4 * Copyright (C) 1999 Linus Torvalds
5 * Copyright (C) 2002 Christoph Hellwig
8 #include <linux/mman.h>
9 #include <linux/pagemap.h>
10 #include <linux/syscalls.h>
11 #include <linux/mempolicy.h>
12 #include <linux/hugetlb.h>
15 * We can potentially split a vm area into separate
16 * areas, each area with its own behavior.
18 static long madvise_behavior(struct vm_area_struct * vma,
19 struct vm_area_struct **prev,
20 unsigned long start, unsigned long end, int behavior)
22 struct mm_struct * mm = vma->vm_mm;
25 int new_flags = vma->vm_flags;
29 new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ;
32 new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ;
35 new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ;
38 new_flags |= VM_DONTCOPY;
41 new_flags &= ~VM_DONTCOPY;
45 if (new_flags == vma->vm_flags) {
50 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
51 *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma,
52 vma->vm_file, pgoff, vma_policy(vma));
60 if (start != vma->vm_start) {
61 error = split_vma(mm, vma, start, 1);
66 if (end != vma->vm_end) {
67 error = split_vma(mm, vma, end, 0);
74 * vm_flags is protected by the mmap_sem held in write mode.
76 vma->vm_flags = new_flags;
85 * Schedule all required I/O operations. Do not wait for completion.
87 static long madvise_willneed(struct vm_area_struct * vma,
88 struct vm_area_struct ** prev,
89 unsigned long start, unsigned long end)
91 struct file *file = vma->vm_file;
96 if (file->f_mapping->a_ops->get_xip_page) {
97 /* no bad return value, but ignore advice */
102 start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
103 if (end > vma->vm_end)
105 end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
107 force_page_cache_readahead(file->f_mapping,
108 file, start, max_sane_readahead(end - start));
113 * Application no longer needs these pages. If the pages are dirty,
114 * it's OK to just throw them away. The app will be more careful about
115 * data it wants to keep. Be sure to free swap resources too. The
116 * zap_page_range call sets things up for refill_inactive to actually free
117 * these pages later if no one else has touched them in the meantime,
118 * although we could add these pages to a global reuse list for
119 * refill_inactive to pick up before reclaiming other pages.
121 * NB: This interface discards data rather than pushes it out to swap,
122 * as some implementations do. This has performance implications for
123 * applications like large transactional databases which want to discard
124 * pages in anonymous maps after committing to backing store the data
125 * that was kept in them. There is no reason to write this data out to
126 * the swap area if the application is discarding it.
128 * An interface that causes the system to free clean pages and flush
129 * dirty pages is already available as msync(MS_INVALIDATE).
131 static long madvise_dontneed(struct vm_area_struct * vma,
132 struct vm_area_struct ** prev,
133 unsigned long start, unsigned long end)
136 if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
139 if (unlikely(vma->vm_flags & VM_NONLINEAR)) {
140 struct zap_details details = {
141 .nonlinear_vma = vma,
142 .last_index = ULONG_MAX,
144 zap_page_range(vma, start, end - start, &details);
146 zap_page_range(vma, start, end - start, NULL);
151 * Application wants to free up the pages and associated backing store.
152 * This is effectively punching a hole into the middle of a file.
154 * NOTE: Currently, only shmfs/tmpfs is supported for this operation.
155 * Other filesystems return -ENOSYS.
157 static long madvise_remove(struct vm_area_struct *vma,
158 struct vm_area_struct **prev,
159 unsigned long start, unsigned long end)
161 struct address_space *mapping;
162 loff_t offset, endoff;
166 if (vma->vm_flags & (VM_LOCKED|VM_NONLINEAR|VM_HUGETLB))
169 if (!vma->vm_file || !vma->vm_file->f_mapping
170 || !vma->vm_file->f_mapping->host) {
174 if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE))
177 mapping = vma->vm_file->f_mapping;
179 offset = (loff_t)(start - vma->vm_start)
180 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
181 endoff = (loff_t)(end - vma->vm_start - 1)
182 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
183 return vmtruncate_range(mapping->host, offset, endoff);
187 madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
188 unsigned long start, unsigned long end, int behavior)
194 if (vma->vm_flags & VM_IO) {
200 case MADV_SEQUENTIAL:
202 error = madvise_behavior(vma, prev, start, end, behavior);
205 error = madvise_remove(vma, prev, start, end);
209 error = madvise_willneed(vma, prev, start, end);
213 error = madvise_dontneed(vma, prev, start, end);
224 * The madvise(2) system call.
226 * Applications can use madvise() to advise the kernel how it should
227 * handle paging I/O in this VM area. The idea is to help the kernel
228 * use appropriate read-ahead and caching techniques. The information
229 * provided is advisory only, and can be safely disregarded by the
230 * kernel without affecting the correct operation of the application.
233 * MADV_NORMAL - the default behavior is to read clusters. This
234 * results in some read-ahead and read-behind.
235 * MADV_RANDOM - the system should read the minimum amount of data
236 * on any access, since it is unlikely that the appli-
237 * cation will need more than what it asks for.
238 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
239 * once, so they can be aggressively read ahead, and
240 * can be freed soon after they are accessed.
241 * MADV_WILLNEED - the application is notifying the system to read
243 * MADV_DONTNEED - the application is finished with the given range,
244 * so the kernel can free resources associated with it.
245 * MADV_REMOVE - the application wants to free up the given range of
246 * pages and associated backing store.
250 * -EINVAL - start + len < 0, start is not page-aligned,
251 * "behavior" is not a valid value, or application
252 * is attempting to release locked or shared pages.
253 * -ENOMEM - addresses in the specified range are not currently
254 * mapped, or are outside the AS of the process.
255 * -EIO - an I/O error occurred while paging in data.
256 * -EBADF - map exists, but area maps something that isn't a file.
257 * -EAGAIN - a kernel resource was temporarily unavailable.
259 asmlinkage long sys_madvise(unsigned long start, size_t len_in, int behavior)
261 unsigned long end, tmp;
262 struct vm_area_struct * vma, *prev;
263 int unmapped_error = 0;
267 down_write(¤t->mm->mmap_sem);
269 if (start & ~PAGE_MASK)
271 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
273 /* Check to see whether len was rounded up from small -ve to zero */
286 * If the interval [start,end) covers some unmapped address
287 * ranges, just ignore them, but return -ENOMEM at the end.
288 * - different from the way of handling in mlock etc.
290 vma = find_vma_prev(current->mm, start, &prev);
291 if (vma && start > vma->vm_start)
295 /* Still start < end. */
300 /* Here start < (end|vma->vm_end). */
301 if (start < vma->vm_start) {
302 unmapped_error = -ENOMEM;
303 start = vma->vm_start;
308 /* Here vma->vm_start <= start < (end|vma->vm_end) */
313 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
314 error = madvise_vma(vma, &prev, start, tmp, behavior);
318 if (start < prev->vm_end)
319 start = prev->vm_end;
320 error = unmapped_error;
326 up_write(¤t->mm->mmap_sem);