285 unsigned long retval = -EBADF;
1286
1287 audit_mmap_fd(fd, flags);
1288 if (!(flags & MAP_ANONYMOUS)) {
1289 file = fget(fd);
1290 if (!file)
1291 goto out;
1292 }
1293
1294 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1295
1296 if (file)
1297 fput(file);
1298out:
1299 return retval;
1300}
1301
1302SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1303 unsigned long, prot, unsigned long, flags,
1304 unsigned long, fd, unsigned long, pgoff)
1305{
1306 return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1307}
1308
1309#ifdef __ARCH_WANT_SYS_OLD_MMAP
1310struct mmap_arg_struct {
1311 unsigned long addr;
1312 unsigned long len;
1313 unsigned long prot;
1314 unsigned long flags;
1315 unsigned long fd;
1316 unsigned long offset;
1317};
1318
1319SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1320{
1321 struct mmap_arg_struct a;
1322
1323 if (copy_from_user(&a, arg, sizeof(a)))
1324 return -EFAULT;
1325 if (offset_in_page(a.offset))
1326 return -EINVAL;
1327
1328 return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1329 a.offset >> PAGE_SHIFT);
1330}
1331#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1332
1333/*
1334 * split a vma into two pieces at address 'addr', a new vma is allocated either
1335 * for the first part or the tail.
1336 */
1337int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1338 unsigned long addr, int new_below)
1339{
1340 struct vm_area_struct *new;
1341 struct vm_region *region;
1342 unsigned long npages;
1343 MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_end);
1344
1345 /* we're only permitted to split anonymous regions (these should have
1346 * only a single usage on the region) */
1347 if (vma->vm_file)
1348 return -ENOMEM;
1349
1350 mm = vma->vm_mm;
1351 if (mm->map_count >= sysctl_max_map_count)
1352 return -ENOMEM;
1353
1354 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1355 if (!region)
1356 return -ENOMEM;
1357
1358 new = vm_area_dup(vma);
1359 if (!new)
1360 goto err_vma_dup;
1361
1362 if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
1363 pr_warn("Allocation of vma tree for process %d failed\n",
1364 current->pid);
1365 goto err_mas_preallocate;
1366 }
1367
1368 /* most fields are the same, copy all, and then fixup */
1369 *region = *vma->vm_region;
1370 new->vm_region = region;
1371
1372 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1373
1374 if (new_below) {
1375 region->vm_top = region->vm_end = new->vm_end = addr;
1376 } else {
1377 region->vm_start = new->vm_start = addr;
1378 region->vm_pgoff = new->vm_pgoff += npages;
1379 }
1380
1381 if (new->vm_ops && new->vm_ops->open)
1382 new->vm_ops->open(new);
1383
1384 down_write(&nommu_region_sem);
1385 delete_nommu_region(vma->vm_region);
1386 if (new_below) {
1387 vma->vm_region->vm_start = vma->vm_start = addr;
1388 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1389 } else {
1390 vma->vm_region->vm_end = vma->vm_end = addr;
1391 vma->vm_region->vm_top = addr;
1392 }
1393 add_nommu_region(vma->vm_region);
1394 add_nommu_region(new->vm_region);
1395 up_write(&nommu_region_sem);
1396
1397 setup_vma_to_mm(vma, mm);
1398 setup_vma_to_mm(new, mm);
1399 mas_set_range(&mas, vma->vm_start, vma->vm_end - 1);
1400 mas_store(&mas, vma);
1401 vma_mas_store(new, &mas);
1402 mm->map_count++;
1403 return 0;
1404
1405err_mas_preallocate:
1406 vm_area_free(new);
1407err_vma_dup:
1408 kmem_cache_free(vm_region_jar, region);
1409 return -ENOMEM;
1410}
1411
1412/*
1413 * shrink a VMA by removing the specified chunk from either the beginning or
1414 * the end
1415 */
1416static int shrink_vma(struct mm_struct *mm,
1417 struct vm_area_struct *vma,
1418 unsigned long from, unsigned long to)
1419{
1420 struct vm_region *region;
1421
1422 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1423 * and list */
1424 if (delete_vma_from_mm(vma))
1425 return -ENOMEM;
1426 if (from > vma->vm_start)
1427 vma->vm_end = from;
1428 else
1429 vma->vm_start = to;
1430 if (add_vma_to_mm(mm, vma))
1431 return -ENOMEM;
1432
1433 /* cut the backing region down to size */
1434 region = vma->vm_region;
1435 BUG_ON(region->vm_usage != 1);
1436
1437 down_write(&nommu_region_sem);
1438 delete_nommu_region(region);
1439 if (from > region->vm_start) {
1440 to = region->vm_top;
1441 region->vm_top = region->vm_end = from;
1442 } else {
1443 region->vm_start = to;
1444 }
1445 add_nommu_region(region);
1446 up_write(&nommu_region_sem);
1447
1448 free_page_series(from, to);
1449 return 0;
1450}
1451
1452/*
1453 * release a mapping
1454 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1455 * VMA, though it need not cover the whole VMA
1456 */
1457int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
1458{
1459 MA_STATE(mas, &mm->mm_mt, start, start);
1460 struct vm_area_struct *vma;
1461 unsigned long end;
1462 int ret = 0;
1463
1464 len = PAGE_ALIGN(len);
1465 if (len == 0)
1466 return -EINVAL;
1467
1468 end = start + len;
1469
1470 /* find the first potentially overlapping VMA */
1471 vma = mas_find(&mas, end - 1);
1472 if (!vma) {
1473 static int limit;
1474 if (limit < 5) {
1475 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1476 current->pid, current->comm,
1477 start, start + len - 1);
1478 limit++;
1479 }
1480 return -EINVAL;
1481 }
1482
1483 /* we're allowed to split an anonymous VMA but not a file-backed one */
1484 if (vma->vm_file) {
1485 do {
1486 if (start > vma->vm_start)
1487 return -EINVAL;
1488 if (end == vma->vm_end)
1489 goto erase_whole_vma;
1490 vma = mas_next(&mas, end - 1);
1491 } while (vma);
1492 return -EINVAL;
1493 } else {
1494 /* the chunk must be a subset of the VMA found */
1495 if (start == vma->vm_start && end == vma->vm_end)
1496 goto erase_whole_vma;
1497 if (start < vma->vm_start || end > vma->vm_end)
1498 return -EINVAL;
1499 if (offset_in_page(start))
1500 return -EINVAL;
1501 if (end != vma->vm_end && offset_in_page(end))
1502 return -EINVAL;
1503 if (start != vma->vm_start && end != vma->vm_end) {
1504 ret = split_vma(mm, vma, start, 1);
1505 if (ret < 0)
1506 return ret;
1507 }
1508 return shrink_vma(mm, vma, start, end);
1509 }
1510
1511erase_whole_vma:
1512 if (delete_vma_from_mm(vma))
1513 ret = -ENOMEM;
1514 else
1515 delete_vma(mm, vma);
1516 return ret;
1517}
1518
1519int vm_munmap(unsigned long addr, size_t len)
1520{
1521 struct mm_struct *mm = current->mm;
1522 int ret;
1523
1524 mmap_write_lock(mm);
1525 ret = do_munmap(mm, addr, len, NULL);
1526 mmap_write_unlock(mm);
1527 return ret;
1528}
1529EXPORT_SYMBOL(vm_munmap);
1530
1531SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1532{
1533 return vm_munmap(addr, len);
1534}
1535
1536/*
1537 * release all the mappings made in a process's VM space
1538 */
1539void exit_mmap(struct mm_struct *mm)
1540{
1541 VMA_ITERATOR(vmi, mm, 0);
1542 struct vm_area_struct *vma;
1543
1544 if (!mm)
1545 return;
1546
1547 mm->total_vm = 0;
1548
1549 /*
1550 * Lock the mm to avoid assert complaining even though this is the only
1551 * user of the mm
1552 */
1553 mmap_write_lock(mm);
1554 for_each_vma(vmi, vma) {
1555 cleanup_vma_from_mm(vma);
1556 delete_vma(mm, vma);
1557 cond_resched();
1558 }
1559 __mt_destroy(&mm->mm_mt);
1560 mmap_write_unlock(mm);
1561}
1562
1563int vm_brk(unsigned long addr, unsigned long len)
1564{
1565 return -ENOMEM;
1566}
1567
1568/*
1569 * expand (or shrink) an existing mapping, potentially moving it at the same
1570 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1571 *
1572 * under NOMMU conditions, we only permit changing a mapping's size, and only
1573 * as long as it stays within the region allocated by do_mmap_private() and the
1574 * block is not shareable
1575 *
1576 * MREMAP_FIXED is not supported under NOMMU conditions
1577 */
1578static unsigned long do_mremap(unsigned long addr,
1579 unsigned long old_len, unsigned long new_len,
1580 unsigned long flags, unsigned long new_addr)
1581{
1582 struct vm_area_struct *vma;
1583
1584 /* insanity checks first */
1585 old_len = PAGE_ALIGN(old_len);
1586 new_len = PAGE_ALIGN(new_len);
1587 if (old_len == 0 || new_len == 0)
1588 return (unsigned long) -EINVAL;
1589
1590 if (offset_in_page(addr))
1591 return -EINVAL;
1592
1593 if (flags & MREMAP_FIXED && new_addr != addr)
1594 return (unsigned long) -EINVAL;
1595
1596 vma = find_vma_exact(current->mm, addr, old_len);
1597 if (!vma)
1598 return (unsigned long) -EINVAL;
1599
1600 if (vma->vm_end != vma->vm_start + old_len)
1601 return (unsigned long) -EFAULT;
1602
1603 if (vma->vm_flags & VM_MAYSHARE)
1604 return (unsigned long) -EPERM;
1605
1606 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1607 return (unsigned long) -ENOMEM;
1608
1609 /* all checks complete - do it */
1610 vma->vm_end = vma->vm_start + new_len;
1611 return vma->vm_start;
1612}
1613
1614SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1615 unsigned long, new_len, unsigned long, flags,
1616 unsigned long, new_addr)
1617{
1618 unsigned long ret;
1619
1620 mmap_write_lock(current->mm);
1621 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1622 mmap_write_unlock(current->mm);
1623 return ret;
1624}
1625
1626struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1627 unsigned int foll_flags)
1628{
1629 return NULL;
1630}
1631
1632int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1633 unsigned long pfn, unsigned long size, pgprot_t prot)
1634{
1635 if (addr != (pfn << PAGE_SHIFT))
1636 return -EINVAL;
1637
1638 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1639 return 0;
1640}
1641EXPORT_SYMBOL(remap_pfn_range);
1642
1643int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1644{
1645 unsigned long pfn = start >> PAGE_SHIFT;
1646 unsigned long vm_len = vma->vm_end - vma->vm_start;
1647
1648 pfn += vma->vm_pgoff;
1649 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1650}
1651EXPORT_SYMBOL(vm_iomap_memory);
1652
1653int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1654 unsigned long pgoff)
1655{
1656 unsigned int size = vma->vm_end - vma->vm_start;
1657
1658 if (!(vma->vm_flags & VM_USERMAP))
1659 return -EINVAL;
1660
1661 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1662 vma->vm_end = vma->vm_start + size;
1663
1664 return 0;
1665}
1666EXPORT_SYMBOL(remap_vmalloc_range);
1667
1668vm_fault_t filemap_fault(struct vm_fault *vmf)
1669{
1670 BUG();
1671 return 0;
1672}
1673EXPORT_SYMBOL(filemap_fault);
1674
1675vm_fault_t filemap_map_pages(struct vm_fault *vmf,
1676 pgoff_t start_pgoff, pgoff_t end_pgoff)
1677{
1678 BUG();
1679 return 0;
1680}
1681EXPORT_SYMBOL(filemap_map_pages);
1682
1683int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf,
1684 int len, unsigned int gup_flags)
1685{
1686 struct vm_area_struct *vma;
1687 int write = gup_flags & FOLL_WRITE;
1688
1689 if (mmap_read_lock_killable(mm))
1690 return 0;
1691
1692 /* the access must start within one of the target process's mappings */
1693 vma = find_vma(mm, addr);
1694 if (vma) {
1695 /* don't overrun this mapping */
1696 if (addr + len >= vma->vm_end)
1697 len = vma->vm_end - addr;
1698
1699 /* only read or write mappings where it is permitted */
1700 if (write && vma->vm_flags & VM_MAYWRITE)
1701 copy_to_user_page(vma, NULL, addr,
1702 (void *) addr, buf, len);
1703 else if (!write && vma->vm_flags & VM_MAYREAD)
1704 copy_from_user_page(vma, NULL, addr,
1705 buf, (void *) addr, len);
1706 else
1707 len = 0;
1708 } else {
1709 len = 0;
1710 }
1711
1712 mmap_read_unlock(mm);
1713
1714 return len;
1715}
1716
1717/**
1718 * access_remote_vm - access another process' address space
1719 * @mm: the mm_struct of the target address space
1720 * @addr: start address to access
1721 * @buf: source or destination buffer
1722 * @len: number of bytes to transfer
1723 * @gup_flags: flags modifying lookup behaviour
1724 *
1725 * The caller must hold a reference on @mm.
1726 */
1727int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1728 void *buf, int len, unsigned int gup_flags)
1729{
1730 return __access_remote_vm(mm, addr, buf, len, gup_flags);
1731}
1732
1733/*
1734 * Access another process' address space.
1735 * - source/target buffer must be kernel space
1736 */
1737int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1738 unsigned int gup_flags)
1739{
1740 struct mm_struct *mm;
1741
1742 if (addr + len < addr)
1743 return 0;
1744
1745 mm = get_task_mm(tsk);
1746 if (!mm)
1747 return 0;
1748
1749 len = __access_remote_vm(mm, addr, buf, len, gup_flags);
1750
1751 mmput(mm);
1752 return len;
1753}
1754EXPORT_SYMBOL_GPL(access_process_vm);
1755
1756/**
1757 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1758 * @inode: The inode to check
1759 * @size: The current filesize of the inode
1760 * @newsize: The proposed filesize of the inode
1761 *
1762 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1763 * make sure that any outstanding VMAs aren't broken and then shrink the
1764 * vm_regions that extend beyond so that do_mmap() doesn't
1765 * automatically grant mappings that are too large.
1766 */
1767int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1768 size_t newsize)
1769{
1770 struct vm_area_struct *vma;
1771 struct vm_region *region;
1772 pgoff_t low, high;
1773 size_t r_size, r_top;
1774
1775 low = newsize >> PAGE_SHIFT;
1776 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1777
1778 down_write(&nommu_region_sem);
1779 i_mmap_lock_read(inode->i_mapping);
1780
1781 /* search for VMAs that fall within the dead zone */
1782 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1783 /* found one - only interested if it's shared out of the page
1784 * cache */
1785 if (vma->vm_flags & VM_SHARED) {
1786 i_mmap_unlock_read(inode->i_mapping);
1787 up_write(&nommu_region_sem);
1788 return -ETXTBSY; /* not quite true, but near enough */
1789 }
1790 }
1791
1792 /* reduce any regions that overlap the dead zone - if in existence,
1793 * these will be pointed to by VMAs that don't overlap the dead zone
1794 *
1795 * we don't check for any regions that start beyond the EOF as there
1796 * shouldn't be any
1797 */
1798 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1799 if (!(vma->vm_flags & VM_SHARED))
1800 continue;
1801
1802 region = vma->vm_region;
1803 r_size = region->vm_top - region->vm_start;
1804 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1805
1806 if (r_top > newsize) {
1807 region->vm_top -= r_top - newsize;
1808 if (region->vm_end > region->vm_top)
1809 region->vm_end = region->vm_top;
1810 }
1811 }
1812
1813 i_mmap_unlock_read(inode->i_mapping);
1814 up_write(&nommu_region_sem);
1815 return 0;
1816}
1817
1818/*
1819 * Initialise sysctl_user_reserve_kbytes.
1820 *
1821 * This is intended to prevent a user from starting a single memory hogging
1822 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1823 * mode.
1824 *
1825 * The default value is min(3% of free memory, 128MB)
1826 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1827 */
1828static int __meminit init_user_reserve(void)
1829{
1830 unsigned long free_kbytes;
1831
1832 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1833
1834 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1835 return 0;
1836}
1837subsys_initcall(init_user_reserve);
1838
1839/*
1840 * Initialise sysctl_admin_reserve_kbytes.
1841 *
1842 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1843 * to log in and kill a memory hogging process.
1844 *
1845 * Systems with more than 256MB will reserve 8MB, enough to recover
1846 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1847 * only reserve 3% of free pages by default.
1848 */
1849static int __meminit init_admin_reserve(void)
1850{
1851 unsigned long free_kbytes;
1852
1853 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1854
1855 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1856 return 0;
1857}
1858subsys_initcall(init_admin_reserve);
1/*
2 * linux/mm/nommu.c
3 *
4 * Replacement code for mm functions to support CPU's that don't
5 * have any form of memory management unit (thus no virtual memory).
6 *
7 * See Documentation/nommu-mmap.txt
8 *
9 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
10 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
13 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
14 */
15
16#include <linux/export.h>
17#include <linux/mm.h>
18#include <linux/vmacache.h>
19#include <linux/mman.h>
20#include <linux/swap.h>
21#include <linux/file.h>
22#include <linux/highmem.h>
23#include <linux/pagemap.h>
24#include <linux/slab.h>
25#include <linux/vmalloc.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/compiler.h>
29#include <linux/mount.h>
30#include <linux/personality.h>
31#include <linux/security.h>
32#include <linux/syscalls.h>
33#include <linux/audit.h>
34#include <linux/sched/sysctl.h>
35
36#include <asm/uaccess.h>
37#include <asm/tlb.h>
38#include <asm/tlbflush.h>
39#include <asm/mmu_context.h>
40#include "internal.h"
41
42#if 0
43#define kenter(FMT, ...) \
44 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
45#define kleave(FMT, ...) \
46 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
47#define kdebug(FMT, ...) \
48 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
49#else
50#define kenter(FMT, ...) \
51 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
52#define kleave(FMT, ...) \
53 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
54#define kdebug(FMT, ...) \
55 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
56#endif
57
58void *high_memory;
59struct page *mem_map;
60unsigned long max_mapnr;
61unsigned long highest_memmap_pfn;
62struct percpu_counter vm_committed_as;
63int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
64int sysctl_overcommit_ratio = 50; /* default is 50% */
65unsigned long sysctl_overcommit_kbytes __read_mostly;
66int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
67int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
68unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
69unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
70int heap_stack_gap = 0;
71
72atomic_long_t mmap_pages_allocated;
73
74/*
75 * The global memory commitment made in the system can be a metric
76 * that can be used to drive ballooning decisions when Linux is hosted
77 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
78 * balancing memory across competing virtual machines that are hosted.
79 * Several metrics drive this policy engine including the guest reported
80 * memory commitment.
81 */
82unsigned long vm_memory_committed(void)
83{
84 return percpu_counter_read_positive(&vm_committed_as);
85}
86
87EXPORT_SYMBOL_GPL(vm_memory_committed);
88
89EXPORT_SYMBOL(mem_map);
90
91/* list of mapped, potentially shareable regions */
92static struct kmem_cache *vm_region_jar;
93struct rb_root nommu_region_tree = RB_ROOT;
94DECLARE_RWSEM(nommu_region_sem);
95
96const struct vm_operations_struct generic_file_vm_ops = {
97};
98
99/*
100 * Return the total memory allocated for this pointer, not
101 * just what the caller asked for.
102 *
103 * Doesn't have to be accurate, i.e. may have races.
104 */
105unsigned int kobjsize(const void *objp)
106{
107 struct page *page;
108
109 /*
110 * If the object we have should not have ksize performed on it,
111 * return size of 0
112 */
113 if (!objp || !virt_addr_valid(objp))
114 return 0;
115
116 page = virt_to_head_page(objp);
117
118 /*
119 * If the allocator sets PageSlab, we know the pointer came from
120 * kmalloc().
121 */
122 if (PageSlab(page))
123 return ksize(objp);
124
125 /*
126 * If it's not a compound page, see if we have a matching VMA
127 * region. This test is intentionally done in reverse order,
128 * so if there's no VMA, we still fall through and hand back
129 * PAGE_SIZE for 0-order pages.
130 */
131 if (!PageCompound(page)) {
132 struct vm_area_struct *vma;
133
134 vma = find_vma(current->mm, (unsigned long)objp);
135 if (vma)
136 return vma->vm_end - vma->vm_start;
137 }
138
139 /*
140 * The ksize() function is only guaranteed to work for pointers
141 * returned by kmalloc(). So handle arbitrary pointers here.
142 */
143 return PAGE_SIZE << compound_order(page);
144}
145
146long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
147 unsigned long start, unsigned long nr_pages,
148 unsigned int foll_flags, struct page **pages,
149 struct vm_area_struct **vmas, int *nonblocking)
150{
151 struct vm_area_struct *vma;
152 unsigned long vm_flags;
153 int i;
154
155 /* calculate required read or write permissions.
156 * If FOLL_FORCE is set, we only require the "MAY" flags.
157 */
158 vm_flags = (foll_flags & FOLL_WRITE) ?
159 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
160 vm_flags &= (foll_flags & FOLL_FORCE) ?
161 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
162
163 for (i = 0; i < nr_pages; i++) {
164 vma = find_vma(mm, start);
165 if (!vma)
166 goto finish_or_fault;
167
168 /* protect what we can, including chardevs */
169 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
170 !(vm_flags & vma->vm_flags))
171 goto finish_or_fault;
172
173 if (pages) {
174 pages[i] = virt_to_page(start);
175 if (pages[i])
176 page_cache_get(pages[i]);
177 }
178 if (vmas)
179 vmas[i] = vma;
180 start = (start + PAGE_SIZE) & PAGE_MASK;
181 }
182
183 return i;
184
185finish_or_fault:
186 return i ? : -EFAULT;
187}
188
189/*
190 * get a list of pages in an address range belonging to the specified process
191 * and indicate the VMA that covers each page
192 * - this is potentially dodgy as we may end incrementing the page count of a
193 * slab page or a secondary page from a compound page
194 * - don't permit access to VMAs that don't support it, such as I/O mappings
195 */
196long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
197 unsigned long start, unsigned long nr_pages,
198 int write, int force, struct page **pages,
199 struct vm_area_struct **vmas)
200{
201 int flags = 0;
202
203 if (write)
204 flags |= FOLL_WRITE;
205 if (force)
206 flags |= FOLL_FORCE;
207
208 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
209 NULL);
210}
211EXPORT_SYMBOL(get_user_pages);
212
213/**
214 * follow_pfn - look up PFN at a user virtual address
215 * @vma: memory mapping
216 * @address: user virtual address
217 * @pfn: location to store found PFN
218 *
219 * Only IO mappings and raw PFN mappings are allowed.
220 *
221 * Returns zero and the pfn at @pfn on success, -ve otherwise.
222 */
223int follow_pfn(struct vm_area_struct *vma, unsigned long address,
224 unsigned long *pfn)
225{
226 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
227 return -EINVAL;
228
229 *pfn = address >> PAGE_SHIFT;
230 return 0;
231}
232EXPORT_SYMBOL(follow_pfn);
233
234LIST_HEAD(vmap_area_list);
235
236void vfree(const void *addr)
237{
238 kfree(addr);
239}
240EXPORT_SYMBOL(vfree);
241
242void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
243{
244 /*
245 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
246 * returns only a logical address.
247 */
248 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
249}
250EXPORT_SYMBOL(__vmalloc);
251
252void *vmalloc_user(unsigned long size)
253{
254 void *ret;
255
256 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
257 PAGE_KERNEL);
258 if (ret) {
259 struct vm_area_struct *vma;
260
261 down_write(¤t->mm->mmap_sem);
262 vma = find_vma(current->mm, (unsigned long)ret);
263 if (vma)
264 vma->vm_flags |= VM_USERMAP;
265 up_write(¤t->mm->mmap_sem);
266 }
267
268 return ret;
269}
270EXPORT_SYMBOL(vmalloc_user);
271
272struct page *vmalloc_to_page(const void *addr)
273{
274 return virt_to_page(addr);
275}
276EXPORT_SYMBOL(vmalloc_to_page);
277
278unsigned long vmalloc_to_pfn(const void *addr)
279{
280 return page_to_pfn(virt_to_page(addr));
281}
282EXPORT_SYMBOL(vmalloc_to_pfn);
283
284long vread(char *buf, char *addr, unsigned long count)
285{
286 /* Don't allow overflow */
287 if ((unsigned long) buf + count < count)
288 count = -(unsigned long) buf;
289
290 memcpy(buf, addr, count);
291 return count;
292}
293
294long vwrite(char *buf, char *addr, unsigned long count)
295{
296 /* Don't allow overflow */
297 if ((unsigned long) addr + count < count)
298 count = -(unsigned long) addr;
299
300 memcpy(addr, buf, count);
301 return count;
302}
303
304/*
305 * vmalloc - allocate virtually continguos memory
306 *
307 * @size: allocation size
308 *
309 * Allocate enough pages to cover @size from the page level
310 * allocator and map them into continguos kernel virtual space.
311 *
312 * For tight control over page level allocator and protection flags
313 * use __vmalloc() instead.
314 */
315void *vmalloc(unsigned long size)
316{
317 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
318}
319EXPORT_SYMBOL(vmalloc);
320
321/*
322 * vzalloc - allocate virtually continguos memory with zero fill
323 *
324 * @size: allocation size
325 *
326 * Allocate enough pages to cover @size from the page level
327 * allocator and map them into continguos kernel virtual space.
328 * The memory allocated is set to zero.
329 *
330 * For tight control over page level allocator and protection flags
331 * use __vmalloc() instead.
332 */
333void *vzalloc(unsigned long size)
334{
335 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
336 PAGE_KERNEL);
337}
338EXPORT_SYMBOL(vzalloc);
339
340/**
341 * vmalloc_node - allocate memory on a specific node
342 * @size: allocation size
343 * @node: numa node
344 *
345 * Allocate enough pages to cover @size from the page level
346 * allocator and map them into contiguous kernel virtual space.
347 *
348 * For tight control over page level allocator and protection flags
349 * use __vmalloc() instead.
350 */
351void *vmalloc_node(unsigned long size, int node)
352{
353 return vmalloc(size);
354}
355EXPORT_SYMBOL(vmalloc_node);
356
357/**
358 * vzalloc_node - allocate memory on a specific node with zero fill
359 * @size: allocation size
360 * @node: numa node
361 *
362 * Allocate enough pages to cover @size from the page level
363 * allocator and map them into contiguous kernel virtual space.
364 * The memory allocated is set to zero.
365 *
366 * For tight control over page level allocator and protection flags
367 * use __vmalloc() instead.
368 */
369void *vzalloc_node(unsigned long size, int node)
370{
371 return vzalloc(size);
372}
373EXPORT_SYMBOL(vzalloc_node);
374
375#ifndef PAGE_KERNEL_EXEC
376# define PAGE_KERNEL_EXEC PAGE_KERNEL
377#endif
378
379/**
380 * vmalloc_exec - allocate virtually contiguous, executable memory
381 * @size: allocation size
382 *
383 * Kernel-internal function to allocate enough pages to cover @size
384 * the page level allocator and map them into contiguous and
385 * executable kernel virtual space.
386 *
387 * For tight control over page level allocator and protection flags
388 * use __vmalloc() instead.
389 */
390
391void *vmalloc_exec(unsigned long size)
392{
393 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
394}
395
396/**
397 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
398 * @size: allocation size
399 *
400 * Allocate enough 32bit PA addressable pages to cover @size from the
401 * page level allocator and map them into continguos kernel virtual space.
402 */
403void *vmalloc_32(unsigned long size)
404{
405 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
406}
407EXPORT_SYMBOL(vmalloc_32);
408
409/**
410 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
411 * @size: allocation size
412 *
413 * The resulting memory area is 32bit addressable and zeroed so it can be
414 * mapped to userspace without leaking data.
415 *
416 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
417 * remap_vmalloc_range() are permissible.
418 */
419void *vmalloc_32_user(unsigned long size)
420{
421 /*
422 * We'll have to sort out the ZONE_DMA bits for 64-bit,
423 * but for now this can simply use vmalloc_user() directly.
424 */
425 return vmalloc_user(size);
426}
427EXPORT_SYMBOL(vmalloc_32_user);
428
429void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
430{
431 BUG();
432 return NULL;
433}
434EXPORT_SYMBOL(vmap);
435
436void vunmap(const void *addr)
437{
438 BUG();
439}
440EXPORT_SYMBOL(vunmap);
441
442void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
443{
444 BUG();
445 return NULL;
446}
447EXPORT_SYMBOL(vm_map_ram);
448
449void vm_unmap_ram(const void *mem, unsigned int count)
450{
451 BUG();
452}
453EXPORT_SYMBOL(vm_unmap_ram);
454
455void vm_unmap_aliases(void)
456{
457}
458EXPORT_SYMBOL_GPL(vm_unmap_aliases);
459
460/*
461 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
462 * have one.
463 */
464void __weak vmalloc_sync_all(void)
465{
466}
467
468/**
469 * alloc_vm_area - allocate a range of kernel address space
470 * @size: size of the area
471 *
472 * Returns: NULL on failure, vm_struct on success
473 *
474 * This function reserves a range of kernel address space, and
475 * allocates pagetables to map that range. No actual mappings
476 * are created. If the kernel address space is not shared
477 * between processes, it syncs the pagetable across all
478 * processes.
479 */
480struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
481{
482 BUG();
483 return NULL;
484}
485EXPORT_SYMBOL_GPL(alloc_vm_area);
486
487void free_vm_area(struct vm_struct *area)
488{
489 BUG();
490}
491EXPORT_SYMBOL_GPL(free_vm_area);
492
493int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
494 struct page *page)
495{
496 return -EINVAL;
497}
498EXPORT_SYMBOL(vm_insert_page);
499
500/*
501 * sys_brk() for the most part doesn't need the global kernel
502 * lock, except when an application is doing something nasty
503 * like trying to un-brk an area that has already been mapped
504 * to a regular file. in this case, the unmapping will need
505 * to invoke file system routines that need the global lock.
506 */
507SYSCALL_DEFINE1(brk, unsigned long, brk)
508{
509 struct mm_struct *mm = current->mm;
510
511 if (brk < mm->start_brk || brk > mm->context.end_brk)
512 return mm->brk;
513
514 if (mm->brk == brk)
515 return mm->brk;
516
517 /*
518 * Always allow shrinking brk
519 */
520 if (brk <= mm->brk) {
521 mm->brk = brk;
522 return brk;
523 }
524
525 /*
526 * Ok, looks good - let it rip.
527 */
528 flush_icache_range(mm->brk, brk);
529 return mm->brk = brk;
530}
531
532/*
533 * initialise the VMA and region record slabs
534 */
535void __init mmap_init(void)
536{
537 int ret;
538
539 ret = percpu_counter_init(&vm_committed_as, 0);
540 VM_BUG_ON(ret);
541 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
542}
543
544/*
545 * validate the region tree
546 * - the caller must hold the region lock
547 */
548#ifdef CONFIG_DEBUG_NOMMU_REGIONS
549static noinline void validate_nommu_regions(void)
550{
551 struct vm_region *region, *last;
552 struct rb_node *p, *lastp;
553
554 lastp = rb_first(&nommu_region_tree);
555 if (!lastp)
556 return;
557
558 last = rb_entry(lastp, struct vm_region, vm_rb);
559 BUG_ON(unlikely(last->vm_end <= last->vm_start));
560 BUG_ON(unlikely(last->vm_top < last->vm_end));
561
562 while ((p = rb_next(lastp))) {
563 region = rb_entry(p, struct vm_region, vm_rb);
564 last = rb_entry(lastp, struct vm_region, vm_rb);
565
566 BUG_ON(unlikely(region->vm_end <= region->vm_start));
567 BUG_ON(unlikely(region->vm_top < region->vm_end));
568 BUG_ON(unlikely(region->vm_start < last->vm_top));
569
570 lastp = p;
571 }
572}
573#else
574static void validate_nommu_regions(void)
575{
576}
577#endif
578
579/*
580 * add a region into the global tree
581 */
582static void add_nommu_region(struct vm_region *region)
583{
584 struct vm_region *pregion;
585 struct rb_node **p, *parent;
586
587 validate_nommu_regions();
588
589 parent = NULL;
590 p = &nommu_region_tree.rb_node;
591 while (*p) {
592 parent = *p;
593 pregion = rb_entry(parent, struct vm_region, vm_rb);
594 if (region->vm_start < pregion->vm_start)
595 p = &(*p)->rb_left;
596 else if (region->vm_start > pregion->vm_start)
597 p = &(*p)->rb_right;
598 else if (pregion == region)
599 return;
600 else
601 BUG();
602 }
603
604 rb_link_node(®ion->vm_rb, parent, p);
605 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
606
607 validate_nommu_regions();
608}
609
610/*
611 * delete a region from the global tree
612 */
613static void delete_nommu_region(struct vm_region *region)
614{
615 BUG_ON(!nommu_region_tree.rb_node);
616
617 validate_nommu_regions();
618 rb_erase(®ion->vm_rb, &nommu_region_tree);
619 validate_nommu_regions();
620}
621
622/*
623 * free a contiguous series of pages
624 */
625static void free_page_series(unsigned long from, unsigned long to)
626{
627 for (; from < to; from += PAGE_SIZE) {
628 struct page *page = virt_to_page(from);
629
630 kdebug("- free %lx", from);
631 atomic_long_dec(&mmap_pages_allocated);
632 if (page_count(page) != 1)
633 kdebug("free page %p: refcount not one: %d",
634 page, page_count(page));
635 put_page(page);
636 }
637}
638
639/*
640 * release a reference to a region
641 * - the caller must hold the region semaphore for writing, which this releases
642 * - the region may not have been added to the tree yet, in which case vm_top
643 * will equal vm_start
644 */
645static void __put_nommu_region(struct vm_region *region)
646 __releases(nommu_region_sem)
647{
648 kenter("%p{%d}", region, region->vm_usage);
649
650 BUG_ON(!nommu_region_tree.rb_node);
651
652 if (--region->vm_usage == 0) {
653 if (region->vm_top > region->vm_start)
654 delete_nommu_region(region);
655 up_write(&nommu_region_sem);
656
657 if (region->vm_file)
658 fput(region->vm_file);
659
660 /* IO memory and memory shared directly out of the pagecache
661 * from ramfs/tmpfs mustn't be released here */
662 if (region->vm_flags & VM_MAPPED_COPY) {
663 kdebug("free series");
664 free_page_series(region->vm_start, region->vm_top);
665 }
666 kmem_cache_free(vm_region_jar, region);
667 } else {
668 up_write(&nommu_region_sem);
669 }
670}
671
672/*
673 * release a reference to a region
674 */
675static void put_nommu_region(struct vm_region *region)
676{
677 down_write(&nommu_region_sem);
678 __put_nommu_region(region);
679}
680
681/*
682 * update protection on a vma
683 */
684static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
685{
686#ifdef CONFIG_MPU
687 struct mm_struct *mm = vma->vm_mm;
688 long start = vma->vm_start & PAGE_MASK;
689 while (start < vma->vm_end) {
690 protect_page(mm, start, flags);
691 start += PAGE_SIZE;
692 }
693 update_protections(mm);
694#endif
695}
696
697/*
698 * add a VMA into a process's mm_struct in the appropriate place in the list
699 * and tree and add to the address space's page tree also if not an anonymous
700 * page
701 * - should be called with mm->mmap_sem held writelocked
702 */
703static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
704{
705 struct vm_area_struct *pvma, *prev;
706 struct address_space *mapping;
707 struct rb_node **p, *parent, *rb_prev;
708
709 kenter(",%p", vma);
710
711 BUG_ON(!vma->vm_region);
712
713 mm->map_count++;
714 vma->vm_mm = mm;
715
716 protect_vma(vma, vma->vm_flags);
717
718 /* add the VMA to the mapping */
719 if (vma->vm_file) {
720 mapping = vma->vm_file->f_mapping;
721
722 mutex_lock(&mapping->i_mmap_mutex);
723 flush_dcache_mmap_lock(mapping);
724 vma_interval_tree_insert(vma, &mapping->i_mmap);
725 flush_dcache_mmap_unlock(mapping);
726 mutex_unlock(&mapping->i_mmap_mutex);
727 }
728
729 /* add the VMA to the tree */
730 parent = rb_prev = NULL;
731 p = &mm->mm_rb.rb_node;
732 while (*p) {
733 parent = *p;
734 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
735
736 /* sort by: start addr, end addr, VMA struct addr in that order
737 * (the latter is necessary as we may get identical VMAs) */
738 if (vma->vm_start < pvma->vm_start)
739 p = &(*p)->rb_left;
740 else if (vma->vm_start > pvma->vm_start) {
741 rb_prev = parent;
742 p = &(*p)->rb_right;
743 } else if (vma->vm_end < pvma->vm_end)
744 p = &(*p)->rb_left;
745 else if (vma->vm_end > pvma->vm_end) {
746 rb_prev = parent;
747 p = &(*p)->rb_right;
748 } else if (vma < pvma)
749 p = &(*p)->rb_left;
750 else if (vma > pvma) {
751 rb_prev = parent;
752 p = &(*p)->rb_right;
753 } else
754 BUG();
755 }
756
757 rb_link_node(&vma->vm_rb, parent, p);
758 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
759
760 /* add VMA to the VMA list also */
761 prev = NULL;
762 if (rb_prev)
763 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
764
765 __vma_link_list(mm, vma, prev, parent);
766}
767
768/*
769 * delete a VMA from its owning mm_struct and address space
770 */
771static void delete_vma_from_mm(struct vm_area_struct *vma)
772{
773 int i;
774 struct address_space *mapping;
775 struct mm_struct *mm = vma->vm_mm;
776 struct task_struct *curr = current;
777
778 kenter("%p", vma);
779
780 protect_vma(vma, 0);
781
782 mm->map_count--;
783 for (i = 0; i < VMACACHE_SIZE; i++) {
784 /* if the vma is cached, invalidate the entire cache */
785 if (curr->vmacache[i] == vma) {
786 vmacache_invalidate(curr->mm);
787 break;
788 }
789 }
790
791 /* remove the VMA from the mapping */
792 if (vma->vm_file) {
793 mapping = vma->vm_file->f_mapping;
794
795 mutex_lock(&mapping->i_mmap_mutex);
796 flush_dcache_mmap_lock(mapping);
797 vma_interval_tree_remove(vma, &mapping->i_mmap);
798 flush_dcache_mmap_unlock(mapping);
799 mutex_unlock(&mapping->i_mmap_mutex);
800 }
801
802 /* remove from the MM's tree and list */
803 rb_erase(&vma->vm_rb, &mm->mm_rb);
804
805 if (vma->vm_prev)
806 vma->vm_prev->vm_next = vma->vm_next;
807 else
808 mm->mmap = vma->vm_next;
809
810 if (vma->vm_next)
811 vma->vm_next->vm_prev = vma->vm_prev;
812}
813
814/*
815 * destroy a VMA record
816 */
817static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
818{
819 kenter("%p", vma);
820 if (vma->vm_ops && vma->vm_ops->close)
821 vma->vm_ops->close(vma);
822 if (vma->vm_file)
823 fput(vma->vm_file);
824 put_nommu_region(vma->vm_region);
825 kmem_cache_free(vm_area_cachep, vma);
826}
827
828/*
829 * look up the first VMA in which addr resides, NULL if none
830 * - should be called with mm->mmap_sem at least held readlocked
831 */
832struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
833{
834 struct vm_area_struct *vma;
835
836 /* check the cache first */
837 vma = vmacache_find(mm, addr);
838 if (likely(vma))
839 return vma;
840
841 /* trawl the list (there may be multiple mappings in which addr
842 * resides) */
843 for (vma = mm->mmap; vma; vma = vma->vm_next) {
844 if (vma->vm_start > addr)
845 return NULL;
846 if (vma->vm_end > addr) {
847 vmacache_update(addr, vma);
848 return vma;
849 }
850 }
851
852 return NULL;
853}
854EXPORT_SYMBOL(find_vma);
855
856/*
857 * find a VMA
858 * - we don't extend stack VMAs under NOMMU conditions
859 */
860struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
861{
862 return find_vma(mm, addr);
863}
864
865/*
866 * expand a stack to a given address
867 * - not supported under NOMMU conditions
868 */
869int expand_stack(struct vm_area_struct *vma, unsigned long address)
870{
871 return -ENOMEM;
872}
873
874/*
875 * look up the first VMA exactly that exactly matches addr
876 * - should be called with mm->mmap_sem at least held readlocked
877 */
878static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
879 unsigned long addr,
880 unsigned long len)
881{
882 struct vm_area_struct *vma;
883 unsigned long end = addr + len;
884
885 /* check the cache first */
886 vma = vmacache_find_exact(mm, addr, end);
887 if (vma)
888 return vma;
889
890 /* trawl the list (there may be multiple mappings in which addr
891 * resides) */
892 for (vma = mm->mmap; vma; vma = vma->vm_next) {
893 if (vma->vm_start < addr)
894 continue;
895 if (vma->vm_start > addr)
896 return NULL;
897 if (vma->vm_end == end) {
898 vmacache_update(addr, vma);
899 return vma;
900 }
901 }
902
903 return NULL;
904}
905
906/*
907 * determine whether a mapping should be permitted and, if so, what sort of
908 * mapping we're capable of supporting
909 */
910static int validate_mmap_request(struct file *file,
911 unsigned long addr,
912 unsigned long len,
913 unsigned long prot,
914 unsigned long flags,
915 unsigned long pgoff,
916 unsigned long *_capabilities)
917{
918 unsigned long capabilities, rlen;
919 int ret;
920
921 /* do the simple checks first */
922 if (flags & MAP_FIXED) {
923 printk(KERN_DEBUG
924 "%d: Can't do fixed-address/overlay mmap of RAM\n",
925 current->pid);
926 return -EINVAL;
927 }
928
929 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
930 (flags & MAP_TYPE) != MAP_SHARED)
931 return -EINVAL;
932
933 if (!len)
934 return -EINVAL;
935
936 /* Careful about overflows.. */
937 rlen = PAGE_ALIGN(len);
938 if (!rlen || rlen > TASK_SIZE)
939 return -ENOMEM;
940
941 /* offset overflow? */
942 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
943 return -EOVERFLOW;
944
945 if (file) {
946 /* validate file mapping requests */
947 struct address_space *mapping;
948
949 /* files must support mmap */
950 if (!file->f_op->mmap)
951 return -ENODEV;
952
953 /* work out if what we've got could possibly be shared
954 * - we support chardevs that provide their own "memory"
955 * - we support files/blockdevs that are memory backed
956 */
957 mapping = file->f_mapping;
958 if (!mapping)
959 mapping = file_inode(file)->i_mapping;
960
961 capabilities = 0;
962 if (mapping && mapping->backing_dev_info)
963 capabilities = mapping->backing_dev_info->capabilities;
964
965 if (!capabilities) {
966 /* no explicit capabilities set, so assume some
967 * defaults */
968 switch (file_inode(file)->i_mode & S_IFMT) {
969 case S_IFREG:
970 case S_IFBLK:
971 capabilities = BDI_CAP_MAP_COPY;
972 break;
973
974 case S_IFCHR:
975 capabilities =
976 BDI_CAP_MAP_DIRECT |
977 BDI_CAP_READ_MAP |
978 BDI_CAP_WRITE_MAP;
979 break;
980
981 default:
982 return -EINVAL;
983 }
984 }
985
986 /* eliminate any capabilities that we can't support on this
987 * device */
988 if (!file->f_op->get_unmapped_area)
989 capabilities &= ~BDI_CAP_MAP_DIRECT;
990 if (!file->f_op->read)
991 capabilities &= ~BDI_CAP_MAP_COPY;
992
993 /* The file shall have been opened with read permission. */
994 if (!(file->f_mode & FMODE_READ))
995 return -EACCES;
996
997 if (flags & MAP_SHARED) {
998 /* do checks for writing, appending and locking */
999 if ((prot & PROT_WRITE) &&
1000 !(file->f_mode & FMODE_WRITE))
1001 return -EACCES;
1002
1003 if (IS_APPEND(file_inode(file)) &&
1004 (file->f_mode & FMODE_WRITE))
1005 return -EACCES;
1006
1007 if (locks_verify_locked(file))
1008 return -EAGAIN;
1009
1010 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1011 return -ENODEV;
1012
1013 /* we mustn't privatise shared mappings */
1014 capabilities &= ~BDI_CAP_MAP_COPY;
1015 } else {
1016 /* we're going to read the file into private memory we
1017 * allocate */
1018 if (!(capabilities & BDI_CAP_MAP_COPY))
1019 return -ENODEV;
1020
1021 /* we don't permit a private writable mapping to be
1022 * shared with the backing device */
1023 if (prot & PROT_WRITE)
1024 capabilities &= ~BDI_CAP_MAP_DIRECT;
1025 }
1026
1027 if (capabilities & BDI_CAP_MAP_DIRECT) {
1028 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
1029 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1030 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
1031 ) {
1032 capabilities &= ~BDI_CAP_MAP_DIRECT;
1033 if (flags & MAP_SHARED) {
1034 printk(KERN_WARNING
1035 "MAP_SHARED not completely supported on !MMU\n");
1036 return -EINVAL;
1037 }
1038 }
1039 }
1040
1041 /* handle executable mappings and implied executable
1042 * mappings */
1043 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1044 if (prot & PROT_EXEC)
1045 return -EPERM;
1046 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1047 /* handle implication of PROT_EXEC by PROT_READ */
1048 if (current->personality & READ_IMPLIES_EXEC) {
1049 if (capabilities & BDI_CAP_EXEC_MAP)
1050 prot |= PROT_EXEC;
1051 }
1052 } else if ((prot & PROT_READ) &&
1053 (prot & PROT_EXEC) &&
1054 !(capabilities & BDI_CAP_EXEC_MAP)
1055 ) {
1056 /* backing file is not executable, try to copy */
1057 capabilities &= ~BDI_CAP_MAP_DIRECT;
1058 }
1059 } else {
1060 /* anonymous mappings are always memory backed and can be
1061 * privately mapped
1062 */
1063 capabilities = BDI_CAP_MAP_COPY;
1064
1065 /* handle PROT_EXEC implication by PROT_READ */
1066 if ((prot & PROT_READ) &&
1067 (current->personality & READ_IMPLIES_EXEC))
1068 prot |= PROT_EXEC;
1069 }
1070
1071 /* allow the security API to have its say */
1072 ret = security_mmap_addr(addr);
1073 if (ret < 0)
1074 return ret;
1075
1076 /* looks okay */
1077 *_capabilities = capabilities;
1078 return 0;
1079}
1080
1081/*
1082 * we've determined that we can make the mapping, now translate what we
1083 * now know into VMA flags
1084 */
1085static unsigned long determine_vm_flags(struct file *file,
1086 unsigned long prot,
1087 unsigned long flags,
1088 unsigned long capabilities)
1089{
1090 unsigned long vm_flags;
1091
1092 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1093 /* vm_flags |= mm->def_flags; */
1094
1095 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1096 /* attempt to share read-only copies of mapped file chunks */
1097 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1098 if (file && !(prot & PROT_WRITE))
1099 vm_flags |= VM_MAYSHARE;
1100 } else {
1101 /* overlay a shareable mapping on the backing device or inode
1102 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1103 * romfs/cramfs */
1104 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1105 if (flags & MAP_SHARED)
1106 vm_flags |= VM_SHARED;
1107 }
1108
1109 /* refuse to let anyone share private mappings with this process if
1110 * it's being traced - otherwise breakpoints set in it may interfere
1111 * with another untraced process
1112 */
1113 if ((flags & MAP_PRIVATE) && current->ptrace)
1114 vm_flags &= ~VM_MAYSHARE;
1115
1116 return vm_flags;
1117}
1118
1119/*
1120 * set up a shared mapping on a file (the driver or filesystem provides and
1121 * pins the storage)
1122 */
1123static int do_mmap_shared_file(struct vm_area_struct *vma)
1124{
1125 int ret;
1126
1127 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1128 if (ret == 0) {
1129 vma->vm_region->vm_top = vma->vm_region->vm_end;
1130 return 0;
1131 }
1132 if (ret != -ENOSYS)
1133 return ret;
1134
1135 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1136 * opposed to tried but failed) so we can only give a suitable error as
1137 * it's not possible to make a private copy if MAP_SHARED was given */
1138 return -ENODEV;
1139}
1140
1141/*
1142 * set up a private mapping or an anonymous shared mapping
1143 */
1144static int do_mmap_private(struct vm_area_struct *vma,
1145 struct vm_region *region,
1146 unsigned long len,
1147 unsigned long capabilities)
1148{
1149 struct page *pages;
1150 unsigned long total, point, n;
1151 void *base;
1152 int ret, order;
1153
1154 /* invoke the file's mapping function so that it can keep track of
1155 * shared mappings on devices or memory
1156 * - VM_MAYSHARE will be set if it may attempt to share
1157 */
1158 if (capabilities & BDI_CAP_MAP_DIRECT) {
1159 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1160 if (ret == 0) {
1161 /* shouldn't return success if we're not sharing */
1162 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1163 vma->vm_region->vm_top = vma->vm_region->vm_end;
1164 return 0;
1165 }
1166 if (ret != -ENOSYS)
1167 return ret;
1168
1169 /* getting an ENOSYS error indicates that direct mmap isn't
1170 * possible (as opposed to tried but failed) so we'll try to
1171 * make a private copy of the data and map that instead */
1172 }
1173
1174
1175 /* allocate some memory to hold the mapping
1176 * - note that this may not return a page-aligned address if the object
1177 * we're allocating is smaller than a page
1178 */
1179 order = get_order(len);
1180 kdebug("alloc order %d for %lx", order, len);
1181
1182 pages = alloc_pages(GFP_KERNEL, order);
1183 if (!pages)
1184 goto enomem;
1185
1186 total = 1 << order;
1187 atomic_long_add(total, &mmap_pages_allocated);
1188
1189 point = len >> PAGE_SHIFT;
1190
1191 /* we allocated a power-of-2 sized page set, so we may want to trim off
1192 * the excess */
1193 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1194 while (total > point) {
1195 order = ilog2(total - point);
1196 n = 1 << order;
1197 kdebug("shave %lu/%lu @%lu", n, total - point, total);
1198 atomic_long_sub(n, &mmap_pages_allocated);
1199 total -= n;
1200 set_page_refcounted(pages + total);
1201 __free_pages(pages + total, order);
1202 }
1203 }
1204
1205 for (point = 1; point < total; point++)
1206 set_page_refcounted(&pages[point]);
1207
1208 base = page_address(pages);
1209 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1210 region->vm_start = (unsigned long) base;
1211 region->vm_end = region->vm_start + len;
1212 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1213
1214 vma->vm_start = region->vm_start;
1215 vma->vm_end = region->vm_start + len;
1216
1217 if (vma->vm_file) {
1218 /* read the contents of a file into the copy */
1219 mm_segment_t old_fs;
1220 loff_t fpos;
1221
1222 fpos = vma->vm_pgoff;
1223 fpos <<= PAGE_SHIFT;
1224
1225 old_fs = get_fs();
1226 set_fs(KERNEL_DS);
1227 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
1228 set_fs(old_fs);
1229
1230 if (ret < 0)
1231 goto error_free;
1232
1233 /* clear the last little bit */
1234 if (ret < len)
1235 memset(base + ret, 0, len - ret);
1236
1237 }
1238
1239 return 0;
1240
1241error_free:
1242 free_page_series(region->vm_start, region->vm_top);
1243 region->vm_start = vma->vm_start = 0;
1244 region->vm_end = vma->vm_end = 0;
1245 region->vm_top = 0;
1246 return ret;
1247
1248enomem:
1249 printk("Allocation of length %lu from process %d (%s) failed\n",
1250 len, current->pid, current->comm);
1251 show_free_areas(0);
1252 return -ENOMEM;
1253}
1254
1255/*
1256 * handle mapping creation for uClinux
1257 */
1258unsigned long do_mmap_pgoff(struct file *file,
1259 unsigned long addr,
1260 unsigned long len,
1261 unsigned long prot,
1262 unsigned long flags,
1263 unsigned long pgoff,
1264 unsigned long *populate)
1265{
1266 struct vm_area_struct *vma;
1267 struct vm_region *region;
1268 struct rb_node *rb;
1269 unsigned long capabilities, vm_flags, result;
1270 int ret;
1271
1272 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1273
1274 *populate = 0;
1275
1276 /* decide whether we should attempt the mapping, and if so what sort of
1277 * mapping */
1278 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1279 &capabilities);
1280 if (ret < 0) {
1281 kleave(" = %d [val]", ret);
1282 return ret;
1283 }
1284
1285 /* we ignore the address hint */
1286 addr = 0;
1287 len = PAGE_ALIGN(len);
1288
1289 /* we've determined that we can make the mapping, now translate what we
1290 * now know into VMA flags */
1291 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1292
1293 /* we're going to need to record the mapping */
1294 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1295 if (!region)
1296 goto error_getting_region;
1297
1298 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1299 if (!vma)
1300 goto error_getting_vma;
1301
1302 region->vm_usage = 1;
1303 region->vm_flags = vm_flags;
1304 region->vm_pgoff = pgoff;
1305
1306 INIT_LIST_HEAD(&vma->anon_vma_chain);
1307 vma->vm_flags = vm_flags;
1308 vma->vm_pgoff = pgoff;
1309
1310 if (file) {
1311 region->vm_file = get_file(file);
1312 vma->vm_file = get_file(file);
1313 }
1314
1315 down_write(&nommu_region_sem);
1316
1317 /* if we want to share, we need to check for regions created by other
1318 * mmap() calls that overlap with our proposed mapping
1319 * - we can only share with a superset match on most regular files
1320 * - shared mappings on character devices and memory backed files are
1321 * permitted to overlap inexactly as far as we are concerned for in
1322 * these cases, sharing is handled in the driver or filesystem rather
1323 * than here
1324 */
1325 if (vm_flags & VM_MAYSHARE) {
1326 struct vm_region *pregion;
1327 unsigned long pglen, rpglen, pgend, rpgend, start;
1328
1329 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1330 pgend = pgoff + pglen;
1331
1332 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1333 pregion = rb_entry(rb, struct vm_region, vm_rb);
1334
1335 if (!(pregion->vm_flags & VM_MAYSHARE))
1336 continue;
1337
1338 /* search for overlapping mappings on the same file */
1339 if (file_inode(pregion->vm_file) !=
1340 file_inode(file))
1341 continue;
1342
1343 if (pregion->vm_pgoff >= pgend)
1344 continue;
1345
1346 rpglen = pregion->vm_end - pregion->vm_start;
1347 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1348 rpgend = pregion->vm_pgoff + rpglen;
1349 if (pgoff >= rpgend)
1350 continue;
1351
1352 /* handle inexactly overlapping matches between
1353 * mappings */
1354 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1355 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1356 /* new mapping is not a subset of the region */
1357 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1358 goto sharing_violation;
1359 continue;
1360 }
1361
1362 /* we've found a region we can share */
1363 pregion->vm_usage++;
1364 vma->vm_region = pregion;
1365 start = pregion->vm_start;
1366 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1367 vma->vm_start = start;
1368 vma->vm_end = start + len;
1369
1370 if (pregion->vm_flags & VM_MAPPED_COPY) {
1371 kdebug("share copy");
1372 vma->vm_flags |= VM_MAPPED_COPY;
1373 } else {
1374 kdebug("share mmap");
1375 ret = do_mmap_shared_file(vma);
1376 if (ret < 0) {
1377 vma->vm_region = NULL;
1378 vma->vm_start = 0;
1379 vma->vm_end = 0;
1380 pregion->vm_usage--;
1381 pregion = NULL;
1382 goto error_just_free;
1383 }
1384 }
1385 fput(region->vm_file);
1386 kmem_cache_free(vm_region_jar, region);
1387 region = pregion;
1388 result = start;
1389 goto share;
1390 }
1391
1392 /* obtain the address at which to make a shared mapping
1393 * - this is the hook for quasi-memory character devices to
1394 * tell us the location of a shared mapping
1395 */
1396 if (capabilities & BDI_CAP_MAP_DIRECT) {
1397 addr = file->f_op->get_unmapped_area(file, addr, len,
1398 pgoff, flags);
1399 if (IS_ERR_VALUE(addr)) {
1400 ret = addr;
1401 if (ret != -ENOSYS)
1402 goto error_just_free;
1403
1404 /* the driver refused to tell us where to site
1405 * the mapping so we'll have to attempt to copy
1406 * it */
1407 ret = -ENODEV;
1408 if (!(capabilities & BDI_CAP_MAP_COPY))
1409 goto error_just_free;
1410
1411 capabilities &= ~BDI_CAP_MAP_DIRECT;
1412 } else {
1413 vma->vm_start = region->vm_start = addr;
1414 vma->vm_end = region->vm_end = addr + len;
1415 }
1416 }
1417 }
1418
1419 vma->vm_region = region;
1420
1421 /* set up the mapping
1422 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1423 */
1424 if (file && vma->vm_flags & VM_SHARED)
1425 ret = do_mmap_shared_file(vma);
1426 else
1427 ret = do_mmap_private(vma, region, len, capabilities);
1428 if (ret < 0)
1429 goto error_just_free;
1430 add_nommu_region(region);
1431
1432 /* clear anonymous mappings that don't ask for uninitialized data */
1433 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1434 memset((void *)region->vm_start, 0,
1435 region->vm_end - region->vm_start);
1436
1437 /* okay... we have a mapping; now we have to register it */
1438 result = vma->vm_start;
1439
1440 current->mm->total_vm += len >> PAGE_SHIFT;
1441
1442share:
1443 add_vma_to_mm(current->mm, vma);
1444
1445 /* we flush the region from the icache only when the first executable
1446 * mapping of it is made */
1447 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1448 flush_icache_range(region->vm_start, region->vm_end);
1449 region->vm_icache_flushed = true;
1450 }
1451
1452 up_write(&nommu_region_sem);
1453
1454 kleave(" = %lx", result);
1455 return result;
1456
1457error_just_free:
1458 up_write(&nommu_region_sem);
1459error:
1460 if (region->vm_file)
1461 fput(region->vm_file);
1462 kmem_cache_free(vm_region_jar, region);
1463 if (vma->vm_file)
1464 fput(vma->vm_file);
1465 kmem_cache_free(vm_area_cachep, vma);
1466 kleave(" = %d", ret);
1467 return ret;
1468
1469sharing_violation:
1470 up_write(&nommu_region_sem);
1471 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1472 ret = -EINVAL;
1473 goto error;
1474
1475error_getting_vma:
1476 kmem_cache_free(vm_region_jar, region);
1477 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1478 " from process %d failed\n",
1479 len, current->pid);
1480 show_free_areas(0);
1481 return -ENOMEM;
1482
1483error_getting_region:
1484 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1485 " from process %d failed\n",
1486 len, current->pid);
1487 show_free_areas(0);
1488 return -ENOMEM;
1489}
1490
1491SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1492 unsigned long, prot, unsigned long, flags,
1493 unsigned long, fd, unsigned long, pgoff)
1494{
1495 struct file *file = NULL;
1496 unsigned long retval = -EBADF;
1497
1498 audit_mmap_fd(fd, flags);
1499 if (!(flags & MAP_ANONYMOUS)) {
1500 file = fget(fd);
1501 if (!file)
1502 goto out;
1503 }
1504
1505 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1506
1507 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1508
1509 if (file)
1510 fput(file);
1511out:
1512 return retval;
1513}
1514
1515#ifdef __ARCH_WANT_SYS_OLD_MMAP
1516struct mmap_arg_struct {
1517 unsigned long addr;
1518 unsigned long len;
1519 unsigned long prot;
1520 unsigned long flags;
1521 unsigned long fd;
1522 unsigned long offset;
1523};
1524
1525SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1526{
1527 struct mmap_arg_struct a;
1528
1529 if (copy_from_user(&a, arg, sizeof(a)))
1530 return -EFAULT;
1531 if (a.offset & ~PAGE_MASK)
1532 return -EINVAL;
1533
1534 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1535 a.offset >> PAGE_SHIFT);
1536}
1537#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1538
1539/*
1540 * split a vma into two pieces at address 'addr', a new vma is allocated either
1541 * for the first part or the tail.
1542 */
1543int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1544 unsigned long addr, int new_below)
1545{
1546 struct vm_area_struct *new;
1547 struct vm_region *region;
1548 unsigned long npages;
1549
1550 kenter("");
1551
1552 /* we're only permitted to split anonymous regions (these should have
1553 * only a single usage on the region) */
1554 if (vma->vm_file)
1555 return -ENOMEM;
1556
1557 if (mm->map_count >= sysctl_max_map_count)
1558 return -ENOMEM;
1559
1560 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1561 if (!region)
1562 return -ENOMEM;
1563
1564 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1565 if (!new) {
1566 kmem_cache_free(vm_region_jar, region);
1567 return -ENOMEM;
1568 }
1569
1570 /* most fields are the same, copy all, and then fixup */
1571 *new = *vma;
1572 *region = *vma->vm_region;
1573 new->vm_region = region;
1574
1575 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1576
1577 if (new_below) {
1578 region->vm_top = region->vm_end = new->vm_end = addr;
1579 } else {
1580 region->vm_start = new->vm_start = addr;
1581 region->vm_pgoff = new->vm_pgoff += npages;
1582 }
1583
1584 if (new->vm_ops && new->vm_ops->open)
1585 new->vm_ops->open(new);
1586
1587 delete_vma_from_mm(vma);
1588 down_write(&nommu_region_sem);
1589 delete_nommu_region(vma->vm_region);
1590 if (new_below) {
1591 vma->vm_region->vm_start = vma->vm_start = addr;
1592 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1593 } else {
1594 vma->vm_region->vm_end = vma->vm_end = addr;
1595 vma->vm_region->vm_top = addr;
1596 }
1597 add_nommu_region(vma->vm_region);
1598 add_nommu_region(new->vm_region);
1599 up_write(&nommu_region_sem);
1600 add_vma_to_mm(mm, vma);
1601 add_vma_to_mm(mm, new);
1602 return 0;
1603}
1604
1605/*
1606 * shrink a VMA by removing the specified chunk from either the beginning or
1607 * the end
1608 */
1609static int shrink_vma(struct mm_struct *mm,
1610 struct vm_area_struct *vma,
1611 unsigned long from, unsigned long to)
1612{
1613 struct vm_region *region;
1614
1615 kenter("");
1616
1617 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1618 * and list */
1619 delete_vma_from_mm(vma);
1620 if (from > vma->vm_start)
1621 vma->vm_end = from;
1622 else
1623 vma->vm_start = to;
1624 add_vma_to_mm(mm, vma);
1625
1626 /* cut the backing region down to size */
1627 region = vma->vm_region;
1628 BUG_ON(region->vm_usage != 1);
1629
1630 down_write(&nommu_region_sem);
1631 delete_nommu_region(region);
1632 if (from > region->vm_start) {
1633 to = region->vm_top;
1634 region->vm_top = region->vm_end = from;
1635 } else {
1636 region->vm_start = to;
1637 }
1638 add_nommu_region(region);
1639 up_write(&nommu_region_sem);
1640
1641 free_page_series(from, to);
1642 return 0;
1643}
1644
1645/*
1646 * release a mapping
1647 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1648 * VMA, though it need not cover the whole VMA
1649 */
1650int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1651{
1652 struct vm_area_struct *vma;
1653 unsigned long end;
1654 int ret;
1655
1656 kenter(",%lx,%zx", start, len);
1657
1658 len = PAGE_ALIGN(len);
1659 if (len == 0)
1660 return -EINVAL;
1661
1662 end = start + len;
1663
1664 /* find the first potentially overlapping VMA */
1665 vma = find_vma(mm, start);
1666 if (!vma) {
1667 static int limit;
1668 if (limit < 5) {
1669 printk(KERN_WARNING
1670 "munmap of memory not mmapped by process %d"
1671 " (%s): 0x%lx-0x%lx\n",
1672 current->pid, current->comm,
1673 start, start + len - 1);
1674 limit++;
1675 }
1676 return -EINVAL;
1677 }
1678
1679 /* we're allowed to split an anonymous VMA but not a file-backed one */
1680 if (vma->vm_file) {
1681 do {
1682 if (start > vma->vm_start) {
1683 kleave(" = -EINVAL [miss]");
1684 return -EINVAL;
1685 }
1686 if (end == vma->vm_end)
1687 goto erase_whole_vma;
1688 vma = vma->vm_next;
1689 } while (vma);
1690 kleave(" = -EINVAL [split file]");
1691 return -EINVAL;
1692 } else {
1693 /* the chunk must be a subset of the VMA found */
1694 if (start == vma->vm_start && end == vma->vm_end)
1695 goto erase_whole_vma;
1696 if (start < vma->vm_start || end > vma->vm_end) {
1697 kleave(" = -EINVAL [superset]");
1698 return -EINVAL;
1699 }
1700 if (start & ~PAGE_MASK) {
1701 kleave(" = -EINVAL [unaligned start]");
1702 return -EINVAL;
1703 }
1704 if (end != vma->vm_end && end & ~PAGE_MASK) {
1705 kleave(" = -EINVAL [unaligned split]");
1706 return -EINVAL;
1707 }
1708 if (start != vma->vm_start && end != vma->vm_end) {
1709 ret = split_vma(mm, vma, start, 1);
1710 if (ret < 0) {
1711 kleave(" = %d [split]", ret);
1712 return ret;
1713 }
1714 }
1715 return shrink_vma(mm, vma, start, end);
1716 }
1717
1718erase_whole_vma:
1719 delete_vma_from_mm(vma);
1720 delete_vma(mm, vma);
1721 kleave(" = 0");
1722 return 0;
1723}
1724EXPORT_SYMBOL(do_munmap);
1725
1726int vm_munmap(unsigned long addr, size_t len)
1727{
1728 struct mm_struct *mm = current->mm;
1729 int ret;
1730
1731 down_write(&mm->mmap_sem);
1732 ret = do_munmap(mm, addr, len);
1733 up_write(&mm->mmap_sem);
1734 return ret;
1735}
1736EXPORT_SYMBOL(vm_munmap);
1737
1738SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1739{
1740 return vm_munmap(addr, len);
1741}
1742
1743/*
1744 * release all the mappings made in a process's VM space
1745 */
1746void exit_mmap(struct mm_struct *mm)
1747{
1748 struct vm_area_struct *vma;
1749
1750 if (!mm)
1751 return;
1752
1753 kenter("");
1754
1755 mm->total_vm = 0;
1756
1757 while ((vma = mm->mmap)) {
1758 mm->mmap = vma->vm_next;
1759 delete_vma_from_mm(vma);
1760 delete_vma(mm, vma);
1761 cond_resched();
1762 }
1763
1764 kleave("");
1765}
1766
1767unsigned long vm_brk(unsigned long addr, unsigned long len)
1768{
1769 return -ENOMEM;
1770}
1771
1772/*
1773 * expand (or shrink) an existing mapping, potentially moving it at the same
1774 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1775 *
1776 * under NOMMU conditions, we only permit changing a mapping's size, and only
1777 * as long as it stays within the region allocated by do_mmap_private() and the
1778 * block is not shareable
1779 *
1780 * MREMAP_FIXED is not supported under NOMMU conditions
1781 */
1782static unsigned long do_mremap(unsigned long addr,
1783 unsigned long old_len, unsigned long new_len,
1784 unsigned long flags, unsigned long new_addr)
1785{
1786 struct vm_area_struct *vma;
1787
1788 /* insanity checks first */
1789 old_len = PAGE_ALIGN(old_len);
1790 new_len = PAGE_ALIGN(new_len);
1791 if (old_len == 0 || new_len == 0)
1792 return (unsigned long) -EINVAL;
1793
1794 if (addr & ~PAGE_MASK)
1795 return -EINVAL;
1796
1797 if (flags & MREMAP_FIXED && new_addr != addr)
1798 return (unsigned long) -EINVAL;
1799
1800 vma = find_vma_exact(current->mm, addr, old_len);
1801 if (!vma)
1802 return (unsigned long) -EINVAL;
1803
1804 if (vma->vm_end != vma->vm_start + old_len)
1805 return (unsigned long) -EFAULT;
1806
1807 if (vma->vm_flags & VM_MAYSHARE)
1808 return (unsigned long) -EPERM;
1809
1810 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1811 return (unsigned long) -ENOMEM;
1812
1813 /* all checks complete - do it */
1814 vma->vm_end = vma->vm_start + new_len;
1815 return vma->vm_start;
1816}
1817
1818SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1819 unsigned long, new_len, unsigned long, flags,
1820 unsigned long, new_addr)
1821{
1822 unsigned long ret;
1823
1824 down_write(¤t->mm->mmap_sem);
1825 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1826 up_write(¤t->mm->mmap_sem);
1827 return ret;
1828}
1829
1830struct page *follow_page_mask(struct vm_area_struct *vma,
1831 unsigned long address, unsigned int flags,
1832 unsigned int *page_mask)
1833{
1834 *page_mask = 0;
1835 return NULL;
1836}
1837
1838int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1839 unsigned long pfn, unsigned long size, pgprot_t prot)
1840{
1841 if (addr != (pfn << PAGE_SHIFT))
1842 return -EINVAL;
1843
1844 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1845 return 0;
1846}
1847EXPORT_SYMBOL(remap_pfn_range);
1848
1849int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1850{
1851 unsigned long pfn = start >> PAGE_SHIFT;
1852 unsigned long vm_len = vma->vm_end - vma->vm_start;
1853
1854 pfn += vma->vm_pgoff;
1855 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1856}
1857EXPORT_SYMBOL(vm_iomap_memory);
1858
1859int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1860 unsigned long pgoff)
1861{
1862 unsigned int size = vma->vm_end - vma->vm_start;
1863
1864 if (!(vma->vm_flags & VM_USERMAP))
1865 return -EINVAL;
1866
1867 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1868 vma->vm_end = vma->vm_start + size;
1869
1870 return 0;
1871}
1872EXPORT_SYMBOL(remap_vmalloc_range);
1873
1874unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1875 unsigned long len, unsigned long pgoff, unsigned long flags)
1876{
1877 return -ENOMEM;
1878}
1879
1880void unmap_mapping_range(struct address_space *mapping,
1881 loff_t const holebegin, loff_t const holelen,
1882 int even_cows)
1883{
1884}
1885EXPORT_SYMBOL(unmap_mapping_range);
1886
1887/*
1888 * Check that a process has enough memory to allocate a new virtual
1889 * mapping. 0 means there is enough memory for the allocation to
1890 * succeed and -ENOMEM implies there is not.
1891 *
1892 * We currently support three overcommit policies, which are set via the
1893 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1894 *
1895 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1896 * Additional code 2002 Jul 20 by Robert Love.
1897 *
1898 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1899 *
1900 * Note this is a helper function intended to be used by LSMs which
1901 * wish to use this logic.
1902 */
1903int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1904{
1905 unsigned long free, allowed, reserve;
1906
1907 vm_acct_memory(pages);
1908
1909 /*
1910 * Sometimes we want to use more memory than we have
1911 */
1912 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1913 return 0;
1914
1915 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1916 free = global_page_state(NR_FREE_PAGES);
1917 free += global_page_state(NR_FILE_PAGES);
1918
1919 /*
1920 * shmem pages shouldn't be counted as free in this
1921 * case, they can't be purged, only swapped out, and
1922 * that won't affect the overall amount of available
1923 * memory in the system.
1924 */
1925 free -= global_page_state(NR_SHMEM);
1926
1927 free += get_nr_swap_pages();
1928
1929 /*
1930 * Any slabs which are created with the
1931 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1932 * which are reclaimable, under pressure. The dentry
1933 * cache and most inode caches should fall into this
1934 */
1935 free += global_page_state(NR_SLAB_RECLAIMABLE);
1936
1937 /*
1938 * Leave reserved pages. The pages are not for anonymous pages.
1939 */
1940 if (free <= totalreserve_pages)
1941 goto error;
1942 else
1943 free -= totalreserve_pages;
1944
1945 /*
1946 * Reserve some for root
1947 */
1948 if (!cap_sys_admin)
1949 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1950
1951 if (free > pages)
1952 return 0;
1953
1954 goto error;
1955 }
1956
1957 allowed = vm_commit_limit();
1958 /*
1959 * Reserve some 3% for root
1960 */
1961 if (!cap_sys_admin)
1962 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1963
1964 /*
1965 * Don't let a single process grow so big a user can't recover
1966 */
1967 if (mm) {
1968 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1969 allowed -= min(mm->total_vm / 32, reserve);
1970 }
1971
1972 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1973 return 0;
1974
1975error:
1976 vm_unacct_memory(pages);
1977
1978 return -ENOMEM;
1979}
1980
1981int in_gate_area_no_mm(unsigned long addr)
1982{
1983 return 0;
1984}
1985
1986int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1987{
1988 BUG();
1989 return 0;
1990}
1991EXPORT_SYMBOL(filemap_fault);
1992
1993void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1994{
1995 BUG();
1996}
1997EXPORT_SYMBOL(filemap_map_pages);
1998
1999int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr,
2000 unsigned long size, pgoff_t pgoff)
2001{
2002 BUG();
2003 return 0;
2004}
2005EXPORT_SYMBOL(generic_file_remap_pages);
2006
2007static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
2008 unsigned long addr, void *buf, int len, int write)
2009{
2010 struct vm_area_struct *vma;
2011
2012 down_read(&mm->mmap_sem);
2013
2014 /* the access must start within one of the target process's mappings */
2015 vma = find_vma(mm, addr);
2016 if (vma) {
2017 /* don't overrun this mapping */
2018 if (addr + len >= vma->vm_end)
2019 len = vma->vm_end - addr;
2020
2021 /* only read or write mappings where it is permitted */
2022 if (write && vma->vm_flags & VM_MAYWRITE)
2023 copy_to_user_page(vma, NULL, addr,
2024 (void *) addr, buf, len);
2025 else if (!write && vma->vm_flags & VM_MAYREAD)
2026 copy_from_user_page(vma, NULL, addr,
2027 buf, (void *) addr, len);
2028 else
2029 len = 0;
2030 } else {
2031 len = 0;
2032 }
2033
2034 up_read(&mm->mmap_sem);
2035
2036 return len;
2037}
2038
2039/**
2040 * @access_remote_vm - access another process' address space
2041 * @mm: the mm_struct of the target address space
2042 * @addr: start address to access
2043 * @buf: source or destination buffer
2044 * @len: number of bytes to transfer
2045 * @write: whether the access is a write
2046 *
2047 * The caller must hold a reference on @mm.
2048 */
2049int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2050 void *buf, int len, int write)
2051{
2052 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2053}
2054
2055/*
2056 * Access another process' address space.
2057 * - source/target buffer must be kernel space
2058 */
2059int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2060{
2061 struct mm_struct *mm;
2062
2063 if (addr + len < addr)
2064 return 0;
2065
2066 mm = get_task_mm(tsk);
2067 if (!mm)
2068 return 0;
2069
2070 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2071
2072 mmput(mm);
2073 return len;
2074}
2075
2076/**
2077 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2078 * @inode: The inode to check
2079 * @size: The current filesize of the inode
2080 * @newsize: The proposed filesize of the inode
2081 *
2082 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2083 * make sure that that any outstanding VMAs aren't broken and then shrink the
2084 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2085 * automatically grant mappings that are too large.
2086 */
2087int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2088 size_t newsize)
2089{
2090 struct vm_area_struct *vma;
2091 struct vm_region *region;
2092 pgoff_t low, high;
2093 size_t r_size, r_top;
2094
2095 low = newsize >> PAGE_SHIFT;
2096 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2097
2098 down_write(&nommu_region_sem);
2099 mutex_lock(&inode->i_mapping->i_mmap_mutex);
2100
2101 /* search for VMAs that fall within the dead zone */
2102 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
2103 /* found one - only interested if it's shared out of the page
2104 * cache */
2105 if (vma->vm_flags & VM_SHARED) {
2106 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2107 up_write(&nommu_region_sem);
2108 return -ETXTBSY; /* not quite true, but near enough */
2109 }
2110 }
2111
2112 /* reduce any regions that overlap the dead zone - if in existence,
2113 * these will be pointed to by VMAs that don't overlap the dead zone
2114 *
2115 * we don't check for any regions that start beyond the EOF as there
2116 * shouldn't be any
2117 */
2118 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap,
2119 0, ULONG_MAX) {
2120 if (!(vma->vm_flags & VM_SHARED))
2121 continue;
2122
2123 region = vma->vm_region;
2124 r_size = region->vm_top - region->vm_start;
2125 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2126
2127 if (r_top > newsize) {
2128 region->vm_top -= r_top - newsize;
2129 if (region->vm_end > region->vm_top)
2130 region->vm_end = region->vm_top;
2131 }
2132 }
2133
2134 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2135 up_write(&nommu_region_sem);
2136 return 0;
2137}
2138
2139/*
2140 * Initialise sysctl_user_reserve_kbytes.
2141 *
2142 * This is intended to prevent a user from starting a single memory hogging
2143 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2144 * mode.
2145 *
2146 * The default value is min(3% of free memory, 128MB)
2147 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2148 */
2149static int __meminit init_user_reserve(void)
2150{
2151 unsigned long free_kbytes;
2152
2153 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2154
2155 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2156 return 0;
2157}
2158module_init(init_user_reserve)
2159
2160/*
2161 * Initialise sysctl_admin_reserve_kbytes.
2162 *
2163 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2164 * to log in and kill a memory hogging process.
2165 *
2166 * Systems with more than 256MB will reserve 8MB, enough to recover
2167 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2168 * only reserve 3% of free pages by default.
2169 */
2170static int __meminit init_admin_reserve(void)
2171{
2172 unsigned long free_kbytes;
2173
2174 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2175
2176 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2177 return 0;
2178}
2179module_init(init_admin_reserve)