diff options
| author | hyeongseok.kim <hyeongseok@gmail.com> | 2020-03-30 16:22:42 +0900 |
|---|---|---|
| committer | hyeongseok.kim <hyeongseok@gmail.com> | 2020-04-08 09:28:23 +0900 |
| commit | 41c6a50a702f1a8d0e2f2337520a5b815fb55d9f (patch) | |
| tree | 4d555410b44e3bff30af80e70a9833f6004f1b02 | |
| parent | 07723d7610873a82ab9ec4c28a85e439713963be (diff) | |
| download | newfs_msdos-41c6a50a702f1a8d0e2f2337520a5b815fb55d9f.tar.gz | |
set bdev->bd_block_size to reduce formatting time
When bdev->bd_block_size is bigger than bytesPerSec, e.g. 1024 bytes,
block write should read first to sync-up the mapped block.
This fix could prevent unnecessary read by temporarily change
bd_block_size to bytesPerSec size while flushing fat data.
bd_block_size would be reconfigured again at open() or mount() time.
Test: newfs_msdos -A /dev/<foo> (/dev/block/sdh1)
[2TB disk / 32kb cluster] : FAT#1/#2
476879KiB/512 bytes to write,
before : 476864 read I/Os (each 1024 bytes) : takes 180 seconds
after : 0 read I/Os : takes 20 seconds
Bug: 152893650
Change-Id: Idaa8d73e056cf740647345457ef9061c429f35a7
Signed-off-by: hyeongseok.kim <hyeongseok@gmail.com>
| -rw-r--r-- | mkfs_msdos.c | 6 |
1 files changed, 6 insertions, 0 deletions
diff --git a/mkfs_msdos.c b/mkfs_msdos.c index 04afc84..0ef30f2 100644 --- a/mkfs_msdos.c +++ b/mkfs_msdos.c @@ -640,6 +640,12 @@ mkfs_msdos(const char *fname, const char *dtype, const struct msdos_options *op) warn("sigaction SIGINFO"); goto done; } + +#if defined(__linux__) + if (ioctl(fd, BLKBSZSET, &bpb.bpbBytesPerSec)) + printf("BLKBSZSET to %u failed\n", bpb.bpbBytesPerSec); +#endif + for (lsn = 0; lsn < dir + (fat == 32 ? bpb.bpbSecPerClust : rds); lsn++) { if (got_siginfo) { fprintf(stderr,"%s: writing sector %u of %u (%u%%)\n", |