xref: /openbmc/linux/fs/iomap/swapfile.c (revision 35267cea)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2018 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
5  */
6 #include <linux/module.h>
7 #include <linux/compiler.h>
8 #include <linux/fs.h>
9 #include <linux/iomap.h>
10 #include <linux/swap.h>
11 
12 /* Swapfile activation */
13 
14 struct iomap_swapfile_info {
15 	struct iomap iomap;		/* accumulated iomap */
16 	struct swap_info_struct *sis;
17 	uint64_t lowest_ppage;		/* lowest physical addr seen (pages) */
18 	uint64_t highest_ppage;		/* highest physical addr seen (pages) */
19 	unsigned long nr_pages;		/* number of pages collected */
20 	int nr_extents;			/* extent count */
21 	struct file *file;
22 };
23 
24 /*
25  * Collect physical extents for this swap file.  Physical extents reported to
26  * the swap code must be trimmed to align to a page boundary.  The logical
27  * offset within the file is irrelevant since the swapfile code maps logical
28  * page numbers of the swap device to the physical page-aligned extents.
29  */
30 static int iomap_swapfile_add_extent(struct iomap_swapfile_info *isi)
31 {
32 	struct iomap *iomap = &isi->iomap;
33 	unsigned long nr_pages;
34 	uint64_t first_ppage;
35 	uint64_t first_ppage_reported;
36 	uint64_t next_ppage;
37 	int error;
38 
39 	/*
40 	 * Round the start up and the end down so that the physical
41 	 * extent aligns to a page boundary.
42 	 */
43 	first_ppage = ALIGN(iomap->addr, PAGE_SIZE) >> PAGE_SHIFT;
44 	next_ppage = ALIGN_DOWN(iomap->addr + iomap->length, PAGE_SIZE) >>
45 			PAGE_SHIFT;
46 
47 	/* Skip too-short physical extents. */
48 	if (first_ppage >= next_ppage)
49 		return 0;
50 	nr_pages = next_ppage - first_ppage;
51 
52 	/*
53 	 * Calculate how much swap space we're adding; the first page contains
54 	 * the swap header and doesn't count.  The mm still wants that first
55 	 * page fed to add_swap_extent, however.
56 	 */
57 	first_ppage_reported = first_ppage;
58 	if (iomap->offset == 0)
59 		first_ppage_reported++;
60 	if (isi->lowest_ppage > first_ppage_reported)
61 		isi->lowest_ppage = first_ppage_reported;
62 	if (isi->highest_ppage < (next_ppage - 1))
63 		isi->highest_ppage = next_ppage - 1;
64 
65 	/* Add extent, set up for the next call. */
66 	error = add_swap_extent(isi->sis, isi->nr_pages, nr_pages, first_ppage);
67 	if (error < 0)
68 		return error;
69 	isi->nr_extents += error;
70 	isi->nr_pages += nr_pages;
71 	return 0;
72 }
73 
74 static int iomap_swapfile_fail(struct iomap_swapfile_info *isi, const char *str)
75 {
76 	char *buf, *p = ERR_PTR(-ENOMEM);
77 
78 	buf = kmalloc(PATH_MAX, GFP_KERNEL);
79 	if (buf)
80 		p = file_path(isi->file, buf, PATH_MAX);
81 	pr_err("swapon: file %s %s\n", IS_ERR(p) ? "<unknown>" : p, str);
82 	kfree(buf);
83 	return -EINVAL;
84 }
85 
86 /*
87  * Accumulate iomaps for this swap file.  We have to accumulate iomaps because
88  * swap only cares about contiguous page-aligned physical extents and makes no
89  * distinction between written and unwritten extents.
90  */
91 static loff_t iomap_swapfile_activate_actor(struct inode *inode, loff_t pos,
92 		loff_t count, void *data, struct iomap *iomap,
93 		struct iomap *srcmap)
94 {
95 	struct iomap_swapfile_info *isi = data;
96 	int error;
97 
98 	switch (iomap->type) {
99 	case IOMAP_MAPPED:
100 	case IOMAP_UNWRITTEN:
101 		/* Only real or unwritten extents. */
102 		break;
103 	case IOMAP_INLINE:
104 		/* No inline data. */
105 		return iomap_swapfile_fail(isi, "is inline");
106 	default:
107 		return iomap_swapfile_fail(isi, "has unallocated extents");
108 	}
109 
110 	/* No uncommitted metadata or shared blocks. */
111 	if (iomap->flags & IOMAP_F_DIRTY)
112 		return iomap_swapfile_fail(isi, "is not committed");
113 	if (iomap->flags & IOMAP_F_SHARED)
114 		return iomap_swapfile_fail(isi, "has shared extents");
115 
116 	/* Only one bdev per swap file. */
117 	if (iomap->bdev != isi->sis->bdev)
118 		return iomap_swapfile_fail(isi, "outside the main device");
119 
120 	if (isi->iomap.length == 0) {
121 		/* No accumulated extent, so just store it. */
122 		memcpy(&isi->iomap, iomap, sizeof(isi->iomap));
123 	} else if (isi->iomap.addr + isi->iomap.length == iomap->addr) {
124 		/* Append this to the accumulated extent. */
125 		isi->iomap.length += iomap->length;
126 	} else {
127 		/* Otherwise, add the retained iomap and store this one. */
128 		error = iomap_swapfile_add_extent(isi);
129 		if (error)
130 			return error;
131 		memcpy(&isi->iomap, iomap, sizeof(isi->iomap));
132 	}
133 	return count;
134 }
135 
136 /*
137  * Iterate a swap file's iomaps to construct physical extents that can be
138  * passed to the swapfile subsystem.
139  */
140 int iomap_swapfile_activate(struct swap_info_struct *sis,
141 		struct file *swap_file, sector_t *pagespan,
142 		const struct iomap_ops *ops)
143 {
144 	struct iomap_swapfile_info isi = {
145 		.sis = sis,
146 		.lowest_ppage = (sector_t)-1ULL,
147 		.file = swap_file,
148 	};
149 	struct address_space *mapping = swap_file->f_mapping;
150 	struct inode *inode = mapping->host;
151 	loff_t pos = 0;
152 	loff_t len = ALIGN_DOWN(i_size_read(inode), PAGE_SIZE);
153 	loff_t ret;
154 
155 	/*
156 	 * Persist all file mapping metadata so that we won't have any
157 	 * IOMAP_F_DIRTY iomaps.
158 	 */
159 	ret = vfs_fsync(swap_file, 1);
160 	if (ret)
161 		return ret;
162 
163 	while (len > 0) {
164 		ret = iomap_apply(inode, pos, len, IOMAP_REPORT,
165 				ops, &isi, iomap_swapfile_activate_actor);
166 		if (ret <= 0)
167 			return ret;
168 
169 		pos += ret;
170 		len -= ret;
171 	}
172 
173 	if (isi.iomap.length) {
174 		ret = iomap_swapfile_add_extent(&isi);
175 		if (ret)
176 			return ret;
177 	}
178 
179 	/*
180 	 * If this swapfile doesn't contain even a single page-aligned
181 	 * contiguous range of blocks, reject this useless swapfile to
182 	 * prevent confusion later on.
183 	 */
184 	if (isi.nr_pages == 0) {
185 		pr_warn("swapon: Cannot find a single usable page in file.\n");
186 		return -EINVAL;
187 	}
188 
189 	*pagespan = 1 + isi.highest_ppage - isi.lowest_ppage;
190 	sis->max = isi.nr_pages;
191 	sis->pages = isi.nr_pages - 1;
192 	sis->highest_bit = isi.nr_pages - 1;
193 	return isi.nr_extents;
194 }
195 EXPORT_SYMBOL_GPL(iomap_swapfile_activate);
196