Red Hat Bugzilla – Bug 84126
rpm-4.2 hang on futex during update or install
Last modified: 2007-04-18 12:51:02 EDT
Description of problem: [root@localhost newfiles]# rpm -Uvh groff-1.18.1-15.i386.rpm Preparing... ########################################### [100%] 1:groff ########################################### [100%] ( hangs here for minutes , no CPU activity , no disk activity ) [root@localhost root]# ps aux|grep rpm root 3236 0.3 2.0 9252 5296 pts/1 S 20:34 0:01 rpm -Uvh [root@localhost root]# strace -p 3236 futex(0x405881ec, FUTEX_WAIT, 0, NULL ( cursor stays after "NULL" , no activity ) If I press ctrl-C on strace : [root@localhost root]# strace -p 3236 futex(0x405881ec, FUTEX_WAIT, 0, NULL <unfinished ...> Then I did : [root@localhost newfiles]# kill -9 3236 [root@localhost newfiles]# rm /var/lib/rpm/__db.00* [root@localhost newfiles]# rpm -e groff --nodeps [root@localhost newfiles]# rpm -i groff-1.18.1-15.i386.rpm and it all worked. This seems to happen on rpm -Uvh xxx , if the __db.00? files exits in /var/lib/rpm. As they are left after all rpm invocations, that is almost a given. If I remove them before calling "rpm -Uvh xxx.rpm" then it does not hang. Version-Release number of selected component (if applicable): rpm-4.2-0.56 glibc-2.3.1-36 kernel-2.4.20-2.21.athlon ( also tried 2.4.20-2.41-athlon, same ) How reproducible: Steps to Reproduce: 1. rpm ivh ( or -Uvh ) some.rpm 2. 3. Actual results: it hangs at the end ( the new files are put on the file system ) Expected results: no hang Additional info:
Can you reproduce this problem? If so, please reopen with a description of how to do so, as I doubt I'm going to be able to reproduce otherwise.
After a few minutes of trying, I can't reproduce it any more with : kernel-2.4.20-2.54-athlon glibc-2.3.1-51 rpm-4.2-0.68
I also get the error from time to time on Fedora Core 3 when synaptic for some reason is stuck and I have killed it. I can confirm that removing the /var/lib/__db.00? files "fixes" the problem, however this time I saved them first, and now I can consistently retrigger the error by copying the files back. I will attach a tar file of a couple of /var/lib/rpm configurations with and without hangs (with hard links for common files) so you should be able to recreate. rpm-4.3.2-21 glibc-2.3.5-0.fc3.1 kernel-2.6.9-1.667 The hang is related to glibc nptl: OK: env LD_ASSUME_KERNEL=2.2.5 rpm -qal | head OK: env LD_ASSUME_KERNEL=2.4.1 rpm -qal | head HANG: env LD_ASSUME_KERNEL=2.4.20 rpm -qal | head but that is sort of expected I guess since futexes are firstly used in nptl.
Created attachment 113909 [details] This file only contains the __db.00? files
Created attachment 113910 [details] Full contents of /var/lib/rpm (if needed), part 1
Created attachment 113911 [details] Full contents of /var/lib/rpm (if needed), part 2