On Mon, 2025-12-01 at 22:02 -0500, Benjamin Marzinski wrote:
> When libmpathpersist notifies multipathd that a key has been
> registered,
> cli_setprstatus() calls pr_register_active_paths() with a flag to let
> it
> know that the paths are likely already registered, and it can skip
> re-registering them, as long as the number of active paths matches
> the
> number of registered keys. This shortcut can fail, causing multipathd
> to
> not register needed paths, if either a path becomes usable and
> another
> becomes unusable while libmpathpersist is running or if there already
> were registered keys for I_T Nexus's that don't correspond to path
> devices.
>
> To make this shortcut work in cases like that, this commit adds a new
> multipathd command "setprstatus map <map> pathlist <pathlist>", where
> <pathlist> is a quoted, comma separated list of scsi path devices.
> libmpathpersist will send out the list of paths it registered the key
> on. pr_register_active_paths() will skip calling
> mpath_pr_event_handle()
> for paths on that list.
>
> In order to deal with the possiblity of a preempt occuring while
> libmpathpersist was running, the code still needs to check that it
> has
> the expected number of keys.
>
> Fixes: f7d6cd17 ("multipathd: Fix race while registering PR key")
> Signed-off-by: Benjamin Marzinski <[email protected]>
> ---
> libmpathpersist/mpath_persist_int.c | 6 +--
> libmpathpersist/mpath_updatepr.c | 48 +++++++++++++++++------
> libmpathpersist/mpathpr.h | 4 +-
> multipathd/callbacks.c | 1 +
> multipathd/cli.c | 1 +
> multipathd/cli.h | 2 +
> multipathd/cli_handlers.c | 39 ++++++++++++++++--
> multipathd/main.c | 61 +++++++++++++++++++--------
> --
> multipathd/main.h | 3 +-
> multipathd/multipathd.8.in | 6 +++
> 10 files changed, 132 insertions(+), 39 deletions(-)
I may be missing something, but this is quite a lot of additional
complexity just for the shortcut. Have you considered just not taking
the shortcut in the first place?
Martin
