[RFE] could we see multipath options somewhere in /sys?

[jirib]

Could we see multipath options somewhere in /sys and not to dump configuration via multipath -T and check it?

Today if I see something like this:

# multipath -ll
mpathz (3600601606660450079c71c6173c6a3cc) dm-8 DGC,VRAID
 size=10G features='1 queue_if_no_path' hwhandler='1 alua' wp=rw
 |-+- policy='service-time 0' prio=50 status=active
 | |- 1:0:4:2 sdj 8:144 active ready running
 | `- 3:0:4:2 sdw 65:96 active ready running
 `-+- policy='service-time 0' prio=10 status=enabled
 |- 1:0:5:2 sdr 65:16 active ready running
 `- 3:0:5:2 sdy 65:128 active ready running

that doesn't tells me all multipath options. I need to run multipath -T and locate device section for vendor/model and see the options. Also, queue_if_no_path in the multipath -ll - IIUC - doesn't need to correspond to effective settings, ie. (IIUC) queue_if_no_path in multipath -ll output could be no_path_retry = <value>, right?

IMO if we could see right from a place in /sys the settings, it would be much easier.

[mwilck]

The answer to the first question is simply "no". Under /sys, you see the kernel device mapper status. For example, you see member devices under /sys/block/dm-$N/slaves. But most device mapper properties are only visible through the ioctl API and not through sysfs. This is by design. The generic tool for obtaining information from the kernel device-mapper is dmsetup (dmsetup table, dmsetup info). multipath and multipathd provide a more high-level view specific for dm-multipath.

The high-level management of dm-multipath devices is done in user space. This applies to no_path_retry for example. Device mapper only knows queueing state (queue_if_no_path) or non-queueing state. no_path_retry is a concept of multipathd, which counts seconds after the last path goes offline, and switches to non-queueing mode after the specified time (no_path_retry * polling_interval). Thus, what you see in multipath -ll output is indeed effective - it shows how the kernel is configured. If all paths went down in your sample above, the kernel would start queueing, until multipathd tells it to stop by clearing the queue_if_no_path flag.

The command multipath -ll (or multipathd show topology) is a command crafted to provide the most important information about the currently configured maps at a quick glance. We are most certainly not going to change the format or semantics of multipath -ll output, which users have got used to for more than a decade. If you want more specific information, you can use the multipath show paths format "..." and multipathd show maps format "..." commands, which can print a lot of information about the maps and their components using format wildcards (use multipathd show wildcards to get a list of supported outputs).

It is true that multipathd currently has no command (no wildcard) to print all settings that are effective for a given map. This would be an RFE. I'll set myself a reminder, but the todo list quite long already. @jirib, patches would be welcome.

[jirib]

Thank you very much for explanation. I'm just closing this RFE.