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);
35 * vm_flags is protected by the mmap_sem held in write mode.
37 VM_ClearReadHint(vma);
41 vma->vm_flags |= VM_SEQ_READ;
44 vma->vm_flags |= VM_RAND_READ;
54 * Schedule all required I/O operations. Do not wait for completion.
56 static long madvise_willneed(struct vm_area_struct * vma,
57 unsigned long start, unsigned long end)
59 struct file *file = vma->vm_file;
64 start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
65 if (end > vma->vm_end)
67 end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
69 force_page_cache_readahead(file->f_mapping,
70 file, start, max_sane_readahead(end - start));
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.
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.
90 * An interface that causes the system to free clean pages and flush
91 * dirty pages is already available as msync(MS_INVALIDATE).
93 static long madvise_dontneed(struct vm_area_struct * vma,
94 unsigned long start, unsigned long end)
96 if (vma->vm_flags & VM_LOCKED)
99 if (unlikely(vma->vm_flags & VM_NONLINEAR)) {
100 struct zap_details details = {
101 .nonlinear_vma = vma,
102 .last_index = ULONG_MAX,
104 zap_page_range(vma, start, end - start, &details);
106 zap_page_range(vma, start, end - start, NULL);
110 static long madvise_vma(struct vm_area_struct * vma, unsigned long start,
111 unsigned long end, int behavior)
117 case MADV_SEQUENTIAL:
119 error = madvise_behavior(vma, start, end, behavior);
123 error = madvise_willneed(vma, start, end);
127 error = madvise_dontneed(vma, start, end);
139 * The madvise(2) system call.
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.
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
158 * MADV_DONTNEED - the application is finished with the given range,
159 * so the kernel can free resources associated with it.
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.
172 asmlinkage long sys_madvise(unsigned long start, size_t len_in, int behavior)
175 struct vm_area_struct * vma;
176 int unmapped_error = 0;
180 down_write(¤t->mm->mmap_sem);
182 if (start & ~PAGE_MASK)
184 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
186 /* Check to see whether len was rounded up from small -ve to zero */
199 * If the interval [start,end) covers some unmapped address
200 * ranges, just ignore them, but return -ENOMEM at the end.
202 vma = find_vma(current->mm, start);
204 /* Still start < end. */
209 /* Here start < vma->vm_end. */
210 if (start < vma->vm_start) {
211 unmapped_error = -ENOMEM;
212 start = vma->vm_start;
215 /* Here vma->vm_start <= start < vma->vm_end. */
216 if (end <= vma->vm_end) {
218 error = madvise_vma(vma, start, end,
223 error = unmapped_error;
227 /* Here vma->vm_start <= start < vma->vm_end < end. */
228 error = madvise_vma(vma, start, vma->vm_end, behavior);
236 up_write(¤t->mm->mmap_sem);