| Message ID | 20240103175643.18438-1-bmarzins@redhat.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | change how multipathd deletes maps plus cleanups | expand |
On Wed, 2024-01-03 at 12:56 -0500, Benjamin Marzinski wrote: > The main purpose of this patchset is to make multipathd call > dm_suspend_and_flush_map() for the "del map" and "del maps" > interactive > commands. I ran into an issue where if there was outstanding IO to a > queueing multipath device, dm_flush_map() would hang removing the > last > kpartx device, since that would close the last opener of the > multipath > device, forcing it to wait for the IO to flush. This is avoided by > dm_suspend_and_flush_map(), since it disables queueing first. The > first > three patches do this. The next three are cleanups around removing a > device. The rest of the patches are make multipath return better > errors > when it delegates removes since it doesn't need to failback to doing > its > own removes anymore, make libdmmp handle failures from running > multipathd commands better, and make multipath and multipathc return > proper error codes when multipathd commands fail. > > Differences from v1 (based on suggestions by Martin Wilck) > 0003: simplify dm_flush_maps(), since all callers set need_suspend > 0005: Minor change in dm_flush_maps() to accomodate change in 0003. I > retained the reviewed-by. > 0007: Stripped the initial "fail\n" off messages with additional > failure > information. > 0009: Minor changes in dm_flush_maps() and its callers to accomodate > change in 0003. I retained the reviewed-by. > 0010: changed the log level of the error message to 2. > 0012: search for path by dev_t first, and change the log level of the > error message to 2. The commit message still says "make cli_handlers check for paths by dev and devt". I'll take the freedom to change this to the actual ordering. > 0015: Minor change to calling dm_flush_maps() to accomodate change in > 0003. I retained the reviewed-by You didn't, actually. I'll re-add it. > 0016: multipath repeatedly calls delegate_to_multipathd for all > remove_retries, instead of failing back to locally calling > remove. > 0017: New patch based on patch 0016 discussion. > 0018: New patch to deal with a issue caused by the change to track > the > queue_if_no_path state in mpp->features, from my previous > patchset. > > Benjamin Marzinski (18): > multipathd: remove nopath flushing code from flush_map() > multipathd: make flush_map() delete maps like the multipath command > multipathd: disable queueing when removing unknown maps > multipathd: simplify cli_del_map() > libmultipath: make _dm_flush_map() return an enum > libmultipath: make dm_remove_partmaps() a static function > multipathd: always start failure replies with "fail\n" > multipathd: print extra default reply msg for busy devices > multipathd: handle busy devices in cli_del_maps() > libmultipath: print error when find_mp_by_str() fails. > multipathd: don't open code find_mp_by_str() in client handers > multipathd: make cli_handlers check for paths by dev and devt > multipathd: add cli_handler reply message for missing devs > libdmmp: handle failures in _process_cmd > multipath: get rid of unnecessary retries variable > multipath: Don't locally retry deletgated remove failures > multipath: if delegation times out mark as not delegated > multipathd: sync features on flush_map failure corner case > > libdmmp/libdmmp.c | 71 +++++------- > libmultipath/devmapper.c | 47 ++++---- > libmultipath/devmapper.h | 13 ++- > libmultipath/libmultipath.version | 3 +- > libmultipath/structs.c | 9 +- > multipath/main.c | 72 +++++++----- > multipathd/cli_handlers.c | 182 ++++++++++++++-------------- > -- > multipathd/main.c | 70 ++++-------- > multipathd/main.h | 3 +- > multipathd/uxclnt.c | 9 +- > multipathd/uxlsnr.c | 15 ++- > 11 files changed, 241 insertions(+), 253 deletions(-) For the set: Reviewed-by: Martin Wilck <mwilck@suse.com>
The main purpose of this patchset is to make multipathd call dm_suspend_and_flush_map() for the "del map" and "del maps" interactive commands. I ran into an issue where if there was outstanding IO to a queueing multipath device, dm_flush_map() would hang removing the last kpartx device, since that would close the last opener of the multipath device, forcing it to wait for the IO to flush. This is avoided by dm_suspend_and_flush_map(), since it disables queueing first. The first three patches do this. The next three are cleanups around removing a device. The rest of the patches are make multipath return better errors when it delegates removes since it doesn't need to failback to doing its own removes anymore, make libdmmp handle failures from running multipathd commands better, and make multipath and multipathc return proper error codes when multipathd commands fail. Differences from v1 (based on suggestions by Martin Wilck) 0003: simplify dm_flush_maps(), since all callers set need_suspend 0005: Minor change in dm_flush_maps() to accomodate change in 0003. I retained the reviewed-by. 0007: Stripped the initial "fail\n" off messages with additional failure information. 0009: Minor changes in dm_flush_maps() and its callers to accomodate change in 0003. I retained the reviewed-by. 0010: changed the log level of the error message to 2. 0012: search for path by dev_t first, and change the log level of the error message to 2. 0015: Minor change to calling dm_flush_maps() to accomodate change in 0003. I retained the reviewed-by 0016: multipath repeatedly calls delegate_to_multipathd for all remove_retries, instead of failing back to locally calling remove. 0017: New patch based on patch 0016 discussion. 0018: New patch to deal with a issue caused by the change to track the queue_if_no_path state in mpp->features, from my previous patchset. Benjamin Marzinski (18): multipathd: remove nopath flushing code from flush_map() multipathd: make flush_map() delete maps like the multipath command multipathd: disable queueing when removing unknown maps multipathd: simplify cli_del_map() libmultipath: make _dm_flush_map() return an enum libmultipath: make dm_remove_partmaps() a static function multipathd: always start failure replies with "fail\n" multipathd: print extra default reply msg for busy devices multipathd: handle busy devices in cli_del_maps() libmultipath: print error when find_mp_by_str() fails. multipathd: don't open code find_mp_by_str() in client handers multipathd: make cli_handlers check for paths by dev and devt multipathd: add cli_handler reply message for missing devs libdmmp: handle failures in _process_cmd multipath: get rid of unnecessary retries variable multipath: Don't locally retry deletgated remove failures multipath: if delegation times out mark as not delegated multipathd: sync features on flush_map failure corner case libdmmp/libdmmp.c | 71 +++++------- libmultipath/devmapper.c | 47 ++++---- libmultipath/devmapper.h | 13 ++- libmultipath/libmultipath.version | 3 +- libmultipath/structs.c | 9 +- multipath/main.c | 72 +++++++----- multipathd/cli_handlers.c | 182 ++++++++++++++---------------- multipathd/main.c | 70 ++++-------- multipathd/main.h | 3 +- multipathd/uxclnt.c | 9 +- multipathd/uxlsnr.c | 15 ++- 11 files changed, 241 insertions(+), 253 deletions(-)