https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=281749
--- Comment #20 from Steve Peurifoy <[email protected]> --- I've also just encountered this again while doing a simple 'pkg upgrade'. Was semi-repeatable (although with differing files on subsequent runs). Here's the relevant section of ktrace output (14.4-RELEASE): 62327 pkg 7.033198205 CALL fchownat(0x14,0x2297c6261a81,0,0,0x200<AT_SYMLINK_NOFOLLOW>) 62327 pkg 7.033202207 NAMI "usr/local/include/boost/hana/.pkgtemp.bool.hpp.4FrzSwfmiNKf" 62327 pkg 7.033208622 RET fchownat 0 62327 pkg 7.033212207 CALL fchmodat(0x14,0x2297c6261a81,0100644<S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH|S_IFREG>,0x200<AT_SYMLINK_NOFOLLOW>) 62327 pkg 7.033215893 NAMI "usr/local/include/boost/hana/.pkgtemp.bool.hpp.4FrzSwfmiNKf" 62327 pkg 7.033225237 RET fchmodat -1 errno 9 Bad file descriptor so in the 14 microseconds between the call to fchownat() and fchmodat(), file descriptor 20 for '/' "went bad". This was on a VPS with a quite small zfs.arc_max inherited from its initial installation. I cranked up the setting and the error disappeared. So as someone who only knows enough to be dangerous, my tentative hypothesis is that there's some kind of locking issue and that under memory pressure, zfs is yanking vnodes that it shouldn't. Been meaning to learn the ways of dtrace, but I see that [email protected] has beaten me to the punch. :-) -- You are receiving this mail because: You are the assignee for the bug.
