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 #include <linux/syscalls.h>
  11 #include <linux/hugetlb.h>
  12
  13 /*
  14  * We can potentially split a vm area into separate
  15  * areas, each area with its own behavior.
  16  */
  17 static long madvise_behavior(struct vm_area_struct * vma, unsigned long start,
  18                              unsigned long end, int behavior)
  19 {
  20         struct mm_struct * mm = vma->vm_mm;
  21         int error = 0;
  22         int new_flags = vma->vm_flags & ~VM_READHINTMASK;
  23
  24         switch (behavior) {
  25         case MADV_SEQUENTIAL:
  26                 new_flags |= VM_SEQ_READ;
  27                 break;
  28         case MADV_RANDOM:
  29                 new_flags |= VM_RAND_READ;
  30                 break;
  31         default:
  32                 break;
  33         }
  34
  35         if (new_flags == vma->vm_flags)
  36                 goto out;
  37
  38         if (start != vma->vm_start) {
  39                 error = split_vma(mm, vma, start, 1);
  40                 if (error)
  41                         goto out;
  42         }
  43
  44         if (end != vma->vm_end) {
  45                 error = split_vma(mm, vma, end, 0);
  46                 if (error)
  47                         goto out;
  48         }
  49
  50         /*
  51          * vm_flags is protected by the mmap_sem held in write mode.
  52          */
  53         VM_ClearReadHint(vma);
  54         vma->vm_flags = new_flags;
  55
  56 out:
  57         if (error == -ENOMEM)
  58                 error = -EAGAIN;
  59         return error;
  60 }
  61
  62 /*
  63  * Schedule all required I/O operations.  Do not wait for completion.
  64  */
  65 static long madvise_willneed(struct vm_area_struct * vma,
  66                              unsigned long start, unsigned long end)
  67 {
  68         struct file *file = vma->vm_file;
  69
  70         if (!file)
  71                 return -EBADF;
  72
  73         start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
  74         if (end > vma->vm_end)
  75                 end = vma->vm_end;
  76         end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
  77
  78         force_page_cache_readahead(file->f_mapping,
  79                         file, start, max_sane_readahead(end - start));
  80         return 0;
  81 }
  82
  83 /*
  84  * Application no longer needs these pages.  If the pages are dirty,
  85  * it's OK to just throw them away.  The app will be more careful about
  86  * data it wants to keep.  Be sure to free swap resources too.  The
  87  * zap_page_range call sets things up for refill_inactive to actually free
  88  * these pages later if no one else has touched them in the meantime,
  89  * although we could add these pages to a global reuse list for
  90  * refill_inactive to pick up before reclaiming other pages.
  91  *
  92  * NB: This interface discards data rather than pushes it out to swap,
  93  * as some implementations do.  This has performance implications for
  94  * applications like large transactional databases which want to discard
  95  * pages in anonymous maps after committing to backing store the data
  96  * that was kept in them.  There is no reason to write this data out to
  97  * the swap area if the application is discarding it.
  98  *
  99  * An interface that causes the system to free clean pages and flush
 100  * dirty pages is already available as msync(MS_INVALIDATE).
 101  */
 102 static long madvise_dontneed(struct vm_area_struct * vma,
 103                              unsigned long start, unsigned long end)
 104 {
 105         if ((vma->vm_flags & VM_LOCKED) || is_vm_hugetlb_page(vma))
 106                 return -EINVAL;
 107
 108         if (unlikely(vma->vm_flags & VM_NONLINEAR)) {
 109                 struct zap_details details = {
 110                         .nonlinear_vma = vma,
 111                         .last_index = ULONG_MAX,
 112                 };
 113                 zap_page_range(vma, start, end - start, &details);
 114         } else
 115                 zap_page_range(vma, start, end - start, NULL);
 116         return 0;
 117 }
 118
 119 static long madvise_vma(struct vm_area_struct * vma, unsigned long start,
 120                         unsigned long end, int behavior)
 121 {
 122         long error = -EBADF;
 123
 124         switch (behavior) {
 125         case MADV_NORMAL:
 126         case MADV_SEQUENTIAL:
 127         case MADV_RANDOM:
 128                 error = madvise_behavior(vma, start, end, behavior);
 129                 break;
 130
 131         case MADV_WILLNEED:
 132                 error = madvise_willneed(vma, start, end);
 133                 break;
 134
 135         case MADV_DONTNEED:
 136                 error = madvise_dontneed(vma, start, end);
 137                 break;
 138
 139         default:
 140                 error = -EINVAL;
 141                 break;
 142         }
 143
 144         return error;
 145 }
 146
 147 /*
 148  * The madvise(2) system call.
 149  *
 150  * Applications can use madvise() to advise the kernel how it should
 151  * handle paging I/O in this VM area.  The idea is to help the kernel
 152  * use appropriate read-ahead and caching techniques.  The information
 153  * provided is advisory only, and can be safely disregarded by the
 154  * kernel without affecting the correct operation of the application.
 155  *
 156  * behavior values:
 157  *  MADV_NORMAL - the default behavior is to read clusters.  This
 158  *              results in some read-ahead and read-behind.
 159  *  MADV_RANDOM - the system should read the minimum amount of data
 160  *              on any access, since it is unlikely that the appli-
 161  *              cation will need more than what it asks for.
 162  *  MADV_SEQUENTIAL - pages in the given range will probably be accessed
 163  *              once, so they can be aggressively read ahead, and
 164  *              can be freed soon after they are accessed.
 165  *  MADV_WILLNEED - the application is notifying the system to read
 166  *              some pages ahead.
 167  *  MADV_DONTNEED - the application is finished with the given range,
 168  *              so the kernel can free resources associated with it.
 169  *
 170  * return values:
 171  *  zero    - success
 172  *  -EINVAL - start + len < 0, start is not page-aligned,
 173  *              "behavior" is not a valid value, or application
 174  *              is attempting to release locked or shared pages.
 175  *  -ENOMEM - addresses in the specified range are not currently
 176  *              mapped, or are outside the AS of the process.
 177  *  -EIO    - an I/O error occurred while paging in data.
 178  *  -EBADF  - map exists, but area maps something that isn't a file.
 179  *  -EAGAIN - a kernel resource was temporarily unavailable.
 180  */
 181 asmlinkage long sys_madvise(unsigned long start, size_t len_in, int behavior)
 182 {
 183         unsigned long end;
 184         struct vm_area_struct * vma;
 185         int unmapped_error = 0;
 186         int error = -EINVAL;
 187         size_t len;
 188
 189         down_write(&current->mm->mmap_sem);
 190
 191         if (start & ~PAGE_MASK)
 192                 goto out;
 193         len = (len_in + ~PAGE_MASK) & PAGE_MASK;
 194
 195         /* Check to see whether len was rounded up from small -ve to zero */
 196         if (len_in && !len)
 197                 goto out;
 198
 199         end = start + len;
 200         if (end < start)
 201                 goto out;
 202
 203         error = 0;
 204         if (end == start)
 205                 goto out;
 206
 207         /*
 208          * If the interval [start,end) covers some unmapped address
 209          * ranges, just ignore them, but return -ENOMEM at the end.
 210          */
 211         vma = find_vma(current->mm, start);
 212         for (;;) {
 213                 /* Still start < end. */
 214                 error = -ENOMEM;
 215                 if (!vma)
 216                         goto out;
 217
 218                 /* Here start < vma->vm_end. */
 219                 if (start < vma->vm_start) {
 220                         unmapped_error = -ENOMEM;
 221                         start = vma->vm_start;
 222                 }
 223
 224                 /* Here vma->vm_start <= start < vma->vm_end. */
 225                 if (end <= vma->vm_end) {
 226                         if (start < end) {
 227                                 error = madvise_vma(vma, start, end,
 228                                                         behavior);
 229                                 if (error)
 230                                         goto out;
 231                         }
 232                         error = unmapped_error;
 233                         goto out;
 234                 }
 235
 236                 /* Here vma->vm_start <= start < vma->vm_end < end. */
 237                 error = madvise_vma(vma, start, vma->vm_end, behavior);
 238                 if (error)
 239                         goto out;
 240                 start = vma->vm_end;
 241                 vma = vma->vm_next;
 242         }
 243
 244 out:
 245         up_write(&current->mm->mmap_sem);
 246         return error;
 247 }