Description of problem: Fresh upgrade to Fedora 38. Existing amanda configuration that worked fine under Fedora 37 stopped working, segfault in libamanda. I've tried a few things --- pulled down amanda source from git, did a local build. No joy. Tried downgrading to the fc37 rpms, but those, too, now segfault in libamanda. Version-Release number of selected component: amanda-3.5.3-1.fc38 Additional info: reporter: libreport-2.17.9 reason: amandad killed by SIGSEGV rootdir: / cgroup: 0::/user.slice/user-33.slice/session-c2.scope type: CCpp package: amanda-3.5.3-1.fc38 crash_function: sec_tcp_conn_read_callback comment: Fresh upgrade to Fedora 38. Existing amanda configuration that worked fine under Fedora 37 stopped working, segfault in libamanda. I've tried a few things --- pulled down amanda source from git, did a local build. No joy. Tried downgrading to the fc37 rpms, but those, too, now segfault in libamanda. executable: /usr/lib64/amanda/amandad uid: 33 journald_cursor: s=c3eb572d1dac4d238e91f8bd30e77373;i=2c2966;b=4596403fb84845d1852efe94f5ab0d09;m=d3e9a7282;t=5fa9f43caac06;x=26d4e23ddc007e77 runlevel: N 5 backtrace_rating: 4 kernel: 6.2.13-300.fc38.x86_64 cmdline: /usr/lib64/amanda/amandad -auth=local Truncated backtrace: Thread no. 1 (6 frames) #0 sec_tcp_conn_read_callback at /usr/src/debug/amanda-3.5.3-1.fc38.x86_64/common-src/security-util.c:2449 #1 event_handle_callback at /usr/src/debug/amanda-3.5.3-1.fc38.x86_64/common-src/event.c:118 #5 g_main_context_iterate.isra.0 at ../glib/gmain.c:4276 #6 g_main_context_iteration at ../glib/gmain.c:4343 #7 event_loop_wait at /usr/src/debug/amanda-3.5.3-1.fc38.x86_64/common-src/event.c:427 #8 event_loop at /usr/src/debug/amanda-3.5.3-1.fc38.x86_64/common-src/event.c:328
Created attachment 1961519 [details] File: exploitable
Created attachment 1961520 [details] File: backtrace
Created attachment 1961521 [details] File: dso_list
Created attachment 1961522 [details] File: os_info
Created attachment 1961523 [details] File: cpuinfo
Created attachment 1961524 [details] File: core_backtrace
Created attachment 1961525 [details] File: limits
Created attachment 1961526 [details] File: environ
Created attachment 1961527 [details] File: open_fds
Created attachment 1961528 [details] File: proc_pid_status
Created attachment 1961529 [details] File: maps
Created attachment 1961530 [details] File: mountinfo
The system log shows crashes from amandad and planner, both of which resolve to sec_tcp_conn_read_callback. amcheck also crashes with a segfault in libamanda when I try to run it as user amandabackup but the system journal doesn't show me additional detail. There was a period when I'd upgraded a client (an ancient Sun Ultra 27) to Fed38 but the server was still running Fed37. The amanda client seemed to work ok and there were several successful backups with the Fed37 amanda server talking to the Fed38 amanda client. But once I upgraded the server system to Fed38, amanda stopped working.
It does seem that some library on which Amanda depends (maybe glib2 from the backtrace) either has a bug or changed in an incompatible way. glib2 did go from 2.74.7 to 2.76.2 so there's at least a chance, as Amanda uses the glib2 event loop. It could also be due to the compiler change. Unfortunately this is at a level far deeper than I am experienced at debugging so I will first have to see if I can reproduce this and then I will see if I can find someone who has a bit deeper understanding of what exactly is going on.
As of 230712 3:00 am, this bug has resolved itself --- amanada once again started running as a nightly cron job. I have no idea why, as I was actually away for a few days. The most recent previous software update was 230709 11:30 am and did include glib2, but I can see no reason that the change would take 2.5 days to propagate. A complete mystery to me but the log timestamps don't lie. All is not quite perfect, as the amanda planner now aborts the backup run if it cannot contact a host. Prior to updating, planner would simply soldier on, ignoring any failed hosts. A problem for the amanda mailing lists, I suspect. Might just be an option setting. This bug report can probably be closed.