xref: /openbmc/linux/fs/iomap/swapfile.c (revision 5e012745)
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 };
22 
23 /*
24  * Collect physical extents for this swap file.  Physical extents reported to
25  * the swap code must be trimmed to align to a page boundary.  The logical
26  * offset within the file is irrelevant since the swapfile code maps logical
27  * page numbers of the swap device to the physical page-aligned extents.
28  */
29 static int iomap_swapfile_add_extent(struct iomap_swapfile_info *isi)
30 {
31 	struct iomap *iomap = &isi->iomap;
32 	unsigned long nr_pages;
33 	uint64_t first_ppage;
34 	uint64_t first_ppage_reported;
35 	uint64_t next_ppage;
36 	int error;
37 
38 	/*
39 	 * Round the start up and the end down so that the physical
40 	 * extent aligns to a page boundary.
41 	 */
42 	first_ppage = ALIGN(iomap->addr, PAGE_SIZE) >> PAGE_SHIFT;
43 	next_ppage = ALIGN_DOWN(iomap->addr + iomap->length, PAGE_SIZE) >>
44 			PAGE_SHIFT;
45 
46 	/* Skip too-short physical extents. */
47 	if (first_ppage >= next_ppage)
48 		return 0;
49 	nr_pages = next_ppage - first_ppage;
50 
51 	/*
52 	 * Calculate how much swap space we're adding; the first page contains
53 	 * the swap header and doesn't count.  The mm still wants that first
54 	 * page fed to add_swap_extent, however.
55 	 */
56 	first_ppage_reported = first_ppage;
57 	if (iomap->offset == 0)
58 		first_ppage_reported++;
59 	if (isi->lowest_ppage > first_ppage_reported)
60 		isi->lowest_ppage = first_ppage_reported;
61 	if (isi->highest_ppage < (next_ppage - 1))
62 		isi->highest_ppage = next_ppage - 1;
63 
64 	/* Add extent, set up for the next call. */
65 	error = add_swap_extent(isi->sis, isi->nr_pages, nr_pages, first_ppage);
66 	if (error < 0)
67 		return error;
68 	isi->nr_extents += error;
69 	isi->nr_pages += nr_pages;
70 	return 0;
71 }
72 
73 /*
74  * Accumulate iomaps for this swap file.  We have to accumulate iomaps because
75  * swap only cares about contiguous page-aligned physical extents and makes no
76  * distinction between written and unwritten extents.
77  */
78 static loff_t iomap_swapfile_activate_actor(struct inode *inode, loff_t pos,
79 		loff_t count, void *data, struct iomap *iomap)
80 {
81 	struct iomap_swapfile_info *isi = data;
82 	int error;
83 
84 	switch (iomap->type) {
85 	case IOMAP_MAPPED:
86 	case IOMAP_UNWRITTEN:
87 		/* Only real or unwritten extents. */
88 		break;
89 	case IOMAP_INLINE:
90 		/* No inline data. */
91 		pr_err("swapon: file is inline\n");
92 		return -EINVAL;
93 	default:
94 		pr_err("swapon: file has unallocated extents\n");
95 		return -EINVAL;
96 	}
97 
98 	/* No uncommitted metadata or shared blocks. */
99 	if (iomap->flags & IOMAP_F_DIRTY) {
100 		pr_err("swapon: file is not committed\n");
101 		return -EINVAL;
102 	}
103 	if (iomap->flags & IOMAP_F_SHARED) {
104 		pr_err("swapon: file has shared extents\n");
105 		return -EINVAL;
106 	}
107 
108 	/* Only one bdev per swap file. */
109 	if (iomap->bdev != isi->sis->bdev) {
110 		pr_err("swapon: file is on multiple devices\n");
111 		return -EINVAL;
112 	}
113 
114 	if (isi->iomap.length == 0) {
115 		/* No accumulated extent, so just store it. */
116 		memcpy(&isi->iomap, iomap, sizeof(isi->iomap));
117 	} else if (isi->iomap.addr + isi->iomap.length == iomap->addr) {
118 		/* Append this to the accumulated extent. */
119 		isi->iomap.length += iomap->length;
120 	} else {
121 		/* Otherwise, add the retained iomap and store this one. */
122 		error = iomap_swapfile_add_extent(isi);
123 		if (error)
124 			return error;
125 		memcpy(&isi->iomap, iomap, sizeof(isi->iomap));
126 	}
127 	return count;
128 }
129 
130 /*
131  * Iterate a swap file's iomaps to construct physical extents that can be
132  * passed to the swapfile subsystem.
133  */
134 int iomap_swapfile_activate(struct swap_info_struct *sis,
135 		struct file *swap_file, sector_t *pagespan,
136 		const struct iomap_ops *ops)
137 {
138 	struct iomap_swapfile_info isi = {
139 		.sis = sis,
140 		.lowest_ppage = (sector_t)-1ULL,
141 	};
142 	struct address_space *mapping = swap_file->f_mapping;
143 	struct inode *inode = mapping->host;
144 	loff_t pos = 0;
145 	loff_t len = ALIGN_DOWN(i_size_read(inode), PAGE_SIZE);
146 	loff_t ret;
147 
148 	/*
149 	 * Persist all file mapping metadata so that we won't have any
150 	 * IOMAP_F_DIRTY iomaps.
151 	 */
152 	ret = vfs_fsync(swap_file, 1);
153 	if (ret)
154 		return ret;
155 
156 	while (len > 0) {
157 		ret = iomap_apply(inode, pos, len, IOMAP_REPORT,
158 				ops, &isi, iomap_swapfile_activate_actor);
159 		if (ret <= 0)
160 			return ret;
161 
162 		pos += ret;
163 		len -= ret;
164 	}
165 
166 	if (isi.iomap.length) {
167 		ret = iomap_swapfile_add_extent(&isi);
168 		if (ret)
169 			return ret;
170 	}
171 
172 	*pagespan = 1 + isi.highest_ppage - isi.lowest_ppage;
173 	sis->max = isi.nr_pages;
174 	sis->pages = isi.nr_pages - 1;
175 	sis->highest_bit = isi.nr_pages - 1;
176 	return isi.nr_extents;
177 }
178 EXPORT_SYMBOL_GPL(iomap_swapfile_activate);
179