TL;DR: sudo hdparm --repair-sector SECTOR /dev/sdx
Intro
If S.M.A.R.T. self-tests started failing on you (you're both running and monitoring them, are you?), you might try to force drive to relocate it. Note that proposed method is destructive, meaning whatever was written in this sector is erased with zeros. It can be contents of some file, filesystem structure, etc. Hence, think about preserving your data first! If you care about it, ofc. Hopefully your filesystem supports data checksums, you have backups, and they're located on a filesystem that supports data checksums.
Problem
Let's consider, for example, that output of sudo smartctl -a /dev/sda has this snippet, among others:
<…>
SMART Self-test log structure revision number 1
Num Test_Description Status Remaining LifeTime(hours) LBA_of_first_error
# 1 Extended offline Completed: read failure 90% 5196 318795064
# 2 Short offline Completed without error 00% 5194 -
<…>
Here you can clearly see that:
a) Offline test ended with read failure;
b) It happened at LBA (sector) number 318795064.
Probably this happened due to some local problem on the surface of harddisk platter.
Solution
To force drive into discarding this surface location and use some reserve space for this LBA sector, issue this command:
sudo hdparm --repair-sector 318795064 --yes-i-know-what-i-am-doing /dev/sda
where 318795064 is that number from smartctl output, and /dev/sda is the disk name.
Once again, friendly reminder that it will overwrite this sector with zeros, wiping whatever was stored there previously.
Also worth noting that sometimes surface problems span several LBA sectors, so you might want to run the above hdparm command for several consecutive sectors, starting from the one mentioned in smartctl output.
To test disk surface starting from specific LBA sector, you can use the following command:
smartctl -t select,318795064-max /dev/sda
This StackOverflow answer has a nice script automating this loop of "find bad sector - overwrite it with 0s - repeat".