[Ocfs2-tools-devel] [PATCH] discard unused blocks before mkfs

[Larry Chen]

Hi piaojun,

On 02/24/2018 02:39 PM, piaojun wrote:
> Hi Larry,
>
> On 2018/2/24 12:14, Larry Chen wrote:
>> From: Larry <lchen at suse.com>
>>
>> When using a SSD as a block device, if unused SSD blocks could
>> be discarded in advance, performance will be improved.
>>
>> This patch uses ioctl interface to release unused blocks on SSD
>> layer.
> Will 'mkfs' become slower after this patch applied?
Of course, this patch will cause mkfs slower. But not too much(for 
several seconds on average).
It much depends on SSD controller and the current SSD block layer 
situation.
>
>> Signed-off-by: Larry Chen <lchen at suse.com>
>> ---
>>   mkfs.ocfs2/mkfs.c | 44 ++++++++++++++++++++++++++++++++++++++++++++
>>   mkfs.ocfs2/mkfs.h |  9 +++++++++
>>   2 files changed, 53 insertions(+)
>>
>> diff --git a/mkfs.ocfs2/mkfs.c b/mkfs.ocfs2/mkfs.c
>> index 354b2ee5..e9c8a9fb 100644
>> --- a/mkfs.ocfs2/mkfs.c
>> +++ b/mkfs.ocfs2/mkfs.c
>> @@ -543,6 +543,48 @@ static void finish_normal_format(State *s)
>>   	ocfs2_close(fs);Thanks
>>   }
>>   
>> +static errcode_t _discard_blocks(State *s, uint64_t block,
>> +		uint64_t count)
>> +{
>> +	int ret;
>> +	uint64_t range[2];
>> +
>> +	range[0] = (uint64_t)(block) << s->blocksize_bits;
>> +	range[1] = (uint64_t)(count) << s->blocksize_bits;
>> +
>> +	ret = ioctl(s->fd, BLKDISCARD, &range);
>> +	if (ret < 0) {
>> +		if (errno == EOPNOTSUPP && !s->quiet)
>> +			printf("block discard not supported...skip\n");
>> +		return errno;
> Should we catch other errno except 'EOPNOTSUPP'? And I suggest noticing
> upper user if any error happens.
>
> thanks,
> Jun
Sounds reasonable. I'll do it.
Thanks for your suggestion.
: )

Thanks,
Larry
>
>> +	}
>> +	return 0;
>> +}
>> +
>> +static int discard_blocks(State *s)
>> +{
>> +	uint64_t blocks = s->volume_size_in_blocks;
>> +	uint64_t count = DISCARD_STEP_MB;
>> +	uint64_t cur = 0;
>> +	int retval = 0;
>> +
>> +	count *= (1024 * 1024);
>> +	count >>= s->blocksize_bits;
>> +
>> +	while (cur < blocks) {
>> +		if (cur + count > blocks)
>> +			count = blocks - cur;
>> +
>> +		retval = _discard_blocks(s, cur, count);
>> +		if (retval)
>> +			break;
>> +		cur += count;
>> +	}
>> +
>> +	return retval;
>> +}
>> +
>> +
>>   int
>>   main(int argc, char **argv)
>>   {
>> @@ -615,6 +657,8 @@ main(int argc, char **argv)
>>   		return 0;
>>   	}
>>   
>> +	discard_blocks(s);
>> +
>>   	clear_both_ends(s);
>>   
>>   	init_record(s, &superblock_rec, SFI_OTHER, S_IFREG | 0644);
>> diff --git a/mkfs.ocfs2/mkfs.h b/mkfs.ocfs2/mkfs.h
>> index f9ba4dcf..ac9709f3 100644
>> --- a/mkfs.ocfs2/mkfs.h
>> +++ b/mkfs.ocfs2/mkfs.h
>> @@ -41,6 +41,7 @@
>>   #include <inttypes.h>
>>   #include <ctype.h>
>>   #include <assert.h>
>> +#include <sys/ioctl.h>
>>   
>>   #include <uuid/uuid.h>
>>   
>> @@ -92,6 +93,14 @@
>>   
>>   #define MAX_EXTALLOC_RESERVE_PERCENT	5
>>   
>> +#define DISCARD_STEP_MB         2048
>> +
>> +#if defined(__linux__) && !defined(BLKDISCARD)
>> +#define BLKDISCARD		_IO(0x12,119)
>> +#endif
>> +
>> +
>> +
>>   enum {
>>   	SFI_JOURNAL,
>>   	SFI_CLUSTER,
>>