blob: 3f5f68ad57080f6909e8fcc91a516d946da5d519 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * mm/fadvise.c
4 *
5 * Copyright (C) 2002, Linus Torvalds
6 *
Francois Camie1f8e872008-10-15 22:01:59 -07007 * 11Jan2003 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * 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>
Andrew Mortonebcf28e2006-03-24 03:18:04 -080019#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/syscalls.h>
Mel Gorman67d46b22013-02-22 16:35:59 -080021#include <linux/swap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
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 */
Al Viro4a0fd5b2013-01-21 15:16:58 -050029SYSCALL_DEFINE4(fadvise64_64, int, fd, loff_t, offset, loff_t, len, int, advice)
Linus Torvalds1da177e2005-04-16 15:20:36 -070030{
Al Viro2903ff02012-08-28 12:52:22 -040031 struct fd f = fdget(fd);
Matthew Wilcoxe748dcd2015-02-16 15:59:12 -080032 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 struct address_space *mapping;
34 struct backing_dev_info *bdi;
Andrew Mortonebcf28e2006-03-24 03:18:04 -080035 loff_t endbyte; /* inclusive */
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 pgoff_t start_index;
37 pgoff_t end_index;
38 unsigned long nrpages;
39 int ret = 0;
40
Al Viro2903ff02012-08-28 12:52:22 -040041 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 return -EBADF;
43
Matthew Wilcoxe748dcd2015-02-16 15:59:12 -080044 inode = file_inode(f.file);
45 if (S_ISFIFO(inode->i_mode)) {
Valentine Barshak87ba81d2006-01-08 01:03:44 -080046 ret = -ESPIPE;
47 goto out;
48 }
49
Al Viro2903ff02012-08-28 12:52:22 -040050 mapping = f.file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051 if (!mapping || len < 0) {
52 ret = -EINVAL;
53 goto out;
54 }
55
Shakeel Butt3a77d212017-09-08 16:13:05 -070056 bdi = inode_to_bdi(mapping->host);
57
58 if (IS_DAX(inode) || (bdi == &noop_backing_dev_info)) {
Masatake YAMATOb5beb1c2008-02-04 22:29:31 -080059 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 }
Carsten Ottefe77ba62005-06-23 22:05:29 -070071 goto out;
Masatake YAMATOb5beb1c2008-02-04 22:29:31 -080072 }
Carsten Ottefe77ba62005-06-23 22:05:29 -070073
Andrey Ryabinin45704032018-08-17 15:46:57 -070074 /*
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 if (!len || endbyte < len)
81 endbyte = -1;
Andrew Mortonebcf28e2006-03-24 03:18:04 -080082 else
83 endbyte--; /* inclusive */
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 switch (advice) {
86 case POSIX_FADV_NORMAL:
Al Viro2903ff02012-08-28 12:52:22 -040087 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 break;
92 case POSIX_FADV_RANDOM:
Al Viro2903ff02012-08-28 12:52:22 -040093 spin_lock(&f.file->f_lock);
94 f.file->f_mode |= FMODE_RANDOM;
95 spin_unlock(&f.file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 break;
97 case POSIX_FADV_SEQUENTIAL:
Al Viro2903ff02012-08-28 12:52:22 -040098 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 break;
103 case POSIX_FADV_WILLNEED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 /* First and last PARTIAL page! */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300105 start_index = offset >> PAGE_SHIFT;
106 end_index = endbyte >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
108 /* Careful about overflow on the "+1" */
109 nrpages = end_index - start_index + 1;
110 if (!nrpages)
111 nrpages = ~0UL;
KOSAKI Motohiro3d3727c2012-07-31 16:42:50 -0700112
113 /*
114 * Ignore return value because fadvise() shall return
115 * success even if filesystem can't retrieve a hint,
116 */
Al Viro2903ff02012-08-28 12:52:22 -0400117 force_page_cache_readahead(mapping, f.file, start_index,
KOSAKI Motohiro3d3727c2012-07-31 16:42:50 -0700118 nrpages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 break;
Andrew Morton60c371b2006-08-05 12:14:25 -0700120 case POSIX_FADV_NOREUSE:
121 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 case POSIX_FADV_DONTNEED:
Tejun Heo703c2702015-05-22 17:13:44 -0400123 if (!inode_write_congested(mapping->host))
Shawn Bohrerad8a1b52012-01-10 15:07:35 -0800124 __filemap_fdatawrite_range(mapping, offset, endbyte,
125 WB_SYNC_NONE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Mel Gorman441c2282014-12-12 16:56:33 -0800127 /*
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 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300132 start_index = (offset+(PAGE_SIZE-1)) >> PAGE_SHIFT;
133 end_index = (endbyte >> PAGE_SHIFT);
shidao.ytt3b1d9622018-01-31 16:19:55 -0800134 /*
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) {
Oleg Drokin18aba412016-06-08 15:33:59 -0700143 /* 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 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Mel Gorman67d46b22013-02-22 16:35:59 -0800154 if (end_index >= start_index) {
Johannes Weiner4dd72b42016-12-19 16:23:03 -0800155 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,
Mel Gorman67d46b22013-02-22 16:35:59 -0800169 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,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800180 end_index);
Mel Gorman67d46b22013-02-22 16:35:59 -0800181 }
182 }
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800183 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 default:
185 ret = -EINVAL;
186 }
187out:
Al Viro2903ff02012-08-28 12:52:22 -0400188 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 return ret;
190}
191
192#ifdef __ARCH_WANT_SYS_FADVISE64
193
Al Viro4a0fd5b2013-01-21 15:16:58 -0500194SYSCALL_DEFINE4(fadvise64, int, fd, loff_t, offset, size_t, len, int, advice)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195{
196 return sys_fadvise64_64(fd, offset, len, advice);
197}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
199#endif