device-mapper-multipath/0056-libmultipath-count-pending-paths-as-active-on-loads.patch
Benjamin Marzinski 9abdc502cf device-mapper-multipath-0.8.4-4
Rebased on top of additional commits staged for upstream
  * Previous patches 0048-0060 are now patches 0053-0054 & 0059-0069
Add 0048-libmultipath-add-device-to-hwtable.c.patch
Add 0049-master-libmultipath-fix-use-after-free-when-iscsi-lo.patch
Add 0050-libmultipath-warn-if-freeing-path-that-holds-mpp-hwe.patch
Add 0051-libmultipath-warn-about-NULL-value-of-mpp-hwe.patch
Add 0052-libmultipath-fix-mpp-hwe-handling-in-sync_paths.patch
Add 0055-libmultipath-remove-code-duplication-in-path-countin.patch
Add 0056-libmultipath-count-pending-paths-as-active-on-loads.patch
Add 0057-libmultipath-deal-with-flushing-no-maps.patch
Add 0058-multipath-deal-with-delegation-failures-correctly.patch
Add 0070-multipath-add-libmpathvalid-library.patch
  * adds the libmpathvalid.so library to determine if devices are
    valid multipath paths.
Add 0071-libmultipath-add-uid-failback-for-dasd-devices.patch
Add 0072-libmultipath-add-ignore_udev_uid-option.patch
2020-07-21 17:39:38 -05:00

74 lines
2.9 KiB
Diff

From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Benjamin Marzinski <bmarzins@redhat.com>
Date: Tue, 21 Jul 2020 01:19:30 -0500
Subject: [PATCH] libmultipath: count pending paths as active on loads
When multipath loads a table, it signals to udev if there are no active
paths. Multipath wasn't counting pending paths as active. This meant
that if all the paths were pending, udev would treat the device as not
ready, and not run kpartx on it. Even if the pending paths later
because active and were reinstated, the kernel would not send a new
uevent, because from its point of view, they were always up.
The alternative would be to continue to treat them as failed in the udev
rules, but then also tell the kernel that they were down, so that it
would trigger a uevent when they were reinstated. However, this could
lead to newly created multipath devices failing IO, simply because the
path checkers hadn't returned yet. Having udev assume that the the
device is up, like the kernel does, seems like the safer option.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
---
libmultipath/devmapper.c | 3 ++-
libmultipath/structs.c | 7 +++++++
libmultipath/structs.h | 1 +
3 files changed, 10 insertions(+), 1 deletion(-)
diff --git a/libmultipath/devmapper.c b/libmultipath/devmapper.c
index f597ff8b..126cd728 100644
--- a/libmultipath/devmapper.c
+++ b/libmultipath/devmapper.c
@@ -417,7 +417,8 @@ static uint16_t build_udev_flags(const struct multipath *mpp, int reload)
/* DM_UDEV_DISABLE_LIBRARY_FALLBACK is added in dm_addmap */
return (mpp->skip_kpartx == SKIP_KPARTX_ON ?
MPATH_UDEV_NO_KPARTX_FLAG : 0) |
- ((count_active_paths(mpp) == 0 || mpp->ghost_delay_tick > 0) ?
+ ((count_active_pending_paths(mpp) == 0 ||
+ mpp->ghost_delay_tick > 0) ?
MPATH_UDEV_NO_PATHS_FLAG : 0) |
(reload && !mpp->force_udev_reload ?
MPATH_UDEV_RELOAD_FLAG : 0);
diff --git a/libmultipath/structs.c b/libmultipath/structs.c
index 3eac3d61..0d1f969d 100644
--- a/libmultipath/structs.c
+++ b/libmultipath/structs.c
@@ -491,6 +491,13 @@ int count_active_paths(const struct multipath *mpp)
return do_pathcount(mpp, states, 2);
}
+int count_active_pending_paths(const struct multipath *mpp)
+{
+ int states[] = {PATH_UP, PATH_GHOST, PATH_PENDING};
+
+ return do_pathcount(mpp, states, 3);
+}
+
int pathcmp(const struct pathgroup *pgp, const struct pathgroup *cpgp)
{
int i, j;
diff --git a/libmultipath/structs.h b/libmultipath/structs.h
index 0c03e711..917e4083 100644
--- a/libmultipath/structs.h
+++ b/libmultipath/structs.h
@@ -448,6 +448,7 @@ struct path * first_path (const struct multipath *mpp);
int pathcount (const struct multipath *, int);
int count_active_paths(const struct multipath *);
+int count_active_pending_paths(const struct multipath *);
int pathcmp (const struct pathgroup *, const struct pathgroup *);
int add_feature (char **, const char *);
int remove_feature (char **, const char *);
--
2.17.2