blob: 767887f5f3bfd79b493a731dbc467d96545f8494 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 /* Careful about overflows. Len == 0 means "as much as possible" */
75 endbyte = offset + len;
76 if (!len || endbyte < len)
77 endbyte = -1;
Andrew Mortonebcf28e2006-03-24 03:18:04 -080078 else
79 endbyte--; /* inclusive */
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 switch (advice) {
82 case POSIX_FADV_NORMAL:
Al Viro2903ff02012-08-28 12:52:22 -040083 f.file->f_ra.ra_pages = bdi->ra_pages;
84 spin_lock(&f.file->f_lock);
85 f.file->f_mode &= ~FMODE_RANDOM;
86 spin_unlock(&f.file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 break;
88 case POSIX_FADV_RANDOM:
Al Viro2903ff02012-08-28 12:52:22 -040089 spin_lock(&f.file->f_lock);
90 f.file->f_mode |= FMODE_RANDOM;
91 spin_unlock(&f.file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 break;
93 case POSIX_FADV_SEQUENTIAL:
Al Viro2903ff02012-08-28 12:52:22 -040094 f.file->f_ra.ra_pages = bdi->ra_pages * 2;
95 spin_lock(&f.file->f_lock);
96 f.file->f_mode &= ~FMODE_RANDOM;
97 spin_unlock(&f.file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 break;
99 case POSIX_FADV_WILLNEED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 /* First and last PARTIAL page! */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300101 start_index = offset >> PAGE_SHIFT;
102 end_index = endbyte >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104 /* Careful about overflow on the "+1" */
105 nrpages = end_index - start_index + 1;
106 if (!nrpages)
107 nrpages = ~0UL;
KOSAKI Motohiro3d3727c2012-07-31 16:42:50 -0700108
109 /*
110 * Ignore return value because fadvise() shall return
111 * success even if filesystem can't retrieve a hint,
112 */
Al Viro2903ff02012-08-28 12:52:22 -0400113 force_page_cache_readahead(mapping, f.file, start_index,
KOSAKI Motohiro3d3727c2012-07-31 16:42:50 -0700114 nrpages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 break;
Andrew Morton60c371b2006-08-05 12:14:25 -0700116 case POSIX_FADV_NOREUSE:
117 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 case POSIX_FADV_DONTNEED:
Tejun Heo703c2702015-05-22 17:13:44 -0400119 if (!inode_write_congested(mapping->host))
Shawn Bohrerad8a1b52012-01-10 15:07:35 -0800120 __filemap_fdatawrite_range(mapping, offset, endbyte,
121 WB_SYNC_NONE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Mel Gorman441c2282014-12-12 16:56:33 -0800123 /*
124 * First and last FULL page! Partial pages are deliberately
125 * preserved on the expectation that it is better to preserve
126 * needed memory than to discard unneeded memory.
127 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300128 start_index = (offset+(PAGE_SIZE-1)) >> PAGE_SHIFT;
129 end_index = (endbyte >> PAGE_SHIFT);
shidao.ytta7ab4002018-01-31 16:19:55 -0800130 /*
131 * The page at end_index will be inclusively discarded according
132 * by invalidate_mapping_pages(), so subtracting 1 from
133 * end_index means we will skip the last page. But if endbyte
134 * is page aligned or is at the end of file, we should not skip
135 * that page - discarding the last page is safe enough.
136 */
137 if ((endbyte & ~PAGE_MASK) != ~PAGE_MASK &&
138 endbyte != inode->i_size - 1) {
Oleg Drokin18aba412016-06-08 15:33:59 -0700139 /* First page is tricky as 0 - 1 = -1, but pgoff_t
140 * is unsigned, so the end_index >= start_index
141 * check below would be true and we'll discard the whole
142 * file cache which is not what was asked.
143 */
144 if (end_index == 0)
145 break;
146
147 end_index--;
148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
Mel Gorman67d46b22013-02-22 16:35:59 -0800150 if (end_index >= start_index) {
Johannes Weiner4dd72b42016-12-19 16:23:03 -0800151 unsigned long count;
152
153 /*
154 * It's common to FADV_DONTNEED right after
155 * the read or write that instantiates the
156 * pages, in which case there will be some
157 * sitting on the local LRU cache. Try to
158 * avoid the expensive remote drain and the
159 * second cache tree walk below by flushing
160 * them out right away.
161 */
162 lru_add_drain();
163
164 count = invalidate_mapping_pages(mapping,
Mel Gorman67d46b22013-02-22 16:35:59 -0800165 start_index, end_index);
166
167 /*
168 * If fewer pages were invalidated than expected then
169 * it is possible that some of the pages were on
170 * a per-cpu pagevec for a remote CPU. Drain all
171 * pagevecs and try again.
172 */
173 if (count < (end_index - start_index + 1)) {
174 lru_add_drain_all();
175 invalidate_mapping_pages(mapping, start_index,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800176 end_index);
Mel Gorman67d46b22013-02-22 16:35:59 -0800177 }
178 }
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800179 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 default:
181 ret = -EINVAL;
182 }
183out:
Al Viro2903ff02012-08-28 12:52:22 -0400184 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 return ret;
186}
187
188#ifdef __ARCH_WANT_SYS_FADVISE64
189
Al Viro4a0fd5b2013-01-21 15:16:58 -0500190SYSCALL_DEFINE4(fadvise64, int, fd, loff_t, offset, size_t, len, int, advice)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191{
192 return sys_fadvise64_64(fd, offset, len, advice);
193}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195#endif