#1355 new user config lost after power failure

[Josef Ridky]

Net-SNMP's configuration files are not resilient to power failure.

Users created with net-snmp-create-v3-user are missing from /etc/snmp/snmpd.conf after immidiate power failure event and that /var/lib/net-snmp/snmpd.conf is empty.

How to reproduce:
[root@desktop]# grep testuser /etc/snmp/snmpd.conf /var/lib/net-snmp/snmpd.conf
[root@desktop]# systemctl stop snmpd
[root@desktop]# /usr/bin/net-snmp-create-v3-user -A testpass testuser
[root@desktop]# grep testuser /etc/snmp/snmpd.conf /var/lib/net-snmp/snmpd.conf
/etc/snmp/snmpd.conf:rwuser testuser
/var/lib/net-snmp/snmpd.conf:createUser testuser MD5 "testpass" DES
[root@desktop]# systemctl start snmpd

Force power off and reconnect.

[root@desktop]# grep testuser /etc/snmp/snmpd.conf /var/lib/net-snmp/snmpd.conf
output is empty

Expected results:
/etc/snmp/snmpd.conf contains the newly added user.
/var/lib/net-snmp/snmpd.conf is present on file system.

Additional info:
This behaviour was observed for XFS root filesystems, on Ext's it works as expected.

Patch is attached. Please consider to add this patch to next Net-SNMP release.

[Bill Fenner]

I think you'd better fflush(out) first, right, or there may be data sitting in stdio's buffers that haven't made it to the file descriptor by the time you ask to fsync the file descriptor.

The man page for fsync on Linux says,

Calling fsync() does not necessarily ensure that the entry in the directory containing the file has also reached disk. For that an explicit fsync() on a file descriptor for the directory is also needed.

Do you think net-snmp should do this too?

Also, this adds as many fsync()s as there are rows in the tables being stored - basically every line gets synced to disk individually, possibly slowing this operation down significantly. I realize that this is the most straightforward way to accomplish your goal given the structure of the code, but perhaps it would be better to add an explicit read_config_flush(const char *type) which calls fsync() (and maybe does the magic that the man page describes on the directory too), and call that after all of the callbacks have been called.

(Or a rearchitecting of this system that keeps a file open while it's calling the callbacks, so that we can just call fflush()/fsync() at the end before closing it, which would be the most efficient but that is a major change.)

[Niels Baggesen]

Or, instead of rewriting the file, write a new file and rename it when the file is complete.

[Bart Van Assche]

+1 from me for Niels' proposal.

[Josef Ridky]

I am not sure, if creating temporary file and moving/renaming it to new location is suitable for all calls of read_config_store method due in this case (create new record for user) you just append to the end of file new line with information about new user. Personally, I think the best solution would be to add an explicit function for this (e.g. already mentioned read_config_flush) due this function (for meet goal of this request) is required to call only in usm_save_user method. But I am open to all of yours suggestions. What do you think about it?