Revisit the ordering in factor

Revisit the ordering in factor

I have read the SO thread: Reorder levels of a data frame without changing
order of values, which works fine. However, on the particular use case, I
am puzzled on the output:
> df$mode
[1] write read write_with_journal write
read write_with_journal
[7] write read write_with_journal
Levels: read write write_with_journal
Now, I am changing the factor order from "read write write_with_journal"
to "write read write_with_journal":
> factor(df$mode, levels = c('write', 'read', 'write_with_journal'))
[1] <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA> <NA>
Levels: write read write_with_journal
Notice all the previous category values "write", "read" etc. are replaced
with NA. I am not sure why this is happening. If I manually create the
data frame (instead of reading from a file), then everything is fine.
> p = factor(rep(c("write", "read", "write_with_journal"),3))
> factor(p, levels = c('write', 'read', 'write_with_journal'))
Then everything is fine. Why?