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