4 * Replacement code for mm functions to support CPU's that don't
5 * have any form of memory management unit (thus no virtual memory).
7 * See Documentation/nommu-mmap.txt
9 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
10 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
13 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18 #include <linux/export.h>
20 #include <linux/vmacache.h>
21 #include <linux/mman.h>
22 #include <linux/swap.h>
23 #include <linux/file.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <linux/slab.h>
27 #include <linux/vmalloc.h>
28 #include <linux/blkdev.h>
29 #include <linux/backing-dev.h>
30 #include <linux/compiler.h>
31 #include <linux/mount.h>
32 #include <linux/personality.h>
33 #include <linux/security.h>
34 #include <linux/syscalls.h>
35 #include <linux/audit.h>
36 #include <linux/sched/sysctl.h>
37 #include <linux/printk.h>
39 #include <asm/uaccess.h>
41 #include <asm/tlbflush.h>
42 #include <asm/mmu_context.h>
46 EXPORT_SYMBOL(high_memory);
48 unsigned long max_mapnr;
49 EXPORT_SYMBOL(max_mapnr);
50 unsigned long highest_memmap_pfn;
51 struct percpu_counter vm_committed_as;
52 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
53 int sysctl_overcommit_ratio = 50; /* default is 50% */
54 unsigned long sysctl_overcommit_kbytes __read_mostly;
55 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
56 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
57 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
58 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
59 int heap_stack_gap = 0;
61 atomic_long_t mmap_pages_allocated;
64 * The global memory commitment made in the system can be a metric
65 * that can be used to drive ballooning decisions when Linux is hosted
66 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
67 * balancing memory across competing virtual machines that are hosted.
68 * Several metrics drive this policy engine including the guest reported
71 unsigned long vm_memory_committed(void)
73 return percpu_counter_read_positive(&vm_committed_as);
76 EXPORT_SYMBOL_GPL(vm_memory_committed);
78 EXPORT_SYMBOL(mem_map);
80 /* list of mapped, potentially shareable regions */
81 static struct kmem_cache *vm_region_jar;
82 struct rb_root nommu_region_tree = RB_ROOT;
83 DECLARE_RWSEM(nommu_region_sem);
85 const struct vm_operations_struct generic_file_vm_ops = {
89 * Return the total memory allocated for this pointer, not
90 * just what the caller asked for.
92 * Doesn't have to be accurate, i.e. may have races.
94 unsigned int kobjsize(const void *objp)
99 * If the object we have should not have ksize performed on it,
102 if (!objp || !virt_addr_valid(objp))
105 page = virt_to_head_page(objp);
108 * If the allocator sets PageSlab, we know the pointer came from
115 * If it's not a compound page, see if we have a matching VMA
116 * region. This test is intentionally done in reverse order,
117 * so if there's no VMA, we still fall through and hand back
118 * PAGE_SIZE for 0-order pages.
120 if (!PageCompound(page)) {
121 struct vm_area_struct *vma;
123 vma = find_vma(current->mm, (unsigned long)objp);
125 return vma->vm_end - vma->vm_start;
129 * The ksize() function is only guaranteed to work for pointers
130 * returned by kmalloc(). So handle arbitrary pointers here.
132 return PAGE_SIZE << compound_order(page);
135 long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
136 unsigned long start, unsigned long nr_pages,
137 unsigned int foll_flags, struct page **pages,
138 struct vm_area_struct **vmas, int *nonblocking)
140 struct vm_area_struct *vma;
141 unsigned long vm_flags;
144 /* calculate required read or write permissions.
145 * If FOLL_FORCE is set, we only require the "MAY" flags.
147 vm_flags = (foll_flags & FOLL_WRITE) ?
148 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
149 vm_flags &= (foll_flags & FOLL_FORCE) ?
150 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
152 for (i = 0; i < nr_pages; i++) {
153 vma = find_vma(mm, start);
155 goto finish_or_fault;
157 /* protect what we can, including chardevs */
158 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
159 !(vm_flags & vma->vm_flags))
160 goto finish_or_fault;
163 pages[i] = virt_to_page(start);
165 page_cache_get(pages[i]);
169 start = (start + PAGE_SIZE) & PAGE_MASK;
175 return i ? : -EFAULT;
179 * get a list of pages in an address range belonging to the specified process
180 * and indicate the VMA that covers each page
181 * - this is potentially dodgy as we may end incrementing the page count of a
182 * slab page or a secondary page from a compound page
183 * - don't permit access to VMAs that don't support it, such as I/O mappings
185 long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
186 unsigned long start, unsigned long nr_pages,
187 int write, int force, struct page **pages,
188 struct vm_area_struct **vmas)
197 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
200 EXPORT_SYMBOL(get_user_pages);
202 long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm,
203 unsigned long start, unsigned long nr_pages,
204 int write, int force, struct page **pages,
207 return get_user_pages(tsk, mm, start, nr_pages, write, force,
210 EXPORT_SYMBOL(get_user_pages_locked);
212 long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
213 unsigned long start, unsigned long nr_pages,
214 int write, int force, struct page **pages,
215 unsigned int gup_flags)
218 down_read(&mm->mmap_sem);
219 ret = get_user_pages(tsk, mm, start, nr_pages, write, force,
221 up_read(&mm->mmap_sem);
224 EXPORT_SYMBOL(__get_user_pages_unlocked);
226 long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
227 unsigned long start, unsigned long nr_pages,
228 int write, int force, struct page **pages)
230 return __get_user_pages_unlocked(tsk, mm, start, nr_pages, write,
233 EXPORT_SYMBOL(get_user_pages_unlocked);
236 * follow_pfn - look up PFN at a user virtual address
237 * @vma: memory mapping
238 * @address: user virtual address
239 * @pfn: location to store found PFN
241 * Only IO mappings and raw PFN mappings are allowed.
243 * Returns zero and the pfn at @pfn on success, -ve otherwise.
245 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
248 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
251 *pfn = address >> PAGE_SHIFT;
254 EXPORT_SYMBOL(follow_pfn);
256 LIST_HEAD(vmap_area_list);
258 void vfree(const void *addr)
262 EXPORT_SYMBOL(vfree);
264 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
267 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
268 * returns only a logical address.
270 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
272 EXPORT_SYMBOL(__vmalloc);
274 void *vmalloc_user(unsigned long size)
278 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
281 struct vm_area_struct *vma;
283 down_write(¤t->mm->mmap_sem);
284 vma = find_vma(current->mm, (unsigned long)ret);
286 vma->vm_flags |= VM_USERMAP;
287 up_write(¤t->mm->mmap_sem);
292 EXPORT_SYMBOL(vmalloc_user);
294 struct page *vmalloc_to_page(const void *addr)
296 return virt_to_page(addr);
298 EXPORT_SYMBOL(vmalloc_to_page);
300 unsigned long vmalloc_to_pfn(const void *addr)
302 return page_to_pfn(virt_to_page(addr));
304 EXPORT_SYMBOL(vmalloc_to_pfn);
306 long vread(char *buf, char *addr, unsigned long count)
308 /* Don't allow overflow */
309 if ((unsigned long) buf + count < count)
310 count = -(unsigned long) buf;
312 memcpy(buf, addr, count);
316 long vwrite(char *buf, char *addr, unsigned long count)
318 /* Don't allow overflow */
319 if ((unsigned long) addr + count < count)
320 count = -(unsigned long) addr;
322 memcpy(addr, buf, count);
327 * vmalloc - allocate virtually contiguous memory
329 * @size: allocation size
331 * Allocate enough pages to cover @size from the page level
332 * allocator and map them into contiguous kernel virtual space.
334 * For tight control over page level allocator and protection flags
335 * use __vmalloc() instead.
337 void *vmalloc(unsigned long size)
339 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
341 EXPORT_SYMBOL(vmalloc);
344 * vzalloc - allocate virtually contiguous memory with zero fill
346 * @size: allocation size
348 * Allocate enough pages to cover @size from the page level
349 * allocator and map them into contiguous kernel virtual space.
350 * The memory allocated is set to zero.
352 * For tight control over page level allocator and protection flags
353 * use __vmalloc() instead.
355 void *vzalloc(unsigned long size)
357 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
360 EXPORT_SYMBOL(vzalloc);
363 * vmalloc_node - allocate memory on a specific node
364 * @size: allocation size
367 * Allocate enough pages to cover @size from the page level
368 * allocator and map them into contiguous kernel virtual space.
370 * For tight control over page level allocator and protection flags
371 * use __vmalloc() instead.
373 void *vmalloc_node(unsigned long size, int node)
375 return vmalloc(size);
377 EXPORT_SYMBOL(vmalloc_node);
380 * vzalloc_node - allocate memory on a specific node with zero fill
381 * @size: allocation size
384 * Allocate enough pages to cover @size from the page level
385 * allocator and map them into contiguous kernel virtual space.
386 * The memory allocated is set to zero.
388 * For tight control over page level allocator and protection flags
389 * use __vmalloc() instead.
391 void *vzalloc_node(unsigned long size, int node)
393 return vzalloc(size);
395 EXPORT_SYMBOL(vzalloc_node);
397 #ifndef PAGE_KERNEL_EXEC
398 # define PAGE_KERNEL_EXEC PAGE_KERNEL
402 * vmalloc_exec - allocate virtually contiguous, executable memory
403 * @size: allocation size
405 * Kernel-internal function to allocate enough pages to cover @size
406 * the page level allocator and map them into contiguous and
407 * executable kernel virtual space.
409 * For tight control over page level allocator and protection flags
410 * use __vmalloc() instead.
413 void *vmalloc_exec(unsigned long size)
415 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
419 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
420 * @size: allocation size
422 * Allocate enough 32bit PA addressable pages to cover @size from the
423 * page level allocator and map them into contiguous kernel virtual space.
425 void *vmalloc_32(unsigned long size)
427 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
429 EXPORT_SYMBOL(vmalloc_32);
432 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
433 * @size: allocation size
435 * The resulting memory area is 32bit addressable and zeroed so it can be
436 * mapped to userspace without leaking data.
438 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
439 * remap_vmalloc_range() are permissible.
441 void *vmalloc_32_user(unsigned long size)
444 * We'll have to sort out the ZONE_DMA bits for 64-bit,
445 * but for now this can simply use vmalloc_user() directly.
447 return vmalloc_user(size);
449 EXPORT_SYMBOL(vmalloc_32_user);
451 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
458 void vunmap(const void *addr)
462 EXPORT_SYMBOL(vunmap);
464 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
469 EXPORT_SYMBOL(vm_map_ram);
471 void vm_unmap_ram(const void *mem, unsigned int count)
475 EXPORT_SYMBOL(vm_unmap_ram);
477 void vm_unmap_aliases(void)
480 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
483 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
486 void __weak vmalloc_sync_all(void)
491 * alloc_vm_area - allocate a range of kernel address space
492 * @size: size of the area
494 * Returns: NULL on failure, vm_struct on success
496 * This function reserves a range of kernel address space, and
497 * allocates pagetables to map that range. No actual mappings
498 * are created. If the kernel address space is not shared
499 * between processes, it syncs the pagetable across all
502 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
507 EXPORT_SYMBOL_GPL(alloc_vm_area);
509 void free_vm_area(struct vm_struct *area)
513 EXPORT_SYMBOL_GPL(free_vm_area);
515 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
520 EXPORT_SYMBOL(vm_insert_page);
523 * sys_brk() for the most part doesn't need the global kernel
524 * lock, except when an application is doing something nasty
525 * like trying to un-brk an area that has already been mapped
526 * to a regular file. in this case, the unmapping will need
527 * to invoke file system routines that need the global lock.
529 SYSCALL_DEFINE1(brk, unsigned long, brk)
531 struct mm_struct *mm = current->mm;
533 if (brk < mm->start_brk || brk > mm->context.end_brk)
540 * Always allow shrinking brk
542 if (brk <= mm->brk) {
548 * Ok, looks good - let it rip.
550 flush_icache_range(mm->brk, brk);
551 return mm->brk = brk;
555 * initialise the VMA and region record slabs
557 void __init mmap_init(void)
561 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
563 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
567 * validate the region tree
568 * - the caller must hold the region lock
570 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
571 static noinline void validate_nommu_regions(void)
573 struct vm_region *region, *last;
574 struct rb_node *p, *lastp;
576 lastp = rb_first(&nommu_region_tree);
580 last = rb_entry(lastp, struct vm_region, vm_rb);
581 BUG_ON(last->vm_end <= last->vm_start);
582 BUG_ON(last->vm_top < last->vm_end);
584 while ((p = rb_next(lastp))) {
585 region = rb_entry(p, struct vm_region, vm_rb);
586 last = rb_entry(lastp, struct vm_region, vm_rb);
588 BUG_ON(region->vm_end <= region->vm_start);
589 BUG_ON(region->vm_top < region->vm_end);
590 BUG_ON(region->vm_start < last->vm_top);
596 static void validate_nommu_regions(void)
602 * add a region into the global tree
604 static void add_nommu_region(struct vm_region *region)
606 struct vm_region *pregion;
607 struct rb_node **p, *parent;
609 validate_nommu_regions();
612 p = &nommu_region_tree.rb_node;
615 pregion = rb_entry(parent, struct vm_region, vm_rb);
616 if (region->vm_start < pregion->vm_start)
618 else if (region->vm_start > pregion->vm_start)
620 else if (pregion == region)
626 rb_link_node(®ion->vm_rb, parent, p);
627 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
629 validate_nommu_regions();
633 * delete a region from the global tree
635 static void delete_nommu_region(struct vm_region *region)
637 BUG_ON(!nommu_region_tree.rb_node);
639 validate_nommu_regions();
640 rb_erase(®ion->vm_rb, &nommu_region_tree);
641 validate_nommu_regions();
645 * free a contiguous series of pages
647 static void free_page_series(unsigned long from, unsigned long to)
649 for (; from < to; from += PAGE_SIZE) {
650 struct page *page = virt_to_page(from);
652 atomic_long_dec(&mmap_pages_allocated);
658 * release a reference to a region
659 * - the caller must hold the region semaphore for writing, which this releases
660 * - the region may not have been added to the tree yet, in which case vm_top
661 * will equal vm_start
663 static void __put_nommu_region(struct vm_region *region)
664 __releases(nommu_region_sem)
666 BUG_ON(!nommu_region_tree.rb_node);
668 if (--region->vm_usage == 0) {
669 if (region->vm_top > region->vm_start)
670 delete_nommu_region(region);
671 up_write(&nommu_region_sem);
674 fput(region->vm_file);
676 /* IO memory and memory shared directly out of the pagecache
677 * from ramfs/tmpfs mustn't be released here */
678 if (region->vm_flags & VM_MAPPED_COPY)
679 free_page_series(region->vm_start, region->vm_top);
680 kmem_cache_free(vm_region_jar, region);
682 up_write(&nommu_region_sem);
687 * release a reference to a region
689 static void put_nommu_region(struct vm_region *region)
691 down_write(&nommu_region_sem);
692 __put_nommu_region(region);
696 * update protection on a vma
698 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
701 struct mm_struct *mm = vma->vm_mm;
702 long start = vma->vm_start & PAGE_MASK;
703 while (start < vma->vm_end) {
704 protect_page(mm, start, flags);
707 update_protections(mm);
712 * add a VMA into a process's mm_struct in the appropriate place in the list
713 * and tree and add to the address space's page tree also if not an anonymous
715 * - should be called with mm->mmap_sem held writelocked
717 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
719 struct vm_area_struct *pvma, *prev;
720 struct address_space *mapping;
721 struct rb_node **p, *parent, *rb_prev;
723 BUG_ON(!vma->vm_region);
728 protect_vma(vma, vma->vm_flags);
730 /* add the VMA to the mapping */
732 mapping = vma->vm_file->f_mapping;
734 i_mmap_lock_write(mapping);
735 flush_dcache_mmap_lock(mapping);
736 vma_interval_tree_insert(vma, &mapping->i_mmap);
737 flush_dcache_mmap_unlock(mapping);
738 i_mmap_unlock_write(mapping);
741 /* add the VMA to the tree */
742 parent = rb_prev = NULL;
743 p = &mm->mm_rb.rb_node;
746 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
748 /* sort by: start addr, end addr, VMA struct addr in that order
749 * (the latter is necessary as we may get identical VMAs) */
750 if (vma->vm_start < pvma->vm_start)
752 else if (vma->vm_start > pvma->vm_start) {
755 } else if (vma->vm_end < pvma->vm_end)
757 else if (vma->vm_end > pvma->vm_end) {
760 } else if (vma < pvma)
762 else if (vma > pvma) {
769 rb_link_node(&vma->vm_rb, parent, p);
770 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
772 /* add VMA to the VMA list also */
775 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
777 __vma_link_list(mm, vma, prev, parent);
781 * delete a VMA from its owning mm_struct and address space
783 static void delete_vma_from_mm(struct vm_area_struct *vma)
786 struct address_space *mapping;
787 struct mm_struct *mm = vma->vm_mm;
788 struct task_struct *curr = current;
793 for (i = 0; i < VMACACHE_SIZE; i++) {
794 /* if the vma is cached, invalidate the entire cache */
795 if (curr->vmacache[i] == vma) {
796 vmacache_invalidate(mm);
801 /* remove the VMA from the mapping */
803 mapping = vma->vm_file->f_mapping;
805 i_mmap_lock_write(mapping);
806 flush_dcache_mmap_lock(mapping);
807 vma_interval_tree_remove(vma, &mapping->i_mmap);
808 flush_dcache_mmap_unlock(mapping);
809 i_mmap_unlock_write(mapping);
812 /* remove from the MM's tree and list */
813 rb_erase(&vma->vm_rb, &mm->mm_rb);
816 vma->vm_prev->vm_next = vma->vm_next;
818 mm->mmap = vma->vm_next;
821 vma->vm_next->vm_prev = vma->vm_prev;
825 * destroy a VMA record
827 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
829 if (vma->vm_ops && vma->vm_ops->close)
830 vma->vm_ops->close(vma);
833 put_nommu_region(vma->vm_region);
834 kmem_cache_free(vm_area_cachep, vma);
838 * look up the first VMA in which addr resides, NULL if none
839 * - should be called with mm->mmap_sem at least held readlocked
841 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
843 struct vm_area_struct *vma;
845 /* check the cache first */
846 vma = vmacache_find(mm, addr);
850 /* trawl the list (there may be multiple mappings in which addr
852 for (vma = mm->mmap; vma; vma = vma->vm_next) {
853 if (vma->vm_start > addr)
855 if (vma->vm_end > addr) {
856 vmacache_update(addr, vma);
863 EXPORT_SYMBOL(find_vma);
867 * - we don't extend stack VMAs under NOMMU conditions
869 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
871 return find_vma(mm, addr);
875 * expand a stack to a given address
876 * - not supported under NOMMU conditions
878 int expand_stack(struct vm_area_struct *vma, unsigned long address)
884 * look up the first VMA exactly that exactly matches addr
885 * - should be called with mm->mmap_sem at least held readlocked
887 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
891 struct vm_area_struct *vma;
892 unsigned long end = addr + len;
894 /* check the cache first */
895 vma = vmacache_find_exact(mm, addr, end);
899 /* trawl the list (there may be multiple mappings in which addr
901 for (vma = mm->mmap; vma; vma = vma->vm_next) {
902 if (vma->vm_start < addr)
904 if (vma->vm_start > addr)
906 if (vma->vm_end == end) {
907 vmacache_update(addr, vma);
916 * determine whether a mapping should be permitted and, if so, what sort of
917 * mapping we're capable of supporting
919 static int validate_mmap_request(struct file *file,
925 unsigned long *_capabilities)
927 unsigned long capabilities, rlen;
930 /* do the simple checks first */
931 if (flags & MAP_FIXED)
934 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
935 (flags & MAP_TYPE) != MAP_SHARED)
941 /* Careful about overflows.. */
942 rlen = PAGE_ALIGN(len);
943 if (!rlen || rlen > TASK_SIZE)
946 /* offset overflow? */
947 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
951 /* files must support mmap */
952 if (!file->f_op->mmap)
955 /* work out if what we've got could possibly be shared
956 * - we support chardevs that provide their own "memory"
957 * - we support files/blockdevs that are memory backed
959 if (file->f_op->mmap_capabilities) {
960 capabilities = file->f_op->mmap_capabilities(file);
962 /* no explicit capabilities set, so assume some
964 switch (file_inode(file)->i_mode & S_IFMT) {
967 capabilities = NOMMU_MAP_COPY;
982 /* eliminate any capabilities that we can't support on this
984 if (!file->f_op->get_unmapped_area)
985 capabilities &= ~NOMMU_MAP_DIRECT;
986 if (!(file->f_mode & FMODE_CAN_READ))
987 capabilities &= ~NOMMU_MAP_COPY;
989 /* The file shall have been opened with read permission. */
990 if (!(file->f_mode & FMODE_READ))
993 if (flags & MAP_SHARED) {
994 /* do checks for writing, appending and locking */
995 if ((prot & PROT_WRITE) &&
996 !(file->f_mode & FMODE_WRITE))
999 if (IS_APPEND(file_inode(file)) &&
1000 (file->f_mode & FMODE_WRITE))
1003 if (locks_verify_locked(file))
1006 if (!(capabilities & NOMMU_MAP_DIRECT))
1009 /* we mustn't privatise shared mappings */
1010 capabilities &= ~NOMMU_MAP_COPY;
1012 /* we're going to read the file into private memory we
1014 if (!(capabilities & NOMMU_MAP_COPY))
1017 /* we don't permit a private writable mapping to be
1018 * shared with the backing device */
1019 if (prot & PROT_WRITE)
1020 capabilities &= ~NOMMU_MAP_DIRECT;
1023 if (capabilities & NOMMU_MAP_DIRECT) {
1024 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
1025 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
1026 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
1028 capabilities &= ~NOMMU_MAP_DIRECT;
1029 if (flags & MAP_SHARED) {
1030 pr_warn("MAP_SHARED not completely supported on !MMU\n");
1036 /* handle executable mappings and implied executable
1038 if (path_noexec(&file->f_path)) {
1039 if (prot & PROT_EXEC)
1041 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1042 /* handle implication of PROT_EXEC by PROT_READ */
1043 if (current->personality & READ_IMPLIES_EXEC) {
1044 if (capabilities & NOMMU_MAP_EXEC)
1047 } else if ((prot & PROT_READ) &&
1048 (prot & PROT_EXEC) &&
1049 !(capabilities & NOMMU_MAP_EXEC)
1051 /* backing file is not executable, try to copy */
1052 capabilities &= ~NOMMU_MAP_DIRECT;
1055 /* anonymous mappings are always memory backed and can be
1058 capabilities = NOMMU_MAP_COPY;
1060 /* handle PROT_EXEC implication by PROT_READ */
1061 if ((prot & PROT_READ) &&
1062 (current->personality & READ_IMPLIES_EXEC))
1066 /* allow the security API to have its say */
1067 ret = security_mmap_addr(addr);
1072 *_capabilities = capabilities;
1077 * we've determined that we can make the mapping, now translate what we
1078 * now know into VMA flags
1080 static unsigned long determine_vm_flags(struct file *file,
1082 unsigned long flags,
1083 unsigned long capabilities)
1085 unsigned long vm_flags;
1087 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1088 /* vm_flags |= mm->def_flags; */
1090 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1091 /* attempt to share read-only copies of mapped file chunks */
1092 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1093 if (file && !(prot & PROT_WRITE))
1094 vm_flags |= VM_MAYSHARE;
1096 /* overlay a shareable mapping on the backing device or inode
1097 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1099 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1100 if (flags & MAP_SHARED)
1101 vm_flags |= VM_SHARED;
1104 /* refuse to let anyone share private mappings with this process if
1105 * it's being traced - otherwise breakpoints set in it may interfere
1106 * with another untraced process
1108 if ((flags & MAP_PRIVATE) && current->ptrace)
1109 vm_flags &= ~VM_MAYSHARE;
1115 * set up a shared mapping on a file (the driver or filesystem provides and
1118 static int do_mmap_shared_file(struct vm_area_struct *vma)
1122 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1124 vma->vm_region->vm_top = vma->vm_region->vm_end;
1130 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1131 * opposed to tried but failed) so we can only give a suitable error as
1132 * it's not possible to make a private copy if MAP_SHARED was given */
1137 * set up a private mapping or an anonymous shared mapping
1139 static int do_mmap_private(struct vm_area_struct *vma,
1140 struct vm_region *region,
1142 unsigned long capabilities)
1144 unsigned long total, point;
1148 /* invoke the file's mapping function so that it can keep track of
1149 * shared mappings on devices or memory
1150 * - VM_MAYSHARE will be set if it may attempt to share
1152 if (capabilities & NOMMU_MAP_DIRECT) {
1153 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1155 /* shouldn't return success if we're not sharing */
1156 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1157 vma->vm_region->vm_top = vma->vm_region->vm_end;
1163 /* getting an ENOSYS error indicates that direct mmap isn't
1164 * possible (as opposed to tried but failed) so we'll try to
1165 * make a private copy of the data and map that instead */
1169 /* allocate some memory to hold the mapping
1170 * - note that this may not return a page-aligned address if the object
1171 * we're allocating is smaller than a page
1173 order = get_order(len);
1175 point = len >> PAGE_SHIFT;
1177 /* we don't want to allocate a power-of-2 sized page set */
1178 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1181 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1185 atomic_long_add(total, &mmap_pages_allocated);
1187 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1188 region->vm_start = (unsigned long) base;
1189 region->vm_end = region->vm_start + len;
1190 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1192 vma->vm_start = region->vm_start;
1193 vma->vm_end = region->vm_start + len;
1196 /* read the contents of a file into the copy */
1197 mm_segment_t old_fs;
1200 fpos = vma->vm_pgoff;
1201 fpos <<= PAGE_SHIFT;
1205 ret = __vfs_read(vma->vm_file, base, len, &fpos);
1211 /* clear the last little bit */
1213 memset(base + ret, 0, len - ret);
1220 free_page_series(region->vm_start, region->vm_top);
1221 region->vm_start = vma->vm_start = 0;
1222 region->vm_end = vma->vm_end = 0;
1227 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1228 len, current->pid, current->comm);
1234 * handle mapping creation for uClinux
1236 unsigned long do_mmap(struct file *file,
1240 unsigned long flags,
1241 vm_flags_t vm_flags,
1242 unsigned long pgoff,
1243 unsigned long *populate)
1245 struct vm_area_struct *vma;
1246 struct vm_region *region;
1248 unsigned long capabilities, result;
1253 /* decide whether we should attempt the mapping, and if so what sort of
1255 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1260 /* we ignore the address hint */
1262 len = PAGE_ALIGN(len);
1264 /* we've determined that we can make the mapping, now translate what we
1265 * now know into VMA flags */
1266 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1268 /* we're going to need to record the mapping */
1269 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1271 goto error_getting_region;
1273 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1275 goto error_getting_vma;
1277 region->vm_usage = 1;
1278 region->vm_flags = vm_flags;
1279 region->vm_pgoff = pgoff;
1281 INIT_LIST_HEAD(&vma->anon_vma_chain);
1282 vma->vm_flags = vm_flags;
1283 vma->vm_pgoff = pgoff;
1286 region->vm_file = get_file(file);
1287 vma->vm_file = get_file(file);
1290 down_write(&nommu_region_sem);
1292 /* if we want to share, we need to check for regions created by other
1293 * mmap() calls that overlap with our proposed mapping
1294 * - we can only share with a superset match on most regular files
1295 * - shared mappings on character devices and memory backed files are
1296 * permitted to overlap inexactly as far as we are concerned for in
1297 * these cases, sharing is handled in the driver or filesystem rather
1300 if (vm_flags & VM_MAYSHARE) {
1301 struct vm_region *pregion;
1302 unsigned long pglen, rpglen, pgend, rpgend, start;
1304 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1305 pgend = pgoff + pglen;
1307 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1308 pregion = rb_entry(rb, struct vm_region, vm_rb);
1310 if (!(pregion->vm_flags & VM_MAYSHARE))
1313 /* search for overlapping mappings on the same file */
1314 if (file_inode(pregion->vm_file) !=
1318 if (pregion->vm_pgoff >= pgend)
1321 rpglen = pregion->vm_end - pregion->vm_start;
1322 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1323 rpgend = pregion->vm_pgoff + rpglen;
1324 if (pgoff >= rpgend)
1327 /* handle inexactly overlapping matches between
1329 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1330 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1331 /* new mapping is not a subset of the region */
1332 if (!(capabilities & NOMMU_MAP_DIRECT))
1333 goto sharing_violation;
1337 /* we've found a region we can share */
1338 pregion->vm_usage++;
1339 vma->vm_region = pregion;
1340 start = pregion->vm_start;
1341 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1342 vma->vm_start = start;
1343 vma->vm_end = start + len;
1345 if (pregion->vm_flags & VM_MAPPED_COPY)
1346 vma->vm_flags |= VM_MAPPED_COPY;
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 & NOMMU_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 & NOMMU_MAP_COPY))
1382 goto error_just_free;
1384 capabilities &= ~NOMMU_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 NOMMU_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);
1430 up_write(&nommu_region_sem);
1432 if (region->vm_file)
1433 fput(region->vm_file);
1434 kmem_cache_free(vm_region_jar, region);
1437 kmem_cache_free(vm_area_cachep, vma);
1441 up_write(&nommu_region_sem);
1442 pr_warn("Attempt to share mismatched mappings\n");
1447 kmem_cache_free(vm_region_jar, region);
1448 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1453 error_getting_region:
1454 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1460 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1461 unsigned long, prot, unsigned long, flags,
1462 unsigned long, fd, unsigned long, pgoff)
1464 struct file *file = NULL;
1465 unsigned long retval = -EBADF;
1467 audit_mmap_fd(fd, flags);
1468 if (!(flags & MAP_ANONYMOUS)) {
1474 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1476 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1484 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1485 struct mmap_arg_struct {
1489 unsigned long flags;
1491 unsigned long offset;
1494 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1496 struct mmap_arg_struct a;
1498 if (copy_from_user(&a, arg, sizeof(a)))
1500 if (offset_in_page(a.offset))
1503 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1504 a.offset >> PAGE_SHIFT);
1506 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1509 * split a vma into two pieces at address 'addr', a new vma is allocated either
1510 * for the first part or the tail.
1512 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1513 unsigned long addr, int new_below)
1515 struct vm_area_struct *new;
1516 struct vm_region *region;
1517 unsigned long npages;
1519 /* we're only permitted to split anonymous regions (these should have
1520 * only a single usage on the region) */
1524 if (mm->map_count >= sysctl_max_map_count)
1527 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1531 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1533 kmem_cache_free(vm_region_jar, region);
1537 /* most fields are the same, copy all, and then fixup */
1539 *region = *vma->vm_region;
1540 new->vm_region = region;
1542 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1545 region->vm_top = region->vm_end = new->vm_end = addr;
1547 region->vm_start = new->vm_start = addr;
1548 region->vm_pgoff = new->vm_pgoff += npages;
1551 if (new->vm_ops && new->vm_ops->open)
1552 new->vm_ops->open(new);
1554 delete_vma_from_mm(vma);
1555 down_write(&nommu_region_sem);
1556 delete_nommu_region(vma->vm_region);
1558 vma->vm_region->vm_start = vma->vm_start = addr;
1559 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1561 vma->vm_region->vm_end = vma->vm_end = addr;
1562 vma->vm_region->vm_top = addr;
1564 add_nommu_region(vma->vm_region);
1565 add_nommu_region(new->vm_region);
1566 up_write(&nommu_region_sem);
1567 add_vma_to_mm(mm, vma);
1568 add_vma_to_mm(mm, new);
1573 * shrink a VMA by removing the specified chunk from either the beginning or
1576 static int shrink_vma(struct mm_struct *mm,
1577 struct vm_area_struct *vma,
1578 unsigned long from, unsigned long to)
1580 struct vm_region *region;
1582 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1584 delete_vma_from_mm(vma);
1585 if (from > vma->vm_start)
1589 add_vma_to_mm(mm, vma);
1591 /* cut the backing region down to size */
1592 region = vma->vm_region;
1593 BUG_ON(region->vm_usage != 1);
1595 down_write(&nommu_region_sem);
1596 delete_nommu_region(region);
1597 if (from > region->vm_start) {
1598 to = region->vm_top;
1599 region->vm_top = region->vm_end = from;
1601 region->vm_start = to;
1603 add_nommu_region(region);
1604 up_write(&nommu_region_sem);
1606 free_page_series(from, to);
1612 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1613 * VMA, though it need not cover the whole VMA
1615 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1617 struct vm_area_struct *vma;
1621 len = PAGE_ALIGN(len);
1627 /* find the first potentially overlapping VMA */
1628 vma = find_vma(mm, start);
1632 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1633 current->pid, current->comm,
1634 start, start + len - 1);
1640 /* we're allowed to split an anonymous VMA but not a file-backed one */
1643 if (start > vma->vm_start)
1645 if (end == vma->vm_end)
1646 goto erase_whole_vma;
1651 /* the chunk must be a subset of the VMA found */
1652 if (start == vma->vm_start && end == vma->vm_end)
1653 goto erase_whole_vma;
1654 if (start < vma->vm_start || end > vma->vm_end)
1656 if (offset_in_page(start))
1658 if (end != vma->vm_end && offset_in_page(end))
1660 if (start != vma->vm_start && end != vma->vm_end) {
1661 ret = split_vma(mm, vma, start, 1);
1665 return shrink_vma(mm, vma, start, end);
1669 delete_vma_from_mm(vma);
1670 delete_vma(mm, vma);
1673 EXPORT_SYMBOL(do_munmap);
1675 int vm_munmap(unsigned long addr, size_t len)
1677 struct mm_struct *mm = current->mm;
1680 down_write(&mm->mmap_sem);
1681 ret = do_munmap(mm, addr, len);
1682 up_write(&mm->mmap_sem);
1685 EXPORT_SYMBOL(vm_munmap);
1687 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1689 return vm_munmap(addr, len);
1693 * release all the mappings made in a process's VM space
1695 void exit_mmap(struct mm_struct *mm)
1697 struct vm_area_struct *vma;
1704 while ((vma = mm->mmap)) {
1705 mm->mmap = vma->vm_next;
1706 delete_vma_from_mm(vma);
1707 delete_vma(mm, vma);
1712 unsigned long vm_brk(unsigned long addr, unsigned long len)
1718 * expand (or shrink) an existing mapping, potentially moving it at the same
1719 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1721 * under NOMMU conditions, we only permit changing a mapping's size, and only
1722 * as long as it stays within the region allocated by do_mmap_private() and the
1723 * block is not shareable
1725 * MREMAP_FIXED is not supported under NOMMU conditions
1727 static unsigned long do_mremap(unsigned long addr,
1728 unsigned long old_len, unsigned long new_len,
1729 unsigned long flags, unsigned long new_addr)
1731 struct vm_area_struct *vma;
1733 /* insanity checks first */
1734 old_len = PAGE_ALIGN(old_len);
1735 new_len = PAGE_ALIGN(new_len);
1736 if (old_len == 0 || new_len == 0)
1737 return (unsigned long) -EINVAL;
1739 if (offset_in_page(addr))
1742 if (flags & MREMAP_FIXED && new_addr != addr)
1743 return (unsigned long) -EINVAL;
1745 vma = find_vma_exact(current->mm, addr, old_len);
1747 return (unsigned long) -EINVAL;
1749 if (vma->vm_end != vma->vm_start + old_len)
1750 return (unsigned long) -EFAULT;
1752 if (vma->vm_flags & VM_MAYSHARE)
1753 return (unsigned long) -EPERM;
1755 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1756 return (unsigned long) -ENOMEM;
1758 /* all checks complete - do it */
1759 vma->vm_end = vma->vm_start + new_len;
1760 return vma->vm_start;
1763 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1764 unsigned long, new_len, unsigned long, flags,
1765 unsigned long, new_addr)
1769 down_write(¤t->mm->mmap_sem);
1770 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1771 up_write(¤t->mm->mmap_sem);
1775 struct page *follow_page_mask(struct vm_area_struct *vma,
1776 unsigned long address, unsigned int flags,
1777 unsigned int *page_mask)
1783 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1784 unsigned long pfn, unsigned long size, pgprot_t prot)
1786 if (addr != (pfn << PAGE_SHIFT))
1789 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1792 EXPORT_SYMBOL(remap_pfn_range);
1794 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1796 unsigned long pfn = start >> PAGE_SHIFT;
1797 unsigned long vm_len = vma->vm_end - vma->vm_start;
1799 pfn += vma->vm_pgoff;
1800 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1802 EXPORT_SYMBOL(vm_iomap_memory);
1804 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1805 unsigned long pgoff)
1807 unsigned int size = vma->vm_end - vma->vm_start;
1809 if (!(vma->vm_flags & VM_USERMAP))
1812 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1813 vma->vm_end = vma->vm_start + size;
1817 EXPORT_SYMBOL(remap_vmalloc_range);
1819 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1820 unsigned long len, unsigned long pgoff, unsigned long flags)
1825 void unmap_mapping_range(struct address_space *mapping,
1826 loff_t const holebegin, loff_t const holelen,
1830 EXPORT_SYMBOL(unmap_mapping_range);
1833 * Check that a process has enough memory to allocate a new virtual
1834 * mapping. 0 means there is enough memory for the allocation to
1835 * succeed and -ENOMEM implies there is not.
1837 * We currently support three overcommit policies, which are set via the
1838 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1840 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1841 * Additional code 2002 Jul 20 by Robert Love.
1843 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1845 * Note this is a helper function intended to be used by LSMs which
1846 * wish to use this logic.
1848 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1850 long free, allowed, reserve;
1852 vm_acct_memory(pages);
1855 * Sometimes we want to use more memory than we have
1857 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1860 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1861 free = global_page_state(NR_FREE_PAGES);
1862 free += global_page_state(NR_FILE_PAGES);
1865 * shmem pages shouldn't be counted as free in this
1866 * case, they can't be purged, only swapped out, and
1867 * that won't affect the overall amount of available
1868 * memory in the system.
1870 free -= global_page_state(NR_SHMEM);
1872 free += get_nr_swap_pages();
1875 * Any slabs which are created with the
1876 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1877 * which are reclaimable, under pressure. The dentry
1878 * cache and most inode caches should fall into this
1880 free += global_page_state(NR_SLAB_RECLAIMABLE);
1883 * Leave reserved pages. The pages are not for anonymous pages.
1885 if (free <= totalreserve_pages)
1888 free -= totalreserve_pages;
1891 * Reserve some for root
1894 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1902 allowed = vm_commit_limit();
1904 * Reserve some 3% for root
1907 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1910 * Don't let a single process grow so big a user can't recover
1913 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1914 allowed -= min_t(long, mm->total_vm / 32, reserve);
1917 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1921 vm_unacct_memory(pages);
1926 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1931 EXPORT_SYMBOL(filemap_fault);
1933 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1937 EXPORT_SYMBOL(filemap_map_pages);
1939 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1940 unsigned long addr, void *buf, int len, int write)
1942 struct vm_area_struct *vma;
1944 down_read(&mm->mmap_sem);
1946 /* the access must start within one of the target process's mappings */
1947 vma = find_vma(mm, addr);
1949 /* don't overrun this mapping */
1950 if (addr + len >= vma->vm_end)
1951 len = vma->vm_end - addr;
1953 /* only read or write mappings where it is permitted */
1954 if (write && vma->vm_flags & VM_MAYWRITE)
1955 copy_to_user_page(vma, NULL, addr,
1956 (void *) addr, buf, len);
1957 else if (!write && vma->vm_flags & VM_MAYREAD)
1958 copy_from_user_page(vma, NULL, addr,
1959 buf, (void *) addr, len);
1966 up_read(&mm->mmap_sem);
1972 * @access_remote_vm - access another process' address space
1973 * @mm: the mm_struct of the target address space
1974 * @addr: start address to access
1975 * @buf: source or destination buffer
1976 * @len: number of bytes to transfer
1977 * @write: whether the access is a write
1979 * The caller must hold a reference on @mm.
1981 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1982 void *buf, int len, int write)
1984 return __access_remote_vm(NULL, mm, addr, buf, len, write);
1988 * Access another process' address space.
1989 * - source/target buffer must be kernel space
1991 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1993 struct mm_struct *mm;
1995 if (addr + len < addr)
1998 mm = get_task_mm(tsk);
2002 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2009 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2010 * @inode: The inode to check
2011 * @size: The current filesize of the inode
2012 * @newsize: The proposed filesize of the inode
2014 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2015 * make sure that that any outstanding VMAs aren't broken and then shrink the
2016 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2017 * automatically grant mappings that are too large.
2019 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2022 struct vm_area_struct *vma;
2023 struct vm_region *region;
2025 size_t r_size, r_top;
2027 low = newsize >> PAGE_SHIFT;
2028 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2030 down_write(&nommu_region_sem);
2031 i_mmap_lock_read(inode->i_mapping);
2033 /* search for VMAs that fall within the dead zone */
2034 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
2035 /* found one - only interested if it's shared out of the page
2037 if (vma->vm_flags & VM_SHARED) {
2038 i_mmap_unlock_read(inode->i_mapping);
2039 up_write(&nommu_region_sem);
2040 return -ETXTBSY; /* not quite true, but near enough */
2044 /* reduce any regions that overlap the dead zone - if in existence,
2045 * these will be pointed to by VMAs that don't overlap the dead zone
2047 * we don't check for any regions that start beyond the EOF as there
2050 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
2051 if (!(vma->vm_flags & VM_SHARED))
2054 region = vma->vm_region;
2055 r_size = region->vm_top - region->vm_start;
2056 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2058 if (r_top > newsize) {
2059 region->vm_top -= r_top - newsize;
2060 if (region->vm_end > region->vm_top)
2061 region->vm_end = region->vm_top;
2065 i_mmap_unlock_read(inode->i_mapping);
2066 up_write(&nommu_region_sem);
2071 * Initialise sysctl_user_reserve_kbytes.
2073 * This is intended to prevent a user from starting a single memory hogging
2074 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2077 * The default value is min(3% of free memory, 128MB)
2078 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2080 static int __meminit init_user_reserve(void)
2082 unsigned long free_kbytes;
2084 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2086 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2089 subsys_initcall(init_user_reserve);
2092 * Initialise sysctl_admin_reserve_kbytes.
2094 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2095 * to log in and kill a memory hogging process.
2097 * Systems with more than 256MB will reserve 8MB, enough to recover
2098 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2099 * only reserve 3% of free pages by default.
2101 static int __meminit init_admin_reserve(void)
2103 unsigned long free_kbytes;
2105 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2107 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2110 subsys_initcall(init_admin_reserve);