| 1 |
Alexander Gabert has been working very hard on getting et_dyn (ALSR) |
| 2 |
|
| 3 |
here is my report, with problem classification already: |
| 4 |
|
| 5 |
1) propolice: problems with __guard symbol not referred correctly |
| 6 |
----------------------------------------------------------------- |
| 7 |
grub (the boot loader) misses __guard in some of it's parts when compiling and fails |
| 8 |
cracklib misses __guard which is not in the @GOT in the library referred(?) |
| 9 |
hsi from Adamantix already suggested a workaround for that... |
| 10 |
investigating... Hiroaki has to say something about that... |
| 11 |
|
| 12 |
2) etdyn: Makefiles that forget to define -lcrypt or other -l* |
| 13 |
-------------------------------------------------------------- |
| 14 |
sys-libs/pam gets through when the Makefile is changed to include |
| 15 |
-lcrypt in the _right_ place where gcc is invoked |
| 16 |
funny though that this only uncovers with etdyn :-) |
| 17 |
|
| 18 |
3) etdyn: old style system calls via assembler constructed interface |
| 19 |
-------------------------------------------------------------------- |
| 20 |
lilo uses the old llseek syscall (like fdisk used in util-linux) |
| 21 |
net-ftp/proftpd uses a system call to check for capabilities(?) |
| 22 |
|
| 23 |
4) etdyn: failing configure scripts due to broken libraries |
| 24 |
----------------------------------------------------------- |
| 25 |
net-misc/ntp breaks because -lreadline is unable to resolve some symbols |
| 26 |
when doing a conftest in configure ... here is the fast workaround: |
| 27 |
the mistake only appears with a dumb ./configure conftest.c |
| 28 |
set CC='gcc -static' for the configure part and then let the following |
| 29 |
script rush through the dir: |
| 30 |
find . -name "Makefile" -exec sed -i -s 's:gcc -static:gcc:' {} \; |
| 31 |
then, do make and get etdyn executables though the configure is broken. |
| 32 |
disadvantage: is linux and gcc centric, need to find a better way, |
| 33 |
i have to report to the appropriate upstream! |
| 34 |
|
| 35 |
|
| 36 |
So the conclusion is that using CFLAGS and LDFLAGS to build an et_dyn system |
| 37 |
is not adequate and something else must be done. The current solution |
| 38 |
is to use gcc-config to switch the default spec file but this has a number |
| 39 |
of downfalls, we'd like any other thoughts, opinions, or ideas .. Thanks |
| 40 |
|
| 41 |
|
| 42 |
Joshua Brindle |
| 43 |
|
| 44 |
-- |
| 45 |
gentoo-dev@g.o mailing list |
Archives