Re: [PATCH proxmox-backup v2] fix #5247: relative paths in exclude patterns.

[Manuel Federanko <m.federanko@proxmox.com>]

On 2026-03-31 4:04 PM, Wolfgang Bumiller wrote:
> On Mon, 30 Mar 2026 10:11:10 +0200, Manuel Federanko <m.federanko@proxmox.com> wrote:
>> Patterns which start with ./ or for that matter contain /../ or similar
>> constructs will never match, since they get compared to sanitized paths
>> from directory traversal, sanitize those patterns too. Implement this
>> for both .pxarexclude files and the --exclude command line option.
>> Log a warning if a path was sanitized that was provided via --exclude.
>
> In the `.pxarexclude` case, logging may make sense as well IMO.

ack

> As for `--exclude`, I wonder we should just *bail* instead?
> 
> Although that might "break" some automated
> otherwise-working-but-probably-too-big backups, so we can't... (although
> for `foo/../bar` it's *really* tempting...)

I'm a bit undecided, maybe log a warning with a deprecation notice that
this will lead to an error in the future? Same goes for the cli option
now that I'm thinking about it.

>>
>>
>> diff --git a/pbs-client/src/pxar/tools.rs b/pbs-client/src/pxar/tools.rs
>> index 475c08ad3..7ad4009fe 100644
>> --- a/pbs-client/src/pxar/tools.rs
>> +++ b/pbs-client/src/pxar/tools.rs
>> @@ -76,6 +76,37 @@ fn assert_single_path_component_do(path: &Path) -> Result<(), Error> {
>>      Ok(())
>>  }
>>  
>> +pub fn normalize_lexically<S: AsRef<OsStr> + ?Sized>(path: &S) -> PathBuf {
> 
> (Note: the monomorphization from v1 was actually fine, but if the `_do`
> variant is only used by this one function it could live *inside* here.)
> 
>> +    // FIXME: Once std::path::normalize_lexically is stabilized we can
>> +    // switch to that
>> +    use std::path::Component;
>> +
>> +    let path = Path::new(path);
>> +    let has_trailing_slash = path
>> +        .as_os_str()
>> +        .as_encoded_bytes()
>> +        .last()
>> +        .copied()
>> +        .is_some_and(|c: u8| c == std::path::MAIN_SEPARATOR_STR.bytes().last().unwrap());
> 
> Could just use
> 
>     .as_encoded_bytes()
>     .ends_with(MAIN_SEPARATOR_STR.as_bytes())

ack, will change in v3 once warning/bailing has been decided