All, here is a draft for a rewrite of the device mapper udev rules, as follow up to the previous thread "About DM_UDEV_DISABLE_OTHER_RULES_FLAG and DM_NOSCAN" [1]. While I have taken care to minimize the impact on other udev rule sets, some changes to both the multipath-tools and systemd rules will be necessary. Patches for these are in preparation, but I'd like to reach consensus on the dm part first. As discussed in [1], the idea is to use just one udev property as "API" for later rules to determine whether the device at hand can be probed or otherwise accessed. This API remains DM_UDEV_DISABLE_OTHER_RULES_FLAG, because it is widely used by other rule sets. This means that the flag of the same name that is set in the DM_COOKIE needs to be saved and restored via a different property. For this property, I've chosen the name DM_COOKIE_DISABLE_OTHER_RULES_FLAG, which is hopefully awkward enough that authors of non-dm rules won't consider using it. DM_UDEV_DISABLE_OTHER_RULES_FLAG is a logical "OR" of various conditions that should prevent other rules from accessing a device. This includes the cookie flag, the "suspended" flag, and the "noscan" flag (DM_SUBSYSTEM_UDEV_FLAG0=1). dm-multipath has more such conditions. Because DM_SUSPENDED and DM_NOSCAN are now purely internal and don't need to be restored from the udev db, they are renamed to .DM_SUSPENDED and .DM_NOSCAN, respectively. For follow-up rules, the semantics of DM_UDEV_DISABLE_OTHER_RULES_FLAG=1 is "don't probe, don't attempt IO, and keep existing device properties (if any) until subsequent uevents are received". It doesn't mean the device should be completely ignored, and it doesn't mean that the block layer stack on top of the device should be destroyed. Contrary to what we discussed in [1], I didn't try to come up with flags carrying these other meanings. I the LVM layer doesn't have the necessary information for doing this. The only flag with "destroy" semantics I am aware of is SYSTEMD_READY="0", and it's up to systemd to deal with that. The first two patches are minor fixes that I came up with while working on the rules. [1] https://lore.kernel.org/linux-lvm/9d50edb0-baa4-4a01-a2f0-136dfdb79937@xxxxxxxxxx/T/#t Martin Wilck (7): 13-dm-disk.rules: import ID_FS_TYPE 10-dm.rules: don't deactivate devices for DISK_RO=1 10-dm-rules: don't restore DM_UDEV_DISABLE_OTHER_RULES_FLAG from db 11-dm-lvm.rules: don't restore DM_UDEV_DISABLE_OTHER_RULES_FLAG from db dm udev rules: don't export and save DM_SUSPENDED dm udev rules: don't export and save DM_NOSCAN 10-dm.rules: bump DM_UDEV_RULES_VSN to 3 udev/10-dm.rules.in | 32 +++++++++++++++++++++----------- udev/11-dm-lvm.rules.in | 13 +++++-------- udev/12-dm-permissions.rules | 2 +- udev/13-dm-disk.rules.in | 9 +++++---- 4 files changed, 32 insertions(+), 24 deletions(-) -- 2.43.2
