LXR linux-old/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/pgalloc.h>
  15
  16extern int vm_enough_memory(long pages);
  17
  18static 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                pgd_ERROR(*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                pmd_ERROR(*pmd);
  38                pmd_clear(pmd);
  39                goto end;
  40        }
  41
  42        pte = pte_offset(pmd, addr);
  43        if (pte_none(*pte))
  44                pte = NULL;
  45end:
  46        return pte;
  47}
  48
  49static 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(mm, pgd_offset(mm, addr), addr);
  55        if (pmd)
  56                pte = pte_alloc(mm, pmd, addr);
  57        return pte;
  58}
  59
  60static inline int copy_one_pte(struct mm_struct *mm, pte_t * src, pte_t * dst)
  61{
  62        int error = 0;
  63        pte_t pte;
  64
  65        if (!pte_none(*src)) {
  66                pte = ptep_get_and_clear(src);
  67                if (!dst) {
  68                        /* No dest?  We must put it back. */
  69                        dst = src;
  70                        error++;
  71                }
  72                set_pte(dst, pte);
  73        }
  74        return error;
  75}
  76
  77static int move_one_page(struct mm_struct *mm, unsigned long old_addr, unsigned long new_addr)
  78{
  79        int error = 0;
  80        pte_t * src, * dst;
  81
  82        spin_lock(&mm->page_table_lock);
  83        src = get_one_pte(mm, old_addr);
  84        if (src) {
  85                dst = alloc_one_pte(mm, new_addr);
  86                src = get_one_pte(mm, old_addr);
  87                if (src) 
  88                        error = copy_one_pte(mm, src, dst);
  89        }
  90        spin_unlock(&mm->page_table_lock);
  91        return error;
  92}
  93
  94static int move_page_tables(struct mm_struct * mm,
  95        unsigned long new_addr, unsigned long old_addr, unsigned long len)
  96{
  97        unsigned long offset = len;
  98
  99        flush_cache_range(mm, old_addr, old_addr + len);
 100
 101        /*
 102         * This is not the clever way to do this, but we're taking the
 103         * easy way out on the assumption that most remappings will be
 104         * only a few pages.. This also makes error recovery easier.
 105         */
 106        while (offset) {
 107                offset -= PAGE_SIZE;
 108                if (move_one_page(mm, old_addr + offset, new_addr + offset))
 109                        goto oops_we_failed;
 110        }
 111        flush_tlb_range(mm, old_addr, old_addr + len);
 112        return 0;
 113
 114        /*
 115         * Ok, the move failed because we didn't have enough pages for
 116         * the new page table tree. This is unlikely, but we have to
 117         * take the possibility into account. In that case we just move
 118         * all the pages back (this will work, because we still have
 119         * the old page tables)
 120         */
 121oops_we_failed:
 122        flush_cache_range(mm, new_addr, new_addr + len);
 123        while ((offset += PAGE_SIZE) < len)
 124                move_one_page(mm, new_addr + offset, old_addr + offset);
 125        zap_page_range(mm, new_addr, len);
 126        return -1;
 127}
 128
 129static inline unsigned long move_vma(struct vm_area_struct * vma,
 130        unsigned long addr, unsigned long old_len, unsigned long new_len,
 131        unsigned long new_addr)
 132{
 133        struct mm_struct * mm = vma->vm_mm;
 134        struct vm_area_struct * new_vma, * next, * prev;
 135        int allocated_vma;
 136
 137        new_vma = NULL;
 138        next = find_vma_prev(mm, new_addr, &prev);
 139        if (next) {
 140                if (prev && prev->vm_end == new_addr &&
 141                    can_vma_merge(prev, vma->vm_flags) && !vma->vm_file && !(vma->vm_flags & VM_SHARED)) {
 142                        spin_lock(&mm->page_table_lock);
 143                        prev->vm_end = new_addr + new_len;
 144                        spin_unlock(&mm->page_table_lock);
 145                        new_vma = prev;
 146                        if (next != prev->vm_next)
 147                                BUG();
 148                        if (prev->vm_end == next->vm_start && can_vma_merge(next, prev->vm_flags)) {
 149                                spin_lock(&mm->page_table_lock);
 150                                prev->vm_end = next->vm_end;
 151                                __vma_unlink(mm, next, prev);
 152                                spin_unlock(&mm->page_table_lock);
 153
 154                                mm->map_count--;
 155                                kmem_cache_free(vm_area_cachep, next);
 156                        }
 157                } else if (next->vm_start == new_addr + new_len &&
 158                           can_vma_merge(next, vma->vm_flags) && !vma->vm_file && !(vma->vm_flags & VM_SHARED)) {
 159                        spin_lock(&mm->page_table_lock);
 160                        next->vm_start = new_addr;
 161                        spin_unlock(&mm->page_table_lock);
 162                        new_vma = next;
 163                }
 164        } else {
 165                prev = find_vma(mm, new_addr-1);
 166                if (prev && prev->vm_end == new_addr &&
 167                    can_vma_merge(prev, vma->vm_flags) && !vma->vm_file && !(vma->vm_flags & VM_SHARED)) {
 168                        spin_lock(&mm->page_table_lock);
 169                        prev->vm_end = new_addr + new_len;
 170                        spin_unlock(&mm->page_table_lock);
 171                        new_vma = prev;
 172                }
 173        }
 174
 175        allocated_vma = 0;
 176        if (!new_vma) {
 177                new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 178                if (!new_vma)
 179                        goto out;
 180                allocated_vma = 1;
 181        }
 182
 183        if (!move_page_tables(current->mm, new_addr, addr, old_len)) {
 184                unsigned long vm_locked = vma->vm_flags & VM_LOCKED;
 185
 186                if (allocated_vma) {
 187                        *new_vma = *vma;
 188                        new_vma->vm_start = new_addr;
 189                        new_vma->vm_end = new_addr+new_len;
 190                        new_vma->vm_pgoff += (addr-vma->vm_start) >> PAGE_SHIFT;
 191                        new_vma->vm_raend = 0;
 192                        if (new_vma->vm_file)
 193                                get_file(new_vma->vm_file);
 194                        if (new_vma->vm_ops && new_vma->vm_ops->open)
 195                                new_vma->vm_ops->open(new_vma);
 196                        insert_vm_struct(current->mm, new_vma);
 197                }
 198
 199                do_munmap(current->mm, addr, old_len);
 200
 201                current->mm->total_vm += new_len >> PAGE_SHIFT;
 202                if (vm_locked) {
 203                        current->mm->locked_vm += new_len >> PAGE_SHIFT;
 204                        if (new_len > old_len)
 205                                make_pages_present(new_addr + old_len,
 206                                                   new_addr + new_len);
 207                }
 208                return new_addr;
 209        }
 210        if (allocated_vma)
 211                kmem_cache_free(vm_area_cachep, new_vma);
 212 out:
 213        return -ENOMEM;
 214}
 215
 216/*
 217 * Expand (or shrink) an existing mapping, potentially moving it at the
 218 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
 219 *
 220 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
 221 * This option implies MREMAP_MAYMOVE.
 222 */
 223unsigned long do_mremap(unsigned long addr,
 224        unsigned long old_len, unsigned long new_len,
 225        unsigned long flags, unsigned long new_addr)
 226{
 227        struct vm_area_struct *vma;
 228        unsigned long ret = -EINVAL;
 229
 230        if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
 231                goto out;
 232
 233        if (addr & ~PAGE_MASK)
 234                goto out;
 235
 236        old_len = PAGE_ALIGN(old_len);
 237        new_len = PAGE_ALIGN(new_len);
 238
 239        /* new_addr is only valid if MREMAP_FIXED is specified */
 240        if (flags & MREMAP_FIXED) {
 241                if (new_addr & ~PAGE_MASK)
 242                        goto out;
 243                if (!(flags & MREMAP_MAYMOVE))
 244                        goto out;
 245
 246                if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
 247                        goto out;
 248                /*
 249                 * Allow new_len == 0 only if new_addr == addr
 250                 * to preserve truncation in place (that was working
 251                 * safe and some app may depend on it).
 252                 */
 253                if (unlikely(!new_len && new_addr != addr))
 254                        goto out;
 255
 256                /* Check if the location we're moving into overlaps the
 257                 * old location at all, and fail if it does.
 258                 */
 259                if ((new_addr <= addr) && (new_addr+new_len) > addr)
 260                        goto out;
 261
 262                if ((addr <= new_addr) && (addr+old_len) > new_addr)
 263                        goto out;
 264
 265                ret = do_munmap(current->mm, new_addr, new_len);
 266                if (ret && new_len)
 267                        goto out;
 268        }
 269
 270        /*
 271         * Always allow a shrinking remap: that just unmaps
 272         * the unnecessary pages..
 273         */
 274        if (old_len >= new_len) {
 275                ret = do_munmap(current->mm, addr+new_len, old_len - new_len);
 276                if (ret && old_len != new_len)
 277                        goto out;
 278                ret = addr;
 279                if (!(flags & MREMAP_FIXED) || (new_addr == addr))
 280                        goto out;
 281        }
 282
 283        /*
 284         * Ok, we need to grow..  or relocate.
 285         */
 286        ret = -EFAULT;
 287        vma = find_vma(current->mm, addr);
 288        if (!vma || vma->vm_start > addr)
 289                goto out;
 290        /* We can't remap across vm area boundaries */
 291        if (old_len > vma->vm_end - addr)
 292                goto out;
 293        if (vma->vm_flags & VM_DONTEXPAND) {
 294                if (new_len > old_len)
 295                        goto out;
 296        }
 297        if (vma->vm_flags & VM_LOCKED) {
 298                unsigned long locked = current->mm->locked_vm << PAGE_SHIFT;
 299                locked += new_len - old_len;
 300                ret = -EAGAIN;
 301                if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
 302                        goto out;
 303        }
 304        ret = -ENOMEM;
 305        if ((current->mm->total_vm << PAGE_SHIFT) + (new_len - old_len)
 306            > current->rlim[RLIMIT_AS].rlim_cur)
 307                goto out;
 308        /* Private writable mapping? Check memory availability.. */
 309        if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) == VM_WRITE &&
 310            !(flags & MAP_NORESERVE)                             &&
 311            !vm_enough_memory((new_len - old_len) >> PAGE_SHIFT))
 312                goto out;
 313
 314        /* old_len exactly to the end of the area..
 315         * And we're not relocating the area.
 316         */
 317        if (old_len == vma->vm_end - addr &&
 318            !((flags & MREMAP_FIXED) && (addr != new_addr)) &&
 319            (old_len != new_len || !(flags & MREMAP_MAYMOVE))) {
 320                unsigned long max_addr = TASK_SIZE;
 321                if (vma->vm_next)
 322                        max_addr = vma->vm_next->vm_start;
 323                /* can we just expand the current mapping? */
 324                if (max_addr - addr >= new_len) {
 325                        int pages = (new_len - old_len) >> PAGE_SHIFT;
 326                        spin_lock(&vma->vm_mm->page_table_lock);
 327                        vma->vm_end = addr + new_len;
 328                        spin_unlock(&vma->vm_mm->page_table_lock);
 329                        current->mm->total_vm += pages;
 330                        if (vma->vm_flags & VM_LOCKED) {
 331                                current->mm->locked_vm += pages;
 332                                make_pages_present(addr + old_len,
 333                                                   addr + new_len);
 334                        }
 335                        ret = addr;
 336                        goto out;
 337                }
 338        }
 339
 340        /*
 341         * We weren't able to just expand or shrink the area,
 342         * we need to create a new one and move it..
 343         */
 344        ret = -ENOMEM;
 345        if (flags & MREMAP_MAYMOVE) {
 346                if (!(flags & MREMAP_FIXED)) {
 347                        unsigned long map_flags = 0;
 348                        if (vma->vm_flags & VM_SHARED)
 349                                map_flags |= MAP_SHARED;
 350
 351                        new_addr = get_unmapped_area(vma->vm_file, 0, new_len, vma->vm_pgoff, map_flags);
 352                        ret = new_addr;
 353                        if (new_addr & ~PAGE_MASK)
 354                                goto out;
 355                }
 356                ret = move_vma(vma, addr, old_len, new_len, new_addr);
 357        }
 358out:
 359        return ret;
 360}
 361
 362asmlinkage unsigned long sys_mremap(unsigned long addr,
 363        unsigned long old_len, unsigned long new_len,
 364        unsigned long flags, unsigned long new_addr)
 365{
 366        unsigned long ret;
 367
 368        down_write(&current->mm->mmap_sem);
 369        ret = do_mremap(addr, old_len, new_len, flags, new_addr);
 370        up_write(&current->mm->mmap_sem);
 371        return ret;
 372}
 373