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);
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 int write, int force, struct page **pages,
164 struct vm_area_struct **vmas)
173 return __get_user_pages(current, current->mm, start, nr_pages, flags,
176 EXPORT_SYMBOL(get_user_pages);
178 long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
179 int write, int force, struct page **pages,
182 return get_user_pages(start, nr_pages, write, force, pages, NULL);
184 EXPORT_SYMBOL(get_user_pages_locked);
186 long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
187 unsigned long start, unsigned long nr_pages,
188 struct page **pages, unsigned int gup_flags)
191 down_read(&mm->mmap_sem);
192 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
194 up_read(&mm->mmap_sem);
197 EXPORT_SYMBOL(__get_user_pages_unlocked);
199 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
200 struct page **pages, unsigned int gup_flags)
202 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
205 EXPORT_SYMBOL(get_user_pages_unlocked);
208 * follow_pfn - look up PFN at a user virtual address
209 * @vma: memory mapping
210 * @address: user virtual address
211 * @pfn: location to store found PFN
213 * Only IO mappings and raw PFN mappings are allowed.
215 * Returns zero and the pfn at @pfn on success, -ve otherwise.
217 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
220 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
223 *pfn = address >> PAGE_SHIFT;
226 EXPORT_SYMBOL(follow_pfn);
228 LIST_HEAD(vmap_area_list);
230 void vfree(const void *addr)
234 EXPORT_SYMBOL(vfree);
236 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
239 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
240 * returns only a logical address.
242 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
244 EXPORT_SYMBOL(__vmalloc);
246 void *vmalloc_user(unsigned long size)
250 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
253 struct vm_area_struct *vma;
255 down_write(¤t->mm->mmap_sem);
256 vma = find_vma(current->mm, (unsigned long)ret);
258 vma->vm_flags |= VM_USERMAP;
259 up_write(¤t->mm->mmap_sem);
264 EXPORT_SYMBOL(vmalloc_user);
266 struct page *vmalloc_to_page(const void *addr)
268 return virt_to_page(addr);
270 EXPORT_SYMBOL(vmalloc_to_page);
272 unsigned long vmalloc_to_pfn(const void *addr)
274 return page_to_pfn(virt_to_page(addr));
276 EXPORT_SYMBOL(vmalloc_to_pfn);
278 long vread(char *buf, char *addr, unsigned long count)
280 /* Don't allow overflow */
281 if ((unsigned long) buf + count < count)
282 count = -(unsigned long) buf;
284 memcpy(buf, addr, count);
288 long vwrite(char *buf, char *addr, unsigned long count)
290 /* Don't allow overflow */
291 if ((unsigned long) addr + count < count)
292 count = -(unsigned long) addr;
294 memcpy(addr, buf, count);
299 * vmalloc - allocate virtually contiguous memory
301 * @size: allocation size
303 * Allocate enough pages to cover @size from the page level
304 * allocator and map them into contiguous kernel virtual space.
306 * For tight control over page level allocator and protection flags
307 * use __vmalloc() instead.
309 void *vmalloc(unsigned long size)
311 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
313 EXPORT_SYMBOL(vmalloc);
316 * vzalloc - allocate virtually contiguous memory with zero fill
318 * @size: allocation size
320 * Allocate enough pages to cover @size from the page level
321 * allocator and map them into contiguous kernel virtual space.
322 * The memory allocated is set to zero.
324 * For tight control over page level allocator and protection flags
325 * use __vmalloc() instead.
327 void *vzalloc(unsigned long size)
329 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
332 EXPORT_SYMBOL(vzalloc);
335 * vmalloc_node - allocate memory on a specific node
336 * @size: allocation size
339 * Allocate enough pages to cover @size from the page level
340 * allocator and map them into contiguous kernel virtual space.
342 * For tight control over page level allocator and protection flags
343 * use __vmalloc() instead.
345 void *vmalloc_node(unsigned long size, int node)
347 return vmalloc(size);
349 EXPORT_SYMBOL(vmalloc_node);
352 * vzalloc_node - allocate memory on a specific node with zero fill
353 * @size: allocation size
356 * Allocate enough pages to cover @size from the page level
357 * allocator and map them into contiguous kernel virtual space.
358 * The memory allocated is set to zero.
360 * For tight control over page level allocator and protection flags
361 * use __vmalloc() instead.
363 void *vzalloc_node(unsigned long size, int node)
365 return vzalloc(size);
367 EXPORT_SYMBOL(vzalloc_node);
369 #ifndef PAGE_KERNEL_EXEC
370 # define PAGE_KERNEL_EXEC PAGE_KERNEL
374 * vmalloc_exec - allocate virtually contiguous, executable memory
375 * @size: allocation size
377 * Kernel-internal function to allocate enough pages to cover @size
378 * the page level allocator and map them into contiguous and
379 * executable kernel virtual space.
381 * For tight control over page level allocator and protection flags
382 * use __vmalloc() instead.
385 void *vmalloc_exec(unsigned long size)
387 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
391 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
392 * @size: allocation size
394 * Allocate enough 32bit PA addressable pages to cover @size from the
395 * page level allocator and map them into contiguous kernel virtual space.
397 void *vmalloc_32(unsigned long size)
399 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
401 EXPORT_SYMBOL(vmalloc_32);
404 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
405 * @size: allocation size
407 * The resulting memory area is 32bit addressable and zeroed so it can be
408 * mapped to userspace without leaking data.
410 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
411 * remap_vmalloc_range() are permissible.
413 void *vmalloc_32_user(unsigned long size)
416 * We'll have to sort out the ZONE_DMA bits for 64-bit,
417 * but for now this can simply use vmalloc_user() directly.
419 return vmalloc_user(size);
421 EXPORT_SYMBOL(vmalloc_32_user);
423 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
430 void vunmap(const void *addr)
434 EXPORT_SYMBOL(vunmap);
436 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
441 EXPORT_SYMBOL(vm_map_ram);
443 void vm_unmap_ram(const void *mem, unsigned int count)
447 EXPORT_SYMBOL(vm_unmap_ram);
449 void vm_unmap_aliases(void)
452 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
455 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
458 void __weak vmalloc_sync_all(void)
463 * alloc_vm_area - allocate a range of kernel address space
464 * @size: size of the area
466 * Returns: NULL on failure, vm_struct on success
468 * This function reserves a range of kernel address space, and
469 * allocates pagetables to map that range. No actual mappings
470 * are created. If the kernel address space is not shared
471 * between processes, it syncs the pagetable across all
474 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
479 EXPORT_SYMBOL_GPL(alloc_vm_area);
481 void free_vm_area(struct vm_struct *area)
485 EXPORT_SYMBOL_GPL(free_vm_area);
487 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
492 EXPORT_SYMBOL(vm_insert_page);
495 * sys_brk() for the most part doesn't need the global kernel
496 * lock, except when an application is doing something nasty
497 * like trying to un-brk an area that has already been mapped
498 * to a regular file. in this case, the unmapping will need
499 * to invoke file system routines that need the global lock.
501 SYSCALL_DEFINE1(brk, unsigned long, brk)
503 struct mm_struct *mm = current->mm;
505 if (brk < mm->start_brk || brk > mm->context.end_brk)
512 * Always allow shrinking brk
514 if (brk <= mm->brk) {
520 * Ok, looks good - let it rip.
522 flush_icache_range(mm->brk, brk);
523 return mm->brk = brk;
527 * initialise the VMA and region record slabs
529 void __init mmap_init(void)
533 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
535 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
539 * validate the region tree
540 * - the caller must hold the region lock
542 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
543 static noinline void validate_nommu_regions(void)
545 struct vm_region *region, *last;
546 struct rb_node *p, *lastp;
548 lastp = rb_first(&nommu_region_tree);
552 last = rb_entry(lastp, struct vm_region, vm_rb);
553 BUG_ON(last->vm_end <= last->vm_start);
554 BUG_ON(last->vm_top < last->vm_end);
556 while ((p = rb_next(lastp))) {
557 region = rb_entry(p, struct vm_region, vm_rb);
558 last = rb_entry(lastp, struct vm_region, vm_rb);
560 BUG_ON(region->vm_end <= region->vm_start);
561 BUG_ON(region->vm_top < region->vm_end);
562 BUG_ON(region->vm_start < last->vm_top);
568 static void validate_nommu_regions(void)
574 * add a region into the global tree
576 static void add_nommu_region(struct vm_region *region)
578 struct vm_region *pregion;
579 struct rb_node **p, *parent;
581 validate_nommu_regions();
584 p = &nommu_region_tree.rb_node;
587 pregion = rb_entry(parent, struct vm_region, vm_rb);
588 if (region->vm_start < pregion->vm_start)
590 else if (region->vm_start > pregion->vm_start)
592 else if (pregion == region)
598 rb_link_node(®ion->vm_rb, parent, p);
599 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
601 validate_nommu_regions();
605 * delete a region from the global tree
607 static void delete_nommu_region(struct vm_region *region)
609 BUG_ON(!nommu_region_tree.rb_node);
611 validate_nommu_regions();
612 rb_erase(®ion->vm_rb, &nommu_region_tree);
613 validate_nommu_regions();
617 * free a contiguous series of pages
619 static void free_page_series(unsigned long from, unsigned long to)
621 for (; from < to; from += PAGE_SIZE) {
622 struct page *page = virt_to_page(from);
624 atomic_long_dec(&mmap_pages_allocated);
630 * release a reference to a region
631 * - the caller must hold the region semaphore for writing, which this releases
632 * - the region may not have been added to the tree yet, in which case vm_top
633 * will equal vm_start
635 static void __put_nommu_region(struct vm_region *region)
636 __releases(nommu_region_sem)
638 BUG_ON(!nommu_region_tree.rb_node);
640 if (--region->vm_usage == 0) {
641 if (region->vm_top > region->vm_start)
642 delete_nommu_region(region);
643 up_write(&nommu_region_sem);
646 fput(region->vm_file);
648 /* IO memory and memory shared directly out of the pagecache
649 * from ramfs/tmpfs mustn't be released here */
650 if (region->vm_flags & VM_MAPPED_COPY)
651 free_page_series(region->vm_start, region->vm_top);
652 kmem_cache_free(vm_region_jar, region);
654 up_write(&nommu_region_sem);
659 * release a reference to a region
661 static void put_nommu_region(struct vm_region *region)
663 down_write(&nommu_region_sem);
664 __put_nommu_region(region);
668 * update protection on a vma
670 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
673 struct mm_struct *mm = vma->vm_mm;
674 long start = vma->vm_start & PAGE_MASK;
675 while (start < vma->vm_end) {
676 protect_page(mm, start, flags);
679 update_protections(mm);
684 * add a VMA into a process's mm_struct in the appropriate place in the list
685 * and tree and add to the address space's page tree also if not an anonymous
687 * - should be called with mm->mmap_sem held writelocked
689 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
691 struct vm_area_struct *pvma, *prev;
692 struct address_space *mapping;
693 struct rb_node **p, *parent, *rb_prev;
695 BUG_ON(!vma->vm_region);
700 protect_vma(vma, vma->vm_flags);
702 /* add the VMA to the mapping */
704 mapping = vma->vm_file->f_mapping;
706 i_mmap_lock_write(mapping);
707 flush_dcache_mmap_lock(mapping);
708 vma_interval_tree_insert(vma, &mapping->i_mmap);
709 flush_dcache_mmap_unlock(mapping);
710 i_mmap_unlock_write(mapping);
713 /* add the VMA to the tree */
714 parent = rb_prev = NULL;
715 p = &mm->mm_rb.rb_node;
718 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
720 /* sort by: start addr, end addr, VMA struct addr in that order
721 * (the latter is necessary as we may get identical VMAs) */
722 if (vma->vm_start < pvma->vm_start)
724 else if (vma->vm_start > pvma->vm_start) {
727 } else if (vma->vm_end < pvma->vm_end)
729 else if (vma->vm_end > pvma->vm_end) {
732 } else if (vma < pvma)
734 else if (vma > pvma) {
741 rb_link_node(&vma->vm_rb, parent, p);
742 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
744 /* add VMA to the VMA list also */
747 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
749 __vma_link_list(mm, vma, prev, parent);
753 * delete a VMA from its owning mm_struct and address space
755 static void delete_vma_from_mm(struct vm_area_struct *vma)
758 struct address_space *mapping;
759 struct mm_struct *mm = vma->vm_mm;
760 struct task_struct *curr = current;
765 for (i = 0; i < VMACACHE_SIZE; i++) {
766 /* if the vma is cached, invalidate the entire cache */
767 if (curr->vmacache[i] == vma) {
768 vmacache_invalidate(mm);
773 /* remove the VMA from the mapping */
775 mapping = vma->vm_file->f_mapping;
777 i_mmap_lock_write(mapping);
778 flush_dcache_mmap_lock(mapping);
779 vma_interval_tree_remove(vma, &mapping->i_mmap);
780 flush_dcache_mmap_unlock(mapping);
781 i_mmap_unlock_write(mapping);
784 /* remove from the MM's tree and list */
785 rb_erase(&vma->vm_rb, &mm->mm_rb);
788 vma->vm_prev->vm_next = vma->vm_next;
790 mm->mmap = vma->vm_next;
793 vma->vm_next->vm_prev = vma->vm_prev;
797 * destroy a VMA record
799 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
801 if (vma->vm_ops && vma->vm_ops->close)
802 vma->vm_ops->close(vma);
805 put_nommu_region(vma->vm_region);
806 kmem_cache_free(vm_area_cachep, vma);
810 * look up the first VMA in which addr resides, NULL if none
811 * - should be called with mm->mmap_sem at least held readlocked
813 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
815 struct vm_area_struct *vma;
817 /* check the cache first */
818 vma = vmacache_find(mm, addr);
822 /* trawl the list (there may be multiple mappings in which addr
824 for (vma = mm->mmap; vma; vma = vma->vm_next) {
825 if (vma->vm_start > addr)
827 if (vma->vm_end > addr) {
828 vmacache_update(addr, vma);
835 EXPORT_SYMBOL(find_vma);
839 * - we don't extend stack VMAs under NOMMU conditions
841 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
843 return find_vma(mm, addr);
847 * expand a stack to a given address
848 * - not supported under NOMMU conditions
850 int expand_stack(struct vm_area_struct *vma, unsigned long address)
856 * look up the first VMA exactly that exactly matches addr
857 * - should be called with mm->mmap_sem at least held readlocked
859 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
863 struct vm_area_struct *vma;
864 unsigned long end = addr + len;
866 /* check the cache first */
867 vma = vmacache_find_exact(mm, addr, end);
871 /* trawl the list (there may be multiple mappings in which addr
873 for (vma = mm->mmap; vma; vma = vma->vm_next) {
874 if (vma->vm_start < addr)
876 if (vma->vm_start > addr)
878 if (vma->vm_end == end) {
879 vmacache_update(addr, vma);
888 * determine whether a mapping should be permitted and, if so, what sort of
889 * mapping we're capable of supporting
891 static int validate_mmap_request(struct file *file,
897 unsigned long *_capabilities)
899 unsigned long capabilities, rlen;
902 /* do the simple checks first */
903 if (flags & MAP_FIXED)
906 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
907 (flags & MAP_TYPE) != MAP_SHARED)
913 /* Careful about overflows.. */
914 rlen = PAGE_ALIGN(len);
915 if (!rlen || rlen > TASK_SIZE)
918 /* offset overflow? */
919 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
923 /* files must support mmap */
924 if (!file->f_op->mmap)
927 /* work out if what we've got could possibly be shared
928 * - we support chardevs that provide their own "memory"
929 * - we support files/blockdevs that are memory backed
931 if (file->f_op->mmap_capabilities) {
932 capabilities = file->f_op->mmap_capabilities(file);
934 /* no explicit capabilities set, so assume some
936 switch (file_inode(file)->i_mode & S_IFMT) {
939 capabilities = NOMMU_MAP_COPY;
954 /* eliminate any capabilities that we can't support on this
956 if (!file->f_op->get_unmapped_area)
957 capabilities &= ~NOMMU_MAP_DIRECT;
958 if (!(file->f_mode & FMODE_CAN_READ))
959 capabilities &= ~NOMMU_MAP_COPY;
961 /* The file shall have been opened with read permission. */
962 if (!(file->f_mode & FMODE_READ))
965 if (flags & MAP_SHARED) {
966 /* do checks for writing, appending and locking */
967 if ((prot & PROT_WRITE) &&
968 !(file->f_mode & FMODE_WRITE))
971 if (IS_APPEND(file_inode(file)) &&
972 (file->f_mode & FMODE_WRITE))
975 if (locks_verify_locked(file))
978 if (!(capabilities & NOMMU_MAP_DIRECT))
981 /* we mustn't privatise shared mappings */
982 capabilities &= ~NOMMU_MAP_COPY;
984 /* we're going to read the file into private memory we
986 if (!(capabilities & NOMMU_MAP_COPY))
989 /* we don't permit a private writable mapping to be
990 * shared with the backing device */
991 if (prot & PROT_WRITE)
992 capabilities &= ~NOMMU_MAP_DIRECT;
995 if (capabilities & NOMMU_MAP_DIRECT) {
996 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
997 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
998 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
1000 capabilities &= ~NOMMU_MAP_DIRECT;
1001 if (flags & MAP_SHARED) {
1002 pr_warn("MAP_SHARED not completely supported on !MMU\n");
1008 /* handle executable mappings and implied executable
1010 if (path_noexec(&file->f_path)) {
1011 if (prot & PROT_EXEC)
1013 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1014 /* handle implication of PROT_EXEC by PROT_READ */
1015 if (current->personality & READ_IMPLIES_EXEC) {
1016 if (capabilities & NOMMU_MAP_EXEC)
1019 } else if ((prot & PROT_READ) &&
1020 (prot & PROT_EXEC) &&
1021 !(capabilities & NOMMU_MAP_EXEC)
1023 /* backing file is not executable, try to copy */
1024 capabilities &= ~NOMMU_MAP_DIRECT;
1027 /* anonymous mappings are always memory backed and can be
1030 capabilities = NOMMU_MAP_COPY;
1032 /* handle PROT_EXEC implication by PROT_READ */
1033 if ((prot & PROT_READ) &&
1034 (current->personality & READ_IMPLIES_EXEC))
1038 /* allow the security API to have its say */
1039 ret = security_mmap_addr(addr);
1044 *_capabilities = capabilities;
1049 * we've determined that we can make the mapping, now translate what we
1050 * now know into VMA flags
1052 static unsigned long determine_vm_flags(struct file *file,
1054 unsigned long flags,
1055 unsigned long capabilities)
1057 unsigned long vm_flags;
1059 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1060 /* vm_flags |= mm->def_flags; */
1062 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1063 /* attempt to share read-only copies of mapped file chunks */
1064 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1065 if (file && !(prot & PROT_WRITE))
1066 vm_flags |= VM_MAYSHARE;
1068 /* overlay a shareable mapping on the backing device or inode
1069 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1071 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1072 if (flags & MAP_SHARED)
1073 vm_flags |= VM_SHARED;
1076 /* refuse to let anyone share private mappings with this process if
1077 * it's being traced - otherwise breakpoints set in it may interfere
1078 * with another untraced process
1080 if ((flags & MAP_PRIVATE) && current->ptrace)
1081 vm_flags &= ~VM_MAYSHARE;
1087 * set up a shared mapping on a file (the driver or filesystem provides and
1090 static int do_mmap_shared_file(struct vm_area_struct *vma)
1094 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1096 vma->vm_region->vm_top = vma->vm_region->vm_end;
1102 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1103 * opposed to tried but failed) so we can only give a suitable error as
1104 * it's not possible to make a private copy if MAP_SHARED was given */
1109 * set up a private mapping or an anonymous shared mapping
1111 static int do_mmap_private(struct vm_area_struct *vma,
1112 struct vm_region *region,
1114 unsigned long capabilities)
1116 unsigned long total, point;
1120 /* invoke the file's mapping function so that it can keep track of
1121 * shared mappings on devices or memory
1122 * - VM_MAYSHARE will be set if it may attempt to share
1124 if (capabilities & NOMMU_MAP_DIRECT) {
1125 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1127 /* shouldn't return success if we're not sharing */
1128 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1129 vma->vm_region->vm_top = vma->vm_region->vm_end;
1135 /* getting an ENOSYS error indicates that direct mmap isn't
1136 * possible (as opposed to tried but failed) so we'll try to
1137 * make a private copy of the data and map that instead */
1141 /* allocate some memory to hold the mapping
1142 * - note that this may not return a page-aligned address if the object
1143 * we're allocating is smaller than a page
1145 order = get_order(len);
1147 point = len >> PAGE_SHIFT;
1149 /* we don't want to allocate a power-of-2 sized page set */
1150 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1153 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1157 atomic_long_add(total, &mmap_pages_allocated);
1159 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1160 region->vm_start = (unsigned long) base;
1161 region->vm_end = region->vm_start + len;
1162 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1164 vma->vm_start = region->vm_start;
1165 vma->vm_end = region->vm_start + len;
1168 /* read the contents of a file into the copy */
1169 mm_segment_t old_fs;
1172 fpos = vma->vm_pgoff;
1173 fpos <<= PAGE_SHIFT;
1177 ret = __vfs_read(vma->vm_file, base, len, &fpos);
1183 /* clear the last little bit */
1185 memset(base + ret, 0, len - ret);
1192 free_page_series(region->vm_start, region->vm_top);
1193 region->vm_start = vma->vm_start = 0;
1194 region->vm_end = vma->vm_end = 0;
1199 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1200 len, current->pid, current->comm);
1206 * handle mapping creation for uClinux
1208 unsigned long do_mmap(struct file *file,
1212 unsigned long flags,
1213 vm_flags_t vm_flags,
1214 unsigned long pgoff,
1215 unsigned long *populate)
1217 struct vm_area_struct *vma;
1218 struct vm_region *region;
1220 unsigned long capabilities, result;
1225 /* decide whether we should attempt the mapping, and if so what sort of
1227 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1232 /* we ignore the address hint */
1234 len = PAGE_ALIGN(len);
1236 /* we've determined that we can make the mapping, now translate what we
1237 * now know into VMA flags */
1238 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1240 /* we're going to need to record the mapping */
1241 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1243 goto error_getting_region;
1245 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1247 goto error_getting_vma;
1249 region->vm_usage = 1;
1250 region->vm_flags = vm_flags;
1251 region->vm_pgoff = pgoff;
1253 INIT_LIST_HEAD(&vma->anon_vma_chain);
1254 vma->vm_flags = vm_flags;
1255 vma->vm_pgoff = pgoff;
1258 region->vm_file = get_file(file);
1259 vma->vm_file = get_file(file);
1262 down_write(&nommu_region_sem);
1264 /* if we want to share, we need to check for regions created by other
1265 * mmap() calls that overlap with our proposed mapping
1266 * - we can only share with a superset match on most regular files
1267 * - shared mappings on character devices and memory backed files are
1268 * permitted to overlap inexactly as far as we are concerned for in
1269 * these cases, sharing is handled in the driver or filesystem rather
1272 if (vm_flags & VM_MAYSHARE) {
1273 struct vm_region *pregion;
1274 unsigned long pglen, rpglen, pgend, rpgend, start;
1276 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1277 pgend = pgoff + pglen;
1279 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1280 pregion = rb_entry(rb, struct vm_region, vm_rb);
1282 if (!(pregion->vm_flags & VM_MAYSHARE))
1285 /* search for overlapping mappings on the same file */
1286 if (file_inode(pregion->vm_file) !=
1290 if (pregion->vm_pgoff >= pgend)
1293 rpglen = pregion->vm_end - pregion->vm_start;
1294 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1295 rpgend = pregion->vm_pgoff + rpglen;
1296 if (pgoff >= rpgend)
1299 /* handle inexactly overlapping matches between
1301 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1302 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1303 /* new mapping is not a subset of the region */
1304 if (!(capabilities & NOMMU_MAP_DIRECT))
1305 goto sharing_violation;
1309 /* we've found a region we can share */
1310 pregion->vm_usage++;
1311 vma->vm_region = pregion;
1312 start = pregion->vm_start;
1313 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1314 vma->vm_start = start;
1315 vma->vm_end = start + len;
1317 if (pregion->vm_flags & VM_MAPPED_COPY)
1318 vma->vm_flags |= VM_MAPPED_COPY;
1320 ret = do_mmap_shared_file(vma);
1322 vma->vm_region = NULL;
1325 pregion->vm_usage--;
1327 goto error_just_free;
1330 fput(region->vm_file);
1331 kmem_cache_free(vm_region_jar, region);
1337 /* obtain the address at which to make a shared mapping
1338 * - this is the hook for quasi-memory character devices to
1339 * tell us the location of a shared mapping
1341 if (capabilities & NOMMU_MAP_DIRECT) {
1342 addr = file->f_op->get_unmapped_area(file, addr, len,
1344 if (IS_ERR_VALUE(addr)) {
1347 goto error_just_free;
1349 /* the driver refused to tell us where to site
1350 * the mapping so we'll have to attempt to copy
1353 if (!(capabilities & NOMMU_MAP_COPY))
1354 goto error_just_free;
1356 capabilities &= ~NOMMU_MAP_DIRECT;
1358 vma->vm_start = region->vm_start = addr;
1359 vma->vm_end = region->vm_end = addr + len;
1364 vma->vm_region = region;
1366 /* set up the mapping
1367 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1369 if (file && vma->vm_flags & VM_SHARED)
1370 ret = do_mmap_shared_file(vma);
1372 ret = do_mmap_private(vma, region, len, capabilities);
1374 goto error_just_free;
1375 add_nommu_region(region);
1377 /* clear anonymous mappings that don't ask for uninitialized data */
1378 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1379 memset((void *)region->vm_start, 0,
1380 region->vm_end - region->vm_start);
1382 /* okay... we have a mapping; now we have to register it */
1383 result = vma->vm_start;
1385 current->mm->total_vm += len >> PAGE_SHIFT;
1388 add_vma_to_mm(current->mm, vma);
1390 /* we flush the region from the icache only when the first executable
1391 * mapping of it is made */
1392 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1393 flush_icache_range(region->vm_start, region->vm_end);
1394 region->vm_icache_flushed = true;
1397 up_write(&nommu_region_sem);
1402 up_write(&nommu_region_sem);
1404 if (region->vm_file)
1405 fput(region->vm_file);
1406 kmem_cache_free(vm_region_jar, region);
1409 kmem_cache_free(vm_area_cachep, vma);
1413 up_write(&nommu_region_sem);
1414 pr_warn("Attempt to share mismatched mappings\n");
1419 kmem_cache_free(vm_region_jar, region);
1420 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1425 error_getting_region:
1426 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1432 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1433 unsigned long, prot, unsigned long, flags,
1434 unsigned long, fd, unsigned long, pgoff)
1436 struct file *file = NULL;
1437 unsigned long retval = -EBADF;
1439 audit_mmap_fd(fd, flags);
1440 if (!(flags & MAP_ANONYMOUS)) {
1446 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1448 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1456 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1457 struct mmap_arg_struct {
1461 unsigned long flags;
1463 unsigned long offset;
1466 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1468 struct mmap_arg_struct a;
1470 if (copy_from_user(&a, arg, sizeof(a)))
1472 if (offset_in_page(a.offset))
1475 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1476 a.offset >> PAGE_SHIFT);
1478 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1481 * split a vma into two pieces at address 'addr', a new vma is allocated either
1482 * for the first part or the tail.
1484 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1485 unsigned long addr, int new_below)
1487 struct vm_area_struct *new;
1488 struct vm_region *region;
1489 unsigned long npages;
1491 /* we're only permitted to split anonymous regions (these should have
1492 * only a single usage on the region) */
1496 if (mm->map_count >= sysctl_max_map_count)
1499 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1503 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1505 kmem_cache_free(vm_region_jar, region);
1509 /* most fields are the same, copy all, and then fixup */
1511 *region = *vma->vm_region;
1512 new->vm_region = region;
1514 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1517 region->vm_top = region->vm_end = new->vm_end = addr;
1519 region->vm_start = new->vm_start = addr;
1520 region->vm_pgoff = new->vm_pgoff += npages;
1523 if (new->vm_ops && new->vm_ops->open)
1524 new->vm_ops->open(new);
1526 delete_vma_from_mm(vma);
1527 down_write(&nommu_region_sem);
1528 delete_nommu_region(vma->vm_region);
1530 vma->vm_region->vm_start = vma->vm_start = addr;
1531 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1533 vma->vm_region->vm_end = vma->vm_end = addr;
1534 vma->vm_region->vm_top = addr;
1536 add_nommu_region(vma->vm_region);
1537 add_nommu_region(new->vm_region);
1538 up_write(&nommu_region_sem);
1539 add_vma_to_mm(mm, vma);
1540 add_vma_to_mm(mm, new);
1545 * shrink a VMA by removing the specified chunk from either the beginning or
1548 static int shrink_vma(struct mm_struct *mm,
1549 struct vm_area_struct *vma,
1550 unsigned long from, unsigned long to)
1552 struct vm_region *region;
1554 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1556 delete_vma_from_mm(vma);
1557 if (from > vma->vm_start)
1561 add_vma_to_mm(mm, vma);
1563 /* cut the backing region down to size */
1564 region = vma->vm_region;
1565 BUG_ON(region->vm_usage != 1);
1567 down_write(&nommu_region_sem);
1568 delete_nommu_region(region);
1569 if (from > region->vm_start) {
1570 to = region->vm_top;
1571 region->vm_top = region->vm_end = from;
1573 region->vm_start = to;
1575 add_nommu_region(region);
1576 up_write(&nommu_region_sem);
1578 free_page_series(from, to);
1584 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1585 * VMA, though it need not cover the whole VMA
1587 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1589 struct vm_area_struct *vma;
1593 len = PAGE_ALIGN(len);
1599 /* find the first potentially overlapping VMA */
1600 vma = find_vma(mm, start);
1604 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1605 current->pid, current->comm,
1606 start, start + len - 1);
1612 /* we're allowed to split an anonymous VMA but not a file-backed one */
1615 if (start > vma->vm_start)
1617 if (end == vma->vm_end)
1618 goto erase_whole_vma;
1623 /* the chunk must be a subset of the VMA found */
1624 if (start == vma->vm_start && end == vma->vm_end)
1625 goto erase_whole_vma;
1626 if (start < vma->vm_start || end > vma->vm_end)
1628 if (offset_in_page(start))
1630 if (end != vma->vm_end && offset_in_page(end))
1632 if (start != vma->vm_start && end != vma->vm_end) {
1633 ret = split_vma(mm, vma, start, 1);
1637 return shrink_vma(mm, vma, start, end);
1641 delete_vma_from_mm(vma);
1642 delete_vma(mm, vma);
1645 EXPORT_SYMBOL(do_munmap);
1647 int vm_munmap(unsigned long addr, size_t len)
1649 struct mm_struct *mm = current->mm;
1652 down_write(&mm->mmap_sem);
1653 ret = do_munmap(mm, addr, len);
1654 up_write(&mm->mmap_sem);
1657 EXPORT_SYMBOL(vm_munmap);
1659 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1661 return vm_munmap(addr, len);
1665 * release all the mappings made in a process's VM space
1667 void exit_mmap(struct mm_struct *mm)
1669 struct vm_area_struct *vma;
1676 while ((vma = mm->mmap)) {
1677 mm->mmap = vma->vm_next;
1678 delete_vma_from_mm(vma);
1679 delete_vma(mm, vma);
1684 int vm_brk(unsigned long addr, unsigned long len)
1690 * expand (or shrink) an existing mapping, potentially moving it at the same
1691 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1693 * under NOMMU conditions, we only permit changing a mapping's size, and only
1694 * as long as it stays within the region allocated by do_mmap_private() and the
1695 * block is not shareable
1697 * MREMAP_FIXED is not supported under NOMMU conditions
1699 static unsigned long do_mremap(unsigned long addr,
1700 unsigned long old_len, unsigned long new_len,
1701 unsigned long flags, unsigned long new_addr)
1703 struct vm_area_struct *vma;
1705 /* insanity checks first */
1706 old_len = PAGE_ALIGN(old_len);
1707 new_len = PAGE_ALIGN(new_len);
1708 if (old_len == 0 || new_len == 0)
1709 return (unsigned long) -EINVAL;
1711 if (offset_in_page(addr))
1714 if (flags & MREMAP_FIXED && new_addr != addr)
1715 return (unsigned long) -EINVAL;
1717 vma = find_vma_exact(current->mm, addr, old_len);
1719 return (unsigned long) -EINVAL;
1721 if (vma->vm_end != vma->vm_start + old_len)
1722 return (unsigned long) -EFAULT;
1724 if (vma->vm_flags & VM_MAYSHARE)
1725 return (unsigned long) -EPERM;
1727 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1728 return (unsigned long) -ENOMEM;
1730 /* all checks complete - do it */
1731 vma->vm_end = vma->vm_start + new_len;
1732 return vma->vm_start;
1735 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1736 unsigned long, new_len, unsigned long, flags,
1737 unsigned long, new_addr)
1741 down_write(¤t->mm->mmap_sem);
1742 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1743 up_write(¤t->mm->mmap_sem);
1747 struct page *follow_page_mask(struct vm_area_struct *vma,
1748 unsigned long address, unsigned int flags,
1749 unsigned int *page_mask)
1755 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1756 unsigned long pfn, unsigned long size, pgprot_t prot)
1758 if (addr != (pfn << PAGE_SHIFT))
1761 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1764 EXPORT_SYMBOL(remap_pfn_range);
1766 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1768 unsigned long pfn = start >> PAGE_SHIFT;
1769 unsigned long vm_len = vma->vm_end - vma->vm_start;
1771 pfn += vma->vm_pgoff;
1772 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1774 EXPORT_SYMBOL(vm_iomap_memory);
1776 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1777 unsigned long pgoff)
1779 unsigned int size = vma->vm_end - vma->vm_start;
1781 if (!(vma->vm_flags & VM_USERMAP))
1784 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1785 vma->vm_end = vma->vm_start + size;
1789 EXPORT_SYMBOL(remap_vmalloc_range);
1791 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1792 unsigned long len, unsigned long pgoff, unsigned long flags)
1797 void unmap_mapping_range(struct address_space *mapping,
1798 loff_t const holebegin, loff_t const holelen,
1802 EXPORT_SYMBOL(unmap_mapping_range);
1804 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1809 EXPORT_SYMBOL(filemap_fault);
1811 void filemap_map_pages(struct fault_env *fe,
1812 pgoff_t start_pgoff, pgoff_t end_pgoff)
1816 EXPORT_SYMBOL(filemap_map_pages);
1818 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1819 unsigned long addr, void *buf, int len, int write)
1821 struct vm_area_struct *vma;
1823 down_read(&mm->mmap_sem);
1825 /* the access must start within one of the target process's mappings */
1826 vma = find_vma(mm, addr);
1828 /* don't overrun this mapping */
1829 if (addr + len >= vma->vm_end)
1830 len = vma->vm_end - addr;
1832 /* only read or write mappings where it is permitted */
1833 if (write && vma->vm_flags & VM_MAYWRITE)
1834 copy_to_user_page(vma, NULL, addr,
1835 (void *) addr, buf, len);
1836 else if (!write && vma->vm_flags & VM_MAYREAD)
1837 copy_from_user_page(vma, NULL, addr,
1838 buf, (void *) addr, len);
1845 up_read(&mm->mmap_sem);
1851 * @access_remote_vm - access another process' address space
1852 * @mm: the mm_struct of the target address space
1853 * @addr: start address to access
1854 * @buf: source or destination buffer
1855 * @len: number of bytes to transfer
1856 * @write: whether the access is a write
1858 * The caller must hold a reference on @mm.
1860 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1861 void *buf, int len, int write)
1863 return __access_remote_vm(NULL, mm, addr, buf, len, write);
1867 * Access another process' address space.
1868 * - source/target buffer must be kernel space
1870 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1872 struct mm_struct *mm;
1874 if (addr + len < addr)
1877 mm = get_task_mm(tsk);
1881 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
1888 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1889 * @inode: The inode to check
1890 * @size: The current filesize of the inode
1891 * @newsize: The proposed filesize of the inode
1893 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1894 * make sure that that any outstanding VMAs aren't broken and then shrink the
1895 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1896 * automatically grant mappings that are too large.
1898 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1901 struct vm_area_struct *vma;
1902 struct vm_region *region;
1904 size_t r_size, r_top;
1906 low = newsize >> PAGE_SHIFT;
1907 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1909 down_write(&nommu_region_sem);
1910 i_mmap_lock_read(inode->i_mapping);
1912 /* search for VMAs that fall within the dead zone */
1913 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1914 /* found one - only interested if it's shared out of the page
1916 if (vma->vm_flags & VM_SHARED) {
1917 i_mmap_unlock_read(inode->i_mapping);
1918 up_write(&nommu_region_sem);
1919 return -ETXTBSY; /* not quite true, but near enough */
1923 /* reduce any regions that overlap the dead zone - if in existence,
1924 * these will be pointed to by VMAs that don't overlap the dead zone
1926 * we don't check for any regions that start beyond the EOF as there
1929 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1930 if (!(vma->vm_flags & VM_SHARED))
1933 region = vma->vm_region;
1934 r_size = region->vm_top - region->vm_start;
1935 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1937 if (r_top > newsize) {
1938 region->vm_top -= r_top - newsize;
1939 if (region->vm_end > region->vm_top)
1940 region->vm_end = region->vm_top;
1944 i_mmap_unlock_read(inode->i_mapping);
1945 up_write(&nommu_region_sem);
1950 * Initialise sysctl_user_reserve_kbytes.
1952 * This is intended to prevent a user from starting a single memory hogging
1953 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1956 * The default value is min(3% of free memory, 128MB)
1957 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1959 static int __meminit init_user_reserve(void)
1961 unsigned long free_kbytes;
1963 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1965 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1968 subsys_initcall(init_user_reserve);
1971 * Initialise sysctl_admin_reserve_kbytes.
1973 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1974 * to log in and kill a memory hogging process.
1976 * Systems with more than 256MB will reserve 8MB, enough to recover
1977 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1978 * only reserve 3% of free pages by default.
1980 static int __meminit init_admin_reserve(void)
1982 unsigned long free_kbytes;
1984 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1986 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1989 subsys_initcall(init_admin_reserve);