Christian Franke wrote:
The problem is that no errno is offered for "unspecified" errors
IMHO the problem is the existence of "unspecified" errors. Ddrescue needs
that the system returns 2 different kinds of errors: non-fatal (temporary or
affecting just a sector, for example) and fatal (permanent, like EACCES
"Permission denied").
If the system returns a fatal error code for a non-fatal error, ddrescue
quits prematurely.
If the system returns a non-fatal error code for a fatal error, ddrescue
marks the rest of the drive as "bad-sector" and rescues nothing.
and interestingly also not for "media changed".
Maybe ESTALE could be used for "media changed".
Any suggestion for a better fallback errno?
Once the system mixes fatal and non-fatal errors (for example by returning
ERROR_MEDIA_CHANGED instead of EIO) any mapping will make ddrescue behave
incorrectly one way or the other.
In addition to all this, option '-O, --reopen-on-error' only has effect
during the copying phase, when data are copied in large blocks. Therefore
the proposed '--continue-on-errno' would make ddrescue behave incorrectly
during trimming, scraping, and retrying.
Therefore I propose the following simpler change:
[...]
Great, this would help.
OK. I'll release a test version as soon as I have it implemented.
For diagnostics of possible other "interesting" behavior, it would also
be helpful to see an errno if interpreted as non-fatal but with
unexpected value, e.g. different from EIO.
The problem is to find a way for ddrescue to show such errno values, as
there may be thousands of failed reads per second. (Maybe adding an errno
field to the output of '--log-reads').
Best regards,
Antonio.