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 & ~VM_READHINTMASK;
29 new_flags |= VM_SEQ_READ;
32 new_flags |= VM_RAND_READ;
38 if (new_flags == vma->vm_flags) {
43 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
44 *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma,
45 vma->vm_file, pgoff, vma_policy(vma));
53 if (start != vma->vm_start) {
54 error = split_vma(mm, vma, start, 1);
59 if (end != vma->vm_end) {
60 error = split_vma(mm, vma, end, 0);
67 * vm_flags is protected by the mmap_sem held in write mode.
69 vma->vm_flags = new_flags;
78 * Schedule all required I/O operations. Do not wait for completion.
80 static long madvise_willneed(struct vm_area_struct * vma,
81 struct vm_area_struct ** prev,
82 unsigned long start, unsigned long end)
84 struct file *file = vma->vm_file;
86 if (file->f_mapping->a_ops->get_xip_page) {
87 /* no bad return value, but ignore advice */
92 start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
93 if (end > vma->vm_end)
95 end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
97 force_page_cache_readahead(file->f_mapping,
98 file, start, max_sane_readahead(end - start));
103 * Application no longer needs these pages. If the pages are dirty,
104 * it's OK to just throw them away. The app will be more careful about
105 * data it wants to keep. Be sure to free swap resources too. The
106 * zap_page_range call sets things up for refill_inactive to actually free
107 * these pages later if no one else has touched them in the meantime,
108 * although we could add these pages to a global reuse list for
109 * refill_inactive to pick up before reclaiming other pages.
111 * NB: This interface discards data rather than pushes it out to swap,
112 * as some implementations do. This has performance implications for
113 * applications like large transactional databases which want to discard
114 * pages in anonymous maps after committing to backing store the data
115 * that was kept in them. There is no reason to write this data out to
116 * the swap area if the application is discarding it.
118 * An interface that causes the system to free clean pages and flush
119 * dirty pages is already available as msync(MS_INVALIDATE).
121 static long madvise_dontneed(struct vm_area_struct * vma,
122 struct vm_area_struct ** prev,
123 unsigned long start, unsigned long end)
126 if ((vma->vm_flags & VM_LOCKED) || is_vm_hugetlb_page(vma))
129 if (unlikely(vma->vm_flags & VM_NONLINEAR)) {
130 struct zap_details details = {
131 .nonlinear_vma = vma,
132 .last_index = ULONG_MAX,
134 zap_page_range(vma, start, end - start, &details);
136 zap_page_range(vma, start, end - start, NULL);
141 madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
142 unsigned long start, unsigned long end, int behavior)
144 struct file *filp = vma->vm_file;
152 case MADV_SEQUENTIAL:
154 error = madvise_behavior(vma, prev, start, end, behavior);
158 error = madvise_willneed(vma, prev, start, end);
162 error = madvise_dontneed(vma, prev, start, end);
175 * The madvise(2) system call.
177 * Applications can use madvise() to advise the kernel how it should
178 * handle paging I/O in this VM area. The idea is to help the kernel
179 * use appropriate read-ahead and caching techniques. The information
180 * provided is advisory only, and can be safely disregarded by the
181 * kernel without affecting the correct operation of the application.
184 * MADV_NORMAL - the default behavior is to read clusters. This
185 * results in some read-ahead and read-behind.
186 * MADV_RANDOM - the system should read the minimum amount of data
187 * on any access, since it is unlikely that the appli-
188 * cation will need more than what it asks for.
189 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
190 * once, so they can be aggressively read ahead, and
191 * can be freed soon after they are accessed.
192 * MADV_WILLNEED - the application is notifying the system to read
194 * MADV_DONTNEED - the application is finished with the given range,
195 * so the kernel can free resources associated with it.
199 * -EINVAL - start + len < 0, start is not page-aligned,
200 * "behavior" is not a valid value, or application
201 * is attempting to release locked or shared pages.
202 * -ENOMEM - addresses in the specified range are not currently
203 * mapped, or are outside the AS of the process.
204 * -EIO - an I/O error occurred while paging in data.
205 * -EBADF - map exists, but area maps something that isn't a file.
206 * -EAGAIN - a kernel resource was temporarily unavailable.
208 asmlinkage long sys_madvise(unsigned long start, size_t len_in, int behavior)
210 unsigned long end, tmp;
211 struct vm_area_struct * vma, *prev;
212 int unmapped_error = 0;
216 down_write(¤t->mm->mmap_sem);
218 if (start & ~PAGE_MASK)
220 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
222 /* Check to see whether len was rounded up from small -ve to zero */
235 * If the interval [start,end) covers some unmapped address
236 * ranges, just ignore them, but return -ENOMEM at the end.
237 * - different from the way of handling in mlock etc.
239 vma = find_vma_prev(current->mm, start, &prev);
240 if (vma && start > vma->vm_start)
244 /* Still start < end. */
249 /* Here start < (end|vma->vm_end). */
250 if (start < vma->vm_start) {
251 unmapped_error = -ENOMEM;
252 start = vma->vm_start;
257 /* Here vma->vm_start <= start < (end|vma->vm_end) */
262 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
263 error = madvise_vma(vma, &prev, start, tmp, behavior);
267 if (start < prev->vm_end)
268 start = prev->vm_end;
269 error = unmapped_error;
275 up_write(¤t->mm->mmap_sem);