I've been trying to explore alternative ways to avoid hanging stat() calls on
nfs mounts on linux, especially for "fuser -k -M /mount/path" invocations on
system shutdown. One thought is that you can determine the device IDs for
mounted filesystems from /proc/self/mountinfo. Then perhaps one could do a
strncmp() search though that to determine device numbers? Thoughts?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
But stat() is still called from scan_mounts(), which should no longer be
necessary - should be able to pull that directly from the mountinfo
information. Except for perhaps the inode info - is that needed? Perhaps skip
calling it if the inode info isn't needed?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I've been trying to explore alternative ways to avoid hanging stat() calls on
nfs mounts on linux, especially for "fuser -k -M /mount/path" invocations on
system shutdown. One thought is that you can determine the device IDs for
mounted filesystems from /proc/self/mountinfo. Then perhaps one could do a
strncmp() search though that to determine device numbers? Thoughts?
Wow, looks like this is already in the code with _LISTS_H.
But stat() is still called from scan_mounts(), which should no longer be
necessary - should be able to pull that directly from the mountinfo
information. Except for perhaps the inode info - is that needed? Perhaps skip
calling it if the inode info isn't needed?
I've a patch which uses one pipe and one forked sub process to handle the
stat(2) system call
as well with tested signal handling and restart of the sub process is a
SIGKILL was required.
Have a look at patch 3433393