Veeam Agent for Linux 6 blksnap module - Xen PV/PVHVM filesystem corruption

After Veeam Agent for Linux completes a volume-level backup or file-level backup using a snapshot, filesystem corruption is detected.

/var/log/syslog example:

EXT4-fs error (device dm-0): ext4_lookup:1752: inode #129793: comm ls: iget: checksum invalid
EXT4-fs error (device dm-0): ext4_lookup:1752: inode #129793: comm ls: iget: checksum invalid
EXT4-fs warning (device dm-0): ext4_dirblock_csum_verify:400: inode #12: comm ls: No space for directory leaf checksum. Please run e2fsck -D.
EXT4-fs error (device dm-0): htree_dirblock_to_tree:1042: inode #12: comm ls: Directory block failed checksum

Affected Configuration

Veeam Agent for Linux 6 is installed on a Xen PV/PVHVM virtual machine running a Linux distribution that uses blksnap kernel module by default (RHEL/Oracle Linux 9, SLES 15 SP3+,openSUSE Tumbleweed/Leap 15.3+,Ubuntu 22.04+, or Debian 11). For more information about which distributions use blksnap, refer to KB2804.

Research has determined that this issue is related xen-blkfront handling of read requests with the REQ_FUA bit enabled (implicitly converting these read requests to be zero-write requests, effectively zeroing the block being read).

For more information, refer to this Xenproject patch request: [PATCH] xen/blkfront: Only check REQ_FUA for writes