It seems that server process just crashes, but I am not sure. No segfault is logged in dmesg, nothing is logged in the syslog. The server starts new process for each connection, so this doesn't kill the whole server, only drops the current connection. Reproducible: Always Steps to Reproduce: 1. emerge vsftpd and configure it 2. connect via FTP and upload files one by one Actual Results: when you done uploading 30th file you will not see any files in the directory; if using Midnight Commander built-in FTP client - it will refuse to enter the directory Expected Results: obviously, complete directory listings every time, no matter how many files are inside The specific setup is as such: Gentoo system is in Proxmox VE 4.4 LXC container, created from official PVE template. User noticed the problem with system world updated in summer; recent world update with profile changed to 17 (with complete world rebuild according to the news article) didn't change anything. Every client chroots. Auth with PAM. No clients in exception list in vsftpd config. Every client experiences the problem, no matter if they could enter system with SSH or local or if they can't (/sbin/nologin). Clients are primarily announcemet robots which download clips to play. There isn't an option to retire this setup, as it requires FTP.
Created attachment 515218 [details] emerge --info output
I encountered the same issue on one of my servers and dig out that issue is qsort with enabled seccomp_sandbox (enabled by default). So currently there are two workarounds: 1) seccomp_sandbox=NO or 2) disable server side sorting by commenting out str_list_sort block at the end of vsf_ls_populate_dir_list.
*** This bug has been marked as a duplicate of bug 443898 ***