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/sched/mm.h>
21 #include <linux/vmacache.h>
22 #include <linux/mman.h>
23 #include <linux/swap.h>
24 #include <linux/file.h>
25 #include <linux/highmem.h>
26 #include <linux/pagemap.h>
27 #include <linux/slab.h>
28 #include <linux/vmalloc.h>
29 #include <linux/blkdev.h>
30 #include <linux/backing-dev.h>
31 #include <linux/compiler.h>
32 #include <linux/mount.h>
33 #include <linux/personality.h>
34 #include <linux/security.h>
35 #include <linux/syscalls.h>
36 #include <linux/audit.h>
37 #include <linux/printk.h>
39 #include <linux/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 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
52 int heap_stack_gap = 0;
54 atomic_long_t mmap_pages_allocated;
56 EXPORT_SYMBOL(mem_map);
58 /* list of mapped, potentially shareable regions */
59 static struct kmem_cache *vm_region_jar;
60 struct rb_root nommu_region_tree = RB_ROOT;
61 DECLARE_RWSEM(nommu_region_sem);
63 const struct vm_operations_struct generic_file_vm_ops = {
67 * Return the total memory allocated for this pointer, not
68 * just what the caller asked for.
70 * Doesn't have to be accurate, i.e. may have races.
72 unsigned int kobjsize(const void *objp)
77 * If the object we have should not have ksize performed on it,
80 if (!objp || !virt_addr_valid(objp))
83 page = virt_to_head_page(objp);
86 * If the allocator sets PageSlab, we know the pointer came from
93 * If it's not a compound page, see if we have a matching VMA
94 * region. This test is intentionally done in reverse order,
95 * so if there's no VMA, we still fall through and hand back
96 * PAGE_SIZE for 0-order pages.
98 if (!PageCompound(page)) {
99 struct vm_area_struct *vma;
101 vma = find_vma(current->mm, (unsigned long)objp);
103 return vma->vm_end - vma->vm_start;
107 * The ksize() function is only guaranteed to work for pointers
108 * returned by kmalloc(). So handle arbitrary pointers here.
110 return PAGE_SIZE << compound_order(page);
113 static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
114 unsigned long start, unsigned long nr_pages,
115 unsigned int foll_flags, struct page **pages,
116 struct vm_area_struct **vmas, int *nonblocking)
118 struct vm_area_struct *vma;
119 unsigned long vm_flags;
122 /* calculate required read or write permissions.
123 * If FOLL_FORCE is set, we only require the "MAY" flags.
125 vm_flags = (foll_flags & FOLL_WRITE) ?
126 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
127 vm_flags &= (foll_flags & FOLL_FORCE) ?
128 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
130 for (i = 0; i < nr_pages; i++) {
131 vma = find_vma(mm, start);
133 goto finish_or_fault;
135 /* protect what we can, including chardevs */
136 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
137 !(vm_flags & vma->vm_flags))
138 goto finish_or_fault;
141 pages[i] = virt_to_page(start);
147 start = (start + PAGE_SIZE) & PAGE_MASK;
153 return i ? : -EFAULT;
157 * get a list of pages in an address range belonging to the specified process
158 * and indicate the VMA that covers each page
159 * - this is potentially dodgy as we may end incrementing the page count of a
160 * slab page or a secondary page from a compound page
161 * - don't permit access to VMAs that don't support it, such as I/O mappings
163 long get_user_pages(unsigned long start, unsigned long nr_pages,
164 unsigned int gup_flags, struct page **pages,
165 struct vm_area_struct **vmas)
167 return __get_user_pages(current, current->mm, start, nr_pages,
168 gup_flags, pages, vmas, NULL);
170 EXPORT_SYMBOL(get_user_pages);
172 long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
173 unsigned int gup_flags, struct page **pages,
176 return get_user_pages(start, nr_pages, gup_flags, pages, NULL);
178 EXPORT_SYMBOL(get_user_pages_locked);
180 static long __get_user_pages_unlocked(struct task_struct *tsk,
181 struct mm_struct *mm, unsigned long start,
182 unsigned long nr_pages, struct page **pages,
183 unsigned int gup_flags)
186 down_read(&mm->mmap_sem);
187 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
189 up_read(&mm->mmap_sem);
193 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
194 struct page **pages, unsigned int gup_flags)
196 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
199 EXPORT_SYMBOL(get_user_pages_unlocked);
202 * follow_pfn - look up PFN at a user virtual address
203 * @vma: memory mapping
204 * @address: user virtual address
205 * @pfn: location to store found PFN
207 * Only IO mappings and raw PFN mappings are allowed.
209 * Returns zero and the pfn at @pfn on success, -ve otherwise.
211 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
214 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
217 *pfn = address >> PAGE_SHIFT;
220 EXPORT_SYMBOL(follow_pfn);
222 LIST_HEAD(vmap_area_list);
224 void vfree(const void *addr)
228 EXPORT_SYMBOL(vfree);
230 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
233 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
234 * returns only a logical address.
236 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
238 EXPORT_SYMBOL(__vmalloc);
240 void *__vmalloc_node_flags(unsigned long size, int node, gfp_t flags)
242 return __vmalloc(size, flags, PAGE_KERNEL);
245 void *vmalloc_user(unsigned long size)
249 ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
251 struct vm_area_struct *vma;
253 down_write(¤t->mm->mmap_sem);
254 vma = find_vma(current->mm, (unsigned long)ret);
256 vma->vm_flags |= VM_USERMAP;
257 up_write(¤t->mm->mmap_sem);
262 EXPORT_SYMBOL(vmalloc_user);
264 struct page *vmalloc_to_page(const void *addr)
266 return virt_to_page(addr);
268 EXPORT_SYMBOL(vmalloc_to_page);
270 unsigned long vmalloc_to_pfn(const void *addr)
272 return page_to_pfn(virt_to_page(addr));
274 EXPORT_SYMBOL(vmalloc_to_pfn);
276 long vread(char *buf, char *addr, unsigned long count)
278 /* Don't allow overflow */
279 if ((unsigned long) buf + count < count)
280 count = -(unsigned long) buf;
282 memcpy(buf, addr, count);
286 long vwrite(char *buf, char *addr, unsigned long count)
288 /* Don't allow overflow */
289 if ((unsigned long) addr + count < count)
290 count = -(unsigned long) addr;
292 memcpy(addr, buf, count);
297 * vmalloc - allocate virtually contiguous memory
299 * @size: allocation size
301 * Allocate enough pages to cover @size from the page level
302 * allocator and map them into contiguous kernel virtual space.
304 * For tight control over page level allocator and protection flags
305 * use __vmalloc() instead.
307 void *vmalloc(unsigned long size)
309 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
311 EXPORT_SYMBOL(vmalloc);
314 * vzalloc - allocate virtually contiguous memory with zero fill
316 * @size: allocation size
318 * Allocate enough pages to cover @size from the page level
319 * allocator and map them into contiguous kernel virtual space.
320 * The memory allocated is set to zero.
322 * For tight control over page level allocator and protection flags
323 * use __vmalloc() instead.
325 void *vzalloc(unsigned long size)
327 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
330 EXPORT_SYMBOL(vzalloc);
333 * vmalloc_node - allocate memory on a specific node
334 * @size: allocation size
337 * Allocate enough pages to cover @size from the page level
338 * allocator and map them into contiguous kernel virtual space.
340 * For tight control over page level allocator and protection flags
341 * use __vmalloc() instead.
343 void *vmalloc_node(unsigned long size, int node)
345 return vmalloc(size);
347 EXPORT_SYMBOL(vmalloc_node);
350 * vzalloc_node - allocate memory on a specific node with zero fill
351 * @size: allocation size
354 * Allocate enough pages to cover @size from the page level
355 * allocator and map them into contiguous kernel virtual space.
356 * The memory allocated is set to zero.
358 * For tight control over page level allocator and protection flags
359 * use __vmalloc() instead.
361 void *vzalloc_node(unsigned long size, int node)
363 return vzalloc(size);
365 EXPORT_SYMBOL(vzalloc_node);
367 #ifndef PAGE_KERNEL_EXEC
368 # define PAGE_KERNEL_EXEC PAGE_KERNEL
372 * vmalloc_exec - allocate virtually contiguous, executable memory
373 * @size: allocation size
375 * Kernel-internal function to allocate enough pages to cover @size
376 * the page level allocator and map them into contiguous and
377 * executable kernel virtual space.
379 * For tight control over page level allocator and protection flags
380 * use __vmalloc() instead.
383 void *vmalloc_exec(unsigned long size)
385 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
389 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
390 * @size: allocation size
392 * Allocate enough 32bit PA addressable pages to cover @size from the
393 * page level allocator and map them into contiguous kernel virtual space.
395 void *vmalloc_32(unsigned long size)
397 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
399 EXPORT_SYMBOL(vmalloc_32);
402 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
403 * @size: allocation size
405 * The resulting memory area is 32bit addressable and zeroed so it can be
406 * mapped to userspace without leaking data.
408 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
409 * remap_vmalloc_range() are permissible.
411 void *vmalloc_32_user(unsigned long size)
414 * We'll have to sort out the ZONE_DMA bits for 64-bit,
415 * but for now this can simply use vmalloc_user() directly.
417 return vmalloc_user(size);
419 EXPORT_SYMBOL(vmalloc_32_user);
421 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
428 void vunmap(const void *addr)
432 EXPORT_SYMBOL(vunmap);
434 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
439 EXPORT_SYMBOL(vm_map_ram);
441 void vm_unmap_ram(const void *mem, unsigned int count)
445 EXPORT_SYMBOL(vm_unmap_ram);
447 void vm_unmap_aliases(void)
450 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
453 * Implement a stub for vmalloc_sync_[un]mapping() if the architecture
454 * chose not to have one.
456 void __weak vmalloc_sync_mappings(void)
460 void __weak vmalloc_sync_unmappings(void)
465 * alloc_vm_area - allocate a range of kernel address space
466 * @size: size of the area
468 * Returns: NULL on failure, vm_struct on success
470 * This function reserves a range of kernel address space, and
471 * allocates pagetables to map that range. No actual mappings
472 * are created. If the kernel address space is not shared
473 * between processes, it syncs the pagetable across all
476 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
481 EXPORT_SYMBOL_GPL(alloc_vm_area);
483 void free_vm_area(struct vm_struct *area)
487 EXPORT_SYMBOL_GPL(free_vm_area);
489 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
494 EXPORT_SYMBOL(vm_insert_page);
497 * sys_brk() for the most part doesn't need the global kernel
498 * lock, except when an application is doing something nasty
499 * like trying to un-brk an area that has already been mapped
500 * to a regular file. in this case, the unmapping will need
501 * to invoke file system routines that need the global lock.
503 SYSCALL_DEFINE1(brk, unsigned long, brk)
505 struct mm_struct *mm = current->mm;
507 if (brk < mm->start_brk || brk > mm->context.end_brk)
514 * Always allow shrinking brk
516 if (brk <= mm->brk) {
522 * Ok, looks good - let it rip.
524 flush_icache_range(mm->brk, brk);
525 return mm->brk = brk;
529 * initialise the percpu counter for VM and region record slabs
531 void __init mmap_init(void)
535 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
537 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
541 * validate the region tree
542 * - the caller must hold the region lock
544 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
545 static noinline void validate_nommu_regions(void)
547 struct vm_region *region, *last;
548 struct rb_node *p, *lastp;
550 lastp = rb_first(&nommu_region_tree);
554 last = rb_entry(lastp, struct vm_region, vm_rb);
555 BUG_ON(last->vm_end <= last->vm_start);
556 BUG_ON(last->vm_top < last->vm_end);
558 while ((p = rb_next(lastp))) {
559 region = rb_entry(p, struct vm_region, vm_rb);
560 last = rb_entry(lastp, struct vm_region, vm_rb);
562 BUG_ON(region->vm_end <= region->vm_start);
563 BUG_ON(region->vm_top < region->vm_end);
564 BUG_ON(region->vm_start < last->vm_top);
570 static void validate_nommu_regions(void)
576 * add a region into the global tree
578 static void add_nommu_region(struct vm_region *region)
580 struct vm_region *pregion;
581 struct rb_node **p, *parent;
583 validate_nommu_regions();
586 p = &nommu_region_tree.rb_node;
589 pregion = rb_entry(parent, struct vm_region, vm_rb);
590 if (region->vm_start < pregion->vm_start)
592 else if (region->vm_start > pregion->vm_start)
594 else if (pregion == region)
600 rb_link_node(®ion->vm_rb, parent, p);
601 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
603 validate_nommu_regions();
607 * delete a region from the global tree
609 static void delete_nommu_region(struct vm_region *region)
611 BUG_ON(!nommu_region_tree.rb_node);
613 validate_nommu_regions();
614 rb_erase(®ion->vm_rb, &nommu_region_tree);
615 validate_nommu_regions();
619 * free a contiguous series of pages
621 static void free_page_series(unsigned long from, unsigned long to)
623 for (; from < to; from += PAGE_SIZE) {
624 struct page *page = virt_to_page(from);
626 atomic_long_dec(&mmap_pages_allocated);
632 * release a reference to a region
633 * - the caller must hold the region semaphore for writing, which this releases
634 * - the region may not have been added to the tree yet, in which case vm_top
635 * will equal vm_start
637 static void __put_nommu_region(struct vm_region *region)
638 __releases(nommu_region_sem)
640 BUG_ON(!nommu_region_tree.rb_node);
642 if (--region->vm_usage == 0) {
643 if (region->vm_top > region->vm_start)
644 delete_nommu_region(region);
645 up_write(&nommu_region_sem);
648 fput(region->vm_file);
650 /* IO memory and memory shared directly out of the pagecache
651 * from ramfs/tmpfs mustn't be released here */
652 if (region->vm_flags & VM_MAPPED_COPY)
653 free_page_series(region->vm_start, region->vm_top);
654 kmem_cache_free(vm_region_jar, region);
656 up_write(&nommu_region_sem);
661 * release a reference to a region
663 static void put_nommu_region(struct vm_region *region)
665 down_write(&nommu_region_sem);
666 __put_nommu_region(region);
670 * update protection on a vma
672 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
675 struct mm_struct *mm = vma->vm_mm;
676 long start = vma->vm_start & PAGE_MASK;
677 while (start < vma->vm_end) {
678 protect_page(mm, start, flags);
681 update_protections(mm);
686 * add a VMA into a process's mm_struct in the appropriate place in the list
687 * and tree and add to the address space's page tree also if not an anonymous
689 * - should be called with mm->mmap_sem held writelocked
691 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
693 struct vm_area_struct *pvma, *prev;
694 struct address_space *mapping;
695 struct rb_node **p, *parent, *rb_prev;
697 BUG_ON(!vma->vm_region);
702 protect_vma(vma, vma->vm_flags);
704 /* add the VMA to the mapping */
706 mapping = vma->vm_file->f_mapping;
708 i_mmap_lock_write(mapping);
709 flush_dcache_mmap_lock(mapping);
710 vma_interval_tree_insert(vma, &mapping->i_mmap);
711 flush_dcache_mmap_unlock(mapping);
712 i_mmap_unlock_write(mapping);
715 /* add the VMA to the tree */
716 parent = rb_prev = NULL;
717 p = &mm->mm_rb.rb_node;
720 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
722 /* sort by: start addr, end addr, VMA struct addr in that order
723 * (the latter is necessary as we may get identical VMAs) */
724 if (vma->vm_start < pvma->vm_start)
726 else if (vma->vm_start > pvma->vm_start) {
729 } else if (vma->vm_end < pvma->vm_end)
731 else if (vma->vm_end > pvma->vm_end) {
734 } else if (vma < pvma)
736 else if (vma > pvma) {
743 rb_link_node(&vma->vm_rb, parent, p);
744 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
746 /* add VMA to the VMA list also */
749 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
751 __vma_link_list(mm, vma, prev, parent);
755 * delete a VMA from its owning mm_struct and address space
757 static void delete_vma_from_mm(struct vm_area_struct *vma)
760 struct address_space *mapping;
761 struct mm_struct *mm = vma->vm_mm;
762 struct task_struct *curr = current;
767 for (i = 0; i < VMACACHE_SIZE; i++) {
768 /* if the vma is cached, invalidate the entire cache */
769 if (curr->vmacache.vmas[i] == vma) {
770 vmacache_invalidate(mm);
775 /* remove the VMA from the mapping */
777 mapping = vma->vm_file->f_mapping;
779 i_mmap_lock_write(mapping);
780 flush_dcache_mmap_lock(mapping);
781 vma_interval_tree_remove(vma, &mapping->i_mmap);
782 flush_dcache_mmap_unlock(mapping);
783 i_mmap_unlock_write(mapping);
786 /* remove from the MM's tree and list */
787 rb_erase(&vma->vm_rb, &mm->mm_rb);
790 vma->vm_prev->vm_next = vma->vm_next;
792 mm->mmap = vma->vm_next;
795 vma->vm_next->vm_prev = vma->vm_prev;
799 * destroy a VMA record
801 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
803 if (vma->vm_ops && vma->vm_ops->close)
804 vma->vm_ops->close(vma);
807 put_nommu_region(vma->vm_region);
808 kmem_cache_free(vm_area_cachep, vma);
812 * look up the first VMA in which addr resides, NULL if none
813 * - should be called with mm->mmap_sem at least held readlocked
815 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
817 struct vm_area_struct *vma;
819 /* check the cache first */
820 vma = vmacache_find(mm, addr);
824 /* trawl the list (there may be multiple mappings in which addr
826 for (vma = mm->mmap; vma; vma = vma->vm_next) {
827 if (vma->vm_start > addr)
829 if (vma->vm_end > addr) {
830 vmacache_update(addr, vma);
837 EXPORT_SYMBOL(find_vma);
841 * - we don't extend stack VMAs under NOMMU conditions
843 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
845 return find_vma(mm, addr);
849 * expand a stack to a given address
850 * - not supported under NOMMU conditions
852 int expand_stack(struct vm_area_struct *vma, unsigned long address)
858 * look up the first VMA exactly that exactly matches addr
859 * - should be called with mm->mmap_sem at least held readlocked
861 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
865 struct vm_area_struct *vma;
866 unsigned long end = addr + len;
868 /* check the cache first */
869 vma = vmacache_find_exact(mm, addr, end);
873 /* trawl the list (there may be multiple mappings in which addr
875 for (vma = mm->mmap; vma; vma = vma->vm_next) {
876 if (vma->vm_start < addr)
878 if (vma->vm_start > addr)
880 if (vma->vm_end == end) {
881 vmacache_update(addr, vma);
890 * determine whether a mapping should be permitted and, if so, what sort of
891 * mapping we're capable of supporting
893 static int validate_mmap_request(struct file *file,
899 unsigned long *_capabilities)
901 unsigned long capabilities, rlen;
904 /* do the simple checks first */
905 if (flags & MAP_FIXED)
908 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
909 (flags & MAP_TYPE) != MAP_SHARED)
915 /* Careful about overflows.. */
916 rlen = PAGE_ALIGN(len);
917 if (!rlen || rlen > TASK_SIZE)
920 /* offset overflow? */
921 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
925 /* files must support mmap */
926 if (!file->f_op->mmap)
929 /* work out if what we've got could possibly be shared
930 * - we support chardevs that provide their own "memory"
931 * - we support files/blockdevs that are memory backed
933 if (file->f_op->mmap_capabilities) {
934 capabilities = file->f_op->mmap_capabilities(file);
936 /* no explicit capabilities set, so assume some
938 switch (file_inode(file)->i_mode & S_IFMT) {
941 capabilities = NOMMU_MAP_COPY;
956 /* eliminate any capabilities that we can't support on this
958 if (!file->f_op->get_unmapped_area)
959 capabilities &= ~NOMMU_MAP_DIRECT;
960 if (!(file->f_mode & FMODE_CAN_READ))
961 capabilities &= ~NOMMU_MAP_COPY;
963 /* The file shall have been opened with read permission. */
964 if (!(file->f_mode & FMODE_READ))
967 if (flags & MAP_SHARED) {
968 /* do checks for writing, appending and locking */
969 if ((prot & PROT_WRITE) &&
970 !(file->f_mode & FMODE_WRITE))
973 if (IS_APPEND(file_inode(file)) &&
974 (file->f_mode & FMODE_WRITE))
977 if (locks_verify_locked(file))
980 if (!(capabilities & NOMMU_MAP_DIRECT))
983 /* we mustn't privatise shared mappings */
984 capabilities &= ~NOMMU_MAP_COPY;
986 /* we're going to read the file into private memory we
988 if (!(capabilities & NOMMU_MAP_COPY))
991 /* we don't permit a private writable mapping to be
992 * shared with the backing device */
993 if (prot & PROT_WRITE)
994 capabilities &= ~NOMMU_MAP_DIRECT;
997 if (capabilities & NOMMU_MAP_DIRECT) {
998 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
999 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
1000 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
1002 capabilities &= ~NOMMU_MAP_DIRECT;
1003 if (flags & MAP_SHARED) {
1004 pr_warn("MAP_SHARED not completely supported on !MMU\n");
1010 /* handle executable mappings and implied executable
1012 if (path_noexec(&file->f_path)) {
1013 if (prot & PROT_EXEC)
1015 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1016 /* handle implication of PROT_EXEC by PROT_READ */
1017 if (current->personality & READ_IMPLIES_EXEC) {
1018 if (capabilities & NOMMU_MAP_EXEC)
1021 } else if ((prot & PROT_READ) &&
1022 (prot & PROT_EXEC) &&
1023 !(capabilities & NOMMU_MAP_EXEC)
1025 /* backing file is not executable, try to copy */
1026 capabilities &= ~NOMMU_MAP_DIRECT;
1029 /* anonymous mappings are always memory backed and can be
1032 capabilities = NOMMU_MAP_COPY;
1034 /* handle PROT_EXEC implication by PROT_READ */
1035 if ((prot & PROT_READ) &&
1036 (current->personality & READ_IMPLIES_EXEC))
1040 /* allow the security API to have its say */
1041 ret = security_mmap_addr(addr);
1046 *_capabilities = capabilities;
1051 * we've determined that we can make the mapping, now translate what we
1052 * now know into VMA flags
1054 static unsigned long determine_vm_flags(struct file *file,
1056 unsigned long flags,
1057 unsigned long capabilities)
1059 unsigned long vm_flags;
1061 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1062 /* vm_flags |= mm->def_flags; */
1064 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1065 /* attempt to share read-only copies of mapped file chunks */
1066 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1067 if (file && !(prot & PROT_WRITE))
1068 vm_flags |= VM_MAYSHARE;
1070 /* overlay a shareable mapping on the backing device or inode
1071 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1073 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1074 if (flags & MAP_SHARED)
1075 vm_flags |= VM_SHARED;
1078 /* refuse to let anyone share private mappings with this process if
1079 * it's being traced - otherwise breakpoints set in it may interfere
1080 * with another untraced process
1082 if ((flags & MAP_PRIVATE) && current->ptrace)
1083 vm_flags &= ~VM_MAYSHARE;
1089 * set up a shared mapping on a file (the driver or filesystem provides and
1092 static int do_mmap_shared_file(struct vm_area_struct *vma)
1096 ret = call_mmap(vma->vm_file, vma);
1098 vma->vm_region->vm_top = vma->vm_region->vm_end;
1104 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1105 * opposed to tried but failed) so we can only give a suitable error as
1106 * it's not possible to make a private copy if MAP_SHARED was given */
1111 * set up a private mapping or an anonymous shared mapping
1113 static int do_mmap_private(struct vm_area_struct *vma,
1114 struct vm_region *region,
1116 unsigned long capabilities)
1118 unsigned long total, point;
1122 /* invoke the file's mapping function so that it can keep track of
1123 * shared mappings on devices or memory
1124 * - VM_MAYSHARE will be set if it may attempt to share
1126 if (capabilities & NOMMU_MAP_DIRECT) {
1127 ret = call_mmap(vma->vm_file, vma);
1129 /* shouldn't return success if we're not sharing */
1130 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1131 vma->vm_region->vm_top = vma->vm_region->vm_end;
1137 /* getting an ENOSYS error indicates that direct mmap isn't
1138 * possible (as opposed to tried but failed) so we'll try to
1139 * make a private copy of the data and map that instead */
1143 /* allocate some memory to hold the mapping
1144 * - note that this may not return a page-aligned address if the object
1145 * we're allocating is smaller than a page
1147 order = get_order(len);
1149 point = len >> PAGE_SHIFT;
1151 /* we don't want to allocate a power-of-2 sized page set */
1152 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1155 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1159 atomic_long_add(total, &mmap_pages_allocated);
1161 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1162 region->vm_start = (unsigned long) base;
1163 region->vm_end = region->vm_start + len;
1164 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1166 vma->vm_start = region->vm_start;
1167 vma->vm_end = region->vm_start + len;
1170 /* read the contents of a file into the copy */
1173 fpos = vma->vm_pgoff;
1174 fpos <<= PAGE_SHIFT;
1176 ret = kernel_read(vma->vm_file, base, len, &fpos);
1180 /* clear the last little bit */
1182 memset(base + ret, 0, len - ret);
1189 free_page_series(region->vm_start, region->vm_top);
1190 region->vm_start = vma->vm_start = 0;
1191 region->vm_end = vma->vm_end = 0;
1196 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1197 len, current->pid, current->comm);
1198 show_free_areas(0, NULL);
1203 * handle mapping creation for uClinux
1205 unsigned long do_mmap(struct file *file,
1209 unsigned long flags,
1210 vm_flags_t vm_flags,
1211 unsigned long pgoff,
1212 unsigned long *populate,
1213 struct list_head *uf)
1215 struct vm_area_struct *vma;
1216 struct vm_region *region;
1218 unsigned long capabilities, result;
1223 /* decide whether we should attempt the mapping, and if so what sort of
1225 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1230 /* we ignore the address hint */
1232 len = PAGE_ALIGN(len);
1234 /* we've determined that we can make the mapping, now translate what we
1235 * now know into VMA flags */
1236 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1238 /* we're going to need to record the mapping */
1239 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1241 goto error_getting_region;
1243 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1245 goto error_getting_vma;
1247 region->vm_usage = 1;
1248 region->vm_flags = vm_flags;
1249 region->vm_pgoff = pgoff;
1251 INIT_LIST_HEAD(&vma->anon_vma_chain);
1252 vma->vm_flags = vm_flags;
1253 vma->vm_pgoff = pgoff;
1256 region->vm_file = get_file(file);
1257 vma->vm_file = get_file(file);
1260 down_write(&nommu_region_sem);
1262 /* if we want to share, we need to check for regions created by other
1263 * mmap() calls that overlap with our proposed mapping
1264 * - we can only share with a superset match on most regular files
1265 * - shared mappings on character devices and memory backed files are
1266 * permitted to overlap inexactly as far as we are concerned for in
1267 * these cases, sharing is handled in the driver or filesystem rather
1270 if (vm_flags & VM_MAYSHARE) {
1271 struct vm_region *pregion;
1272 unsigned long pglen, rpglen, pgend, rpgend, start;
1274 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1275 pgend = pgoff + pglen;
1277 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1278 pregion = rb_entry(rb, struct vm_region, vm_rb);
1280 if (!(pregion->vm_flags & VM_MAYSHARE))
1283 /* search for overlapping mappings on the same file */
1284 if (file_inode(pregion->vm_file) !=
1288 if (pregion->vm_pgoff >= pgend)
1291 rpglen = pregion->vm_end - pregion->vm_start;
1292 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1293 rpgend = pregion->vm_pgoff + rpglen;
1294 if (pgoff >= rpgend)
1297 /* handle inexactly overlapping matches between
1299 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1300 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1301 /* new mapping is not a subset of the region */
1302 if (!(capabilities & NOMMU_MAP_DIRECT))
1303 goto sharing_violation;
1307 /* we've found a region we can share */
1308 pregion->vm_usage++;
1309 vma->vm_region = pregion;
1310 start = pregion->vm_start;
1311 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1312 vma->vm_start = start;
1313 vma->vm_end = start + len;
1315 if (pregion->vm_flags & VM_MAPPED_COPY)
1316 vma->vm_flags |= VM_MAPPED_COPY;
1318 ret = do_mmap_shared_file(vma);
1320 vma->vm_region = NULL;
1323 pregion->vm_usage--;
1325 goto error_just_free;
1328 fput(region->vm_file);
1329 kmem_cache_free(vm_region_jar, region);
1335 /* obtain the address at which to make a shared mapping
1336 * - this is the hook for quasi-memory character devices to
1337 * tell us the location of a shared mapping
1339 if (capabilities & NOMMU_MAP_DIRECT) {
1340 addr = file->f_op->get_unmapped_area(file, addr, len,
1342 if (IS_ERR_VALUE(addr)) {
1345 goto error_just_free;
1347 /* the driver refused to tell us where to site
1348 * the mapping so we'll have to attempt to copy
1351 if (!(capabilities & NOMMU_MAP_COPY))
1352 goto error_just_free;
1354 capabilities &= ~NOMMU_MAP_DIRECT;
1356 vma->vm_start = region->vm_start = addr;
1357 vma->vm_end = region->vm_end = addr + len;
1362 vma->vm_region = region;
1364 /* set up the mapping
1365 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1367 if (file && vma->vm_flags & VM_SHARED)
1368 ret = do_mmap_shared_file(vma);
1370 ret = do_mmap_private(vma, region, len, capabilities);
1372 goto error_just_free;
1373 add_nommu_region(region);
1375 /* clear anonymous mappings that don't ask for uninitialized data */
1376 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1377 memset((void *)region->vm_start, 0,
1378 region->vm_end - region->vm_start);
1380 /* okay... we have a mapping; now we have to register it */
1381 result = vma->vm_start;
1383 current->mm->total_vm += len >> PAGE_SHIFT;
1386 add_vma_to_mm(current->mm, vma);
1388 /* we flush the region from the icache only when the first executable
1389 * mapping of it is made */
1390 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1391 flush_icache_range(region->vm_start, region->vm_end);
1392 region->vm_icache_flushed = true;
1395 up_write(&nommu_region_sem);
1400 up_write(&nommu_region_sem);
1402 if (region->vm_file)
1403 fput(region->vm_file);
1404 kmem_cache_free(vm_region_jar, region);
1407 kmem_cache_free(vm_area_cachep, vma);
1411 up_write(&nommu_region_sem);
1412 pr_warn("Attempt to share mismatched mappings\n");
1417 kmem_cache_free(vm_region_jar, region);
1418 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1420 show_free_areas(0, NULL);
1423 error_getting_region:
1424 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1426 show_free_areas(0, NULL);
1430 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1431 unsigned long, prot, unsigned long, flags,
1432 unsigned long, fd, unsigned long, pgoff)
1434 struct file *file = NULL;
1435 unsigned long retval = -EBADF;
1437 audit_mmap_fd(fd, flags);
1438 if (!(flags & MAP_ANONYMOUS)) {
1444 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1446 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1454 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1455 struct mmap_arg_struct {
1459 unsigned long flags;
1461 unsigned long offset;
1464 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1466 struct mmap_arg_struct a;
1468 if (copy_from_user(&a, arg, sizeof(a)))
1470 if (offset_in_page(a.offset))
1473 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1474 a.offset >> PAGE_SHIFT);
1476 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1479 * split a vma into two pieces at address 'addr', a new vma is allocated either
1480 * for the first part or the tail.
1482 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1483 unsigned long addr, int new_below)
1485 struct vm_area_struct *new;
1486 struct vm_region *region;
1487 unsigned long npages;
1489 /* we're only permitted to split anonymous regions (these should have
1490 * only a single usage on the region) */
1494 if (mm->map_count >= sysctl_max_map_count)
1497 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1501 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1503 kmem_cache_free(vm_region_jar, region);
1507 /* most fields are the same, copy all, and then fixup */
1509 *region = *vma->vm_region;
1510 new->vm_region = region;
1512 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1515 region->vm_top = region->vm_end = new->vm_end = addr;
1517 region->vm_start = new->vm_start = addr;
1518 region->vm_pgoff = new->vm_pgoff += npages;
1521 if (new->vm_ops && new->vm_ops->open)
1522 new->vm_ops->open(new);
1524 delete_vma_from_mm(vma);
1525 down_write(&nommu_region_sem);
1526 delete_nommu_region(vma->vm_region);
1528 vma->vm_region->vm_start = vma->vm_start = addr;
1529 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1531 vma->vm_region->vm_end = vma->vm_end = addr;
1532 vma->vm_region->vm_top = addr;
1534 add_nommu_region(vma->vm_region);
1535 add_nommu_region(new->vm_region);
1536 up_write(&nommu_region_sem);
1537 add_vma_to_mm(mm, vma);
1538 add_vma_to_mm(mm, new);
1543 * shrink a VMA by removing the specified chunk from either the beginning or
1546 static int shrink_vma(struct mm_struct *mm,
1547 struct vm_area_struct *vma,
1548 unsigned long from, unsigned long to)
1550 struct vm_region *region;
1552 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1554 delete_vma_from_mm(vma);
1555 if (from > vma->vm_start)
1559 add_vma_to_mm(mm, vma);
1561 /* cut the backing region down to size */
1562 region = vma->vm_region;
1563 BUG_ON(region->vm_usage != 1);
1565 down_write(&nommu_region_sem);
1566 delete_nommu_region(region);
1567 if (from > region->vm_start) {
1568 to = region->vm_top;
1569 region->vm_top = region->vm_end = from;
1571 region->vm_start = to;
1573 add_nommu_region(region);
1574 up_write(&nommu_region_sem);
1576 free_page_series(from, to);
1582 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1583 * VMA, though it need not cover the whole VMA
1585 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
1587 struct vm_area_struct *vma;
1591 len = PAGE_ALIGN(len);
1597 /* find the first potentially overlapping VMA */
1598 vma = find_vma(mm, start);
1602 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1603 current->pid, current->comm,
1604 start, start + len - 1);
1610 /* we're allowed to split an anonymous VMA but not a file-backed one */
1613 if (start > vma->vm_start)
1615 if (end == vma->vm_end)
1616 goto erase_whole_vma;
1621 /* the chunk must be a subset of the VMA found */
1622 if (start == vma->vm_start && end == vma->vm_end)
1623 goto erase_whole_vma;
1624 if (start < vma->vm_start || end > vma->vm_end)
1626 if (offset_in_page(start))
1628 if (end != vma->vm_end && offset_in_page(end))
1630 if (start != vma->vm_start && end != vma->vm_end) {
1631 ret = split_vma(mm, vma, start, 1);
1635 return shrink_vma(mm, vma, start, end);
1639 delete_vma_from_mm(vma);
1640 delete_vma(mm, vma);
1643 EXPORT_SYMBOL(do_munmap);
1645 int vm_munmap(unsigned long addr, size_t len)
1647 struct mm_struct *mm = current->mm;
1650 down_write(&mm->mmap_sem);
1651 ret = do_munmap(mm, addr, len, NULL);
1652 up_write(&mm->mmap_sem);
1655 EXPORT_SYMBOL(vm_munmap);
1657 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1659 return vm_munmap(addr, len);
1663 * release all the mappings made in a process's VM space
1665 void exit_mmap(struct mm_struct *mm)
1667 struct vm_area_struct *vma;
1674 while ((vma = mm->mmap)) {
1675 mm->mmap = vma->vm_next;
1676 delete_vma_from_mm(vma);
1677 delete_vma(mm, vma);
1682 int vm_brk(unsigned long addr, unsigned long len)
1688 * expand (or shrink) an existing mapping, potentially moving it at the same
1689 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1691 * under NOMMU conditions, we only permit changing a mapping's size, and only
1692 * as long as it stays within the region allocated by do_mmap_private() and the
1693 * block is not shareable
1695 * MREMAP_FIXED is not supported under NOMMU conditions
1697 static unsigned long do_mremap(unsigned long addr,
1698 unsigned long old_len, unsigned long new_len,
1699 unsigned long flags, unsigned long new_addr)
1701 struct vm_area_struct *vma;
1703 /* insanity checks first */
1704 old_len = PAGE_ALIGN(old_len);
1705 new_len = PAGE_ALIGN(new_len);
1706 if (old_len == 0 || new_len == 0)
1707 return (unsigned long) -EINVAL;
1709 if (offset_in_page(addr))
1712 if (flags & MREMAP_FIXED && new_addr != addr)
1713 return (unsigned long) -EINVAL;
1715 vma = find_vma_exact(current->mm, addr, old_len);
1717 return (unsigned long) -EINVAL;
1719 if (vma->vm_end != vma->vm_start + old_len)
1720 return (unsigned long) -EFAULT;
1722 if (vma->vm_flags & VM_MAYSHARE)
1723 return (unsigned long) -EPERM;
1725 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1726 return (unsigned long) -ENOMEM;
1728 /* all checks complete - do it */
1729 vma->vm_end = vma->vm_start + new_len;
1730 return vma->vm_start;
1733 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1734 unsigned long, new_len, unsigned long, flags,
1735 unsigned long, new_addr)
1739 down_write(¤t->mm->mmap_sem);
1740 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1741 up_write(¤t->mm->mmap_sem);
1745 struct page *follow_page_mask(struct vm_area_struct *vma,
1746 unsigned long address, unsigned int flags,
1747 unsigned int *page_mask)
1753 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1754 unsigned long pfn, unsigned long size, pgprot_t prot)
1756 if (addr != (pfn << PAGE_SHIFT))
1759 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1762 EXPORT_SYMBOL(remap_pfn_range);
1764 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1766 unsigned long pfn = start >> PAGE_SHIFT;
1767 unsigned long vm_len = vma->vm_end - vma->vm_start;
1769 pfn += vma->vm_pgoff;
1770 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1772 EXPORT_SYMBOL(vm_iomap_memory);
1774 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1775 unsigned long pgoff)
1777 unsigned int size = vma->vm_end - vma->vm_start;
1779 if (!(vma->vm_flags & VM_USERMAP))
1782 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1783 vma->vm_end = vma->vm_start + size;
1787 EXPORT_SYMBOL(remap_vmalloc_range);
1789 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1790 unsigned long len, unsigned long pgoff, unsigned long flags)
1795 void unmap_mapping_range(struct address_space *mapping,
1796 loff_t const holebegin, loff_t const holelen,
1800 EXPORT_SYMBOL(unmap_mapping_range);
1802 int filemap_fault(struct vm_fault *vmf)
1807 EXPORT_SYMBOL(filemap_fault);
1809 void filemap_map_pages(struct vm_fault *vmf,
1810 pgoff_t start_pgoff, pgoff_t end_pgoff)
1814 EXPORT_SYMBOL(filemap_map_pages);
1816 int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1817 unsigned long addr, void *buf, int len, unsigned int gup_flags)
1819 struct vm_area_struct *vma;
1820 int write = gup_flags & FOLL_WRITE;
1822 down_read(&mm->mmap_sem);
1824 /* the access must start within one of the target process's mappings */
1825 vma = find_vma(mm, addr);
1827 /* don't overrun this mapping */
1828 if (addr + len >= vma->vm_end)
1829 len = vma->vm_end - addr;
1831 /* only read or write mappings where it is permitted */
1832 if (write && vma->vm_flags & VM_MAYWRITE)
1833 copy_to_user_page(vma, NULL, addr,
1834 (void *) addr, buf, len);
1835 else if (!write && vma->vm_flags & VM_MAYREAD)
1836 copy_from_user_page(vma, NULL, addr,
1837 buf, (void *) addr, len);
1844 up_read(&mm->mmap_sem);
1850 * @access_remote_vm - access another process' address space
1851 * @mm: the mm_struct of the target address space
1852 * @addr: start address to access
1853 * @buf: source or destination buffer
1854 * @len: number of bytes to transfer
1855 * @gup_flags: flags modifying lookup behaviour
1857 * The caller must hold a reference on @mm.
1859 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1860 void *buf, int len, unsigned int gup_flags)
1862 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
1866 * Access another process' address space.
1867 * - source/target buffer must be kernel space
1869 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1870 unsigned int gup_flags)
1872 struct mm_struct *mm;
1874 if (addr + len < addr)
1877 mm = get_task_mm(tsk);
1881 len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
1886 EXPORT_SYMBOL_GPL(access_process_vm);
1889 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1890 * @inode: The inode to check
1891 * @size: The current filesize of the inode
1892 * @newsize: The proposed filesize of the inode
1894 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1895 * make sure that that any outstanding VMAs aren't broken and then shrink the
1896 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1897 * automatically grant mappings that are too large.
1899 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1902 struct vm_area_struct *vma;
1903 struct vm_region *region;
1905 size_t r_size, r_top;
1907 low = newsize >> PAGE_SHIFT;
1908 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1910 down_write(&nommu_region_sem);
1911 i_mmap_lock_read(inode->i_mapping);
1913 /* search for VMAs that fall within the dead zone */
1914 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1915 /* found one - only interested if it's shared out of the page
1917 if (vma->vm_flags & VM_SHARED) {
1918 i_mmap_unlock_read(inode->i_mapping);
1919 up_write(&nommu_region_sem);
1920 return -ETXTBSY; /* not quite true, but near enough */
1924 /* reduce any regions that overlap the dead zone - if in existence,
1925 * these will be pointed to by VMAs that don't overlap the dead zone
1927 * we don't check for any regions that start beyond the EOF as there
1930 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1931 if (!(vma->vm_flags & VM_SHARED))
1934 region = vma->vm_region;
1935 r_size = region->vm_top - region->vm_start;
1936 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1938 if (r_top > newsize) {
1939 region->vm_top -= r_top - newsize;
1940 if (region->vm_end > region->vm_top)
1941 region->vm_end = region->vm_top;
1945 i_mmap_unlock_read(inode->i_mapping);
1946 up_write(&nommu_region_sem);
1951 * Initialise sysctl_user_reserve_kbytes.
1953 * This is intended to prevent a user from starting a single memory hogging
1954 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1957 * The default value is min(3% of free memory, 128MB)
1958 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1960 static int __meminit init_user_reserve(void)
1962 unsigned long free_kbytes;
1964 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1966 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1969 subsys_initcall(init_user_reserve);
1972 * Initialise sysctl_admin_reserve_kbytes.
1974 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1975 * to log in and kill a memory hogging process.
1977 * Systems with more than 256MB will reserve 8MB, enough to recover
1978 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1979 * only reserve 3% of free pages by default.
1981 static int __meminit init_admin_reserve(void)
1983 unsigned long free_kbytes;
1985 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1987 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1990 subsys_initcall(init_admin_reserve);