[uClibc] PATCH: My patches adapted to current cvs
Joakim Tjernlund
joakim.tjernlund at lumentis.se
Mon Nov 8 15:34:06 UTC 2004
> On Mon, 8 Nov 2004, Joakim Tjernlund wrote:
>
> > > On Behalf Of Peter S. Mazinger
> > > Hello!
> > >
> > > The attached tarball includes all the patches I consider going into 0.9.27
> > > Apply them in the following order.
> > > 1. asflags
> > > Replaces all CFLAGS w/ ASFLAGS, if an object is built from assembler code.
> > > The current functionality is not changed in any way.
> > > I have left the usage of SFLAGS as it was (in context), these should be
> > > considered to be replaced w/ ASFLAGS+= and use ASFLAGS instead as compile
> > > option
> > >
> > > 2. pie-correction
> > > Removes the conditions to create the Scrt1.o file (the files are created
> > > on x86/ppc/frv all the time), and use the option only to build ldd/iconv
> > > as PIE executables.
> > > The current functionality is not changed in any way.
> > >
> > > 3. pie_support-has_pie
> > > Renames UCLIBC_PIE_SUPPORT option to UCLIBC_BUILD_PIE
> > > The current functionality is not changed, but due to the option name
> > > change, buildroot's configs have to adapted
> > >
> > > 4. ssp-correction
> > > Corrects the usage of the SSP_FLAGS
> > > If UCLIBC_PROPOLICE is not enabled, the current functionality is not
> > > changed in any way. If the option is enabled, all libs and execs are built
> > > w/ -fstack-protector and -fstack-protector-all enabling full protection,
> > > unless it is not possible for some reason (libc/ldso)
> > >
> > > 5. propolice-has_ssp
> > > The UCLIBC_PROPOLICE option is splitted to UCLIBC_HAS_SSP and
> > > UCLIBC_BUILD_SSP, meaning:
> > > HAS_SSP: add ssp.c to libc
> > > BUILD_SSP: build all the libs/execs w/ max. ssp protection
> > > Current functionality is not changed, but due to the option name change,
> > > buildroot's configs have to be adapted
> > >
> > > 6. sec
> > > Adds a security submenu to menuconfig, moving PIE and SSP related options
> > > there and adds UCLIBC_BUILD_RELRO and UCLIBC_BUILD_NOW options, meaning
> > > build all w/ -z relro, resp. build all w/ -z now.
> > > The new options default to n.
> > > The current functionality is not changed.
> > > Maybe it would be better to also add a common security option, and if that
> > > one is disabled, all the others in the submenu are disabled as well (if
> > > future changes are done within this submenu, those not setting the common
> > > security option won't see any change, the build systems won't have to
> > > change/adapt configs)
> > >
> > > 7. main
> > > Part of Jocke's .protected main patch
> > >
> > > 8. zdefs
> > > Adds zdefs commonly to all LDFLAGS (can only be used in conjunction
> > > w/ patch 7.
> > > on x86, else libc will fail to build)
> > >
> > > 9. fgets_unlocked
> > > adds extern fgets_unlocked where needed
> > >
> > > 10. no_dl_do_copy_reloc
> > > disables dl_do_copy_reloc, it's not used
> > >
> > > 11. nocache
> > > This patch was already sent to the list, and was not accepted, because it
> > > has too many ifdefs, it makes ld.so smaller if ld.so.cache support is
> > > disabled. Maybe someone can reorganize it, so that fewer ifdefs are used.
> > > Does not influence the current functionality.
> > >
> > > 12. getent
> > > Shell script that simulates glibc's getent binary where possible. Please
> > > consider adding it to docs or debian.
> > >
> > > Peter
> >
> > After a quick look I think 1-6 are OK and could probably go in. Need Eriks
> > blessing first though.
> >
> > 7 and 8: OK, if Erik is OK with this. I am not sure that nothing bad will happen.
> >
> > 9. No comment, I have no idea as to why this is needed.
>
> only to get rid of a warning about implicit use, when the .c file is
> compiled.
hmm, you also change fgets_unlocked to __fgets_unlocked
>
> >
> > 10. Sure, will do.
>
> what about 11 ;)
Forgot that one :) Will look closer a rainy day. Can't find the time to
do that now.
>
> >
> > 12. No comment, I have no idea as to why this is needed.
>
> many packages make use of getent instead of grepping passwd/group to check
> for existence of a user/group in post-install-scripts
> (rpms/debs/tbz2/tgz).
> They use it because the user/group db could be located on an ldap or
> nis server and getent from glibc is ldap/nis aware. uclibc does not
> support (yet) ldap and won't support nis, so this simulates the glibc
> getent binary based on passwd/group/aliases/hosts/networks and so on...
> and allows it to be used as drop in replacement. Currently gentoo makes
> use of it, and I am using it w/ my rpms.
OK, but I will leave this to Erik.
More information about the uClibc
mailing list