4 * Copyright (C) 2002, Linus Torvalds
6 * 11Jan2003 Andrew Morton
10 #include <linux/kernel.h>
11 #include <linux/file.h>
14 #include <linux/pagemap.h>
15 #include <linux/backing-dev.h>
16 #include <linux/pagevec.h>
17 #include <linux/fadvise.h>
18 #include <linux/writeback.h>
19 #include <linux/syscalls.h>
20 #include <linux/swap.h>
22 #include <asm/unistd.h>
25 * POSIX_FADV_WILLNEED could set PG_Referenced, and POSIX_FADV_NOREUSE could
26 * deactivate the pages and clear PG_Referenced.
28 SYSCALL_DEFINE4(fadvise64_64, int, fd, loff_t, offset, loff_t, len, int, advice)
30 struct fd f = fdget(fd);
32 struct address_space *mapping;
33 struct backing_dev_info *bdi;
34 loff_t endbyte; /* inclusive */
37 unsigned long nrpages;
43 inode = file_inode(f.file);
44 if (S_ISFIFO(inode->i_mode)) {
49 mapping = f.file->f_mapping;
50 if (!mapping || len < 0) {
57 case POSIX_FADV_NORMAL:
58 case POSIX_FADV_RANDOM:
59 case POSIX_FADV_SEQUENTIAL:
60 case POSIX_FADV_WILLNEED:
61 case POSIX_FADV_NOREUSE:
62 case POSIX_FADV_DONTNEED:
63 /* no bad return value, but ignore advice */
72 * Careful about overflows. Len == 0 means "as much as possible". Use
73 * unsigned math because signed overflows are undefined and UBSan
76 endbyte = (u64)offset + (u64)len;
77 if (!len || endbyte < len)
80 endbyte--; /* inclusive */
82 bdi = inode_to_bdi(mapping->host);
85 case POSIX_FADV_NORMAL:
86 f.file->f_ra.ra_pages = bdi->ra_pages;
87 spin_lock(&f.file->f_lock);
88 f.file->f_mode &= ~FMODE_RANDOM;
89 spin_unlock(&f.file->f_lock);
91 case POSIX_FADV_RANDOM:
92 spin_lock(&f.file->f_lock);
93 f.file->f_mode |= FMODE_RANDOM;
94 spin_unlock(&f.file->f_lock);
96 case POSIX_FADV_SEQUENTIAL:
97 f.file->f_ra.ra_pages = bdi->ra_pages * 2;
98 spin_lock(&f.file->f_lock);
99 f.file->f_mode &= ~FMODE_RANDOM;
100 spin_unlock(&f.file->f_lock);
102 case POSIX_FADV_WILLNEED:
103 /* First and last PARTIAL page! */
104 start_index = offset >> PAGE_SHIFT;
105 end_index = endbyte >> PAGE_SHIFT;
107 /* Careful about overflow on the "+1" */
108 nrpages = end_index - start_index + 1;
113 * Ignore return value because fadvise() shall return
114 * success even if filesystem can't retrieve a hint,
116 force_page_cache_readahead(mapping, f.file, start_index,
119 case POSIX_FADV_NOREUSE:
121 case POSIX_FADV_DONTNEED:
122 if (!inode_write_congested(mapping->host))
123 __filemap_fdatawrite_range(mapping, offset, endbyte,
127 * First and last FULL page! Partial pages are deliberately
128 * preserved on the expectation that it is better to preserve
129 * needed memory than to discard unneeded memory.
131 start_index = (offset+(PAGE_SIZE-1)) >> PAGE_SHIFT;
132 end_index = (endbyte >> PAGE_SHIFT);
134 * The page at end_index will be inclusively discarded according
135 * by invalidate_mapping_pages(), so subtracting 1 from
136 * end_index means we will skip the last page. But if endbyte
137 * is page aligned or is at the end of file, we should not skip
138 * that page - discarding the last page is safe enough.
140 if ((endbyte & ~PAGE_MASK) != ~PAGE_MASK &&
141 endbyte != inode->i_size - 1) {
142 /* First page is tricky as 0 - 1 = -1, but pgoff_t
143 * is unsigned, so the end_index >= start_index
144 * check below would be true and we'll discard the whole
145 * file cache which is not what was asked.
153 if (end_index >= start_index) {
154 unsigned long count = invalidate_mapping_pages(mapping,
155 start_index, end_index);
158 * If fewer pages were invalidated than expected then
159 * it is possible that some of the pages were on
160 * a per-cpu pagevec for a remote CPU. Drain all
161 * pagevecs and try again.
163 if (count < (end_index - start_index + 1)) {
165 invalidate_mapping_pages(mapping, start_index,
178 #ifdef __ARCH_WANT_SYS_FADVISE64
180 SYSCALL_DEFINE4(fadvise64, int, fd, loff_t, offset, size_t, len, int, advice)
182 return sys_fadvise64_64(fd, offset, len, advice);