mm/mremap.c

   1 /*
   2  *      linux/mm/remap.c
   3  *
   4  *      (C) Copyright 1996 Linus Torvalds
   5  */
   6
   7 #include <linux/slab.h>
   8 #include <linux/smp_lock.h>
   9 #include <linux/shm.h>
  10 #include <linux/mman.h>
  11 #include <linux/swap.h>
  12
  13 #include <asm/uaccess.h>
  14 #include <asm/pgtable.h>
  15
  16 extern int vm_enough_memory(long pages);
  17
  18 static inline pte_t *get_one_pte(struct mm_struct *mm, unsigned long addr)
  19 {
  20         pgd_t * pgd;
  21         pmd_t * pmd;
  22         pte_t * pte = NULL;
  23
  24         pgd = pgd_offset(mm, addr);
  25         if (pgd_none(*pgd))
  26                 goto end;
  27         if (pgd_bad(*pgd)) {
  28                 printk("move_one_page: bad source pgd (%08lx)\n", pgd_val(*pgd));
  29                 pgd_clear(pgd);
  30                 goto end;
  31         }
  32
  33         pmd = pmd_offset(pgd, addr);
  34         if (pmd_none(*pmd))
  35                 goto end;
  36         if (pmd_bad(*pmd)) {
  37                 printk("move_one_page: bad source pmd (%08lx)\n", pmd_val(*pmd));
  38                 pmd_clear(pmd);
  39                 goto end;
  40         }
  41
  42         pte = pte_offset(pmd, addr);
  43         if (pte_none(*pte))
  44                 pte = NULL;
  45 end:
  46         return pte;
  47 }
  48
  49 static inline pte_t *alloc_one_pte(struct mm_struct *mm, unsigned long addr)
  50 {
  51         pmd_t * pmd;
  52         pte_t * pte = NULL;
  53
  54         pmd = pmd_alloc(pgd_offset(mm, addr), addr);
  55         if (pmd)
  56                 pte = pte_alloc(pmd, addr);
  57         return pte;
  58 }
  59
  60 static inline int copy_one_pte(pte_t * src, pte_t * dst)
  61 {
  62         int error = 0;
  63         pte_t pte = *src;
  64
  65         if (!pte_none(pte)) {
  66                 error++;
  67                 if (dst) {
  68                         pte_clear(src);
  69                         set_pte(dst, pte);
  70                         error--;
  71                 }
  72         }
  73         return error;
  74 }
  75
  76 static int move_one_page(struct mm_struct *mm, unsigned long old_addr, unsigned long new_addr)
  77 {
  78         int error = 0;
  79         pte_t * src;
  80
  81         src = get_one_pte(mm, old_addr);
  82         if (src)
  83                 error = copy_one_pte(src, alloc_one_pte(mm, new_addr));
  84         return error;
  85 }
  86
  87 static int move_page_tables(struct mm_struct * mm,
  88         unsigned long new_addr, unsigned long old_addr, unsigned long len)
  89 {
  90         unsigned long offset = len;
  91
  92         flush_cache_range(mm, old_addr, old_addr + len);
  93         flush_tlb_range(mm, old_addr, old_addr + len);
  94
  95         /*
  96          * This is not the clever way to do this, but we're taking the
  97          * easy way out on the assumption that most remappings will be
  98          * only a few pages.. This also makes error recovery easier.
  99          */
 100         while (offset) {
 101                 offset -= PAGE_SIZE;
 102                 if (move_one_page(mm, old_addr + offset, new_addr + offset))
 103                         goto oops_we_failed;
 104         }
 105         return 0;
 106
 107         /*
 108          * Ok, the move failed because we didn't have enough pages for
 109          * the new page table tree. This is unlikely, but we have to
 110          * take the possibility into account. In that case we just move
 111          * all the pages back (this will work, because we still have
 112          * the old page tables)
 113          */
 114 oops_we_failed:
 115         flush_cache_range(mm, new_addr, new_addr + len);
 116         while ((offset += PAGE_SIZE) < len)
 117                 move_one_page(mm, new_addr + offset, old_addr + offset);
 118         zap_page_range(mm, new_addr, new_addr + len);
 119         flush_tlb_range(mm, new_addr, new_addr + len);
 120         return -1;
 121 }
 122
 123 static inline unsigned long move_vma(struct vm_area_struct * vma,
 124         unsigned long addr, unsigned long old_len, unsigned long new_len)
 125 {
 126         struct vm_area_struct * new_vma;
 127
 128         new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 129         if (new_vma) {
 130                 unsigned long new_addr = get_unmapped_area(addr, new_len);
 131
 132                 if (new_addr && !move_page_tables(current->mm, new_addr, addr, old_len)) {
 133                         *new_vma = *vma;
 134                         new_vma->vm_start = new_addr;
 135                         new_vma->vm_end = new_addr+new_len;
 136                         new_vma->vm_offset = vma->vm_offset + (addr - vma->vm_start);
 137                         if (new_vma->vm_file)
 138                                 new_vma->vm_file->f_count++;
 139                         if (new_vma->vm_ops && new_vma->vm_ops->open)
 140                                 new_vma->vm_ops->open(new_vma);
 141                         insert_vm_struct(current->mm, new_vma);
 142                         merge_segments(current->mm, new_vma->vm_start, new_vma->vm_end);
 143                         do_munmap(addr, old_len);
 144                         current->mm->total_vm += new_len >> PAGE_SHIFT;
 145                         if (new_vma->vm_flags & VM_LOCKED) {
 146                                 current->mm->locked_vm += new_len >> PAGE_SHIFT;
 147                                 make_pages_present(new_vma->vm_start,
 148                                                    new_vma->vm_end);
 149                         }
 150                         return new_addr;
 151                 }
 152                 kmem_cache_free(vm_area_cachep, new_vma);
 153         }
 154         return -ENOMEM;
 155 }
 156
 157 /*
 158  * Expand (or shrink) an existing mapping, potentially moving it at the
 159  * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
 160  */
 161 asmlinkage unsigned long sys_mremap(unsigned long addr,
 162         unsigned long old_len, unsigned long new_len,
 163         unsigned long flags)
 164 {
 165         struct vm_area_struct *vma;
 166         unsigned long ret = -EINVAL;
 167
 168         down(&current->mm->mmap_sem);
 169         lock_kernel();
 170         if (addr & ~PAGE_MASK)
 171                 goto out;
 172         old_len = PAGE_ALIGN(old_len);
 173         new_len = PAGE_ALIGN(new_len);
 174
 175         /*
 176          * Always allow a shrinking remap: that just unmaps
 177          * the unnecessary pages..
 178          */
 179         ret = addr;
 180         if (old_len >= new_len) {
 181                 do_munmap(addr+new_len, old_len - new_len);
 182                 goto out;
 183         }
 184
 185         /*
 186          * Ok, we need to grow..
 187          */
 188         ret = -EFAULT;
 189         vma = find_vma(current->mm, addr);
 190         if (!vma || vma->vm_start > addr)
 191                 goto out;
 192         /* We can't remap across vm area boundaries */
 193         if (old_len > vma->vm_end - addr)
 194                 goto out;
 195         if (vma->vm_flags & VM_LOCKED) {
 196                 unsigned long locked = current->mm->locked_vm << PAGE_SHIFT;
 197                 locked += new_len - old_len;
 198                 ret = -EAGAIN;
 199                 if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
 200                         goto out;
 201         }
 202         ret = -ENOMEM;
 203         if ((current->mm->total_vm << PAGE_SHIFT) + (new_len - old_len)
 204             > current->rlim[RLIMIT_AS].rlim_cur)
 205                 goto out;
 206         /* Private writable mapping? Check memory availability.. */
 207         if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) == VM_WRITE &&
 208             !(flags & MAP_NORESERVE)                             &&
 209             !vm_enough_memory((new_len - old_len) >> PAGE_SHIFT))
 210                 goto out;
 211
 212         /* old_len exactly to the end of the area.. */
 213         if (old_len == vma->vm_end - addr &&
 214             (old_len != new_len || !(flags & MREMAP_MAYMOVE))) {
 215                 unsigned long max_addr = TASK_SIZE;
 216                 if (vma->vm_next)
 217                         max_addr = vma->vm_next->vm_start;
 218                 /* can we just expand the current mapping? */
 219                 if (max_addr - addr >= new_len) {
 220                         int pages = (new_len - old_len) >> PAGE_SHIFT;
 221                         vma->vm_end = addr + new_len;
 222                         current->mm->total_vm += pages;
 223                         if (vma->vm_flags & VM_LOCKED) {
 224                                 current->mm->locked_vm += pages;
 225                                 make_pages_present(addr + old_len,
 226                                                    addr + new_len);
 227                         }
 228                         ret = addr;
 229                         goto out;
 230                 }
 231         }
 232
 233         /*
 234          * We weren't able to just expand or shrink the area,
 235          * we need to create a new one and move it..
 236          */
 237         if (flags & MREMAP_MAYMOVE)
 238                 ret = move_vma(vma, addr, old_len, new_len);
 239         else
 240                 ret = -ENOMEM;
 241 out:
 242         unlock_kernel();
 243         up(&current->mm->mmap_sem);
 244         return ret;
 245 }