1 // SPDX-License-Identifier: GPL-2.0-only
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).
8 * See Documentation/admin-guide/mm/nommu-mmap.rst
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>
17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 #include <linux/export.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>
38 #include <linux/uaccess.h>
39 #include <linux/uio.h>
41 #include <asm/tlbflush.h>
42 #include <asm/mmu_context.h>
46 EXPORT_SYMBOL(high_memory);
48 unsigned long max_mapnr;
49 EXPORT_SYMBOL(max_mapnr);
50 unsigned long highest_memmap_pfn;
51 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
52 int heap_stack_gap = 0;
54 atomic_long_t mmap_pages_allocated;
56 EXPORT_SYMBOL(mem_map);
58 /* list of mapped, potentially shareable regions */
59 static struct kmem_cache *vm_region_jar;
60 struct rb_root nommu_region_tree = RB_ROOT;
61 DECLARE_RWSEM(nommu_region_sem);
63 const struct vm_operations_struct generic_file_vm_ops = {
67 * Return the total memory allocated for this pointer, not
68 * just what the caller asked for.
70 * Doesn't have to be accurate, i.e. may have races.
72 unsigned int kobjsize(const void *objp)
77 * If the object we have should not have ksize performed on it,
80 if (!objp || !virt_addr_valid(objp))
83 page = virt_to_head_page(objp);
86 * If the allocator sets PageSlab, we know the pointer came from
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.
98 if (!PageCompound(page)) {
99 struct vm_area_struct *vma;
101 vma = find_vma(current->mm, (unsigned long)objp);
103 return vma->vm_end - vma->vm_start;
107 * The ksize() function is only guaranteed to work for pointers
108 * returned by kmalloc(). So handle arbitrary pointers here.
110 return page_size(page);
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
119 * Only IO mappings and raw PFN mappings are allowed.
121 * Returns zero and the pfn at @pfn on success, -ve otherwise.
123 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
126 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
129 *pfn = address >> PAGE_SHIFT;
132 EXPORT_SYMBOL(follow_pfn);
134 void vfree(const void *addr)
138 EXPORT_SYMBOL(vfree);
140 void *__vmalloc(unsigned long size, gfp_t gfp_mask)
143 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
144 * returns only a logical address.
146 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
148 EXPORT_SYMBOL(__vmalloc);
150 void *__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,
155 return __vmalloc(size, gfp_mask);
158 void *__vmalloc_node(unsigned long size, unsigned long align, gfp_t gfp_mask,
159 int node, const void *caller)
161 return __vmalloc(size, gfp_mask);
164 static void *__vmalloc_user_flags(unsigned long size, gfp_t flags)
168 ret = __vmalloc(size, flags);
170 struct vm_area_struct *vma;
172 mmap_write_lock(current->mm);
173 vma = find_vma(current->mm, (unsigned long)ret);
175 vm_flags_set(vma, VM_USERMAP);
176 mmap_write_unlock(current->mm);
182 void *vmalloc_user(unsigned long size)
184 return __vmalloc_user_flags(size, GFP_KERNEL | __GFP_ZERO);
186 EXPORT_SYMBOL(vmalloc_user);
188 struct page *vmalloc_to_page(const void *addr)
190 return virt_to_page(addr);
192 EXPORT_SYMBOL(vmalloc_to_page);
194 unsigned long vmalloc_to_pfn(const void *addr)
196 return page_to_pfn(virt_to_page(addr));
198 EXPORT_SYMBOL(vmalloc_to_pfn);
200 long vread_iter(struct iov_iter *iter, const char *addr, size_t count)
202 /* Don't allow overflow */
203 if ((unsigned long) addr + count < count)
204 count = -(unsigned long) addr;
206 return copy_to_iter(addr, count, iter);
210 * vmalloc - allocate virtually contiguous memory
212 * @size: allocation size
214 * Allocate enough pages to cover @size from the page level
215 * allocator and map them into contiguous kernel virtual space.
217 * For tight control over page level allocator and protection flags
218 * use __vmalloc() instead.
220 void *vmalloc(unsigned long size)
222 return __vmalloc(size, GFP_KERNEL);
224 EXPORT_SYMBOL(vmalloc);
226 void *vmalloc_huge(unsigned long size, gfp_t gfp_mask) __weak __alias(__vmalloc);
229 * vzalloc - allocate virtually contiguous memory with zero fill
231 * @size: allocation size
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.
237 * For tight control over page level allocator and protection flags
238 * use __vmalloc() instead.
240 void *vzalloc(unsigned long size)
242 return __vmalloc(size, GFP_KERNEL | __GFP_ZERO);
244 EXPORT_SYMBOL(vzalloc);
247 * vmalloc_node - allocate memory on a specific node
248 * @size: allocation size
251 * Allocate enough pages to cover @size from the page level
252 * allocator and map them into contiguous kernel virtual space.
254 * For tight control over page level allocator and protection flags
255 * use __vmalloc() instead.
257 void *vmalloc_node(unsigned long size, int node)
259 return vmalloc(size);
261 EXPORT_SYMBOL(vmalloc_node);
264 * vzalloc_node - allocate memory on a specific node with zero fill
265 * @size: allocation size
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.
272 * For tight control over page level allocator and protection flags
273 * use __vmalloc() instead.
275 void *vzalloc_node(unsigned long size, int node)
277 return vzalloc(size);
279 EXPORT_SYMBOL(vzalloc_node);
282 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
283 * @size: allocation size
285 * Allocate enough 32bit PA addressable pages to cover @size from the
286 * page level allocator and map them into contiguous kernel virtual space.
288 void *vmalloc_32(unsigned long size)
290 return __vmalloc(size, GFP_KERNEL);
292 EXPORT_SYMBOL(vmalloc_32);
295 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
296 * @size: allocation size
298 * The resulting memory area is 32bit addressable and zeroed so it can be
299 * mapped to userspace without leaking data.
301 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
302 * remap_vmalloc_range() are permissible.
304 void *vmalloc_32_user(unsigned long size)
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.
310 return vmalloc_user(size);
312 EXPORT_SYMBOL(vmalloc_32_user);
314 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
321 void vunmap(const void *addr)
325 EXPORT_SYMBOL(vunmap);
327 void *vm_map_ram(struct page **pages, unsigned int count, int node)
332 EXPORT_SYMBOL(vm_map_ram);
334 void vm_unmap_ram(const void *mem, unsigned int count)
338 EXPORT_SYMBOL(vm_unmap_ram);
340 void vm_unmap_aliases(void)
343 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
345 void free_vm_area(struct vm_struct *area)
349 EXPORT_SYMBOL_GPL(free_vm_area);
351 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
356 EXPORT_SYMBOL(vm_insert_page);
358 int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr,
359 struct page **pages, unsigned long *num)
363 EXPORT_SYMBOL(vm_insert_pages);
365 int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
370 EXPORT_SYMBOL(vm_map_pages);
372 int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
377 EXPORT_SYMBOL(vm_map_pages_zero);
380 * sys_brk() for the most part doesn't need the global kernel
381 * lock, except when an application is doing something nasty
382 * like trying to un-brk an area that has already been mapped
383 * to a regular file. in this case, the unmapping will need
384 * to invoke file system routines that need the global lock.
386 SYSCALL_DEFINE1(brk, unsigned long, brk)
388 struct mm_struct *mm = current->mm;
390 if (brk < mm->start_brk || brk > mm->context.end_brk)
397 * Always allow shrinking brk
399 if (brk <= mm->brk) {
405 * Ok, looks good - let it rip.
407 flush_icache_user_range(mm->brk, brk);
408 return mm->brk = brk;
412 * initialise the percpu counter for VM and region record slabs
414 void __init mmap_init(void)
418 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
420 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
424 * validate the region tree
425 * - the caller must hold the region lock
427 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
428 static noinline void validate_nommu_regions(void)
430 struct vm_region *region, *last;
431 struct rb_node *p, *lastp;
433 lastp = rb_first(&nommu_region_tree);
437 last = rb_entry(lastp, struct vm_region, vm_rb);
438 BUG_ON(last->vm_end <= last->vm_start);
439 BUG_ON(last->vm_top < last->vm_end);
441 while ((p = rb_next(lastp))) {
442 region = rb_entry(p, struct vm_region, vm_rb);
443 last = rb_entry(lastp, struct vm_region, vm_rb);
445 BUG_ON(region->vm_end <= region->vm_start);
446 BUG_ON(region->vm_top < region->vm_end);
447 BUG_ON(region->vm_start < last->vm_top);
453 static void validate_nommu_regions(void)
459 * add a region into the global tree
461 static void add_nommu_region(struct vm_region *region)
463 struct vm_region *pregion;
464 struct rb_node **p, *parent;
466 validate_nommu_regions();
469 p = &nommu_region_tree.rb_node;
472 pregion = rb_entry(parent, struct vm_region, vm_rb);
473 if (region->vm_start < pregion->vm_start)
475 else if (region->vm_start > pregion->vm_start)
477 else if (pregion == region)
483 rb_link_node(®ion->vm_rb, parent, p);
484 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
486 validate_nommu_regions();
490 * delete a region from the global tree
492 static void delete_nommu_region(struct vm_region *region)
494 BUG_ON(!nommu_region_tree.rb_node);
496 validate_nommu_regions();
497 rb_erase(®ion->vm_rb, &nommu_region_tree);
498 validate_nommu_regions();
502 * free a contiguous series of pages
504 static void free_page_series(unsigned long from, unsigned long to)
506 for (; from < to; from += PAGE_SIZE) {
507 struct page *page = virt_to_page((void *)from);
509 atomic_long_dec(&mmap_pages_allocated);
515 * release a reference to a region
516 * - the caller must hold the region semaphore for writing, which this releases
517 * - the region may not have been added to the tree yet, in which case vm_top
518 * will equal vm_start
520 static void __put_nommu_region(struct vm_region *region)
521 __releases(nommu_region_sem)
523 BUG_ON(!nommu_region_tree.rb_node);
525 if (--region->vm_usage == 0) {
526 if (region->vm_top > region->vm_start)
527 delete_nommu_region(region);
528 up_write(&nommu_region_sem);
531 fput(region->vm_file);
533 /* IO memory and memory shared directly out of the pagecache
534 * from ramfs/tmpfs mustn't be released here */
535 if (region->vm_flags & VM_MAPPED_COPY)
536 free_page_series(region->vm_start, region->vm_top);
537 kmem_cache_free(vm_region_jar, region);
539 up_write(&nommu_region_sem);
544 * release a reference to a region
546 static void put_nommu_region(struct vm_region *region)
548 down_write(&nommu_region_sem);
549 __put_nommu_region(region);
552 static void setup_vma_to_mm(struct vm_area_struct *vma, struct mm_struct *mm)
556 /* add the VMA to the mapping */
558 struct address_space *mapping = vma->vm_file->f_mapping;
560 i_mmap_lock_write(mapping);
561 flush_dcache_mmap_lock(mapping);
562 vma_interval_tree_insert(vma, &mapping->i_mmap);
563 flush_dcache_mmap_unlock(mapping);
564 i_mmap_unlock_write(mapping);
568 static void cleanup_vma_from_mm(struct vm_area_struct *vma)
570 vma->vm_mm->map_count--;
571 /* remove the VMA from the mapping */
573 struct address_space *mapping;
574 mapping = vma->vm_file->f_mapping;
576 i_mmap_lock_write(mapping);
577 flush_dcache_mmap_lock(mapping);
578 vma_interval_tree_remove(vma, &mapping->i_mmap);
579 flush_dcache_mmap_unlock(mapping);
580 i_mmap_unlock_write(mapping);
585 * delete a VMA from its owning mm_struct and address space
587 static int delete_vma_from_mm(struct vm_area_struct *vma)
589 VMA_ITERATOR(vmi, vma->vm_mm, vma->vm_start);
591 vma_iter_config(&vmi, vma->vm_start, vma->vm_end);
592 if (vma_iter_prealloc(&vmi, vma)) {
593 pr_warn("Allocation of vma tree for process %d failed\n",
597 cleanup_vma_from_mm(vma);
599 /* remove from the MM's tree and list */
600 vma_iter_clear(&vmi);
604 * destroy a VMA record
606 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
608 if (vma->vm_ops && vma->vm_ops->close)
609 vma->vm_ops->close(vma);
612 put_nommu_region(vma->vm_region);
616 struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
617 unsigned long start_addr,
618 unsigned long end_addr)
620 unsigned long index = start_addr;
622 mmap_assert_locked(mm);
623 return mt_find(&mm->mm_mt, &index, end_addr - 1);
625 EXPORT_SYMBOL(find_vma_intersection);
628 * look up the first VMA in which addr resides, NULL if none
629 * - should be called with mm->mmap_lock at least held readlocked
631 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
633 VMA_ITERATOR(vmi, mm, addr);
635 return vma_iter_load(&vmi);
637 EXPORT_SYMBOL(find_vma);
640 * At least xtensa ends up having protection faults even with no
641 * MMU.. No stack expansion, at least.
643 struct vm_area_struct *lock_mm_and_find_vma(struct mm_struct *mm,
644 unsigned long addr, struct pt_regs *regs)
646 struct vm_area_struct *vma;
649 vma = vma_lookup(mm, addr);
651 mmap_read_unlock(mm);
656 * expand a stack to a given address
657 * - not supported under NOMMU conditions
659 int expand_stack_locked(struct vm_area_struct *vma, unsigned long addr)
664 struct vm_area_struct *expand_stack(struct mm_struct *mm, unsigned long addr)
666 mmap_read_unlock(mm);
671 * look up the first VMA exactly that exactly matches addr
672 * - should be called with mm->mmap_lock at least held readlocked
674 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
678 struct vm_area_struct *vma;
679 unsigned long end = addr + len;
680 VMA_ITERATOR(vmi, mm, addr);
682 vma = vma_iter_load(&vmi);
685 if (vma->vm_start != addr)
687 if (vma->vm_end != end)
694 * determine whether a mapping should be permitted and, if so, what sort of
695 * mapping we're capable of supporting
697 static int validate_mmap_request(struct file *file,
703 unsigned long *_capabilities)
705 unsigned long capabilities, rlen;
708 /* do the simple checks first */
709 if (flags & MAP_FIXED)
712 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
713 (flags & MAP_TYPE) != MAP_SHARED)
719 /* Careful about overflows.. */
720 rlen = PAGE_ALIGN(len);
721 if (!rlen || rlen > TASK_SIZE)
724 /* offset overflow? */
725 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
729 /* files must support mmap */
730 if (!file->f_op->mmap)
733 /* work out if what we've got could possibly be shared
734 * - we support chardevs that provide their own "memory"
735 * - we support files/blockdevs that are memory backed
737 if (file->f_op->mmap_capabilities) {
738 capabilities = file->f_op->mmap_capabilities(file);
740 /* no explicit capabilities set, so assume some
742 switch (file_inode(file)->i_mode & S_IFMT) {
745 capabilities = NOMMU_MAP_COPY;
760 /* eliminate any capabilities that we can't support on this
762 if (!file->f_op->get_unmapped_area)
763 capabilities &= ~NOMMU_MAP_DIRECT;
764 if (!(file->f_mode & FMODE_CAN_READ))
765 capabilities &= ~NOMMU_MAP_COPY;
767 /* The file shall have been opened with read permission. */
768 if (!(file->f_mode & FMODE_READ))
771 if (flags & MAP_SHARED) {
772 /* do checks for writing, appending and locking */
773 if ((prot & PROT_WRITE) &&
774 !(file->f_mode & FMODE_WRITE))
777 if (IS_APPEND(file_inode(file)) &&
778 (file->f_mode & FMODE_WRITE))
781 if (!(capabilities & NOMMU_MAP_DIRECT))
784 /* we mustn't privatise shared mappings */
785 capabilities &= ~NOMMU_MAP_COPY;
787 /* we're going to read the file into private memory we
789 if (!(capabilities & NOMMU_MAP_COPY))
792 /* we don't permit a private writable mapping to be
793 * shared with the backing device */
794 if (prot & PROT_WRITE)
795 capabilities &= ~NOMMU_MAP_DIRECT;
798 if (capabilities & NOMMU_MAP_DIRECT) {
799 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
800 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
801 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
803 capabilities &= ~NOMMU_MAP_DIRECT;
804 if (flags & MAP_SHARED) {
805 pr_warn("MAP_SHARED not completely supported on !MMU\n");
811 /* handle executable mappings and implied executable
813 if (path_noexec(&file->f_path)) {
814 if (prot & PROT_EXEC)
816 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
817 /* handle implication of PROT_EXEC by PROT_READ */
818 if (current->personality & READ_IMPLIES_EXEC) {
819 if (capabilities & NOMMU_MAP_EXEC)
822 } else if ((prot & PROT_READ) &&
823 (prot & PROT_EXEC) &&
824 !(capabilities & NOMMU_MAP_EXEC)
826 /* backing file is not executable, try to copy */
827 capabilities &= ~NOMMU_MAP_DIRECT;
830 /* anonymous mappings are always memory backed and can be
833 capabilities = NOMMU_MAP_COPY;
835 /* handle PROT_EXEC implication by PROT_READ */
836 if ((prot & PROT_READ) &&
837 (current->personality & READ_IMPLIES_EXEC))
841 /* allow the security API to have its say */
842 ret = security_mmap_addr(addr);
847 *_capabilities = capabilities;
852 * we've determined that we can make the mapping, now translate what we
853 * now know into VMA flags
855 static unsigned long determine_vm_flags(struct file *file,
858 unsigned long capabilities)
860 unsigned long vm_flags;
862 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
866 * MAP_ANONYMOUS. MAP_SHARED is mapped to MAP_PRIVATE, because
867 * there is no fork().
869 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
870 } else if (flags & MAP_PRIVATE) {
871 /* MAP_PRIVATE file mapping */
872 if (capabilities & NOMMU_MAP_DIRECT)
873 vm_flags |= (capabilities & NOMMU_VMFLAGS);
875 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
877 if (!(prot & PROT_WRITE) && !current->ptrace)
879 * R/O private file mapping which cannot be used to
880 * modify memory, especially also not via active ptrace
881 * (e.g., set breakpoints) or later by upgrading
882 * permissions (no mprotect()). We can try overlaying
883 * the file mapping, which will work e.g., on chardevs,
884 * ramfs/tmpfs/shmfs and romfs/cramf.
886 vm_flags |= VM_MAYOVERLAY;
888 /* MAP_SHARED file mapping: NOMMU_MAP_DIRECT is set. */
889 vm_flags |= VM_SHARED | VM_MAYSHARE |
890 (capabilities & NOMMU_VMFLAGS);
897 * set up a shared mapping on a file (the driver or filesystem provides and
900 static int do_mmap_shared_file(struct vm_area_struct *vma)
904 ret = call_mmap(vma->vm_file, vma);
906 vma->vm_region->vm_top = vma->vm_region->vm_end;
912 /* getting -ENOSYS indicates that direct mmap isn't possible (as
913 * opposed to tried but failed) so we can only give a suitable error as
914 * it's not possible to make a private copy if MAP_SHARED was given */
919 * set up a private mapping or an anonymous shared mapping
921 static int do_mmap_private(struct vm_area_struct *vma,
922 struct vm_region *region,
924 unsigned long capabilities)
926 unsigned long total, point;
931 * Invoke the file's mapping function so that it can keep track of
932 * shared mappings on devices or memory. VM_MAYOVERLAY will be set if
933 * it may attempt to share, which will make is_nommu_shared_mapping()
936 if (capabilities & NOMMU_MAP_DIRECT) {
937 ret = call_mmap(vma->vm_file, vma);
938 /* shouldn't return success if we're not sharing */
939 if (WARN_ON_ONCE(!is_nommu_shared_mapping(vma->vm_flags)))
942 vma->vm_region->vm_top = vma->vm_region->vm_end;
948 /* getting an ENOSYS error indicates that direct mmap isn't
949 * possible (as opposed to tried but failed) so we'll try to
950 * make a private copy of the data and map that instead */
954 /* allocate some memory to hold the mapping
955 * - note that this may not return a page-aligned address if the object
956 * we're allocating is smaller than a page
958 order = get_order(len);
960 point = len >> PAGE_SHIFT;
962 /* we don't want to allocate a power-of-2 sized page set */
963 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
966 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
970 atomic_long_add(total, &mmap_pages_allocated);
972 vm_flags_set(vma, VM_MAPPED_COPY);
973 region->vm_flags = vma->vm_flags;
974 region->vm_start = (unsigned long) base;
975 region->vm_end = region->vm_start + len;
976 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
978 vma->vm_start = region->vm_start;
979 vma->vm_end = region->vm_start + len;
982 /* read the contents of a file into the copy */
985 fpos = vma->vm_pgoff;
988 ret = kernel_read(vma->vm_file, base, len, &fpos);
992 /* clear the last little bit */
994 memset(base + ret, 0, len - ret);
997 vma_set_anonymous(vma);
1003 free_page_series(region->vm_start, region->vm_top);
1004 region->vm_start = vma->vm_start = 0;
1005 region->vm_end = vma->vm_end = 0;
1010 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1011 len, current->pid, current->comm);
1017 * handle mapping creation for uClinux
1019 unsigned long do_mmap(struct file *file,
1023 unsigned long flags,
1024 vm_flags_t vm_flags,
1025 unsigned long pgoff,
1026 unsigned long *populate,
1027 struct list_head *uf)
1029 struct vm_area_struct *vma;
1030 struct vm_region *region;
1032 unsigned long capabilities, result;
1034 VMA_ITERATOR(vmi, current->mm, 0);
1038 /* decide whether we should attempt the mapping, and if so what sort of
1040 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1045 /* we ignore the address hint */
1047 len = PAGE_ALIGN(len);
1049 /* we've determined that we can make the mapping, now translate what we
1050 * now know into VMA flags */
1051 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1054 /* we're going to need to record the mapping */
1055 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1057 goto error_getting_region;
1059 vma = vm_area_alloc(current->mm);
1061 goto error_getting_vma;
1063 region->vm_usage = 1;
1064 region->vm_flags = vm_flags;
1065 region->vm_pgoff = pgoff;
1067 vm_flags_init(vma, vm_flags);
1068 vma->vm_pgoff = pgoff;
1071 region->vm_file = get_file(file);
1072 vma->vm_file = get_file(file);
1075 down_write(&nommu_region_sem);
1077 /* if we want to share, we need to check for regions created by other
1078 * mmap() calls that overlap with our proposed mapping
1079 * - we can only share with a superset match on most regular files
1080 * - shared mappings on character devices and memory backed files are
1081 * permitted to overlap inexactly as far as we are concerned for in
1082 * these cases, sharing is handled in the driver or filesystem rather
1085 if (is_nommu_shared_mapping(vm_flags)) {
1086 struct vm_region *pregion;
1087 unsigned long pglen, rpglen, pgend, rpgend, start;
1089 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1090 pgend = pgoff + pglen;
1092 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1093 pregion = rb_entry(rb, struct vm_region, vm_rb);
1095 if (!is_nommu_shared_mapping(pregion->vm_flags))
1098 /* search for overlapping mappings on the same file */
1099 if (file_inode(pregion->vm_file) !=
1103 if (pregion->vm_pgoff >= pgend)
1106 rpglen = pregion->vm_end - pregion->vm_start;
1107 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1108 rpgend = pregion->vm_pgoff + rpglen;
1109 if (pgoff >= rpgend)
1112 /* handle inexactly overlapping matches between
1114 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1115 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1116 /* new mapping is not a subset of the region */
1117 if (!(capabilities & NOMMU_MAP_DIRECT))
1118 goto sharing_violation;
1122 /* we've found a region we can share */
1123 pregion->vm_usage++;
1124 vma->vm_region = pregion;
1125 start = pregion->vm_start;
1126 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1127 vma->vm_start = start;
1128 vma->vm_end = start + len;
1130 if (pregion->vm_flags & VM_MAPPED_COPY)
1131 vm_flags_set(vma, VM_MAPPED_COPY);
1133 ret = do_mmap_shared_file(vma);
1135 vma->vm_region = NULL;
1138 pregion->vm_usage--;
1140 goto error_just_free;
1143 fput(region->vm_file);
1144 kmem_cache_free(vm_region_jar, region);
1150 /* obtain the address at which to make a shared mapping
1151 * - this is the hook for quasi-memory character devices to
1152 * tell us the location of a shared mapping
1154 if (capabilities & NOMMU_MAP_DIRECT) {
1155 addr = file->f_op->get_unmapped_area(file, addr, len,
1157 if (IS_ERR_VALUE(addr)) {
1160 goto error_just_free;
1162 /* the driver refused to tell us where to site
1163 * the mapping so we'll have to attempt to copy
1166 if (!(capabilities & NOMMU_MAP_COPY))
1167 goto error_just_free;
1169 capabilities &= ~NOMMU_MAP_DIRECT;
1171 vma->vm_start = region->vm_start = addr;
1172 vma->vm_end = region->vm_end = addr + len;
1177 vma->vm_region = region;
1179 /* set up the mapping
1180 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1182 if (file && vma->vm_flags & VM_SHARED)
1183 ret = do_mmap_shared_file(vma);
1185 ret = do_mmap_private(vma, region, len, capabilities);
1187 goto error_just_free;
1188 add_nommu_region(region);
1190 /* clear anonymous mappings that don't ask for uninitialized data */
1191 if (!vma->vm_file &&
1192 (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) ||
1193 !(flags & MAP_UNINITIALIZED)))
1194 memset((void *)region->vm_start, 0,
1195 region->vm_end - region->vm_start);
1197 /* okay... we have a mapping; now we have to register it */
1198 result = vma->vm_start;
1200 current->mm->total_vm += len >> PAGE_SHIFT;
1203 BUG_ON(!vma->vm_region);
1204 vma_iter_config(&vmi, vma->vm_start, vma->vm_end);
1205 if (vma_iter_prealloc(&vmi, vma))
1206 goto error_just_free;
1208 setup_vma_to_mm(vma, current->mm);
1209 current->mm->map_count++;
1210 /* add the VMA to the tree */
1211 vma_iter_store(&vmi, vma);
1213 /* we flush the region from the icache only when the first executable
1214 * mapping of it is made */
1215 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1216 flush_icache_user_range(region->vm_start, region->vm_end);
1217 region->vm_icache_flushed = true;
1220 up_write(&nommu_region_sem);
1225 up_write(&nommu_region_sem);
1227 vma_iter_free(&vmi);
1228 if (region->vm_file)
1229 fput(region->vm_file);
1230 kmem_cache_free(vm_region_jar, region);
1237 up_write(&nommu_region_sem);
1238 pr_warn("Attempt to share mismatched mappings\n");
1243 kmem_cache_free(vm_region_jar, region);
1244 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1249 error_getting_region:
1250 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1256 unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1257 unsigned long prot, unsigned long flags,
1258 unsigned long fd, unsigned long pgoff)
1260 struct file *file = NULL;
1261 unsigned long retval = -EBADF;
1263 audit_mmap_fd(fd, flags);
1264 if (!(flags & MAP_ANONYMOUS)) {
1270 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1278 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1279 unsigned long, prot, unsigned long, flags,
1280 unsigned long, fd, unsigned long, pgoff)
1282 return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1285 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1286 struct mmap_arg_struct {
1290 unsigned long flags;
1292 unsigned long offset;
1295 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1297 struct mmap_arg_struct a;
1299 if (copy_from_user(&a, arg, sizeof(a)))
1301 if (offset_in_page(a.offset))
1304 return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1305 a.offset >> PAGE_SHIFT);
1307 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1310 * split a vma into two pieces at address 'addr', a new vma is allocated either
1311 * for the first part or the tail.
1313 static int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
1314 unsigned long addr, int new_below)
1316 struct vm_area_struct *new;
1317 struct vm_region *region;
1318 unsigned long npages;
1319 struct mm_struct *mm;
1321 /* we're only permitted to split anonymous regions (these should have
1322 * only a single usage on the region) */
1327 if (mm->map_count >= sysctl_max_map_count)
1330 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1334 new = vm_area_dup(vma);
1338 /* most fields are the same, copy all, and then fixup */
1339 *region = *vma->vm_region;
1340 new->vm_region = region;
1342 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1345 region->vm_top = region->vm_end = new->vm_end = addr;
1347 region->vm_start = new->vm_start = addr;
1348 region->vm_pgoff = new->vm_pgoff += npages;
1351 vma_iter_config(vmi, new->vm_start, new->vm_end);
1352 if (vma_iter_prealloc(vmi, vma)) {
1353 pr_warn("Allocation of vma tree for process %d failed\n",
1355 goto err_vmi_preallocate;
1358 if (new->vm_ops && new->vm_ops->open)
1359 new->vm_ops->open(new);
1361 down_write(&nommu_region_sem);
1362 delete_nommu_region(vma->vm_region);
1364 vma->vm_region->vm_start = vma->vm_start = addr;
1365 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1367 vma->vm_region->vm_end = vma->vm_end = addr;
1368 vma->vm_region->vm_top = addr;
1370 add_nommu_region(vma->vm_region);
1371 add_nommu_region(new->vm_region);
1372 up_write(&nommu_region_sem);
1374 setup_vma_to_mm(vma, mm);
1375 setup_vma_to_mm(new, mm);
1376 vma_iter_store(vmi, new);
1380 err_vmi_preallocate:
1383 kmem_cache_free(vm_region_jar, region);
1388 * shrink a VMA by removing the specified chunk from either the beginning or
1391 static int vmi_shrink_vma(struct vma_iterator *vmi,
1392 struct vm_area_struct *vma,
1393 unsigned long from, unsigned long to)
1395 struct vm_region *region;
1397 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1399 if (from > vma->vm_start) {
1400 if (vma_iter_clear_gfp(vmi, from, vma->vm_end, GFP_KERNEL))
1404 if (vma_iter_clear_gfp(vmi, vma->vm_start, to, GFP_KERNEL))
1409 /* cut the backing region down to size */
1410 region = vma->vm_region;
1411 BUG_ON(region->vm_usage != 1);
1413 down_write(&nommu_region_sem);
1414 delete_nommu_region(region);
1415 if (from > region->vm_start) {
1416 to = region->vm_top;
1417 region->vm_top = region->vm_end = from;
1419 region->vm_start = to;
1421 add_nommu_region(region);
1422 up_write(&nommu_region_sem);
1424 free_page_series(from, to);
1430 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1431 * VMA, though it need not cover the whole VMA
1433 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
1435 VMA_ITERATOR(vmi, mm, start);
1436 struct vm_area_struct *vma;
1440 len = PAGE_ALIGN(len);
1446 /* find the first potentially overlapping VMA */
1447 vma = vma_find(&vmi, end);
1451 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1452 current->pid, current->comm,
1453 start, start + len - 1);
1459 /* we're allowed to split an anonymous VMA but not a file-backed one */
1462 if (start > vma->vm_start)
1464 if (end == vma->vm_end)
1465 goto erase_whole_vma;
1466 vma = vma_find(&vmi, end);
1470 /* the chunk must be a subset of the VMA found */
1471 if (start == vma->vm_start && end == vma->vm_end)
1472 goto erase_whole_vma;
1473 if (start < vma->vm_start || end > vma->vm_end)
1475 if (offset_in_page(start))
1477 if (end != vma->vm_end && offset_in_page(end))
1479 if (start != vma->vm_start && end != vma->vm_end) {
1480 ret = split_vma(&vmi, vma, start, 1);
1484 return vmi_shrink_vma(&vmi, vma, start, end);
1488 if (delete_vma_from_mm(vma))
1491 delete_vma(mm, vma);
1495 int vm_munmap(unsigned long addr, size_t len)
1497 struct mm_struct *mm = current->mm;
1500 mmap_write_lock(mm);
1501 ret = do_munmap(mm, addr, len, NULL);
1502 mmap_write_unlock(mm);
1505 EXPORT_SYMBOL(vm_munmap);
1507 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1509 return vm_munmap(addr, len);
1513 * release all the mappings made in a process's VM space
1515 void exit_mmap(struct mm_struct *mm)
1517 VMA_ITERATOR(vmi, mm, 0);
1518 struct vm_area_struct *vma;
1526 * Lock the mm to avoid assert complaining even though this is the only
1529 mmap_write_lock(mm);
1530 for_each_vma(vmi, vma) {
1531 cleanup_vma_from_mm(vma);
1532 delete_vma(mm, vma);
1535 __mt_destroy(&mm->mm_mt);
1536 mmap_write_unlock(mm);
1540 * expand (or shrink) an existing mapping, potentially moving it at the same
1541 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1543 * under NOMMU conditions, we only permit changing a mapping's size, and only
1544 * as long as it stays within the region allocated by do_mmap_private() and the
1545 * block is not shareable
1547 * MREMAP_FIXED is not supported under NOMMU conditions
1549 static unsigned long do_mremap(unsigned long addr,
1550 unsigned long old_len, unsigned long new_len,
1551 unsigned long flags, unsigned long new_addr)
1553 struct vm_area_struct *vma;
1555 /* insanity checks first */
1556 old_len = PAGE_ALIGN(old_len);
1557 new_len = PAGE_ALIGN(new_len);
1558 if (old_len == 0 || new_len == 0)
1559 return (unsigned long) -EINVAL;
1561 if (offset_in_page(addr))
1564 if (flags & MREMAP_FIXED && new_addr != addr)
1565 return (unsigned long) -EINVAL;
1567 vma = find_vma_exact(current->mm, addr, old_len);
1569 return (unsigned long) -EINVAL;
1571 if (vma->vm_end != vma->vm_start + old_len)
1572 return (unsigned long) -EFAULT;
1574 if (is_nommu_shared_mapping(vma->vm_flags))
1575 return (unsigned long) -EPERM;
1577 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1578 return (unsigned long) -ENOMEM;
1580 /* all checks complete - do it */
1581 vma->vm_end = vma->vm_start + new_len;
1582 return vma->vm_start;
1585 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1586 unsigned long, new_len, unsigned long, flags,
1587 unsigned long, new_addr)
1591 mmap_write_lock(current->mm);
1592 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1593 mmap_write_unlock(current->mm);
1597 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1598 unsigned int foll_flags)
1603 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1604 unsigned long pfn, unsigned long size, pgprot_t prot)
1606 if (addr != (pfn << PAGE_SHIFT))
1609 vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP);
1612 EXPORT_SYMBOL(remap_pfn_range);
1614 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1616 unsigned long pfn = start >> PAGE_SHIFT;
1617 unsigned long vm_len = vma->vm_end - vma->vm_start;
1619 pfn += vma->vm_pgoff;
1620 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1622 EXPORT_SYMBOL(vm_iomap_memory);
1624 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1625 unsigned long pgoff)
1627 unsigned int size = vma->vm_end - vma->vm_start;
1629 if (!(vma->vm_flags & VM_USERMAP))
1632 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1633 vma->vm_end = vma->vm_start + size;
1637 EXPORT_SYMBOL(remap_vmalloc_range);
1639 vm_fault_t filemap_fault(struct vm_fault *vmf)
1644 EXPORT_SYMBOL(filemap_fault);
1646 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
1647 pgoff_t start_pgoff, pgoff_t end_pgoff)
1652 EXPORT_SYMBOL(filemap_map_pages);
1654 static int __access_remote_vm(struct mm_struct *mm, unsigned long addr,
1655 void *buf, int len, unsigned int gup_flags)
1657 struct vm_area_struct *vma;
1658 int write = gup_flags & FOLL_WRITE;
1660 if (mmap_read_lock_killable(mm))
1663 /* the access must start within one of the target process's mappings */
1664 vma = find_vma(mm, addr);
1666 /* don't overrun this mapping */
1667 if (addr + len >= vma->vm_end)
1668 len = vma->vm_end - addr;
1670 /* only read or write mappings where it is permitted */
1671 if (write && vma->vm_flags & VM_MAYWRITE)
1672 copy_to_user_page(vma, NULL, addr,
1673 (void *) addr, buf, len);
1674 else if (!write && vma->vm_flags & VM_MAYREAD)
1675 copy_from_user_page(vma, NULL, addr,
1676 buf, (void *) addr, len);
1683 mmap_read_unlock(mm);
1689 * access_remote_vm - access another process' address space
1690 * @mm: the mm_struct of the target address space
1691 * @addr: start address to access
1692 * @buf: source or destination buffer
1693 * @len: number of bytes to transfer
1694 * @gup_flags: flags modifying lookup behaviour
1696 * The caller must hold a reference on @mm.
1698 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1699 void *buf, int len, unsigned int gup_flags)
1701 return __access_remote_vm(mm, addr, buf, len, gup_flags);
1705 * Access another process' address space.
1706 * - source/target buffer must be kernel space
1708 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1709 unsigned int gup_flags)
1711 struct mm_struct *mm;
1713 if (addr + len < addr)
1716 mm = get_task_mm(tsk);
1720 len = __access_remote_vm(mm, addr, buf, len, gup_flags);
1725 EXPORT_SYMBOL_GPL(access_process_vm);
1728 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1729 * @inode: The inode to check
1730 * @size: The current filesize of the inode
1731 * @newsize: The proposed filesize of the inode
1733 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1734 * make sure that any outstanding VMAs aren't broken and then shrink the
1735 * vm_regions that extend beyond so that do_mmap() doesn't
1736 * automatically grant mappings that are too large.
1738 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1741 struct vm_area_struct *vma;
1742 struct vm_region *region;
1744 size_t r_size, r_top;
1746 low = newsize >> PAGE_SHIFT;
1747 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1749 down_write(&nommu_region_sem);
1750 i_mmap_lock_read(inode->i_mapping);
1752 /* search for VMAs that fall within the dead zone */
1753 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1754 /* found one - only interested if it's shared out of the page
1756 if (vma->vm_flags & VM_SHARED) {
1757 i_mmap_unlock_read(inode->i_mapping);
1758 up_write(&nommu_region_sem);
1759 return -ETXTBSY; /* not quite true, but near enough */
1763 /* reduce any regions that overlap the dead zone - if in existence,
1764 * these will be pointed to by VMAs that don't overlap the dead zone
1766 * we don't check for any regions that start beyond the EOF as there
1769 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1770 if (!(vma->vm_flags & VM_SHARED))
1773 region = vma->vm_region;
1774 r_size = region->vm_top - region->vm_start;
1775 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1777 if (r_top > newsize) {
1778 region->vm_top -= r_top - newsize;
1779 if (region->vm_end > region->vm_top)
1780 region->vm_end = region->vm_top;
1784 i_mmap_unlock_read(inode->i_mapping);
1785 up_write(&nommu_region_sem);
1790 * Initialise sysctl_user_reserve_kbytes.
1792 * This is intended to prevent a user from starting a single memory hogging
1793 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1796 * The default value is min(3% of free memory, 128MB)
1797 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1799 static int __meminit init_user_reserve(void)
1801 unsigned long free_kbytes;
1803 free_kbytes = K(global_zone_page_state(NR_FREE_PAGES));
1805 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1808 subsys_initcall(init_user_reserve);
1811 * Initialise sysctl_admin_reserve_kbytes.
1813 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1814 * to log in and kill a memory hogging process.
1816 * Systems with more than 256MB will reserve 8MB, enough to recover
1817 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1818 * only reserve 3% of free pages by default.
1820 static int __meminit init_admin_reserve(void)
1822 unsigned long free_kbytes;
1824 free_kbytes = K(global_zone_page_state(NR_FREE_PAGES));
1826 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1829 subsys_initcall(init_admin_reserve);