Re: [Bug 984910] rm fails to detect errors in readdir(3)

[Pádraig Brady]

On 22/06/16 23:53, Bernhard Voelker wrote:
> On 06/22/2016 03:01 PM, Pádraig Brady wrote:
>> It'll need this on top, to better differentiate FTS_ERR and FTS_DNR,
>> as fts_build may be called multiple times for the same dir.
> 
> Hmm, but rm(1) won't show the errno.

Right, as FTS_DNR is actually passed out as intended in my patch,
and rm will try to rmdir() in that case.

I did consider the more aggressive FTS_ERR + FTS_STOP, though
that would then happen when readdir() returned nothing (with errno).
I was worried that some file systems may return NULL + ENOENT
for empty dirs for example, which would now cause rm to start
erroring where it might not have before.

> I'm getting better results with the attached, i.e., handling the
> readdir() failure like the other failures way down and setting the
> FTS_STOP flag.

Another reason I thought FTS_STOP too aggressive was it was
currently used only for internal errors where we it didn't
make any sense to proceed.

> 
> With Peter's LD_PRELOAD lib, I'm getting this:
> 
>   $ LD_PRELOAD=/tmp/libfake_readdir.so ~/findutils/find/find /tmp/ddd
>   /tmp/ddd
>   Forcing failure of readdir
>   /home/berny/findutils/find/find: failed to read file names from \
>   file system at or below ‘/tmp/ddd’: Too many levels of symbolic links
> 
>   $ LD_PRELOAD=/tmp/libfake_readdir.so ~/coreutils/src/rm -rfv /tmp/ddd
>   Forcing failure of readdir
>   Forcing failure of readdir
>   /home/berny/coreutils/src/rm: fts_read failed: Too many levels of \
>   symbolic links
> 
> WDYT?

I'd modified the LD_PRELOAD code to have 2 paths.
1. to return NULL immediately and thus return FTS_DNR
2. to return NULL after a couple of iterations and thus return FTS_ERR.

If you've objections to my assumptions I'm happy to adjust.

Pádraig