Created attachment 16372 [details]
Samba Log File

Overview:
If samba is up for a while and vfs_virusfilter is used, the virusfilter will stop working after some time.

Steps to Reproduce:
I do not have a definitive 100% working method here. However, on both my productive system (home server) and my laptop, it happened persistently enough to debug a bit into it. Steps that I took:
  1) Start smbd
  2) Open a few Excel files on Windows 10 (on the samba share of course)
  3) Write and close and reopen the Excel-Files in a random fashion
  4) After some minutes (sometimes just seconds) the bug should occur

Actual result:
vfs_virusfilter stops working. According to the log file, it cannot connect to the ClamAV socket anymore. Judging from the log file (see attachment), the memory where the "virusfilter:socket path" configuration is stored gets overwritten by some seemingly random data (yes, I have set it correctly in the configuration file and directly after startup of smbd everything works and smbd communicates properly with the ClamAV socket).

Expected result:
vfs_virusfilter should keep working without overwriting its internal configuration memory.

Build:
Version 4.11.6-Ubuntu

Additional Information:
From my first investigation, it seems that the memory which holds the configuration gets freed and thus gets later allocated somewhere else and thus gets overwritten with some seemingly random data. I made a first attempt in fixing the bug here: https://gitlab.com/fr89k/samba/-/commit/6e0b8468195fac8f5791ca9e8eb25af0321b89ae
I think the configuration char pointers should be copied into a newly allocated memory during initialization of vfs_virusfilter. At least this is the way other vfs modules do it, as far as I have seen. However, due to lack of understanding of the samba software architecture, memory ownership, and memory allocation concepts, I am not sure if I identified the root cause correctly and if my fix is actually the correct way to fix the problem (or if I just introduced a new memory leak).