mm/madvise.c

   1 /*
   2  *      linux/mm/madvise.c
   3  *
   4  * Copyright (C) 1999  Linus Torvalds
   5  * Copyright (C) 2002  Christoph Hellwig
   6  */
   7
   8 #include <linux/mman.h>
   9 #include <linux/pagemap.h>
  10
  11
  12 /*
  13  * We can potentially split a vm area into separate
  14  * areas, each area with its own behavior.
  15  */
  16 static long madvise_behavior(struct vm_area_struct * vma, unsigned long start,
  17                              unsigned long end, int behavior)
  18 {
  19         struct mm_struct * mm = vma->vm_mm;
  20         int error;
  21
  22         if (start != vma->vm_start) {
  23                 error = split_vma(mm, vma, start, 1);
  24                 if (error)
  25                         return -EAGAIN;
  26         }
  27
  28         if (end != vma->vm_end) {
  29                 error = split_vma(mm, vma, end, 0);
  30                 if (error)
  31                         return -EAGAIN;
  32         }
  33
  34         spin_lock(&mm->page_table_lock);
  35         VM_ClearReadHint(vma);
  36
  37         switch (behavior) {
  38         case MADV_SEQUENTIAL:
  39                 vma->vm_flags |= VM_SEQ_READ;
  40                 break;
  41         case MADV_RANDOM:
  42                 vma->vm_flags |= VM_RAND_READ;
  43                 break;
  44         default:
  45                 break;
  46         }
  47         spin_unlock(&mm->page_table_lock);
  48
  49         return 0;
  50 }
  51
  52 /*
  53  * Schedule all required I/O operations.  Do not wait for completion.
  54  */
  55 static long madvise_willneed(struct vm_area_struct * vma,
  56                              unsigned long start, unsigned long end)
  57 {
  58         struct file *file = vma->vm_file;
  59
  60         if (!file)
  61                 return -EBADF;
  62
  63         start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
  64         if (end > vma->vm_end)
  65                 end = vma->vm_end;
  66         end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
  67
  68         do_page_cache_readahead(file->f_dentry->d_inode->i_mapping,
  69                         file, start, max_sane_readahead(end - start));
  70         return 0;
  71 }
  72
  73 /*
  74  * Application no longer needs these pages.  If the pages are dirty,
  75  * it's OK to just throw them away.  The app will be more careful about
  76  * data it wants to keep.  Be sure to free swap resources too.  The
  77  * zap_page_range call sets things up for refill_inactive to actually free
  78  * these pages later if no one else has touched them in the meantime,
  79  * although we could add these pages to a global reuse list for
  80  * refill_inactive to pick up before reclaiming other pages.
  81  *
  82  * NB: This interface discards data rather than pushes it out to swap,
  83  * as some implementations do.  This has performance implications for
  84  * applications like large transactional databases which want to discard
  85  * pages in anonymous maps after committing to backing store the data
  86  * that was kept in them.  There is no reason to write this data out to
  87  * the swap area if the application is discarding it.
  88  *
  89  * An interface that causes the system to free clean pages and flush
  90  * dirty pages is already available as msync(MS_INVALIDATE).
  91  */
  92 static long madvise_dontneed(struct vm_area_struct * vma,
  93                              unsigned long start, unsigned long end)
  94 {
  95         if (vma->vm_flags & VM_LOCKED)
  96                 return -EINVAL;
  97
  98         zap_page_range(vma, start, end - start);
  99         return 0;
 100 }
 101
 102 static long madvise_vma(struct vm_area_struct * vma, unsigned long start,
 103                         unsigned long end, int behavior)
 104 {
 105         long error = -EBADF;
 106
 107         switch (behavior) {
 108         case MADV_NORMAL:
 109         case MADV_SEQUENTIAL:
 110         case MADV_RANDOM:
 111                 error = madvise_behavior(vma, start, end, behavior);
 112                 break;
 113
 114         case MADV_WILLNEED:
 115                 error = madvise_willneed(vma, start, end);
 116                 break;
 117
 118         case MADV_DONTNEED:
 119                 error = madvise_dontneed(vma, start, end);
 120                 break;
 121
 122         default:
 123                 error = -EINVAL;
 124                 break;
 125         }
 126
 127         return error;
 128 }
 129
 130 /*
 131  * The madvise(2) system call.
 132  *
 133  * Applications can use madvise() to advise the kernel how it should
 134  * handle paging I/O in this VM area.  The idea is to help the kernel
 135  * use appropriate read-ahead and caching techniques.  The information
 136  * provided is advisory only, and can be safely disregarded by the
 137  * kernel without affecting the correct operation of the application.
 138  *
 139  * behavior values:
 140  *  MADV_NORMAL - the default behavior is to read clusters.  This
 141  *              results in some read-ahead and read-behind.
 142  *  MADV_RANDOM - the system should read the minimum amount of data
 143  *              on any access, since it is unlikely that the appli-
 144  *              cation will need more than what it asks for.
 145  *  MADV_SEQUENTIAL - pages in the given range will probably be accessed
 146  *              once, so they can be aggressively read ahead, and
 147  *              can be freed soon after they are accessed.
 148  *  MADV_WILLNEED - the application is notifying the system to read
 149  *              some pages ahead.
 150  *  MADV_DONTNEED - the application is finished with the given range,
 151  *              so the kernel can free resources associated with it.
 152  *
 153  * return values:
 154  *  zero    - success
 155  *  -EINVAL - start + len < 0, start is not page-aligned,
 156  *              "behavior" is not a valid value, or application
 157  *              is attempting to release locked or shared pages.
 158  *  -ENOMEM - addresses in the specified range are not currently
 159  *              mapped, or are outside the AS of the process.
 160  *  -EIO    - an I/O error occurred while paging in data.
 161  *  -EBADF  - map exists, but area maps something that isn't a file.
 162  *  -EAGAIN - a kernel resource was temporarily unavailable.
 163  */
 164 asmlinkage long sys_madvise(unsigned long start, size_t len, int behavior)
 165 {
 166         unsigned long end;
 167         struct vm_area_struct * vma;
 168         int unmapped_error = 0;
 169         int error = -EINVAL;
 170
 171         down_write(&current->mm->mmap_sem);
 172
 173         if (start & ~PAGE_MASK)
 174                 goto out;
 175         len = (len + ~PAGE_MASK) & PAGE_MASK;
 176         end = start + len;
 177         if (end < start)
 178                 goto out;
 179
 180         error = 0;
 181         if (end == start)
 182                 goto out;
 183
 184         /*
 185          * If the interval [start,end) covers some unmapped address
 186          * ranges, just ignore them, but return -ENOMEM at the end.
 187          */
 188         vma = find_vma(current->mm, start);
 189         for (;;) {
 190                 /* Still start < end. */
 191                 error = -ENOMEM;
 192                 if (!vma)
 193                         goto out;
 194
 195                 /* Here start < vma->vm_end. */
 196                 if (start < vma->vm_start) {
 197                         unmapped_error = -ENOMEM;
 198                         start = vma->vm_start;
 199                 }
 200
 201                 /* Here vma->vm_start <= start < vma->vm_end. */
 202                 if (end <= vma->vm_end) {
 203                         if (start < end) {
 204                                 error = madvise_vma(vma, start, end,
 205                                                         behavior);
 206                                 if (error)
 207                                         goto out;
 208                         }
 209                         error = unmapped_error;
 210                         goto out;
 211                 }
 212
 213                 /* Here vma->vm_start <= start < vma->vm_end < end. */
 214                 error = madvise_vma(vma, start, vma->vm_end, behavior);
 215                 if (error)
 216                         goto out;
 217                 start = vma->vm_end;
 218                 vma = vma->vm_next;
 219         }
 220
 221 out:
 222         up_write(&current->mm->mmap_sem);
 223         return error;
 224 }