On Fri, 2026-10-30 at 14:09 -0400, Benjamin Marzinski wrote: > On Thu, Oct 29, 2026 at 04:52:41PM -0700, Brian Bunker wrote: > > multipathd takes a path's WWID from udev. When a change uevent > > reports > > a different WWID, uev_update_path() removes the path from its map > > and > > re-adds it through uev_add_path(), which places it in whatever map > > owns > > the new WWID. Nothing checks the new value against the device > > itself; > > recheck_wwid only covers paths coming back up in the checker and > > "multipathd add path". If udev reports the wrong WWID, or the > > uevent is > > stale by the time it is processed, the path goes into the map of a > > different LUN and I/O for one LUN is silently sent to another. > > > > Use SG_IO to read VPD page 0x83 from the device in two places, for > > paths where can_recheck_wwid() is true: > > > > - In uev_update_path(), if the device still reports the cached > > WWID, > > ignore the udev value and leave the path where it is. > > > > - In ev_add_path(), when a path is being re-added because its udev > > data was missing or changed, refuse to add it if the device > > doesn't > > confirm its WWID. The path is orphaned, its WWID is cleared and > > it > > is marked INIT_MISSING_UDEV, so the checker retriggers udev > > (bounded > > by retrigger_tries) and the path is re-added from fresh data. > > > > To tell re-adds from ordinary adds, turn the need_do_map argument > > of > > uev_add_path() and ev_add_path() into enum path_add_mode, with a > > new > > PATH_ADD_REINIT value that behaves like PATH_ADD_MAP and also > > requests > > the WWID check. Ordinary adds are unchanged and never read the WWID > > from the device. > > > > If the SG_IO read fails, behavior is unchanged. > > > > Split the SG_IO read out of check_path_wwid_change() into > > get_sgio_wwid() so all the checks share it. > > > > Signed-off-by: Brian Bunker <[email protected]> > Reviewed-by: Benjamin Marzinski <[email protected]>
You will need convince me that this doesn't do more harm than good. We can't just use udev in one place and other methods elsewhere. I am open for a general redesign that uses sysfs for retrieving WWIDs, rather than udev. But using SG_IO seems wrong to me in general, because it can hang. If anything, we'd need to make this configurable, and we'd then need to use SG_IO consistently throughout the code (and SG_IO has significant disadvantages, as already mentioned). Also, SG_IO works for SCSI only. That may be a minor issue these days, because almost all disks handled by dm-multipath are SCSI disks, but at least formally we still support other block devices like NVMe and DASD. Unfortunately there's already a zoo of options for WWID detection which has grown over time. > About how likely this is: we hit it because a test tool was opening > and closing the sd devices read-write many times a second, and with > udev's "watch" rule every close generates a change uevent that runs > scsi_id against the device. I wouldn't call this a valid test. It's a deliberate abuse of udev functionality. If you do this, you need to disable udev's watch functionality, which is merely a convenience for users anyway. It's questionable if it's a useful feature on production servers in general. Thinking about it, I wonder if we should simply set "nowatch" for multipath members by default. AFAIU that would avoid the issue Brian encountered, at least in this test scenario. Regards Martin -- Dr. Martin Wilck <[email protected]> SUSE Software Solutions Germany GmbH, Frankenstr. 146, 90461 Nürnberg, Germany Geschäftsführer: Stefan Gaiser, Jochen Jaser, Abhinav Puri (HRB 36809,AG Nürnberg)
