[dm-devel] [PATCH] kpartx: use inode to identify loopback mounts

Risto Kankkunen risto.kankkunen at f-secure.com
Mon Jul 6 11:56:59 UTC 2015


This is a patch which I submitted to Launchpad 
(https://bugs.launchpad.net/ubuntu/+source/multipath-tools/+bug/1469143). I 
was told to submit it to this list instead.

Currently kpartx doesn't work correctly with relative path names, 
hardlinks, symlinks or paths longer than 63 characters. This is because 
it tries to identify mounts by a non-unique name it gives to the mount. 
The loopinfo struct contains the device and inode information of the 
backing image, it is better to identify mounts by those.
-------------- next part --------------
From: Risto Kankkunen <risto.kankkunen at f-secure.com>
Date: Mon, 29 Jun 2015 18:39:48 +0300
Subject: Use image file device/inode to find the correct loop device in kpartx

Previously kpartx used the "lo_name" field of struct loop_info to store 
and match the image file name. That field is not intended to store path 
names (it's not big enough). Therefore kpartx was unable to delete 
mappings to file paths longer than 63 characters. It also didn't 
properly handle relative file paths: two different files with identical
relative path names could not be mapped. Instead kpartx would modify
the existing mapping when seeing a new file with the same relative path.

The "loopinfo" structure contains the image file device and inode 
numbers. Use those to uniquely identify the correct loop device.

--- a/kpartx/lopart.c
+++ b/kpartx/lopart.c
@@ -103,6 +103,14 @@
 	int i, j, fd;
 	struct stat statbuf;
 	struct loop_info loopinfo;
+	dev_t file_dev;
+	ino_t file_ino;
+
+	if (stat (filename, &statbuf) != 0) {
+		return NULL;
+	}
+	file_dev = statbuf.st_dev;
+	file_ino = statbuf.st_ino;
 
 	for (j = 0; j < SIZE(loop_formats); j++) {
 
@@ -123,7 +131,7 @@
 				continue;
 			}
 
-			if (0 == strcmp(filename, loopinfo.lo_name)) {
+			if (loopinfo.lo_device == file_dev && loopinfo.lo_inode == file_ino) {
 				close (fd);
 				return xstrdup(dev); /*found */
 			}


More information about the dm-devel mailing list