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>
32 #include <linux/sched/sysctl.h>
34 #include <asm/uaccess.h>
36 #include <asm/tlbflush.h>
37 #include <asm/mmu_context.h>
41 #define kenter(FMT, ...) \
42 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
43 #define kleave(FMT, ...) \
44 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
45 #define kdebug(FMT, ...) \
46 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
48 #define kenter(FMT, ...) \
49 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
50 #define kleave(FMT, ...) \
51 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
52 #define kdebug(FMT, ...) \
53 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
58 unsigned long max_mapnr;
59 unsigned long num_physpages;
60 unsigned long highest_memmap_pfn;
61 struct percpu_counter vm_committed_as;
62 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
63 int sysctl_overcommit_ratio = 50; /* default is 50% */
64 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
65 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
66 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
67 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
68 int heap_stack_gap = 0;
70 atomic_long_t mmap_pages_allocated;
73 * The global memory commitment made in the system can be a metric
74 * that can be used to drive ballooning decisions when Linux is hosted
75 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
76 * balancing memory across competing virtual machines that are hosted.
77 * Several metrics drive this policy engine including the guest reported
80 unsigned long vm_memory_committed(void)
82 return percpu_counter_read_positive(&vm_committed_as);
85 EXPORT_SYMBOL_GPL(vm_memory_committed);
87 EXPORT_SYMBOL(mem_map);
88 EXPORT_SYMBOL(num_physpages);
90 /* list of mapped, potentially shareable regions */
91 static struct kmem_cache *vm_region_jar;
92 struct rb_root nommu_region_tree = RB_ROOT;
93 DECLARE_RWSEM(nommu_region_sem);
95 const struct vm_operations_struct generic_file_vm_ops = {
99 * Return the total memory allocated for this pointer, not
100 * just what the caller asked for.
102 * Doesn't have to be accurate, i.e. may have races.
104 unsigned int kobjsize(const void *objp)
109 * If the object we have should not have ksize performed on it,
112 if (!objp || !virt_addr_valid(objp))
115 page = virt_to_head_page(objp);
118 * If the allocator sets PageSlab, we know the pointer came from
125 * If it's not a compound page, see if we have a matching VMA
126 * region. This test is intentionally done in reverse order,
127 * so if there's no VMA, we still fall through and hand back
128 * PAGE_SIZE for 0-order pages.
130 if (!PageCompound(page)) {
131 struct vm_area_struct *vma;
133 vma = find_vma(current->mm, (unsigned long)objp);
135 return vma->vm_end - vma->vm_start;
139 * The ksize() function is only guaranteed to work for pointers
140 * returned by kmalloc(). So handle arbitrary pointers here.
142 return PAGE_SIZE << compound_order(page);
145 long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
146 unsigned long start, unsigned long nr_pages,
147 unsigned int foll_flags, struct page **pages,
148 struct vm_area_struct **vmas, int *nonblocking)
150 struct vm_area_struct *vma;
151 unsigned long vm_flags;
154 /* calculate required read or write permissions.
155 * If FOLL_FORCE is set, we only require the "MAY" flags.
157 vm_flags = (foll_flags & FOLL_WRITE) ?
158 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
159 vm_flags &= (foll_flags & FOLL_FORCE) ?
160 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
162 for (i = 0; i < nr_pages; i++) {
163 vma = find_vma(mm, start);
165 goto finish_or_fault;
167 /* protect what we can, including chardevs */
168 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
169 !(vm_flags & vma->vm_flags))
170 goto finish_or_fault;
173 pages[i] = virt_to_page(start);
175 page_cache_get(pages[i]);
179 start = (start + PAGE_SIZE) & PAGE_MASK;
185 return i ? : -EFAULT;
189 * get a list of pages in an address range belonging to the specified process
190 * and indicate the VMA that covers each page
191 * - this is potentially dodgy as we may end incrementing the page count of a
192 * slab page or a secondary page from a compound page
193 * - don't permit access to VMAs that don't support it, such as I/O mappings
195 long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
196 unsigned long start, unsigned long nr_pages,
197 int write, int force, struct page **pages,
198 struct vm_area_struct **vmas)
207 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
210 EXPORT_SYMBOL(get_user_pages);
213 * follow_pfn - look up PFN at a user virtual address
214 * @vma: memory mapping
215 * @address: user virtual address
216 * @pfn: location to store found PFN
218 * Only IO mappings and raw PFN mappings are allowed.
220 * Returns zero and the pfn at @pfn on success, -ve otherwise.
222 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
225 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
228 *pfn = address >> PAGE_SHIFT;
231 EXPORT_SYMBOL(follow_pfn);
233 LIST_HEAD(vmap_area_list);
235 void vfree(const void *addr)
239 EXPORT_SYMBOL(vfree);
241 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
244 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
245 * returns only a logical address.
247 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
249 EXPORT_SYMBOL(__vmalloc);
251 void *vmalloc_user(unsigned long size)
255 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
258 struct vm_area_struct *vma;
260 down_write(¤t->mm->mmap_sem);
261 vma = find_vma(current->mm, (unsigned long)ret);
263 vma->vm_flags |= VM_USERMAP;
264 up_write(¤t->mm->mmap_sem);
269 EXPORT_SYMBOL(vmalloc_user);
271 struct page *vmalloc_to_page(const void *addr)
273 return virt_to_page(addr);
275 EXPORT_SYMBOL(vmalloc_to_page);
277 unsigned long vmalloc_to_pfn(const void *addr)
279 return page_to_pfn(virt_to_page(addr));
281 EXPORT_SYMBOL(vmalloc_to_pfn);
283 long vread(char *buf, char *addr, unsigned long count)
285 memcpy(buf, addr, count);
289 long vwrite(char *buf, char *addr, unsigned long count)
291 /* Don't allow overflow */
292 if ((unsigned long) addr + count < count)
293 count = -(unsigned long) addr;
295 memcpy(addr, buf, count);
300 * vmalloc - allocate virtually continguos memory
302 * @size: allocation size
304 * Allocate enough pages to cover @size from the page level
305 * allocator and map them into continguos kernel virtual space.
307 * For tight control over page level allocator and protection flags
308 * use __vmalloc() instead.
310 void *vmalloc(unsigned long size)
312 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
314 EXPORT_SYMBOL(vmalloc);
317 * vzalloc - allocate virtually continguos memory with zero fill
319 * @size: allocation size
321 * Allocate enough pages to cover @size from the page level
322 * allocator and map them into continguos kernel virtual space.
323 * The memory allocated is set to zero.
325 * For tight control over page level allocator and protection flags
326 * use __vmalloc() instead.
328 void *vzalloc(unsigned long size)
330 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
333 EXPORT_SYMBOL(vzalloc);
336 * vmalloc_node - allocate memory on a specific node
337 * @size: allocation size
340 * Allocate enough pages to cover @size from the page level
341 * allocator and map them into contiguous kernel virtual space.
343 * For tight control over page level allocator and protection flags
344 * use __vmalloc() instead.
346 void *vmalloc_node(unsigned long size, int node)
348 return vmalloc(size);
350 EXPORT_SYMBOL(vmalloc_node);
353 * vzalloc_node - allocate memory on a specific node with zero fill
354 * @size: allocation size
357 * Allocate enough pages to cover @size from the page level
358 * allocator and map them into contiguous kernel virtual space.
359 * The memory allocated is set to zero.
361 * For tight control over page level allocator and protection flags
362 * use __vmalloc() instead.
364 void *vzalloc_node(unsigned long size, int node)
366 return vzalloc(size);
368 EXPORT_SYMBOL(vzalloc_node);
370 #ifndef PAGE_KERNEL_EXEC
371 # define PAGE_KERNEL_EXEC PAGE_KERNEL
375 * vmalloc_exec - allocate virtually contiguous, executable memory
376 * @size: allocation size
378 * Kernel-internal function to allocate enough pages to cover @size
379 * the page level allocator and map them into contiguous and
380 * executable kernel virtual space.
382 * For tight control over page level allocator and protection flags
383 * use __vmalloc() instead.
386 void *vmalloc_exec(unsigned long size)
388 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
392 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
393 * @size: allocation size
395 * Allocate enough 32bit PA addressable pages to cover @size from the
396 * page level allocator and map them into continguos kernel virtual space.
398 void *vmalloc_32(unsigned long size)
400 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
402 EXPORT_SYMBOL(vmalloc_32);
405 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
406 * @size: allocation size
408 * The resulting memory area is 32bit addressable and zeroed so it can be
409 * mapped to userspace without leaking data.
411 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
412 * remap_vmalloc_range() are permissible.
414 void *vmalloc_32_user(unsigned long size)
417 * We'll have to sort out the ZONE_DMA bits for 64-bit,
418 * but for now this can simply use vmalloc_user() directly.
420 return vmalloc_user(size);
422 EXPORT_SYMBOL(vmalloc_32_user);
424 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
431 void vunmap(const void *addr)
435 EXPORT_SYMBOL(vunmap);
437 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
442 EXPORT_SYMBOL(vm_map_ram);
444 void vm_unmap_ram(const void *mem, unsigned int count)
448 EXPORT_SYMBOL(vm_unmap_ram);
450 void vm_unmap_aliases(void)
453 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
456 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
459 void __attribute__((weak)) vmalloc_sync_all(void)
464 * alloc_vm_area - allocate a range of kernel address space
465 * @size: size of the area
467 * Returns: NULL on failure, vm_struct on success
469 * This function reserves a range of kernel address space, and
470 * allocates pagetables to map that range. No actual mappings
471 * are created. If the kernel address space is not shared
472 * between processes, it syncs the pagetable across all
475 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
480 EXPORT_SYMBOL_GPL(alloc_vm_area);
482 void free_vm_area(struct vm_struct *area)
486 EXPORT_SYMBOL_GPL(free_vm_area);
488 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
493 EXPORT_SYMBOL(vm_insert_page);
496 * sys_brk() for the most part doesn't need the global kernel
497 * lock, except when an application is doing something nasty
498 * like trying to un-brk an area that has already been mapped
499 * to a regular file. in this case, the unmapping will need
500 * to invoke file system routines that need the global lock.
502 SYSCALL_DEFINE1(brk, unsigned long, brk)
504 struct mm_struct *mm = current->mm;
506 if (brk < mm->start_brk || brk > mm->context.end_brk)
513 * Always allow shrinking brk
515 if (brk <= mm->brk) {
521 * Ok, looks good - let it rip.
523 flush_icache_range(mm->brk, brk);
524 return mm->brk = brk;
528 * initialise the VMA and region record slabs
530 void __init mmap_init(void)
534 ret = percpu_counter_init(&vm_committed_as, 0);
536 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
540 * validate the region tree
541 * - the caller must hold the region lock
543 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
544 static noinline void validate_nommu_regions(void)
546 struct vm_region *region, *last;
547 struct rb_node *p, *lastp;
549 lastp = rb_first(&nommu_region_tree);
553 last = rb_entry(lastp, struct vm_region, vm_rb);
554 BUG_ON(unlikely(last->vm_end <= last->vm_start));
555 BUG_ON(unlikely(last->vm_top < last->vm_end));
557 while ((p = rb_next(lastp))) {
558 region = rb_entry(p, struct vm_region, vm_rb);
559 last = rb_entry(lastp, struct vm_region, vm_rb);
561 BUG_ON(unlikely(region->vm_end <= region->vm_start));
562 BUG_ON(unlikely(region->vm_top < region->vm_end));
563 BUG_ON(unlikely(region->vm_start < last->vm_top));
569 static void validate_nommu_regions(void)
575 * add a region into the global tree
577 static void add_nommu_region(struct vm_region *region)
579 struct vm_region *pregion;
580 struct rb_node **p, *parent;
582 validate_nommu_regions();
585 p = &nommu_region_tree.rb_node;
588 pregion = rb_entry(parent, struct vm_region, vm_rb);
589 if (region->vm_start < pregion->vm_start)
591 else if (region->vm_start > pregion->vm_start)
593 else if (pregion == region)
599 rb_link_node(®ion->vm_rb, parent, p);
600 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
602 validate_nommu_regions();
606 * delete a region from the global tree
608 static void delete_nommu_region(struct vm_region *region)
610 BUG_ON(!nommu_region_tree.rb_node);
612 validate_nommu_regions();
613 rb_erase(®ion->vm_rb, &nommu_region_tree);
614 validate_nommu_regions();
618 * free a contiguous series of pages
620 static void free_page_series(unsigned long from, unsigned long to)
622 for (; from < to; from += PAGE_SIZE) {
623 struct page *page = virt_to_page(from);
625 kdebug("- free %lx", from);
626 atomic_long_dec(&mmap_pages_allocated);
627 if (page_count(page) != 1)
628 kdebug("free page %p: refcount not one: %d",
629 page, page_count(page));
635 * release a reference to a region
636 * - the caller must hold the region semaphore for writing, which this releases
637 * - the region may not have been added to the tree yet, in which case vm_top
638 * will equal vm_start
640 static void __put_nommu_region(struct vm_region *region)
641 __releases(nommu_region_sem)
643 kenter("%p{%d}", region, region->vm_usage);
645 BUG_ON(!nommu_region_tree.rb_node);
647 if (--region->vm_usage == 0) {
648 if (region->vm_top > region->vm_start)
649 delete_nommu_region(region);
650 up_write(&nommu_region_sem);
653 fput(region->vm_file);
655 /* IO memory and memory shared directly out of the pagecache
656 * from ramfs/tmpfs mustn't be released here */
657 if (region->vm_flags & VM_MAPPED_COPY) {
658 kdebug("free series");
659 free_page_series(region->vm_start, region->vm_top);
661 kmem_cache_free(vm_region_jar, region);
663 up_write(&nommu_region_sem);
668 * release a reference to a region
670 static void put_nommu_region(struct vm_region *region)
672 down_write(&nommu_region_sem);
673 __put_nommu_region(region);
677 * update protection on a vma
679 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
682 struct mm_struct *mm = vma->vm_mm;
683 long start = vma->vm_start & PAGE_MASK;
684 while (start < vma->vm_end) {
685 protect_page(mm, start, flags);
688 update_protections(mm);
693 * add a VMA into a process's mm_struct in the appropriate place in the list
694 * and tree and add to the address space's page tree also if not an anonymous
696 * - should be called with mm->mmap_sem held writelocked
698 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
700 struct vm_area_struct *pvma, *prev;
701 struct address_space *mapping;
702 struct rb_node **p, *parent, *rb_prev;
706 BUG_ON(!vma->vm_region);
711 protect_vma(vma, vma->vm_flags);
713 /* add the VMA to the mapping */
715 mapping = vma->vm_file->f_mapping;
717 mutex_lock(&mapping->i_mmap_mutex);
718 flush_dcache_mmap_lock(mapping);
719 vma_interval_tree_insert(vma, &mapping->i_mmap);
720 flush_dcache_mmap_unlock(mapping);
721 mutex_unlock(&mapping->i_mmap_mutex);
724 /* add the VMA to the tree */
725 parent = rb_prev = NULL;
726 p = &mm->mm_rb.rb_node;
729 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
731 /* sort by: start addr, end addr, VMA struct addr in that order
732 * (the latter is necessary as we may get identical VMAs) */
733 if (vma->vm_start < pvma->vm_start)
735 else if (vma->vm_start > pvma->vm_start) {
738 } else if (vma->vm_end < pvma->vm_end)
740 else if (vma->vm_end > pvma->vm_end) {
743 } else if (vma < pvma)
745 else if (vma > pvma) {
752 rb_link_node(&vma->vm_rb, parent, p);
753 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
755 /* add VMA to the VMA list also */
758 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
760 __vma_link_list(mm, vma, prev, parent);
764 * delete a VMA from its owning mm_struct and address space
766 static void delete_vma_from_mm(struct vm_area_struct *vma)
768 struct address_space *mapping;
769 struct mm_struct *mm = vma->vm_mm;
776 if (mm->mmap_cache == vma)
777 mm->mmap_cache = NULL;
779 /* remove the VMA from the mapping */
781 mapping = vma->vm_file->f_mapping;
783 mutex_lock(&mapping->i_mmap_mutex);
784 flush_dcache_mmap_lock(mapping);
785 vma_interval_tree_remove(vma, &mapping->i_mmap);
786 flush_dcache_mmap_unlock(mapping);
787 mutex_unlock(&mapping->i_mmap_mutex);
790 /* remove from the MM's tree and list */
791 rb_erase(&vma->vm_rb, &mm->mm_rb);
794 vma->vm_prev->vm_next = vma->vm_next;
796 mm->mmap = vma->vm_next;
799 vma->vm_next->vm_prev = vma->vm_prev;
803 * destroy a VMA record
805 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
808 if (vma->vm_ops && vma->vm_ops->close)
809 vma->vm_ops->close(vma);
812 put_nommu_region(vma->vm_region);
813 kmem_cache_free(vm_area_cachep, vma);
817 * look up the first VMA in which addr resides, NULL if none
818 * - should be called with mm->mmap_sem at least held readlocked
820 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
822 struct vm_area_struct *vma;
824 /* check the cache first */
825 vma = ACCESS_ONCE(mm->mmap_cache);
826 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
829 /* trawl the list (there may be multiple mappings in which addr
831 for (vma = mm->mmap; vma; vma = vma->vm_next) {
832 if (vma->vm_start > addr)
834 if (vma->vm_end > addr) {
835 mm->mmap_cache = vma;
842 EXPORT_SYMBOL(find_vma);
846 * - we don't extend stack VMAs under NOMMU conditions
848 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
850 return find_vma(mm, addr);
854 * expand a stack to a given address
855 * - not supported under NOMMU conditions
857 int expand_stack(struct vm_area_struct *vma, unsigned long address)
863 * look up the first VMA exactly that exactly matches addr
864 * - should be called with mm->mmap_sem at least held readlocked
866 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
870 struct vm_area_struct *vma;
871 unsigned long end = addr + len;
873 /* check the cache first */
874 vma = mm->mmap_cache;
875 if (vma && vma->vm_start == addr && vma->vm_end == end)
878 /* trawl the list (there may be multiple mappings in which addr
880 for (vma = mm->mmap; vma; vma = vma->vm_next) {
881 if (vma->vm_start < addr)
883 if (vma->vm_start > addr)
885 if (vma->vm_end == end) {
886 mm->mmap_cache = vma;
895 * determine whether a mapping should be permitted and, if so, what sort of
896 * mapping we're capable of supporting
898 static int validate_mmap_request(struct file *file,
904 unsigned long *_capabilities)
906 unsigned long capabilities, rlen;
909 /* do the simple checks first */
910 if (flags & MAP_FIXED) {
912 "%d: Can't do fixed-address/overlay mmap of RAM\n",
917 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
918 (flags & MAP_TYPE) != MAP_SHARED)
924 /* Careful about overflows.. */
925 rlen = PAGE_ALIGN(len);
926 if (!rlen || rlen > TASK_SIZE)
929 /* offset overflow? */
930 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
934 /* validate file mapping requests */
935 struct address_space *mapping;
937 /* files must support mmap */
938 if (!file->f_op || !file->f_op->mmap)
941 /* work out if what we've got could possibly be shared
942 * - we support chardevs that provide their own "memory"
943 * - we support files/blockdevs that are memory backed
945 mapping = file->f_mapping;
947 mapping = file_inode(file)->i_mapping;
950 if (mapping && mapping->backing_dev_info)
951 capabilities = mapping->backing_dev_info->capabilities;
954 /* no explicit capabilities set, so assume some
956 switch (file_inode(file)->i_mode & S_IFMT) {
959 capabilities = BDI_CAP_MAP_COPY;
974 /* eliminate any capabilities that we can't support on this
976 if (!file->f_op->get_unmapped_area)
977 capabilities &= ~BDI_CAP_MAP_DIRECT;
978 if (!file->f_op->read)
979 capabilities &= ~BDI_CAP_MAP_COPY;
981 /* The file shall have been opened with read permission. */
982 if (!(file->f_mode & FMODE_READ))
985 if (flags & MAP_SHARED) {
986 /* do checks for writing, appending and locking */
987 if ((prot & PROT_WRITE) &&
988 !(file->f_mode & FMODE_WRITE))
991 if (IS_APPEND(file_inode(file)) &&
992 (file->f_mode & FMODE_WRITE))
995 if (locks_verify_locked(file_inode(file)))
998 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1001 /* we mustn't privatise shared mappings */
1002 capabilities &= ~BDI_CAP_MAP_COPY;
1005 /* we're going to read the file into private memory we
1007 if (!(capabilities & BDI_CAP_MAP_COPY))
1010 /* we don't permit a private writable mapping to be
1011 * shared with the backing device */
1012 if (prot & PROT_WRITE)
1013 capabilities &= ~BDI_CAP_MAP_DIRECT;
1016 if (capabilities & BDI_CAP_MAP_DIRECT) {
1017 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
1018 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1019 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
1021 capabilities &= ~BDI_CAP_MAP_DIRECT;
1022 if (flags & MAP_SHARED) {
1024 "MAP_SHARED not completely supported on !MMU\n");
1030 /* handle executable mappings and implied executable
1032 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1033 if (prot & PROT_EXEC)
1036 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1037 /* handle implication of PROT_EXEC by PROT_READ */
1038 if (current->personality & READ_IMPLIES_EXEC) {
1039 if (capabilities & BDI_CAP_EXEC_MAP)
1043 else if ((prot & PROT_READ) &&
1044 (prot & PROT_EXEC) &&
1045 !(capabilities & BDI_CAP_EXEC_MAP)
1047 /* backing file is not executable, try to copy */
1048 capabilities &= ~BDI_CAP_MAP_DIRECT;
1052 /* anonymous mappings are always memory backed and can be
1055 capabilities = BDI_CAP_MAP_COPY;
1057 /* handle PROT_EXEC implication by PROT_READ */
1058 if ((prot & PROT_READ) &&
1059 (current->personality & READ_IMPLIES_EXEC))
1063 /* allow the security API to have its say */
1064 ret = security_mmap_addr(addr);
1069 *_capabilities = capabilities;
1074 * we've determined that we can make the mapping, now translate what we
1075 * now know into VMA flags
1077 static unsigned long determine_vm_flags(struct file *file,
1079 unsigned long flags,
1080 unsigned long capabilities)
1082 unsigned long vm_flags;
1084 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1085 /* vm_flags |= mm->def_flags; */
1087 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1088 /* attempt to share read-only copies of mapped file chunks */
1089 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1090 if (file && !(prot & PROT_WRITE))
1091 vm_flags |= VM_MAYSHARE;
1093 /* overlay a shareable mapping on the backing device or inode
1094 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1096 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1097 if (flags & MAP_SHARED)
1098 vm_flags |= VM_SHARED;
1101 /* refuse to let anyone share private mappings with this process if
1102 * it's being traced - otherwise breakpoints set in it may interfere
1103 * with another untraced process
1105 if ((flags & MAP_PRIVATE) && current->ptrace)
1106 vm_flags &= ~VM_MAYSHARE;
1112 * set up a shared mapping on a file (the driver or filesystem provides and
1115 static int do_mmap_shared_file(struct vm_area_struct *vma)
1119 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1121 vma->vm_region->vm_top = vma->vm_region->vm_end;
1127 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1128 * opposed to tried but failed) so we can only give a suitable error as
1129 * it's not possible to make a private copy if MAP_SHARED was given */
1134 * set up a private mapping or an anonymous shared mapping
1136 static int do_mmap_private(struct vm_area_struct *vma,
1137 struct vm_region *region,
1139 unsigned long capabilities)
1142 unsigned long total, point, n;
1146 /* invoke the file's mapping function so that it can keep track of
1147 * shared mappings on devices or memory
1148 * - VM_MAYSHARE will be set if it may attempt to share
1150 if (capabilities & BDI_CAP_MAP_DIRECT) {
1151 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1153 /* shouldn't return success if we're not sharing */
1154 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1155 vma->vm_region->vm_top = vma->vm_region->vm_end;
1161 /* getting an ENOSYS error indicates that direct mmap isn't
1162 * possible (as opposed to tried but failed) so we'll try to
1163 * make a private copy of the data and map that instead */
1167 /* allocate some memory to hold the mapping
1168 * - note that this may not return a page-aligned address if the object
1169 * we're allocating is smaller than a page
1171 order = get_order(len);
1172 kdebug("alloc order %d for %lx", order, len);
1174 pages = alloc_pages(GFP_KERNEL, order);
1179 atomic_long_add(total, &mmap_pages_allocated);
1181 point = len >> PAGE_SHIFT;
1183 /* we allocated a power-of-2 sized page set, so we may want to trim off
1185 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1186 while (total > point) {
1187 order = ilog2(total - point);
1189 kdebug("shave %lu/%lu @%lu", n, total - point, total);
1190 atomic_long_sub(n, &mmap_pages_allocated);
1192 set_page_refcounted(pages + total);
1193 __free_pages(pages + total, order);
1197 for (point = 1; point < total; point++)
1198 set_page_refcounted(&pages[point]);
1200 base = page_address(pages);
1201 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1202 region->vm_start = (unsigned long) base;
1203 region->vm_end = region->vm_start + len;
1204 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1206 vma->vm_start = region->vm_start;
1207 vma->vm_end = region->vm_start + len;
1210 /* read the contents of a file into the copy */
1211 mm_segment_t old_fs;
1214 fpos = vma->vm_pgoff;
1215 fpos <<= PAGE_SHIFT;
1219 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
1225 /* clear the last little bit */
1227 memset(base + ret, 0, len - ret);
1234 free_page_series(region->vm_start, region->vm_top);
1235 region->vm_start = vma->vm_start = 0;
1236 region->vm_end = vma->vm_end = 0;
1241 printk("Allocation of length %lu from process %d (%s) failed\n",
1242 len, current->pid, current->comm);
1248 * handle mapping creation for uClinux
1250 unsigned long do_mmap_pgoff(struct file *file,
1254 unsigned long flags,
1255 unsigned long pgoff,
1256 unsigned long *populate)
1258 struct vm_area_struct *vma;
1259 struct vm_region *region;
1261 unsigned long capabilities, vm_flags, result;
1264 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1268 /* decide whether we should attempt the mapping, and if so what sort of
1270 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1273 kleave(" = %d [val]", ret);
1277 /* we ignore the address hint */
1279 len = PAGE_ALIGN(len);
1281 /* we've determined that we can make the mapping, now translate what we
1282 * now know into VMA flags */
1283 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1285 /* we're going to need to record the mapping */
1286 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1288 goto error_getting_region;
1290 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1292 goto error_getting_vma;
1294 region->vm_usage = 1;
1295 region->vm_flags = vm_flags;
1296 region->vm_pgoff = pgoff;
1298 INIT_LIST_HEAD(&vma->anon_vma_chain);
1299 vma->vm_flags = vm_flags;
1300 vma->vm_pgoff = pgoff;
1303 region->vm_file = get_file(file);
1304 vma->vm_file = get_file(file);
1307 down_write(&nommu_region_sem);
1309 /* if we want to share, we need to check for regions created by other
1310 * mmap() calls that overlap with our proposed mapping
1311 * - we can only share with a superset match on most regular files
1312 * - shared mappings on character devices and memory backed files are
1313 * permitted to overlap inexactly as far as we are concerned for in
1314 * these cases, sharing is handled in the driver or filesystem rather
1317 if (vm_flags & VM_MAYSHARE) {
1318 struct vm_region *pregion;
1319 unsigned long pglen, rpglen, pgend, rpgend, start;
1321 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1322 pgend = pgoff + pglen;
1324 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1325 pregion = rb_entry(rb, struct vm_region, vm_rb);
1327 if (!(pregion->vm_flags & VM_MAYSHARE))
1330 /* search for overlapping mappings on the same file */
1331 if (file_inode(pregion->vm_file) !=
1335 if (pregion->vm_pgoff >= pgend)
1338 rpglen = pregion->vm_end - pregion->vm_start;
1339 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1340 rpgend = pregion->vm_pgoff + rpglen;
1341 if (pgoff >= rpgend)
1344 /* handle inexactly overlapping matches between
1346 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1347 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1348 /* new mapping is not a subset of the region */
1349 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1350 goto sharing_violation;
1354 /* we've found a region we can share */
1355 pregion->vm_usage++;
1356 vma->vm_region = pregion;
1357 start = pregion->vm_start;
1358 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1359 vma->vm_start = start;
1360 vma->vm_end = start + len;
1362 if (pregion->vm_flags & VM_MAPPED_COPY) {
1363 kdebug("share copy");
1364 vma->vm_flags |= VM_MAPPED_COPY;
1366 kdebug("share mmap");
1367 ret = do_mmap_shared_file(vma);
1369 vma->vm_region = NULL;
1372 pregion->vm_usage--;
1374 goto error_just_free;
1377 fput(region->vm_file);
1378 kmem_cache_free(vm_region_jar, region);
1384 /* obtain the address at which to make a shared mapping
1385 * - this is the hook for quasi-memory character devices to
1386 * tell us the location of a shared mapping
1388 if (capabilities & BDI_CAP_MAP_DIRECT) {
1389 addr = file->f_op->get_unmapped_area(file, addr, len,
1391 if (IS_ERR_VALUE(addr)) {
1394 goto error_just_free;
1396 /* the driver refused to tell us where to site
1397 * the mapping so we'll have to attempt to copy
1400 if (!(capabilities & BDI_CAP_MAP_COPY))
1401 goto error_just_free;
1403 capabilities &= ~BDI_CAP_MAP_DIRECT;
1405 vma->vm_start = region->vm_start = addr;
1406 vma->vm_end = region->vm_end = addr + len;
1411 vma->vm_region = region;
1413 /* set up the mapping
1414 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1416 if (file && vma->vm_flags & VM_SHARED)
1417 ret = do_mmap_shared_file(vma);
1419 ret = do_mmap_private(vma, region, len, capabilities);
1421 goto error_just_free;
1422 add_nommu_region(region);
1424 /* clear anonymous mappings that don't ask for uninitialized data */
1425 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1426 memset((void *)region->vm_start, 0,
1427 region->vm_end - region->vm_start);
1429 /* okay... we have a mapping; now we have to register it */
1430 result = vma->vm_start;
1432 current->mm->total_vm += len >> PAGE_SHIFT;
1435 add_vma_to_mm(current->mm, vma);
1437 /* we flush the region from the icache only when the first executable
1438 * mapping of it is made */
1439 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1440 flush_icache_range(region->vm_start, region->vm_end);
1441 region->vm_icache_flushed = true;
1444 up_write(&nommu_region_sem);
1446 kleave(" = %lx", result);
1450 up_write(&nommu_region_sem);
1452 if (region->vm_file)
1453 fput(region->vm_file);
1454 kmem_cache_free(vm_region_jar, region);
1457 kmem_cache_free(vm_area_cachep, vma);
1458 kleave(" = %d", ret);
1462 up_write(&nommu_region_sem);
1463 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1468 kmem_cache_free(vm_region_jar, region);
1469 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1470 " from process %d failed\n",
1475 error_getting_region:
1476 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1477 " from process %d failed\n",
1483 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1484 unsigned long, prot, unsigned long, flags,
1485 unsigned long, fd, unsigned long, pgoff)
1487 struct file *file = NULL;
1488 unsigned long retval = -EBADF;
1490 audit_mmap_fd(fd, flags);
1491 if (!(flags & MAP_ANONYMOUS)) {
1497 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1499 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1507 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1508 struct mmap_arg_struct {
1512 unsigned long flags;
1514 unsigned long offset;
1517 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1519 struct mmap_arg_struct a;
1521 if (copy_from_user(&a, arg, sizeof(a)))
1523 if (a.offset & ~PAGE_MASK)
1526 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1527 a.offset >> PAGE_SHIFT);
1529 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1532 * split a vma into two pieces at address 'addr', a new vma is allocated either
1533 * for the first part or the tail.
1535 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1536 unsigned long addr, int new_below)
1538 struct vm_area_struct *new;
1539 struct vm_region *region;
1540 unsigned long npages;
1544 /* we're only permitted to split anonymous regions (these should have
1545 * only a single usage on the region) */
1549 if (mm->map_count >= sysctl_max_map_count)
1552 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1556 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1558 kmem_cache_free(vm_region_jar, region);
1562 /* most fields are the same, copy all, and then fixup */
1564 *region = *vma->vm_region;
1565 new->vm_region = region;
1567 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1570 region->vm_top = region->vm_end = new->vm_end = addr;
1572 region->vm_start = new->vm_start = addr;
1573 region->vm_pgoff = new->vm_pgoff += npages;
1576 if (new->vm_ops && new->vm_ops->open)
1577 new->vm_ops->open(new);
1579 delete_vma_from_mm(vma);
1580 down_write(&nommu_region_sem);
1581 delete_nommu_region(vma->vm_region);
1583 vma->vm_region->vm_start = vma->vm_start = addr;
1584 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1586 vma->vm_region->vm_end = vma->vm_end = addr;
1587 vma->vm_region->vm_top = addr;
1589 add_nommu_region(vma->vm_region);
1590 add_nommu_region(new->vm_region);
1591 up_write(&nommu_region_sem);
1592 add_vma_to_mm(mm, vma);
1593 add_vma_to_mm(mm, new);
1598 * shrink a VMA by removing the specified chunk from either the beginning or
1601 static int shrink_vma(struct mm_struct *mm,
1602 struct vm_area_struct *vma,
1603 unsigned long from, unsigned long to)
1605 struct vm_region *region;
1609 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1611 delete_vma_from_mm(vma);
1612 if (from > vma->vm_start)
1616 add_vma_to_mm(mm, vma);
1618 /* cut the backing region down to size */
1619 region = vma->vm_region;
1620 BUG_ON(region->vm_usage != 1);
1622 down_write(&nommu_region_sem);
1623 delete_nommu_region(region);
1624 if (from > region->vm_start) {
1625 to = region->vm_top;
1626 region->vm_top = region->vm_end = from;
1628 region->vm_start = to;
1630 add_nommu_region(region);
1631 up_write(&nommu_region_sem);
1633 free_page_series(from, to);
1639 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1640 * VMA, though it need not cover the whole VMA
1642 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1644 struct vm_area_struct *vma;
1648 kenter(",%lx,%zx", start, len);
1650 len = PAGE_ALIGN(len);
1656 /* find the first potentially overlapping VMA */
1657 vma = find_vma(mm, start);
1659 static int limit = 0;
1662 "munmap of memory not mmapped by process %d"
1663 " (%s): 0x%lx-0x%lx\n",
1664 current->pid, current->comm,
1665 start, start + len - 1);
1671 /* we're allowed to split an anonymous VMA but not a file-backed one */
1674 if (start > vma->vm_start) {
1675 kleave(" = -EINVAL [miss]");
1678 if (end == vma->vm_end)
1679 goto erase_whole_vma;
1682 kleave(" = -EINVAL [split file]");
1685 /* the chunk must be a subset of the VMA found */
1686 if (start == vma->vm_start && end == vma->vm_end)
1687 goto erase_whole_vma;
1688 if (start < vma->vm_start || end > vma->vm_end) {
1689 kleave(" = -EINVAL [superset]");
1692 if (start & ~PAGE_MASK) {
1693 kleave(" = -EINVAL [unaligned start]");
1696 if (end != vma->vm_end && end & ~PAGE_MASK) {
1697 kleave(" = -EINVAL [unaligned split]");
1700 if (start != vma->vm_start && end != vma->vm_end) {
1701 ret = split_vma(mm, vma, start, 1);
1703 kleave(" = %d [split]", ret);
1707 return shrink_vma(mm, vma, start, end);
1711 delete_vma_from_mm(vma);
1712 delete_vma(mm, vma);
1716 EXPORT_SYMBOL(do_munmap);
1718 int vm_munmap(unsigned long addr, size_t len)
1720 struct mm_struct *mm = current->mm;
1723 down_write(&mm->mmap_sem);
1724 ret = do_munmap(mm, addr, len);
1725 up_write(&mm->mmap_sem);
1728 EXPORT_SYMBOL(vm_munmap);
1730 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1732 return vm_munmap(addr, len);
1736 * release all the mappings made in a process's VM space
1738 void exit_mmap(struct mm_struct *mm)
1740 struct vm_area_struct *vma;
1749 while ((vma = mm->mmap)) {
1750 mm->mmap = vma->vm_next;
1751 delete_vma_from_mm(vma);
1752 delete_vma(mm, vma);
1759 unsigned long vm_brk(unsigned long addr, unsigned long len)
1765 * expand (or shrink) an existing mapping, potentially moving it at the same
1766 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1768 * under NOMMU conditions, we only permit changing a mapping's size, and only
1769 * as long as it stays within the region allocated by do_mmap_private() and the
1770 * block is not shareable
1772 * MREMAP_FIXED is not supported under NOMMU conditions
1774 static unsigned long do_mremap(unsigned long addr,
1775 unsigned long old_len, unsigned long new_len,
1776 unsigned long flags, unsigned long new_addr)
1778 struct vm_area_struct *vma;
1780 /* insanity checks first */
1781 old_len = PAGE_ALIGN(old_len);
1782 new_len = PAGE_ALIGN(new_len);
1783 if (old_len == 0 || new_len == 0)
1784 return (unsigned long) -EINVAL;
1786 if (addr & ~PAGE_MASK)
1789 if (flags & MREMAP_FIXED && new_addr != addr)
1790 return (unsigned long) -EINVAL;
1792 vma = find_vma_exact(current->mm, addr, old_len);
1794 return (unsigned long) -EINVAL;
1796 if (vma->vm_end != vma->vm_start + old_len)
1797 return (unsigned long) -EFAULT;
1799 if (vma->vm_flags & VM_MAYSHARE)
1800 return (unsigned long) -EPERM;
1802 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1803 return (unsigned long) -ENOMEM;
1805 /* all checks complete - do it */
1806 vma->vm_end = vma->vm_start + new_len;
1807 return vma->vm_start;
1810 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1811 unsigned long, new_len, unsigned long, flags,
1812 unsigned long, new_addr)
1816 down_write(¤t->mm->mmap_sem);
1817 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1818 up_write(¤t->mm->mmap_sem);
1822 struct page *follow_page_mask(struct vm_area_struct *vma,
1823 unsigned long address, unsigned int flags,
1824 unsigned int *page_mask)
1830 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1831 unsigned long pfn, unsigned long size, pgprot_t prot)
1833 if (addr != (pfn << PAGE_SHIFT))
1836 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1839 EXPORT_SYMBOL(remap_pfn_range);
1841 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1843 unsigned long pfn = start >> PAGE_SHIFT;
1844 unsigned long vm_len = vma->vm_end - vma->vm_start;
1846 pfn += vma->vm_pgoff;
1847 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1849 EXPORT_SYMBOL(vm_iomap_memory);
1851 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1852 unsigned long pgoff)
1854 unsigned int size = vma->vm_end - vma->vm_start;
1856 if (!(vma->vm_flags & VM_USERMAP))
1859 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1860 vma->vm_end = vma->vm_start + size;
1864 EXPORT_SYMBOL(remap_vmalloc_range);
1866 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1867 unsigned long len, unsigned long pgoff, unsigned long flags)
1872 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1876 void unmap_mapping_range(struct address_space *mapping,
1877 loff_t const holebegin, loff_t const holelen,
1881 EXPORT_SYMBOL(unmap_mapping_range);
1884 * Check that a process has enough memory to allocate a new virtual
1885 * mapping. 0 means there is enough memory for the allocation to
1886 * succeed and -ENOMEM implies there is not.
1888 * We currently support three overcommit policies, which are set via the
1889 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1891 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1892 * Additional code 2002 Jul 20 by Robert Love.
1894 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1896 * Note this is a helper function intended to be used by LSMs which
1897 * wish to use this logic.
1899 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1901 unsigned long free, allowed, reserve;
1903 vm_acct_memory(pages);
1906 * Sometimes we want to use more memory than we have
1908 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1911 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1912 free = global_page_state(NR_FREE_PAGES);
1913 free += global_page_state(NR_FILE_PAGES);
1916 * shmem pages shouldn't be counted as free in this
1917 * case, they can't be purged, only swapped out, and
1918 * that won't affect the overall amount of available
1919 * memory in the system.
1921 free -= global_page_state(NR_SHMEM);
1923 free += get_nr_swap_pages();
1926 * Any slabs which are created with the
1927 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1928 * which are reclaimable, under pressure. The dentry
1929 * cache and most inode caches should fall into this
1931 free += global_page_state(NR_SLAB_RECLAIMABLE);
1934 * Leave reserved pages. The pages are not for anonymous pages.
1936 if (free <= totalreserve_pages)
1939 free -= totalreserve_pages;
1942 * Reserve some for root
1945 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1953 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1955 * Reserve some 3% for root
1958 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1959 allowed += total_swap_pages;
1962 * Don't let a single process grow so big a user can't recover
1965 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1966 allowed -= min(mm->total_vm / 32, reserve);
1969 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1973 vm_unacct_memory(pages);
1978 int in_gate_area_no_mm(unsigned long addr)
1983 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1988 EXPORT_SYMBOL(filemap_fault);
1990 int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr,
1991 unsigned long size, pgoff_t pgoff)
1996 EXPORT_SYMBOL(generic_file_remap_pages);
1998 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1999 unsigned long addr, void *buf, int len, int write)
2001 struct vm_area_struct *vma;
2003 down_read(&mm->mmap_sem);
2005 /* the access must start within one of the target process's mappings */
2006 vma = find_vma(mm, addr);
2008 /* don't overrun this mapping */
2009 if (addr + len >= vma->vm_end)
2010 len = vma->vm_end - addr;
2012 /* only read or write mappings where it is permitted */
2013 if (write && vma->vm_flags & VM_MAYWRITE)
2014 copy_to_user_page(vma, NULL, addr,
2015 (void *) addr, buf, len);
2016 else if (!write && vma->vm_flags & VM_MAYREAD)
2017 copy_from_user_page(vma, NULL, addr,
2018 buf, (void *) addr, len);
2025 up_read(&mm->mmap_sem);
2031 * @access_remote_vm - access another process' address space
2032 * @mm: the mm_struct of the target address space
2033 * @addr: start address to access
2034 * @buf: source or destination buffer
2035 * @len: number of bytes to transfer
2036 * @write: whether the access is a write
2038 * The caller must hold a reference on @mm.
2040 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2041 void *buf, int len, int write)
2043 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2047 * Access another process' address space.
2048 * - source/target buffer must be kernel space
2050 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2052 struct mm_struct *mm;
2054 if (addr + len < addr)
2057 mm = get_task_mm(tsk);
2061 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2068 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2069 * @inode: The inode to check
2070 * @size: The current filesize of the inode
2071 * @newsize: The proposed filesize of the inode
2073 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2074 * make sure that that any outstanding VMAs aren't broken and then shrink the
2075 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2076 * automatically grant mappings that are too large.
2078 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2081 struct vm_area_struct *vma;
2082 struct vm_region *region;
2084 size_t r_size, r_top;
2086 low = newsize >> PAGE_SHIFT;
2087 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2089 down_write(&nommu_region_sem);
2090 mutex_lock(&inode->i_mapping->i_mmap_mutex);
2092 /* search for VMAs that fall within the dead zone */
2093 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
2094 /* found one - only interested if it's shared out of the page
2096 if (vma->vm_flags & VM_SHARED) {
2097 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2098 up_write(&nommu_region_sem);
2099 return -ETXTBSY; /* not quite true, but near enough */
2103 /* reduce any regions that overlap the dead zone - if in existence,
2104 * these will be pointed to by VMAs that don't overlap the dead zone
2106 * we don't check for any regions that start beyond the EOF as there
2109 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap,
2111 if (!(vma->vm_flags & VM_SHARED))
2114 region = vma->vm_region;
2115 r_size = region->vm_top - region->vm_start;
2116 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2118 if (r_top > newsize) {
2119 region->vm_top -= r_top - newsize;
2120 if (region->vm_end > region->vm_top)
2121 region->vm_end = region->vm_top;
2125 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2126 up_write(&nommu_region_sem);
2131 * Initialise sysctl_user_reserve_kbytes.
2133 * This is intended to prevent a user from starting a single memory hogging
2134 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2137 * The default value is min(3% of free memory, 128MB)
2138 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2140 static int __meminit init_user_reserve(void)
2142 unsigned long free_kbytes;
2144 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2146 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2149 module_init(init_user_reserve)
2152 * Initialise sysctl_admin_reserve_kbytes.
2154 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2155 * to log in and kill a memory hogging process.
2157 * Systems with more than 256MB will reserve 8MB, enough to recover
2158 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2159 * only reserve 3% of free pages by default.
2161 static int __meminit init_admin_reserve(void)
2163 unsigned long free_kbytes;
2165 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2167 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2170 module_init(init_admin_reserve)