File kdump-prefer-by-path-and-device-mapper.patch of Package kdump

Overview Repositories Revisions Requests Users Attributes Meta

File kdump-prefer-by-path-and-device-mapper.patch of Package kdump

From: Petr Tesarik <ptesarik@suse.com>
Date: Mon, 16 Dec 2019 13:33:19 +0100
Subject: Prefer by-path and device-mapper aliases over kernel names
References: bsc#1101149, LTC#168532
Upsream: merged
Git-commit: f153c9ac3f1baf8d1cf66c901b41f7bff19ff528

Mounting by kernel names (e.g. /dev/sda) is generally broken, since
these names are allocated dynamically by the kernel and may change
after a panic kexec. This issue can be usually avoided by using a
more stable tag in /etc/fstab (e.g. UUID=xyz).

However, there are supported ways to mount a filesystem with no
corresponding line in /etc/fstab, and kdump uses /proc/mounts as
fallback. This file shows the block device using the name that was
given as argument to the mount syscall. This name is usually
translated to the kernel name by libblkid(3). As a result, it does
not reflect the original intention, e.g.:

ezekiel:~ # mount UUID=9A4D-9B2B /mnt/data/
    ezekiel:~ # grep /mnt/data /proc/mounts
    /dev/sda /mnt/data vfat rw,relatime,fmask=0022,dmask=0022,codepage=437,iocharset=iso8859-1,shortname=mixed,errors=remount-ro 0 0
    ezekiel:~ # grep /mnt/data /proc/self/mountinfo
    187 94 8:0 / /mnt/data rw,relatime shared:56 - vfat /dev/sda rw,fmask=0022,dmask=0022,codepage=437,iocharset=iso8859-1,shortname=mixed,errors=remount-ro

As the original intention cannot be recovered, it is probably
better to prefer the by-path alias. The semantic of this alias is
closest to the traditional UNIX use of raw block device names: it's
always the drive attached to a known physical connector, regardless
of which medium is inserted.

Signed-off-by: Petr Tesarik <ptesarik@suse.com>

---
 init/module-setup.sh       |    3 ++-
 init/setup-kdump.functions |   27 +++++++++++++++++++++++++++
 2 files changed, 29 insertions(+), 1 deletion(-)

--- a/init/module-setup.sh
+++ b/init/module-setup.sh
@@ -90,7 +90,8 @@ kdump_add_mnt() {
     mkdir -p "$initdir/etc"
     local _passno=2
     [ "${kdump_fstype[_idx]}" = nfs ] && _passno=0
-    echo "${kdump_dev[_idx]} ${kdump_mnt[_idx]} ${kdump_fstype[_idx]} ${kdump_opts[_idx]} 0 $_passno" >> "$initdir/etc/fstab"
+    _dev=$(kdump_mount_dev "${kdump_dev[_idx]}")
+    echo "$_dev ${kdump_mnt[_idx]} ${kdump_fstype[_idx]} ${kdump_opts[_idx]} 0 $_passno" >> "$initdir/etc/fstab"
 }
 
 check() {
--- a/init/setup-kdump.functions
+++ b/init/setup-kdump.functions
@@ -716,6 +716,33 @@ function kdump_get_targets()						   # {
 }									   # }}}
 
 #
+# Get a block device specification that is suitable for use as the
+# first column in /etc/fstab.
+# Since device node names may change after kexec, more stable symlink
+# are preferred (by-path alias or device mapper name).
+# Parameters:
+#   1) _dev: block device specification
+# Output:
+#   block device specification to be used in /etc/fstab
+function kdump_mount_dev()						   # {{{
+{
+    local _dev="$1"
+
+    if [ ! -L "$_dev" -a -b "$_dev" ] ; then
+	local _symlink
+	for _symlink in $(udevadm info --root --query=symlink "$_dev")
+	do
+	    case "$_symlink" in
+		*/by-path/*|*/mapper/*)
+		    _dev="$_symlink"
+		    break
+	    esac
+	done
+    fi
+    echo "$_dev"
+}									   # }}}
+
+#
 # Read and normalize /etc/fstab and /proc/mounts (if exists).
 # The following transformations are done:
 #   - initial TABs and SPACEs are removed

Places

File kdump-prefer-by-path-and-device-mapper.patch of Package kdump

Places