while investigating #2991 (and probably related #1911) I straced pvedaemon noticing that the syncfs-calls all failed (errno EBADF).
This happens because we open the mountpoint path with O_PATH (open(2)), which causes syncfs to fail. Since I expect syncfs failing to be a potential cause for the inconstent state of containers on ceph-storages while backing up in snapshot mode this patchset hopefully mitigates the reported issues. (if this is not the solution, I was also investigating calling ioctl(fd, FIFREEZE) on the mountpoint before creating the snapshot) The patch to pve-container is maybe not necessary, but should save us a few syscalls, which don't help. Would be grateful for review from more experienced eyes - I ran limted tests with the patches applied - creating snapshot backups on a container with lvmthin backed rootfs and on one with ceph backed rootfs. pve-common Stoiko Ivanov (1): sync_mountpoint: open path so that sync works src/PVE/Tools.pm | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) pve-container: Stoiko Ivanov (1): sync_container_namespace: skip virtual filesystems src/PVE/LXC.pm | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) -- 2.20.1 _______________________________________________ pve-devel mailing list [email protected] https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
