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) 2004-2005 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>
16 #include <linux/mman.h>
17 #include <linux/swap.h>
18 #include <linux/file.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ptrace.h>
24 #include <linux/blkdev.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mount.h>
27 #include <linux/personality.h>
28 #include <linux/security.h>
29 #include <linux/syscalls.h>
31 #include <asm/uaccess.h>
33 #include <asm/tlbflush.h>
37 unsigned long max_mapnr;
38 unsigned long num_physpages;
39 unsigned long askedalloc, realalloc;
40 atomic_t vm_committed_space = ATOMIC_INIT(0);
41 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
42 int sysctl_overcommit_ratio = 50; /* default is 50% */
43 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
44 int heap_stack_gap = 0;
46 EXPORT_SYMBOL(mem_map);
47 EXPORT_SYMBOL(__vm_enough_memory);
49 /* list of shareable VMAs */
50 struct rb_root nommu_vma_tree = RB_ROOT;
51 DECLARE_RWSEM(nommu_vma_sem);
53 struct vm_operations_struct generic_file_vm_ops = {
57 EXPORT_SYMBOL(vmalloc_to_page);
58 EXPORT_SYMBOL(vmalloc_32);
60 EXPORT_SYMBOL(vunmap);
63 * Handle all mappings that got truncated by a "truncate()"
66 * NOTE! We have to be ready to update the memory sharing
67 * between the file and the memory map for a potential last
68 * incomplete page. Ugly, but necessary.
70 int vmtruncate(struct inode *inode, loff_t offset)
72 struct address_space *mapping = inode->i_mapping;
75 if (inode->i_size < offset)
77 i_size_write(inode, offset);
79 truncate_inode_pages(mapping, offset);
83 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
84 if (limit != RLIM_INFINITY && offset > limit)
86 if (offset > inode->i_sb->s_maxbytes)
88 i_size_write(inode, offset);
91 if (inode->i_op && inode->i_op->truncate)
92 inode->i_op->truncate(inode);
95 send_sig(SIGXFSZ, current, 0);
100 EXPORT_SYMBOL(vmtruncate);
103 * Return the total memory allocated for this pointer, not
104 * just what the caller asked for.
106 * Doesn't have to be accurate, i.e. may have races.
108 unsigned int kobjsize(const void *objp)
112 if (!objp || !((page = virt_to_page(objp))))
118 BUG_ON(page->index < 0);
119 BUG_ON(page->index >= MAX_ORDER);
121 return (PAGE_SIZE << page->index);
125 * get a list of pages in an address range belonging to the specified process
126 * and indicate the VMA that covers each page
127 * - this is potentially dodgy as we may end incrementing the page count of a
128 * slab page or a secondary page from a compound page
129 * - don't permit access to VMAs that don't support it, such as I/O mappings
131 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
132 unsigned long start, int len, int write, int force,
133 struct page **pages, struct vm_area_struct **vmas)
135 struct vm_area_struct *vma;
136 unsigned long vm_flags;
139 /* calculate required read or write permissions.
140 * - if 'force' is set, we only require the "MAY" flags.
142 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
143 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
145 for (i = 0; i < len; i++) {
146 vma = find_vma(mm, start);
148 goto finish_or_fault;
150 /* protect what we can, including chardevs */
151 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
152 !(vm_flags & vma->vm_flags))
153 goto finish_or_fault;
156 pages[i] = virt_to_page(start);
158 page_cache_get(pages[i]);
168 return i ? : -EFAULT;
171 EXPORT_SYMBOL(get_user_pages);
173 DEFINE_RWLOCK(vmlist_lock);
174 struct vm_struct *vmlist;
176 void vfree(void *addr)
181 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
184 * kmalloc doesn't like __GFP_HIGHMEM for some reason
186 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
189 struct page * vmalloc_to_page(void *addr)
191 return virt_to_page(addr);
194 unsigned long vmalloc_to_pfn(void *addr)
196 return page_to_pfn(virt_to_page(addr));
200 long vread(char *buf, char *addr, unsigned long count)
202 memcpy(buf, addr, count);
206 long vwrite(char *buf, char *addr, unsigned long count)
208 /* Don't allow overflow */
209 if ((unsigned long) addr + count < count)
210 count = -(unsigned long) addr;
212 memcpy(addr, buf, count);
217 * vmalloc - allocate virtually continguos memory
219 * @size: allocation size
221 * Allocate enough pages to cover @size from the page level
222 * allocator and map them into continguos kernel virtual space.
224 * For tight cotrol over page level allocator and protection flags
225 * use __vmalloc() instead.
227 void *vmalloc(unsigned long size)
229 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
231 EXPORT_SYMBOL(vmalloc);
233 void *vmalloc_node(unsigned long size, int node)
235 return vmalloc(size);
237 EXPORT_SYMBOL(vmalloc_node);
240 * vmalloc_32 - allocate virtually continguos memory (32bit addressable)
242 * @size: allocation size
244 * Allocate enough 32bit PA addressable pages to cover @size from the
245 * page level allocator and map them into continguos kernel virtual space.
247 void *vmalloc_32(unsigned long size)
249 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
252 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
258 void vunmap(void *addr)
264 * sys_brk() for the most part doesn't need the global kernel
265 * lock, except when an application is doing something nasty
266 * like trying to un-brk an area that has already been mapped
267 * to a regular file. in this case, the unmapping will need
268 * to invoke file system routines that need the global lock.
270 asmlinkage unsigned long sys_brk(unsigned long brk)
272 struct mm_struct *mm = current->mm;
274 if (brk < mm->start_brk || brk > mm->context.end_brk)
281 * Always allow shrinking brk
283 if (brk <= mm->brk) {
289 * Ok, looks good - let it rip.
291 return mm->brk = brk;
295 static void show_process_blocks(void)
297 struct vm_list_struct *vml;
299 printk("Process blocks %d:", current->pid);
301 for (vml = ¤t->mm->context.vmlist; vml; vml = vml->next) {
302 printk(" %p: %p", vml, vml->vma);
304 printk(" (%d @%lx #%d)",
305 kobjsize((void *) vml->vma->vm_start),
307 atomic_read(&vml->vma->vm_usage));
308 printk(vml->next ? " ->" : ".\n");
313 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
315 struct vm_area_struct *vma;
316 struct rb_node *n = nommu_vma_tree.rb_node;
319 vma = rb_entry(n, struct vm_area_struct, vm_rb);
321 if (start < vma->vm_start)
323 else if (start > vma->vm_start)
332 static void add_nommu_vma(struct vm_area_struct *vma)
334 struct vm_area_struct *pvma;
335 struct address_space *mapping;
336 struct rb_node **p = &nommu_vma_tree.rb_node;
337 struct rb_node *parent = NULL;
339 /* add the VMA to the mapping */
341 mapping = vma->vm_file->f_mapping;
343 flush_dcache_mmap_lock(mapping);
344 vma_prio_tree_insert(vma, &mapping->i_mmap);
345 flush_dcache_mmap_unlock(mapping);
348 /* add the VMA to the master list */
351 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
353 if (vma->vm_start < pvma->vm_start) {
356 else if (vma->vm_start > pvma->vm_start) {
360 /* mappings are at the same address - this can only
361 * happen for shared-mem chardevs and shared file
362 * mappings backed by ramfs/tmpfs */
363 BUG_ON(!(pvma->vm_flags & VM_SHARED));
374 rb_link_node(&vma->vm_rb, parent, p);
375 rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
378 static void delete_nommu_vma(struct vm_area_struct *vma)
380 struct address_space *mapping;
382 /* remove the VMA from the mapping */
384 mapping = vma->vm_file->f_mapping;
386 flush_dcache_mmap_lock(mapping);
387 vma_prio_tree_remove(vma, &mapping->i_mmap);
388 flush_dcache_mmap_unlock(mapping);
391 /* remove from the master list */
392 rb_erase(&vma->vm_rb, &nommu_vma_tree);
396 * determine whether a mapping should be permitted and, if so, what sort of
397 * mapping we're capable of supporting
399 static int validate_mmap_request(struct file *file,
405 unsigned long *_capabilities)
407 unsigned long capabilities;
408 unsigned long reqprot = prot;
411 /* do the simple checks first */
412 if (flags & MAP_FIXED || addr) {
414 "%d: Can't do fixed-address/overlay mmap of RAM\n",
419 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
420 (flags & MAP_TYPE) != MAP_SHARED)
423 if (PAGE_ALIGN(len) == 0)
429 /* offset overflow? */
430 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
434 /* validate file mapping requests */
435 struct address_space *mapping;
437 /* files must support mmap */
438 if (!file->f_op || !file->f_op->mmap)
441 /* work out if what we've got could possibly be shared
442 * - we support chardevs that provide their own "memory"
443 * - we support files/blockdevs that are memory backed
445 mapping = file->f_mapping;
447 mapping = file->f_dentry->d_inode->i_mapping;
450 if (mapping && mapping->backing_dev_info)
451 capabilities = mapping->backing_dev_info->capabilities;
454 /* no explicit capabilities set, so assume some
456 switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
459 capabilities = BDI_CAP_MAP_COPY;
474 /* eliminate any capabilities that we can't support on this
476 if (!file->f_op->get_unmapped_area)
477 capabilities &= ~BDI_CAP_MAP_DIRECT;
478 if (!file->f_op->read)
479 capabilities &= ~BDI_CAP_MAP_COPY;
481 if (flags & MAP_SHARED) {
482 /* do checks for writing, appending and locking */
483 if ((prot & PROT_WRITE) &&
484 !(file->f_mode & FMODE_WRITE))
487 if (IS_APPEND(file->f_dentry->d_inode) &&
488 (file->f_mode & FMODE_WRITE))
491 if (locks_verify_locked(file->f_dentry->d_inode))
494 if (!(capabilities & BDI_CAP_MAP_DIRECT))
497 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
498 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
499 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
501 printk("MAP_SHARED not completely supported on !MMU\n");
505 /* we mustn't privatise shared mappings */
506 capabilities &= ~BDI_CAP_MAP_COPY;
509 /* we're going to read the file into private memory we
511 if (!(capabilities & BDI_CAP_MAP_COPY))
514 /* we don't permit a private writable mapping to be
515 * shared with the backing device */
516 if (prot & PROT_WRITE)
517 capabilities &= ~BDI_CAP_MAP_DIRECT;
520 /* handle executable mappings and implied executable
522 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
523 if (prot & PROT_EXEC)
526 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
527 /* handle implication of PROT_EXEC by PROT_READ */
528 if (current->personality & READ_IMPLIES_EXEC) {
529 if (capabilities & BDI_CAP_EXEC_MAP)
533 else if ((prot & PROT_READ) &&
534 (prot & PROT_EXEC) &&
535 !(capabilities & BDI_CAP_EXEC_MAP)
537 /* backing file is not executable, try to copy */
538 capabilities &= ~BDI_CAP_MAP_DIRECT;
542 /* anonymous mappings are always memory backed and can be
545 capabilities = BDI_CAP_MAP_COPY;
547 /* handle PROT_EXEC implication by PROT_READ */
548 if ((prot & PROT_READ) &&
549 (current->personality & READ_IMPLIES_EXEC))
553 /* allow the security API to have its say */
554 ret = security_file_mmap(file, reqprot, prot, flags);
559 *_capabilities = capabilities;
564 * we've determined that we can make the mapping, now translate what we
565 * now know into VMA flags
567 static unsigned long determine_vm_flags(struct file *file,
570 unsigned long capabilities)
572 unsigned long vm_flags;
574 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
575 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
576 /* vm_flags |= mm->def_flags; */
578 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
579 /* attempt to share read-only copies of mapped file chunks */
580 if (file && !(prot & PROT_WRITE))
581 vm_flags |= VM_MAYSHARE;
584 /* overlay a shareable mapping on the backing device or inode
585 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
587 if (flags & MAP_SHARED)
588 vm_flags |= VM_MAYSHARE | VM_SHARED;
589 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
590 vm_flags |= VM_MAYSHARE;
593 /* refuse to let anyone share private mappings with this process if
594 * it's being traced - otherwise breakpoints set in it may interfere
595 * with another untraced process
597 if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
598 vm_flags &= ~VM_MAYSHARE;
604 * set up a shared mapping on a file
606 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
610 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
614 /* getting an ENOSYS error indicates that direct mmap isn't
615 * possible (as opposed to tried but failed) so we'll fall
616 * through to making a private copy of the data and mapping
622 * set up a private mapping or an anonymous shared mapping
624 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
629 /* invoke the file's mapping function so that it can keep track of
630 * shared mappings on devices or memory
631 * - VM_MAYSHARE will be set if it may attempt to share
634 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
635 if (ret != -ENOSYS) {
636 /* shouldn't return success if we're not sharing */
637 BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
638 return ret; /* success or a real error */
641 /* getting an ENOSYS error indicates that direct mmap isn't
642 * possible (as opposed to tried but failed) so we'll try to
643 * make a private copy of the data and map that instead */
646 /* allocate some memory to hold the mapping
647 * - note that this may not return a page-aligned address if the object
648 * we're allocating is smaller than a page
650 base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
654 vma->vm_start = (unsigned long) base;
655 vma->vm_end = vma->vm_start + len;
656 vma->vm_flags |= VM_MAPPED_COPY;
659 if (len + WARN_ON_SLACK <= kobjsize(result))
660 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
661 len, current->pid, kobjsize(result) - len);
665 /* read the contents of a file into the copy */
669 fpos = vma->vm_pgoff;
674 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
680 /* clear the last little bit */
682 memset(base + ret, 0, len - ret);
685 /* if it's an anonymous mapping, then just clear it */
686 memset(base, 0, len);
697 printk("Allocation of length %lu from process %d failed\n",
704 * handle mapping creation for uClinux
706 unsigned long do_mmap_pgoff(struct file *file,
713 struct vm_list_struct *vml = NULL;
714 struct vm_area_struct *vma = NULL;
716 unsigned long capabilities, vm_flags;
720 /* decide whether we should attempt the mapping, and if so what sort of
722 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
727 /* we've determined that we can make the mapping, now translate what we
728 * now know into VMA flags */
729 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
731 /* we're going to need to record the mapping if it works */
732 vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
734 goto error_getting_vml;
735 memset(vml, 0, sizeof(*vml));
737 down_write(&nommu_vma_sem);
739 /* if we want to share, we need to check for VMAs created by other
740 * mmap() calls that overlap with our proposed mapping
741 * - we can only share with an exact match on most regular files
742 * - shared mappings on character devices and memory backed files are
743 * permitted to overlap inexactly as far as we are concerned for in
744 * these cases, sharing is handled in the driver or filesystem rather
747 if (vm_flags & VM_MAYSHARE) {
748 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
749 unsigned long vmpglen;
751 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
752 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
754 if (!(vma->vm_flags & VM_MAYSHARE))
757 /* search for overlapping mappings on the same file */
758 if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
761 if (vma->vm_pgoff >= pgoff + pglen)
764 vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
765 vmpglen >>= PAGE_SHIFT;
766 if (pgoff >= vma->vm_pgoff + vmpglen)
769 /* handle inexactly overlapping matches between mappings */
770 if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
771 if (!(capabilities & BDI_CAP_MAP_DIRECT))
772 goto sharing_violation;
776 /* we've found a VMA we can share */
777 atomic_inc(&vma->vm_usage);
780 result = (void *) vma->vm_start;
786 /* obtain the address at which to make a shared mapping
787 * - this is the hook for quasi-memory character devices to
788 * tell us the location of a shared mapping
790 if (file && file->f_op->get_unmapped_area) {
791 addr = file->f_op->get_unmapped_area(file, addr, len,
793 if (IS_ERR((void *) addr)) {
795 if (ret != (unsigned long) -ENOSYS)
798 /* the driver refused to tell us where to site
799 * the mapping so we'll have to attempt to copy
801 ret = (unsigned long) -ENODEV;
802 if (!(capabilities & BDI_CAP_MAP_COPY))
805 capabilities &= ~BDI_CAP_MAP_DIRECT;
810 /* we're going to need a VMA struct as well */
811 vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
813 goto error_getting_vma;
815 memset(vma, 0, sizeof(*vma));
816 INIT_LIST_HEAD(&vma->anon_vma_node);
817 atomic_set(&vma->vm_usage, 1);
821 vma->vm_flags = vm_flags;
822 vma->vm_start = addr;
823 vma->vm_end = addr + len;
824 vma->vm_pgoff = pgoff;
828 /* set up the mapping */
829 if (file && vma->vm_flags & VM_SHARED)
830 ret = do_mmap_shared_file(vma, len);
832 ret = do_mmap_private(vma, len);
836 /* okay... we have a mapping; now we have to register it */
837 result = (void *) vma->vm_start;
839 if (vma->vm_flags & VM_MAPPED_COPY) {
840 realalloc += kobjsize(result);
844 realalloc += kobjsize(vma);
845 askedalloc += sizeof(*vma);
847 current->mm->total_vm += len >> PAGE_SHIFT;
852 realalloc += kobjsize(vml);
853 askedalloc += sizeof(*vml);
855 vml->next = current->mm->context.vmlist;
856 current->mm->context.vmlist = vml;
858 up_write(&nommu_vma_sem);
860 if (prot & PROT_EXEC)
861 flush_icache_range((unsigned long) result,
862 (unsigned long) result + len);
865 printk("do_mmap:\n");
866 show_process_blocks();
869 return (unsigned long) result;
872 up_write(&nommu_vma_sem);
881 up_write(&nommu_vma_sem);
882 printk("Attempt to share mismatched mappings\n");
887 up_write(&nommu_vma_sem);
889 printk("Allocation of vma for %lu byte allocation from process %d failed\n",
895 printk("Allocation of vml for %lu byte allocation from process %d failed\n",
902 * handle mapping disposal for uClinux
904 static void put_vma(struct vm_area_struct *vma)
907 down_write(&nommu_vma_sem);
909 if (atomic_dec_and_test(&vma->vm_usage)) {
910 delete_nommu_vma(vma);
912 if (vma->vm_ops && vma->vm_ops->close)
913 vma->vm_ops->close(vma);
915 /* IO memory and memory shared directly out of the pagecache from
916 * ramfs/tmpfs mustn't be released here */
917 if (vma->vm_flags & VM_MAPPED_COPY) {
918 realalloc -= kobjsize((void *) vma->vm_start);
919 askedalloc -= vma->vm_end - vma->vm_start;
920 kfree((void *) vma->vm_start);
923 realalloc -= kobjsize(vma);
924 askedalloc -= sizeof(*vma);
931 up_write(&nommu_vma_sem);
935 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
937 struct vm_list_struct *vml, **parent;
938 unsigned long end = addr + len;
941 printk("do_munmap:\n");
944 for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next)
945 if ((*parent)->vma->vm_start == addr &&
946 ((len == 0) || ((*parent)->vma->vm_end == end)))
949 printk("munmap of non-mmaped memory by process %d (%s): %p\n",
950 current->pid, current->comm, (void *) addr);
959 realalloc -= kobjsize(vml);
960 askedalloc -= sizeof(*vml);
963 update_hiwater_vm(mm);
964 mm->total_vm -= len >> PAGE_SHIFT;
967 show_process_blocks();
973 /* Release all mmaps. */
974 void exit_mmap(struct mm_struct * mm)
976 struct vm_list_struct *tmp;
980 printk("Exit_mmap:\n");
985 while ((tmp = mm->context.vmlist)) {
986 mm->context.vmlist = tmp->next;
989 realalloc -= kobjsize(tmp);
990 askedalloc -= sizeof(*tmp);
995 show_process_blocks();
1000 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1003 struct mm_struct *mm = current->mm;
1005 down_write(&mm->mmap_sem);
1006 ret = do_munmap(mm, addr, len);
1007 up_write(&mm->mmap_sem);
1011 unsigned long do_brk(unsigned long addr, unsigned long len)
1017 * Expand (or shrink) an existing mapping, potentially moving it at the
1018 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1020 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
1021 * This option implies MREMAP_MAYMOVE.
1023 * on uClinux, we only permit changing a mapping's size, and only as long as it stays within the
1024 * hole allocated by the kmalloc() call in do_mmap_pgoff() and the block is not shareable
1026 unsigned long do_mremap(unsigned long addr,
1027 unsigned long old_len, unsigned long new_len,
1028 unsigned long flags, unsigned long new_addr)
1030 struct vm_list_struct *vml = NULL;
1032 /* insanity checks first */
1034 return (unsigned long) -EINVAL;
1036 if (flags & MREMAP_FIXED && new_addr != addr)
1037 return (unsigned long) -EINVAL;
1039 for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1040 if (vml->vma->vm_start == addr)
1043 return (unsigned long) -EINVAL;
1046 if (vml->vma->vm_end != vml->vma->vm_start + old_len)
1047 return (unsigned long) -EFAULT;
1049 if (vml->vma->vm_flags & VM_MAYSHARE)
1050 return (unsigned long) -EPERM;
1052 if (new_len > kobjsize((void *) addr))
1053 return (unsigned long) -ENOMEM;
1055 /* all checks complete - do it */
1056 vml->vma->vm_end = vml->vma->vm_start + new_len;
1058 askedalloc -= old_len;
1059 askedalloc += new_len;
1061 return vml->vma->vm_start;
1065 * Look up the first VMA which satisfies addr < vm_end, NULL if none
1066 * - should be called with mm->mmap_sem at least readlocked
1068 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1070 struct vm_list_struct *vml;
1072 for (vml = mm->context.vmlist; vml; vml = vml->next)
1073 if (addr >= vml->vma->vm_start && addr < vml->vma->vm_end)
1079 EXPORT_SYMBOL(find_vma);
1081 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1082 unsigned int foll_flags)
1087 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
1092 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1093 unsigned long to, unsigned long size, pgprot_t prot)
1095 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1098 EXPORT_SYMBOL(remap_pfn_range);
1100 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1104 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1105 unsigned long len, unsigned long pgoff, unsigned long flags)
1110 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1114 void unmap_mapping_range(struct address_space *mapping,
1115 loff_t const holebegin, loff_t const holelen,
1119 EXPORT_SYMBOL(unmap_mapping_range);
1122 * Check that a process has enough memory to allocate a new virtual
1123 * mapping. 0 means there is enough memory for the allocation to
1124 * succeed and -ENOMEM implies there is not.
1126 * We currently support three overcommit policies, which are set via the
1127 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1129 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1130 * Additional code 2002 Jul 20 by Robert Love.
1132 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1134 * Note this is a helper function intended to be used by LSMs which
1135 * wish to use this logic.
1137 int __vm_enough_memory(long pages, int cap_sys_admin)
1139 unsigned long free, allowed;
1141 vm_acct_memory(pages);
1144 * Sometimes we want to use more memory than we have
1146 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1149 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1152 free = global_page_state(NR_FILE_PAGES);
1153 free += nr_swap_pages;
1156 * Any slabs which are created with the
1157 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1158 * which are reclaimable, under pressure. The dentry
1159 * cache and most inode caches should fall into this
1161 free += global_page_state(NR_SLAB_RECLAIMABLE);
1164 * Leave the last 3% for root
1173 * nr_free_pages() is very expensive on large systems,
1174 * only call if we're about to fail.
1176 n = nr_free_pages();
1179 * Leave reserved pages. The pages are not for anonymous pages.
1181 if (n <= totalreserve_pages)
1184 n -= totalreserve_pages;
1187 * Leave the last 3% for root
1199 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1201 * Leave the last 3% for root
1204 allowed -= allowed / 32;
1205 allowed += total_swap_pages;
1207 /* Don't let a single process grow too big:
1208 leave 3% of the size of this process for other processes */
1209 allowed -= current->mm->total_vm / 32;
1212 * cast `allowed' as a signed long because vm_committed_space
1213 * sometimes has a negative value
1215 if (atomic_read(&vm_committed_space) < (long)allowed)
1218 vm_unacct_memory(pages);
1223 int in_gate_area_no_task(unsigned long addr)
1228 struct page *filemap_nopage(struct vm_area_struct *area,
1229 unsigned long address, int *type)
1236 * Access another process' address space.
1237 * - source/target buffer must be kernel space
1239 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1241 struct vm_area_struct *vma;
1242 struct mm_struct *mm;
1244 if (addr + len < addr)
1247 mm = get_task_mm(tsk);
1251 down_read(&mm->mmap_sem);
1253 /* the access must start within one of the target process's mappings */
1254 vma = find_vma(mm, addr);
1256 /* don't overrun this mapping */
1257 if (addr + len >= vma->vm_end)
1258 len = vma->vm_end - addr;
1260 /* only read or write mappings where it is permitted */
1261 if (write && vma->vm_flags & VM_MAYWRITE)
1262 len -= copy_to_user((void *) addr, buf, len);
1263 else if (!write && vma->vm_flags & VM_MAYREAD)
1264 len -= copy_from_user(buf, (void *) addr, len);
1271 up_read(&mm->mmap_sem);