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>
13 #include <linux/sched.h>
16 * Any behaviour which results in changes to the vma->vm_flags needs to
17 * take mmap_sem for writing. Others, which simply traverse vmas, need
18 * to only take it for reading.
20 static int madvise_need_mmap_write(int behavior)
28 /* be safe, default to 1. list exceptions explicitly */
34 * We can potentially split a vm area into separate
35 * areas, each area with its own behavior.
37 static long madvise_behavior(struct vm_area_struct * vma,
38 struct vm_area_struct **prev,
39 unsigned long start, unsigned long end, int behavior)
41 struct mm_struct * mm = vma->vm_mm;
44 int new_flags = vma->vm_flags;
48 new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ;
51 new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ;
54 new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ;
57 new_flags |= VM_DONTCOPY;
60 new_flags &= ~VM_DONTCOPY;
64 if (new_flags == vma->vm_flags) {
69 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
70 *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma,
71 vma->vm_file, pgoff, vma_policy(vma));
79 if (start != vma->vm_start) {
80 error = split_vma(mm, vma, start, 1);
85 if (end != vma->vm_end) {
86 error = split_vma(mm, vma, end, 0);
93 * vm_flags is protected by the mmap_sem held in write mode.
95 vma->vm_flags = new_flags;
104 * Schedule all required I/O operations. Do not wait for completion.
106 static long madvise_willneed(struct vm_area_struct * vma,
107 struct vm_area_struct ** prev,
108 unsigned long start, unsigned long end)
110 struct file *file = vma->vm_file;
115 if (file->f_mapping->a_ops->get_xip_mem) {
116 /* no bad return value, but ignore advice */
121 start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
122 if (end > vma->vm_end)
124 end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
126 force_page_cache_readahead(file->f_mapping, file, start, end - start);
131 * Application no longer needs these pages. If the pages are dirty,
132 * it's OK to just throw them away. The app will be more careful about
133 * data it wants to keep. Be sure to free swap resources too. The
134 * zap_page_range call sets things up for shrink_active_list to actually free
135 * these pages later if no one else has touched them in the meantime,
136 * although we could add these pages to a global reuse list for
137 * shrink_active_list to pick up before reclaiming other pages.
139 * NB: This interface discards data rather than pushes it out to swap,
140 * as some implementations do. This has performance implications for
141 * applications like large transactional databases which want to discard
142 * pages in anonymous maps after committing to backing store the data
143 * that was kept in them. There is no reason to write this data out to
144 * the swap area if the application is discarding it.
146 * An interface that causes the system to free clean pages and flush
147 * dirty pages is already available as msync(MS_INVALIDATE).
149 static long madvise_dontneed(struct vm_area_struct * vma,
150 struct vm_area_struct ** prev,
151 unsigned long start, unsigned long end)
154 if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
157 if (unlikely(vma->vm_flags & VM_NONLINEAR)) {
158 struct zap_details details = {
159 .nonlinear_vma = vma,
160 .last_index = ULONG_MAX,
162 zap_page_range(vma, start, end - start, &details);
164 zap_page_range(vma, start, end - start, NULL);
169 * Application wants to free up the pages and associated backing store.
170 * This is effectively punching a hole into the middle of a file.
172 * NOTE: Currently, only shmfs/tmpfs is supported for this operation.
173 * Other filesystems return -ENOSYS.
175 static long madvise_remove(struct vm_area_struct *vma,
176 struct vm_area_struct **prev,
177 unsigned long start, unsigned long end)
179 struct address_space *mapping;
180 loff_t offset, endoff;
183 *prev = NULL; /* tell sys_madvise we drop mmap_sem */
185 if (vma->vm_flags & (VM_LOCKED|VM_NONLINEAR|VM_HUGETLB))
188 if (!vma->vm_file || !vma->vm_file->f_mapping
189 || !vma->vm_file->f_mapping->host) {
193 if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE))
196 mapping = vma->vm_file->f_mapping;
198 offset = (loff_t)(start - vma->vm_start)
199 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
200 endoff = (loff_t)(end - vma->vm_start - 1)
201 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
203 /* vmtruncate_range needs to take i_mutex and i_alloc_sem */
204 up_read(¤t->mm->mmap_sem);
205 error = vmtruncate_range(mapping->host, offset, endoff);
206 down_read(¤t->mm->mmap_sem);
211 madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
212 unsigned long start, unsigned long end, int behavior)
218 if (vma->vm_flags & VM_IO) {
224 case MADV_SEQUENTIAL:
226 error = madvise_behavior(vma, prev, start, end, behavior);
229 error = madvise_remove(vma, prev, start, end);
233 error = madvise_willneed(vma, prev, start, end);
237 error = madvise_dontneed(vma, prev, start, end);
248 madvise_behavior_valid(int behavior)
254 case MADV_SEQUENTIAL:
266 * The madvise(2) system call.
268 * Applications can use madvise() to advise the kernel how it should
269 * handle paging I/O in this VM area. The idea is to help the kernel
270 * use appropriate read-ahead and caching techniques. The information
271 * provided is advisory only, and can be safely disregarded by the
272 * kernel without affecting the correct operation of the application.
275 * MADV_NORMAL - the default behavior is to read clusters. This
276 * results in some read-ahead and read-behind.
277 * MADV_RANDOM - the system should read the minimum amount of data
278 * on any access, since it is unlikely that the appli-
279 * cation will need more than what it asks for.
280 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
281 * once, so they can be aggressively read ahead, and
282 * can be freed soon after they are accessed.
283 * MADV_WILLNEED - the application is notifying the system to read
285 * MADV_DONTNEED - the application is finished with the given range,
286 * so the kernel can free resources associated with it.
287 * MADV_REMOVE - the application wants to free up the given range of
288 * pages and associated backing store.
292 * -EINVAL - start + len < 0, start is not page-aligned,
293 * "behavior" is not a valid value, or application
294 * is attempting to release locked or shared pages.
295 * -ENOMEM - addresses in the specified range are not currently
296 * mapped, or are outside the AS of the process.
297 * -EIO - an I/O error occurred while paging in data.
298 * -EBADF - map exists, but area maps something that isn't a file.
299 * -EAGAIN - a kernel resource was temporarily unavailable.
301 SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
303 unsigned long end, tmp;
304 struct vm_area_struct * vma, *prev;
305 int unmapped_error = 0;
310 if (!madvise_behavior_valid(behavior))
313 write = madvise_need_mmap_write(behavior);
315 down_write(¤t->mm->mmap_sem);
317 down_read(¤t->mm->mmap_sem);
319 if (start & ~PAGE_MASK)
321 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
323 /* Check to see whether len was rounded up from small -ve to zero */
336 * If the interval [start,end) covers some unmapped address
337 * ranges, just ignore them, but return -ENOMEM at the end.
338 * - different from the way of handling in mlock etc.
340 vma = find_vma_prev(current->mm, start, &prev);
341 if (vma && start > vma->vm_start)
345 /* Still start < end. */
350 /* Here start < (end|vma->vm_end). */
351 if (start < vma->vm_start) {
352 unmapped_error = -ENOMEM;
353 start = vma->vm_start;
358 /* Here vma->vm_start <= start < (end|vma->vm_end) */
363 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
364 error = madvise_vma(vma, &prev, start, tmp, behavior);
368 if (prev && start < prev->vm_end)
369 start = prev->vm_end;
370 error = unmapped_error;
375 else /* madvise_remove dropped mmap_sem */
376 vma = find_vma(current->mm, start);
380 up_write(¤t->mm->mmap_sem);
382 up_read(¤t->mm->mmap_sem);