Created attachment 5082 [details] Fix for field confusion in control files *Below is a cover letter I wrote before I learned opgk patches don't go through the mailing list. Attached is patch with proposed fix for described issue* A little bit of background: we have a custom field in our ipk control files named `Package_name` with a value of the ipk filename, e.g.: Package_name: libc6_2.35-r0.ipk The problem addressed in this patch is happening, because opkg only checks if first `n` bytes match `Package` string and doesn't care whether that string is actually followed by a colon (`:`) or not. This was especially bad in our case as because of our custom field opkg was cutting the field at the underscore in `Package_name` and was using what followed as the actual package name. So in that example of `Package_name: libc6_2.35-r0.ipk` resulting name of the installed package was `name: libc6_2.35-r0.ipk`, which of course is totally wrong. What's more, because of the space in that parsed package name, all (pre|post)(install|rm) hooks were broken, as they go through shell in `pkg.c:pkg_run_script()`, which uses `-c` argument. Because there is no escaping, it led to error like "no file named `name:`". You can see this is caused by a space in the wrongly parsed package name, which is used as a prefix in name for all `postinstall` hooks and similar. In my opinion that `-c` should be either removed mandating writing those hooks as a shell scripts or the whole shell call replaced with a direct binary execution. As I am not yet sure, which approach is right here, I am only tackling name confusion issue here.
Hey Michal; thanks for the patch. Patches *do* actually go through the mailing list at [1]. Though I could pull it from here, I would prefer that it go to the list, just to give others in the community the opportunity to speak about it. Could you resubmit there? [1] https://lists.yoctoproject.org/g/opkg Regarding this version of the patch: Overall, the issue and fix make sense. What happens if someone calls this `is_field()` implementation with `type` and `line` values that are the same, and do not end with a colon? eg. ``` is_field("foo", "foo); ``` It seems like the new `strncmp("foo", "foo", 3)` would succeed and pass through to the colon checks, where the second case (`line[3] == ':'`) would try to index into an undefined memory region. So you might have to do some more processing to avoid an index error.
> Patches *do* actually go through the mailing list I really don't know how I missed that... > Could you resubmit there? Sure, just did that. > What happens if someone calls this `is_field()` implementation with `type` > and `line` values that are the same, and do not end with a colon? eg. > > ``` > is_field("foo", "foo); > ``` > > It seems like the new `strncmp("foo", "foo", 3)` would succeed and pass > through to the colon checks, where the second case (`line[3] == ':'`) would > try to index into an undefined memory region. So you might have to do some > more processing to avoid an index error. Well, it wouldn't access undefined memory, as it would read the NULL terminator of the `line`, which would be different from `:` and in that case we would conclude the field doesn't match.
Hmm... I was thinking of situations where a null-terminator might not be applied to the end of the string. But on further research, none of those should apply here. So, you're correct. Objection dropped.
Fixed as of opkg v0.8.0, by commit c87188d7535684fddb8cf80993c147b215602b63. https://git.yoctoproject.org/opkg/commit/?id=c87188d7535684fddb8cf80993c147b215602b63