4 * Copyright (C) 1999 Linus Torvalds
5 * Copyright (C) 2002 Christoph Hellwig
8 #include <linux/mman.h>
9 #include <linux/pagemap.h>
13 * We can potentially split a vm area into separate
14 * areas, each area with its own behavior.
16 static long madvise_behavior(struct vm_area_struct * vma, unsigned long start,
17 unsigned long end, int behavior)
19 struct mm_struct * mm = vma->vm_mm;
22 if (start != vma->vm_start) {
23 error = split_vma(mm, vma, start, 1);
28 if (end != vma->vm_end) {
29 error = split_vma(mm, vma, end, 0);
34 spin_lock(&mm->page_table_lock);
35 VM_ClearReadHint(vma);
39 vma->vm_flags |= VM_SEQ_READ;
42 vma->vm_flags |= VM_RAND_READ;
47 spin_unlock(&mm->page_table_lock);
53 * Schedule all required I/O operations. Do not wait for completion.
55 static long madvise_willneed(struct vm_area_struct * vma,
56 unsigned long start, unsigned long end)
58 struct file *file = vma->vm_file;
63 start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
64 if (end > vma->vm_end)
66 end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
68 do_page_cache_readahead(file->f_dentry->d_inode->i_mapping,
69 file, start, max_sane_readahead(end - start));
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.
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.
89 * An interface that causes the system to free clean pages and flush
90 * dirty pages is already available as msync(MS_INVALIDATE).
92 static long madvise_dontneed(struct vm_area_struct * vma,
93 unsigned long start, unsigned long end)
95 if (vma->vm_flags & VM_LOCKED)
98 zap_page_range(vma, start, end - start);
102 static long madvise_vma(struct vm_area_struct * vma, unsigned long start,
103 unsigned long end, int behavior)
109 case MADV_SEQUENTIAL:
111 error = madvise_behavior(vma, start, end, behavior);
115 error = madvise_willneed(vma, start, end);
119 error = madvise_dontneed(vma, start, end);
131 * The madvise(2) system call.
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.
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
150 * MADV_DONTNEED - the application is finished with the given range,
151 * so the kernel can free resources associated with it.
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.
164 asmlinkage long sys_madvise(unsigned long start, size_t len, int behavior)
167 struct vm_area_struct * vma;
168 int unmapped_error = 0;
171 down_write(¤t->mm->mmap_sem);
173 if (start & ~PAGE_MASK)
175 len = (len + ~PAGE_MASK) & PAGE_MASK;
185 * If the interval [start,end) covers some unmapped address
186 * ranges, just ignore them, but return -ENOMEM at the end.
188 vma = find_vma(current->mm, start);
190 /* Still start < end. */
195 /* Here start < vma->vm_end. */
196 if (start < vma->vm_start) {
197 unmapped_error = -ENOMEM;
198 start = vma->vm_start;
201 /* Here vma->vm_start <= start < vma->vm_end. */
202 if (end <= vma->vm_end) {
204 error = madvise_vma(vma, start, end,
209 error = unmapped_error;
213 /* Here vma->vm_start <= start < vma->vm_end < end. */
214 error = madvise_vma(vma, start, vma->vm_end, behavior);
222 up_write(¤t->mm->mmap_sem);