Hide Forgot
Description of problem: When 'ltrace -f' is run on a binary which uses pointers in the thread function, the ltrace dies with 134 errorcode. ======= Here is the nesty thread function: ======== void *TaskCode(void *argument) { unsigned int a = 0xDEADBEEF; unsigned int *ptr = &a; printf("I am a thread... a = %u *ptr = %u ptr = %p \n", a, *ptr, ptr); return NULL; } ======= Here is another thread function, which passes well: ======= void *TaskCode(void *argument) { unsigned int a = 0xDEADBEEF; unsigned int *ptr = &a; printf("I am a thread... a = %u *ptr = %u \n", a, *ptr); return NULL; } ======= The difference: ======= --- wrong.c 2013-11-12 13:09:04.870404143 +0100 +++ right.c 2013-11-12 13:12:02.108676760 +0100 @@ -12,7 +12,7 @@ unsigned int a = 0xDEADBEEF; unsigned int *ptr = &a; - printf("I am a thread... a = %u *ptr = %u ptr = %p \n", a, *ptr, ptr); + printf("I am a thread... a = %u *ptr = %u \n", a, *ptr); return NULL; } ........ so it seems that when I want to print the pointer's value (the target address), it causes ltrace to crash... The attached reproducer contains both files - right.c and wrong.c - and this is the diff between them. The reproducer.sh script compiles them both and runs ltrace on them. The ltrace in the second case (wrong.c) falls down with 134 exitcode and some dump. Version-Release number of selected component (if applicable): ltrace-0.7.2-5.el7.x86_64 kernel-3.10.0-33.el7.x86_64 How reproducible: Unpack & run the attached reproducer... or ... Steps to Reproduce: 1. gcc -o wrong -lpthread wrong.c 2. ltrace -o ltrace.log -f ./wrong Actual results: ltrace dies with 134 and dump like this: *** glibc detected *** ltrace: free(): invalid pointer: 0x0000000000629120 *** ======= Backtrace: ========= /lib64/libc.so.6[0x394ee7cb3e] ltrace[0x409b28] ltrace[0x40a3ab] ltrace[0x40c041] ltrace[0x40b3ee] ltrace[0x416db8] ltrace[0x417a79] ltrace[0x4034ec] ltrace[0x402f2e] /lib64/libc.so.6(__libc_start_main+0xf5)[0x394ee21a05] ltrace[0x402f61] ======= Memory map: ======== 00400000-00426000 r-xp 00000000 08:01 668887 /usr/bin/ltrace 00625000-00626000 r--p 00025000 08:01 668887 /usr/bin/ltrace 00626000-00628000 rw-p 00026000 08:01 668887 /usr/bin/ltrace 00628000-0062a000 rw-p 00000000 00:00 0 00827000-00829000 rw-p 00027000 08:01 668887 /usr/bin/ltrace 00e42000-00ea5000 rw-p 00000000 00:00 0 [heap] ... Expected results: The ltrace don't crash and traces as expected. Additional info: * When traced under strace, everything goes well. * The bug occurs also on Fedora 18 with ltrace-0.7.2-1.fc18.x86_64
Created attachment 822938 [details] Reproducer script with necessary data
The problem probably is somewhere in the printf-tracing. This code is enough to cause the ltrace's fail, no threads are necessary: ----------- #include <stdio.h> #include <stdlib.h> int main(void) { unsigned int a = 0xDEADBEEF; unsigned int *ptr = &a; printf("a = %u *ptr = %u ptr = %p \n", a, *ptr, ptr); return 0; } ----------- gcc wrong.c ltrace ./a.out ....
I believe this is fixed in ltrace-0.7.91-1.el7. Upstream this was fixed in commits cae76962 (the fix) and cb943f1d (test cases).
This request was resolved in Red Hat Enterprise Linux 7.0. Contact your manager or support representative in case you have further questions about the request.
filezilla-3.7.3-3.el6 has been submitted as an update to Fedora EPEL 6. https://bodhi.fedoraproject.org/updates/FEDORA-EPEL-2015-e180c5b56d