xref: /linux/mm/fadvise.c (revision ebaeabfa5ab711a9b69b686d58329e258fdae75f)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * mm/fadvise.c
4  *
5  * Copyright (C) 2002, Linus Torvalds
6  *
7  * 11Jan2003	Andrew Morton
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/fadvise.h>
18 #include <linux/writeback.h>
19 #include <linux/syscalls.h>
20 #include <linux/swap.h>
21 
22 #include <asm/unistd.h>
23 
24 #include "internal.h"
25 
26 /*
27  * POSIX_FADV_WILLNEED could set PG_Referenced, and POSIX_FADV_NOREUSE could
28  * deactivate the pages and clear PG_Referenced.
29  */
30 
generic_fadvise(struct file * file,loff_t offset,loff_t len,int advice)31 int generic_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
32 {
33 	struct inode *inode;
34 	struct address_space *mapping;
35 	struct backing_dev_info *bdi;
36 	loff_t endbyte;			/* inclusive */
37 	pgoff_t start_index;
38 	pgoff_t end_index;
39 	unsigned long nrpages;
40 
41 	inode = file_inode(file);
42 	if (S_ISFIFO(inode->i_mode))
43 		return -ESPIPE;
44 
45 	mapping = file->f_mapping;
46 	if (!mapping || len < 0)
47 		return -EINVAL;
48 
49 	bdi = inode_to_bdi(mapping->host);
50 
51 	if (IS_DAX(inode) || (bdi == &noop_backing_dev_info)) {
52 		switch (advice) {
53 		case POSIX_FADV_NORMAL:
54 		case POSIX_FADV_RANDOM:
55 		case POSIX_FADV_SEQUENTIAL:
56 		case POSIX_FADV_WILLNEED:
57 		case POSIX_FADV_NOREUSE:
58 		case POSIX_FADV_DONTNEED:
59 			/* no bad return value, but ignore advice */
60 			break;
61 		default:
62 			return -EINVAL;
63 		}
64 		return 0;
65 	}
66 
67 	/*
68 	 * Careful about overflows. Len == 0 means "as much as possible".  Use
69 	 * unsigned math because signed overflows are undefined and UBSan
70 	 * complains.
71 	 */
72 	endbyte = (u64)offset + (u64)len;
73 	if (!len || endbyte < len)
74 		endbyte = LLONG_MAX;
75 	else
76 		endbyte--;		/* inclusive */
77 
78 	switch (advice) {
79 	case POSIX_FADV_NORMAL:
80 		file->f_ra.ra_pages = bdi->ra_pages;
81 		spin_lock(&file->f_lock);
82 		file->f_mode &= ~(FMODE_RANDOM | FMODE_NOREUSE);
83 		spin_unlock(&file->f_lock);
84 		break;
85 	case POSIX_FADV_RANDOM:
86 		spin_lock(&file->f_lock);
87 		file->f_mode |= FMODE_RANDOM;
88 		spin_unlock(&file->f_lock);
89 		break;
90 	case POSIX_FADV_SEQUENTIAL:
91 		file->f_ra.ra_pages = bdi->ra_pages * 2;
92 		spin_lock(&file->f_lock);
93 		file->f_mode &= ~FMODE_RANDOM;
94 		spin_unlock(&file->f_lock);
95 		break;
96 	case POSIX_FADV_WILLNEED:
97 		/* First and last PARTIAL page! */
98 		start_index = offset >> PAGE_SHIFT;
99 		end_index = endbyte >> PAGE_SHIFT;
100 
101 		/* Careful about overflow on the "+1" */
102 		nrpages = end_index - start_index + 1;
103 		if (!nrpages)
104 			nrpages = ~0UL;
105 
106 		force_page_cache_readahead(mapping, file, start_index, nrpages);
107 		break;
108 	case POSIX_FADV_NOREUSE:
109 		spin_lock(&file->f_lock);
110 		file->f_mode |= FMODE_NOREUSE;
111 		spin_unlock(&file->f_lock);
112 		break;
113 	case POSIX_FADV_DONTNEED:
114 		filemap_flush_range(mapping, offset, endbyte);
115 
116 		/*
117 		 * First and last FULL page! Partial pages are deliberately
118 		 * preserved on the expectation that it is better to preserve
119 		 * needed memory than to discard unneeded memory.
120 		 */
121 		start_index = (offset+(PAGE_SIZE-1)) >> PAGE_SHIFT;
122 		end_index = (endbyte >> PAGE_SHIFT);
123 		/*
124 		 * The page at end_index will be inclusively discarded according
125 		 * by invalidate_mapping_pages(), so subtracting 1 from
126 		 * end_index means we will skip the last page.  But if endbyte
127 		 * is page aligned or is at the end of file, we should not skip
128 		 * that page - discarding the last page is safe enough.
129 		 */
130 		if ((endbyte & ~PAGE_MASK) != ~PAGE_MASK &&
131 				endbyte != inode->i_size - 1) {
132 			/* First page is tricky as 0 - 1 = -1, but pgoff_t
133 			 * is unsigned, so the end_index >= start_index
134 			 * check below would be true and we'll discard the whole
135 			 * file cache which is not what was asked.
136 			 */
137 			if (end_index == 0)
138 				break;
139 
140 			end_index--;
141 		}
142 
143 		if (end_index >= start_index) {
144 			unsigned long nr_failed = 0;
145 
146 			/*
147 			 * It's common to FADV_DONTNEED right after
148 			 * the read or write that instantiates the
149 			 * pages, in which case there will be some
150 			 * sitting on the local LRU cache. Try to
151 			 * avoid the expensive remote drain and the
152 			 * second cache tree walk below by flushing
153 			 * them out right away.
154 			 */
155 			lru_add_drain();
156 
157 			mapping_try_invalidate(mapping, start_index, end_index,
158 					&nr_failed);
159 
160 			/*
161 			 * The failures may be due to the folio being
162 			 * in the LRU cache of a remote CPU. Drain all
163 			 * caches and try again.
164 			 */
165 			if (nr_failed) {
166 				lru_add_drain_all();
167 				invalidate_mapping_pages(mapping, start_index,
168 						end_index);
169 			}
170 		}
171 		break;
172 	default:
173 		return -EINVAL;
174 	}
175 	return 0;
176 }
177 EXPORT_SYMBOL(generic_fadvise);
178 
vfs_fadvise(struct file * file,loff_t offset,loff_t len,int advice)179 int vfs_fadvise(struct file *file, loff_t offset, loff_t len, int advice)
180 {
181 	if (file->f_op->fadvise)
182 		return file->f_op->fadvise(file, offset, len, advice);
183 
184 	return generic_fadvise(file, offset, len, advice);
185 }
186 EXPORT_SYMBOL(vfs_fadvise);
187 
188 #ifdef CONFIG_ADVISE_SYSCALLS
189 
ksys_fadvise64_64(int fd,loff_t offset,loff_t len,int advice)190 int ksys_fadvise64_64(int fd, loff_t offset, loff_t len, int advice)
191 {
192 	CLASS(fd, f)(fd);
193 
194 	if (fd_empty(f))
195 		return -EBADF;
196 
197 	return vfs_fadvise(fd_file(f), offset, len, advice);
198 }
199 
SYSCALL_DEFINE4(fadvise64_64,int,fd,loff_t,offset,loff_t,len,int,advice)200 SYSCALL_DEFINE4(fadvise64_64, int, fd, loff_t, offset, loff_t, len, int, advice)
201 {
202 	return ksys_fadvise64_64(fd, offset, len, advice);
203 }
204 
205 #ifdef __ARCH_WANT_SYS_FADVISE64
206 
SYSCALL_DEFINE4(fadvise64,int,fd,loff_t,offset,size_t,len,int,advice)207 SYSCALL_DEFINE4(fadvise64, int, fd, loff_t, offset, size_t, len, int, advice)
208 {
209 	return ksys_fadvise64_64(fd, offset, len, advice);
210 }
211 
212 #endif
213 
214 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FADVISE64_64)
215 
COMPAT_SYSCALL_DEFINE6(fadvise64_64,int,fd,compat_arg_u64_dual (offset),compat_arg_u64_dual (len),int,advice)216 COMPAT_SYSCALL_DEFINE6(fadvise64_64, int, fd, compat_arg_u64_dual(offset),
217 		       compat_arg_u64_dual(len), int, advice)
218 {
219 	return ksys_fadvise64_64(fd, compat_arg_u64_glue(offset),
220 				 compat_arg_u64_glue(len), advice);
221 }
222 
223 #endif
224 #endif
225