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/page-isolation.h>
13 #include <linux/hugetlb.h>
14 #include <linux/falloc.h>
15 #include <linux/sched.h>
16 #include <linux/ksm.h>
18 #include <linux/file.h>
21 * Any behaviour which results in changes to the vma->vm_flags needs to
22 * take mmap_sem for writing. Others, which simply traverse vmas, need
23 * to only take it for reading.
25 static int madvise_need_mmap_write(int behavior)
33 /* be safe, default to 1. list exceptions explicitly */
39 * We can potentially split a vm area into separate
40 * areas, each area with its own behavior.
42 static long madvise_behavior(struct vm_area_struct * vma,
43 struct vm_area_struct **prev,
44 unsigned long start, unsigned long end, int behavior)
46 struct mm_struct * mm = vma->vm_mm;
49 unsigned long new_flags = vma->vm_flags;
53 new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ;
56 new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ;
59 new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ;
62 new_flags |= VM_DONTCOPY;
65 if (vma->vm_flags & VM_IO) {
69 new_flags &= ~VM_DONTCOPY;
72 new_flags |= VM_NODUMP;
75 new_flags &= ~VM_NODUMP;
78 case MADV_UNMERGEABLE:
79 error = ksm_madvise(vma, start, end, behavior, &new_flags);
85 error = hugepage_madvise(vma, &new_flags, behavior);
91 if (new_flags == vma->vm_flags) {
96 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
97 *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma,
98 vma->vm_file, pgoff, vma_policy(vma));
106 if (start != vma->vm_start) {
107 error = split_vma(mm, vma, start, 1);
112 if (end != vma->vm_end) {
113 error = split_vma(mm, vma, end, 0);
120 * vm_flags is protected by the mmap_sem held in write mode.
122 vma->vm_flags = new_flags;
125 if (error == -ENOMEM)
131 * Schedule all required I/O operations. Do not wait for completion.
133 static long madvise_willneed(struct vm_area_struct * vma,
134 struct vm_area_struct ** prev,
135 unsigned long start, unsigned long end)
137 struct file *file = vma->vm_file;
142 if (file->f_mapping->a_ops->get_xip_mem) {
143 /* no bad return value, but ignore advice */
148 start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
149 if (end > vma->vm_end)
151 end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
153 force_page_cache_readahead(file->f_mapping, file, start, end - start);
158 * Application no longer needs these pages. If the pages are dirty,
159 * it's OK to just throw them away. The app will be more careful about
160 * data it wants to keep. Be sure to free swap resources too. The
161 * zap_page_range call sets things up for shrink_active_list to actually free
162 * these pages later if no one else has touched them in the meantime,
163 * although we could add these pages to a global reuse list for
164 * shrink_active_list to pick up before reclaiming other pages.
166 * NB: This interface discards data rather than pushes it out to swap,
167 * as some implementations do. This has performance implications for
168 * applications like large transactional databases which want to discard
169 * pages in anonymous maps after committing to backing store the data
170 * that was kept in them. There is no reason to write this data out to
171 * the swap area if the application is discarding it.
173 * An interface that causes the system to free clean pages and flush
174 * dirty pages is already available as msync(MS_INVALIDATE).
176 static long madvise_dontneed(struct vm_area_struct * vma,
177 struct vm_area_struct ** prev,
178 unsigned long start, unsigned long end)
181 if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
184 if (unlikely(vma->vm_flags & VM_NONLINEAR)) {
185 struct zap_details details = {
186 .nonlinear_vma = vma,
187 .last_index = ULONG_MAX,
189 zap_page_range(vma, start, end - start, &details);
191 zap_page_range(vma, start, end - start, NULL);
196 * Application wants to free up the pages and associated backing store.
197 * This is effectively punching a hole into the middle of a file.
199 * NOTE: Currently, only shmfs/tmpfs is supported for this operation.
200 * Other filesystems return -ENOSYS.
202 static long madvise_remove(struct vm_area_struct *vma,
203 struct vm_area_struct **prev,
204 unsigned long start, unsigned long end)
210 *prev = NULL; /* tell sys_madvise we drop mmap_sem */
212 if (vma->vm_flags & (VM_LOCKED|VM_NONLINEAR|VM_HUGETLB))
217 if (!f || !f->f_mapping || !f->f_mapping->host) {
221 if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE))
224 offset = (loff_t)(start - vma->vm_start)
225 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
228 * Filesystem's fallocate may need to take i_mutex. We need to
229 * explicitly grab a reference because the vma (and hence the
230 * vma's reference to the file) can go away as soon as we drop
234 up_read(¤t->mm->mmap_sem);
235 error = do_fallocate(f,
236 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
237 offset, end - start);
239 down_read(¤t->mm->mmap_sem);
243 #ifdef CONFIG_MEMORY_FAILURE
245 * Error injection support for memory error handling.
247 static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
251 if (!capable(CAP_SYS_ADMIN))
253 for (; start < end; start += PAGE_SIZE) {
255 int ret = get_user_pages_fast(start, 1, 0, &p);
258 if (bhv == MADV_SOFT_OFFLINE) {
259 printk(KERN_INFO "Soft offlining page %lx at %lx\n",
260 page_to_pfn(p), start);
261 ret = soft_offline_page(p, MF_COUNT_INCREASED);
266 printk(KERN_INFO "Injecting memory failure for page %lx at %lx\n",
267 page_to_pfn(p), start);
268 /* Ignore return value for now */
269 memory_failure(page_to_pfn(p), 0, MF_COUNT_INCREASED);
276 madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
277 unsigned long start, unsigned long end, int behavior)
281 return madvise_remove(vma, prev, start, end);
283 return madvise_willneed(vma, prev, start, end);
285 return madvise_dontneed(vma, prev, start, end);
287 return madvise_behavior(vma, prev, start, end, behavior);
292 madvise_behavior_valid(int behavior)
298 case MADV_SEQUENTIAL:
305 case MADV_UNMERGEABLE:
307 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
309 case MADV_NOHUGEPAGE:
321 * The madvise(2) system call.
323 * Applications can use madvise() to advise the kernel how it should
324 * handle paging I/O in this VM area. The idea is to help the kernel
325 * use appropriate read-ahead and caching techniques. The information
326 * provided is advisory only, and can be safely disregarded by the
327 * kernel without affecting the correct operation of the application.
330 * MADV_NORMAL - the default behavior is to read clusters. This
331 * results in some read-ahead and read-behind.
332 * MADV_RANDOM - the system should read the minimum amount of data
333 * on any access, since it is unlikely that the appli-
334 * cation will need more than what it asks for.
335 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
336 * once, so they can be aggressively read ahead, and
337 * can be freed soon after they are accessed.
338 * MADV_WILLNEED - the application is notifying the system to read
340 * MADV_DONTNEED - the application is finished with the given range,
341 * so the kernel can free resources associated with it.
342 * MADV_REMOVE - the application wants to free up the given range of
343 * pages and associated backing store.
344 * MADV_DONTFORK - omit this area from child's address space when forking:
345 * typically, to avoid COWing pages pinned by get_user_pages().
346 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
347 * MADV_MERGEABLE - the application recommends that KSM try to merge pages in
348 * this area with pages of identical content from other such areas.
349 * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others.
353 * -EINVAL - start + len < 0, start is not page-aligned,
354 * "behavior" is not a valid value, or application
355 * is attempting to release locked or shared pages.
356 * -ENOMEM - addresses in the specified range are not currently
357 * mapped, or are outside the AS of the process.
358 * -EIO - an I/O error occurred while paging in data.
359 * -EBADF - map exists, but area maps something that isn't a file.
360 * -EAGAIN - a kernel resource was temporarily unavailable.
362 SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
364 unsigned long end, tmp;
365 struct vm_area_struct * vma, *prev;
366 int unmapped_error = 0;
371 #ifdef CONFIG_MEMORY_FAILURE
372 if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE)
373 return madvise_hwpoison(behavior, start, start+len_in);
375 if (!madvise_behavior_valid(behavior))
378 write = madvise_need_mmap_write(behavior);
380 down_write(¤t->mm->mmap_sem);
382 down_read(¤t->mm->mmap_sem);
384 if (start & ~PAGE_MASK)
386 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
388 /* Check to see whether len was rounded up from small -ve to zero */
401 * If the interval [start,end) covers some unmapped address
402 * ranges, just ignore them, but return -ENOMEM at the end.
403 * - different from the way of handling in mlock etc.
405 vma = find_vma_prev(current->mm, start, &prev);
406 if (vma && start > vma->vm_start)
410 /* Still start < end. */
415 /* Here start < (end|vma->vm_end). */
416 if (start < vma->vm_start) {
417 unmapped_error = -ENOMEM;
418 start = vma->vm_start;
423 /* Here vma->vm_start <= start < (end|vma->vm_end) */
428 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
429 error = madvise_vma(vma, &prev, start, tmp, behavior);
433 if (prev && start < prev->vm_end)
434 start = prev->vm_end;
435 error = unmapped_error;
440 else /* madvise_remove dropped mmap_sem */
441 vma = find_vma(current->mm, start);
445 up_write(¤t->mm->mmap_sem);
447 up_read(¤t->mm->mmap_sem);