__chk_fail() is defined by LSB, and some "portable" binaries require it. For some reason that doesn't appear to be documented, Gentoo renames it to __stack_chk_fail, breaking compatibility with such binaries. Is there a reason we can't have __chk_fail aliased to __stack_chk_fail? (Simply renaming it would probably break ABIs all over the place when upgrading..)
(In reply to Luke-Jr from comment #0) > __chk_fail() is defined by LSB, and some "portable" binaries require it. For > some reason that doesn't appear to be documented, Gentoo renames it to > __stack_chk_fail, breaking compatibility with such binaries. We'll need some evidence. Preferably as a build failure and 'emerge --info' output. $ nm -D /lib/libc.so.6 | fgrep __chk_fail 0000000000117860 T __chk_fail
(In reply to Sergei Trofimovich from comment #1) > (In reply to Luke-Jr from comment #0) > > __chk_fail() is defined by LSB, and some "portable" binaries require it. For > > some reason that doesn't appear to be documented, Gentoo renames it to > > __stack_chk_fail, breaking compatibility with such binaries. > > We'll need some evidence. Preferably as a build failure and 'emerge --info' > output. > > $ nm -D /lib/libc.so.6 | fgrep __chk_fail > 0000000000117860 T __chk_fail Might only apply with USE=hardened, from the looks of it. The problematic patch is sys-libs/glibc/files/2.25/glibc-2.25-gentoo-chk_fail.c which contains: #ifdef GENTOO_SSP_HANDLER void __stack_chk_fail(void) #else void __chk_fail(void) #endif There is no build failure. The problem is running foreign binaries. For example, http://bitcoinknots.org/files/0.17.x/0.17.1.knots20181229/bitcoin-0.17.1.knots20181229-x86_64-linux-gnu.tar.gz (or any other deterministic-built Bitcoin node binary).
(In reply to Luke-Jr from comment #2) > (In reply to Sergei Trofimovich from comment #1) > > (In reply to Luke-Jr from comment #0) > > > __chk_fail() is defined by LSB, and some "portable" binaries require it. For > > > some reason that doesn't appear to be documented, Gentoo renames it to > > > __stack_chk_fail, breaking compatibility with such binaries. > > > > We'll need some evidence. Preferably as a build failure and 'emerge --info' > > output. > > > > $ nm -D /lib/libc.so.6 | fgrep __chk_fail > > 0000000000117860 T __chk_fail > > Might only apply with USE=hardened, from the looks of it. > > The problematic patch is > sys-libs/glibc/files/2.25/glibc-2.25-gentoo-chk_fail.c which contains: Could be it. This change is not applied to stable non-masked glibc versions though. > There is no build failure. The problem is running foreign binaries. I still don't see 'emerge --info'. Runtime failure is also fine to post. LD_DEBUG=all <tool-to-run> should shouw missing bindings.
Updating glibc seems to have resolve the issue.
