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) 2007-2009 Paul Mundt <lethal@linux-sh.org>
16 #include <linux/module.h>
18 #include <linux/mman.h>
19 #include <linux/swap.h>
20 #include <linux/file.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/tracehook.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mount.h>
29 #include <linux/personality.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
33 #include <asm/uaccess.h>
35 #include <asm/tlbflush.h>
36 #include <asm/mmu_context.h>
39 static inline __attribute__((format(printf, 1, 2)))
40 void no_printk(const char *fmt, ...)
45 #define kenter(FMT, ...) \
46 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
47 #define kleave(FMT, ...) \
48 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
49 #define kdebug(FMT, ...) \
50 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
52 #define kenter(FMT, ...) \
53 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
54 #define kleave(FMT, ...) \
55 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
56 #define kdebug(FMT, ...) \
57 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
62 unsigned long max_mapnr;
63 unsigned long num_physpages;
64 unsigned long highest_memmap_pfn;
65 struct percpu_counter vm_committed_as;
66 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
67 int sysctl_overcommit_ratio = 50; /* default is 50% */
68 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
69 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
70 int heap_stack_gap = 0;
72 atomic_long_t mmap_pages_allocated;
74 EXPORT_SYMBOL(mem_map);
75 EXPORT_SYMBOL(num_physpages);
77 /* list of mapped, potentially shareable regions */
78 static struct kmem_cache *vm_region_jar;
79 struct rb_root nommu_region_tree = RB_ROOT;
80 DECLARE_RWSEM(nommu_region_sem);
82 const struct vm_operations_struct generic_file_vm_ops = {
86 * Return the total memory allocated for this pointer, not
87 * just what the caller asked for.
89 * Doesn't have to be accurate, i.e. may have races.
91 unsigned int kobjsize(const void *objp)
96 * If the object we have should not have ksize performed on it,
99 if (!objp || !virt_addr_valid(objp))
102 page = virt_to_head_page(objp);
105 * If the allocator sets PageSlab, we know the pointer came from
112 * If it's not a compound page, see if we have a matching VMA
113 * region. This test is intentionally done in reverse order,
114 * so if there's no VMA, we still fall through and hand back
115 * PAGE_SIZE for 0-order pages.
117 if (!PageCompound(page)) {
118 struct vm_area_struct *vma;
120 vma = find_vma(current->mm, (unsigned long)objp);
122 return vma->vm_end - vma->vm_start;
126 * The ksize() function is only guaranteed to work for pointers
127 * returned by kmalloc(). So handle arbitrary pointers here.
129 return PAGE_SIZE << compound_order(page);
132 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
133 unsigned long start, int nr_pages, unsigned int foll_flags,
134 struct page **pages, struct vm_area_struct **vmas)
136 struct vm_area_struct *vma;
137 unsigned long vm_flags;
140 /* calculate required read or write permissions.
141 * If FOLL_FORCE is set, we only require the "MAY" flags.
143 vm_flags = (foll_flags & FOLL_WRITE) ?
144 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
145 vm_flags &= (foll_flags & FOLL_FORCE) ?
146 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
148 for (i = 0; i < nr_pages; i++) {
149 vma = find_vma(mm, start);
151 goto finish_or_fault;
153 /* protect what we can, including chardevs */
154 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
155 !(vm_flags & vma->vm_flags))
156 goto finish_or_fault;
159 pages[i] = virt_to_page(start);
161 page_cache_get(pages[i]);
165 start = (start + PAGE_SIZE) & PAGE_MASK;
171 return i ? : -EFAULT;
175 * get a list of pages in an address range belonging to the specified process
176 * and indicate the VMA that covers each page
177 * - this is potentially dodgy as we may end incrementing the page count of a
178 * slab page or a secondary page from a compound page
179 * - don't permit access to VMAs that don't support it, such as I/O mappings
181 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
182 unsigned long start, int nr_pages, int write, int force,
183 struct page **pages, struct vm_area_struct **vmas)
192 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
194 EXPORT_SYMBOL(get_user_pages);
197 * follow_pfn - look up PFN at a user virtual address
198 * @vma: memory mapping
199 * @address: user virtual address
200 * @pfn: location to store found PFN
202 * Only IO mappings and raw PFN mappings are allowed.
204 * Returns zero and the pfn at @pfn on success, -ve otherwise.
206 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
209 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
212 *pfn = address >> PAGE_SHIFT;
215 EXPORT_SYMBOL(follow_pfn);
217 DEFINE_RWLOCK(vmlist_lock);
218 struct vm_struct *vmlist;
220 void vfree(const void *addr)
224 EXPORT_SYMBOL(vfree);
226 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
229 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
230 * returns only a logical address.
232 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
234 EXPORT_SYMBOL(__vmalloc);
236 void *vmalloc_user(unsigned long size)
240 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
243 struct vm_area_struct *vma;
245 down_write(¤t->mm->mmap_sem);
246 vma = find_vma(current->mm, (unsigned long)ret);
248 vma->vm_flags |= VM_USERMAP;
249 up_write(¤t->mm->mmap_sem);
254 EXPORT_SYMBOL(vmalloc_user);
256 struct page *vmalloc_to_page(const void *addr)
258 return virt_to_page(addr);
260 EXPORT_SYMBOL(vmalloc_to_page);
262 unsigned long vmalloc_to_pfn(const void *addr)
264 return page_to_pfn(virt_to_page(addr));
266 EXPORT_SYMBOL(vmalloc_to_pfn);
268 long vread(char *buf, char *addr, unsigned long count)
270 memcpy(buf, addr, count);
274 long vwrite(char *buf, char *addr, unsigned long count)
276 /* Don't allow overflow */
277 if ((unsigned long) addr + count < count)
278 count = -(unsigned long) addr;
280 memcpy(addr, buf, count);
285 * vmalloc - allocate virtually continguos memory
287 * @size: allocation size
289 * Allocate enough pages to cover @size from the page level
290 * allocator and map them into continguos kernel virtual space.
292 * For tight control over page level allocator and protection flags
293 * use __vmalloc() instead.
295 void *vmalloc(unsigned long size)
297 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
299 EXPORT_SYMBOL(vmalloc);
301 void *vmalloc_node(unsigned long size, int node)
303 return vmalloc(size);
305 EXPORT_SYMBOL(vmalloc_node);
307 #ifndef PAGE_KERNEL_EXEC
308 # define PAGE_KERNEL_EXEC PAGE_KERNEL
312 * vmalloc_exec - allocate virtually contiguous, executable memory
313 * @size: allocation size
315 * Kernel-internal function to allocate enough pages to cover @size
316 * the page level allocator and map them into contiguous and
317 * executable kernel virtual space.
319 * For tight control over page level allocator and protection flags
320 * use __vmalloc() instead.
323 void *vmalloc_exec(unsigned long size)
325 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
329 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
330 * @size: allocation size
332 * Allocate enough 32bit PA addressable pages to cover @size from the
333 * page level allocator and map them into continguos kernel virtual space.
335 void *vmalloc_32(unsigned long size)
337 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
339 EXPORT_SYMBOL(vmalloc_32);
342 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
343 * @size: allocation size
345 * The resulting memory area is 32bit addressable and zeroed so it can be
346 * mapped to userspace without leaking data.
348 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
349 * remap_vmalloc_range() are permissible.
351 void *vmalloc_32_user(unsigned long size)
354 * We'll have to sort out the ZONE_DMA bits for 64-bit,
355 * but for now this can simply use vmalloc_user() directly.
357 return vmalloc_user(size);
359 EXPORT_SYMBOL(vmalloc_32_user);
361 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
368 void vunmap(const void *addr)
372 EXPORT_SYMBOL(vunmap);
374 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
379 EXPORT_SYMBOL(vm_map_ram);
381 void vm_unmap_ram(const void *mem, unsigned int count)
385 EXPORT_SYMBOL(vm_unmap_ram);
387 void vm_unmap_aliases(void)
390 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
393 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
396 void __attribute__((weak)) vmalloc_sync_all(void)
400 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
405 EXPORT_SYMBOL(vm_insert_page);
408 * sys_brk() for the most part doesn't need the global kernel
409 * lock, except when an application is doing something nasty
410 * like trying to un-brk an area that has already been mapped
411 * to a regular file. in this case, the unmapping will need
412 * to invoke file system routines that need the global lock.
414 SYSCALL_DEFINE1(brk, unsigned long, brk)
416 struct mm_struct *mm = current->mm;
418 if (brk < mm->start_brk || brk > mm->context.end_brk)
425 * Always allow shrinking brk
427 if (brk <= mm->brk) {
433 * Ok, looks good - let it rip.
435 flush_icache_range(mm->brk, brk);
436 return mm->brk = brk;
440 * initialise the VMA and region record slabs
442 void __init mmap_init(void)
446 ret = percpu_counter_init(&vm_committed_as, 0);
448 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
452 * validate the region tree
453 * - the caller must hold the region lock
455 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
456 static noinline void validate_nommu_regions(void)
458 struct vm_region *region, *last;
459 struct rb_node *p, *lastp;
461 lastp = rb_first(&nommu_region_tree);
465 last = rb_entry(lastp, struct vm_region, vm_rb);
466 BUG_ON(unlikely(last->vm_end <= last->vm_start));
467 BUG_ON(unlikely(last->vm_top < last->vm_end));
469 while ((p = rb_next(lastp))) {
470 region = rb_entry(p, struct vm_region, vm_rb);
471 last = rb_entry(lastp, struct vm_region, vm_rb);
473 BUG_ON(unlikely(region->vm_end <= region->vm_start));
474 BUG_ON(unlikely(region->vm_top < region->vm_end));
475 BUG_ON(unlikely(region->vm_start < last->vm_top));
481 static void validate_nommu_regions(void)
487 * add a region into the global tree
489 static void add_nommu_region(struct vm_region *region)
491 struct vm_region *pregion;
492 struct rb_node **p, *parent;
494 validate_nommu_regions();
497 p = &nommu_region_tree.rb_node;
500 pregion = rb_entry(parent, struct vm_region, vm_rb);
501 if (region->vm_start < pregion->vm_start)
503 else if (region->vm_start > pregion->vm_start)
505 else if (pregion == region)
511 rb_link_node(®ion->vm_rb, parent, p);
512 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
514 validate_nommu_regions();
518 * delete a region from the global tree
520 static void delete_nommu_region(struct vm_region *region)
522 BUG_ON(!nommu_region_tree.rb_node);
524 validate_nommu_regions();
525 rb_erase(®ion->vm_rb, &nommu_region_tree);
526 validate_nommu_regions();
530 * free a contiguous series of pages
532 static void free_page_series(unsigned long from, unsigned long to)
534 for (; from < to; from += PAGE_SIZE) {
535 struct page *page = virt_to_page(from);
537 kdebug("- free %lx", from);
538 atomic_long_dec(&mmap_pages_allocated);
539 if (page_count(page) != 1)
540 kdebug("free page %p: refcount not one: %d",
541 page, page_count(page));
547 * release a reference to a region
548 * - the caller must hold the region semaphore for writing, which this releases
549 * - the region may not have been added to the tree yet, in which case vm_top
550 * will equal vm_start
552 static void __put_nommu_region(struct vm_region *region)
553 __releases(nommu_region_sem)
555 kenter("%p{%d}", region, region->vm_usage);
557 BUG_ON(!nommu_region_tree.rb_node);
559 if (--region->vm_usage == 0) {
560 if (region->vm_top > region->vm_start)
561 delete_nommu_region(region);
562 up_write(&nommu_region_sem);
565 fput(region->vm_file);
567 /* IO memory and memory shared directly out of the pagecache
568 * from ramfs/tmpfs mustn't be released here */
569 if (region->vm_flags & VM_MAPPED_COPY) {
570 kdebug("free series");
571 free_page_series(region->vm_start, region->vm_top);
573 kmem_cache_free(vm_region_jar, region);
575 up_write(&nommu_region_sem);
580 * release a reference to a region
582 static void put_nommu_region(struct vm_region *region)
584 down_write(&nommu_region_sem);
585 __put_nommu_region(region);
589 * update protection on a vma
591 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
594 struct mm_struct *mm = vma->vm_mm;
595 long start = vma->vm_start & PAGE_MASK;
596 while (start < vma->vm_end) {
597 protect_page(mm, start, flags);
600 update_protections(mm);
605 * add a VMA into a process's mm_struct in the appropriate place in the list
606 * and tree and add to the address space's page tree also if not an anonymous
608 * - should be called with mm->mmap_sem held writelocked
610 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
612 struct vm_area_struct *pvma, **pp, *next;
613 struct address_space *mapping;
614 struct rb_node **p, *parent;
618 BUG_ON(!vma->vm_region);
623 protect_vma(vma, vma->vm_flags);
625 /* add the VMA to the mapping */
627 mapping = vma->vm_file->f_mapping;
629 flush_dcache_mmap_lock(mapping);
630 vma_prio_tree_insert(vma, &mapping->i_mmap);
631 flush_dcache_mmap_unlock(mapping);
634 /* add the VMA to the tree */
636 p = &mm->mm_rb.rb_node;
639 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
641 /* sort by: start addr, end addr, VMA struct addr in that order
642 * (the latter is necessary as we may get identical VMAs) */
643 if (vma->vm_start < pvma->vm_start)
645 else if (vma->vm_start > pvma->vm_start)
647 else if (vma->vm_end < pvma->vm_end)
649 else if (vma->vm_end > pvma->vm_end)
659 rb_link_node(&vma->vm_rb, parent, p);
660 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
662 /* add VMA to the VMA list also */
663 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
664 if (pvma->vm_start > vma->vm_start)
666 if (pvma->vm_start < vma->vm_start)
668 if (pvma->vm_end < vma->vm_end)
680 * delete a VMA from its owning mm_struct and address space
682 static void delete_vma_from_mm(struct vm_area_struct *vma)
684 struct vm_area_struct **pp;
685 struct address_space *mapping;
686 struct mm_struct *mm = vma->vm_mm;
693 if (mm->mmap_cache == vma)
694 mm->mmap_cache = NULL;
696 /* remove the VMA from the mapping */
698 mapping = vma->vm_file->f_mapping;
700 flush_dcache_mmap_lock(mapping);
701 vma_prio_tree_remove(vma, &mapping->i_mmap);
702 flush_dcache_mmap_unlock(mapping);
705 /* remove from the MM's tree and list */
706 rb_erase(&vma->vm_rb, &mm->mm_rb);
707 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
718 * destroy a VMA record
720 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
723 if (vma->vm_ops && vma->vm_ops->close)
724 vma->vm_ops->close(vma);
727 if (vma->vm_flags & VM_EXECUTABLE)
728 removed_exe_file_vma(mm);
730 put_nommu_region(vma->vm_region);
731 kmem_cache_free(vm_area_cachep, vma);
735 * look up the first VMA in which addr resides, NULL if none
736 * - should be called with mm->mmap_sem at least held readlocked
738 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
740 struct vm_area_struct *vma;
741 struct rb_node *n = mm->mm_rb.rb_node;
743 /* check the cache first */
744 vma = mm->mmap_cache;
745 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
748 /* trawl the tree (there may be multiple mappings in which addr
750 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
751 vma = rb_entry(n, struct vm_area_struct, vm_rb);
752 if (vma->vm_start > addr)
754 if (vma->vm_end > addr) {
755 mm->mmap_cache = vma;
762 EXPORT_SYMBOL(find_vma);
766 * - we don't extend stack VMAs under NOMMU conditions
768 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
770 return find_vma(mm, addr);
774 * expand a stack to a given address
775 * - not supported under NOMMU conditions
777 int expand_stack(struct vm_area_struct *vma, unsigned long address)
783 * look up the first VMA exactly that exactly matches addr
784 * - should be called with mm->mmap_sem at least held readlocked
786 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
790 struct vm_area_struct *vma;
791 struct rb_node *n = mm->mm_rb.rb_node;
792 unsigned long end = addr + len;
794 /* check the cache first */
795 vma = mm->mmap_cache;
796 if (vma && vma->vm_start == addr && vma->vm_end == end)
799 /* trawl the tree (there may be multiple mappings in which addr
801 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
802 vma = rb_entry(n, struct vm_area_struct, vm_rb);
803 if (vma->vm_start < addr)
805 if (vma->vm_start > addr)
807 if (vma->vm_end == end) {
808 mm->mmap_cache = vma;
817 * determine whether a mapping should be permitted and, if so, what sort of
818 * mapping we're capable of supporting
820 static int validate_mmap_request(struct file *file,
826 unsigned long *_capabilities)
828 unsigned long capabilities, rlen;
829 unsigned long reqprot = prot;
832 /* do the simple checks first */
833 if (flags & MAP_FIXED) {
835 "%d: Can't do fixed-address/overlay mmap of RAM\n",
840 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
841 (flags & MAP_TYPE) != MAP_SHARED)
847 /* Careful about overflows.. */
848 rlen = PAGE_ALIGN(len);
849 if (!rlen || rlen > TASK_SIZE)
852 /* offset overflow? */
853 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
857 /* validate file mapping requests */
858 struct address_space *mapping;
860 /* files must support mmap */
861 if (!file->f_op || !file->f_op->mmap)
864 /* work out if what we've got could possibly be shared
865 * - we support chardevs that provide their own "memory"
866 * - we support files/blockdevs that are memory backed
868 mapping = file->f_mapping;
870 mapping = file->f_path.dentry->d_inode->i_mapping;
873 if (mapping && mapping->backing_dev_info)
874 capabilities = mapping->backing_dev_info->capabilities;
877 /* no explicit capabilities set, so assume some
879 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
882 capabilities = BDI_CAP_MAP_COPY;
897 /* eliminate any capabilities that we can't support on this
899 if (!file->f_op->get_unmapped_area)
900 capabilities &= ~BDI_CAP_MAP_DIRECT;
901 if (!file->f_op->read)
902 capabilities &= ~BDI_CAP_MAP_COPY;
904 /* The file shall have been opened with read permission. */
905 if (!(file->f_mode & FMODE_READ))
908 if (flags & MAP_SHARED) {
909 /* do checks for writing, appending and locking */
910 if ((prot & PROT_WRITE) &&
911 !(file->f_mode & FMODE_WRITE))
914 if (IS_APPEND(file->f_path.dentry->d_inode) &&
915 (file->f_mode & FMODE_WRITE))
918 if (locks_verify_locked(file->f_path.dentry->d_inode))
921 if (!(capabilities & BDI_CAP_MAP_DIRECT))
924 /* we mustn't privatise shared mappings */
925 capabilities &= ~BDI_CAP_MAP_COPY;
928 /* we're going to read the file into private memory we
930 if (!(capabilities & BDI_CAP_MAP_COPY))
933 /* we don't permit a private writable mapping to be
934 * shared with the backing device */
935 if (prot & PROT_WRITE)
936 capabilities &= ~BDI_CAP_MAP_DIRECT;
939 if (capabilities & BDI_CAP_MAP_DIRECT) {
940 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
941 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
942 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
944 capabilities &= ~BDI_CAP_MAP_DIRECT;
945 if (flags & MAP_SHARED) {
947 "MAP_SHARED not completely supported on !MMU\n");
953 /* handle executable mappings and implied executable
955 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
956 if (prot & PROT_EXEC)
959 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
960 /* handle implication of PROT_EXEC by PROT_READ */
961 if (current->personality & READ_IMPLIES_EXEC) {
962 if (capabilities & BDI_CAP_EXEC_MAP)
966 else if ((prot & PROT_READ) &&
967 (prot & PROT_EXEC) &&
968 !(capabilities & BDI_CAP_EXEC_MAP)
970 /* backing file is not executable, try to copy */
971 capabilities &= ~BDI_CAP_MAP_DIRECT;
975 /* anonymous mappings are always memory backed and can be
978 capabilities = BDI_CAP_MAP_COPY;
980 /* handle PROT_EXEC implication by PROT_READ */
981 if ((prot & PROT_READ) &&
982 (current->personality & READ_IMPLIES_EXEC))
986 /* allow the security API to have its say */
987 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
992 *_capabilities = capabilities;
997 * we've determined that we can make the mapping, now translate what we
998 * now know into VMA flags
1000 static unsigned long determine_vm_flags(struct file *file,
1002 unsigned long flags,
1003 unsigned long capabilities)
1005 unsigned long vm_flags;
1007 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1008 /* vm_flags |= mm->def_flags; */
1010 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1011 /* attempt to share read-only copies of mapped file chunks */
1012 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1013 if (file && !(prot & PROT_WRITE))
1014 vm_flags |= VM_MAYSHARE;
1016 /* overlay a shareable mapping on the backing device or inode
1017 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1019 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1020 if (flags & MAP_SHARED)
1021 vm_flags |= VM_SHARED;
1024 /* refuse to let anyone share private mappings with this process if
1025 * it's being traced - otherwise breakpoints set in it may interfere
1026 * with another untraced process
1028 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1029 vm_flags &= ~VM_MAYSHARE;
1035 * set up a shared mapping on a file (the driver or filesystem provides and
1038 static int do_mmap_shared_file(struct vm_area_struct *vma)
1042 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1044 vma->vm_region->vm_top = vma->vm_region->vm_end;
1050 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1051 * opposed to tried but failed) so we can only give a suitable error as
1052 * it's not possible to make a private copy if MAP_SHARED was given */
1057 * set up a private mapping or an anonymous shared mapping
1059 static int do_mmap_private(struct vm_area_struct *vma,
1060 struct vm_region *region,
1062 unsigned long capabilities)
1065 unsigned long total, point, n, rlen;
1069 /* invoke the file's mapping function so that it can keep track of
1070 * shared mappings on devices or memory
1071 * - VM_MAYSHARE will be set if it may attempt to share
1073 if (capabilities & BDI_CAP_MAP_DIRECT) {
1074 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1076 /* shouldn't return success if we're not sharing */
1077 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1078 vma->vm_region->vm_top = vma->vm_region->vm_end;
1084 /* getting an ENOSYS error indicates that direct mmap isn't
1085 * possible (as opposed to tried but failed) so we'll try to
1086 * make a private copy of the data and map that instead */
1089 rlen = PAGE_ALIGN(len);
1091 /* allocate some memory to hold the mapping
1092 * - note that this may not return a page-aligned address if the object
1093 * we're allocating is smaller than a page
1095 order = get_order(rlen);
1096 kdebug("alloc order %d for %lx", order, len);
1098 pages = alloc_pages(GFP_KERNEL, order);
1103 atomic_long_add(total, &mmap_pages_allocated);
1105 point = rlen >> PAGE_SHIFT;
1107 /* we allocated a power-of-2 sized page set, so we may want to trim off
1109 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1110 while (total > point) {
1111 order = ilog2(total - point);
1113 kdebug("shave %lu/%lu @%lu", n, total - point, total);
1114 atomic_long_sub(n, &mmap_pages_allocated);
1116 set_page_refcounted(pages + total);
1117 __free_pages(pages + total, order);
1121 for (point = 1; point < total; point++)
1122 set_page_refcounted(&pages[point]);
1124 base = page_address(pages);
1125 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1126 region->vm_start = (unsigned long) base;
1127 region->vm_end = region->vm_start + rlen;
1128 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1130 vma->vm_start = region->vm_start;
1131 vma->vm_end = region->vm_start + len;
1134 /* read the contents of a file into the copy */
1135 mm_segment_t old_fs;
1138 fpos = vma->vm_pgoff;
1139 fpos <<= PAGE_SHIFT;
1143 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1149 /* clear the last little bit */
1151 memset(base + ret, 0, rlen - ret);
1158 free_page_series(region->vm_start, region->vm_end);
1159 region->vm_start = vma->vm_start = 0;
1160 region->vm_end = vma->vm_end = 0;
1165 printk("Allocation of length %lu from process %d (%s) failed\n",
1166 len, current->pid, current->comm);
1172 * handle mapping creation for uClinux
1174 unsigned long do_mmap_pgoff(struct file *file,
1178 unsigned long flags,
1179 unsigned long pgoff)
1181 struct vm_area_struct *vma;
1182 struct vm_region *region;
1184 unsigned long capabilities, vm_flags, result;
1187 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1189 /* decide whether we should attempt the mapping, and if so what sort of
1191 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1194 kleave(" = %d [val]", ret);
1198 /* we ignore the address hint */
1201 /* we've determined that we can make the mapping, now translate what we
1202 * now know into VMA flags */
1203 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1205 /* we're going to need to record the mapping */
1206 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1208 goto error_getting_region;
1210 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1212 goto error_getting_vma;
1214 region->vm_usage = 1;
1215 region->vm_flags = vm_flags;
1216 region->vm_pgoff = pgoff;
1218 INIT_LIST_HEAD(&vma->anon_vma_chain);
1219 vma->vm_flags = vm_flags;
1220 vma->vm_pgoff = pgoff;
1223 region->vm_file = file;
1225 vma->vm_file = file;
1227 if (vm_flags & VM_EXECUTABLE) {
1228 added_exe_file_vma(current->mm);
1229 vma->vm_mm = current->mm;
1233 down_write(&nommu_region_sem);
1235 /* if we want to share, we need to check for regions created by other
1236 * mmap() calls that overlap with our proposed mapping
1237 * - we can only share with a superset match on most regular files
1238 * - shared mappings on character devices and memory backed files are
1239 * permitted to overlap inexactly as far as we are concerned for in
1240 * these cases, sharing is handled in the driver or filesystem rather
1243 if (vm_flags & VM_MAYSHARE) {
1244 struct vm_region *pregion;
1245 unsigned long pglen, rpglen, pgend, rpgend, start;
1247 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1248 pgend = pgoff + pglen;
1250 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1251 pregion = rb_entry(rb, struct vm_region, vm_rb);
1253 if (!(pregion->vm_flags & VM_MAYSHARE))
1256 /* search for overlapping mappings on the same file */
1257 if (pregion->vm_file->f_path.dentry->d_inode !=
1258 file->f_path.dentry->d_inode)
1261 if (pregion->vm_pgoff >= pgend)
1264 rpglen = pregion->vm_end - pregion->vm_start;
1265 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1266 rpgend = pregion->vm_pgoff + rpglen;
1267 if (pgoff >= rpgend)
1270 /* handle inexactly overlapping matches between
1272 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1273 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1274 /* new mapping is not a subset of the region */
1275 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1276 goto sharing_violation;
1280 /* we've found a region we can share */
1281 pregion->vm_usage++;
1282 vma->vm_region = pregion;
1283 start = pregion->vm_start;
1284 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1285 vma->vm_start = start;
1286 vma->vm_end = start + len;
1288 if (pregion->vm_flags & VM_MAPPED_COPY) {
1289 kdebug("share copy");
1290 vma->vm_flags |= VM_MAPPED_COPY;
1292 kdebug("share mmap");
1293 ret = do_mmap_shared_file(vma);
1295 vma->vm_region = NULL;
1298 pregion->vm_usage--;
1300 goto error_just_free;
1303 fput(region->vm_file);
1304 kmem_cache_free(vm_region_jar, region);
1310 /* obtain the address at which to make a shared mapping
1311 * - this is the hook for quasi-memory character devices to
1312 * tell us the location of a shared mapping
1314 if (capabilities & BDI_CAP_MAP_DIRECT) {
1315 addr = file->f_op->get_unmapped_area(file, addr, len,
1317 if (IS_ERR((void *) addr)) {
1319 if (ret != (unsigned long) -ENOSYS)
1320 goto error_just_free;
1322 /* the driver refused to tell us where to site
1323 * the mapping so we'll have to attempt to copy
1325 ret = (unsigned long) -ENODEV;
1326 if (!(capabilities & BDI_CAP_MAP_COPY))
1327 goto error_just_free;
1329 capabilities &= ~BDI_CAP_MAP_DIRECT;
1331 vma->vm_start = region->vm_start = addr;
1332 vma->vm_end = region->vm_end = addr + len;
1337 vma->vm_region = region;
1339 /* set up the mapping
1340 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1342 if (file && vma->vm_flags & VM_SHARED)
1343 ret = do_mmap_shared_file(vma);
1345 ret = do_mmap_private(vma, region, len, capabilities);
1347 goto error_just_free;
1348 add_nommu_region(region);
1350 /* clear anonymous mappings that don't ask for uninitialized data */
1351 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1352 memset((void *)region->vm_start, 0,
1353 region->vm_end - region->vm_start);
1355 /* okay... we have a mapping; now we have to register it */
1356 result = vma->vm_start;
1358 current->mm->total_vm += len >> PAGE_SHIFT;
1361 add_vma_to_mm(current->mm, vma);
1363 /* we flush the region from the icache only when the first executable
1364 * mapping of it is made */
1365 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1366 flush_icache_range(region->vm_start, region->vm_end);
1367 region->vm_icache_flushed = true;
1370 up_write(&nommu_region_sem);
1372 kleave(" = %lx", result);
1376 up_write(&nommu_region_sem);
1378 if (region->vm_file)
1379 fput(region->vm_file);
1380 kmem_cache_free(vm_region_jar, region);
1383 if (vma->vm_flags & VM_EXECUTABLE)
1384 removed_exe_file_vma(vma->vm_mm);
1385 kmem_cache_free(vm_area_cachep, vma);
1386 kleave(" = %d", ret);
1390 up_write(&nommu_region_sem);
1391 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1396 kmem_cache_free(vm_region_jar, region);
1397 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1398 " from process %d failed\n",
1403 error_getting_region:
1404 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1405 " from process %d failed\n",
1410 EXPORT_SYMBOL(do_mmap_pgoff);
1412 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1413 unsigned long, prot, unsigned long, flags,
1414 unsigned long, fd, unsigned long, pgoff)
1416 struct file *file = NULL;
1417 unsigned long retval = -EBADF;
1419 if (!(flags & MAP_ANONYMOUS)) {
1425 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1427 down_write(¤t->mm->mmap_sem);
1428 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1429 up_write(¤t->mm->mmap_sem);
1437 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1438 struct mmap_arg_struct {
1442 unsigned long flags;
1444 unsigned long offset;
1447 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1449 struct mmap_arg_struct a;
1451 if (copy_from_user(&a, arg, sizeof(a)))
1453 if (a.offset & ~PAGE_MASK)
1456 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1457 a.offset >> PAGE_SHIFT);
1459 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1462 * split a vma into two pieces at address 'addr', a new vma is allocated either
1463 * for the first part or the tail.
1465 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1466 unsigned long addr, int new_below)
1468 struct vm_area_struct *new;
1469 struct vm_region *region;
1470 unsigned long npages;
1474 /* we're only permitted to split anonymous regions (these should have
1475 * only a single usage on the region) */
1479 if (mm->map_count >= sysctl_max_map_count)
1482 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1486 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1488 kmem_cache_free(vm_region_jar, region);
1492 /* most fields are the same, copy all, and then fixup */
1494 *region = *vma->vm_region;
1495 new->vm_region = region;
1497 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1500 region->vm_top = region->vm_end = new->vm_end = addr;
1502 region->vm_start = new->vm_start = addr;
1503 region->vm_pgoff = new->vm_pgoff += npages;
1506 if (new->vm_ops && new->vm_ops->open)
1507 new->vm_ops->open(new);
1509 delete_vma_from_mm(vma);
1510 down_write(&nommu_region_sem);
1511 delete_nommu_region(vma->vm_region);
1513 vma->vm_region->vm_start = vma->vm_start = addr;
1514 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1516 vma->vm_region->vm_end = vma->vm_end = addr;
1517 vma->vm_region->vm_top = addr;
1519 add_nommu_region(vma->vm_region);
1520 add_nommu_region(new->vm_region);
1521 up_write(&nommu_region_sem);
1522 add_vma_to_mm(mm, vma);
1523 add_vma_to_mm(mm, new);
1528 * shrink a VMA by removing the specified chunk from either the beginning or
1531 static int shrink_vma(struct mm_struct *mm,
1532 struct vm_area_struct *vma,
1533 unsigned long from, unsigned long to)
1535 struct vm_region *region;
1539 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1541 delete_vma_from_mm(vma);
1542 if (from > vma->vm_start)
1546 add_vma_to_mm(mm, vma);
1548 /* cut the backing region down to size */
1549 region = vma->vm_region;
1550 BUG_ON(region->vm_usage != 1);
1552 down_write(&nommu_region_sem);
1553 delete_nommu_region(region);
1554 if (from > region->vm_start) {
1555 to = region->vm_top;
1556 region->vm_top = region->vm_end = from;
1558 region->vm_start = to;
1560 add_nommu_region(region);
1561 up_write(&nommu_region_sem);
1563 free_page_series(from, to);
1569 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1570 * VMA, though it need not cover the whole VMA
1572 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1574 struct vm_area_struct *vma;
1576 unsigned long end = start + len;
1579 kenter(",%lx,%zx", start, len);
1584 /* find the first potentially overlapping VMA */
1585 vma = find_vma(mm, start);
1587 static int limit = 0;
1590 "munmap of memory not mmapped by process %d"
1591 " (%s): 0x%lx-0x%lx\n",
1592 current->pid, current->comm,
1593 start, start + len - 1);
1599 /* we're allowed to split an anonymous VMA but not a file-backed one */
1602 if (start > vma->vm_start) {
1603 kleave(" = -EINVAL [miss]");
1606 if (end == vma->vm_end)
1607 goto erase_whole_vma;
1608 rb = rb_next(&vma->vm_rb);
1609 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1611 kleave(" = -EINVAL [split file]");
1614 /* the chunk must be a subset of the VMA found */
1615 if (start == vma->vm_start && end == vma->vm_end)
1616 goto erase_whole_vma;
1617 if (start < vma->vm_start || end > vma->vm_end) {
1618 kleave(" = -EINVAL [superset]");
1621 if (start & ~PAGE_MASK) {
1622 kleave(" = -EINVAL [unaligned start]");
1625 if (end != vma->vm_end && end & ~PAGE_MASK) {
1626 kleave(" = -EINVAL [unaligned split]");
1629 if (start != vma->vm_start && end != vma->vm_end) {
1630 ret = split_vma(mm, vma, start, 1);
1632 kleave(" = %d [split]", ret);
1636 return shrink_vma(mm, vma, start, end);
1640 delete_vma_from_mm(vma);
1641 delete_vma(mm, vma);
1645 EXPORT_SYMBOL(do_munmap);
1647 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1650 struct mm_struct *mm = current->mm;
1652 down_write(&mm->mmap_sem);
1653 ret = do_munmap(mm, addr, len);
1654 up_write(&mm->mmap_sem);
1659 * release all the mappings made in a process's VM space
1661 void exit_mmap(struct mm_struct *mm)
1663 struct vm_area_struct *vma;
1672 while ((vma = mm->mmap)) {
1673 mm->mmap = vma->vm_next;
1674 delete_vma_from_mm(vma);
1675 delete_vma(mm, vma);
1682 unsigned long do_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 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 if (old_len == 0 || new_len == 0)
1705 return (unsigned long) -EINVAL;
1707 if (addr & ~PAGE_MASK)
1710 if (flags & MREMAP_FIXED && new_addr != addr)
1711 return (unsigned long) -EINVAL;
1713 vma = find_vma_exact(current->mm, addr, old_len);
1715 return (unsigned long) -EINVAL;
1717 if (vma->vm_end != vma->vm_start + old_len)
1718 return (unsigned long) -EFAULT;
1720 if (vma->vm_flags & VM_MAYSHARE)
1721 return (unsigned long) -EPERM;
1723 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1724 return (unsigned long) -ENOMEM;
1726 /* all checks complete - do it */
1727 vma->vm_end = vma->vm_start + new_len;
1728 return vma->vm_start;
1730 EXPORT_SYMBOL(do_mremap);
1732 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1733 unsigned long, new_len, unsigned long, flags,
1734 unsigned long, new_addr)
1738 down_write(¤t->mm->mmap_sem);
1739 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1740 up_write(¤t->mm->mmap_sem);
1744 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1745 unsigned int foll_flags)
1750 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1751 unsigned long to, unsigned long size, pgprot_t prot)
1753 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1756 EXPORT_SYMBOL(remap_pfn_range);
1758 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1759 unsigned long pgoff)
1761 unsigned int size = vma->vm_end - vma->vm_start;
1763 if (!(vma->vm_flags & VM_USERMAP))
1766 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1767 vma->vm_end = vma->vm_start + size;
1771 EXPORT_SYMBOL(remap_vmalloc_range);
1773 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1777 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1778 unsigned long len, unsigned long pgoff, unsigned long flags)
1783 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1787 void unmap_mapping_range(struct address_space *mapping,
1788 loff_t const holebegin, loff_t const holelen,
1792 EXPORT_SYMBOL(unmap_mapping_range);
1795 * Check that a process has enough memory to allocate a new virtual
1796 * mapping. 0 means there is enough memory for the allocation to
1797 * succeed and -ENOMEM implies there is not.
1799 * We currently support three overcommit policies, which are set via the
1800 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1802 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1803 * Additional code 2002 Jul 20 by Robert Love.
1805 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1807 * Note this is a helper function intended to be used by LSMs which
1808 * wish to use this logic.
1810 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1812 unsigned long free, allowed;
1814 vm_acct_memory(pages);
1817 * Sometimes we want to use more memory than we have
1819 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1822 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1825 free = global_page_state(NR_FILE_PAGES);
1826 free += nr_swap_pages;
1829 * Any slabs which are created with the
1830 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1831 * which are reclaimable, under pressure. The dentry
1832 * cache and most inode caches should fall into this
1834 free += global_page_state(NR_SLAB_RECLAIMABLE);
1837 * Leave the last 3% for root
1846 * nr_free_pages() is very expensive on large systems,
1847 * only call if we're about to fail.
1849 n = nr_free_pages();
1852 * Leave reserved pages. The pages are not for anonymous pages.
1854 if (n <= totalreserve_pages)
1857 n -= totalreserve_pages;
1860 * Leave the last 3% for root
1872 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1874 * Leave the last 3% for root
1877 allowed -= allowed / 32;
1878 allowed += total_swap_pages;
1880 /* Don't let a single process grow too big:
1881 leave 3% of the size of this process for other processes */
1883 allowed -= mm->total_vm / 32;
1885 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1889 vm_unacct_memory(pages);
1894 int in_gate_area_no_task(unsigned long addr)
1899 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1904 EXPORT_SYMBOL(filemap_fault);
1907 * Access another process' address space.
1908 * - source/target buffer must be kernel space
1910 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1912 struct vm_area_struct *vma;
1913 struct mm_struct *mm;
1915 if (addr + len < addr)
1918 mm = get_task_mm(tsk);
1922 down_read(&mm->mmap_sem);
1924 /* the access must start within one of the target process's mappings */
1925 vma = find_vma(mm, addr);
1927 /* don't overrun this mapping */
1928 if (addr + len >= vma->vm_end)
1929 len = vma->vm_end - addr;
1931 /* only read or write mappings where it is permitted */
1932 if (write && vma->vm_flags & VM_MAYWRITE)
1933 copy_to_user_page(vma, NULL, addr,
1934 (void *) addr, buf, len);
1935 else if (!write && vma->vm_flags & VM_MAYREAD)
1936 copy_from_user_page(vma, NULL, addr,
1937 buf, (void *) addr, len);
1944 up_read(&mm->mmap_sem);
1950 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1951 * @inode: The inode to check
1952 * @size: The current filesize of the inode
1953 * @newsize: The proposed filesize of the inode
1955 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1956 * make sure that that any outstanding VMAs aren't broken and then shrink the
1957 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1958 * automatically grant mappings that are too large.
1960 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1963 struct vm_area_struct *vma;
1964 struct prio_tree_iter iter;
1965 struct vm_region *region;
1967 size_t r_size, r_top;
1969 low = newsize >> PAGE_SHIFT;
1970 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1972 down_write(&nommu_region_sem);
1974 /* search for VMAs that fall within the dead zone */
1975 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
1977 /* found one - only interested if it's shared out of the page
1979 if (vma->vm_flags & VM_SHARED) {
1980 up_write(&nommu_region_sem);
1981 return -ETXTBSY; /* not quite true, but near enough */
1985 /* reduce any regions that overlap the dead zone - if in existence,
1986 * these will be pointed to by VMAs that don't overlap the dead zone
1988 * we don't check for any regions that start beyond the EOF as there
1991 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
1993 if (!(vma->vm_flags & VM_SHARED))
1996 region = vma->vm_region;
1997 r_size = region->vm_top - region->vm_start;
1998 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2000 if (r_top > newsize) {
2001 region->vm_top -= r_top - newsize;
2002 if (region->vm_end > region->vm_top)
2003 region->vm_end = region->vm_top;
2007 up_write(&nommu_region_sem);