Jean-Christian de Rivaz wrote: > Jean-Christian de Rivaz a écrit : >> The first 5MB device have, acoording to parted-2.3, a cylinder size >> (16065) not smaller than the device length (10240), causing the >> PedGeometry start_geom of the function _primary_constraint() (file >> libparted/labels/dos.c) to stay uninitialized. I suspect that the >> condition "cylinder_size < dev->length" have bee introduced for the >> memory-mapped test as explained into the comment above the >> condition. Unfortunately, it seem that the true real world contain >> at least a type of USB key sold with a storage device that go wrong >> with this condition. > > It seem that the only maintained source archive for GNU parted is > git://git.debian.org/git/parted/parted.git . Right ?
That is the so-called upstream repository. > I found the this bug was introduced by this commit: Thank you for the bug report and for tracking down the offending commit. That change ensures that this unusual condition is caught early enough. Before that change, passing a negative length to ped_geometry_init would have triggered a failed assertion or something similarly unpleasant. Thus, simply reverting it just reintroduces the original problem. > commit c79d91ec71882a1673daae0482aa90c514c63cc1 > Author: Jim Meyering <meyer...@redhat.com> > Date: Tue Mar 30 16:50:59 2010 +0200 > > dos: accommodate very small devices (useful for testing) > > * libparted/labels/dos.c (_primary_constraint): Don't pass a negative > "device_length" to ped_geometry_init when the underlying device > has fewer sectors than a "cylinder". Here's a patch to solve the problem. It relaxes the constraint to use a starting sector of "1", if necessary. Please let me know if it works for you: >From 9f5b0608611eed40ef33be2096f5d482710602e5 Mon Sep 17 00:00:00 2001 From: Jim Meyering <meyer...@redhat.com> Date: Tue, 9 Nov 2010 10:25:36 +0100 Subject: [PATCH] dos: fix a bug affecting very small devices (smaller than 1 cylinder) This bug was introduced in commit c79d91ec, "dos: accommodate very small devices (useful for testing)". * libparted/labels/dos.c (_primary_constraint): The bug was to skip setting start_geom for small devices. That led to a used- uninitialized bug in the subsequent ped_constraint_new call. The fix is to relax the constraint to use a starting sector of "1", if necessary. Report and diagnosis by Jean-Christian de Rivaz in http://thread.gmane.org/gmane.comp.gnu.parted.bugs/10178 * NEWS (Bug fixes): Mention it. --- NEWS | 4 ++++ libparted/labels/dos.c | 14 +++++++------- 2 files changed, 11 insertions(+), 7 deletions(-) diff --git a/NEWS b/NEWS index 5dd7dca..4979b90 100644 --- a/NEWS +++ b/NEWS @@ -14,6 +14,10 @@ GNU parted NEWS -*- outline -*- libparted now recognizes scsi disks with a high major (128-135) as scsi disks + an msdos partition table on a very small device (smaller than one cylinder) + is now recognized. [bug introduced in parted-2.2] + + * Noteworthy changes in release 2.3 (2010-05-28) [stable] ** New features diff --git a/libparted/labels/dos.c b/libparted/labels/dos.c index d9e7d4a..578180b 100644 --- a/libparted/labels/dos.c +++ b/libparted/labels/dos.c @@ -1713,13 +1713,13 @@ _primary_constraint (const PedDisk* disk, const PedCHSGeometry* bios_geom, dev->length - min_geom->end)) return NULL; } else { - /* Do not assume that length is larger than 1 cylinder's - worth of sectors. This is useful when testing with - a memory-mapped "disk" (a la scsi_debug) that is say, - 2048 sectors long. */ - if (cylinder_size < dev->length - && !ped_geometry_init (&start_geom, dev, cylinder_size, - dev->length - cylinder_size)) + /* Use cylinder_size as the starting sector number + when the device is large enough to accommodate that. + Otherwise, use sector 1. */ + PedSector start = (cylinder_size < dev->length + ? cylinder_size : 1); + if (!ped_geometry_init (&start_geom, dev, start, + dev->length - start)) return NULL; if (!ped_geometry_init (&end_geom, dev, 0, dev->length)) return NULL; -- 1.7.3.2.2.gaf77a _______________________________________________ bug-parted mailing list bug-parted@gnu.org http://lists.gnu.org/mailman/listinfo/bug-parted
