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         /*
  35          * vm_flags is protected by the mmap_sem held in write mode.
  36          */
  37         VM_ClearReadHint(vma);
  38
  39         switch (behavior) {
  40         case MADV_SEQUENTIAL:
  41                 vma->vm_flags |= VM_SEQ_READ;
  42                 break;
  43         case MADV_RANDOM:
  44                 vma->vm_flags |= VM_RAND_READ;
  45                 break;
  46         default:
  47                 break;
  48         }
  49
  50         return 0;
  51 }
  52
  53 /*
  54  * Schedule all required I/O operations.  Do not wait for completion.
  55  */
  56 static long madvise_willneed(struct vm_area_struct * vma,
  57                              unsigned long start, unsigned long end)
  58 {
  59         struct file *file = vma->vm_file;
  60
  61         if (!file)
  62                 return -EBADF;
  63
  64         start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
  65         if (end > vma->vm_end)
  66                 end = vma->vm_end;
  67         end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
  68
  69         force_page_cache_readahead(file->f_mapping,
  70                         file, start, max_sane_readahead(end - start));
  71         return 0;
  72 }
  73
  74 /*
  75  * Application no longer needs these pages.  If the pages are dirty,
  76  * it's OK to just throw them away.  The app will be more careful about
  77  * data it wants to keep.  Be sure to free swap resources too.  The
  78  * zap_page_range call sets things up for refill_inactive to actually free
  79  * these pages later if no one else has touched them in the meantime,
  80  * although we could add these pages to a global reuse list for
  81  * refill_inactive to pick up before reclaiming other pages.
  82  *
  83  * NB: This interface discards data rather than pushes it out to swap,
  84  * as some implementations do.  This has performance implications for
  85  * applications like large transactional databases which want to discard
  86  * pages in anonymous maps after committing to backing store the data
  87  * that was kept in them.  There is no reason to write this data out to
  88  * the swap area if the application is discarding it.
  89  *
  90  * An interface that causes the system to free clean pages and flush
  91  * dirty pages is already available as msync(MS_INVALIDATE).
  92  */
  93 static long madvise_dontneed(struct vm_area_struct * vma,
  94                              unsigned long start, unsigned long end)
  95 {
  96         if (vma->vm_flags & VM_LOCKED)
  97                 return -EINVAL;
  98
  99         if (unlikely(vma->vm_flags & VM_NONLINEAR)) {
 100                 struct zap_details details = {
 101                         .nonlinear_vma = vma,
 102                         .last_index = ULONG_MAX,
 103                 };
 104                 zap_page_range(vma, start, end - start, &details);
 105         } else
 106                 zap_page_range(vma, start, end - start, NULL);
 107         return 0;
 108 }
 109
 110 static long madvise_vma(struct vm_area_struct * vma, unsigned long start,
 111                         unsigned long end, int behavior)
 112 {
 113         long error = -EBADF;
 114
 115         switch (behavior) {
 116         case MADV_NORMAL:
 117         case MADV_SEQUENTIAL:
 118         case MADV_RANDOM:
 119                 error = madvise_behavior(vma, start, end, behavior);
 120                 break;
 121
 122         case MADV_WILLNEED:
 123                 error = madvise_willneed(vma, start, end);
 124                 break;
 125
 126         case MADV_DONTNEED:
 127                 error = madvise_dontneed(vma, start, end);
 128                 break;
 129
 130         default:
 131                 error = -EINVAL;
 132                 break;
 133         }
 134
 135         return error;
 136 }
 137
 138 /*
 139  * The madvise(2) system call.
 140  *
 141  * Applications can use madvise() to advise the kernel how it should
 142  * handle paging I/O in this VM area.  The idea is to help the kernel
 143  * use appropriate read-ahead and caching techniques.  The information
 144  * provided is advisory only, and can be safely disregarded by the
 145  * kernel without affecting the correct operation of the application.
 146  *
 147  * behavior values:
 148  *  MADV_NORMAL - the default behavior is to read clusters.  This
 149  *              results in some read-ahead and read-behind.
 150  *  MADV_RANDOM - the system should read the minimum amount of data
 151  *              on any access, since it is unlikely that the appli-
 152  *              cation will need more than what it asks for.
 153  *  MADV_SEQUENTIAL - pages in the given range will probably be accessed
 154  *              once, so they can be aggressively read ahead, and
 155  *              can be freed soon after they are accessed.
 156  *  MADV_WILLNEED - the application is notifying the system to read
 157  *              some pages ahead.
 158  *  MADV_DONTNEED - the application is finished with the given range,
 159  *              so the kernel can free resources associated with it.
 160  *
 161  * return values:
 162  *  zero    - success
 163  *  -EINVAL - start + len < 0, start is not page-aligned,
 164  *              "behavior" is not a valid value, or application
 165  *              is attempting to release locked or shared pages.
 166  *  -ENOMEM - addresses in the specified range are not currently
 167  *              mapped, or are outside the AS of the process.
 168  *  -EIO    - an I/O error occurred while paging in data.
 169  *  -EBADF  - map exists, but area maps something that isn't a file.
 170  *  -EAGAIN - a kernel resource was temporarily unavailable.
 171  */
 172 asmlinkage long sys_madvise(unsigned long start, size_t len_in, int behavior)
 173 {
 174         unsigned long end;
 175         struct vm_area_struct * vma;
 176         int unmapped_error = 0;
 177         int error = -EINVAL;
 178         size_t len;
 179
 180         down_write(&current->mm->mmap_sem);
 181
 182         if (start & ~PAGE_MASK)
 183                 goto out;
 184         len = (len_in + ~PAGE_MASK) & PAGE_MASK;
 185
 186         /* Check to see whether len was rounded up from small -ve to zero */
 187         if (len_in && !len)
 188                 goto out;
 189
 190         end = start + len;
 191         if (end < start)
 192                 goto out;
 193
 194         error = 0;
 195         if (end == start)
 196                 goto out;
 197
 198         /*
 199          * If the interval [start,end) covers some unmapped address
 200          * ranges, just ignore them, but return -ENOMEM at the end.
 201          */
 202         vma = find_vma(current->mm, start);
 203         for (;;) {
 204                 /* Still start < end. */
 205                 error = -ENOMEM;
 206                 if (!vma)
 207                         goto out;
 208
 209                 /* Here start < vma->vm_end. */
 210                 if (start < vma->vm_start) {
 211                         unmapped_error = -ENOMEM;
 212                         start = vma->vm_start;
 213                 }
 214
 215                 /* Here vma->vm_start <= start < vma->vm_end. */
 216                 if (end <= vma->vm_end) {
 217                         if (start < end) {
 218                                 error = madvise_vma(vma, start, end,
 219                                                         behavior);
 220                                 if (error)
 221                                         goto out;
 222                         }
 223                         error = unmapped_error;
 224                         goto out;
 225                 }
 226
 227                 /* Here vma->vm_start <= start < vma->vm_end < end. */
 228                 error = madvise_vma(vma, start, vma->vm_end, behavior);
 229                 if (error)
 230                         goto out;
 231                 start = vma->vm_end;
 232                 vma = vma->vm_next;
 233         }
 234
 235 out:
 236         up_write(&current->mm->mmap_sem);
 237         return error;
 238 }