Comment 0 for bug 2042366

[Impact]

 * There are corner cases in which multipathd
   might segfault if device-mapper parameters
   for a multipath map are returned as NULL.

 * This is apparently a rare error condition,
   but it's been reported and fixed upstream.

 * However, older releases need the patch too,
   for another SRU, so we take it individually
   to perform synthetic tests on the commit.

 * commit 7439a41 ("dm_get_map: fix segfault
   when can't found target") [1]

[Test Steps]

 * Synthetic reproducer available with GDB,
   (comments #1 and #2).

 * Create a VM with a multipath disk.
 * Attach GDB to multipathd.
 * Break on the device-mapper function
   that returns a map's info/parameters.
 * Let it finish that function, and set
   parameters to NULL.
 * Let multipathd run.

 * Expected: multipathd continues to run.
 * Actual: multipathd hits a segfault.

[Regression Potential]

 * The change is just a NULL pointer check,
   which returns early from the function.

 * Theoretically, regressions would manifest
   in the dm_get_map() function, and likely
   show as incomplete information on status
   of multipath maps or paths.

 * It's not expected to regress to another
   segmentation fault, as the function path
   is shorter if the condition happens, and
   only accesses memory that was initialized
   and checked before (the dm_task variable).

 * Any regression should be easy to catch,
   as dm_get_map() runs periodically on the
   path checkers in multipathd (the daemon)
   and commands in multipath (the cli tool).

[Other Info]

 * The commit [1] is present in Mantic and later.

   $ git describe --contains 7439a41
   0.9.0^2~8

   $ rmadison -a source multipath-tools
   ...
    multipath-tools | 0.8.8-1ubuntu2.2 | lunar-proposed | source
    multipath-tools | 0.9.4-5ubuntu3 | mantic | source
    multipath-tools | 0.9.4-5ubuntu3 | noble | source

[1] https://github.com/opensvc/multipath-tools/commit/7439a41