On Mon, 02 Jun 2014 15:02:59 +0200 Artur Paszkiewicz
<artur.paszkiewicz@xxxxxxxxx> wrote:
> On 06/02/2014 04:36 AM, NeilBrown wrote:
> > On Fri, 30 May 2014 15:18:33 +0200 Artur Paszkiewicz
> > <artur.paszkiewicz@xxxxxxxxx> wrote:
> >
> >> If the checksum verification fails in mdadm and mdmon is running, retry
> >> the load to get a consistent snapshot of the mpb.
> >>
> >> Based on db575f3b
> >>
> >> Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@xxxxxxxxx>
> >> Reviewed-by: Pawel Baldysiak <pawel.baldysiak@xxxxxxxxx>
> >> ---
> >> super-intel.c | 17 +++++++++++++++++
> >> 1 file changed, 17 insertions(+)
> >>
> >> diff --git a/super-intel.c b/super-intel.c
> >> index f0a7ab5..037c018 100644
> >> --- a/super-intel.c
> >> +++ b/super-intel.c
> >> @@ -4422,6 +4422,7 @@ static int load_super_imsm(struct supertype *st, int fd, char *devname)
> >> {
> >> struct intel_super *super;
> >> int rv;
> >> + int retry;
> >>
> >> if (test_partition(fd))
> >> /* IMSM not allowed on partitions */
> >> @@ -4444,6 +4445,22 @@ static int load_super_imsm(struct supertype *st, int fd, char *devname)
> >> }
> >> rv = load_and_parse_mpb(fd, super, devname, 0);
> >>
> >> + /* retry the load if we might have raced against mdmon */
> >> + if (rv == 3) {
> >> + struct mdstat_ent *mdstat = mdstat_by_component(fd2devnm(fd));
> >> +
> >> + if (mdmon_running(mdstat->devnm) && getpid() != mdmon_pid(mdstat->devnm)) {
> >> + for (retry = 0; retry < 3; retry++) {
> >> + usleep(3000);
> >> + rv = load_and_parse_mpb(fd, super, devname, 0);
> >> + if (rv != 3)
> >> + break;
> >> + }
> >> + }
> >
> > The only thing you use from mdstat is devnm, and that is the thing you passed
> > to mdstat_by_component to get mdstat....
> >
> > Can you just do
> > char *devnm = fd2devnm(fd);
> > if (mdmon_running(devnm) && ......)
> >
> > ??
> >
> I can't do that because mdmon_running and mdmon_pid need a devnm of a
> container device, and the only thing we have here is the file descriptor
> of a component device. So I used mdstat_by_component to get the
> container devnm. Do you have an idea how to get that reliably without
> reading mdstat?
>
> I have overlooked that mdstat_by_component can return NULL here. I've
> added a check for this in the patch below.
Right, of course, yes.
I've applied your patch - and thanks for the updated version.
NeilBrown
Attachment:
signature.asc
Description: PGP signature
