This script works on FreeBSD and repairs faulty sectors by using the S.M.A.R.T. replacement mechanisms of your harddrive. It only works on harddrives with 4k sectors (for now).
Be aware that running it on a drive containing a filesystem will DESTROY data, metadata and perhaps even the entire filesystem.
MAKE SURE YOU HAVE BACKUPS!
The command will run over entire drives or parts of it and check if there are any I/O errors. If it finds any, it will overwrite one 4k sector with zeros and then continue scanning from there on.
The operation is slow, because it reads single LBAs (512 bytes). If it finds an error, the repair procedure takes about 30s per sector. It also makes sure that a single sector read causes an error and aborts the entire procedure in cases that are unexpected.
This script destroys the filesystem. Yes, this is not a joke. It selectively overwrites parts of the filesystem on the harddrive with zeros.
After running this script you will need to repair the filesystem, if there
was any (use fsck or zpool scrub).
Most filesystems and device drivers do not enforce S.M.A.R.T. replacement. This is especially annoying when you have data that is read often, but it is not written again. The harddrive won't replace it, because it never writes to the faulty sectors. You always get read errors. It gets even more horrible on copy-on-write filesystems, where sectors are not reused but allocated at other offsets. In this case you lose track of this faulty sector that might appear later again.
The script has not been run on many harddrives, yet. Please consider it experimental and even worse, keep in mind that it really destroys data to repair a harddrive.
This script can be used when S.M.A.R.T. detect errors on a drive and there are still replacement sectors available. Watch this S.M.A.R.T. attribute:
# smartctl -A /dev/ada0
.
.
.
197 Current_Pending_Sector 0x0032 200 200 000 Old_age Always - 213
.
.
.
It shows that there are 213 sectors to replace, but S.M.A.R.T. has not replaced
them, yet.
When you are not sure what this script does and when you are not sure what the source of the errors is. A harddrive is a part of a complex piece of hardware. There might be multiple problems that lead to misbehavior of a harddrive. It starts by flaky cables, instable power sources, vibrations and much more.
DMA transfer errors are clearly not caused by faulty sectors.
./heal-harddrive.sh /dev/ada0
Scan device ada0 for errors and heal them.
./heal-harddrive.sh /dev/ada1 100000
Scan device ada1 for errors starting from LBA 100000 and heal them.