Re: [PATCH 1/3] fio: Fix error string not matching errno

[Damien Le Moal <dlemoal@kernel.org>]

On 2/7/26 01:40, Niklas Cassel wrote:
> When using --error_dump=1 together with --continue_on_error=io and
> --ignore_error=62, we can get an inconsistent error in the summary line
>  produced by __show_run_stats():
> 
> Before patch:
> test: (groupid=0, jobs=1): err= 5 (file:io_u.c:2012, func=io_u error, error=Timer expired): pid=30925: Thu Feb  5 09:15:15 2026
> ...
>   IO depths    : 1=0.8%, 2=1.6%, 4=3.3%, 8=6.6%, 16=13.1%, 32=74.6%, >=64=0.0%
>      submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
>      complete  : 0=0.0%, 4=98.9%, 8=0.0%, 16=0.0%, 32=1.1%, 64=0.0%, >=64=0.0%
>      issued rwts: total=0,122,0,0 short=0,0,0,0 dropped=0,0,0,0
>      errors    : total=4, first_error=62/<Timer expired>
>      latency   : target=0, window=0, percentile=100.00%, depth=32
> 
> The string for errno "err= 5" is incorrectly printed as "Timer expired".
> (The correct string for "err= 5" is "Input/output error".)
> 
> There is thus a mismatch between the errno and the verbose string for the
> errno.
> 
> This problem is this code in __td_verror():
> 
> td->error = ____e;
> if (!td->first_error)
> 	nowarn_snprintf(td->verror, ..);
> 
> I.e. td->error (errno) is updated unconditionally, while td->verror (the
> verbose error string), is updated only if td->first_error is not set.
> 
> Thus, if you get a non-fatal error, it will set td->error and td->error.
> Later, for a non-fatal error, io_completed() will call update_error_count()
> which sets td->first_error, followed by td_clear_error() which will clear
> td->error.
> 
> Thus a second error (fatal or non-fatal) will set td->error, but since
> td->first_error is now set, it will not update td->verror.
> 
> Since td->verror contains the string representation of the errno stored in
> td->error, these two struct members should obviously always be updated at
> the same time.
> 
> If you look at the example print above, you can see that there is another
> print: first_error=62/<Timer expired>
> 
> Which prints td->first_error, and does not even use td->verror. Instead
> show_thread_status_normal() calls strerror(td->first_error) to get the
> error string for td->first_error.
> 
> There is thus absolutely no reason to not set td->verror every time
> td->error is set.
> 
> Remove the useless guard such that the error string will always correspond
> to errno.
> 
> Fixes: f2bba1820a56 ("Add a 'continue_on_error' option to fio")
> Signed-off-by: Niklas Cassel <cassel@kernel.org>

Looks OK to me.

Reviewed-by: Damien Le Moal <dlemoal@kernel.org>


-- 
Damien Le Moal
Western Digital Research