A tmpfs filesystem containing a rrds log (via gmetad) and a newsyslog rotated verbose error log seems to be leaking memory at a constant rate (4K per second) while being unaccounted for in the actual filesystem. netdrive# df -h Filesystem Size Used Avail Capacity Mounted on tmpfs 1.1G 631M 504M 56% /tmp netdrive# du -hd 0 /tmp/ 2.6M /tmp/ netdrive# fstat -f /tmp/. USER CMD PID FD MOUNT INUM MODE SZ|DV R/W The memory df claims is being used is allocated, as the total used space matches the used space plus the running process's working set.
Try % fstat | less and search for ?(tmpfs). It looks "fstat" doesn't report a filename. For example, it did 'tail -f /tmp/x' and got hiro tail 19521 root / 2 drwxr-xr-x 512 r hiro tail 19521 wd - - ?(null) - hiro tail 19521 text /usr 525051 -r-xr-xr-x 16980 r hiro tail 19521 0 /dev 132 crw-rw-rw- ttyp5 rw hiro tail 19521 1 /dev 132 crw-rw-rw- ttyp5 rw hiro tail 19521 2 /dev 132 crw-rw-rw- ttyp5 rw hiro tail 19521 3 - - ?(tmpfs) - I do not know how rrds or newsyslog rotates files. But, http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/131356 may also be related.
Responsible Changed From-To: freebsd-bugs->freebsd-fs Over to maintainer(s).
State Changed From-To: open->feedback Did you try Yoshihiro Ota's suggestion? Do you still see this problem?
Responsible Changed From-To: freebsd-fs->jh Track.
State Changed From-To: feedback->closed Feedback timeout. Possibly not a tmpfs bug.