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