cloud-utils-growpart: Workaround for timeout

Regression identified when using flock disk locking with 5.10 kernel and
systemd-239. During unlock_disk_and_settle(), udevadm settle will not
complete and eventually timeout after 2 minutes. When looking at the
systemd-udevd logs, the daemon crashes and produces the following stack
trace:

    Stack trace of thread 1531:
    #0  0x00007fd73d9be405 recvmsg (libpthread.so.0)
    #1  0x00007fd73dab33b8 udev_monitor_receive_device (libsystemd-shared-239.so)
    #2  0x0000600347316201 on_uevent (systemd-udevd)
    #3  0x0000600347316667 on_inotify (systemd-udevd)
    #4  0x00007fd73dbad6d7 source_dispatch (libsystemd-shared-239.so)
    #5  0x00007fd73dbaf4e5 sd_event_dispatch (libsystemd-shared-239.so)
    #6  0x00007fd73dbaf678 sd_event_run (libsystemd-shared-239.so)
    #7  0x00007fd73dbaf89f sd_event_loop (libsystemd-shared-239.so)
    #8  0x00006003473132df run (systemd-udevd)
    #9  0x00007fd73d80e133 __libc_start_main (libc.so.6)
    #10 0x0000600347313efe _start (systemd-udevd)

The failing behavior appears to be directly linked to the "exec FD"
actions. A quick way to replicate this issue in the repro environment:
    exec 9<>$disk
    exec 9>&-
    udevadm settle

This patch comments out the initial lock_disk() call, which makes
unlock_disk_and_settle() return early because ${FLOCK_DISK_FD} is not
set to a valid FD, avoiding the file descriptor actions that lead to
the failing behavior.

Note that this change does re-introduce the possibility of udev race
conditions during the disk operations, effectively reverting this
behavior to pre-0.32 behavior.

Signed-off-by: Chris Co <chrco@microsoft.com>

Author: christopherco

SPECS/cloud-utils-growpart/cloud-utils-growpart.spec

@@ -1,7 +1,7 @@
 Summary:        Shell script to auto detect free size on disk and grow partition.
 Name:           cloud-utils-growpart
 Version:        0.32
-Release:        1%{?dist}
+Release:        2%{?dist}
 License:        GPLv3
 Group:          System Environment
 Source0:        https://launchpad.net/cloud-utils/trunk/%{version}/+download/cloud-utils-%{version}.tar.gz
@@ -12,6 +12,7 @@ Requires:       gptfdisk
 Requires:       gawk
 Requires:       util-linux
 BuildArch:      noarch
+Patch0:         growpart-remove-flock-disk-locking.patch
 
 %description
 Cloud-utils brings in growpart script. This script is very useful for
@@ -20,6 +21,7 @@ This is generally used by cloud-init for disk space manangement on cloud images.
 
 %prep
 %setup -q -n cloud-utils-%{version}
+%patch0 -p1
 
 %build
 %install
@@ -34,6 +36,8 @@ cp man/growpart.* $RPM_BUILD_ROOT/%{_mandir}/man1/
 %doc %{_mandir}/man1/growpart.*
 
 %changelog
+*   Tue Apr 27 2021 Chris Co <chrco@microsoft.com> - 0.32-2
+-   Add patch to handle unexpected timeout
 *   Sat Mar 13 2021 Henry Beberman <henry.beberman@microsoft.com> 0.32-1
 -   Update to version 0.32 for more robust parsing of kernel version in growpart
 *   Sat May 09 2020 Nick Samson <nisamson@microsoft.com> 0.30-6

SPECS/cloud-utils-growpart/growpart-remove-flock-disk-locking.patch

@@ -0,0 +1,60 @@
+From 1baf9aed12e35e7c032499081ea1c9887e65626a Mon Sep 17 00:00:00 2001
+From: Chris Co <chrco@microsoft.com>
+Date: Wed, 28 Apr 2021 05:58:02 +0000
+Subject: [PATCH] growpart: remove flock disk locking
+
+Regression identified when using flock disk locking with 5.10 kernel and
+systemd-239. During unlock_disk_and_settle(), udevadm settle will not
+complete and eventually timeout after 2 minutes. When looking at the
+systemd-udevd logs, the daemon crashes and produces the following stack
+trace:
+
+    Stack trace of thread 1531:
+    #0  0x00007fd73d9be405 recvmsg (libpthread.so.0)
+    #1  0x00007fd73dab33b8 udev_monitor_receive_device (libsystemd-shared-239.so)
+    #2  0x0000600347316201 on_uevent (systemd-udevd)
+    #3  0x0000600347316667 on_inotify (systemd-udevd)
+    #4  0x00007fd73dbad6d7 source_dispatch (libsystemd-shared-239.so)
+    #5  0x00007fd73dbaf4e5 sd_event_dispatch (libsystemd-shared-239.so)
+    #6  0x00007fd73dbaf678 sd_event_run (libsystemd-shared-239.so)
+    #7  0x00007fd73dbaf89f sd_event_loop (libsystemd-shared-239.so)
+    #8  0x00006003473132df run (systemd-udevd)
+    #9  0x00007fd73d80e133 __libc_start_main (libc.so.6)
+    #10 0x0000600347313efe _start (systemd-udevd)
+
+The failing behavior appears to be directly linked to the "exec FD"
+actions. A quick way to replicate this issue in the repro environment:
+    exec 9<>$disk
+    exec 9>&-
+    udevadm settle
+
+This patch comments out the initial lock_disk() call, which makes
+unlock_disk_and_settle() return early because ${FLOCK_DISK_FD} is not
+set to a valid FD, avoiding the file descriptor actions that lead to
+the failing behavior.
+
+Note that this change does re-introduce the possibility of udev race
+conditions during the disk operations, effectively reverting this
+behavior to pre-0.32 behavior.
+
+Signed-off-by: Chris Co <chrco@microsoft.com>
+---
+ bin/growpart | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/bin/growpart b/bin/growpart
+index 994b258..90fa016 100755
+--- a/bin/growpart
++++ b/bin/growpart
+@@ -954,7 +954,7 @@ get_resizer "$format" "$resizer" ||
+ 	fail "failed to get a resizer for format '$format'"
+ resizer=$_RET
+ 
+-lock_disk $DISK
++#lock_disk $DISK
+ debug 1 "resizing $PART on $DISK using $resizer"
+ "$resizer"
+ ret=$?
+-- 
+2.17.1
+