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 static 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);
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(unsigned long start, unsigned long nr_pages,
163 unsigned int gup_flags, struct page **pages,
164 struct vm_area_struct **vmas)
166 return __get_user_pages(current, current->mm, start, nr_pages,
167 gup_flags, pages, vmas, NULL);
169 EXPORT_SYMBOL(get_user_pages);
171 long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
172 unsigned int gup_flags, struct page **pages,
175 return get_user_pages(start, nr_pages, gup_flags, pages, NULL);
177 EXPORT_SYMBOL(get_user_pages_locked);
179 long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
180 unsigned long start, unsigned long nr_pages,
181 struct page **pages, unsigned int gup_flags)
184 down_read(&mm->mmap_sem);
185 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
187 up_read(&mm->mmap_sem);
190 EXPORT_SYMBOL(__get_user_pages_unlocked);
192 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
193 struct page **pages, unsigned int gup_flags)
195 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
198 EXPORT_SYMBOL(get_user_pages_unlocked);
201 * follow_pfn - look up PFN at a user virtual address
202 * @vma: memory mapping
203 * @address: user virtual address
204 * @pfn: location to store found PFN
206 * Only IO mappings and raw PFN mappings are allowed.
208 * Returns zero and the pfn at @pfn on success, -ve otherwise.
210 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
213 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
216 *pfn = address >> PAGE_SHIFT;
219 EXPORT_SYMBOL(follow_pfn);
221 LIST_HEAD(vmap_area_list);
223 void vfree(const void *addr)
227 EXPORT_SYMBOL(vfree);
229 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
232 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
233 * returns only a logical address.
235 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
237 EXPORT_SYMBOL(__vmalloc);
239 void *vmalloc_user(unsigned long size)
243 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
246 struct vm_area_struct *vma;
248 down_write(¤t->mm->mmap_sem);
249 vma = find_vma(current->mm, (unsigned long)ret);
251 vma->vm_flags |= VM_USERMAP;
252 up_write(¤t->mm->mmap_sem);
257 EXPORT_SYMBOL(vmalloc_user);
259 struct page *vmalloc_to_page(const void *addr)
261 return virt_to_page(addr);
263 EXPORT_SYMBOL(vmalloc_to_page);
265 unsigned long vmalloc_to_pfn(const void *addr)
267 return page_to_pfn(virt_to_page(addr));
269 EXPORT_SYMBOL(vmalloc_to_pfn);
271 long vread(char *buf, char *addr, unsigned long count)
273 /* Don't allow overflow */
274 if ((unsigned long) buf + count < count)
275 count = -(unsigned long) buf;
277 memcpy(buf, addr, count);
281 long vwrite(char *buf, char *addr, unsigned long count)
283 /* Don't allow overflow */
284 if ((unsigned long) addr + count < count)
285 count = -(unsigned long) addr;
287 memcpy(addr, buf, count);
292 * vmalloc - allocate virtually contiguous memory
294 * @size: allocation size
296 * Allocate enough pages to cover @size from the page level
297 * allocator and map them into contiguous kernel virtual space.
299 * For tight control over page level allocator and protection flags
300 * use __vmalloc() instead.
302 void *vmalloc(unsigned long size)
304 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
306 EXPORT_SYMBOL(vmalloc);
309 * vzalloc - allocate virtually contiguous memory with zero fill
311 * @size: allocation size
313 * Allocate enough pages to cover @size from the page level
314 * allocator and map them into contiguous kernel virtual space.
315 * The memory allocated is set to zero.
317 * For tight control over page level allocator and protection flags
318 * use __vmalloc() instead.
320 void *vzalloc(unsigned long size)
322 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
325 EXPORT_SYMBOL(vzalloc);
328 * vmalloc_node - allocate memory on a specific node
329 * @size: allocation size
332 * Allocate enough pages to cover @size from the page level
333 * allocator and map them into contiguous kernel virtual space.
335 * For tight control over page level allocator and protection flags
336 * use __vmalloc() instead.
338 void *vmalloc_node(unsigned long size, int node)
340 return vmalloc(size);
342 EXPORT_SYMBOL(vmalloc_node);
345 * vzalloc_node - allocate memory on a specific node with zero fill
346 * @size: allocation size
349 * Allocate enough pages to cover @size from the page level
350 * allocator and map them into contiguous kernel virtual space.
351 * The memory allocated is set to zero.
353 * For tight control over page level allocator and protection flags
354 * use __vmalloc() instead.
356 void *vzalloc_node(unsigned long size, int node)
358 return vzalloc(size);
360 EXPORT_SYMBOL(vzalloc_node);
362 #ifndef PAGE_KERNEL_EXEC
363 # define PAGE_KERNEL_EXEC PAGE_KERNEL
367 * vmalloc_exec - allocate virtually contiguous, executable memory
368 * @size: allocation size
370 * Kernel-internal function to allocate enough pages to cover @size
371 * the page level allocator and map them into contiguous and
372 * executable kernel virtual space.
374 * For tight control over page level allocator and protection flags
375 * use __vmalloc() instead.
378 void *vmalloc_exec(unsigned long size)
380 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
384 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
385 * @size: allocation size
387 * Allocate enough 32bit PA addressable pages to cover @size from the
388 * page level allocator and map them into contiguous kernel virtual space.
390 void *vmalloc_32(unsigned long size)
392 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
394 EXPORT_SYMBOL(vmalloc_32);
397 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
398 * @size: allocation size
400 * The resulting memory area is 32bit addressable and zeroed so it can be
401 * mapped to userspace without leaking data.
403 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
404 * remap_vmalloc_range() are permissible.
406 void *vmalloc_32_user(unsigned long size)
409 * We'll have to sort out the ZONE_DMA bits for 64-bit,
410 * but for now this can simply use vmalloc_user() directly.
412 return vmalloc_user(size);
414 EXPORT_SYMBOL(vmalloc_32_user);
416 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
423 void vunmap(const void *addr)
427 EXPORT_SYMBOL(vunmap);
429 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
434 EXPORT_SYMBOL(vm_map_ram);
436 void vm_unmap_ram(const void *mem, unsigned int count)
440 EXPORT_SYMBOL(vm_unmap_ram);
442 void vm_unmap_aliases(void)
445 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
448 * Implement a stub for vmalloc_sync_[un]mapping() if the architecture
449 * chose not to have one.
451 void __weak vmalloc_sync_mappings(void)
455 void __weak vmalloc_sync_unmappings(void)
460 * alloc_vm_area - allocate a range of kernel address space
461 * @size: size of the area
463 * Returns: NULL on failure, vm_struct on success
465 * This function reserves a range of kernel address space, and
466 * allocates pagetables to map that range. No actual mappings
467 * are created. If the kernel address space is not shared
468 * between processes, it syncs the pagetable across all
471 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
476 EXPORT_SYMBOL_GPL(alloc_vm_area);
478 void free_vm_area(struct vm_struct *area)
482 EXPORT_SYMBOL_GPL(free_vm_area);
484 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
489 EXPORT_SYMBOL(vm_insert_page);
492 * sys_brk() for the most part doesn't need the global kernel
493 * lock, except when an application is doing something nasty
494 * like trying to un-brk an area that has already been mapped
495 * to a regular file. in this case, the unmapping will need
496 * to invoke file system routines that need the global lock.
498 SYSCALL_DEFINE1(brk, unsigned long, brk)
500 struct mm_struct *mm = current->mm;
502 if (brk < mm->start_brk || brk > mm->context.end_brk)
509 * Always allow shrinking brk
511 if (brk <= mm->brk) {
517 * Ok, looks good - let it rip.
519 flush_icache_range(mm->brk, brk);
520 return mm->brk = brk;
524 * initialise the VMA and region record slabs
526 void __init mmap_init(void)
530 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
532 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
536 * validate the region tree
537 * - the caller must hold the region lock
539 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
540 static noinline void validate_nommu_regions(void)
542 struct vm_region *region, *last;
543 struct rb_node *p, *lastp;
545 lastp = rb_first(&nommu_region_tree);
549 last = rb_entry(lastp, struct vm_region, vm_rb);
550 BUG_ON(last->vm_end <= last->vm_start);
551 BUG_ON(last->vm_top < last->vm_end);
553 while ((p = rb_next(lastp))) {
554 region = rb_entry(p, struct vm_region, vm_rb);
555 last = rb_entry(lastp, struct vm_region, vm_rb);
557 BUG_ON(region->vm_end <= region->vm_start);
558 BUG_ON(region->vm_top < region->vm_end);
559 BUG_ON(region->vm_start < last->vm_top);
565 static void validate_nommu_regions(void)
571 * add a region into the global tree
573 static void add_nommu_region(struct vm_region *region)
575 struct vm_region *pregion;
576 struct rb_node **p, *parent;
578 validate_nommu_regions();
581 p = &nommu_region_tree.rb_node;
584 pregion = rb_entry(parent, struct vm_region, vm_rb);
585 if (region->vm_start < pregion->vm_start)
587 else if (region->vm_start > pregion->vm_start)
589 else if (pregion == region)
595 rb_link_node(®ion->vm_rb, parent, p);
596 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
598 validate_nommu_regions();
602 * delete a region from the global tree
604 static void delete_nommu_region(struct vm_region *region)
606 BUG_ON(!nommu_region_tree.rb_node);
608 validate_nommu_regions();
609 rb_erase(®ion->vm_rb, &nommu_region_tree);
610 validate_nommu_regions();
614 * free a contiguous series of pages
616 static void free_page_series(unsigned long from, unsigned long to)
618 for (; from < to; from += PAGE_SIZE) {
619 struct page *page = virt_to_page(from);
621 atomic_long_dec(&mmap_pages_allocated);
627 * release a reference to a region
628 * - the caller must hold the region semaphore for writing, which this releases
629 * - the region may not have been added to the tree yet, in which case vm_top
630 * will equal vm_start
632 static void __put_nommu_region(struct vm_region *region)
633 __releases(nommu_region_sem)
635 BUG_ON(!nommu_region_tree.rb_node);
637 if (--region->vm_usage == 0) {
638 if (region->vm_top > region->vm_start)
639 delete_nommu_region(region);
640 up_write(&nommu_region_sem);
643 fput(region->vm_file);
645 /* IO memory and memory shared directly out of the pagecache
646 * from ramfs/tmpfs mustn't be released here */
647 if (region->vm_flags & VM_MAPPED_COPY)
648 free_page_series(region->vm_start, region->vm_top);
649 kmem_cache_free(vm_region_jar, region);
651 up_write(&nommu_region_sem);
656 * release a reference to a region
658 static void put_nommu_region(struct vm_region *region)
660 down_write(&nommu_region_sem);
661 __put_nommu_region(region);
665 * update protection on a vma
667 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
670 struct mm_struct *mm = vma->vm_mm;
671 long start = vma->vm_start & PAGE_MASK;
672 while (start < vma->vm_end) {
673 protect_page(mm, start, flags);
676 update_protections(mm);
681 * add a VMA into a process's mm_struct in the appropriate place in the list
682 * and tree and add to the address space's page tree also if not an anonymous
684 * - should be called with mm->mmap_sem held writelocked
686 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
688 struct vm_area_struct *pvma, *prev;
689 struct address_space *mapping;
690 struct rb_node **p, *parent, *rb_prev;
692 BUG_ON(!vma->vm_region);
697 protect_vma(vma, vma->vm_flags);
699 /* add the VMA to the mapping */
701 mapping = vma->vm_file->f_mapping;
703 i_mmap_lock_write(mapping);
704 flush_dcache_mmap_lock(mapping);
705 vma_interval_tree_insert(vma, &mapping->i_mmap);
706 flush_dcache_mmap_unlock(mapping);
707 i_mmap_unlock_write(mapping);
710 /* add the VMA to the tree */
711 parent = rb_prev = NULL;
712 p = &mm->mm_rb.rb_node;
715 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
717 /* sort by: start addr, end addr, VMA struct addr in that order
718 * (the latter is necessary as we may get identical VMAs) */
719 if (vma->vm_start < pvma->vm_start)
721 else if (vma->vm_start > pvma->vm_start) {
724 } else if (vma->vm_end < pvma->vm_end)
726 else if (vma->vm_end > pvma->vm_end) {
729 } else if (vma < pvma)
731 else if (vma > pvma) {
738 rb_link_node(&vma->vm_rb, parent, p);
739 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
741 /* add VMA to the VMA list also */
744 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
746 __vma_link_list(mm, vma, prev, parent);
750 * delete a VMA from its owning mm_struct and address space
752 static void delete_vma_from_mm(struct vm_area_struct *vma)
755 struct address_space *mapping;
756 struct mm_struct *mm = vma->vm_mm;
757 struct task_struct *curr = current;
762 for (i = 0; i < VMACACHE_SIZE; i++) {
763 /* if the vma is cached, invalidate the entire cache */
764 if (curr->vmacache[i] == vma) {
765 vmacache_invalidate(mm);
770 /* remove the VMA from the mapping */
772 mapping = vma->vm_file->f_mapping;
774 i_mmap_lock_write(mapping);
775 flush_dcache_mmap_lock(mapping);
776 vma_interval_tree_remove(vma, &mapping->i_mmap);
777 flush_dcache_mmap_unlock(mapping);
778 i_mmap_unlock_write(mapping);
781 /* remove from the MM's tree and list */
782 rb_erase(&vma->vm_rb, &mm->mm_rb);
785 vma->vm_prev->vm_next = vma->vm_next;
787 mm->mmap = vma->vm_next;
790 vma->vm_next->vm_prev = vma->vm_prev;
794 * destroy a VMA record
796 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
798 if (vma->vm_ops && vma->vm_ops->close)
799 vma->vm_ops->close(vma);
802 put_nommu_region(vma->vm_region);
803 kmem_cache_free(vm_area_cachep, vma);
807 * look up the first VMA in which addr resides, NULL if none
808 * - should be called with mm->mmap_sem at least held readlocked
810 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
812 struct vm_area_struct *vma;
814 /* check the cache first */
815 vma = vmacache_find(mm, addr);
819 /* trawl the list (there may be multiple mappings in which addr
821 for (vma = mm->mmap; vma; vma = vma->vm_next) {
822 if (vma->vm_start > addr)
824 if (vma->vm_end > addr) {
825 vmacache_update(addr, vma);
832 EXPORT_SYMBOL(find_vma);
836 * - we don't extend stack VMAs under NOMMU conditions
838 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
840 return find_vma(mm, addr);
844 * expand a stack to a given address
845 * - not supported under NOMMU conditions
847 int expand_stack(struct vm_area_struct *vma, unsigned long address)
853 * look up the first VMA exactly that exactly matches addr
854 * - should be called with mm->mmap_sem at least held readlocked
856 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
860 struct vm_area_struct *vma;
861 unsigned long end = addr + len;
863 /* check the cache first */
864 vma = vmacache_find_exact(mm, addr, end);
868 /* trawl the list (there may be multiple mappings in which addr
870 for (vma = mm->mmap; vma; vma = vma->vm_next) {
871 if (vma->vm_start < addr)
873 if (vma->vm_start > addr)
875 if (vma->vm_end == end) {
876 vmacache_update(addr, vma);
885 * determine whether a mapping should be permitted and, if so, what sort of
886 * mapping we're capable of supporting
888 static int validate_mmap_request(struct file *file,
894 unsigned long *_capabilities)
896 unsigned long capabilities, rlen;
899 /* do the simple checks first */
900 if (flags & MAP_FIXED)
903 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
904 (flags & MAP_TYPE) != MAP_SHARED)
910 /* Careful about overflows.. */
911 rlen = PAGE_ALIGN(len);
912 if (!rlen || rlen > TASK_SIZE)
915 /* offset overflow? */
916 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
920 /* files must support mmap */
921 if (!file->f_op->mmap)
924 /* work out if what we've got could possibly be shared
925 * - we support chardevs that provide their own "memory"
926 * - we support files/blockdevs that are memory backed
928 if (file->f_op->mmap_capabilities) {
929 capabilities = file->f_op->mmap_capabilities(file);
931 /* no explicit capabilities set, so assume some
933 switch (file_inode(file)->i_mode & S_IFMT) {
936 capabilities = NOMMU_MAP_COPY;
951 /* eliminate any capabilities that we can't support on this
953 if (!file->f_op->get_unmapped_area)
954 capabilities &= ~NOMMU_MAP_DIRECT;
955 if (!(file->f_mode & FMODE_CAN_READ))
956 capabilities &= ~NOMMU_MAP_COPY;
958 /* The file shall have been opened with read permission. */
959 if (!(file->f_mode & FMODE_READ))
962 if (flags & MAP_SHARED) {
963 /* do checks for writing, appending and locking */
964 if ((prot & PROT_WRITE) &&
965 !(file->f_mode & FMODE_WRITE))
968 if (IS_APPEND(file_inode(file)) &&
969 (file->f_mode & FMODE_WRITE))
972 if (locks_verify_locked(file))
975 if (!(capabilities & NOMMU_MAP_DIRECT))
978 /* we mustn't privatise shared mappings */
979 capabilities &= ~NOMMU_MAP_COPY;
981 /* we're going to read the file into private memory we
983 if (!(capabilities & NOMMU_MAP_COPY))
986 /* we don't permit a private writable mapping to be
987 * shared with the backing device */
988 if (prot & PROT_WRITE)
989 capabilities &= ~NOMMU_MAP_DIRECT;
992 if (capabilities & NOMMU_MAP_DIRECT) {
993 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
994 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
995 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
997 capabilities &= ~NOMMU_MAP_DIRECT;
998 if (flags & MAP_SHARED) {
999 pr_warn("MAP_SHARED not completely supported on !MMU\n");
1005 /* handle executable mappings and implied executable
1007 if (path_noexec(&file->f_path)) {
1008 if (prot & PROT_EXEC)
1010 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1011 /* handle implication of PROT_EXEC by PROT_READ */
1012 if (current->personality & READ_IMPLIES_EXEC) {
1013 if (capabilities & NOMMU_MAP_EXEC)
1016 } else if ((prot & PROT_READ) &&
1017 (prot & PROT_EXEC) &&
1018 !(capabilities & NOMMU_MAP_EXEC)
1020 /* backing file is not executable, try to copy */
1021 capabilities &= ~NOMMU_MAP_DIRECT;
1024 /* anonymous mappings are always memory backed and can be
1027 capabilities = NOMMU_MAP_COPY;
1029 /* handle PROT_EXEC implication by PROT_READ */
1030 if ((prot & PROT_READ) &&
1031 (current->personality & READ_IMPLIES_EXEC))
1035 /* allow the security API to have its say */
1036 ret = security_mmap_addr(addr);
1041 *_capabilities = capabilities;
1046 * we've determined that we can make the mapping, now translate what we
1047 * now know into VMA flags
1049 static unsigned long determine_vm_flags(struct file *file,
1051 unsigned long flags,
1052 unsigned long capabilities)
1054 unsigned long vm_flags;
1056 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1057 /* vm_flags |= mm->def_flags; */
1059 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1060 /* attempt to share read-only copies of mapped file chunks */
1061 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1062 if (file && !(prot & PROT_WRITE))
1063 vm_flags |= VM_MAYSHARE;
1065 /* overlay a shareable mapping on the backing device or inode
1066 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1068 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1069 if (flags & MAP_SHARED)
1070 vm_flags |= VM_SHARED;
1073 /* refuse to let anyone share private mappings with this process if
1074 * it's being traced - otherwise breakpoints set in it may interfere
1075 * with another untraced process
1077 if ((flags & MAP_PRIVATE) && current->ptrace)
1078 vm_flags &= ~VM_MAYSHARE;
1084 * set up a shared mapping on a file (the driver or filesystem provides and
1087 static int do_mmap_shared_file(struct vm_area_struct *vma)
1091 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1093 vma->vm_region->vm_top = vma->vm_region->vm_end;
1099 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1100 * opposed to tried but failed) so we can only give a suitable error as
1101 * it's not possible to make a private copy if MAP_SHARED was given */
1106 * set up a private mapping or an anonymous shared mapping
1108 static int do_mmap_private(struct vm_area_struct *vma,
1109 struct vm_region *region,
1111 unsigned long capabilities)
1113 unsigned long total, point;
1117 /* invoke the file's mapping function so that it can keep track of
1118 * shared mappings on devices or memory
1119 * - VM_MAYSHARE will be set if it may attempt to share
1121 if (capabilities & NOMMU_MAP_DIRECT) {
1122 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1124 /* shouldn't return success if we're not sharing */
1125 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1126 vma->vm_region->vm_top = vma->vm_region->vm_end;
1132 /* getting an ENOSYS error indicates that direct mmap isn't
1133 * possible (as opposed to tried but failed) so we'll try to
1134 * make a private copy of the data and map that instead */
1138 /* allocate some memory to hold the mapping
1139 * - note that this may not return a page-aligned address if the object
1140 * we're allocating is smaller than a page
1142 order = get_order(len);
1144 point = len >> PAGE_SHIFT;
1146 /* we don't want to allocate a power-of-2 sized page set */
1147 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1150 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1154 atomic_long_add(total, &mmap_pages_allocated);
1156 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1157 region->vm_start = (unsigned long) base;
1158 region->vm_end = region->vm_start + len;
1159 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1161 vma->vm_start = region->vm_start;
1162 vma->vm_end = region->vm_start + len;
1165 /* read the contents of a file into the copy */
1166 mm_segment_t old_fs;
1169 fpos = vma->vm_pgoff;
1170 fpos <<= PAGE_SHIFT;
1174 ret = __vfs_read(vma->vm_file, base, len, &fpos);
1180 /* clear the last little bit */
1182 memset(base + ret, 0, len - ret);
1189 free_page_series(region->vm_start, region->vm_top);
1190 region->vm_start = vma->vm_start = 0;
1191 region->vm_end = vma->vm_end = 0;
1196 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1197 len, current->pid, current->comm);
1203 * handle mapping creation for uClinux
1205 unsigned long do_mmap(struct file *file,
1209 unsigned long flags,
1210 vm_flags_t vm_flags,
1211 unsigned long pgoff,
1212 unsigned long *populate)
1214 struct vm_area_struct *vma;
1215 struct vm_region *region;
1217 unsigned long capabilities, result;
1222 /* decide whether we should attempt the mapping, and if so what sort of
1224 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1229 /* we ignore the address hint */
1231 len = PAGE_ALIGN(len);
1233 /* we've determined that we can make the mapping, now translate what we
1234 * now know into VMA flags */
1235 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1237 /* we're going to need to record the mapping */
1238 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1240 goto error_getting_region;
1242 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1244 goto error_getting_vma;
1246 region->vm_usage = 1;
1247 region->vm_flags = vm_flags;
1248 region->vm_pgoff = pgoff;
1250 INIT_LIST_HEAD(&vma->anon_vma_chain);
1251 vma->vm_flags = vm_flags;
1252 vma->vm_pgoff = pgoff;
1255 region->vm_file = get_file(file);
1256 vma->vm_file = get_file(file);
1259 down_write(&nommu_region_sem);
1261 /* if we want to share, we need to check for regions created by other
1262 * mmap() calls that overlap with our proposed mapping
1263 * - we can only share with a superset match on most regular files
1264 * - shared mappings on character devices and memory backed files are
1265 * permitted to overlap inexactly as far as we are concerned for in
1266 * these cases, sharing is handled in the driver or filesystem rather
1269 if (vm_flags & VM_MAYSHARE) {
1270 struct vm_region *pregion;
1271 unsigned long pglen, rpglen, pgend, rpgend, start;
1273 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1274 pgend = pgoff + pglen;
1276 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1277 pregion = rb_entry(rb, struct vm_region, vm_rb);
1279 if (!(pregion->vm_flags & VM_MAYSHARE))
1282 /* search for overlapping mappings on the same file */
1283 if (file_inode(pregion->vm_file) !=
1287 if (pregion->vm_pgoff >= pgend)
1290 rpglen = pregion->vm_end - pregion->vm_start;
1291 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1292 rpgend = pregion->vm_pgoff + rpglen;
1293 if (pgoff >= rpgend)
1296 /* handle inexactly overlapping matches between
1298 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1299 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1300 /* new mapping is not a subset of the region */
1301 if (!(capabilities & NOMMU_MAP_DIRECT))
1302 goto sharing_violation;
1306 /* we've found a region we can share */
1307 pregion->vm_usage++;
1308 vma->vm_region = pregion;
1309 start = pregion->vm_start;
1310 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1311 vma->vm_start = start;
1312 vma->vm_end = start + len;
1314 if (pregion->vm_flags & VM_MAPPED_COPY)
1315 vma->vm_flags |= VM_MAPPED_COPY;
1317 ret = do_mmap_shared_file(vma);
1319 vma->vm_region = NULL;
1322 pregion->vm_usage--;
1324 goto error_just_free;
1327 fput(region->vm_file);
1328 kmem_cache_free(vm_region_jar, region);
1334 /* obtain the address at which to make a shared mapping
1335 * - this is the hook for quasi-memory character devices to
1336 * tell us the location of a shared mapping
1338 if (capabilities & NOMMU_MAP_DIRECT) {
1339 addr = file->f_op->get_unmapped_area(file, addr, len,
1341 if (IS_ERR_VALUE(addr)) {
1344 goto error_just_free;
1346 /* the driver refused to tell us where to site
1347 * the mapping so we'll have to attempt to copy
1350 if (!(capabilities & NOMMU_MAP_COPY))
1351 goto error_just_free;
1353 capabilities &= ~NOMMU_MAP_DIRECT;
1355 vma->vm_start = region->vm_start = addr;
1356 vma->vm_end = region->vm_end = addr + len;
1361 vma->vm_region = region;
1363 /* set up the mapping
1364 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1366 if (file && vma->vm_flags & VM_SHARED)
1367 ret = do_mmap_shared_file(vma);
1369 ret = do_mmap_private(vma, region, len, capabilities);
1371 goto error_just_free;
1372 add_nommu_region(region);
1374 /* clear anonymous mappings that don't ask for uninitialized data */
1375 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1376 memset((void *)region->vm_start, 0,
1377 region->vm_end - region->vm_start);
1379 /* okay... we have a mapping; now we have to register it */
1380 result = vma->vm_start;
1382 current->mm->total_vm += len >> PAGE_SHIFT;
1385 add_vma_to_mm(current->mm, vma);
1387 /* we flush the region from the icache only when the first executable
1388 * mapping of it is made */
1389 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1390 flush_icache_range(region->vm_start, region->vm_end);
1391 region->vm_icache_flushed = true;
1394 up_write(&nommu_region_sem);
1399 up_write(&nommu_region_sem);
1401 if (region->vm_file)
1402 fput(region->vm_file);
1403 kmem_cache_free(vm_region_jar, region);
1406 kmem_cache_free(vm_area_cachep, vma);
1410 up_write(&nommu_region_sem);
1411 pr_warn("Attempt to share mismatched mappings\n");
1416 kmem_cache_free(vm_region_jar, region);
1417 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1422 error_getting_region:
1423 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1429 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1430 unsigned long, prot, unsigned long, flags,
1431 unsigned long, fd, unsigned long, pgoff)
1433 struct file *file = NULL;
1434 unsigned long retval = -EBADF;
1436 audit_mmap_fd(fd, flags);
1437 if (!(flags & MAP_ANONYMOUS)) {
1443 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1445 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1453 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1454 struct mmap_arg_struct {
1458 unsigned long flags;
1460 unsigned long offset;
1463 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1465 struct mmap_arg_struct a;
1467 if (copy_from_user(&a, arg, sizeof(a)))
1469 if (offset_in_page(a.offset))
1472 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1473 a.offset >> PAGE_SHIFT);
1475 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1478 * split a vma into two pieces at address 'addr', a new vma is allocated either
1479 * for the first part or the tail.
1481 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1482 unsigned long addr, int new_below)
1484 struct vm_area_struct *new;
1485 struct vm_region *region;
1486 unsigned long npages;
1488 /* we're only permitted to split anonymous regions (these should have
1489 * only a single usage on the region) */
1493 if (mm->map_count >= sysctl_max_map_count)
1496 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1500 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1502 kmem_cache_free(vm_region_jar, region);
1506 /* most fields are the same, copy all, and then fixup */
1508 *region = *vma->vm_region;
1509 new->vm_region = region;
1511 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1514 region->vm_top = region->vm_end = new->vm_end = addr;
1516 region->vm_start = new->vm_start = addr;
1517 region->vm_pgoff = new->vm_pgoff += npages;
1520 if (new->vm_ops && new->vm_ops->open)
1521 new->vm_ops->open(new);
1523 delete_vma_from_mm(vma);
1524 down_write(&nommu_region_sem);
1525 delete_nommu_region(vma->vm_region);
1527 vma->vm_region->vm_start = vma->vm_start = addr;
1528 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1530 vma->vm_region->vm_end = vma->vm_end = addr;
1531 vma->vm_region->vm_top = addr;
1533 add_nommu_region(vma->vm_region);
1534 add_nommu_region(new->vm_region);
1535 up_write(&nommu_region_sem);
1536 add_vma_to_mm(mm, vma);
1537 add_vma_to_mm(mm, new);
1542 * shrink a VMA by removing the specified chunk from either the beginning or
1545 static int shrink_vma(struct mm_struct *mm,
1546 struct vm_area_struct *vma,
1547 unsigned long from, unsigned long to)
1549 struct vm_region *region;
1551 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1553 delete_vma_from_mm(vma);
1554 if (from > vma->vm_start)
1558 add_vma_to_mm(mm, vma);
1560 /* cut the backing region down to size */
1561 region = vma->vm_region;
1562 BUG_ON(region->vm_usage != 1);
1564 down_write(&nommu_region_sem);
1565 delete_nommu_region(region);
1566 if (from > region->vm_start) {
1567 to = region->vm_top;
1568 region->vm_top = region->vm_end = from;
1570 region->vm_start = to;
1572 add_nommu_region(region);
1573 up_write(&nommu_region_sem);
1575 free_page_series(from, to);
1581 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1582 * VMA, though it need not cover the whole VMA
1584 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1586 struct vm_area_struct *vma;
1590 len = PAGE_ALIGN(len);
1596 /* find the first potentially overlapping VMA */
1597 vma = find_vma(mm, start);
1601 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1602 current->pid, current->comm,
1603 start, start + len - 1);
1609 /* we're allowed to split an anonymous VMA but not a file-backed one */
1612 if (start > vma->vm_start)
1614 if (end == vma->vm_end)
1615 goto erase_whole_vma;
1620 /* the chunk must be a subset of the VMA found */
1621 if (start == vma->vm_start && end == vma->vm_end)
1622 goto erase_whole_vma;
1623 if (start < vma->vm_start || end > vma->vm_end)
1625 if (offset_in_page(start))
1627 if (end != vma->vm_end && offset_in_page(end))
1629 if (start != vma->vm_start && end != vma->vm_end) {
1630 ret = split_vma(mm, vma, start, 1);
1634 return shrink_vma(mm, vma, start, end);
1638 delete_vma_from_mm(vma);
1639 delete_vma(mm, vma);
1642 EXPORT_SYMBOL(do_munmap);
1644 int vm_munmap(unsigned long addr, size_t len)
1646 struct mm_struct *mm = current->mm;
1649 down_write(&mm->mmap_sem);
1650 ret = do_munmap(mm, addr, len);
1651 up_write(&mm->mmap_sem);
1654 EXPORT_SYMBOL(vm_munmap);
1656 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1658 return vm_munmap(addr, len);
1662 * release all the mappings made in a process's VM space
1664 void exit_mmap(struct mm_struct *mm)
1666 struct vm_area_struct *vma;
1673 while ((vma = mm->mmap)) {
1674 mm->mmap = vma->vm_next;
1675 delete_vma_from_mm(vma);
1676 delete_vma(mm, vma);
1681 int vm_brk(unsigned long addr, unsigned long len)
1687 * expand (or shrink) an existing mapping, potentially moving it at the same
1688 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1690 * under NOMMU conditions, we only permit changing a mapping's size, and only
1691 * as long as it stays within the region allocated by do_mmap_private() and the
1692 * block is not shareable
1694 * MREMAP_FIXED is not supported under NOMMU conditions
1696 static unsigned long do_mremap(unsigned long addr,
1697 unsigned long old_len, unsigned long new_len,
1698 unsigned long flags, unsigned long new_addr)
1700 struct vm_area_struct *vma;
1702 /* insanity checks first */
1703 old_len = PAGE_ALIGN(old_len);
1704 new_len = PAGE_ALIGN(new_len);
1705 if (old_len == 0 || new_len == 0)
1706 return (unsigned long) -EINVAL;
1708 if (offset_in_page(addr))
1711 if (flags & MREMAP_FIXED && new_addr != addr)
1712 return (unsigned long) -EINVAL;
1714 vma = find_vma_exact(current->mm, addr, old_len);
1716 return (unsigned long) -EINVAL;
1718 if (vma->vm_end != vma->vm_start + old_len)
1719 return (unsigned long) -EFAULT;
1721 if (vma->vm_flags & VM_MAYSHARE)
1722 return (unsigned long) -EPERM;
1724 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1725 return (unsigned long) -ENOMEM;
1727 /* all checks complete - do it */
1728 vma->vm_end = vma->vm_start + new_len;
1729 return vma->vm_start;
1732 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1733 unsigned long, new_len, unsigned long, flags,
1734 unsigned long, new_addr)
1738 down_write(¤t->mm->mmap_sem);
1739 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1740 up_write(¤t->mm->mmap_sem);
1744 struct page *follow_page_mask(struct vm_area_struct *vma,
1745 unsigned long address, unsigned int flags,
1746 unsigned int *page_mask)
1752 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1753 unsigned long pfn, unsigned long size, pgprot_t prot)
1755 if (addr != (pfn << PAGE_SHIFT))
1758 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1761 EXPORT_SYMBOL(remap_pfn_range);
1763 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1765 unsigned long pfn = start >> PAGE_SHIFT;
1766 unsigned long vm_len = vma->vm_end - vma->vm_start;
1768 pfn += vma->vm_pgoff;
1769 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1771 EXPORT_SYMBOL(vm_iomap_memory);
1773 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1774 unsigned long pgoff)
1776 unsigned int size = vma->vm_end - vma->vm_start;
1778 if (!(vma->vm_flags & VM_USERMAP))
1781 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1782 vma->vm_end = vma->vm_start + size;
1786 EXPORT_SYMBOL(remap_vmalloc_range);
1788 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1789 unsigned long len, unsigned long pgoff, unsigned long flags)
1794 void unmap_mapping_range(struct address_space *mapping,
1795 loff_t const holebegin, loff_t const holelen,
1799 EXPORT_SYMBOL(unmap_mapping_range);
1801 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1806 EXPORT_SYMBOL(filemap_fault);
1808 void filemap_map_pages(struct fault_env *fe,
1809 pgoff_t start_pgoff, pgoff_t end_pgoff)
1813 EXPORT_SYMBOL(filemap_map_pages);
1815 int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1816 unsigned long addr, void *buf, int len, unsigned int gup_flags)
1818 struct vm_area_struct *vma;
1819 int write = gup_flags & FOLL_WRITE;
1821 down_read(&mm->mmap_sem);
1823 /* the access must start within one of the target process's mappings */
1824 vma = find_vma(mm, addr);
1826 /* don't overrun this mapping */
1827 if (addr + len >= vma->vm_end)
1828 len = vma->vm_end - addr;
1830 /* only read or write mappings where it is permitted */
1831 if (write && vma->vm_flags & VM_MAYWRITE)
1832 copy_to_user_page(vma, NULL, addr,
1833 (void *) addr, buf, len);
1834 else if (!write && vma->vm_flags & VM_MAYREAD)
1835 copy_from_user_page(vma, NULL, addr,
1836 buf, (void *) addr, len);
1843 up_read(&mm->mmap_sem);
1849 * @access_remote_vm - access another process' address space
1850 * @mm: the mm_struct of the target address space
1851 * @addr: start address to access
1852 * @buf: source or destination buffer
1853 * @len: number of bytes to transfer
1854 * @gup_flags: flags modifying lookup behaviour
1856 * The caller must hold a reference on @mm.
1858 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1859 void *buf, int len, unsigned int gup_flags)
1861 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
1865 * Access another process' address space.
1866 * - source/target buffer must be kernel space
1868 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1869 unsigned int gup_flags)
1871 struct mm_struct *mm;
1873 if (addr + len < addr)
1876 mm = get_task_mm(tsk);
1880 len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
1887 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1888 * @inode: The inode to check
1889 * @size: The current filesize of the inode
1890 * @newsize: The proposed filesize of the inode
1892 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1893 * make sure that that any outstanding VMAs aren't broken and then shrink the
1894 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1895 * automatically grant mappings that are too large.
1897 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1900 struct vm_area_struct *vma;
1901 struct vm_region *region;
1903 size_t r_size, r_top;
1905 low = newsize >> PAGE_SHIFT;
1906 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1908 down_write(&nommu_region_sem);
1909 i_mmap_lock_read(inode->i_mapping);
1911 /* search for VMAs that fall within the dead zone */
1912 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1913 /* found one - only interested if it's shared out of the page
1915 if (vma->vm_flags & VM_SHARED) {
1916 i_mmap_unlock_read(inode->i_mapping);
1917 up_write(&nommu_region_sem);
1918 return -ETXTBSY; /* not quite true, but near enough */
1922 /* reduce any regions that overlap the dead zone - if in existence,
1923 * these will be pointed to by VMAs that don't overlap the dead zone
1925 * we don't check for any regions that start beyond the EOF as there
1928 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1929 if (!(vma->vm_flags & VM_SHARED))
1932 region = vma->vm_region;
1933 r_size = region->vm_top - region->vm_start;
1934 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1936 if (r_top > newsize) {
1937 region->vm_top -= r_top - newsize;
1938 if (region->vm_end > region->vm_top)
1939 region->vm_end = region->vm_top;
1943 i_mmap_unlock_read(inode->i_mapping);
1944 up_write(&nommu_region_sem);
1949 * Initialise sysctl_user_reserve_kbytes.
1951 * This is intended to prevent a user from starting a single memory hogging
1952 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1955 * The default value is min(3% of free memory, 128MB)
1956 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1958 static int __meminit init_user_reserve(void)
1960 unsigned long free_kbytes;
1962 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1964 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1967 subsys_initcall(init_user_reserve);
1970 * Initialise sysctl_admin_reserve_kbytes.
1972 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1973 * to log in and kill a memory hogging process.
1975 * Systems with more than 256MB will reserve 8MB, enough to recover
1976 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1977 * only reserve 3% of free pages by default.
1979 static int __meminit init_admin_reserve(void)
1981 unsigned long free_kbytes;
1983 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1985 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1988 subsys_initcall(init_admin_reserve);