Add metrics for scan_stats scanned and scan_stats issued

[RaptahJezus]

Hello,

ZFS On Linux 0.80 implemented sequential scrub and resilvers. SSDs may scan its metadata at 2.5+ GB/sec, and complete in a matter of seconds, while the subsequent scrub I/O may take many times longer. This was resulting in innaccurate ETAs for completion, as well as inaccurate stats about how fast the scrub was running.

Some metrics have been renamed to clarify them as "scan", and additional metrics have been added for the "issue" phase.

[richardelling]

Hi @RaptahJezus thanks for the PR! I'd like to have this code also compile for previous versions
of ZFS, so we'll need some way to determine at compile time that those struct entries exist.
Since C doesn't have reflection and it doesn't seem that the original PR for "Sequential scrub and
resilver" as a new feature #define, we'll need to find something that can be easily checked.
Can you look for such a #define?

[RaptahJezus]

Sure, one that sticks out is the addition of SSD TRIM in ZFS 0.80 RC-5. The following definitions were added in include/sys/fs/zfs.h

• ZPOOL_TRIM_COMMAND
• ZPOOL_TRIM_VDEVS
• ZPOOL_TRIM_RATE
• ZPOOL_TRIM_SECURE

How do you feel about those?

[richardelling]

@RaptahJezus I don't see anything else that would correlate better. Let's go with ZPOOL_TRIM_COMMAND and put a comment that explains why.