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> | |
11 | #include <linux/hugetlb.h> | |
12 | ||
13 | /* | |
14 | * We can potentially split a vm area into separate | |
15 | * areas, each area with its own behavior. | |
16 | */ | |
17 | static long madvise_behavior(struct vm_area_struct * vma, unsigned long start, | |
18 | unsigned long end, int behavior) | |
19 | { | |
20 | struct mm_struct * mm = vma->vm_mm; | |
21 | int error = 0; | |
22 | ||
23 | if (start != vma->vm_start) { | |
24 | error = split_vma(mm, vma, start, 1); | |
25 | if (error) | |
26 | goto out; | |
27 | } | |
28 | ||
29 | if (end != vma->vm_end) { | |
30 | error = split_vma(mm, vma, end, 0); | |
31 | if (error) | |
32 | goto out; | |
33 | } | |
34 | ||
35 | /* | |
36 | * vm_flags is protected by the mmap_sem held in write mode. | |
37 | */ | |
38 | VM_ClearReadHint(vma); | |
39 | ||
40 | switch (behavior) { | |
41 | case MADV_SEQUENTIAL: | |
42 | vma->vm_flags |= VM_SEQ_READ; | |
43 | break; | |
44 | case MADV_RANDOM: | |
45 | vma->vm_flags |= VM_RAND_READ; | |
46 | break; | |
47 | default: | |
48 | break; | |
49 | } | |
50 | ||
51 | out: | |
52 | if (error == -ENOMEM) | |
53 | error = -EAGAIN; | |
54 | return error; | |
55 | } | |
56 | ||
57 | /* | |
58 | * Schedule all required I/O operations. Do not wait for completion. | |
59 | */ | |
60 | static long madvise_willneed(struct vm_area_struct * vma, | |
61 | unsigned long start, unsigned long end) | |
62 | { | |
63 | struct file *file = vma->vm_file; | |
64 | ||
65 | if (!file) | |
66 | return -EBADF; | |
67 | ||
68 | start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
69 | if (end > vma->vm_end) | |
70 | end = vma->vm_end; | |
71 | end = ((end - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; | |
72 | ||
73 | force_page_cache_readahead(file->f_mapping, | |
74 | file, start, max_sane_readahead(end - start)); | |
75 | return 0; | |
76 | } | |
77 | ||
78 | /* | |
79 | * Application no longer needs these pages. If the pages are dirty, | |
80 | * it's OK to just throw them away. The app will be more careful about | |
81 | * data it wants to keep. Be sure to free swap resources too. The | |
82 | * zap_page_range call sets things up for refill_inactive to actually free | |
83 | * these pages later if no one else has touched them in the meantime, | |
84 | * although we could add these pages to a global reuse list for | |
85 | * refill_inactive to pick up before reclaiming other pages. | |
86 | * | |
87 | * NB: This interface discards data rather than pushes it out to swap, | |
88 | * as some implementations do. This has performance implications for | |
89 | * applications like large transactional databases which want to discard | |
90 | * pages in anonymous maps after committing to backing store the data | |
91 | * that was kept in them. There is no reason to write this data out to | |
92 | * the swap area if the application is discarding it. | |
93 | * | |
94 | * An interface that causes the system to free clean pages and flush | |
95 | * dirty pages is already available as msync(MS_INVALIDATE). | |
96 | */ | |
97 | static long madvise_dontneed(struct vm_area_struct * vma, | |
98 | unsigned long start, unsigned long end) | |
99 | { | |
100 | if ((vma->vm_flags & VM_LOCKED) || is_vm_hugetlb_page(vma)) | |
101 | return -EINVAL; | |
102 | ||
103 | if (unlikely(vma->vm_flags & VM_NONLINEAR)) { | |
104 | struct zap_details details = { | |
105 | .nonlinear_vma = vma, | |
106 | .last_index = ULONG_MAX, | |
107 | }; | |
108 | zap_page_range(vma, start, end - start, &details); | |
109 | } else | |
110 | zap_page_range(vma, start, end - start, NULL); | |
111 | return 0; | |
112 | } | |
113 | ||
114 | static long madvise_vma(struct vm_area_struct * vma, unsigned long start, | |
115 | unsigned long end, int behavior) | |
116 | { | |
117 | long error = -EBADF; | |
118 | ||
119 | switch (behavior) { | |
120 | case MADV_NORMAL: | |
121 | case MADV_SEQUENTIAL: | |
122 | case MADV_RANDOM: | |
123 | error = madvise_behavior(vma, start, end, behavior); | |
124 | break; | |
125 | ||
126 | case MADV_WILLNEED: | |
127 | error = madvise_willneed(vma, start, end); | |
128 | break; | |
129 | ||
130 | case MADV_DONTNEED: | |
131 | error = madvise_dontneed(vma, start, end); | |
132 | break; | |
133 | ||
134 | default: | |
135 | error = -EINVAL; | |
136 | break; | |
137 | } | |
138 | ||
139 | return error; | |
140 | } | |
141 | ||
142 | /* | |
143 | * The madvise(2) system call. | |
144 | * | |
145 | * Applications can use madvise() to advise the kernel how it should | |
146 | * handle paging I/O in this VM area. The idea is to help the kernel | |
147 | * use appropriate read-ahead and caching techniques. The information | |
148 | * provided is advisory only, and can be safely disregarded by the | |
149 | * kernel without affecting the correct operation of the application. | |
150 | * | |
151 | * behavior values: | |
152 | * MADV_NORMAL - the default behavior is to read clusters. This | |
153 | * results in some read-ahead and read-behind. | |
154 | * MADV_RANDOM - the system should read the minimum amount of data | |
155 | * on any access, since it is unlikely that the appli- | |
156 | * cation will need more than what it asks for. | |
157 | * MADV_SEQUENTIAL - pages in the given range will probably be accessed | |
158 | * once, so they can be aggressively read ahead, and | |
159 | * can be freed soon after they are accessed. | |
160 | * MADV_WILLNEED - the application is notifying the system to read | |
161 | * some pages ahead. | |
162 | * MADV_DONTNEED - the application is finished with the given range, | |
163 | * so the kernel can free resources associated with it. | |
164 | * | |
165 | * return values: | |
166 | * zero - success | |
167 | * -EINVAL - start + len < 0, start is not page-aligned, | |
168 | * "behavior" is not a valid value, or application | |
169 | * is attempting to release locked or shared pages. | |
170 | * -ENOMEM - addresses in the specified range are not currently | |
171 | * mapped, or are outside the AS of the process. | |
172 | * -EIO - an I/O error occurred while paging in data. | |
173 | * -EBADF - map exists, but area maps something that isn't a file. | |
174 | * -EAGAIN - a kernel resource was temporarily unavailable. | |
175 | */ | |
176 | asmlinkage long sys_madvise(unsigned long start, size_t len_in, int behavior) | |
177 | { | |
178 | unsigned long end; | |
179 | struct vm_area_struct * vma; | |
180 | int unmapped_error = 0; | |
181 | int error = -EINVAL; | |
182 | size_t len; | |
183 | ||
184 | down_write(¤t->mm->mmap_sem); | |
185 | ||
186 | if (start & ~PAGE_MASK) | |
187 | goto out; | |
188 | len = (len_in + ~PAGE_MASK) & PAGE_MASK; | |
189 | ||
190 | /* Check to see whether len was rounded up from small -ve to zero */ | |
191 | if (len_in && !len) | |
192 | goto out; | |
193 | ||
194 | end = start + len; | |
195 | if (end < start) | |
196 | goto out; | |
197 | ||
198 | error = 0; | |
199 | if (end == start) | |
200 | goto out; | |
201 | ||
202 | /* | |
203 | * If the interval [start,end) covers some unmapped address | |
204 | * ranges, just ignore them, but return -ENOMEM at the end. | |
205 | */ | |
206 | vma = find_vma(current->mm, start); | |
207 | for (;;) { | |
208 | /* Still start < end. */ | |
209 | error = -ENOMEM; | |
210 | if (!vma) | |
211 | goto out; | |
212 | ||
213 | /* Here start < vma->vm_end. */ | |
214 | if (start < vma->vm_start) { | |
215 | unmapped_error = -ENOMEM; | |
216 | start = vma->vm_start; | |
217 | } | |
218 | ||
219 | /* Here vma->vm_start <= start < vma->vm_end. */ | |
220 | if (end <= vma->vm_end) { | |
221 | if (start < end) { | |
222 | error = madvise_vma(vma, start, end, | |
223 | behavior); | |
224 | if (error) | |
225 | goto out; | |
226 | } | |
227 | error = unmapped_error; | |
228 | goto out; | |
229 | } | |
230 | ||
231 | /* Here vma->vm_start <= start < vma->vm_end < end. */ | |
232 | error = madvise_vma(vma, start, vma->vm_end, behavior); | |
233 | if (error) | |
234 | goto out; | |
235 | start = vma->vm_end; | |
236 | vma = vma->vm_next; | |
237 | } | |
238 | ||
239 | out: | |
240 | up_write(¤t->mm->mmap_sem); | |
241 | return error; | |
242 | } |