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/vmacache.h>
21 #include <linux/mman.h>
22 #include <linux/swap.h>
23 #include <linux/file.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <linux/slab.h>
27 #include <linux/vmalloc.h>
28 #include <linux/blkdev.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 <asm/uaccess.h>
40 #include <asm/tlbflush.h>
41 #include <asm/mmu_context.h>
45 EXPORT_SYMBOL(high_memory);
47 unsigned long max_mapnr;
48 EXPORT_SYMBOL(max_mapnr);
49 unsigned long highest_memmap_pfn;
50 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
51 int heap_stack_gap = 0;
53 atomic_long_t mmap_pages_allocated;
55 EXPORT_SYMBOL(mem_map);
57 /* list of mapped, potentially shareable regions */
58 static struct kmem_cache *vm_region_jar;
59 struct rb_root nommu_region_tree = RB_ROOT;
60 DECLARE_RWSEM(nommu_region_sem);
62 const struct vm_operations_struct generic_file_vm_ops = {
66 * Return the total memory allocated for this pointer, not
67 * just what the caller asked for.
69 * Doesn't have to be accurate, i.e. may have races.
71 unsigned int kobjsize(const void *objp)
76 * If the object we have should not have ksize performed on it,
79 if (!objp || !virt_addr_valid(objp))
82 page = virt_to_head_page(objp);
85 * If the allocator sets PageSlab, we know the pointer came from
92 * If it's not a compound page, see if we have a matching VMA
93 * region. This test is intentionally done in reverse order,
94 * so if there's no VMA, we still fall through and hand back
95 * PAGE_SIZE for 0-order pages.
97 if (!PageCompound(page)) {
98 struct vm_area_struct *vma;
100 vma = find_vma(current->mm, (unsigned long)objp);
102 return vma->vm_end - vma->vm_start;
106 * The ksize() function is only guaranteed to work for pointers
107 * returned by kmalloc(). So handle arbitrary pointers here.
109 return PAGE_SIZE << compound_order(page);
112 long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
113 unsigned long start, unsigned long nr_pages,
114 unsigned int foll_flags, struct page **pages,
115 struct vm_area_struct **vmas, int *nonblocking)
117 struct vm_area_struct *vma;
118 unsigned long vm_flags;
121 /* calculate required read or write permissions.
122 * If FOLL_FORCE is set, we only require the "MAY" flags.
124 vm_flags = (foll_flags & FOLL_WRITE) ?
125 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
126 vm_flags &= (foll_flags & FOLL_FORCE) ?
127 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
129 for (i = 0; i < nr_pages; i++) {
130 vma = find_vma(mm, start);
132 goto finish_or_fault;
134 /* protect what we can, including chardevs */
135 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
136 !(vm_flags & vma->vm_flags))
137 goto finish_or_fault;
140 pages[i] = virt_to_page(start);
142 page_cache_get(pages[i]);
146 start = (start + PAGE_SIZE) & PAGE_MASK;
152 return i ? : -EFAULT;
156 * get a list of pages in an address range belonging to the specified process
157 * and indicate the VMA that covers each page
158 * - this is potentially dodgy as we may end incrementing the page count of a
159 * slab page or a secondary page from a compound page
160 * - don't permit access to VMAs that don't support it, such as I/O mappings
162 long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
163 unsigned long start, unsigned long nr_pages,
164 int write, int force, struct page **pages,
165 struct vm_area_struct **vmas)
174 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
177 EXPORT_SYMBOL(get_user_pages);
179 long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm,
180 unsigned long start, unsigned long nr_pages,
181 int write, int force, struct page **pages,
184 return get_user_pages(tsk, mm, start, nr_pages, write, force,
187 EXPORT_SYMBOL(get_user_pages_locked);
189 long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
190 unsigned long start, unsigned long nr_pages,
191 int write, int force, struct page **pages,
192 unsigned int gup_flags)
195 down_read(&mm->mmap_sem);
196 ret = get_user_pages(tsk, mm, start, nr_pages, write, force,
198 up_read(&mm->mmap_sem);
201 EXPORT_SYMBOL(__get_user_pages_unlocked);
203 long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
204 unsigned long start, unsigned long nr_pages,
205 int write, int force, struct page **pages)
207 return __get_user_pages_unlocked(tsk, mm, start, nr_pages, write,
210 EXPORT_SYMBOL(get_user_pages_unlocked);
213 * follow_pfn - look up PFN at a user virtual address
214 * @vma: memory mapping
215 * @address: user virtual address
216 * @pfn: location to store found PFN
218 * Only IO mappings and raw PFN mappings are allowed.
220 * Returns zero and the pfn at @pfn on success, -ve otherwise.
222 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
225 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
228 *pfn = address >> PAGE_SHIFT;
231 EXPORT_SYMBOL(follow_pfn);
233 LIST_HEAD(vmap_area_list);
235 void vfree(const void *addr)
239 EXPORT_SYMBOL(vfree);
241 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
244 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
245 * returns only a logical address.
247 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
249 EXPORT_SYMBOL(__vmalloc);
251 void *vmalloc_user(unsigned long size)
255 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
258 struct vm_area_struct *vma;
260 down_write(¤t->mm->mmap_sem);
261 vma = find_vma(current->mm, (unsigned long)ret);
263 vma->vm_flags |= VM_USERMAP;
264 up_write(¤t->mm->mmap_sem);
269 EXPORT_SYMBOL(vmalloc_user);
271 struct page *vmalloc_to_page(const void *addr)
273 return virt_to_page(addr);
275 EXPORT_SYMBOL(vmalloc_to_page);
277 unsigned long vmalloc_to_pfn(const void *addr)
279 return page_to_pfn(virt_to_page(addr));
281 EXPORT_SYMBOL(vmalloc_to_pfn);
283 long vread(char *buf, char *addr, unsigned long count)
285 /* Don't allow overflow */
286 if ((unsigned long) buf + count < count)
287 count = -(unsigned long) buf;
289 memcpy(buf, addr, count);
293 long vwrite(char *buf, char *addr, unsigned long count)
295 /* Don't allow overflow */
296 if ((unsigned long) addr + count < count)
297 count = -(unsigned long) addr;
299 memcpy(addr, buf, count);
304 * vmalloc - allocate virtually contiguous memory
306 * @size: allocation size
308 * Allocate enough pages to cover @size from the page level
309 * allocator and map them into contiguous kernel virtual space.
311 * For tight control over page level allocator and protection flags
312 * use __vmalloc() instead.
314 void *vmalloc(unsigned long size)
316 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
318 EXPORT_SYMBOL(vmalloc);
321 * vzalloc - allocate virtually contiguous memory with zero fill
323 * @size: allocation size
325 * Allocate enough pages to cover @size from the page level
326 * allocator and map them into contiguous kernel virtual space.
327 * The memory allocated is set to zero.
329 * For tight control over page level allocator and protection flags
330 * use __vmalloc() instead.
332 void *vzalloc(unsigned long size)
334 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
337 EXPORT_SYMBOL(vzalloc);
340 * vmalloc_node - allocate memory on a specific node
341 * @size: allocation size
344 * Allocate enough pages to cover @size from the page level
345 * allocator and map them into contiguous kernel virtual space.
347 * For tight control over page level allocator and protection flags
348 * use __vmalloc() instead.
350 void *vmalloc_node(unsigned long size, int node)
352 return vmalloc(size);
354 EXPORT_SYMBOL(vmalloc_node);
357 * vzalloc_node - allocate memory on a specific node with zero fill
358 * @size: allocation size
361 * Allocate enough pages to cover @size from the page level
362 * allocator and map them into contiguous kernel virtual space.
363 * The memory allocated is set to zero.
365 * For tight control over page level allocator and protection flags
366 * use __vmalloc() instead.
368 void *vzalloc_node(unsigned long size, int node)
370 return vzalloc(size);
372 EXPORT_SYMBOL(vzalloc_node);
374 #ifndef PAGE_KERNEL_EXEC
375 # define PAGE_KERNEL_EXEC PAGE_KERNEL
379 * vmalloc_exec - allocate virtually contiguous, executable memory
380 * @size: allocation size
382 * Kernel-internal function to allocate enough pages to cover @size
383 * the page level allocator and map them into contiguous and
384 * executable kernel virtual space.
386 * For tight control over page level allocator and protection flags
387 * use __vmalloc() instead.
390 void *vmalloc_exec(unsigned long size)
392 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
396 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
397 * @size: allocation size
399 * Allocate enough 32bit PA addressable pages to cover @size from the
400 * page level allocator and map them into contiguous kernel virtual space.
402 void *vmalloc_32(unsigned long size)
404 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
406 EXPORT_SYMBOL(vmalloc_32);
409 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
410 * @size: allocation size
412 * The resulting memory area is 32bit addressable and zeroed so it can be
413 * mapped to userspace without leaking data.
415 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
416 * remap_vmalloc_range() are permissible.
418 void *vmalloc_32_user(unsigned long size)
421 * We'll have to sort out the ZONE_DMA bits for 64-bit,
422 * but for now this can simply use vmalloc_user() directly.
424 return vmalloc_user(size);
426 EXPORT_SYMBOL(vmalloc_32_user);
428 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
435 void vunmap(const void *addr)
439 EXPORT_SYMBOL(vunmap);
441 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
446 EXPORT_SYMBOL(vm_map_ram);
448 void vm_unmap_ram(const void *mem, unsigned int count)
452 EXPORT_SYMBOL(vm_unmap_ram);
454 void vm_unmap_aliases(void)
457 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
460 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
463 void __weak vmalloc_sync_all(void)
468 * alloc_vm_area - allocate a range of kernel address space
469 * @size: size of the area
471 * Returns: NULL on failure, vm_struct on success
473 * This function reserves a range of kernel address space, and
474 * allocates pagetables to map that range. No actual mappings
475 * are created. If the kernel address space is not shared
476 * between processes, it syncs the pagetable across all
479 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
484 EXPORT_SYMBOL_GPL(alloc_vm_area);
486 void free_vm_area(struct vm_struct *area)
490 EXPORT_SYMBOL_GPL(free_vm_area);
492 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
497 EXPORT_SYMBOL(vm_insert_page);
500 * sys_brk() for the most part doesn't need the global kernel
501 * lock, except when an application is doing something nasty
502 * like trying to un-brk an area that has already been mapped
503 * to a regular file. in this case, the unmapping will need
504 * to invoke file system routines that need the global lock.
506 SYSCALL_DEFINE1(brk, unsigned long, brk)
508 struct mm_struct *mm = current->mm;
510 if (brk < mm->start_brk || brk > mm->context.end_brk)
517 * Always allow shrinking brk
519 if (brk <= mm->brk) {
525 * Ok, looks good - let it rip.
527 flush_icache_range(mm->brk, brk);
528 return mm->brk = brk;
532 * initialise the VMA and region record slabs
534 void __init mmap_init(void)
538 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
540 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
544 * validate the region tree
545 * - the caller must hold the region lock
547 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
548 static noinline void validate_nommu_regions(void)
550 struct vm_region *region, *last;
551 struct rb_node *p, *lastp;
553 lastp = rb_first(&nommu_region_tree);
557 last = rb_entry(lastp, struct vm_region, vm_rb);
558 BUG_ON(last->vm_end <= last->vm_start);
559 BUG_ON(last->vm_top < last->vm_end);
561 while ((p = rb_next(lastp))) {
562 region = rb_entry(p, struct vm_region, vm_rb);
563 last = rb_entry(lastp, struct vm_region, vm_rb);
565 BUG_ON(region->vm_end <= region->vm_start);
566 BUG_ON(region->vm_top < region->vm_end);
567 BUG_ON(region->vm_start < last->vm_top);
573 static void validate_nommu_regions(void)
579 * add a region into the global tree
581 static void add_nommu_region(struct vm_region *region)
583 struct vm_region *pregion;
584 struct rb_node **p, *parent;
586 validate_nommu_regions();
589 p = &nommu_region_tree.rb_node;
592 pregion = rb_entry(parent, struct vm_region, vm_rb);
593 if (region->vm_start < pregion->vm_start)
595 else if (region->vm_start > pregion->vm_start)
597 else if (pregion == region)
603 rb_link_node(®ion->vm_rb, parent, p);
604 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
606 validate_nommu_regions();
610 * delete a region from the global tree
612 static void delete_nommu_region(struct vm_region *region)
614 BUG_ON(!nommu_region_tree.rb_node);
616 validate_nommu_regions();
617 rb_erase(®ion->vm_rb, &nommu_region_tree);
618 validate_nommu_regions();
622 * free a contiguous series of pages
624 static void free_page_series(unsigned long from, unsigned long to)
626 for (; from < to; from += PAGE_SIZE) {
627 struct page *page = virt_to_page(from);
629 atomic_long_dec(&mmap_pages_allocated);
635 * release a reference to a region
636 * - the caller must hold the region semaphore for writing, which this releases
637 * - the region may not have been added to the tree yet, in which case vm_top
638 * will equal vm_start
640 static void __put_nommu_region(struct vm_region *region)
641 __releases(nommu_region_sem)
643 BUG_ON(!nommu_region_tree.rb_node);
645 if (--region->vm_usage == 0) {
646 if (region->vm_top > region->vm_start)
647 delete_nommu_region(region);
648 up_write(&nommu_region_sem);
651 fput(region->vm_file);
653 /* IO memory and memory shared directly out of the pagecache
654 * from ramfs/tmpfs mustn't be released here */
655 if (region->vm_flags & VM_MAPPED_COPY)
656 free_page_series(region->vm_start, region->vm_top);
657 kmem_cache_free(vm_region_jar, region);
659 up_write(&nommu_region_sem);
664 * release a reference to a region
666 static void put_nommu_region(struct vm_region *region)
668 down_write(&nommu_region_sem);
669 __put_nommu_region(region);
673 * update protection on a vma
675 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
678 struct mm_struct *mm = vma->vm_mm;
679 long start = vma->vm_start & PAGE_MASK;
680 while (start < vma->vm_end) {
681 protect_page(mm, start, flags);
684 update_protections(mm);
689 * add a VMA into a process's mm_struct in the appropriate place in the list
690 * and tree and add to the address space's page tree also if not an anonymous
692 * - should be called with mm->mmap_sem held writelocked
694 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
696 struct vm_area_struct *pvma, *prev;
697 struct address_space *mapping;
698 struct rb_node **p, *parent, *rb_prev;
700 BUG_ON(!vma->vm_region);
705 protect_vma(vma, vma->vm_flags);
707 /* add the VMA to the mapping */
709 mapping = vma->vm_file->f_mapping;
711 i_mmap_lock_write(mapping);
712 flush_dcache_mmap_lock(mapping);
713 vma_interval_tree_insert(vma, &mapping->i_mmap);
714 flush_dcache_mmap_unlock(mapping);
715 i_mmap_unlock_write(mapping);
718 /* add the VMA to the tree */
719 parent = rb_prev = NULL;
720 p = &mm->mm_rb.rb_node;
723 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
725 /* sort by: start addr, end addr, VMA struct addr in that order
726 * (the latter is necessary as we may get identical VMAs) */
727 if (vma->vm_start < pvma->vm_start)
729 else if (vma->vm_start > pvma->vm_start) {
732 } else if (vma->vm_end < pvma->vm_end)
734 else if (vma->vm_end > pvma->vm_end) {
737 } else if (vma < pvma)
739 else if (vma > pvma) {
746 rb_link_node(&vma->vm_rb, parent, p);
747 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
749 /* add VMA to the VMA list also */
752 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
754 __vma_link_list(mm, vma, prev, parent);
758 * delete a VMA from its owning mm_struct and address space
760 static void delete_vma_from_mm(struct vm_area_struct *vma)
763 struct address_space *mapping;
764 struct mm_struct *mm = vma->vm_mm;
765 struct task_struct *curr = current;
770 for (i = 0; i < VMACACHE_SIZE; i++) {
771 /* if the vma is cached, invalidate the entire cache */
772 if (curr->vmacache[i] == vma) {
773 vmacache_invalidate(mm);
778 /* remove the VMA from the mapping */
780 mapping = vma->vm_file->f_mapping;
782 i_mmap_lock_write(mapping);
783 flush_dcache_mmap_lock(mapping);
784 vma_interval_tree_remove(vma, &mapping->i_mmap);
785 flush_dcache_mmap_unlock(mapping);
786 i_mmap_unlock_write(mapping);
789 /* remove from the MM's tree and list */
790 rb_erase(&vma->vm_rb, &mm->mm_rb);
793 vma->vm_prev->vm_next = vma->vm_next;
795 mm->mmap = vma->vm_next;
798 vma->vm_next->vm_prev = vma->vm_prev;
802 * destroy a VMA record
804 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
806 if (vma->vm_ops && vma->vm_ops->close)
807 vma->vm_ops->close(vma);
810 put_nommu_region(vma->vm_region);
811 kmem_cache_free(vm_area_cachep, vma);
815 * look up the first VMA in which addr resides, NULL if none
816 * - should be called with mm->mmap_sem at least held readlocked
818 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
820 struct vm_area_struct *vma;
822 /* check the cache first */
823 vma = vmacache_find(mm, addr);
827 /* trawl the list (there may be multiple mappings in which addr
829 for (vma = mm->mmap; vma; vma = vma->vm_next) {
830 if (vma->vm_start > addr)
832 if (vma->vm_end > addr) {
833 vmacache_update(addr, vma);
840 EXPORT_SYMBOL(find_vma);
844 * - we don't extend stack VMAs under NOMMU conditions
846 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
848 return find_vma(mm, addr);
852 * expand a stack to a given address
853 * - not supported under NOMMU conditions
855 int expand_stack(struct vm_area_struct *vma, unsigned long address)
861 * look up the first VMA exactly that exactly matches addr
862 * - should be called with mm->mmap_sem at least held readlocked
864 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
868 struct vm_area_struct *vma;
869 unsigned long end = addr + len;
871 /* check the cache first */
872 vma = vmacache_find_exact(mm, addr, end);
876 /* trawl the list (there may be multiple mappings in which addr
878 for (vma = mm->mmap; vma; vma = vma->vm_next) {
879 if (vma->vm_start < addr)
881 if (vma->vm_start > addr)
883 if (vma->vm_end == end) {
884 vmacache_update(addr, vma);
893 * determine whether a mapping should be permitted and, if so, what sort of
894 * mapping we're capable of supporting
896 static int validate_mmap_request(struct file *file,
902 unsigned long *_capabilities)
904 unsigned long capabilities, rlen;
907 /* do the simple checks first */
908 if (flags & MAP_FIXED)
911 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
912 (flags & MAP_TYPE) != MAP_SHARED)
918 /* Careful about overflows.. */
919 rlen = PAGE_ALIGN(len);
920 if (!rlen || rlen > TASK_SIZE)
923 /* offset overflow? */
924 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
928 /* files must support mmap */
929 if (!file->f_op->mmap)
932 /* work out if what we've got could possibly be shared
933 * - we support chardevs that provide their own "memory"
934 * - we support files/blockdevs that are memory backed
936 if (file->f_op->mmap_capabilities) {
937 capabilities = file->f_op->mmap_capabilities(file);
939 /* no explicit capabilities set, so assume some
941 switch (file_inode(file)->i_mode & S_IFMT) {
944 capabilities = NOMMU_MAP_COPY;
959 /* eliminate any capabilities that we can't support on this
961 if (!file->f_op->get_unmapped_area)
962 capabilities &= ~NOMMU_MAP_DIRECT;
963 if (!(file->f_mode & FMODE_CAN_READ))
964 capabilities &= ~NOMMU_MAP_COPY;
966 /* The file shall have been opened with read permission. */
967 if (!(file->f_mode & FMODE_READ))
970 if (flags & MAP_SHARED) {
971 /* do checks for writing, appending and locking */
972 if ((prot & PROT_WRITE) &&
973 !(file->f_mode & FMODE_WRITE))
976 if (IS_APPEND(file_inode(file)) &&
977 (file->f_mode & FMODE_WRITE))
980 if (locks_verify_locked(file))
983 if (!(capabilities & NOMMU_MAP_DIRECT))
986 /* we mustn't privatise shared mappings */
987 capabilities &= ~NOMMU_MAP_COPY;
989 /* we're going to read the file into private memory we
991 if (!(capabilities & NOMMU_MAP_COPY))
994 /* we don't permit a private writable mapping to be
995 * shared with the backing device */
996 if (prot & PROT_WRITE)
997 capabilities &= ~NOMMU_MAP_DIRECT;
1000 if (capabilities & NOMMU_MAP_DIRECT) {
1001 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
1002 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
1003 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
1005 capabilities &= ~NOMMU_MAP_DIRECT;
1006 if (flags & MAP_SHARED) {
1007 pr_warn("MAP_SHARED not completely supported on !MMU\n");
1013 /* handle executable mappings and implied executable
1015 if (path_noexec(&file->f_path)) {
1016 if (prot & PROT_EXEC)
1018 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1019 /* handle implication of PROT_EXEC by PROT_READ */
1020 if (current->personality & READ_IMPLIES_EXEC) {
1021 if (capabilities & NOMMU_MAP_EXEC)
1024 } else if ((prot & PROT_READ) &&
1025 (prot & PROT_EXEC) &&
1026 !(capabilities & NOMMU_MAP_EXEC)
1028 /* backing file is not executable, try to copy */
1029 capabilities &= ~NOMMU_MAP_DIRECT;
1032 /* anonymous mappings are always memory backed and can be
1035 capabilities = NOMMU_MAP_COPY;
1037 /* handle PROT_EXEC implication by PROT_READ */
1038 if ((prot & PROT_READ) &&
1039 (current->personality & READ_IMPLIES_EXEC))
1043 /* allow the security API to have its say */
1044 ret = security_mmap_addr(addr);
1049 *_capabilities = capabilities;
1054 * we've determined that we can make the mapping, now translate what we
1055 * now know into VMA flags
1057 static unsigned long determine_vm_flags(struct file *file,
1059 unsigned long flags,
1060 unsigned long capabilities)
1062 unsigned long vm_flags;
1064 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1065 /* vm_flags |= mm->def_flags; */
1067 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1068 /* attempt to share read-only copies of mapped file chunks */
1069 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1070 if (file && !(prot & PROT_WRITE))
1071 vm_flags |= VM_MAYSHARE;
1073 /* overlay a shareable mapping on the backing device or inode
1074 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1076 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1077 if (flags & MAP_SHARED)
1078 vm_flags |= VM_SHARED;
1081 /* refuse to let anyone share private mappings with this process if
1082 * it's being traced - otherwise breakpoints set in it may interfere
1083 * with another untraced process
1085 if ((flags & MAP_PRIVATE) && current->ptrace)
1086 vm_flags &= ~VM_MAYSHARE;
1092 * set up a shared mapping on a file (the driver or filesystem provides and
1095 static int do_mmap_shared_file(struct vm_area_struct *vma)
1099 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1101 vma->vm_region->vm_top = vma->vm_region->vm_end;
1107 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1108 * opposed to tried but failed) so we can only give a suitable error as
1109 * it's not possible to make a private copy if MAP_SHARED was given */
1114 * set up a private mapping or an anonymous shared mapping
1116 static int do_mmap_private(struct vm_area_struct *vma,
1117 struct vm_region *region,
1119 unsigned long capabilities)
1121 unsigned long total, point;
1125 /* invoke the file's mapping function so that it can keep track of
1126 * shared mappings on devices or memory
1127 * - VM_MAYSHARE will be set if it may attempt to share
1129 if (capabilities & NOMMU_MAP_DIRECT) {
1130 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1132 /* shouldn't return success if we're not sharing */
1133 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1134 vma->vm_region->vm_top = vma->vm_region->vm_end;
1140 /* getting an ENOSYS error indicates that direct mmap isn't
1141 * possible (as opposed to tried but failed) so we'll try to
1142 * make a private copy of the data and map that instead */
1146 /* allocate some memory to hold the mapping
1147 * - note that this may not return a page-aligned address if the object
1148 * we're allocating is smaller than a page
1150 order = get_order(len);
1152 point = len >> PAGE_SHIFT;
1154 /* we don't want to allocate a power-of-2 sized page set */
1155 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1158 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1162 atomic_long_add(total, &mmap_pages_allocated);
1164 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1165 region->vm_start = (unsigned long) base;
1166 region->vm_end = region->vm_start + len;
1167 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1169 vma->vm_start = region->vm_start;
1170 vma->vm_end = region->vm_start + len;
1173 /* read the contents of a file into the copy */
1174 mm_segment_t old_fs;
1177 fpos = vma->vm_pgoff;
1178 fpos <<= PAGE_SHIFT;
1182 ret = __vfs_read(vma->vm_file, base, len, &fpos);
1188 /* clear the last little bit */
1190 memset(base + ret, 0, len - ret);
1197 free_page_series(region->vm_start, region->vm_top);
1198 region->vm_start = vma->vm_start = 0;
1199 region->vm_end = vma->vm_end = 0;
1204 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1205 len, current->pid, current->comm);
1211 * handle mapping creation for uClinux
1213 unsigned long do_mmap(struct file *file,
1217 unsigned long flags,
1218 vm_flags_t vm_flags,
1219 unsigned long pgoff,
1220 unsigned long *populate)
1222 struct vm_area_struct *vma;
1223 struct vm_region *region;
1225 unsigned long capabilities, result;
1230 /* decide whether we should attempt the mapping, and if so what sort of
1232 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1237 /* we ignore the address hint */
1239 len = PAGE_ALIGN(len);
1241 /* we've determined that we can make the mapping, now translate what we
1242 * now know into VMA flags */
1243 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1245 /* we're going to need to record the mapping */
1246 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1248 goto error_getting_region;
1250 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1252 goto error_getting_vma;
1254 region->vm_usage = 1;
1255 region->vm_flags = vm_flags;
1256 region->vm_pgoff = pgoff;
1258 INIT_LIST_HEAD(&vma->anon_vma_chain);
1259 vma->vm_flags = vm_flags;
1260 vma->vm_pgoff = pgoff;
1263 region->vm_file = get_file(file);
1264 vma->vm_file = get_file(file);
1267 down_write(&nommu_region_sem);
1269 /* if we want to share, we need to check for regions created by other
1270 * mmap() calls that overlap with our proposed mapping
1271 * - we can only share with a superset match on most regular files
1272 * - shared mappings on character devices and memory backed files are
1273 * permitted to overlap inexactly as far as we are concerned for in
1274 * these cases, sharing is handled in the driver or filesystem rather
1277 if (vm_flags & VM_MAYSHARE) {
1278 struct vm_region *pregion;
1279 unsigned long pglen, rpglen, pgend, rpgend, start;
1281 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1282 pgend = pgoff + pglen;
1284 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1285 pregion = rb_entry(rb, struct vm_region, vm_rb);
1287 if (!(pregion->vm_flags & VM_MAYSHARE))
1290 /* search for overlapping mappings on the same file */
1291 if (file_inode(pregion->vm_file) !=
1295 if (pregion->vm_pgoff >= pgend)
1298 rpglen = pregion->vm_end - pregion->vm_start;
1299 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1300 rpgend = pregion->vm_pgoff + rpglen;
1301 if (pgoff >= rpgend)
1304 /* handle inexactly overlapping matches between
1306 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1307 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1308 /* new mapping is not a subset of the region */
1309 if (!(capabilities & NOMMU_MAP_DIRECT))
1310 goto sharing_violation;
1314 /* we've found a region we can share */
1315 pregion->vm_usage++;
1316 vma->vm_region = pregion;
1317 start = pregion->vm_start;
1318 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1319 vma->vm_start = start;
1320 vma->vm_end = start + len;
1322 if (pregion->vm_flags & VM_MAPPED_COPY)
1323 vma->vm_flags |= VM_MAPPED_COPY;
1325 ret = do_mmap_shared_file(vma);
1327 vma->vm_region = NULL;
1330 pregion->vm_usage--;
1332 goto error_just_free;
1335 fput(region->vm_file);
1336 kmem_cache_free(vm_region_jar, region);
1342 /* obtain the address at which to make a shared mapping
1343 * - this is the hook for quasi-memory character devices to
1344 * tell us the location of a shared mapping
1346 if (capabilities & NOMMU_MAP_DIRECT) {
1347 addr = file->f_op->get_unmapped_area(file, addr, len,
1349 if (IS_ERR_VALUE(addr)) {
1352 goto error_just_free;
1354 /* the driver refused to tell us where to site
1355 * the mapping so we'll have to attempt to copy
1358 if (!(capabilities & NOMMU_MAP_COPY))
1359 goto error_just_free;
1361 capabilities &= ~NOMMU_MAP_DIRECT;
1363 vma->vm_start = region->vm_start = addr;
1364 vma->vm_end = region->vm_end = addr + len;
1369 vma->vm_region = region;
1371 /* set up the mapping
1372 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1374 if (file && vma->vm_flags & VM_SHARED)
1375 ret = do_mmap_shared_file(vma);
1377 ret = do_mmap_private(vma, region, len, capabilities);
1379 goto error_just_free;
1380 add_nommu_region(region);
1382 /* clear anonymous mappings that don't ask for uninitialized data */
1383 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1384 memset((void *)region->vm_start, 0,
1385 region->vm_end - region->vm_start);
1387 /* okay... we have a mapping; now we have to register it */
1388 result = vma->vm_start;
1390 current->mm->total_vm += len >> PAGE_SHIFT;
1393 add_vma_to_mm(current->mm, vma);
1395 /* we flush the region from the icache only when the first executable
1396 * mapping of it is made */
1397 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1398 flush_icache_range(region->vm_start, region->vm_end);
1399 region->vm_icache_flushed = true;
1402 up_write(&nommu_region_sem);
1407 up_write(&nommu_region_sem);
1409 if (region->vm_file)
1410 fput(region->vm_file);
1411 kmem_cache_free(vm_region_jar, region);
1414 kmem_cache_free(vm_area_cachep, vma);
1418 up_write(&nommu_region_sem);
1419 pr_warn("Attempt to share mismatched mappings\n");
1424 kmem_cache_free(vm_region_jar, region);
1425 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1430 error_getting_region:
1431 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1437 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1438 unsigned long, prot, unsigned long, flags,
1439 unsigned long, fd, unsigned long, pgoff)
1441 struct file *file = NULL;
1442 unsigned long retval = -EBADF;
1444 audit_mmap_fd(fd, flags);
1445 if (!(flags & MAP_ANONYMOUS)) {
1451 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1453 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1461 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1462 struct mmap_arg_struct {
1466 unsigned long flags;
1468 unsigned long offset;
1471 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1473 struct mmap_arg_struct a;
1475 if (copy_from_user(&a, arg, sizeof(a)))
1477 if (offset_in_page(a.offset))
1480 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1481 a.offset >> PAGE_SHIFT);
1483 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1486 * split a vma into two pieces at address 'addr', a new vma is allocated either
1487 * for the first part or the tail.
1489 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1490 unsigned long addr, int new_below)
1492 struct vm_area_struct *new;
1493 struct vm_region *region;
1494 unsigned long npages;
1496 /* we're only permitted to split anonymous regions (these should have
1497 * only a single usage on the region) */
1501 if (mm->map_count >= sysctl_max_map_count)
1504 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1508 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1510 kmem_cache_free(vm_region_jar, region);
1514 /* most fields are the same, copy all, and then fixup */
1516 *region = *vma->vm_region;
1517 new->vm_region = region;
1519 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1522 region->vm_top = region->vm_end = new->vm_end = addr;
1524 region->vm_start = new->vm_start = addr;
1525 region->vm_pgoff = new->vm_pgoff += npages;
1528 if (new->vm_ops && new->vm_ops->open)
1529 new->vm_ops->open(new);
1531 delete_vma_from_mm(vma);
1532 down_write(&nommu_region_sem);
1533 delete_nommu_region(vma->vm_region);
1535 vma->vm_region->vm_start = vma->vm_start = addr;
1536 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1538 vma->vm_region->vm_end = vma->vm_end = addr;
1539 vma->vm_region->vm_top = addr;
1541 add_nommu_region(vma->vm_region);
1542 add_nommu_region(new->vm_region);
1543 up_write(&nommu_region_sem);
1544 add_vma_to_mm(mm, vma);
1545 add_vma_to_mm(mm, new);
1550 * shrink a VMA by removing the specified chunk from either the beginning or
1553 static int shrink_vma(struct mm_struct *mm,
1554 struct vm_area_struct *vma,
1555 unsigned long from, unsigned long to)
1557 struct vm_region *region;
1559 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1561 delete_vma_from_mm(vma);
1562 if (from > vma->vm_start)
1566 add_vma_to_mm(mm, vma);
1568 /* cut the backing region down to size */
1569 region = vma->vm_region;
1570 BUG_ON(region->vm_usage != 1);
1572 down_write(&nommu_region_sem);
1573 delete_nommu_region(region);
1574 if (from > region->vm_start) {
1575 to = region->vm_top;
1576 region->vm_top = region->vm_end = from;
1578 region->vm_start = to;
1580 add_nommu_region(region);
1581 up_write(&nommu_region_sem);
1583 free_page_series(from, to);
1589 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1590 * VMA, though it need not cover the whole VMA
1592 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1594 struct vm_area_struct *vma;
1598 len = PAGE_ALIGN(len);
1604 /* find the first potentially overlapping VMA */
1605 vma = find_vma(mm, start);
1609 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1610 current->pid, current->comm,
1611 start, start + len - 1);
1617 /* we're allowed to split an anonymous VMA but not a file-backed one */
1620 if (start > vma->vm_start)
1622 if (end == vma->vm_end)
1623 goto erase_whole_vma;
1628 /* the chunk must be a subset of the VMA found */
1629 if (start == vma->vm_start && end == vma->vm_end)
1630 goto erase_whole_vma;
1631 if (start < vma->vm_start || end > vma->vm_end)
1633 if (offset_in_page(start))
1635 if (end != vma->vm_end && offset_in_page(end))
1637 if (start != vma->vm_start && end != vma->vm_end) {
1638 ret = split_vma(mm, vma, start, 1);
1642 return shrink_vma(mm, vma, start, end);
1646 delete_vma_from_mm(vma);
1647 delete_vma(mm, vma);
1650 EXPORT_SYMBOL(do_munmap);
1652 int vm_munmap(unsigned long addr, size_t len)
1654 struct mm_struct *mm = current->mm;
1657 down_write(&mm->mmap_sem);
1658 ret = do_munmap(mm, addr, len);
1659 up_write(&mm->mmap_sem);
1662 EXPORT_SYMBOL(vm_munmap);
1664 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1666 return vm_munmap(addr, len);
1670 * release all the mappings made in a process's VM space
1672 void exit_mmap(struct mm_struct *mm)
1674 struct vm_area_struct *vma;
1681 while ((vma = mm->mmap)) {
1682 mm->mmap = vma->vm_next;
1683 delete_vma_from_mm(vma);
1684 delete_vma(mm, vma);
1689 unsigned long vm_brk(unsigned long addr, unsigned long len)
1695 * expand (or shrink) an existing mapping, potentially moving it at the same
1696 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1698 * under NOMMU conditions, we only permit changing a mapping's size, and only
1699 * as long as it stays within the region allocated by do_mmap_private() and the
1700 * block is not shareable
1702 * MREMAP_FIXED is not supported under NOMMU conditions
1704 static unsigned long do_mremap(unsigned long addr,
1705 unsigned long old_len, unsigned long new_len,
1706 unsigned long flags, unsigned long new_addr)
1708 struct vm_area_struct *vma;
1710 /* insanity checks first */
1711 old_len = PAGE_ALIGN(old_len);
1712 new_len = PAGE_ALIGN(new_len);
1713 if (old_len == 0 || new_len == 0)
1714 return (unsigned long) -EINVAL;
1716 if (offset_in_page(addr))
1719 if (flags & MREMAP_FIXED && new_addr != addr)
1720 return (unsigned long) -EINVAL;
1722 vma = find_vma_exact(current->mm, addr, old_len);
1724 return (unsigned long) -EINVAL;
1726 if (vma->vm_end != vma->vm_start + old_len)
1727 return (unsigned long) -EFAULT;
1729 if (vma->vm_flags & VM_MAYSHARE)
1730 return (unsigned long) -EPERM;
1732 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1733 return (unsigned long) -ENOMEM;
1735 /* all checks complete - do it */
1736 vma->vm_end = vma->vm_start + new_len;
1737 return vma->vm_start;
1740 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1741 unsigned long, new_len, unsigned long, flags,
1742 unsigned long, new_addr)
1746 down_write(¤t->mm->mmap_sem);
1747 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1748 up_write(¤t->mm->mmap_sem);
1752 struct page *follow_page_mask(struct vm_area_struct *vma,
1753 unsigned long address, unsigned int flags,
1754 unsigned int *page_mask)
1760 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1761 unsigned long pfn, unsigned long size, pgprot_t prot)
1763 if (addr != (pfn << PAGE_SHIFT))
1766 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1769 EXPORT_SYMBOL(remap_pfn_range);
1771 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1773 unsigned long pfn = start >> PAGE_SHIFT;
1774 unsigned long vm_len = vma->vm_end - vma->vm_start;
1776 pfn += vma->vm_pgoff;
1777 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1779 EXPORT_SYMBOL(vm_iomap_memory);
1781 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1782 unsigned long pgoff)
1784 unsigned int size = vma->vm_end - vma->vm_start;
1786 if (!(vma->vm_flags & VM_USERMAP))
1789 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1790 vma->vm_end = vma->vm_start + size;
1794 EXPORT_SYMBOL(remap_vmalloc_range);
1796 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1797 unsigned long len, unsigned long pgoff, unsigned long flags)
1802 void unmap_mapping_range(struct address_space *mapping,
1803 loff_t const holebegin, loff_t const holelen,
1807 EXPORT_SYMBOL(unmap_mapping_range);
1809 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1814 EXPORT_SYMBOL(filemap_fault);
1816 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1820 EXPORT_SYMBOL(filemap_map_pages);
1822 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1823 unsigned long addr, void *buf, int len, int write)
1825 struct vm_area_struct *vma;
1827 down_read(&mm->mmap_sem);
1829 /* the access must start within one of the target process's mappings */
1830 vma = find_vma(mm, addr);
1832 /* don't overrun this mapping */
1833 if (addr + len >= vma->vm_end)
1834 len = vma->vm_end - addr;
1836 /* only read or write mappings where it is permitted */
1837 if (write && vma->vm_flags & VM_MAYWRITE)
1838 copy_to_user_page(vma, NULL, addr,
1839 (void *) addr, buf, len);
1840 else if (!write && vma->vm_flags & VM_MAYREAD)
1841 copy_from_user_page(vma, NULL, addr,
1842 buf, (void *) addr, len);
1849 up_read(&mm->mmap_sem);
1855 * @access_remote_vm - access another process' address space
1856 * @mm: the mm_struct of the target address space
1857 * @addr: start address to access
1858 * @buf: source or destination buffer
1859 * @len: number of bytes to transfer
1860 * @write: whether the access is a write
1862 * The caller must hold a reference on @mm.
1864 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1865 void *buf, int len, int write)
1867 return __access_remote_vm(NULL, mm, addr, buf, len, write);
1871 * Access another process' address space.
1872 * - source/target buffer must be kernel space
1874 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1876 struct mm_struct *mm;
1878 if (addr + len < addr)
1881 mm = get_task_mm(tsk);
1885 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
1892 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1893 * @inode: The inode to check
1894 * @size: The current filesize of the inode
1895 * @newsize: The proposed filesize of the inode
1897 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1898 * make sure that that any outstanding VMAs aren't broken and then shrink the
1899 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1900 * automatically grant mappings that are too large.
1902 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1905 struct vm_area_struct *vma;
1906 struct vm_region *region;
1908 size_t r_size, r_top;
1910 low = newsize >> PAGE_SHIFT;
1911 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1913 down_write(&nommu_region_sem);
1914 i_mmap_lock_read(inode->i_mapping);
1916 /* search for VMAs that fall within the dead zone */
1917 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1918 /* found one - only interested if it's shared out of the page
1920 if (vma->vm_flags & VM_SHARED) {
1921 i_mmap_unlock_read(inode->i_mapping);
1922 up_write(&nommu_region_sem);
1923 return -ETXTBSY; /* not quite true, but near enough */
1927 /* reduce any regions that overlap the dead zone - if in existence,
1928 * these will be pointed to by VMAs that don't overlap the dead zone
1930 * we don't check for any regions that start beyond the EOF as there
1933 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1934 if (!(vma->vm_flags & VM_SHARED))
1937 region = vma->vm_region;
1938 r_size = region->vm_top - region->vm_start;
1939 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1941 if (r_top > newsize) {
1942 region->vm_top -= r_top - newsize;
1943 if (region->vm_end > region->vm_top)
1944 region->vm_end = region->vm_top;
1948 i_mmap_unlock_read(inode->i_mapping);
1949 up_write(&nommu_region_sem);
1954 * Initialise sysctl_user_reserve_kbytes.
1956 * This is intended to prevent a user from starting a single memory hogging
1957 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1960 * The default value is min(3% of free memory, 128MB)
1961 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1963 static int __meminit init_user_reserve(void)
1965 unsigned long free_kbytes;
1967 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1969 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1972 subsys_initcall(init_user_reserve);
1975 * Initialise sysctl_admin_reserve_kbytes.
1977 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1978 * to log in and kill a memory hogging process.
1980 * Systems with more than 256MB will reserve 8MB, enough to recover
1981 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1982 * only reserve 3% of free pages by default.
1984 static int __meminit init_admin_reserve(void)
1986 unsigned long free_kbytes;
1988 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1990 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1993 subsys_initcall(init_admin_reserve);