]> git.ipfire.org Git - thirdparty/e2fsprogs.git/blob - e2fsck/readahead.c
Merge branch 'maint' into next
[thirdparty/e2fsprogs.git] / e2fsck / readahead.c
1 /*
2 * readahead.c -- Prefetch filesystem metadata to speed up fsck.
3 *
4 * Copyright (C) 2014 Oracle.
5 *
6 * %Begin-Header%
7 * This file may be redistributed under the terms of the GNU Library
8 * General Public License, version 2.
9 * %End-Header%
10 */
11
12 #include "config.h"
13 #include <string.h>
14
15 #include "e2fsck.h"
16
17 #undef DEBUG
18
19 #ifdef DEBUG
20 # define dbg_printf(f, a...) do {printf(f, ## a); fflush(stdout); } while (0)
21 #else
22 # define dbg_printf(f, a...)
23 #endif
24
25 struct read_dblist {
26 errcode_t err;
27 blk64_t run_start;
28 blk64_t run_len;
29 int flags;
30 };
31
32 static int readahead_dir_block(ext2_filsys fs, struct ext2_db_entry2 *db,
33 void *priv_data)
34 {
35 struct read_dblist *pr = priv_data;
36 e2_blkcnt_t count = (pr->flags & E2FSCK_RA_DBLIST_IGNORE_BLOCKCNT ?
37 1 : db->blockcnt);
38
39 if (!pr->run_len || db->blk != pr->run_start + pr->run_len) {
40 if (pr->run_len) {
41 pr->err = io_channel_cache_readahead(fs->io,
42 pr->run_start,
43 pr->run_len);
44 dbg_printf("readahead start=%llu len=%llu err=%d\n",
45 pr->run_start, pr->run_len,
46 (int)pr->err);
47 }
48 pr->run_start = db->blk;
49 pr->run_len = 0;
50 }
51 pr->run_len += count;
52
53 return pr->err ? DBLIST_ABORT : 0;
54 }
55
56 errcode_t e2fsck_readahead_dblist(ext2_filsys fs, int flags,
57 ext2_dblist dblist,
58 unsigned long long start,
59 unsigned long long count)
60 {
61 errcode_t err;
62 struct read_dblist pr;
63
64 dbg_printf("%s: flags=0x%x\n", __func__, flags);
65 if (flags & ~E2FSCK_RA_DBLIST_ALL_FLAGS)
66 return EXT2_ET_INVALID_ARGUMENT;
67
68 memset(&pr, 0, sizeof(pr));
69 pr.flags = flags;
70 err = ext2fs_dblist_iterate3(dblist, readahead_dir_block, start,
71 count, &pr);
72 if (pr.err)
73 return pr.err;
74 if (err)
75 return err;
76
77 if (pr.run_len)
78 err = io_channel_cache_readahead(fs->io, pr.run_start,
79 pr.run_len);
80
81 return err;
82 }
83
84 static errcode_t e2fsck_readahead_bitmap(ext2_filsys fs,
85 ext2fs_block_bitmap ra_map)
86 {
87 blk64_t start, end, out;
88 errcode_t err;
89
90 start = 1;
91 end = ext2fs_blocks_count(fs->super) - 1;
92
93 err = ext2fs_find_first_set_block_bitmap2(ra_map, start, end, &out);
94 while (err == 0) {
95 start = out;
96 err = ext2fs_find_first_zero_block_bitmap2(ra_map, start, end,
97 &out);
98 if (err == ENOENT) {
99 out = end;
100 err = 0;
101 } else if (err)
102 break;
103
104 err = io_channel_cache_readahead(fs->io, start, out - start);
105 if (err)
106 break;
107 start = out;
108 err = ext2fs_find_first_set_block_bitmap2(ra_map, start, end,
109 &out);
110 }
111
112 if (err == ENOENT)
113 err = 0;
114
115 return err;
116 }
117
118 /* Try not to spew bitmap range errors for readahead */
119 static errcode_t mark_bmap_range(ext2fs_block_bitmap map,
120 blk64_t blk, unsigned int num)
121 {
122 if (blk >= ext2fs_get_generic_bmap_start(map) &&
123 blk + num <= ext2fs_get_generic_bmap_end(map))
124 ext2fs_mark_block_bitmap_range2(map, blk, num);
125 else
126 return EXT2_ET_INVALID_ARGUMENT;
127 return 0;
128 }
129
130 static errcode_t mark_bmap(ext2fs_block_bitmap map, blk64_t blk)
131 {
132 if (blk >= ext2fs_get_generic_bmap_start(map) &&
133 blk <= ext2fs_get_generic_bmap_end(map))
134 ext2fs_mark_block_bitmap2(map, blk);
135 else
136 return EXT2_ET_INVALID_ARGUMENT;
137 return 0;
138 }
139
140 errcode_t e2fsck_readahead(ext2_filsys fs, int flags, dgrp_t start,
141 dgrp_t ngroups)
142 {
143 blk64_t super, old_gdt, new_gdt;
144 blk_t blocks;
145 dgrp_t i;
146 ext2fs_block_bitmap ra_map = NULL;
147 dgrp_t end = start + ngroups;
148 errcode_t err = 0;
149
150 dbg_printf("%s: flags=0x%x start=%d groups=%d\n", __func__, flags,
151 start, ngroups);
152 if (flags & ~E2FSCK_READA_ALL_FLAGS)
153 return EXT2_ET_INVALID_ARGUMENT;
154
155 if (end > fs->group_desc_count)
156 end = fs->group_desc_count;
157
158 if (flags == 0)
159 return 0;
160
161 err = ext2fs_allocate_block_bitmap(fs, "readahead bitmap",
162 &ra_map);
163 if (err)
164 return err;
165
166 for (i = start; i < end; i++) {
167 err = ext2fs_super_and_bgd_loc2(fs, i, &super, &old_gdt,
168 &new_gdt, &blocks);
169 if (err)
170 break;
171
172 if (flags & E2FSCK_READA_SUPER) {
173 err = mark_bmap(ra_map, super);
174 if (err)
175 break;
176 }
177
178 if (flags & E2FSCK_READA_GDT) {
179 err = mark_bmap_range(ra_map,
180 old_gdt ? old_gdt : new_gdt,
181 blocks);
182 if (err)
183 break;
184 }
185
186 if ((flags & E2FSCK_READA_BBITMAP) &&
187 !ext2fs_bg_flags_test(fs, i, EXT2_BG_BLOCK_UNINIT) &&
188 ext2fs_bg_free_blocks_count(fs, i) <
189 fs->super->s_blocks_per_group) {
190 super = ext2fs_block_bitmap_loc(fs, i);
191 err = mark_bmap(ra_map, super);
192 if (err)
193 break;
194 }
195
196 if ((flags & E2FSCK_READA_IBITMAP) &&
197 !ext2fs_bg_flags_test(fs, i, EXT2_BG_INODE_UNINIT) &&
198 ext2fs_bg_free_inodes_count(fs, i) <
199 fs->super->s_inodes_per_group) {
200 super = ext2fs_inode_bitmap_loc(fs, i);
201 err = mark_bmap(ra_map, super);
202 if (err)
203 break;
204 }
205
206 if ((flags & E2FSCK_READA_ITABLE) &&
207 ext2fs_bg_free_inodes_count(fs, i) <
208 fs->super->s_inodes_per_group) {
209 super = ext2fs_inode_table_loc(fs, i);
210 blocks = fs->inode_blocks_per_group -
211 (ext2fs_bg_itable_unused(fs, i) *
212 EXT2_INODE_SIZE(fs->super) / fs->blocksize);
213 err = mark_bmap_range(ra_map, super, blocks);
214 if (err)
215 break;
216 }
217 }
218
219 if (!err)
220 err = e2fsck_readahead_bitmap(fs, ra_map);
221
222 ext2fs_free_block_bitmap(ra_map);
223 return err;
224 }
225
226 int e2fsck_can_readahead(ext2_filsys fs)
227 {
228 errcode_t err;
229
230 err = io_channel_cache_readahead(fs->io, 0, 1);
231 dbg_printf("%s: supp=%d\n", __func__, err != EXT2_ET_OP_NOT_SUPPORTED);
232 return err != EXT2_ET_OP_NOT_SUPPORTED;
233 }
234
235 unsigned long long e2fsck_guess_readahead(ext2_filsys fs)
236 {
237 unsigned long long guess;
238
239 /*
240 * The optimal readahead sizes were experimentally determined by
241 * djwong in August 2014. Setting the RA size to two block groups'
242 * worth of inode table blocks seems to yield the largest reductions
243 * in e2fsck runtime.
244 */
245 guess = 2ULL * fs->blocksize * fs->inode_blocks_per_group;
246
247 /* Disable RA if it'd use more 1/50th of RAM. */
248 if (get_memory_size() > (guess * 50))
249 return guess / 1024;
250
251 return 0;
252 }