| Summary: | GKrellMD incorrectly classifies automounted filesystems as local filesystems [explanation and proposed fix included] | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | Kevin Pyle <gentoo.8eaf7cd8e5128d8191fe> |
| Component: | Current packages | Assignee: | Jim Ramsay (lack) (RETIRED) <lack> |
| Status: | RESOLVED DUPLICATE | ||
| Severity: | minor | CC: | desktop-dock |
| Priority: | High | ||
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
A new version of gkrellm has been added to the tree (unstable as of this moment), so please try gkrellm-2.3.4. I don't see anything specific in the Changelog that would likely address your issue, but I'd like to just make sure this is the case. I looked at the source for 2.3.4 when it came out, and it definitely does not have the fix I proposed here. I did not do an exhaustive check of whether it was fixed in some other way. I will try it out within a few days and report back. Fixed in 416141! *** This bug has been marked as a duplicate of bug 416141 *** |
I reported this issue upstream to billw@ earlier this month, but have not received a response. I am reposting the report here, for the benefit of other Gentoo users until he has a chance to issue a release with the fix. This was observed on a stable amd64. A few non-system packages, such as autofs, were installed from unstable. There is no unstable release of gkrellm at this time. -- Begin excerpt -- Recently, I noticed unwarranted network activity when a system running gkrellmd automounts an NFS export. My configuration file sets both fs-interval and nfs-interval to 0. The latter is supposed to be supported, and I misread the comment as stating that the former was also supported. Running strace -e trace=statfs -p $(pgrep gkrellmd) showed that the gkrellmd was calling statfs both for my local mounts and for my NFS mount. I attached gdb to the gkrellmd, which confirmed that my setting of nfs-interval=0 was translated to _GK.nfs_interval=0, and alerted me that I had misread the comment, since _GK.fs_interval was set to 2. Since I do not use the file system monitor feature, I patched the gkrellmd code to allow fs-interval=0, and fixed a divide-by-zero crash that occurred in server/monitor.c!update_fs as a result of code that divides by _GK.fs_interval. The shipped gkrellmd never allowed _GK.fs_interval to be zero, so the crash was introduced by my other change. The patch consists of two sed commands executed by the ebuild: sed -e '/_GK.fs_interval < 1 || _GK.fs_interval > 1000/s/1/0/' \ -i server/main.c || die "sed server/main.c failed" sed -e 's/!(check_tick % _GK.fs_interval)/_GK.fs_interval ? & : 0/'\ -i server/monitor.c || die "sed server/monitor.c failed" The first changes the lower boundary from 1 to 0. The second fixes the divide-by-zero crash that is created by the first command. Debugging of the daemon revealed that the statfs of the NFS mount is because my /proc/mounts lists the entry twice, first with a type of "autofs" and then again with a type of "nfs". The daemon correctly did not statfs the entry marked as "nfs", but did try to statfs the entry marked "autofs". On a related note, the daemon currently classifies CIFS mounts as local, but SMBFS mounts as NFS. As a quick fix, adding "cifs" and "autofs" to server/monitor.c!remote_fs_types[] should address both these issues. I have not done or tested this, since disabling all statfs usage was adequate for my purposes. -- End excerpt --