[elbe-devel] [PATCH 01/10] elbepack: imgutils: properly wait for all udev events in losetup()

[Thomas Weißschuh]

grub-install called inside the target expects the /dev/disk/by-uuid/ links to
be set up properly. Otherwise plain device names are used, which in the case of
loop devices does not work at all.
Depending on the timing, the current code could block execution of udev and
therefore prevent the creation of the symlinks.

Reported-by: Thomas Bonnefille <thomas.bonnefille at bootlin.com>
Closes: https://github.com/Linutronix/elbe/issues/431
Fixes: c5e73aadb421 ("elbepack: imgutils: protect against race between losetup() and udev")
Signed-off-by: Thomas Weißschuh <thomas.weissschuh at linutronix.de>
---
 elbepack/imgutils.py                   | 28 ++++++++++++++++++++++++----
 newsfragments/+losetup-udev.bugfix.rst |  1 +
 2 files changed, 25 insertions(+), 4 deletions(-)

diff --git a/elbepack/imgutils.py b/elbepack/imgutils.py
index b060121c4590d4235a0fbc65c97c0b0c82cd95ce..4f4a5ac65d6dcf7424e13c950b65e621b6ac17af 100644
--- a/elbepack/imgutils.py
+++ b/elbepack/imgutils.py
@@ -4,11 +4,33 @@
 
 import contextlib
 import fcntl
+import pathlib
 import subprocess
 
 from elbepack.shellhelper import ELBE_LOGGING, do, run
 
 
+def _wait_on_udev_for_device_and_partitions(device):
+    # The callers expect the udev symlinks of the loop device and its
+    # partitions to be present.
+
+    device_name = pathlib.Path(device).name
+    with open(device) as f:
+        # The partition entries in /sys/class/blocks are created by the kernel
+        # and guaranteed to exist after "losetup" returns.
+        # However udev processing triggers a rescan of the partitions, removing
+        # the entries for a short time. Prevent udev from doing so while we iterate.
+        fcntl.flock(f, fcntl.LOCK_EX)
+        partitions = [
+            '/dev/' + entry.name
+            for entry in pathlib.Path('/sys/class/block', device_name).iterdir()
+            if entry.name.startswith(device_name)
+        ]
+
+    # All partitions need to be mentioned explicitly.
+    subprocess.check_call(['udevadm', 'wait', device, *partitions])
+
+
 @contextlib.contextmanager
 def losetup(dev, extra_args=[]):
     loopdev = run(
@@ -17,10 +39,8 @@ def losetup(dev, extra_args=[]):
     ).stdout.decode('ascii').rstrip('\n')
 
     try:
-        with open(loopdev) as f:
-            # protect against races with udev
-            fcntl.flock(f, fcntl.LOCK_EX)
-            yield loopdev
+        _wait_on_udev_for_device_and_partitions(loopdev)
+        yield loopdev
     finally:
         do(['losetup', '--detach', loopdev], check=False)
 
diff --git a/newsfragments/+losetup-udev.bugfix.rst b/newsfragments/+losetup-udev.bugfix.rst
new file mode 100644
index 0000000000000000000000000000000000000000..177b28f0725e20aebddea9d5260b7337d645e806
--- /dev/null
+++ b/newsfragments/+losetup-udev.bugfix.rst
@@ -0,0 +1 @@
+Fix synchronization between losetup and udev (2nd try).

-- 
2.48.1