multipath-tools: cleanup uevent generation on startup
cover multipath-tools: cleanup uevent generation on startup
From: Martin Wilck <[email protected]>
This series undoes several changes I did myself in an attempt to fix issues with multipath maps incompletely initialized in udev. They are now all superseded by 0d66e03 ("libmultipath: force map reload if udev incomplete"). Triggering artificial (spurious) uevents for dm devices is useless in almost all cases. If some map is found in an inconsistent or incomplete state, the only action that would fix it is a full map reload.
While the code reverted here doesn't do actual harm, it unnecessarily complicates matters, wastes resources, and may leave readers of the code with wrong ideas how to handle udev issues in multipathd.
Martin Wilck (3):
libmultipath: select_action(): skip is_mpp_known_to_udev() test
libmultipath: coalesce_paths(): stop triggering spurious uevents Revert "multipathd: uev_trigger(): handle incomplete ADD events"
libmultipath/configure.c | 39 --------------------------------------- multipathd/main.c | 25 ------------------------- 2 files changed, 64 deletions(-)
Signature · 7 lines
2.29.2 -- dm-devel mailing list [email protected] https://www.redhat.com/mailman/listinfo/dm-devel
PATCH 1/3 libmultipath: select_action(): skip is_mpp_known_to_udev() test
From: Martin Wilck <[email protected]>
This test is now superseded by the check introduced in 0d66e03 ("libmultipath: force map reload if udev incomplete").
--- libmultipath/configure.c | 15 --------------- 1 file changed, 15 deletions(-)
libmultipath/configure.c +0 -15
@@ -635,15 +635,6 @@ trigger_paths_udev_change(struct multipath *mpp, bool is_mpath) mpp->needs_paths_uevent =0 ; } -static int -is_mpp_known_to_udev (const struct multipath *mpp) -{ -struct udev_device *udd =get_udev_for_mpp (mpp); -int ret = (udd !=NULL ); -udev_device_unref (udd); -return ret; -} -static int sysfs_set_max_sectors_kb (struct multipath *mpp,int is_reload) {
@@ -865,12 +856,6 @@ void select_action (struct multipath *mpp, const struct _vector *curmp, mpp->alias);return ; } -if (!is_mpp_known_to_udev (cmpp)) { - mpp->action =ACT_RELOAD ; -condlog (3 ,"%s: set ACT_RELOAD (udev device not initialized)" , - mpp->alias); -return ; - } mpp->action =ACT_NOTHING ;condlog (3 ,"%s: set ACT_NOTHING (map unchanged)" , mpp->alias);
Signature · 7 lines
2.29.2 -- dm-devel mailing list [email protected] https://www.redhat.com/mailman/listinfo/dm-devel
PATCH 2/3 libmultipath: coalesce_paths(): stop triggering spurious uevents
In reply to [email protected]From: Martin Wilck <[email protected]>
Since 0d66e03 ("libmultipath: force map reload if udev incomplete"), we force-reload maps that we find incompletely initialized by udev. If select_action returns ACT_NOTHING nonetheless, the map must be initialized in udev, and thus and "add" uevent must have been seen already. Triggering this event once more is unlikely to fix anything for real.
Reverts: b516118 ("libmultipath: coalesce_paths: trigger uevent if nothing done")
--- libmultipath/configure.c | 24 ------------------------ 1 file changed, 24 deletions(-)
libmultipath/configure.c +0 -24
@@ -527,18 +527,6 @@ get_udev_for_mpp(const struct multipath *mpp)return udd; } -static void -trigger_udev_change (const struct multipath *mpp) -{ -static const char change[] ="change" ; -struct udev_device *udd =get_udev_for_mpp (mpp); -if (!udd) -return ; -condlog (3 ,"triggering %s uevent for %s" , change, mpp->alias); -sysfs_attr_set_value (udd,"uevent" , change,sizeof (change)-1 ); -udev_device_unref (udd); -} -static void trigger_partitions_udev_change (struct udev_device *dev,const char *action,int len) {
@@ -1297,18 +1285,6 @@ int coalesce_paths (struct vectors *vecs, vector mpvec, char *refwwid,continue ; } -if (r ==DOMAP_EXIST && mpp->action ==ACT_NOTHING && - force_reload ==FORCE_RELOAD_WEAK ) -/* -* First time we're called, and no changes applied. -* domap() was a noop. But we can't be sure that -* udev has already finished setting up this device -* (udev in initrd may have been shut down while -* processing this device or its children). -* Trigger a change event, just in case. -*/ -trigger_udev_change (find_mp_by_wwid (curmp, mpp->wwid)); - conf =get_multipath_config (); allow_queueing = conf->allow_queueing;put_multipath_config (conf);
Signature · 7 lines
2.29.2 -- dm-devel mailing list [email protected] https://www.redhat.com/mailman/listinfo/dm-devel
PATCH 3/3 Revert "multipathd: uev_trigger(): handle incomplete ADD events"
In reply to [email protected]From: Martin Wilck <[email protected]>
cb10d38 ("multipathd: uev_trigger(): handle incomplete ADD events") was an attempt to fix issues with incompletely initialized multipath maps observed in various scenarious. However, that patch was wrong. Spurious "change" events as this patch would generate have no effect, because they are ignored by the device-mapper udev rules. The correct fix for the problem we were facing is 0d66e03 ("libmultipath: force map reload if udev incomplete"), which forces a full map reload.
Reverts: cb10d38 ("multipathd: uev_trigger(): handle incomplete ADD events")
--- multipathd/main.c | 25 ------------------------- 1 file changed, 25 deletions(-)
multipathd/main.c +0 -25
@@ -1499,31 +1499,6 @@ uev_trigger (struct uevent * uev, void * trigger_data)uev_pathfail_check (uev, vecs); }else if (!strncmp (uev->action,"remove" ,6 )) { r =uev_remove_map (uev, vecs); - }else if (!strncmp (uev->action,"add" ,3 )) { -const char *ev_name; -char *dm_name; -int major =-1 , minor =-1 ; - -/* -* If DM_NAME is not set for a valid map, trigger a -* change event. This can happen during coldplug -* if udev was killed between handling the 'add' and -* 'change' events before. -*/ - ev_name =uevent_get_dm_name (uev); -if (!ev_name) { - major =uevent_get_major (uev); - minor =uevent_get_minor (uev); - dm_name =dm_mapname (major, minor); -if (dm_name && *dm_name) { -condlog (2 ,"%s: received incomplete 'add' uevent, triggering change" , - dm_name); -udev_device_set_sysattr_value (uev->udev, -"uevent" , -"change" ); -free (dm_name); - } - } } goto out; }
Signature · 7 lines
2.29.2 -- dm-devel mailing list [email protected] https://www.redhat.com/mailman/listinfo/dm-devel
Quoted 1 line · On Fri, Jan 08, 2021 at 06:00:41PM +0100, [email protected] wrote:
From: Martin Wilck <[email protected]>
For the set
Quoted 13 lines
This series undoes several changes I did myself in an attempt to fix issues with multipath maps incompletely initialized in udev. They are now all superseded by 0d66e03 ("libmultipath: force map reload if udev incomplete"). Triggering artificial (spurious) uevents for dm devices is useless in almost all cases. If some map is found in an inconsistent or incomplete state, the only action that would fix it is a full map reload.
While the code reverted here doesn't do actual harm, it unnecessarily complicates matters, wastes resources, and may leave readers of the code with wrong ideas how to handle udev issues in multipathd.
Quoted 11 lines
Martin Wilck (3):
libmultipath:select_action(): skipis_mpp_known_to_udev()test
libmultipath:coalesce_paths(): stop triggering spurious uevents Revert "multipathd:uev_trigger(): handle incomplete ADD events"libmultipath/configure.c | 39 --------------------------------------- multipathd/main.c | 25 ------------------------- 2 files changed, 64 deletions(-)Signature · 1 line
2.29.2
--
dm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/dm-devel