4 * Replacement code for mm functions to support CPU's that don't
5 * have any form of memory management unit (thus no virtual memory).
7 * See Documentation/nommu-mmap.txt
9 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
10 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
13 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18 #include <linux/export.h>
20 #include <linux/sched/mm.h>
21 #include <linux/vmacache.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/blkdev.h>
30 #include <linux/backing-dev.h>
31 #include <linux/compiler.h>
32 #include <linux/mount.h>
33 #include <linux/personality.h>
34 #include <linux/security.h>
35 #include <linux/syscalls.h>
36 #include <linux/audit.h>
37 #include <linux/printk.h>
39 #include <linux/uaccess.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 << compound_order(page);
113 static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
114 unsigned long start, unsigned long nr_pages,
115 unsigned int foll_flags, struct page **pages,
116 struct vm_area_struct **vmas, int *nonblocking)
118 struct vm_area_struct *vma;
119 unsigned long vm_flags;
122 /* calculate required read or write permissions.
123 * If FOLL_FORCE is set, we only require the "MAY" flags.
125 vm_flags = (foll_flags & FOLL_WRITE) ?
126 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
127 vm_flags &= (foll_flags & FOLL_FORCE) ?
128 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
130 for (i = 0; i < nr_pages; i++) {
131 vma = find_vma(mm, start);
133 goto finish_or_fault;
135 /* protect what we can, including chardevs */
136 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
137 !(vm_flags & vma->vm_flags))
138 goto finish_or_fault;
141 pages[i] = virt_to_page(start);
147 start = (start + PAGE_SIZE) & PAGE_MASK;
153 return i ? : -EFAULT;
157 * get a list of pages in an address range belonging to the specified process
158 * and indicate the VMA that covers each page
159 * - this is potentially dodgy as we may end incrementing the page count of a
160 * slab page or a secondary page from a compound page
161 * - don't permit access to VMAs that don't support it, such as I/O mappings
163 long get_user_pages(unsigned long start, unsigned long nr_pages,
164 unsigned int gup_flags, struct page **pages,
165 struct vm_area_struct **vmas)
167 return __get_user_pages(current, current->mm, start, nr_pages,
168 gup_flags, pages, vmas, NULL);
170 EXPORT_SYMBOL(get_user_pages);
172 long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
173 unsigned int gup_flags, struct page **pages,
176 return get_user_pages(start, nr_pages, gup_flags, pages, NULL);
178 EXPORT_SYMBOL(get_user_pages_locked);
180 static long __get_user_pages_unlocked(struct task_struct *tsk,
181 struct mm_struct *mm, unsigned long start,
182 unsigned long nr_pages, struct page **pages,
183 unsigned int gup_flags)
186 down_read(&mm->mmap_sem);
187 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
189 up_read(&mm->mmap_sem);
193 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
194 struct page **pages, unsigned int gup_flags)
196 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
199 EXPORT_SYMBOL(get_user_pages_unlocked);
202 * follow_pfn - look up PFN at a user virtual address
203 * @vma: memory mapping
204 * @address: user virtual address
205 * @pfn: location to store found PFN
207 * Only IO mappings and raw PFN mappings are allowed.
209 * Returns zero and the pfn at @pfn on success, -ve otherwise.
211 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
214 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
217 *pfn = address >> PAGE_SHIFT;
220 EXPORT_SYMBOL(follow_pfn);
222 LIST_HEAD(vmap_area_list);
224 void vfree(const void *addr)
228 EXPORT_SYMBOL(vfree);
230 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
233 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
234 * returns only a logical address.
236 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
238 EXPORT_SYMBOL(__vmalloc);
240 void *__vmalloc_node_flags(unsigned long size, int node, gfp_t flags)
242 return __vmalloc(size, flags, PAGE_KERNEL);
245 void *vmalloc_user(unsigned long size)
249 ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
251 struct vm_area_struct *vma;
253 down_write(¤t->mm->mmap_sem);
254 vma = find_vma(current->mm, (unsigned long)ret);
256 vma->vm_flags |= VM_USERMAP;
257 up_write(¤t->mm->mmap_sem);
262 EXPORT_SYMBOL(vmalloc_user);
264 struct page *vmalloc_to_page(const void *addr)
266 return virt_to_page(addr);
268 EXPORT_SYMBOL(vmalloc_to_page);
270 unsigned long vmalloc_to_pfn(const void *addr)
272 return page_to_pfn(virt_to_page(addr));
274 EXPORT_SYMBOL(vmalloc_to_pfn);
276 long vread(char *buf, char *addr, unsigned long count)
278 /* Don't allow overflow */
279 if ((unsigned long) buf + count < count)
280 count = -(unsigned long) buf;
282 memcpy(buf, addr, count);
286 long vwrite(char *buf, char *addr, unsigned long count)
288 /* Don't allow overflow */
289 if ((unsigned long) addr + count < count)
290 count = -(unsigned long) addr;
292 memcpy(addr, buf, count);
297 * vmalloc - allocate virtually contiguous memory
299 * @size: allocation size
301 * Allocate enough pages to cover @size from the page level
302 * allocator and map them into contiguous kernel virtual space.
304 * For tight control over page level allocator and protection flags
305 * use __vmalloc() instead.
307 void *vmalloc(unsigned long size)
309 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
311 EXPORT_SYMBOL(vmalloc);
314 * vzalloc - allocate virtually contiguous memory with zero fill
316 * @size: allocation size
318 * Allocate enough pages to cover @size from the page level
319 * allocator and map them into contiguous kernel virtual space.
320 * The memory allocated is set to zero.
322 * For tight control over page level allocator and protection flags
323 * use __vmalloc() instead.
325 void *vzalloc(unsigned long size)
327 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
330 EXPORT_SYMBOL(vzalloc);
333 * vmalloc_node - allocate memory on a specific node
334 * @size: allocation size
337 * Allocate enough pages to cover @size from the page level
338 * allocator and map them into contiguous kernel virtual space.
340 * For tight control over page level allocator and protection flags
341 * use __vmalloc() instead.
343 void *vmalloc_node(unsigned long size, int node)
345 return vmalloc(size);
347 EXPORT_SYMBOL(vmalloc_node);
350 * vzalloc_node - allocate memory on a specific node with zero fill
351 * @size: allocation size
354 * Allocate enough pages to cover @size from the page level
355 * allocator and map them into contiguous kernel virtual space.
356 * The memory allocated is set to zero.
358 * For tight control over page level allocator and protection flags
359 * use __vmalloc() instead.
361 void *vzalloc_node(unsigned long size, int node)
363 return vzalloc(size);
365 EXPORT_SYMBOL(vzalloc_node);
367 #ifndef PAGE_KERNEL_EXEC
368 # define PAGE_KERNEL_EXEC PAGE_KERNEL
372 * vmalloc_exec - allocate virtually contiguous, executable memory
373 * @size: allocation size
375 * Kernel-internal function to allocate enough pages to cover @size
376 * the page level allocator and map them into contiguous and
377 * executable kernel virtual space.
379 * For tight control over page level allocator and protection flags
380 * use __vmalloc() instead.
383 void *vmalloc_exec(unsigned long size)
385 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
389 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
390 * @size: allocation size
392 * Allocate enough 32bit PA addressable pages to cover @size from the
393 * page level allocator and map them into contiguous kernel virtual space.
395 void *vmalloc_32(unsigned long size)
397 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
399 EXPORT_SYMBOL(vmalloc_32);
402 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
403 * @size: allocation size
405 * The resulting memory area is 32bit addressable and zeroed so it can be
406 * mapped to userspace without leaking data.
408 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
409 * remap_vmalloc_range() are permissible.
411 void *vmalloc_32_user(unsigned long size)
414 * We'll have to sort out the ZONE_DMA bits for 64-bit,
415 * but for now this can simply use vmalloc_user() directly.
417 return vmalloc_user(size);
419 EXPORT_SYMBOL(vmalloc_32_user);
421 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
428 void vunmap(const void *addr)
432 EXPORT_SYMBOL(vunmap);
434 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
439 EXPORT_SYMBOL(vm_map_ram);
441 void vm_unmap_ram(const void *mem, unsigned int count)
445 EXPORT_SYMBOL(vm_unmap_ram);
447 void vm_unmap_aliases(void)
450 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
453 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
456 void __weak vmalloc_sync_all(void)
461 * alloc_vm_area - allocate a range of kernel address space
462 * @size: size of the area
464 * Returns: NULL on failure, vm_struct on success
466 * This function reserves a range of kernel address space, and
467 * allocates pagetables to map that range. No actual mappings
468 * are created. If the kernel address space is not shared
469 * between processes, it syncs the pagetable across all
472 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
477 EXPORT_SYMBOL_GPL(alloc_vm_area);
479 void free_vm_area(struct vm_struct *area)
483 EXPORT_SYMBOL_GPL(free_vm_area);
485 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
490 EXPORT_SYMBOL(vm_insert_page);
493 * sys_brk() for the most part doesn't need the global kernel
494 * lock, except when an application is doing something nasty
495 * like trying to un-brk an area that has already been mapped
496 * to a regular file. in this case, the unmapping will need
497 * to invoke file system routines that need the global lock.
499 SYSCALL_DEFINE1(brk, unsigned long, brk)
501 struct mm_struct *mm = current->mm;
503 if (brk < mm->start_brk || brk > mm->context.end_brk)
510 * Always allow shrinking brk
512 if (brk <= mm->brk) {
518 * Ok, looks good - let it rip.
520 flush_icache_range(mm->brk, brk);
521 return mm->brk = brk;
525 * initialise the percpu counter for VM and region record slabs
527 void __init mmap_init(void)
531 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
533 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
537 * validate the region tree
538 * - the caller must hold the region lock
540 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
541 static noinline void validate_nommu_regions(void)
543 struct vm_region *region, *last;
544 struct rb_node *p, *lastp;
546 lastp = rb_first(&nommu_region_tree);
550 last = rb_entry(lastp, struct vm_region, vm_rb);
551 BUG_ON(last->vm_end <= last->vm_start);
552 BUG_ON(last->vm_top < last->vm_end);
554 while ((p = rb_next(lastp))) {
555 region = rb_entry(p, struct vm_region, vm_rb);
556 last = rb_entry(lastp, struct vm_region, vm_rb);
558 BUG_ON(region->vm_end <= region->vm_start);
559 BUG_ON(region->vm_top < region->vm_end);
560 BUG_ON(region->vm_start < last->vm_top);
566 static void validate_nommu_regions(void)
572 * add a region into the global tree
574 static void add_nommu_region(struct vm_region *region)
576 struct vm_region *pregion;
577 struct rb_node **p, *parent;
579 validate_nommu_regions();
582 p = &nommu_region_tree.rb_node;
585 pregion = rb_entry(parent, struct vm_region, vm_rb);
586 if (region->vm_start < pregion->vm_start)
588 else if (region->vm_start > pregion->vm_start)
590 else if (pregion == region)
596 rb_link_node(®ion->vm_rb, parent, p);
597 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
599 validate_nommu_regions();
603 * delete a region from the global tree
605 static void delete_nommu_region(struct vm_region *region)
607 BUG_ON(!nommu_region_tree.rb_node);
609 validate_nommu_regions();
610 rb_erase(®ion->vm_rb, &nommu_region_tree);
611 validate_nommu_regions();
615 * free a contiguous series of pages
617 static void free_page_series(unsigned long from, unsigned long to)
619 for (; from < to; from += PAGE_SIZE) {
620 struct page *page = virt_to_page(from);
622 atomic_long_dec(&mmap_pages_allocated);
628 * release a reference to a region
629 * - the caller must hold the region semaphore for writing, which this releases
630 * - the region may not have been added to the tree yet, in which case vm_top
631 * will equal vm_start
633 static void __put_nommu_region(struct vm_region *region)
634 __releases(nommu_region_sem)
636 BUG_ON(!nommu_region_tree.rb_node);
638 if (--region->vm_usage == 0) {
639 if (region->vm_top > region->vm_start)
640 delete_nommu_region(region);
641 up_write(&nommu_region_sem);
644 fput(region->vm_file);
646 /* IO memory and memory shared directly out of the pagecache
647 * from ramfs/tmpfs mustn't be released here */
648 if (region->vm_flags & VM_MAPPED_COPY)
649 free_page_series(region->vm_start, region->vm_top);
650 kmem_cache_free(vm_region_jar, region);
652 up_write(&nommu_region_sem);
657 * release a reference to a region
659 static void put_nommu_region(struct vm_region *region)
661 down_write(&nommu_region_sem);
662 __put_nommu_region(region);
666 * update protection on a vma
668 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
671 struct mm_struct *mm = vma->vm_mm;
672 long start = vma->vm_start & PAGE_MASK;
673 while (start < vma->vm_end) {
674 protect_page(mm, start, flags);
677 update_protections(mm);
682 * add a VMA into a process's mm_struct in the appropriate place in the list
683 * and tree and add to the address space's page tree also if not an anonymous
685 * - should be called with mm->mmap_sem held writelocked
687 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
689 struct vm_area_struct *pvma, *prev;
690 struct address_space *mapping;
691 struct rb_node **p, *parent, *rb_prev;
693 BUG_ON(!vma->vm_region);
698 protect_vma(vma, vma->vm_flags);
700 /* add the VMA to the mapping */
702 mapping = vma->vm_file->f_mapping;
704 i_mmap_lock_write(mapping);
705 flush_dcache_mmap_lock(mapping);
706 vma_interval_tree_insert(vma, &mapping->i_mmap);
707 flush_dcache_mmap_unlock(mapping);
708 i_mmap_unlock_write(mapping);
711 /* add the VMA to the tree */
712 parent = rb_prev = NULL;
713 p = &mm->mm_rb.rb_node;
716 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
718 /* sort by: start addr, end addr, VMA struct addr in that order
719 * (the latter is necessary as we may get identical VMAs) */
720 if (vma->vm_start < pvma->vm_start)
722 else if (vma->vm_start > pvma->vm_start) {
725 } else if (vma->vm_end < pvma->vm_end)
727 else if (vma->vm_end > pvma->vm_end) {
730 } else if (vma < pvma)
732 else if (vma > pvma) {
739 rb_link_node(&vma->vm_rb, parent, p);
740 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
742 /* add VMA to the VMA list also */
745 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
747 __vma_link_list(mm, vma, prev, parent);
751 * delete a VMA from its owning mm_struct and address space
753 static void delete_vma_from_mm(struct vm_area_struct *vma)
756 struct address_space *mapping;
757 struct mm_struct *mm = vma->vm_mm;
758 struct task_struct *curr = current;
763 for (i = 0; i < VMACACHE_SIZE; i++) {
764 /* if the vma is cached, invalidate the entire cache */
765 if (curr->vmacache.vmas[i] == vma) {
766 vmacache_invalidate(mm);
771 /* remove the VMA from the mapping */
773 mapping = vma->vm_file->f_mapping;
775 i_mmap_lock_write(mapping);
776 flush_dcache_mmap_lock(mapping);
777 vma_interval_tree_remove(vma, &mapping->i_mmap);
778 flush_dcache_mmap_unlock(mapping);
779 i_mmap_unlock_write(mapping);
782 /* remove from the MM's tree and list */
783 rb_erase(&vma->vm_rb, &mm->mm_rb);
786 vma->vm_prev->vm_next = vma->vm_next;
788 mm->mmap = vma->vm_next;
791 vma->vm_next->vm_prev = vma->vm_prev;
795 * destroy a VMA record
797 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
799 if (vma->vm_ops && vma->vm_ops->close)
800 vma->vm_ops->close(vma);
803 put_nommu_region(vma->vm_region);
804 kmem_cache_free(vm_area_cachep, vma);
808 * look up the first VMA in which addr resides, NULL if none
809 * - should be called with mm->mmap_sem at least held readlocked
811 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
813 struct vm_area_struct *vma;
815 /* check the cache first */
816 vma = vmacache_find(mm, addr);
820 /* trawl the list (there may be multiple mappings in which addr
822 for (vma = mm->mmap; vma; vma = vma->vm_next) {
823 if (vma->vm_start > addr)
825 if (vma->vm_end > addr) {
826 vmacache_update(addr, vma);
833 EXPORT_SYMBOL(find_vma);
837 * - we don't extend stack VMAs under NOMMU conditions
839 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
841 return find_vma(mm, addr);
845 * expand a stack to a given address
846 * - not supported under NOMMU conditions
848 int expand_stack(struct vm_area_struct *vma, unsigned long address)
854 * look up the first VMA exactly that exactly matches addr
855 * - should be called with mm->mmap_sem at least held readlocked
857 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
861 struct vm_area_struct *vma;
862 unsigned long end = addr + len;
864 /* check the cache first */
865 vma = vmacache_find_exact(mm, addr, end);
869 /* trawl the list (there may be multiple mappings in which addr
871 for (vma = mm->mmap; vma; vma = vma->vm_next) {
872 if (vma->vm_start < addr)
874 if (vma->vm_start > addr)
876 if (vma->vm_end == end) {
877 vmacache_update(addr, vma);
886 * determine whether a mapping should be permitted and, if so, what sort of
887 * mapping we're capable of supporting
889 static int validate_mmap_request(struct file *file,
895 unsigned long *_capabilities)
897 unsigned long capabilities, rlen;
900 /* do the simple checks first */
901 if (flags & MAP_FIXED)
904 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
905 (flags & MAP_TYPE) != MAP_SHARED)
911 /* Careful about overflows.. */
912 rlen = PAGE_ALIGN(len);
913 if (!rlen || rlen > TASK_SIZE)
916 /* offset overflow? */
917 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
921 /* files must support mmap */
922 if (!file->f_op->mmap)
925 /* work out if what we've got could possibly be shared
926 * - we support chardevs that provide their own "memory"
927 * - we support files/blockdevs that are memory backed
929 if (file->f_op->mmap_capabilities) {
930 capabilities = file->f_op->mmap_capabilities(file);
932 /* no explicit capabilities set, so assume some
934 switch (file_inode(file)->i_mode & S_IFMT) {
937 capabilities = NOMMU_MAP_COPY;
952 /* eliminate any capabilities that we can't support on this
954 if (!file->f_op->get_unmapped_area)
955 capabilities &= ~NOMMU_MAP_DIRECT;
956 if (!(file->f_mode & FMODE_CAN_READ))
957 capabilities &= ~NOMMU_MAP_COPY;
959 /* The file shall have been opened with read permission. */
960 if (!(file->f_mode & FMODE_READ))
963 if (flags & MAP_SHARED) {
964 /* do checks for writing, appending and locking */
965 if ((prot & PROT_WRITE) &&
966 !(file->f_mode & FMODE_WRITE))
969 if (IS_APPEND(file_inode(file)) &&
970 (file->f_mode & FMODE_WRITE))
973 if (locks_verify_locked(file))
976 if (!(capabilities & NOMMU_MAP_DIRECT))
979 /* we mustn't privatise shared mappings */
980 capabilities &= ~NOMMU_MAP_COPY;
982 /* we're going to read the file into private memory we
984 if (!(capabilities & NOMMU_MAP_COPY))
987 /* we don't permit a private writable mapping to be
988 * shared with the backing device */
989 if (prot & PROT_WRITE)
990 capabilities &= ~NOMMU_MAP_DIRECT;
993 if (capabilities & NOMMU_MAP_DIRECT) {
994 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
995 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
996 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
998 capabilities &= ~NOMMU_MAP_DIRECT;
999 if (flags & MAP_SHARED) {
1000 pr_warn("MAP_SHARED not completely supported on !MMU\n");
1006 /* handle executable mappings and implied executable
1008 if (path_noexec(&file->f_path)) {
1009 if (prot & PROT_EXEC)
1011 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1012 /* handle implication of PROT_EXEC by PROT_READ */
1013 if (current->personality & READ_IMPLIES_EXEC) {
1014 if (capabilities & NOMMU_MAP_EXEC)
1017 } else if ((prot & PROT_READ) &&
1018 (prot & PROT_EXEC) &&
1019 !(capabilities & NOMMU_MAP_EXEC)
1021 /* backing file is not executable, try to copy */
1022 capabilities &= ~NOMMU_MAP_DIRECT;
1025 /* anonymous mappings are always memory backed and can be
1028 capabilities = NOMMU_MAP_COPY;
1030 /* handle PROT_EXEC implication by PROT_READ */
1031 if ((prot & PROT_READ) &&
1032 (current->personality & READ_IMPLIES_EXEC))
1036 /* allow the security API to have its say */
1037 ret = security_mmap_addr(addr);
1042 *_capabilities = capabilities;
1047 * we've determined that we can make the mapping, now translate what we
1048 * now know into VMA flags
1050 static unsigned long determine_vm_flags(struct file *file,
1052 unsigned long flags,
1053 unsigned long capabilities)
1055 unsigned long vm_flags;
1057 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1058 /* vm_flags |= mm->def_flags; */
1060 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1061 /* attempt to share read-only copies of mapped file chunks */
1062 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1063 if (file && !(prot & PROT_WRITE))
1064 vm_flags |= VM_MAYSHARE;
1066 /* overlay a shareable mapping on the backing device or inode
1067 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1069 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1070 if (flags & MAP_SHARED)
1071 vm_flags |= VM_SHARED;
1074 /* refuse to let anyone share private mappings with this process if
1075 * it's being traced - otherwise breakpoints set in it may interfere
1076 * with another untraced process
1078 if ((flags & MAP_PRIVATE) && current->ptrace)
1079 vm_flags &= ~VM_MAYSHARE;
1085 * set up a shared mapping on a file (the driver or filesystem provides and
1088 static int do_mmap_shared_file(struct vm_area_struct *vma)
1092 ret = call_mmap(vma->vm_file, vma);
1094 vma->vm_region->vm_top = vma->vm_region->vm_end;
1100 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1101 * opposed to tried but failed) so we can only give a suitable error as
1102 * it's not possible to make a private copy if MAP_SHARED was given */
1107 * set up a private mapping or an anonymous shared mapping
1109 static int do_mmap_private(struct vm_area_struct *vma,
1110 struct vm_region *region,
1112 unsigned long capabilities)
1114 unsigned long total, point;
1118 /* invoke the file's mapping function so that it can keep track of
1119 * shared mappings on devices or memory
1120 * - VM_MAYSHARE will be set if it may attempt to share
1122 if (capabilities & NOMMU_MAP_DIRECT) {
1123 ret = call_mmap(vma->vm_file, vma);
1125 /* shouldn't return success if we're not sharing */
1126 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1127 vma->vm_region->vm_top = vma->vm_region->vm_end;
1133 /* getting an ENOSYS error indicates that direct mmap isn't
1134 * possible (as opposed to tried but failed) so we'll try to
1135 * make a private copy of the data and map that instead */
1139 /* allocate some memory to hold the mapping
1140 * - note that this may not return a page-aligned address if the object
1141 * we're allocating is smaller than a page
1143 order = get_order(len);
1145 point = len >> PAGE_SHIFT;
1147 /* we don't want to allocate a power-of-2 sized page set */
1148 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1151 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1155 atomic_long_add(total, &mmap_pages_allocated);
1157 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1158 region->vm_start = (unsigned long) base;
1159 region->vm_end = region->vm_start + len;
1160 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1162 vma->vm_start = region->vm_start;
1163 vma->vm_end = region->vm_start + len;
1166 /* read the contents of a file into the copy */
1169 fpos = vma->vm_pgoff;
1170 fpos <<= PAGE_SHIFT;
1172 ret = kernel_read(vma->vm_file, base, len, &fpos);
1176 /* clear the last little bit */
1178 memset(base + ret, 0, len - ret);
1185 free_page_series(region->vm_start, region->vm_top);
1186 region->vm_start = vma->vm_start = 0;
1187 region->vm_end = vma->vm_end = 0;
1192 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1193 len, current->pid, current->comm);
1194 show_free_areas(0, NULL);
1199 * handle mapping creation for uClinux
1201 unsigned long do_mmap(struct file *file,
1205 unsigned long flags,
1206 vm_flags_t vm_flags,
1207 unsigned long pgoff,
1208 unsigned long *populate,
1209 struct list_head *uf)
1211 struct vm_area_struct *vma;
1212 struct vm_region *region;
1214 unsigned long capabilities, result;
1219 /* decide whether we should attempt the mapping, and if so what sort of
1221 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1226 /* we ignore the address hint */
1228 len = PAGE_ALIGN(len);
1230 /* we've determined that we can make the mapping, now translate what we
1231 * now know into VMA flags */
1232 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1234 /* we're going to need to record the mapping */
1235 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1237 goto error_getting_region;
1239 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1241 goto error_getting_vma;
1243 region->vm_usage = 1;
1244 region->vm_flags = vm_flags;
1245 region->vm_pgoff = pgoff;
1247 INIT_LIST_HEAD(&vma->anon_vma_chain);
1248 vma->vm_flags = vm_flags;
1249 vma->vm_pgoff = pgoff;
1252 region->vm_file = get_file(file);
1253 vma->vm_file = get_file(file);
1256 down_write(&nommu_region_sem);
1258 /* if we want to share, we need to check for regions created by other
1259 * mmap() calls that overlap with our proposed mapping
1260 * - we can only share with a superset match on most regular files
1261 * - shared mappings on character devices and memory backed files are
1262 * permitted to overlap inexactly as far as we are concerned for in
1263 * these cases, sharing is handled in the driver or filesystem rather
1266 if (vm_flags & VM_MAYSHARE) {
1267 struct vm_region *pregion;
1268 unsigned long pglen, rpglen, pgend, rpgend, start;
1270 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1271 pgend = pgoff + pglen;
1273 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1274 pregion = rb_entry(rb, struct vm_region, vm_rb);
1276 if (!(pregion->vm_flags & VM_MAYSHARE))
1279 /* search for overlapping mappings on the same file */
1280 if (file_inode(pregion->vm_file) !=
1284 if (pregion->vm_pgoff >= pgend)
1287 rpglen = pregion->vm_end - pregion->vm_start;
1288 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1289 rpgend = pregion->vm_pgoff + rpglen;
1290 if (pgoff >= rpgend)
1293 /* handle inexactly overlapping matches between
1295 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1296 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1297 /* new mapping is not a subset of the region */
1298 if (!(capabilities & NOMMU_MAP_DIRECT))
1299 goto sharing_violation;
1303 /* we've found a region we can share */
1304 pregion->vm_usage++;
1305 vma->vm_region = pregion;
1306 start = pregion->vm_start;
1307 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1308 vma->vm_start = start;
1309 vma->vm_end = start + len;
1311 if (pregion->vm_flags & VM_MAPPED_COPY)
1312 vma->vm_flags |= VM_MAPPED_COPY;
1314 ret = do_mmap_shared_file(vma);
1316 vma->vm_region = NULL;
1319 pregion->vm_usage--;
1321 goto error_just_free;
1324 fput(region->vm_file);
1325 kmem_cache_free(vm_region_jar, region);
1331 /* obtain the address at which to make a shared mapping
1332 * - this is the hook for quasi-memory character devices to
1333 * tell us the location of a shared mapping
1335 if (capabilities & NOMMU_MAP_DIRECT) {
1336 addr = file->f_op->get_unmapped_area(file, addr, len,
1338 if (IS_ERR_VALUE(addr)) {
1341 goto error_just_free;
1343 /* the driver refused to tell us where to site
1344 * the mapping so we'll have to attempt to copy
1347 if (!(capabilities & NOMMU_MAP_COPY))
1348 goto error_just_free;
1350 capabilities &= ~NOMMU_MAP_DIRECT;
1352 vma->vm_start = region->vm_start = addr;
1353 vma->vm_end = region->vm_end = addr + len;
1358 vma->vm_region = region;
1360 /* set up the mapping
1361 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1363 if (file && vma->vm_flags & VM_SHARED)
1364 ret = do_mmap_shared_file(vma);
1366 ret = do_mmap_private(vma, region, len, capabilities);
1368 goto error_just_free;
1369 add_nommu_region(region);
1371 /* clear anonymous mappings that don't ask for uninitialized data */
1372 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1373 memset((void *)region->vm_start, 0,
1374 region->vm_end - region->vm_start);
1376 /* okay... we have a mapping; now we have to register it */
1377 result = vma->vm_start;
1379 current->mm->total_vm += len >> PAGE_SHIFT;
1382 add_vma_to_mm(current->mm, vma);
1384 /* we flush the region from the icache only when the first executable
1385 * mapping of it is made */
1386 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1387 flush_icache_range(region->vm_start, region->vm_end);
1388 region->vm_icache_flushed = true;
1391 up_write(&nommu_region_sem);
1396 up_write(&nommu_region_sem);
1398 if (region->vm_file)
1399 fput(region->vm_file);
1400 kmem_cache_free(vm_region_jar, region);
1403 kmem_cache_free(vm_area_cachep, vma);
1407 up_write(&nommu_region_sem);
1408 pr_warn("Attempt to share mismatched mappings\n");
1413 kmem_cache_free(vm_region_jar, region);
1414 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1416 show_free_areas(0, NULL);
1419 error_getting_region:
1420 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1422 show_free_areas(0, NULL);
1426 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1427 unsigned long, prot, unsigned long, flags,
1428 unsigned long, fd, unsigned long, pgoff)
1430 struct file *file = NULL;
1431 unsigned long retval = -EBADF;
1433 audit_mmap_fd(fd, flags);
1434 if (!(flags & MAP_ANONYMOUS)) {
1440 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1442 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1450 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1451 struct mmap_arg_struct {
1455 unsigned long flags;
1457 unsigned long offset;
1460 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1462 struct mmap_arg_struct a;
1464 if (copy_from_user(&a, arg, sizeof(a)))
1466 if (offset_in_page(a.offset))
1469 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1470 a.offset >> PAGE_SHIFT);
1472 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1475 * split a vma into two pieces at address 'addr', a new vma is allocated either
1476 * for the first part or the tail.
1478 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1479 unsigned long addr, int new_below)
1481 struct vm_area_struct *new;
1482 struct vm_region *region;
1483 unsigned long npages;
1485 /* we're only permitted to split anonymous regions (these should have
1486 * only a single usage on the region) */
1490 if (mm->map_count >= sysctl_max_map_count)
1493 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1497 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1499 kmem_cache_free(vm_region_jar, region);
1503 /* most fields are the same, copy all, and then fixup */
1505 *region = *vma->vm_region;
1506 new->vm_region = region;
1508 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1511 region->vm_top = region->vm_end = new->vm_end = addr;
1513 region->vm_start = new->vm_start = addr;
1514 region->vm_pgoff = new->vm_pgoff += npages;
1517 if (new->vm_ops && new->vm_ops->open)
1518 new->vm_ops->open(new);
1520 delete_vma_from_mm(vma);
1521 down_write(&nommu_region_sem);
1522 delete_nommu_region(vma->vm_region);
1524 vma->vm_region->vm_start = vma->vm_start = addr;
1525 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1527 vma->vm_region->vm_end = vma->vm_end = addr;
1528 vma->vm_region->vm_top = addr;
1530 add_nommu_region(vma->vm_region);
1531 add_nommu_region(new->vm_region);
1532 up_write(&nommu_region_sem);
1533 add_vma_to_mm(mm, vma);
1534 add_vma_to_mm(mm, new);
1539 * shrink a VMA by removing the specified chunk from either the beginning or
1542 static int shrink_vma(struct mm_struct *mm,
1543 struct vm_area_struct *vma,
1544 unsigned long from, unsigned long to)
1546 struct vm_region *region;
1548 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1550 delete_vma_from_mm(vma);
1551 if (from > vma->vm_start)
1555 add_vma_to_mm(mm, vma);
1557 /* cut the backing region down to size */
1558 region = vma->vm_region;
1559 BUG_ON(region->vm_usage != 1);
1561 down_write(&nommu_region_sem);
1562 delete_nommu_region(region);
1563 if (from > region->vm_start) {
1564 to = region->vm_top;
1565 region->vm_top = region->vm_end = from;
1567 region->vm_start = to;
1569 add_nommu_region(region);
1570 up_write(&nommu_region_sem);
1572 free_page_series(from, to);
1578 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1579 * VMA, though it need not cover the whole VMA
1581 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
1583 struct vm_area_struct *vma;
1587 len = PAGE_ALIGN(len);
1593 /* find the first potentially overlapping VMA */
1594 vma = find_vma(mm, start);
1598 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1599 current->pid, current->comm,
1600 start, start + len - 1);
1606 /* we're allowed to split an anonymous VMA but not a file-backed one */
1609 if (start > vma->vm_start)
1611 if (end == vma->vm_end)
1612 goto erase_whole_vma;
1617 /* the chunk must be a subset of the VMA found */
1618 if (start == vma->vm_start && end == vma->vm_end)
1619 goto erase_whole_vma;
1620 if (start < vma->vm_start || end > vma->vm_end)
1622 if (offset_in_page(start))
1624 if (end != vma->vm_end && offset_in_page(end))
1626 if (start != vma->vm_start && end != vma->vm_end) {
1627 ret = split_vma(mm, vma, start, 1);
1631 return shrink_vma(mm, vma, start, end);
1635 delete_vma_from_mm(vma);
1636 delete_vma(mm, vma);
1639 EXPORT_SYMBOL(do_munmap);
1641 int vm_munmap(unsigned long addr, size_t len)
1643 struct mm_struct *mm = current->mm;
1646 down_write(&mm->mmap_sem);
1647 ret = do_munmap(mm, addr, len, NULL);
1648 up_write(&mm->mmap_sem);
1651 EXPORT_SYMBOL(vm_munmap);
1653 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1655 return vm_munmap(addr, len);
1659 * release all the mappings made in a process's VM space
1661 void exit_mmap(struct mm_struct *mm)
1663 struct vm_area_struct *vma;
1670 while ((vma = mm->mmap)) {
1671 mm->mmap = vma->vm_next;
1672 delete_vma_from_mm(vma);
1673 delete_vma(mm, vma);
1678 int vm_brk(unsigned long addr, unsigned long len)
1684 * expand (or shrink) an existing mapping, potentially moving it at the same
1685 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1687 * under NOMMU conditions, we only permit changing a mapping's size, and only
1688 * as long as it stays within the region allocated by do_mmap_private() and the
1689 * block is not shareable
1691 * MREMAP_FIXED is not supported under NOMMU conditions
1693 static unsigned long do_mremap(unsigned long addr,
1694 unsigned long old_len, unsigned long new_len,
1695 unsigned long flags, unsigned long new_addr)
1697 struct vm_area_struct *vma;
1699 /* insanity checks first */
1700 old_len = PAGE_ALIGN(old_len);
1701 new_len = PAGE_ALIGN(new_len);
1702 if (old_len == 0 || new_len == 0)
1703 return (unsigned long) -EINVAL;
1705 if (offset_in_page(addr))
1708 if (flags & MREMAP_FIXED && new_addr != addr)
1709 return (unsigned long) -EINVAL;
1711 vma = find_vma_exact(current->mm, addr, old_len);
1713 return (unsigned long) -EINVAL;
1715 if (vma->vm_end != vma->vm_start + old_len)
1716 return (unsigned long) -EFAULT;
1718 if (vma->vm_flags & VM_MAYSHARE)
1719 return (unsigned long) -EPERM;
1721 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1722 return (unsigned long) -ENOMEM;
1724 /* all checks complete - do it */
1725 vma->vm_end = vma->vm_start + new_len;
1726 return vma->vm_start;
1729 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1730 unsigned long, new_len, unsigned long, flags,
1731 unsigned long, new_addr)
1735 down_write(¤t->mm->mmap_sem);
1736 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1737 up_write(¤t->mm->mmap_sem);
1741 struct page *follow_page_mask(struct vm_area_struct *vma,
1742 unsigned long address, unsigned int flags,
1743 unsigned int *page_mask)
1749 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1750 unsigned long pfn, unsigned long size, pgprot_t prot)
1752 if (addr != (pfn << PAGE_SHIFT))
1755 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1758 EXPORT_SYMBOL(remap_pfn_range);
1760 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1762 unsigned long pfn = start >> PAGE_SHIFT;
1763 unsigned long vm_len = vma->vm_end - vma->vm_start;
1765 pfn += vma->vm_pgoff;
1766 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1768 EXPORT_SYMBOL(vm_iomap_memory);
1770 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1771 unsigned long pgoff)
1773 unsigned int size = vma->vm_end - vma->vm_start;
1775 if (!(vma->vm_flags & VM_USERMAP))
1778 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1779 vma->vm_end = vma->vm_start + size;
1783 EXPORT_SYMBOL(remap_vmalloc_range);
1785 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1786 unsigned long len, unsigned long pgoff, unsigned long flags)
1791 void unmap_mapping_range(struct address_space *mapping,
1792 loff_t const holebegin, loff_t const holelen,
1796 EXPORT_SYMBOL(unmap_mapping_range);
1798 int filemap_fault(struct vm_fault *vmf)
1803 EXPORT_SYMBOL(filemap_fault);
1805 void filemap_map_pages(struct vm_fault *vmf,
1806 pgoff_t start_pgoff, pgoff_t end_pgoff)
1810 EXPORT_SYMBOL(filemap_map_pages);
1812 int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1813 unsigned long addr, void *buf, int len, unsigned int gup_flags)
1815 struct vm_area_struct *vma;
1816 int write = gup_flags & FOLL_WRITE;
1818 down_read(&mm->mmap_sem);
1820 /* the access must start within one of the target process's mappings */
1821 vma = find_vma(mm, addr);
1823 /* don't overrun this mapping */
1824 if (addr + len >= vma->vm_end)
1825 len = vma->vm_end - addr;
1827 /* only read or write mappings where it is permitted */
1828 if (write && vma->vm_flags & VM_MAYWRITE)
1829 copy_to_user_page(vma, NULL, addr,
1830 (void *) addr, buf, len);
1831 else if (!write && vma->vm_flags & VM_MAYREAD)
1832 copy_from_user_page(vma, NULL, addr,
1833 buf, (void *) addr, len);
1840 up_read(&mm->mmap_sem);
1846 * @access_remote_vm - access another process' address space
1847 * @mm: the mm_struct of the target address space
1848 * @addr: start address to access
1849 * @buf: source or destination buffer
1850 * @len: number of bytes to transfer
1851 * @gup_flags: flags modifying lookup behaviour
1853 * The caller must hold a reference on @mm.
1855 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1856 void *buf, int len, unsigned int gup_flags)
1858 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
1862 * Access another process' address space.
1863 * - source/target buffer must be kernel space
1865 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1866 unsigned int gup_flags)
1868 struct mm_struct *mm;
1870 if (addr + len < addr)
1873 mm = get_task_mm(tsk);
1877 len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
1882 EXPORT_SYMBOL_GPL(access_process_vm);
1885 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1886 * @inode: The inode to check
1887 * @size: The current filesize of the inode
1888 * @newsize: The proposed filesize of the inode
1890 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1891 * make sure that that any outstanding VMAs aren't broken and then shrink the
1892 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1893 * automatically grant mappings that are too large.
1895 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1898 struct vm_area_struct *vma;
1899 struct vm_region *region;
1901 size_t r_size, r_top;
1903 low = newsize >> PAGE_SHIFT;
1904 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1906 down_write(&nommu_region_sem);
1907 i_mmap_lock_read(inode->i_mapping);
1909 /* search for VMAs that fall within the dead zone */
1910 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1911 /* found one - only interested if it's shared out of the page
1913 if (vma->vm_flags & VM_SHARED) {
1914 i_mmap_unlock_read(inode->i_mapping);
1915 up_write(&nommu_region_sem);
1916 return -ETXTBSY; /* not quite true, but near enough */
1920 /* reduce any regions that overlap the dead zone - if in existence,
1921 * these will be pointed to by VMAs that don't overlap the dead zone
1923 * we don't check for any regions that start beyond the EOF as there
1926 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1927 if (!(vma->vm_flags & VM_SHARED))
1930 region = vma->vm_region;
1931 r_size = region->vm_top - region->vm_start;
1932 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1934 if (r_top > newsize) {
1935 region->vm_top -= r_top - newsize;
1936 if (region->vm_end > region->vm_top)
1937 region->vm_end = region->vm_top;
1941 i_mmap_unlock_read(inode->i_mapping);
1942 up_write(&nommu_region_sem);
1947 * Initialise sysctl_user_reserve_kbytes.
1949 * This is intended to prevent a user from starting a single memory hogging
1950 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1953 * The default value is min(3% of free memory, 128MB)
1954 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1956 static int __meminit init_user_reserve(void)
1958 unsigned long free_kbytes;
1960 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1962 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1965 subsys_initcall(init_user_reserve);
1968 * Initialise sysctl_admin_reserve_kbytes.
1970 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1971 * to log in and kill a memory hogging process.
1973 * Systems with more than 256MB will reserve 8MB, enough to recover
1974 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1975 * only reserve 3% of free pages by default.
1977 static int __meminit init_admin_reserve(void)
1979 unsigned long free_kbytes;
1981 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1983 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1986 subsys_initcall(init_admin_reserve);