* GNU/Hurd: ld.so.1 crashes while resolving symbols
@ 2021-12-15 18:30 Andrea Monaco
2021-12-16 9:22 ` Samuel Thibault
0 siblings, 1 reply; 3+ messages in thread
From: Andrea Monaco @ 2021-12-15 18:30 UTC (permalink / raw)
To: libc-alpha
After successfully building latest glibc, I test it with the following:
GLIBC=/root/glibc/build
GCONV_PATH=${GLIBC}/iconvdata LD_DEBUG=all LC_ALL=C
${GLIBC}/elf/ld.so.1 --library-path
${GLIBC}:${GLIBC}/math:${GLIBC}/elf:${GLIBC}/dlfcn:${GLIBC}/nss:${GLIBC}/nis:${GLIBC}/rt:${GLIBC}/resolv:${GLIBC}/crypt:${GLIBC}/nptl:${GLIBC}/dfp:${GLIBC}/mach:${GLIBC}/hurd
/root/glibc/bug-fstat
Shared objects are found and linked, but then the loader crashes.
Here's the last part of the output
0: symbol=__access_noerrno; lookup in file=./bug-fstat [0]
0: symbol=__access_noerrno; lookup in file=/root/glibc/build/libc.so.0.3 [0]
0: binding file /root/glibc/build/elf/ld.so.1 [0] to /root/glibc/build/libc.so.0.3 [0]: normal symbol `__access_noerrno' [GLIBC_PRIVATE]
0: symbol=_dl_init_first; lookup in file=./bug-fstat [0]
0: symbol=_dl_init_first; lookup in file=/root/glibc/build/libc.so.0.3 [0]
0: binding file /root/glibc/build/elf/ld.so.1 [0] to /root/glibc/build/libc.so.0.3 [0]: normal symbol `_dl_init_first' [GLIBC_PRIVATE]
0: symbol=_dl_signal_error; lookup in file=./bug-fstat [0]
0: symbol=_dl_signal_error; lookup in file=/root/glibc/build/libc.so.0.3 [0]
0: binding file /root/glibc/build/elf/ld.so.1 [0] to /root/glibc/build/libc.so.0.3 [0]: normal symbol `_dl_signal_error' [GLIBC_PRIVATE]
0: symbol=__errno_location; lookup in file=./bug-fstat [0]
0: symbol=__errno_location; lookup in file=/root/glibc/build/libc.so.0.3 [0]
0: binding file /root/glibc/build/elf/ld.so.1 [0] to /root/glibc/build/libc.so.0.3 [0]: normal symbol `__errno_location' [GLIBC_2.2.6]
0: symbol=__writev; lookup in file=./bug-fstat [0]
0: symbol=__writev; lookup in file=/root/glibc/build/libc.so.0.3 [0]
0: binding file /root/glibc/build/elf/ld.so.1 [0] to /root/glibc/build/libc.so.0.3 [0]: normal symbol `__writev' [GLIBC_2.2.6]
Killed
"bug-fstat" is my test program, but every program I tried does the same,
it crashes right after finding symbol "__writev".
After many printf, I singled out the exact instruction that causes the
crash, which is line 361 in sysdeps/i386/dl-machine.h (marked with >>):
switch (r_type)
{
# ifndef RTLD_BOOTSTRAP
case R_386_SIZE32:
/* Set to symbol size plus addend. */
*reloc_addr += sym->st_size;
break;
# endif
case R_386_GLOB_DAT:
case R_386_JMP_SLOT:
>> *reloc_addr = value;
break;
Is this obvious? Even if I made mistakes while building, I don't think
a crash is expected.
Thanks,
Andrea Monaco
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: GNU/Hurd: ld.so.1 crashes while resolving symbols
2021-12-15 18:30 GNU/Hurd: ld.so.1 crashes while resolving symbols Andrea Monaco
@ 2021-12-16 9:22 ` Samuel Thibault
0 siblings, 0 replies; 3+ messages in thread
From: Samuel Thibault @ 2021-12-16 9:22 UTC (permalink / raw)
To: Andrea Monaco; +Cc: libc-alpha
Hello,
Andrea Monaco via Libc-alpha, le mer. 15 déc. 2021 19:30:19 +0100, a ecrit:
> GLIBC=/root/glibc/build
>
> GCONV_PATH=${GLIBC}/iconvdata LD_DEBUG=all LC_ALL=C
> ${GLIBC}/elf/ld.so.1 --library-path
> ${GLIBC}:${GLIBC}/math:${GLIBC}/elf:${GLIBC}/dlfcn:${GLIBC}/nss:${GLIBC}/nis:${GLIBC}/rt:${GLIBC}/resolv:${GLIBC}/crypt:${GLIBC}/nptl:${GLIBC}/dfp:${GLIBC}/mach:${GLIBC}/hurd
> /root/glibc/bug-fstat
Rather than specifying all of this by hand, use the testrun.sh script
that is generated when running make check. Notably you are using a line
that is valid for Linux only (nptl).
Samuel
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: GNU/Hurd: ld.so.1 crashes while resolving symbols
@ 2021-12-16 23:59 Andrea Monaco
0 siblings, 0 replies; 3+ messages in thread
From: Andrea Monaco @ 2021-12-16 23:59 UTC (permalink / raw)
To: samuel.thibault; +Cc: libc-alpha
> Rather than specifying all of this by hand, use the testrun.sh
> script that is generated when running make check.
Thanks! Using the script solved the issue.
But it was LD_DEBUG=all that caused the crash. LD_DEBUG makes the
loader verbose, and it takes many values turning on various classes of
messages; "all" activates them all. More specifically, I found out that
the values of "bindings", "reloc" and "symbols" to LD_DEBUG make the
loader crash, while the others don't.
I got the same behavior on my already installed glibc, so it's a true
bug of glibc on GNU/Hurd.
Cheers,
Andrea Monaco
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2021-12-16 23:59 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-12-15 18:30 GNU/Hurd: ld.so.1 crashes while resolving symbols Andrea Monaco
2021-12-16 9:22 ` Samuel Thibault
2021-12-16 23:59 Andrea Monaco
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).