thanks for the info re timestamp.
The reason why my buffer is getting filled up so quickly is that I'm
getting lots of the below messages:
th xxxx tm xxxx: quorum_ioctl_with_retries: ioctl DKIOCGGEOM returned
error (48). Will retry in 2 seconds.
th xxxx tm xxxx: quorum_ioctl_with_retries: ioctl DKIOCGGEOM returned
error (48). Will retry in 2 seconds.
th xxxx tm xxxx: quorum_ioctl_with_retries: ioctl DKIOCGGEOM returned
error (48). Will retry in 2 seconds.
th xxxx tm xxxx: quorum_ioctl_with_retries: ioctl DKIOCGGEOM returned
error (48).
as timestamps are in microseconds the entries with "Will retry in 2
seconds." do indeed happen at every 2s, while the other entry happens at
about every 306s. I also double checked it with dtrace. I also get
stack() and ustack() when quorum_ioctl_with_retries() returns with
ENOTSUP (48). All of them happen when scqdmd (Quorum Device Monitor
Daemon) is checking qourum device and inside a kernel the
quorum_ioctl_with_retries() is then called from
quorum_scsi_get_sblkno(). By looking at the source code I can see that
if quorum_ioctl_with_retries() returns with ENOTSUP (48) and SC is
compiled with SC_EFI_SUPPORT then it will try to get the PGRE area by
calling get_efi_reserved(). I checked with dtrace and the
get_efi_reserved() returns with 0 all the time (so it succeeds).
To summarize:
scqdmd is checking quorum device status every 300s by default
scsi_get_sblkno() is called to get PGRE reserved area
quorum_ioctl_with_retries(DKIOCGGEOM) is called but fails with
ENOTSUP (48) if quorum is on a disk with EFI label
if quorum_ioctl_with_retries() returned with 48 then
SC_EFI_SUPPORT() is called to get the PGRE area
it means that the cmm_dbg_buf will get filled up relatively quickly on
every system with a quorum device on a lun with an EFI label. While not
a big issue imho the code should be fixed so no unnecesary warning
messages are written all the time - perhaps if QD is on an efi lun it
should be marked and code should call SC_EFI_SUPPORT() without calling
quorum_ioctl_with_retries(DKIOCGGEOM) in such a case.
Anyway, while it was a fun it didn't really help me :(
But at least I know why I'm getting these warning messages.
--
Robert Milkowski
http://milek.blogspot.com
