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 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>
38 static inline __attribute__((format(printf, 1, 2)))
39 void no_printk(const char *fmt, ...)
44 #define kenter(FMT, ...) \
45 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
46 #define kleave(FMT, ...) \
47 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
48 #define kdebug(FMT, ...) \
49 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
51 #define kenter(FMT, ...) \
52 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
53 #define kleave(FMT, ...) \
54 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
55 #define kdebug(FMT, ...) \
56 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
63 unsigned long max_mapnr;
64 unsigned long num_physpages;
65 atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
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 heap_stack_gap = 0;
71 atomic_t mmap_pages_allocated;
73 EXPORT_SYMBOL(mem_map);
74 EXPORT_SYMBOL(num_physpages);
76 /* list of mapped, potentially shareable regions */
77 static struct kmem_cache *vm_region_jar;
78 struct rb_root nommu_region_tree = RB_ROOT;
79 DECLARE_RWSEM(nommu_region_sem);
81 struct vm_operations_struct generic_file_vm_ops = {
85 * Handle all mappings that got truncated by a "truncate()"
88 * NOTE! We have to be ready to update the memory sharing
89 * between the file and the memory map for a potential last
90 * incomplete page. Ugly, but necessary.
92 int vmtruncate(struct inode *inode, loff_t offset)
94 struct address_space *mapping = inode->i_mapping;
97 if (inode->i_size < offset)
99 i_size_write(inode, offset);
101 truncate_inode_pages(mapping, offset);
105 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
106 if (limit != RLIM_INFINITY && offset > limit)
108 if (offset > inode->i_sb->s_maxbytes)
110 i_size_write(inode, offset);
113 if (inode->i_op->truncate)
114 inode->i_op->truncate(inode);
117 send_sig(SIGXFSZ, current, 0);
122 EXPORT_SYMBOL(vmtruncate);
125 * Return the total memory allocated for this pointer, not
126 * just what the caller asked for.
128 * Doesn't have to be accurate, i.e. may have races.
130 unsigned int kobjsize(const void *objp)
135 * If the object we have should not have ksize performed on it,
138 if (!objp || !virt_addr_valid(objp))
141 page = virt_to_head_page(objp);
144 * If the allocator sets PageSlab, we know the pointer came from
151 * The ksize() function is only guaranteed to work for pointers
152 * returned by kmalloc(). So handle arbitrary pointers here.
154 return PAGE_SIZE << compound_order(page);
157 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
158 unsigned long start, int len, int flags,
159 struct page **pages, struct vm_area_struct **vmas)
161 struct vm_area_struct *vma;
162 unsigned long vm_flags;
164 int write = !!(flags & GUP_FLAGS_WRITE);
165 int force = !!(flags & GUP_FLAGS_FORCE);
166 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
168 /* calculate required read or write permissions.
169 * - if 'force' is set, we only require the "MAY" flags.
171 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
172 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
174 for (i = 0; i < len; i++) {
175 vma = find_vma(mm, start);
177 goto finish_or_fault;
179 /* protect what we can, including chardevs */
180 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
181 (!ignore && !(vm_flags & vma->vm_flags)))
182 goto finish_or_fault;
185 pages[i] = virt_to_page(start);
187 page_cache_get(pages[i]);
197 return i ? : -EFAULT;
202 * get a list of pages in an address range belonging to the specified process
203 * and indicate the VMA that covers each page
204 * - this is potentially dodgy as we may end incrementing the page count of a
205 * slab page or a secondary page from a compound page
206 * - don't permit access to VMAs that don't support it, such as I/O mappings
208 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
209 unsigned long start, int len, int write, int force,
210 struct page **pages, struct vm_area_struct **vmas)
215 flags |= GUP_FLAGS_WRITE;
217 flags |= GUP_FLAGS_FORCE;
219 return __get_user_pages(tsk, mm,
223 EXPORT_SYMBOL(get_user_pages);
225 DEFINE_RWLOCK(vmlist_lock);
226 struct vm_struct *vmlist;
228 void vfree(const void *addr)
232 EXPORT_SYMBOL(vfree);
234 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
237 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
238 * returns only a logical address.
240 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
242 EXPORT_SYMBOL(__vmalloc);
244 void *vmalloc_user(unsigned long size)
248 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
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 memcpy(buf, addr, count);
282 long vwrite(char *buf, char *addr, unsigned long count)
284 /* Don't allow overflow */
285 if ((unsigned long) addr + count < count)
286 count = -(unsigned long) addr;
288 memcpy(addr, buf, count);
293 * vmalloc - allocate virtually continguos memory
295 * @size: allocation size
297 * Allocate enough pages to cover @size from the page level
298 * allocator and map them into continguos kernel virtual space.
300 * For tight control over page level allocator and protection flags
301 * use __vmalloc() instead.
303 void *vmalloc(unsigned long size)
305 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
307 EXPORT_SYMBOL(vmalloc);
309 void *vmalloc_node(unsigned long size, int node)
311 return vmalloc(size);
313 EXPORT_SYMBOL(vmalloc_node);
315 #ifndef PAGE_KERNEL_EXEC
316 # define PAGE_KERNEL_EXEC PAGE_KERNEL
320 * vmalloc_exec - allocate virtually contiguous, executable memory
321 * @size: allocation size
323 * Kernel-internal function to allocate enough pages to cover @size
324 * the page level allocator and map them into contiguous and
325 * executable kernel virtual space.
327 * For tight control over page level allocator and protection flags
328 * use __vmalloc() instead.
331 void *vmalloc_exec(unsigned long size)
333 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
337 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
338 * @size: allocation size
340 * Allocate enough 32bit PA addressable pages to cover @size from the
341 * page level allocator and map them into continguos kernel virtual space.
343 void *vmalloc_32(unsigned long size)
345 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
347 EXPORT_SYMBOL(vmalloc_32);
350 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
351 * @size: allocation size
353 * The resulting memory area is 32bit addressable and zeroed so it can be
354 * mapped to userspace without leaking data.
356 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
357 * remap_vmalloc_range() are permissible.
359 void *vmalloc_32_user(unsigned long size)
362 * We'll have to sort out the ZONE_DMA bits for 64-bit,
363 * but for now this can simply use vmalloc_user() directly.
365 return vmalloc_user(size);
367 EXPORT_SYMBOL(vmalloc_32_user);
369 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
376 void vunmap(const void *addr)
380 EXPORT_SYMBOL(vunmap);
383 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
386 void __attribute__((weak)) vmalloc_sync_all(void)
390 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
395 EXPORT_SYMBOL(vm_insert_page);
398 * sys_brk() for the most part doesn't need the global kernel
399 * lock, except when an application is doing something nasty
400 * like trying to un-brk an area that has already been mapped
401 * to a regular file. in this case, the unmapping will need
402 * to invoke file system routines that need the global lock.
404 asmlinkage unsigned long sys_brk(unsigned long brk)
406 struct mm_struct *mm = current->mm;
408 if (brk < mm->start_brk || brk > mm->context.end_brk)
415 * Always allow shrinking brk
417 if (brk <= mm->brk) {
423 * Ok, looks good - let it rip.
425 return mm->brk = brk;
429 * initialise the VMA and region record slabs
431 void __init mmap_init(void)
433 vm_region_jar = kmem_cache_create("vm_region_jar",
434 sizeof(struct vm_region), 0,
436 vm_area_cachep = kmem_cache_create("vm_area_struct",
437 sizeof(struct vm_area_struct), 0,
442 * validate the region tree
443 * - the caller must hold the region lock
445 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
446 static noinline void validate_nommu_regions(void)
448 struct vm_region *region, *last;
449 struct rb_node *p, *lastp;
451 lastp = rb_first(&nommu_region_tree);
455 last = rb_entry(lastp, struct vm_region, vm_rb);
456 if (unlikely(last->vm_end <= last->vm_start))
459 while ((p = rb_next(lastp))) {
460 region = rb_entry(p, struct vm_region, vm_rb);
461 last = rb_entry(lastp, struct vm_region, vm_rb);
463 if (unlikely(region->vm_end <= region->vm_start))
465 if (unlikely(region->vm_start < last->vm_end))
472 #define validate_nommu_regions() do {} while(0)
476 * add a region into the global tree
478 static void add_nommu_region(struct vm_region *region)
480 struct vm_region *pregion;
481 struct rb_node **p, *parent;
483 validate_nommu_regions();
485 BUG_ON(region->vm_start & ~PAGE_MASK);
488 p = &nommu_region_tree.rb_node;
491 pregion = rb_entry(parent, struct vm_region, vm_rb);
492 if (region->vm_start < pregion->vm_start)
494 else if (region->vm_start > pregion->vm_start)
496 else if (pregion == region)
502 rb_link_node(®ion->vm_rb, parent, p);
503 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
505 validate_nommu_regions();
509 * delete a region from the global tree
511 static void delete_nommu_region(struct vm_region *region)
513 BUG_ON(!nommu_region_tree.rb_node);
515 validate_nommu_regions();
516 rb_erase(®ion->vm_rb, &nommu_region_tree);
517 validate_nommu_regions();
521 * free a contiguous series of pages
523 static void free_page_series(unsigned long from, unsigned long to)
525 for (; from < to; from += PAGE_SIZE) {
526 struct page *page = virt_to_page(from);
528 kdebug("- free %lx", from);
529 atomic_dec(&mmap_pages_allocated);
530 if (page_count(page) != 1)
531 kdebug("free page %p [%d]", page, page_count(page));
537 * release a reference to a region
538 * - the caller must hold the region semaphore, which this releases
539 * - the region may not have been added to the tree yet, in which case vm_end
540 * will equal vm_start
542 static void __put_nommu_region(struct vm_region *region)
543 __releases(nommu_region_sem)
545 kenter("%p{%d}", region, atomic_read(®ion->vm_usage));
547 BUG_ON(!nommu_region_tree.rb_node);
549 if (atomic_dec_and_test(®ion->vm_usage)) {
550 if (region->vm_end > region->vm_start)
551 delete_nommu_region(region);
552 up_write(&nommu_region_sem);
555 fput(region->vm_file);
557 /* IO memory and memory shared directly out of the pagecache
558 * from ramfs/tmpfs mustn't be released here */
559 if (region->vm_flags & VM_MAPPED_COPY) {
560 kdebug("free series");
561 free_page_series(region->vm_start, region->vm_end);
563 kmem_cache_free(vm_region_jar, region);
565 up_write(&nommu_region_sem);
570 * release a reference to a region
572 static void put_nommu_region(struct vm_region *region)
574 down_write(&nommu_region_sem);
575 __put_nommu_region(region);
579 * add a VMA into a process's mm_struct in the appropriate place in the list
580 * and tree and add to the address space's page tree also if not an anonymous
582 * - should be called with mm->mmap_sem held writelocked
584 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
586 struct vm_area_struct *pvma, **pp;
587 struct address_space *mapping;
588 struct rb_node **p, *parent;
592 BUG_ON(!vma->vm_region);
597 /* add the VMA to the mapping */
599 mapping = vma->vm_file->f_mapping;
601 flush_dcache_mmap_lock(mapping);
602 vma_prio_tree_insert(vma, &mapping->i_mmap);
603 flush_dcache_mmap_unlock(mapping);
606 /* add the VMA to the tree */
608 p = &mm->mm_rb.rb_node;
611 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
613 /* sort by: start addr, end addr, VMA struct addr in that order
614 * (the latter is necessary as we may get identical VMAs) */
615 if (vma->vm_start < pvma->vm_start)
617 else if (vma->vm_start > pvma->vm_start)
619 else if (vma->vm_end < pvma->vm_end)
621 else if (vma->vm_end > pvma->vm_end)
631 rb_link_node(&vma->vm_rb, parent, p);
632 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
634 /* add VMA to the VMA list also */
635 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
636 if (pvma->vm_start > vma->vm_start)
638 if (pvma->vm_start < vma->vm_start)
640 if (pvma->vm_end < vma->vm_end)
649 * delete a VMA from its owning mm_struct and address space
651 static void delete_vma_from_mm(struct vm_area_struct *vma)
653 struct vm_area_struct **pp;
654 struct address_space *mapping;
655 struct mm_struct *mm = vma->vm_mm;
660 if (mm->mmap_cache == vma)
661 mm->mmap_cache = NULL;
663 /* remove the VMA from the mapping */
665 mapping = vma->vm_file->f_mapping;
667 flush_dcache_mmap_lock(mapping);
668 vma_prio_tree_remove(vma, &mapping->i_mmap);
669 flush_dcache_mmap_unlock(mapping);
672 /* remove from the MM's tree and list */
673 rb_erase(&vma->vm_rb, &mm->mm_rb);
674 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
685 * destroy a VMA record
687 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
690 if (vma->vm_ops && vma->vm_ops->close)
691 vma->vm_ops->close(vma);
694 if (vma->vm_flags & VM_EXECUTABLE)
695 removed_exe_file_vma(mm);
697 put_nommu_region(vma->vm_region);
698 kmem_cache_free(vm_area_cachep, vma);
702 * look up the first VMA in which addr resides, NULL if none
703 * - should be called with mm->mmap_sem at least held readlocked
705 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
707 struct vm_area_struct *vma;
708 struct rb_node *n = mm->mm_rb.rb_node;
710 /* check the cache first */
711 vma = mm->mmap_cache;
712 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
715 /* trawl the tree (there may be multiple mappings in which addr
717 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
718 vma = rb_entry(n, struct vm_area_struct, vm_rb);
719 if (vma->vm_start > addr)
721 if (vma->vm_end > addr) {
722 mm->mmap_cache = vma;
729 EXPORT_SYMBOL(find_vma);
733 * - we don't extend stack VMAs under NOMMU conditions
735 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
737 return find_vma(mm, addr);
741 * expand a stack to a given address
742 * - not supported under NOMMU conditions
744 int expand_stack(struct vm_area_struct *vma, unsigned long address)
750 * look up the first VMA exactly that exactly matches addr
751 * - should be called with mm->mmap_sem at least held readlocked
753 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
757 struct vm_area_struct *vma;
758 struct rb_node *n = mm->mm_rb.rb_node;
759 unsigned long end = addr + len;
761 /* check the cache first */
762 vma = mm->mmap_cache;
763 if (vma && vma->vm_start == addr && vma->vm_end == end)
766 /* trawl the tree (there may be multiple mappings in which addr
768 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
769 vma = rb_entry(n, struct vm_area_struct, vm_rb);
770 if (vma->vm_start < addr)
772 if (vma->vm_start > addr)
774 if (vma->vm_end == end) {
775 mm->mmap_cache = vma;
784 * determine whether a mapping should be permitted and, if so, what sort of
785 * mapping we're capable of supporting
787 static int validate_mmap_request(struct file *file,
793 unsigned long *_capabilities)
795 unsigned long capabilities, rlen;
796 unsigned long reqprot = prot;
799 /* do the simple checks first */
800 if (flags & MAP_FIXED || addr) {
802 "%d: Can't do fixed-address/overlay mmap of RAM\n",
807 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
808 (flags & MAP_TYPE) != MAP_SHARED)
814 /* Careful about overflows.. */
815 rlen = PAGE_ALIGN(len);
816 if (!rlen || rlen > TASK_SIZE)
819 /* offset overflow? */
820 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
824 /* validate file mapping requests */
825 struct address_space *mapping;
827 /* files must support mmap */
828 if (!file->f_op || !file->f_op->mmap)
831 /* work out if what we've got could possibly be shared
832 * - we support chardevs that provide their own "memory"
833 * - we support files/blockdevs that are memory backed
835 mapping = file->f_mapping;
837 mapping = file->f_path.dentry->d_inode->i_mapping;
840 if (mapping && mapping->backing_dev_info)
841 capabilities = mapping->backing_dev_info->capabilities;
844 /* no explicit capabilities set, so assume some
846 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
849 capabilities = BDI_CAP_MAP_COPY;
864 /* eliminate any capabilities that we can't support on this
866 if (!file->f_op->get_unmapped_area)
867 capabilities &= ~BDI_CAP_MAP_DIRECT;
868 if (!file->f_op->read)
869 capabilities &= ~BDI_CAP_MAP_COPY;
871 if (flags & MAP_SHARED) {
872 /* do checks for writing, appending and locking */
873 if ((prot & PROT_WRITE) &&
874 !(file->f_mode & FMODE_WRITE))
877 if (IS_APPEND(file->f_path.dentry->d_inode) &&
878 (file->f_mode & FMODE_WRITE))
881 if (locks_verify_locked(file->f_path.dentry->d_inode))
884 if (!(capabilities & BDI_CAP_MAP_DIRECT))
887 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
888 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
889 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
891 printk("MAP_SHARED not completely supported on !MMU\n");
895 /* we mustn't privatise shared mappings */
896 capabilities &= ~BDI_CAP_MAP_COPY;
899 /* we're going to read the file into private memory we
901 if (!(capabilities & BDI_CAP_MAP_COPY))
904 /* we don't permit a private writable mapping to be
905 * shared with the backing device */
906 if (prot & PROT_WRITE)
907 capabilities &= ~BDI_CAP_MAP_DIRECT;
910 /* handle executable mappings and implied executable
912 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
913 if (prot & PROT_EXEC)
916 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
917 /* handle implication of PROT_EXEC by PROT_READ */
918 if (current->personality & READ_IMPLIES_EXEC) {
919 if (capabilities & BDI_CAP_EXEC_MAP)
923 else if ((prot & PROT_READ) &&
924 (prot & PROT_EXEC) &&
925 !(capabilities & BDI_CAP_EXEC_MAP)
927 /* backing file is not executable, try to copy */
928 capabilities &= ~BDI_CAP_MAP_DIRECT;
932 /* anonymous mappings are always memory backed and can be
935 capabilities = BDI_CAP_MAP_COPY;
937 /* handle PROT_EXEC implication by PROT_READ */
938 if ((prot & PROT_READ) &&
939 (current->personality & READ_IMPLIES_EXEC))
943 /* allow the security API to have its say */
944 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
949 *_capabilities = capabilities;
954 * we've determined that we can make the mapping, now translate what we
955 * now know into VMA flags
957 static unsigned long determine_vm_flags(struct file *file,
960 unsigned long capabilities)
962 unsigned long vm_flags;
964 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
965 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
966 /* vm_flags |= mm->def_flags; */
968 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
969 /* attempt to share read-only copies of mapped file chunks */
970 if (file && !(prot & PROT_WRITE))
971 vm_flags |= VM_MAYSHARE;
974 /* overlay a shareable mapping on the backing device or inode
975 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
977 if (flags & MAP_SHARED)
978 vm_flags |= VM_MAYSHARE | VM_SHARED;
979 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
980 vm_flags |= VM_MAYSHARE;
983 /* refuse to let anyone share private mappings with this process if
984 * it's being traced - otherwise breakpoints set in it may interfere
985 * with another untraced process
987 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
988 vm_flags &= ~VM_MAYSHARE;
994 * set up a shared mapping on a file (the driver or filesystem provides and
997 static int do_mmap_shared_file(struct vm_area_struct *vma)
1001 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1005 /* getting an ENOSYS error indicates that direct mmap isn't
1006 * possible (as opposed to tried but failed) so we'll fall
1007 * through to making a private copy of the data and mapping
1013 * set up a private mapping or an anonymous shared mapping
1015 static int do_mmap_private(struct vm_area_struct *vma,
1016 struct vm_region *region,
1020 unsigned long total, point, n, rlen;
1024 /* invoke the file's mapping function so that it can keep track of
1025 * shared mappings on devices or memory
1026 * - VM_MAYSHARE will be set if it may attempt to share
1029 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1030 if (ret != -ENOSYS) {
1031 /* shouldn't return success if we're not sharing */
1032 BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
1033 return ret; /* success or a real error */
1036 /* getting an ENOSYS error indicates that direct mmap isn't
1037 * possible (as opposed to tried but failed) so we'll try to
1038 * make a private copy of the data and map that instead */
1041 rlen = PAGE_ALIGN(len);
1043 /* allocate some memory to hold the mapping
1044 * - note that this may not return a page-aligned address if the object
1045 * we're allocating is smaller than a page
1047 order = get_order(rlen);
1048 kdebug("alloc order %d for %lx", order, len);
1050 pages = alloc_pages(GFP_KERNEL, order);
1054 /* we allocated a power-of-2 sized page set, so we need to trim off the
1057 atomic_add(total, &mmap_pages_allocated);
1059 point = rlen >> PAGE_SHIFT;
1060 while (total > point) {
1061 order = ilog2(total - point);
1063 kdebug("shave %lu/%lu @%lu", n, total - point, total);
1064 atomic_sub(n, &mmap_pages_allocated);
1066 set_page_refcounted(pages + total);
1067 __free_pages(pages + total, order);
1070 total = rlen >> PAGE_SHIFT;
1071 for (point = 1; point < total; point++)
1072 set_page_refcounted(&pages[point]);
1074 base = page_address(pages);
1075 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1076 region->vm_start = (unsigned long) base;
1077 region->vm_end = region->vm_start + rlen;
1079 vma->vm_start = region->vm_start;
1080 vma->vm_end = region->vm_start + len;
1083 /* read the contents of a file into the copy */
1084 mm_segment_t old_fs;
1087 fpos = vma->vm_pgoff;
1088 fpos <<= PAGE_SHIFT;
1092 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1098 /* clear the last little bit */
1100 memset(base + ret, 0, rlen - ret);
1103 /* if it's an anonymous mapping, then just clear it */
1104 memset(base, 0, rlen);
1110 free_page_series(region->vm_start, region->vm_end);
1111 region->vm_start = vma->vm_start = 0;
1112 region->vm_end = vma->vm_end = 0;
1116 printk("Allocation of length %lu from process %d failed\n",
1123 * handle mapping creation for uClinux
1125 unsigned long do_mmap_pgoff(struct file *file,
1129 unsigned long flags,
1130 unsigned long pgoff)
1132 struct vm_area_struct *vma;
1133 struct vm_region *region;
1135 unsigned long capabilities, vm_flags, result;
1138 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1140 if (!(flags & MAP_FIXED))
1141 addr = round_hint_to_min(addr);
1143 /* decide whether we should attempt the mapping, and if so what sort of
1145 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1148 kleave(" = %d [val]", ret);
1152 /* we've determined that we can make the mapping, now translate what we
1153 * now know into VMA flags */
1154 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1156 /* we're going to need to record the mapping */
1157 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1159 goto error_getting_region;
1161 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1163 goto error_getting_vma;
1165 atomic_set(®ion->vm_usage, 1);
1166 region->vm_flags = vm_flags;
1167 region->vm_pgoff = pgoff;
1169 INIT_LIST_HEAD(&vma->anon_vma_node);
1170 vma->vm_flags = vm_flags;
1171 vma->vm_pgoff = pgoff;
1174 region->vm_file = file;
1176 vma->vm_file = file;
1178 if (vm_flags & VM_EXECUTABLE) {
1179 added_exe_file_vma(current->mm);
1180 vma->vm_mm = current->mm;
1184 down_write(&nommu_region_sem);
1186 /* if we want to share, we need to check for regions created by other
1187 * mmap() calls that overlap with our proposed mapping
1188 * - we can only share with a superset match on most regular files
1189 * - shared mappings on character devices and memory backed files are
1190 * permitted to overlap inexactly as far as we are concerned for in
1191 * these cases, sharing is handled in the driver or filesystem rather
1194 if (vm_flags & VM_MAYSHARE) {
1195 struct vm_region *pregion;
1196 unsigned long pglen, rpglen, pgend, rpgend, start;
1198 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1199 pgend = pgoff + pglen;
1201 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1202 pregion = rb_entry(rb, struct vm_region, vm_rb);
1204 if (!(pregion->vm_flags & VM_MAYSHARE))
1207 /* search for overlapping mappings on the same file */
1208 if (pregion->vm_file->f_path.dentry->d_inode !=
1209 file->f_path.dentry->d_inode)
1212 if (pregion->vm_pgoff >= pgend)
1215 rpglen = pregion->vm_end - pregion->vm_start;
1216 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1217 rpgend = pregion->vm_pgoff + rpglen;
1218 if (pgoff >= rpgend)
1221 /* handle inexactly overlapping matches between
1223 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1224 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1225 /* new mapping is not a subset of the region */
1226 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1227 goto sharing_violation;
1231 /* we've found a region we can share */
1232 atomic_inc(&pregion->vm_usage);
1233 vma->vm_region = pregion;
1234 start = pregion->vm_start;
1235 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1236 vma->vm_start = start;
1237 vma->vm_end = start + len;
1239 if (pregion->vm_flags & VM_MAPPED_COPY) {
1240 kdebug("share copy");
1241 vma->vm_flags |= VM_MAPPED_COPY;
1243 kdebug("share mmap");
1244 ret = do_mmap_shared_file(vma);
1246 vma->vm_region = NULL;
1249 atomic_dec(&pregion->vm_usage);
1251 goto error_just_free;
1254 fput(region->vm_file);
1255 kmem_cache_free(vm_region_jar, region);
1261 /* obtain the address at which to make a shared mapping
1262 * - this is the hook for quasi-memory character devices to
1263 * tell us the location of a shared mapping
1265 if (file && file->f_op->get_unmapped_area) {
1266 addr = file->f_op->get_unmapped_area(file, addr, len,
1268 if (IS_ERR((void *) addr)) {
1270 if (ret != (unsigned long) -ENOSYS)
1271 goto error_just_free;
1273 /* the driver refused to tell us where to site
1274 * the mapping so we'll have to attempt to copy
1276 ret = (unsigned long) -ENODEV;
1277 if (!(capabilities & BDI_CAP_MAP_COPY))
1278 goto error_just_free;
1280 capabilities &= ~BDI_CAP_MAP_DIRECT;
1282 vma->vm_start = region->vm_start = addr;
1283 vma->vm_end = region->vm_end = addr + len;
1288 vma->vm_region = region;
1290 /* set up the mapping */
1291 if (file && vma->vm_flags & VM_SHARED)
1292 ret = do_mmap_shared_file(vma);
1294 ret = do_mmap_private(vma, region, len);
1296 goto error_put_region;
1298 add_nommu_region(region);
1300 /* okay... we have a mapping; now we have to register it */
1301 result = vma->vm_start;
1303 current->mm->total_vm += len >> PAGE_SHIFT;
1306 add_vma_to_mm(current->mm, vma);
1308 up_write(&nommu_region_sem);
1310 if (prot & PROT_EXEC)
1311 flush_icache_range(result, result + len);
1313 kleave(" = %lx", result);
1317 __put_nommu_region(region);
1321 if (vma->vm_flags & VM_EXECUTABLE)
1322 removed_exe_file_vma(vma->vm_mm);
1324 kmem_cache_free(vm_area_cachep, vma);
1326 kleave(" = %d [pr]", ret);
1330 up_write(&nommu_region_sem);
1332 fput(region->vm_file);
1333 kmem_cache_free(vm_region_jar, region);
1335 if (vma->vm_flags & VM_EXECUTABLE)
1336 removed_exe_file_vma(vma->vm_mm);
1337 kmem_cache_free(vm_area_cachep, vma);
1338 kleave(" = %d", ret);
1342 up_write(&nommu_region_sem);
1343 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1348 kmem_cache_free(vm_region_jar, region);
1349 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1350 " from process %d failed\n",
1355 error_getting_region:
1356 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1357 " from process %d failed\n",
1362 EXPORT_SYMBOL(do_mmap_pgoff);
1365 * split a vma into two pieces at address 'addr', a new vma is allocated either
1366 * for the first part or the tail.
1368 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1369 unsigned long addr, int new_below)
1371 struct vm_area_struct *new;
1372 struct vm_region *region;
1373 unsigned long npages;
1377 /* we're only permitted to split anonymous regions that have a single
1380 atomic_read(&vma->vm_region->vm_usage) != 1)
1383 if (mm->map_count >= sysctl_max_map_count)
1386 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1390 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1392 kmem_cache_free(vm_region_jar, region);
1396 /* most fields are the same, copy all, and then fixup */
1398 *region = *vma->vm_region;
1399 new->vm_region = region;
1401 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1404 region->vm_end = new->vm_end = addr;
1406 region->vm_start = new->vm_start = addr;
1407 region->vm_pgoff = new->vm_pgoff += npages;
1410 if (new->vm_ops && new->vm_ops->open)
1411 new->vm_ops->open(new);
1413 delete_vma_from_mm(vma);
1414 down_write(&nommu_region_sem);
1415 delete_nommu_region(vma->vm_region);
1417 vma->vm_region->vm_start = vma->vm_start = addr;
1418 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1420 vma->vm_region->vm_end = vma->vm_end = addr;
1422 add_nommu_region(vma->vm_region);
1423 add_nommu_region(new->vm_region);
1424 up_write(&nommu_region_sem);
1425 add_vma_to_mm(mm, vma);
1426 add_vma_to_mm(mm, new);
1431 * shrink a VMA by removing the specified chunk from either the beginning or
1434 static int shrink_vma(struct mm_struct *mm,
1435 struct vm_area_struct *vma,
1436 unsigned long from, unsigned long to)
1438 struct vm_region *region;
1442 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1444 delete_vma_from_mm(vma);
1445 if (from > vma->vm_start)
1449 add_vma_to_mm(mm, vma);
1451 /* cut the backing region down to size */
1452 region = vma->vm_region;
1453 BUG_ON(atomic_read(®ion->vm_usage) != 1);
1455 down_write(&nommu_region_sem);
1456 delete_nommu_region(region);
1457 if (from > region->vm_start)
1458 region->vm_end = from;
1460 region->vm_start = to;
1461 add_nommu_region(region);
1462 up_write(&nommu_region_sem);
1464 free_page_series(from, to);
1470 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1471 * VMA, though it need not cover the whole VMA
1473 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1475 struct vm_area_struct *vma;
1477 unsigned long end = start + len;
1480 kenter(",%lx,%zx", start, len);
1485 /* find the first potentially overlapping VMA */
1486 vma = find_vma(mm, start);
1489 "munmap of memory not mmapped by process %d (%s):"
1491 current->pid, current->comm, start, start + len - 1);
1495 /* we're allowed to split an anonymous VMA but not a file-backed one */
1498 if (start > vma->vm_start) {
1499 kleave(" = -EINVAL [miss]");
1502 if (end == vma->vm_end)
1503 goto erase_whole_vma;
1504 rb = rb_next(&vma->vm_rb);
1505 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1507 kleave(" = -EINVAL [split file]");
1510 /* the chunk must be a subset of the VMA found */
1511 if (start == vma->vm_start && end == vma->vm_end)
1512 goto erase_whole_vma;
1513 if (start < vma->vm_start || end > vma->vm_end) {
1514 kleave(" = -EINVAL [superset]");
1517 if (start & ~PAGE_MASK) {
1518 kleave(" = -EINVAL [unaligned start]");
1521 if (end != vma->vm_end && end & ~PAGE_MASK) {
1522 kleave(" = -EINVAL [unaligned split]");
1525 if (start != vma->vm_start && end != vma->vm_end) {
1526 ret = split_vma(mm, vma, start, 1);
1528 kleave(" = %d [split]", ret);
1532 return shrink_vma(mm, vma, start, end);
1536 delete_vma_from_mm(vma);
1537 delete_vma(mm, vma);
1541 EXPORT_SYMBOL(do_munmap);
1543 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1546 struct mm_struct *mm = current->mm;
1548 down_write(&mm->mmap_sem);
1549 ret = do_munmap(mm, addr, len);
1550 up_write(&mm->mmap_sem);
1555 * release all the mappings made in a process's VM space
1557 void exit_mmap(struct mm_struct *mm)
1559 struct vm_area_struct *vma;
1568 while ((vma = mm->mmap)) {
1569 mm->mmap = vma->vm_next;
1570 delete_vma_from_mm(vma);
1571 delete_vma(mm, vma);
1577 unsigned long do_brk(unsigned long addr, unsigned long len)
1583 * expand (or shrink) an existing mapping, potentially moving it at the same
1584 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1586 * under NOMMU conditions, we only permit changing a mapping's size, and only
1587 * as long as it stays within the region allocated by do_mmap_private() and the
1588 * block is not shareable
1590 * MREMAP_FIXED is not supported under NOMMU conditions
1592 unsigned long do_mremap(unsigned long addr,
1593 unsigned long old_len, unsigned long new_len,
1594 unsigned long flags, unsigned long new_addr)
1596 struct vm_area_struct *vma;
1598 /* insanity checks first */
1599 if (old_len == 0 || new_len == 0)
1600 return (unsigned long) -EINVAL;
1602 if (addr & ~PAGE_MASK)
1605 if (flags & MREMAP_FIXED && new_addr != addr)
1606 return (unsigned long) -EINVAL;
1608 vma = find_vma_exact(current->mm, addr, old_len);
1610 return (unsigned long) -EINVAL;
1612 if (vma->vm_end != vma->vm_start + old_len)
1613 return (unsigned long) -EFAULT;
1615 if (vma->vm_flags & VM_MAYSHARE)
1616 return (unsigned long) -EPERM;
1618 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1619 return (unsigned long) -ENOMEM;
1621 /* all checks complete - do it */
1622 vma->vm_end = vma->vm_start + new_len;
1623 return vma->vm_start;
1625 EXPORT_SYMBOL(do_mremap);
1628 unsigned long sys_mremap(unsigned long addr,
1629 unsigned long old_len, unsigned long new_len,
1630 unsigned long flags, unsigned long new_addr)
1634 down_write(¤t->mm->mmap_sem);
1635 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1636 up_write(¤t->mm->mmap_sem);
1640 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1641 unsigned int foll_flags)
1646 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1647 unsigned long to, unsigned long size, pgprot_t prot)
1649 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1652 EXPORT_SYMBOL(remap_pfn_range);
1654 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1655 unsigned long pgoff)
1657 unsigned int size = vma->vm_end - vma->vm_start;
1659 if (!(vma->vm_flags & VM_USERMAP))
1662 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1663 vma->vm_end = vma->vm_start + size;
1667 EXPORT_SYMBOL(remap_vmalloc_range);
1669 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1673 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1674 unsigned long len, unsigned long pgoff, unsigned long flags)
1679 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1683 void unmap_mapping_range(struct address_space *mapping,
1684 loff_t const holebegin, loff_t const holelen,
1688 EXPORT_SYMBOL(unmap_mapping_range);
1691 * ask for an unmapped area at which to create a mapping on a file
1693 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1694 unsigned long len, unsigned long pgoff,
1695 unsigned long flags)
1697 unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1698 unsigned long, unsigned long);
1700 get_area = current->mm->get_unmapped_area;
1701 if (file && file->f_op && file->f_op->get_unmapped_area)
1702 get_area = file->f_op->get_unmapped_area;
1707 return get_area(file, addr, len, pgoff, flags);
1709 EXPORT_SYMBOL(get_unmapped_area);
1712 * Check that a process has enough memory to allocate a new virtual
1713 * mapping. 0 means there is enough memory for the allocation to
1714 * succeed and -ENOMEM implies there is not.
1716 * We currently support three overcommit policies, which are set via the
1717 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1719 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1720 * Additional code 2002 Jul 20 by Robert Love.
1722 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1724 * Note this is a helper function intended to be used by LSMs which
1725 * wish to use this logic.
1727 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1729 unsigned long free, allowed;
1731 vm_acct_memory(pages);
1734 * Sometimes we want to use more memory than we have
1736 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1739 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1742 free = global_page_state(NR_FILE_PAGES);
1743 free += nr_swap_pages;
1746 * Any slabs which are created with the
1747 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1748 * which are reclaimable, under pressure. The dentry
1749 * cache and most inode caches should fall into this
1751 free += global_page_state(NR_SLAB_RECLAIMABLE);
1754 * Leave the last 3% for root
1763 * nr_free_pages() is very expensive on large systems,
1764 * only call if we're about to fail.
1766 n = nr_free_pages();
1769 * Leave reserved pages. The pages are not for anonymous pages.
1771 if (n <= totalreserve_pages)
1774 n -= totalreserve_pages;
1777 * Leave the last 3% for root
1789 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1791 * Leave the last 3% for root
1794 allowed -= allowed / 32;
1795 allowed += total_swap_pages;
1797 /* Don't let a single process grow too big:
1798 leave 3% of the size of this process for other processes */
1800 allowed -= mm->total_vm / 32;
1803 * cast `allowed' as a signed long because vm_committed_space
1804 * sometimes has a negative value
1806 if (atomic_long_read(&vm_committed_space) < (long)allowed)
1809 vm_unacct_memory(pages);
1814 int in_gate_area_no_task(unsigned long addr)
1819 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1824 EXPORT_SYMBOL(filemap_fault);
1827 * Access another process' address space.
1828 * - source/target buffer must be kernel space
1830 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1832 struct vm_area_struct *vma;
1833 struct mm_struct *mm;
1835 if (addr + len < addr)
1838 mm = get_task_mm(tsk);
1842 down_read(&mm->mmap_sem);
1844 /* the access must start within one of the target process's mappings */
1845 vma = find_vma(mm, addr);
1847 /* don't overrun this mapping */
1848 if (addr + len >= vma->vm_end)
1849 len = vma->vm_end - addr;
1851 /* only read or write mappings where it is permitted */
1852 if (write && vma->vm_flags & VM_MAYWRITE)
1853 len -= copy_to_user((void *) addr, buf, len);
1854 else if (!write && vma->vm_flags & VM_MAYREAD)
1855 len -= copy_from_user(buf, (void *) addr, len);
1862 up_read(&mm->mmap_sem);