On Thu, Feb 27, 2020 at 1:41 PM William Roberts <bill.c.roberts@xxxxxxxxx> wrote: > It doesn't get us around all the issues, anything marked deprecated in selinux.h > and marked hidden_proto in selinux_internal.h, will still need a pragma. > > This would be much simpler if we drop the whole hidden_proto stuff, I guess > kloczek is proposing patches? Who is that, I see the Github name as: > Tomasz Kłoczko > > But I can't find any matching author in the git logs to CC them. I don't actually know him myself but he has opened a number of issues on the selinux userspace and appears to be involved in Fedora in some capacity. > I'm looking at those macros: > hidden_def > hidden_proto > > They both seem to take a function and create an _internal symbol, and > set the global symbol equal to the internal one. Essentially, both the > normal and _internal functions are pointing to the same address. > > $ readelf -s ./src/libselinux.so | grep setexeccon_raw > 176: 0000000000017c26 63 FUNC GLOBAL DEFAULT 12 setexeccon_raw > 279: 0000000000017c26 63 FUNC LOCAL DEFAULT 12 > setexeccon_raw_internal > 606: 0000000000017c26 63 FUNC GLOBAL DEFAULT 12 setexeccon_raw > > The hidden one, additionally > sets the visibility to hidden, which if IIUC, just requires one to > statically link. We can see above the > visibility is local. > > What I don't understand, is why one would do this? What actual > performance enhancement do we get? The hidden_def/hidden_proto stuff originally came from Ulrich Drepper, glibc maintainer at the time, to eliminate unnecessary runtime relocations and PLT entries being used for local symbols. Per the comments in https://github.com/SELinuxProject/selinux/issues/204, these might not be needed anymore and are breaking building with LTO. Willing to get rid of them if it doesn't produce a significant regression.
