[Toybox] ls -Z with relative paths

[Rob Landley]

On 01/09/2016 03:38 PM, enh wrote:
> i have the following internal bug:
> 
> ls -laZ doesn't properly return the SELinux label for files where a
> non-fully qualified path is provided.
> 
> Steps to reproduce (on Nexus 9):
> 
>   adb shell ls -laZ /sys/kernel/debug/tracing/tracing_on
>   adb shell ls -laZ /sys/kernel/debug/tracing/ | grep tracing_on
>   adb shell 'cd /sys/kernel/debug/tracing ; ls -laZ tracing_on'
> 
> Expected: The output of all commands should yield the same SELinux
> label: "u:object_r:debugfs_tracing:s0"

I have now ordered a 1 gig quad processor raspberry pi "b" to try to
fling marshmallow upon. (I really need an android test environment, and
that's powerful enough to potentially self-host, so...)

(I would have ordered a b+ but those are all 512M with 700 mhz processor
and I think it's only dual core... Not sure how "+" works into it here.)

> something like this seems sensible for ls:
> 
> diff --git a/toys/posix/ls.c b/toys/posix/ls.c
> index 2f44d24..08ae695 100644
> --- a/toys/posix/ls.c
> +++ b/toys/posix/ls.c
> @@ -541,6 +541,7 @@ void ls_main(void)
>    // Iterate through command line arguments, collecting directories and files.
>    // Non-absolute paths are relative to current directory.
>    TT.files = dirtree_start(0, 0);
> +  TT.files->dirfd = AT_FDCWD;
>    for (s = *toys.optargs ? toys.optargs : noargs; *s; s++) {
>      dt = dirtree_start(*s, !(toys.optflags&(FLAG_l|FLAG_d|FLAG_F)) ||
>                              (toys.optflags&(FLAG_L|FLAG_H)));

Yes, that's the correct fix. (I just independently did that before
noticing the second half of your message. :)

> but i'm not sure whether the semantics of dirtree_start are supposed
> to imply this? certainly defaulting dirfd to -1 rather than 0 might
> lead to fewer surprises in future.

The ls recursion is unusual because ls's recursion semantics are funky.
(It does two passes over the data to pad out columns, the "label this
directory or not" logic involves knowledge of other arguments, and so on.)

So to implement what posix wants, ls takes manual control over recursion
(calling dirtree_recurse() itself in listfiles()), which uses mostly the
same data structure and mostly the same functions, but just... connects
them up strangely so it can do crazy things that straddle nodes.

This is why there's a fake node at the top of the ls tree representing
".", which is not usually the case. (The manual recursion in ls requires
a parent node to operate, the common infrastructure doesn't). So our
"did we fall off the top of the tree logic and have no ->parent,
therefore AT_CWD" logic doesn't trigger, and sticking AT_CWD into the
fake top of tree node as its dirfd is the correct response.

Rob