It's counter-intuitive, but no top-level API should return
LFS_ERR_CORRUPT. Instead, if we can't make progress because of a corrupt
block, we should return LFS_ERR_NOSPC. This makes it easier for users to
write code that is well behaved even when a device is end-of-life.
It's up to our mroot extension algorithm to make sure this case can't be
reached in normal operation unless the device is _actually_ at
end-of-life.
Because mroot extension is a bit of a special case, we weren't
converting these corrupt errors to nospc errors consistently. This is
fixed now, along with a couple more hopefully-useful logging statements.
Found while playing around with test_exhaustion + block_recycles=-1.
This should assert on bad wear-leveling, but LFS_ERR_CORRUPT was
unexpected. Added an explicit test because this is an easy thing to let
split through:
- test_badblocks_mrootanchor_wear
Code changes were surprisingly minimal, I wonder if constants are being
swapped out somewhere low-level?
code stack
before: 33766 2600
after: 33770 (+0.0%) 2600 (+0.0%)
Christopher Haster
76ffb0e7b6