]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * mm/fadvise.c | |
4 | * | |
5 | * Copyright (C) 2002, Linus Torvalds | |
6 | * | |
e1f8e874 | 7 | * 11Jan2003 Andrew Morton |
1da177e4 LT |
8 | * Initial version. |
9 | */ | |
10 | ||
11 | #include <linux/kernel.h> | |
12 | #include <linux/file.h> | |
13 | #include <linux/fs.h> | |
14 | #include <linux/mm.h> | |
15 | #include <linux/pagemap.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/pagevec.h> | |
18 | #include <linux/fadvise.h> | |
ebcf28e1 | 19 | #include <linux/writeback.h> |
1da177e4 | 20 | #include <linux/syscalls.h> |
67d46b29 | 21 | #include <linux/swap.h> |
1da177e4 LT |
22 | |
23 | #include <asm/unistd.h> | |
24 | ||
25 | /* | |
26 | * POSIX_FADV_WILLNEED could set PG_Referenced, and POSIX_FADV_NOREUSE could | |
27 | * deactivate the pages and clear PG_Referenced. | |
28 | */ | |
4a0fd5bf | 29 | SYSCALL_DEFINE4(fadvise64_64, int, fd, loff_t, offset, loff_t, len, int, advice) |
1da177e4 | 30 | { |
2903ff01 | 31 | struct fd f = fdget(fd); |
e748dcd0 | 32 | struct inode *inode; |
1da177e4 LT |
33 | struct address_space *mapping; |
34 | struct backing_dev_info *bdi; | |
ebcf28e1 | 35 | loff_t endbyte; /* inclusive */ |
1da177e4 LT |
36 | pgoff_t start_index; |
37 | pgoff_t end_index; | |
38 | unsigned long nrpages; | |
39 | int ret = 0; | |
40 | ||
2903ff01 | 41 | if (!f.file) |
1da177e4 LT |
42 | return -EBADF; |
43 | ||
e748dcd0 MW |
44 | inode = file_inode(f.file); |
45 | if (S_ISFIFO(inode->i_mode)) { | |
87ba81db VB |
46 | ret = -ESPIPE; |
47 | goto out; | |
48 | } | |
49 | ||
2903ff01 | 50 | mapping = f.file->f_mapping; |
1da177e4 LT |
51 | if (!mapping || len < 0) { |
52 | ret = -EINVAL; | |
53 | goto out; | |
54 | } | |
55 | ||
3a77d214 SB |
56 | bdi = inode_to_bdi(mapping->host); |
57 | ||
58 | if (IS_DAX(inode) || (bdi == &noop_backing_dev_info)) { | |
b5beb1ca MY |
59 | switch (advice) { |
60 | case POSIX_FADV_NORMAL: | |
61 | case POSIX_FADV_RANDOM: | |
62 | case POSIX_FADV_SEQUENTIAL: | |
63 | case POSIX_FADV_WILLNEED: | |
64 | case POSIX_FADV_NOREUSE: | |
65 | case POSIX_FADV_DONTNEED: | |
66 | /* no bad return value, but ignore advice */ | |
67 | break; | |
68 | default: | |
69 | ret = -EINVAL; | |
70 | } | |
fe77ba6f | 71 | goto out; |
b5beb1ca | 72 | } |
fe77ba6f | 73 | |
b899d03b AR |
74 | /* |
75 | * Careful about overflows. Len == 0 means "as much as possible". Use | |
76 | * unsigned math because signed overflows are undefined and UBSan | |
77 | * complains. | |
78 | */ | |
79 | endbyte = (u64)offset + (u64)len; | |
1da177e4 LT |
80 | if (!len || endbyte < len) |
81 | endbyte = -1; | |
ebcf28e1 AM |
82 | else |
83 | endbyte--; /* inclusive */ | |
1da177e4 | 84 | |
1da177e4 LT |
85 | switch (advice) { |
86 | case POSIX_FADV_NORMAL: | |
2903ff01 AV |
87 | f.file->f_ra.ra_pages = bdi->ra_pages; |
88 | spin_lock(&f.file->f_lock); | |
89 | f.file->f_mode &= ~FMODE_RANDOM; | |
90 | spin_unlock(&f.file->f_lock); | |
1da177e4 LT |
91 | break; |
92 | case POSIX_FADV_RANDOM: | |
2903ff01 AV |
93 | spin_lock(&f.file->f_lock); |
94 | f.file->f_mode |= FMODE_RANDOM; | |
95 | spin_unlock(&f.file->f_lock); | |
1da177e4 LT |
96 | break; |
97 | case POSIX_FADV_SEQUENTIAL: | |
2903ff01 AV |
98 | f.file->f_ra.ra_pages = bdi->ra_pages * 2; |
99 | spin_lock(&f.file->f_lock); | |
100 | f.file->f_mode &= ~FMODE_RANDOM; | |
101 | spin_unlock(&f.file->f_lock); | |
1da177e4 LT |
102 | break; |
103 | case POSIX_FADV_WILLNEED: | |
1da177e4 | 104 | /* First and last PARTIAL page! */ |
09cbfeaf KS |
105 | start_index = offset >> PAGE_SHIFT; |
106 | end_index = endbyte >> PAGE_SHIFT; | |
1da177e4 LT |
107 | |
108 | /* Careful about overflow on the "+1" */ | |
109 | nrpages = end_index - start_index + 1; | |
110 | if (!nrpages) | |
111 | nrpages = ~0UL; | |
3d3727cd KM |
112 | |
113 | /* | |
114 | * Ignore return value because fadvise() shall return | |
115 | * success even if filesystem can't retrieve a hint, | |
116 | */ | |
2903ff01 | 117 | force_page_cache_readahead(mapping, f.file, start_index, |
3d3727cd | 118 | nrpages); |
1da177e4 | 119 | break; |
60c371bc AM |
120 | case POSIX_FADV_NOREUSE: |
121 | break; | |
1da177e4 | 122 | case POSIX_FADV_DONTNEED: |
703c2708 | 123 | if (!inode_write_congested(mapping->host)) |
ad8a1b55 SB |
124 | __filemap_fdatawrite_range(mapping, offset, endbyte, |
125 | WB_SYNC_NONE); | |
1da177e4 | 126 | |
441c228f MG |
127 | /* |
128 | * First and last FULL page! Partial pages are deliberately | |
129 | * preserved on the expectation that it is better to preserve | |
130 | * needed memory than to discard unneeded memory. | |
131 | */ | |
09cbfeaf KS |
132 | start_index = (offset+(PAGE_SIZE-1)) >> PAGE_SHIFT; |
133 | end_index = (endbyte >> PAGE_SHIFT); | |
5674b344 | 134 | /* |
135 | * The page at end_index will be inclusively discarded according | |
136 | * by invalidate_mapping_pages(), so subtracting 1 from | |
137 | * end_index means we will skip the last page. But if endbyte | |
138 | * is page aligned or is at the end of file, we should not skip | |
139 | * that page - discarding the last page is safe enough. | |
140 | */ | |
141 | if ((endbyte & ~PAGE_MASK) != ~PAGE_MASK && | |
142 | endbyte != inode->i_size - 1) { | |
18aba41c OD |
143 | /* First page is tricky as 0 - 1 = -1, but pgoff_t |
144 | * is unsigned, so the end_index >= start_index | |
145 | * check below would be true and we'll discard the whole | |
146 | * file cache which is not what was asked. | |
147 | */ | |
148 | if (end_index == 0) | |
149 | break; | |
150 | ||
151 | end_index--; | |
152 | } | |
1da177e4 | 153 | |
67d46b29 | 154 | if (end_index >= start_index) { |
4dd72b4a JW |
155 | unsigned long count; |
156 | ||
157 | /* | |
158 | * It's common to FADV_DONTNEED right after | |
159 | * the read or write that instantiates the | |
160 | * pages, in which case there will be some | |
161 | * sitting on the local LRU cache. Try to | |
162 | * avoid the expensive remote drain and the | |
163 | * second cache tree walk below by flushing | |
164 | * them out right away. | |
165 | */ | |
166 | lru_add_drain(); | |
167 | ||
168 | count = invalidate_mapping_pages(mapping, | |
67d46b29 MG |
169 | start_index, end_index); |
170 | ||
171 | /* | |
172 | * If fewer pages were invalidated than expected then | |
173 | * it is possible that some of the pages were on | |
174 | * a per-cpu pagevec for a remote CPU. Drain all | |
175 | * pagevecs and try again. | |
176 | */ | |
177 | if (count < (end_index - start_index + 1)) { | |
178 | lru_add_drain_all(); | |
179 | invalidate_mapping_pages(mapping, start_index, | |
ebcf28e1 | 180 | end_index); |
67d46b29 MG |
181 | } |
182 | } | |
ebcf28e1 | 183 | break; |
1da177e4 LT |
184 | default: |
185 | ret = -EINVAL; | |
186 | } | |
187 | out: | |
2903ff01 | 188 | fdput(f); |
1da177e4 LT |
189 | return ret; |
190 | } | |
191 | ||
192 | #ifdef __ARCH_WANT_SYS_FADVISE64 | |
193 | ||
4a0fd5bf | 194 | SYSCALL_DEFINE4(fadvise64, int, fd, loff_t, offset, size_t, len, int, advice) |
1da177e4 LT |
195 | { |
196 | return sys_fadvise64_64(fd, offset, len, advice); | |
197 | } | |
198 | ||
199 | #endif |