1 /* I/O block size definitions for coreutils
2 Copyright (C) 1989-2023 Free Software Foundation, Inc.
3
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
8
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
13
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <https://www.gnu.org/licenses/>. */
16
17 /* Include this file _after_ system headers if possible. */
18
19 /* sys/stat.h and minmax.h will already have been included by system.h. */
20 #include "count-leading-zeros.h"
21 #include "stat-size.h"
22
23
24 /* As of May 2014, 128KiB is determined to be the minimum
25 blksize to best minimize system call overhead.
26 This can be tested with this script:
27
28 for i in $(seq 0 10); do
29 bs=$((1024*2**$i))
30 printf "%7s=" $bs
31 timeout --foreground -sINT 2 \
32 dd bs=$bs if=/dev/zero of=/dev/null 2>&1 \
33 | sed -n 's/.* \([0-9.]* [GM]B\/s\)/\1/p'
34 done
35
36 With the results shown for these systems:
37 system #1: 1.7GHz pentium-m with 400MHz DDR2 RAM, arch=i686
38 system #2: 2.1GHz i3-2310M with 1333MHz DDR3 RAM, arch=x86_64
39 system #3: 3.2GHz i7-970 with 1333MHz DDR3, arch=x86_64
40 system #4: 2.20GHz Xeon E5-2660 with 1333MHz DDR3, arch=x86_64
41 system #5: 2.30GHz i7-3615QM with 1600MHz DDR3, arch=x86_64
42 system #6: 1.30GHz i5-4250U with 1-channel 1600MHz DDR3, arch=x86_64
43 system #7: 3.55GHz IBM,8231-E2B with 1066MHz DDR3, POWER7 revision 2.1
44
45 per-system transfer rate (GB/s)
46 blksize #1 #2 #3 #4 #5 #6 #7
47 ------------------------------------------------------------------------
48 1024 .73 1.7 2.6 .64 1.0 2.5 1.3
49 2048 1.3 3.0 4.4 1.2 2.0 4.4 2.5
50 4096 2.4 5.1 6.5 2.3 3.7 7.4 4.8
51 8192 3.5 7.3 8.5 4.0 6.0 10.4 9.2
52 16384 3.9 9.4 10.1 6.3 8.3 13.3 16.8
53 32768 5.2 9.9 11.1 8.1 10.7 13.2 28.0
54 65536 5.3 11.2 12.0 10.6 12.8 16.1 41.4
55 131072 5.5 11.8 12.3 12.1 14.0 16.7 54.8
56 262144 5.7 11.6 12.5 12.3 14.7 16.4 40.0
57 524288 5.7 11.4 12.5 12.1 14.7 15.5 34.5
58 1048576 5.8 11.4 12.6 12.2 14.9 15.7 36.5
59
60
61 Note that this is to minimize system call overhead.
62 Other values may be appropriate to minimize file system
63 overhead. For example on my current GNU/Linux system
64 the readahead setting is 128KiB which was read using:
65
66 file="."
67 device=$(df --output=source --local "$file" | tail -n1)
68 echo $(( $(blockdev --getra $device) * 512 ))
69
70 However there isn't a portable way to get the above.
71 In the future we could use the above method if available
72 and default to io_blksize() if not.
73 */
74 enum { IO_BUFSIZE = 128 * 1024 };
75 static inline idx_t
io_blksize(struct stat const * st)76 io_blksize (struct stat const *st)
77 {
78 /* Treat impossible blocksizes as if they were IO_BUFSIZE. */
79 idx_t blocksize = STP_BLKSIZE (st) <= 0 ? IO_BUFSIZE : STP_BLKSIZE (st);
80
81 /* Use a blocksize of at least IO_BUFSIZE bytes, keeping it a
82 multiple of the original blocksize. */
83 blocksize += (IO_BUFSIZE - 1) - (IO_BUFSIZE - 1) % blocksize;
84
85 /* For regular files we can ignore the blocksize if we think we know better.
86 ZFS sometimes understates the blocksize, because it thinks
87 apps stupidly allocate a block that large even for small files.
88 This misinformation can cause coreutils to use wrong-sized blocks.
89 Work around some of the performance bug by substituting the next
90 power of two when the reported blocksize is not a power of two. */
91 if (S_ISREG (st->st_mode) && blocksize & (blocksize - 1))
92 {
93 int leading_zeros = count_leading_zeros_ll (blocksize);
94 if (IDX_MAX < ULLONG_MAX || leading_zeros)
95 {
96 unsigned long long power = 1ull << (ULLONG_WIDTH - leading_zeros);
97 if (power <= IDX_MAX)
98 blocksize = power;
99 }
100 }
101
102 /* Don’t go above the largest power of two that fits in idx_t and size_t,
103 as that is asking for trouble. */
104 return MIN (MIN (IDX_MAX, SIZE_MAX) / 2 + 1,
105 blocksize);
106 }
107