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/sched.h>
15 #include <linux/ksm.h>
16 #include <linux/file.h>
19 * Any behaviour which results in changes to the vma->vm_flags needs to
20 * take mmap_sem for writing. Others, which simply traverse vmas, need
21 * to only take it for reading.
23 static int madvise_need_mmap_write(int behavior)
31 /* be safe, default to 1. list exceptions explicitly */
37 * We can potentially split a vm area into separate
38 * areas, each area with its own behavior.
40 static long madvise_behavior(struct vm_area_struct * vma,
41 struct vm_area_struct **prev,
42 unsigned long start, unsigned long end, int behavior)
44 struct mm_struct * mm = vma->vm_mm;
47 unsigned long new_flags = vma->vm_flags;
51 new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ;
54 new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ;
57 new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ;
60 new_flags |= VM_DONTCOPY;
63 if (vma->vm_flags & VM_IO) {
67 new_flags &= ~VM_DONTCOPY;
70 case MADV_UNMERGEABLE:
71 error = ksm_madvise(vma, start, end, behavior, &new_flags);
77 error = hugepage_madvise(vma, &new_flags, behavior);
83 if (new_flags == vma->vm_flags) {
88 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
89 *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma,
90 vma->vm_file, pgoff, vma_policy(vma));
98 if (start != vma->vm_start) {
99 error = split_vma(mm, vma, start, 1);
104 if (end != vma->vm_end) {
105 error = split_vma(mm, vma, end, 0);
112 * vm_flags is protected by the mmap_sem held in write mode.
114 vma->vm_flags = new_flags;
117 if (error == -ENOMEM)
123 * Schedule all required I/O operations. Do not wait for completion.
125 static long madvise_willneed(struct vm_area_struct * vma,
126 struct vm_area_struct ** prev,
127 unsigned long start, unsigned long end)
129 struct file *file = vma->vm_file;
134 if (file->f_mapping->a_ops->get_xip_mem) {
135 /* no bad return value, but ignore advice */
140 start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
141 if (end > vma->vm_end)
143 end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
145 force_page_cache_readahead(file->f_mapping, file, start, end - start);
150 * Application no longer needs these pages. If the pages are dirty,
151 * it's OK to just throw them away. The app will be more careful about
152 * data it wants to keep. Be sure to free swap resources too. The
153 * zap_page_range call sets things up for shrink_active_list to actually free
154 * these pages later if no one else has touched them in the meantime,
155 * although we could add these pages to a global reuse list for
156 * shrink_active_list to pick up before reclaiming other pages.
158 * NB: This interface discards data rather than pushes it out to swap,
159 * as some implementations do. This has performance implications for
160 * applications like large transactional databases which want to discard
161 * pages in anonymous maps after committing to backing store the data
162 * that was kept in them. There is no reason to write this data out to
163 * the swap area if the application is discarding it.
165 * An interface that causes the system to free clean pages and flush
166 * dirty pages is already available as msync(MS_INVALIDATE).
168 static long madvise_dontneed(struct vm_area_struct * vma,
169 struct vm_area_struct ** prev,
170 unsigned long start, unsigned long end)
173 if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
176 if (unlikely(vma->vm_flags & VM_NONLINEAR)) {
177 struct zap_details details = {
178 .nonlinear_vma = vma,
179 .last_index = ULONG_MAX,
181 zap_page_range(vma, start, end - start, &details);
183 zap_page_range(vma, start, end - start, NULL);
188 * Application wants to free up the pages and associated backing store.
189 * This is effectively punching a hole into the middle of a file.
191 * NOTE: Currently, only shmfs/tmpfs is supported for this operation.
192 * Other filesystems return -ENOSYS.
194 static long madvise_remove(struct vm_area_struct *vma,
195 struct vm_area_struct **prev,
196 unsigned long start, unsigned long end)
198 struct address_space *mapping;
199 loff_t offset, endoff;
203 *prev = NULL; /* tell sys_madvise we drop mmap_sem */
205 if (vma->vm_flags & (VM_LOCKED|VM_NONLINEAR|VM_HUGETLB))
210 if (!f || !f->f_mapping || !f->f_mapping->host) {
214 if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE))
217 mapping = vma->vm_file->f_mapping;
219 offset = (loff_t)(start - vma->vm_start)
220 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
221 endoff = (loff_t)(end - vma->vm_start - 1)
222 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
225 * vmtruncate_range may need to take i_mutex. We need to
226 * explicitly grab a reference because the vma (and hence the
227 * vma's reference to the file) can go away as soon as we drop
231 up_read(¤t->mm->mmap_sem);
232 error = vmtruncate_range(mapping->host, offset, endoff);
234 down_read(¤t->mm->mmap_sem);
238 #ifdef CONFIG_MEMORY_FAILURE
240 * Error injection support for memory error handling.
242 static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
246 if (!capable(CAP_SYS_ADMIN))
248 for (; start < end; start += PAGE_SIZE) {
250 int ret = get_user_pages_fast(start, 1, 0, &p);
253 if (bhv == MADV_SOFT_OFFLINE) {
254 printk(KERN_INFO "Soft offlining page %lx at %lx\n",
255 page_to_pfn(p), start);
256 ret = soft_offline_page(p, MF_COUNT_INCREASED);
261 printk(KERN_INFO "Injecting memory failure for page %lx at %lx\n",
262 page_to_pfn(p), start);
263 /* Ignore return value for now */
264 __memory_failure(page_to_pfn(p), 0, MF_COUNT_INCREASED);
271 madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
272 unsigned long start, unsigned long end, int behavior)
276 return madvise_remove(vma, prev, start, end);
278 return madvise_willneed(vma, prev, start, end);
280 return madvise_dontneed(vma, prev, start, end);
282 return madvise_behavior(vma, prev, start, end, behavior);
287 madvise_behavior_valid(int behavior)
293 case MADV_SEQUENTIAL:
300 case MADV_UNMERGEABLE:
302 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
304 case MADV_NOHUGEPAGE:
314 * The madvise(2) system call.
316 * Applications can use madvise() to advise the kernel how it should
317 * handle paging I/O in this VM area. The idea is to help the kernel
318 * use appropriate read-ahead and caching techniques. The information
319 * provided is advisory only, and can be safely disregarded by the
320 * kernel without affecting the correct operation of the application.
323 * MADV_NORMAL - the default behavior is to read clusters. This
324 * results in some read-ahead and read-behind.
325 * MADV_RANDOM - the system should read the minimum amount of data
326 * on any access, since it is unlikely that the appli-
327 * cation will need more than what it asks for.
328 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
329 * once, so they can be aggressively read ahead, and
330 * can be freed soon after they are accessed.
331 * MADV_WILLNEED - the application is notifying the system to read
333 * MADV_DONTNEED - the application is finished with the given range,
334 * so the kernel can free resources associated with it.
335 * MADV_REMOVE - the application wants to free up the given range of
336 * pages and associated backing store.
337 * MADV_DONTFORK - omit this area from child's address space when forking:
338 * typically, to avoid COWing pages pinned by get_user_pages().
339 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
340 * MADV_MERGEABLE - the application recommends that KSM try to merge pages in
341 * this area with pages of identical content from other such areas.
342 * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others.
346 * -EINVAL - start + len < 0, start is not page-aligned,
347 * "behavior" is not a valid value, or application
348 * is attempting to release locked or shared pages.
349 * -ENOMEM - addresses in the specified range are not currently
350 * mapped, or are outside the AS of the process.
351 * -EIO - an I/O error occurred while paging in data.
352 * -EBADF - map exists, but area maps something that isn't a file.
353 * -EAGAIN - a kernel resource was temporarily unavailable.
355 SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
357 unsigned long end, tmp;
358 struct vm_area_struct * vma, *prev;
359 int unmapped_error = 0;
364 #ifdef CONFIG_MEMORY_FAILURE
365 if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE)
366 return madvise_hwpoison(behavior, start, start+len_in);
368 if (!madvise_behavior_valid(behavior))
371 write = madvise_need_mmap_write(behavior);
373 down_write(¤t->mm->mmap_sem);
375 down_read(¤t->mm->mmap_sem);
377 if (start & ~PAGE_MASK)
379 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
381 /* Check to see whether len was rounded up from small -ve to zero */
394 * If the interval [start,end) covers some unmapped address
395 * ranges, just ignore them, but return -ENOMEM at the end.
396 * - different from the way of handling in mlock etc.
398 vma = find_vma_prev(current->mm, start, &prev);
399 if (vma && start > vma->vm_start)
403 /* Still start < end. */
408 /* Here start < (end|vma->vm_end). */
409 if (start < vma->vm_start) {
410 unmapped_error = -ENOMEM;
411 start = vma->vm_start;
416 /* Here vma->vm_start <= start < (end|vma->vm_end) */
421 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
422 error = madvise_vma(vma, &prev, start, tmp, behavior);
426 if (prev && start < prev->vm_end)
427 start = prev->vm_end;
428 error = unmapped_error;
433 else /* madvise_remove dropped mmap_sem */
434 vma = find_vma(current->mm, start);
438 up_write(¤t->mm->mmap_sem);
440 up_read(¤t->mm->mmap_sem);