blob: 74bf193eff04fdeea6c433675a3705adbeb3c0f1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/madvise.c
3 *
4 * Copyright (C) 1999 Linus Torvalds
5 * Copyright (C) 2002 Christoph Hellwig
6 */
7
8#include <linux/mman.h>
9#include <linux/pagemap.h>
10#include <linux/syscalls.h>
Prasanna Meda05b74382005-06-21 17:14:37 -070011#include <linux/mempolicy.h>
Andi Kleenafcf9382009-12-16 12:20:00 +010012#include <linux/page-isolation.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/hugetlb.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040014#include <linux/sched.h>
Hugh Dickinsf8af4da2009-09-21 17:01:57 -070015#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016
17/*
Nick Piggin0a27a142007-05-06 14:49:53 -070018 * Any behaviour which results in changes to the vma->vm_flags needs to
19 * take mmap_sem for writing. Others, which simply traverse vmas, need
20 * to only take it for reading.
21 */
22static int madvise_need_mmap_write(int behavior)
23{
24 switch (behavior) {
25 case MADV_REMOVE:
26 case MADV_WILLNEED:
27 case MADV_DONTNEED:
28 return 0;
29 default:
30 /* be safe, default to 1. list exceptions explicitly */
31 return 1;
32 }
33}
34
35/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 * We can potentially split a vm area into separate
37 * areas, each area with its own behavior.
38 */
Prasanna Meda05b74382005-06-21 17:14:37 -070039static long madvise_behavior(struct vm_area_struct * vma,
40 struct vm_area_struct **prev,
41 unsigned long start, unsigned long end, int behavior)
Linus Torvalds1da177e2005-04-16 15:20:36 -070042{
43 struct mm_struct * mm = vma->vm_mm;
44 int error = 0;
Prasanna Meda05b74382005-06-21 17:14:37 -070045 pgoff_t pgoff;
Hugh Dickins3866ea92009-09-21 17:01:52 -070046 unsigned long new_flags = vma->vm_flags;
Prasanna Medae798c6e2005-06-21 17:14:36 -070047
48 switch (behavior) {
Michael S. Tsirkinf8225662006-02-14 13:53:08 -080049 case MADV_NORMAL:
50 new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ;
51 break;
Prasanna Medae798c6e2005-06-21 17:14:36 -070052 case MADV_SEQUENTIAL:
Michael S. Tsirkinf8225662006-02-14 13:53:08 -080053 new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ;
Prasanna Medae798c6e2005-06-21 17:14:36 -070054 break;
55 case MADV_RANDOM:
Michael S. Tsirkinf8225662006-02-14 13:53:08 -080056 new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ;
Prasanna Medae798c6e2005-06-21 17:14:36 -070057 break;
Michael S. Tsirkinf8225662006-02-14 13:53:08 -080058 case MADV_DONTFORK:
59 new_flags |= VM_DONTCOPY;
60 break;
61 case MADV_DOFORK:
Hugh Dickins3866ea92009-09-21 17:01:52 -070062 if (vma->vm_flags & VM_IO) {
63 error = -EINVAL;
64 goto out;
65 }
Michael S. Tsirkinf8225662006-02-14 13:53:08 -080066 new_flags &= ~VM_DONTCOPY;
Prasanna Medae798c6e2005-06-21 17:14:36 -070067 break;
Hugh Dickinsf8af4da2009-09-21 17:01:57 -070068 case MADV_MERGEABLE:
69 case MADV_UNMERGEABLE:
70 error = ksm_madvise(vma, start, end, behavior, &new_flags);
71 if (error)
72 goto out;
73 break;
Andrea Arcangeli0af4e982011-01-13 15:46:55 -080074 case MADV_HUGEPAGE:
Andrea Arcangelia664b2d2011-01-13 15:47:17 -080075 case MADV_NOHUGEPAGE:
Andrea Arcangeli60ab3242011-01-13 15:47:18 -080076 error = hugepage_madvise(vma, &new_flags, behavior);
Andrea Arcangeli0af4e982011-01-13 15:46:55 -080077 if (error)
78 goto out;
79 break;
Prasanna Medae798c6e2005-06-21 17:14:36 -070080 }
81
Prasanna Meda05b74382005-06-21 17:14:37 -070082 if (new_flags == vma->vm_flags) {
83 *prev = vma;
Hugh Dickins836d5ff2005-09-03 15:54:53 -070084 goto out;
Prasanna Meda05b74382005-06-21 17:14:37 -070085 }
86
87 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
88 *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma,
89 vma->vm_file, pgoff, vma_policy(vma));
90 if (*prev) {
91 vma = *prev;
92 goto success;
93 }
94
95 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97 if (start != vma->vm_start) {
98 error = split_vma(mm, vma, start, 1);
99 if (error)
100 goto out;
101 }
102
103 if (end != vma->vm_end) {
104 error = split_vma(mm, vma, end, 0);
105 if (error)
106 goto out;
107 }
108
Hugh Dickins836d5ff2005-09-03 15:54:53 -0700109success:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 /*
111 * vm_flags is protected by the mmap_sem held in write mode.
112 */
Prasanna Medae798c6e2005-06-21 17:14:36 -0700113 vma->vm_flags = new_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
115out:
116 if (error == -ENOMEM)
117 error = -EAGAIN;
118 return error;
119}
120
121/*
122 * Schedule all required I/O operations. Do not wait for completion.
123 */
124static long madvise_willneed(struct vm_area_struct * vma,
Prasanna Meda05b74382005-06-21 17:14:37 -0700125 struct vm_area_struct ** prev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 unsigned long start, unsigned long end)
127{
128 struct file *file = vma->vm_file;
129
Suzuki1bef4002005-10-11 08:29:06 -0700130 if (!file)
131 return -EBADF;
132
Nick Piggin70688e42008-04-28 02:13:02 -0700133 if (file->f_mapping->a_ops->get_xip_mem) {
Carsten Ottefe77ba62005-06-23 22:05:29 -0700134 /* no bad return value, but ignore advice */
135 return 0;
136 }
137
Prasanna Meda05b74382005-06-21 17:14:37 -0700138 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
140 if (end > vma->vm_end)
141 end = vma->vm_end;
142 end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
143
Wu Fengguangf7e839d2009-06-16 15:31:20 -0700144 force_page_cache_readahead(file->f_mapping, file, start, end - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 return 0;
146}
147
148/*
149 * Application no longer needs these pages. If the pages are dirty,
150 * it's OK to just throw them away. The app will be more careful about
151 * data it wants to keep. Be sure to free swap resources too. The
Fernando Luis Vazquez Cao7e6cbea2008-07-29 22:33:39 -0700152 * zap_page_range call sets things up for shrink_active_list to actually free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 * these pages later if no one else has touched them in the meantime,
154 * although we could add these pages to a global reuse list for
Fernando Luis Vazquez Cao7e6cbea2008-07-29 22:33:39 -0700155 * shrink_active_list to pick up before reclaiming other pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 *
157 * NB: This interface discards data rather than pushes it out to swap,
158 * as some implementations do. This has performance implications for
159 * applications like large transactional databases which want to discard
160 * pages in anonymous maps after committing to backing store the data
161 * that was kept in them. There is no reason to write this data out to
162 * the swap area if the application is discarding it.
163 *
164 * An interface that causes the system to free clean pages and flush
165 * dirty pages is already available as msync(MS_INVALIDATE).
166 */
167static long madvise_dontneed(struct vm_area_struct * vma,
Prasanna Meda05b74382005-06-21 17:14:37 -0700168 struct vm_area_struct ** prev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 unsigned long start, unsigned long end)
170{
Prasanna Meda05b74382005-06-21 17:14:37 -0700171 *prev = vma;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800172 if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 return -EINVAL;
174
175 if (unlikely(vma->vm_flags & VM_NONLINEAR)) {
176 struct zap_details details = {
177 .nonlinear_vma = vma,
178 .last_index = ULONG_MAX,
179 };
180 zap_page_range(vma, start, end - start, &details);
181 } else
182 zap_page_range(vma, start, end - start, NULL);
183 return 0;
184}
185
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800186/*
187 * Application wants to free up the pages and associated backing store.
188 * This is effectively punching a hole into the middle of a file.
189 *
190 * NOTE: Currently, only shmfs/tmpfs is supported for this operation.
191 * Other filesystems return -ENOSYS.
192 */
193static long madvise_remove(struct vm_area_struct *vma,
Nick Piggin00e9fa22007-03-16 13:38:10 -0800194 struct vm_area_struct **prev,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800195 unsigned long start, unsigned long end)
196{
197 struct address_space *mapping;
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700198 loff_t offset, endoff;
199 int error;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800200
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700201 *prev = NULL; /* tell sys_madvise we drop mmap_sem */
Nick Piggin00e9fa22007-03-16 13:38:10 -0800202
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800203 if (vma->vm_flags & (VM_LOCKED|VM_NONLINEAR|VM_HUGETLB))
204 return -EINVAL;
205
206 if (!vma->vm_file || !vma->vm_file->f_mapping
207 || !vma->vm_file->f_mapping->host) {
208 return -EINVAL;
209 }
210
Hugh Dickins69cf0fa2006-04-17 22:46:32 +0100211 if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE))
212 return -EACCES;
213
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800214 mapping = vma->vm_file->f_mapping;
215
216 offset = (loff_t)(start - vma->vm_start)
217 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
218 endoff = (loff_t)(end - vma->vm_start - 1)
219 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700220
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -0400221 /* vmtruncate_range needs to take i_mutex */
Nick Piggin0a27a142007-05-06 14:49:53 -0700222 up_read(&current->mm->mmap_sem);
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700223 error = vmtruncate_range(mapping->host, offset, endoff);
Nick Piggin0a27a142007-05-06 14:49:53 -0700224 down_read(&current->mm->mmap_sem);
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700225 return error;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800226}
227
Andi Kleen9893e492009-09-16 11:50:17 +0200228#ifdef CONFIG_MEMORY_FAILURE
229/*
230 * Error injection support for memory error handling.
231 */
Andi Kleenafcf9382009-12-16 12:20:00 +0100232static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end)
Andi Kleen9893e492009-09-16 11:50:17 +0200233{
234 int ret = 0;
235
236 if (!capable(CAP_SYS_ADMIN))
237 return -EPERM;
238 for (; start < end; start += PAGE_SIZE) {
239 struct page *p;
Andi Kleend15f1072009-12-16 12:20:00 +0100240 int ret = get_user_pages_fast(start, 1, 0, &p);
Andi Kleen9893e492009-09-16 11:50:17 +0200241 if (ret != 1)
242 return ret;
Andi Kleenafcf9382009-12-16 12:20:00 +0100243 if (bhv == MADV_SOFT_OFFLINE) {
244 printk(KERN_INFO "Soft offlining page %lx at %lx\n",
245 page_to_pfn(p), start);
246 ret = soft_offline_page(p, MF_COUNT_INCREASED);
247 if (ret)
248 break;
249 continue;
250 }
Andi Kleen9893e492009-09-16 11:50:17 +0200251 printk(KERN_INFO "Injecting memory failure for page %lx at %lx\n",
252 page_to_pfn(p), start);
253 /* Ignore return value for now */
Andi Kleen82ba0112009-12-16 12:19:57 +0100254 __memory_failure(page_to_pfn(p), 0, MF_COUNT_INCREASED);
Andi Kleen9893e492009-09-16 11:50:17 +0200255 }
256 return ret;
257}
258#endif
259
suzuki165cd402005-07-27 11:43:59 -0700260static long
261madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
262 unsigned long start, unsigned long end, int behavior)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 switch (behavior) {
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800265 case MADV_REMOVE:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700266 return madvise_remove(vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 case MADV_WILLNEED:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700268 return madvise_willneed(vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 case MADV_DONTNEED:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700270 return madvise_dontneed(vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 default:
Hugh Dickins3866ea92009-09-21 17:01:52 -0700272 return madvise_behavior(vma, prev, start, end, behavior);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274}
275
Nick Piggin75927af2009-06-16 15:32:38 -0700276static int
277madvise_behavior_valid(int behavior)
278{
279 switch (behavior) {
280 case MADV_DOFORK:
281 case MADV_DONTFORK:
282 case MADV_NORMAL:
283 case MADV_SEQUENTIAL:
284 case MADV_RANDOM:
285 case MADV_REMOVE:
286 case MADV_WILLNEED:
287 case MADV_DONTNEED:
Hugh Dickinsf8af4da2009-09-21 17:01:57 -0700288#ifdef CONFIG_KSM
289 case MADV_MERGEABLE:
290 case MADV_UNMERGEABLE:
291#endif
Andrea Arcangeli0af4e982011-01-13 15:46:55 -0800292#ifdef CONFIG_TRANSPARENT_HUGEPAGE
293 case MADV_HUGEPAGE:
Andrea Arcangelia664b2d2011-01-13 15:47:17 -0800294 case MADV_NOHUGEPAGE:
Andrea Arcangeli0af4e982011-01-13 15:46:55 -0800295#endif
Nick Piggin75927af2009-06-16 15:32:38 -0700296 return 1;
297
298 default:
299 return 0;
300 }
301}
Hugh Dickins3866ea92009-09-21 17:01:52 -0700302
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303/*
304 * The madvise(2) system call.
305 *
306 * Applications can use madvise() to advise the kernel how it should
307 * handle paging I/O in this VM area. The idea is to help the kernel
308 * use appropriate read-ahead and caching techniques. The information
309 * provided is advisory only, and can be safely disregarded by the
310 * kernel without affecting the correct operation of the application.
311 *
312 * behavior values:
313 * MADV_NORMAL - the default behavior is to read clusters. This
314 * results in some read-ahead and read-behind.
315 * MADV_RANDOM - the system should read the minimum amount of data
316 * on any access, since it is unlikely that the appli-
317 * cation will need more than what it asks for.
318 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
319 * once, so they can be aggressively read ahead, and
320 * can be freed soon after they are accessed.
321 * MADV_WILLNEED - the application is notifying the system to read
322 * some pages ahead.
323 * MADV_DONTNEED - the application is finished with the given range,
324 * so the kernel can free resources associated with it.
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800325 * MADV_REMOVE - the application wants to free up the given range of
326 * pages and associated backing store.
Hugh Dickins3866ea92009-09-21 17:01:52 -0700327 * MADV_DONTFORK - omit this area from child's address space when forking:
328 * typically, to avoid COWing pages pinned by get_user_pages().
329 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
Hugh Dickinsf8af4da2009-09-21 17:01:57 -0700330 * MADV_MERGEABLE - the application recommends that KSM try to merge pages in
331 * this area with pages of identical content from other such areas.
332 * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 *
334 * return values:
335 * zero - success
336 * -EINVAL - start + len < 0, start is not page-aligned,
337 * "behavior" is not a valid value, or application
338 * is attempting to release locked or shared pages.
339 * -ENOMEM - addresses in the specified range are not currently
340 * mapped, or are outside the AS of the process.
341 * -EIO - an I/O error occurred while paging in data.
342 * -EBADF - map exists, but area maps something that isn't a file.
343 * -EAGAIN - a kernel resource was temporarily unavailable.
344 */
Heiko Carstens3480b252009-01-14 14:14:16 +0100345SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
Prasanna Meda05b74382005-06-21 17:14:37 -0700347 unsigned long end, tmp;
348 struct vm_area_struct * vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 int unmapped_error = 0;
350 int error = -EINVAL;
Jason Baronf7977792007-07-15 23:38:21 -0700351 int write;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 size_t len;
353
Andi Kleen9893e492009-09-16 11:50:17 +0200354#ifdef CONFIG_MEMORY_FAILURE
Andi Kleenafcf9382009-12-16 12:20:00 +0100355 if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE)
356 return madvise_hwpoison(behavior, start, start+len_in);
Andi Kleen9893e492009-09-16 11:50:17 +0200357#endif
Nick Piggin75927af2009-06-16 15:32:38 -0700358 if (!madvise_behavior_valid(behavior))
359 return error;
360
Jason Baronf7977792007-07-15 23:38:21 -0700361 write = madvise_need_mmap_write(behavior);
362 if (write)
Nick Piggin0a27a142007-05-06 14:49:53 -0700363 down_write(&current->mm->mmap_sem);
364 else
365 down_read(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
367 if (start & ~PAGE_MASK)
368 goto out;
369 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
370
371 /* Check to see whether len was rounded up from small -ve to zero */
372 if (len_in && !len)
373 goto out;
374
375 end = start + len;
376 if (end < start)
377 goto out;
378
379 error = 0;
380 if (end == start)
381 goto out;
382
383 /*
384 * If the interval [start,end) covers some unmapped address
385 * ranges, just ignore them, but return -ENOMEM at the end.
Prasanna Meda05b74382005-06-21 17:14:37 -0700386 * - different from the way of handling in mlock etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 */
Prasanna Meda05b74382005-06-21 17:14:37 -0700388 vma = find_vma_prev(current->mm, start, &prev);
Hugh Dickins836d5ff2005-09-03 15:54:53 -0700389 if (vma && start > vma->vm_start)
390 prev = vma;
391
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 for (;;) {
393 /* Still start < end. */
394 error = -ENOMEM;
395 if (!vma)
396 goto out;
397
Prasanna Meda05b74382005-06-21 17:14:37 -0700398 /* Here start < (end|vma->vm_end). */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 if (start < vma->vm_start) {
400 unmapped_error = -ENOMEM;
401 start = vma->vm_start;
Prasanna Meda05b74382005-06-21 17:14:37 -0700402 if (start >= end)
403 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 }
405
Prasanna Meda05b74382005-06-21 17:14:37 -0700406 /* Here vma->vm_start <= start < (end|vma->vm_end) */
407 tmp = vma->vm_end;
408 if (end < tmp)
409 tmp = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Prasanna Meda05b74382005-06-21 17:14:37 -0700411 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
412 error = madvise_vma(vma, &prev, start, tmp, behavior);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 if (error)
414 goto out;
Prasanna Meda05b74382005-06-21 17:14:37 -0700415 start = tmp;
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700416 if (prev && start < prev->vm_end)
Prasanna Meda05b74382005-06-21 17:14:37 -0700417 start = prev->vm_end;
418 error = unmapped_error;
419 if (start >= end)
420 goto out;
Hugh Dickins90ed52e2007-03-29 01:20:38 -0700421 if (prev)
422 vma = prev->vm_next;
423 else /* madvise_remove dropped mmap_sem */
424 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426out:
Jason Baronf7977792007-07-15 23:38:21 -0700427 if (write)
Nick Piggin0a27a142007-05-06 14:49:53 -0700428 up_write(&current->mm->mmap_sem);
429 else
430 up_read(&current->mm->mmap_sem);
431
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 return error;
433}