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 #include <linux/export.h>
18 #include <linux/mman.h>
19 #include <linux/swap.h>
20 #include <linux/file.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/blkdev.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mount.h>
28 #include <linux/personality.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
31 #include <linux/audit.h>
33 #include <asm/uaccess.h>
35 #include <asm/tlbflush.h>
36 #include <asm/mmu_context.h>
40 #define kenter(FMT, ...) \
41 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
42 #define kleave(FMT, ...) \
43 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
44 #define kdebug(FMT, ...) \
45 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
47 #define kenter(FMT, ...) \
48 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
49 #define kleave(FMT, ...) \
50 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
51 #define kdebug(FMT, ...) \
52 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
57 unsigned long max_mapnr;
58 unsigned long num_physpages;
59 unsigned long highest_memmap_pfn;
60 struct percpu_counter vm_committed_as;
61 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
62 int sysctl_overcommit_ratio = 50; /* default is 50% */
63 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
64 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
65 int heap_stack_gap = 0;
67 atomic_long_t mmap_pages_allocated;
69 EXPORT_SYMBOL(mem_map);
70 EXPORT_SYMBOL(num_physpages);
72 /* list of mapped, potentially shareable regions */
73 static struct kmem_cache *vm_region_jar;
74 struct rb_root nommu_region_tree = RB_ROOT;
75 DECLARE_RWSEM(nommu_region_sem);
77 const struct vm_operations_struct generic_file_vm_ops = {
81 * Return the total memory allocated for this pointer, not
82 * just what the caller asked for.
84 * Doesn't have to be accurate, i.e. may have races.
86 unsigned int kobjsize(const void *objp)
91 * If the object we have should not have ksize performed on it,
94 if (!objp || !virt_addr_valid(objp))
97 page = virt_to_head_page(objp);
100 * If the allocator sets PageSlab, we know the pointer came from
107 * If it's not a compound page, see if we have a matching VMA
108 * region. This test is intentionally done in reverse order,
109 * so if there's no VMA, we still fall through and hand back
110 * PAGE_SIZE for 0-order pages.
112 if (!PageCompound(page)) {
113 struct vm_area_struct *vma;
115 vma = find_vma(current->mm, (unsigned long)objp);
117 return vma->vm_end - vma->vm_start;
121 * The ksize() function is only guaranteed to work for pointers
122 * returned by kmalloc(). So handle arbitrary pointers here.
124 return PAGE_SIZE << compound_order(page);
127 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
128 unsigned long start, int nr_pages, unsigned int foll_flags,
129 struct page **pages, struct vm_area_struct **vmas,
132 struct vm_area_struct *vma;
133 unsigned long vm_flags;
136 /* calculate required read or write permissions.
137 * If FOLL_FORCE is set, we only require the "MAY" flags.
139 vm_flags = (foll_flags & FOLL_WRITE) ?
140 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
141 vm_flags &= (foll_flags & FOLL_FORCE) ?
142 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
144 for (i = 0; i < nr_pages; i++) {
145 vma = find_vma(mm, start);
147 goto finish_or_fault;
149 /* protect what we can, including chardevs */
150 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
151 !(vm_flags & vma->vm_flags))
152 goto finish_or_fault;
155 pages[i] = virt_to_page(start);
157 page_cache_get(pages[i]);
161 start = (start + PAGE_SIZE) & PAGE_MASK;
167 return i ? : -EFAULT;
171 * get a list of pages in an address range belonging to the specified process
172 * and indicate the VMA that covers each page
173 * - this is potentially dodgy as we may end incrementing the page count of a
174 * slab page or a secondary page from a compound page
175 * - don't permit access to VMAs that don't support it, such as I/O mappings
177 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
178 unsigned long start, int nr_pages, int write, int force,
179 struct page **pages, struct vm_area_struct **vmas)
188 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
191 EXPORT_SYMBOL(get_user_pages);
194 * follow_pfn - look up PFN at a user virtual address
195 * @vma: memory mapping
196 * @address: user virtual address
197 * @pfn: location to store found PFN
199 * Only IO mappings and raw PFN mappings are allowed.
201 * Returns zero and the pfn at @pfn on success, -ve otherwise.
203 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
206 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
209 *pfn = address >> PAGE_SHIFT;
212 EXPORT_SYMBOL(follow_pfn);
214 DEFINE_RWLOCK(vmlist_lock);
215 struct vm_struct *vmlist;
217 void vfree(const void *addr)
221 EXPORT_SYMBOL(vfree);
223 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
226 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
227 * returns only a logical address.
229 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
231 EXPORT_SYMBOL(__vmalloc);
233 void *vmalloc_user(unsigned long size)
237 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
240 struct vm_area_struct *vma;
242 down_write(¤t->mm->mmap_sem);
243 vma = find_vma(current->mm, (unsigned long)ret);
245 vma->vm_flags |= VM_USERMAP;
246 up_write(¤t->mm->mmap_sem);
251 EXPORT_SYMBOL(vmalloc_user);
253 struct page *vmalloc_to_page(const void *addr)
255 return virt_to_page(addr);
257 EXPORT_SYMBOL(vmalloc_to_page);
259 unsigned long vmalloc_to_pfn(const void *addr)
261 return page_to_pfn(virt_to_page(addr));
263 EXPORT_SYMBOL(vmalloc_to_pfn);
265 long vread(char *buf, char *addr, unsigned long count)
267 memcpy(buf, addr, count);
271 long vwrite(char *buf, char *addr, unsigned long count)
273 /* Don't allow overflow */
274 if ((unsigned long) addr + count < count)
275 count = -(unsigned long) addr;
277 memcpy(addr, buf, count);
282 * vmalloc - allocate virtually continguos memory
284 * @size: allocation size
286 * Allocate enough pages to cover @size from the page level
287 * allocator and map them into continguos kernel virtual space.
289 * For tight control over page level allocator and protection flags
290 * use __vmalloc() instead.
292 void *vmalloc(unsigned long size)
294 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
296 EXPORT_SYMBOL(vmalloc);
299 * vzalloc - allocate virtually continguos memory with zero fill
301 * @size: allocation size
303 * Allocate enough pages to cover @size from the page level
304 * allocator and map them into continguos kernel virtual space.
305 * The memory allocated is set to zero.
307 * For tight control over page level allocator and protection flags
308 * use __vmalloc() instead.
310 void *vzalloc(unsigned long size)
312 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
315 EXPORT_SYMBOL(vzalloc);
318 * vmalloc_node - allocate memory on a specific node
319 * @size: allocation size
322 * Allocate enough pages to cover @size from the page level
323 * allocator and map them into contiguous kernel virtual space.
325 * For tight control over page level allocator and protection flags
326 * use __vmalloc() instead.
328 void *vmalloc_node(unsigned long size, int node)
330 return vmalloc(size);
332 EXPORT_SYMBOL(vmalloc_node);
335 * vzalloc_node - allocate memory on a specific node with zero fill
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.
341 * The memory allocated is set to zero.
343 * For tight control over page level allocator and protection flags
344 * use __vmalloc() instead.
346 void *vzalloc_node(unsigned long size, int node)
348 return vzalloc(size);
350 EXPORT_SYMBOL(vzalloc_node);
352 #ifndef PAGE_KERNEL_EXEC
353 # define PAGE_KERNEL_EXEC PAGE_KERNEL
357 * vmalloc_exec - allocate virtually contiguous, executable memory
358 * @size: allocation size
360 * Kernel-internal function to allocate enough pages to cover @size
361 * the page level allocator and map them into contiguous and
362 * executable kernel virtual space.
364 * For tight control over page level allocator and protection flags
365 * use __vmalloc() instead.
368 void *vmalloc_exec(unsigned long size)
370 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
374 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
375 * @size: allocation size
377 * Allocate enough 32bit PA addressable pages to cover @size from the
378 * page level allocator and map them into continguos kernel virtual space.
380 void *vmalloc_32(unsigned long size)
382 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
384 EXPORT_SYMBOL(vmalloc_32);
387 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
388 * @size: allocation size
390 * The resulting memory area is 32bit addressable and zeroed so it can be
391 * mapped to userspace without leaking data.
393 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
394 * remap_vmalloc_range() are permissible.
396 void *vmalloc_32_user(unsigned long size)
399 * We'll have to sort out the ZONE_DMA bits for 64-bit,
400 * but for now this can simply use vmalloc_user() directly.
402 return vmalloc_user(size);
404 EXPORT_SYMBOL(vmalloc_32_user);
406 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
413 void vunmap(const void *addr)
417 EXPORT_SYMBOL(vunmap);
419 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
424 EXPORT_SYMBOL(vm_map_ram);
426 void vm_unmap_ram(const void *mem, unsigned int count)
430 EXPORT_SYMBOL(vm_unmap_ram);
432 void vm_unmap_aliases(void)
435 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
438 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
441 void __attribute__((weak)) vmalloc_sync_all(void)
446 * alloc_vm_area - allocate a range of kernel address space
447 * @size: size of the area
449 * Returns: NULL on failure, vm_struct on success
451 * This function reserves a range of kernel address space, and
452 * allocates pagetables to map that range. No actual mappings
453 * are created. If the kernel address space is not shared
454 * between processes, it syncs the pagetable across all
457 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
462 EXPORT_SYMBOL_GPL(alloc_vm_area);
464 void free_vm_area(struct vm_struct *area)
468 EXPORT_SYMBOL_GPL(free_vm_area);
470 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
475 EXPORT_SYMBOL(vm_insert_page);
478 * sys_brk() for the most part doesn't need the global kernel
479 * lock, except when an application is doing something nasty
480 * like trying to un-brk an area that has already been mapped
481 * to a regular file. in this case, the unmapping will need
482 * to invoke file system routines that need the global lock.
484 SYSCALL_DEFINE1(brk, unsigned long, brk)
486 struct mm_struct *mm = current->mm;
488 if (brk < mm->start_brk || brk > mm->context.end_brk)
495 * Always allow shrinking brk
497 if (brk <= mm->brk) {
503 * Ok, looks good - let it rip.
505 flush_icache_range(mm->brk, brk);
506 return mm->brk = brk;
510 * initialise the VMA and region record slabs
512 void __init mmap_init(void)
516 ret = percpu_counter_init(&vm_committed_as, 0);
518 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
522 * validate the region tree
523 * - the caller must hold the region lock
525 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
526 static noinline void validate_nommu_regions(void)
528 struct vm_region *region, *last;
529 struct rb_node *p, *lastp;
531 lastp = rb_first(&nommu_region_tree);
535 last = rb_entry(lastp, struct vm_region, vm_rb);
536 BUG_ON(unlikely(last->vm_end <= last->vm_start));
537 BUG_ON(unlikely(last->vm_top < last->vm_end));
539 while ((p = rb_next(lastp))) {
540 region = rb_entry(p, struct vm_region, vm_rb);
541 last = rb_entry(lastp, struct vm_region, vm_rb);
543 BUG_ON(unlikely(region->vm_end <= region->vm_start));
544 BUG_ON(unlikely(region->vm_top < region->vm_end));
545 BUG_ON(unlikely(region->vm_start < last->vm_top));
551 static void validate_nommu_regions(void)
557 * add a region into the global tree
559 static void add_nommu_region(struct vm_region *region)
561 struct vm_region *pregion;
562 struct rb_node **p, *parent;
564 validate_nommu_regions();
567 p = &nommu_region_tree.rb_node;
570 pregion = rb_entry(parent, struct vm_region, vm_rb);
571 if (region->vm_start < pregion->vm_start)
573 else if (region->vm_start > pregion->vm_start)
575 else if (pregion == region)
581 rb_link_node(®ion->vm_rb, parent, p);
582 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
584 validate_nommu_regions();
588 * delete a region from the global tree
590 static void delete_nommu_region(struct vm_region *region)
592 BUG_ON(!nommu_region_tree.rb_node);
594 validate_nommu_regions();
595 rb_erase(®ion->vm_rb, &nommu_region_tree);
596 validate_nommu_regions();
600 * free a contiguous series of pages
602 static void free_page_series(unsigned long from, unsigned long to)
604 for (; from < to; from += PAGE_SIZE) {
605 struct page *page = virt_to_page(from);
607 kdebug("- free %lx", from);
608 atomic_long_dec(&mmap_pages_allocated);
609 if (page_count(page) != 1)
610 kdebug("free page %p: refcount not one: %d",
611 page, page_count(page));
617 * release a reference to a region
618 * - the caller must hold the region semaphore for writing, which this releases
619 * - the region may not have been added to the tree yet, in which case vm_top
620 * will equal vm_start
622 static void __put_nommu_region(struct vm_region *region)
623 __releases(nommu_region_sem)
625 kenter("%p{%d}", region, region->vm_usage);
627 BUG_ON(!nommu_region_tree.rb_node);
629 if (--region->vm_usage == 0) {
630 if (region->vm_top > region->vm_start)
631 delete_nommu_region(region);
632 up_write(&nommu_region_sem);
635 fput(region->vm_file);
637 /* IO memory and memory shared directly out of the pagecache
638 * from ramfs/tmpfs mustn't be released here */
639 if (region->vm_flags & VM_MAPPED_COPY) {
640 kdebug("free series");
641 free_page_series(region->vm_start, region->vm_top);
643 kmem_cache_free(vm_region_jar, region);
645 up_write(&nommu_region_sem);
650 * release a reference to a region
652 static void put_nommu_region(struct vm_region *region)
654 down_write(&nommu_region_sem);
655 __put_nommu_region(region);
659 * update protection on a vma
661 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
664 struct mm_struct *mm = vma->vm_mm;
665 long start = vma->vm_start & PAGE_MASK;
666 while (start < vma->vm_end) {
667 protect_page(mm, start, flags);
670 update_protections(mm);
675 * add a VMA into a process's mm_struct in the appropriate place in the list
676 * and tree and add to the address space's page tree also if not an anonymous
678 * - should be called with mm->mmap_sem held writelocked
680 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
682 struct vm_area_struct *pvma, *prev;
683 struct address_space *mapping;
684 struct rb_node **p, *parent, *rb_prev;
688 BUG_ON(!vma->vm_region);
693 protect_vma(vma, vma->vm_flags);
695 /* add the VMA to the mapping */
697 mapping = vma->vm_file->f_mapping;
699 mutex_lock(&mapping->i_mmap_mutex);
700 flush_dcache_mmap_lock(mapping);
701 vma_prio_tree_insert(vma, &mapping->i_mmap);
702 flush_dcache_mmap_unlock(mapping);
703 mutex_unlock(&mapping->i_mmap_mutex);
706 /* add the VMA to the tree */
707 parent = rb_prev = NULL;
708 p = &mm->mm_rb.rb_node;
711 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
713 /* sort by: start addr, end addr, VMA struct addr in that order
714 * (the latter is necessary as we may get identical VMAs) */
715 if (vma->vm_start < pvma->vm_start)
717 else if (vma->vm_start > pvma->vm_start) {
720 } else if (vma->vm_end < pvma->vm_end)
722 else if (vma->vm_end > pvma->vm_end) {
725 } else if (vma < pvma)
727 else if (vma > pvma) {
734 rb_link_node(&vma->vm_rb, parent, p);
735 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
737 /* add VMA to the VMA list also */
740 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
742 __vma_link_list(mm, vma, prev, parent);
746 * delete a VMA from its owning mm_struct and address space
748 static void delete_vma_from_mm(struct vm_area_struct *vma)
750 struct address_space *mapping;
751 struct mm_struct *mm = vma->vm_mm;
758 if (mm->mmap_cache == vma)
759 mm->mmap_cache = NULL;
761 /* remove the VMA from the mapping */
763 mapping = vma->vm_file->f_mapping;
765 mutex_lock(&mapping->i_mmap_mutex);
766 flush_dcache_mmap_lock(mapping);
767 vma_prio_tree_remove(vma, &mapping->i_mmap);
768 flush_dcache_mmap_unlock(mapping);
769 mutex_unlock(&mapping->i_mmap_mutex);
772 /* remove from the MM's tree and list */
773 rb_erase(&vma->vm_rb, &mm->mm_rb);
776 vma->vm_prev->vm_next = vma->vm_next;
778 mm->mmap = vma->vm_next;
781 vma->vm_next->vm_prev = vma->vm_prev;
785 * destroy a VMA record
787 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
790 if (vma->vm_ops && vma->vm_ops->close)
791 vma->vm_ops->close(vma);
794 put_nommu_region(vma->vm_region);
795 kmem_cache_free(vm_area_cachep, vma);
799 * look up the first VMA in which addr resides, NULL if none
800 * - should be called with mm->mmap_sem at least held readlocked
802 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
804 struct vm_area_struct *vma;
806 /* check the cache first */
807 vma = mm->mmap_cache;
808 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
811 /* trawl the list (there may be multiple mappings in which addr
813 for (vma = mm->mmap; vma; vma = vma->vm_next) {
814 if (vma->vm_start > addr)
816 if (vma->vm_end > addr) {
817 mm->mmap_cache = vma;
824 EXPORT_SYMBOL(find_vma);
828 * - we don't extend stack VMAs under NOMMU conditions
830 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
832 return find_vma(mm, addr);
836 * expand a stack to a given address
837 * - not supported under NOMMU conditions
839 int expand_stack(struct vm_area_struct *vma, unsigned long address)
845 * look up the first VMA exactly that exactly matches addr
846 * - should be called with mm->mmap_sem at least held readlocked
848 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
852 struct vm_area_struct *vma;
853 unsigned long end = addr + len;
855 /* check the cache first */
856 vma = mm->mmap_cache;
857 if (vma && vma->vm_start == addr && vma->vm_end == end)
860 /* trawl the list (there may be multiple mappings in which addr
862 for (vma = mm->mmap; vma; vma = vma->vm_next) {
863 if (vma->vm_start < addr)
865 if (vma->vm_start > addr)
867 if (vma->vm_end == end) {
868 mm->mmap_cache = vma;
877 * determine whether a mapping should be permitted and, if so, what sort of
878 * mapping we're capable of supporting
880 static int validate_mmap_request(struct file *file,
886 unsigned long *_capabilities)
888 unsigned long capabilities, rlen;
891 /* do the simple checks first */
892 if (flags & MAP_FIXED) {
894 "%d: Can't do fixed-address/overlay mmap of RAM\n",
899 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
900 (flags & MAP_TYPE) != MAP_SHARED)
906 /* Careful about overflows.. */
907 rlen = PAGE_ALIGN(len);
908 if (!rlen || rlen > TASK_SIZE)
911 /* offset overflow? */
912 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
916 /* validate file mapping requests */
917 struct address_space *mapping;
919 /* files must support mmap */
920 if (!file->f_op || !file->f_op->mmap)
923 /* work out if what we've got could possibly be shared
924 * - we support chardevs that provide their own "memory"
925 * - we support files/blockdevs that are memory backed
927 mapping = file->f_mapping;
929 mapping = file->f_path.dentry->d_inode->i_mapping;
932 if (mapping && mapping->backing_dev_info)
933 capabilities = mapping->backing_dev_info->capabilities;
936 /* no explicit capabilities set, so assume some
938 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
941 capabilities = BDI_CAP_MAP_COPY;
956 /* eliminate any capabilities that we can't support on this
958 if (!file->f_op->get_unmapped_area)
959 capabilities &= ~BDI_CAP_MAP_DIRECT;
960 if (!file->f_op->read)
961 capabilities &= ~BDI_CAP_MAP_COPY;
963 /* The file shall have been opened with read permission. */
964 if (!(file->f_mode & FMODE_READ))
967 if (flags & MAP_SHARED) {
968 /* do checks for writing, appending and locking */
969 if ((prot & PROT_WRITE) &&
970 !(file->f_mode & FMODE_WRITE))
973 if (IS_APPEND(file->f_path.dentry->d_inode) &&
974 (file->f_mode & FMODE_WRITE))
977 if (locks_verify_locked(file->f_path.dentry->d_inode))
980 if (!(capabilities & BDI_CAP_MAP_DIRECT))
983 /* we mustn't privatise shared mappings */
984 capabilities &= ~BDI_CAP_MAP_COPY;
987 /* we're going to read the file into private memory we
989 if (!(capabilities & BDI_CAP_MAP_COPY))
992 /* we don't permit a private writable mapping to be
993 * shared with the backing device */
994 if (prot & PROT_WRITE)
995 capabilities &= ~BDI_CAP_MAP_DIRECT;
998 if (capabilities & BDI_CAP_MAP_DIRECT) {
999 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
1000 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1001 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
1003 capabilities &= ~BDI_CAP_MAP_DIRECT;
1004 if (flags & MAP_SHARED) {
1006 "MAP_SHARED not completely supported on !MMU\n");
1012 /* handle executable mappings and implied executable
1014 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1015 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 & BDI_CAP_EXEC_MAP)
1025 else if ((prot & PROT_READ) &&
1026 (prot & PROT_EXEC) &&
1027 !(capabilities & BDI_CAP_EXEC_MAP)
1029 /* backing file is not executable, try to copy */
1030 capabilities &= ~BDI_CAP_MAP_DIRECT;
1034 /* anonymous mappings are always memory backed and can be
1037 capabilities = BDI_CAP_MAP_COPY;
1039 /* handle PROT_EXEC implication by PROT_READ */
1040 if ((prot & PROT_READ) &&
1041 (current->personality & READ_IMPLIES_EXEC))
1045 /* allow the security API to have its say */
1046 ret = security_mmap_addr(addr);
1051 *_capabilities = capabilities;
1056 * we've determined that we can make the mapping, now translate what we
1057 * now know into VMA flags
1059 static unsigned long determine_vm_flags(struct file *file,
1061 unsigned long flags,
1062 unsigned long capabilities)
1064 unsigned long vm_flags;
1066 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1067 /* vm_flags |= mm->def_flags; */
1069 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1070 /* attempt to share read-only copies of mapped file chunks */
1071 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1072 if (file && !(prot & PROT_WRITE))
1073 vm_flags |= VM_MAYSHARE;
1075 /* overlay a shareable mapping on the backing device or inode
1076 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1078 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1079 if (flags & MAP_SHARED)
1080 vm_flags |= VM_SHARED;
1083 /* refuse to let anyone share private mappings with this process if
1084 * it's being traced - otherwise breakpoints set in it may interfere
1085 * with another untraced process
1087 if ((flags & MAP_PRIVATE) && current->ptrace)
1088 vm_flags &= ~VM_MAYSHARE;
1094 * set up a shared mapping on a file (the driver or filesystem provides and
1097 static int do_mmap_shared_file(struct vm_area_struct *vma)
1101 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1103 vma->vm_region->vm_top = vma->vm_region->vm_end;
1109 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1110 * opposed to tried but failed) so we can only give a suitable error as
1111 * it's not possible to make a private copy if MAP_SHARED was given */
1116 * set up a private mapping or an anonymous shared mapping
1118 static int do_mmap_private(struct vm_area_struct *vma,
1119 struct vm_region *region,
1121 unsigned long capabilities)
1124 unsigned long total, point, n;
1128 /* invoke the file's mapping function so that it can keep track of
1129 * shared mappings on devices or memory
1130 * - VM_MAYSHARE will be set if it may attempt to share
1132 if (capabilities & BDI_CAP_MAP_DIRECT) {
1133 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1135 /* shouldn't return success if we're not sharing */
1136 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1137 vma->vm_region->vm_top = vma->vm_region->vm_end;
1143 /* getting an ENOSYS error indicates that direct mmap isn't
1144 * possible (as opposed to tried but failed) so we'll try to
1145 * make a private copy of the data and map that instead */
1149 /* allocate some memory to hold the mapping
1150 * - note that this may not return a page-aligned address if the object
1151 * we're allocating is smaller than a page
1153 order = get_order(len);
1154 kdebug("alloc order %d for %lx", order, len);
1156 pages = alloc_pages(GFP_KERNEL, order);
1161 atomic_long_add(total, &mmap_pages_allocated);
1163 point = len >> PAGE_SHIFT;
1165 /* we allocated a power-of-2 sized page set, so we may want to trim off
1167 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1168 while (total > point) {
1169 order = ilog2(total - point);
1171 kdebug("shave %lu/%lu @%lu", n, total - point, total);
1172 atomic_long_sub(n, &mmap_pages_allocated);
1174 set_page_refcounted(pages + total);
1175 __free_pages(pages + total, order);
1179 for (point = 1; point < total; point++)
1180 set_page_refcounted(&pages[point]);
1182 base = page_address(pages);
1183 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1184 region->vm_start = (unsigned long) base;
1185 region->vm_end = region->vm_start + len;
1186 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1188 vma->vm_start = region->vm_start;
1189 vma->vm_end = region->vm_start + len;
1192 /* read the contents of a file into the copy */
1193 mm_segment_t old_fs;
1196 fpos = vma->vm_pgoff;
1197 fpos <<= PAGE_SHIFT;
1201 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
1207 /* clear the last little bit */
1209 memset(base + ret, 0, len - ret);
1216 free_page_series(region->vm_start, region->vm_top);
1217 region->vm_start = vma->vm_start = 0;
1218 region->vm_end = vma->vm_end = 0;
1223 printk("Allocation of length %lu from process %d (%s) failed\n",
1224 len, current->pid, current->comm);
1230 * handle mapping creation for uClinux
1232 unsigned long do_mmap_pgoff(struct file *file,
1236 unsigned long flags,
1237 unsigned long pgoff)
1239 struct vm_area_struct *vma;
1240 struct vm_region *region;
1242 unsigned long capabilities, vm_flags, result;
1245 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1247 /* decide whether we should attempt the mapping, and if so what sort of
1249 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1252 kleave(" = %d [val]", ret);
1256 /* we ignore the address hint */
1258 len = PAGE_ALIGN(len);
1260 /* we've determined that we can make the mapping, now translate what we
1261 * now know into VMA flags */
1262 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1264 /* we're going to need to record the mapping */
1265 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1267 goto error_getting_region;
1269 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1271 goto error_getting_vma;
1273 region->vm_usage = 1;
1274 region->vm_flags = vm_flags;
1275 region->vm_pgoff = pgoff;
1277 INIT_LIST_HEAD(&vma->anon_vma_chain);
1278 vma->vm_flags = vm_flags;
1279 vma->vm_pgoff = pgoff;
1282 region->vm_file = file;
1284 vma->vm_file = file;
1288 down_write(&nommu_region_sem);
1290 /* if we want to share, we need to check for regions created by other
1291 * mmap() calls that overlap with our proposed mapping
1292 * - we can only share with a superset match on most regular files
1293 * - shared mappings on character devices and memory backed files are
1294 * permitted to overlap inexactly as far as we are concerned for in
1295 * these cases, sharing is handled in the driver or filesystem rather
1298 if (vm_flags & VM_MAYSHARE) {
1299 struct vm_region *pregion;
1300 unsigned long pglen, rpglen, pgend, rpgend, start;
1302 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1303 pgend = pgoff + pglen;
1305 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1306 pregion = rb_entry(rb, struct vm_region, vm_rb);
1308 if (!(pregion->vm_flags & VM_MAYSHARE))
1311 /* search for overlapping mappings on the same file */
1312 if (pregion->vm_file->f_path.dentry->d_inode !=
1313 file->f_path.dentry->d_inode)
1316 if (pregion->vm_pgoff >= pgend)
1319 rpglen = pregion->vm_end - pregion->vm_start;
1320 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1321 rpgend = pregion->vm_pgoff + rpglen;
1322 if (pgoff >= rpgend)
1325 /* handle inexactly overlapping matches between
1327 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1328 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1329 /* new mapping is not a subset of the region */
1330 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1331 goto sharing_violation;
1335 /* we've found a region we can share */
1336 pregion->vm_usage++;
1337 vma->vm_region = pregion;
1338 start = pregion->vm_start;
1339 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1340 vma->vm_start = start;
1341 vma->vm_end = start + len;
1343 if (pregion->vm_flags & VM_MAPPED_COPY) {
1344 kdebug("share copy");
1345 vma->vm_flags |= VM_MAPPED_COPY;
1347 kdebug("share mmap");
1348 ret = do_mmap_shared_file(vma);
1350 vma->vm_region = NULL;
1353 pregion->vm_usage--;
1355 goto error_just_free;
1358 fput(region->vm_file);
1359 kmem_cache_free(vm_region_jar, region);
1365 /* obtain the address at which to make a shared mapping
1366 * - this is the hook for quasi-memory character devices to
1367 * tell us the location of a shared mapping
1369 if (capabilities & BDI_CAP_MAP_DIRECT) {
1370 addr = file->f_op->get_unmapped_area(file, addr, len,
1372 if (IS_ERR_VALUE(addr)) {
1375 goto error_just_free;
1377 /* the driver refused to tell us where to site
1378 * the mapping so we'll have to attempt to copy
1381 if (!(capabilities & BDI_CAP_MAP_COPY))
1382 goto error_just_free;
1384 capabilities &= ~BDI_CAP_MAP_DIRECT;
1386 vma->vm_start = region->vm_start = addr;
1387 vma->vm_end = region->vm_end = addr + len;
1392 vma->vm_region = region;
1394 /* set up the mapping
1395 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1397 if (file && vma->vm_flags & VM_SHARED)
1398 ret = do_mmap_shared_file(vma);
1400 ret = do_mmap_private(vma, region, len, capabilities);
1402 goto error_just_free;
1403 add_nommu_region(region);
1405 /* clear anonymous mappings that don't ask for uninitialized data */
1406 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1407 memset((void *)region->vm_start, 0,
1408 region->vm_end - region->vm_start);
1410 /* okay... we have a mapping; now we have to register it */
1411 result = vma->vm_start;
1413 current->mm->total_vm += len >> PAGE_SHIFT;
1416 add_vma_to_mm(current->mm, vma);
1418 /* we flush the region from the icache only when the first executable
1419 * mapping of it is made */
1420 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1421 flush_icache_range(region->vm_start, region->vm_end);
1422 region->vm_icache_flushed = true;
1425 up_write(&nommu_region_sem);
1427 kleave(" = %lx", result);
1431 up_write(&nommu_region_sem);
1433 if (region->vm_file)
1434 fput(region->vm_file);
1435 kmem_cache_free(vm_region_jar, region);
1438 kmem_cache_free(vm_area_cachep, vma);
1439 kleave(" = %d", ret);
1443 up_write(&nommu_region_sem);
1444 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1449 kmem_cache_free(vm_region_jar, region);
1450 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1451 " from process %d failed\n",
1456 error_getting_region:
1457 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1458 " from process %d failed\n",
1464 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1465 unsigned long, prot, unsigned long, flags,
1466 unsigned long, fd, unsigned long, pgoff)
1468 struct file *file = NULL;
1469 unsigned long retval = -EBADF;
1471 audit_mmap_fd(fd, flags);
1472 if (!(flags & MAP_ANONYMOUS)) {
1478 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1480 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1488 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1489 struct mmap_arg_struct {
1493 unsigned long flags;
1495 unsigned long offset;
1498 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1500 struct mmap_arg_struct a;
1502 if (copy_from_user(&a, arg, sizeof(a)))
1504 if (a.offset & ~PAGE_MASK)
1507 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1508 a.offset >> PAGE_SHIFT);
1510 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1513 * split a vma into two pieces at address 'addr', a new vma is allocated either
1514 * for the first part or the tail.
1516 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1517 unsigned long addr, int new_below)
1519 struct vm_area_struct *new;
1520 struct vm_region *region;
1521 unsigned long npages;
1525 /* we're only permitted to split anonymous regions (these should have
1526 * only a single usage on the region) */
1530 if (mm->map_count >= sysctl_max_map_count)
1533 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1537 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1539 kmem_cache_free(vm_region_jar, region);
1543 /* most fields are the same, copy all, and then fixup */
1545 *region = *vma->vm_region;
1546 new->vm_region = region;
1548 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1551 region->vm_top = region->vm_end = new->vm_end = addr;
1553 region->vm_start = new->vm_start = addr;
1554 region->vm_pgoff = new->vm_pgoff += npages;
1557 if (new->vm_ops && new->vm_ops->open)
1558 new->vm_ops->open(new);
1560 delete_vma_from_mm(vma);
1561 down_write(&nommu_region_sem);
1562 delete_nommu_region(vma->vm_region);
1564 vma->vm_region->vm_start = vma->vm_start = addr;
1565 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1567 vma->vm_region->vm_end = vma->vm_end = addr;
1568 vma->vm_region->vm_top = addr;
1570 add_nommu_region(vma->vm_region);
1571 add_nommu_region(new->vm_region);
1572 up_write(&nommu_region_sem);
1573 add_vma_to_mm(mm, vma);
1574 add_vma_to_mm(mm, new);
1579 * shrink a VMA by removing the specified chunk from either the beginning or
1582 static int shrink_vma(struct mm_struct *mm,
1583 struct vm_area_struct *vma,
1584 unsigned long from, unsigned long to)
1586 struct vm_region *region;
1590 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1592 delete_vma_from_mm(vma);
1593 if (from > vma->vm_start)
1597 add_vma_to_mm(mm, vma);
1599 /* cut the backing region down to size */
1600 region = vma->vm_region;
1601 BUG_ON(region->vm_usage != 1);
1603 down_write(&nommu_region_sem);
1604 delete_nommu_region(region);
1605 if (from > region->vm_start) {
1606 to = region->vm_top;
1607 region->vm_top = region->vm_end = from;
1609 region->vm_start = to;
1611 add_nommu_region(region);
1612 up_write(&nommu_region_sem);
1614 free_page_series(from, to);
1620 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1621 * VMA, though it need not cover the whole VMA
1623 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1625 struct vm_area_struct *vma;
1629 kenter(",%lx,%zx", start, len);
1631 len = PAGE_ALIGN(len);
1637 /* find the first potentially overlapping VMA */
1638 vma = find_vma(mm, start);
1640 static int limit = 0;
1643 "munmap of memory not mmapped by process %d"
1644 " (%s): 0x%lx-0x%lx\n",
1645 current->pid, current->comm,
1646 start, start + len - 1);
1652 /* we're allowed to split an anonymous VMA but not a file-backed one */
1655 if (start > vma->vm_start) {
1656 kleave(" = -EINVAL [miss]");
1659 if (end == vma->vm_end)
1660 goto erase_whole_vma;
1663 kleave(" = -EINVAL [split file]");
1666 /* the chunk must be a subset of the VMA found */
1667 if (start == vma->vm_start && end == vma->vm_end)
1668 goto erase_whole_vma;
1669 if (start < vma->vm_start || end > vma->vm_end) {
1670 kleave(" = -EINVAL [superset]");
1673 if (start & ~PAGE_MASK) {
1674 kleave(" = -EINVAL [unaligned start]");
1677 if (end != vma->vm_end && end & ~PAGE_MASK) {
1678 kleave(" = -EINVAL [unaligned split]");
1681 if (start != vma->vm_start && end != vma->vm_end) {
1682 ret = split_vma(mm, vma, start, 1);
1684 kleave(" = %d [split]", ret);
1688 return shrink_vma(mm, vma, start, end);
1692 delete_vma_from_mm(vma);
1693 delete_vma(mm, vma);
1697 EXPORT_SYMBOL(do_munmap);
1699 int vm_munmap(unsigned long addr, size_t len)
1701 struct mm_struct *mm = current->mm;
1704 down_write(&mm->mmap_sem);
1705 ret = do_munmap(mm, addr, len);
1706 up_write(&mm->mmap_sem);
1709 EXPORT_SYMBOL(vm_munmap);
1711 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1713 return vm_munmap(addr, len);
1717 * release all the mappings made in a process's VM space
1719 void exit_mmap(struct mm_struct *mm)
1721 struct vm_area_struct *vma;
1730 while ((vma = mm->mmap)) {
1731 mm->mmap = vma->vm_next;
1732 delete_vma_from_mm(vma);
1733 delete_vma(mm, vma);
1740 unsigned long vm_brk(unsigned long addr, unsigned long len)
1746 * expand (or shrink) an existing mapping, potentially moving it at the same
1747 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1749 * under NOMMU conditions, we only permit changing a mapping's size, and only
1750 * as long as it stays within the region allocated by do_mmap_private() and the
1751 * block is not shareable
1753 * MREMAP_FIXED is not supported under NOMMU conditions
1755 unsigned long do_mremap(unsigned long addr,
1756 unsigned long old_len, unsigned long new_len,
1757 unsigned long flags, unsigned long new_addr)
1759 struct vm_area_struct *vma;
1761 /* insanity checks first */
1762 old_len = PAGE_ALIGN(old_len);
1763 new_len = PAGE_ALIGN(new_len);
1764 if (old_len == 0 || new_len == 0)
1765 return (unsigned long) -EINVAL;
1767 if (addr & ~PAGE_MASK)
1770 if (flags & MREMAP_FIXED && new_addr != addr)
1771 return (unsigned long) -EINVAL;
1773 vma = find_vma_exact(current->mm, addr, old_len);
1775 return (unsigned long) -EINVAL;
1777 if (vma->vm_end != vma->vm_start + old_len)
1778 return (unsigned long) -EFAULT;
1780 if (vma->vm_flags & VM_MAYSHARE)
1781 return (unsigned long) -EPERM;
1783 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1784 return (unsigned long) -ENOMEM;
1786 /* all checks complete - do it */
1787 vma->vm_end = vma->vm_start + new_len;
1788 return vma->vm_start;
1790 EXPORT_SYMBOL(do_mremap);
1792 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1793 unsigned long, new_len, unsigned long, flags,
1794 unsigned long, new_addr)
1798 down_write(¤t->mm->mmap_sem);
1799 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1800 up_write(¤t->mm->mmap_sem);
1804 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1805 unsigned int foll_flags)
1810 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1811 unsigned long pfn, unsigned long size, pgprot_t prot)
1813 if (addr != (pfn << PAGE_SHIFT))
1816 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1819 EXPORT_SYMBOL(remap_pfn_range);
1821 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1822 unsigned long pgoff)
1824 unsigned int size = vma->vm_end - vma->vm_start;
1826 if (!(vma->vm_flags & VM_USERMAP))
1829 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1830 vma->vm_end = vma->vm_start + size;
1834 EXPORT_SYMBOL(remap_vmalloc_range);
1836 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1837 unsigned long len, unsigned long pgoff, unsigned long flags)
1842 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1846 void unmap_mapping_range(struct address_space *mapping,
1847 loff_t const holebegin, loff_t const holelen,
1851 EXPORT_SYMBOL(unmap_mapping_range);
1854 * Check that a process has enough memory to allocate a new virtual
1855 * mapping. 0 means there is enough memory for the allocation to
1856 * succeed and -ENOMEM implies there is not.
1858 * We currently support three overcommit policies, which are set via the
1859 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1861 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1862 * Additional code 2002 Jul 20 by Robert Love.
1864 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1866 * Note this is a helper function intended to be used by LSMs which
1867 * wish to use this logic.
1869 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1871 unsigned long free, allowed;
1873 vm_acct_memory(pages);
1876 * Sometimes we want to use more memory than we have
1878 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1881 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1882 free = global_page_state(NR_FREE_PAGES);
1883 free += global_page_state(NR_FILE_PAGES);
1886 * shmem pages shouldn't be counted as free in this
1887 * case, they can't be purged, only swapped out, and
1888 * that won't affect the overall amount of available
1889 * memory in the system.
1891 free -= global_page_state(NR_SHMEM);
1893 free += nr_swap_pages;
1896 * Any slabs which are created with the
1897 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1898 * which are reclaimable, under pressure. The dentry
1899 * cache and most inode caches should fall into this
1901 free += global_page_state(NR_SLAB_RECLAIMABLE);
1904 * Leave reserved pages. The pages are not for anonymous pages.
1906 if (free <= totalreserve_pages)
1909 free -= totalreserve_pages;
1912 * Leave the last 3% for root
1923 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1925 * Leave the last 3% for root
1928 allowed -= allowed / 32;
1929 allowed += total_swap_pages;
1931 /* Don't let a single process grow too big:
1932 leave 3% of the size of this process for other processes */
1934 allowed -= mm->total_vm / 32;
1936 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1940 vm_unacct_memory(pages);
1945 int in_gate_area_no_mm(unsigned long addr)
1950 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1955 EXPORT_SYMBOL(filemap_fault);
1957 int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr,
1958 unsigned long size, pgoff_t pgoff)
1963 EXPORT_SYMBOL(generic_file_remap_pages);
1965 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1966 unsigned long addr, void *buf, int len, int write)
1968 struct vm_area_struct *vma;
1970 down_read(&mm->mmap_sem);
1972 /* the access must start within one of the target process's mappings */
1973 vma = find_vma(mm, addr);
1975 /* don't overrun this mapping */
1976 if (addr + len >= vma->vm_end)
1977 len = vma->vm_end - addr;
1979 /* only read or write mappings where it is permitted */
1980 if (write && vma->vm_flags & VM_MAYWRITE)
1981 copy_to_user_page(vma, NULL, addr,
1982 (void *) addr, buf, len);
1983 else if (!write && vma->vm_flags & VM_MAYREAD)
1984 copy_from_user_page(vma, NULL, addr,
1985 buf, (void *) addr, len);
1992 up_read(&mm->mmap_sem);
1998 * @access_remote_vm - access another process' address space
1999 * @mm: the mm_struct of the target address space
2000 * @addr: start address to access
2001 * @buf: source or destination buffer
2002 * @len: number of bytes to transfer
2003 * @write: whether the access is a write
2005 * The caller must hold a reference on @mm.
2007 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2008 void *buf, int len, int write)
2010 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2014 * Access another process' address space.
2015 * - source/target buffer must be kernel space
2017 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2019 struct mm_struct *mm;
2021 if (addr + len < addr)
2024 mm = get_task_mm(tsk);
2028 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2035 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2036 * @inode: The inode to check
2037 * @size: The current filesize of the inode
2038 * @newsize: The proposed filesize of the inode
2040 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2041 * make sure that that any outstanding VMAs aren't broken and then shrink the
2042 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2043 * automatically grant mappings that are too large.
2045 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2048 struct vm_area_struct *vma;
2049 struct prio_tree_iter iter;
2050 struct vm_region *region;
2052 size_t r_size, r_top;
2054 low = newsize >> PAGE_SHIFT;
2055 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2057 down_write(&nommu_region_sem);
2058 mutex_lock(&inode->i_mapping->i_mmap_mutex);
2060 /* search for VMAs that fall within the dead zone */
2061 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2063 /* found one - only interested if it's shared out of the page
2065 if (vma->vm_flags & VM_SHARED) {
2066 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2067 up_write(&nommu_region_sem);
2068 return -ETXTBSY; /* not quite true, but near enough */
2072 /* reduce any regions that overlap the dead zone - if in existence,
2073 * these will be pointed to by VMAs that don't overlap the dead zone
2075 * we don't check for any regions that start beyond the EOF as there
2078 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2080 if (!(vma->vm_flags & VM_SHARED))
2083 region = vma->vm_region;
2084 r_size = region->vm_top - region->vm_start;
2085 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2087 if (r_top > newsize) {
2088 region->vm_top -= r_top - newsize;
2089 if (region->vm_end > region->vm_top)
2090 region->vm_end = region->vm_top;
2094 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2095 up_write(&nommu_region_sem);