Hi, On Mon, 2020-10-12 at 00:35 -0700, Panu Matilainen wrote: > The problem here is that we're trying to classify a hermaphrodite > entity into one of two sexes. ET_EXEC is clear, but ET_DYN can be > either an executable, a dynamic shared library *or both*. Until we > embrace this fact, we'll get it wrong half of the time. > > So we need to stop trying to decide what *it is* and instead ask it > what it's *capable of*, and act accordingly. > libmagic doesn't know nearly enough to handle this, so the logic > needs to live inside elfdeps or we need to enhance the classifier > somehow (wouldn't be a bad thing at all but perhaps out of scope > here). Which means a split on the .attr level likely isn't going to > be meaningful. > > If we want to preserve the behavior where executables can disable > dependency generation by stripping x, we need to come up with a new > way of defining an executable. Given the nature of modern ELF, I > think the only meaningful way is to draw the line at: regardless of > it's possible other capabilities, does it do something if executed? > Which AIUI in technical terms comes down to: does it have PT_INTERP? > elfdeps looks that info up as it is, but it's not wired in a > meaningful way at the moment.
That is indeed a tricky question to answer. Note that elfutils since 0.177 comes with a tool eu-elfclassify to answer precisely these questions: https://gnu.wildebeest.org/blog/mjw/2019/08/15/elfutils-0-177-released-with-eu-elfclassify/ You might want to use that or study the code to see how it determines some of the corner cases. Specifically you may want to lookup is_executable () and is_program (), is_library () and is_shared () for some of the tricky corner cases: --executable File is (primarily) an ELF program executable (not primarily a DSO) --program File is an ELF program executable (might also be a DSO) --library File is an ELF shared object (DSO) (might also be an executable) --shared File is (primarily) an ELF shared object (DSO) (not primarily an executable) Cheers, Mark _______________________________________________ Rpm-maint mailing list Rpm-maint@lists.rpm.org http://lists.rpm.org/mailman/listinfo/rpm-maint
