Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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> | |
05b74384 | 11 | #include <linux/mempolicy.h> |
1da177e4 | 12 | #include <linux/hugetlb.h> |
e8edc6e0 | 13 | #include <linux/sched.h> |
1da177e4 | 14 | |
0a27a14a NP |
15 | /* |
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. | |
19 | */ | |
20 | static int madvise_need_mmap_write(int behavior) | |
21 | { | |
22 | switch (behavior) { | |
23 | case MADV_REMOVE: | |
24 | case MADV_WILLNEED: | |
25 | case MADV_DONTNEED: | |
26 | return 0; | |
27 | default: | |
28 | /* be safe, default to 1. list exceptions explicitly */ | |
29 | return 1; | |
30 | } | |
31 | } | |
32 | ||
1da177e4 LT |
33 | /* |
34 | * We can potentially split a vm area into separate | |
35 | * areas, each area with its own behavior. | |
36 | */ | |
05b74384 PM |
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) | |
1da177e4 LT |
40 | { |
41 | struct mm_struct * mm = vma->vm_mm; | |
42 | int error = 0; | |
05b74384 | 43 | pgoff_t pgoff; |
f8225661 | 44 | int new_flags = vma->vm_flags; |
e798c6e8 PM |
45 | |
46 | switch (behavior) { | |
f8225661 MT |
47 | case MADV_NORMAL: |
48 | new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ; | |
49 | break; | |
e798c6e8 | 50 | case MADV_SEQUENTIAL: |
f8225661 | 51 | new_flags = (new_flags & ~VM_RAND_READ) | VM_SEQ_READ; |
e798c6e8 PM |
52 | break; |
53 | case MADV_RANDOM: | |
f8225661 | 54 | new_flags = (new_flags & ~VM_SEQ_READ) | VM_RAND_READ; |
e798c6e8 | 55 | break; |
f8225661 MT |
56 | case MADV_DONTFORK: |
57 | new_flags |= VM_DONTCOPY; | |
58 | break; | |
59 | case MADV_DOFORK: | |
60 | new_flags &= ~VM_DONTCOPY; | |
e798c6e8 PM |
61 | break; |
62 | } | |
63 | ||
05b74384 PM |
64 | if (new_flags == vma->vm_flags) { |
65 | *prev = vma; | |
836d5ffd | 66 | goto out; |
05b74384 PM |
67 | } |
68 | ||
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)); | |
72 | if (*prev) { | |
73 | vma = *prev; | |
74 | goto success; | |
75 | } | |
76 | ||
77 | *prev = vma; | |
1da177e4 LT |
78 | |
79 | if (start != vma->vm_start) { | |
80 | error = split_vma(mm, vma, start, 1); | |
81 | if (error) | |
82 | goto out; | |
83 | } | |
84 | ||
85 | if (end != vma->vm_end) { | |
86 | error = split_vma(mm, vma, end, 0); | |
87 | if (error) | |
88 | goto out; | |
89 | } | |
90 | ||
836d5ffd | 91 | success: |
1da177e4 LT |
92 | /* |
93 | * vm_flags is protected by the mmap_sem held in write mode. | |
94 | */ | |
e798c6e8 | 95 | vma->vm_flags = new_flags; |
1da177e4 LT |
96 | |
97 | out: | |
98 | if (error == -ENOMEM) | |
99 | error = -EAGAIN; | |
100 | return error; | |
101 | } | |
102 | ||
103 | /* | |
104 | * Schedule all required I/O operations. Do not wait for completion. | |
105 | */ | |
106 | static long madvise_willneed(struct vm_area_struct * vma, | |
05b74384 | 107 | struct vm_area_struct ** prev, |
1da177e4 LT |
108 | unsigned long start, unsigned long end) |
109 | { | |
110 | struct file *file = vma->vm_file; | |
111 | ||
1bef4003 S |
112 | if (!file) |
113 | return -EBADF; | |
114 | ||
70688e4d | 115 | if (file->f_mapping->a_ops->get_xip_mem) { |
fe77ba6f CO |
116 | /* no bad return value, but ignore advice */ |
117 | return 0; | |
118 | } | |
119 | ||
05b74384 | 120 | *prev = vma; |
1da177e4 LT |
121 | start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
122 | if (end > vma->vm_end) | |
123 | end = vma->vm_end; | |
124 | end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
125 | ||
f7e839dd | 126 | force_page_cache_readahead(file->f_mapping, file, start, end - start); |
1da177e4 LT |
127 | return 0; |
128 | } | |
129 | ||
130 | /* | |
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 | |
7e6cbea3 | 134 | * zap_page_range call sets things up for shrink_active_list to actually free |
1da177e4 LT |
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 | |
7e6cbea3 | 137 | * shrink_active_list to pick up before reclaiming other pages. |
1da177e4 LT |
138 | * |
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. | |
145 | * | |
146 | * An interface that causes the system to free clean pages and flush | |
147 | * dirty pages is already available as msync(MS_INVALIDATE). | |
148 | */ | |
149 | static long madvise_dontneed(struct vm_area_struct * vma, | |
05b74384 | 150 | struct vm_area_struct ** prev, |
1da177e4 LT |
151 | unsigned long start, unsigned long end) |
152 | { | |
05b74384 | 153 | *prev = vma; |
6aab341e | 154 | if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP)) |
1da177e4 LT |
155 | return -EINVAL; |
156 | ||
157 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) { | |
158 | struct zap_details details = { | |
159 | .nonlinear_vma = vma, | |
160 | .last_index = ULONG_MAX, | |
161 | }; | |
162 | zap_page_range(vma, start, end - start, &details); | |
163 | } else | |
164 | zap_page_range(vma, start, end - start, NULL); | |
165 | return 0; | |
166 | } | |
167 | ||
f6b3ec23 BP |
168 | /* |
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. | |
171 | * | |
172 | * NOTE: Currently, only shmfs/tmpfs is supported for this operation. | |
173 | * Other filesystems return -ENOSYS. | |
174 | */ | |
175 | static long madvise_remove(struct vm_area_struct *vma, | |
00e9fa2d | 176 | struct vm_area_struct **prev, |
f6b3ec23 BP |
177 | unsigned long start, unsigned long end) |
178 | { | |
179 | struct address_space *mapping; | |
90ed52eb HD |
180 | loff_t offset, endoff; |
181 | int error; | |
f6b3ec23 | 182 | |
90ed52eb | 183 | *prev = NULL; /* tell sys_madvise we drop mmap_sem */ |
00e9fa2d | 184 | |
f6b3ec23 BP |
185 | if (vma->vm_flags & (VM_LOCKED|VM_NONLINEAR|VM_HUGETLB)) |
186 | return -EINVAL; | |
187 | ||
188 | if (!vma->vm_file || !vma->vm_file->f_mapping | |
189 | || !vma->vm_file->f_mapping->host) { | |
190 | return -EINVAL; | |
191 | } | |
192 | ||
69cf0fac HD |
193 | if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE)) |
194 | return -EACCES; | |
195 | ||
f6b3ec23 BP |
196 | mapping = vma->vm_file->f_mapping; |
197 | ||
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); | |
90ed52eb HD |
202 | |
203 | /* vmtruncate_range needs to take i_mutex and i_alloc_sem */ | |
0a27a14a | 204 | up_read(¤t->mm->mmap_sem); |
90ed52eb | 205 | error = vmtruncate_range(mapping->host, offset, endoff); |
0a27a14a | 206 | down_read(¤t->mm->mmap_sem); |
90ed52eb | 207 | return error; |
f6b3ec23 BP |
208 | } |
209 | ||
165cd402 | 210 | static long |
211 | madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev, | |
212 | unsigned long start, unsigned long end, int behavior) | |
1da177e4 | 213 | { |
1bef4003 | 214 | long error; |
165cd402 | 215 | |
1da177e4 | 216 | switch (behavior) { |
f8225661 MT |
217 | case MADV_DOFORK: |
218 | if (vma->vm_flags & VM_IO) { | |
219 | error = -EINVAL; | |
220 | break; | |
221 | } | |
222 | case MADV_DONTFORK: | |
1da177e4 LT |
223 | case MADV_NORMAL: |
224 | case MADV_SEQUENTIAL: | |
225 | case MADV_RANDOM: | |
05b74384 | 226 | error = madvise_behavior(vma, prev, start, end, behavior); |
1da177e4 | 227 | break; |
f6b3ec23 | 228 | case MADV_REMOVE: |
00e9fa2d | 229 | error = madvise_remove(vma, prev, start, end); |
f6b3ec23 | 230 | break; |
1da177e4 LT |
231 | |
232 | case MADV_WILLNEED: | |
05b74384 | 233 | error = madvise_willneed(vma, prev, start, end); |
1da177e4 LT |
234 | break; |
235 | ||
236 | case MADV_DONTNEED: | |
05b74384 | 237 | error = madvise_dontneed(vma, prev, start, end); |
1da177e4 LT |
238 | break; |
239 | ||
240 | default: | |
75927af8 | 241 | BUG(); |
1da177e4 LT |
242 | break; |
243 | } | |
1da177e4 LT |
244 | return error; |
245 | } | |
246 | ||
75927af8 NP |
247 | static int |
248 | madvise_behavior_valid(int behavior) | |
249 | { | |
250 | switch (behavior) { | |
251 | case MADV_DOFORK: | |
252 | case MADV_DONTFORK: | |
253 | case MADV_NORMAL: | |
254 | case MADV_SEQUENTIAL: | |
255 | case MADV_RANDOM: | |
256 | case MADV_REMOVE: | |
257 | case MADV_WILLNEED: | |
258 | case MADV_DONTNEED: | |
259 | return 1; | |
260 | ||
261 | default: | |
262 | return 0; | |
263 | } | |
264 | } | |
1da177e4 LT |
265 | /* |
266 | * The madvise(2) system call. | |
267 | * | |
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. | |
273 | * | |
274 | * behavior values: | |
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 | |
284 | * some pages ahead. | |
285 | * MADV_DONTNEED - the application is finished with the given range, | |
286 | * so the kernel can free resources associated with it. | |
f6b3ec23 BP |
287 | * MADV_REMOVE - the application wants to free up the given range of |
288 | * pages and associated backing store. | |
1da177e4 LT |
289 | * |
290 | * return values: | |
291 | * zero - success | |
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. | |
300 | */ | |
3480b257 | 301 | SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior) |
1da177e4 | 302 | { |
05b74384 PM |
303 | unsigned long end, tmp; |
304 | struct vm_area_struct * vma, *prev; | |
1da177e4 LT |
305 | int unmapped_error = 0; |
306 | int error = -EINVAL; | |
f7977793 | 307 | int write; |
1da177e4 LT |
308 | size_t len; |
309 | ||
75927af8 NP |
310 | if (!madvise_behavior_valid(behavior)) |
311 | return error; | |
312 | ||
f7977793 JB |
313 | write = madvise_need_mmap_write(behavior); |
314 | if (write) | |
0a27a14a NP |
315 | down_write(¤t->mm->mmap_sem); |
316 | else | |
317 | down_read(¤t->mm->mmap_sem); | |
1da177e4 LT |
318 | |
319 | if (start & ~PAGE_MASK) | |
320 | goto out; | |
321 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; | |
322 | ||
323 | /* Check to see whether len was rounded up from small -ve to zero */ | |
324 | if (len_in && !len) | |
325 | goto out; | |
326 | ||
327 | end = start + len; | |
328 | if (end < start) | |
329 | goto out; | |
330 | ||
331 | error = 0; | |
332 | if (end == start) | |
333 | goto out; | |
334 | ||
335 | /* | |
336 | * If the interval [start,end) covers some unmapped address | |
337 | * ranges, just ignore them, but return -ENOMEM at the end. | |
05b74384 | 338 | * - different from the way of handling in mlock etc. |
1da177e4 | 339 | */ |
05b74384 | 340 | vma = find_vma_prev(current->mm, start, &prev); |
836d5ffd HD |
341 | if (vma && start > vma->vm_start) |
342 | prev = vma; | |
343 | ||
1da177e4 LT |
344 | for (;;) { |
345 | /* Still start < end. */ | |
346 | error = -ENOMEM; | |
347 | if (!vma) | |
348 | goto out; | |
349 | ||
05b74384 | 350 | /* Here start < (end|vma->vm_end). */ |
1da177e4 LT |
351 | if (start < vma->vm_start) { |
352 | unmapped_error = -ENOMEM; | |
353 | start = vma->vm_start; | |
05b74384 PM |
354 | if (start >= end) |
355 | goto out; | |
1da177e4 LT |
356 | } |
357 | ||
05b74384 PM |
358 | /* Here vma->vm_start <= start < (end|vma->vm_end) */ |
359 | tmp = vma->vm_end; | |
360 | if (end < tmp) | |
361 | tmp = end; | |
1da177e4 | 362 | |
05b74384 PM |
363 | /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ |
364 | error = madvise_vma(vma, &prev, start, tmp, behavior); | |
1da177e4 LT |
365 | if (error) |
366 | goto out; | |
05b74384 | 367 | start = tmp; |
90ed52eb | 368 | if (prev && start < prev->vm_end) |
05b74384 PM |
369 | start = prev->vm_end; |
370 | error = unmapped_error; | |
371 | if (start >= end) | |
372 | goto out; | |
90ed52eb HD |
373 | if (prev) |
374 | vma = prev->vm_next; | |
375 | else /* madvise_remove dropped mmap_sem */ | |
376 | vma = find_vma(current->mm, start); | |
1da177e4 | 377 | } |
1da177e4 | 378 | out: |
f7977793 | 379 | if (write) |
0a27a14a NP |
380 | up_write(¤t->mm->mmap_sem); |
381 | else | |
382 | up_read(¤t->mm->mmap_sem); | |
383 | ||
1da177e4 LT |
384 | return error; |
385 | } |